]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
2 | // vim: ts=8 sw=2 smarttab | |
3 | ||
31f18b77 | 4 | #include "include/compat.h" |
7c673cae FG |
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" | |
c07f9fc5 | 43 | #include "osd/osd_types.h" |
7c673cae FG |
44 | |
45 | #include "rgw_tools.h" | |
46 | #include "rgw_coroutine.h" | |
47 | #include "rgw_compression.h" | |
48 | ||
7c673cae FG |
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" | |
31f18b77 | 73 | #include "rgw_reshard.h" |
7c673cae FG |
74 | |
75 | #include "compressor/Compressor.h" | |
76 | ||
7c673cae FG |
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 { | |
31f18b77 | 133 | *pool = placement.get_data_extra_pool(); |
7c673cae FG |
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()); | |
28e407b8 AA |
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 | } | |
7c673cae FG |
181 | if (r < 0 && r != -EEXIST) { |
182 | return r; | |
183 | } | |
184 | ||
185 | r = rados->ioctx_create(pool.name.c_str(), ioctx); | |
c07f9fc5 FG |
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) { | |
7c673cae FG |
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; | |
7c673cae FG |
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); | |
b32b8144 | 491 | // no default realm exist |
7c673cae | 492 | if (ret < 0) { |
b32b8144 | 493 | return read_id(default_zonegroup_name, default_id); |
7c673cae FG |
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 | ||
7c673cae FG |
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 | ||
224ce89b WB |
1134 | int RGWPeriod::read_latest_epoch(RGWPeriodLatestEpochInfo& info, |
1135 | RGWObjVersionTracker *objv) | |
7c673cae FG |
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); | |
224ce89b | 1142 | int ret = rgw_get_system_obj(store, obj_ctx, pool, oid, bl, objv, nullptr); |
7c673cae FG |
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 | ||
224ce89b WB |
1185 | int RGWPeriod::set_latest_epoch(epoch_t epoch, bool exclusive, |
1186 | RGWObjVersionTracker *objv) | |
7c673cae FG |
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(), | |
224ce89b WB |
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 | |
7c673cae FG |
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; | |
224ce89b | 1314 | return ret; |
7c673cae FG |
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 | { | |
7c673cae FG |
1327 | rgw_pool pool(get_pool(cct)); |
1328 | ||
1329 | string oid = get_period_oid(); | |
1330 | bufferlist bl; | |
1331 | ::encode(*this, bl); | |
224ce89b WB |
1332 | |
1333 | return rgw_put_system_obj(store, pool, oid, bl.c_str(), bl.length(), | |
1334 | exclusive, NULL, real_time(), NULL); | |
7c673cae FG |
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 | ||
7c673cae FG |
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.is_master_zonegroup()) { | |
1392 | master_zonegroup = zg.get_id(); | |
1393 | master_zone = zg.master_zone; | |
1394 | } | |
1395 | ||
1396 | int ret = period_map.update(zg, cct); | |
1397 | if (ret < 0) { | |
1398 | return ret; | |
1399 | } | |
1400 | } | |
1401 | ||
1402 | ret = period_config.read(store, realm_id); | |
1403 | if (ret < 0 && ret != -ENOENT) { | |
1404 | ldout(cct, 0) << "ERROR: failed to read period config: " | |
1405 | << cpp_strerror(ret) << dendl; | |
1406 | return ret; | |
1407 | } | |
1408 | return 0; | |
1409 | } | |
1410 | ||
1411 | int RGWPeriod::reflect() | |
1412 | { | |
1413 | for (auto& iter : period_map.zonegroups) { | |
1414 | RGWZoneGroup& zg = iter.second; | |
1415 | zg.reinit_instance(cct, store); | |
1416 | int r = zg.write(false); | |
1417 | if (r < 0) { | |
1418 | ldout(cct, 0) << "ERROR: failed to store zonegroup info for zonegroup=" << iter.first << ": " << cpp_strerror(-r) << dendl; | |
1419 | return r; | |
1420 | } | |
1421 | if (zg.is_master_zonegroup()) { | |
1422 | // set master as default if no default exists | |
1423 | r = zg.set_as_default(true); | |
1424 | if (r == 0) { | |
1425 | ldout(cct, 1) << "Set the period's master zonegroup " << zg.get_id() | |
1426 | << " as the default" << dendl; | |
1427 | } | |
1428 | } | |
1429 | } | |
1430 | ||
1431 | int r = period_config.write(store, realm_id); | |
1432 | if (r < 0) { | |
1433 | ldout(cct, 0) << "ERROR: failed to store period config: " | |
1434 | << cpp_strerror(-r) << dendl; | |
1435 | return r; | |
1436 | } | |
1437 | return 0; | |
1438 | } | |
1439 | ||
1440 | void RGWPeriod::fork() | |
1441 | { | |
1442 | ldout(cct, 20) << __func__ << " realm " << realm_id << " period " << id << dendl; | |
1443 | predecessor_uuid = id; | |
1444 | id = get_staging_id(realm_id); | |
1445 | period_map.reset(); | |
1446 | realm_epoch++; | |
1447 | } | |
1448 | ||
1449 | static int read_sync_status(RGWRados *store, rgw_meta_sync_status *sync_status) | |
1450 | { | |
1451 | // initialize a sync status manager to read the status | |
1452 | RGWMetaSyncStatusManager mgr(store, store->get_async_rados()); | |
1453 | int r = mgr.init(); | |
1454 | if (r < 0) { | |
1455 | return r; | |
1456 | } | |
1457 | r = mgr.read_sync_status(sync_status); | |
1458 | mgr.stop(); | |
1459 | return r; | |
1460 | } | |
1461 | ||
1462 | int RGWPeriod::update_sync_status(const RGWPeriod ¤t_period, | |
1463 | std::ostream& error_stream, | |
1464 | bool force_if_stale) | |
1465 | { | |
1466 | rgw_meta_sync_status status; | |
1467 | int r = read_sync_status(store, &status); | |
1468 | if (r < 0) { | |
1469 | ldout(cct, 0) << "period failed to read sync status: " | |
1470 | << cpp_strerror(-r) << dendl; | |
1471 | return r; | |
1472 | } | |
1473 | ||
1474 | std::vector<std::string> markers; | |
1475 | ||
1476 | const auto current_epoch = current_period.get_realm_epoch(); | |
1477 | if (current_epoch != status.sync_info.realm_epoch) { | |
1478 | // no sync status markers for the current period | |
1479 | assert(current_epoch > status.sync_info.realm_epoch); | |
1480 | const int behind = current_epoch - status.sync_info.realm_epoch; | |
1481 | if (!force_if_stale && current_epoch > 1) { | |
1482 | error_stream << "ERROR: This zone is " << behind << " period(s) behind " | |
1483 | "the current master zone in metadata sync. If this zone is promoted " | |
1484 | "to master, any metadata changes during that time are likely to " | |
1485 | "be lost.\n" | |
1486 | "Waiting for this zone to catch up on metadata sync (see " | |
1487 | "'radosgw-admin sync status') is recommended.\n" | |
1488 | "To promote this zone to master anyway, add the flag " | |
1489 | "--yes-i-really-mean-it." << std::endl; | |
1490 | return -EINVAL; | |
1491 | } | |
1492 | // empty sync status markers - other zones will skip this period during | |
1493 | // incremental metadata sync | |
1494 | markers.resize(status.sync_info.num_shards); | |
1495 | } else { | |
1496 | markers.reserve(status.sync_info.num_shards); | |
1497 | for (auto& i : status.sync_markers) { | |
1498 | auto& marker = i.second; | |
1499 | // filter out markers from other periods | |
1500 | if (marker.realm_epoch != current_epoch) { | |
1501 | marker.marker.clear(); | |
1502 | } | |
1503 | markers.emplace_back(std::move(marker.marker)); | |
1504 | } | |
1505 | } | |
1506 | ||
1507 | std::swap(sync_status, markers); | |
1508 | return 0; | |
1509 | } | |
1510 | ||
1511 | int RGWPeriod::commit(RGWRealm& realm, const RGWPeriod& current_period, | |
1512 | std::ostream& error_stream, bool force_if_stale) | |
1513 | { | |
1514 | ldout(cct, 20) << __func__ << " realm " << realm.get_id() << " period " << current_period.get_id() << dendl; | |
1515 | // gateway must be in the master zone to commit | |
1516 | if (master_zone != store->get_zone_params().get_id()) { | |
1517 | error_stream << "Cannot commit period on zone " | |
1518 | << store->get_zone_params().get_id() << ", it must be sent to " | |
1519 | "the period's master zone " << master_zone << '.' << std::endl; | |
1520 | return -EINVAL; | |
1521 | } | |
1522 | // period predecessor must match current period | |
1523 | if (predecessor_uuid != current_period.get_id()) { | |
1524 | error_stream << "Period predecessor " << predecessor_uuid | |
1525 | << " does not match current period " << current_period.get_id() | |
1526 | << ". Use 'period pull' to get the latest period from the master, " | |
1527 | "reapply your changes, and try again." << std::endl; | |
1528 | return -EINVAL; | |
1529 | } | |
1530 | // realm epoch must be 1 greater than current period | |
1531 | if (realm_epoch != current_period.get_realm_epoch() + 1) { | |
1532 | error_stream << "Period's realm epoch " << realm_epoch | |
1533 | << " does not come directly after current realm epoch " | |
1534 | << current_period.get_realm_epoch() << ". Use 'realm pull' to get the " | |
1535 | "latest realm and period from the master zone, reapply your changes, " | |
1536 | "and try again." << std::endl; | |
1537 | return -EINVAL; | |
1538 | } | |
1539 | // did the master zone change? | |
1540 | if (master_zone != current_period.get_master_zone()) { | |
1541 | // store the current metadata sync status in the period | |
1542 | int r = update_sync_status(current_period, error_stream, force_if_stale); | |
1543 | if (r < 0) { | |
1544 | ldout(cct, 0) << "failed to update metadata sync status: " | |
1545 | << cpp_strerror(-r) << dendl; | |
1546 | return r; | |
1547 | } | |
1548 | // create an object with a new period id | |
1549 | r = create(true); | |
1550 | if (r < 0) { | |
1551 | ldout(cct, 0) << "failed to create new period: " << cpp_strerror(-r) << dendl; | |
1552 | return r; | |
1553 | } | |
1554 | // set as current period | |
1555 | r = realm.set_current_period(*this); | |
1556 | if (r < 0) { | |
1557 | ldout(cct, 0) << "failed to update realm's current period: " | |
1558 | << cpp_strerror(-r) << dendl; | |
1559 | return r; | |
1560 | } | |
1561 | ldout(cct, 4) << "Promoted to master zone and committed new period " | |
1562 | << id << dendl; | |
1563 | realm.notify_new_period(*this); | |
1564 | return 0; | |
1565 | } | |
1566 | // period must be based on current epoch | |
1567 | if (epoch != current_period.get_epoch()) { | |
1568 | error_stream << "Period epoch " << epoch << " does not match " | |
1569 | "predecessor epoch " << current_period.get_epoch() | |
1570 | << ". Use 'period pull' to get the latest epoch from the master zone, " | |
1571 | "reapply your changes, and try again." << std::endl; | |
1572 | return -EINVAL; | |
1573 | } | |
1574 | // set period as next epoch | |
1575 | set_id(current_period.get_id()); | |
1576 | set_epoch(current_period.get_epoch() + 1); | |
1577 | set_predecessor(current_period.get_predecessor()); | |
1578 | realm_epoch = current_period.get_realm_epoch(); | |
1579 | // write the period to rados | |
1580 | int r = store_info(false); | |
1581 | if (r < 0) { | |
1582 | ldout(cct, 0) << "failed to store period: " << cpp_strerror(-r) << dendl; | |
1583 | return r; | |
1584 | } | |
1585 | // set as latest epoch | |
224ce89b WB |
1586 | r = update_latest_epoch(epoch); |
1587 | if (r == -EEXIST) { | |
1588 | // already have this epoch (or a more recent one) | |
1589 | return 0; | |
1590 | } | |
7c673cae FG |
1591 | if (r < 0) { |
1592 | ldout(cct, 0) << "failed to set latest epoch: " << cpp_strerror(-r) << dendl; | |
1593 | return r; | |
1594 | } | |
1595 | r = reflect(); | |
1596 | if (r < 0) { | |
1597 | ldout(cct, 0) << "failed to update local objects: " << cpp_strerror(-r) << dendl; | |
1598 | return r; | |
1599 | } | |
1600 | ldout(cct, 4) << "Committed new epoch " << epoch | |
1601 | << " for period " << id << dendl; | |
1602 | realm.notify_new_period(*this); | |
1603 | return 0; | |
1604 | } | |
1605 | ||
1606 | int RGWZoneParams::create_default(bool old_format) | |
1607 | { | |
1608 | name = default_zone_name; | |
1609 | ||
1610 | int r = create(); | |
1611 | if (r < 0) { | |
1612 | return r; | |
1613 | } | |
1614 | ||
1615 | if (old_format) { | |
1616 | name = id; | |
1617 | } | |
1618 | ||
1619 | return r; | |
1620 | } | |
1621 | ||
1622 | ||
1623 | int get_zones_pool_set(CephContext* cct, | |
1624 | RGWRados* store, | |
1625 | const list<string>& zones, | |
1626 | const string& my_zone_id, | |
1627 | set<rgw_pool>& pool_names) | |
1628 | { | |
1629 | for(auto const& iter : zones) { | |
1630 | RGWZoneParams zone(iter); | |
1631 | int r = zone.init(cct, store); | |
1632 | if (r < 0) { | |
1633 | ldout(cct, 0) << "Error: init zone " << iter << ":" << cpp_strerror(-r) << dendl; | |
1634 | return r; | |
1635 | } | |
1636 | if (zone.get_id() != my_zone_id) { | |
1637 | pool_names.insert(zone.domain_root); | |
1638 | pool_names.insert(zone.metadata_heap); | |
1639 | pool_names.insert(zone.control_pool); | |
1640 | pool_names.insert(zone.gc_pool); | |
1641 | pool_names.insert(zone.log_pool); | |
1642 | pool_names.insert(zone.intent_log_pool); | |
1643 | pool_names.insert(zone.usage_log_pool); | |
1644 | pool_names.insert(zone.user_keys_pool); | |
1645 | pool_names.insert(zone.user_email_pool); | |
1646 | pool_names.insert(zone.user_swift_pool); | |
1647 | pool_names.insert(zone.user_uid_pool); | |
1648 | pool_names.insert(zone.roles_pool); | |
31f18b77 | 1649 | pool_names.insert(zone.reshard_pool); |
7c673cae FG |
1650 | for(auto& iter : zone.placement_pools) { |
1651 | pool_names.insert(iter.second.index_pool); | |
1652 | pool_names.insert(iter.second.data_pool); | |
1653 | pool_names.insert(iter.second.data_extra_pool); | |
1654 | } | |
1655 | } | |
1656 | } | |
1657 | return 0; | |
1658 | } | |
1659 | ||
1660 | rgw_pool fix_zone_pool_dup(set<rgw_pool> pools, | |
1661 | const string& default_prefix, | |
1662 | const string& default_suffix, | |
1663 | const rgw_pool& suggested_pool) | |
1664 | { | |
1665 | string suggested_name = suggested_pool.to_str(); | |
1666 | ||
1667 | string prefix = default_prefix; | |
1668 | string suffix = default_suffix; | |
1669 | ||
1670 | if (!suggested_pool.empty()) { | |
1671 | prefix = suggested_name.substr(0, suggested_name.find(".")); | |
1672 | suffix = suggested_name.substr(prefix.length()); | |
1673 | } | |
1674 | ||
1675 | rgw_pool pool(prefix + suffix); | |
1676 | ||
1677 | if (pools.find(pool) == pools.end()) { | |
1678 | return pool; | |
1679 | } else { | |
1680 | while(true) { | |
1681 | pool = prefix + "_" + std::to_string(std::rand()) + suffix; | |
1682 | if (pools.find(pool) == pools.end()) { | |
1683 | return pool; | |
1684 | } | |
1685 | } | |
1686 | } | |
1687 | } | |
1688 | ||
1689 | int RGWZoneParams::fix_pool_names() | |
1690 | { | |
1691 | ||
1692 | list<string> zones; | |
1693 | int r = store->list_zones(zones); | |
1694 | if (r < 0) { | |
1695 | ldout(cct, 10) << "WARNING: store->list_zones() returned r=" << r << dendl; | |
1696 | } | |
1697 | ||
1698 | set<rgw_pool> pools; | |
1699 | r = get_zones_pool_set(cct, store, zones, id, pools); | |
1700 | if (r < 0) { | |
1701 | ldout(cct, 0) << "Error: get_zones_pool_names" << r << dendl; | |
1702 | return r; | |
1703 | } | |
1704 | ||
1705 | domain_root = fix_zone_pool_dup(pools, name, ".rgw.meta:root", domain_root); | |
1706 | if (!metadata_heap.name.empty()) { | |
1707 | metadata_heap = fix_zone_pool_dup(pools, name, ".rgw.meta:heap", metadata_heap); | |
1708 | } | |
1709 | control_pool = fix_zone_pool_dup(pools, name, ".rgw.control", control_pool); | |
1710 | gc_pool = fix_zone_pool_dup(pools, name ,".rgw.log:gc", gc_pool); | |
1711 | lc_pool = fix_zone_pool_dup(pools, name ,".rgw.log:lc", lc_pool); | |
1712 | log_pool = fix_zone_pool_dup(pools, name, ".rgw.log", log_pool); | |
1713 | intent_log_pool = fix_zone_pool_dup(pools, name, ".rgw.log:intent", intent_log_pool); | |
1714 | usage_log_pool = fix_zone_pool_dup(pools, name, ".rgw.log:usage", usage_log_pool); | |
1715 | user_keys_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:users.keys", user_keys_pool); | |
1716 | user_email_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:users.email", user_email_pool); | |
1717 | user_swift_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:users.swift", user_swift_pool); | |
1718 | user_uid_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:users.uid", user_uid_pool); | |
1719 | roles_pool = fix_zone_pool_dup(pools, name, ".rgw.meta:roles", roles_pool); | |
31f18b77 | 1720 | reshard_pool = fix_zone_pool_dup(pools, name, ".rgw.log:reshard", reshard_pool); |
7c673cae FG |
1721 | |
1722 | for(auto& iter : placement_pools) { | |
1723 | iter.second.index_pool = fix_zone_pool_dup(pools, name, "." + default_bucket_index_pool_suffix, | |
1724 | iter.second.index_pool); | |
1725 | iter.second.data_pool = fix_zone_pool_dup(pools, name, "." + default_storage_pool_suffix, | |
1726 | iter.second.data_pool); | |
1727 | iter.second.data_extra_pool= fix_zone_pool_dup(pools, name, "." + default_storage_extra_pool_suffix, | |
1728 | iter.second.data_extra_pool); | |
1729 | } | |
1730 | ||
1731 | return 0; | |
1732 | } | |
1733 | ||
1734 | int RGWZoneParams::create(bool exclusive) | |
1735 | { | |
1736 | /* check for old pools config */ | |
1737 | rgw_raw_obj obj(domain_root, avail_pools); | |
1738 | int r = store->raw_obj_stat(obj, NULL, NULL, NULL, NULL, NULL, NULL); | |
1739 | if (r < 0) { | |
1740 | ldout(store->ctx(), 10) << "couldn't find old data placement pools config, setting up new ones for the zone" << dendl; | |
1741 | /* a new system, let's set new placement info */ | |
1742 | RGWZonePlacementInfo default_placement; | |
1743 | default_placement.index_pool = name + "." + default_bucket_index_pool_suffix; | |
1744 | default_placement.data_pool = name + "." + default_storage_pool_suffix; | |
1745 | default_placement.data_extra_pool = name + "." + default_storage_extra_pool_suffix; | |
1746 | placement_pools["default-placement"] = default_placement; | |
1747 | } | |
1748 | ||
1749 | r = fix_pool_names(); | |
1750 | if (r < 0) { | |
1751 | ldout(cct, 0) << "ERROR: fix_pool_names returned r=" << r << dendl; | |
1752 | return r; | |
1753 | } | |
1754 | ||
1755 | r = RGWSystemMetaObj::create(exclusive); | |
1756 | if (r < 0) { | |
1757 | return r; | |
1758 | } | |
1759 | ||
1760 | // try to set as default. may race with another create, so pass exclusive=true | |
1761 | // so we don't override an existing default | |
1762 | r = set_as_default(true); | |
1763 | if (r < 0 && r != -EEXIST) { | |
1764 | ldout(cct, 10) << "WARNING: failed to set zone as default, r=" << r << dendl; | |
1765 | } | |
1766 | ||
1767 | return 0; | |
1768 | } | |
1769 | ||
1770 | rgw_pool RGWZoneParams::get_pool(CephContext *cct) | |
1771 | { | |
1772 | if (cct->_conf->rgw_zone_root_pool.empty()) { | |
1773 | return rgw_pool(RGW_DEFAULT_ZONE_ROOT_POOL); | |
1774 | } | |
1775 | ||
1776 | return rgw_pool(cct->_conf->rgw_zone_root_pool); | |
1777 | } | |
1778 | ||
1779 | const string RGWZoneParams::get_default_oid(bool old_format) | |
1780 | { | |
1781 | if (old_format) { | |
1782 | return cct->_conf->rgw_default_zone_info_oid; | |
1783 | } | |
1784 | ||
1785 | return cct->_conf->rgw_default_zone_info_oid + "." + realm_id; | |
1786 | } | |
1787 | ||
1788 | const string& RGWZoneParams::get_names_oid_prefix() | |
1789 | { | |
1790 | return zone_names_oid_prefix; | |
1791 | } | |
1792 | ||
1793 | const string& RGWZoneParams::get_info_oid_prefix(bool old_format) | |
1794 | { | |
1795 | return zone_info_oid_prefix; | |
1796 | } | |
1797 | ||
1798 | const string& RGWZoneParams::get_predefined_name(CephContext *cct) { | |
1799 | return cct->_conf->rgw_zone; | |
1800 | } | |
1801 | ||
1802 | int RGWZoneParams::init(CephContext *cct, RGWRados *store, bool setup_obj, bool old_format) | |
1803 | { | |
1804 | if (name.empty()) { | |
1805 | name = cct->_conf->rgw_zone; | |
1806 | } | |
1807 | ||
1808 | return RGWSystemMetaObj::init(cct, store, setup_obj, old_format); | |
1809 | } | |
1810 | ||
1811 | int RGWZoneParams::read_default_id(string& default_id, bool old_format) | |
1812 | { | |
1813 | if (realm_id.empty()) { | |
1814 | /* try using default realm */ | |
1815 | RGWRealm realm; | |
1816 | int ret = realm.init(cct, store); | |
b32b8144 | 1817 | //no default realm exist |
7c673cae | 1818 | if (ret < 0) { |
b32b8144 | 1819 | return read_id(default_zone_name, default_id); |
7c673cae FG |
1820 | } |
1821 | realm_id = realm.get_id(); | |
1822 | } | |
1823 | ||
1824 | return RGWSystemMetaObj::read_default_id(default_id, old_format); | |
1825 | } | |
1826 | ||
1827 | ||
1828 | int RGWZoneParams::set_as_default(bool exclusive) | |
1829 | { | |
1830 | if (realm_id.empty()) { | |
1831 | /* try using default realm */ | |
1832 | RGWRealm realm; | |
1833 | int ret = realm.init(cct, store); | |
1834 | if (ret < 0) { | |
1835 | ldout(cct, 10) << "could not read realm id: " << cpp_strerror(-ret) << dendl; | |
1836 | return -EINVAL; | |
1837 | } | |
1838 | realm_id = realm.get_id(); | |
1839 | } | |
1840 | ||
1841 | return RGWSystemMetaObj::set_as_default(exclusive); | |
1842 | } | |
1843 | ||
1844 | const string& RGWZoneParams::get_compression_type(const string& placement_rule) const | |
1845 | { | |
1846 | static const std::string NONE{"none"}; | |
1847 | auto p = placement_pools.find(placement_rule); | |
1848 | if (p == placement_pools.end()) { | |
1849 | return NONE; | |
1850 | } | |
1851 | const auto& type = p->second.compression_type; | |
1852 | return !type.empty() ? type : NONE; | |
1853 | } | |
1854 | ||
1855 | void RGWPeriodMap::encode(bufferlist& bl) const { | |
1856 | ENCODE_START(2, 1, bl); | |
1857 | ::encode(id, bl); | |
1858 | ::encode(zonegroups, bl); | |
1859 | ::encode(master_zonegroup, bl); | |
1860 | ::encode(short_zone_ids, bl); | |
1861 | ENCODE_FINISH(bl); | |
1862 | } | |
1863 | ||
1864 | void RGWPeriodMap::decode(bufferlist::iterator& bl) { | |
1865 | DECODE_START(2, bl); | |
1866 | ::decode(id, bl); | |
1867 | ::decode(zonegroups, bl); | |
1868 | ::decode(master_zonegroup, bl); | |
1869 | if (struct_v >= 2) { | |
1870 | ::decode(short_zone_ids, bl); | |
1871 | } | |
1872 | DECODE_FINISH(bl); | |
1873 | ||
1874 | zonegroups_by_api.clear(); | |
1875 | for (map<string, RGWZoneGroup>::iterator iter = zonegroups.begin(); | |
1876 | iter != zonegroups.end(); ++iter) { | |
1877 | RGWZoneGroup& zonegroup = iter->second; | |
1878 | zonegroups_by_api[zonegroup.api_name] = zonegroup; | |
31f18b77 | 1879 | if (zonegroup.is_master_zonegroup()) { |
7c673cae FG |
1880 | master_zonegroup = zonegroup.get_id(); |
1881 | } | |
1882 | } | |
1883 | } | |
1884 | ||
1885 | // run an MD5 hash on the zone_id and return the first 32 bits | |
1886 | static uint32_t gen_short_zone_id(const std::string zone_id) | |
1887 | { | |
1888 | unsigned char md5[CEPH_CRYPTO_MD5_DIGESTSIZE]; | |
1889 | MD5 hash; | |
1890 | hash.Update((const byte *)zone_id.c_str(), zone_id.size()); | |
1891 | hash.Final(md5); | |
1892 | ||
1893 | uint32_t short_id; | |
1894 | memcpy((char *)&short_id, md5, sizeof(short_id)); | |
1895 | return std::max(short_id, 1u); | |
1896 | } | |
1897 | ||
1898 | int RGWPeriodMap::update(const RGWZoneGroup& zonegroup, CephContext *cct) | |
1899 | { | |
31f18b77 | 1900 | if (zonegroup.is_master_zonegroup() && (!master_zonegroup.empty() && zonegroup.get_id() != master_zonegroup)) { |
7c673cae FG |
1901 | ldout(cct,0) << "Error updating periodmap, multiple master zonegroups configured "<< dendl; |
1902 | ldout(cct,0) << "master zonegroup: " << master_zonegroup << " and " << zonegroup.get_id() <<dendl; | |
1903 | return -EINVAL; | |
1904 | } | |
1905 | map<string, RGWZoneGroup>::iterator iter = zonegroups.find(zonegroup.get_id()); | |
1906 | if (iter != zonegroups.end()) { | |
1907 | RGWZoneGroup& old_zonegroup = iter->second; | |
1908 | if (!old_zonegroup.api_name.empty()) { | |
1909 | zonegroups_by_api.erase(old_zonegroup.api_name); | |
1910 | } | |
1911 | } | |
1912 | zonegroups[zonegroup.get_id()] = zonegroup; | |
1913 | ||
1914 | if (!zonegroup.api_name.empty()) { | |
1915 | zonegroups_by_api[zonegroup.api_name] = zonegroup; | |
1916 | } | |
1917 | ||
31f18b77 | 1918 | if (zonegroup.is_master_zonegroup()) { |
7c673cae FG |
1919 | master_zonegroup = zonegroup.get_id(); |
1920 | } else if (master_zonegroup == zonegroup.get_id()) { | |
1921 | master_zonegroup = ""; | |
1922 | } | |
1923 | ||
1924 | for (auto& i : zonegroup.zones) { | |
1925 | auto& zone = i.second; | |
1926 | if (short_zone_ids.find(zone.id) != short_zone_ids.end()) { | |
1927 | continue; | |
1928 | } | |
1929 | // calculate the zone's short id | |
1930 | uint32_t short_id = gen_short_zone_id(zone.id); | |
1931 | ||
1932 | // search for an existing zone with the same short id | |
1933 | for (auto& s : short_zone_ids) { | |
1934 | if (s.second == short_id) { | |
1935 | ldout(cct, 0) << "New zone '" << zone.name << "' (" << zone.id | |
1936 | << ") generates the same short_zone_id " << short_id | |
1937 | << " as existing zone id " << s.first << dendl; | |
1938 | return -EEXIST; | |
1939 | } | |
1940 | } | |
1941 | ||
1942 | short_zone_ids[zone.id] = short_id; | |
1943 | } | |
1944 | ||
1945 | return 0; | |
1946 | } | |
1947 | ||
1948 | uint32_t RGWPeriodMap::get_zone_short_id(const string& zone_id) const | |
1949 | { | |
1950 | auto i = short_zone_ids.find(zone_id); | |
1951 | if (i == short_zone_ids.end()) { | |
1952 | return 0; | |
1953 | } | |
1954 | return i->second; | |
1955 | } | |
1956 | ||
1957 | int RGWZoneGroupMap::read(CephContext *cct, RGWRados *store) | |
1958 | { | |
1959 | ||
1960 | RGWPeriod period; | |
1961 | int ret = period.init(cct, store); | |
1962 | if (ret < 0) { | |
1963 | cerr << "failed to read current period info: " << cpp_strerror(ret); | |
1964 | return ret; | |
1965 | } | |
1966 | ||
1967 | bucket_quota = period.get_config().bucket_quota; | |
1968 | user_quota = period.get_config().user_quota; | |
1969 | zonegroups = period.get_map().zonegroups; | |
1970 | zonegroups_by_api = period.get_map().zonegroups_by_api; | |
1971 | master_zonegroup = period.get_map().master_zonegroup; | |
1972 | ||
1973 | return 0; | |
1974 | } | |
1975 | ||
1976 | void RGWRegionMap::encode(bufferlist& bl) const { | |
1977 | ENCODE_START( 3, 1, bl); | |
1978 | ::encode(regions, bl); | |
1979 | ::encode(master_region, bl); | |
1980 | ::encode(bucket_quota, bl); | |
1981 | ::encode(user_quota, bl); | |
1982 | ENCODE_FINISH(bl); | |
1983 | } | |
1984 | ||
1985 | void RGWRegionMap::decode(bufferlist::iterator& bl) { | |
1986 | DECODE_START(3, bl); | |
1987 | ::decode(regions, bl); | |
1988 | ::decode(master_region, bl); | |
1989 | if (struct_v >= 2) | |
1990 | ::decode(bucket_quota, bl); | |
1991 | if (struct_v >= 3) | |
1992 | ::decode(user_quota, bl); | |
1993 | DECODE_FINISH(bl); | |
1994 | } | |
1995 | ||
1996 | void RGWZoneGroupMap::encode(bufferlist& bl) const { | |
1997 | ENCODE_START( 3, 1, bl); | |
1998 | ::encode(zonegroups, bl); | |
1999 | ::encode(master_zonegroup, bl); | |
2000 | ::encode(bucket_quota, bl); | |
2001 | ::encode(user_quota, bl); | |
2002 | ENCODE_FINISH(bl); | |
2003 | } | |
2004 | ||
2005 | void RGWZoneGroupMap::decode(bufferlist::iterator& bl) { | |
2006 | DECODE_START(3, bl); | |
2007 | ::decode(zonegroups, bl); | |
2008 | ::decode(master_zonegroup, bl); | |
2009 | if (struct_v >= 2) | |
2010 | ::decode(bucket_quota, bl); | |
2011 | if (struct_v >= 3) | |
2012 | ::decode(user_quota, bl); | |
2013 | DECODE_FINISH(bl); | |
2014 | ||
2015 | zonegroups_by_api.clear(); | |
2016 | for (map<string, RGWZoneGroup>::iterator iter = zonegroups.begin(); | |
2017 | iter != zonegroups.end(); ++iter) { | |
2018 | RGWZoneGroup& zonegroup = iter->second; | |
2019 | zonegroups_by_api[zonegroup.api_name] = zonegroup; | |
31f18b77 | 2020 | if (zonegroup.is_master_zonegroup()) { |
7c673cae FG |
2021 | master_zonegroup = zonegroup.get_name(); |
2022 | } | |
2023 | } | |
2024 | } | |
2025 | ||
2026 | void RGWObjVersionTracker::prepare_op_for_read(ObjectReadOperation *op) | |
2027 | { | |
2028 | obj_version *check_objv = version_for_check(); | |
2029 | ||
2030 | if (check_objv) { | |
2031 | cls_version_check(*op, *check_objv, VER_COND_EQ); | |
2032 | } | |
2033 | ||
2034 | cls_version_read(*op, &read_version); | |
2035 | } | |
2036 | ||
2037 | void RGWObjVersionTracker::prepare_op_for_write(ObjectWriteOperation *op) | |
2038 | { | |
2039 | obj_version *check_objv = version_for_check(); | |
2040 | obj_version *modify_version = version_for_write(); | |
2041 | ||
2042 | if (check_objv) { | |
2043 | cls_version_check(*op, *check_objv, VER_COND_EQ); | |
2044 | } | |
2045 | ||
2046 | if (modify_version) { | |
2047 | cls_version_set(*op, *modify_version); | |
2048 | } else { | |
2049 | cls_version_inc(*op); | |
2050 | } | |
2051 | } | |
2052 | ||
2053 | void RGWObjManifest::obj_iterator::operator++() | |
2054 | { | |
2055 | if (manifest->explicit_objs) { | |
2056 | ++explicit_iter; | |
2057 | ||
2058 | if (explicit_iter == manifest->objs.end()) { | |
2059 | ofs = manifest->obj_size; | |
2060 | return; | |
2061 | } | |
2062 | ||
2063 | update_explicit_pos(); | |
2064 | ||
2065 | update_location(); | |
2066 | return; | |
2067 | } | |
2068 | ||
2069 | uint64_t obj_size = manifest->get_obj_size(); | |
2070 | uint64_t head_size = manifest->get_head_size(); | |
2071 | ||
2072 | if (ofs == obj_size) { | |
2073 | return; | |
2074 | } | |
2075 | ||
2076 | if (manifest->rules.empty()) { | |
2077 | return; | |
2078 | } | |
2079 | ||
2080 | /* are we still pointing at the head? */ | |
2081 | if (ofs < head_size) { | |
2082 | rule_iter = manifest->rules.begin(); | |
2083 | RGWObjManifestRule *rule = &rule_iter->second; | |
2084 | ofs = MIN(head_size, obj_size); | |
2085 | stripe_ofs = ofs; | |
2086 | cur_stripe = 1; | |
2087 | stripe_size = MIN(obj_size - ofs, rule->stripe_max_size); | |
2088 | if (rule->part_size > 0) { | |
2089 | stripe_size = MIN(stripe_size, rule->part_size); | |
2090 | } | |
2091 | update_location(); | |
2092 | return; | |
2093 | } | |
2094 | ||
2095 | RGWObjManifestRule *rule = &rule_iter->second; | |
2096 | ||
2097 | stripe_ofs += rule->stripe_max_size; | |
2098 | cur_stripe++; | |
2099 | dout(20) << "RGWObjManifest::operator++(): rule->part_size=" << rule->part_size << " rules.size()=" << manifest->rules.size() << dendl; | |
2100 | ||
2101 | if (rule->part_size > 0) { | |
2102 | /* multi part, multi stripes object */ | |
2103 | ||
2104 | dout(20) << "RGWObjManifest::operator++(): stripe_ofs=" << stripe_ofs << " part_ofs=" << part_ofs << " rule->part_size=" << rule->part_size << dendl; | |
2105 | ||
2106 | if (stripe_ofs >= part_ofs + rule->part_size) { | |
2107 | /* moved to the next part */ | |
2108 | cur_stripe = 0; | |
2109 | part_ofs += rule->part_size; | |
2110 | stripe_ofs = part_ofs; | |
2111 | ||
2112 | bool last_rule = (next_rule_iter == manifest->rules.end()); | |
2113 | /* move to the next rule? */ | |
2114 | if (!last_rule && stripe_ofs >= next_rule_iter->second.start_ofs) { | |
2115 | rule_iter = next_rule_iter; | |
2116 | last_rule = (next_rule_iter == manifest->rules.end()); | |
2117 | if (!last_rule) { | |
2118 | ++next_rule_iter; | |
2119 | } | |
2120 | cur_part_id = rule_iter->second.start_part_num; | |
2121 | } else { | |
2122 | cur_part_id++; | |
2123 | } | |
2124 | ||
2125 | rule = &rule_iter->second; | |
2126 | } | |
2127 | ||
2128 | stripe_size = MIN(rule->part_size - (stripe_ofs - part_ofs), rule->stripe_max_size); | |
2129 | } | |
2130 | ||
2131 | cur_override_prefix = rule->override_prefix; | |
2132 | ||
2133 | ofs = stripe_ofs; | |
2134 | if (ofs > obj_size) { | |
2135 | ofs = obj_size; | |
2136 | stripe_ofs = ofs; | |
2137 | stripe_size = 0; | |
2138 | } | |
2139 | ||
2140 | dout(20) << "RGWObjManifest::operator++(): result: ofs=" << ofs << " stripe_ofs=" << stripe_ofs << " part_ofs=" << part_ofs << " rule->part_size=" << rule->part_size << dendl; | |
2141 | update_location(); | |
2142 | } | |
2143 | ||
2144 | int RGWObjManifest::generator::create_begin(CephContext *cct, RGWObjManifest *_m, const string& placement_rule, rgw_bucket& _b, rgw_obj& _obj) | |
2145 | { | |
2146 | manifest = _m; | |
2147 | ||
2148 | manifest->set_tail_placement(placement_rule, _b); | |
2149 | manifest->set_head(placement_rule, _obj, 0); | |
2150 | last_ofs = 0; | |
2151 | ||
2152 | if (manifest->get_prefix().empty()) { | |
2153 | char buf[33]; | |
2154 | gen_rand_alphanumeric(cct, buf, sizeof(buf) - 1); | |
2155 | ||
2156 | string oid_prefix = "."; | |
2157 | oid_prefix.append(buf); | |
2158 | oid_prefix.append("_"); | |
2159 | ||
2160 | manifest->set_prefix(oid_prefix); | |
2161 | } | |
2162 | ||
2163 | bool found = manifest->get_rule(0, &rule); | |
2164 | if (!found) { | |
2165 | derr << "ERROR: manifest->get_rule() could not find rule" << dendl; | |
2166 | return -EIO; | |
2167 | } | |
2168 | ||
2169 | uint64_t head_size = manifest->get_head_size(); | |
2170 | ||
2171 | if (head_size > 0) { | |
2172 | cur_stripe_size = head_size; | |
2173 | } else { | |
2174 | cur_stripe_size = rule.stripe_max_size; | |
2175 | } | |
2176 | ||
2177 | cur_part_id = rule.start_part_num; | |
2178 | ||
2179 | manifest->get_implicit_location(cur_part_id, cur_stripe, 0, NULL, &cur_obj); | |
2180 | ||
2181 | // Normal object which not generated through copy operation | |
2182 | manifest->set_tail_instance(_obj.key.instance); | |
2183 | ||
2184 | manifest->update_iterators(); | |
2185 | ||
2186 | return 0; | |
2187 | } | |
2188 | ||
2189 | int RGWObjManifest::generator::create_next(uint64_t ofs) | |
2190 | { | |
2191 | if (ofs < last_ofs) /* only going forward */ | |
2192 | return -EINVAL; | |
2193 | ||
2194 | uint64_t max_head_size = manifest->get_max_head_size(); | |
2195 | ||
2196 | if (ofs < max_head_size) { | |
2197 | manifest->set_head_size(ofs); | |
2198 | } | |
2199 | ||
2200 | if (ofs >= max_head_size) { | |
2201 | manifest->set_head_size(max_head_size); | |
2202 | cur_stripe = (ofs - max_head_size) / rule.stripe_max_size; | |
2203 | cur_stripe_size = rule.stripe_max_size; | |
2204 | ||
2205 | if (cur_part_id == 0 && max_head_size > 0) { | |
2206 | cur_stripe++; | |
2207 | } | |
2208 | } | |
2209 | ||
2210 | last_ofs = ofs; | |
2211 | manifest->set_obj_size(ofs); | |
2212 | ||
2213 | manifest->get_implicit_location(cur_part_id, cur_stripe, ofs, NULL, &cur_obj); | |
2214 | ||
2215 | manifest->update_iterators(); | |
2216 | ||
2217 | return 0; | |
2218 | } | |
2219 | ||
2220 | const RGWObjManifest::obj_iterator& RGWObjManifest::obj_begin() | |
2221 | { | |
2222 | return begin_iter; | |
2223 | } | |
2224 | ||
2225 | const RGWObjManifest::obj_iterator& RGWObjManifest::obj_end() | |
2226 | { | |
2227 | return end_iter; | |
2228 | } | |
2229 | ||
2230 | RGWObjManifest::obj_iterator RGWObjManifest::obj_find(uint64_t ofs) | |
2231 | { | |
2232 | if (ofs > obj_size) { | |
2233 | ofs = obj_size; | |
2234 | } | |
2235 | RGWObjManifest::obj_iterator iter(this); | |
2236 | iter.seek(ofs); | |
2237 | return iter; | |
2238 | } | |
2239 | ||
2240 | int RGWObjManifest::append(RGWObjManifest& m, RGWZoneGroup& zonegroup, RGWZoneParams& zone_params) | |
2241 | { | |
2242 | if (explicit_objs || m.explicit_objs) { | |
2243 | return append_explicit(m, zonegroup, zone_params); | |
2244 | } | |
2245 | ||
2246 | if (rules.empty()) { | |
2247 | *this = m; | |
2248 | return 0; | |
2249 | } | |
2250 | ||
2251 | string override_prefix; | |
2252 | ||
2253 | if (prefix.empty()) { | |
2254 | prefix = m.prefix; | |
2255 | } | |
2256 | ||
2257 | if (prefix != m.prefix) { | |
2258 | override_prefix = m.prefix; | |
2259 | } | |
2260 | ||
2261 | map<uint64_t, RGWObjManifestRule>::iterator miter = m.rules.begin(); | |
2262 | if (miter == m.rules.end()) { | |
2263 | return append_explicit(m, zonegroup, zone_params); | |
2264 | } | |
2265 | ||
2266 | for (; miter != m.rules.end(); ++miter) { | |
2267 | map<uint64_t, RGWObjManifestRule>::reverse_iterator last_rule = rules.rbegin(); | |
2268 | ||
2269 | RGWObjManifestRule& rule = last_rule->second; | |
2270 | ||
2271 | if (rule.part_size == 0) { | |
2272 | rule.part_size = obj_size - rule.start_ofs; | |
2273 | } | |
2274 | ||
2275 | RGWObjManifestRule& next_rule = miter->second; | |
2276 | if (!next_rule.part_size) { | |
2277 | next_rule.part_size = m.obj_size - next_rule.start_ofs; | |
2278 | } | |
2279 | ||
2280 | string rule_prefix = prefix; | |
2281 | if (!rule.override_prefix.empty()) { | |
2282 | rule_prefix = rule.override_prefix; | |
2283 | } | |
2284 | ||
2285 | string next_rule_prefix = m.prefix; | |
2286 | if (!next_rule.override_prefix.empty()) { | |
2287 | next_rule_prefix = next_rule.override_prefix; | |
2288 | } | |
2289 | ||
2290 | if (rule.part_size != next_rule.part_size || | |
2291 | rule.stripe_max_size != next_rule.stripe_max_size || | |
2292 | rule_prefix != next_rule_prefix) { | |
2293 | if (next_rule_prefix != prefix) { | |
2294 | append_rules(m, miter, &next_rule_prefix); | |
2295 | } else { | |
2296 | append_rules(m, miter, NULL); | |
2297 | } | |
2298 | break; | |
2299 | } | |
2300 | ||
2301 | uint64_t expected_part_num = rule.start_part_num + 1; | |
2302 | if (rule.part_size > 0) { | |
2303 | expected_part_num = rule.start_part_num + (obj_size + next_rule.start_ofs - rule.start_ofs) / rule.part_size; | |
2304 | } | |
2305 | ||
2306 | if (expected_part_num != next_rule.start_part_num) { | |
2307 | append_rules(m, miter, NULL); | |
2308 | break; | |
2309 | } | |
2310 | } | |
2311 | ||
2312 | set_obj_size(obj_size + m.obj_size); | |
2313 | ||
2314 | return 0; | |
2315 | } | |
2316 | ||
2317 | int RGWObjManifest::append(RGWObjManifest& m, RGWRados *store) | |
2318 | { | |
2319 | return append(m, store->get_zonegroup(), store->get_zone_params()); | |
2320 | } | |
2321 | ||
2322 | void RGWObjManifest::append_rules(RGWObjManifest& m, map<uint64_t, RGWObjManifestRule>::iterator& miter, | |
2323 | string *override_prefix) | |
2324 | { | |
2325 | for (; miter != m.rules.end(); ++miter) { | |
2326 | RGWObjManifestRule rule = miter->second; | |
2327 | rule.start_ofs += obj_size; | |
2328 | if (override_prefix) | |
2329 | rule.override_prefix = *override_prefix; | |
2330 | rules[rule.start_ofs] = rule; | |
2331 | } | |
2332 | } | |
2333 | ||
2334 | void RGWObjManifest::convert_to_explicit(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params) | |
2335 | { | |
2336 | if (explicit_objs) { | |
2337 | return; | |
2338 | } | |
2339 | obj_iterator iter = obj_begin(); | |
2340 | ||
2341 | while (iter != obj_end()) { | |
2342 | RGWObjManifestPart& part = objs[iter.get_stripe_ofs()]; | |
2343 | const rgw_obj_select& os = iter.get_location(); | |
2344 | const rgw_raw_obj& raw_loc = os.get_raw_obj(zonegroup, zone_params); | |
2345 | part.loc_ofs = 0; | |
2346 | ||
2347 | uint64_t ofs = iter.get_stripe_ofs(); | |
2348 | ||
2349 | if (ofs == 0) { | |
2350 | part.loc = obj; | |
2351 | } else { | |
2352 | rgw_raw_obj_to_obj(tail_placement.bucket, raw_loc, &part.loc); | |
2353 | } | |
2354 | ++iter; | |
2355 | uint64_t next_ofs = iter.get_stripe_ofs(); | |
2356 | ||
2357 | part.size = next_ofs - ofs; | |
2358 | } | |
2359 | ||
2360 | explicit_objs = true; | |
2361 | rules.clear(); | |
2362 | prefix.clear(); | |
2363 | } | |
2364 | ||
2365 | int RGWObjManifest::append_explicit(RGWObjManifest& m, const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params) | |
2366 | { | |
2367 | if (!explicit_objs) { | |
2368 | convert_to_explicit(zonegroup, zone_params); | |
2369 | } | |
2370 | if (!m.explicit_objs) { | |
2371 | m.convert_to_explicit(zonegroup, zone_params); | |
2372 | } | |
2373 | map<uint64_t, RGWObjManifestPart>::iterator iter; | |
2374 | uint64_t base = obj_size; | |
2375 | for (iter = m.objs.begin(); iter != m.objs.end(); ++iter) { | |
2376 | RGWObjManifestPart& part = iter->second; | |
2377 | objs[base + iter->first] = part; | |
2378 | } | |
2379 | obj_size += m.obj_size; | |
2380 | ||
2381 | return 0; | |
2382 | } | |
2383 | ||
2384 | bool RGWObjManifest::get_rule(uint64_t ofs, RGWObjManifestRule *rule) | |
2385 | { | |
2386 | if (rules.empty()) { | |
2387 | return false; | |
2388 | } | |
2389 | ||
2390 | map<uint64_t, RGWObjManifestRule>::iterator iter = rules.upper_bound(ofs); | |
2391 | if (iter != rules.begin()) { | |
2392 | --iter; | |
2393 | } | |
2394 | ||
2395 | *rule = iter->second; | |
2396 | ||
2397 | return true; | |
2398 | } | |
2399 | ||
2400 | void RGWObjVersionTracker::generate_new_write_ver(CephContext *cct) | |
2401 | { | |
2402 | write_version.ver = 1; | |
2403 | #define TAG_LEN 24 | |
2404 | ||
2405 | write_version.tag.clear(); | |
2406 | append_rand_alpha(cct, write_version.tag, write_version.tag, TAG_LEN); | |
2407 | } | |
2408 | ||
2409 | int RGWPutObjProcessor::complete(size_t accounted_size, const string& etag, | |
2410 | real_time *mtime, real_time set_mtime, | |
2411 | map<string, bufferlist>& attrs, real_time delete_at, | |
31f18b77 FG |
2412 | const char *if_match, const char *if_nomatch, const string *user_data, |
2413 | rgw_zone_set *zones_trace) | |
7c673cae | 2414 | { |
31f18b77 | 2415 | int r = do_complete(accounted_size, etag, mtime, set_mtime, attrs, delete_at, if_match, if_nomatch, user_data, zones_trace); |
7c673cae FG |
2416 | if (r < 0) |
2417 | return r; | |
2418 | ||
2419 | is_complete = !canceled; | |
2420 | return 0; | |
2421 | } | |
2422 | ||
2423 | CephContext *RGWPutObjProcessor::ctx() | |
2424 | { | |
2425 | return store->ctx(); | |
2426 | } | |
2427 | ||
2428 | RGWPutObjProcessor_Aio::~RGWPutObjProcessor_Aio() | |
2429 | { | |
2430 | drain_pending(); | |
2431 | ||
2432 | if (is_complete) | |
2433 | return; | |
2434 | ||
2435 | set<rgw_raw_obj>::iterator iter; | |
2436 | bool need_to_remove_head = false; | |
2437 | rgw_raw_obj raw_head; | |
2438 | ||
2439 | if (!head_obj.empty()) { | |
2440 | store->obj_to_raw(bucket_info.placement_rule, head_obj, &raw_head); | |
2441 | } | |
2442 | ||
2443 | /** | |
2444 | * We should delete the object in the "multipart" namespace to avoid race condition. | |
2445 | * Such race condition is caused by the fact that the multipart object is the gatekeeper of a multipart | |
2446 | * upload, when it is deleted, a second upload would start with the same suffix("2/"), therefore, objects | |
2447 | * written by the second upload may be deleted by the first upload. | |
2448 | * details is describled on #11749 | |
2449 | * | |
2450 | * The above comment still stands, but instead of searching for a specific object in the multipart | |
2451 | * namespace, we just make sure that we remove the object that is marked as the head object after | |
2452 | * we remove all the other raw objects. Note that we use different call to remove the head object, | |
2453 | * as this one needs to go via the bucket index prepare/complete 2-phase commit scheme. | |
2454 | */ | |
2455 | for (iter = written_objs.begin(); iter != written_objs.end(); ++iter) { | |
2456 | const rgw_raw_obj& obj = *iter; | |
2457 | if (!head_obj.empty() && obj == raw_head) { | |
2458 | ldout(store->ctx(), 5) << "NOTE: we should not process the head object (" << obj << ") here" << dendl; | |
2459 | need_to_remove_head = true; | |
2460 | continue; | |
2461 | } | |
2462 | ||
2463 | int r = store->delete_raw_obj(obj); | |
2464 | if (r < 0 && r != -ENOENT) { | |
2465 | ldout(store->ctx(), 5) << "WARNING: failed to remove obj (" << obj << "), leaked" << dendl; | |
2466 | } | |
2467 | } | |
2468 | ||
2469 | if (need_to_remove_head) { | |
2470 | ldout(store->ctx(), 5) << "NOTE: we are going to process the head obj (" << raw_head << ")" << dendl; | |
2471 | int r = store->delete_obj(obj_ctx, bucket_info, head_obj, 0, 0); | |
2472 | if (r < 0 && r != -ENOENT) { | |
2473 | ldout(store->ctx(), 0) << "WARNING: failed to remove obj (" << raw_head << "), leaked" << dendl; | |
2474 | } | |
2475 | } | |
2476 | } | |
2477 | ||
2478 | int RGWPutObjProcessor_Aio::handle_obj_data(rgw_raw_obj& obj, bufferlist& bl, off_t ofs, off_t abs_ofs, void **phandle, bool exclusive) | |
2479 | { | |
2480 | if ((uint64_t)abs_ofs + bl.length() > obj_len) | |
2481 | obj_len = abs_ofs + bl.length(); | |
2482 | ||
2483 | if (!(obj == last_written_obj)) { | |
2484 | last_written_obj = obj; | |
2485 | } | |
2486 | ||
2487 | // For the first call pass -1 as the offset to | |
2488 | // do a write_full. | |
2489 | return store->aio_put_obj_data(NULL, obj, bl, ((ofs != 0) ? ofs : -1), exclusive, phandle); | |
2490 | } | |
2491 | ||
2492 | struct put_obj_aio_info RGWPutObjProcessor_Aio::pop_pending() | |
2493 | { | |
2494 | struct put_obj_aio_info info; | |
2495 | info = pending.front(); | |
2496 | pending.pop_front(); | |
2497 | pending_size -= info.size; | |
2498 | return info; | |
2499 | } | |
2500 | ||
2501 | int RGWPutObjProcessor_Aio::wait_pending_front() | |
2502 | { | |
2503 | if (pending.empty()) { | |
2504 | return 0; | |
2505 | } | |
2506 | struct put_obj_aio_info info = pop_pending(); | |
2507 | int ret = store->aio_wait(info.handle); | |
2508 | ||
2509 | if (ret >= 0) { | |
2510 | add_written_obj(info.obj); | |
2511 | } | |
2512 | ||
2513 | return ret; | |
2514 | } | |
2515 | ||
2516 | bool RGWPutObjProcessor_Aio::pending_has_completed() | |
2517 | { | |
2518 | if (pending.empty()) | |
2519 | return false; | |
2520 | ||
2521 | struct put_obj_aio_info& info = pending.front(); | |
2522 | return store->aio_completed(info.handle); | |
2523 | } | |
2524 | ||
2525 | int RGWPutObjProcessor_Aio::drain_pending() | |
2526 | { | |
2527 | int ret = 0; | |
2528 | while (!pending.empty()) { | |
2529 | int r = wait_pending_front(); | |
2530 | if (r < 0) | |
2531 | ret = r; | |
2532 | } | |
2533 | return ret; | |
2534 | } | |
2535 | ||
2536 | int RGWPutObjProcessor_Aio::throttle_data(void *handle, const rgw_raw_obj& obj, uint64_t size, bool need_to_wait) | |
2537 | { | |
2538 | bool _wait = need_to_wait; | |
2539 | ||
2540 | if (handle) { | |
2541 | struct put_obj_aio_info info; | |
2542 | info.handle = handle; | |
2543 | info.obj = obj; | |
2544 | info.size = size; | |
2545 | pending_size += size; | |
2546 | pending.push_back(info); | |
2547 | } | |
2548 | size_t orig_size = pending_size; | |
2549 | ||
2550 | /* first drain complete IOs */ | |
2551 | while (pending_has_completed()) { | |
2552 | int r = wait_pending_front(); | |
2553 | if (r < 0) | |
2554 | return r; | |
2555 | ||
2556 | _wait = false; | |
2557 | } | |
2558 | ||
2559 | /* resize window in case messages are draining too fast */ | |
2560 | if (orig_size - pending_size >= window_size) { | |
2561 | window_size += store->ctx()->_conf->rgw_max_chunk_size; | |
2562 | uint64_t max_window_size = store->ctx()->_conf->rgw_put_obj_max_window_size; | |
2563 | if (window_size > max_window_size) { | |
2564 | window_size = max_window_size; | |
2565 | } | |
2566 | } | |
2567 | ||
2568 | /* now throttle. Note that need_to_wait should only affect the first IO operation */ | |
2569 | if (pending_size > window_size || _wait) { | |
2570 | int r = wait_pending_front(); | |
2571 | if (r < 0) | |
2572 | return r; | |
2573 | } | |
2574 | return 0; | |
2575 | } | |
2576 | ||
2577 | int RGWPutObjProcessor_Atomic::write_data(bufferlist& bl, off_t ofs, void **phandle, rgw_raw_obj *pobj, bool exclusive) | |
2578 | { | |
2579 | if (ofs >= next_part_ofs) { | |
2580 | int r = prepare_next_part(ofs); | |
2581 | if (r < 0) { | |
2582 | return r; | |
2583 | } | |
2584 | } | |
2585 | ||
2586 | *pobj = cur_obj; | |
2587 | ||
224ce89b WB |
2588 | if (!bl.length()) { |
2589 | *phandle = nullptr; | |
7c673cae | 2590 | return 0; |
224ce89b | 2591 | } |
7c673cae FG |
2592 | |
2593 | return RGWPutObjProcessor_Aio::handle_obj_data(cur_obj, bl, ofs - cur_part_ofs, ofs, phandle, exclusive); | |
2594 | } | |
2595 | ||
2596 | int RGWPutObjProcessor_Aio::prepare(RGWRados *store, string *oid_rand) | |
2597 | { | |
2598 | RGWPutObjProcessor::prepare(store, oid_rand); | |
2599 | ||
2600 | window_size = store->ctx()->_conf->rgw_put_obj_min_window_size; | |
2601 | ||
2602 | return 0; | |
2603 | } | |
2604 | ||
2605 | int RGWPutObjProcessor_Atomic::handle_data(bufferlist& bl, off_t ofs, void **phandle, rgw_raw_obj *pobj, bool *again) | |
2606 | { | |
2607 | *phandle = NULL; | |
2608 | uint64_t max_write_size = MIN(max_chunk_size, (uint64_t)next_part_ofs - data_ofs); | |
2609 | ||
2610 | pending_data_bl.claim_append(bl); | |
2611 | if (pending_data_bl.length() < max_write_size) { | |
2612 | *again = false; | |
2613 | return 0; | |
2614 | } | |
2615 | ||
2616 | pending_data_bl.splice(0, max_write_size, &bl); | |
2617 | ||
2618 | /* do we have enough data pending accumulated that needs to be written? */ | |
2619 | *again = (pending_data_bl.length() >= max_chunk_size); | |
2620 | ||
2621 | if (!data_ofs && !immutable_head()) { | |
2622 | first_chunk.claim(bl); | |
2623 | obj_len = (uint64_t)first_chunk.length(); | |
2624 | int r = prepare_next_part(obj_len); | |
2625 | if (r < 0) { | |
2626 | return r; | |
2627 | } | |
2628 | data_ofs = obj_len; | |
2629 | return 0; | |
2630 | } | |
2631 | off_t write_ofs = data_ofs; | |
2632 | data_ofs = write_ofs + bl.length(); | |
2633 | bool exclusive = (!write_ofs && immutable_head()); /* immutable head object, need to verify nothing exists there | |
2634 | we could be racing with another upload, to the same | |
2635 | object and cleanup can be messy */ | |
2636 | int ret = write_data(bl, write_ofs, phandle, pobj, exclusive); | |
2637 | if (ret >= 0) { /* we might return, need to clear bl as it was already sent */ | |
2638 | bl.clear(); | |
2639 | } | |
2640 | return ret; | |
2641 | } | |
2642 | ||
2643 | ||
2644 | int RGWPutObjProcessor_Atomic::prepare_init(RGWRados *store, string *oid_rand) | |
2645 | { | |
2646 | RGWPutObjProcessor_Aio::prepare(store, oid_rand); | |
2647 | ||
2648 | int r = store->get_max_chunk_size(bucket_info.placement_rule, head_obj, &max_chunk_size); | |
2649 | if (r < 0) { | |
2650 | return r; | |
2651 | } | |
2652 | ||
2653 | return 0; | |
2654 | } | |
2655 | ||
2656 | int RGWPutObjProcessor_Atomic::prepare(RGWRados *store, string *oid_rand) | |
2657 | { | |
2658 | head_obj.init(bucket, obj_str); | |
2659 | ||
2660 | int r = prepare_init(store, oid_rand); | |
2661 | if (r < 0) { | |
2662 | return r; | |
2663 | } | |
2664 | ||
2665 | if (!version_id.empty()) { | |
2666 | head_obj.key.set_instance(version_id); | |
2667 | } else if (versioned_object) { | |
2668 | store->gen_rand_obj_instance_name(&head_obj); | |
2669 | } | |
2670 | ||
2671 | manifest.set_trivial_rule(max_chunk_size, store->ctx()->_conf->rgw_obj_stripe_size); | |
2672 | ||
2673 | r = manifest_gen.create_begin(store->ctx(), &manifest, bucket_info.placement_rule, head_obj.bucket, head_obj); | |
2674 | if (r < 0) { | |
2675 | return r; | |
2676 | } | |
2677 | ||
2678 | return 0; | |
2679 | } | |
2680 | ||
2681 | int RGWPutObjProcessor_Atomic::prepare_next_part(off_t ofs) { | |
2682 | ||
2683 | int ret = manifest_gen.create_next(ofs); | |
2684 | if (ret < 0) { | |
2685 | lderr(store->ctx()) << "ERROR: manifest_gen.create_next() returned ret=" << ret << dendl; | |
2686 | return ret; | |
2687 | } | |
2688 | cur_part_ofs = ofs; | |
2689 | next_part_ofs = ofs + manifest_gen.cur_stripe_max_size(); | |
2690 | cur_obj = manifest_gen.get_cur_obj(store); | |
2691 | ||
2692 | return 0; | |
2693 | } | |
2694 | ||
2695 | int RGWPutObjProcessor_Atomic::complete_parts() | |
2696 | { | |
2697 | if (obj_len > (uint64_t)cur_part_ofs) { | |
2698 | return prepare_next_part(obj_len); | |
2699 | } | |
2700 | return 0; | |
2701 | } | |
2702 | ||
2703 | int RGWPutObjProcessor_Atomic::complete_writing_data() | |
2704 | { | |
2705 | if (!data_ofs && !immutable_head()) { | |
2706 | /* only claim if pending_data_bl() is not empty. This is needed because we might be called twice | |
2707 | * (e.g., when a retry due to race happens). So a second call to first_chunk.claim() would | |
2708 | * clobber first_chunk | |
2709 | */ | |
2710 | if (pending_data_bl.length() > 0) { | |
2711 | first_chunk.claim(pending_data_bl); | |
2712 | } | |
2713 | obj_len = (uint64_t)first_chunk.length(); | |
2714 | } | |
2715 | while (pending_data_bl.length()) { | |
224ce89b | 2716 | void *handle = nullptr; |
7c673cae FG |
2717 | rgw_raw_obj obj; |
2718 | uint64_t max_write_size = MIN(max_chunk_size, (uint64_t)next_part_ofs - data_ofs); | |
2719 | if (max_write_size > pending_data_bl.length()) { | |
2720 | max_write_size = pending_data_bl.length(); | |
2721 | } | |
2722 | bufferlist bl; | |
2723 | pending_data_bl.splice(0, max_write_size, &bl); | |
2724 | uint64_t write_len = bl.length(); | |
2725 | int r = write_data(bl, data_ofs, &handle, &obj, false); | |
2726 | if (r < 0) { | |
2727 | ldout(store->ctx(), 0) << "ERROR: write_data() returned " << r << dendl; | |
2728 | return r; | |
2729 | } | |
2730 | data_ofs += write_len; | |
2731 | r = throttle_data(handle, obj, write_len, false); | |
2732 | if (r < 0) { | |
2733 | ldout(store->ctx(), 0) << "ERROR: throttle_data() returned " << r << dendl; | |
2734 | return r; | |
2735 | } | |
2736 | ||
2737 | if (data_ofs >= next_part_ofs) { | |
2738 | r = prepare_next_part(data_ofs); | |
2739 | if (r < 0) { | |
2740 | ldout(store->ctx(), 0) << "ERROR: prepare_next_part() returned " << r << dendl; | |
2741 | return r; | |
2742 | } | |
2743 | } | |
2744 | } | |
2745 | int r = complete_parts(); | |
2746 | if (r < 0) { | |
2747 | return r; | |
2748 | } | |
2749 | ||
2750 | r = drain_pending(); | |
2751 | if (r < 0) | |
2752 | return r; | |
2753 | ||
2754 | return 0; | |
2755 | } | |
2756 | ||
2757 | int RGWPutObjProcessor_Atomic::do_complete(size_t accounted_size, const string& etag, | |
2758 | real_time *mtime, real_time set_mtime, | |
2759 | map<string, bufferlist>& attrs, | |
2760 | real_time delete_at, | |
2761 | const char *if_match, | |
31f18b77 FG |
2762 | const char *if_nomatch, const string *user_data, |
2763 | rgw_zone_set *zones_trace) { | |
7c673cae FG |
2764 | int r = complete_writing_data(); |
2765 | if (r < 0) | |
2766 | return r; | |
2767 | ||
2768 | obj_ctx.obj.set_atomic(head_obj); | |
2769 | ||
2770 | RGWRados::Object op_target(store, bucket_info, obj_ctx, head_obj); | |
2771 | ||
2772 | /* some object types shouldn't be versioned, e.g., multipart parts */ | |
2773 | op_target.set_versioning_disabled(!versioned_object); | |
2774 | ||
2775 | RGWRados::Object::Write obj_op(&op_target); | |
2776 | ||
2777 | obj_op.meta.data = &first_chunk; | |
2778 | obj_op.meta.manifest = &manifest; | |
2779 | obj_op.meta.ptag = &unique_tag; /* use req_id as operation tag */ | |
2780 | obj_op.meta.if_match = if_match; | |
2781 | obj_op.meta.if_nomatch = if_nomatch; | |
2782 | obj_op.meta.mtime = mtime; | |
2783 | obj_op.meta.set_mtime = set_mtime; | |
2784 | obj_op.meta.owner = bucket_info.owner; | |
2785 | obj_op.meta.flags = PUT_OBJ_CREATE; | |
2786 | obj_op.meta.olh_epoch = olh_epoch; | |
2787 | obj_op.meta.delete_at = delete_at; | |
2788 | obj_op.meta.user_data = user_data; | |
31f18b77 | 2789 | obj_op.meta.zones_trace = zones_trace; |
181888fb | 2790 | obj_op.meta.modify_tail = true; |
7c673cae FG |
2791 | |
2792 | r = obj_op.write_meta(obj_len, accounted_size, attrs); | |
2793 | if (r < 0) { | |
2794 | return r; | |
2795 | } | |
2796 | ||
2797 | canceled = obj_op.meta.canceled; | |
2798 | ||
2799 | return 0; | |
2800 | } | |
2801 | ||
3a9019d9 FG |
2802 | const char* RGWRados::admin_commands[4][3] = { |
2803 | { "cache list", | |
2804 | "cache list name=filter,type=CephString,req=false", | |
2805 | "cache list [filter_str]: list object cache, possibly matching substrings" }, | |
2806 | { "cache inspect", | |
2807 | "cache inspect name=target,type=CephString,req=true", | |
2808 | "cache inspect target: print cache element" }, | |
2809 | { "cache erase", | |
2810 | "cache erase name=target,type=CephString,req=true", | |
2811 | "cache erase target: erase element from cache" }, | |
2812 | { "cache zap", | |
2813 | "cache zap", | |
2814 | "cache zap: erase all elements from cache" } | |
2815 | }; | |
2816 | ||
2817 | ||
7c673cae FG |
2818 | int RGWRados::watch(const string& oid, uint64_t *watch_handle, librados::WatchCtx2 *ctx) { |
2819 | int r = control_pool_ctx.watch2(oid, watch_handle, ctx); | |
2820 | if (r < 0) | |
2821 | return r; | |
2822 | return 0; | |
2823 | } | |
2824 | ||
2825 | int RGWRados::unwatch(uint64_t watch_handle) | |
2826 | { | |
2827 | int r = control_pool_ctx.unwatch2(watch_handle); | |
2828 | if (r < 0) { | |
2829 | ldout(cct, 0) << "ERROR: rados->unwatch2() returned r=" << r << dendl; | |
2830 | return r; | |
2831 | } | |
2832 | r = rados[0].watch_flush(); | |
2833 | if (r < 0) { | |
2834 | ldout(cct, 0) << "ERROR: rados->watch_flush() returned r=" << r << dendl; | |
2835 | return r; | |
2836 | } | |
2837 | return 0; | |
2838 | } | |
2839 | ||
2840 | void RGWRados::add_watcher(int i) | |
2841 | { | |
2842 | ldout(cct, 20) << "add_watcher() i=" << i << dendl; | |
2843 | Mutex::Locker l(watchers_lock); | |
2844 | watchers_set.insert(i); | |
2845 | if (watchers_set.size() == (size_t)num_watchers) { | |
2846 | ldout(cct, 2) << "all " << num_watchers << " watchers are set, enabling cache" << dendl; | |
2847 | set_cache_enabled(true); | |
2848 | } | |
2849 | } | |
2850 | ||
2851 | void RGWRados::remove_watcher(int i) | |
2852 | { | |
2853 | ldout(cct, 20) << "remove_watcher() i=" << i << dendl; | |
2854 | Mutex::Locker l(watchers_lock); | |
2855 | size_t orig_size = watchers_set.size(); | |
2856 | watchers_set.erase(i); | |
2857 | if (orig_size == (size_t)num_watchers && | |
2858 | watchers_set.size() < orig_size) { /* actually removed */ | |
2859 | ldout(cct, 2) << "removed watcher, disabling cache" << dendl; | |
2860 | set_cache_enabled(false); | |
2861 | } | |
2862 | } | |
2863 | ||
2864 | class RGWWatcher : public librados::WatchCtx2 { | |
2865 | RGWRados *rados; | |
2866 | int index; | |
2867 | string oid; | |
2868 | uint64_t watch_handle; | |
2869 | ||
2870 | class C_ReinitWatch : public Context { | |
2871 | RGWWatcher *watcher; | |
2872 | public: | |
2873 | explicit C_ReinitWatch(RGWWatcher *_watcher) : watcher(_watcher) {} | |
2874 | void finish(int r) override { | |
2875 | watcher->reinit(); | |
2876 | } | |
2877 | }; | |
2878 | public: | |
2879 | RGWWatcher(RGWRados *r, int i, const string& o) : rados(r), index(i), oid(o), watch_handle(0) {} | |
2880 | void handle_notify(uint64_t notify_id, | |
2881 | uint64_t cookie, | |
2882 | uint64_t notifier_id, | |
2883 | bufferlist& bl) override { | |
2884 | ldout(rados->ctx(), 10) << "RGWWatcher::handle_notify() " | |
2885 | << " notify_id " << notify_id | |
2886 | << " cookie " << cookie | |
2887 | << " notifier " << notifier_id | |
2888 | << " bl.length()=" << bl.length() << dendl; | |
2889 | rados->watch_cb(notify_id, cookie, notifier_id, bl); | |
2890 | ||
2891 | bufferlist reply_bl; // empty reply payload | |
2892 | rados->control_pool_ctx.notify_ack(oid, notify_id, cookie, reply_bl); | |
2893 | } | |
2894 | void handle_error(uint64_t cookie, int err) override { | |
2895 | lderr(rados->ctx()) << "RGWWatcher::handle_error cookie " << cookie | |
2896 | << " err " << cpp_strerror(err) << dendl; | |
2897 | rados->remove_watcher(index); | |
2898 | rados->schedule_context(new C_ReinitWatch(this)); | |
2899 | } | |
2900 | ||
2901 | void reinit() { | |
2902 | int ret = unregister_watch(); | |
2903 | if (ret < 0) { | |
2904 | ldout(rados->ctx(), 0) << "ERROR: unregister_watch() returned ret=" << ret << dendl; | |
2905 | return; | |
2906 | } | |
2907 | ret = register_watch(); | |
2908 | if (ret < 0) { | |
2909 | ldout(rados->ctx(), 0) << "ERROR: register_watch() returned ret=" << ret << dendl; | |
2910 | return; | |
2911 | } | |
2912 | } | |
2913 | ||
2914 | int unregister_watch() { | |
2915 | int r = rados->unwatch(watch_handle); | |
2916 | if (r < 0) { | |
2917 | return r; | |
2918 | } | |
2919 | rados->remove_watcher(index); | |
2920 | return 0; | |
2921 | } | |
2922 | ||
2923 | int register_watch() { | |
2924 | int r = rados->watch(oid, &watch_handle, this); | |
2925 | if (r < 0) { | |
2926 | return r; | |
2927 | } | |
2928 | rados->add_watcher(index); | |
2929 | return 0; | |
2930 | } | |
2931 | }; | |
2932 | ||
2933 | class RGWMetaNotifierManager : public RGWCoroutinesManager { | |
2934 | RGWRados *store; | |
2935 | RGWHTTPManager http_manager; | |
2936 | ||
2937 | public: | |
2938 | RGWMetaNotifierManager(RGWRados *_store) : RGWCoroutinesManager(_store->ctx(), _store->get_cr_registry()), store(_store), | |
2939 | http_manager(store->ctx(), completion_mgr) { | |
2940 | http_manager.set_threaded(); | |
2941 | } | |
2942 | ||
2943 | int notify_all(map<string, RGWRESTConn *>& conn_map, set<int>& shards) { | |
2944 | rgw_http_param_pair pairs[] = { { "type", "metadata" }, | |
2945 | { "notify", NULL }, | |
2946 | { NULL, NULL } }; | |
2947 | ||
2948 | list<RGWCoroutinesStack *> stacks; | |
2949 | for (map<string, RGWRESTConn *>::iterator iter = conn_map.begin(); iter != conn_map.end(); ++iter) { | |
2950 | RGWRESTConn *conn = iter->second; | |
2951 | RGWCoroutinesStack *stack = new RGWCoroutinesStack(store->ctx(), this); | |
2952 | stack->call(new RGWPostRESTResourceCR<set<int>, int>(store->ctx(), conn, &http_manager, "/admin/log", pairs, shards, NULL)); | |
2953 | ||
2954 | stacks.push_back(stack); | |
2955 | } | |
2956 | return run(stacks); | |
2957 | } | |
2958 | }; | |
2959 | ||
2960 | class RGWDataNotifierManager : public RGWCoroutinesManager { | |
2961 | RGWRados *store; | |
2962 | RGWHTTPManager http_manager; | |
2963 | ||
2964 | public: | |
2965 | RGWDataNotifierManager(RGWRados *_store) : RGWCoroutinesManager(_store->ctx(), _store->get_cr_registry()), store(_store), | |
2966 | http_manager(store->ctx(), completion_mgr) { | |
2967 | http_manager.set_threaded(); | |
2968 | } | |
2969 | ||
2970 | int notify_all(map<string, RGWRESTConn *>& conn_map, map<int, set<string> >& shards) { | |
2971 | rgw_http_param_pair pairs[] = { { "type", "data" }, | |
2972 | { "notify", NULL }, | |
2973 | { "source-zone", store->get_zone_params().get_id().c_str() }, | |
2974 | { NULL, NULL } }; | |
2975 | ||
2976 | list<RGWCoroutinesStack *> stacks; | |
2977 | for (map<string, RGWRESTConn *>::iterator iter = conn_map.begin(); iter != conn_map.end(); ++iter) { | |
2978 | RGWRESTConn *conn = iter->second; | |
2979 | RGWCoroutinesStack *stack = new RGWCoroutinesStack(store->ctx(), this); | |
2980 | stack->call(new RGWPostRESTResourceCR<map<int, set<string> >, int>(store->ctx(), conn, &http_manager, "/admin/log", pairs, shards, NULL)); | |
2981 | ||
2982 | stacks.push_back(stack); | |
2983 | } | |
2984 | return run(stacks); | |
2985 | } | |
2986 | }; | |
2987 | ||
2988 | class RGWRadosThread { | |
2989 | class Worker : public Thread { | |
2990 | CephContext *cct; | |
2991 | RGWRadosThread *processor; | |
2992 | Mutex lock; | |
2993 | Cond cond; | |
2994 | ||
31f18b77 FG |
2995 | void wait() { |
2996 | Mutex::Locker l(lock); | |
2997 | cond.Wait(lock); | |
2998 | }; | |
2999 | ||
3000 | void wait_interval(const utime_t& wait_time) { | |
3001 | Mutex::Locker l(lock); | |
3002 | cond.WaitInterval(lock, wait_time); | |
3003 | } | |
3004 | ||
7c673cae FG |
3005 | public: |
3006 | Worker(CephContext *_cct, RGWRadosThread *_p) : cct(_cct), processor(_p), lock("RGWRadosThread::Worker") {} | |
3007 | void *entry() override; | |
31f18b77 | 3008 | void signal() { |
7c673cae FG |
3009 | Mutex::Locker l(lock); |
3010 | cond.Signal(); | |
3011 | } | |
3012 | }; | |
3013 | ||
3014 | Worker *worker; | |
3015 | ||
3016 | protected: | |
3017 | CephContext *cct; | |
3018 | RGWRados *store; | |
3019 | ||
3020 | std::atomic<bool> down_flag = { false }; | |
3021 | ||
3022 | string thread_name; | |
3023 | ||
3024 | virtual uint64_t interval_msec() = 0; | |
3025 | virtual void stop_process() {} | |
3026 | public: | |
3027 | RGWRadosThread(RGWRados *_store, const string& thread_name = "radosgw") | |
3028 | : worker(NULL), cct(_store->ctx()), store(_store), thread_name(thread_name) {} | |
3029 | virtual ~RGWRadosThread() { | |
3030 | stop(); | |
3031 | } | |
3032 | ||
3033 | virtual int init() { return 0; } | |
3034 | virtual int process() = 0; | |
3035 | ||
3036 | bool going_down() { return down_flag; } | |
3037 | ||
3038 | void start(); | |
3039 | void stop(); | |
31f18b77 FG |
3040 | |
3041 | void signal() { | |
3042 | if (worker) { | |
3043 | worker->signal(); | |
3044 | } | |
3045 | } | |
7c673cae FG |
3046 | }; |
3047 | ||
3048 | void RGWRadosThread::start() | |
3049 | { | |
3050 | worker = new Worker(cct, this); | |
3051 | worker->create(thread_name.c_str()); | |
3052 | } | |
3053 | ||
3054 | void RGWRadosThread::stop() | |
3055 | { | |
3056 | down_flag = true; | |
3057 | stop_process(); | |
3058 | if (worker) { | |
31f18b77 | 3059 | worker->signal(); |
7c673cae FG |
3060 | worker->join(); |
3061 | } | |
3062 | delete worker; | |
3063 | worker = NULL; | |
3064 | } | |
3065 | ||
3066 | void *RGWRadosThread::Worker::entry() { | |
3067 | uint64_t msec = processor->interval_msec(); | |
3068 | utime_t interval = utime_t(msec / 1000, (msec % 1000) * 1000000); | |
3069 | ||
3070 | do { | |
3071 | utime_t start = ceph_clock_now(); | |
3072 | int r = processor->process(); | |
3073 | if (r < 0) { | |
3074 | dout(0) << "ERROR: processor->process() returned error r=" << r << dendl; | |
3075 | } | |
3076 | ||
3077 | if (processor->going_down()) | |
3078 | break; | |
3079 | ||
3080 | utime_t end = ceph_clock_now(); | |
3081 | end -= start; | |
3082 | ||
3083 | uint64_t cur_msec = processor->interval_msec(); | |
3084 | if (cur_msec != msec) { /* was it reconfigured? */ | |
3085 | msec = cur_msec; | |
3086 | interval = utime_t(msec / 1000, (msec % 1000) * 1000000); | |
3087 | } | |
3088 | ||
3089 | if (cur_msec > 0) { | |
3090 | if (interval <= end) | |
3091 | continue; // next round | |
3092 | ||
3093 | utime_t wait_time = interval; | |
3094 | wait_time -= end; | |
3095 | ||
31f18b77 | 3096 | wait_interval(wait_time); |
7c673cae | 3097 | } else { |
31f18b77 | 3098 | wait(); |
7c673cae FG |
3099 | } |
3100 | } while (!processor->going_down()); | |
3101 | ||
3102 | return NULL; | |
3103 | } | |
3104 | ||
3105 | class RGWMetaNotifier : public RGWRadosThread { | |
3106 | RGWMetaNotifierManager notify_mgr; | |
3107 | RGWMetadataLog *const log; | |
3108 | ||
3109 | uint64_t interval_msec() override { | |
3110 | return cct->_conf->rgw_md_notify_interval_msec; | |
3111 | } | |
3112 | public: | |
3113 | RGWMetaNotifier(RGWRados *_store, RGWMetadataLog* log) | |
3114 | : RGWRadosThread(_store, "meta-notifier"), notify_mgr(_store), log(log) {} | |
3115 | ||
3116 | int process() override; | |
3117 | }; | |
3118 | ||
3119 | int RGWMetaNotifier::process() | |
3120 | { | |
3121 | set<int> shards; | |
3122 | ||
3123 | log->read_clear_modified(shards); | |
3124 | ||
3125 | if (shards.empty()) { | |
3126 | return 0; | |
3127 | } | |
3128 | ||
3129 | for (set<int>::iterator iter = shards.begin(); iter != shards.end(); ++iter) { | |
3130 | ldout(cct, 20) << __func__ << "(): notifying mdlog change, shard_id=" << *iter << dendl; | |
3131 | } | |
3132 | ||
3133 | notify_mgr.notify_all(store->zone_conn_map, shards); | |
3134 | ||
3135 | return 0; | |
3136 | } | |
3137 | ||
3138 | class RGWDataNotifier : public RGWRadosThread { | |
3139 | RGWDataNotifierManager notify_mgr; | |
3140 | ||
3141 | uint64_t interval_msec() override { | |
d2e6a577 | 3142 | return cct->_conf->get_val<int64_t>("rgw_data_notify_interval_msec"); |
7c673cae FG |
3143 | } |
3144 | public: | |
3145 | RGWDataNotifier(RGWRados *_store) : RGWRadosThread(_store, "data-notifier"), notify_mgr(_store) {} | |
3146 | ||
3147 | int process() override; | |
3148 | }; | |
3149 | ||
3150 | int RGWDataNotifier::process() | |
3151 | { | |
3152 | if (!store->data_log) { | |
3153 | return 0; | |
3154 | } | |
3155 | ||
3156 | map<int, set<string> > shards; | |
3157 | ||
3158 | store->data_log->read_clear_modified(shards); | |
3159 | ||
3160 | if (shards.empty()) { | |
3161 | return 0; | |
3162 | } | |
3163 | ||
3164 | for (map<int, set<string> >::iterator iter = shards.begin(); iter != shards.end(); ++iter) { | |
3165 | ldout(cct, 20) << __func__ << "(): notifying datalog change, shard_id=" << iter->first << ": " << iter->second << dendl; | |
3166 | } | |
3167 | ||
3168 | notify_mgr.notify_all(store->zone_data_notify_to_map, shards); | |
3169 | ||
3170 | return 0; | |
3171 | } | |
3172 | ||
3173 | class RGWSyncProcessorThread : public RGWRadosThread { | |
3174 | public: | |
3175 | RGWSyncProcessorThread(RGWRados *_store, const string& thread_name = "radosgw") : RGWRadosThread(_store, thread_name) {} | |
3176 | RGWSyncProcessorThread(RGWRados *_store) : RGWRadosThread(_store) {} | |
3177 | ~RGWSyncProcessorThread() override {} | |
3178 | int init() override = 0 ; | |
3179 | int process() override = 0; | |
3180 | }; | |
3181 | ||
3182 | class RGWMetaSyncProcessorThread : public RGWSyncProcessorThread | |
3183 | { | |
3184 | RGWMetaSyncStatusManager sync; | |
3185 | ||
3186 | uint64_t interval_msec() override { | |
3187 | return 0; /* no interval associated, it'll run once until stopped */ | |
3188 | } | |
3189 | void stop_process() override { | |
3190 | sync.stop(); | |
3191 | } | |
3192 | public: | |
3193 | RGWMetaSyncProcessorThread(RGWRados *_store, RGWAsyncRadosProcessor *async_rados) | |
3194 | : RGWSyncProcessorThread(_store, "meta-sync"), sync(_store, async_rados) {} | |
3195 | ||
3196 | void wakeup_sync_shards(set<int>& shard_ids) { | |
3197 | for (set<int>::iterator iter = shard_ids.begin(); iter != shard_ids.end(); ++iter) { | |
3198 | sync.wakeup(*iter); | |
3199 | } | |
3200 | } | |
3201 | RGWMetaSyncStatusManager* get_manager() { return &sync; } | |
3202 | ||
3203 | int init() override { | |
3204 | int ret = sync.init(); | |
3205 | if (ret < 0) { | |
3206 | ldout(store->ctx(), 0) << "ERROR: sync.init() returned " << ret << dendl; | |
3207 | return ret; | |
3208 | } | |
3209 | return 0; | |
3210 | } | |
3211 | ||
3212 | int process() override { | |
3213 | sync.run(); | |
3214 | return 0; | |
3215 | } | |
3216 | }; | |
3217 | ||
3218 | class RGWDataSyncProcessorThread : public RGWSyncProcessorThread | |
3219 | { | |
3220 | RGWDataSyncStatusManager sync; | |
3221 | bool initialized; | |
3222 | ||
3223 | uint64_t interval_msec() override { | |
3224 | if (initialized) { | |
3225 | return 0; /* no interval associated, it'll run once until stopped */ | |
3226 | } else { | |
3227 | #define DATA_SYNC_INIT_WAIT_SEC 20 | |
3228 | return DATA_SYNC_INIT_WAIT_SEC * 1000; | |
3229 | } | |
3230 | } | |
3231 | void stop_process() override { | |
3232 | sync.stop(); | |
3233 | } | |
3234 | public: | |
3235 | RGWDataSyncProcessorThread(RGWRados *_store, RGWAsyncRadosProcessor *async_rados, | |
b32b8144 FG |
3236 | const string& _source_zone, |
3237 | rgw::BucketChangeObserver *observer) | |
3238 | : RGWSyncProcessorThread(_store, "data-sync"), | |
3239 | sync(_store, async_rados, _source_zone, observer), | |
7c673cae FG |
3240 | initialized(false) {} |
3241 | ||
3242 | void wakeup_sync_shards(map<int, set<string> >& shard_ids) { | |
3243 | for (map<int, set<string> >::iterator iter = shard_ids.begin(); iter != shard_ids.end(); ++iter) { | |
3244 | sync.wakeup(iter->first, iter->second); | |
3245 | } | |
3246 | } | |
3247 | RGWDataSyncStatusManager* get_manager() { return &sync; } | |
3248 | ||
3249 | int init() override { | |
3250 | return 0; | |
3251 | } | |
3252 | ||
3253 | int process() override { | |
3254 | while (!initialized) { | |
3255 | if (going_down()) { | |
3256 | return 0; | |
3257 | } | |
3258 | int ret = sync.init(); | |
3259 | if (ret >= 0) { | |
3260 | initialized = true; | |
3261 | break; | |
3262 | } | |
3263 | /* we'll be back! */ | |
3264 | return 0; | |
3265 | } | |
3266 | sync.run(); | |
3267 | return 0; | |
3268 | } | |
3269 | }; | |
3270 | ||
3271 | class RGWSyncLogTrimThread : public RGWSyncProcessorThread | |
3272 | { | |
3273 | RGWCoroutinesManager crs; | |
3274 | RGWRados *store; | |
b32b8144 | 3275 | rgw::BucketTrimManager *bucket_trim; |
7c673cae FG |
3276 | RGWHTTPManager http; |
3277 | const utime_t trim_interval; | |
3278 | ||
3279 | uint64_t interval_msec() override { return 0; } | |
3280 | void stop_process() override { crs.stop(); } | |
3281 | public: | |
b32b8144 FG |
3282 | RGWSyncLogTrimThread(RGWRados *store, rgw::BucketTrimManager *bucket_trim, |
3283 | int interval) | |
7c673cae FG |
3284 | : RGWSyncProcessorThread(store, "sync-log-trim"), |
3285 | crs(store->ctx(), store->get_cr_registry()), store(store), | |
b32b8144 | 3286 | bucket_trim(bucket_trim), |
7c673cae FG |
3287 | http(store->ctx(), crs.get_completion_mgr()), |
3288 | trim_interval(interval, 0) | |
3289 | {} | |
3290 | ||
3291 | int init() override { | |
3292 | return http.set_threaded(); | |
3293 | } | |
3294 | int process() override { | |
3295 | list<RGWCoroutinesStack*> stacks; | |
3296 | auto meta = new RGWCoroutinesStack(store->ctx(), &crs); | |
3297 | meta->call(create_meta_log_trim_cr(store, &http, | |
3298 | cct->_conf->rgw_md_log_max_shards, | |
3299 | trim_interval)); | |
3300 | stacks.push_back(meta); | |
3301 | ||
3302 | auto data = new RGWCoroutinesStack(store->ctx(), &crs); | |
3303 | data->call(create_data_log_trim_cr(store, &http, | |
3304 | cct->_conf->rgw_data_log_num_shards, | |
3305 | trim_interval)); | |
3306 | stacks.push_back(data); | |
3307 | ||
b32b8144 FG |
3308 | auto bucket = new RGWCoroutinesStack(store->ctx(), &crs); |
3309 | bucket->call(bucket_trim->create_bucket_trim_cr(&http)); | |
3310 | stacks.push_back(bucket); | |
3311 | ||
7c673cae FG |
3312 | crs.run(stacks); |
3313 | return 0; | |
3314 | } | |
3315 | }; | |
3316 | ||
3317 | void RGWRados::wakeup_meta_sync_shards(set<int>& shard_ids) | |
3318 | { | |
3319 | Mutex::Locker l(meta_sync_thread_lock); | |
3320 | if (meta_sync_processor_thread) { | |
3321 | meta_sync_processor_thread->wakeup_sync_shards(shard_ids); | |
3322 | } | |
3323 | } | |
3324 | ||
3325 | void RGWRados::wakeup_data_sync_shards(const string& source_zone, map<int, set<string> >& shard_ids) | |
3326 | { | |
3327 | ldout(ctx(), 20) << __func__ << ": source_zone=" << source_zone << ", shard_ids=" << shard_ids << dendl; | |
3328 | Mutex::Locker l(data_sync_thread_lock); | |
3329 | map<string, RGWDataSyncProcessorThread *>::iterator iter = data_sync_processor_threads.find(source_zone); | |
3330 | if (iter == data_sync_processor_threads.end()) { | |
3331 | ldout(ctx(), 10) << __func__ << ": couldn't find sync thread for zone " << source_zone << ", skipping async data sync processing" << dendl; | |
3332 | return; | |
3333 | } | |
3334 | ||
3335 | RGWDataSyncProcessorThread *thread = iter->second; | |
3336 | assert(thread); | |
3337 | thread->wakeup_sync_shards(shard_ids); | |
3338 | } | |
3339 | ||
3340 | RGWMetaSyncStatusManager* RGWRados::get_meta_sync_manager() | |
3341 | { | |
3342 | Mutex::Locker l(meta_sync_thread_lock); | |
3343 | if (meta_sync_processor_thread) { | |
3344 | return meta_sync_processor_thread->get_manager(); | |
3345 | } | |
3346 | return nullptr; | |
3347 | } | |
3348 | ||
3349 | RGWDataSyncStatusManager* RGWRados::get_data_sync_manager(const std::string& source_zone) | |
3350 | { | |
3351 | Mutex::Locker l(data_sync_thread_lock); | |
3352 | auto thread = data_sync_processor_threads.find(source_zone); | |
3353 | if (thread == data_sync_processor_threads.end()) { | |
3354 | return nullptr; | |
3355 | } | |
3356 | return thread->second->get_manager(); | |
3357 | } | |
3358 | ||
3359 | int RGWRados::get_required_alignment(const rgw_pool& pool, uint64_t *alignment) | |
3360 | { | |
3361 | IoCtx ioctx; | |
3362 | int r = open_pool_ctx(pool, ioctx); | |
3363 | if (r < 0) { | |
3364 | ldout(cct, 0) << "ERROR: open_pool_ctx() returned " << r << dendl; | |
3365 | return r; | |
3366 | } | |
3367 | ||
3368 | bool requires; | |
3369 | r = ioctx.pool_requires_alignment2(&requires); | |
3370 | if (r < 0) { | |
3371 | ldout(cct, 0) << "ERROR: ioctx.pool_requires_alignment2() returned " | |
3372 | << r << dendl; | |
3373 | return r; | |
3374 | } | |
3375 | ||
3376 | if (!requires) { | |
3377 | *alignment = 0; | |
3378 | return 0; | |
3379 | } | |
3380 | ||
3381 | uint64_t align; | |
3382 | r = ioctx.pool_required_alignment2(&align); | |
3383 | if (r < 0) { | |
3384 | ldout(cct, 0) << "ERROR: ioctx.pool_required_alignment2() returned " | |
3385 | << r << dendl; | |
3386 | return r; | |
3387 | } | |
3388 | if (align != 0) { | |
3389 | ldout(cct, 20) << "required alignment=" << align << dendl; | |
3390 | } | |
3391 | *alignment = align; | |
3392 | return 0; | |
3393 | } | |
3394 | ||
3395 | int RGWRados::get_max_chunk_size(const rgw_pool& pool, uint64_t *max_chunk_size) | |
3396 | { | |
224ce89b | 3397 | uint64_t alignment = 0; |
7c673cae FG |
3398 | int r = get_required_alignment(pool, &alignment); |
3399 | if (r < 0) { | |
3400 | return r; | |
3401 | } | |
3402 | ||
3403 | uint64_t config_chunk_size = cct->_conf->rgw_max_chunk_size; | |
3404 | ||
3405 | if (alignment == 0) { | |
3406 | *max_chunk_size = config_chunk_size; | |
3407 | return 0; | |
3408 | } | |
3409 | ||
3410 | if (config_chunk_size <= alignment) { | |
3411 | *max_chunk_size = alignment; | |
3412 | return 0; | |
3413 | } | |
3414 | ||
3415 | *max_chunk_size = config_chunk_size - (config_chunk_size % alignment); | |
3416 | ||
3417 | ldout(cct, 20) << "max_chunk_size=" << *max_chunk_size << dendl; | |
3418 | ||
3419 | return 0; | |
3420 | } | |
3421 | ||
3422 | int RGWRados::get_max_chunk_size(const string& placement_rule, const rgw_obj& obj, uint64_t *max_chunk_size) | |
3423 | { | |
3424 | rgw_pool pool; | |
3425 | if (!get_obj_data_pool(placement_rule, obj, &pool)) { | |
3426 | ldout(cct, 0) << "ERROR: failed to get data pool for object " << obj << dendl; | |
3427 | return -EIO; | |
3428 | } | |
3429 | return get_max_chunk_size(pool, max_chunk_size); | |
3430 | } | |
3431 | ||
31f18b77 FG |
3432 | class RGWIndexCompletionManager; |
3433 | ||
3434 | struct complete_op_data { | |
3435 | Mutex lock{"complete_op_data"}; | |
3436 | AioCompletion *rados_completion{nullptr}; | |
3437 | int manager_shard_id{-1}; | |
3438 | RGWIndexCompletionManager *manager{nullptr}; | |
3439 | rgw_obj obj; | |
3440 | RGWModifyOp op; | |
3441 | string tag; | |
3442 | rgw_bucket_entry_ver ver; | |
3443 | cls_rgw_obj_key key; | |
3444 | rgw_bucket_dir_entry_meta dir_meta; | |
3445 | list<cls_rgw_obj_key> remove_objs; | |
3446 | bool log_op; | |
3447 | uint16_t bilog_op; | |
3448 | rgw_zone_set zones_trace; | |
3449 | ||
3450 | bool stopped{false}; | |
3451 | ||
3452 | void stop() { | |
3453 | Mutex::Locker l(lock); | |
3454 | stopped = true; | |
3455 | } | |
3456 | }; | |
3457 | ||
3458 | class RGWIndexCompletionThread : public RGWRadosThread { | |
3459 | RGWRados *store; | |
3460 | ||
3461 | uint64_t interval_msec() override { | |
3462 | return 0; | |
3463 | } | |
3464 | ||
3465 | list<complete_op_data *> completions; | |
3466 | ||
3467 | Mutex completions_lock; | |
3468 | public: | |
3469 | RGWIndexCompletionThread(RGWRados *_store) | |
3470 | : RGWRadosThread(_store, "index-complete"), store(_store), completions_lock("RGWIndexCompletionThread::completions_lock") {} | |
3471 | ||
3472 | int process() override; | |
3473 | ||
3474 | void add_completion(complete_op_data *completion) { | |
3475 | { | |
3476 | Mutex::Locker l(completions_lock); | |
3477 | completions.push_back(completion); | |
3478 | } | |
3479 | ||
3480 | signal(); | |
3481 | } | |
3482 | }; | |
3483 | ||
3484 | int RGWIndexCompletionThread::process() | |
3485 | { | |
3486 | list<complete_op_data *> comps; | |
3487 | ||
3488 | { | |
3489 | Mutex::Locker l(completions_lock); | |
3490 | completions.swap(comps); | |
3491 | } | |
3492 | ||
3493 | for (auto c : comps) { | |
3494 | std::unique_ptr<complete_op_data> up{c}; | |
3495 | ||
3496 | if (going_down()) { | |
3497 | continue; | |
3498 | } | |
3499 | ldout(store->ctx(), 20) << __func__ << "(): handling completion for key=" << c->key << dendl; | |
3500 | ||
3501 | RGWRados::BucketShard bs(store); | |
3502 | ||
3503 | int r = bs.init(c->obj.bucket, c->obj); | |
3504 | if (r < 0) { | |
3505 | ldout(cct, 0) << "ERROR: " << __func__ << "(): failed to initialize BucketShard, obj=" << c->obj << " r=" << r << dendl; | |
3506 | /* not much to do */ | |
3507 | continue; | |
3508 | } | |
3509 | ||
3510 | r = store->guard_reshard(&bs, c->obj, [&](RGWRados::BucketShard *bs) -> int { | |
3511 | librados::ObjectWriteOperation o; | |
3512 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); | |
3513 | cls_rgw_bucket_complete_op(o, c->op, c->tag, c->ver, c->key, c->dir_meta, &c->remove_objs, | |
3514 | c->log_op, c->bilog_op, &c->zones_trace); | |
3515 | ||
3516 | return bs->index_ctx.operate(bs->bucket_obj, &o); | |
3517 | }); | |
3518 | if (r < 0) { | |
3519 | ldout(cct, 0) << "ERROR: " << __func__ << "(): bucket index completion failed, obj=" << c->obj << " r=" << r << dendl; | |
3520 | /* ignoring error, can't do anything about it */ | |
3521 | continue; | |
3522 | } | |
3523 | r = store->data_log->add_entry(bs.bucket, bs.shard_id); | |
3524 | if (r < 0) { | |
3525 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
3526 | } | |
3527 | } | |
3528 | ||
3529 | return 0; | |
3530 | } | |
3531 | ||
3532 | class RGWIndexCompletionManager { | |
3533 | RGWRados *store{nullptr}; | |
3534 | vector<Mutex *> locks; | |
3535 | vector<set<complete_op_data *> > completions; | |
3536 | ||
3537 | RGWIndexCompletionThread *completion_thread{nullptr}; | |
3538 | ||
3539 | int num_shards; | |
3540 | ||
3541 | std::atomic<int> cur_shard {0}; | |
3542 | ||
3543 | ||
3544 | public: | |
3545 | RGWIndexCompletionManager(RGWRados *_store) : store(_store) { | |
3546 | num_shards = store->ctx()->_conf->rgw_thread_pool_size; | |
3547 | ||
3548 | for (int i = 0; i < num_shards; i++) { | |
3549 | char buf[64]; | |
3550 | snprintf(buf, sizeof(buf), "RGWIndexCompletionManager::lock::%d", i); | |
3551 | locks.push_back(new Mutex(buf)); | |
3552 | } | |
3553 | ||
3554 | completions.resize(num_shards); | |
3555 | } | |
3556 | ~RGWIndexCompletionManager() { | |
3557 | stop(); | |
3558 | ||
3559 | for (auto l : locks) { | |
3560 | delete l; | |
3561 | } | |
3562 | } | |
3563 | ||
3564 | int next_shard() { | |
3565 | int result = cur_shard % num_shards; | |
3566 | cur_shard++; | |
3567 | return result; | |
3568 | } | |
3569 | ||
3570 | void create_completion(const rgw_obj& obj, | |
3571 | RGWModifyOp op, string& tag, | |
3572 | rgw_bucket_entry_ver& ver, | |
3573 | const cls_rgw_obj_key& key, | |
3574 | rgw_bucket_dir_entry_meta& dir_meta, | |
3575 | list<cls_rgw_obj_key> *remove_objs, bool log_op, | |
3576 | uint16_t bilog_op, | |
3577 | rgw_zone_set *zones_trace, | |
3578 | complete_op_data **result); | |
3579 | bool handle_completion(completion_t cb, complete_op_data *arg); | |
3580 | ||
3581 | int start() { | |
3582 | completion_thread = new RGWIndexCompletionThread(store); | |
3583 | int ret = completion_thread->init(); | |
3584 | if (ret < 0) { | |
3585 | return ret; | |
3586 | } | |
3587 | completion_thread->start(); | |
3588 | return 0; | |
3589 | } | |
3590 | void stop() { | |
3591 | if (completion_thread) { | |
3592 | completion_thread->stop(); | |
3593 | delete completion_thread; | |
3594 | } | |
3595 | ||
3596 | for (int i = 0; i < num_shards; ++i) { | |
3597 | Mutex::Locker l(*locks[i]); | |
3598 | for (auto c : completions[i]) { | |
3599 | Mutex::Locker cl(c->lock); | |
3600 | c->stop(); | |
3601 | } | |
3602 | } | |
3603 | completions.clear(); | |
3604 | } | |
3605 | }; | |
3606 | ||
3607 | static void obj_complete_cb(completion_t cb, void *arg) | |
3608 | { | |
3609 | complete_op_data *completion = (complete_op_data *)arg; | |
3610 | completion->lock.Lock(); | |
3611 | if (completion->stopped) { | |
3612 | completion->lock.Unlock(); /* can drop lock, no one else is referencing us */ | |
3613 | delete completion; | |
3614 | return; | |
3615 | } | |
3616 | bool need_delete = completion->manager->handle_completion(cb, completion); | |
3617 | completion->lock.Unlock(); | |
3618 | if (need_delete) { | |
3619 | delete completion; | |
3620 | } | |
3621 | } | |
3622 | ||
3623 | ||
3624 | void RGWIndexCompletionManager::create_completion(const rgw_obj& obj, | |
3625 | RGWModifyOp op, string& tag, | |
3626 | rgw_bucket_entry_ver& ver, | |
3627 | const cls_rgw_obj_key& key, | |
3628 | rgw_bucket_dir_entry_meta& dir_meta, | |
3629 | list<cls_rgw_obj_key> *remove_objs, bool log_op, | |
3630 | uint16_t bilog_op, | |
3631 | rgw_zone_set *zones_trace, | |
3632 | complete_op_data **result) | |
3633 | { | |
3634 | complete_op_data *entry = new complete_op_data; | |
3635 | ||
3636 | int shard_id = next_shard(); | |
3637 | ||
3638 | entry->manager_shard_id = shard_id; | |
3639 | entry->manager = this; | |
3640 | entry->obj = obj; | |
3641 | entry->op = op; | |
3642 | entry->tag = tag; | |
3643 | entry->ver = ver; | |
3644 | entry->key = key; | |
3645 | entry->dir_meta = dir_meta; | |
3646 | entry->log_op = log_op; | |
3647 | entry->bilog_op = bilog_op; | |
3648 | ||
3649 | if (remove_objs) { | |
3650 | for (auto iter = remove_objs->begin(); iter != remove_objs->end(); ++iter) { | |
3651 | entry->remove_objs.push_back(*iter); | |
3652 | } | |
3653 | } | |
3654 | ||
3655 | if (zones_trace) { | |
3656 | entry->zones_trace = *zones_trace; | |
3657 | } else { | |
3658 | entry->zones_trace.insert(store->get_zone().id); | |
3659 | } | |
3660 | ||
3661 | *result = entry; | |
3662 | ||
3663 | entry->rados_completion = librados::Rados::aio_create_completion(entry, NULL, obj_complete_cb); | |
3664 | ||
3665 | Mutex::Locker l(*locks[shard_id]); | |
3666 | completions[shard_id].insert(entry); | |
3667 | } | |
3668 | ||
3669 | bool RGWIndexCompletionManager::handle_completion(completion_t cb, complete_op_data *arg) | |
3670 | { | |
3671 | int shard_id = arg->manager_shard_id; | |
3672 | { | |
3673 | Mutex::Locker l(*locks[shard_id]); | |
3674 | ||
3675 | auto& comps = completions[shard_id]; | |
3676 | ||
3677 | auto iter = comps.find(arg); | |
3678 | if (iter == comps.end()) { | |
3679 | return true; | |
3680 | } | |
3681 | ||
3682 | comps.erase(iter); | |
3683 | } | |
3684 | ||
3685 | int r = rados_aio_get_return_value(cb); | |
3686 | if (r != -ERR_BUSY_RESHARDING) { | |
3687 | return true; | |
3688 | } | |
3689 | completion_thread->add_completion(arg); | |
3690 | return false; | |
3691 | } | |
3692 | ||
7c673cae FG |
3693 | void RGWRados::finalize() |
3694 | { | |
3a9019d9 FG |
3695 | auto admin_socket = cct->get_admin_socket(); |
3696 | for (auto cmd : admin_commands) { | |
3697 | int r = admin_socket->unregister_command(cmd[0]); | |
3698 | if (r < 0) { | |
3699 | lderr(cct) << "ERROR: fail to unregister admin socket command (r=" << r | |
3700 | << ")" << dendl; | |
3701 | } | |
3702 | } | |
3703 | ||
7c673cae FG |
3704 | if (run_sync_thread) { |
3705 | Mutex::Locker l(meta_sync_thread_lock); | |
3706 | meta_sync_processor_thread->stop(); | |
3707 | ||
3708 | Mutex::Locker dl(data_sync_thread_lock); | |
3709 | for (auto iter : data_sync_processor_threads) { | |
3710 | RGWDataSyncProcessorThread *thread = iter.second; | |
3711 | thread->stop(); | |
3712 | } | |
3713 | if (sync_log_trimmer) { | |
3714 | sync_log_trimmer->stop(); | |
3715 | } | |
3716 | } | |
3717 | if (async_rados) { | |
3718 | async_rados->stop(); | |
3719 | } | |
3720 | if (run_sync_thread) { | |
3721 | delete meta_sync_processor_thread; | |
3722 | meta_sync_processor_thread = NULL; | |
3723 | Mutex::Locker dl(data_sync_thread_lock); | |
3724 | for (auto iter : data_sync_processor_threads) { | |
3725 | RGWDataSyncProcessorThread *thread = iter.second; | |
3726 | delete thread; | |
3727 | } | |
3728 | data_sync_processor_threads.clear(); | |
3729 | delete sync_log_trimmer; | |
3730 | sync_log_trimmer = nullptr; | |
b32b8144 | 3731 | bucket_trim = boost::none; |
7c673cae FG |
3732 | } |
3733 | if (finisher) { | |
3734 | finisher->stop(); | |
3735 | } | |
3736 | if (need_watch_notify()) { | |
3737 | finalize_watch(); | |
3738 | } | |
3739 | if (finisher) { | |
3740 | /* delete finisher only after cleaning up watches, as watch error path might call | |
3741 | * into finisher. We stop finisher before finalizing watch to make sure we don't | |
3742 | * actually handle any racing work | |
3743 | */ | |
3744 | delete finisher; | |
3745 | } | |
3746 | if (meta_notifier) { | |
3747 | meta_notifier->stop(); | |
3748 | delete meta_notifier; | |
3749 | } | |
3750 | if (data_notifier) { | |
3751 | data_notifier->stop(); | |
3752 | delete data_notifier; | |
3753 | } | |
3754 | delete data_log; | |
3755 | if (async_rados) { | |
3756 | delete async_rados; | |
3757 | } | |
224ce89b | 3758 | |
c07f9fc5 FG |
3759 | delete lc; |
3760 | lc = NULL; | |
3761 | ||
7c673cae FG |
3762 | delete gc; |
3763 | gc = NULL; | |
3764 | ||
7c673cae FG |
3765 | delete obj_expirer; |
3766 | obj_expirer = NULL; | |
3767 | ||
3768 | delete rest_master_conn; | |
3769 | ||
3770 | map<string, RGWRESTConn *>::iterator iter; | |
3771 | for (iter = zone_conn_map.begin(); iter != zone_conn_map.end(); ++iter) { | |
3772 | RGWRESTConn *conn = iter->second; | |
3773 | delete conn; | |
3774 | } | |
3775 | ||
3776 | for (iter = zonegroup_conn_map.begin(); iter != zonegroup_conn_map.end(); ++iter) { | |
3777 | RGWRESTConn *conn = iter->second; | |
3778 | delete conn; | |
3779 | } | |
3780 | RGWQuotaHandler::free_handler(quota_handler); | |
3781 | if (cr_registry) { | |
3782 | cr_registry->put(); | |
3783 | } | |
3784 | delete meta_mgr; | |
3785 | delete binfo_cache; | |
3786 | delete obj_tombstone_cache; | |
3787 | delete sync_modules_manager; | |
31f18b77 FG |
3788 | |
3789 | if (reshard_wait.get()) { | |
3790 | reshard_wait->stop(); | |
3791 | reshard_wait.reset(); | |
3792 | } | |
3793 | ||
3794 | if (run_reshard_thread) { | |
3795 | reshard->stop_processor(); | |
3796 | } | |
3797 | delete reshard; | |
3798 | delete index_completion_manager; | |
7c673cae FG |
3799 | } |
3800 | ||
3801 | /** | |
3802 | * Initialize the RADOS instance and prepare to do other ops | |
3803 | * Returns 0 on success, -ERR# on failure. | |
3804 | */ | |
3805 | int RGWRados::init_rados() | |
3806 | { | |
3807 | int ret = 0; | |
3a9019d9 FG |
3808 | auto admin_socket = cct->get_admin_socket(); |
3809 | for (auto cmd : admin_commands) { | |
3810 | int r = admin_socket->register_command(cmd[0], cmd[1], this, | |
3811 | cmd[2]); | |
3812 | if (r < 0) { | |
3813 | lderr(cct) << "ERROR: fail to register admin socket command (r=" << r | |
3814 | << ")" << dendl; | |
3815 | return r; | |
3816 | } | |
3817 | } | |
3818 | ||
7c673cae FG |
3819 | auto handles = std::vector<librados::Rados>{cct->_conf->rgw_num_rados_handles}; |
3820 | ||
3821 | for (auto& r : handles) { | |
3822 | ret = r.init_with_context(cct); | |
3823 | if (ret < 0) { | |
3824 | return ret; | |
3825 | } | |
7c673cae FG |
3826 | ret = r.connect(); |
3827 | if (ret < 0) { | |
3828 | return ret; | |
3829 | } | |
3830 | } | |
3831 | ||
3832 | sync_modules_manager = new RGWSyncModulesManager(); | |
3833 | ||
3834 | rgw_register_sync_modules(sync_modules_manager); | |
3835 | ||
3836 | auto crs = std::unique_ptr<RGWCoroutinesManagerRegistry>{ | |
3837 | new RGWCoroutinesManagerRegistry(cct)}; | |
3838 | ret = crs->hook_to_admin_command("cr dump"); | |
3839 | if (ret < 0) { | |
3840 | return ret; | |
3841 | } | |
3842 | ||
3843 | meta_mgr = new RGWMetadataManager(cct, this); | |
3844 | data_log = new RGWDataChangesLog(cct, this); | |
3845 | cr_registry = crs.release(); | |
3846 | ||
3847 | std::swap(handles, rados); | |
3848 | return ret; | |
3849 | } | |
3850 | ||
224ce89b WB |
3851 | |
3852 | int RGWRados::register_to_service_map(const string& daemon_type, const map<string, string>& meta) | |
3853 | { | |
3854 | map<string,string> metadata = meta; | |
3855 | metadata["num_handles"] = stringify(rados.size()); | |
3856 | metadata["zonegroup_id"] = zonegroup.get_id(); | |
3857 | metadata["zonegroup_name"] = zonegroup.get_name(); | |
3858 | metadata["zone_name"] = zone_name(); | |
3859 | metadata["zone_id"] = zone_id();; | |
3860 | string name = cct->_conf->name.get_id(); | |
3861 | if (name.find("rgw.") == 0) { | |
3862 | name = name.substr(4); | |
3863 | } | |
3864 | int ret = rados[0].service_daemon_register(daemon_type, name, metadata); | |
3865 | if (ret < 0) { | |
3866 | ldout(cct, 0) << "ERROR: service_daemon_register() returned ret=" << ret << ": " << cpp_strerror(-ret) << dendl; | |
3867 | return ret; | |
3868 | } | |
3869 | ||
3870 | return 0; | |
3871 | } | |
3872 | ||
7c673cae FG |
3873 | /** |
3874 | * Add new connection to connections map | |
3875 | * @param zonegroup_conn_map map which new connection will be added to | |
3876 | * @param zonegroup zonegroup which new connection will connect to | |
3877 | * @param new_connection pointer to new connection instance | |
3878 | */ | |
3879 | static void add_new_connection_to_map(map<string, RGWRESTConn *> &zonegroup_conn_map, | |
3880 | const RGWZoneGroup &zonegroup, RGWRESTConn *new_connection) | |
3881 | { | |
3882 | // Delete if connection is already exists | |
3883 | map<string, RGWRESTConn *>::iterator iterZoneGroup = zonegroup_conn_map.find(zonegroup.get_id()); | |
3884 | if (iterZoneGroup != zonegroup_conn_map.end()) { | |
3885 | delete iterZoneGroup->second; | |
3886 | } | |
3887 | ||
3888 | // Add new connection to connections map | |
3889 | zonegroup_conn_map[zonegroup.get_id()] = new_connection; | |
3890 | } | |
3891 | ||
3892 | int RGWRados::convert_regionmap() | |
3893 | { | |
3894 | RGWZoneGroupMap zonegroupmap; | |
3895 | ||
3896 | string pool_name = cct->_conf->rgw_zone_root_pool; | |
3897 | if (pool_name.empty()) { | |
3898 | pool_name = RGW_DEFAULT_ZONE_ROOT_POOL; | |
3899 | } | |
3900 | string oid = region_map_oid; | |
3901 | ||
3902 | rgw_pool pool(pool_name); | |
3903 | bufferlist bl; | |
3904 | RGWObjectCtx obj_ctx(this); | |
3905 | int ret = rgw_get_system_obj(this, obj_ctx, pool, oid, bl, NULL, NULL); | |
3906 | if (ret < 0 && ret != -ENOENT) { | |
3907 | return ret; | |
3908 | } else if (ret == -ENOENT) { | |
3909 | return 0; | |
3910 | } | |
3911 | ||
3912 | try { | |
3913 | bufferlist::iterator iter = bl.begin(); | |
3914 | ::decode(zonegroupmap, iter); | |
3915 | } catch (buffer::error& err) { | |
3916 | ldout(cct, 0) << "error decoding regionmap from " << pool << ":" << oid << dendl; | |
3917 | return -EIO; | |
3918 | } | |
3919 | ||
3920 | for (map<string, RGWZoneGroup>::iterator iter = zonegroupmap.zonegroups.begin(); | |
3921 | iter != zonegroupmap.zonegroups.end(); ++iter) { | |
3922 | RGWZoneGroup& zonegroup = iter->second; | |
3923 | ret = zonegroup.init(cct, this, false); | |
3924 | ret = zonegroup.update(); | |
3925 | if (ret < 0 && ret != -ENOENT) { | |
3926 | ldout(cct, 0) << "Error could not update zonegroup " << zonegroup.get_name() << ": " << | |
3927 | cpp_strerror(-ret) << dendl; | |
3928 | return ret; | |
3929 | } else if (ret == -ENOENT) { | |
3930 | ret = zonegroup.create(); | |
3931 | if (ret < 0) { | |
3932 | ldout(cct, 0) << "Error could not create " << zonegroup.get_name() << ": " << | |
3933 | cpp_strerror(-ret) << dendl; | |
3934 | return ret; | |
3935 | } | |
3936 | } | |
3937 | } | |
3938 | ||
3939 | current_period.set_user_quota(zonegroupmap.user_quota); | |
3940 | current_period.set_bucket_quota(zonegroupmap.bucket_quota); | |
3941 | ||
3942 | // remove the region_map so we don't try to convert again | |
3943 | rgw_raw_obj obj(pool, oid); | |
3944 | ret = delete_system_obj(obj); | |
3945 | if (ret < 0) { | |
3946 | ldout(cct, 0) << "Error could not remove " << obj | |
3947 | << " after upgrading to zonegroup map: " << cpp_strerror(ret) << dendl; | |
3948 | return ret; | |
3949 | } | |
3950 | ||
3951 | return 0; | |
3952 | } | |
3953 | ||
3954 | /** | |
3955 | * Replace all region configuration with zonegroup for | |
3956 | * backward compatability | |
3957 | * Returns 0 on success, -ERR# on failure. | |
3958 | */ | |
3959 | int RGWRados::replace_region_with_zonegroup() | |
3960 | { | |
3961 | /* copy default region */ | |
3962 | /* convert default region to default zonegroup */ | |
3963 | string default_oid = cct->_conf->rgw_default_region_info_oid; | |
3964 | if (default_oid.empty()) { | |
3965 | default_oid = default_region_info_oid; | |
3966 | } | |
3967 | ||
3968 | ||
3969 | RGWZoneGroup default_zonegroup; | |
3970 | rgw_pool pool{default_zonegroup.get_pool(cct)}; | |
3971 | string oid = "converted"; | |
3972 | bufferlist bl; | |
3973 | RGWObjectCtx obj_ctx(this); | |
3974 | ||
3975 | int ret = rgw_get_system_obj(this, obj_ctx, pool ,oid, bl, NULL, NULL); | |
3976 | if (ret < 0 && ret != -ENOENT) { | |
3977 | ldout(cct, 0) << __func__ << " failed to read converted: ret "<< ret << " " << cpp_strerror(-ret) | |
3978 | << dendl; | |
3979 | return ret; | |
3980 | } else if (ret != -ENOENT) { | |
3981 | ldout(cct, 20) << "System already converted " << dendl; | |
3982 | return 0; | |
3983 | } | |
3984 | ||
3985 | string default_region; | |
3986 | ret = default_zonegroup.init(cct, this, false, true); | |
3987 | if (ret < 0) { | |
3988 | ldout(cct, 0) << __func__ << " failed init default region: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
3989 | return ret; | |
3990 | } | |
3991 | ret = default_zonegroup.read_default_id(default_region, true); | |
3992 | if (ret < 0 && ret != -ENOENT) { | |
3993 | ldout(cct, 0) << __func__ << " failed reading old default region: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
3994 | return ret; | |
3995 | } | |
3996 | ||
3997 | /* convert regions to zonegroups */ | |
3998 | list<string> regions; | |
3999 | ret = list_regions(regions); | |
4000 | if (ret < 0 && ret != -ENOENT) { | |
4001 | ldout(cct, 0) << __func__ << " failed to list regions: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4002 | return ret; | |
4003 | } else if (ret == -ENOENT || regions.empty()) { | |
4004 | RGWZoneParams zoneparams(default_zone_name); | |
4005 | int ret = zoneparams.init(cct, this); | |
4006 | if (ret < 0 && ret != -ENOENT) { | |
4007 | ldout(cct, 0) << __func__ << ": error initializing default zone params: " << cpp_strerror(-ret) << dendl; | |
4008 | return ret; | |
4009 | } | |
4010 | /* update master zone */ | |
4011 | RGWZoneGroup default_zg(default_zonegroup_name); | |
4012 | ret = default_zg.init(cct, this); | |
4013 | if (ret < 0 && ret != -ENOENT) { | |
4014 | ldout(cct, 0) << __func__ << ": error in initializing default zonegroup: " << cpp_strerror(-ret) << dendl; | |
4015 | return ret; | |
4016 | } | |
4017 | if (ret != -ENOENT && default_zg.master_zone.empty()) { | |
4018 | default_zg.master_zone = zoneparams.get_id(); | |
4019 | return default_zg.update(); | |
4020 | } | |
4021 | return 0; | |
4022 | } | |
4023 | ||
4024 | string master_region, master_zone; | |
4025 | for (list<string>::iterator iter = regions.begin(); iter != regions.end(); ++iter) { | |
4026 | if (*iter != default_zonegroup_name){ | |
4027 | RGWZoneGroup region(*iter); | |
4028 | int ret = region.init(cct, this, true, true); | |
4029 | if (ret < 0) { | |
4030 | ldout(cct, 0) << __func__ << " failed init region "<< *iter << ": " << cpp_strerror(-ret) << dendl; | |
4031 | return ret; | |
4032 | } | |
31f18b77 | 4033 | if (region.is_master_zonegroup()) { |
7c673cae FG |
4034 | master_region = region.get_id(); |
4035 | master_zone = region.master_zone; | |
4036 | } | |
4037 | } | |
4038 | } | |
4039 | ||
4040 | /* create realm if there is none. | |
4041 | The realm name will be the region and zone concatenated | |
4042 | realm id will be mds of its name */ | |
4043 | if (realm.get_id().empty() && !master_region.empty() && !master_zone.empty()) { | |
4044 | string new_realm_name = master_region + "." + master_zone; | |
4045 | unsigned char md5[CEPH_CRYPTO_MD5_DIGESTSIZE]; | |
4046 | char md5_str[CEPH_CRYPTO_MD5_DIGESTSIZE * 2 + 1]; | |
4047 | MD5 hash; | |
4048 | hash.Update((const byte *)new_realm_name.c_str(), new_realm_name.length()); | |
4049 | hash.Final(md5); | |
4050 | buf_to_hex(md5, CEPH_CRYPTO_MD5_DIGESTSIZE, md5_str); | |
4051 | string new_realm_id(md5_str); | |
4052 | RGWRealm new_realm(new_realm_id,new_realm_name); | |
4053 | ret = new_realm.init(cct, this, false); | |
4054 | if (ret < 0) { | |
4055 | ldout(cct, 0) << __func__ << " Error initing new realm: " << cpp_strerror(-ret) << dendl; | |
4056 | return ret; | |
4057 | } | |
4058 | ret = new_realm.create(); | |
4059 | if (ret < 0 && ret != -EEXIST) { | |
4060 | ldout(cct, 0) << __func__ << " Error creating new realm: " << cpp_strerror(-ret) << dendl; | |
4061 | return ret; | |
4062 | } | |
4063 | ret = new_realm.set_as_default(); | |
4064 | if (ret < 0) { | |
4065 | ldout(cct, 0) << __func__ << " Error setting realm as default: " << cpp_strerror(-ret) << dendl; | |
4066 | return ret; | |
4067 | } | |
4068 | ret = realm.init(cct, this); | |
4069 | if (ret < 0) { | |
4070 | ldout(cct, 0) << __func__ << " Error initing realm: " << cpp_strerror(-ret) << dendl; | |
4071 | return ret; | |
4072 | } | |
4073 | ret = current_period.init(cct, this, realm.get_id(), realm.get_name()); | |
4074 | if (ret < 0) { | |
4075 | ldout(cct, 0) << __func__ << " Error initing current period: " << cpp_strerror(-ret) << dendl; | |
4076 | return ret; | |
4077 | } | |
4078 | } | |
4079 | ||
4080 | list<string>::iterator iter; | |
4081 | /* create zonegroups */ | |
4082 | for (iter = regions.begin(); iter != regions.end(); ++iter) | |
4083 | { | |
4084 | ldout(cct, 0) << __func__ << "Converting " << *iter << dendl; | |
4085 | /* check to see if we don't have already a zonegroup with this name */ | |
4086 | RGWZoneGroup new_zonegroup(*iter); | |
4087 | ret = new_zonegroup.init(cct , this); | |
4088 | if (ret == 0 && new_zonegroup.get_id() != *iter) { | |
4089 | ldout(cct, 0) << __func__ << " zonegroup "<< *iter << " already exists id " << new_zonegroup.get_id () << | |
4090 | " skipping conversion " << dendl; | |
4091 | continue; | |
4092 | } | |
4093 | RGWZoneGroup zonegroup(*iter); | |
4094 | zonegroup.set_id(*iter); | |
4095 | int ret = zonegroup.init(cct, this, true, true); | |
4096 | if (ret < 0) { | |
4097 | ldout(cct, 0) << __func__ << " failed init zonegroup: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4098 | return ret; | |
4099 | } | |
4100 | zonegroup.realm_id = realm.get_id(); | |
4101 | /* fix default region master zone */ | |
4102 | if (*iter == default_zonegroup_name && zonegroup.master_zone.empty()) { | |
4103 | ldout(cct, 0) << __func__ << " Setting default zone as master for default region" << dendl; | |
4104 | zonegroup.master_zone = default_zone_name; | |
4105 | } | |
4106 | ret = zonegroup.update(); | |
4107 | if (ret < 0 && ret != -EEXIST) { | |
4108 | ldout(cct, 0) << __func__ << " failed to update zonegroup " << *iter << ": ret "<< ret << " " << cpp_strerror(-ret) | |
4109 | << dendl; | |
4110 | return ret; | |
4111 | } | |
4112 | ret = zonegroup.update_name(); | |
4113 | if (ret < 0 && ret != -EEXIST) { | |
4114 | ldout(cct, 0) << __func__ << " failed to update_name for zonegroup " << *iter << ": ret "<< ret << " " << cpp_strerror(-ret) | |
4115 | << dendl; | |
4116 | return ret; | |
4117 | } | |
4118 | if (zonegroup.get_name() == default_region) { | |
4119 | ret = zonegroup.set_as_default(); | |
4120 | if (ret < 0) { | |
4121 | ldout(cct, 0) << __func__ << " failed to set_as_default " << *iter << ": ret "<< ret << " " << cpp_strerror(-ret) | |
4122 | << dendl; | |
4123 | return ret; | |
4124 | } | |
4125 | } | |
4126 | for (map<string, RGWZone>::const_iterator iter = zonegroup.zones.begin(); iter != zonegroup.zones.end(); | |
4127 | ++iter) { | |
4128 | ldout(cct, 0) << __func__ << " Converting zone" << iter->first << dendl; | |
4129 | RGWZoneParams zoneparams(iter->first, iter->first); | |
4130 | zoneparams.set_id(iter->first); | |
4131 | zoneparams.realm_id = realm.get_id(); | |
4132 | ret = zoneparams.init(cct, this); | |
4133 | if (ret < 0 && ret != -ENOENT) { | |
4134 | ldout(cct, 0) << __func__ << " failed to init zoneparams " << iter->first << ": " << cpp_strerror(-ret) << dendl; | |
4135 | return ret; | |
4136 | } else if (ret == -ENOENT) { | |
4137 | ldout(cct, 0) << __func__ << " zone is part of another cluster " << iter->first << " skipping " << dendl; | |
4138 | continue; | |
4139 | } | |
4140 | zonegroup.realm_id = realm.get_id(); | |
4141 | ret = zoneparams.update(); | |
4142 | if (ret < 0 && ret != -EEXIST) { | |
4143 | ldout(cct, 0) << __func__ << " failed to update zoneparams " << iter->first << ": " << cpp_strerror(-ret) << dendl; | |
4144 | return ret; | |
4145 | } | |
4146 | ret = zoneparams.update_name(); | |
4147 | if (ret < 0 && ret != -EEXIST) { | |
4148 | ldout(cct, 0) << __func__ << " failed to init zoneparams " << iter->first << ": " << cpp_strerror(-ret) << dendl; | |
4149 | return ret; | |
4150 | } | |
4151 | } | |
4152 | ||
4153 | if (!current_period.get_id().empty()) { | |
4154 | ret = current_period.add_zonegroup(zonegroup); | |
4155 | if (ret < 0) { | |
4156 | ldout(cct, 0) << __func__ << " failed to add zonegroup to current_period: " << cpp_strerror(-ret) << dendl; | |
4157 | return ret; | |
4158 | } | |
4159 | } | |
4160 | } | |
4161 | ||
4162 | if (!current_period.get_id().empty()) { | |
4163 | ret = current_period.update(); | |
4164 | if (ret < 0) { | |
4165 | ldout(cct, 0) << __func__ << " failed to update new period: " << cpp_strerror(-ret) << dendl; | |
4166 | return ret; | |
4167 | } | |
4168 | ret = current_period.store_info(false); | |
4169 | if (ret < 0) { | |
4170 | ldout(cct, 0) << __func__ << " failed to store new period: " << cpp_strerror(-ret) << dendl; | |
4171 | return ret; | |
4172 | } | |
4173 | ret = current_period.reflect(); | |
4174 | if (ret < 0) { | |
4175 | ldout(cct, 0) << __func__ << " failed to update local objects: " << cpp_strerror(-ret) << dendl; | |
4176 | return ret; | |
4177 | } | |
4178 | } | |
4179 | ||
4180 | for (auto const& iter : regions) { | |
4181 | RGWZoneGroup zonegroup(iter); | |
4182 | int ret = zonegroup.init(cct, this, true, true); | |
4183 | if (ret < 0) { | |
4184 | ldout(cct, 0) << __func__ << " failed init zonegroup" << iter << ": ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4185 | return ret; | |
4186 | } | |
4187 | ret = zonegroup.delete_obj(true); | |
4188 | if (ret < 0 && ret != -ENOENT) { | |
4189 | ldout(cct, 0) << __func__ << " failed to delete region " << iter << ": ret "<< ret << " " << cpp_strerror(-ret) | |
4190 | << dendl; | |
4191 | return ret; | |
4192 | } | |
4193 | } | |
4194 | ||
4195 | /* mark as converted */ | |
4196 | ret = rgw_put_system_obj(this, pool, oid, bl.c_str(), bl.length(), | |
4197 | true, NULL, real_time(), NULL); | |
4198 | if (ret < 0 ) { | |
4199 | ldout(cct, 0) << __func__ << " failed to mark cluster as converted: ret "<< ret << " " << cpp_strerror(-ret) | |
4200 | << dendl; | |
4201 | return ret; | |
4202 | } | |
4203 | ||
4204 | return 0; | |
4205 | } | |
4206 | ||
4207 | int RGWRados::init_zg_from_period(bool *initialized) | |
4208 | { | |
4209 | *initialized = false; | |
4210 | ||
4211 | if (current_period.get_id().empty()) { | |
4212 | return 0; | |
4213 | } | |
4214 | ||
4215 | int ret = zonegroup.init(cct, this); | |
4216 | ldout(cct, 20) << "period zonegroup init ret " << ret << dendl; | |
4217 | if (ret == -ENOENT) { | |
4218 | return 0; | |
4219 | } | |
4220 | if (ret < 0) { | |
4221 | ldout(cct, 0) << "failed reading zonegroup info: " << cpp_strerror(-ret) << dendl; | |
4222 | return ret; | |
4223 | } | |
4224 | ldout(cct, 20) << "period zonegroup name " << zonegroup.get_name() << dendl; | |
4225 | ||
4226 | map<string, RGWZoneGroup>::const_iterator iter = | |
4227 | current_period.get_map().zonegroups.find(zonegroup.get_id()); | |
4228 | ||
4229 | if (iter != current_period.get_map().zonegroups.end()) { | |
4230 | ldout(cct, 20) << "using current period zonegroup " << zonegroup.get_name() << dendl; | |
4231 | zonegroup = iter->second; | |
4232 | ret = zonegroup.init(cct, this, false); | |
4233 | if (ret < 0) { | |
4234 | ldout(cct, 0) << "failed init zonegroup: " << " " << cpp_strerror(-ret) << dendl; | |
4235 | return ret; | |
4236 | } | |
4237 | ret = zone_params.init(cct, this); | |
4238 | if (ret < 0 && ret != -ENOENT) { | |
4239 | ldout(cct, 0) << "failed reading zone params info: " << " " << cpp_strerror(-ret) << dendl; | |
4240 | return ret; | |
4241 | } if (ret ==-ENOENT && zonegroup.get_name() == default_zonegroup_name) { | |
4242 | ldout(cct, 10) << " Using default name "<< default_zone_name << dendl; | |
4243 | zone_params.set_name(default_zone_name); | |
4244 | ret = zone_params.init(cct, this); | |
4245 | if (ret < 0 && ret != -ENOENT) { | |
4246 | ldout(cct, 0) << "failed reading zone params info: " << " " << cpp_strerror(-ret) << dendl; | |
4247 | return ret; | |
4248 | } | |
4249 | } | |
4250 | } | |
4251 | for (iter = current_period.get_map().zonegroups.begin(); | |
4252 | iter != current_period.get_map().zonegroups.end(); ++iter){ | |
4253 | const RGWZoneGroup& zg = iter->second; | |
4254 | // use endpoints from the zonegroup's master zone | |
4255 | auto master = zg.zones.find(zg.master_zone); | |
4256 | if (master == zg.zones.end()) { | |
4257 | // fix missing master zone for a single zone zonegroup | |
4258 | if (zg.master_zone.empty() && zg.zones.size() == 1) { | |
4259 | master = zg.zones.begin(); | |
4260 | ldout(cct, 0) << "zonegroup " << zg.get_name() << " missing master_zone, setting zone " << | |
4261 | master->second.name << " id:" << master->second.id << " as master" << dendl; | |
4262 | if (zonegroup.get_id() == zg.get_id()) { | |
4263 | zonegroup.master_zone = master->second.id; | |
4264 | ret = zonegroup.update(); | |
4265 | if (ret < 0) { | |
4266 | ldout(cct, 0) << "error updating zonegroup : " << cpp_strerror(-ret) << dendl; | |
4267 | return ret; | |
4268 | } | |
4269 | } else { | |
4270 | RGWZoneGroup fixed_zg(zg.get_id(),zg.get_name()); | |
4271 | ret = fixed_zg.init(cct, this); | |
4272 | if (ret < 0) { | |
4273 | ldout(cct, 0) << "error initializing zonegroup : " << cpp_strerror(-ret) << dendl; | |
4274 | return ret; | |
4275 | } | |
4276 | fixed_zg.master_zone = master->second.id; | |
4277 | ret = fixed_zg.update(); | |
4278 | if (ret < 0) { | |
4279 | ldout(cct, 0) << "error initializing zonegroup : " << cpp_strerror(-ret) << dendl; | |
4280 | return ret; | |
4281 | } | |
4282 | } | |
4283 | } else { | |
4284 | ldout(cct, 0) << "zonegroup " << zg.get_name() << " missing zone for master_zone=" << | |
4285 | zg.master_zone << dendl; | |
4286 | return -EINVAL; | |
4287 | } | |
4288 | } | |
4289 | const auto& endpoints = master->second.endpoints; | |
4290 | add_new_connection_to_map(zonegroup_conn_map, zg, new RGWRESTConn(cct, this, zg.get_id(), endpoints)); | |
4291 | if (!current_period.get_master_zonegroup().empty() && | |
4292 | zg.get_id() == current_period.get_master_zonegroup()) { | |
4293 | rest_master_conn = new RGWRESTConn(cct, this, zg.get_id(), endpoints); | |
4294 | } | |
4295 | } | |
4296 | ||
4297 | *initialized = true; | |
4298 | ||
4299 | return 0; | |
4300 | } | |
4301 | ||
4302 | int RGWRados::init_zg_from_local(bool *creating_defaults) | |
4303 | { | |
4304 | int ret = zonegroup.init(cct, this); | |
4305 | if ( (ret < 0 && ret != -ENOENT) || (ret == -ENOENT && !cct->_conf->rgw_zonegroup.empty())) { | |
4306 | ldout(cct, 0) << "failed reading zonegroup info: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4307 | return ret; | |
4308 | } else if (ret == -ENOENT) { | |
4309 | *creating_defaults = true; | |
4310 | ldout(cct, 10) << "Creating default zonegroup " << dendl; | |
4311 | ret = zonegroup.create_default(); | |
4312 | if (ret < 0) { | |
4313 | ldout(cct, 0) << "failure in zonegroup create_default: ret "<< ret << " " << cpp_strerror(-ret) | |
4314 | << dendl; | |
4315 | return ret; | |
4316 | } | |
4317 | ret = zonegroup.init(cct, this); | |
4318 | if (ret < 0) { | |
4319 | ldout(cct, 0) << "failure in zonegroup create_default: ret "<< ret << " " << cpp_strerror(-ret) | |
4320 | << dendl; | |
4321 | return ret; | |
4322 | } | |
4323 | } | |
4324 | ldout(cct, 20) << "zonegroup " << zonegroup.get_name() << dendl; | |
31f18b77 | 4325 | if (zonegroup.is_master_zonegroup()) { |
7c673cae FG |
4326 | // use endpoints from the zonegroup's master zone |
4327 | auto master = zonegroup.zones.find(zonegroup.master_zone); | |
4328 | if (master == zonegroup.zones.end()) { | |
4329 | // fix missing master zone for a single zone zonegroup | |
4330 | if (zonegroup.master_zone.empty() && zonegroup.zones.size() == 1) { | |
4331 | master = zonegroup.zones.begin(); | |
4332 | ldout(cct, 0) << "zonegroup " << zonegroup.get_name() << " missing master_zone, setting zone " << | |
4333 | master->second.name << " id:" << master->second.id << " as master" << dendl; | |
4334 | zonegroup.master_zone = master->second.id; | |
4335 | ret = zonegroup.update(); | |
4336 | if (ret < 0) { | |
4337 | ldout(cct, 0) << "error initializing zonegroup : " << cpp_strerror(-ret) << dendl; | |
4338 | return ret; | |
4339 | } | |
4340 | } else { | |
4341 | ldout(cct, 0) << "zonegroup " << zonegroup.get_name() << " missing zone for " | |
4342 | "master_zone=" << zonegroup.master_zone << dendl; | |
4343 | return -EINVAL; | |
4344 | } | |
4345 | } | |
4346 | const auto& endpoints = master->second.endpoints; | |
4347 | rest_master_conn = new RGWRESTConn(cct, this, zonegroup.get_id(), endpoints); | |
4348 | } | |
4349 | ||
4350 | return 0; | |
4351 | } | |
4352 | ||
4353 | ||
4354 | bool RGWRados::zone_syncs_from(RGWZone& target_zone, RGWZone& source_zone) | |
4355 | { | |
4356 | return target_zone.syncs_from(source_zone.name) && | |
4357 | sync_modules_manager->supports_data_export(source_zone.tier_type); | |
4358 | } | |
4359 | ||
4360 | /** | |
4361 | * Initialize the RADOS instance and prepare to do other ops | |
4362 | * Returns 0 on success, -ERR# on failure. | |
4363 | */ | |
4364 | int RGWRados::init_complete() | |
4365 | { | |
4366 | int ret = realm.init(cct, this); | |
4367 | if (ret < 0 && ret != -ENOENT) { | |
4368 | ldout(cct, 0) << "failed reading realm info: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4369 | return ret; | |
4370 | } else if (ret != -ENOENT) { | |
4371 | ldout(cct, 20) << "realm " << realm.get_name() << " " << realm.get_id() << dendl; | |
4372 | ret = current_period.init(cct, this, realm.get_id(), realm.get_name()); | |
4373 | if (ret < 0 && ret != -ENOENT) { | |
4374 | ldout(cct, 0) << "failed reading current period info: " << " " << cpp_strerror(-ret) << dendl; | |
4375 | return ret; | |
4376 | } | |
4377 | ldout(cct, 20) << "current period " << current_period.get_id() << dendl; | |
4378 | } | |
4379 | ||
4380 | ret = replace_region_with_zonegroup(); | |
4381 | if (ret < 0) { | |
4382 | lderr(cct) << "failed converting region to zonegroup : ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4383 | return ret; | |
4384 | } | |
4385 | ||
4386 | ret = convert_regionmap(); | |
4387 | if (ret < 0) { | |
4388 | lderr(cct) << "failed converting regionmap: " << cpp_strerror(-ret) << dendl; | |
4389 | return ret; | |
4390 | } | |
4391 | ||
4392 | bool zg_initialized = false; | |
4393 | ||
4394 | if (!current_period.get_id().empty()) { | |
4395 | ret = init_zg_from_period(&zg_initialized); | |
4396 | if (ret < 0) { | |
4397 | return ret; | |
4398 | } | |
4399 | } | |
4400 | ||
4401 | bool creating_defaults = false; | |
4402 | bool using_local = (!zg_initialized); | |
4403 | if (using_local) { | |
4404 | ldout(cct, 10) << " cannot find current period zonegroup using local zonegroup" << dendl; | |
4405 | ret = init_zg_from_local(&creating_defaults); | |
4406 | if (ret < 0) { | |
4407 | return ret; | |
4408 | } | |
4409 | // read period_config into current_period | |
4410 | auto& period_config = current_period.get_config(); | |
4411 | ret = period_config.read(this, zonegroup.realm_id); | |
4412 | if (ret < 0 && ret != -ENOENT) { | |
4413 | ldout(cct, 0) << "ERROR: failed to read period config: " | |
4414 | << cpp_strerror(ret) << dendl; | |
4415 | return ret; | |
4416 | } | |
4417 | } | |
4418 | ||
4419 | ldout(cct, 10) << "Cannot find current period zone using local zone" << dendl; | |
4420 | if (creating_defaults && cct->_conf->rgw_zone.empty()) { | |
4421 | ldout(cct, 10) << " Using default name "<< default_zone_name << dendl; | |
4422 | zone_params.set_name(default_zone_name); | |
4423 | } | |
4424 | ||
4425 | ret = zone_params.init(cct, this); | |
4426 | if (ret < 0 && ret != -ENOENT) { | |
4427 | lderr(cct) << "failed reading zone info: ret "<< ret << " " << cpp_strerror(-ret) << dendl; | |
4428 | return ret; | |
4429 | } | |
4430 | map<string, RGWZone>::iterator zone_iter = get_zonegroup().zones.find(zone_params.get_id()); | |
4431 | if (zone_iter == get_zonegroup().zones.end()) { | |
4432 | if (using_local) { | |
4433 | lderr(cct) << "Cannot find zone id=" << zone_params.get_id() << " (name=" << zone_params.get_name() << ")" << dendl; | |
4434 | return -EINVAL; | |
4435 | } | |
4436 | ldout(cct, 1) << "Cannot find zone id=" << zone_params.get_id() << " (name=" << zone_params.get_name() << "), switching to local zonegroup configuration" << dendl; | |
4437 | ret = init_zg_from_local(&creating_defaults); | |
4438 | if (ret < 0) { | |
4439 | return ret; | |
4440 | } | |
4441 | zone_iter = get_zonegroup().zones.find(zone_params.get_id()); | |
4442 | } | |
4443 | if (zone_iter != get_zonegroup().zones.end()) { | |
4444 | zone_public_config = zone_iter->second; | |
4445 | ldout(cct, 20) << "zone " << zone_params.get_name() << dendl; | |
4446 | } else { | |
4447 | lderr(cct) << "Cannot find zone id=" << zone_params.get_id() << " (name=" << zone_params.get_name() << ")" << dendl; | |
4448 | return -EINVAL; | |
4449 | } | |
4450 | ||
4451 | zone_short_id = current_period.get_map().get_zone_short_id(zone_params.get_id()); | |
4452 | ||
31f18b77 FG |
4453 | if (run_sync_thread) { |
4454 | ret = sync_modules_manager->create_instance(cct, zone_public_config.tier_type, zone_params.tier_config, &sync_module); | |
4455 | if (ret < 0) { | |
4456 | lderr(cct) << "ERROR: failed to init sync module instance, ret=" << ret << dendl; | |
4457 | return ret; | |
4458 | } | |
7c673cae FG |
4459 | } |
4460 | ||
4461 | writeable_zone = (zone_public_config.tier_type.empty() || zone_public_config.tier_type == "rgw"); | |
4462 | ||
4463 | init_unique_trans_id_deps(); | |
4464 | ||
4465 | finisher = new Finisher(cct); | |
4466 | finisher->start(); | |
4467 | ||
4468 | period_puller.reset(new RGWPeriodPuller(this)); | |
4469 | period_history.reset(new RGWPeriodHistory(cct, period_puller.get(), | |
4470 | current_period)); | |
4471 | ||
4472 | if (need_watch_notify()) { | |
4473 | ret = init_watch(); | |
4474 | if (ret < 0) { | |
4475 | lderr(cct) << "ERROR: failed to initialize watch: " << cpp_strerror(-ret) << dendl; | |
4476 | return ret; | |
4477 | } | |
4478 | } | |
4479 | ||
4480 | /* first build all zones index */ | |
4481 | for (auto ziter : get_zonegroup().zones) { | |
4482 | const string& id = ziter.first; | |
4483 | RGWZone& z = ziter.second; | |
4484 | zone_id_by_name[z.name] = id; | |
4485 | zone_by_id[id] = z; | |
4486 | } | |
31f18b77 | 4487 | |
7c673cae FG |
4488 | if (zone_by_id.find(zone_id()) == zone_by_id.end()) { |
4489 | ldout(cct, 0) << "WARNING: could not find zone config in zonegroup for local zone (" << zone_id() << "), will use defaults" << dendl; | |
4490 | } | |
4491 | zone_public_config = zone_by_id[zone_id()]; | |
4492 | for (auto ziter : get_zonegroup().zones) { | |
4493 | const string& id = ziter.first; | |
4494 | RGWZone& z = ziter.second; | |
4495 | if (id == zone_id()) { | |
4496 | continue; | |
4497 | } | |
4498 | if (z.endpoints.empty()) { | |
4499 | ldout(cct, 0) << "WARNING: can't generate connection for zone " << z.id << " id " << z.name << ": no endpoints defined" << dendl; | |
4500 | continue; | |
4501 | } | |
4502 | ldout(cct, 20) << "generating connection object for zone " << z.name << " id " << z.id << dendl; | |
4503 | RGWRESTConn *conn = new RGWRESTConn(cct, this, z.id, z.endpoints); | |
4504 | zone_conn_map[id] = conn; | |
4505 | if (zone_syncs_from(zone_public_config, z) || | |
4506 | zone_syncs_from(z, zone_public_config)) { | |
4507 | if (zone_syncs_from(zone_public_config, z)) { | |
4508 | zone_data_sync_from_map[id] = conn; | |
4509 | } | |
4510 | if (zone_syncs_from(z, zone_public_config)) { | |
4511 | zone_data_notify_to_map[id] = conn; | |
4512 | } | |
4513 | } else { | |
4514 | ldout(cct, 20) << "NOTICE: not syncing to/from zone " << z.name << " id " << z.id << dendl; | |
4515 | } | |
4516 | } | |
4517 | ||
4518 | ret = open_root_pool_ctx(); | |
4519 | if (ret < 0) | |
4520 | return ret; | |
4521 | ||
4522 | ret = open_gc_pool_ctx(); | |
4523 | if (ret < 0) | |
4524 | return ret; | |
4525 | ||
4526 | ret = open_lc_pool_ctx(); | |
4527 | if (ret < 0) | |
4528 | return ret; | |
4529 | ||
4530 | ret = open_objexp_pool_ctx(); | |
4531 | if (ret < 0) | |
4532 | return ret; | |
4533 | ||
31f18b77 FG |
4534 | ret = open_reshard_pool_ctx(); |
4535 | if (ret < 0) | |
4536 | return ret; | |
4537 | ||
7c673cae FG |
4538 | pools_initialized = true; |
4539 | ||
4540 | gc = new RGWGC(); | |
4541 | gc->initialize(cct, this); | |
4542 | ||
4543 | obj_expirer = new RGWObjectExpirer(this); | |
4544 | ||
4545 | if (use_gc_thread) { | |
4546 | gc->start_processor(); | |
4547 | obj_expirer->start_processor(); | |
4548 | } | |
4549 | ||
7c673cae FG |
4550 | /* no point of running sync thread if we don't have a master zone configured |
4551 | or there is no rest_master_conn */ | |
4552 | if (get_zonegroup().master_zone.empty() || !rest_master_conn | |
4553 | || current_period.get_id().empty()) { | |
4554 | run_sync_thread = false; | |
4555 | } | |
4556 | ||
b32b8144 FG |
4557 | if (run_sync_thread) { |
4558 | // initialize the log period history | |
4559 | meta_mgr->init_oldest_log_period(); | |
4560 | } | |
4561 | ||
7c673cae FG |
4562 | async_rados = new RGWAsyncRadosProcessor(this, cct->_conf->rgw_num_async_rados_threads); |
4563 | async_rados->start(); | |
4564 | ||
4565 | ret = meta_mgr->init(current_period.get_id()); | |
4566 | if (ret < 0) { | |
4567 | lderr(cct) << "ERROR: failed to initialize metadata log: " | |
4568 | << cpp_strerror(-ret) << dendl; | |
4569 | return ret; | |
4570 | } | |
4571 | ||
4572 | if (is_meta_master()) { | |
4573 | auto md_log = meta_mgr->get_log(current_period.get_id()); | |
4574 | meta_notifier = new RGWMetaNotifier(this, md_log); | |
4575 | meta_notifier->start(); | |
4576 | } | |
4577 | ||
4578 | if (run_sync_thread) { | |
4579 | Mutex::Locker l(meta_sync_thread_lock); | |
4580 | meta_sync_processor_thread = new RGWMetaSyncProcessorThread(this, async_rados); | |
4581 | ret = meta_sync_processor_thread->init(); | |
4582 | if (ret < 0) { | |
4583 | ldout(cct, 0) << "ERROR: failed to initialize meta sync thread" << dendl; | |
4584 | return ret; | |
4585 | } | |
4586 | meta_sync_processor_thread->start(); | |
4587 | ||
b32b8144 FG |
4588 | // configure the bucket trim manager |
4589 | rgw::BucketTrimConfig config; | |
4590 | rgw::configure_bucket_trim(cct, config); | |
4591 | ||
4592 | bucket_trim.emplace(this, config); | |
4593 | ret = bucket_trim->init(); | |
4594 | if (ret < 0) { | |
4595 | ldout(cct, 0) << "ERROR: failed to start bucket trim manager" << dendl; | |
4596 | return ret; | |
4597 | } | |
4598 | ||
7c673cae FG |
4599 | Mutex::Locker dl(data_sync_thread_lock); |
4600 | for (auto iter : zone_data_sync_from_map) { | |
4601 | ldout(cct, 5) << "starting data sync thread for zone " << iter.first << dendl; | |
b32b8144 FG |
4602 | auto *thread = new RGWDataSyncProcessorThread(this, async_rados, iter.first, |
4603 | &*bucket_trim); | |
7c673cae FG |
4604 | ret = thread->init(); |
4605 | if (ret < 0) { | |
4606 | ldout(cct, 0) << "ERROR: failed to initialize data sync thread" << dendl; | |
4607 | return ret; | |
4608 | } | |
4609 | thread->start(); | |
4610 | data_sync_processor_threads[iter.first] = thread; | |
4611 | } | |
4612 | auto interval = cct->_conf->rgw_sync_log_trim_interval; | |
4613 | if (interval > 0) { | |
b32b8144 | 4614 | sync_log_trimmer = new RGWSyncLogTrimThread(this, &*bucket_trim, interval); |
7c673cae FG |
4615 | ret = sync_log_trimmer->init(); |
4616 | if (ret < 0) { | |
4617 | ldout(cct, 0) << "ERROR: failed to initialize sync log trim thread" << dendl; | |
4618 | return ret; | |
4619 | } | |
4620 | sync_log_trimmer->start(); | |
4621 | } | |
4622 | } | |
4623 | data_notifier = new RGWDataNotifier(this); | |
4624 | data_notifier->start(); | |
4625 | ||
4626 | lc = new RGWLC(); | |
4627 | lc->initialize(cct, this); | |
31f18b77 | 4628 | |
7c673cae FG |
4629 | if (use_lc_thread) |
4630 | lc->start_processor(); | |
31f18b77 | 4631 | |
7c673cae FG |
4632 | quota_handler = RGWQuotaHandler::generate_handler(this, quota_threads); |
4633 | ||
4634 | bucket_index_max_shards = (cct->_conf->rgw_override_bucket_index_max_shards ? cct->_conf->rgw_override_bucket_index_max_shards : | |
4635 | get_zone().bucket_index_max_shards); | |
31f18b77 FG |
4636 | if (bucket_index_max_shards > get_max_bucket_shards()) { |
4637 | bucket_index_max_shards = get_max_bucket_shards(); | |
7c673cae | 4638 | ldout(cct, 1) << __func__ << " bucket index max shards is too large, reset to value: " |
31f18b77 | 4639 | << get_max_bucket_shards() << dendl; |
7c673cae FG |
4640 | } |
4641 | ldout(cct, 20) << __func__ << " bucket index max shards: " << bucket_index_max_shards << dendl; | |
4642 | ||
4643 | binfo_cache = new RGWChainedCacheImpl<bucket_info_entry>; | |
4644 | binfo_cache->init(this); | |
4645 | ||
4646 | bool need_tombstone_cache = !zone_data_notify_to_map.empty(); /* have zones syncing from us */ | |
4647 | ||
4648 | if (need_tombstone_cache) { | |
4649 | obj_tombstone_cache = new tombstone_cache_t(cct->_conf->rgw_obj_tombstone_cache_size); | |
4650 | } | |
4651 | ||
31f18b77 FG |
4652 | reshard_wait = std::make_shared<RGWReshardWait>(this); |
4653 | ||
4654 | reshard = new RGWReshard(this); | |
4655 | ||
4656 | /* only the master zone in the zonegroup reshards buckets */ | |
4657 | run_reshard_thread = run_reshard_thread && (get_zonegroup().master_zone == zone_public_config.id); | |
4658 | if (run_reshard_thread) { | |
4659 | reshard->start_processor(); | |
4660 | } | |
4661 | ||
4662 | index_completion_manager = new RGWIndexCompletionManager(this); | |
4663 | ret = index_completion_manager->start(); | |
4664 | ||
7c673cae FG |
4665 | return ret; |
4666 | } | |
4667 | ||
4668 | /** | |
4669 | * Initialize the RADOS instance and prepare to do other ops | |
4670 | * Returns 0 on success, -ERR# on failure. | |
4671 | */ | |
4672 | int RGWRados::initialize() | |
4673 | { | |
4674 | int ret; | |
4675 | ||
4676 | ret = init_rados(); | |
4677 | if (ret < 0) | |
4678 | return ret; | |
4679 | ||
4680 | return init_complete(); | |
4681 | } | |
4682 | ||
4683 | void RGWRados::finalize_watch() | |
4684 | { | |
4685 | for (int i = 0; i < num_watchers; i++) { | |
4686 | RGWWatcher *watcher = watchers[i]; | |
4687 | watcher->unregister_watch(); | |
4688 | delete watcher; | |
4689 | } | |
4690 | ||
4691 | delete[] notify_oids; | |
4692 | delete[] watchers; | |
4693 | } | |
4694 | ||
4695 | void RGWRados::schedule_context(Context *c) { | |
4696 | finisher->queue(c); | |
4697 | } | |
4698 | ||
4699 | int RGWRados::list_raw_prefixed_objs(const rgw_pool& pool, const string& prefix, list<string>& result) | |
4700 | { | |
4701 | bool is_truncated; | |
4702 | RGWListRawObjsCtx ctx; | |
4703 | do { | |
4704 | list<string> oids; | |
4705 | int r = list_raw_objects(pool, prefix, 1000, | |
4706 | ctx, oids, &is_truncated); | |
4707 | if (r < 0) { | |
4708 | return r; | |
4709 | } | |
4710 | list<string>::iterator iter; | |
4711 | for (iter = oids.begin(); iter != oids.end(); ++iter) { | |
4712 | string& val = *iter; | |
4713 | if (val.size() > prefix.size()) | |
4714 | result.push_back(val.substr(prefix.size())); | |
4715 | } | |
4716 | } while (is_truncated); | |
4717 | ||
4718 | return 0; | |
4719 | } | |
4720 | ||
4721 | int RGWRados::list_regions(list<string>& regions) | |
4722 | { | |
4723 | RGWZoneGroup zonegroup; | |
4724 | ||
4725 | return list_raw_prefixed_objs(zonegroup.get_pool(cct), region_info_oid_prefix, regions); | |
4726 | } | |
4727 | ||
4728 | int RGWRados::list_zonegroups(list<string>& zonegroups) | |
4729 | { | |
4730 | RGWZoneGroup zonegroup; | |
4731 | ||
4732 | return list_raw_prefixed_objs(zonegroup.get_pool(cct), zonegroup_names_oid_prefix, zonegroups); | |
4733 | } | |
4734 | ||
4735 | int RGWRados::list_zones(list<string>& zones) | |
4736 | { | |
4737 | RGWZoneParams zoneparams; | |
4738 | ||
4739 | return list_raw_prefixed_objs(zoneparams.get_pool(cct), zone_names_oid_prefix, zones); | |
4740 | } | |
4741 | ||
4742 | int RGWRados::list_realms(list<string>& realms) | |
4743 | { | |
4744 | RGWRealm realm(cct, this); | |
4745 | return list_raw_prefixed_objs(realm.get_pool(cct), realm_names_oid_prefix, realms); | |
4746 | } | |
4747 | ||
4748 | int RGWRados::list_periods(list<string>& periods) | |
4749 | { | |
4750 | RGWPeriod period; | |
4751 | list<string> raw_periods; | |
4752 | int ret = list_raw_prefixed_objs(period.get_pool(cct), period.get_info_oid_prefix(), raw_periods); | |
4753 | if (ret < 0) { | |
4754 | return ret; | |
4755 | } | |
4756 | for (const auto& oid : raw_periods) { | |
4757 | size_t pos = oid.find("."); | |
4758 | if (pos != std::string::npos) { | |
4759 | periods.push_back(oid.substr(0, pos)); | |
4760 | } else { | |
4761 | periods.push_back(oid); | |
4762 | } | |
4763 | } | |
4764 | periods.sort(); // unique() only detects duplicates if they're adjacent | |
4765 | periods.unique(); | |
4766 | return 0; | |
4767 | } | |
4768 | ||
4769 | ||
4770 | int RGWRados::list_periods(const string& current_period, list<string>& periods) | |
4771 | { | |
4772 | int ret = 0; | |
4773 | string period_id = current_period; | |
4774 | while(!period_id.empty()) { | |
4775 | RGWPeriod period(period_id); | |
4776 | ret = period.init(cct, this); | |
4777 | if (ret < 0) { | |
4778 | return ret; | |
4779 | } | |
4780 | periods.push_back(period.get_id()); | |
4781 | period_id = period.get_predecessor(); | |
4782 | } | |
4783 | ||
4784 | return ret; | |
4785 | } | |
4786 | ||
4787 | /** | |
4788 | * Open the pool used as root for this gateway | |
4789 | * Returns: 0 on success, -ERR# otherwise. | |
4790 | */ | |
4791 | int RGWRados::open_root_pool_ctx() | |
4792 | { | |
4793 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().domain_root, root_pool_ctx, true); | |
4794 | } | |
4795 | ||
4796 | int RGWRados::open_gc_pool_ctx() | |
4797 | { | |
4798 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().gc_pool, gc_pool_ctx, true); | |
4799 | } | |
4800 | ||
4801 | int RGWRados::open_lc_pool_ctx() | |
4802 | { | |
4803 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().lc_pool, lc_pool_ctx, true); | |
4804 | } | |
4805 | ||
4806 | int RGWRados::open_objexp_pool_ctx() | |
4807 | { | |
4808 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, objexp_pool_ctx, true); | |
4809 | } | |
4810 | ||
31f18b77 FG |
4811 | int RGWRados::open_reshard_pool_ctx() |
4812 | { | |
4813 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().reshard_pool, reshard_pool_ctx, true); | |
4814 | } | |
4815 | ||
7c673cae FG |
4816 | int RGWRados::init_watch() |
4817 | { | |
4818 | int r = rgw_init_ioctx(&rados[0], get_zone_params().control_pool, control_pool_ctx, true); | |
4819 | if (r < 0) { | |
4820 | return r; | |
4821 | } | |
4822 | ||
4823 | num_watchers = cct->_conf->rgw_num_control_oids; | |
4824 | ||
4825 | bool compat_oid = (num_watchers == 0); | |
4826 | ||
4827 | if (num_watchers <= 0) | |
4828 | num_watchers = 1; | |
4829 | ||
4830 | notify_oids = new string[num_watchers]; | |
4831 | watchers = new RGWWatcher *[num_watchers]; | |
4832 | ||
4833 | for (int i=0; i < num_watchers; i++) { | |
4834 | string& notify_oid = notify_oids[i]; | |
4835 | notify_oid = notify_oid_prefix; | |
4836 | if (!compat_oid) { | |
4837 | char buf[16]; | |
4838 | snprintf(buf, sizeof(buf), ".%d", i); | |
4839 | notify_oid.append(buf); | |
4840 | } | |
4841 | r = control_pool_ctx.create(notify_oid, false); | |
4842 | if (r < 0 && r != -EEXIST) | |
4843 | return r; | |
4844 | ||
4845 | RGWWatcher *watcher = new RGWWatcher(this, i, notify_oid); | |
4846 | watchers[i] = watcher; | |
4847 | ||
4848 | r = watcher->register_watch(); | |
4849 | if (r < 0) | |
4850 | return r; | |
4851 | } | |
4852 | ||
4853 | watch_initialized = true; | |
4854 | ||
4855 | set_cache_enabled(true); | |
4856 | ||
4857 | return 0; | |
4858 | } | |
4859 | ||
4860 | void RGWRados::pick_control_oid(const string& key, string& notify_oid) | |
4861 | { | |
4862 | uint32_t r = ceph_str_hash_linux(key.c_str(), key.size()); | |
4863 | ||
4864 | int i = r % num_watchers; | |
4865 | char buf[16]; | |
4866 | snprintf(buf, sizeof(buf), ".%d", i); | |
4867 | ||
4868 | notify_oid = notify_oid_prefix; | |
4869 | notify_oid.append(buf); | |
4870 | } | |
4871 | ||
28e407b8 | 4872 | int RGWRados::open_pool_ctx(const rgw_pool& pool, librados::IoCtx& io_ctx) |
7c673cae | 4873 | { |
28e407b8 AA |
4874 | constexpr bool create = true; // create the pool if it doesn't exist |
4875 | return rgw_init_ioctx(get_rados_handle(), pool, io_ctx, create); | |
7c673cae FG |
4876 | } |
4877 | ||
4878 | void RGWRados::build_bucket_index_marker(const string& shard_id_str, const string& shard_marker, | |
4879 | string *marker) { | |
4880 | if (marker) { | |
4881 | *marker = shard_id_str; | |
4882 | marker->append(BucketIndexShardsManager::KEY_VALUE_SEPARATOR); | |
4883 | marker->append(shard_marker); | |
4884 | } | |
4885 | } | |
4886 | ||
4887 | int RGWRados::open_bucket_index_ctx(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx) | |
4888 | { | |
3a9019d9 FG |
4889 | const rgw_pool& explicit_pool = bucket_info.bucket.explicit_placement.index_pool; |
4890 | ||
4891 | if (!explicit_pool.empty()) { | |
4892 | return open_pool_ctx(explicit_pool, index_ctx); | |
4893 | } | |
4894 | ||
7c673cae FG |
4895 | const string *rule = &bucket_info.placement_rule; |
4896 | if (rule->empty()) { | |
4897 | rule = &zonegroup.default_placement; | |
4898 | } | |
4899 | auto iter = zone_params.placement_pools.find(*rule); | |
4900 | if (iter == zone_params.placement_pools.end()) { | |
4901 | ldout(cct, 0) << "could not find placement rule " << *rule << " within zonegroup " << dendl; | |
4902 | return -EINVAL; | |
4903 | } | |
4904 | ||
4905 | int r = open_pool_ctx(iter->second.index_pool, index_ctx); | |
4906 | if (r < 0) | |
4907 | return r; | |
4908 | ||
4909 | return 0; | |
4910 | } | |
4911 | ||
4912 | /** | |
4913 | * set up a bucket listing. | |
4914 | * handle is filled in. | |
4915 | * Returns 0 on success, -ERR# otherwise. | |
4916 | */ | |
4917 | int RGWRados::list_buckets_init(RGWAccessHandle *handle) | |
4918 | { | |
4919 | librados::NObjectIterator *state = new librados::NObjectIterator(root_pool_ctx.nobjects_begin()); | |
4920 | *handle = (RGWAccessHandle)state; | |
4921 | return 0; | |
4922 | } | |
4923 | ||
4924 | /** | |
4925 | * get the next bucket in the listing. | |
4926 | * obj is filled in, | |
4927 | * handle is updated. | |
4928 | * returns 0 on success, -ERR# otherwise. | |
4929 | */ | |
4930 | int RGWRados::list_buckets_next(rgw_bucket_dir_entry& obj, RGWAccessHandle *handle) | |
4931 | { | |
4932 | librados::NObjectIterator *state = (librados::NObjectIterator *)*handle; | |
4933 | ||
4934 | do { | |
4935 | if (*state == root_pool_ctx.nobjects_end()) { | |
4936 | delete state; | |
4937 | return -ENOENT; | |
4938 | } | |
4939 | ||
4940 | obj.key.name = (*state)->get_oid(); | |
4941 | if (obj.key.name[0] == '_') { | |
4942 | obj.key.name = obj.key.name.substr(1); | |
4943 | } | |
4944 | ||
4945 | (*state)++; | |
4946 | } while (obj.key.name[0] == '.'); /* skip all entries starting with '.' */ | |
4947 | ||
4948 | return 0; | |
4949 | } | |
4950 | ||
4951 | ||
4952 | /**** logs ****/ | |
4953 | ||
4954 | struct log_list_state { | |
4955 | string prefix; | |
4956 | librados::IoCtx io_ctx; | |
4957 | librados::NObjectIterator obit; | |
4958 | }; | |
4959 | ||
4960 | int RGWRados::log_list_init(const string& prefix, RGWAccessHandle *handle) | |
4961 | { | |
4962 | log_list_state *state = new log_list_state; | |
4963 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, state->io_ctx); | |
4964 | if (r < 0) { | |
4965 | delete state; | |
4966 | return r; | |
4967 | } | |
4968 | state->prefix = prefix; | |
4969 | state->obit = state->io_ctx.nobjects_begin(); | |
4970 | *handle = (RGWAccessHandle)state; | |
4971 | return 0; | |
4972 | } | |
4973 | ||
4974 | int RGWRados::log_list_next(RGWAccessHandle handle, string *name) | |
4975 | { | |
4976 | log_list_state *state = static_cast<log_list_state *>(handle); | |
4977 | while (true) { | |
4978 | if (state->obit == state->io_ctx.nobjects_end()) { | |
4979 | delete state; | |
4980 | return -ENOENT; | |
4981 | } | |
4982 | if (state->prefix.length() && | |
4983 | state->obit->get_oid().find(state->prefix) != 0) { | |
4984 | state->obit++; | |
4985 | continue; | |
4986 | } | |
4987 | *name = state->obit->get_oid(); | |
4988 | state->obit++; | |
4989 | break; | |
4990 | } | |
4991 | return 0; | |
4992 | } | |
4993 | ||
4994 | int RGWRados::log_remove(const string& name) | |
4995 | { | |
4996 | librados::IoCtx io_ctx; | |
4997 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
4998 | if (r < 0) | |
4999 | return r; | |
5000 | return io_ctx.remove(name); | |
5001 | } | |
5002 | ||
5003 | struct log_show_state { | |
5004 | librados::IoCtx io_ctx; | |
5005 | bufferlist bl; | |
5006 | bufferlist::iterator p; | |
5007 | string name; | |
5008 | uint64_t pos; | |
5009 | bool eof; | |
5010 | log_show_state() : pos(0), eof(false) {} | |
5011 | }; | |
5012 | ||
5013 | int RGWRados::log_show_init(const string& name, RGWAccessHandle *handle) | |
5014 | { | |
5015 | log_show_state *state = new log_show_state; | |
5016 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, state->io_ctx); | |
5017 | if (r < 0) { | |
5018 | delete state; | |
5019 | return r; | |
5020 | } | |
5021 | state->name = name; | |
5022 | *handle = (RGWAccessHandle)state; | |
5023 | return 0; | |
5024 | } | |
5025 | ||
5026 | int RGWRados::log_show_next(RGWAccessHandle handle, rgw_log_entry *entry) | |
5027 | { | |
5028 | log_show_state *state = static_cast<log_show_state *>(handle); | |
5029 | off_t off = state->p.get_off(); | |
5030 | ||
5031 | ldout(cct, 10) << "log_show_next pos " << state->pos << " bl " << state->bl.length() | |
5032 | << " off " << off | |
5033 | << " eof " << (int)state->eof | |
5034 | << dendl; | |
5035 | // read some? | |
5036 | unsigned chunk = 1024*1024; | |
5037 | if ((state->bl.length() - off) < chunk/2 && !state->eof) { | |
5038 | bufferlist more; | |
5039 | int r = state->io_ctx.read(state->name, more, chunk, state->pos); | |
5040 | if (r < 0) | |
5041 | return r; | |
5042 | state->pos += r; | |
5043 | bufferlist old; | |
5044 | try { | |
5045 | old.substr_of(state->bl, off, state->bl.length() - off); | |
5046 | } catch (buffer::error& err) { | |
5047 | return -EINVAL; | |
5048 | } | |
5049 | state->bl.clear(); | |
5050 | state->bl.claim(old); | |
5051 | state->bl.claim_append(more); | |
5052 | state->p = state->bl.begin(); | |
5053 | if ((unsigned)r < chunk) | |
5054 | state->eof = true; | |
5055 | ldout(cct, 10) << " read " << r << dendl; | |
5056 | } | |
5057 | ||
5058 | if (state->p.end()) | |
5059 | return 0; // end of file | |
5060 | try { | |
5061 | ::decode(*entry, state->p); | |
5062 | } | |
5063 | catch (const buffer::error &e) { | |
5064 | return -EINVAL; | |
5065 | } | |
5066 | return 1; | |
5067 | } | |
5068 | ||
5069 | /** | |
5070 | * usage_log_hash: get usage log key hash, based on name and index | |
5071 | * | |
5072 | * Get the usage object name. Since a user may have more than 1 | |
5073 | * object holding that info (multiple shards), we use index to | |
5074 | * specify that shard number. Once index exceeds max shards it | |
5075 | * wraps. | |
5076 | * If name is not being set, results for all users will be returned | |
5077 | * and index will wrap only after total shards number. | |
5078 | * | |
5079 | * @param cct [in] ceph context | |
5080 | * @param name [in] user name | |
5081 | * @param hash [out] hash value | |
5082 | * @param index [in] shard index number | |
5083 | */ | |
5084 | static void usage_log_hash(CephContext *cct, const string& name, string& hash, uint32_t index) | |
5085 | { | |
5086 | uint32_t val = index; | |
5087 | ||
5088 | if (!name.empty()) { | |
c07f9fc5 | 5089 | int max_user_shards = cct->_conf->rgw_usage_max_user_shards; |
7c673cae FG |
5090 | val %= max_user_shards; |
5091 | val += ceph_str_hash_linux(name.c_str(), name.size()); | |
5092 | } | |
5093 | char buf[17]; | |
c07f9fc5 | 5094 | int max_shards = cct->_conf->rgw_usage_max_shards; |
7c673cae FG |
5095 | snprintf(buf, sizeof(buf), RGW_USAGE_OBJ_PREFIX "%u", (unsigned)(val % max_shards)); |
5096 | hash = buf; | |
5097 | } | |
5098 | ||
5099 | int RGWRados::log_usage(map<rgw_user_bucket, RGWUsageBatch>& usage_info) | |
5100 | { | |
5101 | uint32_t index = 0; | |
5102 | ||
5103 | map<string, rgw_usage_log_info> log_objs; | |
5104 | ||
5105 | string hash; | |
5106 | string last_user; | |
5107 | ||
5108 | /* restructure usage map, zone by object hash */ | |
5109 | map<rgw_user_bucket, RGWUsageBatch>::iterator iter; | |
5110 | for (iter = usage_info.begin(); iter != usage_info.end(); ++iter) { | |
5111 | const rgw_user_bucket& ub = iter->first; | |
5112 | RGWUsageBatch& info = iter->second; | |
5113 | ||
5114 | if (ub.user.empty()) { | |
5115 | ldout(cct, 0) << "WARNING: RGWRados::log_usage(): user name empty (bucket=" << ub.bucket << "), skipping" << dendl; | |
5116 | continue; | |
5117 | } | |
5118 | ||
5119 | if (ub.user != last_user) { | |
5120 | /* index *should* be random, but why waste extra cycles | |
5121 | in most cases max user shards is not going to exceed 1, | |
5122 | so just incrementing it */ | |
5123 | usage_log_hash(cct, ub.user, hash, index++); | |
5124 | } | |
5125 | last_user = ub.user; | |
5126 | vector<rgw_usage_log_entry>& v = log_objs[hash].entries; | |
5127 | ||
5128 | for (auto miter = info.m.begin(); miter != info.m.end(); ++miter) { | |
5129 | v.push_back(miter->second); | |
5130 | } | |
5131 | } | |
5132 | ||
5133 | map<string, rgw_usage_log_info>::iterator liter; | |
5134 | ||
5135 | for (liter = log_objs.begin(); liter != log_objs.end(); ++liter) { | |
5136 | int r = cls_obj_usage_log_add(liter->first, liter->second); | |
5137 | if (r < 0) | |
5138 | return r; | |
5139 | } | |
5140 | return 0; | |
5141 | } | |
5142 | ||
5143 | int RGWRados::read_usage(const rgw_user& user, uint64_t start_epoch, uint64_t end_epoch, uint32_t max_entries, | |
5144 | bool *is_truncated, RGWUsageIter& usage_iter, map<rgw_user_bucket, rgw_usage_log_entry>& usage) | |
5145 | { | |
5146 | uint32_t num = max_entries; | |
5147 | string hash, first_hash; | |
5148 | string user_str = user.to_str(); | |
5149 | usage_log_hash(cct, user_str, first_hash, 0); | |
5150 | ||
5151 | if (usage_iter.index) { | |
5152 | usage_log_hash(cct, user_str, hash, usage_iter.index); | |
5153 | } else { | |
5154 | hash = first_hash; | |
5155 | } | |
5156 | ||
5157 | usage.clear(); | |
5158 | ||
5159 | do { | |
5160 | map<rgw_user_bucket, rgw_usage_log_entry> ret_usage; | |
5161 | map<rgw_user_bucket, rgw_usage_log_entry>::iterator iter; | |
5162 | ||
5163 | int ret = cls_obj_usage_log_read(hash, user_str, start_epoch, end_epoch, num, | |
5164 | usage_iter.read_iter, ret_usage, is_truncated); | |
5165 | if (ret == -ENOENT) | |
5166 | goto next; | |
5167 | ||
5168 | if (ret < 0) | |
5169 | return ret; | |
5170 | ||
5171 | num -= ret_usage.size(); | |
5172 | ||
5173 | for (iter = ret_usage.begin(); iter != ret_usage.end(); ++iter) { | |
5174 | usage[iter->first].aggregate(iter->second); | |
5175 | } | |
5176 | ||
5177 | next: | |
5178 | if (!*is_truncated) { | |
5179 | usage_iter.read_iter.clear(); | |
5180 | usage_log_hash(cct, user_str, hash, ++usage_iter.index); | |
5181 | } | |
5182 | } while (num && !*is_truncated && hash != first_hash); | |
5183 | return 0; | |
5184 | } | |
5185 | ||
5186 | int RGWRados::trim_usage(rgw_user& user, uint64_t start_epoch, uint64_t end_epoch) | |
5187 | { | |
5188 | uint32_t index = 0; | |
5189 | string hash, first_hash; | |
5190 | string user_str = user.to_str(); | |
5191 | usage_log_hash(cct, user_str, first_hash, index); | |
5192 | ||
5193 | hash = first_hash; | |
7c673cae FG |
5194 | do { |
5195 | int ret = cls_obj_usage_log_trim(hash, user_str, start_epoch, end_epoch); | |
7c673cae | 5196 | |
b32b8144 | 5197 | if (ret < 0 && ret != -ENOENT) |
7c673cae FG |
5198 | return ret; |
5199 | ||
7c673cae FG |
5200 | usage_log_hash(cct, user_str, hash, ++index); |
5201 | } while (hash != first_hash); | |
5202 | ||
5203 | return 0; | |
5204 | } | |
5205 | ||
7c673cae FG |
5206 | int RGWRados::key_to_shard_id(const string& key, int max_shards) |
5207 | { | |
31f18b77 | 5208 | return rgw_shards_hash(key, max_shards); |
7c673cae FG |
5209 | } |
5210 | ||
5211 | void RGWRados::shard_name(const string& prefix, unsigned max_shards, const string& key, string& name, int *shard_id) | |
5212 | { | |
5213 | uint32_t val = ceph_str_hash_linux(key.c_str(), key.size()); | |
5214 | char buf[16]; | |
5215 | if (shard_id) { | |
5216 | *shard_id = val % max_shards; | |
5217 | } | |
5218 | snprintf(buf, sizeof(buf), "%u", (unsigned)(val % max_shards)); | |
5219 | name = prefix + buf; | |
5220 | } | |
5221 | ||
5222 | void RGWRados::shard_name(const string& prefix, unsigned max_shards, const string& section, const string& key, string& name) | |
5223 | { | |
5224 | uint32_t val = ceph_str_hash_linux(key.c_str(), key.size()); | |
5225 | val ^= ceph_str_hash_linux(section.c_str(), section.size()); | |
5226 | char buf[16]; | |
5227 | snprintf(buf, sizeof(buf), "%u", (unsigned)(val % max_shards)); | |
5228 | name = prefix + buf; | |
5229 | } | |
5230 | ||
5231 | void RGWRados::shard_name(const string& prefix, unsigned shard_id, string& name) | |
5232 | { | |
5233 | char buf[16]; | |
5234 | snprintf(buf, sizeof(buf), "%u", shard_id); | |
5235 | name = prefix + buf; | |
5236 | ||
5237 | } | |
5238 | ||
5239 | void RGWRados::time_log_prepare_entry(cls_log_entry& entry, const real_time& ut, const string& section, const string& key, bufferlist& bl) | |
5240 | { | |
5241 | cls_log_add_prepare_entry(entry, utime_t(ut), section, key, bl); | |
5242 | } | |
5243 | ||
5244 | int RGWRados::time_log_add_init(librados::IoCtx& io_ctx) | |
5245 | { | |
5246 | return rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx, true); | |
5247 | ||
5248 | } | |
5249 | ||
5250 | int RGWRados::time_log_add(const string& oid, const real_time& ut, const string& section, const string& key, bufferlist& bl) | |
5251 | { | |
5252 | librados::IoCtx io_ctx; | |
5253 | ||
5254 | int r = time_log_add_init(io_ctx); | |
5255 | if (r < 0) { | |
5256 | return r; | |
5257 | } | |
5258 | ||
5259 | ObjectWriteOperation op; | |
5260 | utime_t t(ut); | |
5261 | cls_log_add(op, t, section, key, bl); | |
5262 | ||
5263 | return io_ctx.operate(oid, &op); | |
5264 | } | |
5265 | ||
5266 | int RGWRados::time_log_add(const string& oid, list<cls_log_entry>& entries, | |
5267 | librados::AioCompletion *completion, bool monotonic_inc) | |
5268 | { | |
5269 | librados::IoCtx io_ctx; | |
5270 | ||
5271 | int r = time_log_add_init(io_ctx); | |
5272 | if (r < 0) { | |
5273 | return r; | |
5274 | } | |
5275 | ||
5276 | ObjectWriteOperation op; | |
5277 | cls_log_add(op, entries, monotonic_inc); | |
5278 | ||
5279 | if (!completion) { | |
5280 | r = io_ctx.operate(oid, &op); | |
5281 | } else { | |
5282 | r = io_ctx.aio_operate(oid, completion, &op); | |
5283 | } | |
5284 | return r; | |
5285 | } | |
5286 | ||
5287 | int RGWRados::time_log_list(const string& oid, const real_time& start_time, const real_time& end_time, | |
5288 | int max_entries, list<cls_log_entry>& entries, | |
5289 | const string& marker, | |
5290 | string *out_marker, | |
5291 | bool *truncated) | |
5292 | { | |
5293 | librados::IoCtx io_ctx; | |
5294 | ||
5295 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5296 | if (r < 0) | |
5297 | return r; | |
5298 | librados::ObjectReadOperation op; | |
5299 | ||
5300 | utime_t st(start_time); | |
5301 | utime_t et(end_time); | |
5302 | ||
5303 | cls_log_list(op, st, et, marker, max_entries, entries, | |
5304 | out_marker, truncated); | |
5305 | ||
5306 | bufferlist obl; | |
5307 | ||
5308 | int ret = io_ctx.operate(oid, &op, &obl); | |
5309 | if (ret < 0) | |
5310 | return ret; | |
5311 | ||
5312 | return 0; | |
5313 | } | |
5314 | ||
5315 | int RGWRados::time_log_info(const string& oid, cls_log_header *header) | |
5316 | { | |
5317 | librados::IoCtx io_ctx; | |
5318 | ||
5319 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5320 | if (r < 0) | |
5321 | return r; | |
5322 | librados::ObjectReadOperation op; | |
5323 | ||
5324 | cls_log_info(op, header); | |
5325 | ||
5326 | bufferlist obl; | |
5327 | ||
5328 | int ret = io_ctx.operate(oid, &op, &obl); | |
5329 | if (ret < 0) | |
5330 | return ret; | |
5331 | ||
5332 | return 0; | |
5333 | } | |
5334 | ||
5335 | int RGWRados::time_log_info_async(librados::IoCtx& io_ctx, const string& oid, cls_log_header *header, librados::AioCompletion *completion) | |
5336 | { | |
5337 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5338 | if (r < 0) | |
5339 | return r; | |
5340 | ||
5341 | librados::ObjectReadOperation op; | |
5342 | ||
5343 | cls_log_info(op, header); | |
5344 | ||
5345 | int ret = io_ctx.aio_operate(oid, completion, &op, NULL); | |
5346 | if (ret < 0) | |
5347 | return ret; | |
5348 | ||
5349 | return 0; | |
5350 | } | |
5351 | ||
5352 | int RGWRados::time_log_trim(const string& oid, const real_time& start_time, const real_time& end_time, | |
5353 | const string& from_marker, const string& to_marker, | |
5354 | librados::AioCompletion *completion) | |
5355 | { | |
5356 | librados::IoCtx io_ctx; | |
5357 | ||
5358 | int r = rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool, io_ctx); | |
5359 | if (r < 0) | |
5360 | return r; | |
5361 | ||
5362 | utime_t st(start_time); | |
5363 | utime_t et(end_time); | |
5364 | ||
5365 | ObjectWriteOperation op; | |
5366 | cls_log_trim(op, st, et, from_marker, to_marker); | |
5367 | ||
5368 | if (!completion) { | |
5369 | r = io_ctx.operate(oid, &op); | |
5370 | } else { | |
5371 | r = io_ctx.aio_operate(oid, completion, &op); | |
5372 | } | |
5373 | return r; | |
5374 | } | |
5375 | ||
5376 | string RGWRados::objexp_hint_get_shardname(int shard_num) | |
5377 | { | |
5378 | char buf[32]; | |
5379 | snprintf(buf, sizeof(buf), "%010u", (unsigned)shard_num); | |
5380 | ||
5381 | string objname("obj_delete_at_hint."); | |
5382 | return objname + buf; | |
5383 | } | |
5384 | ||
7c673cae FG |
5385 | int RGWRados::objexp_key_shard(const rgw_obj_index_key& key) |
5386 | { | |
5387 | string obj_key = key.name + key.instance; | |
5388 | int num_shards = cct->_conf->rgw_objexp_hints_num_shards; | |
5389 | uint32_t sid = ceph_str_hash_linux(obj_key.c_str(), obj_key.size()); | |
5390 | uint32_t sid2 = sid ^ ((sid & 0xFF) << 24); | |
31f18b77 FG |
5391 | sid = rgw_shards_mod(sid2, num_shards); |
5392 | return sid; | |
7c673cae FG |
5393 | } |
5394 | ||
5395 | static string objexp_hint_get_keyext(const string& tenant_name, | |
5396 | const string& bucket_name, | |
5397 | const string& bucket_id, | |
5398 | const rgw_obj_key& obj_key) | |
5399 | { | |
5400 | return tenant_name + (tenant_name.empty() ? "" : ":") + bucket_name + ":" + bucket_id + | |
5401 | ":" + obj_key.name + ":" + obj_key.instance; | |
5402 | } | |
5403 | ||
5404 | int RGWRados::objexp_hint_add(const ceph::real_time& delete_at, | |
5405 | const string& tenant_name, | |
5406 | const string& bucket_name, | |
5407 | const string& bucket_id, | |
5408 | const rgw_obj_index_key& obj_key) | |
5409 | { | |
5410 | const string keyext = objexp_hint_get_keyext(tenant_name, bucket_name, | |
5411 | bucket_id, obj_key); | |
5412 | objexp_hint_entry he = { | |
5413 | .tenant = tenant_name, | |
5414 | .bucket_name = bucket_name, | |
5415 | .bucket_id = bucket_id, | |
5416 | .obj_key = obj_key, | |
5417 | .exp_time = delete_at }; | |
5418 | bufferlist hebl; | |
5419 | ::encode(he, hebl); | |
5420 | ObjectWriteOperation op; | |
5421 | cls_timeindex_add(op, utime_t(delete_at), keyext, hebl); | |
5422 | ||
5423 | string shard_name = objexp_hint_get_shardname(objexp_key_shard(obj_key)); | |
5424 | return objexp_pool_ctx.operate(shard_name, &op); | |
5425 | } | |
5426 | ||
5427 | void RGWRados::objexp_get_shard(int shard_num, | |
5428 | string& shard) /* out */ | |
5429 | { | |
5430 | shard = objexp_hint_get_shardname(shard_num); | |
5431 | } | |
5432 | ||
5433 | int RGWRados::objexp_hint_list(const string& oid, | |
5434 | const ceph::real_time& start_time, | |
5435 | const ceph::real_time& end_time, | |
5436 | const int max_entries, | |
5437 | const string& marker, | |
5438 | list<cls_timeindex_entry>& entries, /* out */ | |
5439 | string *out_marker, /* out */ | |
5440 | bool *truncated) /* out */ | |
5441 | { | |
5442 | librados::ObjectReadOperation op; | |
5443 | cls_timeindex_list(op, utime_t(start_time), utime_t(end_time), marker, max_entries, entries, | |
5444 | out_marker, truncated); | |
5445 | ||
5446 | bufferlist obl; | |
5447 | int ret = objexp_pool_ctx.operate(oid, &op, &obl); | |
5448 | ||
5449 | if ((ret < 0 ) && (ret != -ENOENT)) { | |
5450 | return ret; | |
5451 | } | |
5452 | ||
5453 | if ((ret == -ENOENT) && truncated) { | |
5454 | *truncated = false; | |
5455 | } | |
5456 | ||
5457 | return 0; | |
5458 | } | |
5459 | ||
5460 | int RGWRados::objexp_hint_parse(cls_timeindex_entry &ti_entry, /* in */ | |
5461 | objexp_hint_entry& hint_entry) /* out */ | |
5462 | { | |
5463 | try { | |
5464 | bufferlist::iterator iter = ti_entry.value.begin(); | |
5465 | ::decode(hint_entry, iter); | |
5466 | } catch (buffer::error& err) { | |
5467 | ldout(cct, 0) << "ERROR: couldn't decode avail_pools" << dendl; | |
5468 | } | |
5469 | ||
5470 | return 0; | |
5471 | } | |
5472 | ||
5473 | int RGWRados::objexp_hint_trim(const string& oid, | |
5474 | const ceph::real_time& start_time, | |
5475 | const ceph::real_time& end_time, | |
5476 | const string& from_marker, | |
5477 | const string& to_marker) | |
5478 | { | |
5479 | int ret = cls_timeindex_trim(objexp_pool_ctx, oid, utime_t(start_time), utime_t(end_time), | |
5480 | from_marker, to_marker); | |
5481 | if ((ret < 0 ) && (ret != -ENOENT)) { | |
5482 | return ret; | |
5483 | } | |
5484 | ||
5485 | return 0; | |
5486 | } | |
5487 | ||
5488 | int RGWRados::lock_exclusive(rgw_pool& pool, const string& oid, timespan& duration, | |
5489 | string& zone_id, string& owner_id) { | |
5490 | librados::IoCtx io_ctx; | |
5491 | ||
5492 | int r = rgw_init_ioctx(get_rados_handle(), pool, io_ctx); | |
5493 | if (r < 0) { | |
5494 | return r; | |
5495 | } | |
5496 | uint64_t msec = std::chrono::duration_cast<std::chrono::milliseconds>(duration).count(); | |
5497 | utime_t ut(msec / 1000, msec % 1000); | |
5498 | ||
5499 | rados::cls::lock::Lock l(log_lock_name); | |
5500 | l.set_duration(ut); | |
5501 | l.set_cookie(owner_id); | |
5502 | l.set_tag(zone_id); | |
5503 | l.set_renew(true); | |
5504 | ||
5505 | return l.lock_exclusive(&io_ctx, oid); | |
5506 | } | |
5507 | ||
5508 | int RGWRados::unlock(rgw_pool& pool, const string& oid, string& zone_id, string& owner_id) { | |
5509 | librados::IoCtx io_ctx; | |
5510 | ||
5511 | int r = rgw_init_ioctx(get_rados_handle(), pool, io_ctx); | |
5512 | if (r < 0) { | |
5513 | return r; | |
5514 | } | |
5515 | ||
5516 | rados::cls::lock::Lock l(log_lock_name); | |
5517 | l.set_tag(zone_id); | |
5518 | l.set_cookie(owner_id); | |
5519 | ||
5520 | return l.unlock(&io_ctx, oid); | |
5521 | } | |
5522 | ||
5523 | int RGWRados::decode_policy(bufferlist& bl, ACLOwner *owner) | |
5524 | { | |
5525 | bufferlist::iterator i = bl.begin(); | |
5526 | RGWAccessControlPolicy policy(cct); | |
5527 | try { | |
5528 | policy.decode_owner(i); | |
5529 | } catch (buffer::error& err) { | |
5530 | ldout(cct, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl; | |
5531 | return -EIO; | |
5532 | } | |
5533 | *owner = policy.get_owner(); | |
5534 | return 0; | |
5535 | } | |
5536 | ||
5537 | int rgw_policy_from_attrset(CephContext *cct, map<string, bufferlist>& attrset, RGWAccessControlPolicy *policy) | |
5538 | { | |
5539 | map<string, bufferlist>::iterator aiter = attrset.find(RGW_ATTR_ACL); | |
5540 | if (aiter == attrset.end()) | |
5541 | return -EIO; | |
5542 | ||
5543 | bufferlist& bl = aiter->second; | |
5544 | bufferlist::iterator iter = bl.begin(); | |
5545 | try { | |
5546 | policy->decode(iter); | |
5547 | } catch (buffer::error& err) { | |
5548 | ldout(cct, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl; | |
5549 | return -EIO; | |
5550 | } | |
5551 | if (cct->_conf->subsys.should_gather(ceph_subsys_rgw, 15)) { | |
5552 | RGWAccessControlPolicy_S3 *s3policy = static_cast<RGWAccessControlPolicy_S3 *>(policy); | |
5553 | ldout(cct, 15) << __func__ << " Read AccessControlPolicy"; | |
5554 | s3policy->to_xml(*_dout); | |
5555 | *_dout << dendl; | |
5556 | } | |
5557 | return 0; | |
5558 | } | |
5559 | ||
5560 | ||
31f18b77 FG |
5561 | int RGWRados::Bucket::update_bucket_id(const string& new_bucket_id) |
5562 | { | |
5563 | rgw_bucket bucket = bucket_info.bucket; | |
5564 | bucket.update_bucket_id(new_bucket_id); | |
5565 | ||
5566 | RGWObjectCtx obj_ctx(store); | |
5567 | ||
5568 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, nullptr, nullptr); | |
5569 | if (ret < 0) { | |
5570 | return ret; | |
5571 | } | |
5572 | ||
5573 | return 0; | |
5574 | } | |
5575 | ||
7c673cae FG |
5576 | /** |
5577 | * get listing of the objects in a bucket. | |
5578 | * | |
5579 | * max: maximum number of results to return | |
5580 | * bucket: bucket to list contents of | |
5581 | * prefix: only return results that match this prefix | |
5582 | * delim: do not include results that match this string. | |
5583 | * Any skipped results will have the matching portion of their name | |
5584 | * inserted in common_prefixes with a "true" mark. | |
5585 | * marker: if filled in, begin the listing with this object. | |
5586 | * end_marker: if filled in, end the listing with this object. | |
5587 | * result: the objects are put in here. | |
5588 | * common_prefixes: if delim is filled in, any matching prefixes are placed here. | |
5589 | * is_truncated: if number of objects in the bucket is bigger than max, then truncated. | |
5590 | */ | |
c07f9fc5 FG |
5591 | int RGWRados::Bucket::List::list_objects(int64_t max, |
5592 | vector<rgw_bucket_dir_entry> *result, | |
7c673cae FG |
5593 | map<string, bool> *common_prefixes, |
5594 | bool *is_truncated) | |
5595 | { | |
5596 | RGWRados *store = target->get_store(); | |
5597 | CephContext *cct = store->ctx(); | |
5598 | int shard_id = target->get_shard_id(); | |
5599 | ||
5600 | int count = 0; | |
5601 | bool truncated = true; | |
5602 | int read_ahead = std::max(cct->_conf->rgw_list_bucket_min_readahead,max); | |
5603 | ||
5604 | result->clear(); | |
5605 | ||
5606 | rgw_obj_key marker_obj(params.marker.name, params.marker.instance, params.ns); | |
7c673cae FG |
5607 | rgw_obj_index_key cur_marker; |
5608 | marker_obj.get_index_key(&cur_marker); | |
5609 | ||
3efd9988 FG |
5610 | rgw_obj_key end_marker_obj(params.end_marker.name, params.end_marker.instance, |
5611 | params.ns); | |
5612 | rgw_obj_index_key cur_end_marker; | |
5613 | end_marker_obj.get_index_key(&cur_end_marker); | |
7c673cae FG |
5614 | const bool cur_end_marker_valid = !params.end_marker.empty(); |
5615 | ||
5616 | rgw_obj_key prefix_obj(params.prefix); | |
5617 | prefix_obj.ns = params.ns; | |
5618 | string cur_prefix = prefix_obj.get_index_key_name(); | |
5619 | ||
5620 | string bigger_than_delim; | |
5621 | ||
5622 | if (!params.delim.empty()) { | |
5623 | unsigned long val = decode_utf8((unsigned char *)params.delim.c_str(), params.delim.size()); | |
5624 | char buf[params.delim.size() + 16]; | |
5625 | int r = encode_utf8(val + 1, (unsigned char *)buf); | |
5626 | if (r < 0) { | |
5627 | ldout(cct,0) << "ERROR: encode_utf8() failed" << dendl; | |
5628 | return -EINVAL; | |
5629 | } | |
5630 | buf[r] = '\0'; | |
5631 | ||
5632 | bigger_than_delim = buf; | |
5633 | ||
5634 | /* if marker points at a common prefix, fast forward it into its upperbound string */ | |
224ce89b | 5635 | int delim_pos = cur_marker.name.find(params.delim, cur_prefix.size()); |
7c673cae FG |
5636 | if (delim_pos >= 0) { |
5637 | string s = cur_marker.name.substr(0, delim_pos); | |
5638 | s.append(bigger_than_delim); | |
5639 | cur_marker = s; | |
5640 | } | |
5641 | } | |
5642 | ||
5643 | string skip_after_delim; | |
5644 | while (truncated && count <= max) { | |
5645 | if (skip_after_delim > cur_marker.name) { | |
5646 | cur_marker = skip_after_delim; | |
5647 | ldout(cct, 20) << "setting cur_marker=" << cur_marker.name << "[" << cur_marker.instance << "]" << dendl; | |
5648 | } | |
5649 | std::map<string, rgw_bucket_dir_entry> ent_map; | |
5650 | int r = store->cls_bucket_list(target->get_bucket_info(), shard_id, cur_marker, cur_prefix, | |
5651 | read_ahead + 1 - count, params.list_versions, ent_map, | |
5652 | &truncated, &cur_marker); | |
5653 | if (r < 0) | |
5654 | return r; | |
5655 | ||
5656 | std::map<string, rgw_bucket_dir_entry>::iterator eiter; | |
5657 | for (eiter = ent_map.begin(); eiter != ent_map.end(); ++eiter) { | |
5658 | rgw_bucket_dir_entry& entry = eiter->second; | |
5659 | rgw_obj_index_key index_key = entry.key; | |
5660 | ||
5661 | rgw_obj_key obj(index_key); | |
5662 | ||
5663 | /* note that parse_raw_oid() here will not set the correct object's instance, as | |
5664 | * rgw_obj_index_key encodes that separately. We don't need to set the instance because it's | |
5665 | * not needed for the checks here and we end up using the raw entry for the return vector | |
5666 | */ | |
5667 | bool valid = rgw_obj_key::parse_raw_oid(index_key.name, &obj); | |
5668 | if (!valid) { | |
5669 | ldout(cct, 0) << "ERROR: could not parse object name: " << obj.name << dendl; | |
5670 | continue; | |
5671 | } | |
5672 | bool check_ns = (obj.ns == params.ns); | |
5673 | if (!params.list_versions && !entry.is_visible()) { | |
5674 | continue; | |
5675 | } | |
5676 | ||
5677 | if (params.enforce_ns && !check_ns) { | |
5678 | if (!params.ns.empty()) { | |
5679 | /* we've iterated past the namespace we're searching -- done now */ | |
5680 | truncated = false; | |
5681 | goto done; | |
5682 | } | |
5683 | ||
5684 | /* we're not looking at the namespace this object is in, next! */ | |
5685 | continue; | |
5686 | } | |
5687 | ||
5688 | if (cur_end_marker_valid && cur_end_marker <= index_key) { | |
5689 | truncated = false; | |
5690 | goto done; | |
5691 | } | |
5692 | ||
5693 | if (count < max) { | |
5694 | params.marker = index_key; | |
5695 | next_marker = index_key; | |
5696 | } | |
5697 | ||
5698 | if (params.filter && !params.filter->filter(obj.name, index_key.name)) | |
5699 | continue; | |
5700 | ||
5701 | if (params.prefix.size() && (obj.name.compare(0, params.prefix.size(), params.prefix) != 0)) | |
5702 | continue; | |
5703 | ||
5704 | if (!params.delim.empty()) { | |
5705 | int delim_pos = obj.name.find(params.delim, params.prefix.size()); | |
5706 | ||
5707 | if (delim_pos >= 0) { | |
5708 | string prefix_key = obj.name.substr(0, delim_pos + 1); | |
5709 | ||
5710 | if (common_prefixes && | |
5711 | common_prefixes->find(prefix_key) == common_prefixes->end()) { | |
5712 | if (count >= max) { | |
5713 | truncated = true; | |
5714 | goto done; | |
5715 | } | |
5716 | next_marker = prefix_key; | |
5717 | (*common_prefixes)[prefix_key] = true; | |
5718 | ||
224ce89b WB |
5719 | int marker_delim_pos = cur_marker.name.find(params.delim, cur_prefix.size()); |
5720 | ||
5721 | skip_after_delim = cur_marker.name.substr(0, marker_delim_pos); | |
7c673cae FG |
5722 | skip_after_delim.append(bigger_than_delim); |
5723 | ||
5724 | ldout(cct, 20) << "skip_after_delim=" << skip_after_delim << dendl; | |
5725 | ||
5726 | count++; | |
5727 | } | |
5728 | ||
5729 | continue; | |
5730 | } | |
5731 | } | |
5732 | ||
5733 | if (count >= max) { | |
5734 | truncated = true; | |
5735 | goto done; | |
5736 | } | |
5737 | ||
5738 | result->emplace_back(std::move(entry)); | |
5739 | count++; | |
5740 | } | |
5741 | ||
5742 | // Either the back-end telling us truncated, or we don't consume all | |
5743 | // items returned per the amount caller request | |
5744 | truncated = (truncated || eiter != ent_map.end()); | |
5745 | } | |
5746 | ||
5747 | done: | |
5748 | if (is_truncated) | |
5749 | *is_truncated = truncated; | |
5750 | ||
5751 | return 0; | |
5752 | } | |
5753 | ||
5754 | /** | |
5755 | * create a rados pool, associated meta info | |
5756 | * returns 0 on success, -ERR# otherwise. | |
5757 | */ | |
5758 | int RGWRados::create_pool(const rgw_pool& pool) | |
5759 | { | |
c07f9fc5 | 5760 | librados::IoCtx io_ctx; |
28e407b8 AA |
5761 | constexpr bool create = true; |
5762 | return rgw_init_ioctx(get_rados_handle(), pool, io_ctx, create); | |
7c673cae FG |
5763 | } |
5764 | ||
5765 | int RGWRados::init_bucket_index(RGWBucketInfo& bucket_info, int num_shards) | |
5766 | { | |
5767 | librados::IoCtx index_ctx; // context for new bucket | |
5768 | ||
31f18b77 | 5769 | string dir_oid = dir_oid_prefix; |
7c673cae | 5770 | int r = open_bucket_index_ctx(bucket_info, index_ctx); |
31f18b77 | 5771 | if (r < 0) { |
7c673cae | 5772 | return r; |
31f18b77 | 5773 | } |
7c673cae | 5774 | |
7c673cae FG |
5775 | dir_oid.append(bucket_info.bucket.bucket_id); |
5776 | ||
5777 | map<int, string> bucket_objs; | |
5778 | get_bucket_index_objects(dir_oid, num_shards, bucket_objs); | |
5779 | ||
5780 | return CLSRGWIssueBucketIndexInit(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); | |
5781 | } | |
5782 | ||
5783 | void RGWRados::create_bucket_id(string *bucket_id) | |
5784 | { | |
5785 | uint64_t iid = instance_id(); | |
5786 | uint64_t bid = next_bucket_id(); | |
5787 | char buf[get_zone_params().get_id().size() + 48]; | |
5788 | snprintf(buf, sizeof(buf), "%s.%llu.%llu", get_zone_params().get_id().c_str(), (long long)iid, (long long)bid); | |
5789 | *bucket_id = buf; | |
5790 | } | |
5791 | ||
7c673cae FG |
5792 | int RGWRados::create_bucket(RGWUserInfo& owner, rgw_bucket& bucket, |
5793 | const string& zonegroup_id, | |
5794 | const string& placement_rule, | |
5795 | const string& swift_ver_location, | |
5796 | const RGWQuotaInfo * pquota_info, | |
5797 | map<std::string, bufferlist>& attrs, | |
5798 | RGWBucketInfo& info, | |
5799 | obj_version *pobjv, | |
5800 | obj_version *pep_objv, | |
5801 | real_time creation_time, | |
5802 | rgw_bucket *pmaster_bucket, | |
5803 | uint32_t *pmaster_num_shards, | |
5804 | bool exclusive) | |
5805 | { | |
5806 | #define MAX_CREATE_RETRIES 20 /* need to bound retries */ | |
5807 | string selected_placement_rule_name; | |
5808 | RGWZonePlacementInfo rule_info; | |
5809 | ||
5810 | for (int i = 0; i < MAX_CREATE_RETRIES; i++) { | |
5811 | int ret = 0; | |
5812 | ret = select_bucket_placement(owner, zonegroup_id, placement_rule, | |
5813 | &selected_placement_rule_name, &rule_info); | |
5814 | if (ret < 0) | |
5815 | return ret; | |
5816 | ||
5817 | if (!pmaster_bucket) { | |
5818 | create_bucket_id(&bucket.marker); | |
5819 | bucket.bucket_id = bucket.marker; | |
5820 | } else { | |
5821 | bucket.marker = pmaster_bucket->marker; | |
5822 | bucket.bucket_id = pmaster_bucket->bucket_id; | |
5823 | } | |
5824 | ||
5825 | RGWObjVersionTracker& objv_tracker = info.objv_tracker; | |
5826 | ||
5827 | if (pobjv) { | |
5828 | objv_tracker.write_version = *pobjv; | |
5829 | } else { | |
5830 | objv_tracker.generate_new_write_ver(cct); | |
5831 | } | |
5832 | ||
5833 | info.bucket = bucket; | |
5834 | info.owner = owner.user_id; | |
5835 | info.zonegroup = zonegroup_id; | |
5836 | info.placement_rule = selected_placement_rule_name; | |
5837 | info.index_type = rule_info.index_type; | |
5838 | info.swift_ver_location = swift_ver_location; | |
5839 | info.swift_versioning = (!swift_ver_location.empty()); | |
5840 | if (pmaster_num_shards) { | |
5841 | info.num_shards = *pmaster_num_shards; | |
5842 | } else { | |
5843 | info.num_shards = bucket_index_max_shards; | |
5844 | } | |
5845 | info.bucket_index_shard_hash_type = RGWBucketInfo::MOD; | |
5846 | info.requester_pays = false; | |
5847 | if (real_clock::is_zero(creation_time)) { | |
5848 | info.creation_time = ceph::real_clock::now(); | |
5849 | } else { | |
5850 | info.creation_time = creation_time; | |
5851 | } | |
5852 | if (pquota_info) { | |
5853 | info.quota = *pquota_info; | |
5854 | } | |
5855 | ||
5856 | int r = init_bucket_index(info, info.num_shards); | |
5857 | if (r < 0) { | |
5858 | return r; | |
5859 | } | |
5860 | ||
5861 | ret = put_linked_bucket_info(info, exclusive, ceph::real_time(), pep_objv, &attrs, true); | |
5862 | if (ret == -EEXIST) { | |
5863 | librados::IoCtx index_ctx; | |
5864 | map<int, string> bucket_objs; | |
5865 | int r = open_bucket_index(info, index_ctx, bucket_objs); | |
5866 | if (r < 0) | |
5867 | return r; | |
5868 | ||
5869 | /* we need to reread the info and return it, caller will have a use for it */ | |
5870 | RGWObjVersionTracker instance_ver = info.objv_tracker; | |
5871 | info.objv_tracker.clear(); | |
5872 | RGWObjectCtx obj_ctx(this); | |
5873 | r = get_bucket_info(obj_ctx, bucket.tenant, bucket.name, info, NULL, NULL); | |
5874 | if (r < 0) { | |
5875 | if (r == -ENOENT) { | |
5876 | continue; | |
5877 | } | |
5878 | ldout(cct, 0) << "get_bucket_info returned " << r << dendl; | |
5879 | return r; | |
5880 | } | |
5881 | ||
5882 | /* only remove it if it's a different bucket instance */ | |
5883 | if (info.bucket.bucket_id != bucket.bucket_id) { | |
5884 | /* remove bucket meta instance */ | |
5885 | string entry = bucket.get_key(); | |
5886 | r = rgw_bucket_instance_remove_entry(this, entry, &instance_ver); | |
5887 | if (r < 0) | |
5888 | return r; | |
5889 | ||
5890 | map<int, string>::const_iterator biter; | |
5891 | for (biter = bucket_objs.begin(); biter != bucket_objs.end(); ++biter) { | |
5892 | // Do best effort removal | |
5893 | index_ctx.remove(biter->second); | |
5894 | } | |
5895 | } | |
5896 | /* ret == -ENOENT here */ | |
5897 | } | |
5898 | return ret; | |
5899 | } | |
5900 | ||
5901 | /* this is highly unlikely */ | |
5902 | ldout(cct, 0) << "ERROR: could not create bucket, continuously raced with bucket creation and removal" << dendl; | |
5903 | return -ENOENT; | |
5904 | } | |
5905 | ||
5906 | int RGWRados::select_new_bucket_location(RGWUserInfo& user_info, const string& zonegroup_id, const string& request_rule, | |
5907 | string *pselected_rule_name, RGWZonePlacementInfo *rule_info) | |
5908 | ||
5909 | { | |
c07f9fc5 | 5910 | /* first check that zonegroup exists within current period. */ |
7c673cae FG |
5911 | RGWZoneGroup zonegroup; |
5912 | int ret = get_zonegroup(zonegroup_id, zonegroup); | |
5913 | if (ret < 0) { | |
5914 | ldout(cct, 0) << "could not find zonegroup " << zonegroup_id << " in current period" << dendl; | |
5915 | return ret; | |
5916 | } | |
5917 | ||
7c673cae | 5918 | /* find placement rule. Hierarchy: request rule > user default rule > zonegroup default rule */ |
c07f9fc5 FG |
5919 | std::map<std::string, RGWZoneGroupPlacementTarget>::const_iterator titer; |
5920 | ||
5921 | if (!request_rule.empty()) { | |
5922 | titer = zonegroup.placement_targets.find(request_rule); | |
5923 | if (titer == zonegroup.placement_targets.end()) { | |
5924 | ldout(cct, 0) << "could not find requested placement id " << request_rule | |
5925 | << " within zonegroup " << dendl; | |
5926 | return -ERR_INVALID_LOCATION_CONSTRAINT; | |
5927 | } | |
5928 | } else if (!user_info.default_placement.empty()) { | |
5929 | titer = zonegroup.placement_targets.find(user_info.default_placement); | |
5930 | if (titer == zonegroup.placement_targets.end()) { | |
5931 | ldout(cct, 0) << "could not find user default placement id " << user_info.default_placement | |
5932 | << " within zonegroup " << dendl; | |
5933 | return -ERR_INVALID_LOCATION_CONSTRAINT; | |
5934 | } | |
5935 | } else { | |
5936 | if (zonegroup.default_placement.empty()) { // zonegroup default rule as fallback, it should not be empty. | |
5937 | ldout(cct, 0) << "misconfiguration, zonegroup default placement id should not be empty." << dendl; | |
5938 | return -ERR_ZONEGROUP_DEFAULT_PLACEMENT_MISCONFIGURATION; | |
5939 | } else { | |
5940 | titer = zonegroup.placement_targets.find(zonegroup.default_placement); | |
5941 | if (titer == zonegroup.placement_targets.end()) { | |
5942 | ldout(cct, 0) << "could not find zonegroup default placement id " << zonegroup.default_placement | |
5943 | << " within zonegroup " << dendl; | |
5944 | return -ERR_INVALID_LOCATION_CONSTRAINT; | |
5945 | } | |
5946 | } | |
7c673cae FG |
5947 | } |
5948 | ||
5949 | /* now check tag for the rule, whether user is permitted to use rule */ | |
c07f9fc5 | 5950 | const auto& target_rule = titer->second; |
7c673cae | 5951 | if (!target_rule.user_permitted(user_info.placement_tags)) { |
c07f9fc5 | 5952 | ldout(cct, 0) << "user not permitted to use placement rule " << titer->first << dendl; |
7c673cae FG |
5953 | return -EPERM; |
5954 | } | |
5955 | ||
5956 | if (pselected_rule_name) | |
c07f9fc5 | 5957 | *pselected_rule_name = titer->first; |
7c673cae | 5958 | |
c07f9fc5 | 5959 | return select_bucket_location_by_rule(titer->first, rule_info); |
7c673cae FG |
5960 | } |
5961 | ||
5962 | int RGWRados::select_bucket_location_by_rule(const string& location_rule, RGWZonePlacementInfo *rule_info) | |
5963 | { | |
5964 | if (location_rule.empty()) { | |
5965 | /* we can only reach here if we're trying to set a bucket location from a bucket | |
5966 | * created on a different zone, using a legacy / default pool configuration | |
5967 | */ | |
5968 | return select_legacy_bucket_placement(rule_info); | |
5969 | } | |
5970 | ||
5971 | /* | |
5972 | * make sure that zone has this rule configured. We're | |
5973 | * checking it for the local zone, because that's where this bucket object is going to | |
5974 | * reside. | |
5975 | */ | |
5976 | map<string, RGWZonePlacementInfo>::iterator piter = get_zone_params().placement_pools.find(location_rule); | |
5977 | if (piter == get_zone_params().placement_pools.end()) { | |
5978 | /* couldn't find, means we cannot really place data for this bucket in this zone */ | |
224ce89b | 5979 | if (get_zonegroup().equals(zonegroup.get_id())) { |
7c673cae FG |
5980 | /* that's a configuration error, zone should have that rule, as we're within the requested |
5981 | * zonegroup */ | |
5982 | return -EINVAL; | |
5983 | } else { | |
5984 | /* oh, well, data is not going to be placed here, bucket object is just a placeholder */ | |
5985 | return 0; | |
5986 | } | |
5987 | } | |
5988 | ||
5989 | RGWZonePlacementInfo& placement_info = piter->second; | |
5990 | ||
5991 | if (rule_info) { | |
5992 | *rule_info = placement_info; | |
5993 | } | |
5994 | ||
5995 | return 0; | |
5996 | } | |
5997 | ||
5998 | int RGWRados::select_bucket_placement(RGWUserInfo& user_info, const string& zonegroup_id, const string& placement_rule, | |
5999 | string *pselected_rule_name, RGWZonePlacementInfo *rule_info) | |
6000 | { | |
6001 | if (!get_zone_params().placement_pools.empty()) { | |
6002 | return select_new_bucket_location(user_info, zonegroup_id, placement_rule, | |
6003 | pselected_rule_name, rule_info); | |
6004 | } | |
6005 | ||
6006 | if (pselected_rule_name) { | |
6007 | pselected_rule_name->clear(); | |
6008 | } | |
6009 | ||
6010 | return select_legacy_bucket_placement(rule_info); | |
6011 | } | |
6012 | ||
6013 | int RGWRados::select_legacy_bucket_placement(RGWZonePlacementInfo *rule_info) | |
6014 | { | |
6015 | bufferlist map_bl; | |
6016 | map<string, bufferlist> m; | |
6017 | string pool_name; | |
6018 | bool write_map = false; | |
6019 | ||
6020 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6021 | ||
6022 | RGWObjectCtx obj_ctx(this); | |
6023 | int ret = rgw_get_system_obj(this, obj_ctx, get_zone_params().domain_root, avail_pools, map_bl, NULL, NULL); | |
6024 | if (ret < 0) { | |
6025 | goto read_omap; | |
6026 | } | |
6027 | ||
6028 | try { | |
6029 | bufferlist::iterator iter = map_bl.begin(); | |
6030 | ::decode(m, iter); | |
6031 | } catch (buffer::error& err) { | |
6032 | ldout(cct, 0) << "ERROR: couldn't decode avail_pools" << dendl; | |
6033 | } | |
6034 | ||
6035 | read_omap: | |
6036 | if (m.empty()) { | |
6037 | bufferlist header; | |
6038 | ret = omap_get_all(obj, header, m); | |
6039 | ||
6040 | write_map = true; | |
6041 | } | |
6042 | ||
6043 | if (ret < 0 || m.empty()) { | |
6044 | vector<rgw_pool> pools; | |
6045 | string s = string("default.") + default_storage_pool_suffix; | |
6046 | pools.push_back(rgw_pool(s)); | |
6047 | vector<int> retcodes; | |
6048 | bufferlist bl; | |
6049 | ret = create_pools(pools, retcodes); | |
6050 | if (ret < 0) | |
6051 | return ret; | |
6052 | ret = omap_set(obj, s, bl); | |
6053 | if (ret < 0) | |
6054 | return ret; | |
6055 | m[s] = bl; | |
6056 | } | |
6057 | ||
6058 | if (write_map) { | |
6059 | bufferlist new_bl; | |
6060 | ::encode(m, new_bl); | |
6061 | ret = put_system_obj_data(NULL, obj, new_bl, -1, false); | |
6062 | if (ret < 0) { | |
6063 | ldout(cct, 0) << "WARNING: could not save avail pools map info ret=" << ret << dendl; | |
6064 | } | |
6065 | } | |
6066 | ||
6067 | map<string, bufferlist>::iterator miter; | |
6068 | if (m.size() > 1) { | |
6069 | vector<string> v; | |
6070 | for (miter = m.begin(); miter != m.end(); ++miter) { | |
6071 | v.push_back(miter->first); | |
6072 | } | |
6073 | ||
6074 | uint32_t r; | |
6075 | ret = get_random_bytes((char *)&r, sizeof(r)); | |
6076 | if (ret < 0) | |
6077 | return ret; | |
6078 | ||
6079 | int i = r % v.size(); | |
6080 | pool_name = v[i]; | |
6081 | } else { | |
6082 | miter = m.begin(); | |
6083 | pool_name = miter->first; | |
6084 | } | |
6085 | ||
6086 | rule_info->data_pool = pool_name; | |
6087 | rule_info->data_extra_pool = pool_name; | |
6088 | rule_info->index_pool = pool_name; | |
6089 | rule_info->index_type = RGWBIType_Normal; | |
6090 | ||
6091 | return 0; | |
6092 | } | |
6093 | ||
6094 | bool RGWRados::get_obj_data_pool(const string& placement_rule, const rgw_obj& obj, rgw_pool *pool) | |
6095 | { | |
6096 | return rgw_get_obj_data_pool(zonegroup, zone_params, placement_rule, obj, pool); | |
6097 | } | |
6098 | ||
6099 | bool RGWRados::obj_to_raw(const string& placement_rule, const rgw_obj& obj, rgw_raw_obj *raw_obj) | |
6100 | { | |
6101 | get_obj_bucket_and_oid_loc(obj, raw_obj->oid, raw_obj->loc); | |
6102 | ||
6103 | return get_obj_data_pool(placement_rule, obj, &raw_obj->pool); | |
6104 | } | |
6105 | ||
6106 | int RGWRados::update_placement_map() | |
6107 | { | |
6108 | bufferlist header; | |
6109 | map<string, bufferlist> m; | |
6110 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6111 | int ret = omap_get_all(obj, header, m); | |
6112 | if (ret < 0) | |
6113 | return ret; | |
6114 | ||
6115 | bufferlist new_bl; | |
6116 | ::encode(m, new_bl); | |
6117 | ret = put_system_obj_data(NULL, obj, new_bl, -1, false); | |
6118 | if (ret < 0) { | |
6119 | ldout(cct, 0) << "WARNING: could not save avail pools map info ret=" << ret << dendl; | |
6120 | } | |
6121 | ||
6122 | return ret; | |
6123 | } | |
6124 | ||
6125 | int RGWRados::add_bucket_placement(const rgw_pool& new_pool) | |
6126 | { | |
6127 | librados::Rados *rad = get_rados_handle(); | |
6128 | int ret = rad->pool_lookup(new_pool.name.c_str()); | |
6129 | if (ret < 0) // DNE, or something | |
6130 | return ret; | |
6131 | ||
6132 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6133 | bufferlist empty_bl; | |
6134 | ret = omap_set(obj, new_pool.to_str(), empty_bl); | |
6135 | ||
6136 | // don't care about return value | |
6137 | update_placement_map(); | |
6138 | ||
6139 | return ret; | |
6140 | } | |
6141 | ||
6142 | int RGWRados::remove_bucket_placement(const rgw_pool& old_pool) | |
6143 | { | |
6144 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6145 | int ret = omap_del(obj, old_pool.to_str()); | |
6146 | ||
6147 | // don't care about return value | |
6148 | update_placement_map(); | |
6149 | ||
6150 | return ret; | |
6151 | } | |
6152 | ||
6153 | int RGWRados::list_placement_set(set<rgw_pool>& names) | |
6154 | { | |
6155 | bufferlist header; | |
6156 | map<string, bufferlist> m; | |
6157 | ||
6158 | rgw_raw_obj obj(get_zone_params().domain_root, avail_pools); | |
6159 | int ret = omap_get_all(obj, header, m); | |
6160 | if (ret < 0) | |
6161 | return ret; | |
6162 | ||
6163 | names.clear(); | |
6164 | map<string, bufferlist>::iterator miter; | |
6165 | for (miter = m.begin(); miter != m.end(); ++miter) { | |
6166 | names.insert(rgw_pool(miter->first)); | |
6167 | } | |
6168 | ||
6169 | return names.size(); | |
6170 | } | |
6171 | ||
6172 | int RGWRados::create_pools(vector<rgw_pool>& pools, vector<int>& retcodes) | |
6173 | { | |
6174 | vector<librados::PoolAsyncCompletion *> completions; | |
6175 | vector<int> rets; | |
6176 | ||
6177 | librados::Rados *rad = get_rados_handle(); | |
6178 | for (auto iter = pools.begin(); iter != pools.end(); ++iter) { | |
6179 | librados::PoolAsyncCompletion *c = librados::Rados::pool_async_create_completion(); | |
6180 | completions.push_back(c); | |
6181 | rgw_pool& pool = *iter; | |
6182 | int ret = rad->pool_create_async(pool.name.c_str(), c); | |
6183 | rets.push_back(ret); | |
6184 | } | |
6185 | ||
6186 | vector<int>::iterator riter; | |
6187 | vector<librados::PoolAsyncCompletion *>::iterator citer; | |
6188 | ||
c07f9fc5 | 6189 | bool error = false; |
7c673cae FG |
6190 | assert(rets.size() == completions.size()); |
6191 | for (riter = rets.begin(), citer = completions.begin(); riter != rets.end(); ++riter, ++citer) { | |
6192 | int r = *riter; | |
6193 | PoolAsyncCompletion *c = *citer; | |
6194 | if (r == 0) { | |
6195 | c->wait(); | |
6196 | r = c->get_return_value(); | |
6197 | if (r < 0) { | |
6198 | ldout(cct, 0) << "WARNING: async pool_create returned " << r << dendl; | |
c07f9fc5 | 6199 | error = true; |
7c673cae FG |
6200 | } |
6201 | } | |
6202 | c->release(); | |
6203 | retcodes.push_back(r); | |
6204 | } | |
c07f9fc5 FG |
6205 | if (error) { |
6206 | return 0; | |
6207 | } | |
6208 | ||
6209 | std::vector<librados::IoCtx> io_ctxs; | |
6210 | retcodes.clear(); | |
6211 | for (auto pool : pools) { | |
6212 | io_ctxs.emplace_back(); | |
6213 | int ret = rad->ioctx_create(pool.name.c_str(), io_ctxs.back()); | |
6214 | if (ret < 0) { | |
6215 | ldout(cct, 0) << "WARNING: ioctx_create returned " << ret << dendl; | |
6216 | error = true; | |
6217 | } | |
6218 | retcodes.push_back(ret); | |
6219 | } | |
6220 | if (error) { | |
6221 | return 0; | |
6222 | } | |
6223 | ||
6224 | completions.clear(); | |
6225 | for (auto &io_ctx : io_ctxs) { | |
6226 | librados::PoolAsyncCompletion *c = | |
6227 | librados::Rados::pool_async_create_completion(); | |
6228 | completions.push_back(c); | |
6229 | int ret = io_ctx.application_enable_async(pg_pool_t::APPLICATION_NAME_RGW, | |
6230 | false, c); | |
6231 | assert(ret == 0); | |
6232 | } | |
6233 | ||
6234 | retcodes.clear(); | |
6235 | for (auto c : completions) { | |
6236 | c->wait(); | |
6237 | int ret = c->get_return_value(); | |
6238 | if (ret == -EOPNOTSUPP) { | |
6239 | ret = 0; | |
6240 | } else if (ret < 0) { | |
6241 | ldout(cct, 0) << "WARNING: async application_enable returned " << ret | |
6242 | << dendl; | |
6243 | error = true; | |
6244 | } | |
6245 | c->release(); | |
6246 | retcodes.push_back(ret); | |
6247 | } | |
7c673cae FG |
6248 | return 0; |
6249 | } | |
6250 | ||
6251 | int RGWRados::get_obj_head_ioctx(const RGWBucketInfo& bucket_info, const rgw_obj& obj, librados::IoCtx *ioctx) | |
6252 | { | |
6253 | string oid, key; | |
6254 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
6255 | ||
6256 | rgw_pool pool; | |
6257 | if (!get_obj_data_pool(bucket_info.placement_rule, obj, &pool)) { | |
6258 | ldout(cct, 0) << "ERROR: cannot get data pool for obj=" << obj << ", probably misconfiguration" << dendl; | |
6259 | return -EIO; | |
6260 | } | |
6261 | ||
6262 | int r = open_pool_ctx(pool, *ioctx); | |
6263 | if (r < 0) { | |
6264 | return r; | |
6265 | } | |
6266 | ||
6267 | ioctx->locator_set_key(key); | |
6268 | ||
6269 | return 0; | |
6270 | } | |
6271 | ||
6272 | int RGWRados::get_obj_head_ref(const RGWBucketInfo& bucket_info, const rgw_obj& obj, rgw_rados_ref *ref) | |
6273 | { | |
6274 | get_obj_bucket_and_oid_loc(obj, ref->oid, ref->key); | |
6275 | ||
6276 | rgw_pool pool; | |
6277 | if (!get_obj_data_pool(bucket_info.placement_rule, obj, &pool)) { | |
6278 | ldout(cct, 0) << "ERROR: cannot get data pool for obj=" << obj << ", probably misconfiguration" << dendl; | |
6279 | return -EIO; | |
6280 | } | |
6281 | ||
6282 | int r = open_pool_ctx(pool, ref->ioctx); | |
6283 | if (r < 0) { | |
6284 | return r; | |
6285 | } | |
6286 | ||
6287 | ref->ioctx.locator_set_key(ref->key); | |
6288 | ||
6289 | return 0; | |
6290 | } | |
6291 | ||
224ce89b | 6292 | int RGWRados::get_raw_obj_ref(const rgw_raw_obj& obj, rgw_rados_ref *ref) |
7c673cae FG |
6293 | { |
6294 | ref->oid = obj.oid; | |
6295 | ref->key = obj.loc; | |
6296 | ||
6297 | int r; | |
6298 | ||
6299 | if (ref->oid.empty()) { | |
6300 | ref->oid = obj.pool.to_str(); | |
6301 | ref->pool = get_zone_params().domain_root; | |
6302 | } else { | |
6303 | ref->pool = obj.pool; | |
6304 | } | |
7c673cae FG |
6305 | r = open_pool_ctx(ref->pool, ref->ioctx); |
6306 | if (r < 0) | |
6307 | return r; | |
6308 | ||
6309 | ref->ioctx.locator_set_key(ref->key); | |
6310 | ||
6311 | return 0; | |
6312 | } | |
6313 | ||
224ce89b | 6314 | int RGWRados::get_system_obj_ref(const rgw_raw_obj& obj, rgw_rados_ref *ref) |
7c673cae | 6315 | { |
224ce89b | 6316 | return get_raw_obj_ref(obj, ref); |
7c673cae FG |
6317 | } |
6318 | ||
6319 | /* | |
6320 | * fixes an issue where head objects were supposed to have a locator created, but ended | |
6321 | * up without one | |
6322 | */ | |
6323 | int RGWRados::fix_head_obj_locator(const RGWBucketInfo& bucket_info, bool copy_obj, bool remove_bad, rgw_obj_key& key) | |
6324 | { | |
6325 | const rgw_bucket& bucket = bucket_info.bucket; | |
6326 | string oid; | |
6327 | string locator; | |
6328 | ||
6329 | rgw_obj obj(bucket, key); | |
6330 | ||
6331 | get_obj_bucket_and_oid_loc(obj, oid, locator); | |
6332 | ||
6333 | if (locator.empty()) { | |
6334 | ldout(cct, 20) << "object does not have a locator, nothing to fix" << dendl; | |
6335 | return 0; | |
6336 | } | |
6337 | ||
6338 | librados::IoCtx ioctx; | |
6339 | ||
6340 | int ret = get_obj_head_ioctx(bucket_info, obj, &ioctx); | |
6341 | if (ret < 0) { | |
6342 | cerr << "ERROR: get_obj_head_ioctx() returned ret=" << ret << std::endl; | |
6343 | return ret; | |
6344 | } | |
6345 | ioctx.locator_set_key(string()); /* override locator for this object, use empty locator */ | |
6346 | ||
6347 | uint64_t size; | |
6348 | bufferlist data; | |
6349 | ||
6350 | struct timespec mtime_ts; | |
6351 | map<string, bufferlist> attrs; | |
6352 | librados::ObjectReadOperation op; | |
6353 | op.getxattrs(&attrs, NULL); | |
6354 | op.stat2(&size, &mtime_ts, NULL); | |
6355 | #define HEAD_SIZE 512 * 1024 | |
6356 | op.read(0, HEAD_SIZE, &data, NULL); | |
6357 | ||
6358 | ret = ioctx.operate(oid, &op, NULL); | |
6359 | if (ret < 0) { | |
6360 | lderr(cct) << "ERROR: ioctx.operate(oid=" << oid << ") returned ret=" << ret << dendl; | |
6361 | return ret; | |
6362 | } | |
6363 | ||
6364 | if (size > HEAD_SIZE) { | |
6365 | lderr(cct) << "ERROR: returned object size (" << size << ") > HEAD_SIZE (" << HEAD_SIZE << ")" << dendl; | |
6366 | return -EIO; | |
6367 | } | |
6368 | ||
6369 | if (size != data.length()) { | |
6370 | lderr(cct) << "ERROR: returned object size (" << size << ") != data.length() (" << data.length() << ")" << dendl; | |
6371 | return -EIO; | |
6372 | } | |
6373 | ||
6374 | if (copy_obj) { | |
6375 | librados::ObjectWriteOperation wop; | |
6376 | ||
6377 | wop.mtime2(&mtime_ts); | |
6378 | ||
6379 | map<string, bufferlist>::iterator iter; | |
6380 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
6381 | wop.setxattr(iter->first.c_str(), iter->second); | |
6382 | } | |
6383 | ||
6384 | wop.write(0, data); | |
6385 | ||
6386 | ioctx.locator_set_key(locator); | |
6387 | ioctx.operate(oid, &wop); | |
6388 | } | |
6389 | ||
6390 | if (remove_bad) { | |
6391 | ioctx.locator_set_key(string()); | |
6392 | ||
6393 | ret = ioctx.remove(oid); | |
6394 | if (ret < 0) { | |
6395 | lderr(cct) << "ERROR: failed to remove original bad object" << dendl; | |
6396 | return ret; | |
6397 | } | |
6398 | } | |
6399 | ||
6400 | return 0; | |
6401 | } | |
6402 | ||
6403 | int RGWRados::move_rados_obj(librados::IoCtx& src_ioctx, | |
6404 | const string& src_oid, const string& src_locator, | |
6405 | librados::IoCtx& dst_ioctx, | |
6406 | const string& dst_oid, const string& dst_locator) | |
6407 | { | |
6408 | ||
6409 | #define COPY_BUF_SIZE (4 * 1024 * 1024) | |
6410 | bool done = false; | |
6411 | uint64_t chunk_size = COPY_BUF_SIZE; | |
6412 | uint64_t ofs = 0; | |
6413 | int ret = 0; | |
6414 | real_time mtime; | |
6415 | struct timespec mtime_ts; | |
6416 | uint64_t size; | |
6417 | ||
6418 | if (src_oid == dst_oid && src_locator == dst_locator) { | |
6419 | return 0; | |
6420 | } | |
6421 | ||
6422 | src_ioctx.locator_set_key(src_locator); | |
6423 | dst_ioctx.locator_set_key(dst_locator); | |
6424 | ||
6425 | do { | |
6426 | bufferlist data; | |
6427 | ObjectReadOperation rop; | |
6428 | ObjectWriteOperation wop; | |
6429 | ||
6430 | if (ofs == 0) { | |
6431 | rop.stat2(&size, &mtime_ts, NULL); | |
6432 | mtime = real_clock::from_timespec(mtime_ts); | |
6433 | } | |
6434 | rop.read(ofs, chunk_size, &data, NULL); | |
6435 | ret = src_ioctx.operate(src_oid, &rop, NULL); | |
6436 | if (ret < 0) { | |
6437 | goto done_err; | |
6438 | } | |
6439 | ||
6440 | if (data.length() == 0) { | |
6441 | break; | |
6442 | } | |
6443 | ||
6444 | if (ofs == 0) { | |
6445 | wop.create(true); /* make it exclusive */ | |
6446 | wop.mtime2(&mtime_ts); | |
6447 | mtime = real_clock::from_timespec(mtime_ts); | |
6448 | } | |
6449 | wop.write(ofs, data); | |
6450 | ret = dst_ioctx.operate(dst_oid, &wop); | |
6451 | ofs += data.length(); | |
6452 | done = data.length() != chunk_size; | |
6453 | } while (!done); | |
6454 | ||
6455 | if (ofs != size) { | |
6456 | lderr(cct) << "ERROR: " << __func__ << ": copying " << src_oid << " -> " << dst_oid | |
6457 | << ": expected " << size << " bytes to copy, ended up with " << ofs << dendl; | |
6458 | ret = -EIO; | |
6459 | goto done_err; | |
6460 | } | |
6461 | ||
6462 | src_ioctx.remove(src_oid); | |
6463 | ||
6464 | return 0; | |
6465 | ||
6466 | done_err: | |
6467 | lderr(cct) << "ERROR: failed to copy " << src_oid << " -> " << dst_oid << dendl; | |
6468 | return ret; | |
6469 | } | |
6470 | ||
6471 | /* | |
6472 | * fixes an issue where head objects were supposed to have a locator created, but ended | |
6473 | * up without one | |
6474 | */ | |
6475 | int RGWRados::fix_tail_obj_locator(const RGWBucketInfo& bucket_info, rgw_obj_key& key, bool fix, bool *need_fix) | |
6476 | { | |
6477 | const rgw_bucket& bucket = bucket_info.bucket; | |
6478 | rgw_obj obj(bucket, key); | |
6479 | ||
6480 | if (need_fix) { | |
6481 | *need_fix = false; | |
6482 | } | |
6483 | ||
6484 | rgw_rados_ref ref; | |
6485 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
6486 | if (r < 0) { | |
6487 | return r; | |
6488 | } | |
6489 | ||
6490 | RGWObjState *astate = NULL; | |
6491 | RGWObjectCtx rctx(this); | |
6492 | r = get_obj_state(&rctx, bucket_info, obj, &astate, false); | |
6493 | if (r < 0) | |
6494 | return r; | |
6495 | ||
6496 | if (astate->has_manifest) { | |
6497 | RGWObjManifest::obj_iterator miter; | |
6498 | RGWObjManifest& manifest = astate->manifest; | |
6499 | for (miter = manifest.obj_begin(); miter != manifest.obj_end(); ++miter) { | |
6500 | rgw_raw_obj raw_loc = miter.get_location().get_raw_obj(this); | |
6501 | rgw_obj loc; | |
6502 | string oid; | |
6503 | string locator; | |
6504 | ||
6505 | rgw_raw_obj_to_obj(manifest.get_tail_placement().bucket, raw_loc, &loc); | |
6506 | ||
6507 | if (loc.key.ns.empty()) { | |
6508 | /* continue, we're only interested in tail objects */ | |
6509 | continue; | |
6510 | } | |
6511 | ||
6512 | get_obj_bucket_and_oid_loc(loc, oid, locator); | |
6513 | ref.ioctx.locator_set_key(locator); | |
6514 | ||
6515 | ldout(cct, 20) << __func__ << ": key=" << key << " oid=" << oid << " locator=" << locator << dendl; | |
6516 | ||
6517 | r = ref.ioctx.stat(oid, NULL, NULL); | |
6518 | if (r != -ENOENT) { | |
6519 | continue; | |
6520 | } | |
6521 | ||
6522 | string bad_loc; | |
6523 | prepend_bucket_marker(bucket, loc.key.name, bad_loc); | |
6524 | ||
6525 | /* create a new ioctx with the bad locator */ | |
6526 | librados::IoCtx src_ioctx; | |
6527 | src_ioctx.dup(ref.ioctx); | |
6528 | src_ioctx.locator_set_key(bad_loc); | |
6529 | ||
6530 | r = src_ioctx.stat(oid, NULL, NULL); | |
6531 | if (r != 0) { | |
6532 | /* cannot find a broken part */ | |
6533 | continue; | |
6534 | } | |
6535 | ldout(cct, 20) << __func__ << ": found bad object part: " << loc << dendl; | |
6536 | if (need_fix) { | |
6537 | *need_fix = true; | |
6538 | } | |
6539 | if (fix) { | |
6540 | r = move_rados_obj(src_ioctx, oid, bad_loc, ref.ioctx, oid, locator); | |
6541 | if (r < 0) { | |
6542 | lderr(cct) << "ERROR: copy_rados_obj() on oid=" << oid << " returned r=" << r << dendl; | |
6543 | } | |
6544 | } | |
6545 | } | |
6546 | } | |
6547 | ||
6548 | return 0; | |
6549 | } | |
6550 | ||
6551 | int RGWRados::BucketShard::init(const rgw_bucket& _bucket, const rgw_obj& obj) | |
6552 | { | |
6553 | bucket = _bucket; | |
6554 | ||
6555 | RGWObjectCtx obj_ctx(store); | |
6556 | ||
6557 | RGWBucketInfo bucket_info; | |
6558 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL); | |
6559 | if (ret < 0) { | |
6560 | return ret; | |
6561 | } | |
6562 | ||
6563 | ret = store->open_bucket_index_shard(bucket_info, index_ctx, obj.get_hash_object(), &bucket_obj, &shard_id); | |
6564 | if (ret < 0) { | |
6565 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
6566 | return ret; | |
6567 | } | |
6568 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
6569 | ||
6570 | return 0; | |
6571 | } | |
6572 | ||
6573 | int RGWRados::BucketShard::init(const rgw_bucket& _bucket, int sid) | |
6574 | { | |
6575 | bucket = _bucket; | |
6576 | shard_id = sid; | |
6577 | ||
6578 | RGWObjectCtx obj_ctx(store); | |
6579 | ||
6580 | RGWBucketInfo bucket_info; | |
6581 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL); | |
6582 | if (ret < 0) { | |
6583 | return ret; | |
6584 | } | |
6585 | ||
6586 | ret = store->open_bucket_index_shard(bucket_info, index_ctx, shard_id, &bucket_obj); | |
6587 | if (ret < 0) { | |
6588 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
6589 | return ret; | |
6590 | } | |
6591 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
6592 | ||
6593 | return 0; | |
6594 | } | |
6595 | ||
b32b8144 FG |
6596 | int RGWRados::BucketShard::init(const RGWBucketInfo& bucket_info, int sid) |
6597 | { | |
6598 | bucket = bucket_info.bucket; | |
6599 | shard_id = sid; | |
6600 | ||
6601 | int ret = store->open_bucket_index_shard(bucket_info, index_ctx, shard_id, &bucket_obj); | |
6602 | if (ret < 0) { | |
6603 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
6604 | return ret; | |
6605 | } | |
6606 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
6607 | ||
6608 | return 0; | |
6609 | } | |
6610 | ||
7c673cae FG |
6611 | |
6612 | /* Execute @handler on last item in bucket listing for bucket specified | |
6613 | * in @bucket_info. @obj_prefix and @obj_delim narrow down the listing | |
6614 | * to objects matching these criterias. */ | |
6615 | int RGWRados::on_last_entry_in_listing(RGWBucketInfo& bucket_info, | |
6616 | const std::string& obj_prefix, | |
6617 | const std::string& obj_delim, | |
6618 | std::function<int(const rgw_bucket_dir_entry&)> handler) | |
6619 | { | |
6620 | RGWRados::Bucket target(this, bucket_info); | |
6621 | RGWRados::Bucket::List list_op(&target); | |
6622 | ||
6623 | list_op.params.prefix = obj_prefix; | |
6624 | list_op.params.delim = obj_delim; | |
6625 | ||
6626 | ldout(cct, 20) << "iterating listing for bucket=" << bucket_info.bucket.name | |
6627 | << ", obj_prefix=" << obj_prefix | |
6628 | << ", obj_delim=" << obj_delim | |
6629 | << dendl; | |
6630 | ||
6631 | bool is_truncated = false; | |
6632 | ||
6633 | boost::optional<rgw_bucket_dir_entry> last_entry; | |
6634 | /* We need to rewind to the last object in a listing. */ | |
6635 | do { | |
6636 | /* List bucket entries in chunks. */ | |
6637 | static constexpr int MAX_LIST_OBJS = 100; | |
6638 | std::vector<rgw_bucket_dir_entry> entries(MAX_LIST_OBJS); | |
6639 | ||
6640 | int ret = list_op.list_objects(MAX_LIST_OBJS, &entries, nullptr, | |
6641 | &is_truncated); | |
6642 | if (ret < 0) { | |
6643 | return ret; | |
6644 | } else if (!entries.empty()) { | |
6645 | last_entry = entries.back(); | |
6646 | } | |
6647 | } while (is_truncated); | |
6648 | ||
6649 | if (last_entry) { | |
6650 | return handler(*last_entry); | |
6651 | } | |
6652 | ||
6653 | /* Empty listing - no items we can run handler on. */ | |
6654 | return 0; | |
6655 | } | |
6656 | ||
6657 | ||
6658 | int RGWRados::swift_versioning_copy(RGWObjectCtx& obj_ctx, | |
6659 | const rgw_user& user, | |
6660 | RGWBucketInfo& bucket_info, | |
6661 | rgw_obj& obj) | |
6662 | { | |
6663 | if (! swift_versioning_enabled(bucket_info)) { | |
6664 | return 0; | |
6665 | } | |
6666 | ||
6667 | obj_ctx.obj.set_atomic(obj); | |
6668 | ||
6669 | RGWObjState * state = nullptr; | |
6670 | int r = get_obj_state(&obj_ctx, bucket_info, obj, &state, false); | |
6671 | if (r < 0) { | |
6672 | return r; | |
6673 | } | |
6674 | ||
6675 | if (!state->exists) { | |
6676 | return 0; | |
6677 | } | |
6678 | ||
6679 | string client_id; | |
6680 | string op_id; | |
6681 | ||
6682 | const string& src_name = obj.get_oid(); | |
6683 | char buf[src_name.size() + 32]; | |
6684 | struct timespec ts = ceph::real_clock::to_timespec(state->mtime); | |
6685 | snprintf(buf, sizeof(buf), "%03x%s/%lld.%06ld", (int)src_name.size(), | |
6686 | src_name.c_str(), (long long)ts.tv_sec, ts.tv_nsec / 1000); | |
6687 | ||
6688 | RGWBucketInfo dest_bucket_info; | |
6689 | ||
6690 | r = get_bucket_info(obj_ctx, bucket_info.bucket.tenant, bucket_info.swift_ver_location, dest_bucket_info, NULL, NULL); | |
6691 | if (r < 0) { | |
6692 | ldout(cct, 10) << "failed to read dest bucket info: r=" << r << dendl; | |
6693 | if (r == -ENOENT) { | |
6694 | return -ERR_PRECONDITION_FAILED; | |
6695 | } | |
6696 | return r; | |
6697 | } | |
6698 | ||
6699 | if (dest_bucket_info.owner != bucket_info.owner) { | |
6700 | return -ERR_PRECONDITION_FAILED; | |
6701 | } | |
6702 | ||
6703 | rgw_obj dest_obj(dest_bucket_info.bucket, buf); | |
6704 | obj_ctx.obj.set_atomic(dest_obj); | |
6705 | ||
6706 | string no_zone; | |
6707 | ||
6708 | r = copy_obj(obj_ctx, | |
6709 | user, | |
6710 | client_id, | |
6711 | op_id, | |
6712 | NULL, /* req_info *info */ | |
6713 | no_zone, | |
6714 | dest_obj, | |
6715 | obj, | |
6716 | dest_bucket_info, | |
6717 | bucket_info, | |
6718 | NULL, /* time_t *src_mtime */ | |
6719 | NULL, /* time_t *mtime */ | |
6720 | NULL, /* const time_t *mod_ptr */ | |
6721 | NULL, /* const time_t *unmod_ptr */ | |
6722 | false, /* bool high_precision_time */ | |
6723 | NULL, /* const char *if_match */ | |
6724 | NULL, /* const char *if_nomatch */ | |
6725 | RGWRados::ATTRSMOD_NONE, | |
6726 | true, /* bool copy_if_newer */ | |
6727 | state->attrset, | |
6728 | RGW_OBJ_CATEGORY_MAIN, | |
6729 | 0, /* uint64_t olh_epoch */ | |
6730 | real_time(), /* time_t delete_at */ | |
6731 | NULL, /* string *version_id */ | |
6732 | NULL, /* string *ptag */ | |
6733 | NULL, /* string *petag */ | |
7c673cae FG |
6734 | NULL, /* void (*progress_cb)(off_t, void *) */ |
6735 | NULL); /* void *progress_data */ | |
6736 | if (r == -ECANCELED || r == -ENOENT) { | |
6737 | /* Has already been overwritten, meaning another rgw process already | |
6738 | * copied it out */ | |
6739 | return 0; | |
6740 | } | |
6741 | ||
6742 | return r; | |
6743 | } | |
6744 | ||
6745 | int RGWRados::swift_versioning_restore(RGWObjectCtx& obj_ctx, | |
6746 | const rgw_user& user, | |
6747 | RGWBucketInfo& bucket_info, | |
6748 | rgw_obj& obj, | |
6749 | bool& restored) /* out */ | |
6750 | { | |
6751 | if (! swift_versioning_enabled(bucket_info)) { | |
6752 | return 0; | |
6753 | } | |
6754 | ||
6755 | /* Bucket info of the bucket that stores previous versions of our object. */ | |
6756 | RGWBucketInfo archive_binfo; | |
6757 | ||
6758 | int ret = get_bucket_info(obj_ctx, bucket_info.bucket.tenant, | |
6759 | bucket_info.swift_ver_location, archive_binfo, | |
6760 | nullptr, nullptr); | |
6761 | if (ret < 0) { | |
6762 | return ret; | |
6763 | } | |
6764 | ||
6765 | /* Abort the operation if the bucket storing our archive belongs to someone | |
6766 | * else. This is a limitation in comparison to Swift as we aren't taking ACLs | |
6767 | * into consideration. For we can live with that. | |
6768 | * | |
6769 | * TODO: delegate this check to un upper layer and compare with ACLs. */ | |
6770 | if (bucket_info.owner != archive_binfo.owner) { | |
6771 | return -EPERM; | |
6772 | } | |
6773 | ||
6774 | /* This code will be executed on latest version of the object. */ | |
6775 | const auto handler = [&](const rgw_bucket_dir_entry& entry) -> int { | |
6776 | std::string no_client_id; | |
6777 | std::string no_op_id; | |
6778 | std::string no_zone; | |
6779 | ||
6780 | /* We don't support object versioning of Swift API on those buckets that | |
6781 | * are already versioned using the S3 mechanism. This affects also bucket | |
6782 | * storing archived objects. Otherwise the delete operation would create | |
6783 | * a deletion marker. */ | |
6784 | if (archive_binfo.versioned()) { | |
6785 | restored = false; | |
6786 | return -ERR_PRECONDITION_FAILED; | |
6787 | } | |
6788 | ||
6789 | /* We are requesting ATTRSMOD_NONE so the attr attribute is perfectly | |
6790 | * irrelevant and may be safely skipped. */ | |
6791 | std::map<std::string, ceph::bufferlist> no_attrs; | |
6792 | ||
6793 | rgw_obj archive_obj(archive_binfo.bucket, entry.key); | |
6794 | obj_ctx.obj.set_atomic(archive_obj); | |
6795 | obj_ctx.obj.set_atomic(obj); | |
6796 | ||
6797 | int ret = copy_obj(obj_ctx, | |
6798 | user, | |
6799 | no_client_id, | |
6800 | no_op_id, | |
6801 | nullptr, /* req_info *info */ | |
6802 | no_zone, | |
6803 | obj, /* dest obj */ | |
6804 | archive_obj, /* src obj */ | |
6805 | bucket_info, /* dest bucket info */ | |
6806 | archive_binfo, /* src bucket info */ | |
6807 | nullptr, /* time_t *src_mtime */ | |
6808 | nullptr, /* time_t *mtime */ | |
6809 | nullptr, /* const time_t *mod_ptr */ | |
6810 | nullptr, /* const time_t *unmod_ptr */ | |
6811 | false, /* bool high_precision_time */ | |
6812 | nullptr, /* const char *if_match */ | |
6813 | nullptr, /* const char *if_nomatch */ | |
6814 | RGWRados::ATTRSMOD_NONE, | |
6815 | true, /* bool copy_if_newer */ | |
6816 | no_attrs, | |
6817 | RGW_OBJ_CATEGORY_MAIN, | |
6818 | 0, /* uint64_t olh_epoch */ | |
6819 | real_time(), /* time_t delete_at */ | |
6820 | nullptr, /* string *version_id */ | |
6821 | nullptr, /* string *ptag */ | |
6822 | nullptr, /* string *petag */ | |
7c673cae FG |
6823 | nullptr, /* void (*progress_cb)(off_t, void *) */ |
6824 | nullptr); /* void *progress_data */ | |
6825 | if (ret == -ECANCELED || ret == -ENOENT) { | |
6826 | /* Has already been overwritten, meaning another rgw process already | |
6827 | * copied it out */ | |
6828 | return 0; | |
6829 | } else if (ret < 0) { | |
6830 | return ret; | |
6831 | } else { | |
6832 | restored = true; | |
6833 | } | |
6834 | ||
6835 | /* Need to remove the archived copy. */ | |
6836 | ret = delete_obj(obj_ctx, archive_binfo, archive_obj, | |
6837 | archive_binfo.versioning_status()); | |
6838 | ||
6839 | return ret; | |
6840 | }; | |
6841 | ||
6842 | const std::string& obj_name = obj.get_oid(); | |
6843 | const auto prefix = boost::str(boost::format("%03x%s") % obj_name.size() | |
6844 | % obj_name); | |
6845 | ||
6846 | return on_last_entry_in_listing(archive_binfo, prefix, std::string(), | |
6847 | handler); | |
6848 | } | |
6849 | ||
6850 | /** | |
6851 | * Write/overwrite an object to the bucket storage. | |
6852 | * bucket: the bucket to store the object in | |
6853 | * obj: the object name/key | |
6854 | * data: the object contents/value | |
6855 | * size: the amount of data to write (data must be this long) | |
6856 | * accounted_size: original size of data before compression, encryption | |
6857 | * mtime: if non-NULL, writes the given mtime to the bucket storage | |
6858 | * attrs: all the given attrs are written to bucket storage for the given object | |
6859 | * exclusive: create object exclusively | |
6860 | * Returns: 0 on success, -ERR# otherwise. | |
6861 | */ | |
6862 | int RGWRados::Object::Write::_do_write_meta(uint64_t size, uint64_t accounted_size, | |
181888fb FG |
6863 | map<string, bufferlist>& attrs, |
6864 | bool assume_noent, bool modify_tail, | |
7c673cae FG |
6865 | void *_index_op) |
6866 | { | |
6867 | RGWRados::Bucket::UpdateIndex *index_op = static_cast<RGWRados::Bucket::UpdateIndex *>(_index_op); | |
7c673cae FG |
6868 | RGWRados *store = target->get_store(); |
6869 | ||
6870 | ObjectWriteOperation op; | |
6871 | ||
6872 | RGWObjState *state; | |
6873 | int r = target->get_state(&state, false, assume_noent); | |
6874 | if (r < 0) | |
6875 | return r; | |
6876 | ||
6877 | rgw_obj& obj = target->get_obj(); | |
6878 | ||
6879 | if (obj.get_oid().empty()) { | |
6880 | ldout(store->ctx(), 0) << "ERROR: " << __func__ << "(): cannot write object with empty name" << dendl; | |
6881 | return -EIO; | |
6882 | } | |
6883 | ||
224ce89b | 6884 | rgw_rados_ref ref; |
7c673cae FG |
6885 | r = store->get_obj_head_ref(target->get_bucket_info(), obj, &ref); |
6886 | if (r < 0) | |
6887 | return r; | |
6888 | ||
6889 | bool is_olh = state->is_olh; | |
6890 | ||
6891 | bool reset_obj = (meta.flags & PUT_OBJ_CREATE) != 0; | |
6892 | ||
6893 | const string *ptag = meta.ptag; | |
6894 | if (!ptag && !index_op->get_optag()->empty()) { | |
6895 | ptag = index_op->get_optag(); | |
6896 | } | |
181888fb | 6897 | r = target->prepare_atomic_modification(op, reset_obj, ptag, meta.if_match, meta.if_nomatch, false, modify_tail); |
7c673cae FG |
6898 | if (r < 0) |
6899 | return r; | |
6900 | ||
6901 | if (real_clock::is_zero(meta.set_mtime)) { | |
6902 | meta.set_mtime = real_clock::now(); | |
6903 | } | |
6904 | ||
6905 | if (state->is_olh) { | |
6906 | op.setxattr(RGW_ATTR_OLH_ID_TAG, state->olh_tag); | |
6907 | } | |
6908 | ||
6909 | struct timespec mtime_ts = real_clock::to_timespec(meta.set_mtime); | |
6910 | op.mtime2(&mtime_ts); | |
6911 | ||
6912 | if (meta.data) { | |
6913 | /* if we want to overwrite the data, we also want to overwrite the | |
6914 | xattrs, so just remove the object */ | |
6915 | op.write_full(*meta.data); | |
6916 | } | |
6917 | ||
6918 | string etag; | |
6919 | string content_type; | |
6920 | bufferlist acl_bl; | |
6921 | ||
6922 | map<string, bufferlist>::iterator iter; | |
6923 | if (meta.rmattrs) { | |
6924 | for (iter = meta.rmattrs->begin(); iter != meta.rmattrs->end(); ++iter) { | |
6925 | const string& name = iter->first; | |
6926 | op.rmxattr(name.c_str()); | |
6927 | } | |
6928 | } | |
6929 | ||
6930 | if (meta.manifest) { | |
6931 | /* remove existing manifest attr */ | |
6932 | iter = attrs.find(RGW_ATTR_MANIFEST); | |
6933 | if (iter != attrs.end()) | |
6934 | attrs.erase(iter); | |
6935 | ||
6936 | bufferlist bl; | |
6937 | ::encode(*meta.manifest, bl); | |
6938 | op.setxattr(RGW_ATTR_MANIFEST, bl); | |
6939 | } | |
6940 | ||
6941 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
6942 | const string& name = iter->first; | |
6943 | bufferlist& bl = iter->second; | |
6944 | ||
6945 | if (!bl.length()) | |
6946 | continue; | |
6947 | ||
6948 | op.setxattr(name.c_str(), bl); | |
6949 | ||
6950 | if (name.compare(RGW_ATTR_ETAG) == 0) { | |
6951 | etag = bl.c_str(); | |
6952 | } else if (name.compare(RGW_ATTR_CONTENT_TYPE) == 0) { | |
6953 | content_type = bl.c_str(); | |
6954 | } else if (name.compare(RGW_ATTR_ACL) == 0) { | |
6955 | acl_bl = bl; | |
6956 | } | |
6957 | } | |
6958 | if (attrs.find(RGW_ATTR_PG_VER) == attrs.end()) { | |
6959 | cls_rgw_obj_store_pg_ver(op, RGW_ATTR_PG_VER); | |
6960 | } | |
6961 | ||
6962 | if (attrs.find(RGW_ATTR_SOURCE_ZONE) == attrs.end()) { | |
6963 | bufferlist bl; | |
6964 | ::encode(store->get_zone_short_id(), bl); | |
6965 | op.setxattr(RGW_ATTR_SOURCE_ZONE, bl); | |
6966 | } | |
6967 | ||
6968 | if (!op.size()) | |
6969 | return 0; | |
6970 | ||
6971 | uint64_t epoch; | |
6972 | int64_t poolid; | |
224ce89b WB |
6973 | bool orig_exists; |
6974 | uint64_t orig_size; | |
6975 | ||
6976 | if (!reset_obj) { //Multipart upload, it has immutable head. | |
6977 | orig_exists = false; | |
6978 | orig_size = 0; | |
6979 | } else { | |
6980 | orig_exists = state->exists; | |
6981 | orig_size = state->accounted_size; | |
6982 | } | |
7c673cae FG |
6983 | |
6984 | bool versioned_target = (meta.olh_epoch > 0 || !obj.key.instance.empty()); | |
6985 | ||
6986 | bool versioned_op = (target->versioning_enabled() || is_olh || versioned_target); | |
6987 | ||
6988 | if (versioned_op) { | |
6989 | index_op->set_bilog_flags(RGW_BILOG_FLAG_VERSIONED_OP); | |
6990 | } | |
6991 | ||
6992 | if (!index_op->is_prepared()) { | |
6993 | r = index_op->prepare(CLS_RGW_OP_ADD, &state->write_tag); | |
6994 | if (r < 0) | |
6995 | return r; | |
6996 | } | |
6997 | ||
6998 | r = ref.ioctx.operate(ref.oid, &op); | |
6999 | if (r < 0) { /* we can expect to get -ECANCELED if object was replaced under, | |
7000 | or -ENOENT if was removed, or -EEXIST if it did not exist | |
7001 | before and now it does */ | |
7002 | if (r == -EEXIST && assume_noent) { | |
7003 | target->invalidate_state(); | |
7004 | return r; | |
7005 | } | |
7006 | goto done_cancel; | |
7007 | } | |
7008 | ||
7009 | epoch = ref.ioctx.get_last_version(); | |
7010 | poolid = ref.ioctx.get_id(); | |
7011 | ||
7012 | r = target->complete_atomic_modification(); | |
7013 | if (r < 0) { | |
7014 | ldout(store->ctx(), 0) << "ERROR: complete_atomic_modification returned r=" << r << dendl; | |
7015 | } | |
7016 | ||
7017 | r = index_op->complete(poolid, epoch, size, accounted_size, | |
7018 | meta.set_mtime, etag, content_type, &acl_bl, | |
7019 | meta.category, meta.remove_objs, meta.user_data); | |
7020 | if (r < 0) | |
7021 | goto done_cancel; | |
7022 | ||
7023 | if (meta.mtime) { | |
7024 | *meta.mtime = meta.set_mtime; | |
7025 | } | |
7026 | ||
7027 | /* note that index_op was using state so we couldn't invalidate it earlier */ | |
7028 | target->invalidate_state(); | |
7029 | state = NULL; | |
7030 | ||
7031 | if (versioned_op) { | |
31f18b77 | 7032 | r = store->set_olh(target->get_ctx(), target->get_bucket_info(), obj, false, NULL, meta.olh_epoch, real_time(), false, meta.zones_trace); |
7c673cae FG |
7033 | if (r < 0) { |
7034 | return r; | |
7035 | } | |
7036 | } | |
7037 | ||
7038 | if (!real_clock::is_zero(meta.delete_at)) { | |
7039 | rgw_obj_index_key obj_key; | |
7040 | obj.key.get_index_key(&obj_key); | |
7041 | ||
7042 | r = store->objexp_hint_add(meta.delete_at, | |
7043 | obj.bucket.tenant, obj.bucket.name, obj.bucket.bucket_id, obj_key); | |
7044 | if (r < 0) { | |
7045 | ldout(store->ctx(), 0) << "ERROR: objexp_hint_add() returned r=" << r << ", object will not get removed" << dendl; | |
7046 | /* ignoring error, nothing we can do at this point */ | |
7047 | } | |
7048 | } | |
7049 | meta.canceled = false; | |
7050 | ||
7051 | /* update quota cache */ | |
3efd9988 FG |
7052 | if (meta.completeMultipart){ |
7053 | store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1), | |
7054 | 0, orig_size); | |
7055 | } | |
7056 | else { | |
7057 | store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1), | |
7058 | accounted_size, orig_size); | |
7059 | } | |
7c673cae FG |
7060 | return 0; |
7061 | ||
7062 | done_cancel: | |
7063 | int ret = index_op->cancel(); | |
7064 | if (ret < 0) { | |
7065 | ldout(store->ctx(), 0) << "ERROR: index_op.cancel()() returned ret=" << ret << dendl; | |
7066 | } | |
7067 | ||
7068 | meta.canceled = true; | |
7069 | ||
7070 | /* we lost in a race. There are a few options: | |
7071 | * - existing object was rewritten (ECANCELED) | |
7072 | * - non existing object was created (EEXIST) | |
7073 | * - object was removed (ENOENT) | |
7074 | * should treat it as a success | |
7075 | */ | |
7076 | if (meta.if_match == NULL && meta.if_nomatch == NULL) { | |
7077 | if (r == -ECANCELED || r == -ENOENT || r == -EEXIST) { | |
7078 | r = 0; | |
7079 | } | |
7080 | } else { | |
7081 | if (meta.if_match != NULL) { | |
7082 | // only overwrite existing object | |
7083 | if (strcmp(meta.if_match, "*") == 0) { | |
7084 | if (r == -ENOENT) { | |
7085 | r = -ERR_PRECONDITION_FAILED; | |
7086 | } else if (r == -ECANCELED) { | |
7087 | r = 0; | |
7088 | } | |
7089 | } | |
7090 | } | |
7091 | ||
7092 | if (meta.if_nomatch != NULL) { | |
7093 | // only create a new object | |
7094 | if (strcmp(meta.if_nomatch, "*") == 0) { | |
7095 | if (r == -EEXIST) { | |
7096 | r = -ERR_PRECONDITION_FAILED; | |
7097 | } else if (r == -ENOENT) { | |
7098 | r = 0; | |
7099 | } | |
7100 | } | |
7101 | } | |
7102 | } | |
7103 | ||
7104 | return r; | |
7105 | } | |
7106 | ||
7107 | int RGWRados::Object::Write::write_meta(uint64_t size, uint64_t accounted_size, | |
7108 | map<string, bufferlist>& attrs) | |
7109 | { | |
7110 | RGWBucketInfo& bucket_info = target->get_bucket_info(); | |
7111 | ||
7112 | RGWRados::Bucket bop(target->get_store(), bucket_info); | |
7113 | RGWRados::Bucket::UpdateIndex index_op(&bop, target->get_obj()); | |
31f18b77 FG |
7114 | index_op.set_zones_trace(meta.zones_trace); |
7115 | ||
7c673cae FG |
7116 | bool assume_noent = (meta.if_match == NULL && meta.if_nomatch == NULL); |
7117 | int r; | |
7118 | if (assume_noent) { | |
181888fb | 7119 | r = _do_write_meta(size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op); |
7c673cae FG |
7120 | if (r == -EEXIST) { |
7121 | assume_noent = false; | |
7122 | } | |
7123 | } | |
7124 | if (!assume_noent) { | |
181888fb | 7125 | r = _do_write_meta(size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op); |
7c673cae FG |
7126 | } |
7127 | return r; | |
7128 | } | |
7129 | ||
7130 | /** Write/overwrite a system object. */ | |
7131 | int RGWRados::put_system_obj_impl(rgw_raw_obj& obj, uint64_t size, real_time *mtime, | |
7132 | map<std::string, bufferlist>& attrs, int flags, | |
7133 | bufferlist& data, | |
7134 | RGWObjVersionTracker *objv_tracker, | |
7135 | real_time set_mtime /* 0 for don't set */) | |
7136 | { | |
7c673cae | 7137 | rgw_rados_ref ref; |
224ce89b | 7138 | int r = get_system_obj_ref(obj, &ref); |
7c673cae FG |
7139 | if (r < 0) |
7140 | return r; | |
7141 | ||
7142 | ObjectWriteOperation op; | |
7143 | ||
7144 | if (flags & PUT_OBJ_EXCL) { | |
7145 | if (!(flags & PUT_OBJ_CREATE)) | |
7146 | return -EINVAL; | |
7147 | op.create(true); // exclusive create | |
7148 | } else { | |
7149 | op.remove(); | |
7150 | op.set_op_flags2(LIBRADOS_OP_FLAG_FAILOK); | |
7151 | op.create(false); | |
7152 | } | |
7153 | ||
7154 | if (objv_tracker) { | |
7155 | objv_tracker->prepare_op_for_write(&op); | |
7156 | } | |
7157 | ||
7158 | if (real_clock::is_zero(set_mtime)) { | |
7159 | set_mtime = real_clock::now(); | |
7160 | } | |
7161 | ||
7162 | struct timespec mtime_ts = real_clock::to_timespec(set_mtime); | |
7163 | op.mtime2(&mtime_ts); | |
7164 | op.write_full(data); | |
7165 | ||
7166 | bufferlist acl_bl; | |
7167 | ||
7168 | for (map<string, bufferlist>::iterator iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
7169 | const string& name = iter->first; | |
7170 | bufferlist& bl = iter->second; | |
7171 | ||
7172 | if (!bl.length()) | |
7173 | continue; | |
7174 | ||
7175 | op.setxattr(name.c_str(), bl); | |
7176 | } | |
7177 | ||
7178 | r = ref.ioctx.operate(ref.oid, &op); | |
7179 | if (r < 0) { | |
7180 | return r; | |
7181 | } | |
7182 | ||
7183 | if (objv_tracker) { | |
7184 | objv_tracker->apply_write(); | |
7185 | } | |
7186 | ||
7187 | if (mtime) { | |
7188 | *mtime = set_mtime; | |
7189 | } | |
7190 | ||
7191 | return 0; | |
7192 | } | |
7193 | ||
7194 | int RGWRados::put_system_obj_data(void *ctx, rgw_raw_obj& obj, bufferlist& bl, | |
7195 | off_t ofs, bool exclusive, | |
7196 | RGWObjVersionTracker *objv_tracker) | |
7197 | { | |
7198 | rgw_rados_ref ref; | |
224ce89b | 7199 | int r = get_system_obj_ref(obj, &ref); |
7c673cae FG |
7200 | if (r < 0) { |
7201 | return r; | |
7202 | } | |
7203 | ||
7204 | ObjectWriteOperation op; | |
7205 | ||
7206 | if (exclusive) | |
7207 | op.create(true); | |
7208 | ||
7209 | if (objv_tracker) { | |
7210 | objv_tracker->prepare_op_for_write(&op); | |
7211 | } | |
7212 | if (ofs == -1) { | |
7213 | op.write_full(bl); | |
7214 | } else { | |
7215 | op.write(ofs, bl); | |
7216 | } | |
7217 | r = ref.ioctx.operate(ref.oid, &op); | |
7218 | if (r < 0) | |
7219 | return r; | |
7220 | ||
7221 | if (objv_tracker) { | |
7222 | objv_tracker->apply_write(); | |
7223 | } | |
7224 | return 0; | |
7225 | } | |
7226 | ||
7227 | /** | |
7228 | * Write/overwrite an object to the bucket storage. | |
7229 | * bucket: the bucket to store the object in | |
7230 | * obj: the object name/key | |
7231 | * data: the object contents/value | |
7232 | * offset: the offet to write to in the object | |
7233 | * If this is -1, we will overwrite the whole object. | |
7234 | * size: the amount of data to write (data must be this long) | |
7235 | * attrs: all the given attrs are written to bucket storage for the given object | |
7236 | * Returns: 0 on success, -ERR# otherwise. | |
7237 | */ | |
7238 | ||
7239 | int RGWRados::aio_put_obj_data(void *ctx, rgw_raw_obj& obj, bufferlist& bl, | |
7240 | off_t ofs, bool exclusive, | |
7241 | void **handle) | |
7242 | { | |
7243 | rgw_rados_ref ref; | |
7244 | int r = get_raw_obj_ref(obj, &ref); | |
7245 | if (r < 0) { | |
7246 | return r; | |
7247 | } | |
7248 | ||
7249 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
7250 | *handle = c; | |
7251 | ||
7252 | ObjectWriteOperation op; | |
7253 | ||
7254 | if (exclusive) | |
7255 | op.create(true); | |
7256 | ||
7257 | if (ofs == -1) { | |
7258 | op.write_full(bl); | |
7259 | } else { | |
7260 | op.write(ofs, bl); | |
7261 | } | |
7262 | r = ref.ioctx.aio_operate(ref.oid, c, &op); | |
7263 | if (r < 0) | |
7264 | return r; | |
7265 | ||
7266 | return 0; | |
7267 | } | |
7268 | ||
7269 | int RGWRados::aio_wait(void *handle) | |
7270 | { | |
7271 | AioCompletion *c = (AioCompletion *)handle; | |
7272 | c->wait_for_safe(); | |
7273 | int ret = c->get_return_value(); | |
7274 | c->release(); | |
7275 | return ret; | |
7276 | } | |
7277 | ||
7278 | bool RGWRados::aio_completed(void *handle) | |
7279 | { | |
7280 | AioCompletion *c = (AioCompletion *)handle; | |
7281 | return c->is_safe(); | |
7282 | } | |
7283 | ||
28e407b8 AA |
7284 | // PutObj filter that buffers data so we don't try to compress tiny blocks. |
7285 | // libcurl reads in 16k at a time, and we need at least 64k to get a good | |
7286 | // compression ratio | |
7287 | class RGWPutObj_Buffer : public RGWPutObj_Filter { | |
7288 | const unsigned buffer_size; | |
7289 | bufferlist buffer; | |
7290 | public: | |
7291 | RGWPutObj_Buffer(RGWPutObjDataProcessor* next, unsigned buffer_size) | |
7292 | : RGWPutObj_Filter(next), buffer_size(buffer_size) { | |
7293 | assert(ISP2(buffer_size)); // must be power of 2 | |
7294 | } | |
7295 | ||
7296 | int handle_data(bufferlist& bl, off_t ofs, void **phandle, rgw_raw_obj *pobj, | |
7297 | bool *again) override { | |
7298 | if (*again || !bl.length()) { | |
7299 | // flush buffered data | |
7300 | return RGWPutObj_Filter::handle_data(buffer, ofs, phandle, pobj, again); | |
7301 | } | |
7302 | // transform offset to the beginning of the buffer | |
7303 | ofs = ofs - buffer.length(); | |
7304 | buffer.claim_append(bl); | |
7305 | if (buffer.length() < buffer_size) { | |
7306 | *again = false; // don't come back until there's more data | |
7307 | return 0; | |
7308 | } | |
7309 | const auto count = P2ALIGN(buffer.length(), buffer_size); | |
7310 | buffer.splice(0, count, &bl); | |
7311 | return RGWPutObj_Filter::handle_data(bl, ofs, phandle, pobj, again); | |
7312 | } | |
7313 | }; | |
7314 | ||
7c673cae FG |
7315 | class RGWRadosPutObj : public RGWGetDataCB |
7316 | { | |
7317 | CephContext* cct; | |
7318 | rgw_obj obj; | |
7319 | RGWPutObjDataProcessor *filter; | |
7320 | boost::optional<RGWPutObj_Compress>& compressor; | |
28e407b8 | 7321 | boost::optional<RGWPutObj_Buffer> buffering; |
7c673cae FG |
7322 | CompressorRef& plugin; |
7323 | RGWPutObjProcessor_Atomic *processor; | |
7324 | RGWOpStateSingleOp *opstate; | |
7325 | void (*progress_cb)(off_t, void *); | |
7326 | void *progress_data; | |
7327 | bufferlist extra_data_bl; | |
b32b8144 | 7328 | uint64_t extra_data_left; |
7c673cae FG |
7329 | uint64_t data_len; |
7330 | map<string, bufferlist> src_attrs; | |
7331 | public: | |
7332 | RGWRadosPutObj(CephContext* cct, | |
7333 | CompressorRef& plugin, | |
7334 | boost::optional<RGWPutObj_Compress>& compressor, | |
7335 | RGWPutObjProcessor_Atomic *p, | |
7336 | RGWOpStateSingleOp *_ops, | |
7337 | void (*_progress_cb)(off_t, void *), | |
7338 | void *_progress_data) : | |
7339 | cct(cct), | |
7340 | filter(p), | |
7341 | compressor(compressor), | |
7342 | plugin(plugin), | |
7343 | processor(p), | |
7344 | opstate(_ops), | |
7345 | progress_cb(_progress_cb), | |
7346 | progress_data(_progress_data), | |
b32b8144 | 7347 | extra_data_left(0), |
7c673cae FG |
7348 | data_len(0) {} |
7349 | ||
7350 | int process_attrs(void) { | |
7351 | if (extra_data_bl.length()) { | |
7352 | JSONParser jp; | |
7353 | if (!jp.parse(extra_data_bl.c_str(), extra_data_bl.length())) { | |
7354 | ldout(cct, 0) << "failed to parse response extra data. len=" << extra_data_bl.length() << " data=" << extra_data_bl.c_str() << dendl; | |
7355 | return -EIO; | |
7356 | } | |
7357 | ||
7358 | JSONDecoder::decode_json("attrs", src_attrs, &jp); | |
7359 | ||
7360 | src_attrs.erase(RGW_ATTR_COMPRESSION); | |
7361 | src_attrs.erase(RGW_ATTR_MANIFEST); // not interested in original object layout | |
7362 | } | |
7363 | ||
7364 | if (plugin && src_attrs.find(RGW_ATTR_CRYPT_MODE) == src_attrs.end()) { | |
7365 | //do not compress if object is encrypted | |
7366 | compressor = boost::in_place(cct, plugin, filter); | |
28e407b8 AA |
7367 | constexpr unsigned buffer_size = 512 * 1024; |
7368 | buffering = boost::in_place(&*compressor, buffer_size); | |
7369 | filter = &*buffering; | |
7c673cae FG |
7370 | } |
7371 | return 0; | |
7372 | } | |
7373 | ||
7374 | int handle_data(bufferlist& bl, off_t ofs, off_t len) override { | |
7375 | if (progress_cb) { | |
7376 | progress_cb(ofs, progress_data); | |
7377 | } | |
b32b8144 | 7378 | if (extra_data_left) { |
7c673cae | 7379 | size_t extra_len = bl.length(); |
b32b8144 FG |
7380 | if (extra_len > extra_data_left) |
7381 | extra_len = extra_data_left; | |
7c673cae FG |
7382 | |
7383 | bufferlist extra; | |
7384 | bl.splice(0, extra_len, &extra); | |
7385 | extra_data_bl.append(extra); | |
7386 | ||
b32b8144 FG |
7387 | extra_data_left -= extra_len; |
7388 | if (extra_data_left == 0) { | |
7c673cae FG |
7389 | int res = process_attrs(); |
7390 | if (res < 0) | |
7391 | return res; | |
7392 | } | |
7393 | if (bl.length() == 0) { | |
7394 | return 0; | |
7395 | } | |
b32b8144 | 7396 | ofs += extra_len; |
7c673cae | 7397 | } |
b32b8144 FG |
7398 | // adjust ofs based on extra_data_len, so the result is a logical offset |
7399 | // into the object data | |
7400 | assert(uint64_t(ofs) >= extra_data_len); | |
7401 | ofs -= extra_data_len; | |
7402 | ||
7c673cae FG |
7403 | data_len += bl.length(); |
7404 | bool again = false; | |
7405 | ||
7406 | bool need_opstate = true; | |
7407 | ||
7408 | do { | |
7409 | void *handle = NULL; | |
7410 | rgw_raw_obj obj; | |
7411 | uint64_t size = bl.length(); | |
7412 | int ret = filter->handle_data(bl, ofs, &handle, &obj, &again); | |
7413 | if (ret < 0) | |
7414 | return ret; | |
7415 | ||
7416 | if (need_opstate && opstate) { | |
7417 | /* need to update opstate repository with new state. This is ratelimited, so we're not | |
7418 | * really doing it every time | |
7419 | */ | |
7420 | ret = opstate->renew_state(); | |
7421 | if (ret < 0) { | |
7422 | ldout(cct, 0) << "ERROR: RGWRadosPutObj::handle_data(): failed to renew op state ret=" << ret << dendl; | |
7423 | int r = filter->throttle_data(handle, obj, size, false); | |
7424 | if (r < 0) { | |
7425 | ldout(cct, 0) << "ERROR: RGWRadosPutObj::handle_data(): processor->throttle_data() returned " << r << dendl; | |
7426 | } | |
7427 | /* could not renew state! might have been marked as cancelled */ | |
7428 | return ret; | |
7429 | } | |
7430 | need_opstate = false; | |
7431 | } | |
7432 | ||
7433 | ret = filter->throttle_data(handle, obj, size, false); | |
7434 | if (ret < 0) | |
7435 | return ret; | |
7436 | } while (again); | |
7437 | ||
7438 | return 0; | |
7439 | } | |
7440 | ||
28e407b8 AA |
7441 | int flush() { |
7442 | bufferlist bl; | |
7443 | return put_data_and_throttle(filter, bl, 0, false); | |
7444 | } | |
7445 | ||
7c673cae FG |
7446 | bufferlist& get_extra_data() { return extra_data_bl; } |
7447 | ||
7448 | map<string, bufferlist>& get_attrs() { return src_attrs; } | |
7449 | ||
7450 | void set_extra_data_len(uint64_t len) override { | |
b32b8144 FG |
7451 | extra_data_left = len; |
7452 | RGWGetDataCB::set_extra_data_len(len); | |
7c673cae FG |
7453 | } |
7454 | ||
7455 | uint64_t get_data_len() { | |
7456 | return data_len; | |
7457 | } | |
7458 | ||
7459 | int complete(const string& etag, real_time *mtime, real_time set_mtime, | |
31f18b77 FG |
7460 | map<string, bufferlist>& attrs, real_time delete_at, rgw_zone_set *zones_trace) { |
7461 | return processor->complete(data_len, etag, mtime, set_mtime, attrs, delete_at, NULL, NULL, NULL, zones_trace); | |
7c673cae FG |
7462 | } |
7463 | ||
7464 | bool is_canceled() { | |
7465 | return processor->is_canceled(); | |
7466 | } | |
7467 | }; | |
7468 | ||
7469 | /* | |
7470 | * prepare attrset depending on attrs_mod. | |
7471 | */ | |
7472 | static void set_copy_attrs(map<string, bufferlist>& src_attrs, | |
7473 | map<string, bufferlist>& attrs, | |
7474 | RGWRados::AttrsMod attrs_mod) | |
7475 | { | |
7476 | switch (attrs_mod) { | |
7477 | case RGWRados::ATTRSMOD_NONE: | |
7478 | attrs = src_attrs; | |
7479 | break; | |
7480 | case RGWRados::ATTRSMOD_REPLACE: | |
7481 | if (!attrs[RGW_ATTR_ETAG].length()) { | |
7482 | attrs[RGW_ATTR_ETAG] = src_attrs[RGW_ATTR_ETAG]; | |
7483 | } | |
181888fb FG |
7484 | if (!attrs[RGW_ATTR_TAIL_TAG].length()) { |
7485 | auto ttiter = src_attrs.find(RGW_ATTR_TAIL_TAG); | |
7486 | if (ttiter != src_attrs.end()) { | |
7487 | attrs[RGW_ATTR_TAIL_TAG] = src_attrs[RGW_ATTR_TAIL_TAG]; | |
7488 | } | |
7489 | } | |
7c673cae FG |
7490 | break; |
7491 | case RGWRados::ATTRSMOD_MERGE: | |
7492 | for (map<string, bufferlist>::iterator it = src_attrs.begin(); it != src_attrs.end(); ++it) { | |
7493 | if (attrs.find(it->first) == attrs.end()) { | |
7494 | attrs[it->first] = it->second; | |
7495 | } | |
7496 | } | |
7497 | break; | |
7498 | } | |
7499 | } | |
7500 | ||
7501 | int RGWRados::rewrite_obj(RGWBucketInfo& dest_bucket_info, rgw_obj& obj) | |
7502 | { | |
7503 | map<string, bufferlist> attrset; | |
7504 | ||
7505 | real_time mtime; | |
7506 | uint64_t obj_size; | |
7507 | RGWObjectCtx rctx(this); | |
7508 | ||
7509 | RGWRados::Object op_target(this, dest_bucket_info, rctx, obj); | |
7510 | RGWRados::Object::Read read_op(&op_target); | |
7511 | ||
7512 | read_op.params.attrs = &attrset; | |
7513 | read_op.params.lastmod = &mtime; | |
7514 | read_op.params.obj_size = &obj_size; | |
7515 | ||
7516 | int ret = read_op.prepare(); | |
7517 | if (ret < 0) | |
7518 | return ret; | |
7519 | ||
7520 | attrset.erase(RGW_ATTR_ID_TAG); | |
181888fb | 7521 | attrset.erase(RGW_ATTR_TAIL_TAG); |
7c673cae FG |
7522 | |
7523 | uint64_t max_chunk_size; | |
7524 | ||
7525 | ret = get_max_chunk_size(dest_bucket_info.placement_rule, obj, &max_chunk_size); | |
7526 | if (ret < 0) { | |
7527 | ldout(cct, 0) << "ERROR: failed to get max_chunk_size() for bucket " << obj.bucket << dendl; | |
7528 | return ret; | |
7529 | } | |
7530 | ||
b32b8144 FG |
7531 | return copy_obj_data(rctx, dest_bucket_info, read_op, obj_size - 1, obj, obj, |
7532 | max_chunk_size, NULL, mtime, attrset, | |
7533 | RGW_OBJ_CATEGORY_MAIN, 0, real_time(), | |
7534 | (obj.key.instance.empty() ? NULL : &(obj.key.instance)), | |
7535 | NULL, NULL); | |
7c673cae FG |
7536 | } |
7537 | ||
7538 | struct obj_time_weight { | |
7539 | real_time mtime; | |
7540 | uint32_t zone_short_id; | |
7541 | uint64_t pg_ver; | |
7542 | bool high_precision; | |
7543 | ||
7544 | obj_time_weight() : zone_short_id(0), pg_ver(0), high_precision(false) {} | |
7545 | ||
7546 | bool compare_low_precision(const obj_time_weight& rhs) { | |
7547 | struct timespec l = ceph::real_clock::to_timespec(mtime); | |
7548 | struct timespec r = ceph::real_clock::to_timespec(rhs.mtime); | |
7549 | l.tv_nsec = 0; | |
7550 | r.tv_nsec = 0; | |
7551 | if (l > r) { | |
7552 | return false; | |
7553 | } | |
7554 | if (l < r) { | |
7555 | return true; | |
7556 | } | |
7557 | if (zone_short_id != rhs.zone_short_id) { | |
7558 | return (zone_short_id < rhs.zone_short_id); | |
7559 | } | |
7560 | return (pg_ver < rhs.pg_ver); | |
7561 | ||
7562 | } | |
7563 | ||
7564 | bool operator<(const obj_time_weight& rhs) { | |
7565 | if (!high_precision || !rhs.high_precision) { | |
7566 | return compare_low_precision(rhs); | |
7567 | } | |
7568 | if (mtime > rhs.mtime) { | |
7569 | return false; | |
7570 | } | |
7571 | if (mtime < rhs.mtime) { | |
7572 | return true; | |
7573 | } | |
7574 | if (zone_short_id != rhs.zone_short_id) { | |
7575 | return (zone_short_id < rhs.zone_short_id); | |
7576 | } | |
7577 | return (pg_ver < rhs.pg_ver); | |
7578 | } | |
7579 | ||
7580 | void init(const real_time& _mtime, uint32_t _short_id, uint64_t _pg_ver) { | |
7581 | mtime = _mtime; | |
7582 | zone_short_id = _short_id; | |
7583 | pg_ver = _pg_ver; | |
7584 | } | |
7585 | ||
7586 | void init(RGWObjState *state) { | |
7587 | mtime = state->mtime; | |
7588 | zone_short_id = state->zone_short_id; | |
7589 | pg_ver = state->pg_ver; | |
7590 | } | |
7591 | }; | |
7592 | ||
7593 | inline ostream& operator<<(ostream& out, const obj_time_weight &o) { | |
7594 | out << o.mtime; | |
7595 | ||
7596 | if (o.zone_short_id != 0 || o.pg_ver != 0) { | |
7597 | out << "[zid=" << o.zone_short_id << ", pgv=" << o.pg_ver << "]"; | |
7598 | } | |
7599 | ||
7600 | return out; | |
7601 | } | |
7602 | ||
7603 | class RGWGetExtraDataCB : public RGWGetDataCB { | |
7604 | bufferlist extra_data; | |
7605 | public: | |
7606 | RGWGetExtraDataCB() {} | |
7607 | int handle_data(bufferlist& bl, off_t bl_ofs, off_t bl_len) override { | |
7608 | if (extra_data.length() < extra_data_len) { | |
7609 | off_t max = extra_data_len - extra_data.length(); | |
7610 | if (max > bl_len) { | |
7611 | max = bl_len; | |
7612 | } | |
7613 | bl.splice(0, max, &extra_data); | |
7614 | } | |
7615 | return bl_len; | |
7616 | } | |
7617 | ||
7618 | bufferlist& get_extra_data() { | |
7619 | return extra_data; | |
7620 | } | |
7621 | }; | |
7622 | ||
7623 | int RGWRados::stat_remote_obj(RGWObjectCtx& obj_ctx, | |
7624 | const rgw_user& user_id, | |
7625 | const string& client_id, | |
7626 | req_info *info, | |
7627 | const string& source_zone, | |
7628 | rgw_obj& src_obj, | |
7629 | RGWBucketInfo& src_bucket_info, | |
7630 | real_time *src_mtime, | |
7631 | uint64_t *psize, | |
7632 | const real_time *mod_ptr, | |
7633 | const real_time *unmod_ptr, | |
7634 | bool high_precision_time, | |
7635 | const char *if_match, | |
7636 | const char *if_nomatch, | |
7637 | map<string, bufferlist> *pattrs, | |
7638 | string *version_id, | |
7639 | string *ptag, | |
7640 | string *petag) | |
7641 | { | |
7642 | /* source is in a different zonegroup, copy from there */ | |
7643 | ||
7644 | RGWRESTStreamRWRequest *in_stream_req; | |
7645 | string tag; | |
7646 | map<string, bufferlist> src_attrs; | |
7647 | append_rand_alpha(cct, tag, tag, 32); | |
7648 | obj_time_weight set_mtime_weight; | |
7649 | set_mtime_weight.high_precision = high_precision_time; | |
7650 | ||
7651 | RGWRESTConn *conn; | |
7652 | if (source_zone.empty()) { | |
7653 | if (src_bucket_info.zonegroup.empty()) { | |
7654 | /* source is in the master zonegroup */ | |
7655 | conn = rest_master_conn; | |
7656 | } else { | |
7657 | map<string, RGWRESTConn *>::iterator iter = zonegroup_conn_map.find(src_bucket_info.zonegroup); | |
7658 | if (iter == zonegroup_conn_map.end()) { | |
7659 | ldout(cct, 0) << "could not find zonegroup connection to zonegroup: " << source_zone << dendl; | |
7660 | return -ENOENT; | |
7661 | } | |
7662 | conn = iter->second; | |
7663 | } | |
7664 | } else { | |
7665 | map<string, RGWRESTConn *>::iterator iter = zone_conn_map.find(source_zone); | |
7666 | if (iter == zone_conn_map.end()) { | |
7667 | ldout(cct, 0) << "could not find zone connection to zone: " << source_zone << dendl; | |
7668 | return -ENOENT; | |
7669 | } | |
7670 | conn = iter->second; | |
7671 | } | |
7672 | ||
7673 | RGWGetExtraDataCB cb; | |
7674 | string etag; | |
7675 | map<string, string> req_headers; | |
7676 | real_time set_mtime; | |
7677 | ||
7678 | const real_time *pmod = mod_ptr; | |
7679 | ||
7680 | obj_time_weight dest_mtime_weight; | |
7681 | ||
181888fb FG |
7682 | constexpr bool prepend_meta = true; |
7683 | constexpr bool get_op = true; | |
7684 | constexpr bool rgwx_stat = true; | |
7685 | constexpr bool sync_manifest = true; | |
7686 | constexpr bool skip_decrypt = true; | |
7c673cae FG |
7687 | int ret = conn->get_obj(user_id, info, src_obj, pmod, unmod_ptr, |
7688 | dest_mtime_weight.zone_short_id, dest_mtime_weight.pg_ver, | |
181888fb FG |
7689 | prepend_meta, get_op, rgwx_stat, |
7690 | sync_manifest, skip_decrypt, &cb, &in_stream_req); | |
7c673cae FG |
7691 | if (ret < 0) { |
7692 | return ret; | |
7693 | } | |
7694 | ||
7695 | ret = conn->complete_request(in_stream_req, etag, &set_mtime, psize, req_headers); | |
7696 | if (ret < 0) { | |
7697 | return ret; | |
7698 | } | |
7699 | ||
7700 | bufferlist& extra_data_bl = cb.get_extra_data(); | |
7701 | if (extra_data_bl.length()) { | |
7702 | JSONParser jp; | |
7703 | if (!jp.parse(extra_data_bl.c_str(), extra_data_bl.length())) { | |
7704 | ldout(cct, 0) << "failed to parse response extra data. len=" << extra_data_bl.length() << " data=" << extra_data_bl.c_str() << dendl; | |
7705 | return -EIO; | |
7706 | } | |
7707 | ||
7708 | JSONDecoder::decode_json("attrs", src_attrs, &jp); | |
7709 | ||
7710 | src_attrs.erase(RGW_ATTR_MANIFEST); // not interested in original object layout | |
7711 | } | |
7712 | ||
7713 | if (src_mtime) { | |
7714 | *src_mtime = set_mtime; | |
7715 | } | |
7716 | ||
7717 | if (petag) { | |
7718 | map<string, bufferlist>::iterator iter = src_attrs.find(RGW_ATTR_ETAG); | |
7719 | if (iter != src_attrs.end()) { | |
7720 | bufferlist& etagbl = iter->second; | |
7721 | *petag = etagbl.to_str(); | |
7722 | } | |
7723 | } | |
7724 | ||
7725 | if (pattrs) { | |
7726 | *pattrs = src_attrs; | |
7727 | } | |
7728 | ||
7729 | return 0; | |
7730 | } | |
7731 | ||
7732 | int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx, | |
7733 | const rgw_user& user_id, | |
7734 | const string& client_id, | |
7735 | const string& op_id, | |
7736 | bool record_op_state, | |
7737 | req_info *info, | |
7738 | const string& source_zone, | |
7739 | rgw_obj& dest_obj, | |
7740 | rgw_obj& src_obj, | |
7741 | RGWBucketInfo& dest_bucket_info, | |
7742 | RGWBucketInfo& src_bucket_info, | |
7743 | real_time *src_mtime, | |
7744 | real_time *mtime, | |
7745 | const real_time *mod_ptr, | |
7746 | const real_time *unmod_ptr, | |
7747 | bool high_precision_time, | |
7748 | const char *if_match, | |
7749 | const char *if_nomatch, | |
7750 | AttrsMod attrs_mod, | |
7751 | bool copy_if_newer, | |
7752 | map<string, bufferlist>& attrs, | |
7753 | RGWObjCategory category, | |
7754 | uint64_t olh_epoch, | |
7755 | real_time delete_at, | |
7756 | string *version_id, | |
7757 | string *ptag, | |
7758 | ceph::buffer::list *petag, | |
7c673cae | 7759 | void (*progress_cb)(off_t, void *), |
31f18b77 FG |
7760 | void *progress_data, |
7761 | rgw_zone_set *zones_trace) | |
7c673cae FG |
7762 | { |
7763 | /* source is in a different zonegroup, copy from there */ | |
7764 | ||
7765 | RGWRESTStreamRWRequest *in_stream_req; | |
7766 | string tag; | |
7767 | int i; | |
7768 | append_rand_alpha(cct, tag, tag, 32); | |
7769 | obj_time_weight set_mtime_weight; | |
7770 | set_mtime_weight.high_precision = high_precision_time; | |
7771 | ||
7772 | RGWPutObjProcessor_Atomic processor(obj_ctx, | |
7773 | dest_bucket_info, dest_obj.bucket, dest_obj.key.name, | |
7774 | cct->_conf->rgw_obj_stripe_size, tag, dest_bucket_info.versioning_enabled()); | |
7775 | if (version_id && *version_id != "null") { | |
7776 | processor.set_version_id(*version_id); | |
7777 | } | |
7778 | processor.set_olh_epoch(olh_epoch); | |
7779 | int ret = processor.prepare(this, NULL); | |
7780 | if (ret < 0) { | |
7781 | return ret; | |
7782 | } | |
7783 | ||
7784 | RGWRESTConn *conn; | |
7785 | if (source_zone.empty()) { | |
7786 | if (dest_bucket_info.zonegroup.empty()) { | |
7787 | /* source is in the master zonegroup */ | |
7788 | conn = rest_master_conn; | |
7789 | } else { | |
7790 | map<string, RGWRESTConn *>::iterator iter = zonegroup_conn_map.find(src_bucket_info.zonegroup); | |
7791 | if (iter == zonegroup_conn_map.end()) { | |
7792 | ldout(cct, 0) << "could not find zonegroup connection to zonegroup: " << source_zone << dendl; | |
7793 | return -ENOENT; | |
7794 | } | |
7795 | conn = iter->second; | |
7796 | } | |
7797 | } else { | |
7798 | map<string, RGWRESTConn *>::iterator iter = zone_conn_map.find(source_zone); | |
7799 | if (iter == zone_conn_map.end()) { | |
7800 | ldout(cct, 0) << "could not find zone connection to zone: " << source_zone << dendl; | |
7801 | return -ENOENT; | |
7802 | } | |
7803 | conn = iter->second; | |
7804 | } | |
7805 | ||
7806 | string obj_name = dest_obj.bucket.name + "/" + dest_obj.get_oid(); | |
7807 | ||
7808 | RGWOpStateSingleOp *opstate = NULL; | |
7809 | ||
7810 | if (record_op_state) { | |
7811 | opstate = new RGWOpStateSingleOp(this, client_id, op_id, obj_name); | |
7812 | ||
7813 | ret = opstate->set_state(RGWOpState::OPSTATE_IN_PROGRESS); | |
7814 | if (ret < 0) { | |
7815 | ldout(cct, 0) << "ERROR: failed to set opstate ret=" << ret << dendl; | |
7816 | delete opstate; | |
7817 | return ret; | |
7818 | } | |
7819 | } | |
7820 | ||
7821 | boost::optional<RGWPutObj_Compress> compressor; | |
7822 | CompressorRef plugin; | |
7823 | ||
7824 | const auto& compression_type = zone_params.get_compression_type( | |
7825 | dest_bucket_info.placement_rule); | |
7826 | if (compression_type != "none") { | |
7827 | plugin = Compressor::create(cct, compression_type); | |
7828 | if (!plugin) { | |
7829 | ldout(cct, 1) << "Cannot load plugin for compression type " | |
7830 | << compression_type << dendl; | |
7831 | } | |
7832 | } | |
7833 | ||
7834 | RGWRadosPutObj cb(cct, plugin, compressor, &processor, opstate, progress_cb, progress_data); | |
7835 | ||
7836 | string etag; | |
7837 | map<string, string> req_headers; | |
7838 | real_time set_mtime; | |
7839 | ||
7840 | RGWObjState *dest_state = NULL; | |
7841 | ||
7842 | const real_time *pmod = mod_ptr; | |
7843 | ||
7844 | obj_time_weight dest_mtime_weight; | |
7845 | ||
7846 | if (copy_if_newer) { | |
7847 | /* need to get mtime for destination */ | |
7848 | ret = get_obj_state(&obj_ctx, dest_bucket_info, dest_obj, &dest_state, false); | |
7849 | if (ret < 0) | |
7850 | goto set_err_state; | |
7851 | ||
7852 | if (!real_clock::is_zero(dest_state->mtime)) { | |
7853 | dest_mtime_weight.init(dest_state); | |
7854 | pmod = &dest_mtime_weight.mtime; | |
7855 | } | |
7856 | } | |
7857 | ||
181888fb FG |
7858 | static constexpr bool prepend_meta = true; |
7859 | static constexpr bool get_op = true; | |
7860 | static constexpr bool rgwx_stat = false; | |
7861 | static constexpr bool sync_manifest = true; | |
7862 | static constexpr bool skip_decrypt = true; | |
7c673cae FG |
7863 | ret = conn->get_obj(user_id, info, src_obj, pmod, unmod_ptr, |
7864 | dest_mtime_weight.zone_short_id, dest_mtime_weight.pg_ver, | |
181888fb FG |
7865 | prepend_meta, get_op, rgwx_stat, |
7866 | sync_manifest, skip_decrypt, &cb, &in_stream_req); | |
7c673cae FG |
7867 | if (ret < 0) { |
7868 | goto set_err_state; | |
7869 | } | |
7870 | ||
7871 | ret = conn->complete_request(in_stream_req, etag, &set_mtime, nullptr, req_headers); | |
7872 | if (ret < 0) { | |
7873 | goto set_err_state; | |
7874 | } | |
28e407b8 AA |
7875 | ret = cb.flush(); |
7876 | if (ret < 0) { | |
7877 | goto set_err_state; | |
7878 | } | |
7c673cae FG |
7879 | if (compressor && compressor->is_compressed()) { |
7880 | bufferlist tmp; | |
7881 | RGWCompressionInfo cs_info; | |
7882 | cs_info.compression_type = plugin->get_type_name(); | |
7883 | cs_info.orig_size = cb.get_data_len(); | |
7884 | cs_info.blocks = move(compressor->get_compression_blocks()); | |
7885 | ::encode(cs_info, tmp); | |
7886 | cb.get_attrs()[RGW_ATTR_COMPRESSION] = tmp; | |
7887 | } | |
7888 | ||
7889 | if (source_zone.empty()) { /* need to preserve expiration if copy in the same zonegroup */ | |
7890 | cb.get_attrs().erase(RGW_ATTR_DELETE_AT); | |
7891 | } else { | |
7892 | map<string, bufferlist>::iterator iter = cb.get_attrs().find(RGW_ATTR_DELETE_AT); | |
7893 | if (iter != cb.get_attrs().end()) { | |
7894 | try { | |
7895 | ::decode(delete_at, iter->second); | |
7896 | } catch (buffer::error& err) { | |
7897 | ldout(cct, 0) << "ERROR: failed to decode delete_at field in intra zone copy" << dendl; | |
7898 | } | |
7899 | } | |
7900 | } | |
7901 | ||
7902 | if (src_mtime) { | |
7903 | *src_mtime = set_mtime; | |
7904 | } | |
7905 | ||
7906 | if (petag) { | |
7907 | const auto iter = cb.get_attrs().find(RGW_ATTR_ETAG); | |
7908 | if (iter != cb.get_attrs().end()) { | |
7909 | *petag = iter->second; | |
7910 | } | |
7911 | } | |
7912 | ||
7913 | if (source_zone.empty()) { | |
7914 | set_copy_attrs(cb.get_attrs(), attrs, attrs_mod); | |
7915 | } else { | |
7916 | attrs = cb.get_attrs(); | |
7917 | } | |
7918 | ||
7919 | if (copy_if_newer) { | |
7920 | uint64_t pg_ver = 0; | |
7921 | auto i = attrs.find(RGW_ATTR_PG_VER); | |
7922 | if (i != attrs.end() && i->second.length() > 0) { | |
7923 | bufferlist::iterator iter = i->second.begin(); | |
7924 | try { | |
7925 | ::decode(pg_ver, iter); | |
7926 | } catch (buffer::error& err) { | |
7927 | ldout(ctx(), 0) << "ERROR: failed to decode pg ver attribute, ignoring" << dendl; | |
7928 | /* non critical error */ | |
7929 | } | |
7930 | } | |
7931 | set_mtime_weight.init(set_mtime, get_zone_short_id(), pg_ver); | |
7932 | } | |
7933 | ||
7934 | #define MAX_COMPLETE_RETRY 100 | |
7935 | for (i = 0; i < MAX_COMPLETE_RETRY; i++) { | |
31f18b77 | 7936 | ret = cb.complete(etag, mtime, set_mtime, attrs, delete_at, zones_trace); |
7c673cae FG |
7937 | if (ret < 0) { |
7938 | goto set_err_state; | |
7939 | } | |
7940 | if (copy_if_newer && cb.is_canceled()) { | |
7941 | ldout(cct, 20) << "raced with another write of obj: " << dest_obj << dendl; | |
7942 | obj_ctx.obj.invalidate(dest_obj); /* object was overwritten */ | |
7943 | ret = get_obj_state(&obj_ctx, dest_bucket_info, dest_obj, &dest_state, false); | |
7944 | if (ret < 0) { | |
7945 | ldout(cct, 0) << "ERROR: " << __func__ << ": get_err_state() returned ret=" << ret << dendl; | |
7946 | goto set_err_state; | |
7947 | } | |
7948 | dest_mtime_weight.init(dest_state); | |
7949 | dest_mtime_weight.high_precision = high_precision_time; | |
7950 | if (!dest_state->exists || | |
7951 | dest_mtime_weight < set_mtime_weight) { | |
7952 | ldout(cct, 20) << "retrying writing object mtime=" << set_mtime << " dest_state->mtime=" << dest_state->mtime << " dest_state->exists=" << dest_state->exists << dendl; | |
7953 | continue; | |
7954 | } else { | |
7955 | ldout(cct, 20) << "not retrying writing object mtime=" << set_mtime << " dest_state->mtime=" << dest_state->mtime << " dest_state->exists=" << dest_state->exists << dendl; | |
7956 | } | |
7957 | } | |
7958 | break; | |
7959 | } | |
7960 | ||
7961 | if (i == MAX_COMPLETE_RETRY) { | |
7962 | ldout(cct, 0) << "ERROR: retried object completion too many times, something is wrong!" << dendl; | |
7963 | ret = -EIO; | |
7964 | goto set_err_state; | |
7965 | } | |
7966 | ||
7967 | if (opstate) { | |
7968 | ret = opstate->set_state(RGWOpState::OPSTATE_COMPLETE); | |
7969 | if (ret < 0) { | |
7970 | ldout(cct, 0) << "ERROR: failed to set opstate ret=" << ret << dendl; | |
7971 | } | |
7972 | delete opstate; | |
7973 | } | |
7974 | ||
7975 | return 0; | |
7976 | set_err_state: | |
7977 | if (copy_if_newer && ret == -ERR_NOT_MODIFIED) { | |
7978 | ret = 0; | |
7979 | } | |
7980 | if (opstate) { | |
7981 | RGWOpState::OpState state; | |
7982 | if (ret < 0) { | |
7983 | state = RGWOpState::OPSTATE_ERROR; | |
7984 | } else { | |
7985 | state = RGWOpState::OPSTATE_COMPLETE; | |
7986 | } | |
7987 | int r = opstate->set_state(state); | |
7988 | if (r < 0) { | |
7989 | ldout(cct, 0) << "ERROR: failed to set opstate r=" << ret << dendl; | |
7990 | } | |
7991 | delete opstate; | |
7992 | } | |
7993 | return ret; | |
7994 | } | |
7995 | ||
7996 | ||
7997 | int RGWRados::copy_obj_to_remote_dest(RGWObjState *astate, | |
7998 | map<string, bufferlist>& src_attrs, | |
7999 | RGWRados::Object::Read& read_op, | |
8000 | const rgw_user& user_id, | |
8001 | rgw_obj& dest_obj, | |
8002 | real_time *mtime) | |
8003 | { | |
8004 | string etag; | |
8005 | ||
8006 | RGWRESTStreamWriteRequest *out_stream_req; | |
8007 | ||
8008 | int ret = rest_master_conn->put_obj_init(user_id, dest_obj, astate->size, src_attrs, &out_stream_req); | |
8009 | if (ret < 0) { | |
7c673cae FG |
8010 | return ret; |
8011 | } | |
8012 | ||
8013 | ret = read_op.iterate(0, astate->size - 1, out_stream_req->get_out_cb()); | |
224ce89b WB |
8014 | if (ret < 0) { |
8015 | delete out_stream_req; | |
7c673cae | 8016 | return ret; |
224ce89b | 8017 | } |
7c673cae FG |
8018 | |
8019 | ret = rest_master_conn->complete_request(out_stream_req, etag, mtime); | |
8020 | if (ret < 0) | |
8021 | return ret; | |
8022 | ||
8023 | return 0; | |
8024 | } | |
8025 | ||
8026 | /** | |
8027 | * Copy an object. | |
8028 | * dest_obj: the object to copy into | |
8029 | * src_obj: the object to copy from | |
8030 | * attrs: usage depends on attrs_mod parameter | |
8031 | * attrs_mod: the modification mode of the attrs, may have the following values: | |
8032 | * ATTRSMOD_NONE - the attributes of the source object will be | |
8033 | * copied without modifications, attrs parameter is ignored; | |
8034 | * ATTRSMOD_REPLACE - new object will have the attributes provided by attrs | |
8035 | * parameter, source object attributes are not copied; | |
8036 | * ATTRSMOD_MERGE - any conflicting meta keys on the source object's attributes | |
8037 | * are overwritten by values contained in attrs parameter. | |
8038 | * err: stores any errors resulting from the get of the original object | |
8039 | * Returns: 0 on success, -ERR# otherwise. | |
8040 | */ | |
8041 | int RGWRados::copy_obj(RGWObjectCtx& obj_ctx, | |
8042 | const rgw_user& user_id, | |
8043 | const string& client_id, | |
8044 | const string& op_id, | |
8045 | req_info *info, | |
8046 | const string& source_zone, | |
8047 | rgw_obj& dest_obj, | |
8048 | rgw_obj& src_obj, | |
8049 | RGWBucketInfo& dest_bucket_info, | |
8050 | RGWBucketInfo& src_bucket_info, | |
8051 | real_time *src_mtime, | |
8052 | real_time *mtime, | |
8053 | const real_time *mod_ptr, | |
8054 | const real_time *unmod_ptr, | |
8055 | bool high_precision_time, | |
8056 | const char *if_match, | |
8057 | const char *if_nomatch, | |
8058 | AttrsMod attrs_mod, | |
8059 | bool copy_if_newer, | |
8060 | map<string, bufferlist>& attrs, | |
8061 | RGWObjCategory category, | |
8062 | uint64_t olh_epoch, | |
8063 | real_time delete_at, | |
8064 | string *version_id, | |
8065 | string *ptag, | |
8066 | ceph::buffer::list *petag, | |
7c673cae FG |
8067 | void (*progress_cb)(off_t, void *), |
8068 | void *progress_data) | |
8069 | { | |
8070 | int ret; | |
8071 | uint64_t obj_size; | |
8072 | rgw_obj shadow_obj = dest_obj; | |
8073 | string shadow_oid; | |
8074 | ||
8075 | bool remote_src; | |
8076 | bool remote_dest; | |
8077 | ||
8078 | append_rand_alpha(cct, dest_obj.get_oid(), shadow_oid, 32); | |
8079 | shadow_obj.init_ns(dest_obj.bucket, shadow_oid, shadow_ns); | |
8080 | ||
8081 | remote_dest = !get_zonegroup().equals(dest_bucket_info.zonegroup); | |
8082 | remote_src = !get_zonegroup().equals(src_bucket_info.zonegroup); | |
8083 | ||
8084 | if (remote_src && remote_dest) { | |
8085 | ldout(cct, 0) << "ERROR: can't copy object when both src and dest buckets are remote" << dendl; | |
8086 | return -EINVAL; | |
8087 | } | |
8088 | ||
8089 | ldout(cct, 5) << "Copy object " << src_obj.bucket << ":" << src_obj.get_oid() << " => " << dest_obj.bucket << ":" << dest_obj.get_oid() << dendl; | |
8090 | ||
8091 | if (remote_src || !source_zone.empty()) { | |
8092 | return fetch_remote_obj(obj_ctx, user_id, client_id, op_id, true, info, source_zone, | |
8093 | dest_obj, src_obj, dest_bucket_info, src_bucket_info, src_mtime, mtime, mod_ptr, | |
8094 | unmod_ptr, high_precision_time, | |
8095 | if_match, if_nomatch, attrs_mod, copy_if_newer, attrs, category, | |
31f18b77 | 8096 | olh_epoch, delete_at, version_id, ptag, petag, progress_cb, progress_data); |
7c673cae FG |
8097 | } |
8098 | ||
8099 | map<string, bufferlist> src_attrs; | |
8100 | RGWRados::Object src_op_target(this, src_bucket_info, obj_ctx, src_obj); | |
8101 | RGWRados::Object::Read read_op(&src_op_target); | |
8102 | ||
8103 | read_op.conds.mod_ptr = mod_ptr; | |
8104 | read_op.conds.unmod_ptr = unmod_ptr; | |
8105 | read_op.conds.high_precision_time = high_precision_time; | |
8106 | read_op.conds.if_match = if_match; | |
8107 | read_op.conds.if_nomatch = if_nomatch; | |
8108 | read_op.params.attrs = &src_attrs; | |
8109 | read_op.params.lastmod = src_mtime; | |
8110 | read_op.params.obj_size = &obj_size; | |
7c673cae FG |
8111 | |
8112 | ret = read_op.prepare(); | |
8113 | if (ret < 0) { | |
8114 | return ret; | |
8115 | } | |
94b18763 FG |
8116 | if (src_attrs.count(RGW_ATTR_CRYPT_MODE)) { |
8117 | // Current implementation does not follow S3 spec and even | |
8118 | // may result in data corruption silently when copying | |
8119 | // multipart objects acorss pools. So reject COPY operations | |
8120 | //on encrypted objects before it is fully functional. | |
8121 | ldout(cct, 0) << "ERROR: copy op for encrypted object " << src_obj | |
8122 | << " has not been implemented." << dendl; | |
8123 | return -ERR_NOT_IMPLEMENTED; | |
8124 | } | |
7c673cae FG |
8125 | |
8126 | src_attrs[RGW_ATTR_ACL] = attrs[RGW_ATTR_ACL]; | |
8127 | src_attrs.erase(RGW_ATTR_DELETE_AT); | |
8128 | ||
8129 | set_copy_attrs(src_attrs, attrs, attrs_mod); | |
8130 | attrs.erase(RGW_ATTR_ID_TAG); | |
8131 | attrs.erase(RGW_ATTR_PG_VER); | |
8132 | attrs.erase(RGW_ATTR_SOURCE_ZONE); | |
8133 | map<string, bufferlist>::iterator cmp = src_attrs.find(RGW_ATTR_COMPRESSION); | |
8134 | if (cmp != src_attrs.end()) | |
8135 | attrs[RGW_ATTR_COMPRESSION] = cmp->second; | |
8136 | ||
8137 | RGWObjManifest manifest; | |
8138 | RGWObjState *astate = NULL; | |
8139 | ||
8140 | ret = get_obj_state(&obj_ctx, src_bucket_info, src_obj, &astate); | |
8141 | if (ret < 0) { | |
8142 | return ret; | |
8143 | } | |
8144 | ||
8145 | vector<rgw_raw_obj> ref_objs; | |
8146 | ||
8147 | if (remote_dest) { | |
8148 | /* dest is in a different zonegroup, copy it there */ | |
8149 | return copy_obj_to_remote_dest(astate, attrs, read_op, user_id, dest_obj, mtime); | |
8150 | } | |
8151 | uint64_t max_chunk_size; | |
8152 | ||
8153 | ret = get_max_chunk_size(dest_bucket_info.placement_rule, dest_obj, &max_chunk_size); | |
8154 | if (ret < 0) { | |
8155 | ldout(cct, 0) << "ERROR: failed to get max_chunk_size() for bucket " << dest_obj.bucket << dendl; | |
8156 | return ret; | |
8157 | } | |
8158 | ||
8159 | rgw_pool src_pool; | |
8160 | rgw_pool dest_pool; | |
8161 | if (!get_obj_data_pool(src_bucket_info.placement_rule, src_obj, &src_pool)) { | |
8162 | ldout(cct, 0) << "ERROR: failed to locate data pool for " << src_obj << dendl; | |
8163 | return -EIO; | |
8164 | } | |
8165 | if (!get_obj_data_pool(dest_bucket_info.placement_rule, dest_obj, &dest_pool)) { | |
8166 | ldout(cct, 0) << "ERROR: failed to locate data pool for " << dest_obj << dendl; | |
8167 | return -EIO; | |
8168 | } | |
8169 | ||
8170 | ||
8171 | bool copy_data = !astate->has_manifest || (src_pool != dest_pool); | |
8172 | bool copy_first = false; | |
8173 | if (astate->has_manifest) { | |
8174 | if (!astate->manifest.has_tail()) { | |
8175 | copy_data = true; | |
8176 | } else { | |
8177 | uint64_t head_size = astate->manifest.get_head_size(); | |
8178 | ||
8179 | if (head_size > 0) { | |
8180 | if (head_size > max_chunk_size) { | |
8181 | copy_data = true; | |
8182 | } else { | |
8183 | copy_first = true; | |
8184 | } | |
8185 | } | |
8186 | } | |
8187 | } | |
8188 | ||
8189 | if (petag) { | |
8190 | const auto iter = attrs.find(RGW_ATTR_ETAG); | |
8191 | if (iter != attrs.end()) { | |
8192 | *petag = iter->second; | |
8193 | } | |
8194 | } | |
8195 | ||
8196 | if (copy_data) { /* refcounting tail wouldn't work here, just copy the data */ | |
8197 | return copy_obj_data(obj_ctx, dest_bucket_info, read_op, obj_size - 1, dest_obj, src_obj, | |
8198 | max_chunk_size, mtime, real_time(), attrs, category, olh_epoch, delete_at, | |
31f18b77 | 8199 | version_id, ptag, petag); |
7c673cae FG |
8200 | } |
8201 | ||
8202 | RGWObjManifest::obj_iterator miter = astate->manifest.obj_begin(); | |
8203 | ||
8204 | if (copy_first) { // we need to copy first chunk, not increase refcount | |
8205 | ++miter; | |
8206 | } | |
8207 | ||
8208 | rgw_rados_ref ref; | |
8209 | ret = get_raw_obj_ref(miter.get_location().get_raw_obj(this), &ref); | |
8210 | if (ret < 0) { | |
8211 | return ret; | |
8212 | } | |
8213 | ||
8214 | bool versioned_dest = dest_bucket_info.versioning_enabled(); | |
8215 | ||
8216 | if (version_id && !version_id->empty()) { | |
8217 | versioned_dest = true; | |
8218 | dest_obj.key.set_instance(*version_id); | |
8219 | } else if (versioned_dest) { | |
8220 | gen_rand_obj_instance_name(&dest_obj); | |
8221 | } | |
8222 | ||
8223 | bufferlist first_chunk; | |
8224 | ||
8225 | bool copy_itself = (dest_obj == src_obj); | |
8226 | RGWObjManifest *pmanifest; | |
31f18b77 | 8227 | ldout(cct, 20) << "dest_obj=" << dest_obj << " src_obj=" << src_obj << " copy_itself=" << (int)copy_itself << dendl; |
7c673cae FG |
8228 | |
8229 | RGWRados::Object dest_op_target(this, dest_bucket_info, obj_ctx, dest_obj); | |
8230 | RGWRados::Object::Write write_op(&dest_op_target); | |
8231 | ||
8232 | string tag; | |
8233 | ||
8234 | if (ptag) { | |
8235 | tag = *ptag; | |
8236 | } | |
8237 | ||
8238 | if (tag.empty()) { | |
8239 | append_rand_alpha(cct, tag, tag, 32); | |
8240 | } | |
8241 | ||
8242 | if (!copy_itself) { | |
181888fb | 8243 | attrs.erase(RGW_ATTR_TAIL_TAG); |
7c673cae FG |
8244 | manifest = astate->manifest; |
8245 | const rgw_bucket_placement& tail_placement = manifest.get_tail_placement(); | |
8246 | if (tail_placement.bucket.name.empty()) { | |
8247 | manifest.set_tail_placement(tail_placement.placement_rule, src_obj.bucket); | |
8248 | } | |
3efd9988 | 8249 | string ref_tag; |
7c673cae FG |
8250 | for (; miter != astate->manifest.obj_end(); ++miter) { |
8251 | ObjectWriteOperation op; | |
3efd9988 FG |
8252 | ref_tag = tag + '\0'; |
8253 | cls_refcount_get(op, ref_tag, true); | |
7c673cae FG |
8254 | const rgw_raw_obj& loc = miter.get_location().get_raw_obj(this); |
8255 | ref.ioctx.locator_set_key(loc.loc); | |
8256 | ||
8257 | ret = ref.ioctx.operate(loc.oid, &op); | |
8258 | if (ret < 0) { | |
8259 | goto done_ret; | |
8260 | } | |
8261 | ||
8262 | ref_objs.push_back(loc); | |
8263 | } | |
8264 | ||
8265 | pmanifest = &manifest; | |
8266 | } else { | |
8267 | pmanifest = &astate->manifest; | |
8268 | /* don't send the object's tail for garbage collection */ | |
8269 | astate->keep_tail = true; | |
8270 | } | |
8271 | ||
8272 | if (copy_first) { | |
8273 | ret = read_op.read(0, max_chunk_size, first_chunk); | |
8274 | if (ret < 0) { | |
8275 | goto done_ret; | |
8276 | } | |
8277 | ||
8278 | pmanifest->set_head(dest_bucket_info.placement_rule, dest_obj, first_chunk.length()); | |
8279 | } else { | |
8280 | pmanifest->set_head(dest_bucket_info.placement_rule, dest_obj, 0); | |
8281 | } | |
8282 | ||
8283 | write_op.meta.data = &first_chunk; | |
8284 | write_op.meta.manifest = pmanifest; | |
8285 | write_op.meta.ptag = &tag; | |
8286 | write_op.meta.owner = dest_bucket_info.owner; | |
8287 | write_op.meta.mtime = mtime; | |
8288 | write_op.meta.flags = PUT_OBJ_CREATE; | |
8289 | write_op.meta.category = category; | |
8290 | write_op.meta.olh_epoch = olh_epoch; | |
8291 | write_op.meta.delete_at = delete_at; | |
181888fb | 8292 | write_op.meta.modify_tail = !copy_itself; |
7c673cae FG |
8293 | |
8294 | ret = write_op.write_meta(obj_size, astate->accounted_size, attrs); | |
8295 | if (ret < 0) { | |
8296 | goto done_ret; | |
8297 | } | |
8298 | ||
8299 | return 0; | |
8300 | ||
8301 | done_ret: | |
8302 | if (!copy_itself) { | |
8303 | vector<rgw_raw_obj>::iterator riter; | |
8304 | ||
7c673cae FG |
8305 | /* rollback reference */ |
8306 | for (riter = ref_objs.begin(); riter != ref_objs.end(); ++riter) { | |
8307 | ObjectWriteOperation op; | |
8308 | cls_refcount_put(op, tag, true); | |
8309 | ||
8310 | ref.ioctx.locator_set_key(riter->loc); | |
8311 | ||
8312 | int r = ref.ioctx.operate(riter->oid, &op); | |
8313 | if (r < 0) { | |
8314 | ldout(cct, 0) << "ERROR: cleanup after error failed to drop reference on obj=" << *riter << dendl; | |
8315 | } | |
8316 | } | |
8317 | } | |
8318 | return ret; | |
8319 | } | |
8320 | ||
8321 | ||
8322 | int RGWRados::copy_obj_data(RGWObjectCtx& obj_ctx, | |
8323 | RGWBucketInfo& dest_bucket_info, | |
8324 | RGWRados::Object::Read& read_op, off_t end, | |
8325 | rgw_obj& dest_obj, | |
8326 | rgw_obj& src_obj, | |
8327 | uint64_t max_chunk_size, | |
8328 | real_time *mtime, | |
8329 | real_time set_mtime, | |
8330 | map<string, bufferlist>& attrs, | |
8331 | RGWObjCategory category, | |
8332 | uint64_t olh_epoch, | |
8333 | real_time delete_at, | |
8334 | string *version_id, | |
8335 | string *ptag, | |
31f18b77 | 8336 | ceph::buffer::list *petag) |
7c673cae FG |
8337 | { |
8338 | bufferlist first_chunk; | |
8339 | RGWObjManifest manifest; | |
8340 | ||
8341 | string tag; | |
8342 | append_rand_alpha(cct, tag, tag, 32); | |
8343 | ||
8344 | RGWPutObjProcessor_Atomic processor(obj_ctx, | |
b32b8144 | 8345 | dest_bucket_info, dest_obj.bucket, dest_obj.key.name, |
7c673cae FG |
8346 | cct->_conf->rgw_obj_stripe_size, tag, dest_bucket_info.versioning_enabled()); |
8347 | if (version_id) { | |
8348 | processor.set_version_id(*version_id); | |
8349 | } | |
8350 | processor.set_olh_epoch(olh_epoch); | |
8351 | int ret = processor.prepare(this, NULL); | |
8352 | if (ret < 0) | |
8353 | return ret; | |
8354 | ||
8355 | off_t ofs = 0; | |
8356 | ||
8357 | do { | |
8358 | bufferlist bl; | |
8359 | ret = read_op.read(ofs, end, bl); | |
8360 | ||
8361 | uint64_t read_len = ret; | |
8362 | bool again; | |
8363 | ||
8364 | do { | |
8365 | void *handle; | |
8366 | rgw_raw_obj obj; | |
8367 | ||
8368 | ret = processor.handle_data(bl, ofs, &handle, &obj, &again); | |
8369 | if (ret < 0) { | |
8370 | return ret; | |
8371 | } | |
8372 | ret = processor.throttle_data(handle, obj, read_len, false); | |
8373 | if (ret < 0) | |
8374 | return ret; | |
8375 | } while (again); | |
8376 | ||
8377 | ofs += read_len; | |
8378 | } while (ofs <= end); | |
8379 | ||
8380 | string etag; | |
8381 | auto iter = attrs.find(RGW_ATTR_ETAG); | |
8382 | if (iter != attrs.end()) { | |
8383 | bufferlist& bl = iter->second; | |
8384 | etag = string(bl.c_str(), bl.length()); | |
8385 | if (petag) { | |
8386 | *petag = bl; | |
8387 | } | |
8388 | } | |
8389 | ||
8390 | uint64_t accounted_size; | |
8391 | { | |
8392 | bool compressed{false}; | |
8393 | RGWCompressionInfo cs_info; | |
8394 | ret = rgw_compression_info_from_attrset(attrs, compressed, cs_info); | |
8395 | if (ret < 0) { | |
8396 | ldout(cct, 0) << "ERROR: failed to read compression info" << dendl; | |
8397 | return ret; | |
8398 | } | |
8399 | // pass original size if compressed | |
8400 | accounted_size = compressed ? cs_info.orig_size : ofs; | |
8401 | } | |
8402 | ||
8403 | return processor.complete(accounted_size, etag, mtime, set_mtime, attrs, delete_at); | |
8404 | } | |
8405 | ||
8406 | bool RGWRados::is_meta_master() | |
8407 | { | |
31f18b77 | 8408 | if (!get_zonegroup().is_master_zonegroup()) { |
7c673cae FG |
8409 | return false; |
8410 | } | |
8411 | ||
8412 | return (get_zonegroup().master_zone == zone_public_config.id); | |
8413 | } | |
8414 | ||
8415 | /** | |
8416 | * Check to see if the bucket metadata could be synced | |
8417 | * bucket: the bucket to check | |
8418 | * Returns false is the bucket is not synced | |
8419 | */ | |
8420 | bool RGWRados::is_syncing_bucket_meta(const rgw_bucket& bucket) | |
8421 | { | |
8422 | ||
8423 | /* no current period */ | |
8424 | if (current_period.get_id().empty()) { | |
8425 | return false; | |
8426 | } | |
8427 | ||
8428 | /* zonegroup is not master zonegroup */ | |
31f18b77 | 8429 | if (!get_zonegroup().is_master_zonegroup()) { |
7c673cae FG |
8430 | return false; |
8431 | } | |
8432 | ||
8433 | /* single zonegroup and a single zone */ | |
224ce89b | 8434 | if (current_period.is_single_zonegroup() && get_zonegroup().zones.size() == 1) { |
7c673cae FG |
8435 | return false; |
8436 | } | |
8437 | ||
8438 | /* zone is not master */ | |
8439 | if (get_zonegroup().master_zone.compare(zone_public_config.id) != 0) { | |
8440 | return false; | |
8441 | } | |
8442 | ||
8443 | return true; | |
8444 | } | |
8445 | ||
8446 | int RGWRados::check_bucket_empty(RGWBucketInfo& bucket_info) | |
8447 | { | |
8448 | std::map<string, rgw_bucket_dir_entry> ent_map; | |
8449 | rgw_obj_index_key marker; | |
8450 | string prefix; | |
8451 | bool is_truncated; | |
8452 | ||
8453 | do { | |
8454 | #define NUM_ENTRIES 1000 | |
8455 | int r = cls_bucket_list(bucket_info, RGW_NO_SHARD, marker, prefix, NUM_ENTRIES, true, ent_map, | |
8456 | &is_truncated, &marker); | |
8457 | if (r < 0) | |
8458 | return r; | |
8459 | ||
8460 | string ns; | |
8461 | std::map<string, rgw_bucket_dir_entry>::iterator eiter; | |
8462 | for (eiter = ent_map.begin(); eiter != ent_map.end(); ++eiter) { | |
8463 | rgw_obj_key obj; | |
8464 | ||
8465 | if (rgw_obj_key::oid_to_key_in_ns(eiter->second.key.name, &obj, ns)) | |
8466 | return -ENOTEMPTY; | |
8467 | } | |
8468 | } while (is_truncated); | |
8469 | return 0; | |
8470 | } | |
8471 | ||
8472 | /** | |
8473 | * Delete a bucket. | |
8474 | * bucket: the name of the bucket to delete | |
8475 | * Returns 0 on success, -ERR# otherwise. | |
8476 | */ | |
8477 | int RGWRados::delete_bucket(RGWBucketInfo& bucket_info, RGWObjVersionTracker& objv_tracker, bool check_empty) | |
8478 | { | |
8479 | const rgw_bucket& bucket = bucket_info.bucket; | |
8480 | librados::IoCtx index_ctx; | |
8481 | map<int, string> bucket_objs; | |
8482 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs); | |
8483 | if (r < 0) | |
8484 | return r; | |
8485 | ||
8486 | if (check_empty) { | |
8487 | r = check_bucket_empty(bucket_info); | |
8488 | if (r < 0) { | |
8489 | return r; | |
8490 | } | |
8491 | } | |
8492 | ||
8493 | r = rgw_bucket_delete_bucket_obj(this, bucket.tenant, bucket.name, objv_tracker); | |
8494 | if (r < 0) | |
8495 | return r; | |
8496 | ||
8497 | /* if the bucket is not synced we can remove the meta file */ | |
8498 | if (!is_syncing_bucket_meta(bucket)) { | |
8499 | RGWObjVersionTracker objv_tracker; | |
8500 | string entry = bucket.get_key(); | |
8501 | r= rgw_bucket_instance_remove_entry(this, entry, &objv_tracker); | |
8502 | if (r < 0) { | |
8503 | return r; | |
8504 | } | |
8505 | /* remove bucket index objects*/ | |
8506 | map<int, string>::const_iterator biter; | |
8507 | for (biter = bucket_objs.begin(); biter != bucket_objs.end(); ++biter) { | |
8508 | index_ctx.remove(biter->second); | |
8509 | } | |
8510 | } | |
8511 | return 0; | |
8512 | } | |
8513 | ||
8514 | int RGWRados::set_bucket_owner(rgw_bucket& bucket, ACLOwner& owner) | |
8515 | { | |
8516 | RGWBucketInfo info; | |
8517 | map<string, bufferlist> attrs; | |
8518 | RGWObjectCtx obj_ctx(this); | |
31f18b77 FG |
8519 | int r; |
8520 | if (bucket.bucket_id.empty()) { | |
8521 | r = get_bucket_info(obj_ctx, bucket.tenant, bucket.name, info, NULL, &attrs); | |
8522 | } else { | |
8523 | r = get_bucket_instance_info(obj_ctx, bucket, info, nullptr, &attrs); | |
8524 | } | |
7c673cae FG |
8525 | if (r < 0) { |
8526 | ldout(cct, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl; | |
8527 | return r; | |
8528 | } | |
8529 | ||
8530 | info.owner = owner.get_id(); | |
8531 | ||
8532 | r = put_bucket_instance_info(info, false, real_time(), &attrs); | |
8533 | if (r < 0) { | |
8534 | ldout(cct, 0) << "NOTICE: put_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl; | |
8535 | return r; | |
8536 | } | |
8537 | ||
8538 | return 0; | |
8539 | } | |
8540 | ||
8541 | ||
8542 | int RGWRados::set_buckets_enabled(vector<rgw_bucket>& buckets, bool enabled) | |
8543 | { | |
8544 | int ret = 0; | |
8545 | ||
8546 | vector<rgw_bucket>::iterator iter; | |
8547 | ||
8548 | for (iter = buckets.begin(); iter != buckets.end(); ++iter) { | |
8549 | rgw_bucket& bucket = *iter; | |
8550 | if (enabled) | |
8551 | ldout(cct, 20) << "enabling bucket name=" << bucket.name << dendl; | |
8552 | else | |
8553 | ldout(cct, 20) << "disabling bucket name=" << bucket.name << dendl; | |
8554 | ||
8555 | RGWBucketInfo info; | |
8556 | map<string, bufferlist> attrs; | |
8557 | RGWObjectCtx obj_ctx(this); | |
8558 | int r = get_bucket_info(obj_ctx, bucket.tenant, bucket.name, info, NULL, &attrs); | |
8559 | if (r < 0) { | |
8560 | ldout(cct, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl; | |
8561 | ret = r; | |
8562 | continue; | |
8563 | } | |
8564 | if (enabled) { | |
8565 | info.flags &= ~BUCKET_SUSPENDED; | |
8566 | } else { | |
8567 | info.flags |= BUCKET_SUSPENDED; | |
8568 | } | |
8569 | ||
8570 | r = put_bucket_instance_info(info, false, real_time(), &attrs); | |
8571 | if (r < 0) { | |
8572 | ldout(cct, 0) << "NOTICE: put_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl; | |
8573 | ret = r; | |
8574 | continue; | |
8575 | } | |
8576 | } | |
8577 | return ret; | |
8578 | } | |
8579 | ||
8580 | int RGWRados::bucket_suspended(rgw_bucket& bucket, bool *suspended) | |
8581 | { | |
8582 | RGWBucketInfo bucket_info; | |
8583 | RGWObjectCtx obj_ctx(this); | |
8584 | int ret = get_bucket_info(obj_ctx, bucket.tenant, bucket.name, bucket_info, NULL); | |
8585 | if (ret < 0) { | |
8586 | return ret; | |
8587 | } | |
8588 | ||
8589 | *suspended = ((bucket_info.flags & BUCKET_SUSPENDED) != 0); | |
8590 | return 0; | |
8591 | } | |
8592 | ||
8593 | int RGWRados::Object::complete_atomic_modification() | |
8594 | { | |
8595 | if (!state->has_manifest || state->keep_tail) | |
8596 | return 0; | |
8597 | ||
8598 | cls_rgw_obj_chain chain; | |
8599 | store->update_gc_chain(obj, state->manifest, &chain); | |
8600 | ||
8601 | if (chain.empty()) { | |
8602 | return 0; | |
8603 | } | |
8604 | ||
181888fb | 8605 | string tag = (state->tail_tag.length() > 0 ? state->tail_tag.to_str() : state->obj_tag.to_str()); |
7c673cae FG |
8606 | return store->gc->send_chain(chain, tag, false); // do it async |
8607 | } | |
8608 | ||
8609 | void RGWRados::update_gc_chain(rgw_obj& head_obj, RGWObjManifest& manifest, cls_rgw_obj_chain *chain) | |
8610 | { | |
8611 | RGWObjManifest::obj_iterator iter; | |
8612 | rgw_raw_obj raw_head; | |
8613 | obj_to_raw(manifest.get_head_placement_rule(), head_obj, &raw_head); | |
8614 | for (iter = manifest.obj_begin(); iter != manifest.obj_end(); ++iter) { | |
8615 | const rgw_raw_obj& mobj = iter.get_location().get_raw_obj(this); | |
8616 | if (mobj == raw_head) | |
8617 | continue; | |
8618 | cls_rgw_obj_key key(mobj.oid); | |
8619 | chain->push_obj(mobj.pool.to_str(), key, mobj.loc); | |
8620 | } | |
8621 | } | |
8622 | ||
8623 | int RGWRados::send_chain_to_gc(cls_rgw_obj_chain& chain, const string& tag, bool sync) | |
8624 | { | |
8625 | return gc->send_chain(chain, tag, sync); | |
8626 | } | |
8627 | ||
8628 | int RGWRados::open_bucket_index(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, string& bucket_oid) | |
8629 | { | |
8630 | const rgw_bucket& bucket = bucket_info.bucket; | |
8631 | int r = open_bucket_index_ctx(bucket_info, index_ctx); | |
8632 | if (r < 0) | |
8633 | return r; | |
8634 | ||
8635 | if (bucket.bucket_id.empty()) { | |
8636 | ldout(cct, 0) << "ERROR: empty bucket id for bucket operation" << dendl; | |
8637 | return -EIO; | |
8638 | } | |
8639 | ||
8640 | bucket_oid = dir_oid_prefix; | |
8641 | bucket_oid.append(bucket.bucket_id); | |
8642 | ||
8643 | return 0; | |
8644 | } | |
8645 | ||
8646 | int RGWRados::open_bucket_index_base(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, | |
8647 | string& bucket_oid_base) { | |
8648 | const rgw_bucket& bucket = bucket_info.bucket; | |
8649 | int r = open_bucket_index_ctx(bucket_info, index_ctx); | |
8650 | if (r < 0) | |
8651 | return r; | |
8652 | ||
8653 | if (bucket.bucket_id.empty()) { | |
8654 | ldout(cct, 0) << "ERROR: empty bucket_id for bucket operation" << dendl; | |
8655 | return -EIO; | |
8656 | } | |
8657 | ||
8658 | bucket_oid_base = dir_oid_prefix; | |
8659 | bucket_oid_base.append(bucket.bucket_id); | |
8660 | ||
8661 | return 0; | |
8662 | ||
8663 | } | |
8664 | ||
8665 | int RGWRados::open_bucket_index(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, | |
8666 | map<int, string>& bucket_objs, int shard_id, map<int, string> *bucket_instance_ids) { | |
8667 | string bucket_oid_base; | |
8668 | int ret = open_bucket_index_base(bucket_info, index_ctx, bucket_oid_base); | |
8669 | if (ret < 0) { | |
8670 | return ret; | |
8671 | } | |
8672 | ||
8673 | get_bucket_index_objects(bucket_oid_base, bucket_info.num_shards, bucket_objs, shard_id); | |
8674 | if (bucket_instance_ids) { | |
8675 | get_bucket_instance_ids(bucket_info, shard_id, bucket_instance_ids); | |
8676 | } | |
8677 | return 0; | |
8678 | } | |
8679 | ||
8680 | template<typename T> | |
8681 | int RGWRados::open_bucket_index(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, | |
8682 | map<int, string>& oids, map<int, T>& bucket_objs, | |
8683 | int shard_id, map<int, string> *bucket_instance_ids) | |
8684 | { | |
8685 | int ret = open_bucket_index(bucket_info, index_ctx, oids, shard_id, bucket_instance_ids); | |
8686 | if (ret < 0) | |
8687 | return ret; | |
8688 | ||
8689 | map<int, string>::const_iterator iter = oids.begin(); | |
8690 | for (; iter != oids.end(); ++iter) { | |
8691 | bucket_objs[iter->first] = T(); | |
8692 | } | |
8693 | return 0; | |
8694 | } | |
8695 | ||
8696 | int RGWRados::open_bucket_index_shard(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, | |
8697 | const string& obj_key, string *bucket_obj, int *shard_id) | |
8698 | { | |
8699 | string bucket_oid_base; | |
8700 | int ret = open_bucket_index_base(bucket_info, index_ctx, bucket_oid_base); | |
8701 | if (ret < 0) | |
8702 | return ret; | |
8703 | ||
8704 | RGWObjectCtx obj_ctx(this); | |
8705 | ||
8706 | ret = get_bucket_index_object(bucket_oid_base, obj_key, bucket_info.num_shards, | |
8707 | (RGWBucketInfo::BIShardsHashType)bucket_info.bucket_index_shard_hash_type, bucket_obj, shard_id); | |
8708 | if (ret < 0) { | |
8709 | ldout(cct, 10) << "get_bucket_index_object() returned ret=" << ret << dendl; | |
8710 | return ret; | |
8711 | } | |
8712 | return 0; | |
8713 | } | |
8714 | ||
8715 | int RGWRados::open_bucket_index_shard(const RGWBucketInfo& bucket_info, librados::IoCtx& index_ctx, | |
8716 | int shard_id, string *bucket_obj) | |
8717 | { | |
8718 | string bucket_oid_base; | |
8719 | int ret = open_bucket_index_base(bucket_info, index_ctx, bucket_oid_base); | |
8720 | if (ret < 0) | |
8721 | return ret; | |
8722 | ||
8723 | RGWObjectCtx obj_ctx(this); | |
8724 | ||
8725 | get_bucket_index_object(bucket_oid_base, bucket_info.num_shards, | |
8726 | shard_id, bucket_obj); | |
8727 | return 0; | |
8728 | } | |
8729 | ||
8730 | static void accumulate_raw_stats(const rgw_bucket_dir_header& header, | |
8731 | map<RGWObjCategory, RGWStorageStats>& stats) | |
8732 | { | |
8733 | for (const auto& pair : header.stats) { | |
8734 | const RGWObjCategory category = static_cast<RGWObjCategory>(pair.first); | |
8735 | const rgw_bucket_category_stats& header_stats = pair.second; | |
8736 | ||
8737 | RGWStorageStats& s = stats[category]; | |
8738 | ||
8739 | s.category = category; | |
8740 | s.size += header_stats.total_size; | |
8741 | s.size_rounded += header_stats.total_size_rounded; | |
8742 | s.size_utilized += header_stats.actual_size; | |
8743 | s.num_objects += header_stats.num_entries; | |
8744 | } | |
8745 | } | |
8746 | ||
8747 | int RGWRados::bucket_check_index(RGWBucketInfo& bucket_info, | |
8748 | map<RGWObjCategory, RGWStorageStats> *existing_stats, | |
8749 | map<RGWObjCategory, RGWStorageStats> *calculated_stats) | |
8750 | { | |
8751 | librados::IoCtx index_ctx; | |
8752 | // key - bucket index object id | |
8753 | // value - bucket index check OP returned result with the given bucket index object (shard) | |
8754 | map<int, string> oids; | |
8755 | map<int, struct rgw_cls_check_index_ret> bucket_objs_ret; | |
31f18b77 | 8756 | |
7c673cae | 8757 | int ret = open_bucket_index(bucket_info, index_ctx, oids, bucket_objs_ret); |
31f18b77 FG |
8758 | if (ret < 0) { |
8759 | return ret; | |
8760 | } | |
7c673cae FG |
8761 | |
8762 | ret = CLSRGWIssueBucketCheck(index_ctx, oids, bucket_objs_ret, cct->_conf->rgw_bucket_index_max_aio)(); | |
31f18b77 FG |
8763 | if (ret < 0) { |
8764 | return ret; | |
8765 | } | |
7c673cae FG |
8766 | |
8767 | // Aggregate results (from different shards if there is any) | |
8768 | map<int, struct rgw_cls_check_index_ret>::iterator iter; | |
8769 | for (iter = bucket_objs_ret.begin(); iter != bucket_objs_ret.end(); ++iter) { | |
8770 | accumulate_raw_stats(iter->second.existing_header, *existing_stats); | |
8771 | accumulate_raw_stats(iter->second.calculated_header, *calculated_stats); | |
8772 | } | |
8773 | ||
8774 | return 0; | |
8775 | } | |
8776 | ||
8777 | int RGWRados::bucket_rebuild_index(RGWBucketInfo& bucket_info) | |
8778 | { | |
8779 | librados::IoCtx index_ctx; | |
8780 | map<int, string> bucket_objs; | |
31f18b77 | 8781 | |
7c673cae | 8782 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs); |
31f18b77 | 8783 | if (r < 0) { |
7c673cae | 8784 | return r; |
31f18b77 | 8785 | } |
7c673cae FG |
8786 | |
8787 | return CLSRGWIssueBucketRebuild(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); | |
8788 | } | |
8789 | ||
31f18b77 FG |
8790 | int RGWRados::bucket_set_reshard(RGWBucketInfo& bucket_info, const cls_rgw_bucket_instance_entry& entry) |
8791 | { | |
8792 | librados::IoCtx index_ctx; | |
8793 | map<int, string> bucket_objs; | |
8794 | ||
8795 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs); | |
8796 | if (r < 0) { | |
8797 | return r; | |
8798 | } | |
8799 | ||
8800 | return CLSRGWIssueSetBucketResharding(index_ctx, bucket_objs, entry, cct->_conf->rgw_bucket_index_max_aio)(); | |
8801 | } | |
7c673cae FG |
8802 | |
8803 | int RGWRados::defer_gc(void *ctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj) | |
8804 | { | |
8805 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
8806 | std::string oid, key; | |
8807 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
8808 | if (!rctx) | |
8809 | return 0; | |
8810 | ||
8811 | RGWObjState *state = NULL; | |
8812 | ||
8813 | int r = get_obj_state(rctx, bucket_info, obj, &state, false); | |
8814 | if (r < 0) | |
8815 | return r; | |
8816 | ||
8817 | if (!state->is_atomic) { | |
8818 | ldout(cct, 20) << "state for obj=" << obj << " is not atomic, not deferring gc operation" << dendl; | |
8819 | return -EINVAL; | |
8820 | } | |
8821 | ||
181888fb FG |
8822 | string tag; |
8823 | ||
8824 | if (state->tail_tag.length() > 0) { | |
8825 | tag = state->tail_tag.c_str(); | |
8826 | } else if (state->obj_tag.length() > 0) { | |
8827 | tag = state->obj_tag.c_str(); | |
8828 | } else { | |
7c673cae FG |
8829 | ldout(cct, 20) << "state->obj_tag is empty, not deferring gc operation" << dendl; |
8830 | return -EINVAL; | |
8831 | } | |
8832 | ||
7c673cae FG |
8833 | ldout(cct, 0) << "defer chain tag=" << tag << dendl; |
8834 | ||
8835 | return gc->defer_chain(tag, false); | |
8836 | } | |
8837 | ||
8838 | void RGWRados::remove_rgw_head_obj(ObjectWriteOperation& op) | |
8839 | { | |
8840 | list<string> prefixes; | |
8841 | prefixes.push_back(RGW_ATTR_OLH_PREFIX); | |
8842 | cls_rgw_remove_obj(op, prefixes); | |
8843 | } | |
8844 | ||
8845 | void RGWRados::cls_obj_check_prefix_exist(ObjectOperation& op, const string& prefix, bool fail_if_exist) | |
8846 | { | |
8847 | cls_rgw_obj_check_attrs_prefix(op, prefix, fail_if_exist); | |
8848 | } | |
8849 | ||
8850 | void RGWRados::cls_obj_check_mtime(ObjectOperation& op, const real_time& mtime, bool high_precision_time, RGWCheckMTimeType type) | |
8851 | { | |
8852 | cls_rgw_obj_check_mtime(op, mtime, high_precision_time, type); | |
8853 | } | |
8854 | ||
8855 | ||
8856 | /** | |
8857 | * Delete an object. | |
8858 | * bucket: name of the bucket storing the object | |
8859 | * obj: name of the object to delete | |
8860 | * Returns: 0 on success, -ERR# otherwise. | |
8861 | */ | |
8862 | int RGWRados::Object::Delete::delete_obj() | |
8863 | { | |
8864 | RGWRados *store = target->get_store(); | |
8865 | rgw_obj& src_obj = target->get_obj(); | |
8866 | const string& instance = src_obj.key.instance; | |
8867 | rgw_obj obj = src_obj; | |
8868 | ||
8869 | if (instance == "null") { | |
8870 | obj.key.instance.clear(); | |
8871 | } | |
8872 | ||
8873 | bool explicit_marker_version = (!params.marker_version_id.empty()); | |
8874 | ||
8875 | if (params.versioning_status & BUCKET_VERSIONED || explicit_marker_version) { | |
8876 | if (instance.empty() || explicit_marker_version) { | |
8877 | rgw_obj marker = obj; | |
8878 | ||
8879 | if (!params.marker_version_id.empty()) { | |
8880 | if (params.marker_version_id != "null") { | |
8881 | marker.key.set_instance(params.marker_version_id); | |
8882 | } | |
8883 | } else if ((params.versioning_status & BUCKET_VERSIONS_SUSPENDED) == 0) { | |
8884 | store->gen_rand_obj_instance_name(&marker); | |
8885 | } | |
8886 | ||
8887 | result.version_id = marker.key.instance; | |
8888 | result.delete_marker = true; | |
8889 | ||
8890 | struct rgw_bucket_dir_entry_meta meta; | |
8891 | ||
8892 | meta.owner = params.obj_owner.get_id().to_str(); | |
8893 | meta.owner_display_name = params.obj_owner.get_display_name(); | |
8894 | ||
8895 | if (real_clock::is_zero(params.mtime)) { | |
8896 | meta.mtime = real_clock::now(); | |
8897 | } else { | |
8898 | meta.mtime = params.mtime; | |
8899 | } | |
8900 | ||
31f18b77 | 8901 | 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); |
7c673cae FG |
8902 | if (r < 0) { |
8903 | return r; | |
8904 | } | |
8905 | } else { | |
8906 | rgw_bucket_dir_entry dirent; | |
8907 | ||
8908 | int r = store->bi_get_instance(target->get_bucket_info(), obj, &dirent); | |
8909 | if (r < 0) { | |
8910 | return r; | |
8911 | } | |
8912 | result.delete_marker = dirent.is_delete_marker(); | |
31f18b77 | 8913 | r = store->unlink_obj_instance(target->get_ctx(), target->get_bucket_info(), obj, params.olh_epoch, params.zones_trace); |
7c673cae FG |
8914 | if (r < 0) { |
8915 | return r; | |
8916 | } | |
8917 | result.version_id = instance; | |
8918 | } | |
8919 | ||
8920 | BucketShard *bs; | |
8921 | int r = target->get_bucket_shard(&bs); | |
8922 | if (r < 0) { | |
8923 | ldout(store->ctx(), 5) << "failed to get BucketShard object: r=" << r << dendl; | |
8924 | return r; | |
8925 | } | |
8926 | ||
c07f9fc5 FG |
8927 | if (target->bucket_info.datasync_flag_enabled()) { |
8928 | r = store->data_log->add_entry(bs->bucket, bs->shard_id); | |
8929 | if (r < 0) { | |
8930 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
8931 | return r; | |
8932 | } | |
7c673cae FG |
8933 | } |
8934 | ||
8935 | return 0; | |
8936 | } | |
8937 | ||
8938 | rgw_rados_ref ref; | |
8939 | int r = store->get_obj_head_ref(target->get_bucket_info(), obj, &ref); | |
8940 | if (r < 0) { | |
8941 | return r; | |
8942 | } | |
8943 | ||
8944 | RGWObjState *state; | |
8945 | r = target->get_state(&state, false); | |
8946 | if (r < 0) | |
8947 | return r; | |
8948 | ||
8949 | ObjectWriteOperation op; | |
8950 | ||
8951 | if (!real_clock::is_zero(params.unmod_since)) { | |
8952 | struct timespec ctime = ceph::real_clock::to_timespec(state->mtime); | |
8953 | struct timespec unmod = ceph::real_clock::to_timespec(params.unmod_since); | |
8954 | if (!params.high_precision_time) { | |
8955 | ctime.tv_nsec = 0; | |
8956 | unmod.tv_nsec = 0; | |
8957 | } | |
8958 | ||
8959 | ldout(store->ctx(), 10) << "If-UnModified-Since: " << params.unmod_since << " Last-Modified: " << ctime << dendl; | |
8960 | if (ctime > unmod) { | |
8961 | return -ERR_PRECONDITION_FAILED; | |
8962 | } | |
8963 | ||
8964 | /* only delete object if mtime is less than or equal to params.unmod_since */ | |
8965 | store->cls_obj_check_mtime(op, params.unmod_since, params.high_precision_time, CLS_RGW_CHECK_TIME_MTIME_LE); | |
8966 | } | |
8967 | uint64_t obj_size = state->size; | |
8968 | ||
8969 | if (!real_clock::is_zero(params.expiration_time)) { | |
8970 | bufferlist bl; | |
8971 | real_time delete_at; | |
8972 | ||
8973 | if (state->get_attr(RGW_ATTR_DELETE_AT, bl)) { | |
8974 | try { | |
8975 | bufferlist::iterator iter = bl.begin(); | |
8976 | ::decode(delete_at, iter); | |
8977 | } catch (buffer::error& err) { | |
8978 | ldout(store->ctx(), 0) << "ERROR: couldn't decode RGW_ATTR_DELETE_AT" << dendl; | |
8979 | return -EIO; | |
8980 | } | |
8981 | ||
8982 | if (params.expiration_time != delete_at) { | |
8983 | return -ERR_PRECONDITION_FAILED; | |
8984 | } | |
8985 | } else { | |
8986 | return -ERR_PRECONDITION_FAILED; | |
8987 | } | |
8988 | } | |
8989 | ||
8990 | if (!state->exists) { | |
8991 | target->invalidate_state(); | |
8992 | return -ENOENT; | |
8993 | } | |
8994 | ||
181888fb | 8995 | r = target->prepare_atomic_modification(op, false, NULL, NULL, NULL, true, false); |
7c673cae FG |
8996 | if (r < 0) |
8997 | return r; | |
8998 | ||
8999 | RGWBucketInfo& bucket_info = target->get_bucket_info(); | |
9000 | ||
9001 | RGWRados::Bucket bop(store, bucket_info); | |
9002 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
31f18b77 FG |
9003 | |
9004 | index_op.set_zones_trace(params.zones_trace); | |
7c673cae FG |
9005 | index_op.set_bilog_flags(params.bilog_flags); |
9006 | ||
7c673cae FG |
9007 | r = index_op.prepare(CLS_RGW_OP_DEL, &state->write_tag); |
9008 | if (r < 0) | |
9009 | return r; | |
9010 | ||
9011 | store->remove_rgw_head_obj(op); | |
9012 | r = ref.ioctx.operate(ref.oid, &op); | |
94b18763 FG |
9013 | |
9014 | /* raced with another operation, object state is indeterminate */ | |
9015 | const bool need_invalidate = (r == -ECANCELED); | |
7c673cae FG |
9016 | |
9017 | int64_t poolid = ref.ioctx.get_id(); | |
9018 | if (r >= 0) { | |
9019 | tombstone_cache_t *obj_tombstone_cache = store->get_tombstone_cache(); | |
9020 | if (obj_tombstone_cache) { | |
9021 | tombstone_entry entry{*state}; | |
9022 | obj_tombstone_cache->add(obj, entry); | |
9023 | } | |
9024 | r = index_op.complete_del(poolid, ref.ioctx.get_last_version(), state->mtime, params.remove_objs); | |
224ce89b | 9025 | |
7c673cae FG |
9026 | int ret = target->complete_atomic_modification(); |
9027 | if (ret < 0) { | |
9028 | ldout(store->ctx(), 0) << "ERROR: complete_atomic_modification returned ret=" << ret << dendl; | |
9029 | } | |
9030 | /* other than that, no need to propagate error */ | |
224ce89b WB |
9031 | } else { |
9032 | int ret = index_op.cancel(); | |
9033 | if (ret < 0) { | |
9034 | ldout(store->ctx(), 0) << "ERROR: index_op.cancel() returned ret=" << ret << dendl; | |
9035 | } | |
7c673cae FG |
9036 | } |
9037 | ||
9038 | if (need_invalidate) { | |
9039 | target->invalidate_state(); | |
9040 | } | |
9041 | ||
9042 | if (r < 0) | |
9043 | return r; | |
9044 | ||
9045 | /* update quota cache */ | |
9046 | store->quota_handler->update_stats(params.bucket_owner, obj.bucket, -1, 0, obj_size); | |
9047 | ||
9048 | return 0; | |
9049 | } | |
9050 | ||
9051 | int RGWRados::delete_obj(RGWObjectCtx& obj_ctx, | |
9052 | const RGWBucketInfo& bucket_info, | |
9053 | const rgw_obj& obj, | |
9054 | int versioning_status, | |
9055 | uint16_t bilog_flags, | |
31f18b77 FG |
9056 | const real_time& expiration_time, |
9057 | rgw_zone_set *zones_trace) | |
7c673cae FG |
9058 | { |
9059 | RGWRados::Object del_target(this, bucket_info, obj_ctx, obj); | |
9060 | RGWRados::Object::Delete del_op(&del_target); | |
9061 | ||
9062 | del_op.params.bucket_owner = bucket_info.owner; | |
9063 | del_op.params.versioning_status = versioning_status; | |
9064 | del_op.params.bilog_flags = bilog_flags; | |
9065 | del_op.params.expiration_time = expiration_time; | |
31f18b77 | 9066 | del_op.params.zones_trace = zones_trace; |
7c673cae FG |
9067 | |
9068 | return del_op.delete_obj(); | |
9069 | } | |
9070 | ||
9071 | int RGWRados::delete_raw_obj(const rgw_raw_obj& obj) | |
9072 | { | |
9073 | rgw_rados_ref ref; | |
224ce89b | 9074 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
9075 | if (r < 0) { |
9076 | return r; | |
9077 | } | |
9078 | ||
9079 | ObjectWriteOperation op; | |
9080 | ||
9081 | op.remove(); | |
9082 | r = ref.ioctx.operate(ref.oid, &op); | |
9083 | if (r < 0) | |
9084 | return r; | |
9085 | ||
9086 | return 0; | |
9087 | } | |
9088 | ||
9089 | int RGWRados::delete_system_obj(rgw_raw_obj& obj, RGWObjVersionTracker *objv_tracker) | |
9090 | { | |
9091 | if (obj.empty()) { | |
9092 | ldout(cct, 1) << "delete_system_obj got empty object name " | |
9093 | << obj << ", returning EINVAL" << dendl; | |
9094 | return -EINVAL; | |
9095 | } | |
9096 | rgw_rados_ref ref; | |
224ce89b | 9097 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
9098 | if (r < 0) { |
9099 | return r; | |
9100 | } | |
9101 | ||
9102 | ObjectWriteOperation op; | |
9103 | ||
9104 | if (objv_tracker) { | |
9105 | objv_tracker->prepare_op_for_write(&op); | |
9106 | } | |
9107 | ||
9108 | op.remove(); | |
9109 | r = ref.ioctx.operate(ref.oid, &op); | |
9110 | if (r < 0) | |
9111 | return r; | |
9112 | ||
9113 | return 0; | |
9114 | } | |
9115 | ||
9116 | int RGWRados::delete_obj_index(const rgw_obj& obj) | |
9117 | { | |
9118 | std::string oid, key; | |
9119 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
9120 | ||
9121 | RGWObjectCtx obj_ctx(this); | |
9122 | ||
9123 | RGWBucketInfo bucket_info; | |
9124 | int ret = get_bucket_instance_info(obj_ctx, obj.bucket, bucket_info, NULL, NULL); | |
9125 | if (ret < 0) { | |
9126 | ldout(cct, 0) << "ERROR: " << __func__ << "() get_bucket_instance_info(bucket=" << obj.bucket << ") returned ret=" << ret << dendl; | |
9127 | return ret; | |
9128 | } | |
9129 | ||
9130 | RGWRados::Bucket bop(this, bucket_info); | |
9131 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
9132 | ||
9133 | real_time removed_mtime; | |
9134 | int r = index_op.complete_del(-1 /* pool */, 0, removed_mtime, NULL); | |
9135 | ||
9136 | return r; | |
9137 | } | |
9138 | ||
9139 | static void generate_fake_tag(RGWRados *store, map<string, bufferlist>& attrset, RGWObjManifest& manifest, bufferlist& manifest_bl, bufferlist& tag_bl) | |
9140 | { | |
9141 | string tag; | |
9142 | ||
9143 | RGWObjManifest::obj_iterator mi = manifest.obj_begin(); | |
9144 | if (mi != manifest.obj_end()) { | |
9145 | if (manifest.has_tail()) // first object usually points at the head, let's skip to a more unique part | |
9146 | ++mi; | |
9147 | tag = mi.get_location().get_raw_obj(store).oid; | |
9148 | tag.append("_"); | |
9149 | } | |
9150 | ||
9151 | unsigned char md5[CEPH_CRYPTO_MD5_DIGESTSIZE]; | |
9152 | char md5_str[CEPH_CRYPTO_MD5_DIGESTSIZE * 2 + 1]; | |
9153 | MD5 hash; | |
9154 | hash.Update((const byte *)manifest_bl.c_str(), manifest_bl.length()); | |
9155 | ||
9156 | map<string, bufferlist>::iterator iter = attrset.find(RGW_ATTR_ETAG); | |
9157 | if (iter != attrset.end()) { | |
9158 | bufferlist& bl = iter->second; | |
9159 | hash.Update((const byte *)bl.c_str(), bl.length()); | |
9160 | } | |
9161 | ||
9162 | hash.Final(md5); | |
9163 | buf_to_hex(md5, CEPH_CRYPTO_MD5_DIGESTSIZE, md5_str); | |
9164 | tag.append(md5_str); | |
9165 | ||
9166 | ldout(store->ctx(), 10) << "generate_fake_tag new tag=" << tag << dendl; | |
9167 | ||
9168 | tag_bl.append(tag.c_str(), tag.size() + 1); | |
9169 | } | |
9170 | ||
9171 | static bool is_olh(map<string, bufferlist>& attrs) | |
9172 | { | |
9173 | map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_OLH_INFO); | |
9174 | return (iter != attrs.end()); | |
9175 | } | |
9176 | ||
9177 | static bool has_olh_tag(map<string, bufferlist>& attrs) | |
9178 | { | |
9179 | map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_OLH_ID_TAG); | |
9180 | return (iter != attrs.end()); | |
9181 | } | |
9182 | ||
9183 | int RGWRados::get_olh_target_state(RGWObjectCtx& obj_ctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
9184 | RGWObjState *olh_state, RGWObjState **target_state) | |
9185 | { | |
9186 | assert(olh_state->is_olh); | |
9187 | ||
9188 | rgw_obj target; | |
9189 | int r = RGWRados::follow_olh(bucket_info, obj_ctx, olh_state, obj, &target); /* might return -EAGAIN */ | |
9190 | if (r < 0) { | |
9191 | return r; | |
9192 | } | |
9193 | r = get_obj_state(&obj_ctx, bucket_info, target, target_state, false); | |
9194 | if (r < 0) { | |
9195 | return r; | |
9196 | } | |
9197 | ||
9198 | return 0; | |
9199 | } | |
9200 | ||
9201 | int RGWRados::get_system_obj_state_impl(RGWObjectCtx *rctx, rgw_raw_obj& obj, RGWRawObjState **state, RGWObjVersionTracker *objv_tracker) | |
9202 | { | |
9203 | if (obj.empty()) { | |
9204 | return -EINVAL; | |
9205 | } | |
9206 | ||
9207 | RGWRawObjState *s = rctx->raw.get_state(obj); | |
9208 | ldout(cct, 20) << "get_system_obj_state: rctx=" << (void *)rctx << " obj=" << obj << " state=" << (void *)s << " s->prefetch_data=" << s->prefetch_data << dendl; | |
9209 | *state = s; | |
9210 | if (s->has_attrs) { | |
9211 | return 0; | |
9212 | } | |
9213 | ||
9214 | s->obj = obj; | |
9215 | ||
9216 | int r = raw_obj_stat(obj, &s->size, &s->mtime, &s->epoch, &s->attrset, (s->prefetch_data ? &s->data : NULL), objv_tracker); | |
9217 | if (r == -ENOENT) { | |
9218 | s->exists = false; | |
9219 | s->has_attrs = true; | |
9220 | s->mtime = real_time(); | |
9221 | return 0; | |
9222 | } | |
9223 | if (r < 0) | |
9224 | return r; | |
9225 | ||
9226 | s->exists = true; | |
9227 | s->has_attrs = true; | |
9228 | s->obj_tag = s->attrset[RGW_ATTR_ID_TAG]; | |
9229 | ||
9230 | if (s->obj_tag.length()) | |
31f18b77 FG |
9231 | ldout(cct, 20) << "get_system_obj_state: setting s->obj_tag to " |
9232 | << s->obj_tag.c_str() << dendl; | |
7c673cae FG |
9233 | else |
9234 | ldout(cct, 20) << "get_system_obj_state: s->obj_tag was set empty" << dendl; | |
9235 | ||
9236 | return 0; | |
9237 | } | |
9238 | ||
9239 | int RGWRados::get_system_obj_state(RGWObjectCtx *rctx, rgw_raw_obj& obj, RGWRawObjState **state, RGWObjVersionTracker *objv_tracker) | |
9240 | { | |
9241 | int ret; | |
9242 | ||
9243 | do { | |
9244 | ret = get_system_obj_state_impl(rctx, obj, state, objv_tracker); | |
9245 | } while (ret == -EAGAIN); | |
9246 | ||
9247 | return ret; | |
9248 | } | |
9249 | ||
9250 | int RGWRados::get_obj_state_impl(RGWObjectCtx *rctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
9251 | RGWObjState **state, bool follow_olh, bool assume_noent) | |
9252 | { | |
9253 | if (obj.empty()) { | |
9254 | return -EINVAL; | |
9255 | } | |
9256 | ||
9257 | bool need_follow_olh = follow_olh && obj.key.instance.empty(); | |
9258 | ||
9259 | RGWObjState *s = rctx->obj.get_state(obj); | |
9260 | ldout(cct, 20) << "get_obj_state: rctx=" << (void *)rctx << " obj=" << obj << " state=" << (void *)s << " s->prefetch_data=" << s->prefetch_data << dendl; | |
9261 | *state = s; | |
9262 | if (s->has_attrs) { | |
9263 | if (s->is_olh && need_follow_olh) { | |
9264 | return get_olh_target_state(*rctx, bucket_info, obj, s, state); | |
9265 | } | |
9266 | return 0; | |
9267 | } | |
9268 | ||
9269 | s->obj = obj; | |
9270 | ||
9271 | rgw_raw_obj raw_obj; | |
9272 | obj_to_raw(bucket_info.placement_rule, obj, &raw_obj); | |
9273 | ||
9274 | int r = -ENOENT; | |
9275 | ||
9276 | if (!assume_noent) { | |
9277 | r = RGWRados::raw_obj_stat(raw_obj, &s->size, &s->mtime, &s->epoch, &s->attrset, (s->prefetch_data ? &s->data : NULL), NULL); | |
9278 | } | |
9279 | ||
9280 | if (r == -ENOENT) { | |
9281 | s->exists = false; | |
9282 | s->has_attrs = true; | |
9283 | tombstone_entry entry; | |
9284 | if (obj_tombstone_cache && obj_tombstone_cache->find(obj, entry)) { | |
9285 | s->mtime = entry.mtime; | |
9286 | s->zone_short_id = entry.zone_short_id; | |
9287 | s->pg_ver = entry.pg_ver; | |
9288 | ldout(cct, 20) << __func__ << "(): found obj in tombstone cache: obj=" << obj | |
9289 | << " mtime=" << s->mtime << " pgv=" << s->pg_ver << dendl; | |
9290 | } else { | |
9291 | s->mtime = real_time(); | |
9292 | } | |
9293 | return 0; | |
9294 | } | |
9295 | if (r < 0) | |
9296 | return r; | |
9297 | ||
9298 | s->exists = true; | |
9299 | s->has_attrs = true; | |
9300 | s->accounted_size = s->size; | |
9301 | ||
9302 | auto iter = s->attrset.find(RGW_ATTR_COMPRESSION); | |
31f18b77 FG |
9303 | const bool compressed = (iter != s->attrset.end()); |
9304 | if (compressed) { | |
7c673cae FG |
9305 | // use uncompressed size for accounted_size |
9306 | try { | |
9307 | RGWCompressionInfo info; | |
9308 | auto p = iter->second.begin(); | |
9309 | ::decode(info, p); | |
31f18b77 | 9310 | s->accounted_size = info.orig_size; |
7c673cae FG |
9311 | } catch (buffer::error&) { |
9312 | dout(0) << "ERROR: could not decode compression info for object: " << obj << dendl; | |
9313 | return -EIO; | |
9314 | } | |
9315 | } | |
9316 | ||
9317 | iter = s->attrset.find(RGW_ATTR_SHADOW_OBJ); | |
9318 | if (iter != s->attrset.end()) { | |
9319 | bufferlist bl = iter->second; | |
9320 | bufferlist::iterator it = bl.begin(); | |
9321 | it.copy(bl.length(), s->shadow_obj); | |
9322 | s->shadow_obj[bl.length()] = '\0'; | |
9323 | } | |
9324 | s->obj_tag = s->attrset[RGW_ATTR_ID_TAG]; | |
181888fb FG |
9325 | auto ttiter = s->attrset.find(RGW_ATTR_TAIL_TAG); |
9326 | if (ttiter != s->attrset.end()) { | |
9327 | s->tail_tag = s->attrset[RGW_ATTR_TAIL_TAG]; | |
9328 | } | |
7c673cae FG |
9329 | |
9330 | bufferlist manifest_bl = s->attrset[RGW_ATTR_MANIFEST]; | |
9331 | if (manifest_bl.length()) { | |
9332 | bufferlist::iterator miter = manifest_bl.begin(); | |
9333 | try { | |
9334 | ::decode(s->manifest, miter); | |
9335 | s->has_manifest = true; | |
9336 | s->manifest.set_head(bucket_info.placement_rule, obj, s->size); /* patch manifest to reflect the head we just read, some manifests might be | |
9337 | broken due to old bugs */ | |
9338 | s->size = s->manifest.get_obj_size(); | |
31f18b77 FG |
9339 | if (!compressed) |
9340 | s->accounted_size = s->size; | |
7c673cae FG |
9341 | } catch (buffer::error& err) { |
9342 | ldout(cct, 0) << "ERROR: couldn't decode manifest" << dendl; | |
9343 | return -EIO; | |
9344 | } | |
9345 | ldout(cct, 10) << "manifest: total_size = " << s->manifest.get_obj_size() << dendl; | |
9346 | if (cct->_conf->subsys.should_gather(ceph_subsys_rgw, 20) && s->manifest.has_explicit_objs()) { | |
9347 | RGWObjManifest::obj_iterator mi; | |
9348 | for (mi = s->manifest.obj_begin(); mi != s->manifest.obj_end(); ++mi) { | |
9349 | ldout(cct, 20) << "manifest: ofs=" << mi.get_ofs() << " loc=" << mi.get_location().get_raw_obj(this) << dendl; | |
9350 | } | |
9351 | } | |
9352 | ||
9353 | if (!s->obj_tag.length()) { | |
9354 | /* | |
9355 | * Uh oh, something's wrong, object with manifest should have tag. Let's | |
9356 | * create one out of the manifest, would be unique | |
9357 | */ | |
9358 | generate_fake_tag(this, s->attrset, s->manifest, manifest_bl, s->obj_tag); | |
9359 | s->fake_tag = true; | |
9360 | } | |
9361 | } | |
9362 | map<string, bufferlist>::iterator aiter = s->attrset.find(RGW_ATTR_PG_VER); | |
9363 | if (aiter != s->attrset.end()) { | |
9364 | bufferlist& pg_ver_bl = aiter->second; | |
9365 | if (pg_ver_bl.length()) { | |
9366 | bufferlist::iterator pgbl = pg_ver_bl.begin(); | |
9367 | try { | |
9368 | ::decode(s->pg_ver, pgbl); | |
9369 | } catch (buffer::error& err) { | |
9370 | ldout(cct, 0) << "ERROR: couldn't decode pg ver attr for object " << s->obj << ", non-critical error, ignoring" << dendl; | |
9371 | } | |
9372 | } | |
9373 | } | |
9374 | aiter = s->attrset.find(RGW_ATTR_SOURCE_ZONE); | |
9375 | if (aiter != s->attrset.end()) { | |
9376 | bufferlist& zone_short_id_bl = aiter->second; | |
9377 | if (zone_short_id_bl.length()) { | |
9378 | bufferlist::iterator zbl = zone_short_id_bl.begin(); | |
9379 | try { | |
9380 | ::decode(s->zone_short_id, zbl); | |
9381 | } catch (buffer::error& err) { | |
9382 | ldout(cct, 0) << "ERROR: couldn't decode zone short id attr for object " << s->obj << ", non-critical error, ignoring" << dendl; | |
9383 | } | |
9384 | } | |
9385 | } | |
9386 | if (s->obj_tag.length()) | |
31f18b77 | 9387 | ldout(cct, 20) << "get_obj_state: setting s->obj_tag to " << s->obj_tag.c_str() << dendl; |
7c673cae FG |
9388 | else |
9389 | ldout(cct, 20) << "get_obj_state: s->obj_tag was set empty" << dendl; | |
9390 | ||
9391 | /* an object might not be olh yet, but could have olh id tag, so we should set it anyway if | |
9392 | * it exist, and not only if is_olh() returns true | |
9393 | */ | |
9394 | iter = s->attrset.find(RGW_ATTR_OLH_ID_TAG); | |
9395 | if (iter != s->attrset.end()) { | |
9396 | s->olh_tag = iter->second; | |
9397 | } | |
9398 | ||
9399 | if (is_olh(s->attrset)) { | |
9400 | s->is_olh = true; | |
9401 | ||
9402 | ldout(cct, 20) << __func__ << ": setting s->olh_tag to " << string(s->olh_tag.c_str(), s->olh_tag.length()) << dendl; | |
9403 | ||
9404 | if (need_follow_olh) { | |
9405 | return get_olh_target_state(*rctx, bucket_info, obj, s, state); | |
9406 | } | |
9407 | } | |
9408 | ||
9409 | return 0; | |
9410 | } | |
9411 | ||
9412 | int RGWRados::get_obj_state(RGWObjectCtx *rctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, RGWObjState **state, | |
9413 | bool follow_olh, bool assume_noent) | |
9414 | { | |
9415 | int ret; | |
9416 | ||
9417 | do { | |
9418 | ret = get_obj_state_impl(rctx, bucket_info, obj, state, follow_olh, assume_noent); | |
9419 | } while (ret == -EAGAIN); | |
9420 | ||
9421 | return ret; | |
9422 | } | |
9423 | ||
9424 | int RGWRados::Object::get_manifest(RGWObjManifest **pmanifest) | |
9425 | { | |
9426 | RGWObjState *astate; | |
9427 | int r = get_state(&astate, true); | |
9428 | if (r < 0) { | |
9429 | return r; | |
9430 | } | |
9431 | ||
9432 | *pmanifest = &astate->manifest; | |
9433 | ||
9434 | return 0; | |
9435 | } | |
9436 | ||
9437 | int RGWRados::Object::Read::get_attr(const char *name, bufferlist& dest) | |
9438 | { | |
9439 | RGWObjState *state; | |
9440 | int r = source->get_state(&state, true); | |
9441 | if (r < 0) | |
9442 | return r; | |
9443 | if (!state->exists) | |
9444 | return -ENOENT; | |
9445 | if (!state->get_attr(name, dest)) | |
9446 | return -ENODATA; | |
9447 | ||
9448 | return 0; | |
9449 | } | |
9450 | ||
9451 | ||
9452 | int RGWRados::Object::Stat::stat_async() | |
9453 | { | |
9454 | RGWObjectCtx& ctx = source->get_ctx(); | |
9455 | rgw_obj& obj = source->get_obj(); | |
9456 | RGWRados *store = source->get_store(); | |
9457 | ||
9458 | RGWObjState *s = ctx.obj.get_state(obj); /* calling this one directly because otherwise a sync request will be sent */ | |
9459 | result.obj = obj; | |
9460 | if (s->has_attrs) { | |
9461 | state.ret = 0; | |
9462 | result.size = s->size; | |
9463 | result.mtime = ceph::real_clock::to_timespec(s->mtime); | |
9464 | result.attrs = s->attrset; | |
9465 | result.has_manifest = s->has_manifest; | |
9466 | result.manifest = s->manifest; | |
9467 | return 0; | |
9468 | } | |
9469 | ||
9470 | string oid; | |
9471 | string loc; | |
9472 | get_obj_bucket_and_oid_loc(obj, oid, loc); | |
9473 | ||
9474 | int r = store->get_obj_head_ioctx(source->get_bucket_info(), obj, &state.io_ctx); | |
9475 | if (r < 0) { | |
9476 | return r; | |
9477 | } | |
9478 | ||
9479 | librados::ObjectReadOperation op; | |
9480 | op.stat2(&result.size, &result.mtime, NULL); | |
9481 | op.getxattrs(&result.attrs, NULL); | |
9482 | state.completion = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
9483 | state.io_ctx.locator_set_key(loc); | |
9484 | r = state.io_ctx.aio_operate(oid, state.completion, &op, NULL); | |
9485 | if (r < 0) { | |
9486 | ldout(store->ctx(), 5) << __func__ | |
9487 | << ": ERROR: aio_operate() returned ret=" << r | |
9488 | << dendl; | |
9489 | return r; | |
9490 | } | |
9491 | ||
9492 | return 0; | |
9493 | } | |
9494 | ||
9495 | ||
9496 | int RGWRados::Object::Stat::wait() | |
9497 | { | |
9498 | if (!state.completion) { | |
9499 | return state.ret; | |
9500 | } | |
9501 | ||
9502 | state.completion->wait_for_safe(); | |
9503 | state.ret = state.completion->get_return_value(); | |
9504 | state.completion->release(); | |
9505 | ||
9506 | if (state.ret != 0) { | |
9507 | return state.ret; | |
9508 | } | |
9509 | ||
9510 | return finish(); | |
9511 | } | |
9512 | ||
9513 | int RGWRados::Object::Stat::finish() | |
9514 | { | |
9515 | map<string, bufferlist>::iterator iter = result.attrs.find(RGW_ATTR_MANIFEST); | |
9516 | if (iter != result.attrs.end()) { | |
9517 | bufferlist& bl = iter->second; | |
9518 | bufferlist::iterator biter = bl.begin(); | |
9519 | try { | |
9520 | ::decode(result.manifest, biter); | |
9521 | } catch (buffer::error& err) { | |
9522 | RGWRados *store = source->get_store(); | |
9523 | ldout(store->ctx(), 0) << "ERROR: " << __func__ << ": failed to decode manifest" << dendl; | |
9524 | return -EIO; | |
9525 | } | |
9526 | result.has_manifest = true; | |
9527 | } | |
9528 | ||
9529 | return 0; | |
9530 | } | |
9531 | ||
9532 | /** | |
31f18b77 FG |
9533 | * Get an attribute for a system object. |
9534 | * obj: the object to get attr | |
7c673cae FG |
9535 | * name: name of the attr to retrieve |
9536 | * dest: bufferlist to store the result in | |
9537 | * Returns: 0 on success, -ERR# otherwise. | |
9538 | */ | |
9539 | int RGWRados::system_obj_get_attr(rgw_raw_obj& obj, const char *name, bufferlist& dest) | |
9540 | { | |
9541 | rgw_rados_ref ref; | |
224ce89b | 9542 | int r = get_system_obj_ref(obj, &ref); |
7c673cae FG |
9543 | if (r < 0) { |
9544 | return r; | |
9545 | } | |
9546 | ||
9547 | ObjectReadOperation op; | |
9548 | ||
9549 | int rval; | |
9550 | op.getxattr(name, &dest, &rval); | |
9551 | ||
9552 | r = ref.ioctx.operate(ref.oid, &op, NULL); | |
9553 | if (r < 0) | |
9554 | return r; | |
9555 | ||
9556 | return 0; | |
9557 | } | |
9558 | ||
9559 | int RGWRados::append_atomic_test(RGWObjectCtx *rctx, | |
9560 | const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
9561 | ObjectOperation& op, RGWObjState **pstate) | |
9562 | { | |
9563 | if (!rctx) | |
9564 | return 0; | |
9565 | ||
9566 | int r = get_obj_state(rctx, bucket_info, obj, pstate, false); | |
9567 | if (r < 0) | |
9568 | return r; | |
9569 | ||
9570 | RGWObjState *state = *pstate; | |
9571 | ||
9572 | if (!state->is_atomic) { | |
9573 | ldout(cct, 20) << "state for obj=" << obj << " is not atomic, not appending atomic test" << dendl; | |
9574 | return 0; | |
9575 | } | |
9576 | ||
9577 | if (state->obj_tag.length() > 0 && !state->fake_tag) {// check for backward compatibility | |
9578 | op.cmpxattr(RGW_ATTR_ID_TAG, LIBRADOS_CMPXATTR_OP_EQ, state->obj_tag); | |
9579 | } else { | |
9580 | ldout(cct, 20) << "state->obj_tag is empty, not appending atomic test" << dendl; | |
9581 | } | |
9582 | return 0; | |
9583 | } | |
9584 | ||
9585 | int RGWRados::Object::get_state(RGWObjState **pstate, bool follow_olh, bool assume_noent) | |
9586 | { | |
9587 | return store->get_obj_state(&ctx, bucket_info, obj, pstate, follow_olh, assume_noent); | |
9588 | } | |
9589 | ||
9590 | void RGWRados::Object::invalidate_state() | |
9591 | { | |
9592 | ctx.obj.invalidate(obj); | |
9593 | } | |
9594 | ||
9595 | void RGWRados::SystemObject::invalidate_state() | |
9596 | { | |
9597 | ctx.raw.invalidate(obj); | |
9598 | } | |
9599 | ||
9600 | int RGWRados::Object::prepare_atomic_modification(ObjectWriteOperation& op, bool reset_obj, const string *ptag, | |
181888fb FG |
9601 | const char *if_match, const char *if_nomatch, bool removal_op, |
9602 | bool modify_tail) | |
7c673cae FG |
9603 | { |
9604 | int r = get_state(&state, false); | |
9605 | if (r < 0) | |
9606 | return r; | |
9607 | ||
9608 | bool need_guard = (state->has_manifest || (state->obj_tag.length() != 0) || | |
9609 | if_match != NULL || if_nomatch != NULL) && | |
9610 | (!state->fake_tag); | |
9611 | ||
9612 | if (!state->is_atomic) { | |
9613 | ldout(store->ctx(), 20) << "prepare_atomic_modification: state is not atomic. state=" << (void *)state << dendl; | |
9614 | ||
9615 | if (reset_obj) { | |
9616 | op.create(false); | |
9617 | store->remove_rgw_head_obj(op); // we're not dropping reference here, actually removing object | |
9618 | } | |
9619 | ||
9620 | return 0; | |
9621 | } | |
9622 | ||
9623 | if (need_guard) { | |
9624 | /* first verify that the object wasn't replaced under */ | |
9625 | if (if_nomatch == NULL || strcmp(if_nomatch, "*") != 0) { | |
9626 | op.cmpxattr(RGW_ATTR_ID_TAG, LIBRADOS_CMPXATTR_OP_EQ, state->obj_tag); | |
9627 | // FIXME: need to add FAIL_NOTEXIST_OK for racing deletion | |
9628 | } | |
9629 | ||
9630 | if (if_match) { | |
9631 | if (strcmp(if_match, "*") == 0) { | |
9632 | // test the object is existing | |
9633 | if (!state->exists) { | |
9634 | return -ERR_PRECONDITION_FAILED; | |
9635 | } | |
9636 | } else { | |
9637 | bufferlist bl; | |
9638 | if (!state->get_attr(RGW_ATTR_ETAG, bl) || | |
9639 | strncmp(if_match, bl.c_str(), bl.length()) != 0) { | |
9640 | return -ERR_PRECONDITION_FAILED; | |
9641 | } | |
9642 | } | |
9643 | } | |
9644 | ||
9645 | if (if_nomatch) { | |
9646 | if (strcmp(if_nomatch, "*") == 0) { | |
9647 | // test the object is NOT existing | |
9648 | if (state->exists) { | |
9649 | return -ERR_PRECONDITION_FAILED; | |
9650 | } | |
9651 | } else { | |
9652 | bufferlist bl; | |
9653 | if (!state->get_attr(RGW_ATTR_ETAG, bl) || | |
9654 | strncmp(if_nomatch, bl.c_str(), bl.length()) == 0) { | |
9655 | return -ERR_PRECONDITION_FAILED; | |
9656 | } | |
9657 | } | |
9658 | } | |
9659 | } | |
9660 | ||
9661 | if (reset_obj) { | |
9662 | if (state->exists) { | |
9663 | op.create(false); | |
9664 | store->remove_rgw_head_obj(op); | |
9665 | } else { | |
9666 | op.create(true); | |
9667 | } | |
9668 | } | |
9669 | ||
9670 | if (removal_op) { | |
9671 | /* the object is being removed, no need to update its tag */ | |
9672 | return 0; | |
9673 | } | |
9674 | ||
9675 | if (ptag) { | |
9676 | state->write_tag = *ptag; | |
9677 | } else { | |
9678 | append_rand_alpha(store->ctx(), state->write_tag, state->write_tag, 32); | |
9679 | } | |
9680 | bufferlist bl; | |
9681 | bl.append(state->write_tag.c_str(), state->write_tag.size() + 1); | |
9682 | ||
9683 | ldout(store->ctx(), 10) << "setting object write_tag=" << state->write_tag << dendl; | |
9684 | ||
9685 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
181888fb FG |
9686 | if (modify_tail) { |
9687 | op.setxattr(RGW_ATTR_TAIL_TAG, bl); | |
9688 | } | |
7c673cae FG |
9689 | |
9690 | return 0; | |
9691 | } | |
9692 | ||
9693 | int RGWRados::system_obj_set_attr(void *ctx, rgw_raw_obj& obj, const char *name, bufferlist& bl, | |
9694 | RGWObjVersionTracker *objv_tracker) | |
9695 | { | |
9696 | map<string, bufferlist> attrs; | |
9697 | attrs[name] = bl; | |
9698 | return system_obj_set_attrs(ctx, obj, attrs, NULL, objv_tracker); | |
9699 | } | |
9700 | ||
9701 | int RGWRados::system_obj_set_attrs(void *ctx, rgw_raw_obj& obj, | |
9702 | map<string, bufferlist>& attrs, | |
9703 | map<string, bufferlist>* rmattrs, | |
9704 | RGWObjVersionTracker *objv_tracker) | |
9705 | { | |
9706 | rgw_rados_ref ref; | |
224ce89b | 9707 | int r = get_system_obj_ref(obj, &ref); |
7c673cae FG |
9708 | if (r < 0) { |
9709 | return r; | |
9710 | } | |
9711 | ObjectWriteOperation op; | |
9712 | ||
9713 | if (objv_tracker) { | |
9714 | objv_tracker->prepare_op_for_write(&op); | |
9715 | } | |
9716 | ||
9717 | map<string, bufferlist>::iterator iter; | |
9718 | if (rmattrs) { | |
9719 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
9720 | const string& name = iter->first; | |
9721 | op.rmxattr(name.c_str()); | |
9722 | } | |
9723 | } | |
9724 | ||
9725 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
9726 | const string& name = iter->first; | |
9727 | bufferlist& bl = iter->second; | |
9728 | ||
9729 | if (!bl.length()) | |
9730 | continue; | |
9731 | ||
9732 | op.setxattr(name.c_str(), bl); | |
9733 | } | |
9734 | ||
9735 | if (!op.size()) | |
9736 | return 0; | |
9737 | ||
9738 | bufferlist bl; | |
9739 | ||
9740 | r = ref.ioctx.operate(ref.oid, &op); | |
9741 | if (r < 0) | |
9742 | return r; | |
9743 | ||
9744 | return 0; | |
9745 | } | |
9746 | ||
9747 | /** | |
9748 | * Set an attr on an object. | |
9749 | * bucket: name of the bucket holding the object | |
9750 | * obj: name of the object to set the attr on | |
9751 | * name: the attr to set | |
9752 | * bl: the contents of the attr | |
9753 | * Returns: 0 on success, -ERR# otherwise. | |
9754 | */ | |
9755 | int RGWRados::set_attr(void *ctx, const RGWBucketInfo& bucket_info, rgw_obj& obj, const char *name, bufferlist& bl) | |
9756 | { | |
9757 | map<string, bufferlist> attrs; | |
9758 | attrs[name] = bl; | |
9759 | return set_attrs(ctx, bucket_info, obj, attrs, NULL); | |
9760 | } | |
9761 | ||
9762 | int RGWRados::set_attrs(void *ctx, const RGWBucketInfo& bucket_info, rgw_obj& obj, | |
9763 | map<string, bufferlist>& attrs, | |
9764 | map<string, bufferlist>* rmattrs) | |
9765 | { | |
9766 | rgw_rados_ref ref; | |
9767 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
9768 | if (r < 0) { | |
9769 | return r; | |
9770 | } | |
9771 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
9772 | ||
9773 | ObjectWriteOperation op; | |
9774 | RGWObjState *state = NULL; | |
9775 | ||
9776 | r = append_atomic_test(rctx, bucket_info, obj, op, &state); | |
9777 | if (r < 0) | |
9778 | return r; | |
9779 | ||
9780 | map<string, bufferlist>::iterator iter; | |
9781 | if (rmattrs) { | |
9782 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
9783 | const string& name = iter->first; | |
9784 | op.rmxattr(name.c_str()); | |
9785 | } | |
9786 | } | |
9787 | ||
9788 | const rgw_bucket& bucket = obj.bucket; | |
9789 | ||
9790 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
9791 | const string& name = iter->first; | |
9792 | bufferlist& bl = iter->second; | |
9793 | ||
9794 | if (!bl.length()) | |
9795 | continue; | |
9796 | ||
9797 | op.setxattr(name.c_str(), bl); | |
9798 | ||
9799 | if (name.compare(RGW_ATTR_DELETE_AT) == 0) { | |
9800 | real_time ts; | |
9801 | try { | |
9802 | ::decode(ts, bl); | |
9803 | ||
9804 | rgw_obj_index_key obj_key; | |
9805 | obj.key.get_index_key(&obj_key); | |
9806 | ||
9807 | objexp_hint_add(ts, bucket.tenant, bucket.name, bucket.bucket_id, obj_key); | |
9808 | } catch (buffer::error& err) { | |
9809 | ldout(cct, 0) << "ERROR: failed to decode " RGW_ATTR_DELETE_AT << " attr" << dendl; | |
9810 | } | |
9811 | } | |
9812 | } | |
9813 | ||
9814 | if (!op.size()) | |
9815 | return 0; | |
9816 | ||
9817 | RGWObjectCtx obj_ctx(this); | |
9818 | ||
9819 | bufferlist bl; | |
9820 | RGWRados::Bucket bop(this, bucket_info); | |
9821 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
9822 | ||
9823 | if (state) { | |
9824 | string tag; | |
9825 | append_rand_alpha(cct, tag, tag, 32); | |
9826 | state->write_tag = tag; | |
9827 | r = index_op.prepare(CLS_RGW_OP_ADD, &state->write_tag); | |
9828 | ||
9829 | if (r < 0) | |
9830 | return r; | |
9831 | ||
9832 | bl.append(tag.c_str(), tag.size() + 1); | |
7c673cae FG |
9833 | op.setxattr(RGW_ATTR_ID_TAG, bl); |
9834 | } | |
9835 | ||
3efd9988 FG |
9836 | |
9837 | real_time mtime = real_clock::now(); | |
9838 | struct timespec mtime_ts = real_clock::to_timespec(mtime); | |
9839 | op.mtime2(&mtime_ts); | |
7c673cae FG |
9840 | r = ref.ioctx.operate(ref.oid, &op); |
9841 | if (state) { | |
9842 | if (r >= 0) { | |
9843 | bufferlist acl_bl = attrs[RGW_ATTR_ACL]; | |
9844 | bufferlist etag_bl = attrs[RGW_ATTR_ETAG]; | |
9845 | bufferlist content_type_bl = attrs[RGW_ATTR_CONTENT_TYPE]; | |
9846 | string etag(etag_bl.c_str(), etag_bl.length()); | |
9847 | string content_type(content_type_bl.c_str(), content_type_bl.length()); | |
9848 | uint64_t epoch = ref.ioctx.get_last_version(); | |
9849 | int64_t poolid = ref.ioctx.get_id(); | |
7c673cae FG |
9850 | r = index_op.complete(poolid, epoch, state->size, state->accounted_size, |
9851 | mtime, etag, content_type, &acl_bl, | |
9852 | RGW_OBJ_CATEGORY_MAIN, NULL); | |
9853 | } else { | |
9854 | int ret = index_op.cancel(); | |
9855 | if (ret < 0) { | |
9856 | ldout(cct, 0) << "ERROR: complete_update_index_cancel() returned ret=" << ret << dendl; | |
9857 | } | |
9858 | } | |
9859 | } | |
9860 | if (r < 0) | |
9861 | return r; | |
9862 | ||
9863 | if (state) { | |
9864 | state->obj_tag.swap(bl); | |
9865 | if (rmattrs) { | |
9866 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
9867 | state->attrset.erase(iter->first); | |
9868 | } | |
9869 | } | |
9870 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
9871 | state->attrset[iter->first] = iter->second; | |
9872 | } | |
9873 | } | |
9874 | ||
9875 | return 0; | |
9876 | } | |
9877 | ||
7c673cae FG |
9878 | int RGWRados::Object::Read::prepare() |
9879 | { | |
9880 | RGWRados *store = source->get_store(); | |
9881 | CephContext *cct = store->ctx(); | |
9882 | ||
9883 | bufferlist etag; | |
9884 | ||
9885 | map<string, bufferlist>::iterator iter; | |
9886 | ||
9887 | RGWObjState *astate; | |
9888 | int r = source->get_state(&astate, true); | |
9889 | if (r < 0) | |
9890 | return r; | |
9891 | ||
9892 | if (!astate->exists) { | |
9893 | return -ENOENT; | |
9894 | } | |
9895 | ||
9896 | const RGWBucketInfo& bucket_info = source->get_bucket_info(); | |
9897 | ||
9898 | state.obj = astate->obj; | |
9899 | store->obj_to_raw(bucket_info.placement_rule, state.obj, &state.head_obj); | |
9900 | ||
9901 | r = store->get_obj_head_ioctx(bucket_info, state.obj, &state.io_ctx); | |
9902 | if (r < 0) { | |
9903 | return r; | |
9904 | } | |
9905 | if (params.attrs) { | |
9906 | *params.attrs = astate->attrset; | |
9907 | if (cct->_conf->subsys.should_gather(ceph_subsys_rgw, 20)) { | |
9908 | for (iter = params.attrs->begin(); iter != params.attrs->end(); ++iter) { | |
9909 | ldout(cct, 20) << "Read xattr: " << iter->first << dendl; | |
9910 | } | |
9911 | } | |
9912 | } | |
9913 | ||
9914 | /* Convert all times go GMT to make them compatible */ | |
9915 | if (conds.mod_ptr || conds.unmod_ptr) { | |
9916 | obj_time_weight src_weight; | |
9917 | src_weight.init(astate); | |
9918 | src_weight.high_precision = conds.high_precision_time; | |
9919 | ||
9920 | obj_time_weight dest_weight; | |
9921 | dest_weight.high_precision = conds.high_precision_time; | |
9922 | ||
9923 | if (conds.mod_ptr) { | |
9924 | dest_weight.init(*conds.mod_ptr, conds.mod_zone_id, conds.mod_pg_ver); | |
9925 | ldout(cct, 10) << "If-Modified-Since: " << dest_weight << " Last-Modified: " << src_weight << dendl; | |
9926 | if (!(dest_weight < src_weight)) { | |
9927 | return -ERR_NOT_MODIFIED; | |
9928 | } | |
9929 | } | |
9930 | ||
9931 | if (conds.unmod_ptr) { | |
9932 | dest_weight.init(*conds.unmod_ptr, conds.mod_zone_id, conds.mod_pg_ver); | |
9933 | ldout(cct, 10) << "If-UnModified-Since: " << dest_weight << " Last-Modified: " << src_weight << dendl; | |
9934 | if (dest_weight < src_weight) { | |
9935 | return -ERR_PRECONDITION_FAILED; | |
9936 | } | |
9937 | } | |
9938 | } | |
9939 | if (conds.if_match || conds.if_nomatch) { | |
9940 | r = get_attr(RGW_ATTR_ETAG, etag); | |
9941 | if (r < 0) | |
9942 | return r; | |
9943 | ||
9944 | if (conds.if_match) { | |
9945 | string if_match_str = rgw_string_unquote(conds.if_match); | |
9946 | ldout(cct, 10) << "ETag: " << etag.c_str() << " " << " If-Match: " << if_match_str << dendl; | |
9947 | if (if_match_str.compare(etag.c_str()) != 0) { | |
9948 | return -ERR_PRECONDITION_FAILED; | |
9949 | } | |
9950 | } | |
9951 | ||
9952 | if (conds.if_nomatch) { | |
9953 | string if_nomatch_str = rgw_string_unquote(conds.if_nomatch); | |
9954 | ldout(cct, 10) << "ETag: " << etag.c_str() << " " << " If-NoMatch: " << if_nomatch_str << dendl; | |
9955 | if (if_nomatch_str.compare(etag.c_str()) == 0) { | |
9956 | return -ERR_NOT_MODIFIED; | |
9957 | } | |
9958 | } | |
9959 | } | |
9960 | ||
9961 | if (params.obj_size) | |
9962 | *params.obj_size = astate->size; | |
9963 | if (params.lastmod) | |
9964 | *params.lastmod = astate->mtime; | |
9965 | ||
9966 | return 0; | |
9967 | } | |
9968 | ||
9969 | int RGWRados::Object::Read::range_to_ofs(uint64_t obj_size, int64_t &ofs, int64_t &end) | |
9970 | { | |
9971 | if (ofs < 0) { | |
9972 | ofs += obj_size; | |
9973 | if (ofs < 0) | |
9974 | ofs = 0; | |
9975 | end = obj_size - 1; | |
9976 | } else if (end < 0) { | |
9977 | end = obj_size - 1; | |
9978 | } | |
9979 | ||
9980 | if (obj_size > 0) { | |
9981 | if (ofs >= (off_t)obj_size) { | |
9982 | return -ERANGE; | |
9983 | } | |
9984 | if (end >= (off_t)obj_size) { | |
9985 | end = obj_size - 1; | |
9986 | } | |
9987 | } | |
9988 | return 0; | |
9989 | } | |
9990 | ||
9991 | int RGWRados::SystemObject::get_state(RGWRawObjState **pstate, RGWObjVersionTracker *objv_tracker) | |
9992 | { | |
9993 | return store->get_system_obj_state(&ctx, obj, pstate, objv_tracker); | |
9994 | } | |
9995 | ||
9996 | int RGWRados::stat_system_obj(RGWObjectCtx& obj_ctx, | |
9997 | RGWRados::SystemObject::Read::GetObjState& state, | |
9998 | rgw_raw_obj& obj, | |
9999 | map<string, bufferlist> *attrs, | |
10000 | real_time *lastmod, | |
10001 | uint64_t *obj_size, | |
10002 | RGWObjVersionTracker *objv_tracker) | |
10003 | { | |
10004 | RGWRawObjState *astate = NULL; | |
10005 | ||
10006 | int r = get_system_obj_state(&obj_ctx, obj, &astate, objv_tracker); | |
10007 | if (r < 0) | |
10008 | return r; | |
10009 | ||
10010 | if (!astate->exists) { | |
10011 | return -ENOENT; | |
10012 | } | |
10013 | ||
10014 | if (attrs) { | |
10015 | *attrs = astate->attrset; | |
10016 | if (cct->_conf->subsys.should_gather(ceph_subsys_rgw, 20)) { | |
10017 | map<string, bufferlist>::iterator iter; | |
10018 | for (iter = attrs->begin(); iter != attrs->end(); ++iter) { | |
10019 | ldout(cct, 20) << "Read xattr: " << iter->first << dendl; | |
10020 | } | |
10021 | } | |
10022 | } | |
10023 | ||
10024 | if (obj_size) | |
10025 | *obj_size = astate->size; | |
10026 | if (lastmod) | |
10027 | *lastmod = astate->mtime; | |
10028 | ||
10029 | return 0; | |
10030 | } | |
10031 | ||
31f18b77 FG |
10032 | |
10033 | int RGWRados::Bucket::UpdateIndex::guard_reshard(BucketShard **pbs, std::function<int(BucketShard *)> call) | |
10034 | { | |
10035 | RGWRados *store = target->get_store(); | |
10036 | BucketShard *bs; | |
10037 | int r; | |
10038 | ||
10039 | #define NUM_RESHARD_RETRIES 10 | |
10040 | for (int i = 0; i < NUM_RESHARD_RETRIES; ++i) { | |
10041 | int ret = get_bucket_shard(&bs); | |
10042 | if (ret < 0) { | |
10043 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
10044 | return ret; | |
10045 | } | |
10046 | r = call(bs); | |
10047 | if (r != -ERR_BUSY_RESHARDING) { | |
10048 | break; | |
10049 | } | |
10050 | ldout(store->ctx(), 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl; | |
10051 | string new_bucket_id; | |
10052 | r = store->block_while_resharding(bs, &new_bucket_id); | |
10053 | if (r == -ERR_BUSY_RESHARDING) { | |
10054 | continue; | |
10055 | } | |
10056 | if (r < 0) { | |
10057 | return r; | |
10058 | } | |
10059 | ldout(store->ctx(), 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id << dendl; | |
10060 | i = 0; /* resharding is finished, make sure we can retry */ | |
10061 | r = target->update_bucket_id(new_bucket_id); | |
10062 | if (r < 0) { | |
10063 | ldout(store->ctx(), 0) << "ERROR: update_bucket_id() new_bucket_id=" << new_bucket_id << " returned r=" << r << dendl; | |
10064 | return r; | |
10065 | } | |
10066 | invalidate_bs(); | |
10067 | } | |
10068 | ||
10069 | if (r < 0) { | |
10070 | return r; | |
10071 | } | |
10072 | ||
10073 | if (pbs) { | |
10074 | *pbs = bs; | |
10075 | } | |
10076 | ||
10077 | return 0; | |
10078 | } | |
10079 | ||
7c673cae FG |
10080 | int RGWRados::SystemObject::Read::stat(RGWObjVersionTracker *objv_tracker) |
10081 | { | |
10082 | RGWRados *store = source->get_store(); | |
10083 | rgw_raw_obj& obj = source->get_obj(); | |
10084 | ||
10085 | return store->stat_system_obj(source->get_ctx(), state, obj, stat_params.attrs, | |
10086 | stat_params.lastmod, stat_params.obj_size, objv_tracker); | |
10087 | } | |
10088 | ||
10089 | int RGWRados::Bucket::UpdateIndex::prepare(RGWModifyOp op, const string *write_tag) | |
10090 | { | |
10091 | if (blind) { | |
10092 | return 0; | |
10093 | } | |
10094 | RGWRados *store = target->get_store(); | |
7c673cae FG |
10095 | |
10096 | if (write_tag && write_tag->length()) { | |
10097 | optag = string(write_tag->c_str(), write_tag->length()); | |
10098 | } else { | |
10099 | if (optag.empty()) { | |
10100 | append_rand_alpha(store->ctx(), optag, optag, 32); | |
10101 | } | |
10102 | } | |
10103 | ||
31f18b77 FG |
10104 | int r = guard_reshard(nullptr, [&](BucketShard *bs) -> int { |
10105 | return store->cls_obj_prepare_op(*bs, op, optag, obj, bilog_flags, zones_trace); | |
10106 | }); | |
10107 | ||
7c673cae FG |
10108 | if (r < 0) { |
10109 | return r; | |
10110 | } | |
10111 | prepared = true; | |
31f18b77 | 10112 | |
7c673cae FG |
10113 | return 0; |
10114 | } | |
10115 | ||
10116 | int RGWRados::Bucket::UpdateIndex::complete(int64_t poolid, uint64_t epoch, | |
10117 | uint64_t size, uint64_t accounted_size, | |
10118 | ceph::real_time& ut, const string& etag, | |
10119 | const string& content_type, | |
10120 | bufferlist *acl_bl, | |
10121 | RGWObjCategory category, | |
10122 | list<rgw_obj_index_key> *remove_objs, const string *user_data) | |
10123 | { | |
10124 | if (blind) { | |
10125 | return 0; | |
10126 | } | |
10127 | RGWRados *store = target->get_store(); | |
10128 | BucketShard *bs; | |
31f18b77 | 10129 | |
7c673cae FG |
10130 | int ret = get_bucket_shard(&bs); |
10131 | if (ret < 0) { | |
10132 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
10133 | return ret; | |
10134 | } | |
10135 | ||
10136 | rgw_bucket_dir_entry ent; | |
10137 | obj.key.get_index_key(&ent.key); | |
10138 | ent.meta.size = size; | |
10139 | ent.meta.accounted_size = accounted_size; | |
10140 | ent.meta.mtime = ut; | |
10141 | ent.meta.etag = etag; | |
10142 | if (user_data) | |
10143 | ent.meta.user_data = *user_data; | |
10144 | ||
10145 | ACLOwner owner; | |
10146 | if (acl_bl && acl_bl->length()) { | |
10147 | int ret = store->decode_policy(*acl_bl, &owner); | |
10148 | if (ret < 0) { | |
10149 | ldout(store->ctx(), 0) << "WARNING: could not decode policy ret=" << ret << dendl; | |
10150 | } | |
10151 | } | |
10152 | ent.meta.owner = owner.get_id().to_str(); | |
10153 | ent.meta.owner_display_name = owner.get_display_name(); | |
10154 | ent.meta.content_type = content_type; | |
10155 | ||
31f18b77 | 10156 | ret = store->cls_obj_complete_add(*bs, obj, optag, poolid, epoch, ent, category, remove_objs, bilog_flags, zones_trace); |
7c673cae | 10157 | |
c07f9fc5 FG |
10158 | if (target->bucket_info.datasync_flag_enabled()) { |
10159 | int r = store->data_log->add_entry(bs->bucket, bs->shard_id); | |
10160 | if (r < 0) { | |
10161 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
10162 | } | |
7c673cae FG |
10163 | } |
10164 | ||
10165 | return ret; | |
10166 | } | |
10167 | ||
10168 | int RGWRados::Bucket::UpdateIndex::complete_del(int64_t poolid, uint64_t epoch, | |
10169 | real_time& removed_mtime, | |
10170 | list<rgw_obj_index_key> *remove_objs) | |
10171 | { | |
10172 | if (blind) { | |
10173 | return 0; | |
10174 | } | |
10175 | RGWRados *store = target->get_store(); | |
10176 | BucketShard *bs; | |
31f18b77 | 10177 | |
7c673cae FG |
10178 | int ret = get_bucket_shard(&bs); |
10179 | if (ret < 0) { | |
10180 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
10181 | return ret; | |
10182 | } | |
10183 | ||
31f18b77 | 10184 | ret = store->cls_obj_complete_del(*bs, optag, poolid, epoch, obj, removed_mtime, remove_objs, bilog_flags, zones_trace); |
7c673cae | 10185 | |
c07f9fc5 FG |
10186 | if (target->bucket_info.datasync_flag_enabled()) { |
10187 | int r = store->data_log->add_entry(bs->bucket, bs->shard_id); | |
10188 | if (r < 0) { | |
10189 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
10190 | } | |
7c673cae FG |
10191 | } |
10192 | ||
10193 | return ret; | |
10194 | } | |
10195 | ||
10196 | ||
10197 | int RGWRados::Bucket::UpdateIndex::cancel() | |
10198 | { | |
10199 | if (blind) { | |
10200 | return 0; | |
10201 | } | |
10202 | RGWRados *store = target->get_store(); | |
10203 | BucketShard *bs; | |
7c673cae | 10204 | |
31f18b77 FG |
10205 | int ret = guard_reshard(&bs, [&](BucketShard *bs) -> int { |
10206 | return store->cls_obj_complete_cancel(*bs, optag, obj, bilog_flags, zones_trace); | |
10207 | }); | |
7c673cae FG |
10208 | |
10209 | /* | |
10210 | * need to update data log anyhow, so that whoever follows needs to update its internal markers | |
10211 | * for following the specific bucket shard log. Otherwise they end up staying behind, and users | |
10212 | * have no way to tell that they're all caught up | |
10213 | */ | |
c07f9fc5 FG |
10214 | if (target->bucket_info.datasync_flag_enabled()) { |
10215 | int r = store->data_log->add_entry(bs->bucket, bs->shard_id); | |
10216 | if (r < 0) { | |
10217 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
10218 | } | |
7c673cae FG |
10219 | } |
10220 | ||
10221 | return ret; | |
10222 | } | |
10223 | ||
10224 | int RGWRados::Object::Read::read(int64_t ofs, int64_t end, bufferlist& bl) | |
10225 | { | |
10226 | RGWRados *store = source->get_store(); | |
10227 | CephContext *cct = store->ctx(); | |
10228 | ||
7c673cae FG |
10229 | rgw_raw_obj read_obj; |
10230 | uint64_t read_ofs = ofs; | |
10231 | uint64_t len, read_len; | |
10232 | bool reading_from_head = true; | |
10233 | ObjectReadOperation op; | |
10234 | ||
10235 | bool merge_bl = false; | |
10236 | bufferlist *pbl = &bl; | |
10237 | bufferlist read_bl; | |
10238 | uint64_t max_chunk_size; | |
10239 | ||
10240 | RGWObjState *astate; | |
10241 | int r = source->get_state(&astate, true); | |
10242 | if (r < 0) | |
10243 | return r; | |
10244 | ||
10245 | if (end < 0) | |
10246 | len = 0; | |
10247 | else | |
10248 | len = end - ofs + 1; | |
10249 | ||
10250 | if (astate->has_manifest && astate->manifest.has_tail()) { | |
10251 | /* now get the relevant object part */ | |
10252 | RGWObjManifest::obj_iterator iter = astate->manifest.obj_find(ofs); | |
10253 | ||
10254 | uint64_t stripe_ofs = iter.get_stripe_ofs(); | |
10255 | read_obj = iter.get_location().get_raw_obj(store); | |
10256 | len = min(len, iter.get_stripe_size() - (ofs - stripe_ofs)); | |
10257 | read_ofs = iter.location_ofs() + (ofs - stripe_ofs); | |
10258 | reading_from_head = (read_obj == state.head_obj); | |
10259 | } else { | |
10260 | read_obj = state.head_obj; | |
10261 | } | |
10262 | ||
10263 | r = store->get_max_chunk_size(read_obj.pool, &max_chunk_size); | |
10264 | if (r < 0) { | |
10265 | ldout(cct, 0) << "ERROR: failed to get max_chunk_size() for pool " << read_obj.pool << dendl; | |
10266 | return r; | |
10267 | } | |
10268 | ||
10269 | if (len > max_chunk_size) | |
10270 | len = max_chunk_size; | |
10271 | ||
10272 | ||
10273 | state.io_ctx.locator_set_key(read_obj.loc); | |
10274 | ||
10275 | read_len = len; | |
10276 | ||
10277 | if (reading_from_head) { | |
10278 | /* only when reading from the head object do we need to do the atomic test */ | |
10279 | r = store->append_atomic_test(&source->get_ctx(), source->get_bucket_info(), state.obj, op, &astate); | |
10280 | if (r < 0) | |
10281 | return r; | |
10282 | ||
10283 | if (astate && astate->prefetch_data) { | |
10284 | if (!ofs && astate->data.length() >= len) { | |
10285 | bl = astate->data; | |
10286 | return bl.length(); | |
10287 | } | |
10288 | ||
10289 | if (ofs < astate->data.length()) { | |
10290 | unsigned copy_len = min((uint64_t)astate->data.length() - ofs, len); | |
10291 | astate->data.copy(ofs, copy_len, bl); | |
10292 | read_len -= copy_len; | |
10293 | read_ofs += copy_len; | |
10294 | if (!read_len) | |
10295 | return bl.length(); | |
10296 | ||
10297 | merge_bl = true; | |
10298 | pbl = &read_bl; | |
10299 | } | |
10300 | } | |
10301 | } | |
10302 | ||
10303 | ldout(cct, 20) << "rados->read obj-ofs=" << ofs << " read_ofs=" << read_ofs << " read_len=" << read_len << dendl; | |
10304 | op.read(read_ofs, read_len, pbl, NULL); | |
10305 | ||
10306 | r = state.io_ctx.operate(read_obj.oid, &op, NULL); | |
10307 | ldout(cct, 20) << "rados->read r=" << r << " bl.length=" << bl.length() << dendl; | |
10308 | ||
10309 | if (r < 0) { | |
10310 | return r; | |
10311 | } | |
10312 | ||
10313 | if (merge_bl) { | |
10314 | bl.append(read_bl); | |
10315 | } | |
10316 | ||
10317 | return bl.length(); | |
10318 | } | |
10319 | ||
10320 | int RGWRados::SystemObject::Read::GetObjState::get_ref(RGWRados *store, rgw_raw_obj& obj, rgw_rados_ref **pref) | |
10321 | { | |
10322 | if (!has_ref) { | |
224ce89b | 10323 | int r = store->get_raw_obj_ref(obj, &ref); |
7c673cae FG |
10324 | if (r < 0) { |
10325 | return r; | |
10326 | } | |
10327 | has_ref = true; | |
10328 | } | |
10329 | *pref = &ref; | |
10330 | return 0; | |
10331 | ||
10332 | } | |
10333 | ||
10334 | int RGWRados::get_system_obj(RGWObjectCtx& obj_ctx, RGWRados::SystemObject::Read::GetObjState& read_state, | |
10335 | RGWObjVersionTracker *objv_tracker, rgw_raw_obj& obj, | |
10336 | bufferlist& bl, off_t ofs, off_t end, | |
10337 | map<string, bufferlist> *attrs, | |
b32b8144 FG |
10338 | rgw_cache_entry_info *cache_info, |
10339 | boost::optional<obj_version>) | |
7c673cae FG |
10340 | { |
10341 | uint64_t len; | |
10342 | ObjectReadOperation op; | |
10343 | ||
10344 | if (end < 0) | |
10345 | len = 0; | |
10346 | else | |
10347 | len = end - ofs + 1; | |
10348 | ||
10349 | if (objv_tracker) { | |
10350 | objv_tracker->prepare_op_for_read(&op); | |
10351 | } | |
10352 | ||
10353 | ldout(cct, 20) << "rados->read ofs=" << ofs << " len=" << len << dendl; | |
10354 | op.read(ofs, len, &bl, NULL); | |
10355 | ||
10356 | if (attrs) { | |
10357 | op.getxattrs(attrs, NULL); | |
10358 | } | |
10359 | ||
10360 | rgw_rados_ref *ref; | |
10361 | int r = read_state.get_ref(this, obj, &ref); | |
10362 | if (r < 0) { | |
10363 | ldout(cct, 20) << "read_state.get_ref() on obj=" << obj << " returned " << r << dendl; | |
10364 | return r; | |
10365 | } | |
10366 | r = ref->ioctx.operate(ref->oid, &op, NULL); | |
10367 | if (r < 0) { | |
10368 | ldout(cct, 20) << "rados->read r=" << r << " bl.length=" << bl.length() << dendl; | |
10369 | return r; | |
10370 | } | |
10371 | ldout(cct, 20) << "rados->read r=" << r << " bl.length=" << bl.length() << dendl; | |
10372 | ||
10373 | uint64_t op_ver = ref->ioctx.get_last_version(); | |
10374 | ||
10375 | if (read_state.last_ver > 0 && | |
10376 | read_state.last_ver != op_ver) { | |
10377 | ldout(cct, 5) << "raced with an object write, abort" << dendl; | |
10378 | return -ECANCELED; | |
10379 | } | |
10380 | ||
10381 | read_state.last_ver = op_ver; | |
10382 | ||
10383 | return bl.length(); | |
10384 | } | |
10385 | ||
b32b8144 FG |
10386 | int RGWRados::SystemObject::Read::read(int64_t ofs, int64_t end, bufferlist& bl, |
10387 | RGWObjVersionTracker *objv_tracker, | |
10388 | boost::optional<obj_version> refresh_version) | |
7c673cae FG |
10389 | { |
10390 | RGWRados *store = source->get_store(); | |
10391 | rgw_raw_obj& obj = source->get_obj(); | |
10392 | ||
b32b8144 FG |
10393 | return store->get_system_obj(source->get_ctx(), state, objv_tracker, obj, bl, |
10394 | ofs, end, read_params.attrs, | |
10395 | read_params.cache_info, refresh_version); | |
7c673cae FG |
10396 | } |
10397 | ||
10398 | int RGWRados::SystemObject::Read::get_attr(const char *name, bufferlist& dest) | |
10399 | { | |
10400 | RGWRados *store = source->get_store(); | |
10401 | rgw_raw_obj& obj = source->get_obj(); | |
10402 | ||
10403 | return store->system_obj_get_attr(obj, name, dest); | |
10404 | } | |
10405 | ||
10406 | struct get_obj_data; | |
10407 | ||
10408 | struct get_obj_aio_data { | |
10409 | struct get_obj_data *op_data; | |
10410 | off_t ofs; | |
10411 | off_t len; | |
10412 | }; | |
10413 | ||
10414 | struct get_obj_io { | |
10415 | off_t len; | |
10416 | bufferlist bl; | |
10417 | }; | |
10418 | ||
10419 | static void _get_obj_aio_completion_cb(completion_t cb, void *arg); | |
10420 | ||
10421 | struct get_obj_data : public RefCountedObject { | |
10422 | CephContext *cct; | |
10423 | RGWRados *rados; | |
10424 | RGWObjectCtx *ctx; | |
10425 | IoCtx io_ctx; | |
10426 | map<off_t, get_obj_io> io_map; | |
10427 | map<off_t, librados::AioCompletion *> completion_map; | |
10428 | uint64_t total_read; | |
10429 | Mutex lock; | |
10430 | Mutex data_lock; | |
10431 | list<get_obj_aio_data> aio_data; | |
10432 | RGWGetDataCB *client_cb; | |
10433 | std::atomic<bool> cancelled = { false }; | |
10434 | std::atomic<int64_t> err_code = { 0 }; | |
10435 | Throttle throttle; | |
10436 | list<bufferlist> read_list; | |
10437 | ||
10438 | explicit get_obj_data(CephContext *_cct) | |
10439 | : cct(_cct), | |
10440 | rados(NULL), ctx(NULL), | |
10441 | total_read(0), lock("get_obj_data"), data_lock("get_obj_data::data_lock"), | |
10442 | client_cb(NULL), | |
10443 | throttle(cct, "get_obj_data", cct->_conf->rgw_get_obj_window_size, false) {} | |
10444 | ~get_obj_data() override { } | |
10445 | void set_cancelled(int r) { | |
10446 | cancelled = true; | |
10447 | err_code = r; | |
10448 | } | |
10449 | ||
10450 | bool is_cancelled() { | |
10451 | return cancelled; | |
10452 | } | |
10453 | ||
10454 | int get_err_code() { | |
10455 | return err_code; | |
10456 | } | |
10457 | ||
10458 | int wait_next_io(bool *done) { | |
10459 | lock.Lock(); | |
10460 | map<off_t, librados::AioCompletion *>::iterator iter = completion_map.begin(); | |
10461 | if (iter == completion_map.end()) { | |
10462 | *done = true; | |
10463 | lock.Unlock(); | |
10464 | return 0; | |
10465 | } | |
10466 | off_t cur_ofs = iter->first; | |
10467 | librados::AioCompletion *c = iter->second; | |
10468 | lock.Unlock(); | |
10469 | ||
10470 | c->wait_for_safe_and_cb(); | |
10471 | int r = c->get_return_value(); | |
10472 | ||
10473 | lock.Lock(); | |
10474 | completion_map.erase(cur_ofs); | |
10475 | ||
10476 | if (completion_map.empty()) { | |
10477 | *done = true; | |
10478 | } | |
10479 | lock.Unlock(); | |
10480 | ||
10481 | c->release(); | |
10482 | ||
10483 | return r; | |
10484 | } | |
10485 | ||
10486 | void add_io(off_t ofs, off_t len, bufferlist **pbl, AioCompletion **pc) { | |
10487 | Mutex::Locker l(lock); | |
10488 | ||
10489 | const auto& io_iter = io_map.insert( | |
10490 | map<off_t, get_obj_io>::value_type(ofs, get_obj_io())); | |
10491 | ||
10492 | assert(io_iter.second); // assert new insertion | |
10493 | ||
10494 | get_obj_io& io = (io_iter.first)->second; | |
10495 | *pbl = &io.bl; | |
10496 | ||
10497 | struct get_obj_aio_data aio; | |
10498 | aio.ofs = ofs; | |
10499 | aio.len = len; | |
10500 | aio.op_data = this; | |
10501 | ||
10502 | aio_data.push_back(aio); | |
10503 | ||
10504 | struct get_obj_aio_data *paio_data = &aio_data.back(); /* last element */ | |
10505 | ||
10506 | librados::AioCompletion *c = librados::Rados::aio_create_completion((void *)paio_data, NULL, _get_obj_aio_completion_cb); | |
10507 | completion_map[ofs] = c; | |
10508 | ||
10509 | *pc = c; | |
10510 | ||
10511 | /* we have a reference per IO, plus one reference for the calling function. | |
10512 | * reference is dropped for each callback, plus when we're done iterating | |
10513 | * over the parts */ | |
10514 | get(); | |
10515 | } | |
10516 | ||
10517 | void cancel_io(off_t ofs) { | |
10518 | ldout(cct, 20) << "get_obj_data::cancel_io() ofs=" << ofs << dendl; | |
10519 | lock.Lock(); | |
10520 | map<off_t, AioCompletion *>::iterator iter = completion_map.find(ofs); | |
10521 | if (iter != completion_map.end()) { | |
10522 | AioCompletion *c = iter->second; | |
10523 | c->release(); | |
10524 | completion_map.erase(ofs); | |
10525 | io_map.erase(ofs); | |
10526 | } | |
10527 | lock.Unlock(); | |
10528 | ||
10529 | /* we don't drop a reference here -- e.g., not calling d->put(), because we still | |
10530 | * need IoCtx to live, as io callback may still be called | |
10531 | */ | |
10532 | } | |
10533 | ||
10534 | void cancel_all_io() { | |
10535 | ldout(cct, 20) << "get_obj_data::cancel_all_io()" << dendl; | |
10536 | Mutex::Locker l(lock); | |
10537 | for (map<off_t, librados::AioCompletion *>::iterator iter = completion_map.begin(); | |
10538 | iter != completion_map.end(); ++iter) { | |
10539 | librados::AioCompletion *c = iter->second; | |
10540 | c->release(); | |
10541 | } | |
10542 | } | |
10543 | ||
10544 | int get_complete_ios(off_t ofs, list<bufferlist>& bl_list) { | |
10545 | Mutex::Locker l(lock); | |
10546 | ||
10547 | map<off_t, get_obj_io>::iterator liter = io_map.begin(); | |
10548 | ||
10549 | if (liter == io_map.end() || | |
10550 | liter->first != ofs) { | |
10551 | return 0; | |
10552 | } | |
10553 | ||
10554 | map<off_t, librados::AioCompletion *>::iterator aiter; | |
10555 | aiter = completion_map.find(ofs); | |
10556 | if (aiter == completion_map.end()) { | |
10557 | /* completion map does not hold this io, it was cancelled */ | |
10558 | return 0; | |
10559 | } | |
10560 | ||
10561 | AioCompletion *completion = aiter->second; | |
10562 | int r = completion->get_return_value(); | |
10563 | if (r < 0) | |
10564 | return r; | |
10565 | ||
10566 | for (; aiter != completion_map.end(); ++aiter) { | |
10567 | completion = aiter->second; | |
10568 | if (!completion->is_safe()) { | |
10569 | /* reached a request that is not yet complete, stop */ | |
10570 | break; | |
10571 | } | |
10572 | ||
10573 | r = completion->get_return_value(); | |
10574 | if (r < 0) { | |
10575 | set_cancelled(r); /* mark it as cancelled, so that we don't continue processing next operations */ | |
10576 | return r; | |
10577 | } | |
10578 | ||
10579 | total_read += r; | |
10580 | ||
10581 | map<off_t, get_obj_io>::iterator old_liter = liter++; | |
10582 | bl_list.push_back(old_liter->second.bl); | |
10583 | io_map.erase(old_liter); | |
10584 | } | |
10585 | ||
10586 | return 0; | |
10587 | } | |
10588 | }; | |
10589 | ||
10590 | 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) | |
10591 | { | |
10592 | struct get_obj_data *d = (struct get_obj_data *)arg; | |
10593 | ||
10594 | return d->rados->get_obj_iterate_cb(d->ctx, astate, bucket_info, obj, read_obj, obj_ofs, read_ofs, len, is_head_obj, arg); | |
10595 | } | |
10596 | ||
10597 | static void _get_obj_aio_completion_cb(completion_t cb, void *arg) | |
10598 | { | |
10599 | struct get_obj_aio_data *aio_data = (struct get_obj_aio_data *)arg; | |
10600 | struct get_obj_data *d = aio_data->op_data; | |
10601 | ||
10602 | d->rados->get_obj_aio_completion_cb(cb, arg); | |
10603 | } | |
10604 | ||
10605 | ||
10606 | void RGWRados::get_obj_aio_completion_cb(completion_t c, void *arg) | |
10607 | { | |
10608 | struct get_obj_aio_data *aio_data = (struct get_obj_aio_data *)arg; | |
10609 | struct get_obj_data *d = aio_data->op_data; | |
10610 | off_t ofs = aio_data->ofs; | |
10611 | off_t len = aio_data->len; | |
10612 | ||
10613 | list<bufferlist> bl_list; | |
10614 | list<bufferlist>::iterator iter; | |
10615 | int r; | |
10616 | ||
10617 | ldout(cct, 20) << "get_obj_aio_completion_cb: io completion ofs=" << ofs << " len=" << len << dendl; | |
10618 | d->throttle.put(len); | |
10619 | ||
10620 | r = rados_aio_get_return_value(c); | |
10621 | if (r < 0) { | |
10622 | ldout(cct, 0) << "ERROR: got unexpected error when trying to read object: " << r << dendl; | |
10623 | d->set_cancelled(r); | |
10624 | goto done; | |
10625 | } | |
10626 | ||
10627 | if (d->is_cancelled()) { | |
10628 | goto done; | |
10629 | } | |
10630 | ||
10631 | d->data_lock.Lock(); | |
10632 | ||
10633 | r = d->get_complete_ios(ofs, bl_list); | |
10634 | if (r < 0) { | |
10635 | goto done_unlock; | |
10636 | } | |
10637 | ||
10638 | d->read_list.splice(d->read_list.end(), bl_list); | |
10639 | ||
10640 | done_unlock: | |
10641 | d->data_lock.Unlock(); | |
10642 | done: | |
10643 | d->put(); | |
10644 | return; | |
10645 | } | |
10646 | ||
10647 | int RGWRados::flush_read_list(struct get_obj_data *d) | |
10648 | { | |
10649 | d->data_lock.Lock(); | |
10650 | list<bufferlist> l; | |
10651 | l.swap(d->read_list); | |
10652 | d->get(); | |
10653 | d->read_list.clear(); | |
10654 | ||
10655 | d->data_lock.Unlock(); | |
10656 | ||
10657 | int r = 0; | |
10658 | ||
10659 | list<bufferlist>::iterator iter; | |
10660 | for (iter = l.begin(); iter != l.end(); ++iter) { | |
10661 | bufferlist& bl = *iter; | |
10662 | r = d->client_cb->handle_data(bl, 0, bl.length()); | |
10663 | if (r < 0) { | |
10664 | dout(0) << "ERROR: flush_read_list(): d->client_cb->handle_data() returned " << r << dendl; | |
10665 | break; | |
10666 | } | |
10667 | } | |
10668 | ||
10669 | d->data_lock.Lock(); | |
10670 | d->put(); | |
10671 | if (r < 0) { | |
10672 | d->set_cancelled(r); | |
10673 | } | |
10674 | d->data_lock.Unlock(); | |
10675 | return r; | |
10676 | } | |
10677 | ||
10678 | int RGWRados::get_obj_iterate_cb(RGWObjectCtx *ctx, RGWObjState *astate, | |
10679 | const RGWBucketInfo& bucket_info, | |
10680 | const rgw_obj& obj, | |
10681 | const rgw_raw_obj& read_obj, | |
10682 | off_t obj_ofs, | |
10683 | off_t read_ofs, off_t len, | |
10684 | bool is_head_obj, void *arg) | |
10685 | { | |
10686 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
10687 | ObjectReadOperation op; | |
10688 | struct get_obj_data *d = (struct get_obj_data *)arg; | |
10689 | string oid, key; | |
10690 | bufferlist *pbl; | |
10691 | AioCompletion *c; | |
10692 | ||
10693 | int r; | |
10694 | ||
10695 | if (is_head_obj) { | |
10696 | /* only when reading from the head object do we need to do the atomic test */ | |
10697 | r = append_atomic_test(rctx, bucket_info, obj, op, &astate); | |
10698 | if (r < 0) | |
10699 | return r; | |
10700 | ||
10701 | if (astate && | |
10702 | obj_ofs < astate->data.length()) { | |
10703 | unsigned chunk_len = min((uint64_t)astate->data.length() - obj_ofs, (uint64_t)len); | |
10704 | ||
10705 | d->data_lock.Lock(); | |
10706 | r = d->client_cb->handle_data(astate->data, obj_ofs, chunk_len); | |
10707 | d->data_lock.Unlock(); | |
10708 | if (r < 0) | |
10709 | return r; | |
10710 | ||
10711 | d->lock.Lock(); | |
10712 | d->total_read += chunk_len; | |
10713 | d->lock.Unlock(); | |
10714 | ||
10715 | len -= chunk_len; | |
10716 | read_ofs += chunk_len; | |
10717 | obj_ofs += chunk_len; | |
10718 | if (!len) | |
10719 | return 0; | |
10720 | } | |
10721 | } | |
10722 | ||
10723 | d->throttle.get(len); | |
10724 | if (d->is_cancelled()) { | |
10725 | return d->get_err_code(); | |
10726 | } | |
10727 | ||
10728 | /* add io after we check that we're not cancelled, otherwise we're going to have trouble | |
10729 | * cleaning up | |
10730 | */ | |
10731 | d->add_io(obj_ofs, len, &pbl, &c); | |
10732 | ||
10733 | ldout(cct, 20) << "rados->get_obj_iterate_cb oid=" << read_obj.oid << " obj-ofs=" << obj_ofs << " read_ofs=" << read_ofs << " len=" << len << dendl; | |
10734 | op.read(read_ofs, len, pbl, NULL); | |
10735 | ||
10736 | librados::IoCtx io_ctx(d->io_ctx); | |
10737 | io_ctx.locator_set_key(read_obj.loc); | |
10738 | ||
10739 | r = io_ctx.aio_operate(read_obj.oid, c, &op, NULL); | |
10740 | if (r < 0) { | |
10741 | ldout(cct, 0) << "rados->aio_operate r=" << r << dendl; | |
10742 | goto done_err; | |
10743 | } | |
10744 | ||
10745 | // Flush data to client if there is any | |
10746 | r = flush_read_list(d); | |
10747 | if (r < 0) | |
10748 | return r; | |
10749 | ||
10750 | return 0; | |
10751 | ||
10752 | done_err: | |
10753 | ldout(cct, 20) << "cancelling io r=" << r << " obj_ofs=" << obj_ofs << dendl; | |
10754 | d->set_cancelled(r); | |
10755 | d->cancel_io(obj_ofs); | |
10756 | ||
10757 | return r; | |
10758 | } | |
10759 | ||
10760 | int RGWRados::Object::Read::iterate(int64_t ofs, int64_t end, RGWGetDataCB *cb) | |
10761 | { | |
10762 | RGWRados *store = source->get_store(); | |
10763 | CephContext *cct = store->ctx(); | |
10764 | ||
10765 | struct get_obj_data *data = new get_obj_data(cct); | |
10766 | bool done = false; | |
10767 | ||
10768 | RGWObjectCtx& obj_ctx = source->get_ctx(); | |
10769 | ||
10770 | data->rados = store; | |
10771 | data->io_ctx.dup(state.io_ctx); | |
10772 | data->client_cb = cb; | |
10773 | ||
10774 | 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); | |
10775 | if (r < 0) { | |
10776 | data->cancel_all_io(); | |
10777 | goto done; | |
10778 | } | |
10779 | ||
10780 | while (!done) { | |
10781 | r = data->wait_next_io(&done); | |
10782 | if (r < 0) { | |
10783 | dout(10) << "get_obj_iterate() r=" << r << ", canceling all io" << dendl; | |
10784 | data->cancel_all_io(); | |
10785 | break; | |
10786 | } | |
10787 | r = store->flush_read_list(data); | |
10788 | if (r < 0) { | |
10789 | dout(10) << "get_obj_iterate() r=" << r << ", canceling all io" << dendl; | |
10790 | data->cancel_all_io(); | |
10791 | break; | |
10792 | } | |
10793 | } | |
10794 | ||
10795 | done: | |
10796 | data->put(); | |
10797 | return r; | |
10798 | } | |
10799 | ||
10800 | int RGWRados::iterate_obj(RGWObjectCtx& obj_ctx, | |
10801 | const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
10802 | off_t ofs, off_t end, | |
10803 | uint64_t max_chunk_size, | |
10804 | int (*iterate_obj_cb)(const RGWBucketInfo&, const rgw_obj& obj, | |
10805 | const rgw_raw_obj&, off_t, off_t, off_t, bool, | |
10806 | RGWObjState *, void *), | |
10807 | void *arg) | |
10808 | { | |
10809 | rgw_raw_obj head_obj; | |
10810 | rgw_raw_obj read_obj; | |
10811 | uint64_t read_ofs = ofs; | |
10812 | uint64_t len; | |
10813 | bool reading_from_head = true; | |
10814 | RGWObjState *astate = NULL; | |
10815 | ||
10816 | obj_to_raw(bucket_info.placement_rule, obj, &head_obj); | |
10817 | ||
10818 | int r = get_obj_state(&obj_ctx, bucket_info, obj, &astate, false); | |
10819 | if (r < 0) { | |
10820 | return r; | |
10821 | } | |
10822 | ||
10823 | if (end < 0) | |
10824 | len = 0; | |
10825 | else | |
10826 | len = end - ofs + 1; | |
10827 | ||
10828 | if (astate->has_manifest) { | |
10829 | /* now get the relevant object stripe */ | |
10830 | RGWObjManifest::obj_iterator iter = astate->manifest.obj_find(ofs); | |
10831 | ||
10832 | RGWObjManifest::obj_iterator obj_end = astate->manifest.obj_end(); | |
10833 | ||
10834 | for (; iter != obj_end && ofs <= end; ++iter) { | |
10835 | off_t stripe_ofs = iter.get_stripe_ofs(); | |
10836 | off_t next_stripe_ofs = stripe_ofs + iter.get_stripe_size(); | |
10837 | ||
10838 | while (ofs < next_stripe_ofs && ofs <= end) { | |
10839 | read_obj = iter.get_location().get_raw_obj(this); | |
10840 | uint64_t read_len = min(len, iter.get_stripe_size() - (ofs - stripe_ofs)); | |
10841 | read_ofs = iter.location_ofs() + (ofs - stripe_ofs); | |
10842 | ||
10843 | if (read_len > max_chunk_size) { | |
10844 | read_len = max_chunk_size; | |
10845 | } | |
10846 | ||
10847 | reading_from_head = (read_obj == head_obj); | |
10848 | r = iterate_obj_cb(bucket_info, obj, read_obj, ofs, read_ofs, read_len, reading_from_head, astate, arg); | |
10849 | if (r < 0) { | |
10850 | return r; | |
10851 | } | |
10852 | ||
10853 | len -= read_len; | |
10854 | ofs += read_len; | |
10855 | } | |
10856 | } | |
10857 | } else { | |
10858 | while (ofs <= end) { | |
10859 | read_obj = head_obj; | |
10860 | uint64_t read_len = min(len, max_chunk_size); | |
10861 | ||
10862 | r = iterate_obj_cb(bucket_info, obj, read_obj, ofs, ofs, read_len, reading_from_head, astate, arg); | |
10863 | if (r < 0) { | |
10864 | return r; | |
10865 | } | |
10866 | ||
10867 | len -= read_len; | |
10868 | ofs += read_len; | |
10869 | } | |
10870 | } | |
10871 | ||
10872 | return 0; | |
10873 | } | |
10874 | ||
10875 | int RGWRados::obj_operate(const RGWBucketInfo& bucket_info, const rgw_obj& obj, ObjectWriteOperation *op) | |
10876 | { | |
10877 | rgw_rados_ref ref; | |
10878 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
10879 | if (r < 0) { | |
10880 | return r; | |
10881 | } | |
10882 | ||
10883 | return ref.ioctx.operate(ref.oid, op); | |
10884 | } | |
10885 | ||
10886 | int RGWRados::obj_operate(const RGWBucketInfo& bucket_info, const rgw_obj& obj, ObjectReadOperation *op) | |
10887 | { | |
10888 | rgw_rados_ref ref; | |
10889 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
10890 | if (r < 0) { | |
10891 | return r; | |
10892 | } | |
10893 | ||
10894 | bufferlist outbl; | |
10895 | ||
10896 | return ref.ioctx.operate(ref.oid, op, &outbl); | |
10897 | } | |
10898 | ||
10899 | int RGWRados::olh_init_modification_impl(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& olh_obj, string *op_tag) | |
10900 | { | |
10901 | ObjectWriteOperation op; | |
10902 | ||
10903 | assert(olh_obj.key.instance.empty()); | |
10904 | ||
10905 | bool has_tag = (state.exists && has_olh_tag(state.attrset)); | |
10906 | ||
10907 | if (!state.exists) { | |
10908 | op.create(true); | |
10909 | } else { | |
10910 | op.assert_exists(); | |
b32b8144 FG |
10911 | struct timespec mtime_ts = real_clock::to_timespec(state.mtime); |
10912 | op.mtime2(&mtime_ts); | |
7c673cae FG |
10913 | } |
10914 | ||
10915 | /* | |
10916 | * 3 possible cases: olh object doesn't exist, it exists as an olh, it exists as a regular object. | |
10917 | * If it exists as a regular object we'll need to transform it into an olh. We'll do it in two | |
10918 | * steps, first change its tag and set the olh pending attrs. Once write is done we'll need to | |
10919 | * truncate it, remove extra attrs, and send it to the garbage collection. The bucket index olh | |
10920 | * log will reflect that. | |
10921 | * | |
10922 | * Need to generate separate olh and obj tags, as olh can be colocated with object data. obj_tag | |
10923 | * is used for object data instance, olh_tag for olh instance. | |
10924 | */ | |
10925 | if (has_tag) { | |
10926 | /* guard against racing writes */ | |
10927 | bucket_index_guard_olh_op(state, op); | |
10928 | } | |
10929 | ||
10930 | if (!has_tag) { | |
10931 | /* obj tag */ | |
10932 | string obj_tag; | |
10933 | int ret = gen_rand_alphanumeric_lower(cct, &obj_tag, 32); | |
10934 | if (ret < 0) { | |
10935 | ldout(cct, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret << dendl; | |
10936 | return ret; | |
10937 | } | |
10938 | bufferlist bl; | |
10939 | bl.append(obj_tag.c_str(), obj_tag.size()); | |
10940 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
10941 | ||
10942 | state.attrset[RGW_ATTR_ID_TAG] = bl; | |
10943 | state.obj_tag = bl; | |
10944 | ||
10945 | /* olh tag */ | |
10946 | string olh_tag; | |
10947 | ret = gen_rand_alphanumeric_lower(cct, &olh_tag, 32); | |
10948 | if (ret < 0) { | |
10949 | ldout(cct, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret << dendl; | |
10950 | return ret; | |
10951 | } | |
10952 | bufferlist olh_bl; | |
10953 | olh_bl.append(olh_tag.c_str(), olh_tag.size()); | |
10954 | op.setxattr(RGW_ATTR_OLH_ID_TAG, olh_bl); | |
10955 | ||
10956 | state.attrset[RGW_ATTR_OLH_ID_TAG] = olh_bl; | |
10957 | state.olh_tag = olh_bl; | |
10958 | state.is_olh = true; | |
10959 | ||
10960 | bufferlist verbl; | |
10961 | op.setxattr(RGW_ATTR_OLH_VER, verbl); | |
10962 | } | |
10963 | ||
10964 | bufferlist bl; | |
10965 | RGWOLHPendingInfo pending_info; | |
10966 | pending_info.time = real_clock::now(); | |
10967 | ::encode(pending_info, bl); | |
10968 | ||
10969 | #define OLH_PENDING_TAG_LEN 32 | |
10970 | /* tag will start with current time epoch, this so that entries are sorted by time */ | |
10971 | char buf[32]; | |
10972 | utime_t ut(pending_info.time); | |
10973 | snprintf(buf, sizeof(buf), "%016llx", (unsigned long long)ut.sec()); | |
10974 | *op_tag = buf; | |
10975 | ||
10976 | string s; | |
10977 | int ret = gen_rand_alphanumeric_lower(cct, &s, OLH_PENDING_TAG_LEN - op_tag->size()); | |
10978 | if (ret < 0) { | |
10979 | ldout(cct, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret << dendl; | |
10980 | return ret; | |
10981 | } | |
10982 | op_tag->append(s); | |
10983 | ||
10984 | string attr_name = RGW_ATTR_OLH_PENDING_PREFIX; | |
10985 | attr_name.append(*op_tag); | |
10986 | ||
10987 | op.setxattr(attr_name.c_str(), bl); | |
10988 | ||
10989 | ret = obj_operate(bucket_info, olh_obj, &op); | |
10990 | if (ret < 0) { | |
10991 | return ret; | |
10992 | } | |
10993 | ||
10994 | state.exists = true; | |
10995 | state.attrset[attr_name] = bl; | |
10996 | ||
10997 | return 0; | |
10998 | } | |
10999 | ||
11000 | int RGWRados::olh_init_modification(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj, string *op_tag) | |
11001 | { | |
11002 | int ret; | |
11003 | ||
11004 | ret = olh_init_modification_impl(bucket_info, state, obj, op_tag); | |
11005 | if (ret == -EEXIST) { | |
11006 | ret = -ECANCELED; | |
11007 | } | |
11008 | ||
11009 | return ret; | |
11010 | } | |
11011 | ||
31f18b77 FG |
11012 | int RGWRados::guard_reshard(BucketShard *bs, const rgw_obj& obj_instance, std::function<int(BucketShard *)> call) |
11013 | { | |
11014 | rgw_obj obj; | |
11015 | const rgw_obj *pobj = &obj_instance; | |
11016 | int r; | |
11017 | ||
11018 | for (int i = 0; i < NUM_RESHARD_RETRIES; ++i) { | |
11019 | r = bs->init(pobj->bucket, *pobj); | |
11020 | if (r < 0) { | |
11021 | ldout(cct, 5) << "bs.init() returned ret=" << r << dendl; | |
11022 | return r; | |
11023 | } | |
11024 | r = call(bs); | |
11025 | if (r != -ERR_BUSY_RESHARDING) { | |
11026 | break; | |
11027 | } | |
11028 | ldout(cct, 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl; | |
11029 | string new_bucket_id; | |
11030 | r = block_while_resharding(bs, &new_bucket_id); | |
11031 | if (r == -ERR_BUSY_RESHARDING) { | |
11032 | continue; | |
11033 | } | |
11034 | if (r < 0) { | |
11035 | return r; | |
11036 | } | |
11037 | ldout(cct, 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id << dendl; | |
11038 | i = 0; /* resharding is finished, make sure we can retry */ | |
11039 | ||
11040 | obj = *pobj; | |
11041 | obj.bucket.update_bucket_id(new_bucket_id); | |
11042 | pobj = &obj; | |
11043 | } | |
11044 | ||
11045 | if (r < 0) { | |
11046 | return r; | |
11047 | } | |
11048 | ||
11049 | return 0; | |
11050 | } | |
11051 | ||
11052 | int RGWRados::block_while_resharding(RGWRados::BucketShard *bs, string *new_bucket_id) | |
11053 | { | |
11054 | std::shared_ptr<RGWReshardWait> waiter = reshard_wait; | |
11055 | ||
11056 | return waiter->block_while_resharding(bs, new_bucket_id); | |
11057 | } | |
11058 | ||
7c673cae FG |
11059 | int RGWRados::bucket_index_link_olh(const RGWBucketInfo& bucket_info, RGWObjState& olh_state, const rgw_obj& obj_instance, |
11060 | bool delete_marker, | |
11061 | const string& op_tag, | |
11062 | struct rgw_bucket_dir_entry_meta *meta, | |
11063 | uint64_t olh_epoch, | |
31f18b77 | 11064 | real_time unmod_since, bool high_precision_time, rgw_zone_set *_zones_trace) |
7c673cae FG |
11065 | { |
11066 | rgw_rados_ref ref; | |
11067 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11068 | if (r < 0) { | |
11069 | return r; | |
11070 | } | |
11071 | ||
31f18b77 FG |
11072 | rgw_zone_set zones_trace; |
11073 | if (_zones_trace) { | |
11074 | zones_trace = *_zones_trace; | |
11075 | } else { | |
11076 | zones_trace.insert(get_zone().id); | |
7c673cae FG |
11077 | } |
11078 | ||
31f18b77 FG |
11079 | BucketShard bs(this); |
11080 | ||
7c673cae | 11081 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), obj_instance.key.instance); |
31f18b77 FG |
11082 | r = guard_reshard(&bs, obj_instance, [&](BucketShard *bs) -> int { |
11083 | librados::ObjectWriteOperation op; | |
11084 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11085 | return cls_rgw_bucket_link_olh(bs->index_ctx, op, | |
11086 | bs->bucket_obj, key, olh_state.olh_tag, delete_marker, op_tag, meta, olh_epoch, | |
11087 | unmod_since, high_precision_time, | |
11088 | get_zone().log_data, zones_trace); | |
11089 | }); | |
11090 | if (r < 0) { | |
11091 | ldout(cct, 20) << "cls_rgw_bucket_link_olh() returned r=" << r << dendl; | |
11092 | return r; | |
7c673cae FG |
11093 | } |
11094 | ||
11095 | return 0; | |
11096 | } | |
11097 | ||
11098 | void RGWRados::bucket_index_guard_olh_op(RGWObjState& olh_state, ObjectOperation& op) | |
11099 | { | |
11100 | ldout(cct, 20) << __func__ << "(): olh_state.olh_tag=" << string(olh_state.olh_tag.c_str(), olh_state.olh_tag.length()) << dendl; | |
11101 | op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_state.olh_tag); | |
11102 | } | |
11103 | ||
11104 | int RGWRados::bucket_index_unlink_instance(const RGWBucketInfo& bucket_info, const rgw_obj& obj_instance, | |
31f18b77 | 11105 | const string& op_tag, const string& olh_tag, uint64_t olh_epoch, rgw_zone_set *_zones_trace) |
7c673cae FG |
11106 | { |
11107 | rgw_rados_ref ref; | |
11108 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11109 | if (r < 0) { | |
11110 | return r; | |
11111 | } | |
11112 | ||
31f18b77 FG |
11113 | rgw_zone_set zones_trace; |
11114 | if (_zones_trace) { | |
11115 | zones_trace = *_zones_trace; | |
7c673cae | 11116 | } |
31f18b77 FG |
11117 | zones_trace.insert(get_zone().id); |
11118 | ||
11119 | BucketShard bs(this); | |
7c673cae FG |
11120 | |
11121 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), obj_instance.key.instance); | |
31f18b77 FG |
11122 | r = guard_reshard(&bs, obj_instance, [&](BucketShard *bs) -> int { |
11123 | librados::ObjectWriteOperation op; | |
11124 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11125 | return cls_rgw_bucket_unlink_instance(bs->index_ctx, op, bs->bucket_obj, key, op_tag, | |
11126 | olh_tag, olh_epoch, get_zone().log_data, zones_trace); | |
11127 | }); | |
11128 | if (r < 0) { | |
11129 | ldout(cct, 20) << "cls_rgw_bucket_link_olh() returned r=" << r << dendl; | |
11130 | return r; | |
7c673cae FG |
11131 | } |
11132 | ||
11133 | return 0; | |
11134 | } | |
11135 | ||
11136 | int RGWRados::bucket_index_read_olh_log(const RGWBucketInfo& bucket_info, RGWObjState& state, | |
11137 | const rgw_obj& obj_instance, uint64_t ver_marker, | |
11138 | map<uint64_t, vector<rgw_bucket_olh_log_entry> > *log, | |
11139 | bool *is_truncated) | |
11140 | { | |
11141 | rgw_rados_ref ref; | |
11142 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11143 | if (r < 0) { | |
11144 | return r; | |
11145 | } | |
11146 | ||
11147 | BucketShard bs(this); | |
11148 | int ret = bs.init(obj_instance.bucket, obj_instance); | |
11149 | if (ret < 0) { | |
11150 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
11151 | return ret; | |
11152 | } | |
11153 | ||
11154 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
11155 | ||
11156 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
11157 | ||
31f18b77 FG |
11158 | ret = guard_reshard(&bs, obj_instance, [&](BucketShard *bs) -> int { |
11159 | ObjectReadOperation op; | |
11160 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11161 | return cls_rgw_get_olh_log(bs->index_ctx, bs->bucket_obj, op, | |
11162 | key, ver_marker, olh_tag, log, is_truncated); | |
11163 | }); | |
11164 | if (ret < 0) { | |
11165 | ldout(cct, 20) << "cls_rgw_get_olh_log() returned r=" << r << dendl; | |
7c673cae | 11166 | return ret; |
31f18b77 | 11167 | } |
7c673cae FG |
11168 | |
11169 | return 0; | |
11170 | } | |
11171 | ||
11172 | int RGWRados::bucket_index_trim_olh_log(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj_instance, uint64_t ver) | |
11173 | { | |
11174 | rgw_rados_ref ref; | |
11175 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11176 | if (r < 0) { | |
11177 | return r; | |
11178 | } | |
11179 | ||
11180 | BucketShard bs(this); | |
11181 | int ret = bs.init(obj_instance.bucket, obj_instance); | |
11182 | if (ret < 0) { | |
11183 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
11184 | return ret; | |
11185 | } | |
11186 | ||
11187 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
11188 | ||
11189 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
11190 | ||
31f18b77 FG |
11191 | ret = guard_reshard(&bs, obj_instance, [&](BucketShard *pbs) -> int { |
11192 | ObjectWriteOperation op; | |
11193 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11194 | cls_rgw_trim_olh_log(op, key, ver, olh_tag); | |
11195 | return pbs->index_ctx.operate(pbs->bucket_obj, &op); | |
11196 | }); | |
11197 | if (ret < 0) { | |
11198 | ldout(cct, 20) << "cls_rgw_trim_olh_log() returned r=" << ret << dendl; | |
7c673cae | 11199 | return ret; |
31f18b77 | 11200 | } |
7c673cae FG |
11201 | |
11202 | return 0; | |
11203 | } | |
11204 | ||
11205 | int RGWRados::bucket_index_clear_olh(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj_instance) | |
11206 | { | |
11207 | rgw_rados_ref ref; | |
11208 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
11209 | if (r < 0) { | |
11210 | return r; | |
11211 | } | |
11212 | ||
11213 | BucketShard bs(this); | |
7c673cae FG |
11214 | |
11215 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
11216 | ||
11217 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
11218 | ||
31f18b77 FG |
11219 | int ret = guard_reshard(&bs, obj_instance, [&](BucketShard *pbs) -> int { |
11220 | ObjectWriteOperation op; | |
11221 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
11222 | return cls_rgw_clear_olh(pbs->index_ctx, op, pbs->bucket_obj, key, olh_tag); | |
11223 | }); | |
7c673cae FG |
11224 | if (ret < 0) { |
11225 | ldout(cct, 5) << "cls_rgw_clear_olh() returned ret=" << ret << dendl; | |
11226 | return ret; | |
11227 | } | |
11228 | ||
11229 | return 0; | |
11230 | } | |
11231 | ||
11232 | int RGWRados::apply_olh_log(RGWObjectCtx& obj_ctx, RGWObjState& state, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
11233 | bufferlist& olh_tag, map<uint64_t, vector<rgw_bucket_olh_log_entry> >& log, | |
31f18b77 | 11234 | uint64_t *plast_ver, rgw_zone_set* zones_trace) |
7c673cae FG |
11235 | { |
11236 | if (log.empty()) { | |
11237 | return 0; | |
11238 | } | |
11239 | ||
11240 | librados::ObjectWriteOperation op; | |
11241 | ||
11242 | uint64_t last_ver = log.rbegin()->first; | |
11243 | *plast_ver = last_ver; | |
11244 | ||
11245 | map<uint64_t, vector<rgw_bucket_olh_log_entry> >::iterator iter = log.begin(); | |
11246 | ||
11247 | op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_tag); | |
11248 | op.cmpxattr(RGW_ATTR_OLH_VER, CEPH_OSD_CMPXATTR_OP_GT, last_ver); | |
11249 | ||
b32b8144 FG |
11250 | struct timespec mtime_ts = real_clock::to_timespec(state.mtime); |
11251 | op.mtime2(&mtime_ts); | |
11252 | ||
7c673cae FG |
11253 | bool need_to_link = false; |
11254 | cls_rgw_obj_key key; | |
11255 | bool delete_marker = false; | |
11256 | list<cls_rgw_obj_key> remove_instances; | |
11257 | bool need_to_remove = false; | |
11258 | ||
11259 | for (iter = log.begin(); iter != log.end(); ++iter) { | |
11260 | vector<rgw_bucket_olh_log_entry>::iterator viter = iter->second.begin(); | |
11261 | for (; viter != iter->second.end(); ++viter) { | |
11262 | rgw_bucket_olh_log_entry& entry = *viter; | |
11263 | ||
11264 | ldout(cct, 20) << "olh_log_entry: op=" << (int)entry.op | |
11265 | << " key=" << entry.key.name << "[" << entry.key.instance << "] " | |
11266 | << (entry.delete_marker ? "(delete)" : "") << dendl; | |
11267 | switch (entry.op) { | |
11268 | case CLS_RGW_OLH_OP_REMOVE_INSTANCE: | |
11269 | remove_instances.push_back(entry.key); | |
11270 | break; | |
11271 | case CLS_RGW_OLH_OP_LINK_OLH: | |
11272 | need_to_link = true; | |
11273 | need_to_remove = false; | |
11274 | key = entry.key; | |
11275 | delete_marker = entry.delete_marker; | |
11276 | break; | |
11277 | case CLS_RGW_OLH_OP_UNLINK_OLH: | |
11278 | need_to_remove = true; | |
11279 | need_to_link = false; | |
11280 | break; | |
11281 | default: | |
11282 | ldout(cct, 0) << "ERROR: apply_olh_log: invalid op: " << (int)entry.op << dendl; | |
11283 | return -EIO; | |
11284 | } | |
11285 | string attr_name = RGW_ATTR_OLH_PENDING_PREFIX; | |
11286 | attr_name.append(entry.op_tag); | |
11287 | op.rmxattr(attr_name.c_str()); | |
11288 | } | |
11289 | } | |
11290 | ||
11291 | rgw_rados_ref ref; | |
11292 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
11293 | if (r < 0) { | |
11294 | return r; | |
11295 | } | |
11296 | ||
11297 | const rgw_bucket& bucket = obj.bucket; | |
11298 | ||
11299 | if (need_to_link) { | |
11300 | rgw_obj target(bucket, key); | |
11301 | RGWOLHInfo info; | |
11302 | info.target = target; | |
11303 | info.removed = delete_marker; | |
11304 | bufferlist bl; | |
11305 | ::encode(info, bl); | |
11306 | op.setxattr(RGW_ATTR_OLH_INFO, bl); | |
11307 | } | |
11308 | ||
11309 | /* first remove object instances */ | |
11310 | for (list<cls_rgw_obj_key>::iterator liter = remove_instances.begin(); | |
11311 | liter != remove_instances.end(); ++liter) { | |
11312 | cls_rgw_obj_key& key = *liter; | |
11313 | rgw_obj obj_instance(bucket, key); | |
31f18b77 | 11314 | int ret = delete_obj(obj_ctx, bucket_info, obj_instance, 0, RGW_BILOG_FLAG_VERSIONED_OP, ceph::real_time(), zones_trace); |
7c673cae FG |
11315 | if (ret < 0 && ret != -ENOENT) { |
11316 | ldout(cct, 0) << "ERROR: delete_obj() returned " << ret << " obj_instance=" << obj_instance << dendl; | |
11317 | return ret; | |
11318 | } | |
11319 | } | |
11320 | ||
11321 | /* update olh object */ | |
11322 | r = ref.ioctx.operate(ref.oid, &op); | |
11323 | if (r == -ECANCELED) { | |
11324 | r = 0; | |
11325 | } | |
11326 | if (r < 0) { | |
11327 | ldout(cct, 0) << "ERROR: could not apply olh update, r=" << r << dendl; | |
11328 | return r; | |
11329 | } | |
11330 | ||
11331 | r = bucket_index_trim_olh_log(bucket_info, state, obj, last_ver); | |
11332 | if (r < 0) { | |
11333 | ldout(cct, 0) << "ERROR: could not trim olh log, r=" << r << dendl; | |
11334 | return r; | |
11335 | } | |
11336 | ||
11337 | if (need_to_remove) { | |
11338 | ObjectWriteOperation rm_op; | |
11339 | ||
11340 | rm_op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_tag); | |
11341 | rm_op.cmpxattr(RGW_ATTR_OLH_VER, CEPH_OSD_CMPXATTR_OP_GT, last_ver); | |
11342 | cls_obj_check_prefix_exist(rm_op, RGW_ATTR_OLH_PENDING_PREFIX, true); /* fail if found one of these, pending modification */ | |
11343 | rm_op.remove(); | |
11344 | ||
11345 | r = ref.ioctx.operate(ref.oid, &rm_op); | |
11346 | if (r == -ECANCELED) { | |
11347 | return 0; /* someone else won this race */ | |
11348 | } else { | |
11349 | /* | |
11350 | * only clear if was successful, otherwise we might clobber pending operations on this object | |
11351 | */ | |
11352 | r = bucket_index_clear_olh(bucket_info, state, obj); | |
11353 | if (r < 0) { | |
11354 | ldout(cct, 0) << "ERROR: could not clear bucket index olh entries r=" << r << dendl; | |
11355 | return r; | |
11356 | } | |
11357 | } | |
11358 | } | |
11359 | ||
11360 | return 0; | |
11361 | } | |
11362 | ||
11363 | /* | |
11364 | * read olh log and apply it | |
11365 | */ | |
31f18b77 | 11366 | int RGWRados::update_olh(RGWObjectCtx& obj_ctx, RGWObjState *state, const RGWBucketInfo& bucket_info, const rgw_obj& obj, rgw_zone_set *zones_trace) |
7c673cae FG |
11367 | { |
11368 | map<uint64_t, vector<rgw_bucket_olh_log_entry> > log; | |
11369 | bool is_truncated; | |
11370 | uint64_t ver_marker = 0; | |
11371 | ||
11372 | do { | |
11373 | int ret = bucket_index_read_olh_log(bucket_info, *state, obj, ver_marker, &log, &is_truncated); | |
11374 | if (ret < 0) { | |
11375 | return ret; | |
11376 | } | |
31f18b77 | 11377 | ret = apply_olh_log(obj_ctx, *state, bucket_info, obj, state->olh_tag, log, &ver_marker, zones_trace); |
7c673cae FG |
11378 | if (ret < 0) { |
11379 | return ret; | |
11380 | } | |
11381 | } while (is_truncated); | |
11382 | ||
11383 | return 0; | |
11384 | } | |
11385 | ||
11386 | int RGWRados::set_olh(RGWObjectCtx& obj_ctx, RGWBucketInfo& bucket_info, const rgw_obj& target_obj, bool delete_marker, rgw_bucket_dir_entry_meta *meta, | |
31f18b77 | 11387 | uint64_t olh_epoch, real_time unmod_since, bool high_precision_time, rgw_zone_set *zones_trace) |
7c673cae FG |
11388 | { |
11389 | string op_tag; | |
11390 | ||
11391 | rgw_obj olh_obj = target_obj; | |
11392 | olh_obj.key.instance.clear(); | |
11393 | ||
11394 | RGWObjState *state = NULL; | |
11395 | ||
11396 | int ret = 0; | |
11397 | int i; | |
31f18b77 | 11398 | |
7c673cae FG |
11399 | #define MAX_ECANCELED_RETRY 100 |
11400 | for (i = 0; i < MAX_ECANCELED_RETRY; i++) { | |
11401 | if (ret == -ECANCELED) { | |
11402 | obj_ctx.obj.invalidate(olh_obj); | |
11403 | } | |
11404 | ||
11405 | ret = get_obj_state(&obj_ctx, bucket_info, olh_obj, &state, false); /* don't follow olh */ | |
11406 | if (ret < 0) { | |
11407 | return ret; | |
11408 | } | |
11409 | ||
11410 | ret = olh_init_modification(bucket_info, *state, olh_obj, &op_tag); | |
11411 | if (ret < 0) { | |
11412 | ldout(cct, 20) << "olh_init_modification() target_obj=" << target_obj << " delete_marker=" << (int)delete_marker << " returned " << ret << dendl; | |
11413 | if (ret == -ECANCELED) { | |
11414 | continue; | |
11415 | } | |
11416 | return ret; | |
11417 | } | |
31f18b77 | 11418 | ret = bucket_index_link_olh(bucket_info, *state, target_obj, delete_marker, op_tag, meta, olh_epoch, unmod_since, high_precision_time, zones_trace); |
7c673cae FG |
11419 | if (ret < 0) { |
11420 | ldout(cct, 20) << "bucket_index_link_olh() target_obj=" << target_obj << " delete_marker=" << (int)delete_marker << " returned " << ret << dendl; | |
11421 | if (ret == -ECANCELED) { | |
11422 | continue; | |
11423 | } | |
11424 | return ret; | |
11425 | } | |
11426 | break; | |
11427 | } | |
11428 | ||
11429 | if (i == MAX_ECANCELED_RETRY) { | |
11430 | ldout(cct, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl; | |
11431 | return -EIO; | |
11432 | } | |
11433 | ||
11434 | ret = update_olh(obj_ctx, state, bucket_info, olh_obj); | |
11435 | if (ret == -ECANCELED) { /* already did what we needed, no need to retry, raced with another user */ | |
11436 | ret = 0; | |
11437 | } | |
11438 | if (ret < 0) { | |
11439 | ldout(cct, 20) << "update_olh() target_obj=" << target_obj << " returned " << ret << dendl; | |
11440 | return ret; | |
11441 | } | |
11442 | ||
11443 | return 0; | |
11444 | } | |
11445 | ||
11446 | int RGWRados::unlink_obj_instance(RGWObjectCtx& obj_ctx, RGWBucketInfo& bucket_info, const rgw_obj& target_obj, | |
31f18b77 | 11447 | uint64_t olh_epoch, rgw_zone_set *zones_trace) |
7c673cae FG |
11448 | { |
11449 | string op_tag; | |
11450 | ||
11451 | rgw_obj olh_obj = target_obj; | |
11452 | olh_obj.key.instance.clear(); | |
11453 | ||
11454 | RGWObjState *state = NULL; | |
11455 | ||
11456 | int ret = 0; | |
11457 | int i; | |
11458 | ||
11459 | for (i = 0; i < MAX_ECANCELED_RETRY; i++) { | |
11460 | if (ret == -ECANCELED) { | |
11461 | obj_ctx.obj.invalidate(olh_obj); | |
11462 | } | |
11463 | ||
11464 | ret = get_obj_state(&obj_ctx, bucket_info, olh_obj, &state, false); /* don't follow olh */ | |
11465 | if (ret < 0) | |
11466 | return ret; | |
11467 | ||
11468 | ret = olh_init_modification(bucket_info, *state, olh_obj, &op_tag); | |
11469 | if (ret < 0) { | |
11470 | ldout(cct, 20) << "olh_init_modification() target_obj=" << target_obj << " returned " << ret << dendl; | |
11471 | if (ret == -ECANCELED) { | |
11472 | continue; | |
11473 | } | |
11474 | return ret; | |
11475 | } | |
11476 | ||
11477 | string olh_tag(state->olh_tag.c_str(), state->olh_tag.length()); | |
11478 | ||
31f18b77 | 11479 | ret = bucket_index_unlink_instance(bucket_info, target_obj, op_tag, olh_tag, olh_epoch, zones_trace); |
7c673cae FG |
11480 | if (ret < 0) { |
11481 | ldout(cct, 20) << "bucket_index_unlink_instance() target_obj=" << target_obj << " returned " << ret << dendl; | |
11482 | if (ret == -ECANCELED) { | |
11483 | continue; | |
11484 | } | |
11485 | return ret; | |
11486 | } | |
11487 | break; | |
11488 | } | |
11489 | ||
11490 | if (i == MAX_ECANCELED_RETRY) { | |
11491 | ldout(cct, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl; | |
11492 | return -EIO; | |
11493 | } | |
11494 | ||
31f18b77 | 11495 | ret = update_olh(obj_ctx, state, bucket_info, olh_obj, zones_trace); |
7c673cae FG |
11496 | if (ret == -ECANCELED) { /* already did what we needed, no need to retry, raced with another user */ |
11497 | return 0; | |
11498 | } | |
11499 | if (ret < 0) { | |
11500 | ldout(cct, 20) << "update_olh() target_obj=" << target_obj << " returned " << ret << dendl; | |
11501 | return ret; | |
11502 | } | |
11503 | ||
11504 | return 0; | |
11505 | } | |
11506 | ||
11507 | void RGWRados::gen_rand_obj_instance_name(rgw_obj *target_obj) | |
11508 | { | |
11509 | #define OBJ_INSTANCE_LEN 32 | |
11510 | char buf[OBJ_INSTANCE_LEN + 1]; | |
11511 | ||
11512 | gen_rand_alphanumeric_no_underscore(cct, buf, OBJ_INSTANCE_LEN); /* don't want it to get url escaped, | |
11513 | no underscore for instance name due to the way we encode the raw keys */ | |
11514 | ||
11515 | target_obj->key.set_instance(buf); | |
11516 | } | |
11517 | ||
11518 | static void filter_attrset(map<string, bufferlist>& unfiltered_attrset, const string& check_prefix, | |
11519 | map<string, bufferlist> *attrset) | |
11520 | { | |
11521 | attrset->clear(); | |
11522 | map<string, bufferlist>::iterator iter; | |
11523 | for (iter = unfiltered_attrset.lower_bound(check_prefix); | |
11524 | iter != unfiltered_attrset.end(); ++iter) { | |
11525 | if (!boost::algorithm::starts_with(iter->first, check_prefix)) | |
11526 | break; | |
11527 | (*attrset)[iter->first] = iter->second; | |
11528 | } | |
11529 | } | |
11530 | ||
11531 | int RGWRados::get_olh(const RGWBucketInfo& bucket_info, const rgw_obj& obj, RGWOLHInfo *olh) | |
11532 | { | |
11533 | map<string, bufferlist> unfiltered_attrset; | |
11534 | ||
11535 | ObjectReadOperation op; | |
11536 | op.getxattrs(&unfiltered_attrset, NULL); | |
11537 | ||
11538 | bufferlist outbl; | |
11539 | int r = obj_operate(bucket_info, obj, &op); | |
11540 | ||
11541 | if (r < 0) { | |
11542 | return r; | |
11543 | } | |
11544 | map<string, bufferlist> attrset; | |
11545 | ||
11546 | filter_attrset(unfiltered_attrset, RGW_ATTR_OLH_PREFIX, &attrset); | |
11547 | ||
11548 | map<string, bufferlist>::iterator iter = attrset.find(RGW_ATTR_OLH_INFO); | |
11549 | if (iter == attrset.end()) { /* not an olh */ | |
11550 | return -EINVAL; | |
11551 | } | |
11552 | ||
11553 | try { | |
11554 | bufferlist::iterator biter = iter->second.begin(); | |
11555 | ::decode(*olh, biter); | |
11556 | } catch (buffer::error& err) { | |
11557 | ldout(cct, 0) << "ERROR: failed to decode olh info" << dendl; | |
11558 | return -EIO; | |
11559 | } | |
11560 | ||
11561 | return 0; | |
11562 | } | |
11563 | ||
11564 | void RGWRados::check_pending_olh_entries(map<string, bufferlist>& pending_entries, | |
11565 | map<string, bufferlist> *rm_pending_entries) | |
11566 | { | |
11567 | map<string, bufferlist>::iterator iter = pending_entries.begin(); | |
11568 | ||
11569 | real_time now = real_clock::now(); | |
11570 | ||
11571 | while (iter != pending_entries.end()) { | |
11572 | bufferlist::iterator biter = iter->second.begin(); | |
11573 | RGWOLHPendingInfo pending_info; | |
11574 | try { | |
11575 | ::decode(pending_info, biter); | |
11576 | } catch (buffer::error& err) { | |
11577 | /* skipping bad entry, we could remove it but it might hide a bug */ | |
11578 | ldout(cct, 0) << "ERROR: failed to decode pending entry " << iter->first << dendl; | |
11579 | ++iter; | |
11580 | continue; | |
11581 | } | |
11582 | ||
11583 | map<string, bufferlist>::iterator cur_iter = iter; | |
11584 | ++iter; | |
11585 | if (now - pending_info.time >= make_timespan(cct->_conf->rgw_olh_pending_timeout_sec)) { | |
11586 | (*rm_pending_entries)[cur_iter->first] = cur_iter->second; | |
11587 | pending_entries.erase(cur_iter); | |
11588 | } else { | |
11589 | /* entries names are sorted by time (rounded to a second) */ | |
11590 | break; | |
11591 | } | |
11592 | } | |
11593 | } | |
11594 | ||
11595 | int RGWRados::remove_olh_pending_entries(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& olh_obj, map<string, bufferlist>& pending_attrs) | |
11596 | { | |
11597 | ObjectWriteOperation op; | |
11598 | ||
11599 | bucket_index_guard_olh_op(state, op); | |
11600 | ||
11601 | for (map<string, bufferlist>::iterator iter = pending_attrs.begin(); iter != pending_attrs.end(); ++iter) { | |
11602 | op.rmxattr(iter->first.c_str()); | |
11603 | } | |
11604 | ||
11605 | rgw_rados_ref ref; | |
11606 | int r = get_obj_head_ref(bucket_info, olh_obj, &ref); | |
11607 | if (r < 0) { | |
11608 | return r; | |
11609 | } | |
11610 | ||
11611 | /* update olh object */ | |
11612 | r = ref.ioctx.operate(ref.oid, &op); | |
11613 | if (r == -ENOENT || r == -ECANCELED) { | |
11614 | /* raced with some other change, shouldn't sweat about it */ | |
11615 | r = 0; | |
11616 | } | |
11617 | if (r < 0) { | |
11618 | ldout(cct, 0) << "ERROR: could not apply olh update, r=" << r << dendl; | |
11619 | return r; | |
11620 | } | |
11621 | ||
11622 | return 0; | |
11623 | } | |
11624 | ||
11625 | int RGWRados::follow_olh(const RGWBucketInfo& bucket_info, RGWObjectCtx& obj_ctx, RGWObjState *state, const rgw_obj& olh_obj, rgw_obj *target) | |
11626 | { | |
11627 | map<string, bufferlist> pending_entries; | |
11628 | filter_attrset(state->attrset, RGW_ATTR_OLH_PENDING_PREFIX, &pending_entries); | |
11629 | ||
11630 | map<string, bufferlist> rm_pending_entries; | |
11631 | check_pending_olh_entries(pending_entries, &rm_pending_entries); | |
11632 | ||
11633 | if (!rm_pending_entries.empty()) { | |
11634 | int ret = remove_olh_pending_entries(bucket_info, *state, olh_obj, rm_pending_entries); | |
11635 | if (ret < 0) { | |
11636 | ldout(cct, 20) << "ERROR: rm_pending_entries returned ret=" << ret << dendl; | |
11637 | return ret; | |
11638 | } | |
11639 | } | |
11640 | if (!pending_entries.empty()) { | |
11641 | ldout(cct, 20) << __func__ << "(): found pending entries, need to update_olh() on bucket=" << olh_obj.bucket << dendl; | |
11642 | ||
11643 | int ret = update_olh(obj_ctx, state, bucket_info, olh_obj); | |
11644 | if (ret < 0) { | |
11645 | return ret; | |
11646 | } | |
11647 | } | |
11648 | ||
11649 | map<string, bufferlist>::iterator iter = state->attrset.find(RGW_ATTR_OLH_INFO); | |
11650 | assert(iter != state->attrset.end()); | |
11651 | RGWOLHInfo olh; | |
11652 | try { | |
11653 | bufferlist::iterator biter = iter->second.begin(); | |
11654 | ::decode(olh, biter); | |
11655 | } catch (buffer::error& err) { | |
11656 | ldout(cct, 0) << "ERROR: failed to decode olh info" << dendl; | |
11657 | return -EIO; | |
11658 | } | |
11659 | ||
11660 | if (olh.removed) { | |
11661 | return -ENOENT; | |
11662 | } | |
11663 | ||
11664 | *target = olh.target; | |
11665 | ||
11666 | return 0; | |
11667 | } | |
11668 | ||
11669 | int RGWRados::raw_obj_stat(rgw_raw_obj& obj, uint64_t *psize, real_time *pmtime, uint64_t *epoch, | |
11670 | map<string, bufferlist> *attrs, bufferlist *first_chunk, | |
11671 | RGWObjVersionTracker *objv_tracker) | |
11672 | { | |
11673 | rgw_rados_ref ref; | |
11674 | int r = get_raw_obj_ref(obj, &ref); | |
11675 | if (r < 0) { | |
11676 | return r; | |
11677 | } | |
11678 | ||
11679 | map<string, bufferlist> unfiltered_attrset; | |
11680 | uint64_t size = 0; | |
11681 | struct timespec mtime_ts; | |
11682 | ||
11683 | ObjectReadOperation op; | |
11684 | if (objv_tracker) { | |
11685 | objv_tracker->prepare_op_for_read(&op); | |
11686 | } | |
11687 | if (attrs) { | |
11688 | op.getxattrs(&unfiltered_attrset, NULL); | |
11689 | } | |
11690 | if (psize || pmtime) { | |
11691 | op.stat2(&size, &mtime_ts, NULL); | |
11692 | } | |
11693 | if (first_chunk) { | |
11694 | op.read(0, cct->_conf->rgw_max_chunk_size, first_chunk, NULL); | |
11695 | } | |
11696 | bufferlist outbl; | |
11697 | r = ref.ioctx.operate(ref.oid, &op, &outbl); | |
11698 | ||
11699 | if (epoch) { | |
11700 | *epoch = ref.ioctx.get_last_version(); | |
11701 | } | |
11702 | ||
11703 | if (r < 0) | |
11704 | return r; | |
11705 | ||
11706 | if (psize) | |
11707 | *psize = size; | |
11708 | if (pmtime) | |
11709 | *pmtime = ceph::real_clock::from_timespec(mtime_ts); | |
11710 | if (attrs) { | |
11711 | filter_attrset(unfiltered_attrset, RGW_ATTR_PREFIX, attrs); | |
11712 | } | |
11713 | ||
11714 | return 0; | |
11715 | } | |
11716 | ||
11717 | int RGWRados::get_bucket_stats(RGWBucketInfo& bucket_info, int shard_id, string *bucket_ver, string *master_ver, | |
c07f9fc5 | 11718 | map<RGWObjCategory, RGWStorageStats>& stats, string *max_marker, bool *syncstopped) |
7c673cae FG |
11719 | { |
11720 | map<string, rgw_bucket_dir_header> headers; | |
11721 | map<int, string> bucket_instance_ids; | |
11722 | int r = cls_bucket_head(bucket_info, shard_id, headers, &bucket_instance_ids); | |
11723 | if (r < 0) { | |
11724 | return r; | |
11725 | } | |
11726 | ||
11727 | assert(headers.size() == bucket_instance_ids.size()); | |
11728 | ||
11729 | map<string, rgw_bucket_dir_header>::iterator iter = headers.begin(); | |
11730 | map<int, string>::iterator viter = bucket_instance_ids.begin(); | |
11731 | BucketIndexShardsManager ver_mgr; | |
11732 | BucketIndexShardsManager master_ver_mgr; | |
11733 | BucketIndexShardsManager marker_mgr; | |
7c673cae FG |
11734 | char buf[64]; |
11735 | for(; iter != headers.end(); ++iter, ++viter) { | |
11736 | accumulate_raw_stats(iter->second, stats); | |
11737 | snprintf(buf, sizeof(buf), "%lu", (unsigned long)iter->second.ver); | |
11738 | ver_mgr.add(viter->first, string(buf)); | |
11739 | snprintf(buf, sizeof(buf), "%lu", (unsigned long)iter->second.master_ver); | |
11740 | master_ver_mgr.add(viter->first, string(buf)); | |
11741 | if (shard_id >= 0) { | |
11742 | *max_marker = iter->second.max_marker; | |
11743 | } else { | |
11744 | marker_mgr.add(viter->first, iter->second.max_marker); | |
11745 | } | |
c07f9fc5 FG |
11746 | if (syncstopped != NULL) |
11747 | *syncstopped = iter->second.syncstopped; | |
7c673cae FG |
11748 | } |
11749 | ver_mgr.to_string(bucket_ver); | |
11750 | master_ver_mgr.to_string(master_ver); | |
11751 | if (shard_id < 0) { | |
11752 | marker_mgr.to_string(max_marker); | |
11753 | } | |
11754 | return 0; | |
11755 | } | |
11756 | ||
11757 | int RGWRados::get_bi_log_status(RGWBucketInfo& bucket_info, int shard_id, | |
11758 | map<int, string>& markers) | |
11759 | { | |
11760 | map<string, rgw_bucket_dir_header> headers; | |
11761 | map<int, string> bucket_instance_ids; | |
11762 | int r = cls_bucket_head(bucket_info, shard_id, headers, &bucket_instance_ids); | |
11763 | if (r < 0) | |
11764 | return r; | |
11765 | ||
11766 | assert(headers.size() == bucket_instance_ids.size()); | |
11767 | ||
11768 | map<string, rgw_bucket_dir_header>::iterator iter = headers.begin(); | |
11769 | map<int, string>::iterator viter = bucket_instance_ids.begin(); | |
11770 | ||
11771 | for(; iter != headers.end(); ++iter, ++viter) { | |
11772 | if (shard_id >= 0) { | |
11773 | markers[shard_id] = iter->second.max_marker; | |
11774 | } else { | |
11775 | markers[viter->first] = iter->second.max_marker; | |
11776 | } | |
11777 | } | |
11778 | return 0; | |
11779 | } | |
11780 | ||
11781 | class RGWGetBucketStatsContext : public RGWGetDirHeader_CB { | |
11782 | RGWGetBucketStats_CB *cb; | |
11783 | uint32_t pendings; | |
11784 | map<RGWObjCategory, RGWStorageStats> stats; | |
11785 | int ret_code; | |
11786 | bool should_cb; | |
11787 | Mutex lock; | |
11788 | ||
11789 | public: | |
11790 | RGWGetBucketStatsContext(RGWGetBucketStats_CB *_cb, uint32_t _pendings) | |
11791 | : cb(_cb), pendings(_pendings), stats(), ret_code(0), should_cb(true), | |
11792 | lock("RGWGetBucketStatsContext") {} | |
11793 | ||
11794 | void handle_response(int r, rgw_bucket_dir_header& header) override { | |
11795 | Mutex::Locker l(lock); | |
11796 | if (should_cb) { | |
11797 | if ( r >= 0) { | |
11798 | accumulate_raw_stats(header, stats); | |
11799 | } else { | |
11800 | ret_code = r; | |
11801 | } | |
11802 | ||
11803 | // Are we all done? | |
11804 | if (--pendings == 0) { | |
11805 | if (!ret_code) { | |
11806 | cb->set_response(&stats); | |
11807 | } | |
11808 | cb->handle_response(ret_code); | |
11809 | cb->put(); | |
11810 | } | |
11811 | } | |
11812 | } | |
11813 | ||
11814 | void unset_cb() { | |
11815 | Mutex::Locker l(lock); | |
11816 | should_cb = false; | |
11817 | } | |
11818 | }; | |
11819 | ||
11820 | int RGWRados::get_bucket_stats_async(RGWBucketInfo& bucket_info, int shard_id, RGWGetBucketStats_CB *ctx) | |
11821 | { | |
11822 | int num_aio = 0; | |
c07f9fc5 | 11823 | RGWGetBucketStatsContext *get_ctx = new RGWGetBucketStatsContext(ctx, bucket_info.num_shards ? : 1); |
7c673cae FG |
11824 | assert(get_ctx); |
11825 | int r = cls_bucket_head_async(bucket_info, shard_id, get_ctx, &num_aio); | |
7c673cae FG |
11826 | if (r < 0) { |
11827 | ctx->put(); | |
11828 | if (num_aio) { | |
11829 | get_ctx->unset_cb(); | |
11830 | } | |
11831 | } | |
c07f9fc5 | 11832 | get_ctx->put(); |
7c673cae FG |
11833 | return r; |
11834 | } | |
11835 | ||
11836 | class RGWGetUserStatsContext : public RGWGetUserHeader_CB { | |
11837 | RGWGetUserStats_CB *cb; | |
11838 | ||
11839 | public: | |
11840 | explicit RGWGetUserStatsContext(RGWGetUserStats_CB * const cb) | |
11841 | : cb(cb) {} | |
11842 | ||
11843 | void handle_response(int r, cls_user_header& header) override { | |
11844 | const cls_user_stats& hs = header.stats; | |
11845 | if (r >= 0) { | |
11846 | RGWStorageStats stats; | |
11847 | ||
11848 | stats.size = hs.total_bytes; | |
11849 | stats.size_rounded = hs.total_bytes_rounded; | |
11850 | stats.num_objects = hs.total_entries; | |
11851 | ||
11852 | cb->set_response(stats); | |
11853 | } | |
11854 | ||
11855 | cb->handle_response(r); | |
11856 | ||
11857 | cb->put(); | |
11858 | } | |
11859 | }; | |
11860 | ||
11861 | int RGWRados::get_user_stats(const rgw_user& user, RGWStorageStats& stats) | |
11862 | { | |
11863 | string user_str = user.to_str(); | |
11864 | ||
11865 | cls_user_header header; | |
11866 | int r = cls_user_get_header(user_str, &header); | |
11867 | if (r < 0) | |
11868 | return r; | |
11869 | ||
11870 | const cls_user_stats& hs = header.stats; | |
11871 | ||
11872 | stats.size = hs.total_bytes; | |
11873 | stats.size_rounded = hs.total_bytes_rounded; | |
11874 | stats.num_objects = hs.total_entries; | |
11875 | ||
11876 | return 0; | |
11877 | } | |
11878 | ||
11879 | int RGWRados::get_user_stats_async(const rgw_user& user, RGWGetUserStats_CB *ctx) | |
11880 | { | |
11881 | string user_str = user.to_str(); | |
11882 | ||
11883 | RGWGetUserStatsContext *get_ctx = new RGWGetUserStatsContext(ctx); | |
11884 | int r = cls_user_get_header_async(user_str, get_ctx); | |
11885 | if (r < 0) { | |
11886 | ctx->put(); | |
11887 | delete get_ctx; | |
11888 | return r; | |
11889 | } | |
11890 | ||
11891 | return 0; | |
11892 | } | |
11893 | ||
11894 | void RGWRados::get_bucket_meta_oid(const rgw_bucket& bucket, string& oid) | |
11895 | { | |
11896 | oid = RGW_BUCKET_INSTANCE_MD_PREFIX + bucket.get_key(':'); | |
11897 | } | |
11898 | ||
11899 | void RGWRados::get_bucket_instance_obj(const rgw_bucket& bucket, rgw_raw_obj& obj) | |
11900 | { | |
11901 | if (!bucket.oid.empty()) { | |
11902 | obj.init(get_zone_params().domain_root, bucket.oid); | |
11903 | } else { | |
11904 | string oid; | |
11905 | get_bucket_meta_oid(bucket, oid); | |
11906 | obj.init(get_zone_params().domain_root, oid); | |
11907 | } | |
11908 | } | |
11909 | ||
11910 | int RGWRados::get_bucket_instance_info(RGWObjectCtx& obj_ctx, const string& meta_key, RGWBucketInfo& info, | |
11911 | real_time *pmtime, map<string, bufferlist> *pattrs) | |
11912 | { | |
11913 | size_t pos = meta_key.find(':'); | |
11914 | if (pos == string::npos) { | |
11915 | return -EINVAL; | |
11916 | } | |
11917 | string oid = RGW_BUCKET_INSTANCE_MD_PREFIX + meta_key; | |
11918 | rgw_bucket_instance_key_to_oid(oid); | |
11919 | ||
11920 | return get_bucket_instance_from_oid(obj_ctx, oid, info, pmtime, pattrs); | |
11921 | } | |
11922 | ||
11923 | int RGWRados::get_bucket_instance_info(RGWObjectCtx& obj_ctx, const rgw_bucket& bucket, RGWBucketInfo& info, | |
11924 | real_time *pmtime, map<string, bufferlist> *pattrs) | |
11925 | { | |
11926 | string oid; | |
11927 | if (bucket.oid.empty()) { | |
11928 | get_bucket_meta_oid(bucket, oid); | |
11929 | } else { | |
11930 | oid = bucket.oid; | |
11931 | } | |
11932 | ||
11933 | return get_bucket_instance_from_oid(obj_ctx, oid, info, pmtime, pattrs); | |
11934 | } | |
11935 | ||
31f18b77 | 11936 | int RGWRados::get_bucket_instance_from_oid(RGWObjectCtx& obj_ctx, const string& oid, RGWBucketInfo& info, |
7c673cae | 11937 | real_time *pmtime, map<string, bufferlist> *pattrs, |
b32b8144 FG |
11938 | rgw_cache_entry_info *cache_info, |
11939 | boost::optional<obj_version> refresh_version) | |
7c673cae FG |
11940 | { |
11941 | ldout(cct, 20) << "reading from " << get_zone_params().domain_root << ":" << oid << dendl; | |
11942 | ||
11943 | bufferlist epbl; | |
11944 | ||
b32b8144 FG |
11945 | int ret = rgw_get_system_obj(this, obj_ctx, get_zone_params().domain_root, |
11946 | oid, epbl, &info.objv_tracker, pmtime, pattrs, | |
11947 | cache_info, refresh_version); | |
7c673cae FG |
11948 | if (ret < 0) { |
11949 | return ret; | |
11950 | } | |
11951 | ||
11952 | bufferlist::iterator iter = epbl.begin(); | |
11953 | try { | |
11954 | ::decode(info, iter); | |
11955 | } catch (buffer::error& err) { | |
11956 | ldout(cct, 0) << "ERROR: could not decode buffer info, caught buffer::error" << dendl; | |
11957 | return -EIO; | |
11958 | } | |
11959 | info.bucket.oid = oid; | |
11960 | return 0; | |
11961 | } | |
11962 | ||
11963 | int RGWRados::get_bucket_entrypoint_info(RGWObjectCtx& obj_ctx, | |
11964 | const string& tenant_name, | |
11965 | const string& bucket_name, | |
11966 | RGWBucketEntryPoint& entry_point, | |
11967 | RGWObjVersionTracker *objv_tracker, | |
11968 | real_time *pmtime, | |
11969 | map<string, bufferlist> *pattrs, | |
b32b8144 FG |
11970 | rgw_cache_entry_info *cache_info, |
11971 | boost::optional<obj_version> refresh_version) | |
7c673cae FG |
11972 | { |
11973 | bufferlist bl; | |
11974 | string bucket_entry; | |
11975 | ||
11976 | rgw_make_bucket_entry_name(tenant_name, bucket_name, bucket_entry); | |
b32b8144 FG |
11977 | int ret = rgw_get_system_obj(this, obj_ctx, get_zone_params().domain_root, |
11978 | bucket_entry, bl, objv_tracker, pmtime, pattrs, | |
11979 | cache_info, refresh_version); | |
7c673cae FG |
11980 | if (ret < 0) { |
11981 | return ret; | |
11982 | } | |
11983 | ||
11984 | bufferlist::iterator iter = bl.begin(); | |
11985 | try { | |
11986 | ::decode(entry_point, iter); | |
11987 | } catch (buffer::error& err) { | |
11988 | ldout(cct, 0) << "ERROR: could not decode buffer info, caught buffer::error" << dendl; | |
11989 | return -EIO; | |
11990 | } | |
11991 | return 0; | |
11992 | } | |
11993 | ||
11994 | int RGWRados::convert_old_bucket_info(RGWObjectCtx& obj_ctx, | |
11995 | const string& tenant_name, | |
11996 | const string& bucket_name) | |
11997 | { | |
11998 | RGWBucketEntryPoint entry_point; | |
11999 | real_time ep_mtime; | |
12000 | RGWObjVersionTracker ot; | |
12001 | map<string, bufferlist> attrs; | |
12002 | RGWBucketInfo info; | |
12003 | ||
12004 | ldout(cct, 10) << "RGWRados::convert_old_bucket_info(): bucket=" << bucket_name << dendl; | |
12005 | ||
12006 | int ret = get_bucket_entrypoint_info(obj_ctx, tenant_name, bucket_name, entry_point, &ot, &ep_mtime, &attrs); | |
12007 | if (ret < 0) { | |
12008 | ldout(cct, 0) << "ERROR: get_bucket_entrypoint_info() returned " << ret << " bucket=" << bucket_name << dendl; | |
12009 | return ret; | |
12010 | } | |
12011 | ||
12012 | if (!entry_point.has_bucket_info) { | |
12013 | /* already converted! */ | |
12014 | return 0; | |
12015 | } | |
12016 | ||
12017 | info = entry_point.old_bucket_info; | |
12018 | info.bucket.oid = bucket_name; | |
12019 | info.ep_objv = ot.read_version; | |
12020 | ||
12021 | ot.generate_new_write_ver(cct); | |
12022 | ||
12023 | ret = put_linked_bucket_info(info, false, ep_mtime, &ot.write_version, &attrs, true); | |
12024 | if (ret < 0) { | |
12025 | ldout(cct, 0) << "ERROR: failed to put_linked_bucket_info(): " << ret << dendl; | |
12026 | return ret; | |
12027 | } | |
12028 | ||
12029 | return 0; | |
12030 | } | |
12031 | ||
b32b8144 FG |
12032 | int RGWRados::_get_bucket_info(RGWObjectCtx& obj_ctx, |
12033 | const string& tenant, | |
12034 | const string& bucket_name, | |
12035 | RGWBucketInfo& info, | |
12036 | real_time *pmtime, | |
12037 | map<string, bufferlist> *pattrs, | |
12038 | boost::optional<obj_version> refresh_version) | |
7c673cae FG |
12039 | { |
12040 | bucket_info_entry e; | |
12041 | string bucket_entry; | |
12042 | rgw_make_bucket_entry_name(tenant, bucket_name, bucket_entry); | |
12043 | ||
b32b8144 | 12044 | |
7c673cae | 12045 | if (binfo_cache->find(bucket_entry, &e)) { |
b32b8144 FG |
12046 | if (refresh_version && |
12047 | e.info.objv_tracker.read_version.compare(&(*refresh_version))) { | |
12048 | lderr(cct) << "WARNING: The bucket info cache is inconsistent. This is " | |
12049 | << "a failure that should be debugged. I am a nice machine, " | |
12050 | << "so I will try to recover." << dendl; | |
12051 | binfo_cache->invalidate(bucket_entry); | |
12052 | } | |
7c673cae FG |
12053 | info = e.info; |
12054 | if (pattrs) | |
12055 | *pattrs = e.attrs; | |
12056 | if (pmtime) | |
12057 | *pmtime = e.mtime; | |
12058 | return 0; | |
12059 | } | |
12060 | ||
12061 | RGWBucketEntryPoint entry_point; | |
12062 | real_time ep_mtime; | |
12063 | RGWObjVersionTracker ot; | |
12064 | rgw_cache_entry_info entry_cache_info; | |
b32b8144 FG |
12065 | int ret = get_bucket_entrypoint_info(obj_ctx, tenant, bucket_name, |
12066 | entry_point, &ot, &ep_mtime, pattrs, | |
12067 | &entry_cache_info, refresh_version); | |
7c673cae FG |
12068 | if (ret < 0) { |
12069 | /* only init these fields */ | |
12070 | info.bucket.tenant = tenant; | |
12071 | info.bucket.name = bucket_name; | |
12072 | return ret; | |
12073 | } | |
12074 | ||
12075 | if (entry_point.has_bucket_info) { | |
12076 | info = entry_point.old_bucket_info; | |
12077 | info.bucket.oid = bucket_name; | |
12078 | info.bucket.tenant = tenant; | |
12079 | info.ep_objv = ot.read_version; | |
12080 | ldout(cct, 20) << "rgw_get_bucket_info: old bucket info, bucket=" << info.bucket << " owner " << info.owner << dendl; | |
12081 | return 0; | |
12082 | } | |
12083 | ||
12084 | /* data is in the bucket instance object, we need to get attributes from there, clear everything | |
12085 | * that we got | |
12086 | */ | |
12087 | if (pattrs) { | |
12088 | pattrs->clear(); | |
12089 | } | |
12090 | ||
12091 | ldout(cct, 20) << "rgw_get_bucket_info: bucket instance: " << entry_point.bucket << dendl; | |
12092 | ||
12093 | ||
12094 | /* read bucket instance info */ | |
12095 | ||
12096 | string oid; | |
12097 | get_bucket_meta_oid(entry_point.bucket, oid); | |
12098 | ||
12099 | rgw_cache_entry_info cache_info; | |
12100 | ||
b32b8144 FG |
12101 | ret = get_bucket_instance_from_oid(obj_ctx, oid, e.info, &e.mtime, &e.attrs, |
12102 | &cache_info, refresh_version); | |
7c673cae FG |
12103 | e.info.ep_objv = ot.read_version; |
12104 | info = e.info; | |
12105 | if (ret < 0) { | |
b32b8144 | 12106 | lderr(cct) << "ERROR: get_bucket_instance_from_oid failed: " << ret << dendl; |
7c673cae FG |
12107 | info.bucket.tenant = tenant; |
12108 | info.bucket.name = bucket_name; | |
12109 | // XXX and why return anything in case of an error anyway? | |
12110 | return ret; | |
12111 | } | |
12112 | ||
12113 | if (pmtime) | |
12114 | *pmtime = e.mtime; | |
12115 | if (pattrs) | |
12116 | *pattrs = e.attrs; | |
12117 | ||
12118 | list<rgw_cache_entry_info *> cache_info_entries; | |
12119 | cache_info_entries.push_back(&entry_cache_info); | |
12120 | cache_info_entries.push_back(&cache_info); | |
12121 | ||
12122 | ||
12123 | /* chain to both bucket entry point and bucket instance */ | |
12124 | if (!binfo_cache->put(this, bucket_entry, &e, cache_info_entries)) { | |
12125 | ldout(cct, 20) << "couldn't put binfo cache entry, might have raced with data changes" << dendl; | |
12126 | } | |
12127 | ||
b32b8144 FG |
12128 | if (refresh_version && |
12129 | refresh_version->compare(&info.objv_tracker.read_version)) { | |
12130 | lderr(cct) << "WARNING: The OSD has the same version I have. Something may " | |
12131 | << "have gone squirrelly. An administrator may have forced a " | |
12132 | << "change; otherwise there is a problem somewhere." << dendl; | |
12133 | } | |
12134 | ||
7c673cae FG |
12135 | return 0; |
12136 | } | |
12137 | ||
b32b8144 FG |
12138 | int RGWRados::get_bucket_info(RGWObjectCtx& obj_ctx, |
12139 | const string& tenant, const string& bucket_name, | |
12140 | RGWBucketInfo& info, | |
12141 | real_time *pmtime, map<string, bufferlist> *pattrs) | |
12142 | { | |
12143 | return _get_bucket_info(obj_ctx, tenant, bucket_name, info, pmtime, | |
12144 | pattrs, boost::none); | |
12145 | } | |
12146 | ||
12147 | int RGWRados::try_refresh_bucket_info(RGWBucketInfo& info, | |
12148 | ceph::real_time *pmtime, | |
12149 | map<string, bufferlist> *pattrs) | |
12150 | { | |
12151 | RGWObjectCtx obj_ctx(this); | |
12152 | ||
12153 | return _get_bucket_info(obj_ctx, info.bucket.tenant, info.bucket.name, | |
12154 | info, pmtime, pattrs, info.objv_tracker.read_version); | |
12155 | } | |
12156 | ||
7c673cae FG |
12157 | int RGWRados::put_bucket_entrypoint_info(const string& tenant_name, const string& bucket_name, RGWBucketEntryPoint& entry_point, |
12158 | bool exclusive, RGWObjVersionTracker& objv_tracker, real_time mtime, | |
12159 | map<string, bufferlist> *pattrs) | |
12160 | { | |
12161 | bufferlist epbl; | |
12162 | ::encode(entry_point, epbl); | |
12163 | string bucket_entry; | |
12164 | rgw_make_bucket_entry_name(tenant_name, bucket_name, bucket_entry); | |
12165 | return rgw_bucket_store_info(this, bucket_entry, epbl, exclusive, pattrs, &objv_tracker, mtime); | |
12166 | } | |
12167 | ||
12168 | int RGWRados::put_bucket_instance_info(RGWBucketInfo& info, bool exclusive, | |
12169 | real_time mtime, map<string, bufferlist> *pattrs) | |
12170 | { | |
12171 | info.has_instance_obj = true; | |
12172 | bufferlist bl; | |
12173 | ||
12174 | ::encode(info, bl); | |
12175 | ||
12176 | string key = info.bucket.get_key(); /* when we go through meta api, we don't use oid directly */ | |
12177 | int ret = rgw_bucket_instance_store_info(this, key, bl, exclusive, pattrs, &info.objv_tracker, mtime); | |
12178 | if (ret == -EEXIST) { | |
12179 | /* well, if it's exclusive we shouldn't overwrite it, because we might race with another | |
12180 | * bucket operation on this specific bucket (e.g., being synced from the master), but | |
12181 | * since bucket instace meta object is unique for this specific bucket instace, we don't | |
12182 | * need to return an error. | |
12183 | * A scenario where we'd get -EEXIST here, is in a multi-zone config, we're not on the | |
12184 | * master, creating a bucket, sending bucket creation to the master, we create the bucket | |
12185 | * locally, while in the sync thread we sync the new bucket. | |
12186 | */ | |
12187 | ret = 0; | |
12188 | } | |
12189 | return ret; | |
12190 | } | |
12191 | ||
12192 | int RGWRados::put_linked_bucket_info(RGWBucketInfo& info, bool exclusive, real_time mtime, obj_version *pep_objv, | |
12193 | map<string, bufferlist> *pattrs, bool create_entry_point) | |
12194 | { | |
12195 | bool create_head = !info.has_instance_obj || create_entry_point; | |
12196 | ||
12197 | int ret = put_bucket_instance_info(info, exclusive, mtime, pattrs); | |
12198 | if (ret < 0) { | |
12199 | return ret; | |
12200 | } | |
12201 | ||
12202 | if (!create_head) | |
12203 | return 0; /* done! */ | |
12204 | ||
12205 | RGWBucketEntryPoint entry_point; | |
12206 | entry_point.bucket = info.bucket; | |
12207 | entry_point.owner = info.owner; | |
12208 | entry_point.creation_time = info.creation_time; | |
12209 | entry_point.linked = true; | |
12210 | RGWObjVersionTracker ot; | |
12211 | if (pep_objv && !pep_objv->tag.empty()) { | |
12212 | ot.write_version = *pep_objv; | |
12213 | } else { | |
12214 | ot.generate_new_write_ver(cct); | |
12215 | if (pep_objv) { | |
12216 | *pep_objv = ot.write_version; | |
12217 | } | |
12218 | } | |
12219 | ret = put_bucket_entrypoint_info(info.bucket.tenant, info.bucket.name, entry_point, exclusive, ot, mtime, NULL); | |
12220 | if (ret < 0) | |
12221 | return ret; | |
12222 | ||
12223 | return 0; | |
12224 | } | |
12225 | ||
12226 | int RGWRados::omap_get_vals(rgw_raw_obj& obj, bufferlist& header, const string& marker, uint64_t count, std::map<string, bufferlist>& m) | |
12227 | { | |
12228 | rgw_rados_ref ref; | |
12229 | int r = get_raw_obj_ref(obj, &ref); | |
12230 | if (r < 0) { | |
12231 | return r; | |
12232 | } | |
12233 | ||
12234 | r = ref.ioctx.omap_get_vals(ref.oid, marker, count, &m); | |
12235 | if (r < 0) | |
12236 | return r; | |
12237 | ||
12238 | return 0; | |
12239 | ||
12240 | } | |
12241 | ||
12242 | int RGWRados::omap_get_all(rgw_raw_obj& obj, bufferlist& header, | |
12243 | std::map<string, bufferlist>& m) | |
12244 | { | |
12245 | rgw_rados_ref ref; | |
12246 | int r = get_raw_obj_ref(obj, &ref); | |
12247 | if (r < 0) { | |
12248 | return r; | |
12249 | } | |
12250 | ||
12251 | #define MAX_OMAP_GET_ENTRIES 1024 | |
12252 | const int count = MAX_OMAP_GET_ENTRIES; | |
12253 | string start_after; | |
12254 | ||
12255 | while (true) { | |
12256 | std::map<string, bufferlist> t; | |
12257 | r = ref.ioctx.omap_get_vals(ref.oid, start_after, count, &t); | |
12258 | if (r < 0) { | |
12259 | return r; | |
12260 | } | |
12261 | if (t.empty()) { | |
12262 | break; | |
12263 | } | |
12264 | start_after = t.rbegin()->first; | |
12265 | m.insert(t.begin(), t.end()); | |
12266 | } | |
12267 | return 0; | |
12268 | } | |
12269 | ||
12270 | int RGWRados::omap_set(rgw_raw_obj& obj, const std::string& key, bufferlist& bl) | |
12271 | { | |
12272 | rgw_rados_ref ref; | |
12273 | int r = get_raw_obj_ref(obj, &ref); | |
12274 | if (r < 0) { | |
12275 | return r; | |
12276 | } | |
12277 | ldout(cct, 15) << "omap_set obj=" << obj << " key=" << key << dendl; | |
12278 | ||
12279 | map<string, bufferlist> m; | |
12280 | m[key] = bl; | |
12281 | ||
12282 | r = ref.ioctx.omap_set(ref.oid, m); | |
12283 | ||
12284 | return r; | |
12285 | } | |
12286 | ||
12287 | int RGWRados::omap_set(rgw_raw_obj& obj, std::map<std::string, bufferlist>& m) | |
12288 | { | |
12289 | rgw_rados_ref ref; | |
12290 | int r = get_raw_obj_ref(obj, &ref); | |
12291 | if (r < 0) { | |
12292 | return r; | |
12293 | } | |
12294 | ||
12295 | r = ref.ioctx.omap_set(ref.oid, m); | |
12296 | ||
12297 | return r; | |
12298 | } | |
12299 | ||
12300 | int RGWRados::omap_del(rgw_raw_obj& obj, const std::string& key) | |
12301 | { | |
12302 | rgw_rados_ref ref; | |
12303 | int r = get_raw_obj_ref(obj, &ref); | |
12304 | if (r < 0) { | |
12305 | return r; | |
12306 | } | |
12307 | ||
12308 | set<string> k; | |
12309 | k.insert(key); | |
12310 | ||
12311 | r = ref.ioctx.omap_rm_keys(ref.oid, k); | |
12312 | return r; | |
12313 | } | |
12314 | ||
12315 | int RGWRados::update_containers_stats(map<string, RGWBucketEnt>& m) | |
12316 | { | |
12317 | RGWObjectCtx obj_ctx(this); | |
12318 | ||
12319 | map<string, RGWBucketEnt>::iterator iter; | |
12320 | for (iter = m.begin(); iter != m.end(); ++iter) { | |
12321 | RGWBucketEnt& ent = iter->second; | |
12322 | rgw_bucket& bucket = ent.bucket; | |
12323 | ent.count = 0; | |
12324 | ent.size = 0; | |
12325 | ent.size_rounded = 0; | |
12326 | ||
12327 | map<string, rgw_bucket_dir_header> headers; | |
12328 | ||
12329 | RGWBucketInfo bucket_info; | |
12330 | int ret = get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL); | |
12331 | if (ret < 0) { | |
12332 | return ret; | |
12333 | } | |
12334 | ||
12335 | int r = cls_bucket_head(bucket_info, RGW_NO_SHARD, headers); | |
12336 | if (r < 0) | |
12337 | return r; | |
12338 | ||
12339 | map<string, rgw_bucket_dir_header>::iterator hiter = headers.begin(); | |
12340 | for (; hiter != headers.end(); ++hiter) { | |
12341 | RGWObjCategory category = main_category; | |
12342 | map<uint8_t, struct rgw_bucket_category_stats>::iterator iter = (hiter->second.stats).find((uint8_t)category); | |
12343 | if (iter != hiter->second.stats.end()) { | |
12344 | struct rgw_bucket_category_stats& stats = iter->second; | |
12345 | ent.count += stats.num_entries; | |
12346 | ent.size += stats.total_size; | |
12347 | ent.size_rounded += stats.total_size_rounded; | |
12348 | } | |
12349 | } | |
3efd9988 FG |
12350 | |
12351 | // fill in placement_rule from the bucket instance for use in swift's | |
12352 | // per-storage policy statistics | |
12353 | ent.placement_rule = std::move(bucket_info.placement_rule); | |
7c673cae FG |
12354 | } |
12355 | ||
12356 | return m.size(); | |
12357 | } | |
12358 | ||
12359 | int RGWRados::append_async(rgw_raw_obj& obj, size_t size, bufferlist& bl) | |
12360 | { | |
12361 | rgw_rados_ref ref; | |
12362 | int r = get_raw_obj_ref(obj, &ref); | |
12363 | if (r < 0) { | |
12364 | return r; | |
12365 | } | |
12366 | librados::Rados *rad = get_rados_handle(); | |
12367 | librados::AioCompletion *completion = rad->aio_create_completion(NULL, NULL, NULL); | |
12368 | ||
12369 | r = ref.ioctx.aio_append(ref.oid, completion, bl, size); | |
12370 | completion->release(); | |
12371 | return r; | |
12372 | } | |
12373 | ||
12374 | int RGWRados::distribute(const string& key, bufferlist& bl) | |
12375 | { | |
12376 | /* | |
12377 | * we were called before watch was initialized. This can only happen if we're updating some system | |
12378 | * config object (e.g., zone info) during init. Don't try to distribute the cache info for these | |
12379 | * objects, they're currently only read on startup anyway. | |
12380 | */ | |
12381 | if (!watch_initialized) | |
12382 | return 0; | |
12383 | ||
12384 | string notify_oid; | |
12385 | pick_control_oid(key, notify_oid); | |
12386 | ||
12387 | ldout(cct, 10) << "distributing notification oid=" << notify_oid << " bl.length()=" << bl.length() << dendl; | |
12388 | return control_pool_ctx.notify2(notify_oid, bl, 0, NULL); | |
12389 | } | |
12390 | ||
12391 | int RGWRados::pool_iterate_begin(const rgw_pool& pool, RGWPoolIterCtx& ctx) | |
12392 | { | |
12393 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
12394 | librados::NObjectIterator& iter = ctx.iter; | |
12395 | ||
12396 | int r = open_pool_ctx(pool, io_ctx); | |
12397 | if (r < 0) | |
12398 | return r; | |
12399 | ||
12400 | iter = io_ctx.nobjects_begin(); | |
12401 | ||
12402 | return 0; | |
12403 | } | |
12404 | ||
181888fb FG |
12405 | int RGWRados::pool_iterate_begin(const rgw_pool& pool, const string& cursor, RGWPoolIterCtx& ctx) |
12406 | { | |
12407 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
12408 | librados::NObjectIterator& iter = ctx.iter; | |
12409 | ||
12410 | int r = open_pool_ctx(pool, io_ctx); | |
12411 | if (r < 0) | |
12412 | return r; | |
12413 | ||
12414 | librados::ObjectCursor oc; | |
12415 | if (!oc.from_str(cursor)) { | |
12416 | ldout(cct, 10) << "failed to parse cursor: " << cursor << dendl; | |
12417 | return -EINVAL; | |
12418 | } | |
12419 | ||
12420 | iter = io_ctx.nobjects_begin(oc); | |
12421 | ||
12422 | return 0; | |
12423 | } | |
12424 | ||
12425 | string RGWRados::pool_iterate_get_cursor(RGWPoolIterCtx& ctx) | |
12426 | { | |
12427 | return ctx.iter.get_cursor().to_str(); | |
12428 | } | |
12429 | ||
7c673cae FG |
12430 | int RGWRados::pool_iterate(RGWPoolIterCtx& ctx, uint32_t num, vector<rgw_bucket_dir_entry>& objs, |
12431 | bool *is_truncated, RGWAccessListFilter *filter) | |
12432 | { | |
12433 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
12434 | librados::NObjectIterator& iter = ctx.iter; | |
12435 | ||
12436 | if (iter == io_ctx.nobjects_end()) | |
12437 | return -ENOENT; | |
12438 | ||
12439 | uint32_t i; | |
12440 | ||
12441 | for (i = 0; i < num && iter != io_ctx.nobjects_end(); ++i, ++iter) { | |
12442 | rgw_bucket_dir_entry e; | |
12443 | ||
12444 | string oid = iter->get_oid(); | |
12445 | ldout(cct, 20) << "RGWRados::pool_iterate: got " << oid << dendl; | |
12446 | ||
12447 | // fill it in with initial values; we may correct later | |
12448 | if (filter && !filter->filter(oid, oid)) | |
12449 | continue; | |
12450 | ||
12451 | e.key = oid; | |
12452 | objs.push_back(e); | |
12453 | } | |
12454 | ||
12455 | if (is_truncated) | |
12456 | *is_truncated = (iter != io_ctx.nobjects_end()); | |
12457 | ||
12458 | return objs.size(); | |
12459 | } | |
12460 | struct RGWAccessListFilterPrefix : public RGWAccessListFilter { | |
12461 | string prefix; | |
12462 | ||
12463 | explicit RGWAccessListFilterPrefix(const string& _prefix) : prefix(_prefix) {} | |
12464 | bool filter(string& name, string& key) override { | |
12465 | return (prefix.compare(key.substr(0, prefix.size())) == 0); | |
12466 | } | |
12467 | }; | |
12468 | ||
181888fb | 12469 | int RGWRados::list_raw_objects_init(const rgw_pool& pool, const string& marker, RGWListRawObjsCtx *ctx) |
7c673cae | 12470 | { |
181888fb FG |
12471 | if (!ctx->initialized) { |
12472 | int r = pool_iterate_begin(pool, marker, ctx->iter_ctx); | |
7c673cae FG |
12473 | if (r < 0) { |
12474 | ldout(cct, 10) << "failed to list objects pool_iterate_begin() returned r=" << r << dendl; | |
12475 | return r; | |
12476 | } | |
181888fb | 12477 | ctx->initialized = true; |
7c673cae | 12478 | } |
181888fb FG |
12479 | return 0; |
12480 | } | |
7c673cae | 12481 | |
181888fb FG |
12482 | int RGWRados::list_raw_objects_next(const string& prefix_filter, int max, |
12483 | RGWListRawObjsCtx& ctx, list<string>& oids, | |
12484 | bool *is_truncated) | |
12485 | { | |
12486 | if (!ctx.initialized) { | |
12487 | return -EINVAL; | |
12488 | } | |
12489 | RGWAccessListFilterPrefix filter(prefix_filter); | |
7c673cae FG |
12490 | vector<rgw_bucket_dir_entry> objs; |
12491 | int r = pool_iterate(ctx.iter_ctx, max, objs, is_truncated, &filter); | |
12492 | if (r < 0) { | |
12493 | if(r != -ENOENT) | |
12494 | ldout(cct, 10) << "failed to list objects pool_iterate returned r=" << r << dendl; | |
12495 | return r; | |
12496 | } | |
12497 | ||
12498 | vector<rgw_bucket_dir_entry>::iterator iter; | |
12499 | for (iter = objs.begin(); iter != objs.end(); ++iter) { | |
12500 | oids.push_back(iter->key.name); | |
12501 | } | |
12502 | ||
12503 | return oids.size(); | |
12504 | } | |
12505 | ||
181888fb FG |
12506 | int RGWRados::list_raw_objects(const rgw_pool& pool, const string& prefix_filter, |
12507 | int max, RGWListRawObjsCtx& ctx, list<string>& oids, | |
12508 | bool *is_truncated) | |
12509 | { | |
12510 | if (!ctx.initialized) { | |
12511 | int r = list_raw_objects_init(pool, string(), &ctx); | |
12512 | if (r < 0) { | |
12513 | return r; | |
12514 | } | |
12515 | } | |
12516 | ||
12517 | return list_raw_objects_next(prefix_filter, max, ctx, oids, is_truncated); | |
12518 | } | |
12519 | ||
12520 | string RGWRados::list_raw_objs_get_cursor(RGWListRawObjsCtx& ctx) | |
12521 | { | |
12522 | return pool_iterate_get_cursor(ctx.iter_ctx); | |
12523 | } | |
12524 | ||
7c673cae FG |
12525 | int RGWRados::list_bi_log_entries(RGWBucketInfo& bucket_info, int shard_id, string& marker, uint32_t max, |
12526 | std::list<rgw_bi_log_entry>& result, bool *truncated) | |
12527 | { | |
12528 | ldout(cct, 20) << __func__ << ": " << bucket_info.bucket << " marker " << marker << " shard_id=" << shard_id << " max " << max << dendl; | |
12529 | result.clear(); | |
12530 | ||
12531 | librados::IoCtx index_ctx; | |
12532 | map<int, string> oids; | |
12533 | map<int, cls_rgw_bi_log_list_ret> bi_log_lists; | |
12534 | map<int, string> bucket_instance_ids; | |
12535 | int r = open_bucket_index(bucket_info, index_ctx, oids, shard_id, &bucket_instance_ids); | |
12536 | if (r < 0) | |
12537 | return r; | |
12538 | ||
12539 | BucketIndexShardsManager marker_mgr; | |
12540 | bool has_shards = (oids.size() > 1 || shard_id >= 0); | |
12541 | // If there are multiple shards for the bucket index object, the marker | |
12542 | // should have the pattern '{shard_id_1}#{shard_marker_1},{shard_id_2}# | |
12543 | // {shard_marker_2}...', if there is no sharding, the bi_log_list should | |
12544 | // only contain one record, and the key is the bucket instance id. | |
12545 | r = marker_mgr.from_string(marker, shard_id); | |
12546 | if (r < 0) | |
12547 | return r; | |
12548 | ||
12549 | r = CLSRGWIssueBILogList(index_ctx, marker_mgr, max, oids, bi_log_lists, cct->_conf->rgw_bucket_index_max_aio)(); | |
12550 | if (r < 0) | |
12551 | return r; | |
12552 | ||
12553 | map<int, list<rgw_bi_log_entry>::iterator> vcurrents; | |
12554 | map<int, list<rgw_bi_log_entry>::iterator> vends; | |
12555 | if (truncated) { | |
12556 | *truncated = false; | |
12557 | } | |
12558 | map<int, cls_rgw_bi_log_list_ret>::iterator miter = bi_log_lists.begin(); | |
12559 | for (; miter != bi_log_lists.end(); ++miter) { | |
12560 | int shard_id = miter->first; | |
12561 | vcurrents[shard_id] = miter->second.entries.begin(); | |
12562 | vends[shard_id] = miter->second.entries.end(); | |
12563 | if (truncated) { | |
12564 | *truncated = (*truncated || miter->second.truncated); | |
12565 | } | |
12566 | } | |
12567 | ||
12568 | size_t total = 0; | |
12569 | bool has_more = true; | |
12570 | map<int, list<rgw_bi_log_entry>::iterator>::iterator viter; | |
12571 | map<int, list<rgw_bi_log_entry>::iterator>::iterator eiter; | |
12572 | while (total < max && has_more) { | |
12573 | has_more = false; | |
12574 | ||
12575 | viter = vcurrents.begin(); | |
12576 | eiter = vends.begin(); | |
12577 | ||
12578 | for (; total < max && viter != vcurrents.end(); ++viter, ++eiter) { | |
12579 | assert (eiter != vends.end()); | |
12580 | ||
12581 | int shard_id = viter->first; | |
12582 | list<rgw_bi_log_entry>::iterator& liter = viter->second; | |
12583 | ||
12584 | if (liter == eiter->second){ | |
12585 | continue; | |
12586 | } | |
12587 | rgw_bi_log_entry& entry = *(liter); | |
12588 | if (has_shards) { | |
12589 | char buf[16]; | |
12590 | snprintf(buf, sizeof(buf), "%d", shard_id); | |
12591 | string tmp_id; | |
12592 | build_bucket_index_marker(buf, entry.id, &tmp_id); | |
12593 | entry.id.swap(tmp_id); | |
12594 | } | |
12595 | marker_mgr.add(shard_id, entry.id); | |
12596 | result.push_back(entry); | |
12597 | total++; | |
12598 | has_more = true; | |
12599 | ++liter; | |
12600 | } | |
12601 | } | |
12602 | ||
12603 | if (truncated) { | |
12604 | for (viter = vcurrents.begin(), eiter = vends.begin(); viter != vcurrents.end(); ++viter, ++eiter) { | |
12605 | assert (eiter != vends.end()); | |
12606 | *truncated = (*truncated || (viter->second != eiter->second)); | |
12607 | } | |
12608 | } | |
12609 | ||
12610 | // Refresh marker, if there are multiple shards, the output will look like | |
12611 | // '{shard_oid_1}#{shard_marker_1},{shard_oid_2}#{shard_marker_2}...', | |
12612 | // if there is no sharding, the simply marker (without oid) is returned | |
12613 | if (has_shards) { | |
12614 | marker_mgr.to_string(&marker); | |
12615 | } else { | |
12616 | if (!result.empty()) { | |
12617 | marker = result.rbegin()->id; | |
12618 | } | |
12619 | } | |
12620 | ||
12621 | return 0; | |
12622 | } | |
12623 | ||
12624 | int RGWRados::trim_bi_log_entries(RGWBucketInfo& bucket_info, int shard_id, string& start_marker, string& end_marker) | |
12625 | { | |
12626 | librados::IoCtx index_ctx; | |
12627 | map<int, string> bucket_objs; | |
31f18b77 FG |
12628 | |
12629 | BucketIndexShardsManager start_marker_mgr; | |
12630 | BucketIndexShardsManager end_marker_mgr; | |
12631 | ||
7c673cae | 12632 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs, shard_id); |
31f18b77 | 12633 | if (r < 0) { |
7c673cae | 12634 | return r; |
31f18b77 | 12635 | } |
7c673cae | 12636 | |
7c673cae | 12637 | r = start_marker_mgr.from_string(start_marker, shard_id); |
31f18b77 | 12638 | if (r < 0) { |
7c673cae | 12639 | return r; |
31f18b77 FG |
12640 | } |
12641 | ||
7c673cae | 12642 | r = end_marker_mgr.from_string(end_marker, shard_id); |
31f18b77 | 12643 | if (r < 0) { |
7c673cae | 12644 | return r; |
31f18b77 | 12645 | } |
7c673cae FG |
12646 | |
12647 | return CLSRGWIssueBILogTrim(index_ctx, start_marker_mgr, end_marker_mgr, bucket_objs, | |
31f18b77 FG |
12648 | cct->_conf->rgw_bucket_index_max_aio)(); |
12649 | ||
12650 | return r; | |
7c673cae FG |
12651 | } |
12652 | ||
c07f9fc5 FG |
12653 | int RGWRados::resync_bi_log_entries(RGWBucketInfo& bucket_info, int shard_id) |
12654 | { | |
12655 | librados::IoCtx index_ctx; | |
12656 | map<int, string> bucket_objs; | |
12657 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs, shard_id); | |
12658 | if (r < 0) | |
12659 | return r; | |
12660 | ||
12661 | return CLSRGWIssueResyncBucketBILog(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); | |
12662 | } | |
12663 | ||
12664 | int RGWRados::stop_bi_log_entries(RGWBucketInfo& bucket_info, int shard_id) | |
12665 | { | |
12666 | librados::IoCtx index_ctx; | |
12667 | map<int, string> bucket_objs; | |
12668 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs, shard_id); | |
12669 | if (r < 0) | |
12670 | return r; | |
12671 | ||
12672 | return CLSRGWIssueBucketBILogStop(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); | |
12673 | } | |
12674 | ||
7c673cae FG |
12675 | int RGWRados::bi_get_instance(const RGWBucketInfo& bucket_info, rgw_obj& obj, rgw_bucket_dir_entry *dirent) |
12676 | { | |
12677 | rgw_rados_ref ref; | |
12678 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
12679 | if (r < 0) { | |
12680 | return r; | |
12681 | } | |
12682 | ||
12683 | rgw_cls_bi_entry bi_entry; | |
12684 | r = bi_get(obj.bucket, obj, InstanceIdx, &bi_entry); | |
12685 | if (r < 0 && r != -ENOENT) { | |
12686 | ldout(cct, 0) << "ERROR: bi_get() returned r=" << r << dendl; | |
12687 | } | |
12688 | if (r < 0) { | |
12689 | return r; | |
12690 | } | |
12691 | bufferlist::iterator iter = bi_entry.data.begin(); | |
12692 | try { | |
12693 | ::decode(*dirent, iter); | |
12694 | } catch (buffer::error& err) { | |
12695 | ldout(cct, 0) << "ERROR: failed to decode bi_entry()" << dendl; | |
12696 | return -EIO; | |
12697 | } | |
12698 | ||
12699 | return 0; | |
12700 | } | |
12701 | ||
12702 | int RGWRados::bi_get(rgw_bucket& bucket, rgw_obj& obj, BIIndexType index_type, rgw_cls_bi_entry *entry) | |
12703 | { | |
12704 | BucketShard bs(this); | |
12705 | int ret = bs.init(bucket, obj); | |
12706 | if (ret < 0) { | |
12707 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
12708 | return ret; | |
12709 | } | |
12710 | ||
12711 | cls_rgw_obj_key key(obj.key.get_index_key_name(), obj.key.instance); | |
12712 | ||
12713 | ret = cls_rgw_bi_get(bs.index_ctx, bs.bucket_obj, index_type, key, entry); | |
12714 | if (ret < 0) | |
12715 | return ret; | |
12716 | ||
12717 | return 0; | |
12718 | } | |
12719 | ||
12720 | void RGWRados::bi_put(ObjectWriteOperation& op, BucketShard& bs, rgw_cls_bi_entry& entry) | |
12721 | { | |
12722 | cls_rgw_bi_put(op, bs.bucket_obj, entry); | |
12723 | } | |
12724 | ||
12725 | int RGWRados::bi_put(BucketShard& bs, rgw_cls_bi_entry& entry) | |
12726 | { | |
12727 | int ret = cls_rgw_bi_put(bs.index_ctx, bs.bucket_obj, entry); | |
12728 | if (ret < 0) | |
12729 | return ret; | |
12730 | ||
12731 | return 0; | |
12732 | } | |
12733 | ||
12734 | int RGWRados::bi_put(rgw_bucket& bucket, rgw_obj& obj, rgw_cls_bi_entry& entry) | |
12735 | { | |
12736 | BucketShard bs(this); | |
12737 | int ret = bs.init(bucket, obj); | |
12738 | if (ret < 0) { | |
12739 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
12740 | return ret; | |
12741 | } | |
12742 | ||
12743 | return bi_put(bs, entry); | |
12744 | } | |
12745 | ||
12746 | 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) | |
12747 | { | |
12748 | rgw_obj obj(bucket, obj_name); | |
12749 | BucketShard bs(this); | |
12750 | int ret = bs.init(bucket, obj); | |
12751 | if (ret < 0) { | |
12752 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
12753 | return ret; | |
12754 | } | |
12755 | ||
12756 | ret = cls_rgw_bi_list(bs.index_ctx, bs.bucket_obj, obj_name, marker, max, entries, is_truncated); | |
31f18b77 FG |
12757 | if (ret == -ENOENT) { |
12758 | *is_truncated = false; | |
12759 | } | |
7c673cae FG |
12760 | if (ret < 0) |
12761 | return ret; | |
12762 | ||
12763 | return 0; | |
12764 | } | |
12765 | ||
12766 | 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) | |
12767 | { | |
12768 | int ret = cls_rgw_bi_list(bs.index_ctx, bs.bucket_obj, filter_obj, marker, max, entries, is_truncated); | |
12769 | if (ret < 0) | |
12770 | return ret; | |
12771 | ||
12772 | return 0; | |
12773 | } | |
12774 | ||
12775 | int RGWRados::bi_remove(BucketShard& bs) | |
12776 | { | |
12777 | int ret = bs.index_ctx.remove(bs.bucket_obj); | |
12778 | if (ret == -ENOENT) { | |
12779 | ret = 0; | |
12780 | } | |
12781 | if (ret < 0) { | |
12782 | ldout(cct, 5) << "bs.index_ctx.remove(" << bs.bucket_obj << ") returned ret=" << ret << dendl; | |
12783 | return ret; | |
12784 | } | |
12785 | ||
12786 | return 0; | |
12787 | } | |
12788 | ||
12789 | 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) | |
12790 | { | |
12791 | BucketShard bs(this); | |
12792 | int ret = bs.init(bucket, shard_id); | |
12793 | if (ret < 0) { | |
12794 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
12795 | return ret; | |
12796 | } | |
12797 | ||
12798 | return bi_list(bs, filter_obj, marker, max, entries, is_truncated); | |
12799 | } | |
12800 | ||
12801 | int RGWRados::gc_operate(string& oid, librados::ObjectWriteOperation *op) | |
12802 | { | |
12803 | return gc_pool_ctx.operate(oid, op); | |
12804 | } | |
12805 | ||
12806 | int RGWRados::gc_aio_operate(string& oid, librados::ObjectWriteOperation *op) | |
12807 | { | |
12808 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
12809 | int r = gc_pool_ctx.aio_operate(oid, c, op); | |
12810 | c->release(); | |
12811 | return r; | |
12812 | } | |
12813 | ||
12814 | int RGWRados::gc_operate(string& oid, librados::ObjectReadOperation *op, bufferlist *pbl) | |
12815 | { | |
12816 | return gc_pool_ctx.operate(oid, op, pbl); | |
12817 | } | |
12818 | ||
12819 | 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) | |
12820 | { | |
12821 | return gc->list(index, marker, max, expired_only, result, truncated); | |
12822 | } | |
12823 | ||
12824 | int RGWRados::process_gc() | |
12825 | { | |
12826 | return gc->process(); | |
12827 | } | |
12828 | ||
12829 | int RGWRados::list_lc_progress(const string& marker, uint32_t max_entries, map<string, int> *progress_map) | |
12830 | { | |
12831 | return lc->list_lc_progress(marker, max_entries, progress_map); | |
12832 | } | |
12833 | ||
12834 | int RGWRados::process_lc() | |
12835 | { | |
12836 | return lc->process(); | |
12837 | } | |
12838 | ||
12839 | int RGWRados::process_expire_objects() | |
12840 | { | |
12841 | obj_expirer->inspect_all_shards(utime_t(), ceph_clock_now()); | |
12842 | return 0; | |
12843 | } | |
12844 | ||
12845 | int RGWRados::cls_rgw_init_index(librados::IoCtx& index_ctx, librados::ObjectWriteOperation& op, string& oid) | |
12846 | { | |
12847 | bufferlist in; | |
12848 | cls_rgw_bucket_init(op); | |
12849 | return index_ctx.operate(oid, &op); | |
12850 | } | |
12851 | ||
12852 | int RGWRados::cls_obj_prepare_op(BucketShard& bs, RGWModifyOp op, string& tag, | |
31f18b77 | 12853 | rgw_obj& obj, uint16_t bilog_flags, rgw_zone_set *_zones_trace) |
7c673cae | 12854 | { |
31f18b77 FG |
12855 | rgw_zone_set zones_trace; |
12856 | if (_zones_trace) { | |
12857 | zones_trace = *_zones_trace; | |
12858 | } | |
12859 | else { | |
12860 | zones_trace.insert(get_zone().id); | |
12861 | } | |
12862 | ||
7c673cae FG |
12863 | ObjectWriteOperation o; |
12864 | cls_rgw_obj_key key(obj.key.get_index_key_name(), obj.key.instance); | |
31f18b77 FG |
12865 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); |
12866 | cls_rgw_bucket_prepare_op(o, op, tag, key, obj.key.get_loc(), get_zone().log_data, bilog_flags, zones_trace); | |
7c673cae FG |
12867 | return bs.index_ctx.operate(bs.bucket_obj, &o); |
12868 | } | |
12869 | ||
31f18b77 | 12870 | int RGWRados::cls_obj_complete_op(BucketShard& bs, const rgw_obj& obj, RGWModifyOp op, string& tag, |
7c673cae FG |
12871 | int64_t pool, uint64_t epoch, |
12872 | rgw_bucket_dir_entry& ent, RGWObjCategory category, | |
31f18b77 | 12873 | list<rgw_obj_index_key> *remove_objs, uint16_t bilog_flags, rgw_zone_set *_zones_trace) |
7c673cae | 12874 | { |
7c673cae FG |
12875 | ObjectWriteOperation o; |
12876 | rgw_bucket_dir_entry_meta dir_meta; | |
12877 | dir_meta = ent.meta; | |
12878 | dir_meta.category = category; | |
12879 | ||
12880 | rgw_bucket_entry_ver ver; | |
12881 | ver.pool = pool; | |
12882 | ver.epoch = epoch; | |
12883 | cls_rgw_obj_key key(ent.key.name, ent.key.instance); | |
31f18b77 FG |
12884 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); |
12885 | cls_rgw_bucket_complete_op(o, op, tag, ver, key, dir_meta, remove_objs, | |
12886 | get_zone().log_data, bilog_flags, _zones_trace); | |
12887 | complete_op_data *arg; | |
12888 | index_completion_manager->create_completion(obj, op, tag, ver, key, dir_meta, remove_objs, | |
12889 | get_zone().log_data, bilog_flags, _zones_trace, &arg); | |
12890 | librados::AioCompletion *completion = arg->rados_completion; | |
12891 | int ret = bs.index_ctx.aio_operate(bs.bucket_obj, arg->rados_completion, &o); | |
12892 | completion->release(); /* can't reference arg here, as it might have already been released */ | |
7c673cae FG |
12893 | return ret; |
12894 | } | |
12895 | ||
31f18b77 | 12896 | int RGWRados::cls_obj_complete_add(BucketShard& bs, const rgw_obj& obj, string& tag, |
7c673cae FG |
12897 | int64_t pool, uint64_t epoch, |
12898 | rgw_bucket_dir_entry& ent, RGWObjCategory category, | |
31f18b77 | 12899 | list<rgw_obj_index_key> *remove_objs, uint16_t bilog_flags, rgw_zone_set *zones_trace) |
7c673cae | 12900 | { |
31f18b77 | 12901 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_ADD, tag, pool, epoch, ent, category, remove_objs, bilog_flags, zones_trace); |
7c673cae FG |
12902 | } |
12903 | ||
12904 | int RGWRados::cls_obj_complete_del(BucketShard& bs, string& tag, | |
12905 | int64_t pool, uint64_t epoch, | |
12906 | rgw_obj& obj, | |
12907 | real_time& removed_mtime, | |
12908 | list<rgw_obj_index_key> *remove_objs, | |
31f18b77 FG |
12909 | uint16_t bilog_flags, |
12910 | rgw_zone_set *zones_trace) | |
7c673cae FG |
12911 | { |
12912 | rgw_bucket_dir_entry ent; | |
12913 | ent.meta.mtime = removed_mtime; | |
12914 | obj.key.get_index_key(&ent.key); | |
31f18b77 | 12915 | 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); |
7c673cae FG |
12916 | } |
12917 | ||
31f18b77 | 12918 | int RGWRados::cls_obj_complete_cancel(BucketShard& bs, string& tag, rgw_obj& obj, uint16_t bilog_flags, rgw_zone_set *zones_trace) |
7c673cae FG |
12919 | { |
12920 | rgw_bucket_dir_entry ent; | |
12921 | obj.key.get_index_key(&ent.key); | |
31f18b77 | 12922 | 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); |
7c673cae FG |
12923 | } |
12924 | ||
12925 | int RGWRados::cls_obj_set_bucket_tag_timeout(RGWBucketInfo& bucket_info, uint64_t timeout) | |
12926 | { | |
12927 | librados::IoCtx index_ctx; | |
12928 | map<int, string> bucket_objs; | |
12929 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs); | |
12930 | if (r < 0) | |
12931 | return r; | |
12932 | ||
12933 | return CLSRGWIssueSetTagTimeout(index_ctx, bucket_objs, cct->_conf->rgw_bucket_index_max_aio, timeout)(); | |
12934 | } | |
12935 | ||
12936 | int RGWRados::cls_bucket_list(RGWBucketInfo& bucket_info, int shard_id, rgw_obj_index_key& start, const string& prefix, | |
12937 | uint32_t num_entries, bool list_versions, map<string, rgw_bucket_dir_entry>& m, | |
12938 | bool *is_truncated, rgw_obj_index_key *last_entry, | |
12939 | bool (*force_check_filter)(const string& name)) | |
12940 | { | |
12941 | ldout(cct, 10) << "cls_bucket_list " << bucket_info.bucket << " start " << start.name << "[" << start.instance << "] num_entries " << num_entries << dendl; | |
12942 | ||
12943 | librados::IoCtx index_ctx; | |
12944 | // key - oid (for different shards if there is any) | |
12945 | // value - list result for the corresponding oid (shard), it is filled by the AIO callback | |
12946 | map<int, string> oids; | |
12947 | map<int, struct rgw_cls_list_ret> list_results; | |
12948 | int r = open_bucket_index(bucket_info, index_ctx, oids, shard_id); | |
12949 | if (r < 0) | |
12950 | return r; | |
12951 | ||
12952 | cls_rgw_obj_key start_key(start.name, start.instance); | |
12953 | r = CLSRGWIssueBucketList(index_ctx, start_key, prefix, num_entries, list_versions, | |
12954 | oids, list_results, cct->_conf->rgw_bucket_index_max_aio)(); | |
12955 | if (r < 0) | |
12956 | return r; | |
12957 | ||
12958 | // Create a list of iterators that are used to iterate each shard | |
12959 | vector<map<string, struct rgw_bucket_dir_entry>::iterator> vcurrents(list_results.size()); | |
12960 | vector<map<string, struct rgw_bucket_dir_entry>::iterator> vends(list_results.size()); | |
12961 | vector<string> vnames(list_results.size()); | |
12962 | map<int, struct rgw_cls_list_ret>::iterator iter = list_results.begin(); | |
12963 | *is_truncated = false; | |
12964 | for (; iter != list_results.end(); ++iter) { | |
12965 | vcurrents.push_back(iter->second.dir.m.begin()); | |
12966 | vends.push_back(iter->second.dir.m.end()); | |
12967 | vnames.push_back(oids[iter->first]); | |
12968 | *is_truncated = (*is_truncated || iter->second.is_truncated); | |
12969 | } | |
12970 | ||
12971 | // Create a map to track the next candidate entry from each shard, if the entry | |
12972 | // from a specified shard is selected/erased, the next entry from that shard will | |
12973 | // be inserted for next round selection | |
12974 | map<string, size_t> candidates; | |
12975 | for (size_t i = 0; i < vcurrents.size(); ++i) { | |
12976 | if (vcurrents[i] != vends[i]) { | |
12977 | candidates[vcurrents[i]->first] = i; | |
12978 | } | |
12979 | } | |
12980 | ||
12981 | map<string, bufferlist> updates; | |
12982 | uint32_t count = 0; | |
12983 | while (count < num_entries && !candidates.empty()) { | |
12984 | r = 0; | |
12985 | // Select the next one | |
12986 | int pos = candidates.begin()->second; | |
12987 | const string& name = vcurrents[pos]->first; | |
12988 | struct rgw_bucket_dir_entry& dirent = vcurrents[pos]->second; | |
12989 | ||
3efd9988 FG |
12990 | bool force_check = force_check_filter && |
12991 | force_check_filter(dirent.key.name); | |
12992 | if ((!dirent.exists && !dirent.is_delete_marker()) || | |
12993 | !dirent.pending_map.empty() || | |
12994 | force_check) { | |
7c673cae FG |
12995 | /* there are uncommitted ops. We need to check the current state, |
12996 | * and if the tags are old we need to do cleanup as well. */ | |
12997 | librados::IoCtx sub_ctx; | |
12998 | sub_ctx.dup(index_ctx); | |
12999 | r = check_disk_state(sub_ctx, bucket_info, dirent, dirent, updates[vnames[pos]]); | |
13000 | if (r < 0 && r != -ENOENT) { | |
13001 | return r; | |
13002 | } | |
13003 | } | |
13004 | if (r >= 0) { | |
13005 | ldout(cct, 10) << "RGWRados::cls_bucket_list: got " << dirent.key.name << "[" << dirent.key.instance << "]" << dendl; | |
13006 | m[name] = std::move(dirent); | |
13007 | ++count; | |
13008 | } | |
13009 | ||
13010 | // Refresh the candidates map | |
13011 | candidates.erase(candidates.begin()); | |
13012 | ++vcurrents[pos]; | |
13013 | if (vcurrents[pos] != vends[pos]) { | |
13014 | candidates[vcurrents[pos]->first] = pos; | |
13015 | } | |
13016 | } | |
13017 | ||
13018 | // Suggest updates if there is any | |
13019 | map<string, bufferlist>::iterator miter = updates.begin(); | |
13020 | for (; miter != updates.end(); ++miter) { | |
13021 | if (miter->second.length()) { | |
13022 | ObjectWriteOperation o; | |
13023 | cls_rgw_suggest_changes(o, miter->second); | |
13024 | // we don't care if we lose suggested updates, send them off blindly | |
13025 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
13026 | index_ctx.aio_operate(miter->first, c, &o); | |
13027 | c->release(); | |
13028 | } | |
13029 | } | |
13030 | ||
13031 | // Check if all the returned entries are consumed or not | |
13032 | for (size_t i = 0; i < vcurrents.size(); ++i) { | |
13033 | if (vcurrents[i] != vends[i]) | |
13034 | *is_truncated = true; | |
13035 | } | |
13036 | if (!m.empty()) | |
13037 | *last_entry = m.rbegin()->first; | |
13038 | ||
13039 | return 0; | |
13040 | } | |
13041 | ||
13042 | int RGWRados::cls_obj_usage_log_add(const string& oid, rgw_usage_log_info& info) | |
13043 | { | |
13044 | rgw_raw_obj obj(get_zone_params().usage_log_pool, oid); | |
13045 | ||
13046 | rgw_rados_ref ref; | |
224ce89b | 13047 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
13048 | if (r < 0) { |
13049 | return r; | |
13050 | } | |
13051 | ||
13052 | ObjectWriteOperation op; | |
13053 | cls_rgw_usage_log_add(op, info); | |
13054 | ||
13055 | r = ref.ioctx.operate(ref.oid, &op); | |
13056 | return r; | |
13057 | } | |
13058 | ||
13059 | int RGWRados::cls_obj_usage_log_read(string& oid, string& user, uint64_t start_epoch, uint64_t end_epoch, uint32_t max_entries, | |
13060 | string& read_iter, map<rgw_user_bucket, rgw_usage_log_entry>& usage, bool *is_truncated) | |
13061 | { | |
13062 | rgw_raw_obj obj(get_zone_params().usage_log_pool, oid); | |
13063 | ||
13064 | rgw_rados_ref ref; | |
224ce89b | 13065 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
13066 | if (r < 0) { |
13067 | return r; | |
13068 | } | |
13069 | ||
13070 | *is_truncated = false; | |
13071 | ||
13072 | r = cls_rgw_usage_log_read(ref.ioctx, ref.oid, user, start_epoch, end_epoch, | |
13073 | max_entries, read_iter, usage, is_truncated); | |
13074 | ||
13075 | return r; | |
13076 | } | |
13077 | ||
13078 | int RGWRados::cls_obj_usage_log_trim(string& oid, string& user, uint64_t start_epoch, uint64_t end_epoch) | |
13079 | { | |
13080 | rgw_raw_obj obj(get_zone_params().usage_log_pool, oid); | |
13081 | ||
13082 | rgw_rados_ref ref; | |
224ce89b | 13083 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
13084 | if (r < 0) { |
13085 | return r; | |
13086 | } | |
13087 | ||
b32b8144 | 13088 | r = cls_rgw_usage_log_trim(ref.ioctx, ref.oid, user, start_epoch, end_epoch); |
7c673cae FG |
13089 | return r; |
13090 | } | |
13091 | ||
13092 | int RGWRados::remove_objs_from_index(RGWBucketInfo& bucket_info, list<rgw_obj_index_key>& oid_list) | |
13093 | { | |
13094 | librados::IoCtx index_ctx; | |
13095 | string dir_oid; | |
13096 | ||
13097 | uint8_t suggest_flag = (get_zone().log_data ? CEPH_RGW_DIR_SUGGEST_LOG_OP : 0); | |
13098 | ||
13099 | int r = open_bucket_index(bucket_info, index_ctx, dir_oid); | |
13100 | if (r < 0) | |
13101 | return r; | |
13102 | ||
13103 | bufferlist updates; | |
13104 | ||
13105 | for (auto iter = oid_list.begin(); iter != oid_list.end(); ++iter) { | |
13106 | rgw_bucket_dir_entry entry; | |
13107 | entry.key = *iter; | |
13108 | dout(2) << "RGWRados::remove_objs_from_index bucket=" << bucket_info.bucket << " obj=" << entry.key.name << ":" << entry.key.instance << dendl; | |
13109 | entry.ver.epoch = (uint64_t)-1; // ULLONG_MAX, needed to that objclass doesn't skip out request | |
13110 | updates.append(CEPH_RGW_REMOVE | suggest_flag); | |
13111 | ::encode(entry, updates); | |
13112 | } | |
13113 | ||
13114 | bufferlist out; | |
13115 | ||
13116 | r = index_ctx.exec(dir_oid, RGW_CLASS, RGW_DIR_SUGGEST_CHANGES, updates, out); | |
13117 | ||
13118 | return r; | |
13119 | } | |
13120 | ||
13121 | int RGWRados::check_disk_state(librados::IoCtx io_ctx, | |
13122 | const RGWBucketInfo& bucket_info, | |
13123 | rgw_bucket_dir_entry& list_state, | |
13124 | rgw_bucket_dir_entry& object, | |
13125 | bufferlist& suggested_updates) | |
13126 | { | |
13127 | const rgw_bucket& bucket = bucket_info.bucket; | |
13128 | uint8_t suggest_flag = (get_zone().log_data ? CEPH_RGW_DIR_SUGGEST_LOG_OP : 0); | |
13129 | ||
13130 | std::string loc; | |
13131 | ||
13132 | rgw_obj obj(bucket, list_state.key); | |
13133 | ||
13134 | string oid; | |
13135 | get_obj_bucket_and_oid_loc(obj, oid, loc); | |
13136 | ||
13137 | if (loc != list_state.locator) { | |
13138 | ldout(cct, 0) << "WARNING: generated locator (" << loc << ") is different from listed locator (" << list_state.locator << ")" << dendl; | |
13139 | } | |
13140 | ||
13141 | io_ctx.locator_set_key(list_state.locator); | |
13142 | ||
13143 | RGWObjState *astate = NULL; | |
13144 | RGWObjectCtx rctx(this); | |
13145 | int r = get_obj_state(&rctx, bucket_info, obj, &astate, false); | |
13146 | if (r < 0) | |
13147 | return r; | |
13148 | ||
13149 | list_state.pending_map.clear(); // we don't need this and it inflates size | |
13150 | if (!astate->exists) { | |
13151 | /* object doesn't exist right now -- hopefully because it's | |
13152 | * marked as !exists and got deleted */ | |
13153 | if (list_state.exists) { | |
13154 | /* FIXME: what should happen now? Work out if there are any | |
13155 | * non-bad ways this could happen (there probably are, but annoying | |
13156 | * to handle!) */ | |
13157 | } | |
13158 | // encode a suggested removal of that key | |
13159 | list_state.ver.epoch = io_ctx.get_last_version(); | |
13160 | list_state.ver.pool = io_ctx.get_id(); | |
13161 | cls_rgw_encode_suggestion(CEPH_RGW_REMOVE, list_state, suggested_updates); | |
13162 | return -ENOENT; | |
13163 | } | |
13164 | ||
13165 | string etag; | |
13166 | string content_type; | |
13167 | ACLOwner owner; | |
13168 | ||
13169 | object.meta.size = astate->size; | |
13170 | object.meta.accounted_size = astate->accounted_size; | |
13171 | object.meta.mtime = astate->mtime; | |
13172 | ||
13173 | map<string, bufferlist>::iterator iter = astate->attrset.find(RGW_ATTR_ETAG); | |
13174 | if (iter != astate->attrset.end()) { | |
13175 | etag = iter->second.c_str(); | |
13176 | } | |
13177 | iter = astate->attrset.find(RGW_ATTR_CONTENT_TYPE); | |
13178 | if (iter != astate->attrset.end()) { | |
13179 | content_type = iter->second.c_str(); | |
13180 | } | |
13181 | iter = astate->attrset.find(RGW_ATTR_ACL); | |
13182 | if (iter != astate->attrset.end()) { | |
13183 | r = decode_policy(iter->second, &owner); | |
13184 | if (r < 0) { | |
13185 | dout(0) << "WARNING: could not decode policy for object: " << obj << dendl; | |
13186 | } | |
13187 | } | |
13188 | ||
13189 | if (astate->has_manifest) { | |
13190 | RGWObjManifest::obj_iterator miter; | |
13191 | RGWObjManifest& manifest = astate->manifest; | |
13192 | for (miter = manifest.obj_begin(); miter != manifest.obj_end(); ++miter) { | |
13193 | const rgw_raw_obj& raw_loc = miter.get_location().get_raw_obj(this); | |
13194 | rgw_obj loc; | |
13195 | rgw_raw_obj_to_obj(manifest.get_obj().bucket, raw_loc, &loc); | |
13196 | ||
13197 | if (loc.key.ns == RGW_OBJ_NS_MULTIPART) { | |
13198 | dout(10) << "check_disk_state(): removing manifest part from index: " << loc << dendl; | |
13199 | r = delete_obj_index(loc); | |
13200 | if (r < 0) { | |
13201 | dout(0) << "WARNING: delete_obj_index() returned r=" << r << dendl; | |
13202 | } | |
13203 | } | |
13204 | } | |
13205 | } | |
13206 | ||
13207 | object.meta.etag = etag; | |
13208 | object.meta.content_type = content_type; | |
13209 | object.meta.owner = owner.get_id().to_str(); | |
13210 | object.meta.owner_display_name = owner.get_display_name(); | |
13211 | ||
13212 | // encode suggested updates | |
13213 | list_state.ver.pool = io_ctx.get_id(); | |
13214 | list_state.ver.epoch = astate->epoch; | |
13215 | list_state.meta.size = object.meta.size; | |
13216 | list_state.meta.accounted_size = object.meta.accounted_size; | |
13217 | list_state.meta.mtime = object.meta.mtime; | |
13218 | list_state.meta.category = main_category; | |
13219 | list_state.meta.etag = etag; | |
13220 | list_state.meta.content_type = content_type; | |
13221 | if (astate->obj_tag.length() > 0) | |
13222 | list_state.tag = astate->obj_tag.c_str(); | |
13223 | list_state.meta.owner = owner.get_id().to_str(); | |
13224 | list_state.meta.owner_display_name = owner.get_display_name(); | |
13225 | ||
13226 | list_state.exists = true; | |
13227 | cls_rgw_encode_suggestion(CEPH_RGW_UPDATE | suggest_flag, list_state, suggested_updates); | |
13228 | return 0; | |
13229 | } | |
13230 | ||
13231 | int RGWRados::cls_bucket_head(const RGWBucketInfo& bucket_info, int shard_id, map<string, struct rgw_bucket_dir_header>& headers, map<int, string> *bucket_instance_ids) | |
13232 | { | |
13233 | librados::IoCtx index_ctx; | |
13234 | map<int, string> oids; | |
13235 | map<int, struct rgw_cls_list_ret> list_results; | |
13236 | int r = open_bucket_index(bucket_info, index_ctx, oids, list_results, shard_id, bucket_instance_ids); | |
13237 | if (r < 0) | |
13238 | return r; | |
13239 | ||
13240 | r = CLSRGWIssueGetDirHeader(index_ctx, oids, list_results, cct->_conf->rgw_bucket_index_max_aio)(); | |
13241 | if (r < 0) | |
13242 | return r; | |
13243 | ||
13244 | map<int, struct rgw_cls_list_ret>::iterator iter = list_results.begin(); | |
13245 | for(; iter != list_results.end(); ++iter) { | |
13246 | headers[oids[iter->first]] = iter->second.dir.header; | |
13247 | } | |
13248 | return 0; | |
13249 | } | |
13250 | ||
13251 | int RGWRados::cls_bucket_head_async(const RGWBucketInfo& bucket_info, int shard_id, RGWGetDirHeader_CB *ctx, int *num_aio) | |
13252 | { | |
13253 | librados::IoCtx index_ctx; | |
13254 | map<int, string> bucket_objs; | |
13255 | int r = open_bucket_index(bucket_info, index_ctx, bucket_objs, shard_id); | |
13256 | if (r < 0) | |
13257 | return r; | |
13258 | ||
13259 | map<int, string>::iterator iter = bucket_objs.begin(); | |
13260 | for (; iter != bucket_objs.end(); ++iter) { | |
13261 | r = cls_rgw_get_dir_header_async(index_ctx, iter->second, static_cast<RGWGetDirHeader_CB*>(ctx->get())); | |
13262 | if (r < 0) { | |
13263 | ctx->put(); | |
13264 | break; | |
13265 | } else { | |
13266 | (*num_aio)++; | |
13267 | } | |
13268 | } | |
13269 | return r; | |
13270 | } | |
13271 | ||
13272 | int RGWRados::cls_user_get_header(const string& user_id, cls_user_header *header) | |
13273 | { | |
13274 | string buckets_obj_id; | |
13275 | rgw_get_buckets_obj(user_id, buckets_obj_id); | |
13276 | rgw_raw_obj obj(get_zone_params().user_uid_pool, buckets_obj_id); | |
13277 | ||
13278 | rgw_rados_ref ref; | |
224ce89b | 13279 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
13280 | if (r < 0) { |
13281 | return r; | |
13282 | } | |
13283 | ||
13284 | librados::ObjectReadOperation op; | |
13285 | int rc; | |
13286 | ::cls_user_get_header(op, header, &rc); | |
13287 | bufferlist ibl; | |
13288 | r = ref.ioctx.operate(ref.oid, &op, &ibl); | |
13289 | if (r < 0) | |
13290 | return r; | |
13291 | if (rc < 0) | |
13292 | return rc; | |
13293 | ||
13294 | return 0; | |
13295 | } | |
13296 | ||
94b18763 FG |
13297 | int RGWRados::cls_user_reset_stats(const string& user_id) |
13298 | { | |
13299 | string buckets_obj_id; | |
13300 | rgw_get_buckets_obj(user_id, buckets_obj_id); | |
13301 | rgw_raw_obj obj(get_zone_params().user_uid_pool, buckets_obj_id); | |
13302 | ||
13303 | rgw_rados_ref ref; | |
13304 | int r = get_raw_obj_ref(obj, &ref); | |
13305 | if (r < 0) { | |
13306 | return r; | |
13307 | } | |
13308 | ||
13309 | librados::ObjectWriteOperation op; | |
13310 | ::cls_user_reset_stats(op); | |
13311 | return ref.ioctx.operate(ref.oid, &op); | |
13312 | } | |
13313 | ||
7c673cae FG |
13314 | int RGWRados::cls_user_get_header_async(const string& user_id, RGWGetUserHeader_CB *ctx) |
13315 | { | |
13316 | string buckets_obj_id; | |
13317 | rgw_get_buckets_obj(user_id, buckets_obj_id); | |
13318 | rgw_raw_obj obj(get_zone_params().user_uid_pool, buckets_obj_id); | |
13319 | ||
13320 | rgw_rados_ref ref; | |
224ce89b | 13321 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
13322 | if (r < 0) { |
13323 | return r; | |
13324 | } | |
13325 | ||
13326 | r = ::cls_user_get_header_async(ref.ioctx, ref.oid, ctx); | |
13327 | if (r < 0) | |
13328 | return r; | |
13329 | ||
13330 | return 0; | |
13331 | } | |
13332 | ||
13333 | int RGWRados::cls_user_sync_bucket_stats(rgw_raw_obj& user_obj, const RGWBucketInfo& bucket_info) | |
13334 | { | |
13335 | map<string, struct rgw_bucket_dir_header> headers; | |
13336 | int r = cls_bucket_head(bucket_info, RGW_NO_SHARD, headers); | |
13337 | if (r < 0) { | |
13338 | ldout(cct, 20) << "cls_bucket_header() returned " << r << dendl; | |
13339 | return r; | |
13340 | } | |
13341 | ||
13342 | cls_user_bucket_entry entry; | |
13343 | ||
13344 | bucket_info.bucket.convert(&entry.bucket); | |
13345 | ||
c07f9fc5 FG |
13346 | for (const auto& hiter : headers) { |
13347 | for (const auto& iter : hiter.second.stats) { | |
13348 | const struct rgw_bucket_category_stats& header_stats = iter.second; | |
7c673cae FG |
13349 | entry.size += header_stats.total_size; |
13350 | entry.size_rounded += header_stats.total_size_rounded; | |
13351 | entry.count += header_stats.num_entries; | |
13352 | } | |
13353 | } | |
13354 | ||
13355 | list<cls_user_bucket_entry> entries; | |
13356 | entries.push_back(entry); | |
13357 | ||
13358 | r = cls_user_update_buckets(user_obj, entries, false); | |
13359 | if (r < 0) { | |
13360 | ldout(cct, 20) << "cls_user_update_buckets() returned " << r << dendl; | |
13361 | return r; | |
13362 | } | |
13363 | ||
13364 | return 0; | |
13365 | } | |
13366 | ||
c07f9fc5 FG |
13367 | int RGWRados::cls_user_get_bucket_stats(const rgw_bucket& bucket, cls_user_bucket_entry& entry) |
13368 | { | |
13369 | map<string, struct rgw_bucket_dir_header> headers; | |
13370 | RGWBucketInfo bucket_info; | |
13371 | RGWObjectCtx obj_ctx(this); | |
13372 | int ret = get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL); | |
13373 | if (ret < 0) { | |
13374 | return ret; | |
13375 | } | |
13376 | ||
13377 | ret = cls_bucket_head(bucket_info, RGW_NO_SHARD, headers); | |
13378 | if (ret < 0) { | |
13379 | ldout(cct, 20) << "cls_bucket_header() returned " << ret << dendl; | |
13380 | return ret; | |
13381 | } | |
13382 | ||
13383 | bucket.convert(&entry.bucket); | |
13384 | ||
13385 | for (const auto& hiter : headers) { | |
13386 | for (const auto& iter : hiter.second.stats) { | |
13387 | const struct rgw_bucket_category_stats& header_stats = iter.second; | |
13388 | entry.size += header_stats.total_size; | |
13389 | entry.size_rounded += header_stats.total_size_rounded; | |
13390 | entry.count += header_stats.num_entries; | |
13391 | } | |
13392 | } | |
13393 | ||
13394 | return 0; | |
13395 | } | |
13396 | ||
7c673cae FG |
13397 | int RGWRados::cls_user_list_buckets(rgw_raw_obj& obj, |
13398 | const string& in_marker, | |
13399 | const string& end_marker, | |
13400 | const int max_entries, | |
13401 | list<cls_user_bucket_entry>& entries, | |
13402 | string * const out_marker, | |
13403 | bool * const truncated) | |
13404 | { | |
13405 | rgw_rados_ref ref; | |
224ce89b | 13406 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
13407 | if (r < 0) { |
13408 | return r; | |
13409 | } | |
13410 | ||
13411 | librados::ObjectReadOperation op; | |
13412 | int rc; | |
13413 | ||
13414 | cls_user_bucket_list(op, in_marker, end_marker, max_entries, entries, out_marker, truncated, &rc); | |
13415 | bufferlist ibl; | |
13416 | r = ref.ioctx.operate(ref.oid, &op, &ibl); | |
13417 | if (r < 0) | |
13418 | return r; | |
13419 | if (rc < 0) | |
13420 | return rc; | |
13421 | ||
13422 | return 0; | |
13423 | } | |
13424 | ||
13425 | int RGWRados::cls_user_update_buckets(rgw_raw_obj& obj, list<cls_user_bucket_entry>& entries, bool add) | |
13426 | { | |
13427 | rgw_rados_ref ref; | |
224ce89b | 13428 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
13429 | if (r < 0) { |
13430 | return r; | |
13431 | } | |
13432 | ||
13433 | librados::ObjectWriteOperation op; | |
13434 | cls_user_set_buckets(op, entries, add); | |
13435 | r = ref.ioctx.operate(ref.oid, &op); | |
13436 | if (r < 0) | |
13437 | return r; | |
13438 | ||
13439 | return 0; | |
13440 | } | |
13441 | ||
13442 | int RGWRados::complete_sync_user_stats(const rgw_user& user_id) | |
13443 | { | |
13444 | string buckets_obj_id; | |
13445 | rgw_get_buckets_obj(user_id, buckets_obj_id); | |
13446 | rgw_raw_obj obj(get_zone_params().user_uid_pool, buckets_obj_id); | |
13447 | return cls_user_complete_stats_sync(obj); | |
13448 | } | |
13449 | ||
13450 | int RGWRados::cls_user_complete_stats_sync(rgw_raw_obj& obj) | |
13451 | { | |
13452 | rgw_rados_ref ref; | |
224ce89b | 13453 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
13454 | if (r < 0) { |
13455 | return r; | |
13456 | } | |
13457 | ||
13458 | librados::ObjectWriteOperation op; | |
13459 | ::cls_user_complete_stats_sync(op); | |
13460 | r = ref.ioctx.operate(ref.oid, &op); | |
13461 | if (r < 0) | |
13462 | return r; | |
13463 | ||
13464 | return 0; | |
13465 | } | |
13466 | ||
13467 | int RGWRados::cls_user_add_bucket(rgw_raw_obj& obj, const cls_user_bucket_entry& entry) | |
13468 | { | |
13469 | list<cls_user_bucket_entry> l; | |
13470 | l.push_back(entry); | |
13471 | ||
13472 | return cls_user_update_buckets(obj, l, true); | |
13473 | } | |
13474 | ||
13475 | int RGWRados::cls_user_remove_bucket(rgw_raw_obj& obj, const cls_user_bucket& bucket) | |
13476 | { | |
7c673cae | 13477 | rgw_rados_ref ref; |
224ce89b | 13478 | int r = get_system_obj_ref(obj, &ref); |
7c673cae FG |
13479 | if (r < 0) { |
13480 | return r; | |
13481 | } | |
13482 | ||
13483 | librados::ObjectWriteOperation op; | |
13484 | ::cls_user_remove_bucket(op, bucket); | |
13485 | r = ref.ioctx.operate(ref.oid, &op); | |
13486 | if (r < 0) | |
13487 | return r; | |
13488 | ||
13489 | return 0; | |
13490 | } | |
13491 | ||
224ce89b | 13492 | int RGWRados::check_bucket_shards(const RGWBucketInfo& bucket_info, const rgw_bucket& bucket, |
31f18b77 FG |
13493 | RGWQuotaInfo& bucket_quota) |
13494 | { | |
13495 | if (!cct->_conf->rgw_dynamic_resharding) { | |
13496 | return 0; | |
13497 | } | |
13498 | ||
13499 | bool need_resharding = false; | |
13500 | int num_source_shards = (bucket_info.num_shards > 0 ? bucket_info.num_shards : 1); | |
13501 | uint32_t suggested_num_shards; | |
13502 | ||
13503 | int ret = quota_handler->check_bucket_shards((uint64_t)cct->_conf->rgw_max_objs_per_shard, | |
13504 | num_source_shards, bucket_info.owner, bucket, bucket_quota, | |
13505 | 1, need_resharding, &suggested_num_shards); | |
13506 | if (ret < 0) { | |
13507 | return ret; | |
13508 | } | |
13509 | ||
13510 | if (need_resharding) { | |
224ce89b WB |
13511 | ldout(cct, 20) << __func__ << " bucket " << bucket.name << " need resharding " << |
13512 | " old num shards " << bucket_info.num_shards << " new num shards " << suggested_num_shards << | |
13513 | dendl; | |
31f18b77 FG |
13514 | return add_bucket_to_reshard(bucket_info, suggested_num_shards); |
13515 | } | |
13516 | ||
13517 | return ret; | |
13518 | } | |
13519 | ||
13520 | int RGWRados::add_bucket_to_reshard(const RGWBucketInfo& bucket_info, uint32_t new_num_shards) | |
13521 | { | |
13522 | RGWReshard reshard(this); | |
13523 | ||
13524 | uint32_t num_source_shards = (bucket_info.num_shards > 0 ? bucket_info.num_shards : 1); | |
13525 | ||
13526 | new_num_shards = min(new_num_shards, get_max_bucket_shards()); | |
13527 | if (new_num_shards <= num_source_shards) { | |
13528 | ldout(cct, 20) << "not resharding bucket name=" << bucket_info.bucket.name << ", orig_num=" << num_source_shards << ", new_num_shards=" << new_num_shards << dendl; | |
13529 | return 0; | |
13530 | } | |
13531 | ||
13532 | cls_rgw_reshard_entry entry; | |
13533 | entry.time = real_clock::now(); | |
13534 | entry.tenant = bucket_info.owner.tenant; | |
13535 | entry.bucket_name = bucket_info.bucket.name; | |
13536 | entry.bucket_id = bucket_info.bucket.bucket_id; | |
13537 | entry.old_num_shards = num_source_shards; | |
13538 | entry.new_num_shards = new_num_shards; | |
13539 | ||
13540 | return reshard.add(entry); | |
13541 | } | |
13542 | ||
7c673cae FG |
13543 | int RGWRados::check_quota(const rgw_user& bucket_owner, rgw_bucket& bucket, |
13544 | RGWQuotaInfo& user_quota, RGWQuotaInfo& bucket_quota, uint64_t obj_size) | |
13545 | { | |
13546 | return quota_handler->check_quota(bucket_owner, bucket, user_quota, bucket_quota, 1, obj_size); | |
13547 | } | |
13548 | ||
13549 | void RGWRados::get_bucket_index_objects(const string& bucket_oid_base, | |
13550 | uint32_t num_shards, map<int, string>& bucket_objects, int shard_id) | |
13551 | { | |
13552 | if (!num_shards) { | |
13553 | bucket_objects[0] = bucket_oid_base; | |
13554 | } else { | |
13555 | char buf[bucket_oid_base.size() + 32]; | |
13556 | if (shard_id < 0) { | |
13557 | for (uint32_t i = 0; i < num_shards; ++i) { | |
13558 | snprintf(buf, sizeof(buf), "%s.%d", bucket_oid_base.c_str(), i); | |
13559 | bucket_objects[i] = buf; | |
13560 | } | |
13561 | } else { | |
13562 | if ((uint32_t)shard_id > num_shards) { | |
13563 | return; | |
13564 | } | |
13565 | snprintf(buf, sizeof(buf), "%s.%d", bucket_oid_base.c_str(), shard_id); | |
13566 | bucket_objects[shard_id] = buf; | |
13567 | } | |
13568 | } | |
13569 | } | |
13570 | ||
13571 | void RGWRados::get_bucket_instance_ids(const RGWBucketInfo& bucket_info, int shard_id, map<int, string> *result) | |
13572 | { | |
13573 | const rgw_bucket& bucket = bucket_info.bucket; | |
13574 | string plain_id = bucket.name + ":" + bucket.bucket_id; | |
13575 | if (!bucket_info.num_shards) { | |
13576 | (*result)[0] = plain_id; | |
13577 | } else { | |
13578 | char buf[16]; | |
13579 | if (shard_id < 0) { | |
13580 | for (uint32_t i = 0; i < bucket_info.num_shards; ++i) { | |
13581 | snprintf(buf, sizeof(buf), ":%d", i); | |
13582 | (*result)[i] = plain_id + buf; | |
13583 | } | |
13584 | } else { | |
13585 | if ((uint32_t)shard_id > bucket_info.num_shards) { | |
13586 | return; | |
13587 | } | |
13588 | snprintf(buf, sizeof(buf), ":%d", shard_id); | |
13589 | (*result)[shard_id] = plain_id + buf; | |
13590 | } | |
13591 | } | |
13592 | } | |
13593 | ||
13594 | int RGWRados::get_target_shard_id(const RGWBucketInfo& bucket_info, const string& obj_key, | |
13595 | int *shard_id) | |
13596 | { | |
13597 | int r = 0; | |
13598 | switch (bucket_info.bucket_index_shard_hash_type) { | |
13599 | case RGWBucketInfo::MOD: | |
13600 | if (!bucket_info.num_shards) { | |
13601 | if (shard_id) { | |
13602 | *shard_id = -1; | |
13603 | } | |
13604 | } else { | |
13605 | uint32_t sid = ceph_str_hash_linux(obj_key.c_str(), obj_key.size()); | |
13606 | uint32_t sid2 = sid ^ ((sid & 0xFF) << 24); | |
31f18b77 | 13607 | sid = rgw_shards_mod(sid2, bucket_info.num_shards); |
7c673cae FG |
13608 | if (shard_id) { |
13609 | *shard_id = (int)sid; | |
13610 | } | |
13611 | } | |
13612 | break; | |
13613 | default: | |
13614 | r = -ENOTSUP; | |
13615 | } | |
13616 | return r; | |
13617 | } | |
13618 | ||
13619 | void RGWRados::get_bucket_index_object(const string& bucket_oid_base, uint32_t num_shards, | |
13620 | int shard_id, string *bucket_obj) | |
13621 | { | |
13622 | if (!num_shards) { | |
13623 | // By default with no sharding, we use the bucket oid as itself | |
13624 | (*bucket_obj) = bucket_oid_base; | |
13625 | } else { | |
13626 | char buf[bucket_oid_base.size() + 32]; | |
13627 | snprintf(buf, sizeof(buf), "%s.%d", bucket_oid_base.c_str(), shard_id); | |
13628 | (*bucket_obj) = buf; | |
13629 | } | |
13630 | } | |
13631 | ||
13632 | int RGWRados::get_bucket_index_object(const string& bucket_oid_base, const string& obj_key, | |
13633 | uint32_t num_shards, RGWBucketInfo::BIShardsHashType hash_type, string *bucket_obj, int *shard_id) | |
13634 | { | |
13635 | int r = 0; | |
13636 | switch (hash_type) { | |
13637 | case RGWBucketInfo::MOD: | |
13638 | if (!num_shards) { | |
13639 | // By default with no sharding, we use the bucket oid as itself | |
13640 | (*bucket_obj) = bucket_oid_base; | |
13641 | if (shard_id) { | |
13642 | *shard_id = -1; | |
13643 | } | |
13644 | } else { | |
13645 | uint32_t sid = ceph_str_hash_linux(obj_key.c_str(), obj_key.size()); | |
13646 | uint32_t sid2 = sid ^ ((sid & 0xFF) << 24); | |
31f18b77 | 13647 | sid = rgw_shards_mod(sid2, num_shards); |
7c673cae FG |
13648 | char buf[bucket_oid_base.size() + 32]; |
13649 | snprintf(buf, sizeof(buf), "%s.%d", bucket_oid_base.c_str(), sid); | |
13650 | (*bucket_obj) = buf; | |
13651 | if (shard_id) { | |
13652 | *shard_id = (int)sid; | |
13653 | } | |
13654 | } | |
13655 | break; | |
13656 | default: | |
13657 | r = -ENOTSUP; | |
13658 | } | |
13659 | return r; | |
13660 | } | |
13661 | ||
13662 | void RGWStateLog::oid_str(int shard, string& oid) { | |
13663 | oid = RGW_STATELOG_OBJ_PREFIX + module_name + "."; | |
13664 | char buf[16]; | |
13665 | snprintf(buf, sizeof(buf), "%d", shard); | |
13666 | oid += buf; | |
13667 | } | |
13668 | ||
13669 | int RGWStateLog::get_shard_num(const string& object) { | |
13670 | uint32_t val = ceph_str_hash_linux(object.c_str(), object.length()); | |
13671 | return val % num_shards; | |
13672 | } | |
13673 | ||
13674 | string RGWStateLog::get_oid(const string& object) { | |
13675 | int shard = get_shard_num(object); | |
13676 | string oid; | |
13677 | oid_str(shard, oid); | |
13678 | return oid; | |
13679 | } | |
13680 | ||
13681 | int RGWStateLog::open_ioctx(librados::IoCtx& ioctx) { | |
13682 | rgw_pool pool; | |
13683 | store->get_log_pool(pool); | |
13684 | int r = rgw_init_ioctx(store->get_rados_handle(), pool, ioctx); | |
13685 | if (r < 0) { | |
13686 | lderr(store->ctx()) << "ERROR: could not open rados pool" << dendl; | |
13687 | return r; | |
13688 | } | |
13689 | return 0; | |
13690 | } | |
13691 | ||
13692 | int RGWStateLog::store_entry(const string& client_id, const string& op_id, const string& object, | |
13693 | uint32_t state, bufferlist *bl, uint32_t *check_state) | |
13694 | { | |
13695 | if (client_id.empty() || | |
13696 | op_id.empty() || | |
13697 | object.empty()) { | |
13698 | ldout(store->ctx(), 0) << "client_id / op_id / object is empty" << dendl; | |
13699 | } | |
13700 | ||
13701 | librados::IoCtx ioctx; | |
13702 | int r = open_ioctx(ioctx); | |
13703 | if (r < 0) | |
13704 | return r; | |
13705 | ||
13706 | string oid = get_oid(object); | |
13707 | ||
13708 | librados::ObjectWriteOperation op; | |
13709 | if (check_state) { | |
13710 | cls_statelog_check_state(op, client_id, op_id, object, *check_state); | |
13711 | } | |
13712 | utime_t ts = ceph_clock_now(); | |
13713 | bufferlist nobl; | |
13714 | cls_statelog_add(op, client_id, op_id, object, ts, state, (bl ? *bl : nobl)); | |
13715 | r = ioctx.operate(oid, &op); | |
13716 | if (r < 0) { | |
13717 | return r; | |
13718 | } | |
13719 | ||
13720 | return 0; | |
13721 | } | |
13722 | ||
13723 | int RGWStateLog::remove_entry(const string& client_id, const string& op_id, const string& object) | |
13724 | { | |
13725 | if (client_id.empty() || | |
13726 | op_id.empty() || | |
13727 | object.empty()) { | |
13728 | ldout(store->ctx(), 0) << "client_id / op_id / object is empty" << dendl; | |
13729 | } | |
13730 | ||
13731 | librados::IoCtx ioctx; | |
13732 | int r = open_ioctx(ioctx); | |
13733 | if (r < 0) | |
13734 | return r; | |
13735 | ||
13736 | string oid = get_oid(object); | |
13737 | ||
13738 | librados::ObjectWriteOperation op; | |
13739 | cls_statelog_remove_by_object(op, object, op_id); | |
13740 | r = ioctx.operate(oid, &op); | |
13741 | if (r < 0) { | |
13742 | return r; | |
13743 | } | |
13744 | ||
13745 | return 0; | |
13746 | } | |
13747 | ||
13748 | void RGWStateLog::init_list_entries(const string& client_id, const string& op_id, const string& object, | |
13749 | void **handle) | |
13750 | { | |
13751 | list_state *state = new list_state; | |
13752 | state->client_id = client_id; | |
13753 | state->op_id = op_id; | |
13754 | state->object = object; | |
13755 | if (object.empty()) { | |
13756 | state->cur_shard = 0; | |
13757 | state->max_shard = num_shards - 1; | |
13758 | } else { | |
13759 | state->cur_shard = state->max_shard = get_shard_num(object); | |
13760 | } | |
13761 | *handle = (void *)state; | |
13762 | } | |
13763 | ||
13764 | int RGWStateLog::list_entries(void *handle, int max_entries, | |
13765 | list<cls_statelog_entry>& entries, | |
13766 | bool *done) | |
13767 | { | |
13768 | list_state *state = static_cast<list_state *>(handle); | |
13769 | ||
13770 | librados::IoCtx ioctx; | |
13771 | int r = open_ioctx(ioctx); | |
13772 | if (r < 0) | |
13773 | return r; | |
13774 | ||
13775 | entries.clear(); | |
13776 | ||
13777 | for (; state->cur_shard <= state->max_shard && max_entries > 0; ++state->cur_shard) { | |
13778 | string oid; | |
13779 | oid_str(state->cur_shard, oid); | |
13780 | ||
13781 | librados::ObjectReadOperation op; | |
13782 | list<cls_statelog_entry> ents; | |
13783 | bool truncated; | |
13784 | cls_statelog_list(op, state->client_id, state->op_id, state->object, state->marker, | |
13785 | max_entries, ents, &state->marker, &truncated); | |
13786 | bufferlist ibl; | |
13787 | r = ioctx.operate(oid, &op, &ibl); | |
13788 | if (r == -ENOENT) { | |
13789 | truncated = false; | |
13790 | r = 0; | |
13791 | } | |
13792 | if (r < 0) { | |
13793 | ldout(store->ctx(), 0) << "cls_statelog_list returned " << r << dendl; | |
13794 | return r; | |
13795 | } | |
13796 | ||
13797 | if (!truncated) { | |
13798 | state->marker.clear(); | |
13799 | } | |
13800 | ||
13801 | max_entries -= ents.size(); | |
13802 | ||
13803 | entries.splice(entries.end(), ents); | |
13804 | ||
13805 | if (truncated) | |
13806 | break; | |
13807 | } | |
13808 | ||
13809 | *done = (state->cur_shard > state->max_shard); | |
13810 | ||
13811 | return 0; | |
13812 | } | |
13813 | ||
13814 | void RGWStateLog::finish_list_entries(void *handle) | |
13815 | { | |
13816 | list_state *state = static_cast<list_state *>(handle); | |
13817 | delete state; | |
13818 | } | |
13819 | ||
13820 | void RGWStateLog::dump_entry(const cls_statelog_entry& entry, Formatter *f) | |
13821 | { | |
13822 | f->open_object_section("statelog_entry"); | |
13823 | f->dump_string("client_id", entry.client_id); | |
13824 | f->dump_string("op_id", entry.op_id); | |
13825 | f->dump_string("object", entry.object); | |
13826 | entry.timestamp.gmtime_nsec(f->dump_stream("timestamp")); | |
13827 | if (!dump_entry_internal(entry, f)) { | |
13828 | f->dump_int("state", entry.state); | |
13829 | } | |
13830 | f->close_section(); | |
13831 | } | |
13832 | ||
13833 | RGWOpState::RGWOpState(RGWRados *_store) : RGWStateLog(_store, _store->ctx()->_conf->rgw_num_zone_opstate_shards, string("obj_opstate")) | |
13834 | { | |
13835 | } | |
13836 | ||
13837 | bool RGWOpState::dump_entry_internal(const cls_statelog_entry& entry, Formatter *f) | |
13838 | { | |
13839 | string s; | |
13840 | switch ((OpState)entry.state) { | |
13841 | case OPSTATE_UNKNOWN: | |
13842 | s = "unknown"; | |
13843 | break; | |
13844 | case OPSTATE_IN_PROGRESS: | |
13845 | s = "in-progress"; | |
13846 | break; | |
13847 | case OPSTATE_COMPLETE: | |
13848 | s = "complete"; | |
13849 | break; | |
13850 | case OPSTATE_ERROR: | |
13851 | s = "error"; | |
13852 | break; | |
13853 | case OPSTATE_ABORT: | |
13854 | s = "abort"; | |
13855 | break; | |
13856 | case OPSTATE_CANCELLED: | |
13857 | s = "cancelled"; | |
13858 | break; | |
13859 | default: | |
13860 | s = "invalid"; | |
13861 | } | |
13862 | f->dump_string("state", s); | |
13863 | return true; | |
13864 | } | |
13865 | ||
13866 | int RGWOpState::state_from_str(const string& s, OpState *state) | |
13867 | { | |
13868 | if (s == "unknown") { | |
13869 | *state = OPSTATE_UNKNOWN; | |
13870 | } else if (s == "in-progress") { | |
13871 | *state = OPSTATE_IN_PROGRESS; | |
13872 | } else if (s == "complete") { | |
13873 | *state = OPSTATE_COMPLETE; | |
13874 | } else if (s == "error") { | |
13875 | *state = OPSTATE_ERROR; | |
13876 | } else if (s == "abort") { | |
13877 | *state = OPSTATE_ABORT; | |
13878 | } else if (s == "cancelled") { | |
13879 | *state = OPSTATE_CANCELLED; | |
13880 | } else { | |
13881 | return -EINVAL; | |
13882 | } | |
13883 | ||
13884 | return 0; | |
13885 | } | |
13886 | ||
13887 | int RGWOpState::set_state(const string& client_id, const string& op_id, const string& object, OpState state) | |
13888 | { | |
13889 | uint32_t s = (uint32_t)state; | |
13890 | return store_entry(client_id, op_id, object, s, NULL, NULL); | |
13891 | } | |
13892 | ||
13893 | int RGWOpState::renew_state(const string& client_id, const string& op_id, const string& object, OpState state) | |
13894 | { | |
13895 | uint32_t s = (uint32_t)state; | |
13896 | return store_entry(client_id, op_id, object, s, NULL, &s); | |
13897 | } | |
13898 | ||
13899 | RGWOpStateSingleOp::RGWOpStateSingleOp(RGWRados *store, const string& cid, const string& oid, | |
13900 | const string& obj) : os(store), client_id(cid), op_id(oid), object(obj) | |
13901 | { | |
13902 | cct = store->ctx(); | |
13903 | cur_state = RGWOpState::OPSTATE_UNKNOWN; | |
13904 | } | |
13905 | ||
13906 | int RGWOpStateSingleOp::set_state(RGWOpState::OpState state) { | |
13907 | last_update = real_clock::now(); | |
13908 | cur_state = state; | |
13909 | return os.set_state(client_id, op_id, object, state); | |
13910 | } | |
13911 | ||
13912 | int RGWOpStateSingleOp::renew_state() { | |
13913 | real_time now = real_clock::now(); | |
13914 | ||
13915 | int rate_limit_sec = cct->_conf->rgw_opstate_ratelimit_sec; | |
13916 | ||
13917 | if (rate_limit_sec && now - last_update < make_timespan(rate_limit_sec)) { | |
13918 | return 0; | |
13919 | } | |
13920 | ||
13921 | last_update = now; | |
13922 | return os.renew_state(client_id, op_id, object, cur_state); | |
13923 | } | |
13924 | ||
13925 | ||
13926 | uint64_t RGWRados::instance_id() | |
13927 | { | |
13928 | return get_rados_handle()->get_instance_id(); | |
13929 | } | |
13930 | ||
13931 | uint64_t RGWRados::next_bucket_id() | |
13932 | { | |
13933 | Mutex::Locker l(bucket_id_lock); | |
13934 | return ++max_bucket_id; | |
13935 | } | |
13936 | ||
28e407b8 AA |
13937 | RGWRados *RGWStoreManager::init_storage_provider(CephContext *cct, bool use_gc_thread, bool use_lc_thread, |
13938 | bool quota_threads, bool run_sync_thread, bool run_reshard_thread, bool use_cache) | |
7c673cae | 13939 | { |
7c673cae FG |
13940 | RGWRados *store = NULL; |
13941 | if (!use_cache) { | |
13942 | store = new RGWRados; | |
13943 | } else { | |
28e407b8 | 13944 | store = new RGWCache<RGWRados>; |
7c673cae FG |
13945 | } |
13946 | ||
31f18b77 | 13947 | if (store->initialize(cct, use_gc_thread, use_lc_thread, quota_threads, run_sync_thread, run_reshard_thread) < 0) { |
7c673cae FG |
13948 | delete store; |
13949 | return NULL; | |
13950 | } | |
13951 | ||
13952 | return store; | |
13953 | } | |
13954 | ||
13955 | RGWRados *RGWStoreManager::init_raw_storage_provider(CephContext *cct) | |
13956 | { | |
13957 | RGWRados *store = NULL; | |
13958 | store = new RGWRados; | |
13959 | ||
13960 | store->set_context(cct); | |
13961 | ||
13962 | if (store->init_rados() < 0) { | |
13963 | delete store; | |
13964 | return NULL; | |
13965 | } | |
13966 | ||
13967 | return store; | |
13968 | } | |
13969 | ||
13970 | void RGWStoreManager::close_storage(RGWRados *store) | |
13971 | { | |
13972 | if (!store) | |
13973 | return; | |
13974 | ||
13975 | store->finalize(); | |
13976 | ||
13977 | delete store; | |
13978 | } | |
13979 | ||
13980 | librados::Rados* RGWRados::get_rados_handle() | |
13981 | { | |
13982 | if (rados.size() == 1) { | |
13983 | return &rados[0]; | |
13984 | } else { | |
13985 | handle_lock.get_read(); | |
13986 | pthread_t id = pthread_self(); | |
13987 | std::map<pthread_t, int>:: iterator it = rados_map.find(id); | |
13988 | ||
13989 | if (it != rados_map.end()) { | |
13990 | handle_lock.put_read(); | |
13991 | return &rados[it->second]; | |
13992 | } else { | |
13993 | handle_lock.put_read(); | |
13994 | handle_lock.get_write(); | |
13995 | const uint32_t handle = next_rados_handle; | |
13996 | rados_map[id] = handle; | |
13997 | if (++next_rados_handle == rados.size()) { | |
13998 | next_rados_handle = 0; | |
13999 | } | |
14000 | handle_lock.put_write(); | |
14001 | return &rados[handle]; | |
14002 | } | |
14003 | } | |
14004 | } | |
14005 | ||
14006 | int RGWRados::delete_raw_obj_aio(const rgw_raw_obj& obj, list<librados::AioCompletion *>& handles) | |
14007 | { | |
14008 | rgw_rados_ref ref; | |
14009 | int ret = get_raw_obj_ref(obj, &ref); | |
14010 | if (ret < 0) { | |
14011 | lderr(cct) << "ERROR: failed to get obj ref with ret=" << ret << dendl; | |
14012 | return ret; | |
14013 | } | |
14014 | ||
14015 | ObjectWriteOperation op; | |
14016 | list<string> prefixes; | |
14017 | cls_rgw_remove_obj(op, prefixes); | |
14018 | ||
14019 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
14020 | ret = ref.ioctx.aio_operate(ref.oid, c, &op); | |
14021 | if (ret < 0) { | |
14022 | lderr(cct) << "ERROR: AioOperate failed with ret=" << ret << dendl; | |
14023 | c->release(); | |
14024 | return ret; | |
14025 | } | |
14026 | ||
14027 | handles.push_back(c); | |
14028 | ||
14029 | return 0; | |
14030 | } | |
14031 | ||
14032 | int RGWRados::delete_obj_aio(const rgw_obj& obj, | |
14033 | RGWBucketInfo& bucket_info, RGWObjState *astate, | |
14034 | list<librados::AioCompletion *>& handles, bool keep_index_consistent) | |
14035 | { | |
14036 | rgw_rados_ref ref; | |
14037 | int ret = get_obj_head_ref(bucket_info, obj, &ref); | |
14038 | if (ret < 0) { | |
14039 | lderr(cct) << "ERROR: failed to get obj ref with ret=" << ret << dendl; | |
14040 | return ret; | |
14041 | } | |
14042 | ||
14043 | if (keep_index_consistent) { | |
14044 | RGWRados::Bucket bop(this, bucket_info); | |
14045 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
14046 | ||
14047 | ret = index_op.prepare(CLS_RGW_OP_DEL, &astate->write_tag); | |
14048 | if (ret < 0) { | |
14049 | lderr(cct) << "ERROR: failed to prepare index op with ret=" << ret << dendl; | |
14050 | return ret; | |
14051 | } | |
14052 | } | |
14053 | ||
14054 | ObjectWriteOperation op; | |
14055 | list<string> prefixes; | |
14056 | cls_rgw_remove_obj(op, prefixes); | |
14057 | ||
14058 | AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL); | |
14059 | ret = ref.ioctx.aio_operate(ref.oid, c, &op); | |
14060 | if (ret < 0) { | |
14061 | lderr(cct) << "ERROR: AioOperate failed with ret=" << ret << dendl; | |
14062 | c->release(); | |
14063 | return ret; | |
14064 | } | |
14065 | ||
14066 | handles.push_back(c); | |
14067 | ||
14068 | if (keep_index_consistent) { | |
14069 | ret = delete_obj_index(obj); | |
14070 | if (ret < 0) { | |
14071 | lderr(cct) << "ERROR: failed to delete obj index with ret=" << ret << dendl; | |
14072 | return ret; | |
14073 | } | |
14074 | } | |
14075 | return ret; | |
14076 | } | |
14077 | ||
14078 | int rgw_compression_info_from_attrset(map<string, bufferlist>& attrs, bool& need_decompress, RGWCompressionInfo& cs_info) { | |
14079 | map<string, bufferlist>::iterator value = attrs.find(RGW_ATTR_COMPRESSION); | |
14080 | if (value != attrs.end()) { | |
14081 | bufferlist::iterator bliter = value->second.begin(); | |
14082 | try { | |
14083 | ::decode(cs_info, bliter); | |
14084 | } catch (buffer::error& err) { | |
14085 | return -EIO; | |
14086 | } | |
14087 | if (cs_info.blocks.size() == 0) { | |
14088 | return -EIO; | |
14089 | } | |
14090 | if (cs_info.compression_type != "none") | |
14091 | need_decompress = true; | |
14092 | else | |
14093 | need_decompress = false; | |
14094 | return 0; | |
14095 | } else { | |
14096 | need_decompress = false; | |
14097 | return 0; | |
14098 | } | |
14099 | } | |
14100 | ||
3a9019d9 FG |
14101 | bool RGWRados::call(std::string command, cmdmap_t& cmdmap, std::string format, |
14102 | bufferlist& out) | |
14103 | { | |
14104 | if (command == "cache list") { | |
14105 | boost::optional<std::string> filter; | |
14106 | auto i = cmdmap.find("filter"); | |
14107 | if (i != cmdmap.cend()) { | |
14108 | filter = boost::get<std::string>(i->second); | |
14109 | } | |
14110 | std::unique_ptr<Formatter> f(ceph::Formatter::create(format, "table")); | |
14111 | if (f) { | |
14112 | f->open_array_section("cache_entries"); | |
14113 | call_list(filter, f.get()); | |
14114 | f->close_section(); | |
14115 | f->flush(out); | |
14116 | return true; | |
14117 | } else { | |
14118 | out.append("Unable to create Formatter.\n"); | |
14119 | return false; | |
14120 | } | |
14121 | } else if (command == "cache inspect") { | |
14122 | std::unique_ptr<Formatter> f(ceph::Formatter::create(format, "json-pretty")); | |
14123 | if (f) { | |
14124 | const auto& target = boost::get<std::string>(cmdmap["target"]); | |
14125 | if (call_inspect(target, f.get())) { | |
14126 | f->flush(out); | |
14127 | return true; | |
14128 | } else { | |
14129 | out.append(string("Unable to find entry ") + target + string(".\n")); | |
14130 | return false; | |
14131 | } | |
14132 | } else { | |
14133 | out.append("Unable to create Formatter.\n"); | |
14134 | return false; | |
14135 | } | |
14136 | } else if (command == "cache erase") { | |
14137 | const auto& target = boost::get<std::string>(cmdmap["target"]); | |
14138 | if (call_erase(target)) { | |
14139 | return true; | |
14140 | } else { | |
14141 | out.append(string("Unable to find entry ") + target + string(".\n")); | |
14142 | return false; | |
14143 | } | |
14144 | } else if (command == "cache zap") { | |
14145 | call_zap(); | |
14146 | return true; | |
14147 | } | |
14148 | return false; | |
14149 | } | |
14150 | ||
14151 | void RGWRados::call_list(const boost::optional<std::string>&, | |
14152 | ceph::Formatter*) | |
14153 | { | |
14154 | return; | |
14155 | } | |
14156 | ||
14157 | bool RGWRados::call_inspect(const std::string&, Formatter*) | |
14158 | { | |
14159 | return false; | |
14160 | } | |
14161 | ||
14162 | bool RGWRados::call_erase(const std::string&) { | |
14163 | return false; | |
14164 | } | |
14165 | ||
14166 | void RGWRados::call_zap() { | |
14167 | return; | |
14168 | } |