]>
Commit | Line | Data |
---|---|---|
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- | |
2 | // vim: ts=8 sw=2 smarttab | |
3 | ||
4 | #include "include/compat.h" | |
5 | #include <errno.h> | |
6 | #include <stdlib.h> | |
7 | #include <sys/types.h> | |
8 | #include <boost/algorithm/string.hpp> | |
9 | ||
10 | #include <boost/format.hpp> | |
11 | #include <boost/optional.hpp> | |
12 | #include <boost/utility/in_place_factory.hpp> | |
13 | ||
14 | #include "common/ceph_json.h" | |
15 | #include "common/utf8.h" | |
16 | ||
17 | #include "common/errno.h" | |
18 | #include "common/Formatter.h" | |
19 | #include "common/Throttle.h" | |
20 | #include "common/Finisher.h" | |
21 | ||
22 | #include "rgw_rados.h" | |
23 | #include "rgw_cache.h" | |
24 | #include "rgw_acl.h" | |
25 | #include "rgw_acl_s3.h" /* for dumping s3policy in debug log */ | |
26 | #include "rgw_metadata.h" | |
27 | #include "rgw_bucket.h" | |
28 | #include "rgw_rest_conn.h" | |
29 | #include "rgw_cr_rados.h" | |
30 | #include "rgw_cr_rest.h" | |
31 | ||
32 | #include "cls/rgw/cls_rgw_ops.h" | |
33 | #include "cls/rgw/cls_rgw_types.h" | |
34 | #include "cls/rgw/cls_rgw_client.h" | |
35 | #include "cls/rgw/cls_rgw_const.h" | |
36 | #include "cls/refcount/cls_refcount_client.h" | |
37 | #include "cls/version/cls_version_client.h" | |
38 | #include "cls/log/cls_log_client.h" | |
39 | #include "cls/statelog/cls_statelog_client.h" | |
40 | #include "cls/timeindex/cls_timeindex_client.h" | |
41 | #include "cls/lock/cls_lock_client.h" | |
42 | #include "cls/user/cls_user_client.h" | |
43 | #include "osd/osd_types.h" | |
44 | ||
45 | #include "rgw_tools.h" | |
46 | #include "rgw_coroutine.h" | |
47 | #include "rgw_compression.h" | |
48 | ||
49 | #undef fork // fails to compile RGWPeriod::fork() below | |
50 | ||
51 | #include "common/Clock.h" | |
52 | ||
53 | #include "include/rados/librados.hpp" | |
54 | using namespace librados; | |
55 | ||
56 | #include <string> | |
57 | #include <iostream> | |
58 | #include <vector> | |
59 | #include <atomic> | |
60 | #include <list> | |
61 | #include <map> | |
62 | #include "auth/Crypto.h" // get_random_bytes() | |
63 | ||
64 | #include "rgw_log.h" | |
65 | ||
66 | #include "rgw_gc.h" | |
67 | #include "rgw_lc.h" | |
68 | ||
69 | #include "rgw_object_expirer_core.h" | |
70 | #include "rgw_sync.h" | |
71 | #include "rgw_data_sync.h" | |
72 | #include "rgw_realm_watcher.h" | |
73 | #include "rgw_reshard.h" | |
74 | ||
75 | #include "compressor/Compressor.h" | |
76 | ||
77 | #define dout_context g_ceph_context | |
78 | #define dout_subsys ceph_subsys_rgw | |
79 | ||
80 | using namespace std; | |
81 | ||
82 | static string notify_oid_prefix = "notify"; | |
83 | static string *notify_oids = NULL; | |
84 | static string shadow_ns = "shadow"; | |
85 | static string dir_oid_prefix = ".dir."; | |
86 | static string default_storage_pool_suffix = "rgw.buckets.data"; | |
87 | static string default_bucket_index_pool_suffix = "rgw.buckets.index"; | |
88 | static string default_storage_extra_pool_suffix = "rgw.buckets.non-ec"; | |
89 | static string avail_pools = ".pools.avail"; | |
90 | ||
91 | static string zone_info_oid_prefix = "zone_info."; | |
92 | static string zone_names_oid_prefix = "zone_names."; | |
93 | static string region_info_oid_prefix = "region_info."; | |
94 | static string zone_group_info_oid_prefix = "zonegroup_info."; | |
95 | static string realm_names_oid_prefix = "realms_names."; | |
96 | static string realm_info_oid_prefix = "realms."; | |
97 | static string default_region_info_oid = "default.region"; | |
98 | static string default_zone_group_info_oid = "default.zonegroup"; | |
99 | static string period_info_oid_prefix = "periods."; | |
100 | static string period_latest_epoch_info_oid = ".latest_epoch"; | |
101 | static string region_map_oid = "region_map"; | |
102 | static string zonegroup_map_oid = "zonegroup_map"; | |
103 | static string log_lock_name = "rgw_log_lock"; | |
104 | static string default_realm_info_oid = "default.realm"; | |
105 | const string default_zonegroup_name = "default"; | |
106 | const string default_zone_name = "default"; | |
107 | static string zonegroup_names_oid_prefix = "zonegroups_names."; | |
108 | static RGWObjCategory main_category = RGW_OBJ_CATEGORY_MAIN; | |
109 | #define RGW_USAGE_OBJ_PREFIX "usage." | |
110 | #define FIRST_EPOCH 1 | |
111 | static string RGW_DEFAULT_ZONE_ROOT_POOL = "rgw.root"; | |
112 | static string RGW_DEFAULT_ZONEGROUP_ROOT_POOL = "rgw.root"; | |
113 | static string RGW_DEFAULT_REALM_ROOT_POOL = "rgw.root"; | |
114 | static string RGW_DEFAULT_PERIOD_ROOT_POOL = "rgw.root"; | |
115 | ||
116 | #define RGW_STATELOG_OBJ_PREFIX "statelog." | |
117 | ||
118 | #define dout_subsys ceph_subsys_rgw | |
119 | ||
120 | ||
121 | static bool rgw_get_obj_data_pool(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params, | |
122 | const string& placement_id, const rgw_obj& obj, rgw_pool *pool) | |
123 | { | |
124 | if (!zone_params.get_head_data_pool(placement_id, obj, pool)) { | |
125 | RGWZonePlacementInfo placement; | |
126 | if (!zone_params.get_placement(zonegroup.default_placement, &placement)) { | |
127 | return false; | |
128 | } | |
129 | ||
130 | if (!obj.in_extra_data) { | |
131 | *pool = placement.data_pool; | |
132 | } else { | |
133 | *pool = placement.get_data_extra_pool(); | |
134 | } | |
135 | } | |
136 | ||
137 | return true; | |
138 | } | |
139 | ||
140 | static bool rgw_obj_to_raw(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params, | |
141 | const string& placement_id, const rgw_obj& obj, rgw_raw_obj *raw_obj) | |
142 | { | |
143 | get_obj_bucket_and_oid_loc(obj, raw_obj->oid, raw_obj->loc); | |
144 | ||
145 | return rgw_get_obj_data_pool(zonegroup, zone_params, placement_id, obj, &raw_obj->pool); | |
146 | } | |
147 | ||
148 | rgw_raw_obj rgw_obj_select::get_raw_obj(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params) const | |
149 | { | |
150 | if (!is_raw) { | |
151 | rgw_raw_obj r; | |
152 | rgw_obj_to_raw(zonegroup, zone_params, placement_rule, obj, &r); | |
153 | return r; | |
154 | } | |
155 | return raw_obj; | |
156 | } | |
157 | ||
158 | rgw_raw_obj rgw_obj_select::get_raw_obj(RGWRados *store) const | |
159 | { | |
160 | if (!is_raw) { | |
161 | rgw_raw_obj r; | |
162 | store->obj_to_raw(placement_rule, obj, &r); | |
163 | return r; | |
164 | } | |
165 | return raw_obj; | |
166 | } | |
167 | ||
168 | int rgw_init_ioctx(librados::Rados *rados, const rgw_pool& pool, IoCtx& ioctx, bool create) | |
169 | { | |
170 | int r = rados->ioctx_create(pool.name.c_str(), ioctx); | |
171 | if (r == -ENOENT && create) { | |
172 | r = rados->pool_create(pool.name.c_str()); | |
173 | if (r == -ERANGE) { | |
174 | dout(0) | |
175 | << __func__ | |
176 | << " ERROR: librados::Rados::pool_create returned " << cpp_strerror(-r) | |
177 | << " (this can be due to a pool or placement group misconfiguration, e.g." | |
178 | << " pg_num < pgp_num or mon_max_pg_per_osd exceeded)" | |
179 | << dendl; | |
180 | } | |
181 | if (r < 0 && r != -EEXIST) { | |
182 | return r; | |
183 | } | |
184 | ||
185 | r = rados->ioctx_create(pool.name.c_str(), ioctx); | |
186 | if (r < 0) { | |
187 | return r; | |
188 | } | |
189 | ||
190 | r = ioctx.application_enable(pg_pool_t::APPLICATION_NAME_RGW, false); | |
191 | if (r < 0 && r != -EOPNOTSUPP) { | |
192 | return r; | |
193 | } | |
194 | } else if (r < 0) { | |
195 | return r; | |
196 | } | |
197 | if (!pool.ns.empty()) { | |
198 | ioctx.set_namespace(pool.ns); | |
199 | } | |
200 | return 0; | |
201 | } | |
202 | ||
203 | template<> | |
204 | void RGWObjectCtxImpl<rgw_obj, RGWObjState>::invalidate(rgw_obj& obj) { | |
205 | RWLock::WLocker wl(lock); | |
206 | auto iter = objs_state.find(obj); | |
207 | if (iter == objs_state.end()) { | |
208 | return; | |
209 | } | |
210 | bool is_atomic = iter->second.is_atomic; | |
211 | bool prefetch_data = iter->second.prefetch_data; | |
212 | ||
213 | objs_state.erase(iter); | |
214 | ||
215 | if (is_atomic || prefetch_data) { | |
216 | auto& s = objs_state[obj]; | |
217 | s.is_atomic = is_atomic; | |
218 | s.prefetch_data = prefetch_data; | |
219 | } | |
220 | } | |
221 | ||
222 | template<> | |
223 | void RGWObjectCtxImpl<rgw_raw_obj, RGWRawObjState>::invalidate(rgw_raw_obj& obj) { | |
224 | RWLock::WLocker wl(lock); | |
225 | auto iter = objs_state.find(obj); | |
226 | if (iter == objs_state.end()) { | |
227 | return; | |
228 | } | |
229 | ||
230 | objs_state.erase(iter); | |
231 | } | |
232 | ||
233 | void RGWDefaultZoneGroupInfo::dump(Formatter *f) const { | |
234 | encode_json("default_zonegroup", default_zonegroup, f); | |
235 | } | |
236 | ||
237 | void RGWDefaultZoneGroupInfo::decode_json(JSONObj *obj) { | |
238 | ||
239 | JSONDecoder::decode_json("default_zonegroup", default_zonegroup, obj); | |
240 | /* backward compatability with region */ | |
241 | if (default_zonegroup.empty()) { | |
242 | JSONDecoder::decode_json("default_region", default_zonegroup, obj); | |
243 | } | |
244 | } | |
245 | ||
246 | rgw_pool RGWZoneGroup::get_pool(CephContext *cct_) | |
247 | { | |
248 | if (cct_->_conf->rgw_zonegroup_root_pool.empty()) { | |
249 | return rgw_pool(RGW_DEFAULT_ZONEGROUP_ROOT_POOL); | |
250 | } | |
251 | ||
252 | return rgw_pool(cct_->_conf->rgw_zonegroup_root_pool); | |
253 | } | |
254 | ||
255 | int RGWZoneGroup::create_default(bool old_format) | |
256 | { | |
257 | name = default_zonegroup_name; | |
258 | is_master = true; | |
259 | ||
260 | RGWZoneGroupPlacementTarget placement_target; | |
261 | placement_target.name = "default-placement"; | |
262 | placement_targets[placement_target.name] = placement_target; | |
263 | default_placement = "default-placement"; | |
264 | ||
265 | RGWZoneParams zone_params(default_zone_name); | |
266 | ||
267 | int r = zone_params.init(cct, store, false); | |
268 | if (r < 0) { | |
269 | ldout(cct, 0) << "create_default: error initializing zone params: " << cpp_strerror(-r) << dendl; | |
270 | return r; | |
271 | } | |
272 | ||
273 | r = zone_params.create_default(); | |
274 | if (r < 0 && r != -EEXIST) { | |
275 | ldout(cct, 0) << "create_default: error in create_default zone params: " << cpp_strerror(-r) << dendl; | |
276 | return r; | |
277 | } else if (r == -EEXIST) { | |
278 | ldout(cct, 10) << "zone_params::create_default() returned -EEXIST, we raced with another default zone_params creation" << dendl; | |
279 | zone_params.clear_id(); | |
280 | r = zone_params.init(cct, store); | |
281 | if (r < 0) { | |
282 | ldout(cct, 0) << "create_default: error in init existing zone params: " << cpp_strerror(-r) << dendl; | |
283 | return r; | |
284 | } | |
285 | ldout(cct, 20) << "zone_params::create_default() " << zone_params.get_name() << " id " << zone_params.get_id() | |
286 | << dendl; | |
287 | } | |
288 | ||
289 | RGWZone& default_zone = zones[zone_params.get_id()]; | |
290 | default_zone.name = zone_params.get_name(); | |
291 | default_zone.id = zone_params.get_id(); | |
292 | master_zone = default_zone.id; | |
293 | ||
294 | r = create(); | |
295 | if (r < 0 && r != -EEXIST) { | |
296 | ldout(cct, 0) << "error storing zone group info: " << cpp_strerror(-r) << dendl; | |
297 | return r; | |
298 | } | |
299 | ||
300 | if (r == -EEXIST) { | |
301 | ldout(cct, 10) << "create_default() returned -EEXIST, we raced with another zonegroup creation" << dendl; | |
302 | id.clear(); | |
303 | r = init(cct, store); | |
304 | if (r < 0) { | |
305 | return r; | |
306 | } | |
307 | } | |
308 | ||
309 | if (old_format) { | |
310 | name = id; | |
311 | } | |
312 | ||
313 | post_process_params(); | |
314 | ||
315 | return 0; | |
316 | } | |
317 | ||
318 | const string RGWZoneGroup::get_default_oid(bool old_region_format) | |
319 | { | |
320 | if (old_region_format) { | |
321 | if (cct->_conf->rgw_default_region_info_oid.empty()) { | |
322 | return default_region_info_oid; | |
323 | } | |
324 | return cct->_conf->rgw_default_region_info_oid; | |
325 | } | |
326 | ||
327 | string default_oid = cct->_conf->rgw_default_zonegroup_info_oid; | |
328 | ||
329 | if (cct->_conf->rgw_default_zonegroup_info_oid.empty()) { | |
330 | default_oid = default_zone_group_info_oid; | |
331 | } | |
332 | ||
333 | default_oid += "." + realm_id; | |
334 | ||
335 | return default_oid; | |
336 | } | |
337 | ||
338 | const string& RGWZoneGroup::get_info_oid_prefix(bool old_region_format) | |
339 | { | |
340 | if (old_region_format) { | |
341 | return region_info_oid_prefix; | |
342 | } | |
343 | return zone_group_info_oid_prefix; | |
344 | } | |
345 | ||
346 | const string& RGWZoneGroup::get_names_oid_prefix() | |
347 | { | |
348 | return zonegroup_names_oid_prefix; | |
349 | } | |
350 | ||
351 | const string& RGWZoneGroup::get_predefined_name(CephContext *cct) { | |
352 | return cct->_conf->rgw_zonegroup; | |
353 | } | |
354 | ||
355 | int RGWZoneGroup::equals(const string& other_zonegroup) const | |
356 | { | |
357 | if (is_master && other_zonegroup.empty()) | |
358 | return true; | |
359 | ||
360 | return (id == other_zonegroup); | |
361 | } | |
362 | ||
363 | int RGWZoneGroup::add_zone(const RGWZoneParams& zone_params, bool *is_master, bool *read_only, | |
364 | const list<string>& endpoints, const string *ptier_type, | |
365 | bool *psync_from_all, list<string>& sync_from, list<string>& sync_from_rm) | |
366 | { | |
367 | auto& zone_id = zone_params.get_id(); | |
368 | auto& zone_name = zone_params.get_name(); | |
369 | ||
370 | // check for duplicate zone name on insert | |
371 | if (!zones.count(zone_id)) { | |
372 | for (const auto& zone : zones) { | |
373 | if (zone.second.name == zone_name) { | |
374 | ldout(cct, 0) << "ERROR: found existing zone name " << zone_name | |
375 | << " (" << zone.first << ") in zonegroup " << get_name() << dendl; | |
376 | return -EEXIST; | |
377 | } | |
378 | } | |
379 | } | |
380 | ||
381 | if (is_master) { | |
382 | if (*is_master) { | |
383 | if (!master_zone.empty() && master_zone != zone_params.get_id()) { | |
384 | ldout(cct, 0) << "NOTICE: overriding master zone: " << master_zone << dendl; | |
385 | } | |
386 | master_zone = zone_params.get_id(); | |
387 | } else if (master_zone == zone_params.get_id()) { | |
388 | master_zone.clear(); | |
389 | } | |
390 | } | |
391 | ||
392 | RGWZone& zone = zones[zone_params.get_id()]; | |
393 | zone.name = zone_params.get_name(); | |
394 | zone.id = zone_params.get_id(); | |
395 | if (!endpoints.empty()) { | |
396 | zone.endpoints = endpoints; | |
397 | } | |
398 | if (read_only) { | |
399 | zone.read_only = *read_only; | |
400 | } | |
401 | if (ptier_type) { | |
402 | zone.tier_type = *ptier_type; | |
403 | } | |
404 | ||
405 | if (psync_from_all) { | |
406 | zone.sync_from_all = *psync_from_all; | |
407 | } | |
408 | ||
409 | for (auto add : sync_from) { | |
410 | zone.sync_from.insert(add); | |
411 | } | |
412 | ||
413 | for (auto rm : sync_from_rm) { | |
414 | zone.sync_from.erase(rm); | |
415 | } | |
416 | ||
417 | post_process_params(); | |
418 | ||
419 | return update(); | |
420 | } | |
421 | ||
422 | ||
423 | int RGWZoneGroup::rename_zone(const RGWZoneParams& zone_params) | |
424 | { | |
425 | RGWZone& zone = zones[zone_params.get_id()]; | |
426 | zone.name = zone_params.get_name(); | |
427 | ||
428 | return update(); | |
429 | } | |
430 | ||
431 | void RGWZoneGroup::post_process_params() | |
432 | { | |
433 | bool log_data = zones.size() > 1; | |
434 | ||
435 | if (master_zone.empty()) { | |
436 | map<string, RGWZone>::iterator iter = zones.begin(); | |
437 | if (iter != zones.end()) { | |
438 | master_zone = iter->first; | |
439 | } | |
440 | } | |
441 | ||
442 | for (map<string, RGWZone>::iterator iter = zones.begin(); iter != zones.end(); ++iter) { | |
443 | RGWZone& zone = iter->second; | |
444 | zone.log_data = log_data; | |
445 | ||
446 | RGWZoneParams zone_params(zone.id, zone.name); | |
447 | int ret = zone_params.init(cct, store); | |
448 | if (ret < 0) { | |
449 | ldout(cct, 0) << "WARNING: could not read zone params for zone id=" << zone.id << " name=" << zone.name << dendl; | |
450 | continue; | |
451 | } | |
452 | ||
453 | for (map<string, RGWZonePlacementInfo>::iterator iter = zone_params.placement_pools.begin(); | |
454 | iter != zone_params.placement_pools.end(); ++iter) { | |
455 | const string& placement_name = iter->first; | |
456 | if (placement_targets.find(placement_name) == placement_targets.end()) { | |
457 | RGWZoneGroupPlacementTarget placement_target; | |
458 | placement_target.name = placement_name; | |
459 | placement_targets[placement_name] = placement_target; | |
460 | } | |
461 | } | |
462 | } | |
463 | ||
464 | if (default_placement.empty() && !placement_targets.empty()) { | |
465 | default_placement = placement_targets.begin()->first; | |
466 | } | |
467 | } | |
468 | ||
469 | int RGWZoneGroup::remove_zone(const std::string& zone_id) | |
470 | { | |
471 | map<string, RGWZone>::iterator iter = zones.find(zone_id); | |
472 | if (iter == zones.end()) { | |
473 | ldout(cct, 0) << "zone id " << zone_id << " is not a part of zonegroup " | |
474 | << name << dendl; | |
475 | return -ENOENT; | |
476 | } | |
477 | ||
478 | zones.erase(iter); | |
479 | ||
480 | post_process_params(); | |
481 | ||
482 | return update(); | |
483 | } | |
484 | ||
485 | int RGWZoneGroup::read_default_id(string& default_id, bool old_format) | |
486 | { | |
487 | if (realm_id.empty()) { | |
488 | /* try using default realm */ | |
489 | RGWRealm realm; | |
490 | int ret = realm.init(cct, store); | |
491 | // no default realm exist | |
492 | if (ret < 0) { | |
493 | return read_id(default_zonegroup_name, default_id); | |
494 | } | |
495 | realm_id = realm.get_id(); | |
496 | } | |
497 | ||
498 | return RGWSystemMetaObj::read_default_id(default_id, old_format); | |
499 | } | |
500 | ||
501 | int RGWZoneGroup::set_as_default(bool exclusive) | |
502 | { | |
503 | if (realm_id.empty()) { | |
504 | /* try using default realm */ | |
505 | RGWRealm realm; | |
506 | int ret = realm.init(cct, store); | |
507 | if (ret < 0) { | |
508 | ldout(cct, 10) << "could not read realm id: " << cpp_strerror(-ret) << dendl; | |
509 | return -EINVAL; | |
510 | } | |
511 | realm_id = realm.get_id(); | |
512 | } | |
513 | ||
514 | return RGWSystemMetaObj::set_as_default(exclusive); | |
515 | } | |
516 | ||
517 | int RGWSystemMetaObj::init(CephContext *_cct, RGWRados *_store, bool setup_obj, bool old_format) | |
518 | { | |
519 | cct = _cct; | |
520 | store = _store; | |
521 | ||
522 | if (!setup_obj) | |
523 | return 0; | |
524 | ||
525 | if (old_format && id.empty()) { | |
526 | id = name; | |
527 | } | |
528 | ||
529 | if (id.empty()) { | |
530 | int r; | |
531 | if (name.empty()) { | |
532 | name = get_predefined_name(cct); | |
533 | } | |
534 | if (name.empty()) { | |
535 | r = use_default(old_format); | |
536 | if (r < 0) { | |
537 | return r; | |
538 | } | |
539 | } else if (!old_format) { | |
540 | r = read_id(name, id); | |
541 | if (r < 0) { | |
542 | if (r != -ENOENT) { | |
543 | ldout(cct, 0) << "error in read_id for object name: " << name << " : " << cpp_strerror(-r) << dendl; | |
544 | } | |
545 | return r; | |
546 | } | |
547 | } | |
548 | } | |
549 | ||
550 | return read_info(id, old_format); | |
551 | } | |
552 | ||
553 | int RGWSystemMetaObj::read_default(RGWDefaultSystemMetaObjInfo& default_info, const string& oid) | |
554 | { | |
555 | auto pool = get_pool(cct); | |
556 | bufferlist bl; | |
557 | RGWObjectCtx obj_ctx(store); | |
558 | int ret = rgw_get_system_obj(store, obj_ctx, pool, oid, bl, NULL, NULL); | |
559 | if (ret < 0) | |
560 | return ret; | |
561 | ||
562 | try { | |
563 | bufferlist::iterator iter = bl.begin(); | |
564 | ::decode(default_info, iter); | |
565 | } catch (buffer::error& err) { | |
566 | ldout(cct, 0) << "error decoding data from " << pool << ":" << oid << dendl; | |
567 | return -EIO; | |
568 | } | |
569 | ||
570 | return 0; | |
571 | } | |
572 | ||
573 | int RGWSystemMetaObj::read_default_id(string& default_id, bool old_format) | |
574 | { | |
575 | RGWDefaultSystemMetaObjInfo default_info; | |
576 | ||
577 | int ret = read_default(default_info, get_default_oid(old_format)); | |
578 | if (ret < 0) { | |
579 | return ret; | |
580 | } | |
581 | ||
582 | default_id = default_info.default_id; | |
583 | ||
584 | return 0; | |
585 | } | |
586 | ||
587 | int RGWSystemMetaObj::use_default(bool old_format) | |
588 | { | |
589 | return read_default_id(id, old_format); | |
590 | } | |
591 | ||
592 | int RGWSystemMetaObj::set_as_default(bool exclusive) | |
593 | { | |
594 | string oid = get_default_oid(); | |
595 | ||
596 | rgw_pool pool(get_pool(cct)); | |
597 | bufferlist bl; | |
598 | ||
599 | RGWDefaultSystemMetaObjInfo default_info; | |
600 | default_info.default_id = id; | |
601 | ||
602 | ::encode(default_info, bl); | |
603 | ||
604 | int ret = rgw_put_system_obj(store, pool, oid, bl.c_str(), bl.length(), | |
605 | exclusive, NULL, real_time(), NULL); | |
606 | if (ret < 0) | |
607 | return ret; | |
608 | ||
609 | return 0; | |
610 | } | |
611 | ||
612 | int RGWSystemMetaObj::read_id(const string& obj_name, string& object_id) | |
613 | { | |
614 | rgw_pool pool(get_pool(cct)); | |
615 | bufferlist bl; | |
616 | ||
617 | string oid = get_names_oid_prefix() + obj_name; | |
618 | ||
619 | RGWObjectCtx obj_ctx(store); | |
620 | int ret = rgw_get_system_obj(store, obj_ctx, pool, oid, bl, NULL, NULL); | |
621 | if (ret < 0) { | |
622 | return ret; | |
623 | } | |
624 | ||
625 | RGWNameToId nameToId; | |
626 | try { | |
627 | bufferlist::iterator iter = bl.begin(); | |
628 | ::decode(nameToId, iter); | |
629 | } catch (buffer::error& err) { | |
630 | ldout(cct, 0) << "ERROR: failed to decode obj from " << pool << ":" << oid << dendl; | |
631 | return -EIO; | |
632 | } | |
633 | object_id = nameToId.obj_id; | |
634 | return 0; | |
635 | } | |
636 | ||
637 | int RGWSystemMetaObj::delete_obj(bool old_format) | |
638 | { | |
639 | rgw_pool pool(get_pool(cct)); | |
640 | ||
641 | /* check to see if obj is the default */ | |
642 | RGWDefaultSystemMetaObjInfo default_info; | |
643 | int ret = read_default(default_info, get_default_oid(old_format)); | |
644 | if (ret < 0 && ret != -ENOENT) | |
645 | return ret; | |
646 | if (default_info.default_id == id || (old_format && default_info.default_id == name)) { | |
647 | string oid = get_default_oid(old_format); | |
648 | rgw_raw_obj default_named_obj(pool, oid); | |
649 | ret = store->delete_system_obj(default_named_obj); | |
650 | if (ret < 0) { | |
651 | ldout(cct, 0) << "Error delete default obj name " << name << ": " << cpp_strerror(-ret) << dendl; | |
652 | return ret; | |
653 | } | |
654 | } | |
655 | if (!old_format) { | |
656 | string oid = get_names_oid_prefix() + name; | |
657 | rgw_raw_obj object_name(pool, oid); | |
658 | ret = store->delete_system_obj(object_name); | |
659 | if (ret < 0) { | |
660 | ldout(cct, 0) << "Error delete obj name " << name << ": " << cpp_strerror(-ret) << dendl; | |
661 | return ret; | |
662 | } | |
663 | } | |
664 | ||
665 | string oid = get_info_oid_prefix(old_format); | |
666 | if (old_format) { | |
667 | oid += name; | |
668 | } else { | |
669 | oid += id; | |
670 | } | |
671 | ||
672 | rgw_raw_obj object_id(pool, oid); | |
673 | ret = store->delete_system_obj(object_id); | |
674 | if (ret < 0) { | |
675 | ldout(cct, 0) << "Error delete object id " << id << ": " << cpp_strerror(-ret) << dendl; | |
676 | } | |
677 | ||
678 | return ret; | |
679 | } | |
680 | ||
681 | int RGWSystemMetaObj::store_name(bool exclusive) | |
682 | { | |
683 | rgw_pool pool(get_pool(cct)); | |
684 | string oid = get_names_oid_prefix() + name; | |
685 | ||
686 | RGWNameToId nameToId; | |
687 | nameToId.obj_id = id; | |
688 | ||
689 | bufferlist bl; | |
690 | ::encode(nameToId, bl); | |
691 | return rgw_put_system_obj(store, pool, oid, bl.c_str(), bl.length(), exclusive, NULL, real_time(), NULL); | |
692 | } | |
693 | ||
694 | int RGWSystemMetaObj::rename(const string& new_name) | |
695 | { | |
696 | string new_id; | |
697 | int ret = read_id(new_name, new_id); | |
698 | if (!ret) { | |
699 | return -EEXIST; | |
700 | } | |
701 | if (ret < 0 && ret != -ENOENT) { | |
702 | ldout(cct, 0) << "Error read_id " << new_name << ": " << cpp_strerror(-ret) << dendl; | |
703 | return ret; | |
704 | } | |
705 | string old_name = name; | |
706 | name = new_name; | |
707 | ret = update(); | |
708 | if (ret < 0) { | |
709 | ldout(cct, 0) << "Error storing new obj info " << new_name << ": " << cpp_strerror(-ret) << dendl; | |
710 | return ret; | |
711 | } | |
712 | ret = store_name(true); | |
713 | if (ret < 0) { | |
714 | ldout(cct, 0) << "Error storing new name " << new_name << ": " << cpp_strerror(-ret) << dendl; | |
715 | return ret; | |
716 | } | |
717 | /* delete old name */ | |
718 | rgw_pool pool(get_pool(cct)); | |
719 | string oid = get_names_oid_prefix() + old_name; | |
720 | rgw_raw_obj old_name_obj(pool, oid); | |
721 | ret = store->delete_system_obj(old_name_obj); | |
722 | if (ret < 0) { | |
723 | ldout(cct, 0) << "Error delete old obj name " << old_name << ": " << cpp_strerror(-ret) << dendl; | |
724 | return ret; | |
725 | } | |
726 | ||
727 | return ret; | |
728 | } | |
729 | ||
730 | int RGWSystemMetaObj::read_info(const string& obj_id, bool old_format) | |
731 | { | |
732 | rgw_pool pool(get_pool(cct)); | |
733 | ||
734 | bufferlist bl; | |
735 | ||
736 | string oid = get_info_oid_prefix(old_format) + obj_id; | |
737 | ||
738 | RGWObjectCtx obj_ctx(store); | |
739 | int ret = rgw_get_system_obj(store, obj_ctx, pool, oid, bl, NULL, NULL); | |
740 | if (ret < 0) { | |
741 | ldout(cct, 0) << "failed reading obj info from " << pool << ":" << oid << ": " << cpp_strerror(-ret) << dendl; | |
742 | return ret; | |
743 | } | |
744 | ||
745 | try { | |
746 | bufferlist::iterator iter = bl.begin(); | |
747 | ::decode(*this, iter); | |
748 | } catch (buffer::error& err) { | |
749 | ldout(cct, 0) << "ERROR: failed to decode obj from " << pool << ":" << oid << dendl; | |
750 | return -EIO; | |
751 | } | |
752 | ||
753 | return 0; | |
754 | } | |
755 | ||
756 | int RGWSystemMetaObj::read() | |
757 | { | |
758 | int ret = read_id(name, id); | |
759 | if (ret < 0) { | |
760 | return ret; | |
761 | } | |
762 | ||
763 | return read_info(id); | |
764 | } | |
765 | ||
766 | int RGWSystemMetaObj::create(bool exclusive) | |
767 | { | |
768 | int ret; | |
769 | ||
770 | /* check to see the name is not used */ | |
771 | ret = read_id(name, id); | |
772 | if (exclusive && ret == 0) { | |
773 | ldout(cct, 10) << "ERROR: name " << name << " already in use for obj id " << id << dendl; | |
774 | return -EEXIST; | |
775 | } else if ( ret < 0 && ret != -ENOENT) { | |
776 | ldout(cct, 0) << "failed reading obj id " << id << ": " << cpp_strerror(-ret) << dendl; | |
777 | return ret; | |
778 | } | |
779 | ||
780 | if (id.empty()) { | |
781 | /* create unique id */ | |
782 | uuid_d new_uuid; | |
783 | char uuid_str[37]; | |
784 | new_uuid.generate_random(); | |
785 | new_uuid.print(uuid_str); | |
786 | id = uuid_str; | |
787 | } | |
788 | ||
789 | ret = store_info(exclusive); | |
790 | if (ret < 0) { | |
791 | ldout(cct, 0) << "ERROR: storing info for " << id << ": " << cpp_strerror(-ret) << dendl; | |
792 | return ret; | |
793 | } | |
794 | ||
795 | return store_name(exclusive); | |
796 | } | |
797 | ||
798 | int RGWSystemMetaObj::store_info(bool exclusive) | |
799 | { | |
800 | rgw_pool pool(get_pool(cct)); | |
801 | ||
802 | string oid = get_info_oid_prefix() + id; | |
803 | ||
804 | bufferlist bl; | |
805 | ::encode(*this, bl); | |
806 | return rgw_put_system_obj(store, pool, oid, bl.c_str(), bl.length(), exclusive, NULL, real_time(), NULL); | |
807 | } | |
808 | ||
809 | int RGWSystemMetaObj::write(bool exclusive) | |
810 | { | |
811 | int ret = store_info(exclusive); | |
812 | if (ret < 0) { | |
813 | ldout(cct, 20) << __func__ << "(): store_info() returned ret=" << ret << dendl; | |
814 | return ret; | |
815 | } | |
816 | ret = store_name(exclusive); | |
817 | if (ret < 0) { | |
818 | ldout(cct, 20) << __func__ << "(): store_name() returned ret=" << ret << dendl; | |
819 | return ret; | |
820 | } | |
821 | return 0; | |
822 | } | |
823 | ||
824 | ||
825 | const string& RGWRealm::get_predefined_name(CephContext *cct) { | |
826 | return cct->_conf->rgw_realm; | |
827 | } | |
828 | ||
829 | int RGWRealm::create(bool exclusive) | |
830 | { | |
831 | int ret = RGWSystemMetaObj::create(exclusive); | |
832 | if (ret < 0) { | |
833 | ldout(cct, 0) << "ERROR creating new realm object " << name << ": " << cpp_strerror(-ret) << dendl; | |
834 | return ret; | |
835 | } | |
836 | // create the control object for watch/notify | |
837 | ret = create_control(exclusive); | |
838 | if (ret < 0) { | |
839 | ldout(cct, 0) << "ERROR creating control for new realm " << name << ": " << cpp_strerror(-ret) << dendl; | |
840 | return ret; | |
841 | } | |
842 | RGWPeriod period; | |
843 | if (current_period.empty()) { | |
844 | /* create new period for the realm */ | |
845 | ret = period.init(cct, store, id, name, false); | |
846 | if (ret < 0 ) { | |
847 | return ret; | |
848 | } | |
849 | ret = period.create(true); | |
850 | if (ret < 0) { | |
851 | ldout(cct, 0) << "ERROR: creating new period for realm " << name << ": " << cpp_strerror(-ret) << dendl; | |
852 | return ret; | |
853 | } | |
854 | } else { | |
855 | period = RGWPeriod(current_period, 0); | |
856 | int ret = period.init(cct, store, id, name); | |
857 | if (ret < 0) { | |
858 | ldout(cct, 0) << "ERROR: failed to init period " << current_period << dendl; | |
859 | return ret; | |
860 | } | |
861 | } | |
862 | ret = set_current_period(period); | |
863 | if (ret < 0) { | |
864 | ldout(cct, 0) << "ERROR: failed set current period " << current_period << dendl; | |
865 | return ret; | |
866 | } | |
867 | // try to set as default. may race with another create, so pass exclusive=true | |
868 | // so we don't override an existing default | |
869 | ret = set_as_default(true); | |
870 | if (ret < 0 && ret != -EEXIST) { | |
871 | ldout(cct, 0) << "WARNING: failed to set realm as default realm, ret=" << ret << dendl; | |
872 | } | |
873 | ||
874 | return 0; | |
875 | } | |
876 | ||
877 | int RGWRealm::delete_obj() | |
878 | { | |
879 | int ret = RGWSystemMetaObj::delete_obj(); | |
880 | if (ret < 0) { | |
881 | return ret; | |
882 | } | |
883 | return delete_control(); | |
884 | } | |
885 | ||
886 | int RGWRealm::create_control(bool exclusive) | |
887 | { | |
888 | auto pool = rgw_pool{get_pool(cct)}; | |
889 | auto oid = get_control_oid(); | |
890 | return rgw_put_system_obj(store, pool, oid, nullptr, 0, exclusive, | |
891 | nullptr, real_time(), nullptr); | |
892 | } | |
893 | ||
894 | int RGWRealm::delete_control() | |
895 | { | |
896 | auto pool = rgw_pool{get_pool(cct)}; | |
897 | auto obj = rgw_raw_obj{pool, get_control_oid()}; | |
898 | return store->delete_system_obj(obj); | |
899 | } | |
900 | ||
901 | rgw_pool RGWRealm::get_pool(CephContext *cct) | |
902 | { | |
903 | if (cct->_conf->rgw_realm_root_pool.empty()) { | |
904 | return rgw_pool(RGW_DEFAULT_REALM_ROOT_POOL); | |
905 | } | |
906 | return rgw_pool(cct->_conf->rgw_realm_root_pool); | |
907 | } | |
908 | ||
909 | const string RGWRealm::get_default_oid(bool old_format) | |
910 | { | |
911 | if (cct->_conf->rgw_default_realm_info_oid.empty()) { | |
912 | return default_realm_info_oid; | |
913 | } | |
914 | return cct->_conf->rgw_default_realm_info_oid; | |
915 | } | |
916 | ||
917 | const string& RGWRealm::get_names_oid_prefix() | |
918 | { | |
919 | return realm_names_oid_prefix; | |
920 | } | |
921 | ||
922 | const string& RGWRealm::get_info_oid_prefix(bool old_format) | |
923 | { | |
924 | return realm_info_oid_prefix; | |
925 | } | |
926 | ||
927 | int RGWRealm::set_current_period(RGWPeriod& period) | |
928 | { | |
929 | // update realm epoch to match the period's | |
930 | if (epoch > period.get_realm_epoch()) { | |
931 | ldout(cct, 0) << "ERROR: set_current_period with old realm epoch " | |
932 | << period.get_realm_epoch() << ", current epoch=" << epoch << dendl; | |
933 | return -EINVAL; | |
934 | } | |
935 | if (epoch == period.get_realm_epoch() && current_period != period.get_id()) { | |
936 | ldout(cct, 0) << "ERROR: set_current_period with same realm epoch " | |
937 | << period.get_realm_epoch() << ", but different period id " | |
938 | << period.get_id() << " != " << current_period << dendl; | |
939 | return -EINVAL; | |
940 | } | |
941 | ||
942 | epoch = period.get_realm_epoch(); | |
943 | current_period = period.get_id(); | |
944 | ||
945 | int ret = update(); | |
946 | if (ret < 0) { | |
947 | ldout(cct, 0) << "ERROR: period update: " << cpp_strerror(-ret) << dendl; | |
948 | return ret; | |
949 | } | |
950 | ||
951 | ret = period.reflect(); | |
952 | if (ret < 0) { | |
953 | ldout(cct, 0) << "ERROR: period.reflect(): " << cpp_strerror(-ret) << dendl; | |
954 | return ret; | |
955 | } | |
956 | ||
957 | return 0; | |
958 | } | |
959 | ||
960 | string RGWRealm::get_control_oid() | |
961 | { | |
962 | return get_info_oid_prefix() + id + ".control"; | |
963 | } | |
964 | ||
965 | int RGWRealm::notify_zone(bufferlist& bl) | |
966 | { | |
967 | // open a context on the realm's pool | |
968 | rgw_pool pool{get_pool(cct)}; | |
969 | librados::IoCtx ctx; | |
970 | int r = rgw_init_ioctx(store->get_rados_handle(), pool, ctx); | |
971 | if (r < 0) { | |
972 | ldout(cct, 0) << "Failed to open pool " << pool << dendl; | |
973 | return r; | |
974 | } | |
975 | // send a notify on the realm object | |
976 | r = ctx.notify2(get_control_oid(), bl, 0, nullptr); | |
977 | if (r < 0) { | |
978 | ldout(cct, 0) << "Realm notify failed with " << r << dendl; | |
979 | return r; | |
980 | } | |
981 | return 0; | |
982 | } | |
983 | ||
984 | int RGWRealm::notify_new_period(const RGWPeriod& period) | |
985 | { | |
986 | bufferlist bl; | |
987 | // push the period to dependent zonegroups/zones | |
988 | ::encode(RGWRealmNotify::ZonesNeedPeriod, bl); | |
989 | ::encode(period, bl); | |
990 | // reload the gateway with the new period | |
991 | ::encode(RGWRealmNotify::Reload, bl); | |
992 | ||
993 | return notify_zone(bl); | |
994 | } | |
995 | ||
996 | std::string RGWPeriodConfig::get_oid(const std::string& realm_id) | |
997 | { | |
998 | if (realm_id.empty()) { | |
999 | return "period_config.default"; | |
1000 | } | |
1001 | return "period_config." + realm_id; | |
1002 | } | |
1003 | ||
1004 | rgw_pool RGWPeriodConfig::get_pool(CephContext *cct) | |
1005 | { | |
1006 | const auto& pool_name = cct->_conf->rgw_period_root_pool; | |
1007 | if (pool_name.empty()) { | |
1008 | return {RGW_DEFAULT_PERIOD_ROOT_POOL}; | |
1009 | } | |
1010 | return {pool_name}; | |
1011 | } | |
1012 | ||
1013 | int RGWPeriodConfig::read(RGWRados *store, const std::string& realm_id) | |
1014 | { | |
1015 | RGWObjectCtx obj_ctx(store); | |
1016 | const auto& pool = get_pool(store->ctx()); | |
1017 | const auto& oid = get_oid(realm_id); | |
1018 | bufferlist bl; | |
1019 | ||
1020 | int ret = rgw_get_system_obj(store, obj_ctx, pool, oid, bl, nullptr, nullptr); | |
1021 | if (ret < 0) { | |
1022 | return ret; | |
1023 | } | |
1024 | try { | |
1025 | bufferlist::iterator iter = bl.begin(); | |
1026 | ::decode(*this, iter); | |
1027 | } catch (buffer::error& err) { | |
1028 | return -EIO; | |
1029 | } | |
1030 | return 0; | |
1031 | } | |
1032 | ||
1033 | int RGWPeriodConfig::write(RGWRados *store, const std::string& realm_id) | |
1034 | { | |
1035 | const auto& pool = get_pool(store->ctx()); | |
1036 | const auto& oid = get_oid(realm_id); | |
1037 | bufferlist bl; | |
1038 | ::encode(*this, bl); | |
1039 | return rgw_put_system_obj(store, pool, oid, bl.c_str(), bl.length(), | |
1040 | false, nullptr, real_time(), nullptr); | |
1041 | } | |
1042 | ||
1043 | int RGWPeriod::init(CephContext *_cct, RGWRados *_store, const string& period_realm_id, | |
1044 | const string& period_realm_name, bool setup_obj) | |
1045 | { | |
1046 | cct = _cct; | |
1047 | store = _store; | |
1048 | realm_id = period_realm_id; | |
1049 | realm_name = period_realm_name; | |
1050 | ||
1051 | if (!setup_obj) | |
1052 | return 0; | |
1053 | ||
1054 | return init(_cct, _store, setup_obj); | |
1055 | } | |
1056 | ||
1057 | ||
1058 | int RGWPeriod::init(CephContext *_cct, RGWRados *_store, bool setup_obj) | |
1059 | { | |
1060 | cct = _cct; | |
1061 | store = _store; | |
1062 | ||
1063 | if (!setup_obj) | |
1064 | return 0; | |
1065 | ||
1066 | if (id.empty()) { | |
1067 | RGWRealm realm(realm_id, realm_name); | |
1068 | int ret = realm.init(cct, store); | |
1069 | if (ret < 0) { | |
1070 | ldout(cct, 0) << "RGWPeriod::init failed to init realm " << realm_name << " id " << realm_id << " : " << | |
1071 | cpp_strerror(-ret) << dendl; | |
1072 | return ret; | |
1073 | } | |
1074 | id = realm.get_current_period(); | |
1075 | realm_id = realm.get_id(); | |
1076 | } | |
1077 | ||
1078 | if (!epoch) { | |
1079 | int ret = use_latest_epoch(); | |
1080 | if (ret < 0) { | |
1081 | ldout(cct, 0) << "failed to use_latest_epoch period id " << id << " realm " << realm_name << " id " << realm_id | |
1082 | << " : " << cpp_strerror(-ret) << dendl; | |
1083 | return ret; | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | return read_info(); | |
1088 | } | |
1089 | ||
1090 | ||
1091 | int RGWPeriod::get_zonegroup(RGWZoneGroup& zonegroup, const string& zonegroup_id) { | |
1092 | map<string, RGWZoneGroup>::const_iterator iter; | |
1093 | if (!zonegroup_id.empty()) { | |
1094 | iter = period_map.zonegroups.find(zonegroup_id); | |
1095 | } else { | |
1096 | iter = period_map.zonegroups.find("default"); | |
1097 | } | |
1098 | if (iter != period_map.zonegroups.end()) { | |
1099 | zonegroup = iter->second; | |
1100 | return 0; | |
1101 | } | |
1102 | ||
1103 | return -ENOENT; | |
1104 | } | |
1105 | ||
1106 | const string& RGWPeriod::get_latest_epoch_oid() | |
1107 | { | |
1108 | if (cct->_conf->rgw_period_latest_epoch_info_oid.empty()) { | |
1109 | return period_latest_epoch_info_oid; | |
1110 | } | |
1111 | return cct->_conf->rgw_period_latest_epoch_info_oid; | |
1112 | } | |
1113 | ||
1114 | const string& RGWPeriod::get_info_oid_prefix() | |
1115 | { | |
1116 | return period_info_oid_prefix; | |
1117 | } | |
1118 | ||
1119 | const string RGWPeriod::get_period_oid_prefix() | |
1120 | { | |
1121 | return get_info_oid_prefix() + id; | |
1122 | } | |
1123 | ||
1124 | const string RGWPeriod::get_period_oid() | |
1125 | { | |
1126 | std::ostringstream oss; | |
1127 | oss << get_period_oid_prefix(); | |
1128 | // skip the epoch for the staging period | |
1129 | if (id != get_staging_id(realm_id)) | |
1130 | oss << "." << epoch; | |
1131 | return oss.str(); | |
1132 | } | |
1133 | ||
1134 | int RGWPeriod::read_latest_epoch(RGWPeriodLatestEpochInfo& info, | |
1135 | RGWObjVersionTracker *objv) | |
1136 | { | |
1137 | string oid = get_period_oid_prefix() + get_latest_epoch_oid(); | |
1138 | ||
1139 | rgw_pool pool(get_pool(cct)); | |
1140 | bufferlist bl; | |
1141 | RGWObjectCtx obj_ctx(store); | |
1142 | int ret = rgw_get_system_obj(store, obj_ctx, pool, oid, bl, objv, nullptr); | |
1143 | if (ret < 0) { | |
1144 | ldout(cct, 1) << "error read_lastest_epoch " << pool << ":" << oid << dendl; | |
1145 | return ret; | |
1146 | } | |
1147 | try { | |
1148 | bufferlist::iterator iter = bl.begin(); | |
1149 | ::decode(info, iter); | |
1150 | } catch (buffer::error& err) { | |
1151 | ldout(cct, 0) << "error decoding data from " << pool << ":" << oid << dendl; | |
1152 | return -EIO; | |
1153 | } | |
1154 | ||
1155 | return 0; | |
1156 | } | |
1157 | ||
1158 | int RGWPeriod::get_latest_epoch(epoch_t& latest_epoch) | |
1159 | { | |
1160 | RGWPeriodLatestEpochInfo info; | |
1161 | ||
1162 | int ret = read_latest_epoch(info); | |
1163 | if (ret < 0) { | |
1164 | return ret; | |
1165 | } | |
1166 | ||
1167 | latest_epoch = info.epoch; | |
1168 | ||
1169 | return 0; | |
1170 | } | |
1171 | ||
1172 | int RGWPeriod::use_latest_epoch() | |
1173 | { | |
1174 | RGWPeriodLatestEpochInfo info; | |
1175 | int ret = read_latest_epoch(info); | |
1176 | if (ret < 0) { | |
1177 | return ret; | |
1178 | } | |
1179 | ||
1180 | epoch = info.epoch; | |
1181 | ||
1182 | return 0; | |
1183 | } | |
1184 | ||
1185 | int RGWPeriod::set_latest_epoch(epoch_t epoch, bool exclusive, | |
1186 | RGWObjVersionTracker *objv) | |
1187 | { | |
1188 | string oid = get_period_oid_prefix() + get_latest_epoch_oid(); | |
1189 | ||
1190 | rgw_pool pool(get_pool(cct)); | |
1191 | bufferlist bl; | |
1192 | ||
1193 | RGWPeriodLatestEpochInfo info; | |
1194 | info.epoch = epoch; | |
1195 | ||
1196 | ::encode(info, bl); | |
1197 | ||
1198 | return rgw_put_system_obj(store, pool, oid, bl.c_str(), bl.length(), | |
1199 | exclusive, objv, real_time(), nullptr); | |
1200 | } | |
1201 | ||
1202 | int RGWPeriod::update_latest_epoch(epoch_t epoch) | |
1203 | { | |
1204 | static constexpr int MAX_RETRIES = 20; | |
1205 | ||
1206 | for (int i = 0; i < MAX_RETRIES; i++) { | |
1207 | RGWPeriodLatestEpochInfo info; | |
1208 | RGWObjVersionTracker objv; | |
1209 | bool exclusive = false; | |
1210 | ||
1211 | // read existing epoch | |
1212 | int r = read_latest_epoch(info, &objv); | |
1213 | if (r == -ENOENT) { | |
1214 | // use an exclusive create to set the epoch atomically | |
1215 | exclusive = true; | |
1216 | ldout(cct, 20) << "creating initial latest_epoch=" << epoch | |
1217 | << " for period=" << id << dendl; | |
1218 | } else if (r < 0) { | |
1219 | ldout(cct, 0) << "ERROR: failed to read latest_epoch" << dendl; | |
1220 | return r; | |
1221 | } else if (epoch <= info.epoch) { | |
1222 | r = -EEXIST; // fail with EEXIST if epoch is not newer | |
1223 | ldout(cct, 1) << "found existing latest_epoch " << info.epoch | |
1224 | << " >= given epoch " << epoch << ", returning r=" << r << dendl; | |
1225 | return r; | |
1226 | } else { | |
1227 | ldout(cct, 20) << "updating latest_epoch from " << info.epoch | |
1228 | << " -> " << epoch << " on period=" << id << dendl; | |
1229 | } | |
1230 | ||
1231 | r = set_latest_epoch(epoch, exclusive, &objv); | |
1232 | if (r == -EEXIST) { | |
1233 | continue; // exclusive create raced with another update, retry | |
1234 | } else if (r == -ECANCELED) { | |
1235 | continue; // write raced with a conflicting version, retry | |
1236 | } | |
1237 | if (r < 0) { | |
1238 | ldout(cct, 0) << "ERROR: failed to write latest_epoch" << dendl; | |
1239 | return r; | |
1240 | } | |
1241 | return 0; // return success | |
1242 | } | |
1243 | ||
1244 | return -ECANCELED; // fail after max retries | |
1245 | } | |
1246 | ||
1247 | int RGWPeriod::delete_obj() | |
1248 | { | |
1249 | rgw_pool pool(get_pool(cct)); | |
1250 | ||
1251 | // delete the object for each period epoch | |
1252 | for (epoch_t e = 1; e <= epoch; e++) { | |
1253 | RGWPeriod p{get_id(), e}; | |
1254 | rgw_raw_obj oid{pool, p.get_period_oid()}; | |
1255 | int ret = store->delete_system_obj(oid); | |
1256 | if (ret < 0) { | |
1257 | ldout(cct, 0) << "WARNING: failed to delete period object " << oid | |
1258 | << ": " << cpp_strerror(-ret) << dendl; | |
1259 | } | |
1260 | } | |
1261 | ||
1262 | // delete the .latest_epoch object | |
1263 | rgw_raw_obj oid{pool, get_period_oid_prefix() + get_latest_epoch_oid()}; | |
1264 | int ret = store->delete_system_obj(oid); | |
1265 | if (ret < 0) { | |
1266 | ldout(cct, 0) << "WARNING: failed to delete period object " << oid | |
1267 | << ": " << cpp_strerror(-ret) << dendl; | |
1268 | } | |
1269 | return ret; | |
1270 | } | |
1271 | ||
1272 | int RGWPeriod::read_info() | |
1273 | { | |
1274 | rgw_pool pool(get_pool(cct)); | |
1275 | ||
1276 | bufferlist bl; | |
1277 | ||
1278 | RGWObjectCtx obj_ctx(store); | |
1279 | int ret = rgw_get_system_obj(store, obj_ctx, pool, get_period_oid(), bl, NULL, NULL); | |
1280 | if (ret < 0) { | |
1281 | ldout(cct, 0) << "failed reading obj info from " << pool << ":" << get_period_oid() << ": " << cpp_strerror(-ret) << dendl; | |
1282 | return ret; | |
1283 | } | |
1284 | ||
1285 | try { | |
1286 | bufferlist::iterator iter = bl.begin(); | |
1287 | ::decode(*this, iter); | |
1288 | } catch (buffer::error& err) { | |
1289 | ldout(cct, 0) << "ERROR: failed to decode obj from " << pool << ":" << get_period_oid() << dendl; | |
1290 | return -EIO; | |
1291 | } | |
1292 | ||
1293 | return 0; | |
1294 | } | |
1295 | ||
1296 | int RGWPeriod::create(bool exclusive) | |
1297 | { | |
1298 | int ret; | |
1299 | ||
1300 | /* create unique id */ | |
1301 | uuid_d new_uuid; | |
1302 | char uuid_str[37]; | |
1303 | new_uuid.generate_random(); | |
1304 | new_uuid.print(uuid_str); | |
1305 | id = uuid_str; | |
1306 | ||
1307 | epoch = FIRST_EPOCH; | |
1308 | ||
1309 | period_map.id = id; | |
1310 | ||
1311 | ret = store_info(exclusive); | |
1312 | if (ret < 0) { | |
1313 | ldout(cct, 0) << "ERROR: storing info for " << id << ": " << cpp_strerror(-ret) << dendl; | |
1314 | return ret; | |
1315 | } | |
1316 | ||
1317 | ret = set_latest_epoch(epoch); | |
1318 | if (ret < 0) { | |
1319 | ldout(cct, 0) << "ERROR: setting latest epoch " << id << ": " << cpp_strerror(-ret) << dendl; | |
1320 | } | |
1321 | ||
1322 | return ret; | |
1323 | } | |
1324 | ||
1325 | int RGWPeriod::store_info(bool exclusive) | |
1326 | { | |
1327 | rgw_pool pool(get_pool(cct)); | |
1328 | ||
1329 | string oid = get_period_oid(); | |
1330 | bufferlist bl; | |
1331 | ::encode(*this, bl); | |
1332 | ||
1333 | return rgw_put_system_obj(store, pool, oid, bl.c_str(), bl.length(), | |
1334 | exclusive, NULL, real_time(), NULL); | |
1335 | } | |
1336 | ||
1337 | rgw_pool RGWPeriod::get_pool(CephContext *cct) | |
1338 | { | |
1339 | if (cct->_conf->rgw_period_root_pool.empty()) { | |
1340 | return rgw_pool(RGW_DEFAULT_PERIOD_ROOT_POOL); | |
1341 | } | |
1342 | return rgw_pool(cct->_conf->rgw_period_root_pool); | |
1343 | } | |
1344 | ||
1345 | int RGWPeriod::add_zonegroup(const RGWZoneGroup& zonegroup) | |
1346 | { | |
1347 | if (zonegroup.realm_id != realm_id) { | |
1348 | return 0; | |
1349 | } | |
1350 | int ret = period_map.update(zonegroup, cct); | |
1351 | if (ret < 0) { | |
1352 | ldout(cct, 0) << "ERROR: updating period map: " << cpp_strerror(-ret) << dendl; | |
1353 | return ret; | |
1354 | } | |
1355 | ||
1356 | return store_info(false); | |
1357 | } | |
1358 | ||
1359 | int RGWPeriod::update() | |
1360 | { | |
1361 | ldout(cct, 20) << __func__ << " realm " << realm_id << " period " << get_id() << dendl; | |
1362 | list<string> zonegroups; | |
1363 | int ret = store->list_zonegroups(zonegroups); | |
1364 | if (ret < 0) { | |
1365 | ldout(cct, 0) << "ERROR: failed to list zonegroups: " << cpp_strerror(-ret) << dendl; | |
1366 | return ret; | |
1367 | } | |
1368 | ||
1369 | // clear zone short ids of removed zones. period_map.update() will add the | |
1370 | // remaining zones back | |
1371 | period_map.short_zone_ids.clear(); | |
1372 | ||
1373 | for (auto& iter : zonegroups) { | |
1374 | RGWZoneGroup zg(string(), iter); | |
1375 | ret = zg.init(cct, store); | |
1376 | if (ret < 0) { | |
1377 | ldout(cct, 0) << "WARNING: zg.init() failed: " << cpp_strerror(-ret) << dendl; | |
1378 | continue; | |
1379 | } | |
1380 | ||
1381 | if (zg.realm_id != realm_id) { | |
1382 | ldout(cct, 20) << "skipping zonegroup " << zg.get_name() << " zone realm id " << zg.realm_id << ", not on our realm " << realm_id << dendl; | |
1383 | continue; | |
1384 | } | |
1385 | ||
1386 | if (zg.master_zone.empty()) { | |
1387 | ldout(cct, 0) << "ERROR: zonegroup " << zg.get_name() << " should have a master zone " << dendl; | |
1388 | return -EINVAL; | |
1389 | } | |
1390 | ||
1391 | if (zg.zones.find(zg.master_zone) == zg.zones.end()) { | |
1392 | ldout(cct,0) << "ERROR: zonegroup " << zg.get_name() | |
1393 | << " has a non existent master zone "<< dendl; | |
1394 | return -EINVAL; | |
1395 | } | |
1396 | ||
1397 | if (zg.is_master_zonegroup()) { | |
1398 | master_zonegroup = zg.get_id(); | |
1399 | master_zone = zg.master_zone; | |
1400 | } | |
1401 | ||
1402 | int ret = period_map.update(zg, cct); | |
1403 | if (ret < 0) { | |
1404 | return ret; | |
1405 | } | |
1406 | } | |
1407 | ||
1408 | ret = period_config.read(store, realm_id); | |
1409 | if (ret < 0 && ret != -ENOENT) { | |
1410 | ldout(cct, 0) << "ERROR: failed to read period config: " | |
1411 | << cpp_strerror(ret) << dendl; | |
1412 | return ret; | |
1413 | } | |
1414 | return 0; | |
1415 | } | |
1416 | ||
1417 | int RGWPeriod::reflect() | |
1418 | { | |
1419 | for (auto& iter : period_map.zonegroups) { | |
1420 | RGWZoneGroup& zg = iter.second; | |
1421 | zg.reinit_instance(cct, store); | |
1422 | int r = zg.write(false); | |
1423 | if (r < 0) { | |
1424 | ldout(cct, 0) << "ERROR: failed to store zonegroup info for zonegroup=" << iter.first << ": " << cpp_strerror(-r) << dendl; | |
1425 | return r; | |
1426 | } | |
1427 | if (zg.is_master_zonegroup()) { | |
1428 | // set master as default if no default exists | |
1429 | r = zg.set_as_default(true); | |
1430 | if (r == 0) { | |
1431 | ldout(cct, 1) << "Set the period's master zonegroup " << zg.get_id() | |
1432 | << " as the default" << dendl; | |
1433 | } | |
1434 | } | |
1435 | } | |
1436 | ||
1437 | int r = period_config.write(store, realm_id); | |
1438 | if (r < 0) { | |
1439 | ldout(cct, 0) << "ERROR: failed to store period config: " | |
1440 | << cpp_strerror(-r) << dendl; | |
1441 | return r; | |
1442 | } | |
1443 | return 0; | |
1444 | } | |
1445 | ||
1446 | void RGWPeriod::fork() | |
1447 | { | |
1448 | ldout(cct, 20) << __func__ << " realm " << realm_id << " period " << id << dendl; | |
1449 | predecessor_uuid = id; | |
1450 | id = get_staging_id(realm_id); | |
1451 | period_map.reset(); | |
1452 | realm_epoch++; | |
1453 | } | |
1454 | ||
1455 | static int read_sync_status(RGWRados *store, rgw_meta_sync_status *sync_status) | |
1456 | { | |
1457 | // initialize a sync status manager to read the status | |
1458 | RGWMetaSyncStatusManager mgr(store, store->get_async_rados()); | |
1459 | int r = mgr.init(); | |
1460 | if (r < 0) { | |
1461 | return r; | |
1462 | } | |
1463 | r = mgr.read_sync_status(sync_status); | |
1464 | mgr.stop(); | |
1465 | return r; | |
1466 | } | |
1467 | ||
1468 | int RGWPeriod::update_sync_status(const RGWPeriod ¤t_period, | |
1469 | std::ostream& error_stream, | |
1470 | bool force_if_stale) | |
1471 | { | |
1472 | rgw_meta_sync_status status; | |
1473 | int r = read_sync_status(store, &status); | |
1474 | if (r < 0) { | |
1475 | ldout(cct, 0) << "period failed to read sync status: " | |
1476 | << cpp_strerror(-r) << dendl; | |
1477 | return r; | |
1478 | } | |
1479 | ||
1480 | std::vector<std::string> markers; | |
1481 | ||
1482 | const auto current_epoch = current_period.get_realm_epoch(); | |
1483 | if (current_epoch != status.sync_info.realm_epoch) { | |
1484 | // no sync status markers for the current period | |
1485 | assert(current_epoch > status.sync_info.realm_epoch); | |
1486 | const int behind = current_epoch - status.sync_info.realm_epoch; | |
1487 | if (!force_if_stale && current_epoch > 1) { | |
1488 | error_stream << "ERROR: This zone is " << behind << " period(s) behind " | |
1489 | "the current master zone in metadata sync. If this zone is promoted " | |
1490 | "to master, any metadata changes during that time are likely to " | |
1491 | "be lost.\n" | |
1492 | "Waiting for this zone to catch up on metadata sync (see " | |
1493 | "'radosgw-admin sync status') is recommended.\n" | |
1494 | "To promote this zone to master anyway, add the flag " | |
1495 | "--yes-i-really-mean-it." << std::endl; | |
1496 | return -EINVAL; | |
1497 | } | |
1498 | // empty sync status markers - other zones will skip this period during | |
1499 | // incremental metadata sync | |
1500 | markers.resize(status.sync_info.num_shards); | |
1501 | } else { | |
1502 | markers.reserve(status.sync_info.num_shards); | |
1503 | for (auto& i : status.sync_markers) { | |
1504 | auto& marker = i.second; | |
1505 | // filter out markers from other periods | |
1506 | if (marker.realm_epoch != current_epoch) { | |
1507 | marker.marker.clear(); | |
1508 | } | |
1509 | markers.emplace_back(std::move(marker.marker)); | |
1510 | } | |
1511 | } | |
1512 | ||
1513 | std::swap(sync_status, markers); | |
1514 | return 0; | |
1515 | } | |
1516 | ||
1517 | int RGWPeriod::commit(RGWRealm& realm, const RGWPeriod& current_period, | |
1518 | std::ostream& error_stream, bool force_if_stale) | |
1519 | { | |
1520 | ldout(cct, 20) << __func__ << " realm " << realm.get_id() << " period " << current_period.get_id() << dendl; | |
1521 | // gateway must be in the master zone to commit | |
1522 | if (master_zone != store->get_zone_params().get_id()) { | |
1523 | error_stream << "Cannot commit period on zone " | |
1524 | << store->get_zone_params().get_id() << ", it must be sent to " | |
1525 | "the period's master zone " << master_zone << '.' << std::endl; | |
1526 | return -EINVAL; | |
1527 | } | |
1528 | // period predecessor must match current period | |
1529 | if (predecessor_uuid != current_period.get_id()) { | |
1530 | error_stream << "Period predecessor " << predecessor_uuid | |
1531 | << " does not match current period " << current_period.get_id() | |
1532 | << ". Use 'period pull' to get the latest period from the master, " | |
1533 | "reapply your changes, and try again." << std::endl; | |
1534 | return -EINVAL; | |
1535 | } | |
1536 | // realm epoch must be 1 greater than current period | |
1537 | if (realm_epoch != current_period.get_realm_epoch() + 1) { | |
1538 | error_stream << "Period's realm epoch " << realm_epoch | |
1539 | << " does not come directly after current realm epoch " | |
1540 | << current_period.get_realm_epoch() << ". Use 'realm pull' to get the " | |
1541 | "latest realm and period from the master zone, reapply your changes, " | |
1542 | "and try again." << std::endl; | |
1543 | return -EINVAL; | |
1544 | } | |
1545 | // did the master zone change? | |
1546 | if (master_zone != current_period.get_master_zone()) { | |
1547 | // store the current metadata sync status in the period | |
1548 | int r = update_sync_status(current_period, error_stream, force_if_stale); | |
1549 | if (r < 0) { | |
1550 | ldout(cct, 0) << "failed to update metadata sync status: " | |
1551 | << cpp_strerror(-r) << dendl; | |
1552 | return r; | |
1553 | } | |
1554 | // create an object with a new period id | |
1555 | r = create(true); | |
1556 | if (r < 0) { | |
1557 | ldout(cct, 0) << "failed to create new period: " << cpp_strerror(-r) << dendl; | |
1558 | return r; | |
1559 | } | |
1560 | // set as current period | |
1561 | r = realm.set_current_period(*this); | |
1562 | if (r < 0) { | |
1563 | ldout(cct, 0) << "failed to update realm's current period: " | |
1564 | << cpp_strerror(-r) << dendl; | |
1565 | return r; | |
1566 | } | |
1567 | ldout(cct, 4) << "Promoted to master zone and committed new period " | |
1568 | << id << dendl; | |
1569 | realm.notify_new_period(*this); | |
1570 | return 0; | |
1571 | } | |
1572 | // period must be based on current epoch | |
1573 | if (epoch != current_period.get_epoch()) { | |
1574 | error_stream << "Period epoch " << epoch << " does not match " | |
1575 | "predecessor epoch " << current_period.get_epoch() | |
1576 | << ". Use 'period pull' to get the latest epoch from the master zone, " | |
1577 | "reapply your changes, and try again." << std::endl; | |
1578 | return -EINVAL; | |
1579 | } | |
1580 | // set period as next epoch | |
1581 | set_id(current_period.get_id()); | |
1582 | set_epoch(current_period.get_epoch() + 1); | |
1583 | set_predecessor(current_period.get_predecessor()); | |
1584 | realm_epoch = current_period.get_realm_epoch(); | |
1585 | // write the period to rados | |
1586 | int r = store_info(false); | |
1587 | if (r < 0) { | |
1588 | ldout(cct, 0) << "failed to store period: " << cpp_strerror(-r) << dendl; | |
1589 | return r; | |
1590 | } | |
1591 | // set as latest epoch | |
1592 | r = update_latest_epoch(epoch); | |
1593 | if (r == -EEXIST) { | |
1594 | // already have this epoch (or a more recent one) | |
1595 | return 0; | |
1596 | } | |
1597 | if (r < 0) { | |
1598 | ldout(cct, 0) << "failed to set latest epoch: " << cpp_strerror(-r) << dendl; | |
1599 | return r; | |
1600 | } | |
1601 | r = reflect(); | |
1602 | if (r < 0) { | |
1603 | ldout(cct, 0) << "failed to update local objects: " << cpp_strerror(-r) << dendl; | |
1604 | return r; | |
1605 | } | |
1606 | ldout(cct, 4) << "Committed new epoch " << epoch | |
1607 | << " for period " << id << dendl; | |
1608 | realm.notify_new_period(*this); | |
1609 | return 0; | |
1610 | } | |
1611 | ||
1612 | int RGWZoneParams::create_default(bool old_format) | |
1613 | { | |
1614 | name = default_zone_name; | |
1615 | ||
1616 | int r = create(); | |
1617 | if (r < 0) { | |
1618 | return r; | |
1619 | } | |
1620 | ||
1621 | if (old_format) { | |
1622 | name = id; | |
1623 | } | |
1624 | ||
1625 | return r; | |
1626 | } | |
1627 | ||
1628 | ||
1629 | int get_zones_pool_set(CephContext* cct, | |
1630 | RGWRados* store, | |
1631 | const list<string>& zones, | |
1632 | const string& my_zone_id, | |
1633 | set<rgw_pool>& pool_names) | |
1634 | { | |
1635 | for(auto const& iter : zones) { | |
1636 | RGWZoneParams zone(iter); | |
1637 | int r = zone.init(cct, store); | |
1638 | if (r < 0) { | |
1639 | ldout(cct, 0) << "Error: init zone " << iter << ":" << cpp_strerror(-r) << dendl; | |
1640 | return r; | |
1641 | } | |
1642 | if (zone.get_id() != my_zone_id) { | |
1643 | pool_names.insert(zone.domain_root); | |
1644 | pool_names.insert(zone.metadata_heap); | |
1645 | pool_names.insert(zone.control_pool); | |
1646 | pool_names.insert(zone.gc_pool); | |
1647 | pool_names.insert(zone.log_pool); | |
1648 | pool_names.insert(zone.intent_log_pool); | |
1649 | pool_names.insert(zone.usage_log_pool); | |
1650 | pool_names.insert(zone.user_keys_pool); | |
1651 | pool_names.insert(zone.user_email_pool); | |
1652 | pool_names.insert(zone.user_swift_pool); | |
1653 | pool_names.insert(zone.user_uid_pool); | |
1654 | pool_names.insert(zone.roles_pool); | |
1655 | pool_names.insert(zone.reshard_pool); | |
1656 | for(auto& iter : zone.placement_pools) { | |
1657 | pool_names.insert(iter.second.index_pool); | |
1658 | pool_names.insert(iter.second.data_pool); | |
1659 | pool_names.insert(iter.second.data_extra_pool); | |
1660 | } | |
1661 | } | |
1662 | } | |
1663 | return 0; | |
1664 | } | |
1665 | ||
1666 | rgw_pool fix_zone_pool_dup(set<rgw_pool> pools, | |
1667 | const string& default_prefix, | |
1668 | const string& default_suffix, | |
1669 | const rgw_pool& suggested_pool) | |
1670 | { | |
1671 | string suggested_name = suggested_pool.to_str(); | |
1672 | ||
1673 | string prefix = default_prefix; | |
1674 | string suffix = default_suffix; | |
1675 | ||
1676 | if (!suggested_pool.empty()) { | |
1677 | prefix = suggested_name.substr(0, suggested_name.find(".")); | |
1678 | suffix = suggested_name.substr(prefix.length()); | |
1679 | } | |
1680 | ||
1681 | rgw_pool pool(prefix + suffix); | |
1682 | ||
1683 | if (pools.find(pool) == pools.end()) { | |
1684 | return pool; | |
1685 | } else { | |
1686 | while(true) { | |
1687 | pool = prefix + "_" + std::to_string(std::rand()) + suffix; | |
1688 | if (pools.find(pool) == pools.end()) { | |
1689 | return pool; | |
1690 | } | |
1691 | } | |
1692 | } | |
1693 | } | |
1694 | ||
1695 | int RGWZoneParams::fix_pool_names() | |
1696 | { | |
1697 | ||
1698 | list<string> zones; | |
1699 | int r = store->list_zones(zones); | |
1700 | if (r < 0) { | |
1701 | ldout(cct, 10) << "WARNING: store->list_zones() returned r=" << r << dendl; | |
1702 | } | |
1703 | ||
1704 | set<rgw_pool> pools; | |
1705 | r = get_zones_pool_set(cct, store, zones, id, pools); | |
1706 | if (r < 0) { | |
1707 | ldout(cct, 0) << "Error: get_zones_pool_names" << r << dendl; | |
1708 | return r; | |
1709 | } | |
1710 | ||
1711 | domain_root = fix_zone_pool_dup(pools, name, ".rgw.meta:root", domain_root); | |
1712 | if (!metadata_heap.name.empty()) { | |
1713 | metadata_heap = fix_zone_pool_dup(pools, name, ".rgw.meta:heap", metadata_heap); | |
1714 | } | |
1715 | control_pool = fix_zone_pool_dup(pools, name, ".rgw.control", control_pool); | |
1716 | gc_pool = fix_zone_pool_dup(pools, name ,".rgw.log:gc", gc_pool); | |
1717 | lc_pool = fix_zone_pool_dup(pools, name ,".rgw.log:lc", lc_pool); | |
1718 | log_pool = fix_zone_pool_dup(pools, name, ".rgw.log", log_pool); | |
1719 | intent_log_pool = fix_zone_pool_dup(pools, name, ".rgw.log:intent", intent_log_pool); | |
1720 | usage_log_pool = fix_zone_pool_dup(pools, name, ".rgw.log:usage", usage_log_pool); | |
1721 | user_keys_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:users.keys", user_keys_pool); | |
1722 | user_email_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:users.email", user_email_pool); | |
1723 | user_swift_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:users.swift", user_swift_pool); | |
1724 | user_uid_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:users.uid", user_uid_pool); | |
1725 | roles_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:roles", roles_pool); | |
1726 | reshard_pool = fix_zone_pool_dup(pools, name, ".rgw.log:reshard", reshard_pool); | |
1727 | ||
1728 | for(auto& iter : placement_pools) { | |
1729 | iter.second.index_pool = fix_zone_pool_dup(pools, name, "." + default_bucket_index_pool_suffix, | |
1730 | iter.second.index_pool); | |
1731 | iter.second.data_pool = fix_zone_pool_dup(pools, name, "." + default_storage_pool_suffix, | |
1732 | iter.second.data_pool); | |
1733 | iter.second.data_extra_pool= fix_zone_pool_dup(pools, name, "." + default_storage_extra_pool_suffix, | |
1734 | iter.second.data_extra_pool); | |
1735 | } | |
1736 | ||
1737 | return 0; | |
1738 | } | |
1739 | ||
1740 | int RGWZoneParams::create(bool exclusive) | |
1741 | { | |
1742 | /* check for old pools config */ | |
1743 | rgw_raw_obj obj(domain_root, avail_pools); | |
1744 | int r = store->raw_obj_stat(obj, NULL, NULL, NULL, NULL, NULL, NULL); | |
1745 | if (r < 0) { | |
1746 | ldout(store->ctx(), 10) << "couldn't find old data placement pools config, setting up new ones for the zone" << dendl; | |
1747 | /* a new system, let's set new placement info */ | |
1748 | RGWZonePlacementInfo default_placement; | |
1749 | default_placement.index_pool = name + "." + default_bucket_index_pool_suffix; | |
1750 | default_placement.data_pool = name + "." + default_storage_pool_suffix; | |
1751 | default_placement.data_extra_pool = name + "." + default_storage_extra_pool_suffix; | |
1752 | placement_pools["default-placement"] = default_placement; | |
1753 | } | |
1754 | ||
1755 | r = fix_pool_names(); | |
1756 | if (r < 0) { | |
1757 | ldout(cct, 0) << "ERROR: fix_pool_names returned r=" << r << dendl; | |
1758 | return r; | |
1759 | } | |
1760 | ||
1761 | r = RGWSystemMetaObj::create(exclusive); | |
1762 | if (r < 0) { | |
1763 | return r; | |
1764 | } | |
1765 | ||
1766 | // try to set as default. may race with another create, so pass exclusive=true | |
1767 | // so we don't override an existing default | |
1768 | r = set_as_default(true); | |
1769 | if (r < 0 && r != -EEXIST) { | |
1770 | ldout(cct, 10) << "WARNING: failed to set zone as default, r=" << r << dendl; | |
1771 | } | |
1772 | ||
1773 | return 0; | |
1774 | } | |
1775 | ||
1776 | rgw_pool RGWZoneParams::get_pool(CephContext *cct) | |
1777 | { | |
1778 | if (cct->_conf->rgw_zone_root_pool.empty()) { | |
1779 | return rgw_pool(RGW_DEFAULT_ZONE_ROOT_POOL); | |
1780 | } | |
1781 | ||
1782 | return rgw_pool(cct->_conf->rgw_zone_root_pool); | |
1783 | } | |
1784 | ||
1785 | const string RGWZoneParams::get_default_oid(bool old_format) | |
1786 | { | |
1787 | if (old_format) { | |
1788 | return cct->_conf->rgw_default_zone_info_oid; | |
1789 | } | |
1790 | ||
1791 | return cct->_conf->rgw_default_zone_info_oid + "." + realm_id; | |
1792 | } | |
1793 | ||
1794 | const string& RGWZoneParams::get_names_oid_prefix() | |
1795 | { | |
1796 | return zone_names_oid_prefix; | |
1797 | } | |
1798 | ||
1799 | const string& RGWZoneParams::get_info_oid_prefix(bool old_format) | |
1800 | { | |
1801 | return zone_info_oid_prefix; | |
1802 | } | |
1803 | ||
1804 | const string& RGWZoneParams::get_predefined_name(CephContext *cct) { | |
1805 | return cct->_conf->rgw_zone; | |
1806 | } | |
1807 | ||
1808 | int RGWZoneParams::init(CephContext *cct, RGWRados *store, bool setup_obj, bool old_format) | |
1809 | { | |
1810 | if (name.empty()) { | |
1811 | name = cct->_conf->rgw_zone; | |
1812 | } | |
1813 | ||
1814 | return RGWSystemMetaObj::init(cct, store, setup_obj, old_format); | |
1815 | } | |
1816 | ||
1817 | int RGWZoneParams::read_default_id(string& default_id, bool old_format) | |
1818 | { | |
1819 | if (realm_id.empty()) { | |
1820 | /* try using default realm */ | |
1821 | RGWRealm realm; | |
1822 | int ret = realm.init(cct, store); | |
1823 | //no default realm exist | |
1824 | if (ret < 0) { | |
1825 | return read_id(default_zone_name, default_id); | |
1826 | } | |
1827 | realm_id = realm.get_id(); | |
1828 | } | |
1829 | ||
1830 | return RGWSystemMetaObj::read_default_id(default_id, old_format); | |
1831 | } | |
1832 | ||
1833 | ||
1834 | int RGWZoneParams::set_as_default(bool exclusive) | |
1835 | { | |
1836 | if (realm_id.empty()) { | |
1837 | /* try using default realm */ | |
1838 | RGWRealm realm; | |
1839 | int ret = realm.init(cct, store); | |
1840 | if (ret < 0) { | |
1841 | ldout(cct, 10) << "could not read realm id: " << cpp_strerror(-ret) << dendl; | |
1842 | return -EINVAL; | |
1843 | } | |
1844 | realm_id = realm.get_id(); | |
1845 | } | |
1846 | ||
1847 | return RGWSystemMetaObj::set_as_default(exclusive); | |
1848 | } | |
1849 | ||
1850 | const string& RGWZoneParams::get_compression_type(const string& placement_rule) const | |
1851 | { | |
1852 | static const std::string NONE{"none"}; | |
1853 | auto p = placement_pools.find(placement_rule); | |
1854 | if (p == placement_pools.end()) { | |
1855 | return NONE; | |
1856 | } | |
1857 | const auto& type = p->second.compression_type; | |
1858 | return !type.empty() ? type : NONE; | |
1859 | } | |
1860 | ||
1861 | void RGWPeriodMap::encode(bufferlist& bl) const { | |
1862 | ENCODE_START(2, 1, bl); | |
1863 | ::encode(id, bl); | |
1864 | ::encode(zonegroups, bl); | |
1865 | ::encode(master_zonegroup, bl); | |
1866 | ::encode(short_zone_ids, bl); | |
1867 | ENCODE_FINISH(bl); | |
1868 | } | |
1869 | ||
1870 | void RGWPeriodMap::decode(bufferlist::iterator& bl) { | |
1871 | DECODE_START(2, bl); | |
1872 | ::decode(id, bl); | |
1873 | ::decode(zonegroups, bl); | |
1874 | ::decode(master_zonegroup, bl); | |
1875 | if (struct_v >= 2) { | |
1876 | ::decode(short_zone_ids, bl); | |
1877 | } | |
1878 | DECODE_FINISH(bl); | |
1879 | ||
1880 | zonegroups_by_api.clear(); | |
1881 | for (map<string, RGWZoneGroup>::iterator iter = zonegroups.begin(); | |
1882 | iter != zonegroups.end(); ++iter) { | |
1883 | RGWZoneGroup& zonegroup = iter->second; | |
1884 | zonegroups_by_api[zonegroup.api_name] = zonegroup; | |
1885 | if (zonegroup.is_master_zonegroup()) { | |
1886 | master_zonegroup = zonegroup.get_id(); | |
1887 | } | |
1888 | } | |
1889 | } | |
1890 | ||
1891 | // run an MD5 hash on the zone_id and return the first 32 bits | |
1892 | static uint32_t gen_short_zone_id(const std::string zone_id) | |
1893 | { | |
1894 | unsigned char md5[CEPH_CRYPTO_MD5_DIGESTSIZE]; | |
1895 | MD5 hash; | |
1896 | hash.Update((const byte *)zone_id.c_str(), zone_id.size()); | |
1897 | hash.Final(md5); | |
1898 | ||
1899 | uint32_t short_id; | |
1900 | memcpy((char *)&short_id, md5, sizeof(short_id)); | |
1901 | return std::max(short_id, 1u); | |
1902 | } | |
1903 | ||
1904 | int RGWPeriodMap::update(const RGWZoneGroup& zonegroup, CephContext *cct) | |
1905 | { | |
1906 | if (zonegroup.is_master_zonegroup() && (!master_zonegroup.empty() && zonegroup.get_id() != master_zonegroup)) { | |
1907 | ldout(cct,0) << "Error updating periodmap, multiple master zonegroups configured "<< dendl; | |
1908 | ldout(cct,0) << "master zonegroup: " << master_zonegroup << " and " << zonegroup.get_id() <<dendl; | |
1909 | return -EINVAL; | |
1910 | } | |
1911 | map<string, RGWZoneGroup>::iterator iter = zonegroups.find(zonegroup.get_id()); | |
1912 | if (iter != zonegroups.end()) { | |
1913 | RGWZoneGroup& old_zonegroup = iter->second; | |
1914 | if (!old_zonegroup.api_name.empty()) { | |
1915 | zonegroups_by_api.erase(old_zonegroup.api_name); | |
1916 | } | |
1917 | } | |
1918 | zonegroups[zonegroup.get_id()] = zonegroup; | |
1919 | ||
1920 | if (!zonegroup.api_name.empty()) { | |
1921 | zonegroups_by_api[zonegroup.api_name] = zonegroup; | |
1922 | } | |
1923 | ||
1924 | if (zonegroup.is_master_zonegroup()) { | |
1925 | master_zonegroup = zonegroup.get_id(); | |
1926 | } else if (master_zonegroup == zonegroup.get_id()) { | |
1927 | master_zonegroup = ""; | |
1928 | } | |
1929 | ||
1930 | for (auto& i : zonegroup.zones) { | |
1931 | auto& zone = i.second; | |
1932 | if (short_zone_ids.find(zone.id) != short_zone_ids.end()) { | |
1933 | continue; | |
1934 | } | |
1935 | // calculate the zone's short id | |
1936 | uint32_t short_id = gen_short_zone_id(zone.id); | |
1937 | ||
1938 | // search for an existing zone with the same short id | |
1939 | for (auto& s : short_zone_ids) { | |
1940 | if (s.second == short_id) { | |
1941 | ldout(cct, 0) << "New zone '" << zone.name << "' (" << zone.id | |
1942 | << ") generates the same short_zone_id " << short_id | |
1943 | << " as existing zone id " << s.first << dendl; | |
1944 | return -EEXIST; | |
1945 | } | |
1946 | } | |
1947 | ||
1948 | short_zone_ids[zone.id] = short_id; | |
1949 | } | |
1950 | ||
1951 | return 0; | |
1952 | } | |
1953 | ||
1954 | uint32_t RGWPeriodMap::get_zone_short_id(const string& zone_id) const | |
1955 | { | |
1956 | auto i = short_zone_ids.find(zone_id); | |
1957 | if (i == short_zone_ids.end()) { | |
1958 | return 0; | |
1959 | } | |
1960 | return i->second; | |
1961 | } | |
1962 | ||
1963 | int RGWZoneGroupMap::read(CephContext *cct, RGWRados *store) | |
1964 | { | |
1965 | ||
1966 | RGWPeriod period; | |
1967 | int ret = period.init(cct, store); | |
1968 | if (ret < 0) { | |
1969 | cerr << "failed to read current period info: " << cpp_strerror(ret); | |
1970 | return ret; | |
1971 | } | |
1972 | ||
1973 | bucket_quota = period.get_config().bucket_quota; | |
1974 | user_quota = period.get_config().user_quota; | |
1975 | zonegroups = period.get_map().zonegroups; | |
1976 | zonegroups_by_api = period.get_map().zonegroups_by_api; | |
1977 | master_zonegroup = period.get_map().master_zonegroup; | |
1978 | ||
1979 | return 0; | |
1980 | } | |
1981 | ||
1982 | void RGWRegionMap::encode(bufferlist& bl) const { | |
1983 | ENCODE_START( 3, 1, bl); | |
1984 | ::encode(regions, bl); | |
1985 | ::encode(master_region, bl); | |
1986 | ::encode(bucket_quota, bl); | |
1987 | ::encode(user_quota, bl); | |
1988 | ENCODE_FINISH(bl); | |
1989 | } | |
1990 | ||
1991 | void RGWRegionMap::decode(bufferlist::iterator& bl) { | |
1992 | DECODE_START(3, bl); | |
1993 | ::decode(regions, bl); | |
1994 | ::decode(master_region, bl); | |
1995 | if (struct_v >= 2) | |
1996 | ::decode(bucket_quota, bl); | |
1997 | if (struct_v >= 3) | |
1998 | ::decode(user_quota, bl); | |
1999 | DECODE_FINISH(bl); | |
2000 | } | |
2001 | ||
2002 | void RGWZoneGroupMap::encode(bufferlist& bl) const { | |
2003 | ENCODE_START( 3, 1, bl); | |
2004 | ::encode(zonegroups, bl); | |
2005 | ::encode(master_zonegroup, bl); | |
2006 | ::encode(bucket_quota, bl); | |
2007 | ::encode(user_quota, bl); | |
2008 | ENCODE_FINISH(bl); | |
2009 | } | |
2010 | ||
2011 | void RGWZoneGroupMap::decode(bufferlist::iterator& bl) { | |
2012 | DECODE_START(3, bl); | |
2013 | ::decode(zonegroups, bl); | |
2014 | ::decode(master_zonegroup, bl); | |
2015 | if (struct_v >= 2) | |
2016 | ::decode(bucket_quota, bl); | |
2017 | if (struct_v >= 3) | |
2018 | ::decode(user_quota, bl); | |
2019 | DECODE_FINISH(bl); | |
2020 | ||
2021 | zonegroups_by_api.clear(); | |
2022 | for (map<string, RGWZoneGroup>::iterator iter = zonegroups.begin(); | |
2023 | iter != zonegroups.end(); ++iter) { | |
2024 | RGWZoneGroup& zonegroup = iter->second; | |
2025 | zonegroups_by_api[zonegroup.api_name] = zonegroup; | |
2026 | if (zonegroup.is_master_zonegroup()) { | |
2027 | master_zonegroup = zonegroup.get_name(); | |
2028 | } | |
2029 | } | |
2030 | } | |
2031 | ||
2032 | void RGWObjVersionTracker::prepare_op_for_read(ObjectReadOperation *op) | |
2033 | { | |
2034 | obj_version *check_objv = version_for_check(); | |
2035 | ||
2036 | if (check_objv) { | |
2037 | cls_version_check(*op, *check_objv, VER_COND_EQ); | |
2038 | } | |
2039 | ||
2040 | cls_version_read(*op, &read_version); | |
2041 | } | |
2042 | ||
2043 | void RGWObjVersionTracker::prepare_op_for_write(ObjectWriteOperation *op) | |
2044 | { | |
2045 | obj_version *check_objv = version_for_check(); | |
2046 | obj_version *modify_version = version_for_write(); | |
2047 | ||
2048 | if (check_objv) { | |
2049 | cls_version_check(*op, *check_objv, VER_COND_EQ); | |
2050 | } | |
2051 | ||
2052 | if (modify_version) { | |
2053 | cls_version_set(*op, *modify_version); | |
2054 | } else { | |
2055 | cls_version_inc(*op); | |
2056 | } | |
2057 | } | |
2058 | ||
2059 | void RGWObjManifest::obj_iterator::operator++() | |
2060 | { | |
2061 | if (manifest->explicit_objs) { | |
2062 | ++explicit_iter; | |
2063 | ||
2064 | if (explicit_iter == manifest->objs.end()) { | |
2065 | ofs = manifest->obj_size; | |
2066 | return; | |
2067 | } | |
2068 | ||
2069 | update_explicit_pos(); | |
2070 | ||
2071 | update_location(); | |
2072 | return; | |
2073 | } | |
2074 | ||
2075 | uint64_t obj_size = manifest->get_obj_size(); | |
2076 | uint64_t head_size = manifest->get_head_size(); | |
2077 | ||
2078 | if (ofs == obj_size) { | |
2079 | return; | |
2080 | } | |
2081 | ||
2082 | if (manifest->rules.empty()) { | |
2083 | return; | |
2084 | } | |
2085 | ||
2086 | /* are we still pointing at the head? */ | |
2087 | if (ofs < head_size) { | |
2088 | rule_iter = manifest->rules.begin(); | |
2089 | RGWObjManifestRule *rule = &rule_iter->second; | |
2090 | ofs = MIN(head_size, obj_size); | |
2091 | stripe_ofs = ofs; | |
2092 | cur_stripe = 1; | |
2093 | stripe_size = MIN(obj_size - ofs, rule->stripe_max_size); | |
2094 | if (rule->part_size > 0) { | |
2095 | stripe_size = MIN(stripe_size, rule->part_size); | |
2096 | } | |
2097 | update_location(); | |
2098 | return; | |
2099 | } | |
2100 | ||
2101 | RGWObjManifestRule *rule = &rule_iter->second; | |
2102 | ||
2103 | stripe_ofs += rule->stripe_max_size; | |
2104 | cur_stripe++; | |
2105 | dout(20) << "RGWObjManifest::operator++(): rule->part_size=" << rule->part_size << " rules.size()=" << manifest->rules.size() << dendl; | |
2106 | ||
2107 | if (rule->part_size > 0) { | |
2108 | /* multi part, multi stripes object */ | |
2109 | ||
2110 | dout(20) << "RGWObjManifest::operator++(): stripe_ofs=" << stripe_ofs << " part_ofs=" << part_ofs << " rule->part_size=" << rule->part_size << dendl; | |
2111 | ||
2112 | if (stripe_ofs >= part_ofs + rule->part_size) { | |
2113 | /* moved to the next part */ | |
2114 | cur_stripe = 0; | |
2115 | part_ofs += rule->part_size; | |
2116 | stripe_ofs = part_ofs; | |
2117 | ||
2118 | bool last_rule = (next_rule_iter == manifest->rules.end()); | |
2119 | /* move to the next rule? */ | |
2120 | if (!last_rule && stripe_ofs >= next_rule_iter->second.start_ofs) { | |
2121 | rule_iter = next_rule_iter; | |
2122 | last_rule = (next_rule_iter == manifest->rules.end()); | |
2123 | if (!last_rule) { | |
2124 | ++next_rule_iter; | |
2125 | } | |
2126 | cur_part_id = rule_iter->second.start_part_num; | |
2127 | } else { | |
2128 | cur_part_id++; | |
2129 | } | |
2130 | ||
2131 | rule = &rule_iter->second; | |
2132 | } | |
2133 | ||
2134 | stripe_size = MIN(rule->part_size - (stripe_ofs - part_ofs), rule->stripe_max_size); | |
2135 | } | |
2136 | ||
2137 | cur_override_prefix = rule->override_prefix; | |
2138 | ||
2139 | ofs = stripe_ofs; | |
2140 | if (ofs > obj_size) { | |
2141 | ofs = obj_size; | |
2142 | stripe_ofs = ofs; | |
2143 | stripe_size = 0; | |
2144 | } | |
2145 | ||
2146 | dout(20) << "RGWObjManifest::operator++(): result: ofs=" << ofs << " stripe_ofs=" << stripe_ofs << " part_ofs=" << part_ofs << " rule->part_size=" << rule->part_size << dendl; | |
2147 | update_location(); | |
2148 | } | |
2149 | ||
2150 | int RGWObjManifest::generator::create_begin(CephContext *cct, RGWObjManifest *_m, const string& placement_rule, rgw_bucket& _b, rgw_obj& _obj) | |
2151 | { | |
2152 | manifest = _m; | |
2153 | ||
2154 | manifest->set_tail_placement(placement_rule, _b); | |
2155 | manifest->set_head(placement_rule, _obj, 0); | |
2156 | last_ofs = 0; | |
2157 | ||
2158 | if (manifest->get_prefix().empty()) { | |
2159 | char buf[33]; | |
2160 | gen_rand_alphanumeric(cct, buf, sizeof(buf) - 1); | |
2161 | ||
2162 | string oid_prefix = "."; | |
2163 | oid_prefix.append(buf); | |
2164 | oid_prefix.append("_"); | |
2165 | ||
2166 | manifest->set_prefix(oid_prefix); | |
2167 | } | |
2168 | ||
2169 | bool found = manifest->get_rule(0, &rule); | |
2170 | if (!found) { | |
2171 | derr << "ERROR: manifest->get_rule() could not find rule" << dendl; | |
2172 | return -EIO; | |
2173 | } | |
2174 | ||
2175 | uint64_t head_size = manifest->get_head_size(); | |
2176 | ||
2177 | if (head_size > 0) { | |
2178 | cur_stripe_size = head_size; | |
2179 | } else { | |
2180 | cur_stripe_size = rule.stripe_max_size; | |
2181 | } | |
2182 | ||
2183 | cur_part_id = rule.start_part_num; | |
2184 | ||
2185 | manifest->get_implicit_location(cur_part_id, cur_stripe, 0, NULL, &cur_obj); | |
2186 | ||
2187 | // Normal object which not generated through copy operation | |
2188 | manifest->set_tail_instance(_obj.key.instance); | |
2189 | ||
2190 | manifest->update_iterators(); | |
2191 | ||
2192 | return 0; | |
2193 | } | |
2194 | ||
2195 | int RGWObjManifest::generator::create_next(uint64_t ofs) | |
2196 | { | |
2197 | if (ofs < last_ofs) /* only going forward */ | |
2198 | return -EINVAL; | |
2199 | ||
2200 | uint64_t max_head_size = manifest->get_max_head_size(); | |
2201 | ||
2202 | if (ofs < max_head_size) { | |
2203 | manifest->set_head_size(ofs); | |
2204 | } | |
2205 | ||
2206 | if (ofs >= max_head_size) { | |
2207 | manifest->set_head_size(max_head_size); | |
2208 | cur_stripe = (ofs - max_head_size) / rule.stripe_max_size; | |
2209 | cur_stripe_size = rule.stripe_max_size; | |
2210 | ||
2211 | if (cur_part_id == 0 && max_head_size > 0) { | |
2212 | cur_stripe++; | |
2213 | } | |
2214 | } | |
2215 | ||
2216 | last_ofs = ofs; | |
2217 | manifest->set_obj_size(ofs); | |
2218 | ||
2219 | manifest->get_implicit_location(cur_part_id, cur_stripe, ofs, NULL, &cur_obj); | |
2220 | ||
2221 | manifest->update_iterators(); | |
2222 | ||
2223 | return 0; | |
2224 | } | |
2225 | ||
2226 | const RGWObjManifest::obj_iterator& RGWObjManifest::obj_begin() | |
2227 | { | |
2228 | return begin_iter; | |
2229 | } | |
2230 | ||
2231 | const RGWObjManifest::obj_iterator& RGWObjManifest::obj_end() | |
2232 | { | |
2233 | return end_iter; | |
2234 | } | |
2235 | ||
2236 | RGWObjManifest::obj_iterator RGWObjManifest::obj_find(uint64_t ofs) | |
2237 | { | |
2238 | if (ofs > obj_size) { | |
2239 | ofs = obj_size; | |
2240 | } | |
2241 | RGWObjManifest::obj_iterator iter(this); | |
2242 | iter.seek(ofs); | |
2243 | return iter; | |
2244 | } | |
2245 | ||
2246 | int RGWObjManifest::append(RGWObjManifest& m, RGWZoneGroup& zonegroup, RGWZoneParams& zone_params) | |
2247 | { | |
2248 | if (explicit_objs || m.explicit_objs) { | |
2249 | return append_explicit(m, zonegroup, zone_params); | |
2250 | } | |
2251 | ||
2252 | if (rules.empty()) { | |
2253 | *this = m; | |
2254 | return 0; | |
2255 | } | |
2256 | ||
2257 | string override_prefix; | |
2258 | ||
2259 | if (prefix.empty()) { | |
2260 | prefix = m.prefix; | |
2261 | } | |
2262 | ||
2263 | if (prefix != m.prefix) { | |
2264 | override_prefix = m.prefix; | |
2265 | } | |
2266 | ||
2267 | map<uint64_t, RGWObjManifestRule>::iterator miter = m.rules.begin(); | |
2268 | if (miter == m.rules.end()) { | |
2269 | return append_explicit(m, zonegroup, zone_params); | |
2270 | } | |
2271 | ||
2272 | for (; miter != m.rules.end(); ++miter) { | |
2273 | map<uint64_t, RGWObjManifestRule>::reverse_iterator last_rule = rules.rbegin(); | |
2274 | ||
2275 | RGWObjManifestRule& rule = last_rule->second; | |
2276 | ||
2277 | if (rule.part_size == 0) { | |
2278 | rule.part_size = obj_size - rule.start_ofs; | |
2279 | } | |
2280 | ||
2281 | RGWObjManifestRule& next_rule = miter->second; | |
2282 | if (!next_rule.part_size) { | |
2283 | next_rule.part_size = m.obj_size - next_rule.start_ofs; | |
2284 | } | |
2285 | ||
2286 | string rule_prefix = prefix; | |
2287 | if (!rule.override_prefix.empty()) { | |
2288 | rule_prefix = rule.override_prefix; | |
2289 | } | |
2290 | ||
2291 | string next_rule_prefix = m.prefix; | |
2292 | if (!next_rule.override_prefix.empty()) { | |
2293 | next_rule_prefix = next_rule.override_prefix; | |
2294 | } | |
2295 | ||
2296 | if (rule.part_size != next_rule.part_size || | |
2297 | rule.stripe_max_size != next_rule.stripe_max_size || | |
2298 | rule_prefix != next_rule_prefix) { | |
2299 | if (next_rule_prefix != prefix) { | |
2300 | append_rules(m, miter, &next_rule_prefix); | |
2301 | } else { | |
2302 | append_rules(m, miter, NULL); | |
2303 | } | |
2304 | break; | |
2305 | } | |
2306 | ||
2307 | uint64_t expected_part_num = rule.start_part_num + 1; | |
2308 | if (rule.part_size > 0) { | |
2309 | expected_part_num = rule.start_part_num + (obj_size + next_rule.start_ofs - rule.start_ofs) / rule.part_size; | |
2310 | } | |
2311 | ||
2312 | if (expected_part_num != next_rule.start_part_num) { | |
2313 | append_rules(m, miter, NULL); | |
2314 | break; | |
2315 | } | |
2316 | } | |
2317 | ||
2318 | set_obj_size(obj_size + m.obj_size); | |
2319 | ||
2320 | return 0; | |
2321 | } | |
2322 | ||
2323 | int RGWObjManifest::append(RGWObjManifest& m, RGWRados *store) | |
2324 | { | |
2325 | return append(m, store->get_zonegroup(), store->get_zone_params()); | |
2326 | } | |
2327 | ||
2328 | void RGWObjManifest::append_rules(RGWObjManifest& m, map<uint64_t, RGWObjManifestRule>::iterator& miter, | |
2329 | string *override_prefix) | |
2330 | { | |
2331 | for (; miter != m.rules.end(); ++miter) { | |
2332 | RGWObjManifestRule rule = miter->second; | |
2333 | rule.start_ofs += obj_size; | |
2334 | if (override_prefix) | |
2335 | rule.override_prefix = *override_prefix; | |
2336 | rules[rule.start_ofs] = rule; | |
2337 | } | |
2338 | } | |
2339 | ||
2340 | void RGWObjManifest::convert_to_explicit(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params) | |
2341 | { | |
2342 | if (explicit_objs) { | |
2343 | return; | |
2344 | } | |
2345 | obj_iterator iter = obj_begin(); | |
2346 | ||
2347 | while (iter != obj_end()) { | |
2348 | RGWObjManifestPart& part = objs[iter.get_stripe_ofs()]; | |
2349 | const rgw_obj_select& os = iter.get_location(); | |
2350 | const rgw_raw_obj& raw_loc = os.get_raw_obj(zonegroup, zone_params); | |
2351 | part.loc_ofs = 0; | |
2352 | ||
2353 | uint64_t ofs = iter.get_stripe_ofs(); | |
2354 | ||
2355 | if (ofs == 0) { | |
2356 | part.loc = obj; | |
2357 | } else { | |
2358 | rgw_raw_obj_to_obj(tail_placement.bucket, raw_loc, &part.loc); | |
2359 | } | |
2360 | ++iter; | |
2361 | uint64_t next_ofs = iter.get_stripe_ofs(); | |
2362 | ||
2363 | part.size = next_ofs - ofs; | |
2364 | } | |
2365 | ||
2366 | explicit_objs = true; | |
2367 | rules.clear(); | |
2368 | prefix.clear(); | |
2369 | } | |
2370 | ||
2371 | int RGWObjManifest::append_explicit(RGWObjManifest& m, const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params) | |
2372 | { | |
2373 | if (!explicit_objs) { | |
2374 | convert_to_explicit(zonegroup, zone_params); | |
2375 | } | |
2376 | if (!m.explicit_objs) { | |
2377 | m.convert_to_explicit(zonegroup, zone_params); | |
2378 | } | |
2379 | map<uint64_t, RGWObjManifestPart>::iterator iter; | |
2380 | uint64_t base = obj_size; | |
2381 | for (iter = m.objs.begin(); iter != m.objs.end(); ++iter) { | |
2382 | RGWObjManifestPart& part = iter->second; | |
2383 | objs[base + iter->first] = part; | |
2384 | } | |
2385 | obj_size += m.obj_size; | |
2386 | ||
2387 | return 0; | |
2388 | } | |
2389 | ||
2390 | bool RGWObjManifest::get_rule(uint64_t ofs, RGWObjManifestRule *rule) | |
2391 | { | |
2392 | if (rules.empty()) { | |
2393 | return false; | |
2394 | } | |
2395 | ||
2396 | map<uint64_t, RGWObjManifestRule>::iterator iter = rules.upper_bound(ofs); | |
2397 | if (iter != rules.begin()) { | |
2398 | --iter; | |
2399 | } | |
2400 | ||
2401 | *rule = iter->second; | |
2402 | ||
2403 | return true; | |
2404 | } | |
2405 | ||
2406 | void RGWObjVersionTracker::generate_new_write_ver(CephContext *cct) | |
2407 | { | |
2408 | write_version.ver = 1; | |
2409 | #define TAG_LEN 24 | |
2410 | ||
2411 | write_version.tag.clear(); | |
2412 | append_rand_alpha(cct, write_version.tag, write_version.tag, TAG_LEN); | |
2413 | } | |
2414 | ||
2415 | int RGWPutObjProcessor::complete(size_t accounted_size, const string& etag, | |
2416 | real_time *mtime, real_time set_mtime, | |
2417 | map<string, bufferlist>& attrs, real_time delete_at, | |
2418 | const char *if_match, const char *if_nomatch, const string *user_data, | |
2419 | rgw_zone_set *zones_trace) | |
2420 | { | |
2421 | int r = do_complete(accounted_size, etag, mtime, set_mtime, attrs, delete_at, if_match, if_nomatch, user_data, zones_trace); | |
2422 | if (r < 0) | |
2423 | return r; | |
2424 | ||
2425 | is_complete = !canceled; | |
2426 | return 0; | |
2427 | } | |
2428 | ||
2429 | CephContext *RGWPutObjProcessor::ctx() | |
2430 | { | |
2431 | return store->ctx(); | |
2432 | } | |
2433 | ||
2434 | RGWPutObjProcessor_Aio::~RGWPutObjProcessor_Aio() | |
2435 | { | |
2436 | drain_pending(); | |
2437 | ||
2438 | if (is_complete) | |
2439 | return; | |
2440 | ||
2441 | set<rgw_raw_obj>::iterator iter; | |
2442 | bool need_to_remove_head = false; | |
2443 | rgw_raw_obj raw_head; | |
2444 | ||
2445 | if (!head_obj.empty()) { | |
2446 | store->obj_to_raw(bucket_info.placement_rule, head_obj, &raw_head); | |
2447 | } | |
2448 | ||
2449 | /** | |
2450 | * We should delete the object in the "multipart" namespace to avoid race condition. | |
2451 | * Such race condition is caused by the fact that the multipart object is the gatekeeper of a multipart | |
2452 | * upload, when it is deleted, a second upload would start with the same suffix("2/"), therefore, objects | |
2453 | * written by the second upload may be deleted by the first upload. | |
2454 | * details is describled on #11749 | |
2455 | * | |
2456 | * The above comment still stands, but instead of searching for a specific object in the multipart | |
2457 | * namespace, we just make sure that we remove the object that is marked as the head object after | |
2458 | * we remove all the other raw objects. Note that we use different call to remove the head object, | |
2459 | * as this one needs to go via the bucket index prepare/complete 2-phase commit scheme. | |
2460 | */ | |
2461 | for (iter = written_objs.begin(); iter != written_objs.end(); ++iter) { | |
2462 | const rgw_raw_obj& obj = *iter; | |
2463 | if (!head_obj.empty() && obj == raw_head) { | |
2464 | ldout(store->ctx(), 5) << "NOTE: we should not process the head object (" << obj << ") here" << dendl; | |
2465 | need_to_remove_head = true; | |
2466 | continue; | |
2467 | } | |
2468 | ||
2469 | int r = store->delete_raw_obj(obj); | |
2470 | if (r < 0 && r != -ENOENT) { | |
2471 | ldout(store->ctx(), 5) << "WARNING: failed to remove obj (" << obj << "), leaked" << dendl; | |
2472 | } | |
2473 | } | |
2474 | ||
2475 | if (need_to_remove_head) { | |
2476 | ldout(store->ctx(), 5) << "NOTE: we are going to process the head obj (" << raw_head << ")" << dendl; | |
2477 | int r = store->delete_obj(obj_ctx, bucket_info, head_obj, 0, 0); | |
2478 | if (r < 0 && r != -ENOENT) { | |
2479 | ldout(store->ctx(), 0) << "WARNING: failed to remove obj (" << raw_head << "), leaked" << dendl; | |
2480 | } | |
2481 | } | |
2482 | } | |
2483 | ||
2484 | int RGWPutObjProcessor_Aio::handle_obj_data(rgw_raw_obj& obj, bufferlist& bl, off_t ofs, off_t abs_ofs, void **phandle, bool exclusive) | |
2485 | { | |
2486 | if ((uint64_t)abs_ofs + bl.length() > obj_len) | |
2487 | obj_len = abs_ofs + bl.length(); | |
2488 | ||
2489 | if (!(obj == last_written_obj)) { | |
2490 | last_written_obj = obj; | |
2491 | } | |
2492 | ||
2493 | // For the first call pass -1 as the offset to | |
2494 | // do a write_full. | |
2495 | return store->aio_put_obj_data(NULL, obj, bl, ((ofs != 0) ? ofs : -1), exclusive, phandle); | |
2496 | } | |
2497 | ||
2498 | struct put_obj_aio_info RGWPutObjProcessor_Aio::pop_pending() | |
2499 | { | |
2500 | struct put_obj_aio_info info; | |
2501 | info = pending.front(); | |
2502 | pending.pop_front(); | |
2503 | pending_size -= info.size; | |
2504 | return info; | |
2505 | } | |
2506 | ||
2507 | int RGWPutObjProcessor_Aio::wait_pending_front() | |
2508 | { | |
2509 | if (pending.empty()) { | |
2510 | return 0; | |
2511 | } | |
2512 | struct put_obj_aio_info info = pop_pending(); | |
2513 | int ret = store->aio_wait(info.handle); | |
2514 | ||
2515 | if (ret >= 0) { | |
2516 | add_written_obj(info.obj); | |
2517 | } | |
2518 | ||
2519 | return ret; | |
2520 | } | |
2521 | ||
2522 | bool RGWPutObjProcessor_Aio::pending_has_completed() | |
2523 | { | |
2524 | if (pending.empty()) | |
2525 | return false; | |
2526 | ||
2527 | struct put_obj_aio_info& info = pending.front(); | |
2528 | return store->aio_completed(info.handle); | |
2529 | } | |
2530 | ||
2531 | int RGWPutObjProcessor_Aio::drain_pending() | |
2532 | { | |
2533 | int ret = 0; | |
2534 | while (!pending.empty()) { | |
2535 | int r = wait_pending_front(); | |
2536 | if (r < 0) | |
2537 | ret = r; | |
2538 | } | |
2539 | return ret; | |
2540 | } | |
2541 | ||
2542 | int RGWPutObjProcessor_Aio::throttle_data(void *handle, const rgw_raw_obj& obj, uint64_t size, bool need_to_wait) | |
2543 | { | |
2544 | bool _wait = need_to_wait; | |
2545 | ||
2546 | if (handle) { | |
2547 | struct put_obj_aio_info info; | |
2548 | info.handle = handle; | |
2549 | info.obj = obj; | |
2550 | info.size = size; | |
2551 | pending_size += size; | |
2552 | pending.push_back(info); | |
2553 | } | |
2554 | size_t orig_size = pending_size; | |
2555 | ||
2556 | /* first drain complete IOs */ | |
2557 | while (pending_has_completed()) { | |
2558 | int r = wait_pending_front(); | |
2559 | if (r < 0) | |
2560 | return r; | |
2561 | ||
2562 | _wait = false; | |
2563 | } | |
2564 | ||
2565 | /* resize window in case messages are draining too fast */ | |
2566 | if (orig_size - pending_size >= window_size) { | |
2567 | window_size += store->ctx()->_conf->rgw_max_chunk_size; | |
2568 | uint64_t max_window_size = store->ctx()->_conf->rgw_put_obj_max_window_size; | |
2569 | if (window_size > max_window_size) { | |
2570 | window_size = max_window_size; | |
2571 | } | |
2572 | } | |
2573 | ||
2574 | /* now throttle. Note that need_to_wait should only affect the first IO operation */ | |
2575 | if (pending_size > window_size || _wait) { | |
2576 | int r = wait_pending_front(); | |
2577 | if (r < 0) | |
2578 | return r; | |
2579 | } | |
2580 | return 0; | |
2581 | } | |
2582 | ||
2583 | int RGWPutObjProcessor_Atomic::write_data(bufferlist& bl, off_t ofs, void **phandle, rgw_raw_obj *pobj, bool exclusive) | |
2584 | { | |
2585 | if (ofs >= next_part_ofs) { | |
2586 | int r = prepare_next_part(ofs); | |
2587 | if (r < 0) { | |
2588 | return r; | |
2589 | } | |
2590 | } | |
2591 | ||
2592 | *pobj = cur_obj; | |
2593 | ||
2594 | if (!bl.length()) { | |
2595 | *phandle = nullptr; | |
2596 | return 0; | |
2597 | } | |
2598 | ||
2599 | return RGWPutObjProcessor_Aio::handle_obj_data(cur_obj, bl, ofs - cur_part_ofs, ofs, phandle, exclusive); | |
2600 | } | |
2601 | ||
2602 | int RGWPutObjProcessor_Aio::prepare(RGWRados *store, string *oid_rand) | |
2603 | { | |
2604 | RGWPutObjProcessor::prepare(store, oid_rand); | |
2605 | ||
2606 | window_size = store->ctx()->_conf->rgw_put_obj_min_window_size; | |
2607 | ||
2608 | return 0; | |
2609 | } | |
2610 | ||
2611 | int RGWPutObjProcessor_Atomic::handle_data(bufferlist& bl, off_t ofs, void **phandle, rgw_raw_obj *pobj, bool *again) | |
2612 | { | |
2613 | *phandle = NULL; | |
2614 | uint64_t max_write_size = MIN(max_chunk_size, (uint64_t)next_part_ofs - data_ofs); | |
2615 | ||
2616 | pending_data_bl.claim_append(bl); | |
2617 | if (pending_data_bl.length() < max_write_size) { | |
2618 | *again = false; | |
2619 | return 0; | |
2620 | } | |
2621 | ||
2622 | pending_data_bl.splice(0, max_write_size, &bl); | |
2623 | ||
2624 | /* do we have enough data pending accumulated that needs to be written? */ | |
2625 | *again = (pending_data_bl.length() >= max_chunk_size); | |
2626 | ||
2627 | if (!data_ofs && !immutable_head()) { | |
2628 | first_chunk.claim(bl); | |
2629 | obj_len = (uint64_t)first_chunk.length(); | |
2630 | int r = prepare_next_part(obj_len); | |
2631 | if (r < 0) { | |
2632 | return r; | |
2633 | } | |
2634 | data_ofs = obj_len; | |
2635 | return 0; | |
2636 | } | |
2637 | off_t write_ofs = data_ofs; | |
2638 | data_ofs = write_ofs + bl.length(); | |
2639 | bool exclusive = (!write_ofs && immutable_head()); /* immutable head object, need to verify nothing exists there | |
2640 | we could be racing with another upload, to the same | |
2641 | object and cleanup can be messy */ | |
2642 | int ret = write_data(bl, write_ofs, phandle, pobj, exclusive); | |
2643 | if (ret >= 0) { /* we might return, need to clear bl as it was already sent */ | |
2644 | bl.clear(); | |
2645 | } | |
2646 | return ret; | |
2647 | } | |
2648 | ||
2649 | ||
2650 | int RGWPutObjProcessor_Atomic::prepare_init(RGWRados *store, string *oid_rand) | |
2651 | { | |
2652 | RGWPutObjProcessor_Aio::prepare(store, oid_rand); | |
2653 | ||
2654 | int r = store->get_max_chunk_size(bucket_info.placement_rule, head_obj, &max_chunk_size); | |
2655 | if (r < 0) { | |
2656 | return r; | |
2657 | } | |
2658 | ||
2659 | return 0; | |
2660 | } | |
2661 | ||
2662 | int RGWPutObjProcessor_Atomic::prepare(RGWRados *store, string *oid_rand) | |
2663 | { | |
2664 | head_obj.init(bucket, obj_str); | |
2665 | ||
2666 | int r = prepare_init(store, oid_rand); | |
2667 | if (r < 0) { | |
2668 | return r; | |
2669 | } | |
2670 | ||
2671 | if (!version_id.empty()) { | |
2672 | head_obj.key.set_instance(version_id); | |
2673 | } else if (versioned_object) { | |
2674 | store->gen_rand_obj_instance_name(&head_obj); | |
2675 | } | |
2676 | ||
2677 | manifest.set_trivial_rule(max_chunk_size, store->ctx()->_conf->rgw_obj_stripe_size); | |
2678 | ||
2679 | r = manifest_gen.create_begin(store->ctx(), &manifest, bucket_info.placement_rule, head_obj.bucket, head_obj); | |
2680 | if (r < 0) { | |
2681 | return r; | |
2682 | } | |
2683 | ||
2684 | return 0; | |
2685 | } | |
2686 | ||
2687 | int RGWPutObjProcessor_Atomic::prepare_next_part(off_t ofs) { | |
2688 | ||
2689 | int ret = manifest_gen.create_next(ofs); | |
2690 | if (ret < 0) { | |
2691 | lderr(store->ctx()) << "ERROR: manifest_gen.create_next() returned ret=" << ret << dendl; | |
2692 | return ret; | |
2693 | } | |
2694 | cur_part_ofs = ofs; | |
2695 | next_part_ofs = ofs + manifest_gen.cur_stripe_max_size(); | |
2696 | cur_obj = manifest_gen.get_cur_obj(store); | |
2697 | ||
2698 | return 0; | |
2699 | } | |
2700 | ||
2701 | int RGWPutObjProcessor_Atomic::complete_parts() | |
2702 | { | |
2703 | if (obj_len > (uint64_t)cur_part_ofs) { | |
2704 | return prepare_next_part(obj_len); | |
2705 | } | |
2706 | return 0; | |
2707 | } | |
2708 | ||
2709 | int RGWPutObjProcessor_Atomic::complete_writing_data() | |
2710 | { | |
2711 | if (!data_ofs && !immutable_head()) { | |
2712 | /* only claim if pending_data_bl() is not empty. This is needed because we might be called twice | |
2713 | * (e.g., when a retry due to race happens). So a second call to first_chunk.claim() would | |
2714 | * clobber first_chunk | |
2715 | */ | |
2716 | if (pending_data_bl.length() > 0) { | |
2717 | first_chunk.claim(pending_data_bl); | |
2718 | } | |
2719 | obj_len = (uint64_t)first_chunk.length(); | |
2720 | } | |
2721 | while (pending_data_bl.length()) { | |
2722 | void *handle = nullptr; | |
2723 | rgw_raw_obj obj; | |
2724 | uint64_t max_write_size = MIN(max_chunk_size, (uint64_t)next_part_ofs - data_ofs); | |
2725 | if (max_write_size > pending_data_bl.length()) { | |
2726 | max_write_size = pending_data_bl.length(); | |
2727 | } | |
2728 | bufferlist bl; | |
2729 | pending_data_bl.splice(0, max_write_size, &bl); | |
2730 | uint64_t write_len = bl.length(); | |
2731 | int r = write_data(bl, data_ofs, &handle, &obj, false); | |
2732 | if (r < 0) { | |
2733 | ldout(store->ctx(), 0) << "ERROR: write_data() returned " << r << dendl; | |
2734 | return r; | |
2735 | } | |
2736 | data_ofs += write_len; | |
2737 | r = throttle_data(handle, obj, write_len, false); | |
2738 | if (r < 0) { | |
2739 | ldout(store->ctx(), 0) << "ERROR: throttle_data() returned " << r << dendl; | |
2740 | return r; | |
2741 | } | |
2742 | ||
2743 | if (data_ofs >= next_part_ofs) { | |
2744 | r = prepare_next_part(data_ofs); | |
2745 | if (r < 0) { | |
2746 | ldout(store->ctx(), 0) << "ERROR: prepare_next_part() returned " << r << dendl; | |
2747 | return r; | |
2748 | } | |
2749 | } | |
2750 | } | |
2751 | int r = complete_parts(); | |
2752 | if (r < 0) { | |
2753 | return r; | |
2754 | } | |
2755 | ||
2756 | r = drain_pending(); | |
2757 | if (r < 0) | |
2758 | return r; | |
2759 | ||
2760 | return 0; | |
2761 | } | |
2762 | ||
2763 | int RGWPutObjProcessor_Atomic::do_complete(size_t accounted_size, const string& etag, | |
2764 | real_time *mtime, real_time set_mtime, | |
2765 | map<string, bufferlist>& attrs, | |
2766 | real_time delete_at, | |
2767 | const char *if_match, | |
2768 | const char *if_nomatch, const string *user_data, | |
2769 | rgw_zone_set *zones_trace) { | |
2770 | int r = complete_writing_data(); | |
2771 | if (r < 0) | |
2772 | return r; | |
2773 | ||
2774 | obj_ctx.obj.set_atomic(head_obj); | |
2775 | ||
2776 | RGWRados::Object op_target(store, bucket_info, obj_ctx, head_obj); | |
2777 | ||
2778 | /* some object types shouldn't be versioned, e.g., multipart parts */ | |
2779 | op_target.set_versioning_disabled(!versioned_object); | |
2780 | ||
2781 | RGWRados::Object::Write obj_op(&op_target); | |
2782 | ||
2783 | obj_op.meta.data = &first_chunk; | |
2784 | obj_op.meta.manifest = &manifest; | |
2785 | obj_op.meta.ptag = &unique_tag; /* use req_id as operation tag */ | |
2786 | obj_op.meta.if_match = if_match; | |
2787 | obj_op.meta.if_nomatch = if_nomatch; | |
2788 | obj_op.meta.mtime = mtime; | |
2789 | obj_op.meta.set_mtime = set_mtime; | |
2790 | obj_op.meta.owner = bucket_info.owner; | |
2791 | obj_op.meta.flags = PUT_OBJ_CREATE; | |
2792 | obj_op.meta.olh_epoch = olh_epoch; | |
2793 | obj_op.meta.delete_at = delete_at; | |
2794 | obj_op.meta.user_data = user_data; | |
2795 | obj_op.meta.zones_trace = zones_trace; | |
2796 | obj_op.meta.modify_tail = true; | |
2797 | ||
2798 | r = obj_op.write_meta(obj_len, accounted_size, attrs); | |
2799 | if (r < 0) { | |
2800 | return r; | |
2801 | } | |
2802 | ||
2803 | canceled = obj_op.meta.canceled; | |
2804 | ||
2805 | return 0; | |
2806 | } | |
2807 | ||
2808 | const char* RGWRados::admin_commands[4][3] = { | |
2809 | { "cache list", | |
2810 | "cache list name=filter,type=CephString,req=false", | |
2811 | "cache list [filter_str]: list object cache, possibly matching substrings" }, | |
2812 | { "cache inspect", | |
2813 | "cache inspect name=target,type=CephString,req=true", | |
2814 | "cache inspect target: print cache element" }, | |
2815 | { "cache erase", | |
2816 | "cache erase name=target,type=CephString,req=true", | |
2817 | "cache erase target: erase element from cache" }, | |
2818 | { "cache zap", | |
2819 | "cache zap", | |
2820 | "cache zap: erase all elements from cache" } | |
2821 | }; | |
2822 | ||
2823 | ||
2824 | int RGWRados::watch(const string& oid, uint64_t *watch_handle, librados::WatchCtx2 *ctx) { | |
2825 | int r = control_pool_ctx.watch2(oid, watch_handle, ctx); | |
2826 | if (r < 0) | |
2827 | return r; | |
2828 | return 0; | |
2829 | } | |
2830 | ||
2831 | int RGWRados::unwatch(uint64_t watch_handle) | |
2832 | { | |
2833 | int r = control_pool_ctx.unwatch2(watch_handle); | |
2834 | if (r < 0) { | |
2835 | ldout(cct, 0) << "ERROR: rados->unwatch2() returned r=" << r << dendl; | |
2836 | return r; | |
2837 | } | |
2838 | r = rados[0].watch_flush(); | |
2839 | if (r < 0) { | |
2840 | ldout(cct, 0) << "ERROR: rados->watch_flush() returned r=" << r << dendl; | |
2841 | return r; | |
2842 | } | |
2843 | return 0; | |
2844 | } | |
2845 | ||
2846 | void RGWRados::add_watcher(int i) | |
2847 | { | |
2848 | ldout(cct, 20) << "add_watcher() i=" << i << dendl; | |
2849 | Mutex::Locker l(watchers_lock); | |
2850 | watchers_set.insert(i); | |
2851 | if (watchers_set.size() == (size_t)num_watchers) { | |
2852 | ldout(cct, 2) << "all " << num_watchers << " watchers are set, enabling cache" << dendl; | |
2853 | set_cache_enabled(true); | |
2854 | } | |
2855 | } | |
2856 | ||
2857 | void RGWRados::remove_watcher(int i) | |
2858 | { | |
2859 | ldout(cct, 20) << "remove_watcher() i=" << i << dendl; | |
2860 | Mutex::Locker l(watchers_lock); | |
2861 | size_t orig_size = watchers_set.size(); | |
2862 | watchers_set.erase(i); | |
2863 | if (orig_size == (size_t)num_watchers && | |
2864 | watchers_set.size() < orig_size) { /* actually removed */ | |
2865 | ldout(cct, 2) << "removed watcher, disabling cache" << dendl; | |
2866 | set_cache_enabled(false); | |
2867 | } | |
2868 | } | |
2869 | ||
2870 | class RGWWatcher : public librados::WatchCtx2 { | |
2871 | RGWRados *rados; | |
2872 | int index; | |
2873 | string oid; | |
2874 | uint64_t watch_handle; | |
2875 | ||
2876 | class C_ReinitWatch : public Context { | |
2877 | RGWWatcher *watcher; | |
2878 | public: | |
2879 | explicit C_ReinitWatch(RGWWatcher *_watcher) : watcher(_watcher) {} | |
2880 | void finish(int r) override { | |
2881 | watcher->reinit(); | |
2882 | } | |
2883 | }; | |
2884 | public: | |
2885 | RGWWatcher(RGWRados *r, int i, const string& o) : rados(r), index(i), oid(o), watch_handle(0) {} | |
2886 | void handle_notify(uint64_t notify_id, | |
2887 | uint64_t cookie, | |
2888 | uint64_t notifier_id, | |
2889 | bufferlist& bl) override { | |
2890 | ldout(rados->ctx(), 10) << "RGWWatcher::handle_notify() " | |
2891 | << " notify_id " << notify_id | |
2892 | << " cookie " << cookie | |
2893 | << " notifier " << notifier_id | |
2894 | << " bl.length()=" << bl.length() << dendl; | |
2895 | rados->watch_cb(notify_id, cookie, notifier_id, bl); | |
2896 | ||
2897 | bufferlist reply_bl; // empty reply payload | |
2898 | rados->control_pool_ctx.notify_ack(oid, notify_id, cookie, reply_bl); | |
2899 | } | |
2900 | void handle_error(uint64_t cookie, int err) override { | |
2901 | lderr(rados->ctx()) << "RGWWatcher::handle_error cookie " << cookie | |
2902 | << " err " << cpp_strerror(err) << dendl; | |
2903 | rados->remove_watcher(index); | |
2904 | rados->schedule_context(new C_ReinitWatch(this)); | |
2905 | } | |
2906 | ||
2907 | void reinit() { | |
2908 | int ret = unregister_watch(); | |
2909 | if (ret < 0) { | |
2910 | ldout(rados->ctx(), 0) << "ERROR: unregister_watch() returned ret=" << ret << dendl; | |
2911 | return; | |
2912 | } | |
2913 | ret = register_watch(); | |
2914 | if (ret < 0) { | |
2915 | ldout(rados->ctx(), 0) << "ERROR: register_watch() returned ret=" << ret << dendl; | |
2916 | return; | |
2917 | } | |
2918 | } | |
2919 | ||
2920 | int unregister_watch() { | |
2921 | int r = rados->unwatch(watch_handle); | |
2922 | if (r < 0) { | |
2923 | return r; | |
2924 | } | |
2925 | rados->remove_watcher(index); | |
2926 | return 0; | |
2927 | } | |
2928 | ||
2929 | int register_watch() { | |
2930 | int r = rados->watch(oid, &watch_handle, this); | |
2931 | if (r < 0) { | |
2932 | return r; | |
2933 | } | |
2934 | rados->add_watcher(index); | |
2935 | return 0; | |
2936 | } | |
2937 | }; | |
2938 | ||
2939 | class RGWMetaNotifierManager : public RGWCoroutinesManager { | |
2940 | RGWRados *store; | |
2941 | RGWHTTPManager http_manager; | |
2942 | ||
2943 | public: | |
2944 | RGWMetaNotifierManager(RGWRados *_store) : RGWCoroutinesManager(_store->ctx(), _store->get_cr_registry()), store(_store), | |
2945 | http_manager(store->ctx(), completion_mgr) { | |
2946 | http_manager.set_threaded(); | |
2947 | } | |
2948 | ||
2949 | int notify_all(map<string, RGWRESTConn *>& conn_map, set<int>& shards) { | |
2950 | rgw_http_param_pair pairs[] = { { "type", "metadata" }, | |
2951 | { "notify", NULL }, | |
2952 | { NULL, NULL } }; | |
2953 | ||
2954 | list<RGWCoroutinesStack *> stacks; | |
2955 | for (map<string, RGWRESTConn *>::iterator iter = conn_map.begin(); iter != conn_map.end(); ++iter) { | |
2956 | RGWRESTConn *conn = iter->second; | |
2957 | RGWCoroutinesStack *stack = new RGWCoroutinesStack(store->ctx(), this); | |
2958 | stack->call(new RGWPostRESTResourceCR<set<int>, int>(store->ctx(), conn, &http_manager, "/admin/log", pairs, shards, NULL)); | |
2959 | ||
2960 | stacks.push_back(stack); | |
2961 | } | |
2962 | return run(stacks); | |
2963 | } | |
2964 | }; | |
2965 | ||
2966 | class RGWDataNotifierManager : public RGWCoroutinesManager { | |
2967 | RGWRados *store; | |
2968 | RGWHTTPManager http_manager; | |
2969 | ||
2970 | public: | |
2971 | RGWDataNotifierManager(RGWRados *_store) : RGWCoroutinesManager(_store->ctx(), _store->get_cr_registry()), store(_store), | |
2972 | http_manager(store->ctx(), completion_mgr) { | |
2973 | http_manager.set_threaded(); | |
2974 | } | |
2975 | ||
2976 | int notify_all(map<string, RGWRESTConn *>& conn_map, map<int, set<string> >& shards) { | |
2977 | rgw_http_param_pair pairs[] = { { "type", "data" }, | |
2978 | { "notify", NULL }, | |
2979 | { "source-zone", store->get_zone_params().get_id().c_str() }, | |
2980 | { NULL, NULL } }; | |
2981 | ||
2982 | list<RGWCoroutinesStack *> stacks; | |
2983 | for (map<string, RGWRESTConn *>::iterator iter = conn_map.begin(); iter != conn_map.end(); ++iter) { | |
2984 | RGWRESTConn *conn = iter->second; | |
2985 | RGWCoroutinesStack *stack = new RGWCoroutinesStack(store->ctx(), this); | |
2986 | stack->call(new RGWPostRESTResourceCR<map<int, set<string> >, int>(store->ctx(), conn, &http_manager, "/admin/log", pairs, shards, NULL)); | |
2987 | ||
2988 | stacks.push_back(stack); | |
2989 | } | |
2990 | return run(stacks); | |
2991 | } | |
2992 | }; | |
2993 | ||
2994 | class RGWRadosThread { | |
2995 | class Worker : public Thread { | |
2996 | CephContext *cct; | |
2997 | RGWRadosThread *processor; | |
2998 | Mutex lock; | |
2999 | Cond cond; | |
3000 | ||
3001 | void wait() { | |
3002 | Mutex::Locker l(lock); | |
3003 | cond.Wait(lock); | |
3004 | }; | |
3005 | ||
3006 | void wait_interval(const utime_t& wait_time) { | |
3007 | Mutex::Locker l(lock); | |
3008 | cond.WaitInterval(lock, wait_time); | |
3009 | } | |
3010 | ||
3011 | public: | |
3012 | Worker(CephContext *_cct, RGWRadosThread *_p) : cct(_cct), processor(_p), lock("RGWRadosThread::Worker") {} | |
3013 | void *entry() override; | |
3014 | void signal() { | |
3015 | Mutex::Locker l(lock); | |
3016 | cond.Signal(); | |
3017 | } | |
3018 | }; | |
3019 | ||
3020 | Worker *worker; | |
3021 | ||
3022 | protected: | |
3023 | CephContext *cct; | |
3024 | RGWRados *store; | |
3025 | ||
3026 | std::atomic<bool> down_flag = { false }; | |
3027 | ||
3028 | string thread_name; | |
3029 | ||
3030 | virtual uint64_t interval_msec() = 0; | |
3031 | virtual void stop_process() {} | |
3032 | public: | |
3033 | RGWRadosThread(RGWRados *_store, const string& thread_name = "radosgw") | |
3034 | : worker(NULL), cct(_store->ctx()), store(_store), thread_name(thread_name) {} | |
3035 | virtual ~RGWRadosThread() { | |
3036 | stop(); | |
3037 | } | |
3038 | ||
3039 | virtual int init() { return 0; } | |
3040 | virtual int process() = 0; | |
3041 | ||
3042 | bool going_down() { return down_flag; } | |
3043 | ||
3044 | void start(); | |
3045 | void stop(); | |
3046 | ||
3047 | void signal() { | |
3048 | if (worker) { | |
3049 | worker->signal(); | |
3050 | } | |
3051 | } | |
3052 | }; | |
3053 | ||
3054 | void RGWRadosThread::start() | |
3055 | { | |
3056 | worker = new Worker(cct, this); | |
3057 | worker->create(thread_name.c_str()); | |
3058 | } | |
3059 | ||
3060 | void RGWRadosThread::stop() | |
3061 | { | |
3062 | down_flag = true; | |
3063 | stop_process(); | |
3064 | if (worker) { | |
3065 | worker->signal(); | |
3066 | worker->join(); | |
3067 | } | |
3068 | delete worker; | |
3069 | worker = NULL; | |
3070 | } | |
3071 | ||
3072 | void *RGWRadosThread::Worker::entry() { | |
3073 | uint64_t msec = processor->interval_msec(); | |
3074 | utime_t interval = utime_t(msec / 1000, (msec % 1000) * 1000000); | |
3075 | ||
3076 | do { | |
3077 | utime_t start = ceph_clock_now(); | |
3078 | int r = processor->process(); | |
3079 | if (r < 0) { | |
3080 | dout(0) << "ERROR: processor->process() returned error r=" << r << dendl; | |
3081 | } | |
3082 | ||
3083 | if (processor->going_down()) | |
3084 | break; | |
3085 | ||
3086 | utime_t end = ceph_clock_now(); | |
3087 | end -= start; | |
3088 | ||
3089 | uint64_t cur_msec = processor->interval_msec(); | |
3090 | if (cur_msec != msec) { /* was it reconfigured? */ | |
3091 | msec = cur_msec; | |
3092 | interval = utime_t(msec / 1000, (msec % 1000) * 1000000); | |
3093 | } | |
3094 | ||
3095 | if (cur_msec > 0) { | |
3096 | if (interval <= end) | |
3097 | continue; // next round | |
3098 | ||
3099 | utime_t wait_time = interval; | |
3100 | wait_time -= end; | |
3101 | ||
3102 | wait_interval(wait_time); | |
3103 | } else { | |
3104 | wait(); | |
3105 | } | |
3106 | } while (!processor->going_down()); | |
3107 | ||
3108 | return NULL; | |
3109 | } | |
3110 | ||
3111 | class RGWMetaNotifier : public RGWRadosThread { | |
3112 | RGWMetaNotifierManager notify_mgr; | |
3113 | RGWMetadataLog *const log; | |
3114 | ||
3115 | uint64_t interval_msec() override { | |
3116 | return cct->_conf->rgw_md_notify_interval_msec; | |
3117 | } | |
3118 | void stop_process() override { | |
3119 | notify_mgr.stop(); | |
3120 | } | |
3121 | public: | |
3122 | RGWMetaNotifier(RGWRados *_store, RGWMetadataLog* log) | |
3123 | : RGWRadosThread(_store, "meta-notifier"), notify_mgr(_store), log(log) {} | |
3124 | ||
3125 | int process() override; | |
3126 | }; | |
3127 | ||
3128 | int RGWMetaNotifier::process() | |
3129 | { | |
3130 | set<int> shards; | |
3131 | ||
3132 | log->read_clear_modified(shards); | |
3133 | ||
3134 | if (shards.empty()) { | |
3135 | return 0; | |
3136 | } | |
3137 | ||
3138 | for (set<int>::iterator iter = shards.begin(); iter != shards.end(); ++iter) { | |
3139 | ldout(cct, 20) << __func__ << "(): notifying mdlog change, shard_id=" << *iter << dendl; | |
3140 | } | |
3141 | ||
3142 | notify_mgr.notify_all(store->zone_conn_map, shards); | |
3143 | ||
3144 | return 0; | |
3145 | } | |
3146 | ||
3147 | class RGWDataNotifier : public RGWRadosThread { | |
3148 | RGWDataNotifierManager notify_mgr; | |
3149 | ||
3150 | uint64_t interval_msec() override { | |
3151 | return cct->_conf->get_val<int64_t>("rgw_data_notify_interval_msec"); | |
3152 | } | |
3153 | void stop_process() override { | |
3154 | notify_mgr.stop(); | |
3155 | } | |
3156 | public: | |
3157 | RGWDataNotifier(RGWRados *_store) : RGWRadosThread(_store, "data-notifier"), notify_mgr(_store) {} | |
3158 | ||
3159 | int process() override; | |
3160 | }; | |
3161 | ||
3162 | int RGWDataNotifier::process() | |
3163 | { | |
3164 | if (!store->data_log) { | |
3165 | return 0; | |
3166 | } | |
3167 | ||
3168 | map<int, set<string> > shards; | |
3169 | ||
3170 | store->data_log->read_clear_modified(shards); | |
3171 | ||
3172 | if (shards.empty()) { | |
3173 | return 0; | |
3174 | } | |
3175 | ||
3176 | for (map<int, set<string> >::iterator iter = shards.begin(); iter != shards.end(); ++iter) { | |
3177 | ldout(cct, 20) << __func__ << "(): notifying datalog change, shard_id=" << iter->first << ": " << iter->second << dendl; | |
3178 | } | |
3179 | ||
3180 | notify_mgr.notify_all(store->zone_data_notify_to_map, shards); | |
3181 | ||
3182 | return 0; | |
3183 | } | |
3184 | ||
3185 | class RGWSyncProcessorThread : public RGWRadosThread { | |
3186 | public: | |
3187 | RGWSyncProcessorThread(RGWRados *_store, const string& thread_name = "radosgw") : RGWRadosThread(_store, thread_name) {} | |
3188 | RGWSyncProcessorThread(RGWRados *_store) : RGWRadosThread(_store) {} | |
3189 | ~RGWSyncProcessorThread() override {} | |
3190 | int init() override = 0 ; | |
3191 | int process() override = 0; | |
3192 | }; | |
3193 | ||
3194 | class RGWMetaSyncProcessorThread : public RGWSyncProcessorThread | |
3195 | { | |
3196 | RGWMetaSyncStatusManager sync; | |
3197 | ||
3198 | uint64_t interval_msec() override { | |
3199 | return 0; /* no interval associated, it'll run once until stopped */ | |
3200 | } | |
3201 | void stop_process() override { | |
3202 | sync.stop(); | |
3203 | } | |
3204 | public: | |
3205 | RGWMetaSyncProcessorThread(RGWRados *_store, RGWAsyncRadosProcessor *async_rados) | |
3206 | : RGWSyncProcessorThread(_store, "meta-sync"), sync(_store, async_rados) {} | |
3207 | ||
3208 | void wakeup_sync_shards(set<int>& shard_ids) { | |
3209 | for (set<int>::iterator iter = shard_ids.begin(); iter != shard_ids.end(); ++iter) { | |
3210 | sync.wakeup(*iter); | |
3211 | } | |
3212 | } | |
3213 | RGWMetaSyncStatusManager* get_manager() { return &sync; } | |
3214 | ||
3215 | int init() override { | |
3216 | int ret = sync.init(); | |
3217 | if (ret < 0) { | |
3218 | ldout(store->ctx(), 0) << "ERROR: sync.init() returned " << ret << dendl; | |
3219 | return ret; | |
3220 | } | |
3221 | return 0; | |
3222 | } | |
3223 | ||
3224 | int process() override { | |
3225 | sync.run(); | |
3226 | return 0; | |
3227 | } | |
3228 | }; | |
3229 | ||
3230 | class RGWDataSyncProcessorThread : public RGWSyncProcessorThread | |
3231 | { | |
3232 | RGWDataSyncStatusManager sync; | |
3233 | bool initialized; | |
3234 | ||
3235 | uint64_t interval_msec() override { | |
3236 | if (initialized) { | |
3237 | return 0; /* no interval associated, it'll run once until stopped */ | |
3238 | } else { | |
3239 | #define DATA_SYNC_INIT_WAIT_SEC 20 | |
3240 | return DATA_SYNC_INIT_WAIT_SEC * 1000; | |
3241 | } | |
3242 | } | |
3243 | void stop_process() override { | |
3244 | sync.stop(); | |
3245 | } | |
3246 | public: | |
3247 | RGWDataSyncProcessorThread(RGWRados *_store, RGWAsyncRadosProcessor *async_rados, | |
3248 | const string& _source_zone) | |
3249 | : RGWSyncProcessorThread(_store, "data-sync"), | |
3250 | sync(_store, async_rados, _source_zone), | |
3251 | initialized(false) {} | |
3252 | ||
3253 | void wakeup_sync_shards(map<int, set<string> >& shard_ids) { | |
3254 | for (map<int, set<string> >::iterator iter = shard_ids.begin(); iter != shard_ids.end(); ++iter) { | |
3255 | sync.wakeup(iter->first, iter->second); | |
3256 | } | |
3257 | } | |
3258 | RGWDataSyncStatusManager* get_manager() { return &sync; } | |
3259 | ||
3260 | int init() override { | |
3261 | return 0; | |
3262 | } | |
3263 | ||
3264 | int process() override { | |
3265 | while (!initialized) { | |
3266 | if (going_down()) { | |
3267 | return 0; | |
3268 | } | |
3269 | int ret = sync.init(); | |
3270 | if (ret >= 0) { | |
3271 | initialized = true; | |
3272 | break; | |
3273 | } | |
3274 | /* we'll be back! */ | |
3275 | return 0; | |
3276 | } | |
3277 | sync.run(); | |
3278 | return 0; | |
3279 | } | |
3280 | }; | |
3281 | ||
3282 | class RGWSyncLogTrimThread : public RGWSyncProcessorThread | |
3283 | { | |
3284 | RGWCoroutinesManager crs; | |
3285 | RGWRados *store; | |
3286 | rgw::BucketTrimManager *bucket_trim; | |
3287 | RGWHTTPManager http; | |
3288 | const utime_t trim_interval; | |
3289 | ||
3290 | uint64_t interval_msec() override { return 0; } | |
3291 | void stop_process() override { crs.stop(); } | |
3292 | public: | |
3293 | RGWSyncLogTrimThread(RGWRados *store, rgw::BucketTrimManager *bucket_trim, | |
3294 | int interval) | |
3295 | : RGWSyncProcessorThread(store, "sync-log-trim"), | |
3296 | crs(store->ctx(), store->get_cr_registry()), store(store), | |
3297 | bucket_trim(bucket_trim), | |
3298 | http(store->ctx(), crs.get_completion_mgr()), | |
3299 | trim_interval(interval, 0) | |
3300 | {} | |
3301 | ||
3302 | int init() override { | |
3303 | return http.set_threaded(); | |
3304 | } | |
3305 | int process() override { | |
3306 | list<RGWCoroutinesStack*> stacks; | |
3307 | auto meta = new RGWCoroutinesStack(store->ctx(), &crs); | |
3308 | meta->call(create_meta_log_trim_cr(store, &http, | |
3309 | cct->_conf->rgw_md_log_max_shards, | |
3310 | trim_interval)); | |
3311 | stacks.push_back(meta); | |
3312 | ||
3313 | auto data = new RGWCoroutinesStack(store->ctx(), &crs); | |
3314 | data->call(create_data_log_trim_cr(store, &http, | |
3315 | cct->_conf->rgw_data_log_num_shards, | |
3316 | trim_interval)); | |
3317 | stacks.push_back(data); | |
3318 | ||
3319 | auto bucket = new RGWCoroutinesStack(store->ctx(), &crs); | |
3320 | bucket->call(bucket_trim->create_bucket_trim_cr(&http)); | |
3321 | stacks.push_back(bucket); | |
3322 | ||
3323 | crs.run(stacks); | |
3324 | return 0; | |
3325 | } | |
3326 | }; | |
3327 | ||
3328 | void RGWRados::wakeup_meta_sync_shards(set<int>& shard_ids) | |
3329 | { | |
3330 | Mutex::Locker l(meta_sync_thread_lock); | |
3331 | if (meta_sync_processor_thread) { | |
3332 | meta_sync_processor_thread->wakeup_sync_shards(shard_ids); | |
3333 | } | |
3334 | } | |
3335 | ||
3336 | void RGWRados::wakeup_data_sync_shards(const string& source_zone, map<int, set<string> >& shard_ids) | |
3337 | { | |
3338 | ldout(ctx(), 20) << __func__ << ": source_zone=" << source_zone << ", shard_ids=" << shard_ids << dendl; | |
3339 | Mutex::Locker l(data_sync_thread_lock); | |
3340 | map<string, RGWDataSyncProcessorThread *>::iterator iter = data_sync_processor_threads.find(source_zone); | |
3341 | if (iter == data_sync_processor_threads.end()) { | |
3342 | ldout(ctx(), 10) << __func__ << ": couldn't find sync thread for zone " << source_zone << ", skipping async data sync processing" << dendl; | |
3343 | return; | |
3344 | } | |
3345 | ||
3346 | RGWDataSyncProcessorThread *thread = iter->second; | |
3347 | assert(thread); | |
3348 | thread->wakeup_sync_shards(shard_ids); | |
3349 | } | |
3350 | ||
3351 | RGWMetaSyncStatusManager* RGWRados::get_meta_sync_manager() | |
3352 | { | |
3353 | Mutex::Locker l(meta_sync_thread_lock); | |
3354 | if (meta_sync_processor_thread) { | |
3355 | return meta_sync_processor_thread->get_manager(); | |
3356 | } | |
3357 | return nullptr; | |
3358 | } | |
3359 | ||
3360 | RGWDataSyncStatusManager* RGWRados::get_data_sync_manager(const std::string& source_zone) | |
3361 | { | |
3362 | Mutex::Locker l(data_sync_thread_lock); | |
3363 | auto thread = data_sync_processor_threads.find(source_zone); | |
3364 | if (thread == data_sync_processor_threads.end()) { | |
3365 | return nullptr; | |
3366 | } | |
3367 | return thread->second->get_manager(); | |
3368 | } | |
3369 | ||
3370 | int RGWRados::get_required_alignment(const rgw_pool& pool, uint64_t *alignment) | |
3371 | { | |
3372 | IoCtx ioctx; | |
3373 | int r = open_pool_ctx(pool, ioctx); | |
3374 | if (r < 0) { | |
3375 | ldout(cct, 0) << "ERROR: open_pool_ctx() returned " << r << dendl; | |
3376 | return r; | |
3377 | } | |
3378 | ||
3379 | bool requires; | |
3380 | r = ioctx.pool_requires_alignment2(&requires); | |
3381 | if (r < 0) { | |
3382 | ldout(cct, 0) << "ERROR: ioctx.pool_requires_alignment2() returned " | |
3383 | << r << dendl; | |
3384 | return r; | |
3385 | } | |
3386 | ||
3387 | if (!requires) { | |
3388 | *alignment = 0; | |
3389 | return 0; | |
3390 | } | |
3391 | ||
3392 | uint64_t align; | |
3393 | r = ioctx.pool_required_alignment2(&align); | |
3394 | if (r < 0) { | |
3395 | ldout(cct, 0) << "ERROR: ioctx.pool_required_alignment2() returned " | |
3396 | << r << dendl; | |
3397 | return r; | |
3398 | } | |
3399 | if (align != 0) { | |
3400 | ldout(cct, 20) << "required alignment=" << align << dendl; | |
3401 | } | |
3402 | *alignment = align; | |
3403 | return 0; | |
3404 | } | |
3405 | ||
3406 | int RGWRados::get_max_chunk_size(const rgw_pool& pool, uint64_t *max_chunk_size) | |
3407 | { | |
3408 | uint64_t alignment = 0; | |
3409 | int r = get_required_alignment(pool, &alignment); | |
3410 | if (r < 0) { | |
3411 | return r; | |
3412 | } | |
3413 | ||
3414 | uint64_t config_chunk_size = cct->_conf->rgw_max_chunk_size; | |
3415 | ||
3416 | if (alignment == 0) { | |
3417 | *max_chunk_size = config_chunk_size; | |
3418 | return 0; | |
3419 | } | |
3420 | ||
3421 | if (config_chunk_size <= alignment) { | |
3422 | *max_chunk_size = alignment; | |
3423 | return 0; | |
3424 | } | |
3425 | ||
3426 | *max_chunk_size = config_chunk_size - (config_chunk_size % alignment); | |
3427 | ||
3428 | ldout(cct, 20) << "max_chunk_size=" << *max_chunk_size << dendl; | |
3429 | ||
3430 | return 0; | |
3431 | } | |
3432 | ||
3433 | int RGWRados::get_max_chunk_size(const string& placement_rule, const rgw_obj& obj, uint64_t *max_chunk_size) | |
3434 | { | |
3435 | rgw_pool pool; | |
3436 | if (!get_obj_data_pool(placement_rule, obj, &pool)) { | |
3437 | ldout(cct, 0) << "ERROR: failed to get data pool for object " << obj << dendl; | |
3438 | return -EIO; | |
3439 | } | |
3440 | return get_max_chunk_size(pool, max_chunk_size); | |
3441 | } | |
3442 | ||
3443 | class RGWIndexCompletionManager; | |
3444 | ||
3445 | struct complete_op_data { | |
3446 | Mutex lock{"complete_op_data"}; | |
3447 | AioCompletion *rados_completion{nullptr}; | |
3448 | int manager_shard_id{-1}; | |
3449 | RGWIndexCompletionManager *manager{nullptr}; | |
3450 | rgw_obj obj; | |
3451 | RGWModifyOp op; | |
3452 | string tag; | |
3453 | rgw_bucket_entry_ver ver; | |
3454 | cls_rgw_obj_key key; | |
3455 | rgw_bucket_dir_entry_meta dir_meta; | |
3456 | list<cls_rgw_obj_key> remove_objs; | |
3457 | bool log_op; | |
3458 | uint16_t bilog_op; | |
3459 | rgw_zone_set zones_trace; | |
3460 | ||
3461 | bool stopped{false}; | |
3462 | ||
3463 | void stop() { | |
3464 | Mutex::Locker l(lock); | |
3465 | stopped = true; | |
3466 | } | |
3467 | }; | |
3468 | ||
3469 | class RGWIndexCompletionThread : public RGWRadosThread { | |
3470 | RGWRados *store; | |
3471 | ||
3472 | uint64_t interval_msec() override { | |
3473 | return 0; | |
3474 | } | |
3475 | ||
3476 | list<complete_op_data *> completions; | |
3477 | ||
3478 | Mutex completions_lock; | |
3479 | public: | |
3480 | RGWIndexCompletionThread(RGWRados *_store) | |
3481 | : RGWRadosThread(_store, "index-complete"), store(_store), completions_lock("RGWIndexCompletionThread::completions_lock") {} | |
3482 | ||
3483 | int process() override; | |
3484 | ||
3485 | void add_completion(complete_op_data *completion) { | |
3486 | { | |
3487 | Mutex::Locker l(completions_lock); | |
3488 | completions.push_back(completion); | |
3489 | } | |
3490 | ||
3491 | signal(); | |
3492 | } | |
3493 | }; | |
3494 | ||
3495 | int RGWIndexCompletionThread::process() | |
3496 | { | |
3497 | list<complete_op_data *> comps; | |
3498 | ||
3499 | { | |
3500 | Mutex::Locker l(completions_lock); | |
3501 | completions.swap(comps); | |
3502 | } | |
3503 | ||
3504 | for (auto c : comps) { | |
3505 | std::unique_ptr<complete_op_data> up{c}; | |
3506 | ||
3507 | if (going_down()) { | |
3508 | continue; | |
3509 | } | |
3510 | ldout(store->ctx(), 20) << __func__ << "(): handling completion for key=" << c->key << dendl; | |
3511 | ||
3512 | RGWRados::BucketShard bs(store); | |
3513 | RGWBucketInfo bucket_info; | |
3514 | ||
3515 | int r = bs.init(c->obj.bucket, c->obj, &bucket_info); | |
3516 | if (r < 0) { | |
3517 | ldout(cct, 0) << "ERROR: " << __func__ << "(): failed to initialize BucketShard, obj=" << c->obj << " r=" << r << dendl; | |
3518 | /* not much to do */ | |
3519 | continue; | |
3520 | } | |
3521 | ||
3522 | r = store->guard_reshard(&bs, c->obj, bucket_info, | |
3523 | [&](RGWRados::BucketShard *bs) -> int { | |
3524 | librados::ObjectWriteOperation o; | |
3525 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); | |
3526 | cls_rgw_bucket_complete_op(o, c->op, c->tag, c->ver, c->key, c->dir_meta, &c->remove_objs, | |
3527 | c->log_op, c->bilog_op, &c->zones_trace); | |
3528 | return bs->index_ctx.operate(bs->bucket_obj, &o); | |
3529 | }); | |
3530 | if (r < 0) { | |
3531 | ldout(cct, 0) << "ERROR: " << __func__ << "(): bucket index completion failed, obj=" << c->obj << " r=" << r << dendl; | |
3532 | /* ignoring error, can't do anything about it */ | |
3533 | continue; | |
3534 | } | |
3535 | r = store->data_log->add_entry(bs.bucket, bs.shard_id); | |
3536 | if (r < 0) { | |
3537 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
3538 | } | |
3539 | } | |
3540 | ||
3541 | return 0; | |
3542 | } | |
3543 | ||
3544 | class RGWIndexCompletionManager { | |
3545 | RGWRados *store{nullptr}; | |
3546 | vector<Mutex *> locks; | |
3547 | vector<set<complete_op_data *> > completions; | |
3548 | ||
3549 | RGWIndexCompletionThread *completion_thread{nullptr}; | |
3550 | ||
3551 | int num_shards; | |
3552 | ||
3553 | std::atomic<int> cur_shard {0}; | |
3554 | ||
3555 | ||
3556 | public: | |
3557 | RGWIndexCompletionManager(RGWRados *_store) : store(_store) { | |
3558 | num_shards = store->ctx()->_conf->rgw_thread_pool_size; | |
3559 | ||
3560 | for (int i = 0; i < num_shards; i++) { | |
3561 | char buf[64]; | |
3562 | snprintf(buf, sizeof(buf), "RGWIndexCompletionManager::lock::%d", i); | |
3563 | locks.push_back(new Mutex(buf)); | |
3564 | } | |
3565 | ||
3566 | completions.resize(num_shards); | |
3567 | } | |
3568 | ~RGWIndexCompletionManager() { | |
3569 | stop(); | |
3570 | ||
3571 | for (auto l : locks) { | |
3572 | delete l; | |
3573 | } | |
3574 | } | |
3575 | ||
3576 | int next_shard() { | |
3577 | int result = cur_shard % num_shards; | |
3578 | cur_shard++; | |
3579 | return result; | |
3580 | } | |
3581 | ||
3582 | void create_completion(const rgw_obj& obj, | |
3583 | RGWModifyOp op, string& tag, | |
3584 | rgw_bucket_entry_ver& ver, | |
3585 | const cls_rgw_obj_key& key, | |
3586 | rgw_bucket_dir_entry_meta& dir_meta, | |
3587 | list<cls_rgw_obj_key> *remove_objs, bool log_op, | |
3588 | uint16_t bilog_op, | |
3589 | rgw_zone_set *zones_trace, | |
3590 | complete_op_data **result); | |
3591 | bool handle_completion(completion_t cb, complete_op_data *arg); | |
3592 | ||
3593 | int start() { | |
3594 | completion_thread = new RGWIndexCompletionThread(store); | |
3595 | int ret = completion_thread->init(); | |
3596 | if (ret < 0) { | |
3597 | return ret; | |
3598 | } | |
3599 | completion_thread->start(); | |
3600 | return 0; | |
3601 | } | |
3602 | void stop() { | |
3603 | if (completion_thread) { | |
3604 | completion_thread->stop(); | |
3605 | delete completion_thread; | |
3606 | } | |
3607 | ||
3608 | for (int i = 0; i < num_shards; ++i) { | |
3609 | Mutex::Locker l(*locks[i]); | |
3610 | for (auto c : completions[i]) { | |
3611 | c->stop(); | |
3612 | } | |
3613 | } | |
3614 | completions.clear(); | |
3615 | } | |
3616 | }; | |
3617 | ||
3618 | static void obj_complete_cb(completion_t cb, void *arg) | |
3619 | { | |
3620 | complete_op_data *completion = (complete_op_data *)arg; | |
3621 | completion->lock.Lock(); | |
3622 | if (completion->stopped) { | |
3623 | completion->lock.Unlock(); /* can drop lock, no one else is referencing us */ | |
3624 | delete completion; | |
3625 | return; | |
3626 | } | |
3627 | bool need_delete = completion->manager->handle_completion(cb, completion); | |
3628 | completion->lock.Unlock(); | |
3629 | if (need_delete) { | |
3630 | delete completion; | |
3631 | } | |
3632 | } | |
3633 | ||
3634 | ||
3635 | void RGWIndexCompletionManager::create_completion(const rgw_obj& obj, | |
3636 | RGWModifyOp op, string& tag, | |
3637 | rgw_bucket_entry_ver& ver, | |
3638 | const cls_rgw_obj_key& key, | |
3639 | rgw_bucket_dir_entry_meta& dir_meta, | |
3640 | list<cls_rgw_obj_key> *remove_objs, bool log_op, | |
3641 | uint16_t bilog_op, | |
3642 | rgw_zone_set *zones_trace, | |
3643 | complete_op_data **result) | |
3644 | { | |
3645 | complete_op_data *entry = new complete_op_data; | |
3646 | ||
3647 | int shard_id = next_shard(); | |
3648 | ||
3649 | entry->manager_shard_id = shard_id; | |
3650 | entry->manager = this; | |
3651 | entry->obj = obj; | |
3652 | entry->op = op; | |
3653 | entry->tag = tag; | |
3654 | entry->ver = ver; | |
3655 | entry->key = key; | |
3656 | entry->dir_meta = dir_meta; | |
3657 | entry->log_op = log_op; | |
3658 | entry->bilog_op = bilog_op; | |
3659 | ||
3660 | if (remove_objs) { | |
3661 | for (auto iter = remove_objs->begin(); iter != remove_objs->end(); ++iter) { | |
3662 | entry->remove_objs.push_back(*iter); | |
3663 | } | |
3664 | } | |
3665 | ||
3666 | if (zones_trace) { | |
3667 | entry->zones_trace = *zones_trace; | |
3668 | } else { | |
3669 | entry->zones_trace.insert(store->get_zone().id); | |
3670 | } | |
3671 | ||
3672 | *result = entry; | |
3673 | ||
3674 | entry->rados_completion = librados::Rados::aio_create_completion(entry, NULL, obj_complete_cb); | |
3675 | ||
3676 | Mutex::Locker l(*locks[shard_id]); | |
3677 | completions[shard_id].insert(entry); | |
3678 | } | |
3679 | ||
3680 | bool RGWIndexCompletionManager::handle_completion(completion_t cb, complete_op_data *arg) | |
3681 | { | |
3682 | int shard_id = arg->manager_shard_id; | |
3683 | { | |
3684 | Mutex::Locker l(*locks[shard_id]); | |
3685 | ||
3686 | auto& comps = completions[shard_id]; | |
3687 | ||
3688 | auto iter = comps.find(arg); | |
3689 | if (iter == comps.end()) { | |
3690 | return true; | |
3691 | } | |
3692 | ||
3693 | comps.erase(iter); | |
3694 | } | |
3695 | ||
3696 | int r = rados_aio_get_return_value(cb); | |
3697 | if (r != -ERR_BUSY_RESHARDING) { | |
3698 | return true; | |
3699 | } | |
3700 | completion_thread->add_completion(arg); | |
3701 | return false; | |
3702 | } | |
3703 | ||
3704 | void RGWRados::finalize() | |
3705 | { | |
3706 | auto admin_socket = cct->get_admin_socket(); | |
3707 | for (auto cmd : admin_commands) { | |
3708 | int r = admin_socket->unregister_command(cmd[0]); | |
3709 | if (r < 0) { | |
3710 | lderr(cct) << "ERROR: fail to unregister admin socket command (r=" << r | |
3711 | << ")" << dendl; | |
3712 | } | |
3713 | } | |
3714 | ||
3715 | if (run_sync_thread) { | |
3716 | Mutex::Locker l(meta_sync_thread_lock); | |
3717 | meta_sync_processor_thread->stop(); | |
3718 | ||
3719 | Mutex::Locker dl(data_sync_thread_lock); | |
3720 | for (auto iter : data_sync_processor_threads) { | |
3721 | RGWDataSyncProcessorThread *thread = iter.second; | |
3722 | thread->stop(); | |
3723 | } | |
3724 | if (sync_log_trimmer) { | |
3725 | sync_log_trimmer->stop(); | |
3726 | } | |
3727 | } | |
3728 | if (async_rados) { | |
3729 | async_rados->stop(); | |
3730 | } | |
3731 | if (run_sync_thread) { | |
3732 | delete meta_sync_processor_thread; | |
3733 | meta_sync_processor_thread = NULL; | |
3734 | Mutex::Locker dl(data_sync_thread_lock); | |
3735 | for (auto iter : data_sync_processor_threads) { | |
3736 | RGWDataSyncProcessorThread *thread = iter.second; | |
3737 | delete thread; | |
3738 | } | |
3739 | data_sync_processor_threads.clear(); | |
3740 | delete sync_log_trimmer; | |
3741 | sync_log_trimmer = nullptr; | |
3742 | bucket_trim = boost::none; | |
3743 | } | |
3744 | if (finisher) { | |
3745 | finisher->stop(); | |
3746 | } | |
3747 | if (need_watch_notify()) { | |
3748 | finalize_watch(); | |
3749 | } | |
3750 | if (finisher) { | |
3751 | /* delete finisher only after cleaning up watches, as watch error path might call | |
3752 | * into finisher. We stop finisher before finalizing watch to make sure we don't | |
3753 | * actually handle any racing work | |
3754 | */ | |
3755 | delete finisher; | |
3756 | } | |
3757 | if (meta_notifier) { | |
3758 | meta_notifier->stop(); | |
3759 | delete meta_notifier; | |
3760 | } | |
3761 | if (data_notifier) { | |
3762 | data_notifier->stop(); | |
3763 | delete data_notifier; | |
3764 | } | |
3765 | delete data_log; | |
3766 | if (async_rados) { | |
3767 | delete async_rados; | |
3768 | } | |
3769 | ||
3770 | delete lc; | |
3771 | lc = NULL; | |
3772 | ||
3773 | delete gc; | |
3774 | gc = NULL; | |
3775 | ||
3776 | delete obj_expirer; | |
3777 | obj_expirer = NULL; | |
3778 | ||
3779 | delete rest_master_conn; | |
3780 | ||
3781 | map<string, RGWRESTConn *>::iterator iter; | |
3782 | for (iter = zone_conn_map.begin(); iter != zone_conn_map.end(); ++iter) { | |
3783 | RGWRESTConn *conn = iter->second; | |
3784 | delete conn; | |
3785 | } | |
3786 | ||
3787 | for (iter = zonegroup_conn_map.begin(); iter != zonegroup_conn_map.end(); ++iter) { | |
3788 | RGWRESTConn *conn = iter->second; | |
3789 | delete conn; | |
3790 | } | |
3791 | RGWQuotaHandler::free_handler(quota_handler); | |
3792 | if (cr_registry) { | |
3793 | cr_registry->put(); | |
3794 | } | |
3795 | delete meta_mgr; | |
3796 | delete binfo_cache; | |
3797 | delete obj_tombstone_cache; | |
3798 | delete sync_modules_manager; | |
3799 | ||
3800 | if (reshard_wait.get()) { | |
3801 | reshard_wait->stop(); | |
3802 | reshard_wait.reset(); | |
3803 | } | |
3804 | ||
3805 | if (run_reshard_thread) { | |
3806 | reshard->stop_processor(); | |
3807 | } | |
3808 | delete reshard; | |
3809 | delete index_completion_manager; | |
3810 | } | |
3811 | ||
3812 | /** | |
3813 | * Initialize the RADOS instance and prepare to do other ops | |
3814 | * Returns 0 on success, -ERR# on failure. | |
3815 | */ | |
3816 | int RGWRados::init_rados() | |
3817 | { | |
3818 | int ret = 0; | |
3819 | auto admin_socket = cct->get_admin_socket(); | |
3820 | for (auto cmd : admin_commands) { | |
3821 | int r = admin_socket->register_command(cmd[0], cmd[1], this, | |
3822 | cmd[2]); | |
3823 | if (r < 0) { | |
3824 | lderr(cct) << "ERROR: fail to register admin socket command (r=" << r | |
3825 | << ")" << dendl; | |
3826 | return r; | |
3827 | } | |
3828 | } | |
3829 | ||
3830 | auto handles = std::vector<librados::Rados>{cct->_conf->rgw_num_rados_handles}; | |
3831 | ||
3832 | for (auto& r : handles) { | |
3833 | ret = r.init_with_context(cct); | |
3834 | if (ret < 0) { | |
3835 | return ret; | |
3836 | } | |
3837 | ret = r.connect(); | |
3838 | if (ret < 0) { | |
3839 | return ret; | |
3840 | } | |
3841 | } | |
3842 | ||
3843 | sync_modules_manager = new RGWSyncModulesManager(); | |
3844 | ||
3845 | rgw_register_sync_modules(sync_modules_manager); | |
3846 | ||
3847 | auto crs = std::unique_ptr<RGWCoroutinesManagerRegistry>{ | |
3848 | new RGWCoroutinesManagerRegistry(cct)}; | |
3849 | ret = crs->hook_to_admin_command("cr dump"); | |
3850 | if (ret < 0) { | |
3851 | return ret; | |
3852 | } | |
3853 | ||
3854 | meta_mgr = new RGWMetadataManager(cct, this); | |
3855 | data_log = new RGWDataChangesLog(cct, this); | |
3856 | cr_registry = crs.release(); | |
3857 | ||
3858 | std::swap(handles, rados); | |
3859 | return ret; | |
3860 | } | |
3861 | ||
3862 | ||
3863 | int RGWRados::register_to_service_map(const string& daemon_type, const map<string, string>& meta) | |
3864 | { | |
3865 | map<string,string> metadata = meta; | |
3866 | metadata["num_handles"] = stringify(rados.size()); | |
3867 | metadata["zonegroup_id"] = zonegroup.get_id(); | |
3868 | metadata["zonegroup_name"] = zonegroup.get_name(); | |
3869 | metadata["zone_name"] = zone_name(); | |
3870 | metadata["zone_id"] = zone_id();; | |
3871 | string name = cct->_conf->name.get_id(); | |
3872 | if (name.find("rgw.") == 0) { | |
3873 | name = name.substr(4); | |
3874 | } | |
3875 | int ret = rados[0].service_daemon_register(daemon_type, name, metadata); | |
3876 | if (ret < 0) { | |
3877 | ldout(cct, 0) << "ERROR: service_daemon_register() returned ret=" << ret << ": " << cpp_strerror(-ret) << dendl; | |
3878 | return ret; | |
3879 | } | |
3880 | ||
3881 | return 0; | |
3882 | } | |
3883 | ||
3884 | /** | |
3885 | * Add new connection to connections map | |
3886 | * @param zonegroup_conn_map map which new connection will be added to | |
3887 | * @param zonegroup zonegroup which new connection will connect to | |
3888 | * @param new_connection pointer to new connection instance | |
3889 | */ | |
3890 | static void add_new_connection_to_map(map<string, RGWRESTConn *> &zonegroup_conn_map, | |
3891 | const RGWZoneGroup &zonegroup, RGWRESTConn *new_connection) | |
3892 | { | |
3893 | // Delete if connection is already exists | |
3894 | map<string, RGWRESTConn *>::iterator iterZoneGroup = zonegroup_conn_map.find(zonegroup.get_id()); | |
3895 | if (iterZoneGroup != zonegroup_conn_map.end()) { | |
3896 | delete iterZoneGroup->second; | |
3897 | } | |
3898 | ||
3899 | // Add new connection to connections map | |
3900 | zonegroup_conn_map[zonegroup.get_id()] = new_connection; | |
3901 | } | |
3902 | ||
3903 | int RGWRados::convert_regionmap() | |
3904 | { | |
3905 | RGWZoneGroupMap zonegroupmap; | |
3906 | ||
3907 | string pool_name = cct->_conf->rgw_zone_root_pool; | |
3908 | if (pool_name.empty()) { | |
3909 | pool_name = RGW_DEFAULT_ZONE_ROOT_POOL; | |
3910 | } | |
3911 | string oid = region_map_oid; | |
3912 | ||
3913 | rgw_pool pool(pool_name); | |
3914 | bufferlist bl; | |
3915 | RGWObjectCtx obj_ctx(this); | |
3916 | int ret = rgw_get_system_obj(this, obj_ctx, pool, oid, bl, NULL, NULL); | |
3917 | if (ret < 0 && ret != -ENOENT) { | |
3918 | return ret; | |
3919 | } else if (ret == -ENOENT) { | |
3920 | return 0; | |
3921 | } | |
3922 | ||
3923 | try { | |
3924 | bufferlist::iterator iter = bl.begin(); | |
3925 | ::decode(zonegroupmap, iter); | |
3926 | } catch (buffer::error& err) { | |
3927 | ldout(cct, 0) << "error decoding regionmap from " << pool << ":" << oid << dendl; | |
3928 | return -EIO; | |
3929 | } | |
3930 | ||
3931 | for (map<string, RGWZoneGroup>::iterator iter = zonegroupmap.zonegroups.begin(); | |
3932 | iter != zonegroupmap.zonegroups.end(); ++iter) { | |
3933 | RGWZoneGroup& zonegroup = iter->second; | |
3934 | ret = zonegroup.init(cct, this, false); | |
3935 | ret = zonegroup.update(); | |
3936 | if (ret < 0 && ret != -ENOENT) { | |
3937 | ldout(cct, 0) << "Error could not update zonegroup " << zonegroup.get_name() << ": " << | |
3938 | cpp_strerror(-ret) << dendl; | |
3939 | return ret; | |
3940 | } else if (ret == -ENOENT) { | |
3941 | ret = zonegroup.create(); | |
3942 | if (ret < 0) { | |
3943 | ldout(cct, 0) << "Error could not create " << zonegroup.get_name() << ": " << | |
3944 | cpp_strerror(-ret) << dendl; | |
3945 | return ret; | |
3946 | } | |
3947 | } | |
3948 | } | |
3949 | ||
3950 | current_period.set_user_quota(zonegroupmap.user_quota); | |
3951 | current_period.set_bucket_quota(zonegroupmap.bucket_quota); | |
3952 | ||
3953 | // remove the region_map so we don't try to convert again | |
3954 | rgw_raw_obj obj(pool, oid); | |
3955 | ret = delete_system_obj(obj); | |
3956 | if (ret < 0) { | |
3957 | ldout(cct, 0) << "Error could not remove " << obj | |
3958 | << " after upgrading to zonegroup map: " << cpp_strerror(ret) << dendl; | |
3959 | return ret; | |
3960 | } | |
3961 | ||
3962 | return 0; | |
3963 | } | |
3964 | ||
3965 | /** | |
3966 | * Replace all region configuration with zonegroup for | |
3967 | * backward compatability | |
3968 | * Returns 0 on success, -ERR# on failure. | |
3969 | */ | |
3970 | int RGWRados::replace_region_with_zonegroup() | |
3971 | { | |
3972 | /* copy default region */ | |
3973 | /* convert default region to default zonegroup */ | |
3974 | string default_oid = cct->_conf->rgw_default_region_info_oid; | |
3975 | if (default_oid.empty()) { | |
3976 | default_oid = default_region_info_oid; | |
3977 | } | |
3978 | ||
3979 | ||
3980 | RGWZoneGroup default_zonegroup; | |
3981 | rgw_pool pool{default_zonegroup.get_pool(cct)}; | |
3982 | string oid = "converted"; | |
3983 | bufferlist bl; | |
3984 | RGWObjectCtx obj_ctx(this); | |
3985 | ||
3986 | int ret = rgw_get_system_obj(this, obj_ctx, pool ,oid, bl, NULL, NULL); | |
3987 | if (ret < 0 && ret != -ENOENT) { | |
3988 | ldout(cct, 0) << __func__ << " failed to read converted: ret "<< ret << " " << cpp_strerror(-ret) | |
3989 | << dendl; | |
3990 | return ret; | |
3991 | } else if (ret != -ENOENT) { | |
3992 | ldout(cct, 20) << "System already converted " << dendl; | |
3993 | return 0; | |
3994 | } | |
3995 | ||
3996 | string default_region; | |
3997 | ret = default_zonegroup.init(cct, this, false, true); | |
3998 | if (ret < 0) { | |
3999 | ldout(cct, 0) << __func__ << " failed init default region: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4000 | return ret; | |
4001 | } | |
4002 | ret = default_zonegroup.read_default_id(default_region, true); | |
4003 | if (ret < 0 && ret != -ENOENT) { | |
4004 | ldout(cct, 0) << __func__ << " failed reading old default region: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4005 | return ret; | |
4006 | } | |
4007 | ||
4008 | /* convert regions to zonegroups */ | |
4009 | list<string> regions; | |
4010 | ret = list_regions(regions); | |
4011 | if (ret < 0 && ret != -ENOENT) { | |
4012 | ldout(cct, 0) << __func__ << " failed to list regions: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4013 | return ret; | |
4014 | } else if (ret == -ENOENT || regions.empty()) { | |
4015 | RGWZoneParams zoneparams(default_zone_name); | |
4016 | int ret = zoneparams.init(cct, this); | |
4017 | if (ret < 0 && ret != -ENOENT) { | |
4018 | ldout(cct, 0) << __func__ << ": error initializing default zone params: " << cpp_strerror(-ret) << dendl; | |
4019 | return ret; | |
4020 | } | |
4021 | /* update master zone */ | |
4022 | RGWZoneGroup default_zg(default_zonegroup_name); | |
4023 | ret = default_zg.init(cct, this); | |
4024 | if (ret < 0 && ret != -ENOENT) { | |
4025 | ldout(cct, 0) << __func__ << ": error in initializing default zonegroup: " << cpp_strerror(-ret) << dendl; | |
4026 | return ret; | |
4027 | } | |
4028 | if (ret != -ENOENT && default_zg.master_zone.empty()) { | |
4029 | default_zg.master_zone = zoneparams.get_id(); | |
4030 | return default_zg.update(); | |
4031 | } | |
4032 | return 0; | |
4033 | } | |
4034 | ||
4035 | string master_region, master_zone; | |
4036 | for (list<string>::iterator iter = regions.begin(); iter != regions.end(); ++iter) { | |
4037 | if (*iter != default_zonegroup_name){ | |
4038 | RGWZoneGroup region(*iter); | |
4039 | int ret = region.init(cct, this, true, true); | |
4040 | if (ret < 0) { | |
4041 | ldout(cct, 0) << __func__ << " failed init region "<< *iter << ": " << cpp_strerror(-ret) << dendl; | |
4042 | return ret; | |
4043 | } | |
4044 | if (region.is_master_zonegroup()) { | |
4045 | master_region = region.get_id(); | |
4046 | master_zone = region.master_zone; | |
4047 | } | |
4048 | } | |
4049 | } | |
4050 | ||
4051 | /* create realm if there is none. | |
4052 | The realm name will be the region and zone concatenated | |
4053 | realm id will be mds of its name */ | |
4054 | if (realm.get_id().empty() && !master_region.empty() && !master_zone.empty()) { | |
4055 | string new_realm_name = master_region + "." + master_zone; | |
4056 | unsigned char md5[CEPH_CRYPTO_MD5_DIGESTSIZE]; | |
4057 | char md5_str[CEPH_CRYPTO_MD5_DIGESTSIZE * 2 + 1]; | |
4058 | MD5 hash; | |
4059 | hash.Update((const byte *)new_realm_name.c_str(), new_realm_name.length()); | |
4060 | hash.Final(md5); | |
4061 | buf_to_hex(md5, CEPH_CRYPTO_MD5_DIGESTSIZE, md5_str); | |
4062 | string new_realm_id(md5_str); | |
4063 | RGWRealm new_realm(new_realm_id,new_realm_name); | |
4064 | ret = new_realm.init(cct, this, false); | |
4065 | if (ret < 0) { | |
4066 | ldout(cct, 0) << __func__ << " Error initing new realm: " << cpp_strerror(-ret) << dendl; | |
4067 | return ret; | |
4068 | } | |
4069 | ret = new_realm.create(); | |
4070 | if (ret < 0 && ret != -EEXIST) { | |
4071 | ldout(cct, 0) << __func__ << " Error creating new realm: " << cpp_strerror(-ret) << dendl; | |
4072 | return ret; | |
4073 | } | |
4074 | ret = new_realm.set_as_default(); | |
4075 | if (ret < 0) { | |
4076 | ldout(cct, 0) << __func__ << " Error setting realm as default: " << cpp_strerror(-ret) << dendl; | |
4077 | return ret; | |
4078 | } | |
4079 | ret = realm.init(cct, this); | |
4080 | if (ret < 0) { | |
4081 | ldout(cct, 0) << __func__ << " Error initing realm: " << cpp_strerror(-ret) << dendl; | |
4082 | return ret; | |
4083 | } | |
4084 | ret = current_period.init(cct, this, realm.get_id(), realm.get_name()); | |
4085 | if (ret < 0) { | |
4086 | ldout(cct, 0) << __func__ << " Error initing current period: " << cpp_strerror(-ret) << dendl; | |
4087 | return ret; | |
4088 | } | |
4089 | } | |
4090 | ||
4091 | list<string>::iterator iter; | |
4092 | /* create zonegroups */ | |
4093 | for (iter = regions.begin(); iter != regions.end(); ++iter) | |
4094 | { | |
4095 | ldout(cct, 0) << __func__ << "Converting " << *iter << dendl; | |
4096 | /* check to see if we don't have already a zonegroup with this name */ | |
4097 | RGWZoneGroup new_zonegroup(*iter); | |
4098 | ret = new_zonegroup.init(cct , this); | |
4099 | if (ret == 0 && new_zonegroup.get_id() != *iter) { | |
4100 | ldout(cct, 0) << __func__ << " zonegroup "<< *iter << " already exists id " << new_zonegroup.get_id () << | |
4101 | " skipping conversion " << dendl; | |
4102 | continue; | |
4103 | } | |
4104 | RGWZoneGroup zonegroup(*iter); | |
4105 | zonegroup.set_id(*iter); | |
4106 | int ret = zonegroup.init(cct, this, true, true); | |
4107 | if (ret < 0) { | |
4108 | ldout(cct, 0) << __func__ << " failed init zonegroup: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4109 | return ret; | |
4110 | } | |
4111 | zonegroup.realm_id = realm.get_id(); | |
4112 | /* fix default region master zone */ | |
4113 | if (*iter == default_zonegroup_name && zonegroup.master_zone.empty()) { | |
4114 | ldout(cct, 0) << __func__ << " Setting default zone as master for default region" << dendl; | |
4115 | zonegroup.master_zone = default_zone_name; | |
4116 | } | |
4117 | ret = zonegroup.update(); | |
4118 | if (ret < 0 && ret != -EEXIST) { | |
4119 | ldout(cct, 0) << __func__ << " failed to update zonegroup " << *iter << ": ret "<< ret << " " << cpp_strerror(-ret) | |
4120 | << dendl; | |
4121 | return ret; | |
4122 | } | |
4123 | ret = zonegroup.update_name(); | |
4124 | if (ret < 0 && ret != -EEXIST) { | |
4125 | ldout(cct, 0) << __func__ << " failed to update_name for zonegroup " << *iter << ": ret "<< ret << " " << cpp_strerror(-ret) | |
4126 | << dendl; | |
4127 | return ret; | |
4128 | } | |
4129 | if (zonegroup.get_name() == default_region) { | |
4130 | ret = zonegroup.set_as_default(); | |
4131 | if (ret < 0) { | |
4132 | ldout(cct, 0) << __func__ << " failed to set_as_default " << *iter << ": ret "<< ret << " " << cpp_strerror(-ret) | |
4133 | << dendl; | |
4134 | return ret; | |
4135 | } | |
4136 | } | |
4137 | for (map<string, RGWZone>::const_iterator iter = zonegroup.zones.begin(); iter != zonegroup.zones.end(); | |
4138 | ++iter) { | |
4139 | ldout(cct, 0) << __func__ << " Converting zone" << iter->first << dendl; | |
4140 | RGWZoneParams zoneparams(iter->first, iter->first); | |
4141 | zoneparams.set_id(iter->first); | |
4142 | zoneparams.realm_id = realm.get_id(); | |
4143 | ret = zoneparams.init(cct, this); | |
4144 | if (ret < 0 && ret != -ENOENT) { | |
4145 | ldout(cct, 0) << __func__ << " failed to init zoneparams " << iter->first << ": " << cpp_strerror(-ret) << dendl; | |
4146 | return ret; | |
4147 | } else if (ret == -ENOENT) { | |
4148 | ldout(cct, 0) << __func__ << " zone is part of another cluster " << iter->first << " skipping " << dendl; | |
4149 | continue; | |
4150 | } | |
4151 | zonegroup.realm_id = realm.get_id(); | |
4152 | ret = zoneparams.update(); | |
4153 | if (ret < 0 && ret != -EEXIST) { | |
4154 | ldout(cct, 0) << __func__ << " failed to update zoneparams " << iter->first << ": " << cpp_strerror(-ret) << dendl; | |
4155 | return ret; | |
4156 | } | |
4157 | ret = zoneparams.update_name(); | |
4158 | if (ret < 0 && ret != -EEXIST) { | |
4159 | ldout(cct, 0) << __func__ << " failed to init zoneparams " << iter->first << ": " << cpp_strerror(-ret) << dendl; | |
4160 | return ret; | |
4161 | } | |
4162 | } | |
4163 | ||
4164 | if (!current_period.get_id().empty()) { | |
4165 | ret = current_period.add_zonegroup(zonegroup); | |
4166 | if (ret < 0) { | |
4167 | ldout(cct, 0) << __func__ << " failed to add zonegroup to current_period: " << cpp_strerror(-ret) << dendl; | |
4168 | return ret; | |
4169 | } | |
4170 | } | |
4171 | } | |
4172 | ||
4173 | if (!current_period.get_id().empty()) { | |
4174 | ret = current_period.update(); | |
4175 | if (ret < 0) { | |
4176 | ldout(cct, 0) << __func__ << " failed to update new period: " << cpp_strerror(-ret) << dendl; | |
4177 | return ret; | |
4178 | } | |
4179 | ret = current_period.store_info(false); | |
4180 | if (ret < 0) { | |
4181 | ldout(cct, 0) << __func__ << " failed to store new period: " << cpp_strerror(-ret) << dendl; | |
4182 | return ret; | |
4183 | } | |
4184 | ret = current_period.reflect(); | |
4185 | if (ret < 0) { | |
4186 | ldout(cct, 0) << __func__ << " failed to update local objects: " << cpp_strerror(-ret) << dendl; | |
4187 | return ret; | |
4188 | } | |
4189 | } | |
4190 | ||
4191 | for (auto const& iter : regions) { | |
4192 | RGWZoneGroup zonegroup(iter); | |
4193 | int ret = zonegroup.init(cct, this, true, true); | |
4194 | if (ret < 0) { | |
4195 | ldout(cct, 0) << __func__ << " failed init zonegroup" << iter << ": ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4196 | return ret; | |
4197 | } | |
4198 | ret = zonegroup.delete_obj(true); | |
4199 | if (ret < 0 && ret != -ENOENT) { | |
4200 | ldout(cct, 0) << __func__ << " failed to delete region " << iter << ": ret "<< ret << " " << cpp_strerror(-ret) | |
4201 | << dendl; | |
4202 | return ret; | |
4203 | } | |
4204 | } | |
4205 | ||
4206 | /* mark as converted */ | |
4207 | ret = rgw_put_system_obj(this, pool, oid, bl.c_str(), bl.length(), | |
4208 | true, NULL, real_time(), NULL); | |
4209 | if (ret < 0 ) { | |
4210 | ldout(cct, 0) << __func__ << " failed to mark cluster as converted: ret "<< ret << " " << cpp_strerror(-ret) | |
4211 | << dendl; | |
4212 | return ret; | |
4213 | } | |
4214 | ||
4215 | return 0; | |
4216 | } | |
4217 | ||
4218 | int RGWRados::init_zg_from_period(bool *initialized) | |
4219 | { | |
4220 | *initialized = false; | |
4221 | ||
4222 | if (current_period.get_id().empty()) { | |
4223 | return 0; | |
4224 | } | |
4225 | ||
4226 | int ret = zonegroup.init(cct, this); | |
4227 | ldout(cct, 20) << "period zonegroup init ret " << ret << dendl; | |
4228 | if (ret == -ENOENT) { | |
4229 | return 0; | |
4230 | } | |
4231 | if (ret < 0) { | |
4232 | ldout(cct, 0) << "failed reading zonegroup info: " << cpp_strerror(-ret) << dendl; | |
4233 | return ret; | |
4234 | } | |
4235 | ldout(cct, 20) << "period zonegroup name " << zonegroup.get_name() << dendl; | |
4236 | ||
4237 | map<string, RGWZoneGroup>::const_iterator iter = | |
4238 | current_period.get_map().zonegroups.find(zonegroup.get_id()); | |
4239 | ||
4240 | if (iter != current_period.get_map().zonegroups.end()) { | |
4241 | ldout(cct, 20) << "using current period zonegroup " << zonegroup.get_name() << dendl; | |
4242 | zonegroup = iter->second; | |
4243 | ret = zonegroup.init(cct, this, false); | |
4244 | if (ret < 0) { | |
4245 | ldout(cct, 0) << "failed init zonegroup: " << " " << cpp_strerror(-ret) << dendl; | |
4246 | return ret; | |
4247 | } | |
4248 | ret = zone_params.init(cct, this); | |
4249 | if (ret < 0 && ret != -ENOENT) { | |
4250 | ldout(cct, 0) << "failed reading zone params info: " << " " << cpp_strerror(-ret) << dendl; | |
4251 | return ret; | |
4252 | } if (ret ==-ENOENT && zonegroup.get_name() == default_zonegroup_name) { | |
4253 | ldout(cct, 10) << " Using default name "<< default_zone_name << dendl; | |
4254 | zone_params.set_name(default_zone_name); | |
4255 | ret = zone_params.init(cct, this); | |
4256 | if (ret < 0 && ret != -ENOENT) { | |
4257 | ldout(cct, 0) << "failed reading zone params info: " << " " << cpp_strerror(-ret) << dendl; | |
4258 | return ret; | |
4259 | } | |
4260 | } | |
4261 | } | |
4262 | for (iter = current_period.get_map().zonegroups.begin(); | |
4263 | iter != current_period.get_map().zonegroups.end(); ++iter){ | |
4264 | const RGWZoneGroup& zg = iter->second; | |
4265 | // use endpoints from the zonegroup's master zone | |
4266 | auto master = zg.zones.find(zg.master_zone); | |
4267 | if (master == zg.zones.end()) { | |
4268 | // Check for empty zonegroup which can happen if zone was deleted before removal | |
4269 | if (zg.zones.size() == 0) | |
4270 | continue; | |
4271 | // fix missing master zone for a single zone zonegroup | |
4272 | if (zg.master_zone.empty() && zg.zones.size() == 1) { | |
4273 | master = zg.zones.begin(); | |
4274 | ldout(cct, 0) << "zonegroup " << zg.get_name() << " missing master_zone, setting zone " << | |
4275 | master->second.name << " id:" << master->second.id << " as master" << dendl; | |
4276 | if (zonegroup.get_id() == zg.get_id()) { | |
4277 | zonegroup.master_zone = master->second.id; | |
4278 | ret = zonegroup.update(); | |
4279 | if (ret < 0) { | |
4280 | ldout(cct, 0) << "error updating zonegroup : " << cpp_strerror(-ret) << dendl; | |
4281 | return ret; | |
4282 | } | |
4283 | } else { | |
4284 | RGWZoneGroup fixed_zg(zg.get_id(),zg.get_name()); | |
4285 | ret = fixed_zg.init(cct, this); | |
4286 | if (ret < 0) { | |
4287 | ldout(cct, 0) << "error initializing zonegroup : " << cpp_strerror(-ret) << dendl; | |
4288 | return ret; | |
4289 | } | |
4290 | fixed_zg.master_zone = master->second.id; | |
4291 | ret = fixed_zg.update(); | |
4292 | if (ret < 0) { | |
4293 | ldout(cct, 0) << "error initializing zonegroup : " << cpp_strerror(-ret) << dendl; | |
4294 | return ret; | |
4295 | } | |
4296 | } | |
4297 | } else { | |
4298 | ldout(cct, 0) << "zonegroup " << zg.get_name() << " missing zone for master_zone=" << | |
4299 | zg.master_zone << dendl; | |
4300 | return -EINVAL; | |
4301 | } | |
4302 | } | |
4303 | const auto& endpoints = master->second.endpoints; | |
4304 | add_new_connection_to_map(zonegroup_conn_map, zg, new RGWRESTConn(cct, this, zg.get_id(), endpoints)); | |
4305 | if (!current_period.get_master_zonegroup().empty() && | |
4306 | zg.get_id() == current_period.get_master_zonegroup()) { | |
4307 | rest_master_conn = new RGWRESTConn(cct, this, zg.get_id(), endpoints); | |
4308 | } | |
4309 | } | |
4310 | ||
4311 | *initialized = true; | |
4312 | ||
4313 | return 0; | |
4314 | } | |
4315 | ||
4316 | int RGWRados::init_zg_from_local(bool *creating_defaults) | |
4317 | { | |
4318 | int ret = zonegroup.init(cct, this); | |
4319 | if ( (ret < 0 && ret != -ENOENT) || (ret == -ENOENT && !cct->_conf->rgw_zonegroup.empty())) { | |
4320 | ldout(cct, 0) << "failed reading zonegroup info: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4321 | return ret; | |
4322 | } else if (ret == -ENOENT) { | |
4323 | *creating_defaults = true; | |
4324 | ldout(cct, 10) << "Creating default zonegroup " << dendl; | |
4325 | ret = zonegroup.create_default(); | |
4326 | if (ret < 0) { | |
4327 | ldout(cct, 0) << "failure in zonegroup create_default: ret "<< ret << " " << cpp_strerror(-ret) | |
4328 | << dendl; | |
4329 | return ret; | |
4330 | } | |
4331 | ret = zonegroup.init(cct, this); | |
4332 | if (ret < 0) { | |
4333 | ldout(cct, 0) << "failure in zonegroup create_default: ret "<< ret << " " << cpp_strerror(-ret) | |
4334 | << dendl; | |
4335 | return ret; | |
4336 | } | |
4337 | } | |
4338 | ldout(cct, 20) << "zonegroup " << zonegroup.get_name() << dendl; | |
4339 | if (zonegroup.is_master_zonegroup()) { | |
4340 | // use endpoints from the zonegroup's master zone | |
4341 | auto master = zonegroup.zones.find(zonegroup.master_zone); | |
4342 | if (master == zonegroup.zones.end()) { | |
4343 | // fix missing master zone for a single zone zonegroup | |
4344 | if (zonegroup.master_zone.empty() && zonegroup.zones.size() == 1) { | |
4345 | master = zonegroup.zones.begin(); | |
4346 | ldout(cct, 0) << "zonegroup " << zonegroup.get_name() << " missing master_zone, setting zone " << | |
4347 | master->second.name << " id:" << master->second.id << " as master" << dendl; | |
4348 | zonegroup.master_zone = master->second.id; | |
4349 | ret = zonegroup.update(); | |
4350 | if (ret < 0) { | |
4351 | ldout(cct, 0) << "error initializing zonegroup : " << cpp_strerror(-ret) << dendl; | |
4352 | return ret; | |
4353 | } | |
4354 | } else { | |
4355 | ldout(cct, 0) << "zonegroup " << zonegroup.get_name() << " missing zone for " | |
4356 | "master_zone=" << zonegroup.master_zone << dendl; | |
4357 | return -EINVAL; | |
4358 | } | |
4359 | } | |
4360 | const auto& endpoints = master->second.endpoints; | |
4361 | rest_master_conn = new RGWRESTConn(cct, this, zonegroup.get_id(), endpoints); | |
4362 | } | |
4363 | ||
4364 | return 0; | |
4365 | } | |
4366 | ||
4367 | ||
4368 | bool RGWRados::zone_syncs_from(RGWZone& target_zone, RGWZone& source_zone) | |
4369 | { | |
4370 | return target_zone.syncs_from(source_zone.name) && | |
4371 | sync_modules_manager->supports_data_export(source_zone.tier_type); | |
4372 | } | |
4373 | ||
4374 | /** | |
4375 | * Initialize the RADOS instance and prepare to do other ops | |
4376 | * Returns 0 on success, -ERR# on failure. | |
4377 | */ | |
4378 | int RGWRados::init_complete() | |
4379 | { | |
4380 | int ret = realm.init(cct, this); | |
4381 | if (ret < 0 && ret != -ENOENT) { | |
4382 | ldout(cct, 0) << "failed reading realm info: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4383 | return ret; | |
4384 | } else if (ret != -ENOENT) { | |
4385 | ldout(cct, 20) << "realm " << realm.get_name() << " " << realm.get_id() << dendl; | |
4386 | ret = current_period.init(cct, this, realm.get_id(), realm.get_name()); | |
4387 | if (ret < 0 && ret != -ENOENT) { | |
4388 | ldout(cct, 0) << "failed reading current period info: " << " " << cpp_strerror(-ret) << dendl; | |
4389 | return ret; | |
4390 | } | |
4391 | ldout(cct, 20) << "current period " << current_period.get_id() << dendl; | |
4392 | } | |
4393 | ||
4394 | ret = replace_region_with_zonegroup(); | |
4395 | if (ret < 0) { | |
4396 | lderr(cct) << "failed converting region to zonegroup : ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4397 | return ret; | |
4398 | } | |
4399 | ||
4400 | ret = convert_regionmap(); | |
4401 | if (ret < 0) { | |
4402 | lderr(cct) << "failed converting regionmap: " << cpp_strerror(-ret) << dendl; | |
4403 | return ret; | |
4404 | } | |
4405 | ||
4406 | bool zg_initialized = false; | |
4407 | ||
4408 | if (!current_period.get_id().empty()) { | |
4409 | ret = init_zg_from_period(&zg_initialized); | |
4410 | if (ret < 0) { | |
4411 | return ret; | |
4412 | } | |
4413 | } | |
4414 | ||
4415 | bool creating_defaults = false; | |
4416 | bool using_local = (!zg_initialized); | |
4417 | if (using_local) { | |
4418 | ldout(cct, 10) << " cannot find current period zonegroup using local zonegroup" << dendl; | |
4419 | ret = init_zg_from_local(&creating_defaults); | |
4420 | if (ret < 0) { | |
4421 | return ret; | |
4422 | } | |
4423 | // read period_config into current_period | |
4424 | auto& period_config = current_period.get_config(); | |
4425 | ret = period_config.read(this, zonegroup.realm_id); | |
4426 | if (ret < 0 && ret != -ENOENT) { | |
4427 | ldout(cct, 0) << "ERROR: failed to read period config: " | |
4428 | << cpp_strerror(ret) << dendl; | |
4429 | return ret; | |
4430 | } | |
4431 | } | |
4432 | ||
4433 | ldout(cct, 10) << "Cannot find current period zone using local zone" << dendl; | |
4434 | if (creating_defaults && cct->_conf->rgw_zone.empty()) { | |
4435 | ldout(cct, 10) << " Using default name "<< default_zone_name << dendl; | |
4436 | zone_params.set_name(default_zone_name); | |
4437 | } | |
4438 | ||
4439 | ret = zone_params.init(cct, this); | |
4440 | if (ret < 0 && ret != -ENOENT) { | |
4441 | lderr(cct) << "failed reading zone info: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4442 | return ret; | |
4443 | } | |
4444 | map<string, RGWZone>::iterator zone_iter = get_zonegroup().zones.find(zone_params.get_id()); | |
4445 | if (zone_iter == get_zonegroup().zones.end()) { | |
4446 | if (using_local) { | |
4447 | lderr(cct) << "Cannot find zone id=" << zone_params.get_id() << " (name=" << zone_params.get_name() << ")" << dendl; | |
4448 | return -EINVAL; | |
4449 | } | |
4450 | ldout(cct, 1) << "Cannot find zone id=" << zone_params.get_id() << " (name=" << zone_params.get_name() << "), switching to local zonegroup configuration" << dendl; | |
4451 | ret = init_zg_from_local(&creating_defaults); | |
4452 | if (ret < 0) { | |
4453 | return ret; | |
4454 | } | |
4455 | zone_iter = get_zonegroup().zones.find(zone_params.get_id()); | |
4456 | } | |
4457 | if (zone_iter != get_zonegroup().zones.end()) { | |
4458 | zone_public_config = zone_iter->second; | |
4459 | ldout(cct, 20) << "zone " << zone_params.get_name() << dendl; | |
4460 | } else { | |
4461 | lderr(cct) << "Cannot find zone id=" << zone_params.get_id() << " (name=" << zone_params.get_name() << ")" << dendl; | |
4462 | return -EINVAL; | |
4463 | } | |
4464 | ||
4465 | zone_short_id = current_period.get_map().get_zone_short_id(zone_params.get_id()); | |
4466 | ||
4467 | if (run_sync_thread) { | |
4468 | ret = sync_modules_manager->create_instance(cct, zone_public_config.tier_type, zone_params.tier_config, &sync_module); | |
4469 | if (ret < 0) { | |
4470 | lderr(cct) << "ERROR: failed to init sync module instance, ret=" << ret << dendl; | |
4471 | return ret; | |
4472 | } | |
4473 | } | |
4474 | ||
4475 | writeable_zone = (zone_public_config.tier_type.empty() || zone_public_config.tier_type == "rgw"); | |
4476 | ||
4477 | init_unique_trans_id_deps(); | |
4478 | ||
4479 | finisher = new Finisher(cct); | |
4480 | finisher->start(); | |
4481 | ||
4482 | period_puller.reset(new RGWPeriodPuller(this)); | |
4483 | period_history.reset(new RGWPeriodHistory(cct, period_puller.get(), | |
4484 | current_period)); | |
4485 | ||
4486 | if (need_watch_notify()) { | |
4487 | ret = init_watch(); | |
4488 | if (ret < 0) { | |
4489 | lderr(cct) << "ERROR: failed to initialize watch: " << cpp_strerror(-ret) << dendl; | |
4490 | return ret; | |
4491 | } | |
4492 | } | |
4493 | ||
4494 | /* first build all zones index */ | |
4495 | for (auto ziter : get_zonegroup().zones) { | |
4496 | const string& id = ziter.first; | |
4497 | RGWZone& z = ziter.second; | |
4498 | zone_id_by_name[z.name] = id; | |
4499 | zone_by_id[id] = z; | |
4500 | } | |
4501 | ||
4502 | if (zone_by_id.find(zone_id()) == zone_by_id.end()) { | |
4503 | ldout(cct, 0) << "WARNING: could not find zone config in zonegroup for local zone (" << zone_id() << "), will use defaults" << dendl; | |
4504 | } | |
4505 | zone_public_config = zone_by_id[zone_id()]; | |
4506 | for (auto ziter : get_zonegroup().zones) { | |
4507 | const string& id = ziter.first; | |
4508 | RGWZone& z = ziter.second; | |
4509 | if (id == zone_id()) { | |
4510 | continue; | |
4511 | } | |
4512 | if (z.endpoints.empty()) { | |
4513 | ldout(cct, 0) << "WARNING: can't generate connection for zone " << z.id << " id " << z.name << ": no endpoints defined" << dendl; | |
4514 | continue; | |
4515 | } | |
4516 | ldout(cct, 20) << "generating connection object for zone " << z.name << " id " << z.id << dendl; | |
4517 | RGWRESTConn *conn = new RGWRESTConn(cct, this, z.id, z.endpoints); | |
4518 | zone_conn_map[id] = conn; | |
4519 | if (zone_syncs_from(zone_public_config, z) || | |
4520 | zone_syncs_from(z, zone_public_config)) { | |
4521 | if (zone_syncs_from(zone_public_config, z)) { | |
4522 | zone_data_sync_from_map[id] = conn; | |
4523 | } | |
4524 | if (zone_syncs_from(z, zone_public_config)) { | |
4525 | zone_data_notify_to_map[id] = conn; | |
4526 | } | |
4527 | } else { | |
4528 | ldout(cct, 20) << "NOTICE: not syncing to/from zone " << z.name << " id " << z.id << dendl; | |
4529 | } | |
4530 | } | |
4531 | ||
4532 | ret = open_root_pool_ctx(); | |
4533 | if (ret < 0) | |
4534 | return ret; | |
4535 | ||
4536 | ret = open_gc_pool_ctx(); | |
4537 | if (ret < 0) | |
4538 | return ret; | |
4539 | ||
4540 | ret = open_lc_pool_ctx(); | |
4541 | if (ret < 0) | |
4542 | return ret; | |
4543 | ||
4544 | ret = open_objexp_pool_ctx(); | |
4545 | if (ret < 0) | |
4546 | return ret; | |
4547 | ||
4548 | ret = open_reshard_pool_ctx(); | |
4549 | if (ret < 0) | |
4550 | return ret; | |
4551 | ||
4552 | pools_initialized = true; | |
4553 | ||
4554 | gc = new RGWGC(); | |
4555 | gc->initialize(cct, this); | |
4556 | ||
4557 | obj_expirer = new RGWObjectExpirer(this); | |
4558 | ||
4559 | if (use_gc_thread) { | |
4560 | gc->start_processor(); | |
4561 | obj_expirer->start_processor(); | |
4562 | } | |
4563 | ||
4564 | /* no point of running sync thread if we don't have a master zone configured | |
4565 | or there is no rest_master_conn */ | |
4566 | if (get_zonegroup().master_zone.empty() || !rest_master_conn | |
4567 | || current_period.get_id().empty()) { | |
4568 | run_sync_thread = false; | |
4569 | } | |
4570 | ||
4571 | if (run_sync_thread) { | |
4572 | // initialize the log period history | |
4573 | meta_mgr->init_oldest_log_period(); | |
4574 | } | |
4575 | ||
4576 | async_rados = new RGWAsyncRadosProcessor(this, cct->_conf->rgw_num_async_rados_threads); | |
4577 | async_rados->start(); | |
4578 | ||
4579 | ret = meta_mgr->init(current_period.get_id()); | |
4580 | if (ret < 0) { | |
4581 | lderr(cct) << "ERROR: failed to initialize metadata log: " | |
4582 | << cpp_strerror(-ret) << dendl; | |
4583 | return ret; | |
4584 | } | |
4585 | ||
4586 | if (is_meta_master()) { | |
4587 | auto md_log = meta_mgr->get_log(current_period.get_id()); | |
4588 | meta_notifier = new RGWMetaNotifier(this, md_log); | |
4589 | meta_notifier->start(); | |
4590 | } | |
4591 | ||
4592 | if (run_sync_thread) { | |
4593 | Mutex::Locker l(meta_sync_thread_lock); | |
4594 | meta_sync_processor_thread = new RGWMetaSyncProcessorThread(this, async_rados); | |
4595 | ret = meta_sync_processor_thread->init(); | |
4596 | if (ret < 0) { | |
4597 | ldout(cct, 0) << "ERROR: failed to initialize meta sync thread" << dendl; | |
4598 | return ret; | |
4599 | } | |
4600 | meta_sync_processor_thread->start(); | |
4601 | ||
4602 | // configure the bucket trim manager | |
4603 | rgw::BucketTrimConfig config; | |
4604 | rgw::configure_bucket_trim(cct, config); | |
4605 | ||
4606 | bucket_trim.emplace(this, config); | |
4607 | ret = bucket_trim->init(); | |
4608 | if (ret < 0) { | |
4609 | ldout(cct, 0) << "ERROR: failed to start bucket trim manager" << dendl; | |
4610 | return ret; | |
4611 | } | |
4612 | data_log->set_observer(&*bucket_trim); | |
4613 | ||
4614 | Mutex::Locker dl(data_sync_thread_lock); | |
4615 | for (auto iter : zone_data_sync_from_map) { | |
4616 | ldout(cct, 5) << "starting data sync thread for zone " << iter.first << dendl; | |
4617 | auto *thread = new RGWDataSyncProcessorThread(this, async_rados, iter.first); | |
4618 | ret = thread->init(); | |
4619 | if (ret < 0) { | |
4620 | ldout(cct, 0) << "ERROR: failed to initialize data sync thread" << dendl; | |
4621 | return ret; | |
4622 | } | |
4623 | thread->start(); | |
4624 | data_sync_processor_threads[iter.first] = thread; | |
4625 | } | |
4626 | auto interval = cct->_conf->rgw_sync_log_trim_interval; | |
4627 | if (interval > 0) { | |
4628 | sync_log_trimmer = new RGWSyncLogTrimThread(this, &*bucket_trim, interval); | |
4629 | ret = sync_log_trimmer->init(); | |
4630 | if (ret < 0) { | |
4631 | ldout(cct, 0) << "ERROR: failed to initialize sync log trim thread" << dendl; | |
4632 | return ret; | |
4633 | } | |
4634 | sync_log_trimmer->start(); | |
4635 | } | |
4636 | } | |
4637 | data_notifier = new RGWDataNotifier(this); | |
4638 | data_notifier->start(); | |
4639 | ||
4640 | lc = new RGWLC(); | |
4641 | lc->initialize(cct, this); | |
4642 | ||
4643 | if (use_lc_thread) | |
4644 | lc->start_processor(); | |
4645 | ||
4646 | quota_handler = RGWQuotaHandler::generate_handler(this, quota_threads); | |
4647 | ||
4648 | bucket_index_max_shards = (cct->_conf->rgw_override_bucket_index_max_shards ? cct->_conf->rgw_override_bucket_index_max_shards : | |
4649 | get_zone().bucket_index_max_shards); | |
4650 | if (bucket_index_max_shards > get_max_bucket_shards()) { | |
4651 | bucket_index_max_shards = get_max_bucket_shards(); | |
4652 | ldout(cct, 1) << __func__ << " bucket index max shards is too large, reset to value: " | |
4653 | << get_max_bucket_shards() << dendl; | |
4654 | } | |
4655 | ldout(cct, 20) << __func__ << " bucket index max shards: " << bucket_index_max_shards << dendl; | |
4656 | ||
4657 | binfo_cache = new RGWChainedCacheImpl<bucket_info_entry>; | |
4658 | binfo_cache->init(this); | |
4659 | ||
4660 | bool need_tombstone_cache = !zone_data_notify_to_map.empty(); /* have zones syncing from us */ | |
4661 | ||
4662 | if (need_tombstone_cache) { | |
4663 | obj_tombstone_cache = new tombstone_cache_t(cct->_conf->rgw_obj_tombstone_cache_size); | |
4664 | } | |
4665 | ||
4666 | reshard_wait = std::make_shared<RGWReshardWait>(this); | |
4667 | ||
4668 | reshard = new RGWReshard(this); | |
4669 | ||
4670 | /* only the master zone in the zonegroup reshards buckets */ | |
4671 | run_reshard_thread = run_reshard_thread && (get_zonegroup().master_zone == zone_public_config.id); | |
4672 | if (run_reshard_thread) { | |
4673 | reshard->start_processor(); | |
4674 | } | |
4675 | ||
4676 | index_completion_manager = new RGWIndexCompletionManager(this); | |
4677 | ret = index_completion_manager->start(); | |
4678 | ||
4679 | return ret; | |
4680 | } | |
4681 | ||
4682 | /** | |
4683 | * Initialize the RADOS instance and prepare to do other ops | |
4684 | * Returns 0 on success, -ERR# on failure. | |
4685 | */ | |
4686 | int RGWRados::initialize() | |
4687 | { | |
4688 | int ret; | |
4689 | ||
4690 | ret = init_rados(); | |
4691 | if (ret < 0) | |
4692 | return ret; | |
4693 | ||
4694 | return init_complete(); | |
4695 | } | |
4696 | ||
4697 | void RGWRados::finalize_watch() | |
4698 | { | |
4699 | for (int i = 0; i < num_watchers; i++) { | |
4700 | RGWWatcher *watcher = watchers[i]; | |
4701 | watcher->unregister_watch(); | |
4702 | delete watcher; | |
4703 | } | |
4704 | ||
4705 | delete[] notify_oids; | |
4706 | delete[] watchers; | |
4707 | } | |
4708 | ||
4709 | void RGWRados::schedule_context(Context *c) { | |
4710 | finisher->queue(c); | |
4711 | } | |
4712 | ||
4713 | int RGWRados::list_raw_prefixed_objs(const rgw_pool& pool, const string& prefix, list<string>& result) | |
4714 | { | |
4715 | bool is_truncated; | |
4716 | RGWListRawObjsCtx ctx; | |
4717 | do { | |
4718 | list<string> oids; | |
4719 | int r = list_raw_objects(pool, prefix, 1000, | |
4720 | ctx, oids, &is_truncated); | |
4721 | if (r < 0) { | |
4722 | return r; | |
4723 | } | |
4724 | list<string>::iterator iter; | |
4725 | for (iter = oids.begin(); iter != oids.end(); ++iter) { | |
4726 | string& val = *iter; | |
4727 | if (val.size() > prefix.size()) | |
4728 | result.push_back(val.substr(prefix.size())); | |
4729 | } | |
4730 | } while (is_truncated); | |
4731 | ||
4732 | return 0; | |
4733 | } | |
4734 | ||
4735 | int RGWRados::list_regions(list<string>& regions) | |
4736 | { | |
4737 | RGWZoneGroup zonegroup; | |
4738 | ||
4739 | return list_raw_prefixed_objs(zonegroup.get_pool(cct), region_info_oid_prefix, regions); | |
4740 | } | |
4741 | ||
4742 | int RGWRados::list_zonegroups(list<string>& zonegroups) | |
4743 | { | |
4744 | RGWZoneGroup zonegroup; | |
4745 | ||
4746 | return list_raw_prefixed_objs(zonegroup.get_pool(cct), zonegroup_names_oid_prefix, zonegroups); | |
4747 | } | |
4748 | ||
4749 | int RGWRados::list_zones(list<string>& zones) | |
4750 | { | |
4751 | RGWZoneParams zoneparams; | |
4752 | ||
4753 | return list_raw_prefixed_objs(zoneparams.get_pool(cct), zone_names_oid_prefix, zones); | |
4754 | } | |
4755 | ||
4756 | int RGWRados::list_realms(list<string>& realms) | |
4757 | { | |
4758 | RGWRealm realm(cct, this); | |
4759 | return list_raw_prefixed_objs(realm.get_pool(cct), realm_names_oid_prefix, realms); | |
4760 | } | |
4761 | ||
4762 | int RGWRados::list_periods(list<string>& periods) | |
4763 | { | |
4764 | RGWPeriod period; | |
4765 | list<string> raw_periods; | |
4766 | int ret = list_raw_prefixed_objs(period.get_pool(cct), period.get_info_oid_prefix(), raw_periods); | |
4767 | if (ret < 0) { | |
4768 | return ret; | |
4769 | } | |
4770 | for (const auto& oid : raw_periods) { | |
4771 | size_t pos = oid.find("."); | |
4772 | if (pos != std::string::npos) { | |
4773 | periods.push_back(oid.substr(0, pos)); | |
4774 | } else { | |
4775 | periods.push_back(oid); | |
4776 | } | |
4777 | } | |
4778 | periods.sort(); // unique() only detects duplicates if they're adjacent | |
4779 | periods.unique(); | |
4780 | return 0; | |
4781 | } | |
4782 | ||
4783 | ||
4784 | int RGWRados::list_periods(const string& current_period, list<string>& periods) | |
4785 | { | |
4786 | int ret = 0; | |
4787 | string period_id = current_period; | |
4788 | while(!period_id.empty()) { | |
4789 | RGWPeriod period(period_id); | |
4790 | ret = period.init(cct, this); | |
4791 | if (ret < 0) { | |
4792 | return ret; | |
4793 | } | |
4794 | periods.push_back(period.get_id()); | |
4795 | period_id = period.get_predecessor(); | |
4796 | } | |
4797 | ||
4798 | return ret; | |
4799 | } | |
4800 | ||
4801 | /** | |
4802 | * Open the pool used as root for this gateway | |
4803 | * Returns: 0 on success, -ERR# otherwise. | |
4804 | */ | |
4805 | int RGWRados::open_root_pool_ctx() | |
4806 | { | |
4807 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().domain_root, root_pool_ctx, true); | |
4808 | } | |
4809 | ||
4810 | int RGWRados::open_gc_pool_ctx() | |
4811 | { | |
4812 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().gc_pool, gc_pool_ctx, true); | |
4813 | } | |
4814 | ||
4815 | int RGWRados::open_lc_pool_ctx() | |
4816 | { | |
4817 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().lc_pool, lc_pool_ctx, true); | |
4818 | } | |
4819 | ||
4820 | int RGWRados::open_objexp_pool_ctx() | |
4821 | { | |
4822 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, objexp_pool_ctx, true); | |
4823 | } | |
4824 | ||
4825 | int RGWRados::open_reshard_pool_ctx() | |
4826 | { | |
4827 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().reshard_pool, reshard_pool_ctx, true); | |
4828 | } | |
4829 | ||
4830 | int RGWRados::init_watch() | |
4831 | { | |
4832 | int r = rgw_init_ioctx(&rados[0], get_zone_params().control_pool, control_pool_ctx, true); | |
4833 | if (r < 0) { | |
4834 | return r; | |
4835 | } | |
4836 | ||
4837 | num_watchers = cct->_conf->rgw_num_control_oids; | |
4838 | ||
4839 | bool compat_oid = (num_watchers == 0); | |
4840 | ||
4841 | if (num_watchers <= 0) | |
4842 | num_watchers = 1; | |
4843 | ||
4844 | notify_oids = new string[num_watchers]; | |
4845 | watchers = new RGWWatcher *[num_watchers]; | |
4846 | ||
4847 | for (int i=0; i < num_watchers; i++) { | |
4848 | string& notify_oid = notify_oids[i]; | |
4849 | notify_oid = notify_oid_prefix; | |
4850 | if (!compat_oid) { | |
4851 | char buf[16]; | |
4852 | snprintf(buf, sizeof(buf), ".%d", i); | |
4853 | notify_oid.append(buf); | |
4854 | } | |
4855 | r = control_pool_ctx.create(notify_oid, false); | |
4856 | if (r < 0 && r != -EEXIST) | |
4857 | return r; | |
4858 | ||
4859 | RGWWatcher *watcher = new RGWWatcher(this, i, notify_oid); | |
4860 | watchers[i] = watcher; | |
4861 | ||
4862 | r = watcher->register_watch(); | |
4863 | if (r < 0) | |
4864 | return r; | |
4865 | } | |
4866 | ||
4867 | watch_initialized = true; | |
4868 | ||
4869 | set_cache_enabled(true); | |
4870 | ||
4871 | return 0; | |
4872 | } | |
4873 | ||
4874 | void RGWRados::pick_control_oid(const string& key, string& notify_oid) | |
4875 | { | |
4876 | uint32_t r = ceph_str_hash_linux(key.c_str(), key.size()); | |
4877 | ||
4878 | int i = r % num_watchers; | |
4879 | char buf[16]; | |
4880 | snprintf(buf, sizeof(buf), ".%d", i); | |
4881 | ||
4882 | notify_oid = notify_oid_prefix; | |
4883 | notify_oid.append(buf); | |
4884 | } | |
4885 | ||
4886 | int RGWRados::open_pool_ctx(const rgw_pool& pool, librados::IoCtx& io_ctx) | |
4887 | { | |
4888 | constexpr bool create = true; // create the pool if it doesn't exist | |
4889 | return rgw_init_ioctx(get_rados_handle(), pool, io_ctx, create); | |
4890 | } | |
4891 | ||
4892 | void RGWRados::build_bucket_index_marker(const string& shard_id_str, const string& shard_marker, | |
4893 | string *marker) { | |
4894 | if (marker) { | |
4895 | *marker = shard_id_str; | |
4896 | marker->append(BucketIndexShardsManager::KEY_VALUE_SEPARATOR); | |
4897 | marker->append(shard_marker); | |
4898 | } | |
4899 | } | |
4900 | ||
4901 | int RGWRados::open_bucket_index_ctx(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx) | |
4902 | { | |
4903 | const rgw_pool& explicit_pool = bucket_info.bucket.explicit_placement.index_pool; | |
4904 | ||
4905 | if (!explicit_pool.empty()) { | |
4906 | return open_pool_ctx(explicit_pool, index_ctx); | |
4907 | } | |
4908 | ||
4909 | const string *rule = &bucket_info.placement_rule; | |
4910 | if (rule->empty()) { | |
4911 | rule = &zonegroup.default_placement; | |
4912 | } | |
4913 | auto iter = zone_params.placement_pools.find(*rule); | |
4914 | if (iter == zone_params.placement_pools.end()) { | |
4915 | ldout(cct, 0) << "could not find placement rule " << *rule << " within zonegroup " << dendl; | |
4916 | return -EINVAL; | |
4917 | } | |
4918 | ||
4919 | int r = open_pool_ctx(iter->second.index_pool, index_ctx); | |
4920 | if (r < 0) | |
4921 | return r; | |
4922 | ||
4923 | return 0; | |
4924 | } | |
4925 | ||
4926 | /** | |
4927 | * set up a bucket listing. | |
4928 | * handle is filled in. | |
4929 | * Returns 0 on success, -ERR# otherwise. | |
4930 | */ | |
4931 | int RGWRados::list_buckets_init(RGWAccessHandle *handle) | |
4932 | { | |
4933 | try { | |
4934 | auto iter = root_pool_ctx.nobjects_begin(); | |
4935 | librados::NObjectIterator *state = new librados::NObjectIterator(iter); | |
4936 | *handle = (RGWAccessHandle)state; | |
4937 | return 0; | |
4938 | } catch (const std::system_error& e) { | |
4939 | int r = -e.code().value(); | |
4940 | ldout(cct, 10) << "nobjects_begin threw " << e.what() | |
4941 | << ", returning " << r << dendl; | |
4942 | return r; | |
4943 | } catch (const std::exception& e) { | |
4944 | ldout(cct, 10) << "nobjects_begin threw " << e.what() | |
4945 | << ", returning -5" << dendl; | |
4946 | return -EIO; | |
4947 | } | |
4948 | } | |
4949 | ||
4950 | /** | |
4951 | * get the next bucket in the listing. | |
4952 | * obj is filled in, | |
4953 | * handle is updated. | |
4954 | * returns 0 on success, -ERR# otherwise. | |
4955 | */ | |
4956 | int RGWRados::list_buckets_next(rgw_bucket_dir_entry& obj, RGWAccessHandle *handle) | |
4957 | { | |
4958 | librados::NObjectIterator *state = (librados::NObjectIterator *)*handle; | |
4959 | ||
4960 | do { | |
4961 | if (*state == root_pool_ctx.nobjects_end()) { | |
4962 | delete state; | |
4963 | return -ENOENT; | |
4964 | } | |
4965 | ||
4966 | obj.key.name = (*state)->get_oid(); | |
4967 | if (obj.key.name[0] == '_') { | |
4968 | obj.key.name = obj.key.name.substr(1); | |
4969 | } | |
4970 | try { | |
4971 | (*state)++; | |
4972 | } catch (const std::system_error& e) { | |
4973 | int r = -e.code().value(); | |
4974 | ldout(cct, 10) << "nobjects_begin threw " << e.what() | |
4975 | << ", returning " << r << dendl; | |
4976 | return r; | |
4977 | } catch (const std::exception& e) { | |
4978 | ldout(cct, 10) << "nobjects_begin threw " << e.what() | |
4979 | << ", returning -5" << dendl; | |
4980 | return -EIO; | |
4981 | } | |
4982 | } while (obj.key.name[0] == '.'); /* skip all entries starting with '.' */ | |
4983 | ||
4984 | return 0; | |
4985 | } | |
4986 | ||
4987 | ||
4988 | /**** logs ****/ | |
4989 | ||
4990 | struct log_list_state { | |
4991 | string prefix; | |
4992 | librados::IoCtx io_ctx; | |
4993 | librados::NObjectIterator obit; | |
4994 | }; | |
4995 | ||
4996 | int RGWRados::log_list_init(const string& prefix, RGWAccessHandle *handle) | |
4997 | { | |
4998 | log_list_state *state = new log_list_state; | |
4999 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, state->io_ctx); | |
5000 | if (r < 0) { | |
5001 | delete state; | |
5002 | return r; | |
5003 | } | |
5004 | state->prefix = prefix; | |
5005 | state->obit = state->io_ctx.nobjects_begin(); | |
5006 | *handle = (RGWAccessHandle)state; | |
5007 | return 0; | |
5008 | } | |
5009 | ||
5010 | int RGWRados::log_list_next(RGWAccessHandle handle, string *name) | |
5011 | { | |
5012 | log_list_state *state = static_cast<log_list_state *>(handle); | |
5013 | while (true) { | |
5014 | if (state->obit == state->io_ctx.nobjects_end()) { | |
5015 | delete state; | |
5016 | return -ENOENT; | |
5017 | } | |
5018 | if (state->prefix.length() && | |
5019 | state->obit->get_oid().find(state->prefix) != 0) { | |
5020 | state->obit++; | |
5021 | continue; | |
5022 | } | |
5023 | *name = state->obit->get_oid(); | |
5024 | state->obit++; | |
5025 | break; | |
5026 | } | |
5027 | return 0; | |
5028 | } | |
5029 | ||
5030 | int RGWRados::log_remove(const string& name) | |
5031 | { | |
5032 | librados::IoCtx io_ctx; | |
5033 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5034 | if (r < 0) | |
5035 | return r; | |
5036 | return io_ctx.remove(name); | |
5037 | } | |
5038 | ||
5039 | struct log_show_state { | |
5040 | librados::IoCtx io_ctx; | |
5041 | bufferlist bl; | |
5042 | bufferlist::iterator p; | |
5043 | string name; | |
5044 | uint64_t pos; | |
5045 | bool eof; | |
5046 | log_show_state() : pos(0), eof(false) {} | |
5047 | }; | |
5048 | ||
5049 | int RGWRados::log_show_init(const string& name, RGWAccessHandle *handle) | |
5050 | { | |
5051 | log_show_state *state = new log_show_state; | |
5052 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, state->io_ctx); | |
5053 | if (r < 0) { | |
5054 | delete state; | |
5055 | return r; | |
5056 | } | |
5057 | state->name = name; | |
5058 | *handle = (RGWAccessHandle)state; | |
5059 | return 0; | |
5060 | } | |
5061 | ||
5062 | int RGWRados::log_show_next(RGWAccessHandle handle, rgw_log_entry *entry) | |
5063 | { | |
5064 | log_show_state *state = static_cast<log_show_state *>(handle); | |
5065 | off_t off = state->p.get_off(); | |
5066 | ||
5067 | ldout(cct, 10) << "log_show_next pos " << state->pos << " bl " << state->bl.length() | |
5068 | << " off " << off | |
5069 | << " eof " << (int)state->eof | |
5070 | << dendl; | |
5071 | // read some? | |
5072 | unsigned chunk = 1024*1024; | |
5073 | if ((state->bl.length() - off) < chunk/2 && !state->eof) { | |
5074 | bufferlist more; | |
5075 | int r = state->io_ctx.read(state->name, more, chunk, state->pos); | |
5076 | if (r < 0) | |
5077 | return r; | |
5078 | state->pos += r; | |
5079 | bufferlist old; | |
5080 | try { | |
5081 | old.substr_of(state->bl, off, state->bl.length() - off); | |
5082 | } catch (buffer::error& err) { | |
5083 | return -EINVAL; | |
5084 | } | |
5085 | state->bl.clear(); | |
5086 | state->bl.claim(old); | |
5087 | state->bl.claim_append(more); | |
5088 | state->p = state->bl.begin(); | |
5089 | if ((unsigned)r < chunk) | |
5090 | state->eof = true; | |
5091 | ldout(cct, 10) << " read " << r << dendl; | |
5092 | } | |
5093 | ||
5094 | if (state->p.end()) | |
5095 | return 0; // end of file | |
5096 | try { | |
5097 | ::decode(*entry, state->p); | |
5098 | } | |
5099 | catch (const buffer::error &e) { | |
5100 | return -EINVAL; | |
5101 | } | |
5102 | return 1; | |
5103 | } | |
5104 | ||
5105 | /** | |
5106 | * usage_log_hash: get usage log key hash, based on name and index | |
5107 | * | |
5108 | * Get the usage object name. Since a user may have more than 1 | |
5109 | * object holding that info (multiple shards), we use index to | |
5110 | * specify that shard number. Once index exceeds max shards it | |
5111 | * wraps. | |
5112 | * If name is not being set, results for all users will be returned | |
5113 | * and index will wrap only after total shards number. | |
5114 | * | |
5115 | * @param cct [in] ceph context | |
5116 | * @param name [in] user name | |
5117 | * @param hash [out] hash value | |
5118 | * @param index [in] shard index number | |
5119 | */ | |
5120 | static void usage_log_hash(CephContext *cct, const string& name, string& hash, uint32_t index) | |
5121 | { | |
5122 | uint32_t val = index; | |
5123 | ||
5124 | if (!name.empty()) { | |
5125 | int max_user_shards = cct->_conf->rgw_usage_max_user_shards; | |
5126 | val %= max_user_shards; | |
5127 | val += ceph_str_hash_linux(name.c_str(), name.size()); | |
5128 | } | |
5129 | char buf[17]; | |
5130 | int max_shards = cct->_conf->rgw_usage_max_shards; | |
5131 | snprintf(buf, sizeof(buf), RGW_USAGE_OBJ_PREFIX "%u", (unsigned)(val % max_shards)); | |
5132 | hash = buf; | |
5133 | } | |
5134 | ||
5135 | int RGWRados::log_usage(map<rgw_user_bucket, RGWUsageBatch>& usage_info) | |
5136 | { | |
5137 | uint32_t index = 0; | |
5138 | ||
5139 | map<string, rgw_usage_log_info> log_objs; | |
5140 | ||
5141 | string hash; | |
5142 | string last_user; | |
5143 | ||
5144 | /* restructure usage map, zone by object hash */ | |
5145 | map<rgw_user_bucket, RGWUsageBatch>::iterator iter; | |
5146 | for (iter = usage_info.begin(); iter != usage_info.end(); ++iter) { | |
5147 | const rgw_user_bucket& ub = iter->first; | |
5148 | RGWUsageBatch& info = iter->second; | |
5149 | ||
5150 | if (ub.user.empty()) { | |
5151 | ldout(cct, 0) << "WARNING: RGWRados::log_usage(): user name empty (bucket=" << ub.bucket << "), skipping" << dendl; | |
5152 | continue; | |
5153 | } | |
5154 | ||
5155 | if (ub.user != last_user) { | |
5156 | /* index *should* be random, but why waste extra cycles | |
5157 | in most cases max user shards is not going to exceed 1, | |
5158 | so just incrementing it */ | |
5159 | usage_log_hash(cct, ub.user, hash, index++); | |
5160 | } | |
5161 | last_user = ub.user; | |
5162 | vector<rgw_usage_log_entry>& v = log_objs[hash].entries; | |
5163 | ||
5164 | for (auto miter = info.m.begin(); miter != info.m.end(); ++miter) { | |
5165 | v.push_back(miter->second); | |
5166 | } | |
5167 | } | |
5168 | ||
5169 | map<string, rgw_usage_log_info>::iterator liter; | |
5170 | ||
5171 | for (liter = log_objs.begin(); liter != log_objs.end(); ++liter) { | |
5172 | int r = cls_obj_usage_log_add(liter->first, liter->second); | |
5173 | if (r < 0) | |
5174 | return r; | |
5175 | } | |
5176 | return 0; | |
5177 | } | |
5178 | ||
5179 | int RGWRados::read_usage(const rgw_user& user, uint64_t start_epoch, uint64_t end_epoch, uint32_t max_entries, | |
5180 | bool *is_truncated, RGWUsageIter& usage_iter, map<rgw_user_bucket, rgw_usage_log_entry>& usage) | |
5181 | { | |
5182 | uint32_t num = max_entries; | |
5183 | string hash, first_hash; | |
5184 | string user_str = user.to_str(); | |
5185 | usage_log_hash(cct, user_str, first_hash, 0); | |
5186 | ||
5187 | if (usage_iter.index) { | |
5188 | usage_log_hash(cct, user_str, hash, usage_iter.index); | |
5189 | } else { | |
5190 | hash = first_hash; | |
5191 | } | |
5192 | ||
5193 | usage.clear(); | |
5194 | ||
5195 | do { | |
5196 | map<rgw_user_bucket, rgw_usage_log_entry> ret_usage; | |
5197 | map<rgw_user_bucket, rgw_usage_log_entry>::iterator iter; | |
5198 | ||
5199 | int ret = cls_obj_usage_log_read(hash, user_str, start_epoch, end_epoch, num, | |
5200 | usage_iter.read_iter, ret_usage, is_truncated); | |
5201 | if (ret == -ENOENT) | |
5202 | goto next; | |
5203 | ||
5204 | if (ret < 0) | |
5205 | return ret; | |
5206 | ||
5207 | num -= ret_usage.size(); | |
5208 | ||
5209 | for (iter = ret_usage.begin(); iter != ret_usage.end(); ++iter) { | |
5210 | usage[iter->first].aggregate(iter->second); | |
5211 | } | |
5212 | ||
5213 | next: | |
5214 | if (!*is_truncated) { | |
5215 | usage_iter.read_iter.clear(); | |
5216 | usage_log_hash(cct, user_str, hash, ++usage_iter.index); | |
5217 | } | |
5218 | } while (num && !*is_truncated && hash != first_hash); | |
5219 | return 0; | |
5220 | } | |
5221 | ||
5222 | int RGWRados::trim_usage(rgw_user& user, uint64_t start_epoch, uint64_t end_epoch) | |
5223 | { | |
5224 | uint32_t index = 0; | |
5225 | string hash, first_hash; | |
5226 | string user_str = user.to_str(); | |
5227 | usage_log_hash(cct, user_str, first_hash, index); | |
5228 | ||
5229 | hash = first_hash; | |
5230 | do { | |
5231 | int ret = cls_obj_usage_log_trim(hash, user_str, start_epoch, end_epoch); | |
5232 | ||
5233 | if (ret < 0 && ret != -ENOENT) | |
5234 | return ret; | |
5235 | ||
5236 | usage_log_hash(cct, user_str, hash, ++index); | |
5237 | } while (hash != first_hash); | |
5238 | ||
5239 | return 0; | |
5240 | } | |
5241 | ||
5242 | int RGWRados::key_to_shard_id(const string& key, int max_shards) | |
5243 | { | |
5244 | return rgw_shard_id(key, max_shards); | |
5245 | } | |
5246 | ||
5247 | void RGWRados::shard_name(const string& prefix, unsigned max_shards, const string& key, string& name, int *shard_id) | |
5248 | { | |
5249 | uint32_t val = ceph_str_hash_linux(key.c_str(), key.size()); | |
5250 | char buf[16]; | |
5251 | if (shard_id) { | |
5252 | *shard_id = val % max_shards; | |
5253 | } | |
5254 | snprintf(buf, sizeof(buf), "%u", (unsigned)(val % max_shards)); | |
5255 | name = prefix + buf; | |
5256 | } | |
5257 | ||
5258 | void RGWRados::shard_name(const string& prefix, unsigned max_shards, const string& section, const string& key, string& name) | |
5259 | { | |
5260 | uint32_t val = ceph_str_hash_linux(key.c_str(), key.size()); | |
5261 | val ^= ceph_str_hash_linux(section.c_str(), section.size()); | |
5262 | char buf[16]; | |
5263 | snprintf(buf, sizeof(buf), "%u", (unsigned)(val % max_shards)); | |
5264 | name = prefix + buf; | |
5265 | } | |
5266 | ||
5267 | void RGWRados::shard_name(const string& prefix, unsigned shard_id, string& name) | |
5268 | { | |
5269 | char buf[16]; | |
5270 | snprintf(buf, sizeof(buf), "%u", shard_id); | |
5271 | name = prefix + buf; | |
5272 | ||
5273 | } | |
5274 | ||
5275 | void RGWRados::time_log_prepare_entry(cls_log_entry& entry, const real_time& ut, const string& section, const string& key, bufferlist& bl) | |
5276 | { | |
5277 | cls_log_add_prepare_entry(entry, utime_t(ut), section, key, bl); | |
5278 | } | |
5279 | ||
5280 | int RGWRados::time_log_add_init(librados::IoCtx& io_ctx) | |
5281 | { | |
5282 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx, true); | |
5283 | ||
5284 | } | |
5285 | ||
5286 | int RGWRados::time_log_add(const string& oid, const real_time& ut, const string& section, const string& key, bufferlist& bl) | |
5287 | { | |
5288 | librados::IoCtx io_ctx; | |
5289 | ||
5290 | int r = time_log_add_init(io_ctx); | |
5291 | if (r < 0) { | |
5292 | return r; | |
5293 | } | |
5294 | ||
5295 | ObjectWriteOperation op; | |
5296 | utime_t t(ut); | |
5297 | cls_log_add(op, t, section, key, bl); | |
5298 | ||
5299 | return io_ctx.operate(oid, &op); | |
5300 | } | |
5301 | ||
5302 | int RGWRados::time_log_add(const string& oid, list<cls_log_entry>& entries, | |
5303 | librados::AioCompletion *completion, bool monotonic_inc) | |
5304 | { | |
5305 | librados::IoCtx io_ctx; | |
5306 | ||
5307 | int r = time_log_add_init(io_ctx); | |
5308 | if (r < 0) { | |
5309 | return r; | |
5310 | } | |
5311 | ||
5312 | ObjectWriteOperation op; | |
5313 | cls_log_add(op, entries, monotonic_inc); | |
5314 | ||
5315 | if (!completion) { | |
5316 | r = io_ctx.operate(oid, &op); | |
5317 | } else { | |
5318 | r = io_ctx.aio_operate(oid, completion, &op); | |
5319 | } | |
5320 | return r; | |
5321 | } | |
5322 | ||
5323 | int RGWRados::time_log_list(const string& oid, const real_time& start_time, const real_time& end_time, | |
5324 | int max_entries, list<cls_log_entry>& entries, | |
5325 | const string& marker, | |
5326 | string *out_marker, | |
5327 | bool *truncated) | |
5328 | { | |
5329 | librados::IoCtx io_ctx; | |
5330 | ||
5331 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5332 | if (r < 0) | |
5333 | return r; | |
5334 | librados::ObjectReadOperation op; | |
5335 | ||
5336 | utime_t st(start_time); | |
5337 | utime_t et(end_time); | |
5338 | ||
5339 | cls_log_list(op, st, et, marker, max_entries, entries, | |
5340 | out_marker, truncated); | |
5341 | ||
5342 | bufferlist obl; | |
5343 | ||
5344 | int ret = io_ctx.operate(oid, &op, &obl); | |
5345 | if (ret < 0) | |
5346 | return ret; | |
5347 | ||
5348 | return 0; | |
5349 | } | |
5350 | ||
5351 | int RGWRados::time_log_info(const string& oid, cls_log_header *header) | |
5352 | { | |
5353 | librados::IoCtx io_ctx; | |
5354 | ||
5355 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5356 | if (r < 0) | |
5357 | return r; | |
5358 | librados::ObjectReadOperation op; | |
5359 | ||
5360 | cls_log_info(op, header); | |
5361 | ||
5362 | bufferlist obl; | |
5363 | ||
5364 | int ret = io_ctx.operate(oid, &op, &obl); | |
5365 | if (ret < 0) | |
5366 | return ret; | |
5367 | ||
5368 | return 0; | |
5369 | } | |
5370 | ||
5371 | int RGWRados::time_log_info_async(librados::IoCtx& io_ctx, const string& oid, cls_log_header *header, librados::AioCompletion *completion) | |
5372 | { | |
5373 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5374 | if (r < 0) | |
5375 | return r; | |
5376 | ||
5377 | librados::ObjectReadOperation op; | |
5378 | ||
5379 | cls_log_info(op, header); | |
5380 | ||
5381 | int ret = io_ctx.aio_operate(oid, completion, &op, NULL); | |
5382 | if (ret < 0) | |
5383 | return ret; | |
5384 | ||
5385 | return 0; | |
5386 | } | |
5387 | ||
5388 | int RGWRados::time_log_trim(const string& oid, const real_time& start_time, const real_time& end_time, | |
5389 | const string& from_marker, const string& to_marker, | |
5390 | librados::AioCompletion *completion) | |
5391 | { | |
5392 | librados::IoCtx io_ctx; | |
5393 | ||
5394 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5395 | if (r < 0) | |
5396 | return r; | |
5397 | ||
5398 | utime_t st(start_time); | |
5399 | utime_t et(end_time); | |
5400 | ||
5401 | ObjectWriteOperation op; | |
5402 | cls_log_trim(op, st, et, from_marker, to_marker); | |
5403 | ||
5404 | if (!completion) { | |
5405 | r = io_ctx.operate(oid, &op); | |
5406 | } else { | |
5407 | r = io_ctx.aio_operate(oid, completion, &op); | |
5408 | } | |
5409 | return r; | |
5410 | } | |
5411 | ||
5412 | string RGWRados::objexp_hint_get_shardname(int shard_num) | |
5413 | { | |
5414 | char buf[32]; | |
5415 | snprintf(buf, sizeof(buf), "%010u", (unsigned)shard_num); | |
5416 | ||
5417 | string objname("obj_delete_at_hint."); | |
5418 | return objname + buf; | |
5419 | } | |
5420 | ||
5421 | int RGWRados::objexp_key_shard(const rgw_obj_index_key& key) | |
5422 | { | |
5423 | string obj_key = key.name + key.instance; | |
5424 | int num_shards = cct->_conf->rgw_objexp_hints_num_shards; | |
5425 | return rgw_bucket_shard_index(obj_key, num_shards); | |
5426 | } | |
5427 | ||
5428 | static string objexp_hint_get_keyext(const string& tenant_name, | |
5429 | const string& bucket_name, | |
5430 | const string& bucket_id, | |
5431 | const rgw_obj_key& obj_key) | |
5432 | { | |
5433 | return tenant_name + (tenant_name.empty() ? "" : ":") + bucket_name + ":" + bucket_id + | |
5434 | ":" + obj_key.name + ":" + obj_key.instance; | |
5435 | } | |
5436 | ||
5437 | int RGWRados::objexp_hint_add(const ceph::real_time& delete_at, | |
5438 | const string& tenant_name, | |
5439 | const string& bucket_name, | |
5440 | const string& bucket_id, | |
5441 | const rgw_obj_index_key& obj_key) | |
5442 | { | |
5443 | const string keyext = objexp_hint_get_keyext(tenant_name, bucket_name, | |
5444 | bucket_id, obj_key); | |
5445 | objexp_hint_entry he = { | |
5446 | .tenant = tenant_name, | |
5447 | .bucket_name = bucket_name, | |
5448 | .bucket_id = bucket_id, | |
5449 | .obj_key = obj_key, | |
5450 | .exp_time = delete_at }; | |
5451 | bufferlist hebl; | |
5452 | ::encode(he, hebl); | |
5453 | ObjectWriteOperation op; | |
5454 | cls_timeindex_add(op, utime_t(delete_at), keyext, hebl); | |
5455 | ||
5456 | string shard_name = objexp_hint_get_shardname(objexp_key_shard(obj_key)); | |
5457 | return objexp_pool_ctx.operate(shard_name, &op); | |
5458 | } | |
5459 | ||
5460 | void RGWRados::objexp_get_shard(int shard_num, | |
5461 | string& shard) /* out */ | |
5462 | { | |
5463 | shard = objexp_hint_get_shardname(shard_num); | |
5464 | } | |
5465 | ||
5466 | int RGWRados::objexp_hint_list(const string& oid, | |
5467 | const ceph::real_time& start_time, | |
5468 | const ceph::real_time& end_time, | |
5469 | const int max_entries, | |
5470 | const string& marker, | |
5471 | list<cls_timeindex_entry>& entries, /* out */ | |
5472 | string *out_marker, /* out */ | |
5473 | bool *truncated) /* out */ | |
5474 | { | |
5475 | librados::ObjectReadOperation op; | |
5476 | cls_timeindex_list(op, utime_t(start_time), utime_t(end_time), marker, max_entries, entries, | |
5477 | out_marker, truncated); | |
5478 | ||
5479 | bufferlist obl; | |
5480 | int ret = objexp_pool_ctx.operate(oid, &op, &obl); | |
5481 | ||
5482 | if ((ret < 0 ) && (ret != -ENOENT)) { | |
5483 | return ret; | |
5484 | } | |
5485 | ||
5486 | if ((ret == -ENOENT) && truncated) { | |
5487 | *truncated = false; | |
5488 | } | |
5489 | ||
5490 | return 0; | |
5491 | } | |
5492 | ||
5493 | int RGWRados::objexp_hint_parse(cls_timeindex_entry &ti_entry, /* in */ | |
5494 | objexp_hint_entry& hint_entry) /* out */ | |
5495 | { | |
5496 | try { | |
5497 | bufferlist::iterator iter = ti_entry.value.begin(); | |
5498 | ::decode(hint_entry, iter); | |
5499 | } catch (buffer::error& err) { | |
5500 | ldout(cct, 0) << "ERROR: couldn't decode avail_pools" << dendl; | |
5501 | } | |
5502 | ||
5503 | return 0; | |
5504 | } | |
5505 | ||
5506 | int RGWRados::objexp_hint_trim(const string& oid, | |
5507 | const ceph::real_time& start_time, | |
5508 | const ceph::real_time& end_time, | |
5509 | const string& from_marker, | |
5510 | const string& to_marker) | |
5511 | { | |
5512 | int ret = cls_timeindex_trim(objexp_pool_ctx, oid, utime_t(start_time), utime_t(end_time), | |
5513 | from_marker, to_marker); | |
5514 | if ((ret < 0 ) && (ret != -ENOENT)) { | |
5515 | return ret; | |
5516 | } | |
5517 | ||
5518 | return 0; | |
5519 | } | |
5520 | ||
5521 | int RGWRados::lock_exclusive(rgw_pool& pool, const string& oid, timespan& duration, | |
5522 | string& zone_id, string& owner_id) { | |
5523 | librados::IoCtx io_ctx; | |
5524 | ||
5525 | int r = rgw_init_ioctx(get_rados_handle(), pool, io_ctx); | |
5526 | if (r < 0) { | |
5527 | return r; | |
5528 | } | |
5529 | uint64_t msec = std::chrono::duration_cast<std::chrono::milliseconds>(duration).count(); | |
5530 | utime_t ut(msec / 1000, msec % 1000); | |
5531 | ||
5532 | rados::cls::lock::Lock l(log_lock_name); | |
5533 | l.set_duration(ut); | |
5534 | l.set_cookie(owner_id); | |
5535 | l.set_tag(zone_id); | |
5536 | l.set_may_renew(true); | |
5537 | ||
5538 | return l.lock_exclusive(&io_ctx, oid); | |
5539 | } | |
5540 | ||
5541 | int RGWRados::unlock(rgw_pool& pool, const string& oid, string& zone_id, string& owner_id) { | |
5542 | librados::IoCtx io_ctx; | |
5543 | ||
5544 | int r = rgw_init_ioctx(get_rados_handle(), pool, io_ctx); | |
5545 | if (r < 0) { | |
5546 | return r; | |
5547 | } | |
5548 | ||
5549 | rados::cls::lock::Lock l(log_lock_name); | |
5550 | l.set_tag(zone_id); | |
5551 | l.set_cookie(owner_id); | |
5552 | ||
5553 | return l.unlock(&io_ctx, oid); | |
5554 | } | |
5555 | ||
5556 | int RGWRados::decode_policy(bufferlist& bl, ACLOwner *owner) | |
5557 | { | |
5558 | bufferlist::iterator i = bl.begin(); | |
5559 | RGWAccessControlPolicy policy(cct); | |
5560 | try { | |
5561 | policy.decode_owner(i); | |
5562 | } catch (buffer::error& err) { | |
5563 | ldout(cct, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl; | |
5564 | return -EIO; | |
5565 | } | |
5566 | *owner = policy.get_owner(); | |
5567 | return 0; | |
5568 | } | |
5569 | ||
5570 | int rgw_policy_from_attrset(CephContext *cct, map<string, bufferlist>& attrset, RGWAccessControlPolicy *policy) | |
5571 | { | |
5572 | map<string, bufferlist>::iterator aiter = attrset.find(RGW_ATTR_ACL); | |
5573 | if (aiter == attrset.end()) | |
5574 | return -EIO; | |
5575 | ||
5576 | bufferlist& bl = aiter->second; | |
5577 | bufferlist::iterator iter = bl.begin(); | |
5578 | try { | |
5579 | policy->decode(iter); | |
5580 | } catch (buffer::error& err) { | |
5581 | ldout(cct, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl; | |
5582 | return -EIO; | |
5583 | } | |
5584 | if (cct->_conf->subsys.should_gather(ceph_subsys_rgw, 15)) { | |
5585 | RGWAccessControlPolicy_S3 *s3policy = static_cast<RGWAccessControlPolicy_S3 *>(policy); | |
5586 | ldout(cct, 15) << __func__ << " Read AccessControlPolicy"; | |
5587 | s3policy->to_xml(*_dout); | |
5588 | *_dout << dendl; | |
5589 | } | |
5590 | return 0; | |
5591 | } | |
5592 | ||
5593 | ||
5594 | int RGWRados::Bucket::update_bucket_id(const string& new_bucket_id) | |
5595 | { | |
5596 | rgw_bucket bucket = bucket_info.bucket; | |
5597 | bucket.update_bucket_id(new_bucket_id); | |
5598 | ||
5599 | RGWObjectCtx obj_ctx(store); | |
5600 | ||
5601 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, nullptr, nullptr); | |
5602 | if (ret < 0) { | |
5603 | return ret; | |
5604 | } | |
5605 | ||
5606 | return 0; | |
5607 | } | |
5608 | ||
5609 | ||
5610 | /** | |
5611 | * Get ordered listing of the objects in a bucket. | |
5612 | * | |
5613 | * max: maximum number of results to return | |
5614 | * bucket: bucket to list contents of | |
5615 | * prefix: only return results that match this prefix | |
5616 | * delim: do not include results that match this string. | |
5617 | * Any skipped results will have the matching portion of their name | |
5618 | * inserted in common_prefixes with a "true" mark. | |
5619 | * marker: if filled in, begin the listing with this object. | |
5620 | * end_marker: if filled in, end the listing with this object. | |
5621 | * result: the objects are put in here. | |
5622 | * common_prefixes: if delim is filled in, any matching prefixes are placed here. | |
5623 | * is_truncated: if number of objects in the bucket is bigger than max, then truncated. | |
5624 | */ | |
5625 | int RGWRados::Bucket::List::list_objects_ordered(int64_t max, | |
5626 | vector<rgw_bucket_dir_entry> *result, | |
5627 | map<string, bool> *common_prefixes, | |
5628 | bool *is_truncated) | |
5629 | { | |
5630 | RGWRados *store = target->get_store(); | |
5631 | CephContext *cct = store->ctx(); | |
5632 | int shard_id = target->get_shard_id(); | |
5633 | ||
5634 | int count = 0; | |
5635 | bool truncated = true; | |
5636 | int read_ahead = std::max(cct->_conf->rgw_list_bucket_min_readahead,max); | |
5637 | ||
5638 | result->clear(); | |
5639 | ||
5640 | rgw_obj_key marker_obj(params.marker.name, params.marker.instance, params.ns); | |
5641 | rgw_obj_index_key cur_marker; | |
5642 | marker_obj.get_index_key(&cur_marker); | |
5643 | ||
5644 | rgw_obj_key end_marker_obj(params.end_marker.name, params.end_marker.instance, | |
5645 | params.ns); | |
5646 | rgw_obj_index_key cur_end_marker; | |
5647 | end_marker_obj.get_index_key(&cur_end_marker); | |
5648 | const bool cur_end_marker_valid = !params.end_marker.empty(); | |
5649 | ||
5650 | rgw_obj_key prefix_obj(params.prefix); | |
5651 | prefix_obj.ns = params.ns; | |
5652 | string cur_prefix = prefix_obj.get_index_key_name(); | |
5653 | ||
5654 | string bigger_than_delim; | |
5655 | ||
5656 | if (!params.delim.empty()) { | |
5657 | unsigned long val = decode_utf8((unsigned char *)params.delim.c_str(), | |
5658 | params.delim.size()); | |
5659 | char buf[params.delim.size() + 16]; | |
5660 | int r = encode_utf8(val + 1, (unsigned char *)buf); | |
5661 | if (r < 0) { | |
5662 | ldout(cct,0) << "ERROR: encode_utf8() failed" << dendl; | |
5663 | return -EINVAL; | |
5664 | } | |
5665 | buf[r] = '\0'; | |
5666 | ||
5667 | bigger_than_delim = buf; | |
5668 | ||
5669 | /* if marker points at a common prefix, fast forward it into its upperbound string */ | |
5670 | int delim_pos = cur_marker.name.find(params.delim, cur_prefix.size()); | |
5671 | if (delim_pos >= 0) { | |
5672 | string s = cur_marker.name.substr(0, delim_pos); | |
5673 | s.append(bigger_than_delim); | |
5674 | cur_marker = s; | |
5675 | } | |
5676 | } | |
5677 | ||
5678 | string skip_after_delim; | |
5679 | while (truncated && count <= max) { | |
5680 | if (skip_after_delim > cur_marker.name) { | |
5681 | cur_marker = skip_after_delim; | |
5682 | ldout(cct, 20) << "setting cur_marker=" << cur_marker.name << "[" << cur_marker.instance << "]" << dendl; | |
5683 | } | |
5684 | std::map<string, rgw_bucket_dir_entry> ent_map; | |
5685 | int r = store->cls_bucket_list_ordered(target->get_bucket_info(), | |
5686 | shard_id, | |
5687 | cur_marker, | |
5688 | cur_prefix, | |
5689 | read_ahead + 1 - count, | |
5690 | params.list_versions, | |
5691 | ent_map, | |
5692 | &truncated, | |
5693 | &cur_marker); | |
5694 | if (r < 0) | |
5695 | return r; | |
5696 | ||
5697 | for (auto eiter = ent_map.begin(); eiter != ent_map.end(); ++eiter) { | |
5698 | rgw_bucket_dir_entry& entry = eiter->second; | |
5699 | rgw_obj_index_key index_key = entry.key; | |
5700 | ||
5701 | rgw_obj_key obj(index_key); | |
5702 | ||
5703 | /* note that parse_raw_oid() here will not set the correct | |
5704 | * object's instance, as rgw_obj_index_key encodes that | |
5705 | * separately. We don't need to set the instance because it's | |
5706 | * not needed for the checks here and we end up using the raw | |
5707 | * entry for the return vector | |
5708 | */ | |
5709 | bool valid = rgw_obj_key::parse_raw_oid(index_key.name, &obj); | |
5710 | if (!valid) { | |
5711 | ldout(cct, 0) << "ERROR: could not parse object name: " << obj.name << dendl; | |
5712 | continue; | |
5713 | } | |
5714 | bool check_ns = (obj.ns == params.ns); | |
5715 | if (!params.list_versions && !entry.is_visible()) { | |
5716 | continue; | |
5717 | } | |
5718 | ||
5719 | if (params.enforce_ns && !check_ns) { | |
5720 | if (!params.ns.empty()) { | |
5721 | /* we've iterated past the namespace we're searching -- done now */ | |
5722 | truncated = false; | |
5723 | goto done; | |
5724 | } | |
5725 | ||
5726 | /* we're not looking at the namespace this object is in, next! */ | |
5727 | continue; | |
5728 | } | |
5729 | ||
5730 | if (cur_end_marker_valid && cur_end_marker <= index_key) { | |
5731 | truncated = false; | |
5732 | goto done; | |
5733 | } | |
5734 | ||
5735 | if (count < max) { | |
5736 | params.marker = index_key; | |
5737 | next_marker = index_key; | |
5738 | } | |
5739 | ||
5740 | if (params.filter && !params.filter->filter(obj.name, index_key.name)) | |
5741 | continue; | |
5742 | ||
5743 | if (params.prefix.size() && | |
5744 | (obj.name.compare(0, params.prefix.size(), params.prefix) != 0)) | |
5745 | continue; | |
5746 | ||
5747 | if (!params.delim.empty()) { | |
5748 | int delim_pos = obj.name.find(params.delim, params.prefix.size()); | |
5749 | ||
5750 | if (delim_pos >= 0) { | |
5751 | string prefix_key = obj.name.substr(0, delim_pos + 1); | |
5752 | ||
5753 | if (common_prefixes && | |
5754 | common_prefixes->find(prefix_key) == common_prefixes->end()) { | |
5755 | if (count >= max) { | |
5756 | truncated = true; | |
5757 | goto done; | |
5758 | } | |
5759 | next_marker = prefix_key; | |
5760 | (*common_prefixes)[prefix_key] = true; | |
5761 | ||
5762 | int marker_delim_pos = cur_marker.name.find(params.delim, cur_prefix.size()); | |
5763 | ||
5764 | skip_after_delim = cur_marker.name.substr(0, marker_delim_pos); | |
5765 | skip_after_delim.append(bigger_than_delim); | |
5766 | ||
5767 | ldout(cct, 20) << "skip_after_delim=" << skip_after_delim << dendl; | |
5768 | ||
5769 | count++; | |
5770 | } | |
5771 | ||
5772 | continue; | |
5773 | } | |
5774 | } | |
5775 | ||
5776 | if (count >= max) { | |
5777 | truncated = true; | |
5778 | goto done; | |
5779 | } | |
5780 | ||
5781 | result->emplace_back(std::move(entry)); | |
5782 | count++; | |
5783 | } | |
5784 | } | |
5785 | ||
5786 | done: | |
5787 | if (is_truncated) | |
5788 | *is_truncated = truncated; | |
5789 | ||
5790 | return 0; | |
5791 | } // list_objects_ordered | |
5792 | ||
5793 | ||
5794 | /** | |
5795 | * Get listing of the objects in a bucket and allow the results to be out | |
5796 | * of order. | |
5797 | * | |
5798 | * Even though there are key differences with the ordered counterpart, | |
5799 | * the parameters are the same to maintain some compatability. | |
5800 | * | |
5801 | * max: maximum number of results to return | |
5802 | * bucket: bucket to list contents of | |
5803 | * prefix: only return results that match this prefix | |
5804 | * delim: should not be set; if it is we should have indicated an error | |
5805 | * marker: if filled in, begin the listing with this object. | |
5806 | * end_marker: if filled in, end the listing with this object. | |
5807 | * result: the objects are put in here. | |
5808 | * common_prefixes: this is never filled with an unordered list; the param | |
5809 | * is maintained for compatibility | |
5810 | * is_truncated: if number of objects in the bucket is bigger than max, then | |
5811 | * truncated. | |
5812 | */ | |
5813 | int RGWRados::Bucket::List::list_objects_unordered(int64_t max, | |
5814 | vector<rgw_bucket_dir_entry> *result, | |
5815 | map<string, bool> *common_prefixes, | |
5816 | bool *is_truncated) | |
5817 | { | |
5818 | RGWRados *store = target->get_store(); | |
5819 | CephContext *cct = store->ctx(); | |
5820 | int shard_id = target->get_shard_id(); | |
5821 | ||
5822 | int count = 0; | |
5823 | bool truncated = true; | |
5824 | ||
5825 | // read a few extra in each call to cls_bucket_list_unordered in | |
5826 | // case some are filtered out due to namespace matching, versioning, | |
5827 | // filtering, etc. | |
5828 | const int64_t max_read_ahead = 100; | |
5829 | const uint32_t read_ahead = uint32_t(max + std::min(max, max_read_ahead)); | |
5830 | ||
5831 | result->clear(); | |
5832 | ||
5833 | rgw_obj_key marker_obj(params.marker.name, params.marker.instance, params.ns); | |
5834 | rgw_obj_index_key cur_marker; | |
5835 | marker_obj.get_index_key(&cur_marker); | |
5836 | ||
5837 | rgw_obj_key end_marker_obj(params.end_marker.name, params.end_marker.instance, | |
5838 | params.ns); | |
5839 | rgw_obj_index_key cur_end_marker; | |
5840 | end_marker_obj.get_index_key(&cur_end_marker); | |
5841 | const bool cur_end_marker_valid = !params.end_marker.empty(); | |
5842 | ||
5843 | rgw_obj_key prefix_obj(params.prefix); | |
5844 | prefix_obj.ns = params.ns; | |
5845 | string cur_prefix = prefix_obj.get_index_key_name(); | |
5846 | ||
5847 | while (truncated && count <= max) { | |
5848 | std::vector<rgw_bucket_dir_entry> ent_list; | |
5849 | int r = store->cls_bucket_list_unordered(target->get_bucket_info(), | |
5850 | shard_id, | |
5851 | cur_marker, | |
5852 | cur_prefix, | |
5853 | read_ahead, | |
5854 | params.list_versions, | |
5855 | ent_list, | |
5856 | &truncated, | |
5857 | &cur_marker); | |
5858 | if (r < 0) | |
5859 | return r; | |
5860 | ||
5861 | // NB: while regions of ent_list will be sorted, we have no | |
5862 | // guarantee that all items will be sorted since they can cross | |
5863 | // shard boundaries | |
5864 | ||
5865 | for (auto& entry : ent_list) { | |
5866 | rgw_obj_index_key index_key = entry.key; | |
5867 | rgw_obj_key obj(index_key); | |
5868 | ||
5869 | /* note that parse_raw_oid() here will not set the correct | |
5870 | * object's instance, as rgw_obj_index_key encodes that | |
5871 | * separately. We don't need to set the instance because it's | |
5872 | * not needed for the checks here and we end up using the raw | |
5873 | * entry for the return vector | |
5874 | */ | |
5875 | bool valid = rgw_obj_key::parse_raw_oid(index_key.name, &obj); | |
5876 | if (!valid) { | |
5877 | ldout(cct, 0) << "ERROR: could not parse object name: " << | |
5878 | obj.name << dendl; | |
5879 | continue; | |
5880 | } | |
5881 | ||
5882 | if (!params.list_versions && !entry.is_visible()) { | |
5883 | continue; | |
5884 | } | |
5885 | ||
5886 | if (params.enforce_ns && obj.ns != params.ns) { | |
5887 | continue; | |
5888 | } | |
5889 | ||
5890 | if (cur_end_marker_valid && cur_end_marker <= index_key) { | |
5891 | // we're not guaranteed items will come in order, so we have | |
5892 | // to loop through all | |
5893 | continue; | |
5894 | } | |
5895 | ||
5896 | if (count < max) { | |
5897 | params.marker = index_key; | |
5898 | next_marker = index_key; | |
5899 | } | |
5900 | ||
5901 | if (params.filter && !params.filter->filter(obj.name, index_key.name)) | |
5902 | continue; | |
5903 | ||
5904 | if (params.prefix.size() && | |
5905 | (0 != obj.name.compare(0, params.prefix.size(), params.prefix))) | |
5906 | continue; | |
5907 | ||
5908 | if (count >= max) { | |
5909 | truncated = true; | |
5910 | goto done; | |
5911 | } | |
5912 | ||
5913 | result->emplace_back(std::move(entry)); | |
5914 | count++; | |
5915 | } // for (auto& entry : ent_list) | |
5916 | } // while (truncated && count <= max) | |
5917 | ||
5918 | done: | |
5919 | if (is_truncated) | |
5920 | *is_truncated = truncated; | |
5921 | ||
5922 | return 0; | |
5923 | } // list_objects_unordered | |
5924 | ||
5925 | ||
5926 | /** | |
5927 | * create a rados pool, associated meta info | |
5928 | * returns 0 on success, -ERR# otherwise. | |
5929 | */ | |
5930 | int RGWRados::create_pool(const rgw_pool& pool) | |
5931 | { | |
5932 | librados::IoCtx io_ctx; | |
5933 | constexpr bool create = true; | |
5934 | return rgw_init_ioctx(get_rados_handle(), pool, io_ctx, create); | |
5935 | } | |
5936 | ||
5937 | int RGWRados::init_bucket_index(RGWBucketInfo& bucket_info, int num_shards) | |
5938 | { | |
5939 | librados::IoCtx index_ctx; | |
5940 | ||
5941 | string dir_oid = dir_oid_prefix; | |
5942 | int r = open_bucket_index_ctx(bucket_info, index_ctx); | |
5943 | if (r < 0) { | |
5944 | return r; | |
5945 | } | |
5946 | ||
5947 | dir_oid.append(bucket_info.bucket.bucket_id); | |
5948 | ||
5949 | map<int, string> bucket_objs; | |
5950 | get_bucket_index_objects(dir_oid, num_shards, bucket_objs); | |
5951 | ||
5952 | return CLSRGWIssueBucketIndexInit(index_ctx, | |
5953 | bucket_objs, | |
5954 | cct->_conf->rgw_bucket_index_max_aio)(); | |
5955 | } | |
5956 | ||
5957 | int RGWRados::clean_bucket_index(RGWBucketInfo& bucket_info, int num_shards) | |
5958 | { | |
5959 | librados::IoCtx index_ctx; | |
5960 | ||
5961 | std::string dir_oid = dir_oid_prefix; | |
5962 | int r = open_bucket_index_ctx(bucket_info, index_ctx); | |
5963 | if (r < 0) { | |
5964 | return r; | |
5965 | } | |
5966 | ||
5967 | dir_oid.append(bucket_info.bucket.bucket_id); | |
5968 | ||
5969 | std::map<int, std::string> bucket_objs; | |
5970 | get_bucket_index_objects(dir_oid, num_shards, bucket_objs); | |
5971 | ||
5972 | return CLSRGWIssueBucketIndexClean(index_ctx, | |
5973 | bucket_objs, | |
5974 | cct->_conf->rgw_bucket_index_max_aio)(); | |
5975 | } | |
5976 | ||
5977 | void RGWRados::create_bucket_id(string *bucket_id) | |
5978 | { | |
5979 | uint64_t iid = instance_id(); | |
5980 | uint64_t bid = next_bucket_id(); | |
5981 | char buf[get_zone_params().get_id().size() + 48]; | |
5982 | snprintf(buf, sizeof(buf), "%s.%llu.%llu", get_zone_params().get_id().c_str(), (long long)iid, (long long)bid); | |
5983 | *bucket_id = buf; | |
5984 | } | |
5985 | ||
5986 | int RGWRados::create_bucket(RGWUserInfo& owner, rgw_bucket& bucket, | |
5987 | const string& zonegroup_id, | |
5988 | const string& placement_rule, | |
5989 | const string& swift_ver_location, | |
5990 | const RGWQuotaInfo * pquota_info, | |
5991 | map<std::string, bufferlist>& attrs, | |
5992 | RGWBucketInfo& info, | |
5993 | obj_version *pobjv, | |
5994 | obj_version *pep_objv, | |
5995 | real_time creation_time, | |
5996 | rgw_bucket *pmaster_bucket, | |
5997 | uint32_t *pmaster_num_shards, | |
5998 | bool exclusive) | |
5999 | { | |
6000 | #define MAX_CREATE_RETRIES 20 /* need to bound retries */ | |
6001 | string selected_placement_rule_name; | |
6002 | RGWZonePlacementInfo rule_info; | |
6003 | ||
6004 | for (int i = 0; i < MAX_CREATE_RETRIES; i++) { | |
6005 | int ret = 0; | |
6006 | ret = select_bucket_placement(owner, zonegroup_id, placement_rule, | |
6007 | &selected_placement_rule_name, &rule_info); | |
6008 | if (ret < 0) | |
6009 | return ret; | |
6010 | ||
6011 | if (!pmaster_bucket) { | |
6012 | create_bucket_id(&bucket.marker); | |
6013 | bucket.bucket_id = bucket.marker; | |
6014 | } else { | |
6015 | bucket.marker = pmaster_bucket->marker; | |
6016 | bucket.bucket_id = pmaster_bucket->bucket_id; | |
6017 | } | |
6018 | ||
6019 | RGWObjVersionTracker& objv_tracker = info.objv_tracker; | |
6020 | ||
6021 | if (pobjv) { | |
6022 | objv_tracker.write_version = *pobjv; | |
6023 | } else { | |
6024 | objv_tracker.generate_new_write_ver(cct); | |
6025 | } | |
6026 | ||
6027 | info.bucket = bucket; | |
6028 | info.owner = owner.user_id; | |
6029 | info.zonegroup = zonegroup_id; | |
6030 | info.placement_rule = selected_placement_rule_name; | |
6031 | info.index_type = rule_info.index_type; | |
6032 | info.swift_ver_location = swift_ver_location; | |
6033 | info.swift_versioning = (!swift_ver_location.empty()); | |
6034 | if (pmaster_num_shards) { | |
6035 | info.num_shards = *pmaster_num_shards; | |
6036 | } else { | |
6037 | info.num_shards = bucket_index_max_shards; | |
6038 | } | |
6039 | info.bucket_index_shard_hash_type = RGWBucketInfo::MOD; | |
6040 | info.requester_pays = false; | |
6041 | if (real_clock::is_zero(creation_time)) { | |
6042 | info.creation_time = ceph::real_clock::now(); | |
6043 | } else { | |
6044 | info.creation_time = creation_time; | |
6045 | } | |
6046 | if (pquota_info) { | |
6047 | info.quota = *pquota_info; | |
6048 | } | |
6049 | ||
6050 | int r = init_bucket_index(info, info.num_shards); | |
6051 | if (r < 0) { | |
6052 | return r; | |
6053 | } | |
6054 | ||
6055 | ret = put_linked_bucket_info(info, exclusive, ceph::real_time(), pep_objv, &attrs, true); | |
6056 | if (ret == -EEXIST) { | |
6057 | librados::IoCtx index_ctx; | |
6058 | map<int, string> bucket_objs; | |
6059 | int r = open_bucket_index(info, index_ctx, bucket_objs); | |
6060 | if (r < 0) | |
6061 | return r; | |
6062 | ||
6063 | /* we need to reread the info and return it, caller will have a use for it */ | |
6064 | RGWObjVersionTracker instance_ver = info.objv_tracker; | |
6065 | info.objv_tracker.clear(); | |
6066 | RGWObjectCtx obj_ctx(this); | |
6067 | r = get_bucket_info(obj_ctx, bucket.tenant, bucket.name, info, NULL, NULL); | |
6068 | if (r < 0) { | |
6069 | if (r == -ENOENT) { | |
6070 | continue; | |
6071 | } | |
6072 | ldout(cct, 0) << "get_bucket_info returned " << r << dendl; | |
6073 | return r; | |
6074 | } | |
6075 | ||
6076 | /* only remove it if it's a different bucket instance */ | |
6077 | if (info.bucket.bucket_id != bucket.bucket_id) { | |
6078 | /* remove bucket meta instance */ | |
6079 | r = rgw_bucket_instance_remove_entry(this, | |
6080 | bucket.get_key(), | |
6081 | &instance_ver); | |
6082 | if (r < 0) | |
6083 | return r; | |
6084 | ||
6085 | /* remove bucket index objects asynchronously by best effort */ | |
6086 | (void) CLSRGWIssueBucketIndexClean(index_ctx, | |
6087 | bucket_objs, | |
6088 | cct->_conf->rgw_bucket_index_max_aio)(); | |
6089 | } | |
6090 | /* ret == -ENOENT here */ | |
6091 | } | |
6092 | return ret; | |
6093 | } | |
6094 | ||
6095 | /* this is highly unlikely */ | |
6096 | ldout(cct, 0) << "ERROR: could not create bucket, continuously raced with bucket creation and removal" << dendl; | |
6097 | return -ENOENT; | |
6098 | } | |
6099 | ||
6100 | int RGWRados::select_new_bucket_location(RGWUserInfo& user_info, const string& zonegroup_id, const string& request_rule, | |
6101 | string *pselected_rule_name, RGWZonePlacementInfo *rule_info) | |
6102 | ||
6103 | { | |
6104 | /* first check that zonegroup exists within current period. */ | |
6105 | RGWZoneGroup zonegroup; | |
6106 | int ret = get_zonegroup(zonegroup_id, zonegroup); | |
6107 | if (ret < 0) { | |
6108 | ldout(cct, 0) << "could not find zonegroup " << zonegroup_id << " in current period" << dendl; | |
6109 | return ret; | |
6110 | } | |
6111 | ||
6112 | /* find placement rule. Hierarchy: request rule > user default rule > zonegroup default rule */ | |
6113 | std::map<std::string, RGWZoneGroupPlacementTarget>::const_iterator titer; | |
6114 | ||
6115 | if (!request_rule.empty()) { | |
6116 | titer = zonegroup.placement_targets.find(request_rule); | |
6117 | if (titer == zonegroup.placement_targets.end()) { | |
6118 | ldout(cct, 0) << "could not find requested placement id " << request_rule | |
6119 | << " within zonegroup " << dendl; | |
6120 | return -ERR_INVALID_LOCATION_CONSTRAINT; | |
6121 | } | |
6122 | } else if (!user_info.default_placement.empty()) { | |
6123 | titer = zonegroup.placement_targets.find(user_info.default_placement); | |
6124 | if (titer == zonegroup.placement_targets.end()) { | |
6125 | ldout(cct, 0) << "could not find user default placement id " << user_info.default_placement | |
6126 | << " within zonegroup " << dendl; | |
6127 | return -ERR_INVALID_LOCATION_CONSTRAINT; | |
6128 | } | |
6129 | } else { | |
6130 | if (zonegroup.default_placement.empty()) { // zonegroup default rule as fallback, it should not be empty. | |
6131 | ldout(cct, 0) << "misconfiguration, zonegroup default placement id should not be empty." << dendl; | |
6132 | return -ERR_ZONEGROUP_DEFAULT_PLACEMENT_MISCONFIGURATION; | |
6133 | } else { | |
6134 | titer = zonegroup.placement_targets.find(zonegroup.default_placement); | |
6135 | if (titer == zonegroup.placement_targets.end()) { | |
6136 | ldout(cct, 0) << "could not find zonegroup default placement id " << zonegroup.default_placement | |
6137 | << " within zonegroup " << dendl; | |
6138 | return -ERR_INVALID_LOCATION_CONSTRAINT; | |
6139 | } | |
6140 | } | |
6141 | } | |
6142 | ||
6143 | /* now check tag for the rule, whether user is permitted to use rule */ | |
6144 | const auto& target_rule = titer->second; | |
6145 | if (!target_rule.user_permitted(user_info.placement_tags)) { | |
6146 | ldout(cct, 0) << "user not permitted to use placement rule " << titer->first << dendl; | |
6147 | return -EPERM; | |
6148 | } | |
6149 | ||
6150 | if (pselected_rule_name) | |
6151 | *pselected_rule_name = titer->first; | |
6152 | ||
6153 | return select_bucket_location_by_rule(titer->first, rule_info); | |
6154 | } | |
6155 | ||
6156 | int RGWRados::select_bucket_location_by_rule(const string& location_rule, RGWZonePlacementInfo *rule_info) | |
6157 | { | |
6158 | if (location_rule.empty()) { | |
6159 | /* we can only reach here if we're trying to set a bucket location from a bucket | |
6160 | * created on a different zone, using a legacy / default pool configuration | |
6161 | */ | |
6162 | return select_legacy_bucket_placement(rule_info); | |
6163 | } | |
6164 | ||
6165 | /* | |
6166 | * make sure that zone has this rule configured. We're | |
6167 | * checking it for the local zone, because that's where this bucket object is going to | |
6168 | * reside. | |
6169 | */ | |
6170 | map<string, RGWZonePlacementInfo>::iterator piter = get_zone_params().placement_pools.find(location_rule); | |
6171 | if (piter == get_zone_params().placement_pools.end()) { | |
6172 | /* couldn't find, means we cannot really place data for this bucket in this zone */ | |
6173 | if (get_zonegroup().equals(zonegroup.get_id())) { | |
6174 | /* that's a configuration error, zone should have that rule, as we're within the requested | |
6175 | * zonegroup */ | |
6176 | return -EINVAL; | |
6177 | } else { | |
6178 | /* oh, well, data is not going to be placed here, bucket object is just a placeholder */ | |
6179 | return 0; | |
6180 | } | |
6181 | } | |
6182 | ||
6183 | RGWZonePlacementInfo& placement_info = piter->second; | |
6184 | ||
6185 | if (rule_info) { | |
6186 | *rule_info = placement_info; | |
6187 | } | |
6188 | ||
6189 | return 0; | |
6190 | } | |
6191 | ||
6192 | int RGWRados::select_bucket_placement(RGWUserInfo& user_info, const string& zonegroup_id, const string& placement_rule, | |
6193 | string *pselected_rule_name, RGWZonePlacementInfo *rule_info) | |
6194 | { | |
6195 | if (!get_zone_params().placement_pools.empty()) { | |
6196 | return select_new_bucket_location(user_info, zonegroup_id, placement_rule, | |
6197 | pselected_rule_name, rule_info); | |
6198 | } | |
6199 | ||
6200 | if (pselected_rule_name) { | |
6201 | pselected_rule_name->clear(); | |
6202 | } | |
6203 | ||
6204 | return select_legacy_bucket_placement(rule_info); | |
6205 | } | |
6206 | ||
6207 | int RGWRados::select_legacy_bucket_placement(RGWZonePlacementInfo *rule_info) | |
6208 | { | |
6209 | bufferlist map_bl; | |
6210 | map<string, bufferlist> m; | |
6211 | string pool_name; | |
6212 | bool write_map = false; | |
6213 | ||
6214 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6215 | ||
6216 | RGWObjectCtx obj_ctx(this); | |
6217 | int ret = rgw_get_system_obj(this, obj_ctx, get_zone_params().domain_root, avail_pools, map_bl, NULL, NULL); | |
6218 | if (ret < 0) { | |
6219 | goto read_omap; | |
6220 | } | |
6221 | ||
6222 | try { | |
6223 | bufferlist::iterator iter = map_bl.begin(); | |
6224 | ::decode(m, iter); | |
6225 | } catch (buffer::error& err) { | |
6226 | ldout(cct, 0) << "ERROR: couldn't decode avail_pools" << dendl; | |
6227 | } | |
6228 | ||
6229 | read_omap: | |
6230 | if (m.empty()) { | |
6231 | bufferlist header; | |
6232 | ret = omap_get_all(obj, header, m); | |
6233 | ||
6234 | write_map = true; | |
6235 | } | |
6236 | ||
6237 | if (ret < 0 || m.empty()) { | |
6238 | vector<rgw_pool> pools; | |
6239 | string s = string("default.") + default_storage_pool_suffix; | |
6240 | pools.push_back(rgw_pool(s)); | |
6241 | vector<int> retcodes; | |
6242 | bufferlist bl; | |
6243 | ret = create_pools(pools, retcodes); | |
6244 | if (ret < 0) | |
6245 | return ret; | |
6246 | ret = omap_set(obj, s, bl); | |
6247 | if (ret < 0) | |
6248 | return ret; | |
6249 | m[s] = bl; | |
6250 | } | |
6251 | ||
6252 | if (write_map) { | |
6253 | bufferlist new_bl; | |
6254 | ::encode(m, new_bl); | |
6255 | ret = put_system_obj_data(NULL, obj, new_bl, -1, false); | |
6256 | if (ret < 0) { | |
6257 | ldout(cct, 0) << "WARNING: could not save avail pools map info ret=" << ret << dendl; | |
6258 | } | |
6259 | } | |
6260 | ||
6261 | map<string, bufferlist>::iterator miter; | |
6262 | if (m.size() > 1) { | |
6263 | vector<string> v; | |
6264 | for (miter = m.begin(); miter != m.end(); ++miter) { | |
6265 | v.push_back(miter->first); | |
6266 | } | |
6267 | ||
6268 | uint32_t r; | |
6269 | ret = get_random_bytes((char *)&r, sizeof(r)); | |
6270 | if (ret < 0) | |
6271 | return ret; | |
6272 | ||
6273 | int i = r % v.size(); | |
6274 | pool_name = v[i]; | |
6275 | } else { | |
6276 | miter = m.begin(); | |
6277 | pool_name = miter->first; | |
6278 | } | |
6279 | ||
6280 | rule_info->data_pool = pool_name; | |
6281 | rule_info->data_extra_pool = pool_name; | |
6282 | rule_info->index_pool = pool_name; | |
6283 | rule_info->index_type = RGWBIType_Normal; | |
6284 | ||
6285 | return 0; | |
6286 | } | |
6287 | ||
6288 | bool RGWRados::get_obj_data_pool(const string& placement_rule, const rgw_obj& obj, rgw_pool *pool) | |
6289 | { | |
6290 | return rgw_get_obj_data_pool(zonegroup, zone_params, placement_rule, obj, pool); | |
6291 | } | |
6292 | ||
6293 | bool RGWRados::obj_to_raw(const string& placement_rule, const rgw_obj& obj, rgw_raw_obj *raw_obj) | |
6294 | { | |
6295 | get_obj_bucket_and_oid_loc(obj, raw_obj->oid, raw_obj->loc); | |
6296 | ||
6297 | return get_obj_data_pool(placement_rule, obj, &raw_obj->pool); | |
6298 | } | |
6299 | ||
6300 | int RGWRados::update_placement_map() | |
6301 | { | |
6302 | bufferlist header; | |
6303 | map<string, bufferlist> m; | |
6304 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6305 | int ret = omap_get_all(obj, header, m); | |
6306 | if (ret < 0) | |
6307 | return ret; | |
6308 | ||
6309 | bufferlist new_bl; | |
6310 | ::encode(m, new_bl); | |
6311 | ret = put_system_obj_data(NULL, obj, new_bl, -1, false); | |
6312 | if (ret < 0) { | |
6313 | ldout(cct, 0) << "WARNING: could not save avail pools map info ret=" << ret << dendl; | |
6314 | } | |
6315 | ||
6316 | return ret; | |
6317 | } | |
6318 | ||
6319 | int RGWRados::add_bucket_placement(const rgw_pool& new_pool) | |
6320 | { | |
6321 | librados::Rados *rad = get_rados_handle(); | |
6322 | int ret = rad->pool_lookup(new_pool.name.c_str()); | |
6323 | if (ret < 0) // DNE, or something | |
6324 | return ret; | |
6325 | ||
6326 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6327 | bufferlist empty_bl; | |
6328 | ret = omap_set(obj, new_pool.to_str(), empty_bl); | |
6329 | ||
6330 | // don't care about return value | |
6331 | update_placement_map(); | |
6332 | ||
6333 | return ret; | |
6334 | } | |
6335 | ||
6336 | int RGWRados::remove_bucket_placement(const rgw_pool& old_pool) | |
6337 | { | |
6338 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6339 | int ret = omap_del(obj, old_pool.to_str()); | |
6340 | ||
6341 | // don't care about return value | |
6342 | update_placement_map(); | |
6343 | ||
6344 | return ret; | |
6345 | } | |
6346 | ||
6347 | int RGWRados::list_placement_set(set<rgw_pool>& names) | |
6348 | { | |
6349 | bufferlist header; | |
6350 | map<string, bufferlist> m; | |
6351 | ||
6352 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6353 | int ret = omap_get_all(obj, header, m); | |
6354 | if (ret < 0) | |
6355 | return ret; | |
6356 | ||
6357 | names.clear(); | |
6358 | map<string, bufferlist>::iterator miter; | |
6359 | for (miter = m.begin(); miter != m.end(); ++miter) { | |
6360 | names.insert(rgw_pool(miter->first)); | |
6361 | } | |
6362 | ||
6363 | return names.size(); | |
6364 | } | |
6365 | ||
6366 | int RGWRados::create_pools(vector<rgw_pool>& pools, vector<int>& retcodes) | |
6367 | { | |
6368 | vector<librados::PoolAsyncCompletion *> completions; | |
6369 | vector<int> rets; | |
6370 | ||
6371 | librados::Rados *rad = get_rados_handle(); | |
6372 | for (auto iter = pools.begin(); iter != pools.end(); ++iter) { | |
6373 | librados::PoolAsyncCompletion *c = librados::Rados::pool_async_create_completion(); | |
6374 | completions.push_back(c); | |
6375 | rgw_pool& pool = *iter; | |
6376 | int ret = rad->pool_create_async(pool.name.c_str(), c); | |
6377 | rets.push_back(ret); | |
6378 | } | |
6379 | ||
6380 | vector<int>::iterator riter; | |
6381 | vector<librados::PoolAsyncCompletion *>::iterator citer; | |
6382 | ||
6383 | bool error = false; | |
6384 | assert(rets.size() == completions.size()); | |
6385 | for (riter = rets.begin(), citer = completions.begin(); riter != rets.end(); ++riter, ++citer) { | |
6386 | int r = *riter; | |
6387 | PoolAsyncCompletion *c = *citer; | |
6388 | if (r == 0) { | |
6389 | c->wait(); | |
6390 | r = c->get_return_value(); | |
6391 | if (r < 0) { | |
6392 | ldout(cct, 0) << "WARNING: async pool_create returned " << r << dendl; | |
6393 | error = true; | |
6394 | } | |
6395 | } | |
6396 | c->release(); | |
6397 | retcodes.push_back(r); | |
6398 | } | |
6399 | if (error) { | |
6400 | return 0; | |
6401 | } | |
6402 | ||
6403 | std::vector<librados::IoCtx> io_ctxs; | |
6404 | retcodes.clear(); | |
6405 | for (auto pool : pools) { | |
6406 | io_ctxs.emplace_back(); | |
6407 | int ret = rad->ioctx_create(pool.name.c_str(), io_ctxs.back()); | |
6408 | if (ret < 0) { | |
6409 | ldout(cct, 0) << "WARNING: ioctx_create returned " << ret << dendl; | |
6410 | error = true; | |
6411 | } | |
6412 | retcodes.push_back(ret); | |
6413 | } | |
6414 | if (error) { | |
6415 | return 0; | |
6416 | } | |
6417 | ||
6418 | completions.clear(); | |
6419 | for (auto &io_ctx : io_ctxs) { | |
6420 | librados::PoolAsyncCompletion *c = | |
6421 | librados::Rados::pool_async_create_completion(); | |
6422 | completions.push_back(c); | |
6423 | int ret = io_ctx.application_enable_async(pg_pool_t::APPLICATION_NAME_RGW, | |
6424 | false, c); | |
6425 | assert(ret == 0); | |
6426 | } | |
6427 | ||
6428 | retcodes.clear(); | |
6429 | for (auto c : completions) { | |
6430 | c->wait(); | |
6431 | int ret = c->get_return_value(); | |
6432 | if (ret == -EOPNOTSUPP) { | |
6433 | ret = 0; | |
6434 | } else if (ret < 0) { | |
6435 | ldout(cct, 0) << "WARNING: async application_enable returned " << ret | |
6436 | << dendl; | |
6437 | error = true; | |
6438 | } | |
6439 | c->release(); | |
6440 | retcodes.push_back(ret); | |
6441 | } | |
6442 | return 0; | |
6443 | } | |
6444 | ||
6445 | int RGWRados::get_obj_head_ioctx(const RGWBucketInfo& bucket_info, const rgw_obj& obj, librados::IoCtx *ioctx) | |
6446 | { | |
6447 | string oid, key; | |
6448 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
6449 | ||
6450 | rgw_pool pool; | |
6451 | if (!get_obj_data_pool(bucket_info.placement_rule, obj, &pool)) { | |
6452 | ldout(cct, 0) << "ERROR: cannot get data pool for obj=" << obj << ", probably misconfiguration" << dendl; | |
6453 | return -EIO; | |
6454 | } | |
6455 | ||
6456 | int r = open_pool_ctx(pool, *ioctx); | |
6457 | if (r < 0) { | |
6458 | return r; | |
6459 | } | |
6460 | ||
6461 | ioctx->locator_set_key(key); | |
6462 | ||
6463 | return 0; | |
6464 | } | |
6465 | ||
6466 | int RGWRados::get_obj_head_ref(const RGWBucketInfo& bucket_info, const rgw_obj& obj, rgw_rados_ref *ref) | |
6467 | { | |
6468 | get_obj_bucket_and_oid_loc(obj, ref->oid, ref->key); | |
6469 | ||
6470 | rgw_pool pool; | |
6471 | if (!get_obj_data_pool(bucket_info.placement_rule, obj, &pool)) { | |
6472 | ldout(cct, 0) << "ERROR: cannot get data pool for obj=" << obj << ", probably misconfiguration" << dendl; | |
6473 | return -EIO; | |
6474 | } | |
6475 | ||
6476 | int r = open_pool_ctx(pool, ref->ioctx); | |
6477 | if (r < 0) { | |
6478 | return r; | |
6479 | } | |
6480 | ||
6481 | ref->ioctx.locator_set_key(ref->key); | |
6482 | ||
6483 | return 0; | |
6484 | } | |
6485 | ||
6486 | int RGWRados::get_raw_obj_ref(const rgw_raw_obj& obj, rgw_rados_ref *ref) | |
6487 | { | |
6488 | ref->oid = obj.oid; | |
6489 | ref->key = obj.loc; | |
6490 | ||
6491 | int r; | |
6492 | ||
6493 | if (ref->oid.empty()) { | |
6494 | ref->oid = obj.pool.to_str(); | |
6495 | ref->pool = get_zone_params().domain_root; | |
6496 | } else { | |
6497 | ref->pool = obj.pool; | |
6498 | } | |
6499 | r = open_pool_ctx(ref->pool, ref->ioctx); | |
6500 | if (r < 0) | |
6501 | return r; | |
6502 | ||
6503 | ref->ioctx.locator_set_key(ref->key); | |
6504 | ||
6505 | return 0; | |
6506 | } | |
6507 | ||
6508 | int RGWRados::get_system_obj_ref(const rgw_raw_obj& obj, rgw_rados_ref *ref) | |
6509 | { | |
6510 | return get_raw_obj_ref(obj, ref); | |
6511 | } | |
6512 | ||
6513 | /* | |
6514 | * fixes an issue where head objects were supposed to have a locator created, but ended | |
6515 | * up without one | |
6516 | */ | |
6517 | int RGWRados::fix_head_obj_locator(const RGWBucketInfo& bucket_info, bool copy_obj, bool remove_bad, rgw_obj_key& key) | |
6518 | { | |
6519 | const rgw_bucket& bucket = bucket_info.bucket; | |
6520 | string oid; | |
6521 | string locator; | |
6522 | ||
6523 | rgw_obj obj(bucket, key); | |
6524 | ||
6525 | get_obj_bucket_and_oid_loc(obj, oid, locator); | |
6526 | ||
6527 | if (locator.empty()) { | |
6528 | ldout(cct, 20) << "object does not have a locator, nothing to fix" << dendl; | |
6529 | return 0; | |
6530 | } | |
6531 | ||
6532 | librados::IoCtx ioctx; | |
6533 | ||
6534 | int ret = get_obj_head_ioctx(bucket_info, obj, &ioctx); | |
6535 | if (ret < 0) { | |
6536 | cerr << "ERROR: get_obj_head_ioctx() returned ret=" << ret << std::endl; | |
6537 | return ret; | |
6538 | } | |
6539 | ioctx.locator_set_key(string()); /* override locator for this object, use empty locator */ | |
6540 | ||
6541 | uint64_t size; | |
6542 | bufferlist data; | |
6543 | ||
6544 | struct timespec mtime_ts; | |
6545 | map<string, bufferlist> attrs; | |
6546 | librados::ObjectReadOperation op; | |
6547 | op.getxattrs(&attrs, NULL); | |
6548 | op.stat2(&size, &mtime_ts, NULL); | |
6549 | #define HEAD_SIZE 512 * 1024 | |
6550 | op.read(0, HEAD_SIZE, &data, NULL); | |
6551 | ||
6552 | ret = ioctx.operate(oid, &op, NULL); | |
6553 | if (ret < 0) { | |
6554 | lderr(cct) << "ERROR: ioctx.operate(oid=" << oid << ") returned ret=" << ret << dendl; | |
6555 | return ret; | |
6556 | } | |
6557 | ||
6558 | if (size > HEAD_SIZE) { | |
6559 | lderr(cct) << "ERROR: returned object size (" << size << ") > HEAD_SIZE (" << HEAD_SIZE << ")" << dendl; | |
6560 | return -EIO; | |
6561 | } | |
6562 | ||
6563 | if (size != data.length()) { | |
6564 | lderr(cct) << "ERROR: returned object size (" << size << ") != data.length() (" << data.length() << ")" << dendl; | |
6565 | return -EIO; | |
6566 | } | |
6567 | ||
6568 | if (copy_obj) { | |
6569 | librados::ObjectWriteOperation wop; | |
6570 | ||
6571 | wop.mtime2(&mtime_ts); | |
6572 | ||
6573 | map<string, bufferlist>::iterator iter; | |
6574 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
6575 | wop.setxattr(iter->first.c_str(), iter->second); | |
6576 | } | |
6577 | ||
6578 | wop.write(0, data); | |
6579 | ||
6580 | ioctx.locator_set_key(locator); | |
6581 | ioctx.operate(oid, &wop); | |
6582 | } | |
6583 | ||
6584 | if (remove_bad) { | |
6585 | ioctx.locator_set_key(string()); | |
6586 | ||
6587 | ret = ioctx.remove(oid); | |
6588 | if (ret < 0) { | |
6589 | lderr(cct) << "ERROR: failed to remove original bad object" << dendl; | |
6590 | return ret; | |
6591 | } | |
6592 | } | |
6593 | ||
6594 | return 0; | |
6595 | } | |
6596 | ||
6597 | int RGWRados::move_rados_obj(librados::IoCtx& src_ioctx, | |
6598 | const string& src_oid, const string& src_locator, | |
6599 | librados::IoCtx& dst_ioctx, | |
6600 | const string& dst_oid, const string& dst_locator) | |
6601 | { | |
6602 | ||
6603 | #define COPY_BUF_SIZE (4 * 1024 * 1024) | |
6604 | bool done = false; | |
6605 | uint64_t chunk_size = COPY_BUF_SIZE; | |
6606 | uint64_t ofs = 0; | |
6607 | int ret = 0; | |
6608 | real_time mtime; | |
6609 | struct timespec mtime_ts; | |
6610 | uint64_t size; | |
6611 | ||
6612 | if (src_oid == dst_oid && src_locator == dst_locator) { | |
6613 | return 0; | |
6614 | } | |
6615 | ||
6616 | src_ioctx.locator_set_key(src_locator); | |
6617 | dst_ioctx.locator_set_key(dst_locator); | |
6618 | ||
6619 | do { | |
6620 | bufferlist data; | |
6621 | ObjectReadOperation rop; | |
6622 | ObjectWriteOperation wop; | |
6623 | ||
6624 | if (ofs == 0) { | |
6625 | rop.stat2(&size, &mtime_ts, NULL); | |
6626 | mtime = real_clock::from_timespec(mtime_ts); | |
6627 | } | |
6628 | rop.read(ofs, chunk_size, &data, NULL); | |
6629 | ret = src_ioctx.operate(src_oid, &rop, NULL); | |
6630 | if (ret < 0) { | |
6631 | goto done_err; | |
6632 | } | |
6633 | ||
6634 | if (data.length() == 0) { | |
6635 | break; | |
6636 | } | |
6637 | ||
6638 | if (ofs == 0) { | |
6639 | wop.create(true); /* make it exclusive */ | |
6640 | wop.mtime2(&mtime_ts); | |
6641 | mtime = real_clock::from_timespec(mtime_ts); | |
6642 | } | |
6643 | wop.write(ofs, data); | |
6644 | ret = dst_ioctx.operate(dst_oid, &wop); | |
6645 | ofs += data.length(); | |
6646 | done = data.length() != chunk_size; | |
6647 | } while (!done); | |
6648 | ||
6649 | if (ofs != size) { | |
6650 | lderr(cct) << "ERROR: " << __func__ << ": copying " << src_oid << " -> " << dst_oid | |
6651 | << ": expected " << size << " bytes to copy, ended up with " << ofs << dendl; | |
6652 | ret = -EIO; | |
6653 | goto done_err; | |
6654 | } | |
6655 | ||
6656 | src_ioctx.remove(src_oid); | |
6657 | ||
6658 | return 0; | |
6659 | ||
6660 | done_err: | |
6661 | lderr(cct) << "ERROR: failed to copy " << src_oid << " -> " << dst_oid << dendl; | |
6662 | return ret; | |
6663 | } | |
6664 | ||
6665 | /* | |
6666 | * fixes an issue where head objects were supposed to have a locator created, but ended | |
6667 | * up without one | |
6668 | */ | |
6669 | int RGWRados::fix_tail_obj_locator(const RGWBucketInfo& bucket_info, rgw_obj_key& key, bool fix, bool *need_fix) | |
6670 | { | |
6671 | const rgw_bucket& bucket = bucket_info.bucket; | |
6672 | rgw_obj obj(bucket, key); | |
6673 | ||
6674 | if (need_fix) { | |
6675 | *need_fix = false; | |
6676 | } | |
6677 | ||
6678 | rgw_rados_ref ref; | |
6679 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
6680 | if (r < 0) { | |
6681 | return r; | |
6682 | } | |
6683 | ||
6684 | RGWObjState *astate = NULL; | |
6685 | RGWObjectCtx rctx(this); | |
6686 | r = get_obj_state(&rctx, bucket_info, obj, &astate, false); | |
6687 | if (r < 0) | |
6688 | return r; | |
6689 | ||
6690 | if (astate->has_manifest) { | |
6691 | RGWObjManifest::obj_iterator miter; | |
6692 | RGWObjManifest& manifest = astate->manifest; | |
6693 | for (miter = manifest.obj_begin(); miter != manifest.obj_end(); ++miter) { | |
6694 | rgw_raw_obj raw_loc = miter.get_location().get_raw_obj(this); | |
6695 | rgw_obj loc; | |
6696 | string oid; | |
6697 | string locator; | |
6698 | ||
6699 | rgw_raw_obj_to_obj(manifest.get_tail_placement().bucket, raw_loc, &loc); | |
6700 | ||
6701 | if (loc.key.ns.empty()) { | |
6702 | /* continue, we're only interested in tail objects */ | |
6703 | continue; | |
6704 | } | |
6705 | ||
6706 | get_obj_bucket_and_oid_loc(loc, oid, locator); | |
6707 | ref.ioctx.locator_set_key(locator); | |
6708 | ||
6709 | ldout(cct, 20) << __func__ << ": key=" << key << " oid=" << oid << " locator=" << locator << dendl; | |
6710 | ||
6711 | r = ref.ioctx.stat(oid, NULL, NULL); | |
6712 | if (r != -ENOENT) { | |
6713 | continue; | |
6714 | } | |
6715 | ||
6716 | string bad_loc; | |
6717 | prepend_bucket_marker(bucket, loc.key.name, bad_loc); | |
6718 | ||
6719 | /* create a new ioctx with the bad locator */ | |
6720 | librados::IoCtx src_ioctx; | |
6721 | src_ioctx.dup(ref.ioctx); | |
6722 | src_ioctx.locator_set_key(bad_loc); | |
6723 | ||
6724 | r = src_ioctx.stat(oid, NULL, NULL); | |
6725 | if (r != 0) { | |
6726 | /* cannot find a broken part */ | |
6727 | continue; | |
6728 | } | |
6729 | ldout(cct, 20) << __func__ << ": found bad object part: " << loc << dendl; | |
6730 | if (need_fix) { | |
6731 | *need_fix = true; | |
6732 | } | |
6733 | if (fix) { | |
6734 | r = move_rados_obj(src_ioctx, oid, bad_loc, ref.ioctx, oid, locator); | |
6735 | if (r < 0) { | |
6736 | lderr(cct) << "ERROR: copy_rados_obj() on oid=" << oid << " returned r=" << r << dendl; | |
6737 | } | |
6738 | } | |
6739 | } | |
6740 | } | |
6741 | ||
6742 | return 0; | |
6743 | } | |
6744 | ||
6745 | int RGWRados::BucketShard::init(const rgw_bucket& _bucket, | |
6746 | const rgw_obj& obj, | |
6747 | RGWBucketInfo* bucket_info_out) | |
6748 | { | |
6749 | bucket = _bucket; | |
6750 | ||
6751 | RGWObjectCtx obj_ctx(store); | |
6752 | ||
6753 | RGWBucketInfo bucket_info; | |
6754 | RGWBucketInfo* bucket_info_p = | |
6755 | bucket_info_out ? bucket_info_out : &bucket_info; | |
6756 | ||
6757 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, *bucket_info_p, NULL, NULL); | |
6758 | if (ret < 0) { | |
6759 | return ret; | |
6760 | } | |
6761 | ||
6762 | ret = store->open_bucket_index_shard(*bucket_info_p, index_ctx, obj.get_hash_object(), &bucket_obj, &shard_id); | |
6763 | if (ret < 0) { | |
6764 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
6765 | return ret; | |
6766 | } | |
6767 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
6768 | ||
6769 | return 0; | |
6770 | } | |
6771 | ||
6772 | int RGWRados::BucketShard::init(const rgw_bucket& _bucket, | |
6773 | int sid, | |
6774 | RGWBucketInfo* bucket_info_out) | |
6775 | { | |
6776 | bucket = _bucket; | |
6777 | shard_id = sid; | |
6778 | ||
6779 | RGWObjectCtx obj_ctx(store); | |
6780 | ||
6781 | RGWBucketInfo bucket_info; | |
6782 | RGWBucketInfo* bucket_info_p = | |
6783 | bucket_info_out ? bucket_info_out : &bucket_info; | |
6784 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, *bucket_info_p, NULL, NULL); | |
6785 | if (ret < 0) { | |
6786 | return ret; | |
6787 | } | |
6788 | ||
6789 | ret = store->open_bucket_index_shard(*bucket_info_p, index_ctx, shard_id, &bucket_obj); | |
6790 | if (ret < 0) { | |
6791 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
6792 | return ret; | |
6793 | } | |
6794 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
6795 | ||
6796 | return 0; | |
6797 | } | |
6798 | ||
6799 | int RGWRados::BucketShard::init(const RGWBucketInfo& bucket_info, | |
6800 | const rgw_obj& obj) | |
6801 | { | |
6802 | bucket = bucket_info.bucket; | |
6803 | ||
6804 | int ret = store->open_bucket_index_shard(bucket_info, index_ctx, | |
6805 | obj.get_hash_object(), &bucket_obj, | |
6806 | &shard_id); | |
6807 | if (ret < 0) { | |
6808 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
6809 | return ret; | |
6810 | } | |
6811 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
6812 | ||
6813 | return 0; | |
6814 | } | |
6815 | ||
6816 | int RGWRados::BucketShard::init(const RGWBucketInfo& bucket_info, int sid) | |
6817 | { | |
6818 | bucket = bucket_info.bucket; | |
6819 | shard_id = sid; | |
6820 | ||
6821 | int ret = store->open_bucket_index_shard(bucket_info, index_ctx, shard_id, &bucket_obj); | |
6822 | if (ret < 0) { | |
6823 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
6824 | return ret; | |
6825 | } | |
6826 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
6827 | ||
6828 | return 0; | |
6829 | } | |
6830 | ||
6831 | ||
6832 | /* Execute @handler on last item in bucket listing for bucket specified | |
6833 | * in @bucket_info. @obj_prefix and @obj_delim narrow down the listing | |
6834 | * to objects matching these criterias. */ | |
6835 | int RGWRados::on_last_entry_in_listing(RGWBucketInfo& bucket_info, | |
6836 | const std::string& obj_prefix, | |
6837 | const std::string& obj_delim, | |
6838 | std::function<int(const rgw_bucket_dir_entry&)> handler) | |
6839 | { | |
6840 | RGWRados::Bucket target(this, bucket_info); | |
6841 | RGWRados::Bucket::List list_op(&target); | |
6842 | ||
6843 | list_op.params.prefix = obj_prefix; | |
6844 | list_op.params.delim = obj_delim; | |
6845 | ||
6846 | ldout(cct, 20) << "iterating listing for bucket=" << bucket_info.bucket.name | |
6847 | << ", obj_prefix=" << obj_prefix | |
6848 | << ", obj_delim=" << obj_delim | |
6849 | << dendl; | |
6850 | ||
6851 | bool is_truncated = false; | |
6852 | ||
6853 | boost::optional<rgw_bucket_dir_entry> last_entry; | |
6854 | /* We need to rewind to the last object in a listing. */ | |
6855 | do { | |
6856 | /* List bucket entries in chunks. */ | |
6857 | static constexpr int MAX_LIST_OBJS = 100; | |
6858 | std::vector<rgw_bucket_dir_entry> entries(MAX_LIST_OBJS); | |
6859 | ||
6860 | int ret = list_op.list_objects(MAX_LIST_OBJS, &entries, nullptr, | |
6861 | &is_truncated); | |
6862 | if (ret < 0) { | |
6863 | return ret; | |
6864 | } else if (!entries.empty()) { | |
6865 | last_entry = entries.back(); | |
6866 | } | |
6867 | } while (is_truncated); | |
6868 | ||
6869 | if (last_entry) { | |
6870 | return handler(*last_entry); | |
6871 | } | |
6872 | ||
6873 | /* Empty listing - no items we can run handler on. */ | |
6874 | return 0; | |
6875 | } | |
6876 | ||
6877 | ||
6878 | int RGWRados::swift_versioning_copy(RGWObjectCtx& obj_ctx, | |
6879 | const rgw_user& user, | |
6880 | RGWBucketInfo& bucket_info, | |
6881 | rgw_obj& obj) | |
6882 | { | |
6883 | if (! swift_versioning_enabled(bucket_info)) { | |
6884 | return 0; | |
6885 | } | |
6886 | ||
6887 | obj_ctx.obj.set_atomic(obj); | |
6888 | ||
6889 | RGWObjState * state = nullptr; | |
6890 | int r = get_obj_state(&obj_ctx, bucket_info, obj, &state, false); | |
6891 | if (r < 0) { | |
6892 | return r; | |
6893 | } | |
6894 | ||
6895 | if (!state->exists) { | |
6896 | return 0; | |
6897 | } | |
6898 | ||
6899 | string client_id; | |
6900 | string op_id; | |
6901 | ||
6902 | const string& src_name = obj.get_oid(); | |
6903 | char buf[src_name.size() + 32]; | |
6904 | struct timespec ts = ceph::real_clock::to_timespec(state->mtime); | |
6905 | snprintf(buf, sizeof(buf), "%03x%s/%lld.%06ld", (int)src_name.size(), | |
6906 | src_name.c_str(), (long long)ts.tv_sec, ts.tv_nsec / 1000); | |
6907 | ||
6908 | RGWBucketInfo dest_bucket_info; | |
6909 | ||
6910 | r = get_bucket_info(obj_ctx, bucket_info.bucket.tenant, bucket_info.swift_ver_location, dest_bucket_info, NULL, NULL); | |
6911 | if (r < 0) { | |
6912 | ldout(cct, 10) << "failed to read dest bucket info: r=" << r << dendl; | |
6913 | if (r == -ENOENT) { | |
6914 | return -ERR_PRECONDITION_FAILED; | |
6915 | } | |
6916 | return r; | |
6917 | } | |
6918 | ||
6919 | if (dest_bucket_info.owner != bucket_info.owner) { | |
6920 | return -ERR_PRECONDITION_FAILED; | |
6921 | } | |
6922 | ||
6923 | rgw_obj dest_obj(dest_bucket_info.bucket, buf); | |
6924 | obj_ctx.obj.set_atomic(dest_obj); | |
6925 | ||
6926 | string no_zone; | |
6927 | ||
6928 | r = copy_obj(obj_ctx, | |
6929 | user, | |
6930 | client_id, | |
6931 | op_id, | |
6932 | NULL, /* req_info *info */ | |
6933 | no_zone, | |
6934 | dest_obj, | |
6935 | obj, | |
6936 | dest_bucket_info, | |
6937 | bucket_info, | |
6938 | NULL, /* time_t *src_mtime */ | |
6939 | NULL, /* time_t *mtime */ | |
6940 | NULL, /* const time_t *mod_ptr */ | |
6941 | NULL, /* const time_t *unmod_ptr */ | |
6942 | false, /* bool high_precision_time */ | |
6943 | NULL, /* const char *if_match */ | |
6944 | NULL, /* const char *if_nomatch */ | |
6945 | RGWRados::ATTRSMOD_NONE, | |
6946 | true, /* bool copy_if_newer */ | |
6947 | state->attrset, | |
6948 | RGW_OBJ_CATEGORY_MAIN, | |
6949 | 0, /* uint64_t olh_epoch */ | |
6950 | real_time(), /* time_t delete_at */ | |
6951 | NULL, /* string *version_id */ | |
6952 | NULL, /* string *ptag */ | |
6953 | NULL, /* string *petag */ | |
6954 | NULL, /* void (*progress_cb)(off_t, void *) */ | |
6955 | NULL); /* void *progress_data */ | |
6956 | if (r == -ECANCELED || r == -ENOENT) { | |
6957 | /* Has already been overwritten, meaning another rgw process already | |
6958 | * copied it out */ | |
6959 | return 0; | |
6960 | } | |
6961 | ||
6962 | return r; | |
6963 | } | |
6964 | ||
6965 | int RGWRados::swift_versioning_restore(RGWObjectCtx& obj_ctx, | |
6966 | const rgw_user& user, | |
6967 | RGWBucketInfo& bucket_info, | |
6968 | rgw_obj& obj, | |
6969 | bool& restored) /* out */ | |
6970 | { | |
6971 | if (! swift_versioning_enabled(bucket_info)) { | |
6972 | return 0; | |
6973 | } | |
6974 | ||
6975 | /* Bucket info of the bucket that stores previous versions of our object. */ | |
6976 | RGWBucketInfo archive_binfo; | |
6977 | ||
6978 | int ret = get_bucket_info(obj_ctx, bucket_info.bucket.tenant, | |
6979 | bucket_info.swift_ver_location, archive_binfo, | |
6980 | nullptr, nullptr); | |
6981 | if (ret < 0) { | |
6982 | return ret; | |
6983 | } | |
6984 | ||
6985 | /* Abort the operation if the bucket storing our archive belongs to someone | |
6986 | * else. This is a limitation in comparison to Swift as we aren't taking ACLs | |
6987 | * into consideration. For we can live with that. | |
6988 | * | |
6989 | * TODO: delegate this check to un upper layer and compare with ACLs. */ | |
6990 | if (bucket_info.owner != archive_binfo.owner) { | |
6991 | return -EPERM; | |
6992 | } | |
6993 | ||
6994 | /* This code will be executed on latest version of the object. */ | |
6995 | const auto handler = [&](const rgw_bucket_dir_entry& entry) -> int { | |
6996 | std::string no_client_id; | |
6997 | std::string no_op_id; | |
6998 | std::string no_zone; | |
6999 | ||
7000 | /* We don't support object versioning of Swift API on those buckets that | |
7001 | * are already versioned using the S3 mechanism. This affects also bucket | |
7002 | * storing archived objects. Otherwise the delete operation would create | |
7003 | * a deletion marker. */ | |
7004 | if (archive_binfo.versioned()) { | |
7005 | restored = false; | |
7006 | return -ERR_PRECONDITION_FAILED; | |
7007 | } | |
7008 | ||
7009 | /* We are requesting ATTRSMOD_NONE so the attr attribute is perfectly | |
7010 | * irrelevant and may be safely skipped. */ | |
7011 | std::map<std::string, ceph::bufferlist> no_attrs; | |
7012 | ||
7013 | rgw_obj archive_obj(archive_binfo.bucket, entry.key); | |
7014 | obj_ctx.obj.set_atomic(archive_obj); | |
7015 | obj_ctx.obj.set_atomic(obj); | |
7016 | ||
7017 | int ret = copy_obj(obj_ctx, | |
7018 | user, | |
7019 | no_client_id, | |
7020 | no_op_id, | |
7021 | nullptr, /* req_info *info */ | |
7022 | no_zone, | |
7023 | obj, /* dest obj */ | |
7024 | archive_obj, /* src obj */ | |
7025 | bucket_info, /* dest bucket info */ | |
7026 | archive_binfo, /* src bucket info */ | |
7027 | nullptr, /* time_t *src_mtime */ | |
7028 | nullptr, /* time_t *mtime */ | |
7029 | nullptr, /* const time_t *mod_ptr */ | |
7030 | nullptr, /* const time_t *unmod_ptr */ | |
7031 | false, /* bool high_precision_time */ | |
7032 | nullptr, /* const char *if_match */ | |
7033 | nullptr, /* const char *if_nomatch */ | |
7034 | RGWRados::ATTRSMOD_NONE, | |
7035 | true, /* bool copy_if_newer */ | |
7036 | no_attrs, | |
7037 | RGW_OBJ_CATEGORY_MAIN, | |
7038 | 0, /* uint64_t olh_epoch */ | |
7039 | real_time(), /* time_t delete_at */ | |
7040 | nullptr, /* string *version_id */ | |
7041 | nullptr, /* string *ptag */ | |
7042 | nullptr, /* string *petag */ | |
7043 | nullptr, /* void (*progress_cb)(off_t, void *) */ | |
7044 | nullptr); /* void *progress_data */ | |
7045 | if (ret == -ECANCELED || ret == -ENOENT) { | |
7046 | /* Has already been overwritten, meaning another rgw process already | |
7047 | * copied it out */ | |
7048 | return 0; | |
7049 | } else if (ret < 0) { | |
7050 | return ret; | |
7051 | } else { | |
7052 | restored = true; | |
7053 | } | |
7054 | ||
7055 | /* Need to remove the archived copy. */ | |
7056 | ret = delete_obj(obj_ctx, archive_binfo, archive_obj, | |
7057 | archive_binfo.versioning_status()); | |
7058 | ||
7059 | return ret; | |
7060 | }; | |
7061 | ||
7062 | const std::string& obj_name = obj.get_oid(); | |
7063 | const auto prefix = boost::str(boost::format("%03x%s") % obj_name.size() | |
7064 | % obj_name); | |
7065 | ||
7066 | return on_last_entry_in_listing(archive_binfo, prefix, std::string(), | |
7067 | handler); | |
7068 | } | |
7069 | ||
7070 | /** | |
7071 | * Write/overwrite an object to the bucket storage. | |
7072 | * bucket: the bucket to store the object in | |
7073 | * obj: the object name/key | |
7074 | * data: the object contents/value | |
7075 | * size: the amount of data to write (data must be this long) | |
7076 | * accounted_size: original size of data before compression, encryption | |
7077 | * mtime: if non-NULL, writes the given mtime to the bucket storage | |
7078 | * attrs: all the given attrs are written to bucket storage for the given object | |
7079 | * exclusive: create object exclusively | |
7080 | * Returns: 0 on success, -ERR# otherwise. | |
7081 | */ | |
7082 | int RGWRados::Object::Write::_do_write_meta(uint64_t size, uint64_t accounted_size, | |
7083 | map<string, bufferlist>& attrs, | |
7084 | bool assume_noent, bool modify_tail, | |
7085 | void *_index_op) | |
7086 | { | |
7087 | RGWRados::Bucket::UpdateIndex *index_op = static_cast<RGWRados::Bucket::UpdateIndex *>(_index_op); | |
7088 | RGWRados *store = target->get_store(); | |
7089 | ||
7090 | ObjectWriteOperation op; | |
7091 | ||
7092 | RGWObjState *state; | |
7093 | int r = target->get_state(&state, false, assume_noent); | |
7094 | if (r < 0) | |
7095 | return r; | |
7096 | ||
7097 | rgw_obj& obj = target->get_obj(); | |
7098 | ||
7099 | if (obj.get_oid().empty()) { | |
7100 | ldout(store->ctx(), 0) << "ERROR: " << __func__ << "(): cannot write object with empty name" << dendl; | |
7101 | return -EIO; | |
7102 | } | |
7103 | ||
7104 | rgw_rados_ref ref; | |
7105 | r = store->get_obj_head_ref(target->get_bucket_info(), obj, &ref); | |
7106 | if (r < 0) | |
7107 | return r; | |
7108 | ||
7109 | bool is_olh = state->is_olh; | |
7110 | ||
7111 | bool reset_obj = (meta.flags & PUT_OBJ_CREATE) != 0; | |
7112 | ||
7113 | const string *ptag = meta.ptag; | |
7114 | if (!ptag && !index_op->get_optag()->empty()) { | |
7115 | ptag = index_op->get_optag(); | |
7116 | } | |
7117 | r = target->prepare_atomic_modification(op, reset_obj, ptag, meta.if_match, meta.if_nomatch, false, modify_tail); | |
7118 | if (r < 0) | |
7119 | return r; | |
7120 | ||
7121 | if (real_clock::is_zero(meta.set_mtime)) { | |
7122 | meta.set_mtime = real_clock::now(); | |
7123 | } | |
7124 | ||
7125 | if (state->is_olh) { | |
7126 | op.setxattr(RGW_ATTR_OLH_ID_TAG, state->olh_tag); | |
7127 | } | |
7128 | ||
7129 | struct timespec mtime_ts = real_clock::to_timespec(meta.set_mtime); | |
7130 | op.mtime2(&mtime_ts); | |
7131 | ||
7132 | if (meta.data) { | |
7133 | /* if we want to overwrite the data, we also want to overwrite the | |
7134 | xattrs, so just remove the object */ | |
7135 | op.write_full(*meta.data); | |
7136 | } | |
7137 | ||
7138 | string etag; | |
7139 | string content_type; | |
7140 | bufferlist acl_bl; | |
7141 | ||
7142 | map<string, bufferlist>::iterator iter; | |
7143 | if (meta.rmattrs) { | |
7144 | for (iter = meta.rmattrs->begin(); iter != meta.rmattrs->end(); ++iter) { | |
7145 | const string& name = iter->first; | |
7146 | op.rmxattr(name.c_str()); | |
7147 | } | |
7148 | } | |
7149 | ||
7150 | if (meta.manifest) { | |
7151 | /* remove existing manifest attr */ | |
7152 | iter = attrs.find(RGW_ATTR_MANIFEST); | |
7153 | if (iter != attrs.end()) | |
7154 | attrs.erase(iter); | |
7155 | ||
7156 | bufferlist bl; | |
7157 | ::encode(*meta.manifest, bl); | |
7158 | op.setxattr(RGW_ATTR_MANIFEST, bl); | |
7159 | } | |
7160 | ||
7161 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
7162 | const string& name = iter->first; | |
7163 | bufferlist& bl = iter->second; | |
7164 | ||
7165 | if (!bl.length()) | |
7166 | continue; | |
7167 | ||
7168 | op.setxattr(name.c_str(), bl); | |
7169 | ||
7170 | if (name.compare(RGW_ATTR_ETAG) == 0) { | |
7171 | etag = bl.c_str(); | |
7172 | } else if (name.compare(RGW_ATTR_CONTENT_TYPE) == 0) { | |
7173 | content_type = bl.c_str(); | |
7174 | } else if (name.compare(RGW_ATTR_ACL) == 0) { | |
7175 | acl_bl = bl; | |
7176 | } | |
7177 | } | |
7178 | if (attrs.find(RGW_ATTR_PG_VER) == attrs.end()) { | |
7179 | cls_rgw_obj_store_pg_ver(op, RGW_ATTR_PG_VER); | |
7180 | } | |
7181 | ||
7182 | if (attrs.find(RGW_ATTR_SOURCE_ZONE) == attrs.end()) { | |
7183 | bufferlist bl; | |
7184 | ::encode(store->get_zone_short_id(), bl); | |
7185 | op.setxattr(RGW_ATTR_SOURCE_ZONE, bl); | |
7186 | } | |
7187 | ||
7188 | if (!op.size()) | |
7189 | return 0; | |
7190 | ||
7191 | uint64_t epoch; | |
7192 | int64_t poolid; | |
7193 | bool orig_exists; | |
7194 | uint64_t orig_size; | |
7195 | ||
7196 | if (!reset_obj) { //Multipart upload, it has immutable head. | |
7197 | orig_exists = false; | |
7198 | orig_size = 0; | |
7199 | } else { | |
7200 | orig_exists = state->exists; | |
7201 | orig_size = state->accounted_size; | |
7202 | } | |
7203 | ||
7204 | bool versioned_target = (meta.olh_epoch && *meta.olh_epoch > 0) || | |
7205 | !obj.key.instance.empty(); | |
7206 | ||
7207 | bool versioned_op = (target->versioning_enabled() || is_olh || versioned_target); | |
7208 | ||
7209 | if (versioned_op) { | |
7210 | index_op->set_bilog_flags(RGW_BILOG_FLAG_VERSIONED_OP); | |
7211 | } | |
7212 | ||
7213 | if (!index_op->is_prepared()) { | |
7214 | r = index_op->prepare(CLS_RGW_OP_ADD, &state->write_tag); | |
7215 | if (r < 0) | |
7216 | return r; | |
7217 | } | |
7218 | ||
7219 | r = ref.ioctx.operate(ref.oid, &op); | |
7220 | if (r < 0) { /* we can expect to get -ECANCELED if object was replaced under, | |
7221 | or -ENOENT if was removed, or -EEXIST if it did not exist | |
7222 | before and now it does */ | |
7223 | if (r == -EEXIST && assume_noent) { | |
7224 | target->invalidate_state(); | |
7225 | return r; | |
7226 | } | |
7227 | goto done_cancel; | |
7228 | } | |
7229 | ||
7230 | epoch = ref.ioctx.get_last_version(); | |
7231 | poolid = ref.ioctx.get_id(); | |
7232 | ||
7233 | r = target->complete_atomic_modification(); | |
7234 | if (r < 0) { | |
7235 | ldout(store->ctx(), 0) << "ERROR: complete_atomic_modification returned r=" << r << dendl; | |
7236 | } | |
7237 | ||
7238 | r = index_op->complete(poolid, epoch, size, accounted_size, | |
7239 | meta.set_mtime, etag, content_type, &acl_bl, | |
7240 | meta.category, meta.remove_objs, meta.user_data); | |
7241 | if (r < 0) | |
7242 | goto done_cancel; | |
7243 | ||
7244 | if (meta.mtime) { | |
7245 | *meta.mtime = meta.set_mtime; | |
7246 | } | |
7247 | ||
7248 | /* note that index_op was using state so we couldn't invalidate it earlier */ | |
7249 | target->invalidate_state(); | |
7250 | state = NULL; | |
7251 | ||
7252 | if (versioned_op && meta.olh_epoch) { | |
7253 | r = store->set_olh(target->get_ctx(), target->get_bucket_info(), obj, false, NULL, *meta.olh_epoch, real_time(), false, meta.zones_trace); | |
7254 | if (r < 0) { | |
7255 | return r; | |
7256 | } | |
7257 | } | |
7258 | ||
7259 | if (!real_clock::is_zero(meta.delete_at)) { | |
7260 | rgw_obj_index_key obj_key; | |
7261 | obj.key.get_index_key(&obj_key); | |
7262 | ||
7263 | r = store->objexp_hint_add(meta.delete_at, | |
7264 | obj.bucket.tenant, obj.bucket.name, obj.bucket.bucket_id, obj_key); | |
7265 | if (r < 0) { | |
7266 | ldout(store->ctx(), 0) << "ERROR: objexp_hint_add() returned r=" << r << ", object will not get removed" << dendl; | |
7267 | /* ignoring error, nothing we can do at this point */ | |
7268 | } | |
7269 | } | |
7270 | meta.canceled = false; | |
7271 | ||
7272 | /* update quota cache */ | |
7273 | if (meta.completeMultipart){ | |
7274 | store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1), | |
7275 | 0, orig_size); | |
7276 | } | |
7277 | else { | |
7278 | store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1), | |
7279 | accounted_size, orig_size); | |
7280 | } | |
7281 | return 0; | |
7282 | ||
7283 | done_cancel: | |
7284 | int ret = index_op->cancel(); | |
7285 | if (ret < 0) { | |
7286 | ldout(store->ctx(), 0) << "ERROR: index_op.cancel()() returned ret=" << ret << dendl; | |
7287 | } | |
7288 | ||
7289 | meta.canceled = true; | |
7290 | ||
7291 | /* we lost in a race. There are a few options: | |
7292 | * - existing object was rewritten (ECANCELED) | |
7293 | * - non existing object was created (EEXIST) | |
7294 | * - object was removed (ENOENT) | |
7295 | * should treat it as a success | |
7296 | */ | |
7297 | if (meta.if_match == NULL && meta.if_nomatch == NULL) { | |
7298 | if (r == -ECANCELED || r == -ENOENT || r == -EEXIST) { | |
7299 | r = 0; | |
7300 | } | |
7301 | } else { | |
7302 | if (meta.if_match != NULL) { | |
7303 | // only overwrite existing object | |
7304 | if (strcmp(meta.if_match, "*") == 0) { | |
7305 | if (r == -ENOENT) { | |
7306 | r = -ERR_PRECONDITION_FAILED; | |
7307 | } else if (r == -ECANCELED) { | |
7308 | r = 0; | |
7309 | } | |
7310 | } | |
7311 | } | |
7312 | ||
7313 | if (meta.if_nomatch != NULL) { | |
7314 | // only create a new object | |
7315 | if (strcmp(meta.if_nomatch, "*") == 0) { | |
7316 | if (r == -EEXIST) { | |
7317 | r = -ERR_PRECONDITION_FAILED; | |
7318 | } else if (r == -ENOENT) { | |
7319 | r = 0; | |
7320 | } | |
7321 | } | |
7322 | } | |
7323 | } | |
7324 | ||
7325 | return r; | |
7326 | } | |
7327 | ||
7328 | int RGWRados::Object::Write::write_meta(uint64_t size, uint64_t accounted_size, | |
7329 | map<string, bufferlist>& attrs) | |
7330 | { | |
7331 | RGWBucketInfo& bucket_info = target->get_bucket_info(); | |
7332 | ||
7333 | RGWRados::Bucket bop(target->get_store(), bucket_info); | |
7334 | RGWRados::Bucket::UpdateIndex index_op(&bop, target->get_obj()); | |
7335 | index_op.set_zones_trace(meta.zones_trace); | |
7336 | ||
7337 | bool assume_noent = (meta.if_match == NULL && meta.if_nomatch == NULL); | |
7338 | int r; | |
7339 | if (assume_noent) { | |
7340 | r = _do_write_meta(size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op); | |
7341 | if (r == -EEXIST) { | |
7342 | assume_noent = false; | |
7343 | } | |
7344 | } | |
7345 | if (!assume_noent) { | |
7346 | r = _do_write_meta(size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op); | |
7347 | } | |
7348 | return r; | |
7349 | } | |
7350 | ||
7351 | /** Write/overwrite a system object. */ | |
7352 | int RGWRados::put_system_obj_impl(rgw_raw_obj& obj, uint64_t size, real_time *mtime, | |
7353 | map<std::string, bufferlist>& attrs, int flags, | |
7354 | bufferlist& data, | |
7355 | RGWObjVersionTracker *objv_tracker, | |
7356 | real_time set_mtime /* 0 for don't set */) | |
7357 | { | |
7358 | rgw_rados_ref ref; | |
7359 | int r = get_system_obj_ref(obj, &ref); | |
7360 | if (r < 0) | |
7361 | return r; | |
7362 | ||
7363 | ObjectWriteOperation op; | |
7364 | ||
7365 | if (flags & PUT_OBJ_EXCL) { | |
7366 | if (!(flags & PUT_OBJ_CREATE)) | |
7367 | return -EINVAL; | |
7368 | op.create(true); // exclusive create | |
7369 | } else { | |
7370 | op.remove(); | |
7371 | op.set_op_flags2(LIBRADOS_OP_FLAG_FAILOK); | |
7372 | op.create(false); | |
7373 | } | |
7374 | ||
7375 | if (objv_tracker) { | |
7376 | objv_tracker->prepare_op_for_write(&op); | |
7377 | } | |
7378 | ||
7379 | if (real_clock::is_zero(set_mtime)) { | |
7380 | set_mtime = real_clock::now(); | |
7381 | } | |
7382 | ||
7383 | struct timespec mtime_ts = real_clock::to_timespec(set_mtime); | |
7384 | op.mtime2(&mtime_ts); | |
7385 | op.write_full(data); | |
7386 | ||
7387 | bufferlist acl_bl; | |
7388 | ||
7389 | for (map<string, bufferlist>::iterator iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
7390 | const string& name = iter->first; | |
7391 | bufferlist& bl = iter->second; | |
7392 | ||
7393 | if (!bl.length()) | |
7394 | continue; | |
7395 | ||
7396 | op.setxattr(name.c_str(), bl); | |
7397 | } | |
7398 | ||
7399 | r = ref.ioctx.operate(ref.oid, &op); | |
7400 | if (r < 0) { | |
7401 | return r; | |
7402 | } | |
7403 | ||
7404 | if (objv_tracker) { | |
7405 | objv_tracker->apply_write(); | |
7406 | } | |
7407 | ||
7408 | if (mtime) { | |
7409 | *mtime = set_mtime; | |
7410 | } | |
7411 | ||
7412 | return 0; | |
7413 | } | |
7414 | ||
7415 | int RGWRados::put_system_obj_data(void *ctx, rgw_raw_obj& obj, bufferlist& bl, | |
7416 | off_t ofs, bool exclusive, | |
7417 | RGWObjVersionTracker *objv_tracker) | |
7418 | { | |
7419 | rgw_rados_ref ref; | |
7420 | int r = get_system_obj_ref(obj, &ref); | |
7421 | if (r < 0) { | |
7422 | return r; | |
7423 | } | |
7424 | ||
7425 | ObjectWriteOperation op; | |
7426 | ||
7427 | if (exclusive) | |
7428 | op.create(true); | |
7429 | ||
7430 | if (objv_tracker) { | |
7431 | objv_tracker->prepare_op_for_write(&op); | |
7432 | } | |
7433 | if (ofs == -1) { | |
7434 | op.write_full(bl); | |
7435 | } else { | |
7436 | op.write(ofs, bl); | |
7437 | } | |
7438 | r = ref.ioctx.operate(ref.oid, &op); | |
7439 | if (r < 0) | |
7440 | return r; | |
7441 | ||
7442 | if (objv_tracker) { | |
7443 | objv_tracker->apply_write(); | |
7444 | } | |
7445 | return 0; | |
7446 | } | |
7447 | ||
7448 | /** | |
7449 | * Write/overwrite an object to the bucket storage. | |
7450 | * bucket: the bucket to store the object in | |
7451 | * obj: the object name/key | |
7452 | * data: the object contents/value | |
7453 | * offset: the offet to write to in the object | |
7454 | * If this is -1, we will overwrite the whole object. | |
7455 | * size: the amount of data to write (data must be this long) | |
7456 | * attrs: all the given attrs are written to bucket storage for the given object | |
7457 | * Returns: 0 on success, -ERR# otherwise. | |
7458 | */ | |
7459 | ||
7460 | int RGWRados::aio_put_obj_data(void *ctx, rgw_raw_obj& obj, bufferlist& bl, | |
7461 | off_t ofs, bool exclusive, | |
7462 | void **handle) | |
7463 | { | |
7464 | rgw_rados_ref ref; | |
7465 | int r = get_raw_obj_ref(obj, &ref); | |
7466 | if (r < 0) { | |
7467 | return r; | |
7468 | } | |
7469 | ||
7470 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
7471 | *handle = c; | |
7472 | ||
7473 | ObjectWriteOperation op; | |
7474 | ||
7475 | if (exclusive) | |
7476 | op.create(true); | |
7477 | ||
7478 | if (ofs == -1) { | |
7479 | op.write_full(bl); | |
7480 | } else { | |
7481 | op.write(ofs, bl); | |
7482 | } | |
7483 | r = ref.ioctx.aio_operate(ref.oid, c, &op); | |
7484 | if (r < 0) | |
7485 | return r; | |
7486 | ||
7487 | return 0; | |
7488 | } | |
7489 | ||
7490 | int RGWRados::aio_wait(void *handle) | |
7491 | { | |
7492 | AioCompletion *c = (AioCompletion *)handle; | |
7493 | c->wait_for_safe(); | |
7494 | int ret = c->get_return_value(); | |
7495 | c->release(); | |
7496 | return ret; | |
7497 | } | |
7498 | ||
7499 | bool RGWRados::aio_completed(void *handle) | |
7500 | { | |
7501 | AioCompletion *c = (AioCompletion *)handle; | |
7502 | return c->is_safe(); | |
7503 | } | |
7504 | ||
7505 | // PutObj filter that buffers data so we don't try to compress tiny blocks. | |
7506 | // libcurl reads in 16k at a time, and we need at least 64k to get a good | |
7507 | // compression ratio | |
7508 | class RGWPutObj_Buffer : public RGWPutObj_Filter { | |
7509 | const unsigned buffer_size; | |
7510 | bufferlist buffer; | |
7511 | public: | |
7512 | RGWPutObj_Buffer(RGWPutObjDataProcessor* next, unsigned buffer_size) | |
7513 | : RGWPutObj_Filter(next), buffer_size(buffer_size) { | |
7514 | assert(ISP2(buffer_size)); // must be power of 2 | |
7515 | } | |
7516 | ||
7517 | int handle_data(bufferlist& bl, off_t ofs, void **phandle, rgw_raw_obj *pobj, | |
7518 | bool *again) override { | |
7519 | if (*again || !bl.length()) { | |
7520 | // flush buffered data | |
7521 | return RGWPutObj_Filter::handle_data(buffer, ofs, phandle, pobj, again); | |
7522 | } | |
7523 | // transform offset to the beginning of the buffer | |
7524 | ofs = ofs - buffer.length(); | |
7525 | buffer.claim_append(bl); | |
7526 | if (buffer.length() < buffer_size) { | |
7527 | *again = false; // don't come back until there's more data | |
7528 | return 0; | |
7529 | } | |
7530 | const auto count = P2ALIGN(buffer.length(), buffer_size); | |
7531 | buffer.splice(0, count, &bl); | |
7532 | return RGWPutObj_Filter::handle_data(bl, ofs, phandle, pobj, again); | |
7533 | } | |
7534 | }; | |
7535 | ||
7536 | class RGWRadosPutObj : public RGWGetDataCB | |
7537 | { | |
7538 | CephContext* cct; | |
7539 | rgw_obj obj; | |
7540 | RGWPutObjDataProcessor *filter; | |
7541 | boost::optional<RGWPutObj_Compress>& compressor; | |
7542 | boost::optional<RGWPutObj_Buffer> buffering; | |
7543 | CompressorRef& plugin; | |
7544 | RGWPutObjProcessor_Atomic *processor; | |
7545 | RGWOpStateSingleOp *opstate; | |
7546 | void (*progress_cb)(off_t, void *); | |
7547 | void *progress_data; | |
7548 | bufferlist extra_data_bl; | |
7549 | uint64_t extra_data_left; | |
7550 | uint64_t data_len; | |
7551 | map<string, bufferlist> src_attrs; | |
7552 | public: | |
7553 | RGWRadosPutObj(CephContext* cct, | |
7554 | CompressorRef& plugin, | |
7555 | boost::optional<RGWPutObj_Compress>& compressor, | |
7556 | RGWPutObjProcessor_Atomic *p, | |
7557 | RGWOpStateSingleOp *_ops, | |
7558 | void (*_progress_cb)(off_t, void *), | |
7559 | void *_progress_data) : | |
7560 | cct(cct), | |
7561 | filter(p), | |
7562 | compressor(compressor), | |
7563 | plugin(plugin), | |
7564 | processor(p), | |
7565 | opstate(_ops), | |
7566 | progress_cb(_progress_cb), | |
7567 | progress_data(_progress_data), | |
7568 | extra_data_left(0), | |
7569 | data_len(0) {} | |
7570 | ||
7571 | int process_attrs(void) { | |
7572 | if (extra_data_bl.length()) { | |
7573 | JSONParser jp; | |
7574 | if (!jp.parse(extra_data_bl.c_str(), extra_data_bl.length())) { | |
7575 | ldout(cct, 0) << "failed to parse response extra data. len=" << extra_data_bl.length() << " data=" << extra_data_bl.c_str() << dendl; | |
7576 | return -EIO; | |
7577 | } | |
7578 | ||
7579 | JSONDecoder::decode_json("attrs", src_attrs, &jp); | |
7580 | ||
7581 | src_attrs.erase(RGW_ATTR_COMPRESSION); | |
7582 | src_attrs.erase(RGW_ATTR_MANIFEST); // not interested in original object layout | |
7583 | ||
7584 | // filter out olh attributes | |
7585 | auto iter = src_attrs.lower_bound(RGW_ATTR_OLH_PREFIX); | |
7586 | while (iter != src_attrs.end()) { | |
7587 | if (!boost::algorithm::starts_with(iter->first, RGW_ATTR_OLH_PREFIX)) { | |
7588 | break; | |
7589 | } | |
7590 | iter = src_attrs.erase(iter); | |
7591 | } | |
7592 | } | |
7593 | ||
7594 | if (plugin && src_attrs.find(RGW_ATTR_CRYPT_MODE) == src_attrs.end()) { | |
7595 | //do not compress if object is encrypted | |
7596 | compressor = boost::in_place(cct, plugin, filter); | |
7597 | constexpr unsigned buffer_size = 512 * 1024; | |
7598 | buffering = boost::in_place(&*compressor, buffer_size); | |
7599 | filter = &*buffering; | |
7600 | } | |
7601 | return 0; | |
7602 | } | |
7603 | ||
7604 | int handle_data(bufferlist& bl, off_t ofs, off_t len) override { | |
7605 | if (progress_cb) { | |
7606 | progress_cb(ofs, progress_data); | |
7607 | } | |
7608 | if (extra_data_left) { | |
7609 | size_t extra_len = bl.length(); | |
7610 | if (extra_len > extra_data_left) | |
7611 | extra_len = extra_data_left; | |
7612 | ||
7613 | bufferlist extra; | |
7614 | bl.splice(0, extra_len, &extra); | |
7615 | extra_data_bl.append(extra); | |
7616 | ||
7617 | extra_data_left -= extra_len; | |
7618 | if (extra_data_left == 0) { | |
7619 | int res = process_attrs(); | |
7620 | if (res < 0) | |
7621 | return res; | |
7622 | } | |
7623 | if (bl.length() == 0) { | |
7624 | return 0; | |
7625 | } | |
7626 | ofs += extra_len; | |
7627 | } | |
7628 | // adjust ofs based on extra_data_len, so the result is a logical offset | |
7629 | // into the object data | |
7630 | assert(uint64_t(ofs) >= extra_data_len); | |
7631 | ofs -= extra_data_len; | |
7632 | ||
7633 | data_len += bl.length(); | |
7634 | bool again = false; | |
7635 | ||
7636 | bool need_opstate = true; | |
7637 | ||
7638 | do { | |
7639 | void *handle = NULL; | |
7640 | rgw_raw_obj obj; | |
7641 | uint64_t size = bl.length(); | |
7642 | int ret = filter->handle_data(bl, ofs, &handle, &obj, &again); | |
7643 | if (ret < 0) | |
7644 | return ret; | |
7645 | ||
7646 | if (need_opstate && opstate) { | |
7647 | /* need to update opstate repository with new state. This is ratelimited, so we're not | |
7648 | * really doing it every time | |
7649 | */ | |
7650 | ret = opstate->renew_state(); | |
7651 | if (ret < 0) { | |
7652 | ldout(cct, 0) << "ERROR: RGWRadosPutObj::handle_data(): failed to renew op state ret=" << ret << dendl; | |
7653 | int r = filter->throttle_data(handle, obj, size, false); | |
7654 | if (r < 0) { | |
7655 | ldout(cct, 0) << "ERROR: RGWRadosPutObj::handle_data(): processor->throttle_data() returned " << r << dendl; | |
7656 | } | |
7657 | /* could not renew state! might have been marked as cancelled */ | |
7658 | return ret; | |
7659 | } | |
7660 | need_opstate = false; | |
7661 | } | |
7662 | ||
7663 | ret = filter->throttle_data(handle, obj, size, false); | |
7664 | if (ret < 0) | |
7665 | return ret; | |
7666 | } while (again); | |
7667 | ||
7668 | return 0; | |
7669 | } | |
7670 | ||
7671 | int flush() { | |
7672 | bufferlist bl; | |
7673 | return put_data_and_throttle(filter, bl, 0, false); | |
7674 | } | |
7675 | ||
7676 | bufferlist& get_extra_data() { return extra_data_bl; } | |
7677 | ||
7678 | map<string, bufferlist>& get_attrs() { return src_attrs; } | |
7679 | ||
7680 | void set_extra_data_len(uint64_t len) override { | |
7681 | extra_data_left = len; | |
7682 | RGWGetDataCB::set_extra_data_len(len); | |
7683 | } | |
7684 | ||
7685 | uint64_t get_data_len() { | |
7686 | return data_len; | |
7687 | } | |
7688 | ||
7689 | int complete(const string& etag, real_time *mtime, real_time set_mtime, | |
7690 | map<string, bufferlist>& attrs, real_time delete_at, rgw_zone_set *zones_trace) { | |
7691 | return processor->complete(data_len, etag, mtime, set_mtime, attrs, delete_at, NULL, NULL, NULL, zones_trace); | |
7692 | } | |
7693 | ||
7694 | bool is_canceled() { | |
7695 | return processor->is_canceled(); | |
7696 | } | |
7697 | }; | |
7698 | ||
7699 | /* | |
7700 | * prepare attrset depending on attrs_mod. | |
7701 | */ | |
7702 | static void set_copy_attrs(map<string, bufferlist>& src_attrs, | |
7703 | map<string, bufferlist>& attrs, | |
7704 | RGWRados::AttrsMod attrs_mod) | |
7705 | { | |
7706 | switch (attrs_mod) { | |
7707 | case RGWRados::ATTRSMOD_NONE: | |
7708 | attrs = src_attrs; | |
7709 | break; | |
7710 | case RGWRados::ATTRSMOD_REPLACE: | |
7711 | if (!attrs[RGW_ATTR_ETAG].length()) { | |
7712 | attrs[RGW_ATTR_ETAG] = src_attrs[RGW_ATTR_ETAG]; | |
7713 | } | |
7714 | if (!attrs[RGW_ATTR_TAIL_TAG].length()) { | |
7715 | auto ttiter = src_attrs.find(RGW_ATTR_TAIL_TAG); | |
7716 | if (ttiter != src_attrs.end()) { | |
7717 | attrs[RGW_ATTR_TAIL_TAG] = src_attrs[RGW_ATTR_TAIL_TAG]; | |
7718 | } | |
7719 | } | |
7720 | break; | |
7721 | case RGWRados::ATTRSMOD_MERGE: | |
7722 | for (map<string, bufferlist>::iterator it = src_attrs.begin(); it != src_attrs.end(); ++it) { | |
7723 | if (attrs.find(it->first) == attrs.end()) { | |
7724 | attrs[it->first] = it->second; | |
7725 | } | |
7726 | } | |
7727 | break; | |
7728 | } | |
7729 | } | |
7730 | ||
7731 | int RGWRados::rewrite_obj(RGWBucketInfo& dest_bucket_info, rgw_obj& obj) | |
7732 | { | |
7733 | map<string, bufferlist> attrset; | |
7734 | ||
7735 | real_time mtime; | |
7736 | uint64_t obj_size; | |
7737 | RGWObjectCtx rctx(this); | |
7738 | ||
7739 | RGWRados::Object op_target(this, dest_bucket_info, rctx, obj); | |
7740 | RGWRados::Object::Read read_op(&op_target); | |
7741 | ||
7742 | read_op.params.attrs = &attrset; | |
7743 | read_op.params.lastmod = &mtime; | |
7744 | read_op.params.obj_size = &obj_size; | |
7745 | ||
7746 | int ret = read_op.prepare(); | |
7747 | if (ret < 0) | |
7748 | return ret; | |
7749 | ||
7750 | attrset.erase(RGW_ATTR_ID_TAG); | |
7751 | attrset.erase(RGW_ATTR_TAIL_TAG); | |
7752 | ||
7753 | uint64_t max_chunk_size; | |
7754 | ||
7755 | ret = get_max_chunk_size(dest_bucket_info.placement_rule, obj, &max_chunk_size); | |
7756 | if (ret < 0) { | |
7757 | ldout(cct, 0) << "ERROR: failed to get max_chunk_size() for bucket " << obj.bucket << dendl; | |
7758 | return ret; | |
7759 | } | |
7760 | ||
7761 | return copy_obj_data(rctx, dest_bucket_info, read_op, obj_size - 1, obj, obj, | |
7762 | max_chunk_size, NULL, mtime, attrset, | |
7763 | RGW_OBJ_CATEGORY_MAIN, 0, real_time(), | |
7764 | (obj.key.instance.empty() ? NULL : &(obj.key.instance)), | |
7765 | NULL, NULL); | |
7766 | } | |
7767 | ||
7768 | struct obj_time_weight { | |
7769 | real_time mtime; | |
7770 | uint32_t zone_short_id; | |
7771 | uint64_t pg_ver; | |
7772 | bool high_precision; | |
7773 | ||
7774 | obj_time_weight() : zone_short_id(0), pg_ver(0), high_precision(false) {} | |
7775 | ||
7776 | bool compare_low_precision(const obj_time_weight& rhs) { | |
7777 | struct timespec l = ceph::real_clock::to_timespec(mtime); | |
7778 | struct timespec r = ceph::real_clock::to_timespec(rhs.mtime); | |
7779 | l.tv_nsec = 0; | |
7780 | r.tv_nsec = 0; | |
7781 | if (l > r) { | |
7782 | return false; | |
7783 | } | |
7784 | if (l < r) { | |
7785 | return true; | |
7786 | } | |
7787 | if (zone_short_id != rhs.zone_short_id) { | |
7788 | return (zone_short_id < rhs.zone_short_id); | |
7789 | } | |
7790 | return (pg_ver < rhs.pg_ver); | |
7791 | ||
7792 | } | |
7793 | ||
7794 | bool operator<(const obj_time_weight& rhs) { | |
7795 | if (!high_precision || !rhs.high_precision) { | |
7796 | return compare_low_precision(rhs); | |
7797 | } | |
7798 | if (mtime > rhs.mtime) { | |
7799 | return false; | |
7800 | } | |
7801 | if (mtime < rhs.mtime) { | |
7802 | return true; | |
7803 | } | |
7804 | if (zone_short_id != rhs.zone_short_id) { | |
7805 | return (zone_short_id < rhs.zone_short_id); | |
7806 | } | |
7807 | return (pg_ver < rhs.pg_ver); | |
7808 | } | |
7809 | ||
7810 | void init(const real_time& _mtime, uint32_t _short_id, uint64_t _pg_ver) { | |
7811 | mtime = _mtime; | |
7812 | zone_short_id = _short_id; | |
7813 | pg_ver = _pg_ver; | |
7814 | } | |
7815 | ||
7816 | void init(RGWObjState *state) { | |
7817 | mtime = state->mtime; | |
7818 | zone_short_id = state->zone_short_id; | |
7819 | pg_ver = state->pg_ver; | |
7820 | } | |
7821 | }; | |
7822 | ||
7823 | inline ostream& operator<<(ostream& out, const obj_time_weight &o) { | |
7824 | out << o.mtime; | |
7825 | ||
7826 | if (o.zone_short_id != 0 || o.pg_ver != 0) { | |
7827 | out << "[zid=" << o.zone_short_id << ", pgv=" << o.pg_ver << "]"; | |
7828 | } | |
7829 | ||
7830 | return out; | |
7831 | } | |
7832 | ||
7833 | class RGWGetExtraDataCB : public RGWGetDataCB { | |
7834 | bufferlist extra_data; | |
7835 | public: | |
7836 | RGWGetExtraDataCB() {} | |
7837 | int handle_data(bufferlist& bl, off_t bl_ofs, off_t bl_len) override { | |
7838 | if (extra_data.length() < extra_data_len) { | |
7839 | off_t max = extra_data_len - extra_data.length(); | |
7840 | if (max > bl_len) { | |
7841 | max = bl_len; | |
7842 | } | |
7843 | bl.splice(0, max, &extra_data); | |
7844 | } | |
7845 | return bl_len; | |
7846 | } | |
7847 | ||
7848 | bufferlist& get_extra_data() { | |
7849 | return extra_data; | |
7850 | } | |
7851 | }; | |
7852 | ||
7853 | int RGWRados::stat_remote_obj(RGWObjectCtx& obj_ctx, | |
7854 | const rgw_user& user_id, | |
7855 | const string& client_id, | |
7856 | req_info *info, | |
7857 | const string& source_zone, | |
7858 | rgw_obj& src_obj, | |
7859 | RGWBucketInfo& src_bucket_info, | |
7860 | real_time *src_mtime, | |
7861 | uint64_t *psize, | |
7862 | const real_time *mod_ptr, | |
7863 | const real_time *unmod_ptr, | |
7864 | bool high_precision_time, | |
7865 | const char *if_match, | |
7866 | const char *if_nomatch, | |
7867 | map<string, bufferlist> *pattrs, | |
7868 | string *version_id, | |
7869 | string *ptag, | |
7870 | string *petag) | |
7871 | { | |
7872 | /* source is in a different zonegroup, copy from there */ | |
7873 | ||
7874 | RGWRESTStreamRWRequest *in_stream_req; | |
7875 | string tag; | |
7876 | map<string, bufferlist> src_attrs; | |
7877 | append_rand_alpha(cct, tag, tag, 32); | |
7878 | obj_time_weight set_mtime_weight; | |
7879 | set_mtime_weight.high_precision = high_precision_time; | |
7880 | ||
7881 | RGWRESTConn *conn; | |
7882 | if (source_zone.empty()) { | |
7883 | if (src_bucket_info.zonegroup.empty()) { | |
7884 | /* source is in the master zonegroup */ | |
7885 | conn = rest_master_conn; | |
7886 | } else { | |
7887 | map<string, RGWRESTConn *>::iterator iter = zonegroup_conn_map.find(src_bucket_info.zonegroup); | |
7888 | if (iter == zonegroup_conn_map.end()) { | |
7889 | ldout(cct, 0) << "could not find zonegroup connection to zonegroup: " << source_zone << dendl; | |
7890 | return -ENOENT; | |
7891 | } | |
7892 | conn = iter->second; | |
7893 | } | |
7894 | } else { | |
7895 | map<string, RGWRESTConn *>::iterator iter = zone_conn_map.find(source_zone); | |
7896 | if (iter == zone_conn_map.end()) { | |
7897 | ldout(cct, 0) << "could not find zone connection to zone: " << source_zone << dendl; | |
7898 | return -ENOENT; | |
7899 | } | |
7900 | conn = iter->second; | |
7901 | } | |
7902 | ||
7903 | RGWGetExtraDataCB cb; | |
7904 | string etag; | |
7905 | map<string, string> req_headers; | |
7906 | real_time set_mtime; | |
7907 | ||
7908 | const real_time *pmod = mod_ptr; | |
7909 | ||
7910 | obj_time_weight dest_mtime_weight; | |
7911 | ||
7912 | constexpr bool prepend_meta = true; | |
7913 | constexpr bool get_op = true; | |
7914 | constexpr bool rgwx_stat = true; | |
7915 | constexpr bool sync_manifest = true; | |
7916 | constexpr bool skip_decrypt = true; | |
7917 | int ret = conn->get_obj(user_id, info, src_obj, pmod, unmod_ptr, | |
7918 | dest_mtime_weight.zone_short_id, dest_mtime_weight.pg_ver, | |
7919 | prepend_meta, get_op, rgwx_stat, | |
7920 | sync_manifest, skip_decrypt, &cb, &in_stream_req); | |
7921 | if (ret < 0) { | |
7922 | return ret; | |
7923 | } | |
7924 | ||
7925 | ret = conn->complete_request(in_stream_req, etag, &set_mtime, psize, req_headers); | |
7926 | if (ret < 0) { | |
7927 | return ret; | |
7928 | } | |
7929 | ||
7930 | bufferlist& extra_data_bl = cb.get_extra_data(); | |
7931 | if (extra_data_bl.length()) { | |
7932 | JSONParser jp; | |
7933 | if (!jp.parse(extra_data_bl.c_str(), extra_data_bl.length())) { | |
7934 | ldout(cct, 0) << "failed to parse response extra data. len=" << extra_data_bl.length() << " data=" << extra_data_bl.c_str() << dendl; | |
7935 | return -EIO; | |
7936 | } | |
7937 | ||
7938 | JSONDecoder::decode_json("attrs", src_attrs, &jp); | |
7939 | ||
7940 | src_attrs.erase(RGW_ATTR_MANIFEST); // not interested in original object layout | |
7941 | } | |
7942 | ||
7943 | if (src_mtime) { | |
7944 | *src_mtime = set_mtime; | |
7945 | } | |
7946 | ||
7947 | if (petag) { | |
7948 | map<string, bufferlist>::iterator iter = src_attrs.find(RGW_ATTR_ETAG); | |
7949 | if (iter != src_attrs.end()) { | |
7950 | bufferlist& etagbl = iter->second; | |
7951 | *petag = etagbl.to_str(); | |
7952 | } | |
7953 | } | |
7954 | ||
7955 | if (pattrs) { | |
7956 | *pattrs = src_attrs; | |
7957 | } | |
7958 | ||
7959 | return 0; | |
7960 | } | |
7961 | ||
7962 | int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx, | |
7963 | const rgw_user& user_id, | |
7964 | const string& client_id, | |
7965 | const string& op_id, | |
7966 | bool record_op_state, | |
7967 | req_info *info, | |
7968 | const string& source_zone, | |
7969 | rgw_obj& dest_obj, | |
7970 | rgw_obj& src_obj, | |
7971 | RGWBucketInfo& dest_bucket_info, | |
7972 | RGWBucketInfo& src_bucket_info, | |
7973 | real_time *src_mtime, | |
7974 | real_time *mtime, | |
7975 | const real_time *mod_ptr, | |
7976 | const real_time *unmod_ptr, | |
7977 | bool high_precision_time, | |
7978 | const char *if_match, | |
7979 | const char *if_nomatch, | |
7980 | AttrsMod attrs_mod, | |
7981 | bool copy_if_newer, | |
7982 | map<string, bufferlist>& attrs, | |
7983 | RGWObjCategory category, | |
7984 | boost::optional<uint64_t> olh_epoch, | |
7985 | real_time delete_at, | |
7986 | string *version_id, | |
7987 | string *ptag, | |
7988 | ceph::buffer::list *petag, | |
7989 | void (*progress_cb)(off_t, void *), | |
7990 | void *progress_data, | |
7991 | rgw_zone_set *zones_trace) | |
7992 | { | |
7993 | /* source is in a different zonegroup, copy from there */ | |
7994 | ||
7995 | RGWRESTStreamRWRequest *in_stream_req; | |
7996 | string tag; | |
7997 | int i; | |
7998 | append_rand_alpha(cct, tag, tag, 32); | |
7999 | obj_time_weight set_mtime_weight; | |
8000 | set_mtime_weight.high_precision = high_precision_time; | |
8001 | ||
8002 | RGWPutObjProcessor_Atomic processor(obj_ctx, | |
8003 | dest_bucket_info, dest_obj.bucket, dest_obj.key.name, | |
8004 | cct->_conf->rgw_obj_stripe_size, tag, dest_bucket_info.versioning_enabled()); | |
8005 | if (version_id && *version_id != "null") { | |
8006 | processor.set_version_id(*version_id); | |
8007 | } | |
8008 | if (olh_epoch) { | |
8009 | processor.set_olh_epoch(*olh_epoch); | |
8010 | } | |
8011 | int ret = processor.prepare(this, NULL); | |
8012 | if (ret < 0) { | |
8013 | return ret; | |
8014 | } | |
8015 | ||
8016 | RGWRESTConn *conn; | |
8017 | if (source_zone.empty()) { | |
8018 | if (dest_bucket_info.zonegroup.empty()) { | |
8019 | /* source is in the master zonegroup */ | |
8020 | conn = rest_master_conn; | |
8021 | } else { | |
8022 | map<string, RGWRESTConn *>::iterator iter = zonegroup_conn_map.find(src_bucket_info.zonegroup); | |
8023 | if (iter == zonegroup_conn_map.end()) { | |
8024 | ldout(cct, 0) << "could not find zonegroup connection to zonegroup: " << source_zone << dendl; | |
8025 | return -ENOENT; | |
8026 | } | |
8027 | conn = iter->second; | |
8028 | } | |
8029 | } else { | |
8030 | map<string, RGWRESTConn *>::iterator iter = zone_conn_map.find(source_zone); | |
8031 | if (iter == zone_conn_map.end()) { | |
8032 | ldout(cct, 0) << "could not find zone connection to zone: " << source_zone << dendl; | |
8033 | return -ENOENT; | |
8034 | } | |
8035 | conn = iter->second; | |
8036 | } | |
8037 | ||
8038 | string obj_name = dest_obj.bucket.name + "/" + dest_obj.get_oid(); | |
8039 | ||
8040 | RGWOpStateSingleOp *opstate = NULL; | |
8041 | ||
8042 | if (record_op_state) { | |
8043 | opstate = new RGWOpStateSingleOp(this, client_id, op_id, obj_name); | |
8044 | ||
8045 | ret = opstate->set_state(RGWOpState::OPSTATE_IN_PROGRESS); | |
8046 | if (ret < 0) { | |
8047 | ldout(cct, 0) << "ERROR: failed to set opstate ret=" << ret << dendl; | |
8048 | delete opstate; | |
8049 | return ret; | |
8050 | } | |
8051 | } | |
8052 | ||
8053 | boost::optional<RGWPutObj_Compress> compressor; | |
8054 | CompressorRef plugin; | |
8055 | ||
8056 | const auto& compression_type = zone_params.get_compression_type( | |
8057 | dest_bucket_info.placement_rule); | |
8058 | if (compression_type != "none") { | |
8059 | plugin = Compressor::create(cct, compression_type); | |
8060 | if (!plugin) { | |
8061 | ldout(cct, 1) << "Cannot load plugin for compression type " | |
8062 | << compression_type << dendl; | |
8063 | } | |
8064 | } | |
8065 | ||
8066 | RGWRadosPutObj cb(cct, plugin, compressor, &processor, opstate, progress_cb, progress_data); | |
8067 | ||
8068 | string etag; | |
8069 | map<string, string> req_headers; | |
8070 | real_time set_mtime; | |
8071 | ||
8072 | RGWObjState *dest_state = NULL; | |
8073 | ||
8074 | const real_time *pmod = mod_ptr; | |
8075 | ||
8076 | obj_time_weight dest_mtime_weight; | |
8077 | ||
8078 | if (copy_if_newer) { | |
8079 | /* need to get mtime for destination */ | |
8080 | ret = get_obj_state(&obj_ctx, dest_bucket_info, dest_obj, &dest_state, false); | |
8081 | if (ret < 0) | |
8082 | goto set_err_state; | |
8083 | ||
8084 | if (!real_clock::is_zero(dest_state->mtime)) { | |
8085 | dest_mtime_weight.init(dest_state); | |
8086 | pmod = &dest_mtime_weight.mtime; | |
8087 | } | |
8088 | } | |
8089 | ||
8090 | static constexpr bool prepend_meta = true; | |
8091 | static constexpr bool get_op = true; | |
8092 | static constexpr bool rgwx_stat = false; | |
8093 | static constexpr bool sync_manifest = true; | |
8094 | static constexpr bool skip_decrypt = true; | |
8095 | ret = conn->get_obj(user_id, info, src_obj, pmod, unmod_ptr, | |
8096 | dest_mtime_weight.zone_short_id, dest_mtime_weight.pg_ver, | |
8097 | prepend_meta, get_op, rgwx_stat, | |
8098 | sync_manifest, skip_decrypt, &cb, &in_stream_req); | |
8099 | if (ret < 0) { | |
8100 | goto set_err_state; | |
8101 | } | |
8102 | ||
8103 | ret = conn->complete_request(in_stream_req, etag, &set_mtime, nullptr, req_headers); | |
8104 | if (ret < 0) { | |
8105 | goto set_err_state; | |
8106 | } | |
8107 | ret = cb.flush(); | |
8108 | if (ret < 0) { | |
8109 | goto set_err_state; | |
8110 | } | |
8111 | if (compressor && compressor->is_compressed()) { | |
8112 | bufferlist tmp; | |
8113 | RGWCompressionInfo cs_info; | |
8114 | cs_info.compression_type = plugin->get_type_name(); | |
8115 | cs_info.orig_size = cb.get_data_len(); | |
8116 | cs_info.blocks = move(compressor->get_compression_blocks()); | |
8117 | ::encode(cs_info, tmp); | |
8118 | cb.get_attrs()[RGW_ATTR_COMPRESSION] = tmp; | |
8119 | } | |
8120 | ||
8121 | if (source_zone.empty()) { /* need to preserve expiration if copy in the same zonegroup */ | |
8122 | cb.get_attrs().erase(RGW_ATTR_DELETE_AT); | |
8123 | } else { | |
8124 | map<string, bufferlist>::iterator iter = cb.get_attrs().find(RGW_ATTR_DELETE_AT); | |
8125 | if (iter != cb.get_attrs().end()) { | |
8126 | try { | |
8127 | ::decode(delete_at, iter->second); | |
8128 | } catch (buffer::error& err) { | |
8129 | ldout(cct, 0) << "ERROR: failed to decode delete_at field in intra zone copy" << dendl; | |
8130 | } | |
8131 | } | |
8132 | } | |
8133 | ||
8134 | if (src_mtime) { | |
8135 | *src_mtime = set_mtime; | |
8136 | } | |
8137 | ||
8138 | if (petag) { | |
8139 | const auto iter = cb.get_attrs().find(RGW_ATTR_ETAG); | |
8140 | if (iter != cb.get_attrs().end()) { | |
8141 | *petag = iter->second; | |
8142 | } | |
8143 | } | |
8144 | ||
8145 | if (source_zone.empty()) { | |
8146 | set_copy_attrs(cb.get_attrs(), attrs, attrs_mod); | |
8147 | } else { | |
8148 | attrs = cb.get_attrs(); | |
8149 | } | |
8150 | ||
8151 | if (copy_if_newer) { | |
8152 | uint64_t pg_ver = 0; | |
8153 | auto i = attrs.find(RGW_ATTR_PG_VER); | |
8154 | if (i != attrs.end() && i->second.length() > 0) { | |
8155 | bufferlist::iterator iter = i->second.begin(); | |
8156 | try { | |
8157 | ::decode(pg_ver, iter); | |
8158 | } catch (buffer::error& err) { | |
8159 | ldout(ctx(), 0) << "ERROR: failed to decode pg ver attribute, ignoring" << dendl; | |
8160 | /* non critical error */ | |
8161 | } | |
8162 | } | |
8163 | set_mtime_weight.init(set_mtime, get_zone_short_id(), pg_ver); | |
8164 | } | |
8165 | ||
8166 | #define MAX_COMPLETE_RETRY 100 | |
8167 | for (i = 0; i < MAX_COMPLETE_RETRY; i++) { | |
8168 | ret = cb.complete(etag, mtime, set_mtime, attrs, delete_at, zones_trace); | |
8169 | if (ret < 0) { | |
8170 | goto set_err_state; | |
8171 | } | |
8172 | if (copy_if_newer && cb.is_canceled()) { | |
8173 | ldout(cct, 20) << "raced with another write of obj: " << dest_obj << dendl; | |
8174 | obj_ctx.obj.invalidate(dest_obj); /* object was overwritten */ | |
8175 | ret = get_obj_state(&obj_ctx, dest_bucket_info, dest_obj, &dest_state, false); | |
8176 | if (ret < 0) { | |
8177 | ldout(cct, 0) << "ERROR: " << __func__ << ": get_err_state() returned ret=" << ret << dendl; | |
8178 | goto set_err_state; | |
8179 | } | |
8180 | dest_mtime_weight.init(dest_state); | |
8181 | dest_mtime_weight.high_precision = high_precision_time; | |
8182 | if (!dest_state->exists || | |
8183 | dest_mtime_weight < set_mtime_weight) { | |
8184 | ldout(cct, 20) << "retrying writing object mtime=" << set_mtime << " dest_state->mtime=" << dest_state->mtime << " dest_state->exists=" << dest_state->exists << dendl; | |
8185 | continue; | |
8186 | } else { | |
8187 | ldout(cct, 20) << "not retrying writing object mtime=" << set_mtime << " dest_state->mtime=" << dest_state->mtime << " dest_state->exists=" << dest_state->exists << dendl; | |
8188 | } | |
8189 | } | |
8190 | break; | |
8191 | } | |
8192 | ||
8193 | if (i == MAX_COMPLETE_RETRY) { | |
8194 | ldout(cct, 0) << "ERROR: retried object completion too many times, something is wrong!" << dendl; | |
8195 | ret = -EIO; | |
8196 | goto set_err_state; | |
8197 | } | |
8198 | ||
8199 | if (opstate) { | |
8200 | ret = opstate->set_state(RGWOpState::OPSTATE_COMPLETE); | |
8201 | if (ret < 0) { | |
8202 | ldout(cct, 0) << "ERROR: failed to set opstate ret=" << ret << dendl; | |
8203 | } | |
8204 | delete opstate; | |
8205 | } | |
8206 | ||
8207 | return 0; | |
8208 | set_err_state: | |
8209 | if (copy_if_newer && ret == -ERR_NOT_MODIFIED) { | |
8210 | // we may have already fetched during sync of OP_ADD, but were waiting | |
8211 | // for OP_LINK_OLH to call set_olh() with a real olh_epoch | |
8212 | if (olh_epoch && *olh_epoch > 0) { | |
8213 | constexpr bool log_data_change = true; | |
8214 | ret = set_olh(obj_ctx, dest_bucket_info, dest_obj, false, nullptr, | |
8215 | *olh_epoch, real_time(), false, zones_trace, log_data_change); | |
8216 | } else { | |
8217 | // we already have the latest copy | |
8218 | ret = 0; | |
8219 | } | |
8220 | } | |
8221 | if (opstate) { | |
8222 | RGWOpState::OpState state; | |
8223 | if (ret < 0) { | |
8224 | state = RGWOpState::OPSTATE_ERROR; | |
8225 | } else { | |
8226 | state = RGWOpState::OPSTATE_COMPLETE; | |
8227 | } | |
8228 | int r = opstate->set_state(state); | |
8229 | if (r < 0) { | |
8230 | ldout(cct, 0) << "ERROR: failed to set opstate r=" << ret << dendl; | |
8231 | } | |
8232 | delete opstate; | |
8233 | } | |
8234 | return ret; | |
8235 | } | |
8236 | ||
8237 | ||
8238 | int RGWRados::copy_obj_to_remote_dest(RGWObjState *astate, | |
8239 | map<string, bufferlist>& src_attrs, | |
8240 | RGWRados::Object::Read& read_op, | |
8241 | const rgw_user& user_id, | |
8242 | rgw_obj& dest_obj, | |
8243 | real_time *mtime) | |
8244 | { | |
8245 | string etag; | |
8246 | ||
8247 | RGWRESTStreamWriteRequest *out_stream_req; | |
8248 | ||
8249 | int ret = rest_master_conn->put_obj_init(user_id, dest_obj, astate->size, src_attrs, &out_stream_req); | |
8250 | if (ret < 0) { | |
8251 | return ret; | |
8252 | } | |
8253 | ||
8254 | ret = read_op.iterate(0, astate->size - 1, out_stream_req->get_out_cb()); | |
8255 | if (ret < 0) { | |
8256 | delete out_stream_req; | |
8257 | return ret; | |
8258 | } | |
8259 | ||
8260 | ret = rest_master_conn->complete_request(out_stream_req, etag, mtime); | |
8261 | if (ret < 0) | |
8262 | return ret; | |
8263 | ||
8264 | return 0; | |
8265 | } | |
8266 | ||
8267 | /** | |
8268 | * Copy an object. | |
8269 | * dest_obj: the object to copy into | |
8270 | * src_obj: the object to copy from | |
8271 | * attrs: usage depends on attrs_mod parameter | |
8272 | * attrs_mod: the modification mode of the attrs, may have the following values: | |
8273 | * ATTRSMOD_NONE - the attributes of the source object will be | |
8274 | * copied without modifications, attrs parameter is ignored; | |
8275 | * ATTRSMOD_REPLACE - new object will have the attributes provided by attrs | |
8276 | * parameter, source object attributes are not copied; | |
8277 | * ATTRSMOD_MERGE - any conflicting meta keys on the source object's attributes | |
8278 | * are overwritten by values contained in attrs parameter. | |
8279 | * err: stores any errors resulting from the get of the original object | |
8280 | * Returns: 0 on success, -ERR# otherwise. | |
8281 | */ | |
8282 | int RGWRados::copy_obj(RGWObjectCtx& obj_ctx, | |
8283 | const rgw_user& user_id, | |
8284 | const string& client_id, | |
8285 | const string& op_id, | |
8286 | req_info *info, | |
8287 | const string& source_zone, | |
8288 | rgw_obj& dest_obj, | |
8289 | rgw_obj& src_obj, | |
8290 | RGWBucketInfo& dest_bucket_info, | |
8291 | RGWBucketInfo& src_bucket_info, | |
8292 | real_time *src_mtime, | |
8293 | real_time *mtime, | |
8294 | const real_time *mod_ptr, | |
8295 | const real_time *unmod_ptr, | |
8296 | bool high_precision_time, | |
8297 | const char *if_match, | |
8298 | const char *if_nomatch, | |
8299 | AttrsMod attrs_mod, | |
8300 | bool copy_if_newer, | |
8301 | map<string, bufferlist>& attrs, | |
8302 | RGWObjCategory category, | |
8303 | uint64_t olh_epoch, | |
8304 | real_time delete_at, | |
8305 | string *version_id, | |
8306 | string *ptag, | |
8307 | ceph::buffer::list *petag, | |
8308 | void (*progress_cb)(off_t, void *), | |
8309 | void *progress_data) | |
8310 | { | |
8311 | int ret; | |
8312 | uint64_t obj_size; | |
8313 | rgw_obj shadow_obj = dest_obj; | |
8314 | string shadow_oid; | |
8315 | ||
8316 | bool remote_src; | |
8317 | bool remote_dest; | |
8318 | ||
8319 | append_rand_alpha(cct, dest_obj.get_oid(), shadow_oid, 32); | |
8320 | shadow_obj.init_ns(dest_obj.bucket, shadow_oid, shadow_ns); | |
8321 | ||
8322 | remote_dest = !get_zonegroup().equals(dest_bucket_info.zonegroup); | |
8323 | remote_src = !get_zonegroup().equals(src_bucket_info.zonegroup); | |
8324 | ||
8325 | if (remote_src && remote_dest) { | |
8326 | ldout(cct, 0) << "ERROR: can't copy object when both src and dest buckets are remote" << dendl; | |
8327 | return -EINVAL; | |
8328 | } | |
8329 | ||
8330 | ldout(cct, 5) << "Copy object " << src_obj.bucket << ":" << src_obj.get_oid() << " => " << dest_obj.bucket << ":" << dest_obj.get_oid() << dendl; | |
8331 | ||
8332 | if (remote_src || !source_zone.empty()) { | |
8333 | return fetch_remote_obj(obj_ctx, user_id, client_id, op_id, true, info, source_zone, | |
8334 | dest_obj, src_obj, dest_bucket_info, src_bucket_info, src_mtime, mtime, mod_ptr, | |
8335 | unmod_ptr, high_precision_time, | |
8336 | if_match, if_nomatch, attrs_mod, copy_if_newer, attrs, category, | |
8337 | olh_epoch, delete_at, version_id, ptag, petag, progress_cb, progress_data); | |
8338 | } | |
8339 | ||
8340 | map<string, bufferlist> src_attrs; | |
8341 | RGWRados::Object src_op_target(this, src_bucket_info, obj_ctx, src_obj); | |
8342 | RGWRados::Object::Read read_op(&src_op_target); | |
8343 | ||
8344 | read_op.conds.mod_ptr = mod_ptr; | |
8345 | read_op.conds.unmod_ptr = unmod_ptr; | |
8346 | read_op.conds.high_precision_time = high_precision_time; | |
8347 | read_op.conds.if_match = if_match; | |
8348 | read_op.conds.if_nomatch = if_nomatch; | |
8349 | read_op.params.attrs = &src_attrs; | |
8350 | read_op.params.lastmod = src_mtime; | |
8351 | read_op.params.obj_size = &obj_size; | |
8352 | ||
8353 | ret = read_op.prepare(); | |
8354 | if (ret < 0) { | |
8355 | return ret; | |
8356 | } | |
8357 | if (src_attrs.count(RGW_ATTR_CRYPT_MODE)) { | |
8358 | // Current implementation does not follow S3 spec and even | |
8359 | // may result in data corruption silently when copying | |
8360 | // multipart objects acorss pools. So reject COPY operations | |
8361 | //on encrypted objects before it is fully functional. | |
8362 | ldout(cct, 0) << "ERROR: copy op for encrypted object " << src_obj | |
8363 | << " has not been implemented." << dendl; | |
8364 | return -ERR_NOT_IMPLEMENTED; | |
8365 | } | |
8366 | ||
8367 | src_attrs[RGW_ATTR_ACL] = attrs[RGW_ATTR_ACL]; | |
8368 | src_attrs.erase(RGW_ATTR_DELETE_AT); | |
8369 | ||
8370 | set_copy_attrs(src_attrs, attrs, attrs_mod); | |
8371 | attrs.erase(RGW_ATTR_ID_TAG); | |
8372 | attrs.erase(RGW_ATTR_PG_VER); | |
8373 | attrs.erase(RGW_ATTR_SOURCE_ZONE); | |
8374 | map<string, bufferlist>::iterator cmp = src_attrs.find(RGW_ATTR_COMPRESSION); | |
8375 | if (cmp != src_attrs.end()) | |
8376 | attrs[RGW_ATTR_COMPRESSION] = cmp->second; | |
8377 | ||
8378 | RGWObjManifest manifest; | |
8379 | RGWObjState *astate = NULL; | |
8380 | ||
8381 | ret = get_obj_state(&obj_ctx, src_bucket_info, src_obj, &astate); | |
8382 | if (ret < 0) { | |
8383 | return ret; | |
8384 | } | |
8385 | ||
8386 | vector<rgw_raw_obj> ref_objs; | |
8387 | ||
8388 | if (remote_dest) { | |
8389 | /* dest is in a different zonegroup, copy it there */ | |
8390 | return copy_obj_to_remote_dest(astate, attrs, read_op, user_id, dest_obj, mtime); | |
8391 | } | |
8392 | uint64_t max_chunk_size; | |
8393 | ||
8394 | ret = get_max_chunk_size(dest_bucket_info.placement_rule, dest_obj, &max_chunk_size); | |
8395 | if (ret < 0) { | |
8396 | ldout(cct, 0) << "ERROR: failed to get max_chunk_size() for bucket " << dest_obj.bucket << dendl; | |
8397 | return ret; | |
8398 | } | |
8399 | ||
8400 | rgw_pool src_pool; | |
8401 | rgw_pool dest_pool; | |
8402 | if (!get_obj_data_pool(src_bucket_info.placement_rule, src_obj, &src_pool)) { | |
8403 | ldout(cct, 0) << "ERROR: failed to locate data pool for " << src_obj << dendl; | |
8404 | return -EIO; | |
8405 | } | |
8406 | if (!get_obj_data_pool(dest_bucket_info.placement_rule, dest_obj, &dest_pool)) { | |
8407 | ldout(cct, 0) << "ERROR: failed to locate data pool for " << dest_obj << dendl; | |
8408 | return -EIO; | |
8409 | } | |
8410 | ||
8411 | ||
8412 | bool copy_data = !astate->has_manifest || (src_pool != dest_pool); | |
8413 | bool copy_first = false; | |
8414 | if (astate->has_manifest) { | |
8415 | if (!astate->manifest.has_tail()) { | |
8416 | copy_data = true; | |
8417 | } else { | |
8418 | uint64_t head_size = astate->manifest.get_head_size(); | |
8419 | ||
8420 | if (head_size > 0) { | |
8421 | if (head_size > max_chunk_size) { | |
8422 | copy_data = true; | |
8423 | } else { | |
8424 | copy_first = true; | |
8425 | } | |
8426 | } | |
8427 | } | |
8428 | } | |
8429 | ||
8430 | if (petag) { | |
8431 | const auto iter = attrs.find(RGW_ATTR_ETAG); | |
8432 | if (iter != attrs.end()) { | |
8433 | *petag = iter->second; | |
8434 | } | |
8435 | } | |
8436 | ||
8437 | if (copy_data) { /* refcounting tail wouldn't work here, just copy the data */ | |
8438 | return copy_obj_data(obj_ctx, dest_bucket_info, read_op, obj_size - 1, dest_obj, src_obj, | |
8439 | max_chunk_size, mtime, real_time(), attrs, category, olh_epoch, delete_at, | |
8440 | version_id, ptag, petag); | |
8441 | } | |
8442 | ||
8443 | RGWObjManifest::obj_iterator miter = astate->manifest.obj_begin(); | |
8444 | ||
8445 | if (copy_first) { // we need to copy first chunk, not increase refcount | |
8446 | ++miter; | |
8447 | } | |
8448 | ||
8449 | rgw_rados_ref ref; | |
8450 | ret = get_raw_obj_ref(miter.get_location().get_raw_obj(this), &ref); | |
8451 | if (ret < 0) { | |
8452 | return ret; | |
8453 | } | |
8454 | ||
8455 | bool versioned_dest = dest_bucket_info.versioning_enabled(); | |
8456 | ||
8457 | if (version_id && !version_id->empty()) { | |
8458 | versioned_dest = true; | |
8459 | dest_obj.key.set_instance(*version_id); | |
8460 | } else if (versioned_dest) { | |
8461 | gen_rand_obj_instance_name(&dest_obj); | |
8462 | } | |
8463 | ||
8464 | bufferlist first_chunk; | |
8465 | ||
8466 | bool copy_itself = (dest_obj == src_obj); | |
8467 | RGWObjManifest *pmanifest; | |
8468 | ldout(cct, 20) << "dest_obj=" << dest_obj << " src_obj=" << src_obj << " copy_itself=" << (int)copy_itself << dendl; | |
8469 | ||
8470 | RGWRados::Object dest_op_target(this, dest_bucket_info, obj_ctx, dest_obj); | |
8471 | RGWRados::Object::Write write_op(&dest_op_target); | |
8472 | ||
8473 | string tag; | |
8474 | ||
8475 | if (ptag) { | |
8476 | tag = *ptag; | |
8477 | } | |
8478 | ||
8479 | if (tag.empty()) { | |
8480 | append_rand_alpha(cct, tag, tag, 32); | |
8481 | } | |
8482 | ||
8483 | if (!copy_itself) { | |
8484 | attrs.erase(RGW_ATTR_TAIL_TAG); | |
8485 | manifest = astate->manifest; | |
8486 | const rgw_bucket_placement& tail_placement = manifest.get_tail_placement(); | |
8487 | if (tail_placement.bucket.name.empty()) { | |
8488 | manifest.set_tail_placement(tail_placement.placement_rule, src_obj.bucket); | |
8489 | } | |
8490 | string ref_tag; | |
8491 | for (; miter != astate->manifest.obj_end(); ++miter) { | |
8492 | ObjectWriteOperation op; | |
8493 | ref_tag = tag + '\0'; | |
8494 | cls_refcount_get(op, ref_tag, true); | |
8495 | const rgw_raw_obj& loc = miter.get_location().get_raw_obj(this); | |
8496 | ref.ioctx.locator_set_key(loc.loc); | |
8497 | ||
8498 | ret = ref.ioctx.operate(loc.oid, &op); | |
8499 | if (ret < 0) { | |
8500 | goto done_ret; | |
8501 | } | |
8502 | ||
8503 | ref_objs.push_back(loc); | |
8504 | } | |
8505 | ||
8506 | pmanifest = &manifest; | |
8507 | } else { | |
8508 | pmanifest = &astate->manifest; | |
8509 | /* don't send the object's tail for garbage collection */ | |
8510 | astate->keep_tail = true; | |
8511 | } | |
8512 | ||
8513 | if (copy_first) { | |
8514 | ret = read_op.read(0, max_chunk_size, first_chunk); | |
8515 | if (ret < 0) { | |
8516 | goto done_ret; | |
8517 | } | |
8518 | ||
8519 | pmanifest->set_head(dest_bucket_info.placement_rule, dest_obj, first_chunk.length()); | |
8520 | } else { | |
8521 | pmanifest->set_head(dest_bucket_info.placement_rule, dest_obj, 0); | |
8522 | } | |
8523 | ||
8524 | write_op.meta.data = &first_chunk; | |
8525 | write_op.meta.manifest = pmanifest; | |
8526 | write_op.meta.ptag = &tag; | |
8527 | write_op.meta.owner = dest_bucket_info.owner; | |
8528 | write_op.meta.mtime = mtime; | |
8529 | write_op.meta.flags = PUT_OBJ_CREATE; | |
8530 | write_op.meta.category = category; | |
8531 | write_op.meta.olh_epoch = olh_epoch; | |
8532 | write_op.meta.delete_at = delete_at; | |
8533 | write_op.meta.modify_tail = !copy_itself; | |
8534 | ||
8535 | ret = write_op.write_meta(obj_size, astate->accounted_size, attrs); | |
8536 | if (ret < 0) { | |
8537 | goto done_ret; | |
8538 | } | |
8539 | ||
8540 | return 0; | |
8541 | ||
8542 | done_ret: | |
8543 | if (!copy_itself) { | |
8544 | vector<rgw_raw_obj>::iterator riter; | |
8545 | ||
8546 | /* rollback reference */ | |
8547 | for (riter = ref_objs.begin(); riter != ref_objs.end(); ++riter) { | |
8548 | ObjectWriteOperation op; | |
8549 | cls_refcount_put(op, tag, true); | |
8550 | ||
8551 | ref.ioctx.locator_set_key(riter->loc); | |
8552 | ||
8553 | int r = ref.ioctx.operate(riter->oid, &op); | |
8554 | if (r < 0) { | |
8555 | ldout(cct, 0) << "ERROR: cleanup after error failed to drop reference on obj=" << *riter << dendl; | |
8556 | } | |
8557 | } | |
8558 | } | |
8559 | return ret; | |
8560 | } | |
8561 | ||
8562 | ||
8563 | int RGWRados::copy_obj_data(RGWObjectCtx& obj_ctx, | |
8564 | RGWBucketInfo& dest_bucket_info, | |
8565 | RGWRados::Object::Read& read_op, off_t end, | |
8566 | rgw_obj& dest_obj, | |
8567 | rgw_obj& src_obj, | |
8568 | uint64_t max_chunk_size, | |
8569 | real_time *mtime, | |
8570 | real_time set_mtime, | |
8571 | map<string, bufferlist>& attrs, | |
8572 | RGWObjCategory category, | |
8573 | uint64_t olh_epoch, | |
8574 | real_time delete_at, | |
8575 | string *version_id, | |
8576 | string *ptag, | |
8577 | ceph::buffer::list *petag) | |
8578 | { | |
8579 | bufferlist first_chunk; | |
8580 | RGWObjManifest manifest; | |
8581 | ||
8582 | string tag; | |
8583 | append_rand_alpha(cct, tag, tag, 32); | |
8584 | ||
8585 | RGWPutObjProcessor_Atomic processor(obj_ctx, | |
8586 | dest_bucket_info, dest_obj.bucket, dest_obj.key.name, | |
8587 | cct->_conf->rgw_obj_stripe_size, tag, dest_bucket_info.versioning_enabled()); | |
8588 | if (version_id) { | |
8589 | processor.set_version_id(*version_id); | |
8590 | } | |
8591 | processor.set_olh_epoch(olh_epoch); | |
8592 | int ret = processor.prepare(this, NULL); | |
8593 | if (ret < 0) | |
8594 | return ret; | |
8595 | ||
8596 | off_t ofs = 0; | |
8597 | ||
8598 | do { | |
8599 | bufferlist bl; | |
8600 | ret = read_op.read(ofs, end, bl); | |
8601 | ||
8602 | uint64_t read_len = ret; | |
8603 | bool again; | |
8604 | ||
8605 | do { | |
8606 | void *handle; | |
8607 | rgw_raw_obj obj; | |
8608 | ||
8609 | ret = processor.handle_data(bl, ofs, &handle, &obj, &again); | |
8610 | if (ret < 0) { | |
8611 | return ret; | |
8612 | } | |
8613 | ret = processor.throttle_data(handle, obj, read_len, false); | |
8614 | if (ret < 0) | |
8615 | return ret; | |
8616 | } while (again); | |
8617 | ||
8618 | ofs += read_len; | |
8619 | } while (ofs <= end); | |
8620 | ||
8621 | string etag; | |
8622 | auto iter = attrs.find(RGW_ATTR_ETAG); | |
8623 | if (iter != attrs.end()) { | |
8624 | bufferlist& bl = iter->second; | |
8625 | etag = string(bl.c_str(), bl.length()); | |
8626 | if (petag) { | |
8627 | *petag = bl; | |
8628 | } | |
8629 | } | |
8630 | ||
8631 | uint64_t accounted_size; | |
8632 | { | |
8633 | bool compressed{false}; | |
8634 | RGWCompressionInfo cs_info; | |
8635 | ret = rgw_compression_info_from_attrset(attrs, compressed, cs_info); | |
8636 | if (ret < 0) { | |
8637 | ldout(cct, 0) << "ERROR: failed to read compression info" << dendl; | |
8638 | return ret; | |
8639 | } | |
8640 | // pass original size if compressed | |
8641 | accounted_size = compressed ? cs_info.orig_size : ofs; | |
8642 | } | |
8643 | ||
8644 | return processor.complete(accounted_size, etag, mtime, set_mtime, attrs, delete_at); | |
8645 | } | |
8646 | ||
8647 | bool RGWRados::is_meta_master() | |
8648 | { | |
8649 | if (!get_zonegroup().is_master_zonegroup()) { | |
8650 | return false; | |
8651 | } | |
8652 | ||
8653 | return (get_zonegroup().master_zone == zone_public_config.id); | |
8654 | } | |
8655 | ||
8656 | /** | |
8657 | * Check to see if the bucket metadata could be synced | |
8658 | * bucket: the bucket to check | |
8659 | * Returns false is the bucket is not synced | |
8660 | */ | |
8661 | bool RGWRados::is_syncing_bucket_meta(const rgw_bucket& bucket) | |
8662 | { | |
8663 | ||
8664 | /* no current period */ | |
8665 | if (current_period.get_id().empty()) { | |
8666 | return false; | |
8667 | } | |
8668 | ||
8669 | /* zonegroup is not master zonegroup */ | |
8670 | if (!get_zonegroup().is_master_zonegroup()) { | |
8671 | return false; | |
8672 | } | |
8673 | ||
8674 | /* single zonegroup and a single zone */ | |
8675 | if (current_period.is_single_zonegroup() && get_zonegroup().zones.size() == 1) { | |
8676 | return false; | |
8677 | } | |
8678 | ||
8679 | /* zone is not master */ | |
8680 | if (get_zonegroup().master_zone.compare(zone_public_config.id) != 0) { | |
8681 | return false; | |
8682 | } | |
8683 | ||
8684 | return true; | |
8685 | } | |
8686 | ||
8687 | int RGWRados::check_bucket_empty(RGWBucketInfo& bucket_info) | |
8688 | { | |
8689 | std::vector<rgw_bucket_dir_entry> ent_list; | |
8690 | rgw_obj_index_key marker; | |
8691 | string prefix; | |
8692 | bool is_truncated; | |
8693 | ||
8694 | do { | |
8695 | constexpr uint NUM_ENTRIES = 1000u; | |
8696 | int r = cls_bucket_list_unordered(bucket_info, | |
8697 | RGW_NO_SHARD, | |
8698 | marker, | |
8699 | prefix, | |
8700 | NUM_ENTRIES, | |
8701 | true, | |
8702 | ent_list, | |
8703 | &is_truncated, | |
8704 | &marker); | |
8705 | if (r < 0) | |
8706 | return r; | |
8707 | ||
8708 | string ns; | |
8709 | for (auto const& dirent : ent_list) { | |
8710 | rgw_obj_key obj; | |
8711 | ||
8712 | if (rgw_obj_key::oid_to_key_in_ns(dirent.key.name, &obj, ns)) | |
8713 | return -ENOTEMPTY; | |
8714 | } | |
8715 | } while (is_truncated); | |
8716 | ||
8717 | return 0; | |
8718 | } | |
8719 | ||
8720 | /** | |
8721 | * Delete a bucket. | |
8722 | * bucket: the name of the bucket to delete | |
8723 | * Returns 0 on success, -ERR# otherwise. | |
8724 | */ | |
8725 | int RGWRados::delete_bucket(RGWBucketInfo& bucket_info, RGWObjVersionTracker& objv_tracker, bool check_empty) | |
8726 | { | |
8727 | const rgw_bucket& bucket = bucket_info.bucket; | |
8728 | librados::IoCtx index_ctx; | |
8729 | map<int, string> bucket_objs; | |
8730 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs); | |
8731 | if (r < 0) | |
8732 | return r; | |
8733 | ||
8734 | if (check_empty) { | |
8735 | r = check_bucket_empty(bucket_info); | |
8736 | if (r < 0) { | |
8737 | return r; | |
8738 | } | |
8739 | } | |
8740 | ||
8741 | r = rgw_bucket_delete_bucket_obj(this, bucket.tenant, bucket.name, objv_tracker); | |
8742 | if (r < 0) | |
8743 | return r; | |
8744 | ||
8745 | /* if the bucket is not synced we can remove the meta file */ | |
8746 | if (!is_syncing_bucket_meta(bucket)) { | |
8747 | RGWObjVersionTracker objv_tracker; | |
8748 | r = rgw_bucket_instance_remove_entry(this, bucket.get_key(), &objv_tracker); | |
8749 | if (r < 0) { | |
8750 | return r; | |
8751 | } | |
8752 | ||
8753 | /* remove bucket index objects asynchronously by best effort */ | |
8754 | (void) CLSRGWIssueBucketIndexClean(index_ctx, | |
8755 | bucket_objs, | |
8756 | cct->_conf->rgw_bucket_index_max_aio)(); | |
8757 | } | |
8758 | ||
8759 | return 0; | |
8760 | } | |
8761 | ||
8762 | int RGWRados::set_bucket_owner(rgw_bucket& bucket, ACLOwner& owner) | |
8763 | { | |
8764 | RGWBucketInfo info; | |
8765 | map<string, bufferlist> attrs; | |
8766 | RGWObjectCtx obj_ctx(this); | |
8767 | int r; | |
8768 | if (bucket.bucket_id.empty()) { | |
8769 | r = get_bucket_info(obj_ctx, bucket.tenant, bucket.name, info, NULL, &attrs); | |
8770 | } else { | |
8771 | r = get_bucket_instance_info(obj_ctx, bucket, info, nullptr, &attrs); | |
8772 | } | |
8773 | if (r < 0) { | |
8774 | ldout(cct, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl; | |
8775 | return r; | |
8776 | } | |
8777 | ||
8778 | info.owner = owner.get_id(); | |
8779 | ||
8780 | r = put_bucket_instance_info(info, false, real_time(), &attrs); | |
8781 | if (r < 0) { | |
8782 | ldout(cct, 0) << "NOTICE: put_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl; | |
8783 | return r; | |
8784 | } | |
8785 | ||
8786 | return 0; | |
8787 | } | |
8788 | ||
8789 | ||
8790 | int RGWRados::set_buckets_enabled(vector<rgw_bucket>& buckets, bool enabled) | |
8791 | { | |
8792 | int ret = 0; | |
8793 | ||
8794 | vector<rgw_bucket>::iterator iter; | |
8795 | ||
8796 | for (iter = buckets.begin(); iter != buckets.end(); ++iter) { | |
8797 | rgw_bucket& bucket = *iter; | |
8798 | if (enabled) | |
8799 | ldout(cct, 20) << "enabling bucket name=" << bucket.name << dendl; | |
8800 | else | |
8801 | ldout(cct, 20) << "disabling bucket name=" << bucket.name << dendl; | |
8802 | ||
8803 | RGWBucketInfo info; | |
8804 | map<string, bufferlist> attrs; | |
8805 | RGWObjectCtx obj_ctx(this); | |
8806 | int r = get_bucket_info(obj_ctx, bucket.tenant, bucket.name, info, NULL, &attrs); | |
8807 | if (r < 0) { | |
8808 | ldout(cct, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl; | |
8809 | ret = r; | |
8810 | continue; | |
8811 | } | |
8812 | if (enabled) { | |
8813 | info.flags &= ~BUCKET_SUSPENDED; | |
8814 | } else { | |
8815 | info.flags |= BUCKET_SUSPENDED; | |
8816 | } | |
8817 | ||
8818 | r = put_bucket_instance_info(info, false, real_time(), &attrs); | |
8819 | if (r < 0) { | |
8820 | ldout(cct, 0) << "NOTICE: put_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl; | |
8821 | ret = r; | |
8822 | continue; | |
8823 | } | |
8824 | } | |
8825 | return ret; | |
8826 | } | |
8827 | ||
8828 | int RGWRados::bucket_suspended(rgw_bucket& bucket, bool *suspended) | |
8829 | { | |
8830 | RGWBucketInfo bucket_info; | |
8831 | RGWObjectCtx obj_ctx(this); | |
8832 | int ret = get_bucket_info(obj_ctx, bucket.tenant, bucket.name, bucket_info, NULL); | |
8833 | if (ret < 0) { | |
8834 | return ret; | |
8835 | } | |
8836 | ||
8837 | *suspended = ((bucket_info.flags & BUCKET_SUSPENDED) != 0); | |
8838 | return 0; | |
8839 | } | |
8840 | ||
8841 | int RGWRados::Object::complete_atomic_modification() | |
8842 | { | |
8843 | if (!state->has_manifest || state->keep_tail) | |
8844 | return 0; | |
8845 | ||
8846 | cls_rgw_obj_chain chain; | |
8847 | store->update_gc_chain(obj, state->manifest, &chain); | |
8848 | ||
8849 | if (chain.empty()) { | |
8850 | return 0; | |
8851 | } | |
8852 | ||
8853 | string tag = (state->tail_tag.length() > 0 ? state->tail_tag.to_str() : state->obj_tag.to_str()); | |
8854 | return store->gc->send_chain(chain, tag, false); // do it async | |
8855 | } | |
8856 | ||
8857 | void RGWRados::update_gc_chain(rgw_obj& head_obj, RGWObjManifest& manifest, cls_rgw_obj_chain *chain) | |
8858 | { | |
8859 | RGWObjManifest::obj_iterator iter; | |
8860 | rgw_raw_obj raw_head; | |
8861 | obj_to_raw(manifest.get_head_placement_rule(), head_obj, &raw_head); | |
8862 | for (iter = manifest.obj_begin(); iter != manifest.obj_end(); ++iter) { | |
8863 | const rgw_raw_obj& mobj = iter.get_location().get_raw_obj(this); | |
8864 | if (mobj == raw_head) | |
8865 | continue; | |
8866 | cls_rgw_obj_key key(mobj.oid); | |
8867 | chain->push_obj(mobj.pool.to_str(), key, mobj.loc); | |
8868 | } | |
8869 | } | |
8870 | ||
8871 | int RGWRados::send_chain_to_gc(cls_rgw_obj_chain& chain, const string& tag, bool sync) | |
8872 | { | |
8873 | return gc->send_chain(chain, tag, sync); | |
8874 | } | |
8875 | ||
8876 | int RGWRados::open_bucket_index(const RGWBucketInfo& bucket_info, | |
8877 | librados::IoCtx& index_ctx, | |
8878 | string& bucket_oid) | |
8879 | { | |
8880 | const rgw_bucket& bucket = bucket_info.bucket; | |
8881 | int r = open_bucket_index_ctx(bucket_info, index_ctx); | |
8882 | if (r < 0) | |
8883 | return r; | |
8884 | ||
8885 | if (bucket.bucket_id.empty()) { | |
8886 | ldout(cct, 0) << "ERROR: empty bucket id for bucket operation" << dendl; | |
8887 | return -EIO; | |
8888 | } | |
8889 | ||
8890 | bucket_oid = dir_oid_prefix; | |
8891 | bucket_oid.append(bucket.bucket_id); | |
8892 | ||
8893 | return 0; | |
8894 | } | |
8895 | ||
8896 | int RGWRados::open_bucket_index_base(const RGWBucketInfo& bucket_info, | |
8897 | librados::IoCtx& index_ctx, | |
8898 | string& bucket_oid_base) { | |
8899 | const rgw_bucket& bucket = bucket_info.bucket; | |
8900 | int r = open_bucket_index_ctx(bucket_info, index_ctx); | |
8901 | if (r < 0) | |
8902 | return r; | |
8903 | ||
8904 | if (bucket.bucket_id.empty()) { | |
8905 | ldout(cct, 0) << "ERROR: empty bucket_id for bucket operation" << dendl; | |
8906 | return -EIO; | |
8907 | } | |
8908 | ||
8909 | bucket_oid_base = dir_oid_prefix; | |
8910 | bucket_oid_base.append(bucket.bucket_id); | |
8911 | ||
8912 | return 0; | |
8913 | ||
8914 | } | |
8915 | ||
8916 | int RGWRados::open_bucket_index(const RGWBucketInfo& bucket_info, | |
8917 | librados::IoCtx& index_ctx, | |
8918 | map<int, string>& bucket_objs, | |
8919 | int shard_id, | |
8920 | map<int, string> *bucket_instance_ids) { | |
8921 | string bucket_oid_base; | |
8922 | int ret = open_bucket_index_base(bucket_info, index_ctx, bucket_oid_base); | |
8923 | if (ret < 0) { | |
8924 | return ret; | |
8925 | } | |
8926 | ||
8927 | get_bucket_index_objects(bucket_oid_base, bucket_info.num_shards, bucket_objs, shard_id); | |
8928 | if (bucket_instance_ids) { | |
8929 | get_bucket_instance_ids(bucket_info, shard_id, bucket_instance_ids); | |
8930 | } | |
8931 | return 0; | |
8932 | } | |
8933 | ||
8934 | template<typename T> | |
8935 | int RGWRados::open_bucket_index(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, | |
8936 | map<int, string>& oids, map<int, T>& bucket_objs, | |
8937 | int shard_id, map<int, string> *bucket_instance_ids) | |
8938 | { | |
8939 | int ret = open_bucket_index(bucket_info, index_ctx, oids, shard_id, bucket_instance_ids); | |
8940 | if (ret < 0) | |
8941 | return ret; | |
8942 | ||
8943 | map<int, string>::const_iterator iter = oids.begin(); | |
8944 | for (; iter != oids.end(); ++iter) { | |
8945 | bucket_objs[iter->first] = T(); | |
8946 | } | |
8947 | return 0; | |
8948 | } | |
8949 | ||
8950 | int RGWRados::open_bucket_index_shard(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, | |
8951 | const string& obj_key, string *bucket_obj, int *shard_id) | |
8952 | { | |
8953 | string bucket_oid_base; | |
8954 | int ret = open_bucket_index_base(bucket_info, index_ctx, bucket_oid_base); | |
8955 | if (ret < 0) | |
8956 | return ret; | |
8957 | ||
8958 | RGWObjectCtx obj_ctx(this); | |
8959 | ||
8960 | ret = get_bucket_index_object(bucket_oid_base, obj_key, bucket_info.num_shards, | |
8961 | (RGWBucketInfo::BIShardsHashType)bucket_info.bucket_index_shard_hash_type, bucket_obj, shard_id); | |
8962 | if (ret < 0) { | |
8963 | ldout(cct, 10) << "get_bucket_index_object() returned ret=" << ret << dendl; | |
8964 | return ret; | |
8965 | } | |
8966 | return 0; | |
8967 | } | |
8968 | ||
8969 | int RGWRados::open_bucket_index_shard(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, | |
8970 | int shard_id, string *bucket_obj) | |
8971 | { | |
8972 | string bucket_oid_base; | |
8973 | int ret = open_bucket_index_base(bucket_info, index_ctx, bucket_oid_base); | |
8974 | if (ret < 0) | |
8975 | return ret; | |
8976 | ||
8977 | RGWObjectCtx obj_ctx(this); | |
8978 | ||
8979 | get_bucket_index_object(bucket_oid_base, bucket_info.num_shards, | |
8980 | shard_id, bucket_obj); | |
8981 | return 0; | |
8982 | } | |
8983 | ||
8984 | static void accumulate_raw_stats(const rgw_bucket_dir_header& header, | |
8985 | map<RGWObjCategory, RGWStorageStats>& stats) | |
8986 | { | |
8987 | for (const auto& pair : header.stats) { | |
8988 | const RGWObjCategory category = static_cast<RGWObjCategory>(pair.first); | |
8989 | const rgw_bucket_category_stats& header_stats = pair.second; | |
8990 | ||
8991 | RGWStorageStats& s = stats[category]; | |
8992 | ||
8993 | s.category = category; | |
8994 | s.size += header_stats.total_size; | |
8995 | s.size_rounded += header_stats.total_size_rounded; | |
8996 | s.size_utilized += header_stats.actual_size; | |
8997 | s.num_objects += header_stats.num_entries; | |
8998 | } | |
8999 | } | |
9000 | ||
9001 | int RGWRados::bucket_check_index(RGWBucketInfo& bucket_info, | |
9002 | map<RGWObjCategory, RGWStorageStats> *existing_stats, | |
9003 | map<RGWObjCategory, RGWStorageStats> *calculated_stats) | |
9004 | { | |
9005 | librados::IoCtx index_ctx; | |
9006 | // key - bucket index object id | |
9007 | // value - bucket index check OP returned result with the given bucket index object (shard) | |
9008 | map<int, string> oids; | |
9009 | map<int, struct rgw_cls_check_index_ret> bucket_objs_ret; | |
9010 | ||
9011 | int ret = open_bucket_index(bucket_info, index_ctx, oids, bucket_objs_ret); | |
9012 | if (ret < 0) { | |
9013 | return ret; | |
9014 | } | |
9015 | ||
9016 | ret = CLSRGWIssueBucketCheck(index_ctx, oids, bucket_objs_ret, cct->_conf->rgw_bucket_index_max_aio)(); | |
9017 | if (ret < 0) { | |
9018 | return ret; | |
9019 | } | |
9020 | ||
9021 | // Aggregate results (from different shards if there is any) | |
9022 | map<int, struct rgw_cls_check_index_ret>::iterator iter; | |
9023 | for (iter = bucket_objs_ret.begin(); iter != bucket_objs_ret.end(); ++iter) { | |
9024 | accumulate_raw_stats(iter->second.existing_header, *existing_stats); | |
9025 | accumulate_raw_stats(iter->second.calculated_header, *calculated_stats); | |
9026 | } | |
9027 | ||
9028 | return 0; | |
9029 | } | |
9030 | ||
9031 | int RGWRados::bucket_rebuild_index(RGWBucketInfo& bucket_info) | |
9032 | { | |
9033 | librados::IoCtx index_ctx; | |
9034 | map<int, string> bucket_objs; | |
9035 | ||
9036 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs); | |
9037 | if (r < 0) { | |
9038 | return r; | |
9039 | } | |
9040 | ||
9041 | return CLSRGWIssueBucketRebuild(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); | |
9042 | } | |
9043 | ||
9044 | int RGWRados::bucket_set_reshard(const RGWBucketInfo& bucket_info, const cls_rgw_bucket_instance_entry& entry) | |
9045 | { | |
9046 | librados::IoCtx index_ctx; | |
9047 | map<int, string> bucket_objs; | |
9048 | ||
9049 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs); | |
9050 | if (r < 0) { | |
9051 | return r; | |
9052 | } | |
9053 | ||
9054 | return CLSRGWIssueSetBucketResharding(index_ctx, bucket_objs, entry, cct->_conf->rgw_bucket_index_max_aio)(); | |
9055 | } | |
9056 | ||
9057 | int RGWRados::defer_gc(void *ctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj) | |
9058 | { | |
9059 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
9060 | std::string oid, key; | |
9061 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
9062 | if (!rctx) | |
9063 | return 0; | |
9064 | ||
9065 | RGWObjState *state = NULL; | |
9066 | ||
9067 | int r = get_obj_state(rctx, bucket_info, obj, &state, false); | |
9068 | if (r < 0) | |
9069 | return r; | |
9070 | ||
9071 | if (!state->is_atomic) { | |
9072 | ldout(cct, 20) << "state for obj=" << obj << " is not atomic, not deferring gc operation" << dendl; | |
9073 | return -EINVAL; | |
9074 | } | |
9075 | ||
9076 | string tag; | |
9077 | ||
9078 | if (state->tail_tag.length() > 0) { | |
9079 | tag = state->tail_tag.c_str(); | |
9080 | } else if (state->obj_tag.length() > 0) { | |
9081 | tag = state->obj_tag.c_str(); | |
9082 | } else { | |
9083 | ldout(cct, 20) << "state->obj_tag is empty, not deferring gc operation" << dendl; | |
9084 | return -EINVAL; | |
9085 | } | |
9086 | ||
9087 | ldout(cct, 0) << "defer chain tag=" << tag << dendl; | |
9088 | ||
9089 | return gc->defer_chain(tag, false); | |
9090 | } | |
9091 | ||
9092 | void RGWRados::remove_rgw_head_obj(ObjectWriteOperation& op) | |
9093 | { | |
9094 | list<string> prefixes; | |
9095 | prefixes.push_back(RGW_ATTR_OLH_PREFIX); | |
9096 | cls_rgw_remove_obj(op, prefixes); | |
9097 | } | |
9098 | ||
9099 | void RGWRados::cls_obj_check_prefix_exist(ObjectOperation& op, const string& prefix, bool fail_if_exist) | |
9100 | { | |
9101 | cls_rgw_obj_check_attrs_prefix(op, prefix, fail_if_exist); | |
9102 | } | |
9103 | ||
9104 | void RGWRados::cls_obj_check_mtime(ObjectOperation& op, const real_time& mtime, bool high_precision_time, RGWCheckMTimeType type) | |
9105 | { | |
9106 | cls_rgw_obj_check_mtime(op, mtime, high_precision_time, type); | |
9107 | } | |
9108 | ||
9109 | ||
9110 | /** | |
9111 | * Delete an object. | |
9112 | * bucket: name of the bucket storing the object | |
9113 | * obj: name of the object to delete | |
9114 | * Returns: 0 on success, -ERR# otherwise. | |
9115 | */ | |
9116 | int RGWRados::Object::Delete::delete_obj() | |
9117 | { | |
9118 | RGWRados *store = target->get_store(); | |
9119 | rgw_obj& src_obj = target->get_obj(); | |
9120 | const string& instance = src_obj.key.instance; | |
9121 | rgw_obj obj = src_obj; | |
9122 | ||
9123 | if (instance == "null") { | |
9124 | obj.key.instance.clear(); | |
9125 | } | |
9126 | ||
9127 | bool explicit_marker_version = (!params.marker_version_id.empty()); | |
9128 | ||
9129 | if (params.versioning_status & BUCKET_VERSIONED || explicit_marker_version) { | |
9130 | if (instance.empty() || explicit_marker_version) { | |
9131 | rgw_obj marker = obj; | |
9132 | ||
9133 | if (!params.marker_version_id.empty()) { | |
9134 | if (params.marker_version_id != "null") { | |
9135 | marker.key.set_instance(params.marker_version_id); | |
9136 | } | |
9137 | } else if ((params.versioning_status & BUCKET_VERSIONS_SUSPENDED) == 0) { | |
9138 | store->gen_rand_obj_instance_name(&marker); | |
9139 | } | |
9140 | ||
9141 | result.version_id = marker.key.instance; | |
9142 | if (result.version_id.empty()) | |
9143 | result.version_id = "null"; | |
9144 | result.delete_marker = true; | |
9145 | ||
9146 | struct rgw_bucket_dir_entry_meta meta; | |
9147 | ||
9148 | meta.owner = params.obj_owner.get_id().to_str(); | |
9149 | meta.owner_display_name = params.obj_owner.get_display_name(); | |
9150 | ||
9151 | if (real_clock::is_zero(params.mtime)) { | |
9152 | meta.mtime = real_clock::now(); | |
9153 | } else { | |
9154 | meta.mtime = params.mtime; | |
9155 | } | |
9156 | ||
9157 | int r = store->set_olh(target->get_ctx(), target->get_bucket_info(), marker, true, &meta, params.olh_epoch, params.unmod_since, params.high_precision_time, params.zones_trace); | |
9158 | if (r < 0) { | |
9159 | return r; | |
9160 | } | |
9161 | } else { | |
9162 | rgw_bucket_dir_entry dirent; | |
9163 | ||
9164 | int r = store->bi_get_instance(target->get_bucket_info(), obj, &dirent); | |
9165 | if (r < 0) { | |
9166 | return r; | |
9167 | } | |
9168 | result.delete_marker = dirent.is_delete_marker(); | |
9169 | r = store->unlink_obj_instance(target->get_ctx(), target->get_bucket_info(), obj, params.olh_epoch, params.zones_trace); | |
9170 | if (r < 0) { | |
9171 | return r; | |
9172 | } | |
9173 | result.version_id = instance; | |
9174 | } | |
9175 | ||
9176 | BucketShard *bs; | |
9177 | int r = target->get_bucket_shard(&bs); | |
9178 | if (r < 0) { | |
9179 | ldout(store->ctx(), 5) << "failed to get BucketShard object: r=" << r << dendl; | |
9180 | return r; | |
9181 | } | |
9182 | ||
9183 | if (target->bucket_info.datasync_flag_enabled()) { | |
9184 | r = store->data_log->add_entry(bs->bucket, bs->shard_id); | |
9185 | if (r < 0) { | |
9186 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
9187 | return r; | |
9188 | } | |
9189 | } | |
9190 | ||
9191 | return 0; | |
9192 | } | |
9193 | ||
9194 | rgw_rados_ref ref; | |
9195 | int r = store->get_obj_head_ref(target->get_bucket_info(), obj, &ref); | |
9196 | if (r < 0) { | |
9197 | return r; | |
9198 | } | |
9199 | ||
9200 | RGWObjState *state; | |
9201 | r = target->get_state(&state, false); | |
9202 | if (r < 0) | |
9203 | return r; | |
9204 | ||
9205 | ObjectWriteOperation op; | |
9206 | ||
9207 | if (!real_clock::is_zero(params.unmod_since)) { | |
9208 | struct timespec ctime = ceph::real_clock::to_timespec(state->mtime); | |
9209 | struct timespec unmod = ceph::real_clock::to_timespec(params.unmod_since); | |
9210 | if (!params.high_precision_time) { | |
9211 | ctime.tv_nsec = 0; | |
9212 | unmod.tv_nsec = 0; | |
9213 | } | |
9214 | ||
9215 | ldout(store->ctx(), 10) << "If-UnModified-Since: " << params.unmod_since << " Last-Modified: " << ctime << dendl; | |
9216 | if (ctime > unmod) { | |
9217 | return -ERR_PRECONDITION_FAILED; | |
9218 | } | |
9219 | ||
9220 | /* only delete object if mtime is less than or equal to params.unmod_since */ | |
9221 | store->cls_obj_check_mtime(op, params.unmod_since, params.high_precision_time, CLS_RGW_CHECK_TIME_MTIME_LE); | |
9222 | } | |
9223 | uint64_t obj_size = state->size; | |
9224 | ||
9225 | if (!real_clock::is_zero(params.expiration_time)) { | |
9226 | bufferlist bl; | |
9227 | real_time delete_at; | |
9228 | ||
9229 | if (state->get_attr(RGW_ATTR_DELETE_AT, bl)) { | |
9230 | try { | |
9231 | bufferlist::iterator iter = bl.begin(); | |
9232 | ::decode(delete_at, iter); | |
9233 | } catch (buffer::error& err) { | |
9234 | ldout(store->ctx(), 0) << "ERROR: couldn't decode RGW_ATTR_DELETE_AT" << dendl; | |
9235 | return -EIO; | |
9236 | } | |
9237 | ||
9238 | if (params.expiration_time != delete_at) { | |
9239 | return -ERR_PRECONDITION_FAILED; | |
9240 | } | |
9241 | } else { | |
9242 | return -ERR_PRECONDITION_FAILED; | |
9243 | } | |
9244 | } | |
9245 | ||
9246 | if (!state->exists) { | |
9247 | target->invalidate_state(); | |
9248 | return -ENOENT; | |
9249 | } | |
9250 | ||
9251 | r = target->prepare_atomic_modification(op, false, NULL, NULL, NULL, true, false); | |
9252 | if (r < 0) | |
9253 | return r; | |
9254 | ||
9255 | RGWBucketInfo& bucket_info = target->get_bucket_info(); | |
9256 | ||
9257 | RGWRados::Bucket bop(store, bucket_info); | |
9258 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
9259 | ||
9260 | index_op.set_zones_trace(params.zones_trace); | |
9261 | index_op.set_bilog_flags(params.bilog_flags); | |
9262 | ||
9263 | r = index_op.prepare(CLS_RGW_OP_DEL, &state->write_tag); | |
9264 | if (r < 0) | |
9265 | return r; | |
9266 | ||
9267 | store->remove_rgw_head_obj(op); | |
9268 | r = ref.ioctx.operate(ref.oid, &op); | |
9269 | ||
9270 | /* raced with another operation, object state is indeterminate */ | |
9271 | const bool need_invalidate = (r == -ECANCELED); | |
9272 | ||
9273 | int64_t poolid = ref.ioctx.get_id(); | |
9274 | if (r >= 0) { | |
9275 | tombstone_cache_t *obj_tombstone_cache = store->get_tombstone_cache(); | |
9276 | if (obj_tombstone_cache) { | |
9277 | tombstone_entry entry{*state}; | |
9278 | obj_tombstone_cache->add(obj, entry); | |
9279 | } | |
9280 | r = index_op.complete_del(poolid, ref.ioctx.get_last_version(), state->mtime, params.remove_objs); | |
9281 | ||
9282 | int ret = target->complete_atomic_modification(); | |
9283 | if (ret < 0) { | |
9284 | ldout(store->ctx(), 0) << "ERROR: complete_atomic_modification returned ret=" << ret << dendl; | |
9285 | } | |
9286 | /* other than that, no need to propagate error */ | |
9287 | } else { | |
9288 | int ret = index_op.cancel(); | |
9289 | if (ret < 0) { | |
9290 | ldout(store->ctx(), 0) << "ERROR: index_op.cancel() returned ret=" << ret << dendl; | |
9291 | } | |
9292 | } | |
9293 | ||
9294 | if (need_invalidate) { | |
9295 | target->invalidate_state(); | |
9296 | } | |
9297 | ||
9298 | if (r < 0) | |
9299 | return r; | |
9300 | ||
9301 | /* update quota cache */ | |
9302 | store->quota_handler->update_stats(params.bucket_owner, obj.bucket, -1, 0, obj_size); | |
9303 | ||
9304 | return 0; | |
9305 | } | |
9306 | ||
9307 | int RGWRados::delete_obj(RGWObjectCtx& obj_ctx, | |
9308 | const RGWBucketInfo& bucket_info, | |
9309 | const rgw_obj& obj, | |
9310 | int versioning_status, | |
9311 | uint16_t bilog_flags, | |
9312 | const real_time& expiration_time, | |
9313 | rgw_zone_set *zones_trace) | |
9314 | { | |
9315 | RGWRados::Object del_target(this, bucket_info, obj_ctx, obj); | |
9316 | RGWRados::Object::Delete del_op(&del_target); | |
9317 | ||
9318 | del_op.params.bucket_owner = bucket_info.owner; | |
9319 | del_op.params.versioning_status = versioning_status; | |
9320 | del_op.params.bilog_flags = bilog_flags; | |
9321 | del_op.params.expiration_time = expiration_time; | |
9322 | del_op.params.zones_trace = zones_trace; | |
9323 | ||
9324 | return del_op.delete_obj(); | |
9325 | } | |
9326 | ||
9327 | int RGWRados::delete_raw_obj(const rgw_raw_obj& obj) | |
9328 | { | |
9329 | rgw_rados_ref ref; | |
9330 | int r = get_raw_obj_ref(obj, &ref); | |
9331 | if (r < 0) { | |
9332 | return r; | |
9333 | } | |
9334 | ||
9335 | ObjectWriteOperation op; | |
9336 | ||
9337 | op.remove(); | |
9338 | r = ref.ioctx.operate(ref.oid, &op); | |
9339 | if (r < 0) | |
9340 | return r; | |
9341 | ||
9342 | return 0; | |
9343 | } | |
9344 | ||
9345 | int RGWRados::delete_system_obj(rgw_raw_obj& obj, RGWObjVersionTracker *objv_tracker) | |
9346 | { | |
9347 | if (obj.empty()) { | |
9348 | ldout(cct, 1) << "delete_system_obj got empty object name " | |
9349 | << obj << ", returning EINVAL" << dendl; | |
9350 | return -EINVAL; | |
9351 | } | |
9352 | rgw_rados_ref ref; | |
9353 | int r = get_raw_obj_ref(obj, &ref); | |
9354 | if (r < 0) { | |
9355 | return r; | |
9356 | } | |
9357 | ||
9358 | ObjectWriteOperation op; | |
9359 | ||
9360 | if (objv_tracker) { | |
9361 | objv_tracker->prepare_op_for_write(&op); | |
9362 | } | |
9363 | ||
9364 | op.remove(); | |
9365 | r = ref.ioctx.operate(ref.oid, &op); | |
9366 | if (r < 0) | |
9367 | return r; | |
9368 | ||
9369 | return 0; | |
9370 | } | |
9371 | ||
9372 | int RGWRados::delete_obj_index(const rgw_obj& obj) | |
9373 | { | |
9374 | std::string oid, key; | |
9375 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
9376 | ||
9377 | RGWObjectCtx obj_ctx(this); | |
9378 | ||
9379 | RGWBucketInfo bucket_info; | |
9380 | int ret = get_bucket_instance_info(obj_ctx, obj.bucket, bucket_info, NULL, NULL); | |
9381 | if (ret < 0) { | |
9382 | ldout(cct, 0) << "ERROR: " << __func__ << "() get_bucket_instance_info(bucket=" << obj.bucket << ") returned ret=" << ret << dendl; | |
9383 | return ret; | |
9384 | } | |
9385 | ||
9386 | RGWRados::Bucket bop(this, bucket_info); | |
9387 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
9388 | ||
9389 | real_time removed_mtime; | |
9390 | int r = index_op.complete_del(-1 /* pool */, 0, removed_mtime, NULL); | |
9391 | ||
9392 | return r; | |
9393 | } | |
9394 | ||
9395 | static void generate_fake_tag(RGWRados *store, map<string, bufferlist>& attrset, RGWObjManifest& manifest, bufferlist& manifest_bl, bufferlist& tag_bl) | |
9396 | { | |
9397 | string tag; | |
9398 | ||
9399 | RGWObjManifest::obj_iterator mi = manifest.obj_begin(); | |
9400 | if (mi != manifest.obj_end()) { | |
9401 | if (manifest.has_tail()) // first object usually points at the head, let's skip to a more unique part | |
9402 | ++mi; | |
9403 | tag = mi.get_location().get_raw_obj(store).oid; | |
9404 | tag.append("_"); | |
9405 | } | |
9406 | ||
9407 | unsigned char md5[CEPH_CRYPTO_MD5_DIGESTSIZE]; | |
9408 | char md5_str[CEPH_CRYPTO_MD5_DIGESTSIZE * 2 + 1]; | |
9409 | MD5 hash; | |
9410 | hash.Update((const byte *)manifest_bl.c_str(), manifest_bl.length()); | |
9411 | ||
9412 | map<string, bufferlist>::iterator iter = attrset.find(RGW_ATTR_ETAG); | |
9413 | if (iter != attrset.end()) { | |
9414 | bufferlist& bl = iter->second; | |
9415 | hash.Update((const byte *)bl.c_str(), bl.length()); | |
9416 | } | |
9417 | ||
9418 | hash.Final(md5); | |
9419 | buf_to_hex(md5, CEPH_CRYPTO_MD5_DIGESTSIZE, md5_str); | |
9420 | tag.append(md5_str); | |
9421 | ||
9422 | ldout(store->ctx(), 10) << "generate_fake_tag new tag=" << tag << dendl; | |
9423 | ||
9424 | tag_bl.append(tag.c_str(), tag.size() + 1); | |
9425 | } | |
9426 | ||
9427 | static bool is_olh(map<string, bufferlist>& attrs) | |
9428 | { | |
9429 | map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_OLH_INFO); | |
9430 | return (iter != attrs.end()); | |
9431 | } | |
9432 | ||
9433 | static bool has_olh_tag(map<string, bufferlist>& attrs) | |
9434 | { | |
9435 | map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_OLH_ID_TAG); | |
9436 | return (iter != attrs.end()); | |
9437 | } | |
9438 | ||
9439 | int RGWRados::get_olh_target_state(RGWObjectCtx& obj_ctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
9440 | RGWObjState *olh_state, RGWObjState **target_state) | |
9441 | { | |
9442 | assert(olh_state->is_olh); | |
9443 | ||
9444 | rgw_obj target; | |
9445 | int r = RGWRados::follow_olh(bucket_info, obj_ctx, olh_state, obj, &target); /* might return -EAGAIN */ | |
9446 | if (r < 0) { | |
9447 | return r; | |
9448 | } | |
9449 | r = get_obj_state(&obj_ctx, bucket_info, target, target_state, false); | |
9450 | if (r < 0) { | |
9451 | return r; | |
9452 | } | |
9453 | ||
9454 | return 0; | |
9455 | } | |
9456 | ||
9457 | int RGWRados::get_system_obj_state_impl(RGWObjectCtx *rctx, rgw_raw_obj& obj, RGWRawObjState **state, RGWObjVersionTracker *objv_tracker) | |
9458 | { | |
9459 | if (obj.empty()) { | |
9460 | return -EINVAL; | |
9461 | } | |
9462 | ||
9463 | RGWRawObjState *s = rctx->raw.get_state(obj); | |
9464 | ldout(cct, 20) << "get_system_obj_state: rctx=" << (void *)rctx << " obj=" << obj << " state=" << (void *)s << " s->prefetch_data=" << s->prefetch_data << dendl; | |
9465 | *state = s; | |
9466 | if (s->has_attrs) { | |
9467 | return 0; | |
9468 | } | |
9469 | ||
9470 | s->obj = obj; | |
9471 | ||
9472 | int r = raw_obj_stat(obj, &s->size, &s->mtime, &s->epoch, &s->attrset, (s->prefetch_data ? &s->data : NULL), objv_tracker); | |
9473 | if (r == -ENOENT) { | |
9474 | s->exists = false; | |
9475 | s->has_attrs = true; | |
9476 | s->mtime = real_time(); | |
9477 | return 0; | |
9478 | } | |
9479 | if (r < 0) | |
9480 | return r; | |
9481 | ||
9482 | s->exists = true; | |
9483 | s->has_attrs = true; | |
9484 | s->obj_tag = s->attrset[RGW_ATTR_ID_TAG]; | |
9485 | ||
9486 | if (s->obj_tag.length()) | |
9487 | ldout(cct, 20) << "get_system_obj_state: setting s->obj_tag to " | |
9488 | << s->obj_tag.c_str() << dendl; | |
9489 | else | |
9490 | ldout(cct, 20) << "get_system_obj_state: s->obj_tag was set empty" << dendl; | |
9491 | ||
9492 | return 0; | |
9493 | } | |
9494 | ||
9495 | int RGWRados::get_system_obj_state(RGWObjectCtx *rctx, rgw_raw_obj& obj, RGWRawObjState **state, RGWObjVersionTracker *objv_tracker) | |
9496 | { | |
9497 | int ret; | |
9498 | ||
9499 | do { | |
9500 | ret = get_system_obj_state_impl(rctx, obj, state, objv_tracker); | |
9501 | } while (ret == -EAGAIN); | |
9502 | ||
9503 | return ret; | |
9504 | } | |
9505 | ||
9506 | int RGWRados::get_obj_state_impl(RGWObjectCtx *rctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
9507 | RGWObjState **state, bool follow_olh, bool assume_noent) | |
9508 | { | |
9509 | if (obj.empty()) { | |
9510 | return -EINVAL; | |
9511 | } | |
9512 | ||
9513 | bool need_follow_olh = follow_olh && obj.key.instance.empty(); | |
9514 | ||
9515 | RGWObjState *s = rctx->obj.get_state(obj); | |
9516 | ldout(cct, 20) << "get_obj_state: rctx=" << (void *)rctx << " obj=" << obj << " state=" << (void *)s << " s->prefetch_data=" << s->prefetch_data << dendl; | |
9517 | *state = s; | |
9518 | if (s->has_attrs) { | |
9519 | if (s->is_olh && need_follow_olh) { | |
9520 | return get_olh_target_state(*rctx, bucket_info, obj, s, state); | |
9521 | } | |
9522 | return 0; | |
9523 | } | |
9524 | ||
9525 | s->obj = obj; | |
9526 | ||
9527 | rgw_raw_obj raw_obj; | |
9528 | obj_to_raw(bucket_info.placement_rule, obj, &raw_obj); | |
9529 | ||
9530 | int r = -ENOENT; | |
9531 | ||
9532 | if (!assume_noent) { | |
9533 | r = RGWRados::raw_obj_stat(raw_obj, &s->size, &s->mtime, &s->epoch, &s->attrset, (s->prefetch_data ? &s->data : NULL), NULL); | |
9534 | } | |
9535 | ||
9536 | if (r == -ENOENT) { | |
9537 | s->exists = false; | |
9538 | s->has_attrs = true; | |
9539 | tombstone_entry entry; | |
9540 | if (obj_tombstone_cache && obj_tombstone_cache->find(obj, entry)) { | |
9541 | s->mtime = entry.mtime; | |
9542 | s->zone_short_id = entry.zone_short_id; | |
9543 | s->pg_ver = entry.pg_ver; | |
9544 | ldout(cct, 20) << __func__ << "(): found obj in tombstone cache: obj=" << obj | |
9545 | << " mtime=" << s->mtime << " pgv=" << s->pg_ver << dendl; | |
9546 | } else { | |
9547 | s->mtime = real_time(); | |
9548 | } | |
9549 | return 0; | |
9550 | } | |
9551 | if (r < 0) | |
9552 | return r; | |
9553 | ||
9554 | s->exists = true; | |
9555 | s->has_attrs = true; | |
9556 | s->accounted_size = s->size; | |
9557 | ||
9558 | auto iter = s->attrset.find(RGW_ATTR_COMPRESSION); | |
9559 | const bool compressed = (iter != s->attrset.end()); | |
9560 | if (compressed) { | |
9561 | // use uncompressed size for accounted_size | |
9562 | try { | |
9563 | RGWCompressionInfo info; | |
9564 | auto p = iter->second.begin(); | |
9565 | ::decode(info, p); | |
9566 | s->accounted_size = info.orig_size; | |
9567 | } catch (buffer::error&) { | |
9568 | dout(0) << "ERROR: could not decode compression info for object: " << obj << dendl; | |
9569 | return -EIO; | |
9570 | } | |
9571 | } | |
9572 | ||
9573 | iter = s->attrset.find(RGW_ATTR_SHADOW_OBJ); | |
9574 | if (iter != s->attrset.end()) { | |
9575 | bufferlist bl = iter->second; | |
9576 | bufferlist::iterator it = bl.begin(); | |
9577 | it.copy(bl.length(), s->shadow_obj); | |
9578 | s->shadow_obj[bl.length()] = '\0'; | |
9579 | } | |
9580 | s->obj_tag = s->attrset[RGW_ATTR_ID_TAG]; | |
9581 | auto ttiter = s->attrset.find(RGW_ATTR_TAIL_TAG); | |
9582 | if (ttiter != s->attrset.end()) { | |
9583 | s->tail_tag = s->attrset[RGW_ATTR_TAIL_TAG]; | |
9584 | } | |
9585 | ||
9586 | bufferlist manifest_bl = s->attrset[RGW_ATTR_MANIFEST]; | |
9587 | if (manifest_bl.length()) { | |
9588 | bufferlist::iterator miter = manifest_bl.begin(); | |
9589 | try { | |
9590 | ::decode(s->manifest, miter); | |
9591 | s->has_manifest = true; | |
9592 | s->manifest.set_head(bucket_info.placement_rule, obj, s->size); /* patch manifest to reflect the head we just read, some manifests might be | |
9593 | broken due to old bugs */ | |
9594 | s->size = s->manifest.get_obj_size(); | |
9595 | if (!compressed) | |
9596 | s->accounted_size = s->size; | |
9597 | } catch (buffer::error& err) { | |
9598 | ldout(cct, 0) << "ERROR: couldn't decode manifest" << dendl; | |
9599 | return -EIO; | |
9600 | } | |
9601 | ldout(cct, 10) << "manifest: total_size = " << s->manifest.get_obj_size() << dendl; | |
9602 | if (cct->_conf->subsys.should_gather(ceph_subsys_rgw, 20) && s->manifest.has_explicit_objs()) { | |
9603 | RGWObjManifest::obj_iterator mi; | |
9604 | for (mi = s->manifest.obj_begin(); mi != s->manifest.obj_end(); ++mi) { | |
9605 | ldout(cct, 20) << "manifest: ofs=" << mi.get_ofs() << " loc=" << mi.get_location().get_raw_obj(this) << dendl; | |
9606 | } | |
9607 | } | |
9608 | ||
9609 | if (!s->obj_tag.length()) { | |
9610 | /* | |
9611 | * Uh oh, something's wrong, object with manifest should have tag. Let's | |
9612 | * create one out of the manifest, would be unique | |
9613 | */ | |
9614 | generate_fake_tag(this, s->attrset, s->manifest, manifest_bl, s->obj_tag); | |
9615 | s->fake_tag = true; | |
9616 | } | |
9617 | } | |
9618 | map<string, bufferlist>::iterator aiter = s->attrset.find(RGW_ATTR_PG_VER); | |
9619 | if (aiter != s->attrset.end()) { | |
9620 | bufferlist& pg_ver_bl = aiter->second; | |
9621 | if (pg_ver_bl.length()) { | |
9622 | bufferlist::iterator pgbl = pg_ver_bl.begin(); | |
9623 | try { | |
9624 | ::decode(s->pg_ver, pgbl); | |
9625 | } catch (buffer::error& err) { | |
9626 | ldout(cct, 0) << "ERROR: couldn't decode pg ver attr for object " << s->obj << ", non-critical error, ignoring" << dendl; | |
9627 | } | |
9628 | } | |
9629 | } | |
9630 | aiter = s->attrset.find(RGW_ATTR_SOURCE_ZONE); | |
9631 | if (aiter != s->attrset.end()) { | |
9632 | bufferlist& zone_short_id_bl = aiter->second; | |
9633 | if (zone_short_id_bl.length()) { | |
9634 | bufferlist::iterator zbl = zone_short_id_bl.begin(); | |
9635 | try { | |
9636 | ::decode(s->zone_short_id, zbl); | |
9637 | } catch (buffer::error& err) { | |
9638 | ldout(cct, 0) << "ERROR: couldn't decode zone short id attr for object " << s->obj << ", non-critical error, ignoring" << dendl; | |
9639 | } | |
9640 | } | |
9641 | } | |
9642 | if (s->obj_tag.length()) | |
9643 | ldout(cct, 20) << "get_obj_state: setting s->obj_tag to " << s->obj_tag.c_str() << dendl; | |
9644 | else | |
9645 | ldout(cct, 20) << "get_obj_state: s->obj_tag was set empty" << dendl; | |
9646 | ||
9647 | /* an object might not be olh yet, but could have olh id tag, so we should set it anyway if | |
9648 | * it exist, and not only if is_olh() returns true | |
9649 | */ | |
9650 | iter = s->attrset.find(RGW_ATTR_OLH_ID_TAG); | |
9651 | if (iter != s->attrset.end()) { | |
9652 | s->olh_tag = iter->second; | |
9653 | } | |
9654 | ||
9655 | if (is_olh(s->attrset)) { | |
9656 | s->is_olh = true; | |
9657 | ||
9658 | ldout(cct, 20) << __func__ << ": setting s->olh_tag to " << string(s->olh_tag.c_str(), s->olh_tag.length()) << dendl; | |
9659 | ||
9660 | if (need_follow_olh) { | |
9661 | return get_olh_target_state(*rctx, bucket_info, obj, s, state); | |
9662 | } | |
9663 | } | |
9664 | ||
9665 | return 0; | |
9666 | } | |
9667 | ||
9668 | int RGWRados::get_obj_state(RGWObjectCtx *rctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, RGWObjState **state, | |
9669 | bool follow_olh, bool assume_noent) | |
9670 | { | |
9671 | int ret; | |
9672 | ||
9673 | do { | |
9674 | ret = get_obj_state_impl(rctx, bucket_info, obj, state, follow_olh, assume_noent); | |
9675 | } while (ret == -EAGAIN); | |
9676 | ||
9677 | return ret; | |
9678 | } | |
9679 | ||
9680 | int RGWRados::Object::get_manifest(RGWObjManifest **pmanifest) | |
9681 | { | |
9682 | RGWObjState *astate; | |
9683 | int r = get_state(&astate, true); | |
9684 | if (r < 0) { | |
9685 | return r; | |
9686 | } | |
9687 | ||
9688 | *pmanifest = &astate->manifest; | |
9689 | ||
9690 | return 0; | |
9691 | } | |
9692 | ||
9693 | int RGWRados::Object::Read::get_attr(const char *name, bufferlist& dest) | |
9694 | { | |
9695 | RGWObjState *state; | |
9696 | int r = source->get_state(&state, true); | |
9697 | if (r < 0) | |
9698 | return r; | |
9699 | if (!state->exists) | |
9700 | return -ENOENT; | |
9701 | if (!state->get_attr(name, dest)) | |
9702 | return -ENODATA; | |
9703 | ||
9704 | return 0; | |
9705 | } | |
9706 | ||
9707 | ||
9708 | int RGWRados::Object::Stat::stat_async() | |
9709 | { | |
9710 | RGWObjectCtx& ctx = source->get_ctx(); | |
9711 | rgw_obj& obj = source->get_obj(); | |
9712 | RGWRados *store = source->get_store(); | |
9713 | ||
9714 | RGWObjState *s = ctx.obj.get_state(obj); /* calling this one directly because otherwise a sync request will be sent */ | |
9715 | result.obj = obj; | |
9716 | if (s->has_attrs) { | |
9717 | state.ret = 0; | |
9718 | result.size = s->size; | |
9719 | result.mtime = ceph::real_clock::to_timespec(s->mtime); | |
9720 | result.attrs = s->attrset; | |
9721 | result.has_manifest = s->has_manifest; | |
9722 | result.manifest = s->manifest; | |
9723 | return 0; | |
9724 | } | |
9725 | ||
9726 | string oid; | |
9727 | string loc; | |
9728 | get_obj_bucket_and_oid_loc(obj, oid, loc); | |
9729 | ||
9730 | int r = store->get_obj_head_ioctx(source->get_bucket_info(), obj, &state.io_ctx); | |
9731 | if (r < 0) { | |
9732 | return r; | |
9733 | } | |
9734 | ||
9735 | librados::ObjectReadOperation op; | |
9736 | op.stat2(&result.size, &result.mtime, NULL); | |
9737 | op.getxattrs(&result.attrs, NULL); | |
9738 | state.completion = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
9739 | state.io_ctx.locator_set_key(loc); | |
9740 | r = state.io_ctx.aio_operate(oid, state.completion, &op, NULL); | |
9741 | if (r < 0) { | |
9742 | ldout(store->ctx(), 5) << __func__ | |
9743 | << ": ERROR: aio_operate() returned ret=" << r | |
9744 | << dendl; | |
9745 | return r; | |
9746 | } | |
9747 | ||
9748 | return 0; | |
9749 | } | |
9750 | ||
9751 | ||
9752 | int RGWRados::Object::Stat::wait() | |
9753 | { | |
9754 | if (!state.completion) { | |
9755 | return state.ret; | |
9756 | } | |
9757 | ||
9758 | state.completion->wait_for_safe(); | |
9759 | state.ret = state.completion->get_return_value(); | |
9760 | state.completion->release(); | |
9761 | ||
9762 | if (state.ret != 0) { | |
9763 | return state.ret; | |
9764 | } | |
9765 | ||
9766 | return finish(); | |
9767 | } | |
9768 | ||
9769 | int RGWRados::Object::Stat::finish() | |
9770 | { | |
9771 | map<string, bufferlist>::iterator iter = result.attrs.find(RGW_ATTR_MANIFEST); | |
9772 | if (iter != result.attrs.end()) { | |
9773 | bufferlist& bl = iter->second; | |
9774 | bufferlist::iterator biter = bl.begin(); | |
9775 | try { | |
9776 | ::decode(result.manifest, biter); | |
9777 | } catch (buffer::error& err) { | |
9778 | RGWRados *store = source->get_store(); | |
9779 | ldout(store->ctx(), 0) << "ERROR: " << __func__ << ": failed to decode manifest" << dendl; | |
9780 | return -EIO; | |
9781 | } | |
9782 | result.has_manifest = true; | |
9783 | } | |
9784 | ||
9785 | return 0; | |
9786 | } | |
9787 | ||
9788 | /** | |
9789 | * Get an attribute for a system object. | |
9790 | * obj: the object to get attr | |
9791 | * name: name of the attr to retrieve | |
9792 | * dest: bufferlist to store the result in | |
9793 | * Returns: 0 on success, -ERR# otherwise. | |
9794 | */ | |
9795 | int RGWRados::system_obj_get_attr(rgw_raw_obj& obj, const char *name, bufferlist& dest) | |
9796 | { | |
9797 | rgw_rados_ref ref; | |
9798 | int r = get_system_obj_ref(obj, &ref); | |
9799 | if (r < 0) { | |
9800 | return r; | |
9801 | } | |
9802 | ||
9803 | ObjectReadOperation op; | |
9804 | ||
9805 | int rval; | |
9806 | op.getxattr(name, &dest, &rval); | |
9807 | ||
9808 | r = ref.ioctx.operate(ref.oid, &op, NULL); | |
9809 | if (r < 0) | |
9810 | return r; | |
9811 | ||
9812 | return 0; | |
9813 | } | |
9814 | ||
9815 | int RGWRados::append_atomic_test(RGWObjectCtx *rctx, | |
9816 | const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
9817 | ObjectOperation& op, RGWObjState **pstate) | |
9818 | { | |
9819 | if (!rctx) | |
9820 | return 0; | |
9821 | ||
9822 | int r = get_obj_state(rctx, bucket_info, obj, pstate, false); | |
9823 | if (r < 0) | |
9824 | return r; | |
9825 | ||
9826 | RGWObjState *state = *pstate; | |
9827 | ||
9828 | if (!state->is_atomic) { | |
9829 | ldout(cct, 20) << "state for obj=" << obj << " is not atomic, not appending atomic test" << dendl; | |
9830 | return 0; | |
9831 | } | |
9832 | ||
9833 | if (state->obj_tag.length() > 0 && !state->fake_tag) {// check for backward compatibility | |
9834 | op.cmpxattr(RGW_ATTR_ID_TAG, LIBRADOS_CMPXATTR_OP_EQ, state->obj_tag); | |
9835 | } else { | |
9836 | ldout(cct, 20) << "state->obj_tag is empty, not appending atomic test" << dendl; | |
9837 | } | |
9838 | return 0; | |
9839 | } | |
9840 | ||
9841 | int RGWRados::Object::get_state(RGWObjState **pstate, bool follow_olh, bool assume_noent) | |
9842 | { | |
9843 | return store->get_obj_state(&ctx, bucket_info, obj, pstate, follow_olh, assume_noent); | |
9844 | } | |
9845 | ||
9846 | void RGWRados::Object::invalidate_state() | |
9847 | { | |
9848 | ctx.obj.invalidate(obj); | |
9849 | } | |
9850 | ||
9851 | void RGWRados::SystemObject::invalidate_state() | |
9852 | { | |
9853 | ctx.raw.invalidate(obj); | |
9854 | } | |
9855 | ||
9856 | int RGWRados::Object::prepare_atomic_modification(ObjectWriteOperation& op, bool reset_obj, const string *ptag, | |
9857 | const char *if_match, const char *if_nomatch, bool removal_op, | |
9858 | bool modify_tail) | |
9859 | { | |
9860 | int r = get_state(&state, false); | |
9861 | if (r < 0) | |
9862 | return r; | |
9863 | ||
9864 | bool need_guard = (state->has_manifest || (state->obj_tag.length() != 0) || | |
9865 | if_match != NULL || if_nomatch != NULL) && | |
9866 | (!state->fake_tag); | |
9867 | ||
9868 | if (!state->is_atomic) { | |
9869 | ldout(store->ctx(), 20) << "prepare_atomic_modification: state is not atomic. state=" << (void *)state << dendl; | |
9870 | ||
9871 | if (reset_obj) { | |
9872 | op.create(false); | |
9873 | store->remove_rgw_head_obj(op); // we're not dropping reference here, actually removing object | |
9874 | } | |
9875 | ||
9876 | return 0; | |
9877 | } | |
9878 | ||
9879 | if (need_guard) { | |
9880 | /* first verify that the object wasn't replaced under */ | |
9881 | if (if_nomatch == NULL || strcmp(if_nomatch, "*") != 0) { | |
9882 | op.cmpxattr(RGW_ATTR_ID_TAG, LIBRADOS_CMPXATTR_OP_EQ, state->obj_tag); | |
9883 | // FIXME: need to add FAIL_NOTEXIST_OK for racing deletion | |
9884 | } | |
9885 | ||
9886 | if (if_match) { | |
9887 | if (strcmp(if_match, "*") == 0) { | |
9888 | // test the object is existing | |
9889 | if (!state->exists) { | |
9890 | return -ERR_PRECONDITION_FAILED; | |
9891 | } | |
9892 | } else { | |
9893 | bufferlist bl; | |
9894 | if (!state->get_attr(RGW_ATTR_ETAG, bl) || | |
9895 | strncmp(if_match, bl.c_str(), bl.length()) != 0) { | |
9896 | return -ERR_PRECONDITION_FAILED; | |
9897 | } | |
9898 | } | |
9899 | } | |
9900 | ||
9901 | if (if_nomatch) { | |
9902 | if (strcmp(if_nomatch, "*") == 0) { | |
9903 | // test the object is NOT existing | |
9904 | if (state->exists) { | |
9905 | return -ERR_PRECONDITION_FAILED; | |
9906 | } | |
9907 | } else { | |
9908 | bufferlist bl; | |
9909 | if (!state->get_attr(RGW_ATTR_ETAG, bl) || | |
9910 | strncmp(if_nomatch, bl.c_str(), bl.length()) == 0) { | |
9911 | return -ERR_PRECONDITION_FAILED; | |
9912 | } | |
9913 | } | |
9914 | } | |
9915 | } | |
9916 | ||
9917 | if (reset_obj) { | |
9918 | if (state->exists) { | |
9919 | op.create(false); | |
9920 | store->remove_rgw_head_obj(op); | |
9921 | } else { | |
9922 | op.create(true); | |
9923 | } | |
9924 | } | |
9925 | ||
9926 | if (removal_op) { | |
9927 | /* the object is being removed, no need to update its tag */ | |
9928 | return 0; | |
9929 | } | |
9930 | ||
9931 | if (ptag) { | |
9932 | state->write_tag = *ptag; | |
9933 | } else { | |
9934 | append_rand_alpha(store->ctx(), state->write_tag, state->write_tag, 32); | |
9935 | } | |
9936 | bufferlist bl; | |
9937 | bl.append(state->write_tag.c_str(), state->write_tag.size() + 1); | |
9938 | ||
9939 | ldout(store->ctx(), 10) << "setting object write_tag=" << state->write_tag << dendl; | |
9940 | ||
9941 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
9942 | if (modify_tail) { | |
9943 | op.setxattr(RGW_ATTR_TAIL_TAG, bl); | |
9944 | } | |
9945 | ||
9946 | return 0; | |
9947 | } | |
9948 | ||
9949 | int RGWRados::system_obj_set_attr(void *ctx, rgw_raw_obj& obj, const char *name, bufferlist& bl, | |
9950 | RGWObjVersionTracker *objv_tracker) | |
9951 | { | |
9952 | map<string, bufferlist> attrs; | |
9953 | attrs[name] = bl; | |
9954 | return system_obj_set_attrs(ctx, obj, attrs, NULL, objv_tracker); | |
9955 | } | |
9956 | ||
9957 | int RGWRados::system_obj_set_attrs(void *ctx, rgw_raw_obj& obj, | |
9958 | map<string, bufferlist>& attrs, | |
9959 | map<string, bufferlist>* rmattrs, | |
9960 | RGWObjVersionTracker *objv_tracker) | |
9961 | { | |
9962 | rgw_rados_ref ref; | |
9963 | int r = get_system_obj_ref(obj, &ref); | |
9964 | if (r < 0) { | |
9965 | return r; | |
9966 | } | |
9967 | ObjectWriteOperation op; | |
9968 | ||
9969 | if (objv_tracker) { | |
9970 | objv_tracker->prepare_op_for_write(&op); | |
9971 | } | |
9972 | ||
9973 | map<string, bufferlist>::iterator iter; | |
9974 | if (rmattrs) { | |
9975 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
9976 | const string& name = iter->first; | |
9977 | op.rmxattr(name.c_str()); | |
9978 | } | |
9979 | } | |
9980 | ||
9981 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
9982 | const string& name = iter->first; | |
9983 | bufferlist& bl = iter->second; | |
9984 | ||
9985 | if (!bl.length()) | |
9986 | continue; | |
9987 | ||
9988 | op.setxattr(name.c_str(), bl); | |
9989 | } | |
9990 | ||
9991 | if (!op.size()) | |
9992 | return 0; | |
9993 | ||
9994 | bufferlist bl; | |
9995 | ||
9996 | r = ref.ioctx.operate(ref.oid, &op); | |
9997 | if (r < 0) | |
9998 | return r; | |
9999 | ||
10000 | return 0; | |
10001 | } | |
10002 | ||
10003 | /** | |
10004 | * Set an attr on an object. | |
10005 | * bucket: name of the bucket holding the object | |
10006 | * obj: name of the object to set the attr on | |
10007 | * name: the attr to set | |
10008 | * bl: the contents of the attr | |
10009 | * Returns: 0 on success, -ERR# otherwise. | |
10010 | */ | |
10011 | int RGWRados::set_attr(void *ctx, const RGWBucketInfo& bucket_info, rgw_obj& obj, const char *name, bufferlist& bl) | |
10012 | { | |
10013 | map<string, bufferlist> attrs; | |
10014 | attrs[name] = bl; | |
10015 | return set_attrs(ctx, bucket_info, obj, attrs, NULL); | |
10016 | } | |
10017 | ||
10018 | int RGWRados::set_attrs(void *ctx, const RGWBucketInfo& bucket_info, rgw_obj& obj, | |
10019 | map<string, bufferlist>& attrs, | |
10020 | map<string, bufferlist>* rmattrs) | |
10021 | { | |
10022 | rgw_rados_ref ref; | |
10023 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
10024 | if (r < 0) { | |
10025 | return r; | |
10026 | } | |
10027 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
10028 | ||
10029 | ObjectWriteOperation op; | |
10030 | RGWObjState *state = NULL; | |
10031 | ||
10032 | r = append_atomic_test(rctx, bucket_info, obj, op, &state); | |
10033 | if (r < 0) | |
10034 | return r; | |
10035 | ||
10036 | map<string, bufferlist>::iterator iter; | |
10037 | if (rmattrs) { | |
10038 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
10039 | const string& name = iter->first; | |
10040 | op.rmxattr(name.c_str()); | |
10041 | } | |
10042 | } | |
10043 | ||
10044 | const rgw_bucket& bucket = obj.bucket; | |
10045 | ||
10046 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
10047 | const string& name = iter->first; | |
10048 | bufferlist& bl = iter->second; | |
10049 | ||
10050 | if (!bl.length()) | |
10051 | continue; | |
10052 | ||
10053 | op.setxattr(name.c_str(), bl); | |
10054 | ||
10055 | if (name.compare(RGW_ATTR_DELETE_AT) == 0) { | |
10056 | real_time ts; | |
10057 | try { | |
10058 | ::decode(ts, bl); | |
10059 | ||
10060 | rgw_obj_index_key obj_key; | |
10061 | obj.key.get_index_key(&obj_key); | |
10062 | ||
10063 | objexp_hint_add(ts, bucket.tenant, bucket.name, bucket.bucket_id, obj_key); | |
10064 | } catch (buffer::error& err) { | |
10065 | ldout(cct, 0) << "ERROR: failed to decode " RGW_ATTR_DELETE_AT << " attr" << dendl; | |
10066 | } | |
10067 | } | |
10068 | } | |
10069 | ||
10070 | if (!op.size()) | |
10071 | return 0; | |
10072 | ||
10073 | RGWObjectCtx obj_ctx(this); | |
10074 | ||
10075 | bufferlist bl; | |
10076 | RGWRados::Bucket bop(this, bucket_info); | |
10077 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
10078 | ||
10079 | if (state) { | |
10080 | string tag; | |
10081 | append_rand_alpha(cct, tag, tag, 32); | |
10082 | state->write_tag = tag; | |
10083 | r = index_op.prepare(CLS_RGW_OP_ADD, &state->write_tag); | |
10084 | ||
10085 | if (r < 0) | |
10086 | return r; | |
10087 | ||
10088 | bl.append(tag.c_str(), tag.size() + 1); | |
10089 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
10090 | } | |
10091 | ||
10092 | ||
10093 | real_time mtime = real_clock::now(); | |
10094 | struct timespec mtime_ts = real_clock::to_timespec(mtime); | |
10095 | op.mtime2(&mtime_ts); | |
10096 | r = ref.ioctx.operate(ref.oid, &op); | |
10097 | if (state) { | |
10098 | if (r >= 0) { | |
10099 | bufferlist acl_bl = attrs[RGW_ATTR_ACL]; | |
10100 | bufferlist etag_bl = attrs[RGW_ATTR_ETAG]; | |
10101 | bufferlist content_type_bl = attrs[RGW_ATTR_CONTENT_TYPE]; | |
10102 | string etag(etag_bl.c_str(), etag_bl.length()); | |
10103 | string content_type(content_type_bl.c_str(), content_type_bl.length()); | |
10104 | uint64_t epoch = ref.ioctx.get_last_version(); | |
10105 | int64_t poolid = ref.ioctx.get_id(); | |
10106 | r = index_op.complete(poolid, epoch, state->size, state->accounted_size, | |
10107 | mtime, etag, content_type, &acl_bl, | |
10108 | RGW_OBJ_CATEGORY_MAIN, NULL); | |
10109 | } else { | |
10110 | int ret = index_op.cancel(); | |
10111 | if (ret < 0) { | |
10112 | ldout(cct, 0) << "ERROR: complete_update_index_cancel() returned ret=" << ret << dendl; | |
10113 | } | |
10114 | } | |
10115 | } | |
10116 | if (r < 0) | |
10117 | return r; | |
10118 | ||
10119 | if (state) { | |
10120 | state->obj_tag.swap(bl); | |
10121 | if (rmattrs) { | |
10122 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
10123 | state->attrset.erase(iter->first); | |
10124 | } | |
10125 | } | |
10126 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
10127 | state->attrset[iter->first] = iter->second; | |
10128 | } | |
10129 | } | |
10130 | ||
10131 | return 0; | |
10132 | } | |
10133 | ||
10134 | int RGWRados::Object::Read::prepare() | |
10135 | { | |
10136 | RGWRados *store = source->get_store(); | |
10137 | CephContext *cct = store->ctx(); | |
10138 | ||
10139 | bufferlist etag; | |
10140 | ||
10141 | map<string, bufferlist>::iterator iter; | |
10142 | ||
10143 | RGWObjState *astate; | |
10144 | int r = source->get_state(&astate, true); | |
10145 | if (r < 0) | |
10146 | return r; | |
10147 | ||
10148 | if (!astate->exists) { | |
10149 | return -ENOENT; | |
10150 | } | |
10151 | ||
10152 | const RGWBucketInfo& bucket_info = source->get_bucket_info(); | |
10153 | ||
10154 | state.obj = astate->obj; | |
10155 | store->obj_to_raw(bucket_info.placement_rule, state.obj, &state.head_obj); | |
10156 | ||
10157 | r = store->get_obj_head_ioctx(bucket_info, state.obj, &state.io_ctx); | |
10158 | if (r < 0) { | |
10159 | return r; | |
10160 | } | |
10161 | if (params.attrs) { | |
10162 | *params.attrs = astate->attrset; | |
10163 | if (cct->_conf->subsys.should_gather(ceph_subsys_rgw, 20)) { | |
10164 | for (iter = params.attrs->begin(); iter != params.attrs->end(); ++iter) { | |
10165 | ldout(cct, 20) << "Read xattr: " << iter->first << dendl; | |
10166 | } | |
10167 | } | |
10168 | } | |
10169 | ||
10170 | /* Convert all times go GMT to make them compatible */ | |
10171 | if (conds.mod_ptr || conds.unmod_ptr) { | |
10172 | obj_time_weight src_weight; | |
10173 | src_weight.init(astate); | |
10174 | src_weight.high_precision = conds.high_precision_time; | |
10175 | ||
10176 | obj_time_weight dest_weight; | |
10177 | dest_weight.high_precision = conds.high_precision_time; | |
10178 | ||
10179 | if (conds.mod_ptr) { | |
10180 | dest_weight.init(*conds.mod_ptr, conds.mod_zone_id, conds.mod_pg_ver); | |
10181 | ldout(cct, 10) << "If-Modified-Since: " << dest_weight << " Last-Modified: " << src_weight << dendl; | |
10182 | if (!(dest_weight < src_weight)) { | |
10183 | return -ERR_NOT_MODIFIED; | |
10184 | } | |
10185 | } | |
10186 | ||
10187 | if (conds.unmod_ptr) { | |
10188 | dest_weight.init(*conds.unmod_ptr, conds.mod_zone_id, conds.mod_pg_ver); | |
10189 | ldout(cct, 10) << "If-UnModified-Since: " << dest_weight << " Last-Modified: " << src_weight << dendl; | |
10190 | if (dest_weight < src_weight) { | |
10191 | return -ERR_PRECONDITION_FAILED; | |
10192 | } | |
10193 | } | |
10194 | } | |
10195 | if (conds.if_match || conds.if_nomatch) { | |
10196 | r = get_attr(RGW_ATTR_ETAG, etag); | |
10197 | if (r < 0) | |
10198 | return r; | |
10199 | ||
10200 | if (conds.if_match) { | |
10201 | string if_match_str = rgw_string_unquote(conds.if_match); | |
10202 | ldout(cct, 10) << "ETag: " << etag.c_str() << " " << " If-Match: " << if_match_str << dendl; | |
10203 | if (if_match_str.compare(etag.c_str()) != 0) { | |
10204 | return -ERR_PRECONDITION_FAILED; | |
10205 | } | |
10206 | } | |
10207 | ||
10208 | if (conds.if_nomatch) { | |
10209 | string if_nomatch_str = rgw_string_unquote(conds.if_nomatch); | |
10210 | ldout(cct, 10) << "ETag: " << etag.c_str() << " " << " If-NoMatch: " << if_nomatch_str << dendl; | |
10211 | if (if_nomatch_str.compare(etag.c_str()) == 0) { | |
10212 | return -ERR_NOT_MODIFIED; | |
10213 | } | |
10214 | } | |
10215 | } | |
10216 | ||
10217 | if (params.obj_size) | |
10218 | *params.obj_size = astate->size; | |
10219 | if (params.lastmod) | |
10220 | *params.lastmod = astate->mtime; | |
10221 | ||
10222 | return 0; | |
10223 | } | |
10224 | ||
10225 | int RGWRados::Object::Read::range_to_ofs(uint64_t obj_size, int64_t &ofs, int64_t &end) | |
10226 | { | |
10227 | if (ofs < 0) { | |
10228 | ofs += obj_size; | |
10229 | if (ofs < 0) | |
10230 | ofs = 0; | |
10231 | end = obj_size - 1; | |
10232 | } else if (end < 0) { | |
10233 | end = obj_size - 1; | |
10234 | } | |
10235 | ||
10236 | if (obj_size > 0) { | |
10237 | if (ofs >= (off_t)obj_size) { | |
10238 | return -ERANGE; | |
10239 | } | |
10240 | if (end >= (off_t)obj_size) { | |
10241 | end = obj_size - 1; | |
10242 | } | |
10243 | } | |
10244 | return 0; | |
10245 | } | |
10246 | ||
10247 | int RGWRados::SystemObject::get_state(RGWRawObjState **pstate, RGWObjVersionTracker *objv_tracker) | |
10248 | { | |
10249 | return store->get_system_obj_state(&ctx, obj, pstate, objv_tracker); | |
10250 | } | |
10251 | ||
10252 | int RGWRados::stat_system_obj(RGWObjectCtx& obj_ctx, | |
10253 | RGWRados::SystemObject::Read::GetObjState& state, | |
10254 | rgw_raw_obj& obj, | |
10255 | map<string, bufferlist> *attrs, | |
10256 | real_time *lastmod, | |
10257 | uint64_t *obj_size, | |
10258 | RGWObjVersionTracker *objv_tracker) | |
10259 | { | |
10260 | RGWRawObjState *astate = NULL; | |
10261 | ||
10262 | int r = get_system_obj_state(&obj_ctx, obj, &astate, objv_tracker); | |
10263 | if (r < 0) | |
10264 | return r; | |
10265 | ||
10266 | if (!astate->exists) { | |
10267 | return -ENOENT; | |
10268 | } | |
10269 | ||
10270 | if (attrs) { | |
10271 | *attrs = astate->attrset; | |
10272 | if (cct->_conf->subsys.should_gather(ceph_subsys_rgw, 20)) { | |
10273 | map<string, bufferlist>::iterator iter; | |
10274 | for (iter = attrs->begin(); iter != attrs->end(); ++iter) { | |
10275 | ldout(cct, 20) << "Read xattr: " << iter->first << dendl; | |
10276 | } | |
10277 | } | |
10278 | } | |
10279 | ||
10280 | if (obj_size) | |
10281 | *obj_size = astate->size; | |
10282 | if (lastmod) | |
10283 | *lastmod = astate->mtime; | |
10284 | ||
10285 | return 0; | |
10286 | } | |
10287 | ||
10288 | ||
10289 | int RGWRados::Bucket::UpdateIndex::guard_reshard(BucketShard **pbs, std::function<int(BucketShard *)> call) | |
10290 | { | |
10291 | RGWRados *store = target->get_store(); | |
10292 | BucketShard *bs; | |
10293 | int r; | |
10294 | ||
10295 | #define NUM_RESHARD_RETRIES 10 | |
10296 | for (int i = 0; i < NUM_RESHARD_RETRIES; ++i) { | |
10297 | int ret = get_bucket_shard(&bs); | |
10298 | if (ret < 0) { | |
10299 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
10300 | return ret; | |
10301 | } | |
10302 | r = call(bs); | |
10303 | if (r != -ERR_BUSY_RESHARDING) { | |
10304 | break; | |
10305 | } | |
10306 | ldout(store->ctx(), 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl; | |
10307 | string new_bucket_id; | |
10308 | r = store->block_while_resharding(bs, &new_bucket_id, target->bucket_info); | |
10309 | if (r == -ERR_BUSY_RESHARDING) { | |
10310 | continue; | |
10311 | } | |
10312 | if (r < 0) { | |
10313 | return r; | |
10314 | } | |
10315 | ldout(store->ctx(), 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id << dendl; | |
10316 | i = 0; /* resharding is finished, make sure we can retry */ | |
10317 | r = target->update_bucket_id(new_bucket_id); | |
10318 | if (r < 0) { | |
10319 | ldout(store->ctx(), 0) << "ERROR: update_bucket_id() new_bucket_id=" << new_bucket_id << " returned r=" << r << dendl; | |
10320 | return r; | |
10321 | } | |
10322 | invalidate_bs(); | |
10323 | } | |
10324 | ||
10325 | if (r < 0) { | |
10326 | return r; | |
10327 | } | |
10328 | ||
10329 | if (pbs) { | |
10330 | *pbs = bs; | |
10331 | } | |
10332 | ||
10333 | return 0; | |
10334 | } | |
10335 | ||
10336 | int RGWRados::SystemObject::Read::stat(RGWObjVersionTracker *objv_tracker) | |
10337 | { | |
10338 | RGWRados *store = source->get_store(); | |
10339 | rgw_raw_obj& obj = source->get_obj(); | |
10340 | ||
10341 | return store->stat_system_obj(source->get_ctx(), state, obj, stat_params.attrs, | |
10342 | stat_params.lastmod, stat_params.obj_size, objv_tracker); | |
10343 | } | |
10344 | ||
10345 | int RGWRados::Bucket::UpdateIndex::prepare(RGWModifyOp op, const string *write_tag) | |
10346 | { | |
10347 | if (blind) { | |
10348 | return 0; | |
10349 | } | |
10350 | RGWRados *store = target->get_store(); | |
10351 | ||
10352 | if (write_tag && write_tag->length()) { | |
10353 | optag = string(write_tag->c_str(), write_tag->length()); | |
10354 | } else { | |
10355 | if (optag.empty()) { | |
10356 | append_rand_alpha(store->ctx(), optag, optag, 32); | |
10357 | } | |
10358 | } | |
10359 | ||
10360 | int r = guard_reshard(nullptr, [&](BucketShard *bs) -> int { | |
10361 | return store->cls_obj_prepare_op(*bs, op, optag, obj, bilog_flags, zones_trace); | |
10362 | }); | |
10363 | ||
10364 | if (r < 0) { | |
10365 | return r; | |
10366 | } | |
10367 | prepared = true; | |
10368 | ||
10369 | return 0; | |
10370 | } | |
10371 | ||
10372 | int RGWRados::Bucket::UpdateIndex::complete(int64_t poolid, uint64_t epoch, | |
10373 | uint64_t size, uint64_t accounted_size, | |
10374 | ceph::real_time& ut, const string& etag, | |
10375 | const string& content_type, | |
10376 | bufferlist *acl_bl, | |
10377 | RGWObjCategory category, | |
10378 | list<rgw_obj_index_key> *remove_objs, const string *user_data) | |
10379 | { | |
10380 | if (blind) { | |
10381 | return 0; | |
10382 | } | |
10383 | RGWRados *store = target->get_store(); | |
10384 | BucketShard *bs; | |
10385 | ||
10386 | int ret = get_bucket_shard(&bs); | |
10387 | if (ret < 0) { | |
10388 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
10389 | return ret; | |
10390 | } | |
10391 | ||
10392 | rgw_bucket_dir_entry ent; | |
10393 | obj.key.get_index_key(&ent.key); | |
10394 | ent.meta.size = size; | |
10395 | ent.meta.accounted_size = accounted_size; | |
10396 | ent.meta.mtime = ut; | |
10397 | ent.meta.etag = etag; | |
10398 | if (user_data) | |
10399 | ent.meta.user_data = *user_data; | |
10400 | ||
10401 | ACLOwner owner; | |
10402 | if (acl_bl && acl_bl->length()) { | |
10403 | int ret = store->decode_policy(*acl_bl, &owner); | |
10404 | if (ret < 0) { | |
10405 | ldout(store->ctx(), 0) << "WARNING: could not decode policy ret=" << ret << dendl; | |
10406 | } | |
10407 | } | |
10408 | ent.meta.owner = owner.get_id().to_str(); | |
10409 | ent.meta.owner_display_name = owner.get_display_name(); | |
10410 | ent.meta.content_type = content_type; | |
10411 | ||
10412 | ret = store->cls_obj_complete_add(*bs, obj, optag, poolid, epoch, ent, category, remove_objs, bilog_flags, zones_trace); | |
10413 | ||
10414 | if (target->bucket_info.datasync_flag_enabled()) { | |
10415 | int r = store->data_log->add_entry(bs->bucket, bs->shard_id); | |
10416 | if (r < 0) { | |
10417 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
10418 | } | |
10419 | } | |
10420 | ||
10421 | return ret; | |
10422 | } | |
10423 | ||
10424 | int RGWRados::Bucket::UpdateIndex::complete_del(int64_t poolid, uint64_t epoch, | |
10425 | real_time& removed_mtime, | |
10426 | list<rgw_obj_index_key> *remove_objs) | |
10427 | { | |
10428 | if (blind) { | |
10429 | return 0; | |
10430 | } | |
10431 | RGWRados *store = target->get_store(); | |
10432 | BucketShard *bs; | |
10433 | ||
10434 | int ret = get_bucket_shard(&bs); | |
10435 | if (ret < 0) { | |
10436 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
10437 | return ret; | |
10438 | } | |
10439 | ||
10440 | ret = store->cls_obj_complete_del(*bs, optag, poolid, epoch, obj, removed_mtime, remove_objs, bilog_flags, zones_trace); | |
10441 | ||
10442 | if (target->bucket_info.datasync_flag_enabled()) { | |
10443 | int r = store->data_log->add_entry(bs->bucket, bs->shard_id); | |
10444 | if (r < 0) { | |
10445 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
10446 | } | |
10447 | } | |
10448 | ||
10449 | return ret; | |
10450 | } | |
10451 | ||
10452 | ||
10453 | int RGWRados::Bucket::UpdateIndex::cancel() | |
10454 | { | |
10455 | if (blind) { | |
10456 | return 0; | |
10457 | } | |
10458 | RGWRados *store = target->get_store(); | |
10459 | BucketShard *bs; | |
10460 | ||
10461 | int ret = guard_reshard(&bs, [&](BucketShard *bs) -> int { | |
10462 | return store->cls_obj_complete_cancel(*bs, optag, obj, bilog_flags, zones_trace); | |
10463 | }); | |
10464 | ||
10465 | /* | |
10466 | * need to update data log anyhow, so that whoever follows needs to update its internal markers | |
10467 | * for following the specific bucket shard log. Otherwise they end up staying behind, and users | |
10468 | * have no way to tell that they're all caught up | |
10469 | */ | |
10470 | if (target->bucket_info.datasync_flag_enabled()) { | |
10471 | int r = store->data_log->add_entry(bs->bucket, bs->shard_id); | |
10472 | if (r < 0) { | |
10473 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
10474 | } | |
10475 | } | |
10476 | ||
10477 | return ret; | |
10478 | } | |
10479 | ||
10480 | int RGWRados::Object::Read::read(int64_t ofs, int64_t end, bufferlist& bl) | |
10481 | { | |
10482 | RGWRados *store = source->get_store(); | |
10483 | CephContext *cct = store->ctx(); | |
10484 | ||
10485 | rgw_raw_obj read_obj; | |
10486 | uint64_t read_ofs = ofs; | |
10487 | uint64_t len, read_len; | |
10488 | bool reading_from_head = true; | |
10489 | ObjectReadOperation op; | |
10490 | ||
10491 | bool merge_bl = false; | |
10492 | bufferlist *pbl = &bl; | |
10493 | bufferlist read_bl; | |
10494 | uint64_t max_chunk_size; | |
10495 | ||
10496 | RGWObjState *astate; | |
10497 | int r = source->get_state(&astate, true); | |
10498 | if (r < 0) | |
10499 | return r; | |
10500 | ||
10501 | if (end < 0) | |
10502 | len = 0; | |
10503 | else | |
10504 | len = end - ofs + 1; | |
10505 | ||
10506 | if (astate->has_manifest && astate->manifest.has_tail()) { | |
10507 | /* now get the relevant object part */ | |
10508 | RGWObjManifest::obj_iterator iter = astate->manifest.obj_find(ofs); | |
10509 | ||
10510 | uint64_t stripe_ofs = iter.get_stripe_ofs(); | |
10511 | read_obj = iter.get_location().get_raw_obj(store); | |
10512 | len = min(len, iter.get_stripe_size() - (ofs - stripe_ofs)); | |
10513 | read_ofs = iter.location_ofs() + (ofs - stripe_ofs); | |
10514 | reading_from_head = (read_obj == state.head_obj); | |
10515 | } else { | |
10516 | read_obj = state.head_obj; | |
10517 | } | |
10518 | ||
10519 | r = store->get_max_chunk_size(read_obj.pool, &max_chunk_size); | |
10520 | if (r < 0) { | |
10521 | ldout(cct, 0) << "ERROR: failed to get max_chunk_size() for pool " << read_obj.pool << dendl; | |
10522 | return r; | |
10523 | } | |
10524 | ||
10525 | if (len > max_chunk_size) | |
10526 | len = max_chunk_size; | |
10527 | ||
10528 | ||
10529 | state.io_ctx.locator_set_key(read_obj.loc); | |
10530 | ||
10531 | read_len = len; | |
10532 | ||
10533 | if (reading_from_head) { | |
10534 | /* only when reading from the head object do we need to do the atomic test */ | |
10535 | r = store->append_atomic_test(&source->get_ctx(), source->get_bucket_info(), state.obj, op, &astate); | |
10536 | if (r < 0) | |
10537 | return r; | |
10538 | ||
10539 | if (astate && astate->prefetch_data) { | |
10540 | if (!ofs && astate->data.length() >= len) { | |
10541 | bl = astate->data; | |
10542 | return bl.length(); | |
10543 | } | |
10544 | ||
10545 | if (ofs < astate->data.length()) { | |
10546 | unsigned copy_len = min((uint64_t)astate->data.length() - ofs, len); | |
10547 | astate->data.copy(ofs, copy_len, bl); | |
10548 | read_len -= copy_len; | |
10549 | read_ofs += copy_len; | |
10550 | if (!read_len) | |
10551 | return bl.length(); | |
10552 | ||
10553 | merge_bl = true; | |
10554 | pbl = &read_bl; | |
10555 | } | |
10556 | } | |
10557 | } | |
10558 | ||
10559 | ldout(cct, 20) << "rados->read obj-ofs=" << ofs << " read_ofs=" << read_ofs << " read_len=" << read_len << dendl; | |
10560 | op.read(read_ofs, read_len, pbl, NULL); | |
10561 | ||
10562 | r = state.io_ctx.operate(read_obj.oid, &op, NULL); | |
10563 | ldout(cct, 20) << "rados->read r=" << r << " bl.length=" << bl.length() << dendl; | |
10564 | ||
10565 | if (r < 0) { | |
10566 | return r; | |
10567 | } | |
10568 | ||
10569 | if (merge_bl) { | |
10570 | bl.append(read_bl); | |
10571 | } | |
10572 | ||
10573 | return bl.length(); | |
10574 | } | |
10575 | ||
10576 | int RGWRados::SystemObject::Read::GetObjState::get_ref(RGWRados *store, rgw_raw_obj& obj, rgw_rados_ref **pref) | |
10577 | { | |
10578 | if (!has_ref) { | |
10579 | int r = store->get_raw_obj_ref(obj, &ref); | |
10580 | if (r < 0) { | |
10581 | return r; | |
10582 | } | |
10583 | has_ref = true; | |
10584 | } | |
10585 | *pref = &ref; | |
10586 | return 0; | |
10587 | ||
10588 | } | |
10589 | ||
10590 | int RGWRados::get_system_obj(RGWObjectCtx& obj_ctx, RGWRados::SystemObject::Read::GetObjState& read_state, | |
10591 | RGWObjVersionTracker *objv_tracker, rgw_raw_obj& obj, | |
10592 | bufferlist& bl, off_t ofs, off_t end, | |
10593 | map<string, bufferlist> *attrs, | |
10594 | rgw_cache_entry_info *cache_info, | |
10595 | boost::optional<obj_version>) | |
10596 | { | |
10597 | uint64_t len; | |
10598 | ObjectReadOperation op; | |
10599 | ||
10600 | if (end < 0) | |
10601 | len = 0; | |
10602 | else | |
10603 | len = end - ofs + 1; | |
10604 | ||
10605 | if (objv_tracker) { | |
10606 | objv_tracker->prepare_op_for_read(&op); | |
10607 | } | |
10608 | ||
10609 | ldout(cct, 20) << "rados->read ofs=" << ofs << " len=" << len << dendl; | |
10610 | op.read(ofs, len, &bl, NULL); | |
10611 | ||
10612 | if (attrs) { | |
10613 | op.getxattrs(attrs, NULL); | |
10614 | } | |
10615 | ||
10616 | rgw_rados_ref *ref; | |
10617 | int r = read_state.get_ref(this, obj, &ref); | |
10618 | if (r < 0) { | |
10619 | ldout(cct, 20) << "read_state.get_ref() on obj=" << obj << " returned " << r << dendl; | |
10620 | return r; | |
10621 | } | |
10622 | r = ref->ioctx.operate(ref->oid, &op, NULL); | |
10623 | if (r < 0) { | |
10624 | ldout(cct, 20) << "rados->read r=" << r << " bl.length=" << bl.length() << dendl; | |
10625 | return r; | |
10626 | } | |
10627 | ldout(cct, 20) << "rados->read r=" << r << " bl.length=" << bl.length() << dendl; | |
10628 | ||
10629 | uint64_t op_ver = ref->ioctx.get_last_version(); | |
10630 | ||
10631 | if (read_state.last_ver > 0 && | |
10632 | read_state.last_ver != op_ver) { | |
10633 | ldout(cct, 5) << "raced with an object write, abort" << dendl; | |
10634 | return -ECANCELED; | |
10635 | } | |
10636 | ||
10637 | read_state.last_ver = op_ver; | |
10638 | ||
10639 | return bl.length(); | |
10640 | } | |
10641 | ||
10642 | int RGWRados::SystemObject::Read::read(int64_t ofs, int64_t end, bufferlist& bl, | |
10643 | RGWObjVersionTracker *objv_tracker, | |
10644 | boost::optional<obj_version> refresh_version) | |
10645 | { | |
10646 | RGWRados *store = source->get_store(); | |
10647 | rgw_raw_obj& obj = source->get_obj(); | |
10648 | ||
10649 | return store->get_system_obj(source->get_ctx(), state, objv_tracker, obj, bl, | |
10650 | ofs, end, read_params.attrs, | |
10651 | read_params.cache_info, refresh_version); | |
10652 | } | |
10653 | ||
10654 | int RGWRados::SystemObject::Read::get_attr(const char *name, bufferlist& dest) | |
10655 | { | |
10656 | RGWRados *store = source->get_store(); | |
10657 | rgw_raw_obj& obj = source->get_obj(); | |
10658 | ||
10659 | return store->system_obj_get_attr(obj, name, dest); | |
10660 | } | |
10661 | ||
10662 | struct get_obj_data; | |
10663 | ||
10664 | struct get_obj_aio_data { | |
10665 | struct get_obj_data *op_data; | |
10666 | off_t ofs; | |
10667 | off_t len; | |
10668 | }; | |
10669 | ||
10670 | struct get_obj_io { | |
10671 | off_t len; | |
10672 | bufferlist bl; | |
10673 | }; | |
10674 | ||
10675 | static void _get_obj_aio_completion_cb(completion_t cb, void *arg); | |
10676 | ||
10677 | struct get_obj_data : public RefCountedObject { | |
10678 | CephContext *cct; | |
10679 | RGWRados *rados; | |
10680 | RGWObjectCtx *ctx; | |
10681 | IoCtx io_ctx; | |
10682 | map<off_t, get_obj_io> io_map; | |
10683 | map<off_t, librados::AioCompletion *> completion_map; | |
10684 | uint64_t total_read; | |
10685 | Mutex lock; | |
10686 | Mutex data_lock; | |
10687 | list<get_obj_aio_data> aio_data; | |
10688 | RGWGetDataCB *client_cb; | |
10689 | std::atomic<bool> cancelled = { false }; | |
10690 | std::atomic<int64_t> err_code = { 0 }; | |
10691 | Throttle throttle; | |
10692 | list<bufferlist> read_list; | |
10693 | ||
10694 | explicit get_obj_data(CephContext *_cct) | |
10695 | : cct(_cct), | |
10696 | rados(NULL), ctx(NULL), | |
10697 | total_read(0), lock("get_obj_data"), data_lock("get_obj_data::data_lock"), | |
10698 | client_cb(NULL), | |
10699 | throttle(cct, "get_obj_data", cct->_conf->rgw_get_obj_window_size, false) {} | |
10700 | ~get_obj_data() override { } | |
10701 | void set_cancelled(int r) { | |
10702 | cancelled = true; | |
10703 | err_code = r; | |
10704 | } | |
10705 | ||
10706 | bool is_cancelled() { | |
10707 | return cancelled; | |
10708 | } | |
10709 | ||
10710 | int get_err_code() { | |
10711 | return err_code; | |
10712 | } | |
10713 | ||
10714 | int wait_next_io(bool *done) { | |
10715 | lock.Lock(); | |
10716 | map<off_t, librados::AioCompletion *>::iterator iter = completion_map.begin(); | |
10717 | if (iter == completion_map.end()) { | |
10718 | *done = true; | |
10719 | lock.Unlock(); | |
10720 | return 0; | |
10721 | } | |
10722 | off_t cur_ofs = iter->first; | |
10723 | librados::AioCompletion *c = iter->second; | |
10724 | lock.Unlock(); | |
10725 | ||
10726 | c->wait_for_safe_and_cb(); | |
10727 | int r = c->get_return_value(); | |
10728 | ||
10729 | lock.Lock(); | |
10730 | completion_map.erase(cur_ofs); | |
10731 | ||
10732 | if (completion_map.empty()) { | |
10733 | *done = true; | |
10734 | } | |
10735 | lock.Unlock(); | |
10736 | ||
10737 | c->release(); | |
10738 | ||
10739 | return r; | |
10740 | } | |
10741 | ||
10742 | void add_io(off_t ofs, off_t len, bufferlist **pbl, AioCompletion **pc) { | |
10743 | Mutex::Locker l(lock); | |
10744 | ||
10745 | const auto& io_iter = io_map.insert( | |
10746 | map<off_t, get_obj_io>::value_type(ofs, get_obj_io())); | |
10747 | ||
10748 | assert(io_iter.second); // assert new insertion | |
10749 | ||
10750 | get_obj_io& io = (io_iter.first)->second; | |
10751 | *pbl = &io.bl; | |
10752 | ||
10753 | struct get_obj_aio_data aio; | |
10754 | aio.ofs = ofs; | |
10755 | aio.len = len; | |
10756 | aio.op_data = this; | |
10757 | ||
10758 | aio_data.push_back(aio); | |
10759 | ||
10760 | struct get_obj_aio_data *paio_data = &aio_data.back(); /* last element */ | |
10761 | ||
10762 | librados::AioCompletion *c = librados::Rados::aio_create_completion((void *)paio_data, NULL, _get_obj_aio_completion_cb); | |
10763 | completion_map[ofs] = c; | |
10764 | ||
10765 | *pc = c; | |
10766 | ||
10767 | /* we have a reference per IO, plus one reference for the calling function. | |
10768 | * reference is dropped for each callback, plus when we're done iterating | |
10769 | * over the parts */ | |
10770 | get(); | |
10771 | } | |
10772 | ||
10773 | void cancel_io(off_t ofs) { | |
10774 | ldout(cct, 20) << "get_obj_data::cancel_io() ofs=" << ofs << dendl; | |
10775 | lock.Lock(); | |
10776 | map<off_t, AioCompletion *>::iterator iter = completion_map.find(ofs); | |
10777 | if (iter != completion_map.end()) { | |
10778 | AioCompletion *c = iter->second; | |
10779 | c->release(); | |
10780 | completion_map.erase(ofs); | |
10781 | io_map.erase(ofs); | |
10782 | } | |
10783 | lock.Unlock(); | |
10784 | ||
10785 | /* we don't drop a reference here -- e.g., not calling d->put(), because we still | |
10786 | * need IoCtx to live, as io callback may still be called | |
10787 | */ | |
10788 | } | |
10789 | ||
10790 | void cancel_all_io() { | |
10791 | ldout(cct, 20) << "get_obj_data::cancel_all_io()" << dendl; | |
10792 | Mutex::Locker l(lock); | |
10793 | for (map<off_t, librados::AioCompletion *>::iterator iter = completion_map.begin(); | |
10794 | iter != completion_map.end(); ++iter) { | |
10795 | librados::AioCompletion *c = iter->second; | |
10796 | c->release(); | |
10797 | } | |
10798 | } | |
10799 | ||
10800 | int get_complete_ios(off_t ofs, list<bufferlist>& bl_list) { | |
10801 | Mutex::Locker l(lock); | |
10802 | ||
10803 | map<off_t, get_obj_io>::iterator liter = io_map.begin(); | |
10804 | ||
10805 | if (liter == io_map.end() || | |
10806 | liter->first != ofs) { | |
10807 | return 0; | |
10808 | } | |
10809 | ||
10810 | map<off_t, librados::AioCompletion *>::iterator aiter; | |
10811 | aiter = completion_map.find(ofs); | |
10812 | if (aiter == completion_map.end()) { | |
10813 | /* completion map does not hold this io, it was cancelled */ | |
10814 | return 0; | |
10815 | } | |
10816 | ||
10817 | AioCompletion *completion = aiter->second; | |
10818 | int r = completion->get_return_value(); | |
10819 | if (r < 0) | |
10820 | return r; | |
10821 | ||
10822 | for (; aiter != completion_map.end(); ++aiter) { | |
10823 | completion = aiter->second; | |
10824 | if (!completion->is_safe()) { | |
10825 | /* reached a request that is not yet complete, stop */ | |
10826 | break; | |
10827 | } | |
10828 | ||
10829 | r = completion->get_return_value(); | |
10830 | if (r < 0) { | |
10831 | set_cancelled(r); /* mark it as cancelled, so that we don't continue processing next operations */ | |
10832 | return r; | |
10833 | } | |
10834 | ||
10835 | total_read += r; | |
10836 | ||
10837 | map<off_t, get_obj_io>::iterator old_liter = liter++; | |
10838 | bl_list.push_back(old_liter->second.bl); | |
10839 | io_map.erase(old_liter); | |
10840 | } | |
10841 | ||
10842 | return 0; | |
10843 | } | |
10844 | }; | |
10845 | ||
10846 | static int _get_obj_iterate_cb(const RGWBucketInfo& bucket_info, const rgw_obj& obj, const rgw_raw_obj& read_obj, off_t obj_ofs, off_t read_ofs, off_t len, bool is_head_obj, RGWObjState *astate, void *arg) | |
10847 | { | |
10848 | struct get_obj_data *d = (struct get_obj_data *)arg; | |
10849 | ||
10850 | return d->rados->get_obj_iterate_cb(d->ctx, astate, bucket_info, obj, read_obj, obj_ofs, read_ofs, len, is_head_obj, arg); | |
10851 | } | |
10852 | ||
10853 | static void _get_obj_aio_completion_cb(completion_t cb, void *arg) | |
10854 | { | |
10855 | struct get_obj_aio_data *aio_data = (struct get_obj_aio_data *)arg; | |
10856 | struct get_obj_data *d = aio_data->op_data; | |
10857 | ||
10858 | d->rados->get_obj_aio_completion_cb(cb, arg); | |
10859 | } | |
10860 | ||
10861 | ||
10862 | void RGWRados::get_obj_aio_completion_cb(completion_t c, void *arg) | |
10863 | { | |
10864 | struct get_obj_aio_data *aio_data = (struct get_obj_aio_data *)arg; | |
10865 | struct get_obj_data *d = aio_data->op_data; | |
10866 | off_t ofs = aio_data->ofs; | |
10867 | off_t len = aio_data->len; | |
10868 | ||
10869 | list<bufferlist> bl_list; | |
10870 | list<bufferlist>::iterator iter; | |
10871 | int r; | |
10872 | ||
10873 | ldout(cct, 20) << "get_obj_aio_completion_cb: io completion ofs=" << ofs << " len=" << len << dendl; | |
10874 | d->throttle.put(len); | |
10875 | ||
10876 | r = rados_aio_get_return_value(c); | |
10877 | if (r < 0) { | |
10878 | ldout(cct, 0) << "ERROR: got unexpected error when trying to read object: " << r << dendl; | |
10879 | d->set_cancelled(r); | |
10880 | goto done; | |
10881 | } | |
10882 | ||
10883 | if (d->is_cancelled()) { | |
10884 | goto done; | |
10885 | } | |
10886 | ||
10887 | d->data_lock.Lock(); | |
10888 | ||
10889 | r = d->get_complete_ios(ofs, bl_list); | |
10890 | if (r < 0) { | |
10891 | goto done_unlock; | |
10892 | } | |
10893 | ||
10894 | d->read_list.splice(d->read_list.end(), bl_list); | |
10895 | ||
10896 | done_unlock: | |
10897 | d->data_lock.Unlock(); | |
10898 | done: | |
10899 | d->put(); | |
10900 | return; | |
10901 | } | |
10902 | ||
10903 | int RGWRados::flush_read_list(struct get_obj_data *d) | |
10904 | { | |
10905 | d->data_lock.Lock(); | |
10906 | list<bufferlist> l; | |
10907 | l.swap(d->read_list); | |
10908 | d->get(); | |
10909 | d->read_list.clear(); | |
10910 | ||
10911 | d->data_lock.Unlock(); | |
10912 | ||
10913 | int r = 0; | |
10914 | ||
10915 | list<bufferlist>::iterator iter; | |
10916 | for (iter = l.begin(); iter != l.end(); ++iter) { | |
10917 | bufferlist& bl = *iter; | |
10918 | r = d->client_cb->handle_data(bl, 0, bl.length()); | |
10919 | if (r < 0) { | |
10920 | dout(0) << "ERROR: flush_read_list(): d->client_cb->handle_data() returned " << r << dendl; | |
10921 | break; | |
10922 | } | |
10923 | } | |
10924 | ||
10925 | d->data_lock.Lock(); | |
10926 | d->put(); | |
10927 | if (r < 0) { | |
10928 | d->set_cancelled(r); | |
10929 | } | |
10930 | d->data_lock.Unlock(); | |
10931 | return r; | |
10932 | } | |
10933 | ||
10934 | int RGWRados::get_obj_iterate_cb(RGWObjectCtx *ctx, RGWObjState *astate, | |
10935 | const RGWBucketInfo& bucket_info, | |
10936 | const rgw_obj& obj, | |
10937 | const rgw_raw_obj& read_obj, | |
10938 | off_t obj_ofs, | |
10939 | off_t read_ofs, off_t len, | |
10940 | bool is_head_obj, void *arg) | |
10941 | { | |
10942 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
10943 | ObjectReadOperation op; | |
10944 | struct get_obj_data *d = (struct get_obj_data *)arg; | |
10945 | string oid, key; | |
10946 | bufferlist *pbl; | |
10947 | AioCompletion *c; | |
10948 | ||
10949 | int r; | |
10950 | ||
10951 | if (is_head_obj) { | |
10952 | /* only when reading from the head object do we need to do the atomic test */ | |
10953 | r = append_atomic_test(rctx, bucket_info, obj, op, &astate); | |
10954 | if (r < 0) | |
10955 | return r; | |
10956 | ||
10957 | if (astate && | |
10958 | obj_ofs < astate->data.length()) { | |
10959 | unsigned chunk_len = min((uint64_t)astate->data.length() - obj_ofs, (uint64_t)len); | |
10960 | ||
10961 | d->data_lock.Lock(); | |
10962 | r = d->client_cb->handle_data(astate->data, obj_ofs, chunk_len); | |
10963 | d->data_lock.Unlock(); | |
10964 | if (r < 0) | |
10965 | return r; | |
10966 | ||
10967 | d->lock.Lock(); | |
10968 | d->total_read += chunk_len; | |
10969 | d->lock.Unlock(); | |
10970 | ||
10971 | len -= chunk_len; | |
10972 | read_ofs += chunk_len; | |
10973 | obj_ofs += chunk_len; | |
10974 | if (!len) | |
10975 | return 0; | |
10976 | } | |
10977 | } | |
10978 | ||
10979 | d->throttle.get(len); | |
10980 | if (d->is_cancelled()) { | |
10981 | return d->get_err_code(); | |
10982 | } | |
10983 | ||
10984 | /* add io after we check that we're not cancelled, otherwise we're going to have trouble | |
10985 | * cleaning up | |
10986 | */ | |
10987 | d->add_io(obj_ofs, len, &pbl, &c); | |
10988 | ||
10989 | ldout(cct, 20) << "rados->get_obj_iterate_cb oid=" << read_obj.oid << " obj-ofs=" << obj_ofs << " read_ofs=" << read_ofs << " len=" << len << dendl; | |
10990 | op.read(read_ofs, len, pbl, NULL); | |
10991 | ||
10992 | librados::IoCtx io_ctx(d->io_ctx); | |
10993 | io_ctx.locator_set_key(read_obj.loc); | |
10994 | ||
10995 | r = io_ctx.aio_operate(read_obj.oid, c, &op, NULL); | |
10996 | if (r < 0) { | |
10997 | ldout(cct, 0) << "rados->aio_operate r=" << r << dendl; | |
10998 | goto done_err; | |
10999 | } | |
11000 | ||
11001 | // Flush data to client if there is any | |
11002 | r = flush_read_list(d); | |
11003 | if (r < 0) | |
11004 | return r; | |
11005 | ||
11006 | return 0; | |
11007 | ||
11008 | done_err: | |
11009 | ldout(cct, 20) << "cancelling io r=" << r << " obj_ofs=" << obj_ofs << dendl; | |
11010 | d->set_cancelled(r); | |
11011 | d->cancel_io(obj_ofs); | |
11012 | ||
11013 | return r; | |
11014 | } | |
11015 | ||
11016 | int RGWRados::Object::Read::iterate(int64_t ofs, int64_t end, RGWGetDataCB *cb) | |
11017 | { | |
11018 | RGWRados *store = source->get_store(); | |
11019 | CephContext *cct = store->ctx(); | |
11020 | ||
11021 | struct get_obj_data *data = new get_obj_data(cct); | |
11022 | bool done = false; | |
11023 | ||
11024 | RGWObjectCtx& obj_ctx = source->get_ctx(); | |
11025 | ||
11026 | data->rados = store; | |
11027 | data->io_ctx.dup(state.io_ctx); | |
11028 | data->client_cb = cb; | |
11029 | ||
11030 | int r = store->iterate_obj(obj_ctx, source->get_bucket_info(), state.obj, ofs, end, cct->_conf->rgw_get_obj_max_req_size, _get_obj_iterate_cb, (void *)data); | |
11031 | if (r < 0) { | |
11032 | data->cancel_all_io(); | |
11033 | goto done; | |
11034 | } | |
11035 | ||
11036 | while (!done) { | |
11037 | r = data->wait_next_io(&done); | |
11038 | if (r < 0) { | |
11039 | dout(10) << "get_obj_iterate() r=" << r << ", canceling all io" << dendl; | |
11040 | data->cancel_all_io(); | |
11041 | break; | |
11042 | } | |
11043 | r = store->flush_read_list(data); | |
11044 | if (r < 0) { | |
11045 | dout(10) << "get_obj_iterate() r=" << r << ", canceling all io" << dendl; | |
11046 | data->cancel_all_io(); | |
11047 | break; | |
11048 | } | |
11049 | } | |
11050 | ||
11051 | done: | |
11052 | data->put(); | |
11053 | return r; | |
11054 | } | |
11055 | ||
11056 | int RGWRados::iterate_obj(RGWObjectCtx& obj_ctx, | |
11057 | const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
11058 | off_t ofs, off_t end, | |
11059 | uint64_t max_chunk_size, | |
11060 | int (*iterate_obj_cb)(const RGWBucketInfo&, const rgw_obj& obj, | |
11061 | const rgw_raw_obj&, off_t, off_t, off_t, bool, | |
11062 | RGWObjState *, void *), | |
11063 | void *arg) | |
11064 | { | |
11065 | rgw_raw_obj head_obj; | |
11066 | rgw_raw_obj read_obj; | |
11067 | uint64_t read_ofs = ofs; | |
11068 | uint64_t len; | |
11069 | bool reading_from_head = true; | |
11070 | RGWObjState *astate = NULL; | |
11071 | ||
11072 | obj_to_raw(bucket_info.placement_rule, obj, &head_obj); | |
11073 | ||
11074 | int r = get_obj_state(&obj_ctx, bucket_info, obj, &astate, false); | |
11075 | if (r < 0) { | |
11076 | return r; | |
11077 | } | |
11078 | ||
11079 | if (end < 0) | |
11080 | len = 0; | |
11081 | else | |
11082 | len = end - ofs + 1; | |
11083 | ||
11084 | if (astate->has_manifest) { | |
11085 | /* now get the relevant object stripe */ | |
11086 | RGWObjManifest::obj_iterator iter = astate->manifest.obj_find(ofs); | |
11087 | ||
11088 | RGWObjManifest::obj_iterator obj_end = astate->manifest.obj_end(); | |
11089 | ||
11090 | for (; iter != obj_end && ofs <= end; ++iter) { | |
11091 | off_t stripe_ofs = iter.get_stripe_ofs(); | |
11092 | off_t next_stripe_ofs = stripe_ofs + iter.get_stripe_size(); | |
11093 | ||
11094 | while (ofs < next_stripe_ofs && ofs <= end) { | |
11095 | read_obj = iter.get_location().get_raw_obj(this); | |
11096 | uint64_t read_len = min(len, iter.get_stripe_size() - (ofs - stripe_ofs)); | |
11097 | read_ofs = iter.location_ofs() + (ofs - stripe_ofs); | |
11098 | ||
11099 | if (read_len > max_chunk_size) { | |
11100 | read_len = max_chunk_size; | |
11101 | } | |
11102 | ||
11103 | reading_from_head = (read_obj == head_obj); | |
11104 | r = iterate_obj_cb(bucket_info, obj, read_obj, ofs, read_ofs, read_len, reading_from_head, astate, arg); | |
11105 | if (r < 0) { | |
11106 | return r; | |
11107 | } | |
11108 | ||
11109 | len -= read_len; | |
11110 | ofs += read_len; | |
11111 | } | |
11112 | } | |
11113 | } else { | |
11114 | while (ofs <= end) { | |
11115 | read_obj = head_obj; | |
11116 | uint64_t read_len = min(len, max_chunk_size); | |
11117 | ||
11118 | r = iterate_obj_cb(bucket_info, obj, read_obj, ofs, ofs, read_len, reading_from_head, astate, arg); | |
11119 | if (r < 0) { | |
11120 | return r; | |
11121 | } | |
11122 | ||
11123 | len -= read_len; | |
11124 | ofs += read_len; | |
11125 | } | |
11126 | } | |
11127 | ||
11128 | return 0; | |
11129 | } | |
11130 | ||
11131 | int RGWRados::obj_operate(const RGWBucketInfo& bucket_info, const rgw_obj& obj, ObjectWriteOperation *op) | |
11132 | { | |
11133 | rgw_rados_ref ref; | |
11134 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
11135 | if (r < 0) { | |
11136 | return r; | |
11137 | } | |
11138 | ||
11139 | return ref.ioctx.operate(ref.oid, op); | |
11140 | } | |
11141 | ||
11142 | int RGWRados::obj_operate(const RGWBucketInfo& bucket_info, const rgw_obj& obj, ObjectReadOperation *op) | |
11143 | { | |
11144 | rgw_rados_ref ref; | |
11145 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
11146 | if (r < 0) { | |
11147 | return r; | |
11148 | } | |
11149 | ||
11150 | bufferlist outbl; | |
11151 | ||
11152 | return ref.ioctx.operate(ref.oid, op, &outbl); | |
11153 | } | |
11154 | ||
11155 | int RGWRados::olh_init_modification_impl(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& olh_obj, string *op_tag) | |
11156 | { | |
11157 | ObjectWriteOperation op; | |
11158 | ||
11159 | assert(olh_obj.key.instance.empty()); | |
11160 | ||
11161 | bool has_tag = (state.exists && has_olh_tag(state.attrset)); | |
11162 | ||
11163 | if (!state.exists) { | |
11164 | op.create(true); | |
11165 | } else { | |
11166 | op.assert_exists(); | |
11167 | struct timespec mtime_ts = real_clock::to_timespec(state.mtime); | |
11168 | op.mtime2(&mtime_ts); | |
11169 | } | |
11170 | ||
11171 | /* | |
11172 | * 3 possible cases: olh object doesn't exist, it exists as an olh, it exists as a regular object. | |
11173 | * If it exists as a regular object we'll need to transform it into an olh. We'll do it in two | |
11174 | * steps, first change its tag and set the olh pending attrs. Once write is done we'll need to | |
11175 | * truncate it, remove extra attrs, and send it to the garbage collection. The bucket index olh | |
11176 | * log will reflect that. | |
11177 | * | |
11178 | * Need to generate separate olh and obj tags, as olh can be colocated with object data. obj_tag | |
11179 | * is used for object data instance, olh_tag for olh instance. | |
11180 | */ | |
11181 | if (has_tag) { | |
11182 | /* guard against racing writes */ | |
11183 | bucket_index_guard_olh_op(state, op); | |
11184 | } | |
11185 | ||
11186 | if (!has_tag) { | |
11187 | /* obj tag */ | |
11188 | string obj_tag; | |
11189 | int ret = gen_rand_alphanumeric_lower(cct, &obj_tag, 32); | |
11190 | if (ret < 0) { | |
11191 | ldout(cct, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret << dendl; | |
11192 | return ret; | |
11193 | } | |
11194 | bufferlist bl; | |
11195 | bl.append(obj_tag.c_str(), obj_tag.size()); | |
11196 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
11197 | ||
11198 | state.attrset[RGW_ATTR_ID_TAG] = bl; | |
11199 | state.obj_tag = bl; | |
11200 | ||
11201 | /* olh tag */ | |
11202 | string olh_tag; | |
11203 | ret = gen_rand_alphanumeric_lower(cct, &olh_tag, 32); | |
11204 | if (ret < 0) { | |
11205 | ldout(cct, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret << dendl; | |
11206 | return ret; | |
11207 | } | |
11208 | bufferlist olh_bl; | |
11209 | olh_bl.append(olh_tag.c_str(), olh_tag.size()); | |
11210 | op.setxattr(RGW_ATTR_OLH_ID_TAG, olh_bl); | |
11211 | ||
11212 | state.attrset[RGW_ATTR_OLH_ID_TAG] = olh_bl; | |
11213 | state.olh_tag = olh_bl; | |
11214 | state.is_olh = true; | |
11215 | ||
11216 | bufferlist verbl; | |
11217 | op.setxattr(RGW_ATTR_OLH_VER, verbl); | |
11218 | } | |
11219 | ||
11220 | bufferlist bl; | |
11221 | RGWOLHPendingInfo pending_info; | |
11222 | pending_info.time = real_clock::now(); | |
11223 | ::encode(pending_info, bl); | |
11224 | ||
11225 | #define OLH_PENDING_TAG_LEN 32 | |
11226 | /* tag will start with current time epoch, this so that entries are sorted by time */ | |
11227 | char buf[32]; | |
11228 | utime_t ut(pending_info.time); | |
11229 | snprintf(buf, sizeof(buf), "%016llx", (unsigned long long)ut.sec()); | |
11230 | *op_tag = buf; | |
11231 | ||
11232 | string s; | |
11233 | int ret = gen_rand_alphanumeric_lower(cct, &s, OLH_PENDING_TAG_LEN - op_tag->size()); | |
11234 | if (ret < 0) { | |
11235 | ldout(cct, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret << dendl; | |
11236 | return ret; | |
11237 | } | |
11238 | op_tag->append(s); | |
11239 | ||
11240 | string attr_name = RGW_ATTR_OLH_PENDING_PREFIX; | |
11241 | attr_name.append(*op_tag); | |
11242 | ||
11243 | op.setxattr(attr_name.c_str(), bl); | |
11244 | ||
11245 | ret = obj_operate(bucket_info, olh_obj, &op); | |
11246 | if (ret < 0) { | |
11247 | return ret; | |
11248 | } | |
11249 | ||
11250 | state.exists = true; | |
11251 | state.attrset[attr_name] = bl; | |
11252 | ||
11253 | return 0; | |
11254 | } | |
11255 | ||
11256 | int RGWRados::olh_init_modification(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj, string *op_tag) | |
11257 | { | |
11258 | int ret; | |
11259 | ||
11260 | ret = olh_init_modification_impl(bucket_info, state, obj, op_tag); | |
11261 | if (ret == -EEXIST) { | |
11262 | ret = -ECANCELED; | |
11263 | } | |
11264 | ||
11265 | return ret; | |
11266 | } | |
11267 | ||
11268 | int RGWRados::guard_reshard(BucketShard *bs, | |
11269 | const rgw_obj& obj_instance, | |
11270 | const RGWBucketInfo& bucket_info, | |
11271 | std::function<int(BucketShard *)> call) | |
11272 | { | |
11273 | rgw_obj obj; | |
11274 | const rgw_obj *pobj = &obj_instance; | |
11275 | int r; | |
11276 | ||
11277 | for (int i = 0; i < NUM_RESHARD_RETRIES; ++i) { | |
11278 | r = bs->init(pobj->bucket, *pobj, nullptr /* no RGWBucketInfo */); | |
11279 | if (r < 0) { | |
11280 | ldout(cct, 5) << "bs.init() returned ret=" << r << dendl; | |
11281 | return r; | |
11282 | } | |
11283 | r = call(bs); | |
11284 | if (r != -ERR_BUSY_RESHARDING) { | |
11285 | break; | |
11286 | } | |
11287 | ldout(cct, 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl; | |
11288 | string new_bucket_id; | |
11289 | r = block_while_resharding(bs, &new_bucket_id, bucket_info); | |
11290 | if (r == -ERR_BUSY_RESHARDING) { | |
11291 | continue; | |
11292 | } | |
11293 | if (r < 0) { | |
11294 | return r; | |
11295 | } | |
11296 | ldout(cct, 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id << dendl; | |
11297 | i = 0; /* resharding is finished, make sure we can retry */ | |
11298 | ||
11299 | obj = *pobj; | |
11300 | obj.bucket.update_bucket_id(new_bucket_id); | |
11301 | pobj = &obj; | |
11302 | } | |
11303 | ||
11304 | if (r < 0) { | |
11305 | return r; | |
11306 | } | |
11307 | ||
11308 | return 0; | |
11309 | } | |
11310 | ||
11311 | int RGWRados::block_while_resharding(RGWRados::BucketShard *bs, | |
11312 | string *new_bucket_id, | |
11313 | const RGWBucketInfo& bucket_info) | |
11314 | { | |
11315 | std::shared_ptr<RGWReshardWait> waiter = reshard_wait; | |
11316 | ||
11317 | return waiter->block_while_resharding(bs, new_bucket_id, bucket_info); | |
11318 | } | |
11319 | ||
11320 | int RGWRados::bucket_index_link_olh(const RGWBucketInfo& bucket_info, RGWObjState& olh_state, const rgw_obj& obj_instance, | |
11321 | bool delete_marker, | |
11322 | const string& op_tag, | |
11323 | struct rgw_bucket_dir_entry_meta *meta, | |
11324 | uint64_t olh_epoch, | |
11325 | real_time unmod_since, bool high_precision_time, | |
11326 | rgw_zone_set *_zones_trace, bool log_data_change) | |
11327 | { | |
11328 | rgw_rados_ref ref; | |
11329 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11330 | if (r < 0) { | |
11331 | return r; | |
11332 | } | |
11333 | ||
11334 | rgw_zone_set zones_trace; | |
11335 | if (_zones_trace) { | |
11336 | zones_trace = *_zones_trace; | |
11337 | } | |
11338 | zones_trace.insert(get_zone().id); | |
11339 | ||
11340 | BucketShard bs(this); | |
11341 | ||
11342 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), obj_instance.key.instance); | |
11343 | r = guard_reshard(&bs, obj_instance, bucket_info, | |
11344 | [&](BucketShard *bs) -> int { | |
11345 | librados::ObjectWriteOperation op; | |
11346 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11347 | return cls_rgw_bucket_link_olh(bs->index_ctx, op, | |
11348 | bs->bucket_obj, key, olh_state.olh_tag, delete_marker, op_tag, meta, olh_epoch, | |
11349 | unmod_since, high_precision_time, | |
11350 | get_zone().log_data, zones_trace); | |
11351 | }); | |
11352 | if (r < 0) { | |
11353 | ldout(cct, 20) << "cls_rgw_bucket_link_olh() returned r=" << r << dendl; | |
11354 | return r; | |
11355 | } | |
11356 | ||
11357 | if (log_data_change && bucket_info.datasync_flag_enabled()) { | |
11358 | data_log->add_entry(bs.bucket, bs.shard_id); | |
11359 | } | |
11360 | ||
11361 | return 0; | |
11362 | } | |
11363 | ||
11364 | void RGWRados::bucket_index_guard_olh_op(RGWObjState& olh_state, ObjectOperation& op) | |
11365 | { | |
11366 | ldout(cct, 20) << __func__ << "(): olh_state.olh_tag=" << string(olh_state.olh_tag.c_str(), olh_state.olh_tag.length()) << dendl; | |
11367 | op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_state.olh_tag); | |
11368 | } | |
11369 | ||
11370 | int RGWRados::bucket_index_unlink_instance(const RGWBucketInfo& bucket_info, const rgw_obj& obj_instance, | |
11371 | const string& op_tag, const string& olh_tag, uint64_t olh_epoch, rgw_zone_set *_zones_trace) | |
11372 | { | |
11373 | rgw_rados_ref ref; | |
11374 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11375 | if (r < 0) { | |
11376 | return r; | |
11377 | } | |
11378 | ||
11379 | rgw_zone_set zones_trace; | |
11380 | if (_zones_trace) { | |
11381 | zones_trace = *_zones_trace; | |
11382 | } | |
11383 | zones_trace.insert(get_zone().id); | |
11384 | ||
11385 | BucketShard bs(this); | |
11386 | ||
11387 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), obj_instance.key.instance); | |
11388 | r = guard_reshard(&bs, obj_instance, bucket_info, | |
11389 | [&](BucketShard *bs) -> int { | |
11390 | librados::ObjectWriteOperation op; | |
11391 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11392 | return cls_rgw_bucket_unlink_instance(bs->index_ctx, op, bs->bucket_obj, key, op_tag, | |
11393 | olh_tag, olh_epoch, get_zone().log_data, zones_trace); | |
11394 | }); | |
11395 | if (r < 0) { | |
11396 | ldout(cct, 20) << "cls_rgw_bucket_link_olh() returned r=" << r << dendl; | |
11397 | return r; | |
11398 | } | |
11399 | ||
11400 | return 0; | |
11401 | } | |
11402 | ||
11403 | int RGWRados::bucket_index_read_olh_log(const RGWBucketInfo& bucket_info, RGWObjState& state, | |
11404 | const rgw_obj& obj_instance, uint64_t ver_marker, | |
11405 | map<uint64_t, vector<rgw_bucket_olh_log_entry> > *log, | |
11406 | bool *is_truncated) | |
11407 | { | |
11408 | rgw_rados_ref ref; | |
11409 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11410 | if (r < 0) { | |
11411 | return r; | |
11412 | } | |
11413 | ||
11414 | BucketShard bs(this); | |
11415 | int ret = | |
11416 | bs.init(obj_instance.bucket, obj_instance, nullptr /* no RGWBucketInfo */); | |
11417 | if (ret < 0) { | |
11418 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
11419 | return ret; | |
11420 | } | |
11421 | ||
11422 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
11423 | ||
11424 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
11425 | ||
11426 | ret = guard_reshard(&bs, obj_instance, bucket_info, | |
11427 | [&](BucketShard *bs) -> int { | |
11428 | ObjectReadOperation op; | |
11429 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11430 | return cls_rgw_get_olh_log(bs->index_ctx, bs->bucket_obj, op, | |
11431 | key, ver_marker, olh_tag, log, is_truncated); | |
11432 | }); | |
11433 | if (ret < 0) { | |
11434 | ldout(cct, 20) << "cls_rgw_get_olh_log() returned r=" << r << dendl; | |
11435 | return ret; | |
11436 | } | |
11437 | ||
11438 | return 0; | |
11439 | } | |
11440 | ||
11441 | // a multisite sync bug resulted in the OLH head attributes being overwritten by | |
11442 | // the attributes from another zone, causing link_olh() to fail endlessly due to | |
11443 | // olh_tag mismatch. this attempts to detect this case and reconstruct the OLH | |
11444 | // attributes from the bucket index. see http://tracker.ceph.com/issues/37792 | |
11445 | int RGWRados::repair_olh(RGWObjState* state, const RGWBucketInfo& bucket_info, | |
11446 | const rgw_obj& obj) | |
11447 | { | |
11448 | // fetch the current olh entry from the bucket index | |
11449 | rgw_bucket_olh_entry olh; | |
11450 | int r = bi_get_olh(bucket_info, obj, &olh); | |
11451 | if (r < 0) { | |
11452 | ldout(cct, 0) << "repair_olh failed to read olh entry for " << obj << dendl; | |
11453 | return r; | |
11454 | } | |
11455 | if (olh.tag == state->olh_tag.to_str()) { // mismatch already resolved? | |
11456 | return 0; | |
11457 | } | |
11458 | ||
11459 | ldout(cct, 4) << "repair_olh setting olh_tag=" << olh.tag | |
11460 | << " key=" << olh.key << " delete_marker=" << olh.delete_marker << dendl; | |
11461 | ||
11462 | // rewrite OLH_ID_TAG and OLH_INFO from current olh | |
11463 | ObjectWriteOperation op; | |
11464 | // assert this is the same olh tag we think we're fixing | |
11465 | bucket_index_guard_olh_op(*state, op); | |
11466 | // preserve existing mtime | |
11467 | struct timespec mtime_ts = ceph::real_clock::to_timespec(state->mtime); | |
11468 | op.mtime2(&mtime_ts); | |
11469 | { | |
11470 | bufferlist bl; | |
11471 | bl.append(olh.tag.c_str(), olh.tag.size()); | |
11472 | op.setxattr(RGW_ATTR_OLH_ID_TAG, bl); | |
11473 | } | |
11474 | { | |
11475 | RGWOLHInfo info; | |
11476 | info.target = rgw_obj(bucket_info.bucket, olh.key); | |
11477 | info.removed = olh.delete_marker; | |
11478 | bufferlist bl; | |
11479 | encode(info, bl); | |
11480 | op.setxattr(RGW_ATTR_OLH_INFO, bl); | |
11481 | } | |
11482 | rgw_rados_ref ref; | |
11483 | r = get_obj_head_ref(bucket_info, obj, &ref); | |
11484 | if (r < 0) { | |
11485 | return r; | |
11486 | } | |
11487 | r = ref.ioctx.operate(ref.oid, &op); | |
11488 | if (r < 0) { | |
11489 | ldout(cct, 0) << "repair_olh failed to write olh attributes with " | |
11490 | << cpp_strerror(r) << dendl; | |
11491 | return r; | |
11492 | } | |
11493 | return 0; | |
11494 | } | |
11495 | ||
11496 | int RGWRados::bucket_index_trim_olh_log(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj_instance, uint64_t ver) | |
11497 | { | |
11498 | rgw_rados_ref ref; | |
11499 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11500 | if (r < 0) { | |
11501 | return r; | |
11502 | } | |
11503 | ||
11504 | BucketShard bs(this); | |
11505 | int ret = | |
11506 | bs.init(obj_instance.bucket, obj_instance, nullptr /* no RGWBucketInfo */); | |
11507 | if (ret < 0) { | |
11508 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
11509 | return ret; | |
11510 | } | |
11511 | ||
11512 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
11513 | ||
11514 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
11515 | ||
11516 | ret = guard_reshard(&bs, obj_instance, bucket_info, | |
11517 | [&](BucketShard *pbs) -> int { | |
11518 | ObjectWriteOperation op; | |
11519 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11520 | cls_rgw_trim_olh_log(op, key, ver, olh_tag); | |
11521 | return pbs->index_ctx.operate(pbs->bucket_obj, &op); | |
11522 | }); | |
11523 | if (ret < 0) { | |
11524 | ldout(cct, 20) << "cls_rgw_trim_olh_log() returned r=" << ret << dendl; | |
11525 | return ret; | |
11526 | } | |
11527 | ||
11528 | return 0; | |
11529 | } | |
11530 | ||
11531 | int RGWRados::bucket_index_clear_olh(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj_instance) | |
11532 | { | |
11533 | rgw_rados_ref ref; | |
11534 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11535 | if (r < 0) { | |
11536 | return r; | |
11537 | } | |
11538 | ||
11539 | BucketShard bs(this); | |
11540 | ||
11541 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
11542 | ||
11543 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
11544 | ||
11545 | int ret = guard_reshard(&bs, obj_instance, bucket_info, | |
11546 | [&](BucketShard *pbs) -> int { | |
11547 | ObjectWriteOperation op; | |
11548 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11549 | return cls_rgw_clear_olh(pbs->index_ctx, op, pbs->bucket_obj, key, olh_tag); | |
11550 | }); | |
11551 | if (ret < 0) { | |
11552 | ldout(cct, 5) << "cls_rgw_clear_olh() returned ret=" << ret << dendl; | |
11553 | return ret; | |
11554 | } | |
11555 | ||
11556 | return 0; | |
11557 | } | |
11558 | ||
11559 | int RGWRados::apply_olh_log(RGWObjectCtx& obj_ctx, RGWObjState& state, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
11560 | bufferlist& olh_tag, map<uint64_t, vector<rgw_bucket_olh_log_entry> >& log, | |
11561 | uint64_t *plast_ver, rgw_zone_set* zones_trace) | |
11562 | { | |
11563 | if (log.empty()) { | |
11564 | return 0; | |
11565 | } | |
11566 | ||
11567 | librados::ObjectWriteOperation op; | |
11568 | ||
11569 | uint64_t last_ver = log.rbegin()->first; | |
11570 | *plast_ver = last_ver; | |
11571 | ||
11572 | map<uint64_t, vector<rgw_bucket_olh_log_entry> >::iterator iter = log.begin(); | |
11573 | ||
11574 | op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_tag); | |
11575 | op.cmpxattr(RGW_ATTR_OLH_VER, CEPH_OSD_CMPXATTR_OP_GT, last_ver); | |
11576 | ||
11577 | bufferlist ver_bl; | |
11578 | string last_ver_s = to_string(last_ver); | |
11579 | ver_bl.append(last_ver_s.c_str(), last_ver_s.size()); | |
11580 | op.setxattr(RGW_ATTR_OLH_VER, ver_bl); | |
11581 | ||
11582 | struct timespec mtime_ts = real_clock::to_timespec(state.mtime); | |
11583 | op.mtime2(&mtime_ts); | |
11584 | ||
11585 | bool need_to_link = false; | |
11586 | cls_rgw_obj_key key; | |
11587 | bool delete_marker = false; | |
11588 | list<cls_rgw_obj_key> remove_instances; | |
11589 | bool need_to_remove = false; | |
11590 | ||
11591 | for (iter = log.begin(); iter != log.end(); ++iter) { | |
11592 | vector<rgw_bucket_olh_log_entry>::iterator viter = iter->second.begin(); | |
11593 | for (; viter != iter->second.end(); ++viter) { | |
11594 | rgw_bucket_olh_log_entry& entry = *viter; | |
11595 | ||
11596 | ldout(cct, 20) << "olh_log_entry: op=" << (int)entry.op | |
11597 | << " key=" << entry.key.name << "[" << entry.key.instance << "] " | |
11598 | << (entry.delete_marker ? "(delete)" : "") << dendl; | |
11599 | switch (entry.op) { | |
11600 | case CLS_RGW_OLH_OP_REMOVE_INSTANCE: | |
11601 | remove_instances.push_back(entry.key); | |
11602 | break; | |
11603 | case CLS_RGW_OLH_OP_LINK_OLH: | |
11604 | need_to_link = true; | |
11605 | need_to_remove = false; | |
11606 | key = entry.key; | |
11607 | delete_marker = entry.delete_marker; | |
11608 | break; | |
11609 | case CLS_RGW_OLH_OP_UNLINK_OLH: | |
11610 | need_to_remove = true; | |
11611 | need_to_link = false; | |
11612 | break; | |
11613 | default: | |
11614 | ldout(cct, 0) << "ERROR: apply_olh_log: invalid op: " << (int)entry.op << dendl; | |
11615 | return -EIO; | |
11616 | } | |
11617 | string attr_name = RGW_ATTR_OLH_PENDING_PREFIX; | |
11618 | attr_name.append(entry.op_tag); | |
11619 | op.rmxattr(attr_name.c_str()); | |
11620 | } | |
11621 | } | |
11622 | ||
11623 | rgw_rados_ref ref; | |
11624 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
11625 | if (r < 0) { | |
11626 | return r; | |
11627 | } | |
11628 | ||
11629 | const rgw_bucket& bucket = obj.bucket; | |
11630 | ||
11631 | if (need_to_link) { | |
11632 | rgw_obj target(bucket, key); | |
11633 | RGWOLHInfo info; | |
11634 | info.target = target; | |
11635 | info.removed = delete_marker; | |
11636 | bufferlist bl; | |
11637 | ::encode(info, bl); | |
11638 | op.setxattr(RGW_ATTR_OLH_INFO, bl); | |
11639 | } | |
11640 | ||
11641 | /* first remove object instances */ | |
11642 | for (list<cls_rgw_obj_key>::iterator liter = remove_instances.begin(); | |
11643 | liter != remove_instances.end(); ++liter) { | |
11644 | cls_rgw_obj_key& key = *liter; | |
11645 | rgw_obj obj_instance(bucket, key); | |
11646 | int ret = delete_obj(obj_ctx, bucket_info, obj_instance, 0, RGW_BILOG_FLAG_VERSIONED_OP, ceph::real_time(), zones_trace); | |
11647 | if (ret < 0 && ret != -ENOENT) { | |
11648 | ldout(cct, 0) << "ERROR: delete_obj() returned " << ret << " obj_instance=" << obj_instance << dendl; | |
11649 | return ret; | |
11650 | } | |
11651 | } | |
11652 | ||
11653 | /* update olh object */ | |
11654 | r = ref.ioctx.operate(ref.oid, &op); | |
11655 | if (r == -ECANCELED) { | |
11656 | r = 0; | |
11657 | } | |
11658 | if (r < 0) { | |
11659 | ldout(cct, 0) << "ERROR: could not apply olh update, r=" << r << dendl; | |
11660 | return r; | |
11661 | } | |
11662 | ||
11663 | r = bucket_index_trim_olh_log(bucket_info, state, obj, last_ver); | |
11664 | if (r < 0) { | |
11665 | ldout(cct, 0) << "ERROR: could not trim olh log, r=" << r << dendl; | |
11666 | return r; | |
11667 | } | |
11668 | ||
11669 | if (need_to_remove) { | |
11670 | ObjectWriteOperation rm_op; | |
11671 | ||
11672 | rm_op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_tag); | |
11673 | rm_op.cmpxattr(RGW_ATTR_OLH_VER, CEPH_OSD_CMPXATTR_OP_EQ, last_ver); | |
11674 | cls_obj_check_prefix_exist(rm_op, RGW_ATTR_OLH_PENDING_PREFIX, true); /* fail if found one of these, pending modification */ | |
11675 | rm_op.remove(); | |
11676 | ||
11677 | r = ref.ioctx.operate(ref.oid, &rm_op); | |
11678 | if (r == -ECANCELED) { | |
11679 | return 0; /* someone else won this race */ | |
11680 | } else { | |
11681 | /* | |
11682 | * only clear if was successful, otherwise we might clobber pending operations on this object | |
11683 | */ | |
11684 | r = bucket_index_clear_olh(bucket_info, state, obj); | |
11685 | if (r < 0) { | |
11686 | ldout(cct, 0) << "ERROR: could not clear bucket index olh entries r=" << r << dendl; | |
11687 | return r; | |
11688 | } | |
11689 | } | |
11690 | } | |
11691 | ||
11692 | return 0; | |
11693 | } | |
11694 | ||
11695 | /* | |
11696 | * read olh log and apply it | |
11697 | */ | |
11698 | int RGWRados::update_olh(RGWObjectCtx& obj_ctx, RGWObjState *state, const RGWBucketInfo& bucket_info, const rgw_obj& obj, rgw_zone_set *zones_trace) | |
11699 | { | |
11700 | map<uint64_t, vector<rgw_bucket_olh_log_entry> > log; | |
11701 | bool is_truncated; | |
11702 | uint64_t ver_marker = 0; | |
11703 | ||
11704 | do { | |
11705 | int ret = bucket_index_read_olh_log(bucket_info, *state, obj, ver_marker, &log, &is_truncated); | |
11706 | if (ret < 0) { | |
11707 | return ret; | |
11708 | } | |
11709 | ret = apply_olh_log(obj_ctx, *state, bucket_info, obj, state->olh_tag, log, &ver_marker, zones_trace); | |
11710 | if (ret < 0) { | |
11711 | return ret; | |
11712 | } | |
11713 | } while (is_truncated); | |
11714 | ||
11715 | return 0; | |
11716 | } | |
11717 | ||
11718 | int RGWRados::set_olh(RGWObjectCtx& obj_ctx, RGWBucketInfo& bucket_info, const rgw_obj& target_obj, bool delete_marker, rgw_bucket_dir_entry_meta *meta, | |
11719 | uint64_t olh_epoch, real_time unmod_since, bool high_precision_time, | |
11720 | rgw_zone_set *zones_trace, bool log_data_change) | |
11721 | { | |
11722 | string op_tag; | |
11723 | ||
11724 | rgw_obj olh_obj = target_obj; | |
11725 | olh_obj.key.instance.clear(); | |
11726 | ||
11727 | RGWObjState *state = NULL; | |
11728 | ||
11729 | int ret = 0; | |
11730 | int i; | |
11731 | ||
11732 | #define MAX_ECANCELED_RETRY 100 | |
11733 | for (i = 0; i < MAX_ECANCELED_RETRY; i++) { | |
11734 | if (ret == -ECANCELED) { | |
11735 | obj_ctx.obj.invalidate(olh_obj); | |
11736 | } | |
11737 | ||
11738 | ret = get_obj_state(&obj_ctx, bucket_info, olh_obj, &state, false); /* don't follow olh */ | |
11739 | if (ret < 0) { | |
11740 | return ret; | |
11741 | } | |
11742 | ||
11743 | ret = olh_init_modification(bucket_info, *state, olh_obj, &op_tag); | |
11744 | if (ret < 0) { | |
11745 | ldout(cct, 20) << "olh_init_modification() target_obj=" << target_obj << " delete_marker=" << (int)delete_marker << " returned " << ret << dendl; | |
11746 | if (ret == -ECANCELED) { | |
11747 | continue; | |
11748 | } | |
11749 | return ret; | |
11750 | } | |
11751 | ret = bucket_index_link_olh(bucket_info, *state, target_obj, delete_marker, | |
11752 | op_tag, meta, olh_epoch, unmod_since, high_precision_time, | |
11753 | zones_trace, log_data_change); | |
11754 | if (ret < 0) { | |
11755 | ldout(cct, 20) << "bucket_index_link_olh() target_obj=" << target_obj << " delete_marker=" << (int)delete_marker << " returned " << ret << dendl; | |
11756 | if (ret == -ECANCELED) { | |
11757 | // the bucket index rejected the link_olh() due to olh tag mismatch; | |
11758 | // attempt to reconstruct olh head attributes based on the bucket index | |
11759 | int r2 = repair_olh(state, bucket_info, olh_obj); | |
11760 | if (r2 < 0 && r2 != -ECANCELED) { | |
11761 | return r2; | |
11762 | } | |
11763 | continue; | |
11764 | } | |
11765 | return ret; | |
11766 | } | |
11767 | break; | |
11768 | } | |
11769 | ||
11770 | if (i == MAX_ECANCELED_RETRY) { | |
11771 | ldout(cct, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl; | |
11772 | return -EIO; | |
11773 | } | |
11774 | ||
11775 | ret = update_olh(obj_ctx, state, bucket_info, olh_obj); | |
11776 | if (ret == -ECANCELED) { /* already did what we needed, no need to retry, raced with another user */ | |
11777 | ret = 0; | |
11778 | } | |
11779 | if (ret < 0) { | |
11780 | ldout(cct, 20) << "update_olh() target_obj=" << target_obj << " returned " << ret << dendl; | |
11781 | return ret; | |
11782 | } | |
11783 | ||
11784 | return 0; | |
11785 | } | |
11786 | ||
11787 | int RGWRados::unlink_obj_instance(RGWObjectCtx& obj_ctx, RGWBucketInfo& bucket_info, const rgw_obj& target_obj, | |
11788 | uint64_t olh_epoch, rgw_zone_set *zones_trace) | |
11789 | { | |
11790 | string op_tag; | |
11791 | ||
11792 | rgw_obj olh_obj = target_obj; | |
11793 | olh_obj.key.instance.clear(); | |
11794 | ||
11795 | RGWObjState *state = NULL; | |
11796 | ||
11797 | int ret = 0; | |
11798 | int i; | |
11799 | ||
11800 | for (i = 0; i < MAX_ECANCELED_RETRY; i++) { | |
11801 | if (ret == -ECANCELED) { | |
11802 | obj_ctx.obj.invalidate(olh_obj); | |
11803 | } | |
11804 | ||
11805 | ret = get_obj_state(&obj_ctx, bucket_info, olh_obj, &state, false); /* don't follow olh */ | |
11806 | if (ret < 0) | |
11807 | return ret; | |
11808 | ||
11809 | ret = olh_init_modification(bucket_info, *state, olh_obj, &op_tag); | |
11810 | if (ret < 0) { | |
11811 | ldout(cct, 20) << "olh_init_modification() target_obj=" << target_obj << " returned " << ret << dendl; | |
11812 | if (ret == -ECANCELED) { | |
11813 | continue; | |
11814 | } | |
11815 | return ret; | |
11816 | } | |
11817 | ||
11818 | string olh_tag(state->olh_tag.c_str(), state->olh_tag.length()); | |
11819 | ||
11820 | ret = bucket_index_unlink_instance(bucket_info, target_obj, op_tag, olh_tag, olh_epoch, zones_trace); | |
11821 | if (ret < 0) { | |
11822 | ldout(cct, 20) << "bucket_index_unlink_instance() target_obj=" << target_obj << " returned " << ret << dendl; | |
11823 | if (ret == -ECANCELED) { | |
11824 | continue; | |
11825 | } | |
11826 | return ret; | |
11827 | } | |
11828 | break; | |
11829 | } | |
11830 | ||
11831 | if (i == MAX_ECANCELED_RETRY) { | |
11832 | ldout(cct, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl; | |
11833 | return -EIO; | |
11834 | } | |
11835 | ||
11836 | ret = update_olh(obj_ctx, state, bucket_info, olh_obj, zones_trace); | |
11837 | if (ret == -ECANCELED) { /* already did what we needed, no need to retry, raced with another user */ | |
11838 | return 0; | |
11839 | } | |
11840 | if (ret < 0) { | |
11841 | ldout(cct, 20) << "update_olh() target_obj=" << target_obj << " returned " << ret << dendl; | |
11842 | return ret; | |
11843 | } | |
11844 | ||
11845 | return 0; | |
11846 | } | |
11847 | ||
11848 | void RGWRados::gen_rand_obj_instance_name(rgw_obj *target_obj) | |
11849 | { | |
11850 | #define OBJ_INSTANCE_LEN 32 | |
11851 | char buf[OBJ_INSTANCE_LEN + 1]; | |
11852 | ||
11853 | gen_rand_alphanumeric_no_underscore(cct, buf, OBJ_INSTANCE_LEN); /* don't want it to get url escaped, | |
11854 | no underscore for instance name due to the way we encode the raw keys */ | |
11855 | ||
11856 | target_obj->key.set_instance(buf); | |
11857 | } | |
11858 | ||
11859 | static void filter_attrset(map<string, bufferlist>& unfiltered_attrset, const string& check_prefix, | |
11860 | map<string, bufferlist> *attrset) | |
11861 | { | |
11862 | attrset->clear(); | |
11863 | map<string, bufferlist>::iterator iter; | |
11864 | for (iter = unfiltered_attrset.lower_bound(check_prefix); | |
11865 | iter != unfiltered_attrset.end(); ++iter) { | |
11866 | if (!boost::algorithm::starts_with(iter->first, check_prefix)) | |
11867 | break; | |
11868 | (*attrset)[iter->first] = iter->second; | |
11869 | } | |
11870 | } | |
11871 | ||
11872 | int RGWRados::get_olh(const RGWBucketInfo& bucket_info, const rgw_obj& obj, RGWOLHInfo *olh) | |
11873 | { | |
11874 | map<string, bufferlist> unfiltered_attrset; | |
11875 | ||
11876 | ObjectReadOperation op; | |
11877 | op.getxattrs(&unfiltered_attrset, NULL); | |
11878 | ||
11879 | bufferlist outbl; | |
11880 | int r = obj_operate(bucket_info, obj, &op); | |
11881 | ||
11882 | if (r < 0) { | |
11883 | return r; | |
11884 | } | |
11885 | map<string, bufferlist> attrset; | |
11886 | ||
11887 | filter_attrset(unfiltered_attrset, RGW_ATTR_OLH_PREFIX, &attrset); | |
11888 | ||
11889 | map<string, bufferlist>::iterator iter = attrset.find(RGW_ATTR_OLH_INFO); | |
11890 | if (iter == attrset.end()) { /* not an olh */ | |
11891 | return -EINVAL; | |
11892 | } | |
11893 | ||
11894 | try { | |
11895 | bufferlist::iterator biter = iter->second.begin(); | |
11896 | ::decode(*olh, biter); | |
11897 | } catch (buffer::error& err) { | |
11898 | ldout(cct, 0) << "ERROR: failed to decode olh info" << dendl; | |
11899 | return -EIO; | |
11900 | } | |
11901 | ||
11902 | return 0; | |
11903 | } | |
11904 | ||
11905 | void RGWRados::check_pending_olh_entries(map<string, bufferlist>& pending_entries, | |
11906 | map<string, bufferlist> *rm_pending_entries) | |
11907 | { | |
11908 | map<string, bufferlist>::iterator iter = pending_entries.begin(); | |
11909 | ||
11910 | real_time now = real_clock::now(); | |
11911 | ||
11912 | while (iter != pending_entries.end()) { | |
11913 | bufferlist::iterator biter = iter->second.begin(); | |
11914 | RGWOLHPendingInfo pending_info; | |
11915 | try { | |
11916 | ::decode(pending_info, biter); | |
11917 | } catch (buffer::error& err) { | |
11918 | /* skipping bad entry, we could remove it but it might hide a bug */ | |
11919 | ldout(cct, 0) << "ERROR: failed to decode pending entry " << iter->first << dendl; | |
11920 | ++iter; | |
11921 | continue; | |
11922 | } | |
11923 | ||
11924 | map<string, bufferlist>::iterator cur_iter = iter; | |
11925 | ++iter; | |
11926 | if (now - pending_info.time >= make_timespan(cct->_conf->rgw_olh_pending_timeout_sec)) { | |
11927 | (*rm_pending_entries)[cur_iter->first] = cur_iter->second; | |
11928 | pending_entries.erase(cur_iter); | |
11929 | } else { | |
11930 | /* entries names are sorted by time (rounded to a second) */ | |
11931 | break; | |
11932 | } | |
11933 | } | |
11934 | } | |
11935 | ||
11936 | int RGWRados::remove_olh_pending_entries(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& olh_obj, map<string, bufferlist>& pending_attrs) | |
11937 | { | |
11938 | ObjectWriteOperation op; | |
11939 | ||
11940 | bucket_index_guard_olh_op(state, op); | |
11941 | ||
11942 | for (map<string, bufferlist>::iterator iter = pending_attrs.begin(); iter != pending_attrs.end(); ++iter) { | |
11943 | op.rmxattr(iter->first.c_str()); | |
11944 | } | |
11945 | ||
11946 | rgw_rados_ref ref; | |
11947 | int r = get_obj_head_ref(bucket_info, olh_obj, &ref); | |
11948 | if (r < 0) { | |
11949 | return r; | |
11950 | } | |
11951 | ||
11952 | /* update olh object */ | |
11953 | r = ref.ioctx.operate(ref.oid, &op); | |
11954 | if (r == -ENOENT || r == -ECANCELED) { | |
11955 | /* raced with some other change, shouldn't sweat about it */ | |
11956 | r = 0; | |
11957 | } | |
11958 | if (r < 0) { | |
11959 | ldout(cct, 0) << "ERROR: could not apply olh update, r=" << r << dendl; | |
11960 | return r; | |
11961 | } | |
11962 | ||
11963 | return 0; | |
11964 | } | |
11965 | ||
11966 | int RGWRados::follow_olh(const RGWBucketInfo& bucket_info, RGWObjectCtx& obj_ctx, RGWObjState *state, const rgw_obj& olh_obj, rgw_obj *target) | |
11967 | { | |
11968 | map<string, bufferlist> pending_entries; | |
11969 | filter_attrset(state->attrset, RGW_ATTR_OLH_PENDING_PREFIX, &pending_entries); | |
11970 | ||
11971 | map<string, bufferlist> rm_pending_entries; | |
11972 | check_pending_olh_entries(pending_entries, &rm_pending_entries); | |
11973 | ||
11974 | if (!rm_pending_entries.empty()) { | |
11975 | int ret = remove_olh_pending_entries(bucket_info, *state, olh_obj, rm_pending_entries); | |
11976 | if (ret < 0) { | |
11977 | ldout(cct, 20) << "ERROR: rm_pending_entries returned ret=" << ret << dendl; | |
11978 | return ret; | |
11979 | } | |
11980 | } | |
11981 | if (!pending_entries.empty()) { | |
11982 | ldout(cct, 20) << __func__ << "(): found pending entries, need to update_olh() on bucket=" << olh_obj.bucket << dendl; | |
11983 | ||
11984 | int ret = update_olh(obj_ctx, state, bucket_info, olh_obj); | |
11985 | if (ret < 0) { | |
11986 | return ret; | |
11987 | } | |
11988 | } | |
11989 | ||
11990 | map<string, bufferlist>::iterator iter = state->attrset.find(RGW_ATTR_OLH_INFO); | |
11991 | assert(iter != state->attrset.end()); | |
11992 | RGWOLHInfo olh; | |
11993 | try { | |
11994 | bufferlist::iterator biter = iter->second.begin(); | |
11995 | ::decode(olh, biter); | |
11996 | } catch (buffer::error& err) { | |
11997 | ldout(cct, 0) << "ERROR: failed to decode olh info" << dendl; | |
11998 | return -EIO; | |
11999 | } | |
12000 | ||
12001 | if (olh.removed) { | |
12002 | return -ENOENT; | |
12003 | } | |
12004 | ||
12005 | *target = olh.target; | |
12006 | ||
12007 | return 0; | |
12008 | } | |
12009 | ||
12010 | int RGWRados::raw_obj_stat(rgw_raw_obj& obj, uint64_t *psize, real_time *pmtime, uint64_t *epoch, | |
12011 | map<string, bufferlist> *attrs, bufferlist *first_chunk, | |
12012 | RGWObjVersionTracker *objv_tracker) | |
12013 | { | |
12014 | rgw_rados_ref ref; | |
12015 | int r = get_raw_obj_ref(obj, &ref); | |
12016 | if (r < 0) { | |
12017 | return r; | |
12018 | } | |
12019 | ||
12020 | map<string, bufferlist> unfiltered_attrset; | |
12021 | uint64_t size = 0; | |
12022 | struct timespec mtime_ts; | |
12023 | ||
12024 | ObjectReadOperation op; | |
12025 | if (objv_tracker) { | |
12026 | objv_tracker->prepare_op_for_read(&op); | |
12027 | } | |
12028 | if (attrs) { | |
12029 | op.getxattrs(&unfiltered_attrset, NULL); | |
12030 | } | |
12031 | if (psize || pmtime) { | |
12032 | op.stat2(&size, &mtime_ts, NULL); | |
12033 | } | |
12034 | if (first_chunk) { | |
12035 | op.read(0, cct->_conf->rgw_max_chunk_size, first_chunk, NULL); | |
12036 | } | |
12037 | bufferlist outbl; | |
12038 | r = ref.ioctx.operate(ref.oid, &op, &outbl); | |
12039 | ||
12040 | if (epoch) { | |
12041 | *epoch = ref.ioctx.get_last_version(); | |
12042 | } | |
12043 | ||
12044 | if (r < 0) | |
12045 | return r; | |
12046 | ||
12047 | if (psize) | |
12048 | *psize = size; | |
12049 | if (pmtime) | |
12050 | *pmtime = ceph::real_clock::from_timespec(mtime_ts); | |
12051 | if (attrs) { | |
12052 | filter_attrset(unfiltered_attrset, RGW_ATTR_PREFIX, attrs); | |
12053 | } | |
12054 | ||
12055 | return 0; | |
12056 | } | |
12057 | ||
12058 | int RGWRados::get_bucket_stats(RGWBucketInfo& bucket_info, int shard_id, string *bucket_ver, string *master_ver, | |
12059 | map<RGWObjCategory, RGWStorageStats>& stats, string *max_marker, bool *syncstopped) | |
12060 | { | |
12061 | vector<rgw_bucket_dir_header> headers; | |
12062 | map<int, string> bucket_instance_ids; | |
12063 | int r = cls_bucket_head(bucket_info, shard_id, headers, &bucket_instance_ids); | |
12064 | if (r < 0) { | |
12065 | return r; | |
12066 | } | |
12067 | ||
12068 | assert(headers.size() == bucket_instance_ids.size()); | |
12069 | ||
12070 | auto iter = headers.begin(); | |
12071 | map<int, string>::iterator viter = bucket_instance_ids.begin(); | |
12072 | BucketIndexShardsManager ver_mgr; | |
12073 | BucketIndexShardsManager master_ver_mgr; | |
12074 | BucketIndexShardsManager marker_mgr; | |
12075 | char buf[64]; | |
12076 | for(; iter != headers.end(); ++iter, ++viter) { | |
12077 | accumulate_raw_stats(*iter, stats); | |
12078 | snprintf(buf, sizeof(buf), "%lu", (unsigned long)iter->ver); | |
12079 | ver_mgr.add(viter->first, string(buf)); | |
12080 | snprintf(buf, sizeof(buf), "%lu", (unsigned long)iter->master_ver); | |
12081 | master_ver_mgr.add(viter->first, string(buf)); | |
12082 | if (shard_id >= 0) { | |
12083 | *max_marker = iter->max_marker; | |
12084 | } else { | |
12085 | marker_mgr.add(viter->first, iter->max_marker); | |
12086 | } | |
12087 | if (syncstopped != NULL) | |
12088 | *syncstopped = iter->syncstopped; | |
12089 | } | |
12090 | ver_mgr.to_string(bucket_ver); | |
12091 | master_ver_mgr.to_string(master_ver); | |
12092 | if (shard_id < 0) { | |
12093 | marker_mgr.to_string(max_marker); | |
12094 | } | |
12095 | return 0; | |
12096 | } | |
12097 | ||
12098 | int RGWRados::get_bi_log_status(RGWBucketInfo& bucket_info, int shard_id, | |
12099 | map<int, string>& markers) | |
12100 | { | |
12101 | vector<rgw_bucket_dir_header> headers; | |
12102 | map<int, string> bucket_instance_ids; | |
12103 | int r = cls_bucket_head(bucket_info, shard_id, headers, &bucket_instance_ids); | |
12104 | if (r < 0) | |
12105 | return r; | |
12106 | ||
12107 | assert(headers.size() == bucket_instance_ids.size()); | |
12108 | ||
12109 | auto iter = headers.begin(); | |
12110 | map<int, string>::iterator viter = bucket_instance_ids.begin(); | |
12111 | ||
12112 | for(; iter != headers.end(); ++iter, ++viter) { | |
12113 | if (shard_id >= 0) { | |
12114 | markers[shard_id] = iter->max_marker; | |
12115 | } else { | |
12116 | markers[viter->first] = iter->max_marker; | |
12117 | } | |
12118 | } | |
12119 | return 0; | |
12120 | } | |
12121 | ||
12122 | class RGWGetBucketStatsContext : public RGWGetDirHeader_CB { | |
12123 | RGWGetBucketStats_CB *cb; | |
12124 | uint32_t pendings; | |
12125 | map<RGWObjCategory, RGWStorageStats> stats; | |
12126 | int ret_code; | |
12127 | bool should_cb; | |
12128 | Mutex lock; | |
12129 | ||
12130 | public: | |
12131 | RGWGetBucketStatsContext(RGWGetBucketStats_CB *_cb, uint32_t _pendings) | |
12132 | : cb(_cb), pendings(_pendings), stats(), ret_code(0), should_cb(true), | |
12133 | lock("RGWGetBucketStatsContext") {} | |
12134 | ||
12135 | void handle_response(int r, rgw_bucket_dir_header& header) override { | |
12136 | Mutex::Locker l(lock); | |
12137 | if (should_cb) { | |
12138 | if ( r >= 0) { | |
12139 | accumulate_raw_stats(header, stats); | |
12140 | } else { | |
12141 | ret_code = r; | |
12142 | } | |
12143 | ||
12144 | // Are we all done? | |
12145 | if (--pendings == 0) { | |
12146 | if (!ret_code) { | |
12147 | cb->set_response(&stats); | |
12148 | } | |
12149 | cb->handle_response(ret_code); | |
12150 | cb->put(); | |
12151 | } | |
12152 | } | |
12153 | } | |
12154 | ||
12155 | void unset_cb() { | |
12156 | Mutex::Locker l(lock); | |
12157 | should_cb = false; | |
12158 | } | |
12159 | }; | |
12160 | ||
12161 | int RGWRados::get_bucket_stats_async(RGWBucketInfo& bucket_info, int shard_id, RGWGetBucketStats_CB *ctx) | |
12162 | { | |
12163 | int num_aio = 0; | |
12164 | RGWGetBucketStatsContext *get_ctx = new RGWGetBucketStatsContext(ctx, bucket_info.num_shards ? : 1); | |
12165 | assert(get_ctx); | |
12166 | int r = cls_bucket_head_async(bucket_info, shard_id, get_ctx, &num_aio); | |
12167 | if (r < 0) { | |
12168 | ctx->put(); | |
12169 | if (num_aio) { | |
12170 | get_ctx->unset_cb(); | |
12171 | } | |
12172 | } | |
12173 | get_ctx->put(); | |
12174 | return r; | |
12175 | } | |
12176 | ||
12177 | class RGWGetUserStatsContext : public RGWGetUserHeader_CB { | |
12178 | RGWGetUserStats_CB *cb; | |
12179 | ||
12180 | public: | |
12181 | explicit RGWGetUserStatsContext(RGWGetUserStats_CB * const cb) | |
12182 | : cb(cb) {} | |
12183 | ||
12184 | void handle_response(int r, cls_user_header& header) override { | |
12185 | const cls_user_stats& hs = header.stats; | |
12186 | if (r >= 0) { | |
12187 | RGWStorageStats stats; | |
12188 | ||
12189 | stats.size = hs.total_bytes; | |
12190 | stats.size_rounded = hs.total_bytes_rounded; | |
12191 | stats.num_objects = hs.total_entries; | |
12192 | ||
12193 | cb->set_response(stats); | |
12194 | } | |
12195 | ||
12196 | cb->handle_response(r); | |
12197 | ||
12198 | cb->put(); | |
12199 | } | |
12200 | }; | |
12201 | ||
12202 | int RGWRados::get_user_stats(const rgw_user& user, RGWStorageStats& stats) | |
12203 | { | |
12204 | string user_str = user.to_str(); | |
12205 | ||
12206 | cls_user_header header; | |
12207 | int r = cls_user_get_header(user_str, &header); | |
12208 | if (r < 0) | |
12209 | return r; | |
12210 | ||
12211 | const cls_user_stats& hs = header.stats; | |
12212 | ||
12213 | stats.size = hs.total_bytes; | |
12214 | stats.size_rounded = hs.total_bytes_rounded; | |
12215 | stats.num_objects = hs.total_entries; | |
12216 | ||
12217 | return 0; | |
12218 | } | |
12219 | ||
12220 | int RGWRados::get_user_stats_async(const rgw_user& user, RGWGetUserStats_CB *ctx) | |
12221 | { | |
12222 | string user_str = user.to_str(); | |
12223 | ||
12224 | RGWGetUserStatsContext *get_ctx = new RGWGetUserStatsContext(ctx); | |
12225 | int r = cls_user_get_header_async(user_str, get_ctx); | |
12226 | if (r < 0) { | |
12227 | ctx->put(); | |
12228 | delete get_ctx; | |
12229 | return r; | |
12230 | } | |
12231 | ||
12232 | return 0; | |
12233 | } | |
12234 | ||
12235 | void RGWRados::get_bucket_meta_oid(const rgw_bucket& bucket, string& oid) | |
12236 | { | |
12237 | oid = RGW_BUCKET_INSTANCE_MD_PREFIX + bucket.get_key(':'); | |
12238 | } | |
12239 | ||
12240 | void RGWRados::get_bucket_instance_obj(const rgw_bucket& bucket, rgw_raw_obj& obj) | |
12241 | { | |
12242 | if (!bucket.oid.empty()) { | |
12243 | obj.init(get_zone_params().domain_root, bucket.oid); | |
12244 | } else { | |
12245 | string oid; | |
12246 | get_bucket_meta_oid(bucket, oid); | |
12247 | obj.init(get_zone_params().domain_root, oid); | |
12248 | } | |
12249 | } | |
12250 | ||
12251 | int RGWRados::get_bucket_instance_info(RGWObjectCtx& obj_ctx, const string& meta_key, RGWBucketInfo& info, | |
12252 | real_time *pmtime, map<string, bufferlist> *pattrs) | |
12253 | { | |
12254 | size_t pos = meta_key.find(':'); | |
12255 | if (pos == string::npos) { | |
12256 | return -EINVAL; | |
12257 | } | |
12258 | string oid = RGW_BUCKET_INSTANCE_MD_PREFIX + meta_key; | |
12259 | rgw_bucket_instance_key_to_oid(oid); | |
12260 | ||
12261 | return get_bucket_instance_from_oid(obj_ctx, oid, info, pmtime, pattrs); | |
12262 | } | |
12263 | ||
12264 | int RGWRados::get_bucket_instance_info(RGWObjectCtx& obj_ctx, const rgw_bucket& bucket, RGWBucketInfo& info, | |
12265 | real_time *pmtime, map<string, bufferlist> *pattrs) | |
12266 | { | |
12267 | string oid; | |
12268 | if (bucket.oid.empty()) { | |
12269 | get_bucket_meta_oid(bucket, oid); | |
12270 | } else { | |
12271 | oid = bucket.oid; | |
12272 | } | |
12273 | ||
12274 | return get_bucket_instance_from_oid(obj_ctx, oid, info, pmtime, pattrs); | |
12275 | } | |
12276 | ||
12277 | int RGWRados::get_bucket_instance_from_oid(RGWObjectCtx& obj_ctx, const string& oid, RGWBucketInfo& info, | |
12278 | real_time *pmtime, map<string, bufferlist> *pattrs, | |
12279 | rgw_cache_entry_info *cache_info, | |
12280 | boost::optional<obj_version> refresh_version) | |
12281 | { | |
12282 | ldout(cct, 20) << "reading from " << get_zone_params().domain_root << ":" << oid << dendl; | |
12283 | ||
12284 | bufferlist epbl; | |
12285 | ||
12286 | int ret = rgw_get_system_obj(this, obj_ctx, get_zone_params().domain_root, | |
12287 | oid, epbl, &info.objv_tracker, pmtime, pattrs, | |
12288 | cache_info, refresh_version); | |
12289 | if (ret < 0) { | |
12290 | return ret; | |
12291 | } | |
12292 | ||
12293 | bufferlist::iterator iter = epbl.begin(); | |
12294 | try { | |
12295 | ::decode(info, iter); | |
12296 | } catch (buffer::error& err) { | |
12297 | ldout(cct, 0) << "ERROR: could not decode buffer info, caught buffer::error" << dendl; | |
12298 | return -EIO; | |
12299 | } | |
12300 | info.bucket.oid = oid; | |
12301 | return 0; | |
12302 | } | |
12303 | ||
12304 | int RGWRados::get_bucket_entrypoint_info(RGWObjectCtx& obj_ctx, | |
12305 | const string& tenant_name, | |
12306 | const string& bucket_name, | |
12307 | RGWBucketEntryPoint& entry_point, | |
12308 | RGWObjVersionTracker *objv_tracker, | |
12309 | real_time *pmtime, | |
12310 | map<string, bufferlist> *pattrs, | |
12311 | rgw_cache_entry_info *cache_info, | |
12312 | boost::optional<obj_version> refresh_version) | |
12313 | { | |
12314 | bufferlist bl; | |
12315 | string bucket_entry; | |
12316 | ||
12317 | rgw_make_bucket_entry_name(tenant_name, bucket_name, bucket_entry); | |
12318 | int ret = rgw_get_system_obj(this, obj_ctx, get_zone_params().domain_root, | |
12319 | bucket_entry, bl, objv_tracker, pmtime, pattrs, | |
12320 | cache_info, refresh_version); | |
12321 | if (ret < 0) { | |
12322 | return ret; | |
12323 | } | |
12324 | ||
12325 | bufferlist::iterator iter = bl.begin(); | |
12326 | try { | |
12327 | ::decode(entry_point, iter); | |
12328 | } catch (buffer::error& err) { | |
12329 | ldout(cct, 0) << "ERROR: could not decode buffer info, caught buffer::error" << dendl; | |
12330 | return -EIO; | |
12331 | } | |
12332 | return 0; | |
12333 | } | |
12334 | ||
12335 | int RGWRados::convert_old_bucket_info(RGWObjectCtx& obj_ctx, | |
12336 | const string& tenant_name, | |
12337 | const string& bucket_name) | |
12338 | { | |
12339 | RGWBucketEntryPoint entry_point; | |
12340 | real_time ep_mtime; | |
12341 | RGWObjVersionTracker ot; | |
12342 | map<string, bufferlist> attrs; | |
12343 | RGWBucketInfo info; | |
12344 | ||
12345 | ldout(cct, 10) << "RGWRados::convert_old_bucket_info(): bucket=" << bucket_name << dendl; | |
12346 | ||
12347 | int ret = get_bucket_entrypoint_info(obj_ctx, tenant_name, bucket_name, entry_point, &ot, &ep_mtime, &attrs); | |
12348 | if (ret < 0) { | |
12349 | ldout(cct, 0) << "ERROR: get_bucket_entrypoint_info() returned " << ret << " bucket=" << bucket_name << dendl; | |
12350 | return ret; | |
12351 | } | |
12352 | ||
12353 | if (!entry_point.has_bucket_info) { | |
12354 | /* already converted! */ | |
12355 | return 0; | |
12356 | } | |
12357 | ||
12358 | info = entry_point.old_bucket_info; | |
12359 | info.bucket.oid = bucket_name; | |
12360 | info.ep_objv = ot.read_version; | |
12361 | ||
12362 | ot.generate_new_write_ver(cct); | |
12363 | ||
12364 | ret = put_linked_bucket_info(info, false, ep_mtime, &ot.write_version, &attrs, true); | |
12365 | if (ret < 0) { | |
12366 | ldout(cct, 0) << "ERROR: failed to put_linked_bucket_info(): " << ret << dendl; | |
12367 | return ret; | |
12368 | } | |
12369 | ||
12370 | return 0; | |
12371 | } | |
12372 | ||
12373 | int RGWRados::_get_bucket_info(RGWObjectCtx& obj_ctx, | |
12374 | const string& tenant, | |
12375 | const string& bucket_name, | |
12376 | RGWBucketInfo& info, | |
12377 | real_time *pmtime, | |
12378 | map<string, bufferlist> *pattrs, | |
12379 | boost::optional<obj_version> refresh_version) | |
12380 | { | |
12381 | bucket_info_entry e; | |
12382 | string bucket_entry; | |
12383 | rgw_make_bucket_entry_name(tenant, bucket_name, bucket_entry); | |
12384 | ||
12385 | ||
12386 | if (binfo_cache->find(bucket_entry, &e)) { | |
12387 | if (refresh_version && | |
12388 | e.info.objv_tracker.read_version.compare(&(*refresh_version))) { | |
12389 | lderr(cct) << "WARNING: The bucket info cache is inconsistent. This is " | |
12390 | << "a failure that should be debugged. I am a nice machine, " | |
12391 | << "so I will try to recover." << dendl; | |
12392 | binfo_cache->invalidate(bucket_entry); | |
12393 | } | |
12394 | info = e.info; | |
12395 | if (pattrs) | |
12396 | *pattrs = e.attrs; | |
12397 | if (pmtime) | |
12398 | *pmtime = e.mtime; | |
12399 | return 0; | |
12400 | } | |
12401 | ||
12402 | RGWBucketEntryPoint entry_point; | |
12403 | real_time ep_mtime; | |
12404 | RGWObjVersionTracker ot; | |
12405 | rgw_cache_entry_info entry_cache_info; | |
12406 | int ret = get_bucket_entrypoint_info(obj_ctx, tenant, bucket_name, | |
12407 | entry_point, &ot, &ep_mtime, pattrs, | |
12408 | &entry_cache_info, refresh_version); | |
12409 | if (ret < 0) { | |
12410 | /* only init these fields */ | |
12411 | info.bucket.tenant = tenant; | |
12412 | info.bucket.name = bucket_name; | |
12413 | return ret; | |
12414 | } | |
12415 | ||
12416 | if (entry_point.has_bucket_info) { | |
12417 | info = entry_point.old_bucket_info; | |
12418 | info.bucket.oid = bucket_name; | |
12419 | info.bucket.tenant = tenant; | |
12420 | info.ep_objv = ot.read_version; | |
12421 | ldout(cct, 20) << "rgw_get_bucket_info: old bucket info, bucket=" << info.bucket << " owner " << info.owner << dendl; | |
12422 | return 0; | |
12423 | } | |
12424 | ||
12425 | /* data is in the bucket instance object, we need to get attributes from there, clear everything | |
12426 | * that we got | |
12427 | */ | |
12428 | if (pattrs) { | |
12429 | pattrs->clear(); | |
12430 | } | |
12431 | ||
12432 | ldout(cct, 20) << "rgw_get_bucket_info: bucket instance: " << entry_point.bucket << dendl; | |
12433 | ||
12434 | ||
12435 | /* read bucket instance info */ | |
12436 | ||
12437 | string oid; | |
12438 | get_bucket_meta_oid(entry_point.bucket, oid); | |
12439 | ||
12440 | rgw_cache_entry_info cache_info; | |
12441 | ||
12442 | ret = get_bucket_instance_from_oid(obj_ctx, oid, e.info, &e.mtime, &e.attrs, | |
12443 | &cache_info, refresh_version); | |
12444 | e.info.ep_objv = ot.read_version; | |
12445 | info = e.info; | |
12446 | if (ret < 0) { | |
12447 | lderr(cct) << "ERROR: get_bucket_instance_from_oid failed: " << ret << dendl; | |
12448 | info.bucket.tenant = tenant; | |
12449 | info.bucket.name = bucket_name; | |
12450 | // XXX and why return anything in case of an error anyway? | |
12451 | return ret; | |
12452 | } | |
12453 | ||
12454 | if (pmtime) | |
12455 | *pmtime = e.mtime; | |
12456 | if (pattrs) | |
12457 | *pattrs = e.attrs; | |
12458 | ||
12459 | list<rgw_cache_entry_info *> cache_info_entries; | |
12460 | cache_info_entries.push_back(&entry_cache_info); | |
12461 | cache_info_entries.push_back(&cache_info); | |
12462 | ||
12463 | ||
12464 | /* chain to both bucket entry point and bucket instance */ | |
12465 | if (!binfo_cache->put(this, bucket_entry, &e, cache_info_entries)) { | |
12466 | ldout(cct, 20) << "couldn't put binfo cache entry, might have raced with data changes" << dendl; | |
12467 | } | |
12468 | ||
12469 | if (refresh_version && | |
12470 | refresh_version->compare(&info.objv_tracker.read_version)) { | |
12471 | lderr(cct) << "WARNING: The OSD has the same version I have. Something may " | |
12472 | << "have gone squirrelly. An administrator may have forced a " | |
12473 | << "change; otherwise there is a problem somewhere." << dendl; | |
12474 | } | |
12475 | ||
12476 | return 0; | |
12477 | } | |
12478 | ||
12479 | int RGWRados::get_bucket_info(RGWObjectCtx& obj_ctx, | |
12480 | const string& tenant, const string& bucket_name, | |
12481 | RGWBucketInfo& info, | |
12482 | real_time *pmtime, map<string, bufferlist> *pattrs) | |
12483 | { | |
12484 | return _get_bucket_info(obj_ctx, tenant, bucket_name, info, pmtime, | |
12485 | pattrs, boost::none); | |
12486 | } | |
12487 | ||
12488 | int RGWRados::try_refresh_bucket_info(RGWBucketInfo& info, | |
12489 | ceph::real_time *pmtime, | |
12490 | map<string, bufferlist> *pattrs) | |
12491 | { | |
12492 | RGWObjectCtx obj_ctx(this); | |
12493 | ||
12494 | return _get_bucket_info(obj_ctx, info.bucket.tenant, info.bucket.name, | |
12495 | info, pmtime, pattrs, info.objv_tracker.read_version); | |
12496 | } | |
12497 | ||
12498 | int RGWRados::put_bucket_entrypoint_info(const string& tenant_name, const string& bucket_name, RGWBucketEntryPoint& entry_point, | |
12499 | bool exclusive, RGWObjVersionTracker& objv_tracker, real_time mtime, | |
12500 | map<string, bufferlist> *pattrs) | |
12501 | { | |
12502 | bufferlist epbl; | |
12503 | ::encode(entry_point, epbl); | |
12504 | string bucket_entry; | |
12505 | rgw_make_bucket_entry_name(tenant_name, bucket_name, bucket_entry); | |
12506 | return rgw_bucket_store_info(this, bucket_entry, epbl, exclusive, pattrs, &objv_tracker, mtime); | |
12507 | } | |
12508 | ||
12509 | int RGWRados::put_bucket_instance_info(RGWBucketInfo& info, bool exclusive, | |
12510 | real_time mtime, map<string, bufferlist> *pattrs) | |
12511 | { | |
12512 | info.has_instance_obj = true; | |
12513 | bufferlist bl; | |
12514 | ||
12515 | ::encode(info, bl); | |
12516 | ||
12517 | string key = info.bucket.get_key(); /* when we go through meta api, we don't use oid directly */ | |
12518 | int ret = rgw_bucket_instance_store_info(this, key, bl, exclusive, pattrs, &info.objv_tracker, mtime); | |
12519 | if (ret == -EEXIST) { | |
12520 | /* well, if it's exclusive we shouldn't overwrite it, because we might race with another | |
12521 | * bucket operation on this specific bucket (e.g., being synced from the master), but | |
12522 | * since bucket instace meta object is unique for this specific bucket instace, we don't | |
12523 | * need to return an error. | |
12524 | * A scenario where we'd get -EEXIST here, is in a multi-zone config, we're not on the | |
12525 | * master, creating a bucket, sending bucket creation to the master, we create the bucket | |
12526 | * locally, while in the sync thread we sync the new bucket. | |
12527 | */ | |
12528 | ret = 0; | |
12529 | } | |
12530 | return ret; | |
12531 | } | |
12532 | ||
12533 | int RGWRados::put_linked_bucket_info(RGWBucketInfo& info, bool exclusive, real_time mtime, obj_version *pep_objv, | |
12534 | map<string, bufferlist> *pattrs, bool create_entry_point) | |
12535 | { | |
12536 | bool create_head = !info.has_instance_obj || create_entry_point; | |
12537 | ||
12538 | int ret = put_bucket_instance_info(info, exclusive, mtime, pattrs); | |
12539 | if (ret < 0) { | |
12540 | return ret; | |
12541 | } | |
12542 | ||
12543 | if (!create_head) | |
12544 | return 0; /* done! */ | |
12545 | ||
12546 | RGWBucketEntryPoint entry_point; | |
12547 | entry_point.bucket = info.bucket; | |
12548 | entry_point.owner = info.owner; | |
12549 | entry_point.creation_time = info.creation_time; | |
12550 | entry_point.linked = true; | |
12551 | RGWObjVersionTracker ot; | |
12552 | if (pep_objv && !pep_objv->tag.empty()) { | |
12553 | ot.write_version = *pep_objv; | |
12554 | } else { | |
12555 | ot.generate_new_write_ver(cct); | |
12556 | if (pep_objv) { | |
12557 | *pep_objv = ot.write_version; | |
12558 | } | |
12559 | } | |
12560 | ret = put_bucket_entrypoint_info(info.bucket.tenant, info.bucket.name, entry_point, exclusive, ot, mtime, NULL); | |
12561 | if (ret < 0) | |
12562 | return ret; | |
12563 | ||
12564 | return 0; | |
12565 | } | |
12566 | ||
12567 | int RGWRados::omap_get_vals(rgw_raw_obj& obj, bufferlist& header, const string& marker, uint64_t count, std::map<string, bufferlist>& m) | |
12568 | { | |
12569 | rgw_rados_ref ref; | |
12570 | int r = get_raw_obj_ref(obj, &ref); | |
12571 | if (r < 0) { | |
12572 | return r; | |
12573 | } | |
12574 | ||
12575 | r = ref.ioctx.omap_get_vals(ref.oid, marker, count, &m); | |
12576 | if (r < 0) | |
12577 | return r; | |
12578 | ||
12579 | return 0; | |
12580 | ||
12581 | } | |
12582 | ||
12583 | int RGWRados::omap_get_all(rgw_raw_obj& obj, bufferlist& header, | |
12584 | std::map<string, bufferlist>& m) | |
12585 | { | |
12586 | rgw_rados_ref ref; | |
12587 | int r = get_raw_obj_ref(obj, &ref); | |
12588 | if (r < 0) { | |
12589 | return r; | |
12590 | } | |
12591 | ||
12592 | #define MAX_OMAP_GET_ENTRIES 1024 | |
12593 | const int count = MAX_OMAP_GET_ENTRIES; | |
12594 | string start_after; | |
12595 | ||
12596 | while (true) { | |
12597 | std::map<string, bufferlist> t; | |
12598 | r = ref.ioctx.omap_get_vals(ref.oid, start_after, count, &t); | |
12599 | if (r < 0) { | |
12600 | return r; | |
12601 | } | |
12602 | if (t.empty()) { | |
12603 | break; | |
12604 | } | |
12605 | start_after = t.rbegin()->first; | |
12606 | m.insert(t.begin(), t.end()); | |
12607 | } | |
12608 | return 0; | |
12609 | } | |
12610 | ||
12611 | int RGWRados::omap_set(rgw_raw_obj& obj, const std::string& key, bufferlist& bl) | |
12612 | { | |
12613 | rgw_rados_ref ref; | |
12614 | int r = get_raw_obj_ref(obj, &ref); | |
12615 | if (r < 0) { | |
12616 | return r; | |
12617 | } | |
12618 | ldout(cct, 15) << "omap_set obj=" << obj << " key=" << key << dendl; | |
12619 | ||
12620 | map<string, bufferlist> m; | |
12621 | m[key] = bl; | |
12622 | ||
12623 | r = ref.ioctx.omap_set(ref.oid, m); | |
12624 | ||
12625 | return r; | |
12626 | } | |
12627 | ||
12628 | int RGWRados::omap_set(rgw_raw_obj& obj, std::map<std::string, bufferlist>& m) | |
12629 | { | |
12630 | rgw_rados_ref ref; | |
12631 | int r = get_raw_obj_ref(obj, &ref); | |
12632 | if (r < 0) { | |
12633 | return r; | |
12634 | } | |
12635 | ||
12636 | r = ref.ioctx.omap_set(ref.oid, m); | |
12637 | ||
12638 | return r; | |
12639 | } | |
12640 | ||
12641 | int RGWRados::omap_del(rgw_raw_obj& obj, const std::string& key) | |
12642 | { | |
12643 | rgw_rados_ref ref; | |
12644 | int r = get_raw_obj_ref(obj, &ref); | |
12645 | if (r < 0) { | |
12646 | return r; | |
12647 | } | |
12648 | ||
12649 | set<string> k; | |
12650 | k.insert(key); | |
12651 | ||
12652 | r = ref.ioctx.omap_rm_keys(ref.oid, k); | |
12653 | return r; | |
12654 | } | |
12655 | ||
12656 | int RGWRados::update_containers_stats(map<string, RGWBucketEnt>& m) | |
12657 | { | |
12658 | RGWObjectCtx obj_ctx(this); | |
12659 | ||
12660 | map<string, RGWBucketEnt>::iterator iter; | |
12661 | for (iter = m.begin(); iter != m.end(); ++iter) { | |
12662 | RGWBucketEnt& ent = iter->second; | |
12663 | rgw_bucket& bucket = ent.bucket; | |
12664 | ent.count = 0; | |
12665 | ent.size = 0; | |
12666 | ent.size_rounded = 0; | |
12667 | ||
12668 | vector<rgw_bucket_dir_header> headers; | |
12669 | ||
12670 | RGWBucketInfo bucket_info; | |
12671 | int ret = get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL); | |
12672 | if (ret < 0) { | |
12673 | return ret; | |
12674 | } | |
12675 | ||
12676 | int r = cls_bucket_head(bucket_info, RGW_NO_SHARD, headers); | |
12677 | if (r < 0) | |
12678 | return r; | |
12679 | ||
12680 | auto hiter = headers.begin(); | |
12681 | for (; hiter != headers.end(); ++hiter) { | |
12682 | RGWObjCategory category = main_category; | |
12683 | map<uint8_t, struct rgw_bucket_category_stats>::iterator iter = (hiter->stats).find((uint8_t)category); | |
12684 | if (iter != hiter->stats.end()) { | |
12685 | struct rgw_bucket_category_stats& stats = iter->second; | |
12686 | ent.count += stats.num_entries; | |
12687 | ent.size += stats.total_size; | |
12688 | ent.size_rounded += stats.total_size_rounded; | |
12689 | } | |
12690 | } | |
12691 | ||
12692 | // fill in placement_rule from the bucket instance for use in swift's | |
12693 | // per-storage policy statistics | |
12694 | ent.placement_rule = std::move(bucket_info.placement_rule); | |
12695 | } | |
12696 | ||
12697 | return m.size(); | |
12698 | } | |
12699 | ||
12700 | int RGWRados::append_async(rgw_raw_obj& obj, size_t size, bufferlist& bl) | |
12701 | { | |
12702 | rgw_rados_ref ref; | |
12703 | int r = get_raw_obj_ref(obj, &ref); | |
12704 | if (r < 0) { | |
12705 | return r; | |
12706 | } | |
12707 | librados::Rados *rad = get_rados_handle(); | |
12708 | librados::AioCompletion *completion = rad->aio_create_completion(NULL, NULL, NULL); | |
12709 | ||
12710 | r = ref.ioctx.aio_append(ref.oid, completion, bl, size); | |
12711 | completion->release(); | |
12712 | return r; | |
12713 | } | |
12714 | ||
12715 | int RGWRados::distribute(const string& key, bufferlist& bl) | |
12716 | { | |
12717 | /* | |
12718 | * we were called before watch was initialized. This can only happen if we're updating some system | |
12719 | * config object (e.g., zone info) during init. Don't try to distribute the cache info for these | |
12720 | * objects, they're currently only read on startup anyway. | |
12721 | */ | |
12722 | if (!watch_initialized) | |
12723 | return 0; | |
12724 | ||
12725 | string notify_oid; | |
12726 | pick_control_oid(key, notify_oid); | |
12727 | ||
12728 | ldout(cct, 10) << "distributing notification oid=" << notify_oid << " bl.length()=" << bl.length() << dendl; | |
12729 | return control_pool_ctx.notify2(notify_oid, bl, 0, NULL); | |
12730 | } | |
12731 | ||
12732 | int RGWRados::pool_iterate_begin(const rgw_pool& pool, RGWPoolIterCtx& ctx) | |
12733 | { | |
12734 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
12735 | librados::NObjectIterator& iter = ctx.iter; | |
12736 | ||
12737 | int r = open_pool_ctx(pool, io_ctx); | |
12738 | if (r < 0) | |
12739 | return r; | |
12740 | ||
12741 | iter = io_ctx.nobjects_begin(); | |
12742 | ||
12743 | return 0; | |
12744 | } | |
12745 | ||
12746 | int RGWRados::pool_iterate_begin(const rgw_pool& pool, const string& cursor, RGWPoolIterCtx& ctx) | |
12747 | { | |
12748 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
12749 | librados::NObjectIterator& iter = ctx.iter; | |
12750 | ||
12751 | int r = open_pool_ctx(pool, io_ctx); | |
12752 | if (r < 0) | |
12753 | return r; | |
12754 | ||
12755 | librados::ObjectCursor oc; | |
12756 | if (!oc.from_str(cursor)) { | |
12757 | ldout(cct, 10) << "failed to parse cursor: " << cursor << dendl; | |
12758 | return -EINVAL; | |
12759 | } | |
12760 | ||
12761 | try { | |
12762 | iter = io_ctx.nobjects_begin(oc); | |
12763 | return 0; | |
12764 | } catch (const std::system_error& e) { | |
12765 | r = -e.code().value(); | |
12766 | ldout(cct, 10) << "nobjects_begin threw " << e.what() | |
12767 | << ", returning " << r << dendl; | |
12768 | return r; | |
12769 | } catch (const std::exception& e) { | |
12770 | ldout(cct, 10) << "nobjects_begin threw " << e.what() | |
12771 | << ", returning -5" << dendl; | |
12772 | return -EIO; | |
12773 | } | |
12774 | } | |
12775 | ||
12776 | string RGWRados::pool_iterate_get_cursor(RGWPoolIterCtx& ctx) | |
12777 | { | |
12778 | return ctx.iter.get_cursor().to_str(); | |
12779 | } | |
12780 | ||
12781 | static int do_pool_iterate(CephContext* cct, RGWPoolIterCtx& ctx, uint32_t num, | |
12782 | vector<rgw_bucket_dir_entry>& objs, | |
12783 | bool *is_truncated, RGWAccessListFilter *filter) | |
12784 | { | |
12785 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
12786 | librados::NObjectIterator& iter = ctx.iter; | |
12787 | ||
12788 | if (iter == io_ctx.nobjects_end()) | |
12789 | return -ENOENT; | |
12790 | ||
12791 | uint32_t i; | |
12792 | ||
12793 | for (i = 0; i < num && iter != io_ctx.nobjects_end(); ++i, ++iter) { | |
12794 | rgw_bucket_dir_entry e; | |
12795 | ||
12796 | string oid = iter->get_oid(); | |
12797 | ldout(cct, 20) << "RGWRados::pool_iterate: got " << oid << dendl; | |
12798 | ||
12799 | // fill it in with initial values; we may correct later | |
12800 | if (filter && !filter->filter(oid, oid)) | |
12801 | continue; | |
12802 | ||
12803 | e.key = oid; | |
12804 | objs.push_back(e); | |
12805 | } | |
12806 | ||
12807 | if (is_truncated) | |
12808 | *is_truncated = (iter != io_ctx.nobjects_end()); | |
12809 | ||
12810 | return objs.size(); | |
12811 | } | |
12812 | struct RGWAccessListFilterPrefix : public RGWAccessListFilter { | |
12813 | string prefix; | |
12814 | ||
12815 | explicit RGWAccessListFilterPrefix(const string& _prefix) : prefix(_prefix) {} | |
12816 | bool filter(string& name, string& key) override { | |
12817 | return (prefix.compare(key.substr(0, prefix.size())) == 0); | |
12818 | } | |
12819 | }; | |
12820 | ||
12821 | int RGWRados::pool_iterate(RGWPoolIterCtx& ctx, uint32_t num, vector<rgw_bucket_dir_entry>& objs, | |
12822 | bool *is_truncated, RGWAccessListFilter *filter) | |
12823 | { | |
12824 | // catch exceptions from NObjectIterator::operator++() | |
12825 | try { | |
12826 | return do_pool_iterate(cct, ctx, num, objs, is_truncated, filter); | |
12827 | } catch (const std::system_error& e) { | |
12828 | int r = -e.code().value(); | |
12829 | ldout(cct, 10) << "NObjectIterator threw exception " << e.what() | |
12830 | << ", returning " << r << dendl; | |
12831 | return r; | |
12832 | } catch (const std::exception& e) { | |
12833 | ldout(cct, 10) << "NObjectIterator threw exception " << e.what() | |
12834 | << ", returning -5" << dendl; | |
12835 | return -EIO; | |
12836 | } | |
12837 | } | |
12838 | ||
12839 | int RGWRados::list_raw_objects_init(const rgw_pool& pool, const string& marker, RGWListRawObjsCtx *ctx) | |
12840 | { | |
12841 | if (!ctx->initialized) { | |
12842 | int r = pool_iterate_begin(pool, marker, ctx->iter_ctx); | |
12843 | if (r < 0) { | |
12844 | ldout(cct, 10) << "failed to list objects pool_iterate_begin() returned r=" << r << dendl; | |
12845 | return r; | |
12846 | } | |
12847 | ctx->initialized = true; | |
12848 | } | |
12849 | return 0; | |
12850 | } | |
12851 | ||
12852 | int RGWRados::list_raw_objects_next(const string& prefix_filter, int max, | |
12853 | RGWListRawObjsCtx& ctx, list<string>& oids, | |
12854 | bool *is_truncated) | |
12855 | { | |
12856 | if (!ctx.initialized) { | |
12857 | return -EINVAL; | |
12858 | } | |
12859 | RGWAccessListFilterPrefix filter(prefix_filter); | |
12860 | vector<rgw_bucket_dir_entry> objs; | |
12861 | int r = pool_iterate(ctx.iter_ctx, max, objs, is_truncated, &filter); | |
12862 | if (r < 0) { | |
12863 | if(r != -ENOENT) | |
12864 | ldout(cct, 10) << "failed to list objects pool_iterate returned r=" << r << dendl; | |
12865 | return r; | |
12866 | } | |
12867 | ||
12868 | vector<rgw_bucket_dir_entry>::iterator iter; | |
12869 | for (iter = objs.begin(); iter != objs.end(); ++iter) { | |
12870 | oids.push_back(iter->key.name); | |
12871 | } | |
12872 | ||
12873 | return oids.size(); | |
12874 | } | |
12875 | ||
12876 | int RGWRados::list_raw_objects(const rgw_pool& pool, const string& prefix_filter, | |
12877 | int max, RGWListRawObjsCtx& ctx, list<string>& oids, | |
12878 | bool *is_truncated) | |
12879 | { | |
12880 | if (!ctx.initialized) { | |
12881 | int r = list_raw_objects_init(pool, string(), &ctx); | |
12882 | if (r < 0) { | |
12883 | return r; | |
12884 | } | |
12885 | } | |
12886 | ||
12887 | return list_raw_objects_next(prefix_filter, max, ctx, oids, is_truncated); | |
12888 | } | |
12889 | ||
12890 | string RGWRados::list_raw_objs_get_cursor(RGWListRawObjsCtx& ctx) | |
12891 | { | |
12892 | return pool_iterate_get_cursor(ctx.iter_ctx); | |
12893 | } | |
12894 | ||
12895 | int RGWRados::list_bi_log_entries(RGWBucketInfo& bucket_info, int shard_id, string& marker, uint32_t max, | |
12896 | std::list<rgw_bi_log_entry>& result, bool *truncated) | |
12897 | { | |
12898 | ldout(cct, 20) << __func__ << ": " << bucket_info.bucket << " marker " << marker << " shard_id=" << shard_id << " max " << max << dendl; | |
12899 | result.clear(); | |
12900 | ||
12901 | librados::IoCtx index_ctx; | |
12902 | map<int, string> oids; | |
12903 | map<int, cls_rgw_bi_log_list_ret> bi_log_lists; | |
12904 | map<int, string> bucket_instance_ids; | |
12905 | int r = open_bucket_index(bucket_info, index_ctx, oids, shard_id, &bucket_instance_ids); | |
12906 | if (r < 0) | |
12907 | return r; | |
12908 | ||
12909 | BucketIndexShardsManager marker_mgr; | |
12910 | bool has_shards = (oids.size() > 1 || shard_id >= 0); | |
12911 | // If there are multiple shards for the bucket index object, the marker | |
12912 | // should have the pattern '{shard_id_1}#{shard_marker_1},{shard_id_2}# | |
12913 | // {shard_marker_2}...', if there is no sharding, the bi_log_list should | |
12914 | // only contain one record, and the key is the bucket instance id. | |
12915 | r = marker_mgr.from_string(marker, shard_id); | |
12916 | if (r < 0) | |
12917 | return r; | |
12918 | ||
12919 | r = CLSRGWIssueBILogList(index_ctx, marker_mgr, max, oids, bi_log_lists, cct->_conf->rgw_bucket_index_max_aio)(); | |
12920 | if (r < 0) | |
12921 | return r; | |
12922 | ||
12923 | map<int, list<rgw_bi_log_entry>::iterator> vcurrents; | |
12924 | map<int, list<rgw_bi_log_entry>::iterator> vends; | |
12925 | if (truncated) { | |
12926 | *truncated = false; | |
12927 | } | |
12928 | map<int, cls_rgw_bi_log_list_ret>::iterator miter = bi_log_lists.begin(); | |
12929 | for (; miter != bi_log_lists.end(); ++miter) { | |
12930 | int shard_id = miter->first; | |
12931 | vcurrents[shard_id] = miter->second.entries.begin(); | |
12932 | vends[shard_id] = miter->second.entries.end(); | |
12933 | if (truncated) { | |
12934 | *truncated = (*truncated || miter->second.truncated); | |
12935 | } | |
12936 | } | |
12937 | ||
12938 | size_t total = 0; | |
12939 | bool has_more = true; | |
12940 | map<int, list<rgw_bi_log_entry>::iterator>::iterator viter; | |
12941 | map<int, list<rgw_bi_log_entry>::iterator>::iterator eiter; | |
12942 | while (total < max && has_more) { | |
12943 | has_more = false; | |
12944 | ||
12945 | viter = vcurrents.begin(); | |
12946 | eiter = vends.begin(); | |
12947 | ||
12948 | for (; total < max && viter != vcurrents.end(); ++viter, ++eiter) { | |
12949 | assert (eiter != vends.end()); | |
12950 | ||
12951 | int shard_id = viter->first; | |
12952 | list<rgw_bi_log_entry>::iterator& liter = viter->second; | |
12953 | ||
12954 | if (liter == eiter->second){ | |
12955 | continue; | |
12956 | } | |
12957 | rgw_bi_log_entry& entry = *(liter); | |
12958 | if (has_shards) { | |
12959 | char buf[16]; | |
12960 | snprintf(buf, sizeof(buf), "%d", shard_id); | |
12961 | string tmp_id; | |
12962 | build_bucket_index_marker(buf, entry.id, &tmp_id); | |
12963 | entry.id.swap(tmp_id); | |
12964 | } | |
12965 | marker_mgr.add(shard_id, entry.id); | |
12966 | result.push_back(entry); | |
12967 | total++; | |
12968 | has_more = true; | |
12969 | ++liter; | |
12970 | } | |
12971 | } | |
12972 | ||
12973 | if (truncated) { | |
12974 | for (viter = vcurrents.begin(), eiter = vends.begin(); viter != vcurrents.end(); ++viter, ++eiter) { | |
12975 | assert (eiter != vends.end()); | |
12976 | *truncated = (*truncated || (viter->second != eiter->second)); | |
12977 | } | |
12978 | } | |
12979 | ||
12980 | // Refresh marker, if there are multiple shards, the output will look like | |
12981 | // '{shard_oid_1}#{shard_marker_1},{shard_oid_2}#{shard_marker_2}...', | |
12982 | // if there is no sharding, the simply marker (without oid) is returned | |
12983 | if (has_shards) { | |
12984 | marker_mgr.to_string(&marker); | |
12985 | } else { | |
12986 | if (!result.empty()) { | |
12987 | marker = result.rbegin()->id; | |
12988 | } | |
12989 | } | |
12990 | ||
12991 | return 0; | |
12992 | } | |
12993 | ||
12994 | int RGWRados::trim_bi_log_entries(RGWBucketInfo& bucket_info, int shard_id, string& start_marker, string& end_marker) | |
12995 | { | |
12996 | librados::IoCtx index_ctx; | |
12997 | map<int, string> bucket_objs; | |
12998 | ||
12999 | BucketIndexShardsManager start_marker_mgr; | |
13000 | BucketIndexShardsManager end_marker_mgr; | |
13001 | ||
13002 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs, shard_id); | |
13003 | if (r < 0) { | |
13004 | return r; | |
13005 | } | |
13006 | ||
13007 | r = start_marker_mgr.from_string(start_marker, shard_id); | |
13008 | if (r < 0) { | |
13009 | return r; | |
13010 | } | |
13011 | ||
13012 | r = end_marker_mgr.from_string(end_marker, shard_id); | |
13013 | if (r < 0) { | |
13014 | return r; | |
13015 | } | |
13016 | ||
13017 | return CLSRGWIssueBILogTrim(index_ctx, start_marker_mgr, end_marker_mgr, bucket_objs, | |
13018 | cct->_conf->rgw_bucket_index_max_aio)(); | |
13019 | ||
13020 | return r; | |
13021 | } | |
13022 | ||
13023 | int RGWRados::resync_bi_log_entries(RGWBucketInfo& bucket_info, int shard_id) | |
13024 | { | |
13025 | librados::IoCtx index_ctx; | |
13026 | map<int, string> bucket_objs; | |
13027 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs, shard_id); | |
13028 | if (r < 0) | |
13029 | return r; | |
13030 | ||
13031 | return CLSRGWIssueResyncBucketBILog(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); | |
13032 | } | |
13033 | ||
13034 | int RGWRados::stop_bi_log_entries(RGWBucketInfo& bucket_info, int shard_id) | |
13035 | { | |
13036 | librados::IoCtx index_ctx; | |
13037 | map<int, string> bucket_objs; | |
13038 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs, shard_id); | |
13039 | if (r < 0) | |
13040 | return r; | |
13041 | ||
13042 | return CLSRGWIssueBucketBILogStop(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); | |
13043 | } | |
13044 | ||
13045 | int RGWRados::bi_get_instance(const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
13046 | rgw_bucket_dir_entry *dirent) | |
13047 | { | |
13048 | rgw_cls_bi_entry bi_entry; | |
13049 | int r = bi_get(bucket_info, obj, InstanceIdx, &bi_entry); | |
13050 | if (r < 0 && r != -ENOENT) { | |
13051 | ldout(cct, 0) << "ERROR: bi_get() returned r=" << r << dendl; | |
13052 | } | |
13053 | if (r < 0) { | |
13054 | return r; | |
13055 | } | |
13056 | bufferlist::iterator iter = bi_entry.data.begin(); | |
13057 | try { | |
13058 | ::decode(*dirent, iter); | |
13059 | } catch (buffer::error& err) { | |
13060 | ldout(cct, 0) << "ERROR: failed to decode bi_entry()" << dendl; | |
13061 | return -EIO; | |
13062 | } | |
13063 | ||
13064 | return 0; | |
13065 | } | |
13066 | ||
13067 | int RGWRados::bi_get_olh(const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
13068 | rgw_bucket_olh_entry *olh) | |
13069 | { | |
13070 | rgw_cls_bi_entry bi_entry; | |
13071 | int r = bi_get(bucket_info, obj, OLHIdx, &bi_entry); | |
13072 | if (r < 0 && r != -ENOENT) { | |
13073 | ldout(cct, 0) << "ERROR: bi_get() returned r=" << r << dendl; | |
13074 | } | |
13075 | if (r < 0) { | |
13076 | return r; | |
13077 | } | |
13078 | auto iter = bi_entry.data.begin(); | |
13079 | try { | |
13080 | decode(*olh, iter); | |
13081 | } catch (buffer::error& err) { | |
13082 | ldout(cct, 0) << "ERROR: failed to decode bi_entry()" << dendl; | |
13083 | return -EIO; | |
13084 | } | |
13085 | ||
13086 | return 0; | |
13087 | } | |
13088 | ||
13089 | int RGWRados::bi_get(const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
13090 | BIIndexType index_type, rgw_cls_bi_entry *entry) | |
13091 | { | |
13092 | BucketShard bs(this); | |
13093 | int ret = bs.init(bucket_info, obj); | |
13094 | if (ret < 0) { | |
13095 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
13096 | return ret; | |
13097 | } | |
13098 | ||
13099 | cls_rgw_obj_key key(obj.key.get_index_key_name(), obj.key.instance); | |
13100 | ||
13101 | return cls_rgw_bi_get(bs.index_ctx, bs.bucket_obj, index_type, key, entry); | |
13102 | } | |
13103 | ||
13104 | void RGWRados::bi_put(ObjectWriteOperation& op, BucketShard& bs, rgw_cls_bi_entry& entry) | |
13105 | { | |
13106 | cls_rgw_bi_put(op, bs.bucket_obj, entry); | |
13107 | } | |
13108 | ||
13109 | int RGWRados::bi_put(BucketShard& bs, rgw_cls_bi_entry& entry) | |
13110 | { | |
13111 | int ret = cls_rgw_bi_put(bs.index_ctx, bs.bucket_obj, entry); | |
13112 | if (ret < 0) | |
13113 | return ret; | |
13114 | ||
13115 | return 0; | |
13116 | } | |
13117 | ||
13118 | int RGWRados::bi_put(rgw_bucket& bucket, rgw_obj& obj, rgw_cls_bi_entry& entry) | |
13119 | { | |
13120 | BucketShard bs(this); | |
13121 | int ret = bs.init(bucket, obj, nullptr /* no RGWBucketInfo */); | |
13122 | if (ret < 0) { | |
13123 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
13124 | return ret; | |
13125 | } | |
13126 | ||
13127 | return bi_put(bs, entry); | |
13128 | } | |
13129 | ||
13130 | int RGWRados::bi_list(rgw_bucket& bucket, const string& obj_name, const string& marker, uint32_t max, list<rgw_cls_bi_entry> *entries, bool *is_truncated) | |
13131 | { | |
13132 | rgw_obj obj(bucket, obj_name); | |
13133 | BucketShard bs(this); | |
13134 | int ret = bs.init(bucket, obj, nullptr /* no RGWBucketInfo */); | |
13135 | if (ret < 0) { | |
13136 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
13137 | return ret; | |
13138 | } | |
13139 | ||
13140 | ret = cls_rgw_bi_list(bs.index_ctx, bs.bucket_obj, obj_name, marker, max, entries, is_truncated); | |
13141 | if (ret == -ENOENT) { | |
13142 | *is_truncated = false; | |
13143 | } | |
13144 | if (ret < 0) | |
13145 | return ret; | |
13146 | ||
13147 | return 0; | |
13148 | } | |
13149 | ||
13150 | int RGWRados::bi_list(BucketShard& bs, const string& filter_obj, const string& marker, uint32_t max, list<rgw_cls_bi_entry> *entries, bool *is_truncated) | |
13151 | { | |
13152 | int ret = cls_rgw_bi_list(bs.index_ctx, bs.bucket_obj, filter_obj, marker, max, entries, is_truncated); | |
13153 | if (ret < 0) | |
13154 | return ret; | |
13155 | ||
13156 | return 0; | |
13157 | } | |
13158 | ||
13159 | int RGWRados::bi_remove(BucketShard& bs) | |
13160 | { | |
13161 | int ret = bs.index_ctx.remove(bs.bucket_obj); | |
13162 | if (ret == -ENOENT) { | |
13163 | ret = 0; | |
13164 | } | |
13165 | if (ret < 0) { | |
13166 | ldout(cct, 5) << "bs.index_ctx.remove(" << bs.bucket_obj << ") returned ret=" << ret << dendl; | |
13167 | return ret; | |
13168 | } | |
13169 | ||
13170 | return 0; | |
13171 | } | |
13172 | ||
13173 | int RGWRados::bi_list(rgw_bucket& bucket, int shard_id, const string& filter_obj, const string& marker, uint32_t max, list<rgw_cls_bi_entry> *entries, bool *is_truncated) | |
13174 | { | |
13175 | BucketShard bs(this); | |
13176 | int ret = bs.init(bucket, shard_id, nullptr /* no RGWBucketInfo */); | |
13177 | if (ret < 0) { | |
13178 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
13179 | return ret; | |
13180 | } | |
13181 | ||
13182 | return bi_list(bs, filter_obj, marker, max, entries, is_truncated); | |
13183 | } | |
13184 | ||
13185 | int RGWRados::gc_operate(string& oid, librados::ObjectWriteOperation *op) | |
13186 | { | |
13187 | return gc_pool_ctx.operate(oid, op); | |
13188 | } | |
13189 | ||
13190 | int RGWRados::gc_aio_operate(string& oid, librados::ObjectWriteOperation *op) | |
13191 | { | |
13192 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
13193 | int r = gc_pool_ctx.aio_operate(oid, c, op); | |
13194 | c->release(); | |
13195 | return r; | |
13196 | } | |
13197 | ||
13198 | int RGWRados::gc_operate(string& oid, librados::ObjectReadOperation *op, bufferlist *pbl) | |
13199 | { | |
13200 | return gc_pool_ctx.operate(oid, op, pbl); | |
13201 | } | |
13202 | ||
13203 | int RGWRados::list_gc_objs(int *index, string& marker, uint32_t max, bool expired_only, std::list<cls_rgw_gc_obj_info>& result, bool *truncated) | |
13204 | { | |
13205 | return gc->list(index, marker, max, expired_only, result, truncated); | |
13206 | } | |
13207 | ||
13208 | int RGWRados::process_gc() | |
13209 | { | |
13210 | return gc->process(); | |
13211 | } | |
13212 | ||
13213 | int RGWRados::list_lc_progress(const string& marker, uint32_t max_entries, map<string, int> *progress_map) | |
13214 | { | |
13215 | return lc->list_lc_progress(marker, max_entries, progress_map); | |
13216 | } | |
13217 | ||
13218 | int RGWRados::process_lc() | |
13219 | { | |
13220 | return lc->process(); | |
13221 | } | |
13222 | ||
13223 | bool RGWRados::process_expire_objects() | |
13224 | { | |
13225 | return obj_expirer->inspect_all_shards(utime_t(), ceph_clock_now()); | |
13226 | } | |
13227 | ||
13228 | int RGWRados::cls_obj_prepare_op(BucketShard& bs, RGWModifyOp op, string& tag, | |
13229 | rgw_obj& obj, uint16_t bilog_flags, rgw_zone_set *_zones_trace) | |
13230 | { | |
13231 | rgw_zone_set zones_trace; | |
13232 | if (_zones_trace) { | |
13233 | zones_trace = *_zones_trace; | |
13234 | } | |
13235 | zones_trace.insert(get_zone().id); | |
13236 | ||
13237 | ObjectWriteOperation o; | |
13238 | cls_rgw_obj_key key(obj.key.get_index_key_name(), obj.key.instance); | |
13239 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); | |
13240 | cls_rgw_bucket_prepare_op(o, op, tag, key, obj.key.get_loc(), get_zone().log_data, bilog_flags, zones_trace); | |
13241 | return bs.index_ctx.operate(bs.bucket_obj, &o); | |
13242 | } | |
13243 | ||
13244 | int RGWRados::cls_obj_complete_op(BucketShard& bs, const rgw_obj& obj, RGWModifyOp op, string& tag, | |
13245 | int64_t pool, uint64_t epoch, | |
13246 | rgw_bucket_dir_entry& ent, RGWObjCategory category, | |
13247 | list<rgw_obj_index_key> *remove_objs, uint16_t bilog_flags, rgw_zone_set *_zones_trace) | |
13248 | { | |
13249 | ObjectWriteOperation o; | |
13250 | rgw_bucket_dir_entry_meta dir_meta; | |
13251 | dir_meta = ent.meta; | |
13252 | dir_meta.category = category; | |
13253 | ||
13254 | rgw_zone_set zones_trace; | |
13255 | if (_zones_trace) { | |
13256 | zones_trace = *_zones_trace; | |
13257 | } | |
13258 | zones_trace.insert(get_zone().id); | |
13259 | ||
13260 | rgw_bucket_entry_ver ver; | |
13261 | ver.pool = pool; | |
13262 | ver.epoch = epoch; | |
13263 | cls_rgw_obj_key key(ent.key.name, ent.key.instance); | |
13264 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); | |
13265 | cls_rgw_bucket_complete_op(o, op, tag, ver, key, dir_meta, remove_objs, | |
13266 | get_zone().log_data, bilog_flags, &zones_trace); | |
13267 | complete_op_data *arg; | |
13268 | index_completion_manager->create_completion(obj, op, tag, ver, key, dir_meta, remove_objs, | |
13269 | get_zone().log_data, bilog_flags, &zones_trace, &arg); | |
13270 | librados::AioCompletion *completion = arg->rados_completion; | |
13271 | int ret = bs.index_ctx.aio_operate(bs.bucket_obj, arg->rados_completion, &o); | |
13272 | completion->release(); /* can't reference arg here, as it might have already been released */ | |
13273 | return ret; | |
13274 | } | |
13275 | ||
13276 | int RGWRados::cls_obj_complete_add(BucketShard& bs, const rgw_obj& obj, string& tag, | |
13277 | int64_t pool, uint64_t epoch, | |
13278 | rgw_bucket_dir_entry& ent, RGWObjCategory category, | |
13279 | list<rgw_obj_index_key> *remove_objs, uint16_t bilog_flags, rgw_zone_set *zones_trace) | |
13280 | { | |
13281 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_ADD, tag, pool, epoch, ent, category, remove_objs, bilog_flags, zones_trace); | |
13282 | } | |
13283 | ||
13284 | int RGWRados::cls_obj_complete_del(BucketShard& bs, string& tag, | |
13285 | int64_t pool, uint64_t epoch, | |
13286 | rgw_obj& obj, | |
13287 | real_time& removed_mtime, | |
13288 | list<rgw_obj_index_key> *remove_objs, | |
13289 | uint16_t bilog_flags, | |
13290 | rgw_zone_set *zones_trace) | |
13291 | { | |
13292 | rgw_bucket_dir_entry ent; | |
13293 | ent.meta.mtime = removed_mtime; | |
13294 | obj.key.get_index_key(&ent.key); | |
13295 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_DEL, tag, pool, epoch, ent, RGW_OBJ_CATEGORY_NONE, remove_objs, bilog_flags, zones_trace); | |
13296 | } | |
13297 | ||
13298 | int RGWRados::cls_obj_complete_cancel(BucketShard& bs, string& tag, rgw_obj& obj, uint16_t bilog_flags, rgw_zone_set *zones_trace) | |
13299 | { | |
13300 | rgw_bucket_dir_entry ent; | |
13301 | obj.key.get_index_key(&ent.key); | |
13302 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_CANCEL, tag, -1 /* pool id */, 0, ent, RGW_OBJ_CATEGORY_NONE, NULL, bilog_flags, zones_trace); | |
13303 | } | |
13304 | ||
13305 | int RGWRados::cls_obj_set_bucket_tag_timeout(RGWBucketInfo& bucket_info, uint64_t timeout) | |
13306 | { | |
13307 | librados::IoCtx index_ctx; | |
13308 | map<int, string> bucket_objs; | |
13309 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs); | |
13310 | if (r < 0) | |
13311 | return r; | |
13312 | ||
13313 | return CLSRGWIssueSetTagTimeout(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio, timeout)(); | |
13314 | } | |
13315 | ||
13316 | ||
13317 | int RGWRados::cls_bucket_list_ordered(RGWBucketInfo& bucket_info, | |
13318 | int shard_id, | |
13319 | rgw_obj_index_key& start, | |
13320 | const string& prefix, | |
13321 | uint32_t num_entries, | |
13322 | bool list_versions, | |
13323 | map<string, rgw_bucket_dir_entry>& m, | |
13324 | bool *is_truncated, | |
13325 | rgw_obj_index_key *last_entry, | |
13326 | bool (*force_check_filter)(const string& name)) | |
13327 | { | |
13328 | ldout(cct, 10) << "cls_bucket_list_ordered " << bucket_info.bucket << | |
13329 | " start " << start.name << "[" << start.instance << "] num_entries " << | |
13330 | num_entries << dendl; | |
13331 | ||
13332 | librados::IoCtx index_ctx; | |
13333 | // key - oid (for different shards if there is any) | |
13334 | // value - list result for the corresponding oid (shard), it is filled by | |
13335 | // the AIO callback | |
13336 | map<int, string> oids; | |
13337 | map<int, struct rgw_cls_list_ret> list_results; | |
13338 | int r = open_bucket_index(bucket_info, index_ctx, oids, shard_id); | |
13339 | if (r < 0) | |
13340 | return r; | |
13341 | ||
13342 | cls_rgw_obj_key start_key(start.name, start.instance); | |
13343 | r = CLSRGWIssueBucketList(index_ctx, start_key, prefix, num_entries, | |
13344 | list_versions, oids, list_results, | |
13345 | cct->_conf->rgw_bucket_index_max_aio)(); | |
13346 | if (r < 0) | |
13347 | return r; | |
13348 | ||
13349 | // Create a list of iterators that are used to iterate each shard | |
13350 | vector<map<string, struct rgw_bucket_dir_entry>::iterator> vcurrents(list_results.size()); | |
13351 | vector<map<string, struct rgw_bucket_dir_entry>::iterator> vends(list_results.size()); | |
13352 | vector<string> vnames(list_results.size()); | |
13353 | map<int, struct rgw_cls_list_ret>::iterator iter = list_results.begin(); | |
13354 | *is_truncated = false; | |
13355 | for (; iter != list_results.end(); ++iter) { | |
13356 | vcurrents.push_back(iter->second.dir.m.begin()); | |
13357 | vends.push_back(iter->second.dir.m.end()); | |
13358 | vnames.push_back(oids[iter->first]); | |
13359 | *is_truncated = (*is_truncated || iter->second.is_truncated); | |
13360 | } | |
13361 | ||
13362 | // Create a map to track the next candidate entry from each shard, if the entry | |
13363 | // from a specified shard is selected/erased, the next entry from that shard will | |
13364 | // be inserted for next round selection | |
13365 | map<string, size_t> candidates; | |
13366 | for (size_t i = 0; i < vcurrents.size(); ++i) { | |
13367 | if (vcurrents[i] != vends[i]) { | |
13368 | candidates[vcurrents[i]->first] = i; | |
13369 | } | |
13370 | } | |
13371 | ||
13372 | map<string, bufferlist> updates; | |
13373 | uint32_t count = 0; | |
13374 | while (count < num_entries && !candidates.empty()) { | |
13375 | r = 0; | |
13376 | // Select the next one | |
13377 | int pos = candidates.begin()->second; | |
13378 | const string& name = vcurrents[pos]->first; | |
13379 | struct rgw_bucket_dir_entry& dirent = vcurrents[pos]->second; | |
13380 | ||
13381 | bool force_check = force_check_filter && | |
13382 | force_check_filter(dirent.key.name); | |
13383 | if ((!dirent.exists && !dirent.is_delete_marker()) || | |
13384 | !dirent.pending_map.empty() || | |
13385 | force_check) { | |
13386 | /* there are uncommitted ops. We need to check the current state, | |
13387 | * and if the tags are old we need to do cleanup as well. */ | |
13388 | librados::IoCtx sub_ctx; | |
13389 | sub_ctx.dup(index_ctx); | |
13390 | r = check_disk_state(sub_ctx, bucket_info, dirent, dirent, | |
13391 | updates[vnames[pos]]); | |
13392 | if (r < 0 && r != -ENOENT) { | |
13393 | return r; | |
13394 | } | |
13395 | } | |
13396 | if (r >= 0) { | |
13397 | ldout(cct, 10) << "RGWRados::cls_bucket_list_ordered: got " << | |
13398 | dirent.key.name << "[" << dirent.key.instance << "]" << dendl; | |
13399 | m[name] = std::move(dirent); | |
13400 | ++count; | |
13401 | } | |
13402 | ||
13403 | // Refresh the candidates map | |
13404 | candidates.erase(candidates.begin()); | |
13405 | ++vcurrents[pos]; | |
13406 | if (vcurrents[pos] != vends[pos]) { | |
13407 | candidates[vcurrents[pos]->first] = pos; | |
13408 | } | |
13409 | } | |
13410 | ||
13411 | // Suggest updates if there is any | |
13412 | map<string, bufferlist>::iterator miter = updates.begin(); | |
13413 | for (; miter != updates.end(); ++miter) { | |
13414 | if (miter->second.length()) { | |
13415 | ObjectWriteOperation o; | |
13416 | cls_rgw_suggest_changes(o, miter->second); | |
13417 | // we don't care if we lose suggested updates, send them off blindly | |
13418 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
13419 | index_ctx.aio_operate(miter->first, c, &o); | |
13420 | c->release(); | |
13421 | } | |
13422 | } | |
13423 | ||
13424 | // Check if all the returned entries are consumed or not | |
13425 | for (size_t i = 0; i < vcurrents.size(); ++i) { | |
13426 | if (vcurrents[i] != vends[i]) { | |
13427 | *is_truncated = true; | |
13428 | break; | |
13429 | } | |
13430 | } | |
13431 | if (!m.empty()) | |
13432 | *last_entry = m.rbegin()->first; | |
13433 | ||
13434 | return 0; | |
13435 | } | |
13436 | ||
13437 | ||
13438 | int RGWRados::cls_bucket_list_unordered(RGWBucketInfo& bucket_info, | |
13439 | int shard_id, | |
13440 | rgw_obj_index_key& start, | |
13441 | const string& prefix, | |
13442 | uint32_t num_entries, | |
13443 | bool list_versions, | |
13444 | std::vector<rgw_bucket_dir_entry>& ent_list, | |
13445 | bool *is_truncated, | |
13446 | rgw_obj_index_key *last_entry, | |
13447 | bool (*force_check_filter)(const string& name)) { | |
13448 | ldout(cct, 10) << "cls_bucket_list_unordered " << bucket_info.bucket << | |
13449 | " start " << start.name << "[" << start.instance << | |
13450 | "] num_entries " << num_entries << dendl; | |
13451 | ||
13452 | *is_truncated = false; | |
13453 | librados::IoCtx index_ctx; | |
13454 | ||
13455 | rgw_obj_index_key my_start = start; | |
13456 | ||
13457 | map<int, string> oids; | |
13458 | int r = open_bucket_index(bucket_info, index_ctx, oids, shard_id); | |
13459 | if (r < 0) | |
13460 | return r; | |
13461 | const uint32_t num_shards = oids.size(); | |
13462 | ||
13463 | uint32_t current_shard; | |
13464 | if (shard_id >= 0) { | |
13465 | current_shard = shard_id; | |
13466 | } else if (my_start.empty()) { | |
13467 | current_shard = 0u; | |
13468 | } else { | |
13469 | current_shard = | |
13470 | rgw_bucket_shard_index(my_start.name, num_shards); | |
13471 | } | |
13472 | ||
13473 | uint32_t count = 0u; | |
13474 | map<string, bufferlist> updates; | |
13475 | std::string last_added_entry; | |
13476 | while (count <= num_entries && | |
13477 | ((shard_id >= 0 && current_shard == uint32_t(shard_id)) || | |
13478 | current_shard < num_shards)) { | |
13479 | // key - oid (for different shards if there is any) | |
13480 | // value - list result for the corresponding oid (shard), it is filled by | |
13481 | // the AIO callback | |
13482 | map<int, struct rgw_cls_list_ret> list_results; | |
13483 | r = CLSRGWIssueBucketList(index_ctx, my_start, prefix, num_entries, | |
13484 | list_versions, oids, list_results, | |
13485 | cct->_conf->rgw_bucket_index_max_aio)(); | |
13486 | if (r < 0) | |
13487 | return r; | |
13488 | ||
13489 | const std::string& oid = oids[current_shard]; | |
13490 | assert(list_results.find(current_shard) != list_results.end()); | |
13491 | auto& result = list_results[current_shard]; | |
13492 | for (auto& entry : result.dir.m) { | |
13493 | rgw_bucket_dir_entry& dirent = entry.second; | |
13494 | ||
13495 | bool force_check = force_check_filter && | |
13496 | force_check_filter(dirent.key.name); | |
13497 | if ((!dirent.exists && !dirent.is_delete_marker()) || | |
13498 | !dirent.pending_map.empty() || | |
13499 | force_check) { | |
13500 | /* there are uncommitted ops. We need to check the current state, | |
13501 | * and if the tags are old we need to do cleanup as well. */ | |
13502 | librados::IoCtx sub_ctx; | |
13503 | sub_ctx.dup(index_ctx); | |
13504 | r = check_disk_state(sub_ctx, bucket_info, dirent, dirent, updates[oid]); | |
13505 | if (r < 0 && r != -ENOENT) { | |
13506 | return r; | |
13507 | } | |
13508 | } | |
13509 | ||
13510 | // at this point either r >=0 or r == -ENOENT | |
13511 | if (r >= 0) { // i.e., if r != -ENOENT | |
13512 | ldout(cct, 10) << "RGWRados::cls_bucket_list_unordered: got " << | |
13513 | dirent.key.name << "[" << dirent.key.instance << "]" << dendl; | |
13514 | ||
13515 | if (count < num_entries) { | |
13516 | last_added_entry = entry.first; | |
13517 | my_start = dirent.key; | |
13518 | ent_list.emplace_back(std::move(dirent)); | |
13519 | ++count; | |
13520 | } else { | |
13521 | *is_truncated = true; | |
13522 | goto check_updates; | |
13523 | } | |
13524 | } else { // r == -ENOENT | |
13525 | // in the case of -ENOENT, make sure we're advancing marker | |
13526 | // for possible next call to CLSRGWIssueBucketList | |
13527 | my_start = dirent.key; | |
13528 | } | |
13529 | } // entry for loop | |
13530 | ||
13531 | if (!result.is_truncated) { | |
13532 | // if we reached the end of the shard read next shard | |
13533 | ++current_shard; | |
13534 | my_start = rgw_obj_index_key(); | |
13535 | } | |
13536 | } // shard loop | |
13537 | ||
13538 | check_updates: | |
13539 | // suggest updates if there is any | |
13540 | map<string, bufferlist>::iterator miter = updates.begin(); | |
13541 | for (; miter != updates.end(); ++miter) { | |
13542 | if (miter->second.length()) { | |
13543 | ObjectWriteOperation o; | |
13544 | cls_rgw_suggest_changes(o, miter->second); | |
13545 | // we don't care if we lose suggested updates, send them off blindly | |
13546 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
13547 | index_ctx.aio_operate(miter->first, c, &o); | |
13548 | c->release(); | |
13549 | } | |
13550 | } | |
13551 | ||
13552 | if (last_entry && !ent_list.empty()) { | |
13553 | *last_entry = last_added_entry; | |
13554 | } | |
13555 | ||
13556 | return 0; | |
13557 | } | |
13558 | ||
13559 | ||
13560 | int RGWRados::cls_obj_usage_log_add(const string& oid, | |
13561 | rgw_usage_log_info& info) | |
13562 | { | |
13563 | rgw_raw_obj obj(get_zone_params().usage_log_pool, oid); | |
13564 | ||
13565 | rgw_rados_ref ref; | |
13566 | int r = get_raw_obj_ref(obj, &ref); | |
13567 | if (r < 0) { | |
13568 | return r; | |
13569 | } | |
13570 | ||
13571 | ObjectWriteOperation op; | |
13572 | cls_rgw_usage_log_add(op, info); | |
13573 | ||
13574 | r = ref.ioctx.operate(ref.oid, &op); | |
13575 | return r; | |
13576 | } | |
13577 | ||
13578 | int RGWRados::cls_obj_usage_log_read(string& oid, string& user, uint64_t start_epoch, uint64_t end_epoch, uint32_t max_entries, | |
13579 | string& read_iter, map<rgw_user_bucket, rgw_usage_log_entry>& usage, bool *is_truncated) | |
13580 | { | |
13581 | rgw_raw_obj obj(get_zone_params().usage_log_pool, oid); | |
13582 | ||
13583 | rgw_rados_ref ref; | |
13584 | int r = get_raw_obj_ref(obj, &ref); | |
13585 | if (r < 0) { | |
13586 | return r; | |
13587 | } | |
13588 | ||
13589 | *is_truncated = false; | |
13590 | ||
13591 | r = cls_rgw_usage_log_read(ref.ioctx, ref.oid, user, start_epoch, end_epoch, | |
13592 | max_entries, read_iter, usage, is_truncated); | |
13593 | ||
13594 | return r; | |
13595 | } | |
13596 | ||
13597 | int RGWRados::cls_obj_usage_log_trim(string& oid, string& user, uint64_t start_epoch, uint64_t end_epoch) | |
13598 | { | |
13599 | rgw_raw_obj obj(get_zone_params().usage_log_pool, oid); | |
13600 | ||
13601 | rgw_rados_ref ref; | |
13602 | int r = get_raw_obj_ref(obj, &ref); | |
13603 | if (r < 0) { | |
13604 | return r; | |
13605 | } | |
13606 | ||
13607 | r = cls_rgw_usage_log_trim(ref.ioctx, ref.oid, user, start_epoch, end_epoch); | |
13608 | return r; | |
13609 | } | |
13610 | ||
13611 | int RGWRados::remove_objs_from_index(RGWBucketInfo& bucket_info, list<rgw_obj_index_key>& oid_list) | |
13612 | { | |
13613 | librados::IoCtx index_ctx; | |
13614 | string dir_oid; | |
13615 | ||
13616 | uint8_t suggest_flag = (get_zone().log_data ? CEPH_RGW_DIR_SUGGEST_LOG_OP : 0); | |
13617 | ||
13618 | int r = open_bucket_index(bucket_info, index_ctx, dir_oid); | |
13619 | if (r < 0) | |
13620 | return r; | |
13621 | ||
13622 | bufferlist updates; | |
13623 | ||
13624 | for (auto iter = oid_list.begin(); iter != oid_list.end(); ++iter) { | |
13625 | rgw_bucket_dir_entry entry; | |
13626 | entry.key = *iter; | |
13627 | dout(2) << "RGWRados::remove_objs_from_index bucket=" << bucket_info.bucket << " obj=" << entry.key.name << ":" << entry.key.instance << dendl; | |
13628 | entry.ver.epoch = (uint64_t)-1; // ULLONG_MAX, needed to that objclass doesn't skip out request | |
13629 | updates.append(CEPH_RGW_REMOVE | suggest_flag); | |
13630 | ::encode(entry, updates); | |
13631 | } | |
13632 | ||
13633 | bufferlist out; | |
13634 | ||
13635 | r = index_ctx.exec(dir_oid, RGW_CLASS, RGW_DIR_SUGGEST_CHANGES, updates, out); | |
13636 | ||
13637 | return r; | |
13638 | } | |
13639 | ||
13640 | int RGWRados::check_disk_state(librados::IoCtx io_ctx, | |
13641 | const RGWBucketInfo& bucket_info, | |
13642 | rgw_bucket_dir_entry& list_state, | |
13643 | rgw_bucket_dir_entry& object, | |
13644 | bufferlist& suggested_updates) | |
13645 | { | |
13646 | const rgw_bucket& bucket = bucket_info.bucket; | |
13647 | uint8_t suggest_flag = (get_zone().log_data ? CEPH_RGW_DIR_SUGGEST_LOG_OP : 0); | |
13648 | ||
13649 | std::string loc; | |
13650 | ||
13651 | rgw_obj obj(bucket, list_state.key); | |
13652 | ||
13653 | string oid; | |
13654 | get_obj_bucket_and_oid_loc(obj, oid, loc); | |
13655 | ||
13656 | if (loc != list_state.locator) { | |
13657 | ldout(cct, 0) << "WARNING: generated locator (" << loc << ") is different from listed locator (" << list_state.locator << ")" << dendl; | |
13658 | } | |
13659 | ||
13660 | io_ctx.locator_set_key(list_state.locator); | |
13661 | ||
13662 | RGWObjState *astate = NULL; | |
13663 | RGWObjectCtx rctx(this); | |
13664 | int r = get_obj_state(&rctx, bucket_info, obj, &astate, false); | |
13665 | if (r < 0) | |
13666 | return r; | |
13667 | ||
13668 | list_state.pending_map.clear(); // we don't need this and it inflates size | |
13669 | if (!astate->exists) { | |
13670 | /* object doesn't exist right now -- hopefully because it's | |
13671 | * marked as !exists and got deleted */ | |
13672 | if (list_state.exists) { | |
13673 | /* FIXME: what should happen now? Work out if there are any | |
13674 | * non-bad ways this could happen (there probably are, but annoying | |
13675 | * to handle!) */ | |
13676 | } | |
13677 | // encode a suggested removal of that key | |
13678 | list_state.ver.epoch = io_ctx.get_last_version(); | |
13679 | list_state.ver.pool = io_ctx.get_id(); | |
13680 | cls_rgw_encode_suggestion(CEPH_RGW_REMOVE, list_state, suggested_updates); | |
13681 | return -ENOENT; | |
13682 | } | |
13683 | ||
13684 | string etag; | |
13685 | string content_type; | |
13686 | ACLOwner owner; | |
13687 | ||
13688 | object.meta.size = astate->size; | |
13689 | object.meta.accounted_size = astate->accounted_size; | |
13690 | object.meta.mtime = astate->mtime; | |
13691 | ||
13692 | map<string, bufferlist>::iterator iter = astate->attrset.find(RGW_ATTR_ETAG); | |
13693 | if (iter != astate->attrset.end()) { | |
13694 | etag = iter->second.c_str(); | |
13695 | } | |
13696 | iter = astate->attrset.find(RGW_ATTR_CONTENT_TYPE); | |
13697 | if (iter != astate->attrset.end()) { | |
13698 | content_type = iter->second.c_str(); | |
13699 | } | |
13700 | iter = astate->attrset.find(RGW_ATTR_ACL); | |
13701 | if (iter != astate->attrset.end()) { | |
13702 | r = decode_policy(iter->second, &owner); | |
13703 | if (r < 0) { | |
13704 | dout(0) << "WARNING: could not decode policy for object: " << obj << dendl; | |
13705 | } | |
13706 | } | |
13707 | ||
13708 | if (astate->has_manifest) { | |
13709 | RGWObjManifest::obj_iterator miter; | |
13710 | RGWObjManifest& manifest = astate->manifest; | |
13711 | for (miter = manifest.obj_begin(); miter != manifest.obj_end(); ++miter) { | |
13712 | const rgw_raw_obj& raw_loc = miter.get_location().get_raw_obj(this); | |
13713 | rgw_obj loc; | |
13714 | rgw_raw_obj_to_obj(manifest.get_obj().bucket, raw_loc, &loc); | |
13715 | ||
13716 | if (loc.key.ns == RGW_OBJ_NS_MULTIPART) { | |
13717 | dout(10) << "check_disk_state(): removing manifest part from index: " << loc << dendl; | |
13718 | r = delete_obj_index(loc); | |
13719 | if (r < 0) { | |
13720 | dout(0) << "WARNING: delete_obj_index() returned r=" << r << dendl; | |
13721 | } | |
13722 | } | |
13723 | } | |
13724 | } | |
13725 | ||
13726 | object.meta.etag = etag; | |
13727 | object.meta.content_type = content_type; | |
13728 | object.meta.owner = owner.get_id().to_str(); | |
13729 | object.meta.owner_display_name = owner.get_display_name(); | |
13730 | ||
13731 | // encode suggested updates | |
13732 | list_state.ver.pool = io_ctx.get_id(); | |
13733 | list_state.ver.epoch = astate->epoch; | |
13734 | list_state.meta.size = object.meta.size; | |
13735 | list_state.meta.accounted_size = object.meta.accounted_size; | |
13736 | list_state.meta.mtime = object.meta.mtime; | |
13737 | list_state.meta.category = main_category; | |
13738 | list_state.meta.etag = etag; | |
13739 | list_state.meta.content_type = content_type; | |
13740 | if (astate->obj_tag.length() > 0) | |
13741 | list_state.tag = astate->obj_tag.c_str(); | |
13742 | list_state.meta.owner = owner.get_id().to_str(); | |
13743 | list_state.meta.owner_display_name = owner.get_display_name(); | |
13744 | ||
13745 | list_state.exists = true; | |
13746 | cls_rgw_encode_suggestion(CEPH_RGW_UPDATE | suggest_flag, list_state, suggested_updates); | |
13747 | return 0; | |
13748 | } | |
13749 | ||
13750 | int RGWRados::cls_bucket_head(const RGWBucketInfo& bucket_info, int shard_id, vector<rgw_bucket_dir_header>& headers, map<int, string> *bucket_instance_ids) | |
13751 | { | |
13752 | librados::IoCtx index_ctx; | |
13753 | map<int, string> oids; | |
13754 | map<int, struct rgw_cls_list_ret> list_results; | |
13755 | int r = open_bucket_index(bucket_info, index_ctx, oids, list_results, shard_id, bucket_instance_ids); | |
13756 | if (r < 0) | |
13757 | return r; | |
13758 | ||
13759 | r = CLSRGWIssueGetDirHeader(index_ctx, oids, list_results, cct->_conf->rgw_bucket_index_max_aio)(); | |
13760 | if (r < 0) | |
13761 | return r; | |
13762 | ||
13763 | map<int, struct rgw_cls_list_ret>::iterator iter = list_results.begin(); | |
13764 | for(; iter != list_results.end(); ++iter) { | |
13765 | headers.push_back(std::move(iter->second.dir.header)); | |
13766 | } | |
13767 | return 0; | |
13768 | } | |
13769 | ||
13770 | int RGWRados::cls_bucket_head_async(const RGWBucketInfo& bucket_info, int shard_id, RGWGetDirHeader_CB *ctx, int *num_aio) | |
13771 | { | |
13772 | librados::IoCtx index_ctx; | |
13773 | map<int, string> bucket_objs; | |
13774 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs, shard_id); | |
13775 | if (r < 0) | |
13776 | return r; | |
13777 | ||
13778 | map<int, string>::iterator iter = bucket_objs.begin(); | |
13779 | for (; iter != bucket_objs.end(); ++iter) { | |
13780 | r = cls_rgw_get_dir_header_async(index_ctx, iter->second, static_cast<RGWGetDirHeader_CB*>(ctx->get())); | |
13781 | if (r < 0) { | |
13782 | ctx->put(); | |
13783 | break; | |
13784 | } else { | |
13785 | (*num_aio)++; | |
13786 | } | |
13787 | } | |
13788 | return r; | |
13789 | } | |
13790 | ||
13791 | int RGWRados::cls_user_get_header(const string& user_id, cls_user_header *header) | |
13792 | { | |
13793 | string buckets_obj_id; | |
13794 | rgw_get_buckets_obj(user_id, buckets_obj_id); | |
13795 | rgw_raw_obj obj(get_zone_params().user_uid_pool, buckets_obj_id); | |
13796 | ||
13797 | rgw_rados_ref ref; | |
13798 | int r = get_raw_obj_ref(obj, &ref); | |
13799 | if (r < 0) { | |
13800 | return r; | |
13801 | } | |
13802 | ||
13803 | librados::ObjectReadOperation op; | |
13804 | int rc; | |
13805 | ::cls_user_get_header(op, header, &rc); | |
13806 | bufferlist ibl; | |
13807 | r = ref.ioctx.operate(ref.oid, &op, &ibl); | |
13808 | if (r < 0) | |
13809 | return r; | |
13810 | if (rc < 0) | |
13811 | return rc; | |
13812 | ||
13813 | return 0; | |
13814 | } | |
13815 | ||
13816 | int RGWRados::cls_user_reset_stats(const string& user_id) | |
13817 | { | |
13818 | string buckets_obj_id; | |
13819 | rgw_get_buckets_obj(user_id, buckets_obj_id); | |
13820 | rgw_raw_obj obj(get_zone_params().user_uid_pool, buckets_obj_id); | |
13821 | ||
13822 | rgw_rados_ref ref; | |
13823 | int r = get_raw_obj_ref(obj, &ref); | |
13824 | if (r < 0) { | |
13825 | return r; | |
13826 | } | |
13827 | ||
13828 | librados::ObjectWriteOperation op; | |
13829 | ::cls_user_reset_stats(op); | |
13830 | return ref.ioctx.operate(ref.oid, &op); | |
13831 | } | |
13832 | ||
13833 | int RGWRados::cls_user_get_header_async(const string& user_id, RGWGetUserHeader_CB *ctx) | |
13834 | { | |
13835 | string buckets_obj_id; | |
13836 | rgw_get_buckets_obj(user_id, buckets_obj_id); | |
13837 | rgw_raw_obj obj(get_zone_params().user_uid_pool, buckets_obj_id); | |
13838 | ||
13839 | rgw_rados_ref ref; | |
13840 | int r = get_raw_obj_ref(obj, &ref); | |
13841 | if (r < 0) { | |
13842 | return r; | |
13843 | } | |
13844 | ||
13845 | r = ::cls_user_get_header_async(ref.ioctx, ref.oid, ctx); | |
13846 | if (r < 0) | |
13847 | return r; | |
13848 | ||
13849 | return 0; | |
13850 | } | |
13851 | ||
13852 | int RGWRados::cls_user_sync_bucket_stats(rgw_raw_obj& user_obj, const RGWBucketInfo& bucket_info) | |
13853 | { | |
13854 | vector<rgw_bucket_dir_header> headers; | |
13855 | int r = cls_bucket_head(bucket_info, RGW_NO_SHARD, headers); | |
13856 | if (r < 0) { | |
13857 | ldout(cct, 20) << "cls_bucket_header() returned " << r << dendl; | |
13858 | return r; | |
13859 | } | |
13860 | ||
13861 | cls_user_bucket_entry entry; | |
13862 | ||
13863 | bucket_info.bucket.convert(&entry.bucket); | |
13864 | ||
13865 | for (const auto& hiter : headers) { | |
13866 | for (const auto& iter : hiter.stats) { | |
13867 | const struct rgw_bucket_category_stats& header_stats = iter.second; | |
13868 | entry.size += header_stats.total_size; | |
13869 | entry.size_rounded += header_stats.total_size_rounded; | |
13870 | entry.count += header_stats.num_entries; | |
13871 | } | |
13872 | } | |
13873 | ||
13874 | list<cls_user_bucket_entry> entries; | |
13875 | entries.push_back(entry); | |
13876 | ||
13877 | r = cls_user_update_buckets(user_obj, entries, false); | |
13878 | if (r < 0) { | |
13879 | ldout(cct, 20) << "cls_user_update_buckets() returned " << r << dendl; | |
13880 | return r; | |
13881 | } | |
13882 | ||
13883 | return 0; | |
13884 | } | |
13885 | ||
13886 | int RGWRados::cls_user_get_bucket_stats(const rgw_bucket& bucket, cls_user_bucket_entry& entry) | |
13887 | { | |
13888 | vector<rgw_bucket_dir_header> headers; | |
13889 | RGWBucketInfo bucket_info; | |
13890 | RGWObjectCtx obj_ctx(this); | |
13891 | int ret = get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL); | |
13892 | if (ret < 0) { | |
13893 | return ret; | |
13894 | } | |
13895 | ||
13896 | ret = cls_bucket_head(bucket_info, RGW_NO_SHARD, headers); | |
13897 | if (ret < 0) { | |
13898 | ldout(cct, 20) << "cls_bucket_header() returned " << ret << dendl; | |
13899 | return ret; | |
13900 | } | |
13901 | ||
13902 | bucket.convert(&entry.bucket); | |
13903 | ||
13904 | for (const auto& hiter : headers) { | |
13905 | for (const auto& iter : hiter.stats) { | |
13906 | const struct rgw_bucket_category_stats& header_stats = iter.second; | |
13907 | entry.size += header_stats.total_size; | |
13908 | entry.size_rounded += header_stats.total_size_rounded; | |
13909 | entry.count += header_stats.num_entries; | |
13910 | } | |
13911 | } | |
13912 | ||
13913 | return 0; | |
13914 | } | |
13915 | ||
13916 | int RGWRados::cls_user_list_buckets(rgw_raw_obj& obj, | |
13917 | const string& in_marker, | |
13918 | const string& end_marker, | |
13919 | const int max_entries, | |
13920 | list<cls_user_bucket_entry>& entries, | |
13921 | string * const out_marker, | |
13922 | bool * const truncated) | |
13923 | { | |
13924 | rgw_rados_ref ref; | |
13925 | int r = get_raw_obj_ref(obj, &ref); | |
13926 | if (r < 0) { | |
13927 | return r; | |
13928 | } | |
13929 | ||
13930 | librados::ObjectReadOperation op; | |
13931 | int rc; | |
13932 | ||
13933 | cls_user_bucket_list(op, in_marker, end_marker, max_entries, entries, out_marker, truncated, &rc); | |
13934 | bufferlist ibl; | |
13935 | r = ref.ioctx.operate(ref.oid, &op, &ibl); | |
13936 | if (r < 0) | |
13937 | return r; | |
13938 | if (rc < 0) | |
13939 | return rc; | |
13940 | ||
13941 | return 0; | |
13942 | } | |
13943 | ||
13944 | int RGWRados::cls_user_update_buckets(rgw_raw_obj& obj, list<cls_user_bucket_entry>& entries, bool add) | |
13945 | { | |
13946 | rgw_rados_ref ref; | |
13947 | int r = get_raw_obj_ref(obj, &ref); | |
13948 | if (r < 0) { | |
13949 | return r; | |
13950 | } | |
13951 | ||
13952 | librados::ObjectWriteOperation op; | |
13953 | cls_user_set_buckets(op, entries, add); | |
13954 | r = ref.ioctx.operate(ref.oid, &op); | |
13955 | if (r < 0) | |
13956 | return r; | |
13957 | ||
13958 | return 0; | |
13959 | } | |
13960 | ||
13961 | int RGWRados::complete_sync_user_stats(const rgw_user& user_id) | |
13962 | { | |
13963 | string buckets_obj_id; | |
13964 | rgw_get_buckets_obj(user_id, buckets_obj_id); | |
13965 | rgw_raw_obj obj(get_zone_params().user_uid_pool, buckets_obj_id); | |
13966 | return cls_user_complete_stats_sync(obj); | |
13967 | } | |
13968 | ||
13969 | int RGWRados::cls_user_complete_stats_sync(rgw_raw_obj& obj) | |
13970 | { | |
13971 | rgw_rados_ref ref; | |
13972 | int r = get_raw_obj_ref(obj, &ref); | |
13973 | if (r < 0) { | |
13974 | return r; | |
13975 | } | |
13976 | ||
13977 | librados::ObjectWriteOperation op; | |
13978 | ::cls_user_complete_stats_sync(op); | |
13979 | r = ref.ioctx.operate(ref.oid, &op); | |
13980 | if (r < 0) | |
13981 | return r; | |
13982 | ||
13983 | return 0; | |
13984 | } | |
13985 | ||
13986 | int RGWRados::cls_user_add_bucket(rgw_raw_obj& obj, const cls_user_bucket_entry& entry) | |
13987 | { | |
13988 | list<cls_user_bucket_entry> l; | |
13989 | l.push_back(entry); | |
13990 | ||
13991 | return cls_user_update_buckets(obj, l, true); | |
13992 | } | |
13993 | ||
13994 | int RGWRados::cls_user_remove_bucket(rgw_raw_obj& obj, const cls_user_bucket& bucket) | |
13995 | { | |
13996 | rgw_rados_ref ref; | |
13997 | int r = get_system_obj_ref(obj, &ref); | |
13998 | if (r < 0) { | |
13999 | return r; | |
14000 | } | |
14001 | ||
14002 | librados::ObjectWriteOperation op; | |
14003 | ::cls_user_remove_bucket(op, bucket); | |
14004 | r = ref.ioctx.operate(ref.oid, &op); | |
14005 | if (r < 0) | |
14006 | return r; | |
14007 | ||
14008 | return 0; | |
14009 | } | |
14010 | ||
14011 | int RGWRados::check_bucket_shards(const RGWBucketInfo& bucket_info, const rgw_bucket& bucket, | |
14012 | RGWQuotaInfo& bucket_quota) | |
14013 | { | |
14014 | if (!cct->_conf->rgw_dynamic_resharding) { | |
14015 | return 0; | |
14016 | } | |
14017 | ||
14018 | bool need_resharding = false; | |
14019 | int num_source_shards = (bucket_info.num_shards > 0 ? bucket_info.num_shards : 1); | |
14020 | uint32_t suggested_num_shards; | |
14021 | ||
14022 | int ret = quota_handler->check_bucket_shards((uint64_t)cct->_conf->rgw_max_objs_per_shard, | |
14023 | num_source_shards, bucket_info.owner, bucket, bucket_quota, | |
14024 | 1, need_resharding, &suggested_num_shards); | |
14025 | if (ret < 0) { | |
14026 | return ret; | |
14027 | } | |
14028 | ||
14029 | if (need_resharding) { | |
14030 | ldout(cct, 20) << __func__ << " bucket " << bucket.name << " need resharding " << | |
14031 | " old num shards " << bucket_info.num_shards << " new num shards " << suggested_num_shards << | |
14032 | dendl; | |
14033 | return add_bucket_to_reshard(bucket_info, suggested_num_shards); | |
14034 | } | |
14035 | ||
14036 | return ret; | |
14037 | } | |
14038 | ||
14039 | int RGWRados::add_bucket_to_reshard(const RGWBucketInfo& bucket_info, uint32_t new_num_shards) | |
14040 | { | |
14041 | RGWReshard reshard(this); | |
14042 | ||
14043 | uint32_t num_source_shards = (bucket_info.num_shards > 0 ? bucket_info.num_shards : 1); | |
14044 | ||
14045 | new_num_shards = min(new_num_shards, get_max_bucket_shards()); | |
14046 | if (new_num_shards <= num_source_shards) { | |
14047 | ldout(cct, 20) << "not resharding bucket name=" << bucket_info.bucket.name << ", orig_num=" << num_source_shards << ", new_num_shards=" << new_num_shards << dendl; | |
14048 | return 0; | |
14049 | } | |
14050 | ||
14051 | cls_rgw_reshard_entry entry; | |
14052 | entry.time = real_clock::now(); | |
14053 | entry.tenant = bucket_info.owner.tenant; | |
14054 | entry.bucket_name = bucket_info.bucket.name; | |
14055 | entry.bucket_id = bucket_info.bucket.bucket_id; | |
14056 | entry.old_num_shards = num_source_shards; | |
14057 | entry.new_num_shards = new_num_shards; | |
14058 | ||
14059 | return reshard.add(entry); | |
14060 | } | |
14061 | ||
14062 | int RGWRados::check_quota(const rgw_user& bucket_owner, rgw_bucket& bucket, | |
14063 | RGWQuotaInfo& user_quota, RGWQuotaInfo& bucket_quota, uint64_t obj_size) | |
14064 | { | |
14065 | return quota_handler->check_quota(bucket_owner, bucket, user_quota, bucket_quota, 1, obj_size); | |
14066 | } | |
14067 | ||
14068 | void RGWRados::get_bucket_index_objects(const string& bucket_oid_base, | |
14069 | uint32_t num_shards, | |
14070 | map<int, string>& bucket_objects, | |
14071 | int shard_id) { | |
14072 | if (!num_shards) { | |
14073 | bucket_objects[0] = bucket_oid_base; | |
14074 | } else { | |
14075 | char buf[bucket_oid_base.size() + 32]; | |
14076 | if (shard_id < 0) { | |
14077 | for (uint32_t i = 0; i < num_shards; ++i) { | |
14078 | snprintf(buf, sizeof(buf), "%s.%d", bucket_oid_base.c_str(), i); | |
14079 | bucket_objects[i] = buf; | |
14080 | } | |
14081 | } else { | |
14082 | if ((uint32_t)shard_id > num_shards) { | |
14083 | return; | |
14084 | } | |
14085 | snprintf(buf, sizeof(buf), "%s.%d", bucket_oid_base.c_str(), shard_id); | |
14086 | bucket_objects[shard_id] = buf; | |
14087 | } | |
14088 | } | |
14089 | } | |
14090 | ||
14091 | void RGWRados::get_bucket_instance_ids(const RGWBucketInfo& bucket_info, int shard_id, map<int, string> *result) | |
14092 | { | |
14093 | const rgw_bucket& bucket = bucket_info.bucket; | |
14094 | string plain_id = bucket.name + ":" + bucket.bucket_id; | |
14095 | if (!bucket_info.num_shards) { | |
14096 | (*result)[0] = plain_id; | |
14097 | } else { | |
14098 | char buf[16]; | |
14099 | if (shard_id < 0) { | |
14100 | for (uint32_t i = 0; i < bucket_info.num_shards; ++i) { | |
14101 | snprintf(buf, sizeof(buf), ":%d", i); | |
14102 | (*result)[i] = plain_id + buf; | |
14103 | } | |
14104 | } else { | |
14105 | if ((uint32_t)shard_id > bucket_info.num_shards) { | |
14106 | return; | |
14107 | } | |
14108 | snprintf(buf, sizeof(buf), ":%d", shard_id); | |
14109 | (*result)[shard_id] = plain_id + buf; | |
14110 | } | |
14111 | } | |
14112 | } | |
14113 | ||
14114 | int RGWRados::get_target_shard_id(const RGWBucketInfo& bucket_info, const string& obj_key, | |
14115 | int *shard_id) | |
14116 | { | |
14117 | int r = 0; | |
14118 | switch (bucket_info.bucket_index_shard_hash_type) { | |
14119 | case RGWBucketInfo::MOD: | |
14120 | if (!bucket_info.num_shards) { | |
14121 | if (shard_id) { | |
14122 | *shard_id = -1; | |
14123 | } | |
14124 | } else { | |
14125 | uint32_t sid = rgw_bucket_shard_index(obj_key, bucket_info.num_shards); | |
14126 | if (shard_id) { | |
14127 | *shard_id = (int)sid; | |
14128 | } | |
14129 | } | |
14130 | break; | |
14131 | default: | |
14132 | r = -ENOTSUP; | |
14133 | } | |
14134 | return r; | |
14135 | } | |
14136 | ||
14137 | void RGWRados::get_bucket_index_object(const string& bucket_oid_base, uint32_t num_shards, | |
14138 | int shard_id, string *bucket_obj) | |
14139 | { | |
14140 | if (!num_shards) { | |
14141 | // By default with no sharding, we use the bucket oid as itself | |
14142 | (*bucket_obj) = bucket_oid_base; | |
14143 | } else { | |
14144 | char buf[bucket_oid_base.size() + 32]; | |
14145 | snprintf(buf, sizeof(buf), "%s.%d", bucket_oid_base.c_str(), shard_id); | |
14146 | (*bucket_obj) = buf; | |
14147 | } | |
14148 | } | |
14149 | ||
14150 | int RGWRados::get_bucket_index_object(const string& bucket_oid_base, const string& obj_key, | |
14151 | uint32_t num_shards, RGWBucketInfo::BIShardsHashType hash_type, string *bucket_obj, int *shard_id) | |
14152 | { | |
14153 | int r = 0; | |
14154 | switch (hash_type) { | |
14155 | case RGWBucketInfo::MOD: | |
14156 | if (!num_shards) { | |
14157 | // By default with no sharding, we use the bucket oid as itself | |
14158 | (*bucket_obj) = bucket_oid_base; | |
14159 | if (shard_id) { | |
14160 | *shard_id = -1; | |
14161 | } | |
14162 | } else { | |
14163 | uint32_t sid = rgw_bucket_shard_index(obj_key, num_shards); | |
14164 | char buf[bucket_oid_base.size() + 32]; | |
14165 | snprintf(buf, sizeof(buf), "%s.%d", bucket_oid_base.c_str(), sid); | |
14166 | (*bucket_obj) = buf; | |
14167 | if (shard_id) { | |
14168 | *shard_id = (int)sid; | |
14169 | } | |
14170 | } | |
14171 | break; | |
14172 | default: | |
14173 | r = -ENOTSUP; | |
14174 | } | |
14175 | return r; | |
14176 | } | |
14177 | ||
14178 | void RGWStateLog::oid_str(int shard, string& oid) { | |
14179 | oid = RGW_STATELOG_OBJ_PREFIX + module_name + "."; | |
14180 | char buf[16]; | |
14181 | snprintf(buf, sizeof(buf), "%d", shard); | |
14182 | oid += buf; | |
14183 | } | |
14184 | ||
14185 | int RGWStateLog::get_shard_num(const string& object) { | |
14186 | uint32_t val = ceph_str_hash_linux(object.c_str(), object.length()); | |
14187 | return val % num_shards; | |
14188 | } | |
14189 | ||
14190 | string RGWStateLog::get_oid(const string& object) { | |
14191 | int shard = get_shard_num(object); | |
14192 | string oid; | |
14193 | oid_str(shard, oid); | |
14194 | return oid; | |
14195 | } | |
14196 | ||
14197 | int RGWStateLog::open_ioctx(librados::IoCtx& ioctx) { | |
14198 | rgw_pool pool; | |
14199 | store->get_log_pool(pool); | |
14200 | int r = rgw_init_ioctx(store->get_rados_handle(), pool, ioctx); | |
14201 | if (r < 0) { | |
14202 | lderr(store->ctx()) << "ERROR: could not open rados pool" << dendl; | |
14203 | return r; | |
14204 | } | |
14205 | return 0; | |
14206 | } | |
14207 | ||
14208 | int RGWStateLog::store_entry(const string& client_id, const string& op_id, const string& object, | |
14209 | uint32_t state, bufferlist *bl, uint32_t *check_state) | |
14210 | { | |
14211 | if (client_id.empty() || | |
14212 | op_id.empty() || | |
14213 | object.empty()) { | |
14214 | ldout(store->ctx(), 0) << "client_id / op_id / object is empty" << dendl; | |
14215 | } | |
14216 | ||
14217 | librados::IoCtx ioctx; | |
14218 | int r = open_ioctx(ioctx); | |
14219 | if (r < 0) | |
14220 | return r; | |
14221 | ||
14222 | string oid = get_oid(object); | |
14223 | ||
14224 | librados::ObjectWriteOperation op; | |
14225 | if (check_state) { | |
14226 | cls_statelog_check_state(op, client_id, op_id, object, *check_state); | |
14227 | } | |
14228 | utime_t ts = ceph_clock_now(); | |
14229 | bufferlist nobl; | |
14230 | cls_statelog_add(op, client_id, op_id, object, ts, state, (bl ? *bl : nobl)); | |
14231 | r = ioctx.operate(oid, &op); | |
14232 | if (r < 0) { | |
14233 | return r; | |
14234 | } | |
14235 | ||
14236 | return 0; | |
14237 | } | |
14238 | ||
14239 | int RGWStateLog::remove_entry(const string& client_id, const string& op_id, const string& object) | |
14240 | { | |
14241 | if (client_id.empty() || | |
14242 | op_id.empty() || | |
14243 | object.empty()) { | |
14244 | ldout(store->ctx(), 0) << "client_id / op_id / object is empty" << dendl; | |
14245 | } | |
14246 | ||
14247 | librados::IoCtx ioctx; | |
14248 | int r = open_ioctx(ioctx); | |
14249 | if (r < 0) | |
14250 | return r; | |
14251 | ||
14252 | string oid = get_oid(object); | |
14253 | ||
14254 | librados::ObjectWriteOperation op; | |
14255 | cls_statelog_remove_by_object(op, object, op_id); | |
14256 | r = ioctx.operate(oid, &op); | |
14257 | if (r < 0) { | |
14258 | return r; | |
14259 | } | |
14260 | ||
14261 | return 0; | |
14262 | } | |
14263 | ||
14264 | void RGWStateLog::init_list_entries(const string& client_id, const string& op_id, const string& object, | |
14265 | void **handle) | |
14266 | { | |
14267 | list_state *state = new list_state; | |
14268 | state->client_id = client_id; | |
14269 | state->op_id = op_id; | |
14270 | state->object = object; | |
14271 | if (object.empty()) { | |
14272 | state->cur_shard = 0; | |
14273 | state->max_shard = num_shards - 1; | |
14274 | } else { | |
14275 | state->cur_shard = state->max_shard = get_shard_num(object); | |
14276 | } | |
14277 | *handle = (void *)state; | |
14278 | } | |
14279 | ||
14280 | int RGWStateLog::list_entries(void *handle, int max_entries, | |
14281 | list<cls_statelog_entry>& entries, | |
14282 | bool *done) | |
14283 | { | |
14284 | list_state *state = static_cast<list_state *>(handle); | |
14285 | ||
14286 | librados::IoCtx ioctx; | |
14287 | int r = open_ioctx(ioctx); | |
14288 | if (r < 0) | |
14289 | return r; | |
14290 | ||
14291 | entries.clear(); | |
14292 | ||
14293 | for (; state->cur_shard <= state->max_shard && max_entries > 0; ++state->cur_shard) { | |
14294 | string oid; | |
14295 | oid_str(state->cur_shard, oid); | |
14296 | ||
14297 | librados::ObjectReadOperation op; | |
14298 | list<cls_statelog_entry> ents; | |
14299 | bool truncated; | |
14300 | cls_statelog_list(op, state->client_id, state->op_id, state->object, state->marker, | |
14301 | max_entries, ents, &state->marker, &truncated); | |
14302 | bufferlist ibl; | |
14303 | r = ioctx.operate(oid, &op, &ibl); | |
14304 | if (r == -ENOENT) { | |
14305 | truncated = false; | |
14306 | r = 0; | |
14307 | } | |
14308 | if (r < 0) { | |
14309 | ldout(store->ctx(), 0) << "cls_statelog_list returned " << r << dendl; | |
14310 | return r; | |
14311 | } | |
14312 | ||
14313 | if (!truncated) { | |
14314 | state->marker.clear(); | |
14315 | } | |
14316 | ||
14317 | max_entries -= ents.size(); | |
14318 | ||
14319 | entries.splice(entries.end(), ents); | |
14320 | ||
14321 | if (truncated) | |
14322 | break; | |
14323 | } | |
14324 | ||
14325 | *done = (state->cur_shard > state->max_shard); | |
14326 | ||
14327 | return 0; | |
14328 | } | |
14329 | ||
14330 | void RGWStateLog::finish_list_entries(void *handle) | |
14331 | { | |
14332 | list_state *state = static_cast<list_state *>(handle); | |
14333 | delete state; | |
14334 | } | |
14335 | ||
14336 | void RGWStateLog::dump_entry(const cls_statelog_entry& entry, Formatter *f) | |
14337 | { | |
14338 | f->open_object_section("statelog_entry"); | |
14339 | f->dump_string("client_id", entry.client_id); | |
14340 | f->dump_string("op_id", entry.op_id); | |
14341 | f->dump_string("object", entry.object); | |
14342 | entry.timestamp.gmtime_nsec(f->dump_stream("timestamp")); | |
14343 | if (!dump_entry_internal(entry, f)) { | |
14344 | f->dump_int("state", entry.state); | |
14345 | } | |
14346 | f->close_section(); | |
14347 | } | |
14348 | ||
14349 | RGWOpState::RGWOpState(RGWRados *_store) : RGWStateLog(_store, _store->ctx()->_conf->rgw_num_zone_opstate_shards, string("obj_opstate")) | |
14350 | { | |
14351 | } | |
14352 | ||
14353 | bool RGWOpState::dump_entry_internal(const cls_statelog_entry& entry, Formatter *f) | |
14354 | { | |
14355 | string s; | |
14356 | switch ((OpState)entry.state) { | |
14357 | case OPSTATE_UNKNOWN: | |
14358 | s = "unknown"; | |
14359 | break; | |
14360 | case OPSTATE_IN_PROGRESS: | |
14361 | s = "in-progress"; | |
14362 | break; | |
14363 | case OPSTATE_COMPLETE: | |
14364 | s = "complete"; | |
14365 | break; | |
14366 | case OPSTATE_ERROR: | |
14367 | s = "error"; | |
14368 | break; | |
14369 | case OPSTATE_ABORT: | |
14370 | s = "abort"; | |
14371 | break; | |
14372 | case OPSTATE_CANCELLED: | |
14373 | s = "cancelled"; | |
14374 | break; | |
14375 | default: | |
14376 | s = "invalid"; | |
14377 | } | |
14378 | f->dump_string("state", s); | |
14379 | return true; | |
14380 | } | |
14381 | ||
14382 | int RGWOpState::state_from_str(const string& s, OpState *state) | |
14383 | { | |
14384 | if (s == "unknown") { | |
14385 | *state = OPSTATE_UNKNOWN; | |
14386 | } else if (s == "in-progress") { | |
14387 | *state = OPSTATE_IN_PROGRESS; | |
14388 | } else if (s == "complete") { | |
14389 | *state = OPSTATE_COMPLETE; | |
14390 | } else if (s == "error") { | |
14391 | *state = OPSTATE_ERROR; | |
14392 | } else if (s == "abort") { | |
14393 | *state = OPSTATE_ABORT; | |
14394 | } else if (s == "cancelled") { | |
14395 | *state = OPSTATE_CANCELLED; | |
14396 | } else { | |
14397 | return -EINVAL; | |
14398 | } | |
14399 | ||
14400 | return 0; | |
14401 | } | |
14402 | ||
14403 | int RGWOpState::set_state(const string& client_id, const string& op_id, const string& object, OpState state) | |
14404 | { | |
14405 | uint32_t s = (uint32_t)state; | |
14406 | return store_entry(client_id, op_id, object, s, NULL, NULL); | |
14407 | } | |
14408 | ||
14409 | int RGWOpState::renew_state(const string& client_id, const string& op_id, const string& object, OpState state) | |
14410 | { | |
14411 | uint32_t s = (uint32_t)state; | |
14412 | return store_entry(client_id, op_id, object, s, NULL, &s); | |
14413 | } | |
14414 | ||
14415 | RGWOpStateSingleOp::RGWOpStateSingleOp(RGWRados *store, const string& cid, const string& oid, | |
14416 | const string& obj) : os(store), client_id(cid), op_id(oid), object(obj) | |
14417 | { | |
14418 | cct = store->ctx(); | |
14419 | cur_state = RGWOpState::OPSTATE_UNKNOWN; | |
14420 | } | |
14421 | ||
14422 | int RGWOpStateSingleOp::set_state(RGWOpState::OpState state) { | |
14423 | last_update = real_clock::now(); | |
14424 | cur_state = state; | |
14425 | return os.set_state(client_id, op_id, object, state); | |
14426 | } | |
14427 | ||
14428 | int RGWOpStateSingleOp::renew_state() { | |
14429 | real_time now = real_clock::now(); | |
14430 | ||
14431 | int rate_limit_sec = cct->_conf->rgw_opstate_ratelimit_sec; | |
14432 | ||
14433 | if (rate_limit_sec && now - last_update < make_timespan(rate_limit_sec)) { | |
14434 | return 0; | |
14435 | } | |
14436 | ||
14437 | last_update = now; | |
14438 | return os.renew_state(client_id, op_id, object, cur_state); | |
14439 | } | |
14440 | ||
14441 | ||
14442 | uint64_t RGWRados::instance_id() | |
14443 | { | |
14444 | return get_rados_handle()->get_instance_id(); | |
14445 | } | |
14446 | ||
14447 | uint64_t RGWRados::next_bucket_id() | |
14448 | { | |
14449 | Mutex::Locker l(bucket_id_lock); | |
14450 | return ++max_bucket_id; | |
14451 | } | |
14452 | ||
14453 | RGWRados *RGWStoreManager::init_storage_provider(CephContext *cct, bool use_gc_thread, bool use_lc_thread, | |
14454 | bool quota_threads, bool run_sync_thread, bool run_reshard_thread, bool use_cache) | |
14455 | { | |
14456 | RGWRados *store = NULL; | |
14457 | if (!use_cache) { | |
14458 | store = new RGWRados; | |
14459 | } else { | |
14460 | store = new RGWCache<RGWRados>; | |
14461 | } | |
14462 | ||
14463 | if (store->initialize(cct, use_gc_thread, use_lc_thread, quota_threads, run_sync_thread, run_reshard_thread) < 0) { | |
14464 | delete store; | |
14465 | return NULL; | |
14466 | } | |
14467 | ||
14468 | return store; | |
14469 | } | |
14470 | ||
14471 | RGWRados *RGWStoreManager::init_raw_storage_provider(CephContext *cct) | |
14472 | { | |
14473 | RGWRados *store = NULL; | |
14474 | store = new RGWRados; | |
14475 | ||
14476 | store->set_context(cct); | |
14477 | ||
14478 | if (store->init_rados() < 0) { | |
14479 | delete store; | |
14480 | return NULL; | |
14481 | } | |
14482 | ||
14483 | return store; | |
14484 | } | |
14485 | ||
14486 | void RGWStoreManager::close_storage(RGWRados *store) | |
14487 | { | |
14488 | if (!store) | |
14489 | return; | |
14490 | ||
14491 | store->finalize(); | |
14492 | ||
14493 | delete store; | |
14494 | } | |
14495 | ||
14496 | librados::Rados* RGWRados::get_rados_handle() | |
14497 | { | |
14498 | if (rados.size() == 1) { | |
14499 | return &rados[0]; | |
14500 | } else { | |
14501 | handle_lock.get_read(); | |
14502 | pthread_t id = pthread_self(); | |
14503 | std::map<pthread_t, int>:: iterator it = rados_map.find(id); | |
14504 | ||
14505 | if (it != rados_map.end()) { | |
14506 | handle_lock.put_read(); | |
14507 | return &rados[it->second]; | |
14508 | } else { | |
14509 | handle_lock.put_read(); | |
14510 | handle_lock.get_write(); | |
14511 | const uint32_t handle = next_rados_handle; | |
14512 | rados_map[id] = handle; | |
14513 | if (++next_rados_handle == rados.size()) { | |
14514 | next_rados_handle = 0; | |
14515 | } | |
14516 | handle_lock.put_write(); | |
14517 | return &rados[handle]; | |
14518 | } | |
14519 | } | |
14520 | } | |
14521 | ||
14522 | int RGWRados::delete_raw_obj_aio(const rgw_raw_obj& obj, list<librados::AioCompletion *>& handles) | |
14523 | { | |
14524 | rgw_rados_ref ref; | |
14525 | int ret = get_raw_obj_ref(obj, &ref); | |
14526 | if (ret < 0) { | |
14527 | lderr(cct) << "ERROR: failed to get obj ref with ret=" << ret << dendl; | |
14528 | return ret; | |
14529 | } | |
14530 | ||
14531 | ObjectWriteOperation op; | |
14532 | list<string> prefixes; | |
14533 | cls_rgw_remove_obj(op, prefixes); | |
14534 | ||
14535 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
14536 | ret = ref.ioctx.aio_operate(ref.oid, c, &op); | |
14537 | if (ret < 0) { | |
14538 | lderr(cct) << "ERROR: AioOperate failed with ret=" << ret << dendl; | |
14539 | c->release(); | |
14540 | return ret; | |
14541 | } | |
14542 | ||
14543 | handles.push_back(c); | |
14544 | ||
14545 | return 0; | |
14546 | } | |
14547 | ||
14548 | int RGWRados::delete_obj_aio(const rgw_obj& obj, | |
14549 | RGWBucketInfo& bucket_info, RGWObjState *astate, | |
14550 | list<librados::AioCompletion *>& handles, bool keep_index_consistent) | |
14551 | { | |
14552 | rgw_rados_ref ref; | |
14553 | int ret = get_obj_head_ref(bucket_info, obj, &ref); | |
14554 | if (ret < 0) { | |
14555 | lderr(cct) << "ERROR: failed to get obj ref with ret=" << ret << dendl; | |
14556 | return ret; | |
14557 | } | |
14558 | ||
14559 | if (keep_index_consistent) { | |
14560 | RGWRados::Bucket bop(this, bucket_info); | |
14561 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
14562 | ||
14563 | ret = index_op.prepare(CLS_RGW_OP_DEL, &astate->write_tag); | |
14564 | if (ret < 0) { | |
14565 | lderr(cct) << "ERROR: failed to prepare index op with ret=" << ret << dendl; | |
14566 | return ret; | |
14567 | } | |
14568 | } | |
14569 | ||
14570 | ObjectWriteOperation op; | |
14571 | list<string> prefixes; | |
14572 | cls_rgw_remove_obj(op, prefixes); | |
14573 | ||
14574 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
14575 | ret = ref.ioctx.aio_operate(ref.oid, c, &op); | |
14576 | if (ret < 0) { | |
14577 | lderr(cct) << "ERROR: AioOperate failed with ret=" << ret << dendl; | |
14578 | c->release(); | |
14579 | return ret; | |
14580 | } | |
14581 | ||
14582 | handles.push_back(c); | |
14583 | ||
14584 | if (keep_index_consistent) { | |
14585 | ret = delete_obj_index(obj); | |
14586 | if (ret < 0) { | |
14587 | lderr(cct) << "ERROR: failed to delete obj index with ret=" << ret << dendl; | |
14588 | return ret; | |
14589 | } | |
14590 | } | |
14591 | return ret; | |
14592 | } | |
14593 | ||
14594 | int rgw_compression_info_from_attrset(map<string, bufferlist>& attrs, bool& need_decompress, RGWCompressionInfo& cs_info) { | |
14595 | map<string, bufferlist>::iterator value = attrs.find(RGW_ATTR_COMPRESSION); | |
14596 | if (value != attrs.end()) { | |
14597 | bufferlist::iterator bliter = value->second.begin(); | |
14598 | try { | |
14599 | ::decode(cs_info, bliter); | |
14600 | } catch (buffer::error& err) { | |
14601 | return -EIO; | |
14602 | } | |
14603 | if (cs_info.blocks.size() == 0) { | |
14604 | return -EIO; | |
14605 | } | |
14606 | if (cs_info.compression_type != "none") | |
14607 | need_decompress = true; | |
14608 | else | |
14609 | need_decompress = false; | |
14610 | return 0; | |
14611 | } else { | |
14612 | need_decompress = false; | |
14613 | return 0; | |
14614 | } | |
14615 | } | |
14616 | ||
14617 | bool RGWRados::call(std::string command, cmdmap_t& cmdmap, std::string format, | |
14618 | bufferlist& out) | |
14619 | { | |
14620 | if (command == "cache list") { | |
14621 | boost::optional<std::string> filter; | |
14622 | auto i = cmdmap.find("filter"); | |
14623 | if (i != cmdmap.cend()) { | |
14624 | filter = boost::get<std::string>(i->second); | |
14625 | } | |
14626 | std::unique_ptr<Formatter> f(ceph::Formatter::create(format, "table")); | |
14627 | if (f) { | |
14628 | f->open_array_section("cache_entries"); | |
14629 | call_list(filter, f.get()); | |
14630 | f->close_section(); | |
14631 | f->flush(out); | |
14632 | return true; | |
14633 | } else { | |
14634 | out.append("Unable to create Formatter.\n"); | |
14635 | return false; | |
14636 | } | |
14637 | } else if (command == "cache inspect") { | |
14638 | std::unique_ptr<Formatter> f(ceph::Formatter::create(format, "json-pretty")); | |
14639 | if (f) { | |
14640 | const auto& target = boost::get<std::string>(cmdmap["target"]); | |
14641 | if (call_inspect(target, f.get())) { | |
14642 | f->flush(out); | |
14643 | return true; | |
14644 | } else { | |
14645 | out.append(string("Unable to find entry ") + target + string(".\n")); | |
14646 | return false; | |
14647 | } | |
14648 | } else { | |
14649 | out.append("Unable to create Formatter.\n"); | |
14650 | return false; | |
14651 | } | |
14652 | } else if (command == "cache erase") { | |
14653 | const auto& target = boost::get<std::string>(cmdmap["target"]); | |
14654 | if (call_erase(target)) { | |
14655 | return true; | |
14656 | } else { | |
14657 | out.append(string("Unable to find entry ") + target + string(".\n")); | |
14658 | return false; | |
14659 | } | |
14660 | } else if (command == "cache zap") { | |
14661 | call_zap(); | |
14662 | return true; | |
14663 | } | |
14664 | return false; | |
14665 | } | |
14666 | ||
14667 | void RGWRados::call_list(const boost::optional<std::string>&, | |
14668 | ceph::Formatter*) | |
14669 | { | |
14670 | return; | |
14671 | } | |
14672 | ||
14673 | bool RGWRados::call_inspect(const std::string&, Formatter*) | |
14674 | { | |
14675 | return false; | |
14676 | } | |
14677 | ||
14678 | bool RGWRados::call_erase(const std::string&) { | |
14679 | return false; | |
14680 | } | |
14681 | ||
14682 | void RGWRados::call_zap() { | |
14683 | return; | |
14684 | } |