]>
Commit | Line | Data |
---|---|---|
7c673cae | 1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
9f95a23c | 2 | // vim: ts=8 sw=2 smarttab ft=cpp |
7c673cae | 3 | |
31f18b77 | 4 | #include "include/compat.h" |
7c673cae FG |
5 | #include <errno.h> |
6 | #include <stdlib.h> | |
7 | #include <sys/types.h> | |
9f95a23c TL |
8 | #include <sstream> |
9 | ||
7c673cae | 10 | #include <boost/algorithm/string.hpp> |
11fdf7f2 | 11 | #include <string_view> |
7c673cae | 12 | |
11fdf7f2 | 13 | #include <boost/container/flat_set.hpp> |
7c673cae FG |
14 | #include <boost/format.hpp> |
15 | #include <boost/optional.hpp> | |
16 | #include <boost/utility/in_place_factory.hpp> | |
17 | ||
18 | #include "common/ceph_json.h" | |
7c673cae FG |
19 | |
20 | #include "common/errno.h" | |
21 | #include "common/Formatter.h" | |
22 | #include "common/Throttle.h" | |
7c673cae | 23 | |
9f95a23c | 24 | #include "rgw_sal.h" |
11fdf7f2 | 25 | #include "rgw_zone.h" |
7c673cae FG |
26 | #include "rgw_cache.h" |
27 | #include "rgw_acl.h" | |
28 | #include "rgw_acl_s3.h" /* for dumping s3policy in debug log */ | |
11fdf7f2 | 29 | #include "rgw_aio_throttle.h" |
7c673cae FG |
30 | #include "rgw_bucket.h" |
31 | #include "rgw_rest_conn.h" | |
32 | #include "rgw_cr_rados.h" | |
33 | #include "rgw_cr_rest.h" | |
11fdf7f2 | 34 | #include "rgw_putobj_processor.h" |
7c673cae FG |
35 | |
36 | #include "cls/rgw/cls_rgw_ops.h" | |
7c673cae FG |
37 | #include "cls/rgw/cls_rgw_client.h" |
38 | #include "cls/rgw/cls_rgw_const.h" | |
39 | #include "cls/refcount/cls_refcount_client.h" | |
40 | #include "cls/version/cls_version_client.h" | |
c07f9fc5 | 41 | #include "osd/osd_types.h" |
7c673cae FG |
42 | |
43 | #include "rgw_tools.h" | |
44 | #include "rgw_coroutine.h" | |
45 | #include "rgw_compression.h" | |
9f95a23c | 46 | #include "rgw_worker.h" |
7c673cae | 47 | |
7c673cae FG |
48 | #undef fork // fails to compile RGWPeriod::fork() below |
49 | ||
50 | #include "common/Clock.h" | |
51 | ||
7c673cae FG |
52 | using namespace librados; |
53 | ||
54 | #include <string> | |
55 | #include <iostream> | |
56 | #include <vector> | |
57 | #include <atomic> | |
58 | #include <list> | |
59 | #include <map> | |
11fdf7f2 | 60 | #include "include/random.h" |
7c673cae FG |
61 | |
62 | #include "rgw_gc.h" | |
63 | #include "rgw_lc.h" | |
64 | ||
65 | #include "rgw_object_expirer_core.h" | |
66 | #include "rgw_sync.h" | |
81eedcae | 67 | #include "rgw_sync_counters.h" |
11fdf7f2 | 68 | #include "rgw_sync_trace.h" |
9f95a23c TL |
69 | #include "rgw_trim_datalog.h" |
70 | #include "rgw_trim_mdlog.h" | |
7c673cae FG |
71 | #include "rgw_data_sync.h" |
72 | #include "rgw_realm_watcher.h" | |
31f18b77 | 73 | #include "rgw_reshard.h" |
7c673cae | 74 | |
11fdf7f2 TL |
75 | #include "services/svc_zone.h" |
76 | #include "services/svc_zone_utils.h" | |
77 | #include "services/svc_quota.h" | |
78 | #include "services/svc_sync_modules.h" | |
79 | #include "services/svc_sys_obj.h" | |
80 | #include "services/svc_sys_obj_cache.h" | |
9f95a23c TL |
81 | #include "services/svc_bucket.h" |
82 | #include "services/svc_mdlog.h" | |
83 | #include "services/svc_datalog_rados.h" | |
11fdf7f2 | 84 | |
7c673cae FG |
85 | #include "compressor/Compressor.h" |
86 | ||
11fdf7f2 TL |
87 | #ifdef WITH_LTTNG |
88 | #define TRACEPOINT_DEFINE | |
89 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
90 | #include "tracing/rgw_rados.h" | |
91 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
92 | #undef TRACEPOINT_DEFINE | |
93 | #else | |
94 | #define tracepoint(...) | |
95 | #endif | |
96 | ||
7c673cae FG |
97 | #define dout_context g_ceph_context |
98 | #define dout_subsys ceph_subsys_rgw | |
99 | ||
7c673cae | 100 | |
7c673cae | 101 | static string shadow_ns = "shadow"; |
7c673cae FG |
102 | static string default_bucket_index_pool_suffix = "rgw.buckets.index"; |
103 | static string default_storage_extra_pool_suffix = "rgw.buckets.non-ec"; | |
11fdf7f2 | 104 | |
11fdf7f2 | 105 | static RGWObjCategory main_category = RGWObjCategory::Main; |
7c673cae | 106 | #define RGW_USAGE_OBJ_PREFIX "usage." |
7c673cae FG |
107 | |
108 | #define dout_subsys ceph_subsys_rgw | |
109 | ||
110 | ||
111 | static bool rgw_get_obj_data_pool(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params, | |
11fdf7f2 TL |
112 | const rgw_placement_rule& head_placement_rule, |
113 | const rgw_obj& obj, rgw_pool *pool) | |
7c673cae | 114 | { |
11fdf7f2 | 115 | if (!zone_params.get_head_data_pool(head_placement_rule, obj, pool)) { |
7c673cae | 116 | RGWZonePlacementInfo placement; |
11fdf7f2 | 117 | if (!zone_params.get_placement(zonegroup.default_placement.name, &placement)) { |
7c673cae FG |
118 | return false; |
119 | } | |
120 | ||
121 | if (!obj.in_extra_data) { | |
11fdf7f2 | 122 | *pool = placement.get_data_pool(zonegroup.default_placement.storage_class); |
7c673cae | 123 | } else { |
31f18b77 | 124 | *pool = placement.get_data_extra_pool(); |
7c673cae FG |
125 | } |
126 | } | |
127 | ||
128 | return true; | |
129 | } | |
130 | ||
131 | static bool rgw_obj_to_raw(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params, | |
11fdf7f2 TL |
132 | const rgw_placement_rule& head_placement_rule, |
133 | const rgw_obj& obj, rgw_raw_obj *raw_obj) | |
7c673cae FG |
134 | { |
135 | get_obj_bucket_and_oid_loc(obj, raw_obj->oid, raw_obj->loc); | |
136 | ||
11fdf7f2 | 137 | return rgw_get_obj_data_pool(zonegroup, zone_params, head_placement_rule, obj, &raw_obj->pool); |
7c673cae FG |
138 | } |
139 | ||
140 | rgw_raw_obj rgw_obj_select::get_raw_obj(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params) const | |
141 | { | |
142 | if (!is_raw) { | |
143 | rgw_raw_obj r; | |
144 | rgw_obj_to_raw(zonegroup, zone_params, placement_rule, obj, &r); | |
145 | return r; | |
146 | } | |
147 | return raw_obj; | |
148 | } | |
149 | ||
150 | rgw_raw_obj rgw_obj_select::get_raw_obj(RGWRados *store) const | |
151 | { | |
152 | if (!is_raw) { | |
153 | rgw_raw_obj r; | |
154 | store->obj_to_raw(placement_rule, obj, &r); | |
155 | return r; | |
156 | } | |
157 | return raw_obj; | |
158 | } | |
159 | ||
11fdf7f2 TL |
160 | void RGWObjVersionTracker::prepare_op_for_read(ObjectReadOperation *op) |
161 | { | |
162 | obj_version *check_objv = version_for_check(); | |
7c673cae | 163 | |
11fdf7f2 TL |
164 | if (check_objv) { |
165 | cls_version_check(*op, *check_objv, VER_COND_EQ); | |
7c673cae FG |
166 | } |
167 | ||
11fdf7f2 | 168 | cls_version_read(*op, &read_version); |
7c673cae FG |
169 | } |
170 | ||
11fdf7f2 TL |
171 | void RGWObjVersionTracker::prepare_op_for_write(ObjectWriteOperation *op) |
172 | { | |
173 | obj_version *check_objv = version_for_check(); | |
174 | obj_version *modify_version = version_for_write(); | |
7c673cae | 175 | |
11fdf7f2 TL |
176 | if (check_objv) { |
177 | cls_version_check(*op, *check_objv, VER_COND_EQ); | |
7c673cae | 178 | } |
7c673cae | 179 | |
11fdf7f2 TL |
180 | if (modify_version) { |
181 | cls_version_set(*op, *modify_version); | |
182 | } else { | |
183 | cls_version_inc(*op); | |
7c673cae | 184 | } |
7c673cae FG |
185 | } |
186 | ||
f91f0fd5 TL |
187 | void RGWObjVersionTracker::apply_write() |
188 | { | |
189 | const bool checked = (read_version.ver != 0); | |
190 | const bool incremented = (write_version.ver == 0); | |
191 | ||
192 | if (checked && incremented) { | |
193 | // apply cls_version_inc() so our next operation can recheck it | |
194 | ++read_version.ver; | |
195 | } else { | |
196 | read_version = write_version; | |
197 | } | |
198 | write_version = obj_version(); | |
199 | } | |
200 | ||
9f95a23c | 201 | RGWObjState::RGWObjState() { |
7c673cae FG |
202 | } |
203 | ||
9f95a23c | 204 | RGWObjState::~RGWObjState() { |
7c673cae FG |
205 | } |
206 | ||
9f95a23c TL |
207 | RGWObjState::RGWObjState(const RGWObjState& rhs) : obj (rhs.obj) { |
208 | is_atomic = rhs.is_atomic; | |
209 | has_attrs = rhs.has_attrs; | |
210 | exists = rhs.exists; | |
211 | size = rhs.size; | |
212 | accounted_size = rhs.accounted_size; | |
213 | mtime = rhs.mtime; | |
214 | epoch = rhs.epoch; | |
215 | if (rhs.obj_tag.length()) { | |
216 | obj_tag = rhs.obj_tag; | |
7c673cae | 217 | } |
9f95a23c TL |
218 | if (rhs.tail_tag.length()) { |
219 | tail_tag = rhs.tail_tag; | |
7c673cae | 220 | } |
9f95a23c TL |
221 | write_tag = rhs.write_tag; |
222 | fake_tag = rhs.fake_tag; | |
223 | manifest = rhs.manifest; | |
224 | shadow_obj = rhs.shadow_obj; | |
225 | has_data = rhs.has_data; | |
226 | if (rhs.data.length()) { | |
227 | data = rhs.data; | |
7c673cae | 228 | } |
9f95a23c TL |
229 | prefetch_data = rhs.prefetch_data; |
230 | keep_tail = rhs.keep_tail; | |
231 | is_olh = rhs.is_olh; | |
232 | objv_tracker = rhs.objv_tracker; | |
233 | pg_ver = rhs.pg_ver; | |
7c673cae FG |
234 | } |
235 | ||
9f95a23c TL |
236 | RGWObjState *RGWObjectCtx::get_state(const rgw_obj& obj) { |
237 | RGWObjState *result; | |
238 | typename std::map<rgw_obj, RGWObjState>::iterator iter; | |
239 | lock.lock_shared(); | |
240 | assert (!obj.empty()); | |
241 | iter = objs_state.find(obj); | |
242 | if (iter != objs_state.end()) { | |
243 | result = &iter->second; | |
244 | lock.unlock_shared(); | |
245 | } else { | |
246 | lock.unlock_shared(); | |
247 | lock.lock(); | |
248 | result = &objs_state[obj]; | |
249 | lock.unlock(); | |
224ce89b | 250 | } |
9f95a23c | 251 | return result; |
7c673cae FG |
252 | } |
253 | ||
9f95a23c TL |
254 | void RGWObjectCtx::set_atomic(rgw_obj& obj) { |
255 | std::unique_lock wl{lock}; | |
256 | assert (!obj.empty()); | |
257 | objs_state[obj].is_atomic = true; | |
7c673cae | 258 | } |
9f95a23c TL |
259 | void RGWObjectCtx::set_prefetch_data(const rgw_obj& obj) { |
260 | std::unique_lock wl{lock}; | |
261 | assert (!obj.empty()); | |
262 | objs_state[obj].prefetch_data = true; | |
7c673cae FG |
263 | } |
264 | ||
9f95a23c TL |
265 | void RGWObjectCtx::invalidate(const rgw_obj& obj) { |
266 | std::unique_lock wl{lock}; | |
267 | auto iter = objs_state.find(obj); | |
268 | if (iter == objs_state.end()) { | |
11fdf7f2 | 269 | return; |
7c673cae | 270 | } |
9f95a23c TL |
271 | bool is_atomic = iter->second.is_atomic; |
272 | bool prefetch_data = iter->second.prefetch_data; | |
7c673cae | 273 | |
9f95a23c | 274 | objs_state.erase(iter); |
7c673cae | 275 | |
9f95a23c TL |
276 | if (is_atomic || prefetch_data) { |
277 | auto& state = objs_state[obj]; | |
278 | state.is_atomic = is_atomic; | |
279 | state.prefetch_data = prefetch_data; | |
11fdf7f2 | 280 | } |
7c673cae FG |
281 | } |
282 | ||
11fdf7f2 | 283 | void RGWObjVersionTracker::generate_new_write_ver(CephContext *cct) |
7c673cae | 284 | { |
11fdf7f2 TL |
285 | write_version.ver = 1; |
286 | #define TAG_LEN 24 | |
7c673cae | 287 | |
11fdf7f2 TL |
288 | write_version.tag.clear(); |
289 | append_rand_alpha(cct, write_version.tag, write_version.tag, TAG_LEN); | |
7c673cae FG |
290 | } |
291 | ||
7c673cae FG |
292 | class RGWMetaNotifierManager : public RGWCoroutinesManager { |
293 | RGWRados *store; | |
294 | RGWHTTPManager http_manager; | |
295 | ||
296 | public: | |
297 | RGWMetaNotifierManager(RGWRados *_store) : RGWCoroutinesManager(_store->ctx(), _store->get_cr_registry()), store(_store), | |
298 | http_manager(store->ctx(), completion_mgr) { | |
11fdf7f2 | 299 | http_manager.start(); |
7c673cae FG |
300 | } |
301 | ||
9f95a23c | 302 | int notify_all(map<rgw_zone_id, RGWRESTConn *>& conn_map, set<int>& shards) { |
7c673cae FG |
303 | rgw_http_param_pair pairs[] = { { "type", "metadata" }, |
304 | { "notify", NULL }, | |
305 | { NULL, NULL } }; | |
306 | ||
307 | list<RGWCoroutinesStack *> stacks; | |
9f95a23c | 308 | for (auto iter = conn_map.begin(); iter != conn_map.end(); ++iter) { |
7c673cae FG |
309 | RGWRESTConn *conn = iter->second; |
310 | RGWCoroutinesStack *stack = new RGWCoroutinesStack(store->ctx(), this); | |
311 | stack->call(new RGWPostRESTResourceCR<set<int>, int>(store->ctx(), conn, &http_manager, "/admin/log", pairs, shards, NULL)); | |
312 | ||
313 | stacks.push_back(stack); | |
314 | } | |
315 | return run(stacks); | |
316 | } | |
317 | }; | |
318 | ||
319 | class RGWDataNotifierManager : public RGWCoroutinesManager { | |
320 | RGWRados *store; | |
321 | RGWHTTPManager http_manager; | |
322 | ||
323 | public: | |
324 | RGWDataNotifierManager(RGWRados *_store) : RGWCoroutinesManager(_store->ctx(), _store->get_cr_registry()), store(_store), | |
325 | http_manager(store->ctx(), completion_mgr) { | |
11fdf7f2 | 326 | http_manager.start(); |
7c673cae FG |
327 | } |
328 | ||
9f95a23c | 329 | int notify_all(map<rgw_zone_id, RGWRESTConn *>& conn_map, map<int, set<string> >& shards) { |
7c673cae FG |
330 | rgw_http_param_pair pairs[] = { { "type", "data" }, |
331 | { "notify", NULL }, | |
11fdf7f2 | 332 | { "source-zone", store->svc.zone->get_zone_params().get_id().c_str() }, |
7c673cae FG |
333 | { NULL, NULL } }; |
334 | ||
335 | list<RGWCoroutinesStack *> stacks; | |
9f95a23c | 336 | for (auto iter = conn_map.begin(); iter != conn_map.end(); ++iter) { |
7c673cae FG |
337 | RGWRESTConn *conn = iter->second; |
338 | RGWCoroutinesStack *stack = new RGWCoroutinesStack(store->ctx(), this); | |
339 | stack->call(new RGWPostRESTResourceCR<map<int, set<string> >, int>(store->ctx(), conn, &http_manager, "/admin/log", pairs, shards, NULL)); | |
340 | ||
341 | stacks.push_back(stack); | |
342 | } | |
343 | return run(stacks); | |
344 | } | |
345 | }; | |
346 | ||
11fdf7f2 TL |
347 | /* class RGWRadosThread */ |
348 | ||
7c673cae FG |
349 | void RGWRadosThread::start() |
350 | { | |
351 | worker = new Worker(cct, this); | |
352 | worker->create(thread_name.c_str()); | |
353 | } | |
354 | ||
355 | void RGWRadosThread::stop() | |
356 | { | |
357 | down_flag = true; | |
358 | stop_process(); | |
359 | if (worker) { | |
31f18b77 | 360 | worker->signal(); |
7c673cae FG |
361 | worker->join(); |
362 | } | |
363 | delete worker; | |
364 | worker = NULL; | |
365 | } | |
366 | ||
367 | void *RGWRadosThread::Worker::entry() { | |
368 | uint64_t msec = processor->interval_msec(); | |
9f95a23c | 369 | auto interval = std::chrono::milliseconds(msec); |
7c673cae FG |
370 | |
371 | do { | |
9f95a23c | 372 | auto start = ceph::real_clock::now(); |
7c673cae FG |
373 | int r = processor->process(); |
374 | if (r < 0) { | |
375 | dout(0) << "ERROR: processor->process() returned error r=" << r << dendl; | |
376 | } | |
377 | ||
378 | if (processor->going_down()) | |
379 | break; | |
380 | ||
9f95a23c | 381 | auto end = ceph::real_clock::now() - start; |
7c673cae FG |
382 | |
383 | uint64_t cur_msec = processor->interval_msec(); | |
384 | if (cur_msec != msec) { /* was it reconfigured? */ | |
385 | msec = cur_msec; | |
9f95a23c | 386 | interval = std::chrono::milliseconds(msec); |
7c673cae FG |
387 | } |
388 | ||
389 | if (cur_msec > 0) { | |
390 | if (interval <= end) | |
391 | continue; // next round | |
392 | ||
9f95a23c | 393 | auto wait_time = interval - end; |
31f18b77 | 394 | wait_interval(wait_time); |
7c673cae | 395 | } else { |
31f18b77 | 396 | wait(); |
7c673cae FG |
397 | } |
398 | } while (!processor->going_down()); | |
399 | ||
400 | return NULL; | |
401 | } | |
402 | ||
403 | class RGWMetaNotifier : public RGWRadosThread { | |
404 | RGWMetaNotifierManager notify_mgr; | |
405 | RGWMetadataLog *const log; | |
406 | ||
407 | uint64_t interval_msec() override { | |
408 | return cct->_conf->rgw_md_notify_interval_msec; | |
409 | } | |
1adf2230 AA |
410 | void stop_process() override { |
411 | notify_mgr.stop(); | |
412 | } | |
7c673cae FG |
413 | public: |
414 | RGWMetaNotifier(RGWRados *_store, RGWMetadataLog* log) | |
415 | : RGWRadosThread(_store, "meta-notifier"), notify_mgr(_store), log(log) {} | |
416 | ||
417 | int process() override; | |
418 | }; | |
419 | ||
420 | int RGWMetaNotifier::process() | |
421 | { | |
422 | set<int> shards; | |
423 | ||
424 | log->read_clear_modified(shards); | |
425 | ||
426 | if (shards.empty()) { | |
427 | return 0; | |
428 | } | |
429 | ||
430 | for (set<int>::iterator iter = shards.begin(); iter != shards.end(); ++iter) { | |
431 | ldout(cct, 20) << __func__ << "(): notifying mdlog change, shard_id=" << *iter << dendl; | |
432 | } | |
433 | ||
11fdf7f2 | 434 | notify_mgr.notify_all(store->svc.zone->get_zone_conn_map(), shards); |
7c673cae FG |
435 | |
436 | return 0; | |
437 | } | |
438 | ||
439 | class RGWDataNotifier : public RGWRadosThread { | |
440 | RGWDataNotifierManager notify_mgr; | |
441 | ||
442 | uint64_t interval_msec() override { | |
11fdf7f2 | 443 | return cct->_conf.get_val<int64_t>("rgw_data_notify_interval_msec"); |
7c673cae | 444 | } |
1adf2230 AA |
445 | void stop_process() override { |
446 | notify_mgr.stop(); | |
447 | } | |
7c673cae FG |
448 | public: |
449 | RGWDataNotifier(RGWRados *_store) : RGWRadosThread(_store, "data-notifier"), notify_mgr(_store) {} | |
450 | ||
451 | int process() override; | |
452 | }; | |
453 | ||
454 | int RGWDataNotifier::process() | |
455 | { | |
9f95a23c TL |
456 | auto data_log = store->svc.datalog_rados->get_log(); |
457 | if (!data_log) { | |
7c673cae FG |
458 | return 0; |
459 | } | |
460 | ||
461 | map<int, set<string> > shards; | |
462 | ||
9f95a23c | 463 | data_log->read_clear_modified(shards); |
7c673cae FG |
464 | |
465 | if (shards.empty()) { | |
466 | return 0; | |
467 | } | |
468 | ||
469 | for (map<int, set<string> >::iterator iter = shards.begin(); iter != shards.end(); ++iter) { | |
470 | ldout(cct, 20) << __func__ << "(): notifying datalog change, shard_id=" << iter->first << ": " << iter->second << dendl; | |
471 | } | |
472 | ||
11fdf7f2 | 473 | notify_mgr.notify_all(store->svc.zone->get_zone_data_notify_to_map(), shards); |
7c673cae FG |
474 | |
475 | return 0; | |
476 | } | |
477 | ||
478 | class RGWSyncProcessorThread : public RGWRadosThread { | |
479 | public: | |
480 | RGWSyncProcessorThread(RGWRados *_store, const string& thread_name = "radosgw") : RGWRadosThread(_store, thread_name) {} | |
481 | RGWSyncProcessorThread(RGWRados *_store) : RGWRadosThread(_store) {} | |
482 | ~RGWSyncProcessorThread() override {} | |
483 | int init() override = 0 ; | |
484 | int process() override = 0; | |
485 | }; | |
486 | ||
487 | class RGWMetaSyncProcessorThread : public RGWSyncProcessorThread | |
488 | { | |
489 | RGWMetaSyncStatusManager sync; | |
490 | ||
491 | uint64_t interval_msec() override { | |
492 | return 0; /* no interval associated, it'll run once until stopped */ | |
493 | } | |
494 | void stop_process() override { | |
495 | sync.stop(); | |
496 | } | |
497 | public: | |
9f95a23c TL |
498 | RGWMetaSyncProcessorThread(rgw::sal::RGWRadosStore *_store, RGWAsyncRadosProcessor *async_rados) |
499 | : RGWSyncProcessorThread(_store->getRados(), "meta-sync"), sync(_store, async_rados) {} | |
7c673cae FG |
500 | |
501 | void wakeup_sync_shards(set<int>& shard_ids) { | |
502 | for (set<int>::iterator iter = shard_ids.begin(); iter != shard_ids.end(); ++iter) { | |
503 | sync.wakeup(*iter); | |
504 | } | |
505 | } | |
506 | RGWMetaSyncStatusManager* get_manager() { return &sync; } | |
507 | ||
508 | int init() override { | |
509 | int ret = sync.init(); | |
510 | if (ret < 0) { | |
511 | ldout(store->ctx(), 0) << "ERROR: sync.init() returned " << ret << dendl; | |
512 | return ret; | |
513 | } | |
514 | return 0; | |
515 | } | |
516 | ||
517 | int process() override { | |
518 | sync.run(); | |
519 | return 0; | |
520 | } | |
521 | }; | |
522 | ||
523 | class RGWDataSyncProcessorThread : public RGWSyncProcessorThread | |
524 | { | |
81eedcae | 525 | PerfCountersRef counters; |
7c673cae FG |
526 | RGWDataSyncStatusManager sync; |
527 | bool initialized; | |
528 | ||
529 | uint64_t interval_msec() override { | |
530 | if (initialized) { | |
531 | return 0; /* no interval associated, it'll run once until stopped */ | |
532 | } else { | |
533 | #define DATA_SYNC_INIT_WAIT_SEC 20 | |
534 | return DATA_SYNC_INIT_WAIT_SEC * 1000; | |
535 | } | |
536 | } | |
537 | void stop_process() override { | |
538 | sync.stop(); | |
539 | } | |
540 | public: | |
9f95a23c | 541 | RGWDataSyncProcessorThread(rgw::sal::RGWRadosStore *_store, RGWAsyncRadosProcessor *async_rados, |
81eedcae | 542 | const RGWZone* source_zone) |
9f95a23c | 543 | : RGWSyncProcessorThread(_store->getRados(), "data-sync"), |
81eedcae TL |
544 | counters(sync_counters::build(store->ctx(), std::string("data-sync-from-") + source_zone->name)), |
545 | sync(_store, async_rados, source_zone->id, counters.get()), | |
7c673cae FG |
546 | initialized(false) {} |
547 | ||
548 | void wakeup_sync_shards(map<int, set<string> >& shard_ids) { | |
549 | for (map<int, set<string> >::iterator iter = shard_ids.begin(); iter != shard_ids.end(); ++iter) { | |
550 | sync.wakeup(iter->first, iter->second); | |
551 | } | |
552 | } | |
553 | RGWDataSyncStatusManager* get_manager() { return &sync; } | |
554 | ||
555 | int init() override { | |
556 | return 0; | |
557 | } | |
558 | ||
559 | int process() override { | |
560 | while (!initialized) { | |
561 | if (going_down()) { | |
562 | return 0; | |
563 | } | |
564 | int ret = sync.init(); | |
565 | if (ret >= 0) { | |
566 | initialized = true; | |
567 | break; | |
568 | } | |
569 | /* we'll be back! */ | |
570 | return 0; | |
571 | } | |
572 | sync.run(); | |
573 | return 0; | |
574 | } | |
575 | }; | |
576 | ||
11fdf7f2 | 577 | class RGWSyncLogTrimThread : public RGWSyncProcessorThread, DoutPrefixProvider |
7c673cae FG |
578 | { |
579 | RGWCoroutinesManager crs; | |
9f95a23c | 580 | rgw::sal::RGWRadosStore *store; |
b32b8144 | 581 | rgw::BucketTrimManager *bucket_trim; |
7c673cae FG |
582 | RGWHTTPManager http; |
583 | const utime_t trim_interval; | |
584 | ||
585 | uint64_t interval_msec() override { return 0; } | |
586 | void stop_process() override { crs.stop(); } | |
587 | public: | |
9f95a23c | 588 | RGWSyncLogTrimThread(rgw::sal::RGWRadosStore *store, rgw::BucketTrimManager *bucket_trim, |
b32b8144 | 589 | int interval) |
9f95a23c TL |
590 | : RGWSyncProcessorThread(store->getRados(), "sync-log-trim"), |
591 | crs(store->ctx(), store->getRados()->get_cr_registry()), store(store), | |
b32b8144 | 592 | bucket_trim(bucket_trim), |
7c673cae FG |
593 | http(store->ctx(), crs.get_completion_mgr()), |
594 | trim_interval(interval, 0) | |
595 | {} | |
596 | ||
597 | int init() override { | |
11fdf7f2 | 598 | return http.start(); |
7c673cae FG |
599 | } |
600 | int process() override { | |
601 | list<RGWCoroutinesStack*> stacks; | |
602 | auto meta = new RGWCoroutinesStack(store->ctx(), &crs); | |
11fdf7f2 | 603 | meta->call(create_meta_log_trim_cr(this, store, &http, |
7c673cae FG |
604 | cct->_conf->rgw_md_log_max_shards, |
605 | trim_interval)); | |
606 | stacks.push_back(meta); | |
607 | ||
9f95a23c TL |
608 | if (store->svc()->zone->sync_module_exports_data()) { |
609 | auto data = new RGWCoroutinesStack(store->ctx(), &crs); | |
610 | data->call(create_data_log_trim_cr(store, &http, | |
611 | cct->_conf->rgw_data_log_num_shards, | |
612 | trim_interval)); | |
613 | stacks.push_back(data); | |
7c673cae | 614 | |
9f95a23c TL |
615 | auto bucket = new RGWCoroutinesStack(store->ctx(), &crs); |
616 | bucket->call(bucket_trim->create_bucket_trim_cr(&http)); | |
617 | stacks.push_back(bucket); | |
618 | } | |
b32b8144 | 619 | |
7c673cae FG |
620 | crs.run(stacks); |
621 | return 0; | |
622 | } | |
11fdf7f2 TL |
623 | |
624 | // implements DoutPrefixProvider | |
625 | CephContext *get_cct() const override { return store->ctx(); } | |
9f95a23c | 626 | unsigned get_subsys() const override |
11fdf7f2 TL |
627 | { |
628 | return dout_subsys; | |
629 | } | |
630 | ||
9f95a23c | 631 | std::ostream& gen_prefix(std::ostream& out) const override |
11fdf7f2 TL |
632 | { |
633 | return out << "sync log trim: "; | |
634 | } | |
635 | ||
7c673cae FG |
636 | }; |
637 | ||
638 | void RGWRados::wakeup_meta_sync_shards(set<int>& shard_ids) | |
639 | { | |
9f95a23c | 640 | std::lock_guard l{meta_sync_thread_lock}; |
7c673cae FG |
641 | if (meta_sync_processor_thread) { |
642 | meta_sync_processor_thread->wakeup_sync_shards(shard_ids); | |
643 | } | |
644 | } | |
645 | ||
9f95a23c | 646 | void RGWRados::wakeup_data_sync_shards(const rgw_zone_id& source_zone, map<int, set<string> >& shard_ids) |
7c673cae FG |
647 | { |
648 | ldout(ctx(), 20) << __func__ << ": source_zone=" << source_zone << ", shard_ids=" << shard_ids << dendl; | |
9f95a23c TL |
649 | std::lock_guard l{data_sync_thread_lock}; |
650 | auto iter = data_sync_processor_threads.find(source_zone); | |
7c673cae FG |
651 | if (iter == data_sync_processor_threads.end()) { |
652 | ldout(ctx(), 10) << __func__ << ": couldn't find sync thread for zone " << source_zone << ", skipping async data sync processing" << dendl; | |
653 | return; | |
654 | } | |
655 | ||
656 | RGWDataSyncProcessorThread *thread = iter->second; | |
11fdf7f2 | 657 | ceph_assert(thread); |
7c673cae FG |
658 | thread->wakeup_sync_shards(shard_ids); |
659 | } | |
660 | ||
661 | RGWMetaSyncStatusManager* RGWRados::get_meta_sync_manager() | |
662 | { | |
9f95a23c | 663 | std::lock_guard l{meta_sync_thread_lock}; |
7c673cae FG |
664 | if (meta_sync_processor_thread) { |
665 | return meta_sync_processor_thread->get_manager(); | |
666 | } | |
667 | return nullptr; | |
668 | } | |
669 | ||
9f95a23c | 670 | RGWDataSyncStatusManager* RGWRados::get_data_sync_manager(const rgw_zone_id& source_zone) |
7c673cae | 671 | { |
9f95a23c | 672 | std::lock_guard l{data_sync_thread_lock}; |
7c673cae FG |
673 | auto thread = data_sync_processor_threads.find(source_zone); |
674 | if (thread == data_sync_processor_threads.end()) { | |
675 | return nullptr; | |
676 | } | |
677 | return thread->second->get_manager(); | |
678 | } | |
679 | ||
680 | int RGWRados::get_required_alignment(const rgw_pool& pool, uint64_t *alignment) | |
681 | { | |
682 | IoCtx ioctx; | |
494da23a | 683 | int r = open_pool_ctx(pool, ioctx, false); |
7c673cae FG |
684 | if (r < 0) { |
685 | ldout(cct, 0) << "ERROR: open_pool_ctx() returned " << r << dendl; | |
686 | return r; | |
687 | } | |
688 | ||
689 | bool requires; | |
690 | r = ioctx.pool_requires_alignment2(&requires); | |
691 | if (r < 0) { | |
692 | ldout(cct, 0) << "ERROR: ioctx.pool_requires_alignment2() returned " | |
693 | << r << dendl; | |
694 | return r; | |
695 | } | |
696 | ||
697 | if (!requires) { | |
698 | *alignment = 0; | |
699 | return 0; | |
700 | } | |
701 | ||
702 | uint64_t align; | |
703 | r = ioctx.pool_required_alignment2(&align); | |
704 | if (r < 0) { | |
705 | ldout(cct, 0) << "ERROR: ioctx.pool_required_alignment2() returned " | |
706 | << r << dendl; | |
707 | return r; | |
708 | } | |
709 | if (align != 0) { | |
710 | ldout(cct, 20) << "required alignment=" << align << dendl; | |
711 | } | |
712 | *alignment = align; | |
713 | return 0; | |
714 | } | |
715 | ||
11fdf7f2 TL |
716 | void RGWRados::get_max_aligned_size(uint64_t size, uint64_t alignment, uint64_t *max_size) |
717 | { | |
718 | if (alignment == 0) { | |
719 | *max_size = size; | |
720 | return; | |
721 | } | |
722 | ||
723 | if (size <= alignment) { | |
724 | *max_size = alignment; | |
725 | return; | |
726 | } | |
727 | ||
728 | *max_size = size - (size % alignment); | |
729 | } | |
730 | ||
731 | int RGWRados::get_max_chunk_size(const rgw_pool& pool, uint64_t *max_chunk_size, uint64_t *palignment) | |
7c673cae | 732 | { |
11fdf7f2 | 733 | uint64_t alignment; |
7c673cae FG |
734 | int r = get_required_alignment(pool, &alignment); |
735 | if (r < 0) { | |
736 | return r; | |
737 | } | |
738 | ||
11fdf7f2 TL |
739 | if (palignment) { |
740 | *palignment = alignment; | |
7c673cae FG |
741 | } |
742 | ||
11fdf7f2 | 743 | uint64_t config_chunk_size = cct->_conf->rgw_max_chunk_size; |
7c673cae | 744 | |
11fdf7f2 | 745 | get_max_aligned_size(config_chunk_size, alignment, max_chunk_size); |
7c673cae FG |
746 | |
747 | ldout(cct, 20) << "max_chunk_size=" << *max_chunk_size << dendl; | |
748 | ||
749 | return 0; | |
750 | } | |
751 | ||
11fdf7f2 TL |
752 | int RGWRados::get_max_chunk_size(const rgw_placement_rule& placement_rule, const rgw_obj& obj, |
753 | uint64_t *max_chunk_size, uint64_t *palignment) | |
7c673cae FG |
754 | { |
755 | rgw_pool pool; | |
756 | if (!get_obj_data_pool(placement_rule, obj, &pool)) { | |
757 | ldout(cct, 0) << "ERROR: failed to get data pool for object " << obj << dendl; | |
758 | return -EIO; | |
759 | } | |
11fdf7f2 | 760 | return get_max_chunk_size(pool, max_chunk_size, palignment); |
7c673cae FG |
761 | } |
762 | ||
31f18b77 FG |
763 | class RGWIndexCompletionManager; |
764 | ||
765 | struct complete_op_data { | |
9f95a23c | 766 | ceph::mutex lock = ceph::make_mutex("complete_op_data"); |
31f18b77 FG |
767 | AioCompletion *rados_completion{nullptr}; |
768 | int manager_shard_id{-1}; | |
769 | RGWIndexCompletionManager *manager{nullptr}; | |
770 | rgw_obj obj; | |
771 | RGWModifyOp op; | |
772 | string tag; | |
773 | rgw_bucket_entry_ver ver; | |
774 | cls_rgw_obj_key key; | |
775 | rgw_bucket_dir_entry_meta dir_meta; | |
776 | list<cls_rgw_obj_key> remove_objs; | |
777 | bool log_op; | |
778 | uint16_t bilog_op; | |
779 | rgw_zone_set zones_trace; | |
780 | ||
781 | bool stopped{false}; | |
782 | ||
783 | void stop() { | |
9f95a23c | 784 | std::lock_guard l{lock}; |
31f18b77 FG |
785 | stopped = true; |
786 | } | |
787 | }; | |
788 | ||
789 | class RGWIndexCompletionThread : public RGWRadosThread { | |
790 | RGWRados *store; | |
791 | ||
792 | uint64_t interval_msec() override { | |
793 | return 0; | |
794 | } | |
795 | ||
796 | list<complete_op_data *> completions; | |
797 | ||
9f95a23c TL |
798 | ceph::mutex completions_lock = |
799 | ceph::make_mutex("RGWIndexCompletionThread::completions_lock"); | |
31f18b77 FG |
800 | public: |
801 | RGWIndexCompletionThread(RGWRados *_store) | |
9f95a23c | 802 | : RGWRadosThread(_store, "index-complete"), store(_store) {} |
31f18b77 FG |
803 | |
804 | int process() override; | |
805 | ||
806 | void add_completion(complete_op_data *completion) { | |
807 | { | |
9f95a23c | 808 | std::lock_guard l{completions_lock}; |
31f18b77 FG |
809 | completions.push_back(completion); |
810 | } | |
811 | ||
812 | signal(); | |
813 | } | |
814 | }; | |
815 | ||
816 | int RGWIndexCompletionThread::process() | |
817 | { | |
818 | list<complete_op_data *> comps; | |
819 | ||
820 | { | |
9f95a23c | 821 | std::lock_guard l{completions_lock}; |
31f18b77 FG |
822 | completions.swap(comps); |
823 | } | |
824 | ||
825 | for (auto c : comps) { | |
826 | std::unique_ptr<complete_op_data> up{c}; | |
827 | ||
828 | if (going_down()) { | |
829 | continue; | |
830 | } | |
831 | ldout(store->ctx(), 20) << __func__ << "(): handling completion for key=" << c->key << dendl; | |
832 | ||
833 | RGWRados::BucketShard bs(store); | |
f64942e4 | 834 | RGWBucketInfo bucket_info; |
31f18b77 | 835 | |
f64942e4 | 836 | int r = bs.init(c->obj.bucket, c->obj, &bucket_info); |
31f18b77 FG |
837 | if (r < 0) { |
838 | ldout(cct, 0) << "ERROR: " << __func__ << "(): failed to initialize BucketShard, obj=" << c->obj << " r=" << r << dendl; | |
839 | /* not much to do */ | |
840 | continue; | |
841 | } | |
842 | ||
f64942e4 AA |
843 | r = store->guard_reshard(&bs, c->obj, bucket_info, |
844 | [&](RGWRados::BucketShard *bs) -> int { | |
845 | librados::ObjectWriteOperation o; | |
846 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); | |
847 | cls_rgw_bucket_complete_op(o, c->op, c->tag, c->ver, c->key, c->dir_meta, &c->remove_objs, | |
848 | c->log_op, c->bilog_op, &c->zones_trace); | |
9f95a23c | 849 | return bs->bucket_obj.operate(&o, null_yield); |
31f18b77 FG |
850 | }); |
851 | if (r < 0) { | |
852 | ldout(cct, 0) << "ERROR: " << __func__ << "(): bucket index completion failed, obj=" << c->obj << " r=" << r << dendl; | |
853 | /* ignoring error, can't do anything about it */ | |
854 | continue; | |
855 | } | |
9f95a23c | 856 | r = store->svc.datalog_rados->add_entry(bucket_info, bs.shard_id); |
31f18b77 FG |
857 | if (r < 0) { |
858 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
859 | } | |
860 | } | |
861 | ||
862 | return 0; | |
863 | } | |
864 | ||
865 | class RGWIndexCompletionManager { | |
866 | RGWRados *store{nullptr}; | |
9f95a23c | 867 | ceph::containers::tiny_vector<ceph::mutex> locks; |
31f18b77 FG |
868 | vector<set<complete_op_data *> > completions; |
869 | ||
870 | RGWIndexCompletionThread *completion_thread{nullptr}; | |
871 | ||
872 | int num_shards; | |
873 | ||
874 | std::atomic<int> cur_shard {0}; | |
875 | ||
876 | ||
877 | public: | |
9f95a23c TL |
878 | RGWIndexCompletionManager(RGWRados *_store) : |
879 | store(_store), | |
880 | locks{ceph::make_lock_container<ceph::mutex>( | |
881 | store->ctx()->_conf->rgw_thread_pool_size, | |
882 | [](const size_t i) { | |
883 | return ceph::make_mutex("RGWIndexCompletionManager::lock::" + | |
884 | std::to_string(i)); | |
885 | })} | |
886 | { | |
31f18b77 | 887 | num_shards = store->ctx()->_conf->rgw_thread_pool_size; |
31f18b77 FG |
888 | completions.resize(num_shards); |
889 | } | |
890 | ~RGWIndexCompletionManager() { | |
891 | stop(); | |
31f18b77 FG |
892 | } |
893 | ||
894 | int next_shard() { | |
895 | int result = cur_shard % num_shards; | |
896 | cur_shard++; | |
897 | return result; | |
898 | } | |
899 | ||
900 | void create_completion(const rgw_obj& obj, | |
901 | RGWModifyOp op, string& tag, | |
902 | rgw_bucket_entry_ver& ver, | |
903 | const cls_rgw_obj_key& key, | |
904 | rgw_bucket_dir_entry_meta& dir_meta, | |
905 | list<cls_rgw_obj_key> *remove_objs, bool log_op, | |
906 | uint16_t bilog_op, | |
907 | rgw_zone_set *zones_trace, | |
908 | complete_op_data **result); | |
909 | bool handle_completion(completion_t cb, complete_op_data *arg); | |
910 | ||
911 | int start() { | |
912 | completion_thread = new RGWIndexCompletionThread(store); | |
913 | int ret = completion_thread->init(); | |
914 | if (ret < 0) { | |
915 | return ret; | |
916 | } | |
917 | completion_thread->start(); | |
918 | return 0; | |
919 | } | |
920 | void stop() { | |
921 | if (completion_thread) { | |
922 | completion_thread->stop(); | |
923 | delete completion_thread; | |
924 | } | |
925 | ||
926 | for (int i = 0; i < num_shards; ++i) { | |
9f95a23c | 927 | std::lock_guard l{locks[i]}; |
31f18b77 | 928 | for (auto c : completions[i]) { |
31f18b77 FG |
929 | c->stop(); |
930 | } | |
931 | } | |
932 | completions.clear(); | |
933 | } | |
934 | }; | |
935 | ||
936 | static void obj_complete_cb(completion_t cb, void *arg) | |
937 | { | |
938 | complete_op_data *completion = (complete_op_data *)arg; | |
9f95a23c | 939 | completion->lock.lock(); |
31f18b77 | 940 | if (completion->stopped) { |
9f95a23c | 941 | completion->lock.unlock(); /* can drop lock, no one else is referencing us */ |
31f18b77 FG |
942 | delete completion; |
943 | return; | |
944 | } | |
945 | bool need_delete = completion->manager->handle_completion(cb, completion); | |
9f95a23c | 946 | completion->lock.unlock(); |
31f18b77 FG |
947 | if (need_delete) { |
948 | delete completion; | |
949 | } | |
950 | } | |
951 | ||
952 | ||
953 | void RGWIndexCompletionManager::create_completion(const rgw_obj& obj, | |
954 | RGWModifyOp op, string& tag, | |
955 | rgw_bucket_entry_ver& ver, | |
956 | const cls_rgw_obj_key& key, | |
957 | rgw_bucket_dir_entry_meta& dir_meta, | |
958 | list<cls_rgw_obj_key> *remove_objs, bool log_op, | |
959 | uint16_t bilog_op, | |
960 | rgw_zone_set *zones_trace, | |
961 | complete_op_data **result) | |
962 | { | |
963 | complete_op_data *entry = new complete_op_data; | |
964 | ||
965 | int shard_id = next_shard(); | |
966 | ||
967 | entry->manager_shard_id = shard_id; | |
968 | entry->manager = this; | |
969 | entry->obj = obj; | |
970 | entry->op = op; | |
971 | entry->tag = tag; | |
972 | entry->ver = ver; | |
973 | entry->key = key; | |
974 | entry->dir_meta = dir_meta; | |
975 | entry->log_op = log_op; | |
976 | entry->bilog_op = bilog_op; | |
977 | ||
978 | if (remove_objs) { | |
979 | for (auto iter = remove_objs->begin(); iter != remove_objs->end(); ++iter) { | |
980 | entry->remove_objs.push_back(*iter); | |
981 | } | |
982 | } | |
983 | ||
984 | if (zones_trace) { | |
985 | entry->zones_trace = *zones_trace; | |
986 | } else { | |
9f95a23c | 987 | entry->zones_trace.insert(store->svc.zone->get_zone().id, obj.bucket.get_key()); |
31f18b77 FG |
988 | } |
989 | ||
990 | *result = entry; | |
991 | ||
9f95a23c | 992 | entry->rados_completion = librados::Rados::aio_create_completion(entry, obj_complete_cb); |
31f18b77 | 993 | |
9f95a23c | 994 | std::lock_guard l{locks[shard_id]}; |
31f18b77 FG |
995 | completions[shard_id].insert(entry); |
996 | } | |
997 | ||
998 | bool RGWIndexCompletionManager::handle_completion(completion_t cb, complete_op_data *arg) | |
999 | { | |
1000 | int shard_id = arg->manager_shard_id; | |
1001 | { | |
9f95a23c | 1002 | std::lock_guard l{locks[shard_id]}; |
31f18b77 FG |
1003 | |
1004 | auto& comps = completions[shard_id]; | |
1005 | ||
1006 | auto iter = comps.find(arg); | |
1007 | if (iter == comps.end()) { | |
1008 | return true; | |
1009 | } | |
1010 | ||
1011 | comps.erase(iter); | |
1012 | } | |
1013 | ||
1014 | int r = rados_aio_get_return_value(cb); | |
1015 | if (r != -ERR_BUSY_RESHARDING) { | |
1016 | return true; | |
1017 | } | |
1018 | completion_thread->add_completion(arg); | |
1019 | return false; | |
1020 | } | |
1021 | ||
7c673cae FG |
1022 | void RGWRados::finalize() |
1023 | { | |
1024 | if (run_sync_thread) { | |
9f95a23c | 1025 | std::lock_guard l{meta_sync_thread_lock}; |
7c673cae FG |
1026 | meta_sync_processor_thread->stop(); |
1027 | ||
9f95a23c | 1028 | std::lock_guard dl{data_sync_thread_lock}; |
7c673cae FG |
1029 | for (auto iter : data_sync_processor_threads) { |
1030 | RGWDataSyncProcessorThread *thread = iter.second; | |
1031 | thread->stop(); | |
1032 | } | |
1033 | if (sync_log_trimmer) { | |
1034 | sync_log_trimmer->stop(); | |
1035 | } | |
1036 | } | |
7c673cae FG |
1037 | if (run_sync_thread) { |
1038 | delete meta_sync_processor_thread; | |
1039 | meta_sync_processor_thread = NULL; | |
9f95a23c | 1040 | std::lock_guard dl{data_sync_thread_lock}; |
7c673cae FG |
1041 | for (auto iter : data_sync_processor_threads) { |
1042 | RGWDataSyncProcessorThread *thread = iter.second; | |
1043 | delete thread; | |
1044 | } | |
1045 | data_sync_processor_threads.clear(); | |
1046 | delete sync_log_trimmer; | |
1047 | sync_log_trimmer = nullptr; | |
b32b8144 | 1048 | bucket_trim = boost::none; |
7c673cae | 1049 | } |
7c673cae FG |
1050 | if (meta_notifier) { |
1051 | meta_notifier->stop(); | |
1052 | delete meta_notifier; | |
1053 | } | |
1054 | if (data_notifier) { | |
1055 | data_notifier->stop(); | |
1056 | delete data_notifier; | |
1057 | } | |
11fdf7f2 | 1058 | delete sync_tracer; |
11fdf7f2 TL |
1059 | |
1060 | delete lc; | |
1061 | lc = NULL; | |
7c673cae | 1062 | |
11fdf7f2 TL |
1063 | delete gc; |
1064 | gc = NULL; | |
7c673cae | 1065 | |
11fdf7f2 TL |
1066 | delete obj_expirer; |
1067 | obj_expirer = NULL; | |
7c673cae | 1068 | |
11fdf7f2 TL |
1069 | RGWQuotaHandler::free_handler(quota_handler); |
1070 | if (cr_registry) { | |
1071 | cr_registry->put(); | |
7c673cae FG |
1072 | } |
1073 | ||
11fdf7f2 | 1074 | svc.shutdown(); |
7c673cae | 1075 | |
11fdf7f2 TL |
1076 | delete binfo_cache; |
1077 | delete obj_tombstone_cache; | |
7c673cae | 1078 | |
11fdf7f2 TL |
1079 | if (reshard_wait.get()) { |
1080 | reshard_wait->stop(); | |
1081 | reshard_wait.reset(); | |
7c673cae FG |
1082 | } |
1083 | ||
11fdf7f2 TL |
1084 | if (run_reshard_thread) { |
1085 | reshard->stop_processor(); | |
7c673cae | 1086 | } |
11fdf7f2 TL |
1087 | delete reshard; |
1088 | delete index_completion_manager; | |
1089 | } | |
1090 | ||
1091 | /** | |
1092 | * Initialize the RADOS instance and prepare to do other ops | |
1093 | * Returns 0 on success, -ERR# on failure. | |
1094 | */ | |
1095 | int RGWRados::init_rados() | |
1096 | { | |
1097 | int ret = 0; | |
7c673cae | 1098 | |
494da23a TL |
1099 | ret = rados.init_with_context(cct); |
1100 | if (ret < 0) { | |
1101 | return ret; | |
1102 | } | |
1103 | ret = rados.connect(); | |
1104 | if (ret < 0) { | |
1105 | return ret; | |
7c673cae | 1106 | } |
11fdf7f2 TL |
1107 | |
1108 | auto crs = std::unique_ptr<RGWCoroutinesManagerRegistry>{ | |
1109 | new RGWCoroutinesManagerRegistry(cct)}; | |
1110 | ret = crs->hook_to_admin_command("cr dump"); | |
1111 | if (ret < 0) { | |
1112 | return ret; | |
7c673cae FG |
1113 | } |
1114 | ||
11fdf7f2 | 1115 | cr_registry = crs.release(); |
11fdf7f2 | 1116 | return ret; |
7c673cae FG |
1117 | } |
1118 | ||
11fdf7f2 | 1119 | int RGWRados::register_to_service_map(const string& daemon_type, const map<string, string>& meta) |
7c673cae | 1120 | { |
11fdf7f2 | 1121 | map<string,string> metadata = meta; |
494da23a | 1122 | metadata["num_handles"] = "1"s; |
11fdf7f2 TL |
1123 | metadata["zonegroup_id"] = svc.zone->get_zonegroup().get_id(); |
1124 | metadata["zonegroup_name"] = svc.zone->get_zonegroup().get_name(); | |
1125 | metadata["zone_name"] = svc.zone->zone_name(); | |
9f95a23c | 1126 | metadata["zone_id"] = svc.zone->zone_id().id; |
11fdf7f2 TL |
1127 | string name = cct->_conf->name.get_id(); |
1128 | if (name.compare(0, 4, "rgw.") == 0) { | |
1129 | name = name.substr(4); | |
7c673cae | 1130 | } |
494da23a | 1131 | int ret = rados.service_daemon_register(daemon_type, name, metadata); |
11fdf7f2 TL |
1132 | if (ret < 0) { |
1133 | ldout(cct, 0) << "ERROR: service_daemon_register() returned ret=" << ret << ": " << cpp_strerror(-ret) << dendl; | |
1134 | return ret; | |
7c673cae FG |
1135 | } |
1136 | ||
1137 | return 0; | |
1138 | } | |
1139 | ||
11fdf7f2 | 1140 | int RGWRados::update_service_map(std::map<std::string, std::string>&& status) |
7c673cae | 1141 | { |
494da23a | 1142 | int ret = rados.service_daemon_update_status(move(status)); |
11fdf7f2 TL |
1143 | if (ret < 0) { |
1144 | ldout(cct, 0) << "ERROR: service_daemon_update_status() returned ret=" << ret << ": " << cpp_strerror(-ret) << dendl; | |
1145 | return ret; | |
1146 | } | |
1147 | ||
1148 | return 0; | |
7c673cae FG |
1149 | } |
1150 | ||
1151 | /** | |
1152 | * Initialize the RADOS instance and prepare to do other ops | |
1153 | * Returns 0 on success, -ERR# on failure. | |
1154 | */ | |
1155 | int RGWRados::init_complete() | |
1156 | { | |
11fdf7f2 | 1157 | int ret; |
7c673cae | 1158 | |
11fdf7f2 TL |
1159 | /* |
1160 | * create sync module instance even if we don't run sync thread, might need it for radosgw-admin | |
1161 | */ | |
9f95a23c | 1162 | sync_module = svc.sync_modules->get_sync_module(); |
7c673cae FG |
1163 | |
1164 | ret = open_root_pool_ctx(); | |
1165 | if (ret < 0) | |
1166 | return ret; | |
1167 | ||
1168 | ret = open_gc_pool_ctx(); | |
1169 | if (ret < 0) | |
1170 | return ret; | |
1171 | ||
1172 | ret = open_lc_pool_ctx(); | |
1173 | if (ret < 0) | |
1174 | return ret; | |
1175 | ||
1176 | ret = open_objexp_pool_ctx(); | |
1177 | if (ret < 0) | |
1178 | return ret; | |
1179 | ||
31f18b77 FG |
1180 | ret = open_reshard_pool_ctx(); |
1181 | if (ret < 0) | |
1182 | return ret; | |
1183 | ||
7c673cae FG |
1184 | pools_initialized = true; |
1185 | ||
1186 | gc = new RGWGC(); | |
1187 | gc->initialize(cct, this); | |
1188 | ||
9f95a23c | 1189 | obj_expirer = new RGWObjectExpirer(this->store); |
7c673cae FG |
1190 | |
1191 | if (use_gc_thread) { | |
1192 | gc->start_processor(); | |
1193 | obj_expirer->start_processor(); | |
1194 | } | |
1195 | ||
11fdf7f2 TL |
1196 | auto& current_period = svc.zone->get_current_period(); |
1197 | auto& zonegroup = svc.zone->get_zonegroup(); | |
1198 | auto& zone_params = svc.zone->get_zone_params(); | |
1199 | auto& zone = svc.zone->get_zone(); | |
1200 | ||
7c673cae FG |
1201 | /* no point of running sync thread if we don't have a master zone configured |
1202 | or there is no rest_master_conn */ | |
9f95a23c | 1203 | if (!svc.zone->need_to_sync()) { |
7c673cae FG |
1204 | run_sync_thread = false; |
1205 | } | |
1206 | ||
11fdf7f2 | 1207 | if (svc.zone->is_meta_master()) { |
9f95a23c | 1208 | auto md_log = svc.mdlog->get_log(current_period.get_id()); |
7c673cae FG |
1209 | meta_notifier = new RGWMetaNotifier(this, md_log); |
1210 | meta_notifier->start(); | |
1211 | } | |
1212 | ||
11fdf7f2 TL |
1213 | /* init it anyway, might run sync through radosgw-admin explicitly */ |
1214 | sync_tracer = new RGWSyncTraceManager(cct, cct->_conf->rgw_sync_trace_history_size); | |
1215 | sync_tracer->init(this); | |
1216 | ret = sync_tracer->hook_to_admin_command(); | |
1217 | if (ret < 0) { | |
1218 | return ret; | |
1219 | } | |
1220 | ||
7c673cae | 1221 | if (run_sync_thread) { |
11fdf7f2 TL |
1222 | for (const auto &pt: zonegroup.placement_targets) { |
1223 | if (zone_params.placement_pools.find(pt.second.name) | |
1224 | == zone_params.placement_pools.end()){ | |
1225 | ldout(cct, 0) << "WARNING: This zone does not contain the placement target " | |
1226 | << pt.second.name << " present in zonegroup" << dendl; | |
1227 | } | |
1228 | } | |
9f95a23c TL |
1229 | auto async_processor = svc.rados->get_async_processor(); |
1230 | std::lock_guard l{meta_sync_thread_lock}; | |
1231 | meta_sync_processor_thread = new RGWMetaSyncProcessorThread(this->store, async_processor); | |
7c673cae FG |
1232 | ret = meta_sync_processor_thread->init(); |
1233 | if (ret < 0) { | |
1234 | ldout(cct, 0) << "ERROR: failed to initialize meta sync thread" << dendl; | |
1235 | return ret; | |
1236 | } | |
1237 | meta_sync_processor_thread->start(); | |
1238 | ||
b32b8144 FG |
1239 | // configure the bucket trim manager |
1240 | rgw::BucketTrimConfig config; | |
1241 | rgw::configure_bucket_trim(cct, config); | |
1242 | ||
9f95a23c | 1243 | bucket_trim.emplace(this->store, config); |
b32b8144 FG |
1244 | ret = bucket_trim->init(); |
1245 | if (ret < 0) { | |
1246 | ldout(cct, 0) << "ERROR: failed to start bucket trim manager" << dendl; | |
1247 | return ret; | |
1248 | } | |
9f95a23c | 1249 | svc.datalog_rados->set_observer(&*bucket_trim); |
b32b8144 | 1250 | |
9f95a23c | 1251 | std::lock_guard dl{data_sync_thread_lock}; |
81eedcae TL |
1252 | for (auto source_zone : svc.zone->get_data_sync_source_zones()) { |
1253 | ldout(cct, 5) << "starting data sync thread for zone " << source_zone->name << dendl; | |
9f95a23c | 1254 | auto *thread = new RGWDataSyncProcessorThread(this->store, svc.rados->get_async_processor(), source_zone); |
7c673cae FG |
1255 | ret = thread->init(); |
1256 | if (ret < 0) { | |
1257 | ldout(cct, 0) << "ERROR: failed to initialize data sync thread" << dendl; | |
1258 | return ret; | |
1259 | } | |
1260 | thread->start(); | |
9f95a23c | 1261 | data_sync_processor_threads[rgw_zone_id(source_zone->id)] = thread; |
7c673cae FG |
1262 | } |
1263 | auto interval = cct->_conf->rgw_sync_log_trim_interval; | |
1264 | if (interval > 0) { | |
9f95a23c | 1265 | sync_log_trimmer = new RGWSyncLogTrimThread(this->store, &*bucket_trim, interval); |
7c673cae FG |
1266 | ret = sync_log_trimmer->init(); |
1267 | if (ret < 0) { | |
1268 | ldout(cct, 0) << "ERROR: failed to initialize sync log trim thread" << dendl; | |
1269 | return ret; | |
1270 | } | |
1271 | sync_log_trimmer->start(); | |
1272 | } | |
1273 | } | |
1274 | data_notifier = new RGWDataNotifier(this); | |
1275 | data_notifier->start(); | |
1276 | ||
92f5a8d4 TL |
1277 | binfo_cache = new RGWChainedCacheImpl<bucket_info_entry>; |
1278 | binfo_cache->init(svc.cache); | |
1279 | ||
7c673cae | 1280 | lc = new RGWLC(); |
9f95a23c | 1281 | lc->initialize(cct, this->store); |
31f18b77 | 1282 | |
7c673cae FG |
1283 | if (use_lc_thread) |
1284 | lc->start_processor(); | |
31f18b77 | 1285 | |
9f95a23c | 1286 | quota_handler = RGWQuotaHandler::generate_handler(this->store, quota_threads); |
7c673cae FG |
1287 | |
1288 | bucket_index_max_shards = (cct->_conf->rgw_override_bucket_index_max_shards ? cct->_conf->rgw_override_bucket_index_max_shards : | |
11fdf7f2 | 1289 | zone.bucket_index_max_shards); |
31f18b77 FG |
1290 | if (bucket_index_max_shards > get_max_bucket_shards()) { |
1291 | bucket_index_max_shards = get_max_bucket_shards(); | |
7c673cae | 1292 | ldout(cct, 1) << __func__ << " bucket index max shards is too large, reset to value: " |
31f18b77 | 1293 | << get_max_bucket_shards() << dendl; |
7c673cae FG |
1294 | } |
1295 | ldout(cct, 20) << __func__ << " bucket index max shards: " << bucket_index_max_shards << dendl; | |
1296 | ||
11fdf7f2 | 1297 | bool need_tombstone_cache = !svc.zone->get_zone_data_notify_to_map().empty(); /* have zones syncing from us */ |
7c673cae FG |
1298 | |
1299 | if (need_tombstone_cache) { | |
1300 | obj_tombstone_cache = new tombstone_cache_t(cct->_conf->rgw_obj_tombstone_cache_size); | |
1301 | } | |
1302 | ||
11fdf7f2 | 1303 | reshard_wait = std::make_shared<RGWReshardWait>(); |
31f18b77 | 1304 | |
9f95a23c | 1305 | reshard = new RGWReshard(this->store); |
31f18b77 FG |
1306 | |
1307 | /* only the master zone in the zonegroup reshards buckets */ | |
11fdf7f2 | 1308 | run_reshard_thread = run_reshard_thread && (zonegroup.master_zone == zone.id); |
31f18b77 FG |
1309 | if (run_reshard_thread) { |
1310 | reshard->start_processor(); | |
1311 | } | |
1312 | ||
1313 | index_completion_manager = new RGWIndexCompletionManager(this); | |
1314 | ret = index_completion_manager->start(); | |
1315 | ||
7c673cae FG |
1316 | return ret; |
1317 | } | |
1318 | ||
11fdf7f2 TL |
1319 | int RGWRados::init_svc(bool raw) |
1320 | { | |
1321 | if (raw) { | |
1322 | return svc.init_raw(cct, use_cache); | |
1323 | } | |
1324 | ||
9f95a23c TL |
1325 | return svc.init(cct, use_cache, run_sync_thread); |
1326 | } | |
1327 | ||
1328 | int RGWRados::init_ctl() | |
1329 | { | |
1330 | return ctl.init(&svc); | |
11fdf7f2 TL |
1331 | } |
1332 | ||
7c673cae FG |
1333 | /** |
1334 | * Initialize the RADOS instance and prepare to do other ops | |
1335 | * Returns 0 on success, -ERR# on failure. | |
1336 | */ | |
1337 | int RGWRados::initialize() | |
1338 | { | |
1339 | int ret; | |
1340 | ||
11fdf7f2 TL |
1341 | inject_notify_timeout_probability = |
1342 | cct->_conf.get_val<double>("rgw_inject_notify_timeout_probability"); | |
1343 | max_notify_retries = cct->_conf.get_val<uint64_t>("rgw_max_notify_retries"); | |
7c673cae | 1344 | |
11fdf7f2 | 1345 | ret = init_svc(false); |
7c673cae | 1346 | if (ret < 0) { |
11fdf7f2 | 1347 | ldout(cct, 0) << "ERROR: failed to init services (ret=" << cpp_strerror(-ret) << ")" << dendl; |
7c673cae FG |
1348 | return ret; |
1349 | } | |
7c673cae | 1350 | |
9f95a23c TL |
1351 | ret = init_ctl(); |
1352 | if (ret < 0) { | |
1353 | ldout(cct, 0) << "ERROR: failed to init ctls (ret=" << cpp_strerror(-ret) << ")" << dendl; | |
1354 | return ret; | |
1355 | } | |
1356 | ||
11fdf7f2 | 1357 | host_id = svc.zone_utils->gen_host_id(); |
7c673cae | 1358 | |
11fdf7f2 TL |
1359 | ret = init_rados(); |
1360 | if (ret < 0) | |
1361 | return ret; | |
1362 | ||
1363 | return init_complete(); | |
7c673cae FG |
1364 | } |
1365 | ||
1366 | /** | |
1367 | * Open the pool used as root for this gateway | |
1368 | * Returns: 0 on success, -ERR# otherwise. | |
1369 | */ | |
1370 | int RGWRados::open_root_pool_ctx() | |
1371 | { | |
494da23a | 1372 | return rgw_init_ioctx(get_rados_handle(), svc.zone->get_zone_params().domain_root, root_pool_ctx, true, true); |
7c673cae FG |
1373 | } |
1374 | ||
1375 | int RGWRados::open_gc_pool_ctx() | |
1376 | { | |
494da23a | 1377 | return rgw_init_ioctx(get_rados_handle(), svc.zone->get_zone_params().gc_pool, gc_pool_ctx, true, true); |
7c673cae FG |
1378 | } |
1379 | ||
1380 | int RGWRados::open_lc_pool_ctx() | |
1381 | { | |
494da23a | 1382 | return rgw_init_ioctx(get_rados_handle(), svc.zone->get_zone_params().lc_pool, lc_pool_ctx, true, true); |
7c673cae FG |
1383 | } |
1384 | ||
1385 | int RGWRados::open_objexp_pool_ctx() | |
1386 | { | |
494da23a | 1387 | return rgw_init_ioctx(get_rados_handle(), svc.zone->get_zone_params().log_pool, objexp_pool_ctx, true, true); |
7c673cae FG |
1388 | } |
1389 | ||
31f18b77 FG |
1390 | int RGWRados::open_reshard_pool_ctx() |
1391 | { | |
494da23a | 1392 | return rgw_init_ioctx(get_rados_handle(), svc.zone->get_zone_params().reshard_pool, reshard_pool_ctx, true, true); |
7c673cae FG |
1393 | } |
1394 | ||
494da23a TL |
1395 | int RGWRados::open_pool_ctx(const rgw_pool& pool, librados::IoCtx& io_ctx, |
1396 | bool mostly_omap) | |
7c673cae | 1397 | { |
28e407b8 | 1398 | constexpr bool create = true; // create the pool if it doesn't exist |
494da23a | 1399 | return rgw_init_ioctx(get_rados_handle(), pool, io_ctx, create, mostly_omap); |
7c673cae FG |
1400 | } |
1401 | ||
7c673cae FG |
1402 | /**** logs ****/ |
1403 | ||
1404 | struct log_list_state { | |
1405 | string prefix; | |
1406 | librados::IoCtx io_ctx; | |
1407 | librados::NObjectIterator obit; | |
1408 | }; | |
1409 | ||
1410 | int RGWRados::log_list_init(const string& prefix, RGWAccessHandle *handle) | |
1411 | { | |
1412 | log_list_state *state = new log_list_state; | |
11fdf7f2 | 1413 | int r = rgw_init_ioctx(get_rados_handle(), svc.zone->get_zone_params().log_pool, state->io_ctx); |
7c673cae FG |
1414 | if (r < 0) { |
1415 | delete state; | |
1416 | return r; | |
1417 | } | |
1418 | state->prefix = prefix; | |
1419 | state->obit = state->io_ctx.nobjects_begin(); | |
1420 | *handle = (RGWAccessHandle)state; | |
1421 | return 0; | |
1422 | } | |
1423 | ||
1424 | int RGWRados::log_list_next(RGWAccessHandle handle, string *name) | |
1425 | { | |
1426 | log_list_state *state = static_cast<log_list_state *>(handle); | |
1427 | while (true) { | |
1428 | if (state->obit == state->io_ctx.nobjects_end()) { | |
1429 | delete state; | |
1430 | return -ENOENT; | |
1431 | } | |
1432 | if (state->prefix.length() && | |
1433 | state->obit->get_oid().find(state->prefix) != 0) { | |
1434 | state->obit++; | |
1435 | continue; | |
1436 | } | |
1437 | *name = state->obit->get_oid(); | |
1438 | state->obit++; | |
1439 | break; | |
1440 | } | |
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | int RGWRados::log_remove(const string& name) | |
1445 | { | |
1446 | librados::IoCtx io_ctx; | |
11fdf7f2 | 1447 | int r = rgw_init_ioctx(get_rados_handle(), svc.zone->get_zone_params().log_pool, io_ctx); |
7c673cae FG |
1448 | if (r < 0) |
1449 | return r; | |
1450 | return io_ctx.remove(name); | |
1451 | } | |
1452 | ||
1453 | struct log_show_state { | |
1454 | librados::IoCtx io_ctx; | |
1455 | bufferlist bl; | |
11fdf7f2 | 1456 | bufferlist::const_iterator p; |
7c673cae FG |
1457 | string name; |
1458 | uint64_t pos; | |
1459 | bool eof; | |
1460 | log_show_state() : pos(0), eof(false) {} | |
1461 | }; | |
1462 | ||
1463 | int RGWRados::log_show_init(const string& name, RGWAccessHandle *handle) | |
1464 | { | |
1465 | log_show_state *state = new log_show_state; | |
11fdf7f2 | 1466 | int r = rgw_init_ioctx(get_rados_handle(), svc.zone->get_zone_params().log_pool, state->io_ctx); |
7c673cae FG |
1467 | if (r < 0) { |
1468 | delete state; | |
1469 | return r; | |
1470 | } | |
1471 | state->name = name; | |
1472 | *handle = (RGWAccessHandle)state; | |
1473 | return 0; | |
1474 | } | |
1475 | ||
1476 | int RGWRados::log_show_next(RGWAccessHandle handle, rgw_log_entry *entry) | |
1477 | { | |
1478 | log_show_state *state = static_cast<log_show_state *>(handle); | |
1479 | off_t off = state->p.get_off(); | |
1480 | ||
1481 | ldout(cct, 10) << "log_show_next pos " << state->pos << " bl " << state->bl.length() | |
1482 | << " off " << off | |
1483 | << " eof " << (int)state->eof | |
1484 | << dendl; | |
1485 | // read some? | |
1486 | unsigned chunk = 1024*1024; | |
1487 | if ((state->bl.length() - off) < chunk/2 && !state->eof) { | |
1488 | bufferlist more; | |
1489 | int r = state->io_ctx.read(state->name, more, chunk, state->pos); | |
1490 | if (r < 0) | |
1491 | return r; | |
1492 | state->pos += r; | |
1493 | bufferlist old; | |
1494 | try { | |
1495 | old.substr_of(state->bl, off, state->bl.length() - off); | |
1496 | } catch (buffer::error& err) { | |
1497 | return -EINVAL; | |
1498 | } | |
1499 | state->bl.clear(); | |
1500 | state->bl.claim(old); | |
1501 | state->bl.claim_append(more); | |
11fdf7f2 | 1502 | state->p = state->bl.cbegin(); |
7c673cae FG |
1503 | if ((unsigned)r < chunk) |
1504 | state->eof = true; | |
1505 | ldout(cct, 10) << " read " << r << dendl; | |
1506 | } | |
1507 | ||
1508 | if (state->p.end()) | |
1509 | return 0; // end of file | |
1510 | try { | |
11fdf7f2 | 1511 | decode(*entry, state->p); |
7c673cae FG |
1512 | } |
1513 | catch (const buffer::error &e) { | |
1514 | return -EINVAL; | |
1515 | } | |
1516 | return 1; | |
1517 | } | |
1518 | ||
1519 | /** | |
1520 | * usage_log_hash: get usage log key hash, based on name and index | |
1521 | * | |
1522 | * Get the usage object name. Since a user may have more than 1 | |
1523 | * object holding that info (multiple shards), we use index to | |
1524 | * specify that shard number. Once index exceeds max shards it | |
1525 | * wraps. | |
1526 | * If name is not being set, results for all users will be returned | |
1527 | * and index will wrap only after total shards number. | |
1528 | * | |
1529 | * @param cct [in] ceph context | |
1530 | * @param name [in] user name | |
1531 | * @param hash [out] hash value | |
1532 | * @param index [in] shard index number | |
1533 | */ | |
1534 | static void usage_log_hash(CephContext *cct, const string& name, string& hash, uint32_t index) | |
1535 | { | |
1536 | uint32_t val = index; | |
1537 | ||
1538 | if (!name.empty()) { | |
c07f9fc5 | 1539 | int max_user_shards = cct->_conf->rgw_usage_max_user_shards; |
7c673cae FG |
1540 | val %= max_user_shards; |
1541 | val += ceph_str_hash_linux(name.c_str(), name.size()); | |
1542 | } | |
1543 | char buf[17]; | |
c07f9fc5 | 1544 | int max_shards = cct->_conf->rgw_usage_max_shards; |
7c673cae FG |
1545 | snprintf(buf, sizeof(buf), RGW_USAGE_OBJ_PREFIX "%u", (unsigned)(val % max_shards)); |
1546 | hash = buf; | |
1547 | } | |
1548 | ||
1549 | int RGWRados::log_usage(map<rgw_user_bucket, RGWUsageBatch>& usage_info) | |
1550 | { | |
1551 | uint32_t index = 0; | |
1552 | ||
1553 | map<string, rgw_usage_log_info> log_objs; | |
1554 | ||
1555 | string hash; | |
1556 | string last_user; | |
1557 | ||
1558 | /* restructure usage map, zone by object hash */ | |
1559 | map<rgw_user_bucket, RGWUsageBatch>::iterator iter; | |
1560 | for (iter = usage_info.begin(); iter != usage_info.end(); ++iter) { | |
1561 | const rgw_user_bucket& ub = iter->first; | |
1562 | RGWUsageBatch& info = iter->second; | |
1563 | ||
1564 | if (ub.user.empty()) { | |
1565 | ldout(cct, 0) << "WARNING: RGWRados::log_usage(): user name empty (bucket=" << ub.bucket << "), skipping" << dendl; | |
1566 | continue; | |
1567 | } | |
1568 | ||
1569 | if (ub.user != last_user) { | |
1570 | /* index *should* be random, but why waste extra cycles | |
1571 | in most cases max user shards is not going to exceed 1, | |
1572 | so just incrementing it */ | |
1573 | usage_log_hash(cct, ub.user, hash, index++); | |
1574 | } | |
1575 | last_user = ub.user; | |
1576 | vector<rgw_usage_log_entry>& v = log_objs[hash].entries; | |
1577 | ||
1578 | for (auto miter = info.m.begin(); miter != info.m.end(); ++miter) { | |
1579 | v.push_back(miter->second); | |
1580 | } | |
1581 | } | |
1582 | ||
1583 | map<string, rgw_usage_log_info>::iterator liter; | |
1584 | ||
1585 | for (liter = log_objs.begin(); liter != log_objs.end(); ++liter) { | |
1586 | int r = cls_obj_usage_log_add(liter->first, liter->second); | |
1587 | if (r < 0) | |
1588 | return r; | |
1589 | } | |
1590 | return 0; | |
1591 | } | |
1592 | ||
11fdf7f2 TL |
1593 | int RGWRados::read_usage(const rgw_user& user, const string& bucket_name, uint64_t start_epoch, uint64_t end_epoch, |
1594 | uint32_t max_entries, bool *is_truncated, RGWUsageIter& usage_iter, map<rgw_user_bucket, | |
1595 | rgw_usage_log_entry>& usage) | |
7c673cae FG |
1596 | { |
1597 | uint32_t num = max_entries; | |
1598 | string hash, first_hash; | |
1599 | string user_str = user.to_str(); | |
1600 | usage_log_hash(cct, user_str, first_hash, 0); | |
1601 | ||
1602 | if (usage_iter.index) { | |
1603 | usage_log_hash(cct, user_str, hash, usage_iter.index); | |
1604 | } else { | |
1605 | hash = first_hash; | |
1606 | } | |
1607 | ||
1608 | usage.clear(); | |
1609 | ||
1610 | do { | |
1611 | map<rgw_user_bucket, rgw_usage_log_entry> ret_usage; | |
1612 | map<rgw_user_bucket, rgw_usage_log_entry>::iterator iter; | |
1613 | ||
11fdf7f2 | 1614 | int ret = cls_obj_usage_log_read(hash, user_str, bucket_name, start_epoch, end_epoch, num, |
7c673cae FG |
1615 | usage_iter.read_iter, ret_usage, is_truncated); |
1616 | if (ret == -ENOENT) | |
1617 | goto next; | |
1618 | ||
1619 | if (ret < 0) | |
1620 | return ret; | |
1621 | ||
1622 | num -= ret_usage.size(); | |
1623 | ||
1624 | for (iter = ret_usage.begin(); iter != ret_usage.end(); ++iter) { | |
1625 | usage[iter->first].aggregate(iter->second); | |
1626 | } | |
1627 | ||
1628 | next: | |
1629 | if (!*is_truncated) { | |
1630 | usage_iter.read_iter.clear(); | |
1631 | usage_log_hash(cct, user_str, hash, ++usage_iter.index); | |
1632 | } | |
1633 | } while (num && !*is_truncated && hash != first_hash); | |
1634 | return 0; | |
1635 | } | |
1636 | ||
11fdf7f2 | 1637 | int RGWRados::trim_usage(const rgw_user& user, const string& bucket_name, uint64_t start_epoch, uint64_t end_epoch) |
7c673cae FG |
1638 | { |
1639 | uint32_t index = 0; | |
1640 | string hash, first_hash; | |
1641 | string user_str = user.to_str(); | |
1642 | usage_log_hash(cct, user_str, first_hash, index); | |
1643 | ||
1644 | hash = first_hash; | |
7c673cae | 1645 | do { |
11fdf7f2 | 1646 | int ret = cls_obj_usage_log_trim(hash, user_str, bucket_name, start_epoch, end_epoch); |
7c673cae | 1647 | |
b32b8144 | 1648 | if (ret < 0 && ret != -ENOENT) |
7c673cae FG |
1649 | return ret; |
1650 | ||
7c673cae FG |
1651 | usage_log_hash(cct, user_str, hash, ++index); |
1652 | } while (hash != first_hash); | |
1653 | ||
1654 | return 0; | |
1655 | } | |
1656 | ||
11fdf7f2 TL |
1657 | |
1658 | int RGWRados::clear_usage() | |
1659 | { | |
1660 | auto max_shards = cct->_conf->rgw_usage_max_shards; | |
1661 | int ret=0; | |
1662 | for (unsigned i=0; i < max_shards; i++){ | |
1663 | string oid = RGW_USAGE_OBJ_PREFIX + to_string(i); | |
1664 | ret = cls_obj_usage_log_clear(oid); | |
1665 | if (ret < 0){ | |
1666 | ldout(cct,0) << "usage clear on oid="<< oid << "failed with ret=" << ret << dendl; | |
1667 | return ret; | |
1668 | } | |
1669 | } | |
1670 | return ret; | |
1671 | } | |
1672 | ||
9f95a23c | 1673 | int RGWRados::decode_policy(bufferlist& bl, ACLOwner *owner) |
7c673cae | 1674 | { |
9f95a23c TL |
1675 | auto i = bl.cbegin(); |
1676 | RGWAccessControlPolicy policy(cct); | |
1677 | try { | |
1678 | policy.decode_owner(i); | |
1679 | } catch (buffer::error& err) { | |
1680 | ldout(cct, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl; | |
1681 | return -EIO; | |
7c673cae | 1682 | } |
9f95a23c TL |
1683 | *owner = policy.get_owner(); |
1684 | return 0; | |
7c673cae FG |
1685 | } |
1686 | ||
9f95a23c | 1687 | int rgw_policy_from_attrset(CephContext *cct, map<string, bufferlist>& attrset, RGWAccessControlPolicy *policy) |
7c673cae | 1688 | { |
9f95a23c TL |
1689 | map<string, bufferlist>::iterator aiter = attrset.find(RGW_ATTR_ACL); |
1690 | if (aiter == attrset.end()) | |
1691 | return -EIO; | |
7c673cae | 1692 | |
9f95a23c TL |
1693 | bufferlist& bl = aiter->second; |
1694 | auto iter = bl.cbegin(); | |
1695 | try { | |
1696 | policy->decode(iter); | |
1697 | } catch (buffer::error& err) { | |
1698 | ldout(cct, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl; | |
1699 | return -EIO; | |
1700 | } | |
1701 | if (cct->_conf->subsys.should_gather<ceph_subsys_rgw, 15>()) { | |
1702 | RGWAccessControlPolicy_S3 *s3policy = static_cast<RGWAccessControlPolicy_S3 *>(policy); | |
1703 | ldout(cct, 15) << __func__ << " Read AccessControlPolicy"; | |
1704 | s3policy->to_xml(*_dout); | |
1705 | *_dout << dendl; | |
1706 | } | |
1707 | return 0; | |
7c673cae FG |
1708 | } |
1709 | ||
7c673cae | 1710 | |
9f95a23c | 1711 | int RGWRados::Bucket::update_bucket_id(const string& new_bucket_id) |
7c673cae | 1712 | { |
9f95a23c TL |
1713 | rgw_bucket bucket = bucket_info.bucket; |
1714 | bucket.update_bucket_id(new_bucket_id); | |
7c673cae | 1715 | |
9f95a23c | 1716 | auto obj_ctx = store->svc.sysobj->init_obj_ctx(); |
7c673cae | 1717 | |
9f95a23c TL |
1718 | bucket_info.objv_tracker.clear(); |
1719 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, nullptr, nullptr, null_yield); | |
1720 | if (ret < 0) { | |
1721 | return ret; | |
7c673cae FG |
1722 | } |
1723 | ||
9f95a23c | 1724 | return 0; |
eafe8130 TL |
1725 | } |
1726 | ||
1727 | ||
1adf2230 AA |
1728 | /** |
1729 | * Get ordered listing of the objects in a bucket. | |
7c673cae | 1730 | * |
9f95a23c | 1731 | * max_p: maximum number of results to return |
7c673cae FG |
1732 | * bucket: bucket to list contents of |
1733 | * prefix: only return results that match this prefix | |
1734 | * delim: do not include results that match this string. | |
1735 | * Any skipped results will have the matching portion of their name | |
1736 | * inserted in common_prefixes with a "true" mark. | |
1737 | * marker: if filled in, begin the listing with this object. | |
1738 | * end_marker: if filled in, end the listing with this object. | |
1739 | * result: the objects are put in here. | |
11fdf7f2 TL |
1740 | * common_prefixes: if delim is filled in, any matching prefixes are |
1741 | * placed here. | |
1742 | * is_truncated: if number of objects in the bucket is bigger than | |
1743 | * max, then truncated. | |
7c673cae | 1744 | */ |
11fdf7f2 | 1745 | int RGWRados::Bucket::List::list_objects_ordered( |
eafe8130 | 1746 | int64_t max_p, |
11fdf7f2 TL |
1747 | vector<rgw_bucket_dir_entry> *result, |
1748 | map<string, bool> *common_prefixes, | |
9f95a23c TL |
1749 | bool *is_truncated, |
1750 | optional_yield y) | |
7c673cae FG |
1751 | { |
1752 | RGWRados *store = target->get_store(); | |
1753 | CephContext *cct = store->ctx(); | |
1754 | int shard_id = target->get_shard_id(); | |
1755 | ||
1756 | int count = 0; | |
1757 | bool truncated = true; | |
9f95a23c | 1758 | bool cls_filtered = false; |
eafe8130 TL |
1759 | const int64_t max = // protect against memory issues and negative vals |
1760 | std::min(bucket_list_objects_absolute_max, std::max(int64_t(0), max_p)); | |
1761 | int read_ahead = std::max(cct->_conf->rgw_list_bucket_min_readahead, max); | |
7c673cae FG |
1762 | |
1763 | result->clear(); | |
1764 | ||
9f95a23c TL |
1765 | // use a local marker; either the marker will have a previous entry |
1766 | // or it will be empty; either way it's OK to copy | |
1767 | rgw_obj_key marker_obj(params.marker.name, | |
1768 | params.marker.instance, | |
f91f0fd5 | 1769 | params.ns.empty() ? params.marker.ns : params.ns); |
7c673cae FG |
1770 | rgw_obj_index_key cur_marker; |
1771 | marker_obj.get_index_key(&cur_marker); | |
1772 | ||
9f95a23c TL |
1773 | rgw_obj_key end_marker_obj(params.end_marker.name, |
1774 | params.end_marker.instance, | |
f91f0fd5 | 1775 | params.ns.empty() ? params.end_marker.ns : params.ns); |
3efd9988 FG |
1776 | rgw_obj_index_key cur_end_marker; |
1777 | end_marker_obj.get_index_key(&cur_end_marker); | |
7c673cae FG |
1778 | const bool cur_end_marker_valid = !params.end_marker.empty(); |
1779 | ||
1780 | rgw_obj_key prefix_obj(params.prefix); | |
9f95a23c | 1781 | prefix_obj.set_ns(params.ns); |
7c673cae | 1782 | string cur_prefix = prefix_obj.get_index_key_name(); |
11fdf7f2 | 1783 | string after_delim_s; /* needed in !params.delim.empty() AND later */ |
7c673cae FG |
1784 | |
1785 | if (!params.delim.empty()) { | |
9f95a23c | 1786 | after_delim_s = cls_rgw_after_delim(params.delim); |
11fdf7f2 TL |
1787 | /* if marker points at a common prefix, fast forward it into its |
1788 | * upper bound string */ | |
224ce89b | 1789 | int delim_pos = cur_marker.name.find(params.delim, cur_prefix.size()); |
7c673cae FG |
1790 | if (delim_pos >= 0) { |
1791 | string s = cur_marker.name.substr(0, delim_pos); | |
11fdf7f2 | 1792 | s.append(after_delim_s); |
7c673cae FG |
1793 | cur_marker = s; |
1794 | } | |
1795 | } | |
1adf2230 | 1796 | |
9f95a23c | 1797 | rgw_obj_index_key prev_marker; |
f6b5b4d7 | 1798 | for (uint16_t attempt = 1; /* empty */; ++attempt) { |
9f95a23c | 1799 | ldout(cct, 20) << "RGWRados::Bucket::List::" << __func__ << |
f6b5b4d7 | 1800 | " starting attempt " << attempt << dendl; |
9f95a23c TL |
1801 | |
1802 | if (attempt > 1 && !(prev_marker < cur_marker)) { | |
1803 | // we've failed to make forward progress | |
1804 | ldout(cct, 0) << "RGWRados::Bucket::List::" << __func__ << | |
1805 | ": ERROR marker failed to make forward progress; attempt=" << attempt << | |
1806 | ", prev_marker=" << prev_marker << | |
1807 | ", cur_marker=" << cur_marker << dendl; | |
1808 | break; | |
1809 | } | |
1810 | prev_marker = cur_marker; | |
1811 | ||
1812 | ent_map_t ent_map; | |
1813 | ent_map.reserve(read_ahead); | |
1adf2230 AA |
1814 | int r = store->cls_bucket_list_ordered(target->get_bucket_info(), |
1815 | shard_id, | |
1816 | cur_marker, | |
1817 | cur_prefix, | |
9f95a23c | 1818 | params.delim, |
1adf2230 AA |
1819 | read_ahead + 1 - count, |
1820 | params.list_versions, | |
9f95a23c | 1821 | attempt, |
1adf2230 AA |
1822 | ent_map, |
1823 | &truncated, | |
9f95a23c TL |
1824 | &cls_filtered, |
1825 | &cur_marker, | |
1826 | y); | |
1827 | if (r < 0) { | |
7c673cae | 1828 | return r; |
9f95a23c | 1829 | } |
7c673cae | 1830 | |
1adf2230 | 1831 | for (auto eiter = ent_map.begin(); eiter != ent_map.end(); ++eiter) { |
7c673cae FG |
1832 | rgw_bucket_dir_entry& entry = eiter->second; |
1833 | rgw_obj_index_key index_key = entry.key; | |
7c673cae FG |
1834 | rgw_obj_key obj(index_key); |
1835 | ||
9f95a23c TL |
1836 | ldout(cct, 20) << "RGWRados::Bucket::List::" << __func__ << |
1837 | " considering entry " << entry.key << dendl; | |
1838 | ||
1adf2230 AA |
1839 | /* note that parse_raw_oid() here will not set the correct |
1840 | * object's instance, as rgw_obj_index_key encodes that | |
1841 | * separately. We don't need to set the instance because it's | |
1842 | * not needed for the checks here and we end up using the raw | |
1843 | * entry for the return vector | |
7c673cae FG |
1844 | */ |
1845 | bool valid = rgw_obj_key::parse_raw_oid(index_key.name, &obj); | |
1846 | if (!valid) { | |
9f95a23c TL |
1847 | ldout(cct, 0) << "ERROR: could not parse object name: " << |
1848 | obj.name << dendl; | |
7c673cae FG |
1849 | continue; |
1850 | } | |
11fdf7f2 | 1851 | |
9f95a23c | 1852 | bool matched_ns = (obj.ns == params.ns); |
7c673cae FG |
1853 | if (!params.list_versions && !entry.is_visible()) { |
1854 | continue; | |
1855 | } | |
1856 | ||
9f95a23c | 1857 | if (params.enforce_ns && !matched_ns) { |
7c673cae FG |
1858 | if (!params.ns.empty()) { |
1859 | /* we've iterated past the namespace we're searching -- done now */ | |
1860 | truncated = false; | |
1861 | goto done; | |
1862 | } | |
1863 | ||
1864 | /* we're not looking at the namespace this object is in, next! */ | |
1865 | continue; | |
1866 | } | |
1867 | ||
1868 | if (cur_end_marker_valid && cur_end_marker <= index_key) { | |
1869 | truncated = false; | |
1870 | goto done; | |
1871 | } | |
1872 | ||
1873 | if (count < max) { | |
9f95a23c TL |
1874 | params.marker = index_key; |
1875 | next_marker = index_key; | |
7c673cae FG |
1876 | } |
1877 | ||
9f95a23c TL |
1878 | if (params.filter && |
1879 | ! params.filter->filter(obj.name, index_key.name)) { | |
7c673cae | 1880 | continue; |
9f95a23c | 1881 | } |
7c673cae | 1882 | |
1adf2230 | 1883 | if (params.prefix.size() && |
9f95a23c | 1884 | 0 != obj.name.compare(0, params.prefix.size(), params.prefix)) { |
7c673cae | 1885 | continue; |
9f95a23c | 1886 | } |
7c673cae FG |
1887 | |
1888 | if (!params.delim.empty()) { | |
9f95a23c TL |
1889 | const int delim_pos = obj.name.find(params.delim, params.prefix.size()); |
1890 | if (delim_pos >= 0) { | |
1891 | // run either the code where delimiter filtering is done a) | |
1892 | // in the OSD/CLS or b) here. | |
1893 | if (cls_filtered) { | |
1894 | // NOTE: this condition is for the newer versions of the | |
1895 | // OSD that does filtering on the CLS side | |
1896 | ||
1897 | // should only find one delimiter at the end if it finds any | |
1898 | // after the prefix | |
1899 | if (delim_pos != | |
1900 | int(obj.name.length() - params.delim.length())) { | |
1901 | ldout(cct, 0) << | |
1902 | "WARNING: found delimiter in place other than the end of " | |
1903 | "the prefix; obj.name=" << obj.name << | |
1904 | ", prefix=" << params.prefix << dendl; | |
1905 | } | |
1906 | if (common_prefixes) { | |
1907 | if (count >= max) { | |
1908 | truncated = true; | |
1909 | goto done; | |
1910 | } | |
1911 | ||
1912 | (*common_prefixes)[obj.name] = true; | |
1913 | count++; | |
1914 | } | |
1915 | ||
1916 | continue; | |
1917 | } else { | |
1918 | // NOTE: this condition is for older versions of the OSD | |
1919 | // that do not filter on the CLS side, so the following code | |
1920 | // must do the filtering; once we reach version 16 of ceph, | |
1921 | // this code can be removed along with the conditional that | |
1922 | // can lead this way | |
1923 | ||
1924 | /* extract key -with trailing delimiter- for CommonPrefix */ | |
1925 | string prefix_key = | |
1926 | obj.name.substr(0, delim_pos + params.delim.length()); | |
1927 | ||
1928 | if (common_prefixes && | |
1929 | common_prefixes->find(prefix_key) == common_prefixes->end()) { | |
1930 | if (count >= max) { | |
1931 | truncated = true; | |
1932 | goto done; | |
1933 | } | |
1934 | next_marker = prefix_key; | |
1935 | (*common_prefixes)[prefix_key] = true; | |
1936 | ||
1937 | count++; | |
1938 | } | |
1939 | ||
1940 | continue; | |
1941 | } // if we're running an older OSD version | |
1942 | } // if a delimiter was found after prefix | |
1943 | } // if a delimiter was passed in | |
7c673cae FG |
1944 | |
1945 | if (count >= max) { | |
1946 | truncated = true; | |
1947 | goto done; | |
1948 | } | |
1949 | ||
9f95a23c TL |
1950 | ldout(cct, 20) << "RGWRados::Bucket::List::" << __func__ << |
1951 | " adding entry " << entry.key << " to result" << dendl; | |
1952 | ||
7c673cae FG |
1953 | result->emplace_back(std::move(entry)); |
1954 | count++; | |
9f95a23c TL |
1955 | } // eiter for loop |
1956 | ||
1957 | // NOTE: the following conditional is needed by older versions of | |
1958 | // the OSD that don't do delimiter filtering on the CLS side; once | |
1959 | // we reach version 16 of ceph, the following conditional and the | |
1960 | // code within can be removed | |
1961 | if (!cls_filtered && !params.delim.empty()) { | |
1962 | int marker_delim_pos = | |
1963 | cur_marker.name.find(params.delim, cur_prefix.size()); | |
eafe8130 | 1964 | if (marker_delim_pos >= 0) { |
9f95a23c TL |
1965 | std::string skip_after_delim = |
1966 | cur_marker.name.substr(0, marker_delim_pos); | |
eafe8130 TL |
1967 | skip_after_delim.append(after_delim_s); |
1968 | ||
1969 | ldout(cct, 20) << "skip_after_delim=" << skip_after_delim << dendl; | |
1970 | ||
1971 | if (skip_after_delim > cur_marker.name) { | |
1972 | cur_marker = skip_after_delim; | |
1973 | ldout(cct, 20) << "setting cur_marker=" | |
1974 | << cur_marker.name | |
1975 | << "[" << cur_marker.instance << "]" | |
1976 | << dendl; | |
1977 | } | |
1978 | } | |
9f95a23c TL |
1979 | } // if older osd didn't do delimiter filtering |
1980 | ||
1981 | ldout(cct, 20) << "RGWRados::Bucket::List::" << __func__ << | |
1982 | " INFO end of outer loop, truncated=" << truncated << | |
1983 | ", count=" << count << ", attempt=" << attempt << dendl; | |
1984 | ||
1985 | if (!truncated || count >= (max + 1) / 2) { | |
1986 | // if we finished listing, or if we're returning at least half the | |
1987 | // requested entries, that's enough; S3 and swift protocols allow | |
1988 | // returning fewer than max entries | |
1989 | break; | |
1990 | } else if (attempt > 8 && count >= 1) { | |
1991 | // if we've made at least 8 attempts and we have some, but very | |
1992 | // few, results, return with what we have | |
1993 | break; | |
eafe8130 | 1994 | } |
f6b5b4d7 | 1995 | } // for (uint16_t attempt... |
7c673cae FG |
1996 | |
1997 | done: | |
9f95a23c TL |
1998 | |
1999 | if (is_truncated) { | |
7c673cae | 2000 | *is_truncated = truncated; |
9f95a23c | 2001 | } |
7c673cae FG |
2002 | |
2003 | return 0; | |
1adf2230 AA |
2004 | } // list_objects_ordered |
2005 | ||
2006 | ||
2007 | /** | |
2008 | * Get listing of the objects in a bucket and allow the results to be out | |
2009 | * of order. | |
2010 | * | |
2011 | * Even though there are key differences with the ordered counterpart, | |
2012 | * the parameters are the same to maintain some compatability. | |
2013 | * | |
2014 | * max: maximum number of results to return | |
2015 | * bucket: bucket to list contents of | |
2016 | * prefix: only return results that match this prefix | |
2017 | * delim: should not be set; if it is we should have indicated an error | |
2018 | * marker: if filled in, begin the listing with this object. | |
2019 | * end_marker: if filled in, end the listing with this object. | |
2020 | * result: the objects are put in here. | |
2021 | * common_prefixes: this is never filled with an unordered list; the param | |
2022 | * is maintained for compatibility | |
2023 | * is_truncated: if number of objects in the bucket is bigger than max, then | |
2024 | * truncated. | |
2025 | */ | |
eafe8130 | 2026 | int RGWRados::Bucket::List::list_objects_unordered(int64_t max_p, |
1adf2230 AA |
2027 | vector<rgw_bucket_dir_entry> *result, |
2028 | map<string, bool> *common_prefixes, | |
9f95a23c TL |
2029 | bool *is_truncated, |
2030 | optional_yield y) | |
1adf2230 AA |
2031 | { |
2032 | RGWRados *store = target->get_store(); | |
2033 | CephContext *cct = store->ctx(); | |
2034 | int shard_id = target->get_shard_id(); | |
2035 | ||
2036 | int count = 0; | |
2037 | bool truncated = true; | |
2038 | ||
eafe8130 TL |
2039 | const int64_t max = // protect against memory issues and negative vals |
2040 | std::min(bucket_list_objects_absolute_max, std::max(int64_t(0), max_p)); | |
2041 | ||
1adf2230 AA |
2042 | // read a few extra in each call to cls_bucket_list_unordered in |
2043 | // case some are filtered out due to namespace matching, versioning, | |
2044 | // filtering, etc. | |
2045 | const int64_t max_read_ahead = 100; | |
2046 | const uint32_t read_ahead = uint32_t(max + std::min(max, max_read_ahead)); | |
2047 | ||
2048 | result->clear(); | |
2049 | ||
9f95a23c TL |
2050 | // use a local marker; either the marker will have a previous entry |
2051 | // or it will be empty; either way it's OK to copy | |
11fdf7f2 TL |
2052 | rgw_obj_key marker_obj(params.marker.name, |
2053 | params.marker.instance, | |
f91f0fd5 | 2054 | params.ns.empty() ? params.marker.ns : params.ns); |
1adf2230 AA |
2055 | rgw_obj_index_key cur_marker; |
2056 | marker_obj.get_index_key(&cur_marker); | |
2057 | ||
11fdf7f2 TL |
2058 | rgw_obj_key end_marker_obj(params.end_marker.name, |
2059 | params.end_marker.instance, | |
f91f0fd5 | 2060 | params.ns.empty() ? params.end_marker.ns : params.ns); |
1adf2230 AA |
2061 | rgw_obj_index_key cur_end_marker; |
2062 | end_marker_obj.get_index_key(&cur_end_marker); | |
2063 | const bool cur_end_marker_valid = !params.end_marker.empty(); | |
2064 | ||
2065 | rgw_obj_key prefix_obj(params.prefix); | |
9f95a23c | 2066 | prefix_obj.set_ns(params.ns); |
1adf2230 AA |
2067 | string cur_prefix = prefix_obj.get_index_key_name(); |
2068 | ||
2069 | while (truncated && count <= max) { | |
2070 | std::vector<rgw_bucket_dir_entry> ent_list; | |
9f95a23c TL |
2071 | ent_list.reserve(read_ahead); |
2072 | ||
1adf2230 AA |
2073 | int r = store->cls_bucket_list_unordered(target->get_bucket_info(), |
2074 | shard_id, | |
2075 | cur_marker, | |
2076 | cur_prefix, | |
2077 | read_ahead, | |
2078 | params.list_versions, | |
2079 | ent_list, | |
2080 | &truncated, | |
9f95a23c TL |
2081 | &cur_marker, |
2082 | y); | |
1adf2230 AA |
2083 | if (r < 0) |
2084 | return r; | |
2085 | ||
2086 | // NB: while regions of ent_list will be sorted, we have no | |
2087 | // guarantee that all items will be sorted since they can cross | |
2088 | // shard boundaries | |
2089 | ||
2090 | for (auto& entry : ent_list) { | |
2091 | rgw_obj_index_key index_key = entry.key; | |
2092 | rgw_obj_key obj(index_key); | |
2093 | ||
9f95a23c TL |
2094 | if (count < max) { |
2095 | params.marker.set(index_key); | |
2096 | next_marker.set(index_key); | |
2097 | } | |
2098 | ||
1adf2230 AA |
2099 | /* note that parse_raw_oid() here will not set the correct |
2100 | * object's instance, as rgw_obj_index_key encodes that | |
2101 | * separately. We don't need to set the instance because it's | |
2102 | * not needed for the checks here and we end up using the raw | |
2103 | * entry for the return vector | |
2104 | */ | |
2105 | bool valid = rgw_obj_key::parse_raw_oid(index_key.name, &obj); | |
2106 | if (!valid) { | |
2107 | ldout(cct, 0) << "ERROR: could not parse object name: " << | |
2108 | obj.name << dendl; | |
2109 | continue; | |
2110 | } | |
2111 | ||
2112 | if (!params.list_versions && !entry.is_visible()) { | |
2113 | continue; | |
2114 | } | |
2115 | ||
2116 | if (params.enforce_ns && obj.ns != params.ns) { | |
2117 | continue; | |
2118 | } | |
2119 | ||
2120 | if (cur_end_marker_valid && cur_end_marker <= index_key) { | |
2121 | // we're not guaranteed items will come in order, so we have | |
2122 | // to loop through all | |
2123 | continue; | |
2124 | } | |
2125 | ||
1adf2230 AA |
2126 | if (params.filter && !params.filter->filter(obj.name, index_key.name)) |
2127 | continue; | |
2128 | ||
2129 | if (params.prefix.size() && | |
2130 | (0 != obj.name.compare(0, params.prefix.size(), params.prefix))) | |
2131 | continue; | |
2132 | ||
2133 | if (count >= max) { | |
2134 | truncated = true; | |
2135 | goto done; | |
2136 | } | |
2137 | ||
2138 | result->emplace_back(std::move(entry)); | |
2139 | count++; | |
2140 | } // for (auto& entry : ent_list) | |
2141 | } // while (truncated && count <= max) | |
2142 | ||
2143 | done: | |
2144 | if (is_truncated) | |
2145 | *is_truncated = truncated; | |
2146 | ||
2147 | return 0; | |
2148 | } // list_objects_unordered | |
2149 | ||
7c673cae FG |
2150 | |
2151 | /** | |
2152 | * create a rados pool, associated meta info | |
2153 | * returns 0 on success, -ERR# otherwise. | |
2154 | */ | |
2155 | int RGWRados::create_pool(const rgw_pool& pool) | |
2156 | { | |
c07f9fc5 | 2157 | librados::IoCtx io_ctx; |
28e407b8 AA |
2158 | constexpr bool create = true; |
2159 | return rgw_init_ioctx(get_rados_handle(), pool, io_ctx, create); | |
7c673cae FG |
2160 | } |
2161 | ||
9f95a23c | 2162 | void RGWRados::create_bucket_id(string *bucket_id) |
7c673cae | 2163 | { |
9f95a23c TL |
2164 | uint64_t iid = instance_id(); |
2165 | uint64_t bid = next_bucket_id(); | |
2166 | char buf[svc.zone->get_zone_params().get_id().size() + 48]; | |
2167 | snprintf(buf, sizeof(buf), "%s.%" PRIu64 ".%" PRIu64, | |
2168 | svc.zone->get_zone_params().get_id().c_str(), iid, bid); | |
2169 | *bucket_id = buf; | |
2170 | } | |
7c673cae | 2171 | |
11fdf7f2 | 2172 | int RGWRados::create_bucket(const RGWUserInfo& owner, rgw_bucket& bucket, |
7c673cae | 2173 | const string& zonegroup_id, |
11fdf7f2 | 2174 | const rgw_placement_rule& placement_rule, |
7c673cae FG |
2175 | const string& swift_ver_location, |
2176 | const RGWQuotaInfo * pquota_info, | |
2177 | map<std::string, bufferlist>& attrs, | |
2178 | RGWBucketInfo& info, | |
2179 | obj_version *pobjv, | |
2180 | obj_version *pep_objv, | |
2181 | real_time creation_time, | |
2182 | rgw_bucket *pmaster_bucket, | |
2183 | uint32_t *pmaster_num_shards, | |
2184 | bool exclusive) | |
2185 | { | |
2186 | #define MAX_CREATE_RETRIES 20 /* need to bound retries */ | |
11fdf7f2 | 2187 | rgw_placement_rule selected_placement_rule; |
7c673cae FG |
2188 | RGWZonePlacementInfo rule_info; |
2189 | ||
2190 | for (int i = 0; i < MAX_CREATE_RETRIES; i++) { | |
2191 | int ret = 0; | |
11fdf7f2 TL |
2192 | ret = svc.zone->select_bucket_placement(owner, zonegroup_id, placement_rule, |
2193 | &selected_placement_rule, &rule_info); | |
7c673cae FG |
2194 | if (ret < 0) |
2195 | return ret; | |
2196 | ||
2197 | if (!pmaster_bucket) { | |
2198 | create_bucket_id(&bucket.marker); | |
2199 | bucket.bucket_id = bucket.marker; | |
2200 | } else { | |
2201 | bucket.marker = pmaster_bucket->marker; | |
2202 | bucket.bucket_id = pmaster_bucket->bucket_id; | |
2203 | } | |
2204 | ||
2205 | RGWObjVersionTracker& objv_tracker = info.objv_tracker; | |
2206 | ||
9f95a23c TL |
2207 | objv_tracker.read_version.clear(); |
2208 | ||
7c673cae FG |
2209 | if (pobjv) { |
2210 | objv_tracker.write_version = *pobjv; | |
2211 | } else { | |
2212 | objv_tracker.generate_new_write_ver(cct); | |
2213 | } | |
2214 | ||
2215 | info.bucket = bucket; | |
2216 | info.owner = owner.user_id; | |
2217 | info.zonegroup = zonegroup_id; | |
11fdf7f2 | 2218 | info.placement_rule = selected_placement_rule; |
7c673cae FG |
2219 | info.index_type = rule_info.index_type; |
2220 | info.swift_ver_location = swift_ver_location; | |
2221 | info.swift_versioning = (!swift_ver_location.empty()); | |
2222 | if (pmaster_num_shards) { | |
2223 | info.num_shards = *pmaster_num_shards; | |
2224 | } else { | |
2225 | info.num_shards = bucket_index_max_shards; | |
2226 | } | |
2227 | info.bucket_index_shard_hash_type = RGWBucketInfo::MOD; | |
2228 | info.requester_pays = false; | |
2229 | if (real_clock::is_zero(creation_time)) { | |
2230 | info.creation_time = ceph::real_clock::now(); | |
2231 | } else { | |
2232 | info.creation_time = creation_time; | |
2233 | } | |
2234 | if (pquota_info) { | |
2235 | info.quota = *pquota_info; | |
2236 | } | |
2237 | ||
9f95a23c | 2238 | int r = svc.bi->init_index(info); |
11fdf7f2 TL |
2239 | if (r < 0) { |
2240 | return r; | |
2241 | } | |
7c673cae | 2242 | |
11fdf7f2 | 2243 | ret = put_linked_bucket_info(info, exclusive, ceph::real_time(), pep_objv, &attrs, true); |
9f95a23c TL |
2244 | if (ret == -ECANCELED) { |
2245 | ret = -EEXIST; | |
2246 | } | |
11fdf7f2 | 2247 | if (ret == -EEXIST) { |
11fdf7f2 | 2248 | /* we need to reread the info and return it, caller will have a use for it */ |
9f95a23c TL |
2249 | RGWBucketInfo orig_info; |
2250 | r = get_bucket_info(&svc, bucket.tenant, bucket.name, orig_info, NULL, null_yield, NULL); | |
11fdf7f2 TL |
2251 | if (r < 0) { |
2252 | if (r == -ENOENT) { | |
2253 | continue; | |
2254 | } | |
2255 | ldout(cct, 0) << "get_bucket_info returned " << r << dendl; | |
2256 | return r; | |
2257 | } | |
7c673cae | 2258 | |
11fdf7f2 | 2259 | /* only remove it if it's a different bucket instance */ |
9f95a23c TL |
2260 | if (orig_info.bucket.bucket_id != bucket.bucket_id) { |
2261 | int r = svc.bi->clean_index(info); | |
2262 | if (r < 0) { | |
2263 | ldout(cct, 0) << "WARNING: could not remove bucket index (r=" << r << ")" << dendl; | |
2264 | } | |
2265 | r = ctl.bucket->remove_bucket_instance_info(info.bucket, info, null_yield); | |
2266 | if (r < 0) { | |
2267 | ldout(cct, 0) << "WARNING: " << __func__ << "(): failed to remove bucket instance info: bucket instance=" << info.bucket.get_key() << ": r=" << r << dendl; | |
2268 | /* continue anyway */ | |
2269 | } | |
11fdf7f2 | 2270 | } |
9f95a23c TL |
2271 | |
2272 | info = std::move(orig_info); | |
2273 | /* ret == -EEXIST here */ | |
11fdf7f2 | 2274 | } |
7c673cae | 2275 | return ret; |
7c673cae FG |
2276 | } |
2277 | ||
11fdf7f2 TL |
2278 | /* this is highly unlikely */ |
2279 | ldout(cct, 0) << "ERROR: could not create bucket, continuously raced with bucket creation and removal" << dendl; | |
2280 | return -ENOENT; | |
7c673cae FG |
2281 | } |
2282 | ||
11fdf7f2 | 2283 | bool RGWRados::get_obj_data_pool(const rgw_placement_rule& placement_rule, const rgw_obj& obj, rgw_pool *pool) |
7c673cae | 2284 | { |
11fdf7f2 TL |
2285 | return rgw_get_obj_data_pool(svc.zone->get_zonegroup(), svc.zone->get_zone_params(), placement_rule, obj, pool); |
2286 | } | |
c07f9fc5 | 2287 | |
11fdf7f2 TL |
2288 | bool RGWRados::obj_to_raw(const rgw_placement_rule& placement_rule, const rgw_obj& obj, rgw_raw_obj *raw_obj) |
2289 | { | |
2290 | get_obj_bucket_and_oid_loc(obj, raw_obj->oid, raw_obj->loc); | |
c07f9fc5 | 2291 | |
11fdf7f2 | 2292 | return get_obj_data_pool(placement_rule, obj, &raw_obj->pool); |
7c673cae FG |
2293 | } |
2294 | ||
2295 | int RGWRados::get_obj_head_ioctx(const RGWBucketInfo& bucket_info, const rgw_obj& obj, librados::IoCtx *ioctx) | |
2296 | { | |
2297 | string oid, key; | |
2298 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
2299 | ||
2300 | rgw_pool pool; | |
2301 | if (!get_obj_data_pool(bucket_info.placement_rule, obj, &pool)) { | |
2302 | ldout(cct, 0) << "ERROR: cannot get data pool for obj=" << obj << ", probably misconfiguration" << dendl; | |
2303 | return -EIO; | |
2304 | } | |
2305 | ||
494da23a | 2306 | int r = open_pool_ctx(pool, *ioctx, false); |
7c673cae FG |
2307 | if (r < 0) { |
2308 | return r; | |
2309 | } | |
2310 | ||
2311 | ioctx->locator_set_key(key); | |
2312 | ||
2313 | return 0; | |
2314 | } | |
2315 | ||
2316 | int RGWRados::get_obj_head_ref(const RGWBucketInfo& bucket_info, const rgw_obj& obj, rgw_rados_ref *ref) | |
2317 | { | |
11fdf7f2 | 2318 | get_obj_bucket_and_oid_loc(obj, ref->obj.oid, ref->obj.loc); |
7c673cae FG |
2319 | |
2320 | rgw_pool pool; | |
2321 | if (!get_obj_data_pool(bucket_info.placement_rule, obj, &pool)) { | |
2322 | ldout(cct, 0) << "ERROR: cannot get data pool for obj=" << obj << ", probably misconfiguration" << dendl; | |
2323 | return -EIO; | |
2324 | } | |
2325 | ||
9f95a23c TL |
2326 | ref->pool = svc.rados->pool(pool); |
2327 | ||
2328 | int r = ref->pool.open(RGWSI_RADOS::OpenParams() | |
2329 | .set_mostly_omap(false)); | |
7c673cae | 2330 | if (r < 0) { |
9f95a23c | 2331 | ldout(cct, 0) << "ERROR: failed opening data pool (pool=" << pool << "); r=" << r << dendl; |
7c673cae FG |
2332 | return r; |
2333 | } | |
2334 | ||
9f95a23c | 2335 | ref->pool.ioctx().locator_set_key(ref->obj.loc); |
7c673cae FG |
2336 | |
2337 | return 0; | |
2338 | } | |
2339 | ||
224ce89b | 2340 | int RGWRados::get_raw_obj_ref(const rgw_raw_obj& obj, rgw_rados_ref *ref) |
7c673cae | 2341 | { |
11fdf7f2 | 2342 | ref->obj = obj; |
7c673cae | 2343 | |
11fdf7f2 TL |
2344 | if (ref->obj.oid.empty()) { |
2345 | ref->obj.oid = obj.pool.to_str(); | |
2346 | ref->obj.pool = svc.zone->get_zone_params().domain_root; | |
7c673cae | 2347 | } |
9f95a23c TL |
2348 | ref->pool = svc.rados->pool(obj.pool); |
2349 | int r = ref->pool.open(RGWSI_RADOS::OpenParams() | |
2350 | .set_mostly_omap(false)); | |
2351 | if (r < 0) { | |
2352 | ldout(cct, 0) << "ERROR: failed opening pool (pool=" << obj.pool << "); r=" << r << dendl; | |
7c673cae | 2353 | return r; |
9f95a23c | 2354 | } |
7c673cae | 2355 | |
9f95a23c | 2356 | ref->pool.ioctx().locator_set_key(ref->obj.loc); |
7c673cae FG |
2357 | |
2358 | return 0; | |
2359 | } | |
2360 | ||
224ce89b | 2361 | int RGWRados::get_system_obj_ref(const rgw_raw_obj& obj, rgw_rados_ref *ref) |
7c673cae | 2362 | { |
224ce89b | 2363 | return get_raw_obj_ref(obj, ref); |
7c673cae FG |
2364 | } |
2365 | ||
2366 | /* | |
2367 | * fixes an issue where head objects were supposed to have a locator created, but ended | |
2368 | * up without one | |
2369 | */ | |
2370 | int RGWRados::fix_head_obj_locator(const RGWBucketInfo& bucket_info, bool copy_obj, bool remove_bad, rgw_obj_key& key) | |
2371 | { | |
2372 | const rgw_bucket& bucket = bucket_info.bucket; | |
2373 | string oid; | |
2374 | string locator; | |
2375 | ||
2376 | rgw_obj obj(bucket, key); | |
2377 | ||
2378 | get_obj_bucket_and_oid_loc(obj, oid, locator); | |
2379 | ||
2380 | if (locator.empty()) { | |
2381 | ldout(cct, 20) << "object does not have a locator, nothing to fix" << dendl; | |
2382 | return 0; | |
2383 | } | |
2384 | ||
2385 | librados::IoCtx ioctx; | |
2386 | ||
2387 | int ret = get_obj_head_ioctx(bucket_info, obj, &ioctx); | |
2388 | if (ret < 0) { | |
2389 | cerr << "ERROR: get_obj_head_ioctx() returned ret=" << ret << std::endl; | |
2390 | return ret; | |
2391 | } | |
2392 | ioctx.locator_set_key(string()); /* override locator for this object, use empty locator */ | |
2393 | ||
2394 | uint64_t size; | |
2395 | bufferlist data; | |
2396 | ||
2397 | struct timespec mtime_ts; | |
2398 | map<string, bufferlist> attrs; | |
2399 | librados::ObjectReadOperation op; | |
2400 | op.getxattrs(&attrs, NULL); | |
2401 | op.stat2(&size, &mtime_ts, NULL); | |
2402 | #define HEAD_SIZE 512 * 1024 | |
2403 | op.read(0, HEAD_SIZE, &data, NULL); | |
2404 | ||
9f95a23c | 2405 | ret = rgw_rados_operate(ioctx, oid, &op, &data, null_yield); |
7c673cae | 2406 | if (ret < 0) { |
9f95a23c | 2407 | lderr(cct) << "ERROR: rgw_rados_operate(oid=" << oid << ") returned ret=" << ret << dendl; |
7c673cae FG |
2408 | return ret; |
2409 | } | |
2410 | ||
2411 | if (size > HEAD_SIZE) { | |
2412 | lderr(cct) << "ERROR: returned object size (" << size << ") > HEAD_SIZE (" << HEAD_SIZE << ")" << dendl; | |
2413 | return -EIO; | |
2414 | } | |
2415 | ||
2416 | if (size != data.length()) { | |
2417 | lderr(cct) << "ERROR: returned object size (" << size << ") != data.length() (" << data.length() << ")" << dendl; | |
2418 | return -EIO; | |
2419 | } | |
2420 | ||
2421 | if (copy_obj) { | |
2422 | librados::ObjectWriteOperation wop; | |
2423 | ||
2424 | wop.mtime2(&mtime_ts); | |
2425 | ||
2426 | map<string, bufferlist>::iterator iter; | |
2427 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
2428 | wop.setxattr(iter->first.c_str(), iter->second); | |
2429 | } | |
2430 | ||
2431 | wop.write(0, data); | |
2432 | ||
2433 | ioctx.locator_set_key(locator); | |
9f95a23c | 2434 | rgw_rados_operate(ioctx, oid, &wop, null_yield); |
7c673cae FG |
2435 | } |
2436 | ||
2437 | if (remove_bad) { | |
2438 | ioctx.locator_set_key(string()); | |
2439 | ||
2440 | ret = ioctx.remove(oid); | |
2441 | if (ret < 0) { | |
2442 | lderr(cct) << "ERROR: failed to remove original bad object" << dendl; | |
2443 | return ret; | |
2444 | } | |
2445 | } | |
2446 | ||
2447 | return 0; | |
2448 | } | |
2449 | ||
2450 | int RGWRados::move_rados_obj(librados::IoCtx& src_ioctx, | |
2451 | const string& src_oid, const string& src_locator, | |
2452 | librados::IoCtx& dst_ioctx, | |
2453 | const string& dst_oid, const string& dst_locator) | |
2454 | { | |
2455 | ||
2456 | #define COPY_BUF_SIZE (4 * 1024 * 1024) | |
2457 | bool done = false; | |
2458 | uint64_t chunk_size = COPY_BUF_SIZE; | |
2459 | uint64_t ofs = 0; | |
2460 | int ret = 0; | |
2461 | real_time mtime; | |
2462 | struct timespec mtime_ts; | |
2463 | uint64_t size; | |
2464 | ||
2465 | if (src_oid == dst_oid && src_locator == dst_locator) { | |
2466 | return 0; | |
2467 | } | |
2468 | ||
2469 | src_ioctx.locator_set_key(src_locator); | |
2470 | dst_ioctx.locator_set_key(dst_locator); | |
2471 | ||
2472 | do { | |
2473 | bufferlist data; | |
2474 | ObjectReadOperation rop; | |
2475 | ObjectWriteOperation wop; | |
2476 | ||
2477 | if (ofs == 0) { | |
2478 | rop.stat2(&size, &mtime_ts, NULL); | |
2479 | mtime = real_clock::from_timespec(mtime_ts); | |
2480 | } | |
2481 | rop.read(ofs, chunk_size, &data, NULL); | |
9f95a23c | 2482 | ret = rgw_rados_operate(src_ioctx, src_oid, &rop, &data, null_yield); |
7c673cae FG |
2483 | if (ret < 0) { |
2484 | goto done_err; | |
2485 | } | |
2486 | ||
2487 | if (data.length() == 0) { | |
2488 | break; | |
2489 | } | |
2490 | ||
2491 | if (ofs == 0) { | |
2492 | wop.create(true); /* make it exclusive */ | |
2493 | wop.mtime2(&mtime_ts); | |
2494 | mtime = real_clock::from_timespec(mtime_ts); | |
2495 | } | |
2496 | wop.write(ofs, data); | |
9f95a23c | 2497 | ret = rgw_rados_operate(dst_ioctx, dst_oid, &wop, null_yield); |
11fdf7f2 TL |
2498 | if (ret < 0) { |
2499 | goto done_err; | |
2500 | } | |
7c673cae FG |
2501 | ofs += data.length(); |
2502 | done = data.length() != chunk_size; | |
2503 | } while (!done); | |
2504 | ||
2505 | if (ofs != size) { | |
2506 | lderr(cct) << "ERROR: " << __func__ << ": copying " << src_oid << " -> " << dst_oid | |
2507 | << ": expected " << size << " bytes to copy, ended up with " << ofs << dendl; | |
2508 | ret = -EIO; | |
2509 | goto done_err; | |
2510 | } | |
2511 | ||
2512 | src_ioctx.remove(src_oid); | |
2513 | ||
2514 | return 0; | |
2515 | ||
2516 | done_err: | |
11fdf7f2 | 2517 | // TODO: clean up dst_oid if we created it |
7c673cae FG |
2518 | lderr(cct) << "ERROR: failed to copy " << src_oid << " -> " << dst_oid << dendl; |
2519 | return ret; | |
2520 | } | |
2521 | ||
2522 | /* | |
2523 | * fixes an issue where head objects were supposed to have a locator created, but ended | |
2524 | * up without one | |
2525 | */ | |
9f95a23c | 2526 | int RGWRados::fix_tail_obj_locator(const RGWBucketInfo& bucket_info, rgw_obj_key& key, bool fix, bool *need_fix, optional_yield y) |
7c673cae FG |
2527 | { |
2528 | const rgw_bucket& bucket = bucket_info.bucket; | |
2529 | rgw_obj obj(bucket, key); | |
2530 | ||
2531 | if (need_fix) { | |
2532 | *need_fix = false; | |
2533 | } | |
2534 | ||
2535 | rgw_rados_ref ref; | |
2536 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
2537 | if (r < 0) { | |
2538 | return r; | |
2539 | } | |
2540 | ||
2541 | RGWObjState *astate = NULL; | |
9f95a23c TL |
2542 | RGWObjectCtx rctx(this->store); |
2543 | r = get_obj_state(&rctx, bucket_info, obj, &astate, false, y); | |
7c673cae FG |
2544 | if (r < 0) |
2545 | return r; | |
2546 | ||
9f95a23c | 2547 | if (astate->manifest) { |
7c673cae | 2548 | RGWObjManifest::obj_iterator miter; |
9f95a23c | 2549 | RGWObjManifest& manifest = *astate->manifest; |
7c673cae FG |
2550 | for (miter = manifest.obj_begin(); miter != manifest.obj_end(); ++miter) { |
2551 | rgw_raw_obj raw_loc = miter.get_location().get_raw_obj(this); | |
2552 | rgw_obj loc; | |
2553 | string oid; | |
2554 | string locator; | |
2555 | ||
9f95a23c | 2556 | RGWSI_Tier_RADOS::raw_obj_to_obj(manifest.get_tail_placement().bucket, raw_loc, &loc); |
7c673cae FG |
2557 | |
2558 | if (loc.key.ns.empty()) { | |
2559 | /* continue, we're only interested in tail objects */ | |
2560 | continue; | |
2561 | } | |
2562 | ||
9f95a23c TL |
2563 | auto& ioctx = ref.pool.ioctx(); |
2564 | ||
7c673cae | 2565 | get_obj_bucket_and_oid_loc(loc, oid, locator); |
9f95a23c | 2566 | ref.pool.ioctx().locator_set_key(locator); |
7c673cae FG |
2567 | |
2568 | ldout(cct, 20) << __func__ << ": key=" << key << " oid=" << oid << " locator=" << locator << dendl; | |
2569 | ||
9f95a23c | 2570 | r = ioctx.stat(oid, NULL, NULL); |
7c673cae FG |
2571 | if (r != -ENOENT) { |
2572 | continue; | |
2573 | } | |
2574 | ||
2575 | string bad_loc; | |
2576 | prepend_bucket_marker(bucket, loc.key.name, bad_loc); | |
2577 | ||
2578 | /* create a new ioctx with the bad locator */ | |
2579 | librados::IoCtx src_ioctx; | |
9f95a23c | 2580 | src_ioctx.dup(ioctx); |
7c673cae FG |
2581 | src_ioctx.locator_set_key(bad_loc); |
2582 | ||
2583 | r = src_ioctx.stat(oid, NULL, NULL); | |
2584 | if (r != 0) { | |
2585 | /* cannot find a broken part */ | |
2586 | continue; | |
2587 | } | |
2588 | ldout(cct, 20) << __func__ << ": found bad object part: " << loc << dendl; | |
2589 | if (need_fix) { | |
2590 | *need_fix = true; | |
2591 | } | |
2592 | if (fix) { | |
9f95a23c | 2593 | r = move_rados_obj(src_ioctx, oid, bad_loc, ioctx, oid, locator); |
7c673cae FG |
2594 | if (r < 0) { |
2595 | lderr(cct) << "ERROR: copy_rados_obj() on oid=" << oid << " returned r=" << r << dendl; | |
2596 | } | |
2597 | } | |
2598 | } | |
2599 | } | |
2600 | ||
2601 | return 0; | |
2602 | } | |
2603 | ||
f64942e4 AA |
2604 | int RGWRados::BucketShard::init(const rgw_bucket& _bucket, |
2605 | const rgw_obj& obj, | |
2606 | RGWBucketInfo* bucket_info_out) | |
7c673cae FG |
2607 | { |
2608 | bucket = _bucket; | |
2609 | ||
11fdf7f2 | 2610 | auto obj_ctx = store->svc.sysobj->init_obj_ctx(); |
7c673cae FG |
2611 | |
2612 | RGWBucketInfo bucket_info; | |
f64942e4 AA |
2613 | RGWBucketInfo* bucket_info_p = |
2614 | bucket_info_out ? bucket_info_out : &bucket_info; | |
2615 | ||
9f95a23c | 2616 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, *bucket_info_p, NULL, NULL, null_yield); |
7c673cae FG |
2617 | if (ret < 0) { |
2618 | return ret; | |
2619 | } | |
2620 | ||
9f95a23c TL |
2621 | string oid; |
2622 | ||
2623 | ret = store->svc.bi_rados->open_bucket_index_shard(*bucket_info_p, obj.get_hash_object(), &bucket_obj, &shard_id); | |
7c673cae FG |
2624 | if (ret < 0) { |
2625 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
2626 | return ret; | |
2627 | } | |
9f95a23c | 2628 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj.get_raw_obj() << dendl; |
7c673cae FG |
2629 | |
2630 | return 0; | |
2631 | } | |
2632 | ||
f64942e4 AA |
2633 | int RGWRados::BucketShard::init(const rgw_bucket& _bucket, |
2634 | int sid, | |
2635 | RGWBucketInfo* bucket_info_out) | |
7c673cae FG |
2636 | { |
2637 | bucket = _bucket; | |
2638 | shard_id = sid; | |
2639 | ||
11fdf7f2 | 2640 | auto obj_ctx = store->svc.sysobj->init_obj_ctx(); |
7c673cae FG |
2641 | |
2642 | RGWBucketInfo bucket_info; | |
f64942e4 AA |
2643 | RGWBucketInfo* bucket_info_p = |
2644 | bucket_info_out ? bucket_info_out : &bucket_info; | |
9f95a23c | 2645 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, *bucket_info_p, NULL, NULL, null_yield); |
7c673cae FG |
2646 | if (ret < 0) { |
2647 | return ret; | |
2648 | } | |
2649 | ||
9f95a23c TL |
2650 | string oid; |
2651 | ||
2652 | ret = store->svc.bi_rados->open_bucket_index_shard(*bucket_info_p, shard_id, &bucket_obj); | |
7c673cae FG |
2653 | if (ret < 0) { |
2654 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
2655 | return ret; | |
2656 | } | |
9f95a23c | 2657 | ldout(store->ctx(), 20) << " bucket index oid: " << bucket_obj.get_raw_obj() << dendl; |
7c673cae FG |
2658 | |
2659 | return 0; | |
2660 | } | |
2661 | ||
a8e16298 TL |
2662 | int RGWRados::BucketShard::init(const RGWBucketInfo& bucket_info, |
2663 | const rgw_obj& obj) | |
2664 | { | |
2665 | bucket = bucket_info.bucket; | |
2666 | ||
9f95a23c TL |
2667 | int ret = store->svc.bi_rados->open_bucket_index_shard(bucket_info, |
2668 | obj.get_hash_object(), | |
2669 | &bucket_obj, | |
2670 | &shard_id); | |
a8e16298 TL |
2671 | if (ret < 0) { |
2672 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
2673 | return ret; | |
2674 | } | |
2675 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
2676 | ||
2677 | return 0; | |
2678 | } | |
2679 | ||
b32b8144 FG |
2680 | int RGWRados::BucketShard::init(const RGWBucketInfo& bucket_info, int sid) |
2681 | { | |
2682 | bucket = bucket_info.bucket; | |
2683 | shard_id = sid; | |
2684 | ||
9f95a23c | 2685 | int ret = store->svc.bi_rados->open_bucket_index_shard(bucket_info, shard_id, &bucket_obj); |
b32b8144 FG |
2686 | if (ret < 0) { |
2687 | ldout(store->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
2688 | return ret; | |
2689 | } | |
2690 | ldout(store->ctx(), 20) << " bucket index object: " << bucket_obj << dendl; | |
2691 | ||
2692 | return 0; | |
2693 | } | |
2694 | ||
7c673cae FG |
2695 | |
2696 | /* Execute @handler on last item in bucket listing for bucket specified | |
2697 | * in @bucket_info. @obj_prefix and @obj_delim narrow down the listing | |
2698 | * to objects matching these criterias. */ | |
2699 | int RGWRados::on_last_entry_in_listing(RGWBucketInfo& bucket_info, | |
2700 | const std::string& obj_prefix, | |
2701 | const std::string& obj_delim, | |
2702 | std::function<int(const rgw_bucket_dir_entry&)> handler) | |
2703 | { | |
2704 | RGWRados::Bucket target(this, bucket_info); | |
2705 | RGWRados::Bucket::List list_op(&target); | |
2706 | ||
2707 | list_op.params.prefix = obj_prefix; | |
2708 | list_op.params.delim = obj_delim; | |
2709 | ||
2710 | ldout(cct, 20) << "iterating listing for bucket=" << bucket_info.bucket.name | |
2711 | << ", obj_prefix=" << obj_prefix | |
2712 | << ", obj_delim=" << obj_delim | |
2713 | << dendl; | |
2714 | ||
2715 | bool is_truncated = false; | |
2716 | ||
2717 | boost::optional<rgw_bucket_dir_entry> last_entry; | |
2718 | /* We need to rewind to the last object in a listing. */ | |
2719 | do { | |
2720 | /* List bucket entries in chunks. */ | |
2721 | static constexpr int MAX_LIST_OBJS = 100; | |
2722 | std::vector<rgw_bucket_dir_entry> entries(MAX_LIST_OBJS); | |
2723 | ||
2724 | int ret = list_op.list_objects(MAX_LIST_OBJS, &entries, nullptr, | |
9f95a23c | 2725 | &is_truncated, null_yield); |
7c673cae FG |
2726 | if (ret < 0) { |
2727 | return ret; | |
2728 | } else if (!entries.empty()) { | |
2729 | last_entry = entries.back(); | |
2730 | } | |
2731 | } while (is_truncated); | |
2732 | ||
2733 | if (last_entry) { | |
2734 | return handler(*last_entry); | |
2735 | } | |
2736 | ||
2737 | /* Empty listing - no items we can run handler on. */ | |
2738 | return 0; | |
2739 | } | |
2740 | ||
2741 | ||
2742 | int RGWRados::swift_versioning_copy(RGWObjectCtx& obj_ctx, | |
2743 | const rgw_user& user, | |
2744 | RGWBucketInfo& bucket_info, | |
9f95a23c TL |
2745 | rgw_obj& obj, |
2746 | const DoutPrefixProvider *dpp, | |
2747 | optional_yield y) | |
7c673cae FG |
2748 | { |
2749 | if (! swift_versioning_enabled(bucket_info)) { | |
2750 | return 0; | |
2751 | } | |
2752 | ||
11fdf7f2 | 2753 | obj_ctx.set_atomic(obj); |
7c673cae FG |
2754 | |
2755 | RGWObjState * state = nullptr; | |
9f95a23c | 2756 | int r = get_obj_state(&obj_ctx, bucket_info, obj, &state, false, y); |
7c673cae FG |
2757 | if (r < 0) { |
2758 | return r; | |
2759 | } | |
2760 | ||
2761 | if (!state->exists) { | |
2762 | return 0; | |
2763 | } | |
2764 | ||
7c673cae FG |
2765 | const string& src_name = obj.get_oid(); |
2766 | char buf[src_name.size() + 32]; | |
2767 | struct timespec ts = ceph::real_clock::to_timespec(state->mtime); | |
2768 | snprintf(buf, sizeof(buf), "%03x%s/%lld.%06ld", (int)src_name.size(), | |
2769 | src_name.c_str(), (long long)ts.tv_sec, ts.tv_nsec / 1000); | |
2770 | ||
2771 | RGWBucketInfo dest_bucket_info; | |
2772 | ||
9f95a23c | 2773 | r = get_bucket_info(&svc, bucket_info.bucket.tenant, bucket_info.swift_ver_location, dest_bucket_info, NULL, null_yield, NULL); |
7c673cae FG |
2774 | if (r < 0) { |
2775 | ldout(cct, 10) << "failed to read dest bucket info: r=" << r << dendl; | |
2776 | if (r == -ENOENT) { | |
2777 | return -ERR_PRECONDITION_FAILED; | |
2778 | } | |
2779 | return r; | |
2780 | } | |
2781 | ||
2782 | if (dest_bucket_info.owner != bucket_info.owner) { | |
2783 | return -ERR_PRECONDITION_FAILED; | |
2784 | } | |
2785 | ||
2786 | rgw_obj dest_obj(dest_bucket_info.bucket, buf); | |
11fdf7f2 TL |
2787 | |
2788 | if (dest_bucket_info.versioning_enabled()){ | |
2789 | gen_rand_obj_instance_name(&dest_obj); | |
2790 | } | |
2791 | ||
2792 | obj_ctx.set_atomic(dest_obj); | |
7c673cae | 2793 | |
9f95a23c | 2794 | rgw_zone_id no_zone; |
7c673cae FG |
2795 | |
2796 | r = copy_obj(obj_ctx, | |
2797 | user, | |
7c673cae FG |
2798 | NULL, /* req_info *info */ |
2799 | no_zone, | |
2800 | dest_obj, | |
2801 | obj, | |
2802 | dest_bucket_info, | |
2803 | bucket_info, | |
11fdf7f2 | 2804 | bucket_info.placement_rule, |
7c673cae FG |
2805 | NULL, /* time_t *src_mtime */ |
2806 | NULL, /* time_t *mtime */ | |
2807 | NULL, /* const time_t *mod_ptr */ | |
2808 | NULL, /* const time_t *unmod_ptr */ | |
2809 | false, /* bool high_precision_time */ | |
2810 | NULL, /* const char *if_match */ | |
2811 | NULL, /* const char *if_nomatch */ | |
2812 | RGWRados::ATTRSMOD_NONE, | |
2813 | true, /* bool copy_if_newer */ | |
2814 | state->attrset, | |
11fdf7f2 | 2815 | RGWObjCategory::Main, |
7c673cae FG |
2816 | 0, /* uint64_t olh_epoch */ |
2817 | real_time(), /* time_t delete_at */ | |
2818 | NULL, /* string *version_id */ | |
2819 | NULL, /* string *ptag */ | |
2820 | NULL, /* string *petag */ | |
7c673cae | 2821 | NULL, /* void (*progress_cb)(off_t, void *) */ |
9f95a23c TL |
2822 | NULL, /* void *progress_data */ |
2823 | dpp, | |
2824 | null_yield); | |
7c673cae FG |
2825 | if (r == -ECANCELED || r == -ENOENT) { |
2826 | /* Has already been overwritten, meaning another rgw process already | |
2827 | * copied it out */ | |
2828 | return 0; | |
2829 | } | |
2830 | ||
2831 | return r; | |
2832 | } | |
2833 | ||
9f95a23c | 2834 | int RGWRados::swift_versioning_restore(RGWObjectCtx& obj_ctx, |
7c673cae FG |
2835 | const rgw_user& user, |
2836 | RGWBucketInfo& bucket_info, | |
2837 | rgw_obj& obj, | |
9f95a23c TL |
2838 | bool& restored, /* out */ |
2839 | const DoutPrefixProvider *dpp) | |
7c673cae FG |
2840 | { |
2841 | if (! swift_versioning_enabled(bucket_info)) { | |
2842 | return 0; | |
2843 | } | |
2844 | ||
2845 | /* Bucket info of the bucket that stores previous versions of our object. */ | |
2846 | RGWBucketInfo archive_binfo; | |
2847 | ||
9f95a23c | 2848 | int ret = get_bucket_info(&svc, bucket_info.bucket.tenant, |
7c673cae | 2849 | bucket_info.swift_ver_location, archive_binfo, |
9f95a23c | 2850 | nullptr, null_yield, nullptr); |
7c673cae FG |
2851 | if (ret < 0) { |
2852 | return ret; | |
2853 | } | |
2854 | ||
2855 | /* Abort the operation if the bucket storing our archive belongs to someone | |
2856 | * else. This is a limitation in comparison to Swift as we aren't taking ACLs | |
2857 | * into consideration. For we can live with that. | |
2858 | * | |
2859 | * TODO: delegate this check to un upper layer and compare with ACLs. */ | |
2860 | if (bucket_info.owner != archive_binfo.owner) { | |
2861 | return -EPERM; | |
2862 | } | |
2863 | ||
2864 | /* This code will be executed on latest version of the object. */ | |
2865 | const auto handler = [&](const rgw_bucket_dir_entry& entry) -> int { | |
9f95a23c | 2866 | rgw_zone_id no_zone; |
7c673cae FG |
2867 | |
2868 | /* We don't support object versioning of Swift API on those buckets that | |
2869 | * are already versioned using the S3 mechanism. This affects also bucket | |
2870 | * storing archived objects. Otherwise the delete operation would create | |
2871 | * a deletion marker. */ | |
2872 | if (archive_binfo.versioned()) { | |
2873 | restored = false; | |
2874 | return -ERR_PRECONDITION_FAILED; | |
2875 | } | |
2876 | ||
2877 | /* We are requesting ATTRSMOD_NONE so the attr attribute is perfectly | |
2878 | * irrelevant and may be safely skipped. */ | |
2879 | std::map<std::string, ceph::bufferlist> no_attrs; | |
2880 | ||
2881 | rgw_obj archive_obj(archive_binfo.bucket, entry.key); | |
11fdf7f2 TL |
2882 | |
2883 | if (bucket_info.versioning_enabled()){ | |
2884 | gen_rand_obj_instance_name(&obj); | |
2885 | } | |
2886 | ||
2887 | obj_ctx.set_atomic(archive_obj); | |
2888 | obj_ctx.set_atomic(obj); | |
7c673cae FG |
2889 | |
2890 | int ret = copy_obj(obj_ctx, | |
2891 | user, | |
7c673cae FG |
2892 | nullptr, /* req_info *info */ |
2893 | no_zone, | |
2894 | obj, /* dest obj */ | |
2895 | archive_obj, /* src obj */ | |
2896 | bucket_info, /* dest bucket info */ | |
2897 | archive_binfo, /* src bucket info */ | |
11fdf7f2 | 2898 | bucket_info.placement_rule, /* placement_rule */ |
7c673cae FG |
2899 | nullptr, /* time_t *src_mtime */ |
2900 | nullptr, /* time_t *mtime */ | |
2901 | nullptr, /* const time_t *mod_ptr */ | |
2902 | nullptr, /* const time_t *unmod_ptr */ | |
2903 | false, /* bool high_precision_time */ | |
2904 | nullptr, /* const char *if_match */ | |
2905 | nullptr, /* const char *if_nomatch */ | |
2906 | RGWRados::ATTRSMOD_NONE, | |
2907 | true, /* bool copy_if_newer */ | |
2908 | no_attrs, | |
11fdf7f2 | 2909 | RGWObjCategory::Main, |
7c673cae FG |
2910 | 0, /* uint64_t olh_epoch */ |
2911 | real_time(), /* time_t delete_at */ | |
2912 | nullptr, /* string *version_id */ | |
2913 | nullptr, /* string *ptag */ | |
2914 | nullptr, /* string *petag */ | |
7c673cae | 2915 | nullptr, /* void (*progress_cb)(off_t, void *) */ |
9f95a23c TL |
2916 | nullptr, /* void *progress_data */ |
2917 | dpp, | |
2918 | null_yield); | |
7c673cae FG |
2919 | if (ret == -ECANCELED || ret == -ENOENT) { |
2920 | /* Has already been overwritten, meaning another rgw process already | |
2921 | * copied it out */ | |
2922 | return 0; | |
2923 | } else if (ret < 0) { | |
2924 | return ret; | |
2925 | } else { | |
2926 | restored = true; | |
2927 | } | |
2928 | ||
2929 | /* Need to remove the archived copy. */ | |
2930 | ret = delete_obj(obj_ctx, archive_binfo, archive_obj, | |
2931 | archive_binfo.versioning_status()); | |
2932 | ||
2933 | return ret; | |
2934 | }; | |
2935 | ||
2936 | const std::string& obj_name = obj.get_oid(); | |
2937 | const auto prefix = boost::str(boost::format("%03x%s") % obj_name.size() | |
2938 | % obj_name); | |
2939 | ||
2940 | return on_last_entry_in_listing(archive_binfo, prefix, std::string(), | |
2941 | handler); | |
2942 | } | |
2943 | ||
7c673cae | 2944 | int RGWRados::Object::Write::_do_write_meta(uint64_t size, uint64_t accounted_size, |
181888fb FG |
2945 | map<string, bufferlist>& attrs, |
2946 | bool assume_noent, bool modify_tail, | |
9f95a23c | 2947 | void *_index_op, optional_yield y) |
7c673cae FG |
2948 | { |
2949 | RGWRados::Bucket::UpdateIndex *index_op = static_cast<RGWRados::Bucket::UpdateIndex *>(_index_op); | |
7c673cae FG |
2950 | RGWRados *store = target->get_store(); |
2951 | ||
2952 | ObjectWriteOperation op; | |
11fdf7f2 TL |
2953 | #ifdef WITH_LTTNG |
2954 | const struct req_state* s = get_req_state(); | |
2955 | string req_id; | |
2956 | if (!s) { | |
2957 | // fake req_id | |
2958 | req_id = store->svc.zone_utils->unique_id(store->get_new_req_id()); | |
2959 | } else { | |
2960 | req_id = s->req_id; | |
2961 | } | |
2962 | #endif | |
7c673cae FG |
2963 | |
2964 | RGWObjState *state; | |
9f95a23c | 2965 | int r = target->get_state(&state, false, y, assume_noent); |
7c673cae FG |
2966 | if (r < 0) |
2967 | return r; | |
2968 | ||
2969 | rgw_obj& obj = target->get_obj(); | |
2970 | ||
2971 | if (obj.get_oid().empty()) { | |
2972 | ldout(store->ctx(), 0) << "ERROR: " << __func__ << "(): cannot write object with empty name" << dendl; | |
2973 | return -EIO; | |
2974 | } | |
2975 | ||
224ce89b | 2976 | rgw_rados_ref ref; |
7c673cae FG |
2977 | r = store->get_obj_head_ref(target->get_bucket_info(), obj, &ref); |
2978 | if (r < 0) | |
2979 | return r; | |
2980 | ||
2981 | bool is_olh = state->is_olh; | |
2982 | ||
2983 | bool reset_obj = (meta.flags & PUT_OBJ_CREATE) != 0; | |
2984 | ||
2985 | const string *ptag = meta.ptag; | |
2986 | if (!ptag && !index_op->get_optag()->empty()) { | |
2987 | ptag = index_op->get_optag(); | |
2988 | } | |
9f95a23c | 2989 | r = target->prepare_atomic_modification(op, reset_obj, ptag, meta.if_match, meta.if_nomatch, false, modify_tail, y); |
7c673cae FG |
2990 | if (r < 0) |
2991 | return r; | |
2992 | ||
2993 | if (real_clock::is_zero(meta.set_mtime)) { | |
2994 | meta.set_mtime = real_clock::now(); | |
2995 | } | |
2996 | ||
eafe8130 TL |
2997 | if (target->bucket_info.obj_lock_enabled() && target->bucket_info.obj_lock.has_rule() && meta.flags == PUT_OBJ_CREATE) { |
2998 | auto iter = attrs.find(RGW_ATTR_OBJECT_RETENTION); | |
2999 | if (iter == attrs.end()) { | |
3000 | real_time lock_until_date = target->bucket_info.obj_lock.get_lock_until_date(meta.set_mtime); | |
3001 | string mode = target->bucket_info.obj_lock.get_mode(); | |
3002 | RGWObjectRetention obj_retention(mode, lock_until_date); | |
3003 | bufferlist bl; | |
3004 | obj_retention.encode(bl); | |
3005 | op.setxattr(RGW_ATTR_OBJECT_RETENTION, bl); | |
3006 | } | |
3007 | } | |
3008 | ||
7c673cae FG |
3009 | if (state->is_olh) { |
3010 | op.setxattr(RGW_ATTR_OLH_ID_TAG, state->olh_tag); | |
3011 | } | |
3012 | ||
3013 | struct timespec mtime_ts = real_clock::to_timespec(meta.set_mtime); | |
3014 | op.mtime2(&mtime_ts); | |
3015 | ||
3016 | if (meta.data) { | |
3017 | /* if we want to overwrite the data, we also want to overwrite the | |
3018 | xattrs, so just remove the object */ | |
3019 | op.write_full(*meta.data); | |
3020 | } | |
3021 | ||
3022 | string etag; | |
3023 | string content_type; | |
3024 | bufferlist acl_bl; | |
11fdf7f2 | 3025 | string storage_class; |
7c673cae FG |
3026 | |
3027 | map<string, bufferlist>::iterator iter; | |
3028 | if (meta.rmattrs) { | |
3029 | for (iter = meta.rmattrs->begin(); iter != meta.rmattrs->end(); ++iter) { | |
3030 | const string& name = iter->first; | |
3031 | op.rmxattr(name.c_str()); | |
3032 | } | |
3033 | } | |
3034 | ||
3035 | if (meta.manifest) { | |
11fdf7f2 TL |
3036 | storage_class = meta.manifest->get_tail_placement().placement_rule.storage_class; |
3037 | ||
7c673cae FG |
3038 | /* remove existing manifest attr */ |
3039 | iter = attrs.find(RGW_ATTR_MANIFEST); | |
3040 | if (iter != attrs.end()) | |
3041 | attrs.erase(iter); | |
3042 | ||
3043 | bufferlist bl; | |
11fdf7f2 | 3044 | encode(*meta.manifest, bl); |
7c673cae FG |
3045 | op.setxattr(RGW_ATTR_MANIFEST, bl); |
3046 | } | |
3047 | ||
3048 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
3049 | const string& name = iter->first; | |
3050 | bufferlist& bl = iter->second; | |
3051 | ||
3052 | if (!bl.length()) | |
3053 | continue; | |
3054 | ||
3055 | op.setxattr(name.c_str(), bl); | |
3056 | ||
3057 | if (name.compare(RGW_ATTR_ETAG) == 0) { | |
11fdf7f2 | 3058 | etag = rgw_bl_str(bl); |
7c673cae | 3059 | } else if (name.compare(RGW_ATTR_CONTENT_TYPE) == 0) { |
11fdf7f2 | 3060 | content_type = rgw_bl_str(bl); |
7c673cae FG |
3061 | } else if (name.compare(RGW_ATTR_ACL) == 0) { |
3062 | acl_bl = bl; | |
3063 | } | |
3064 | } | |
3065 | if (attrs.find(RGW_ATTR_PG_VER) == attrs.end()) { | |
3066 | cls_rgw_obj_store_pg_ver(op, RGW_ATTR_PG_VER); | |
3067 | } | |
3068 | ||
3069 | if (attrs.find(RGW_ATTR_SOURCE_ZONE) == attrs.end()) { | |
3070 | bufferlist bl; | |
11fdf7f2 | 3071 | encode(store->svc.zone->get_zone_short_id(), bl); |
7c673cae FG |
3072 | op.setxattr(RGW_ATTR_SOURCE_ZONE, bl); |
3073 | } | |
3074 | ||
11fdf7f2 TL |
3075 | if (!storage_class.empty()) { |
3076 | bufferlist bl; | |
3077 | bl.append(storage_class); | |
3078 | op.setxattr(RGW_ATTR_STORAGE_CLASS, bl); | |
3079 | } | |
3080 | ||
7c673cae FG |
3081 | if (!op.size()) |
3082 | return 0; | |
3083 | ||
3084 | uint64_t epoch; | |
3085 | int64_t poolid; | |
224ce89b WB |
3086 | bool orig_exists; |
3087 | uint64_t orig_size; | |
3088 | ||
3089 | if (!reset_obj) { //Multipart upload, it has immutable head. | |
3090 | orig_exists = false; | |
3091 | orig_size = 0; | |
3092 | } else { | |
3093 | orig_exists = state->exists; | |
3094 | orig_size = state->accounted_size; | |
3095 | } | |
7c673cae | 3096 | |
91327a77 AA |
3097 | bool versioned_target = (meta.olh_epoch && *meta.olh_epoch > 0) || |
3098 | !obj.key.instance.empty(); | |
7c673cae FG |
3099 | |
3100 | bool versioned_op = (target->versioning_enabled() || is_olh || versioned_target); | |
3101 | ||
3102 | if (versioned_op) { | |
3103 | index_op->set_bilog_flags(RGW_BILOG_FLAG_VERSIONED_OP); | |
3104 | } | |
3105 | ||
3106 | if (!index_op->is_prepared()) { | |
11fdf7f2 | 3107 | tracepoint(rgw_rados, prepare_enter, req_id.c_str()); |
9f95a23c | 3108 | r = index_op->prepare(CLS_RGW_OP_ADD, &state->write_tag, y); |
11fdf7f2 | 3109 | tracepoint(rgw_rados, prepare_exit, req_id.c_str()); |
7c673cae FG |
3110 | if (r < 0) |
3111 | return r; | |
3112 | } | |
3113 | ||
9f95a23c TL |
3114 | auto& ioctx = ref.pool.ioctx(); |
3115 | ||
11fdf7f2 | 3116 | tracepoint(rgw_rados, operate_enter, req_id.c_str()); |
9f95a23c | 3117 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); |
11fdf7f2 | 3118 | tracepoint(rgw_rados, operate_exit, req_id.c_str()); |
7c673cae FG |
3119 | if (r < 0) { /* we can expect to get -ECANCELED if object was replaced under, |
3120 | or -ENOENT if was removed, or -EEXIST if it did not exist | |
3121 | before and now it does */ | |
3122 | if (r == -EEXIST && assume_noent) { | |
3123 | target->invalidate_state(); | |
3124 | return r; | |
3125 | } | |
3126 | goto done_cancel; | |
3127 | } | |
3128 | ||
9f95a23c TL |
3129 | epoch = ioctx.get_last_version(); |
3130 | poolid = ioctx.get_id(); | |
7c673cae FG |
3131 | |
3132 | r = target->complete_atomic_modification(); | |
3133 | if (r < 0) { | |
3134 | ldout(store->ctx(), 0) << "ERROR: complete_atomic_modification returned r=" << r << dendl; | |
3135 | } | |
3136 | ||
11fdf7f2 | 3137 | tracepoint(rgw_rados, complete_enter, req_id.c_str()); |
7c673cae | 3138 | r = index_op->complete(poolid, epoch, size, accounted_size, |
11fdf7f2 TL |
3139 | meta.set_mtime, etag, content_type, |
3140 | storage_class, &acl_bl, | |
3141 | meta.category, meta.remove_objs, meta.user_data, meta.appendable); | |
3142 | tracepoint(rgw_rados, complete_exit, req_id.c_str()); | |
7c673cae FG |
3143 | if (r < 0) |
3144 | goto done_cancel; | |
3145 | ||
3146 | if (meta.mtime) { | |
3147 | *meta.mtime = meta.set_mtime; | |
3148 | } | |
3149 | ||
3150 | /* note that index_op was using state so we couldn't invalidate it earlier */ | |
3151 | target->invalidate_state(); | |
3152 | state = NULL; | |
3153 | ||
91327a77 | 3154 | if (versioned_op && meta.olh_epoch) { |
9f95a23c | 3155 | r = store->set_olh(target->get_ctx(), target->get_bucket_info(), obj, false, NULL, *meta.olh_epoch, real_time(), false, y, meta.zones_trace); |
7c673cae FG |
3156 | if (r < 0) { |
3157 | return r; | |
3158 | } | |
3159 | } | |
3160 | ||
3161 | if (!real_clock::is_zero(meta.delete_at)) { | |
3162 | rgw_obj_index_key obj_key; | |
3163 | obj.key.get_index_key(&obj_key); | |
3164 | ||
9f95a23c TL |
3165 | r = store->obj_expirer->hint_add(meta.delete_at, obj.bucket.tenant, obj.bucket.name, |
3166 | obj.bucket.bucket_id, obj_key); | |
7c673cae FG |
3167 | if (r < 0) { |
3168 | ldout(store->ctx(), 0) << "ERROR: objexp_hint_add() returned r=" << r << ", object will not get removed" << dendl; | |
3169 | /* ignoring error, nothing we can do at this point */ | |
3170 | } | |
3171 | } | |
3172 | meta.canceled = false; | |
3173 | ||
3174 | /* update quota cache */ | |
3efd9988 FG |
3175 | if (meta.completeMultipart){ |
3176 | store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1), | |
3177 | 0, orig_size); | |
3178 | } | |
3179 | else { | |
3180 | store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1), | |
3181 | accounted_size, orig_size); | |
3182 | } | |
7c673cae FG |
3183 | return 0; |
3184 | ||
3185 | done_cancel: | |
3186 | int ret = index_op->cancel(); | |
3187 | if (ret < 0) { | |
3188 | ldout(store->ctx(), 0) << "ERROR: index_op.cancel()() returned ret=" << ret << dendl; | |
3189 | } | |
3190 | ||
3191 | meta.canceled = true; | |
3192 | ||
3193 | /* we lost in a race. There are a few options: | |
3194 | * - existing object was rewritten (ECANCELED) | |
3195 | * - non existing object was created (EEXIST) | |
3196 | * - object was removed (ENOENT) | |
3197 | * should treat it as a success | |
3198 | */ | |
3199 | if (meta.if_match == NULL && meta.if_nomatch == NULL) { | |
3200 | if (r == -ECANCELED || r == -ENOENT || r == -EEXIST) { | |
3201 | r = 0; | |
3202 | } | |
3203 | } else { | |
3204 | if (meta.if_match != NULL) { | |
3205 | // only overwrite existing object | |
3206 | if (strcmp(meta.if_match, "*") == 0) { | |
3207 | if (r == -ENOENT) { | |
3208 | r = -ERR_PRECONDITION_FAILED; | |
3209 | } else if (r == -ECANCELED) { | |
3210 | r = 0; | |
3211 | } | |
3212 | } | |
3213 | } | |
3214 | ||
3215 | if (meta.if_nomatch != NULL) { | |
3216 | // only create a new object | |
3217 | if (strcmp(meta.if_nomatch, "*") == 0) { | |
3218 | if (r == -EEXIST) { | |
3219 | r = -ERR_PRECONDITION_FAILED; | |
3220 | } else if (r == -ENOENT) { | |
3221 | r = 0; | |
3222 | } | |
3223 | } | |
3224 | } | |
3225 | } | |
3226 | ||
3227 | return r; | |
3228 | } | |
3229 | ||
3230 | int RGWRados::Object::Write::write_meta(uint64_t size, uint64_t accounted_size, | |
9f95a23c | 3231 | map<string, bufferlist>& attrs, optional_yield y) |
7c673cae FG |
3232 | { |
3233 | RGWBucketInfo& bucket_info = target->get_bucket_info(); | |
3234 | ||
3235 | RGWRados::Bucket bop(target->get_store(), bucket_info); | |
3236 | RGWRados::Bucket::UpdateIndex index_op(&bop, target->get_obj()); | |
31f18b77 FG |
3237 | index_op.set_zones_trace(meta.zones_trace); |
3238 | ||
7c673cae FG |
3239 | bool assume_noent = (meta.if_match == NULL && meta.if_nomatch == NULL); |
3240 | int r; | |
3241 | if (assume_noent) { | |
9f95a23c | 3242 | r = _do_write_meta(size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op, y); |
7c673cae FG |
3243 | if (r == -EEXIST) { |
3244 | assume_noent = false; | |
3245 | } | |
3246 | } | |
3247 | if (!assume_noent) { | |
9f95a23c | 3248 | r = _do_write_meta(size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op, y); |
7c673cae FG |
3249 | } |
3250 | return r; | |
3251 | } | |
3252 | ||
11fdf7f2 | 3253 | class RGWRadosPutObj : public RGWHTTPStreamRWRequest::ReceiveCB |
7c673cae FG |
3254 | { |
3255 | CephContext* cct; | |
3256 | rgw_obj obj; | |
11fdf7f2 | 3257 | rgw::putobj::DataProcessor *filter; |
7c673cae | 3258 | boost::optional<RGWPutObj_Compress>& compressor; |
11fdf7f2 | 3259 | boost::optional<rgw::putobj::ChunkProcessor> buffering; |
7c673cae | 3260 | CompressorRef& plugin; |
11fdf7f2 | 3261 | rgw::putobj::ObjectProcessor *processor; |
7c673cae FG |
3262 | void (*progress_cb)(off_t, void *); |
3263 | void *progress_data; | |
3264 | bufferlist extra_data_bl; | |
11fdf7f2 TL |
3265 | uint64_t extra_data_left{0}; |
3266 | bool need_to_process_attrs{true}; | |
3267 | uint64_t data_len{0}; | |
7c673cae | 3268 | map<string, bufferlist> src_attrs; |
11fdf7f2 TL |
3269 | uint64_t ofs{0}; |
3270 | uint64_t lofs{0}; /* logical ofs */ | |
9f95a23c | 3271 | std::function<int(map<string, bufferlist>&)> attrs_handler; |
7c673cae FG |
3272 | public: |
3273 | RGWRadosPutObj(CephContext* cct, | |
3274 | CompressorRef& plugin, | |
3275 | boost::optional<RGWPutObj_Compress>& compressor, | |
11fdf7f2 | 3276 | rgw::putobj::ObjectProcessor *p, |
7c673cae | 3277 | void (*_progress_cb)(off_t, void *), |
11fdf7f2 | 3278 | void *_progress_data, |
9f95a23c | 3279 | std::function<int(map<string, bufferlist>&)> _attrs_handler) : |
7c673cae FG |
3280 | cct(cct), |
3281 | filter(p), | |
3282 | compressor(compressor), | |
3283 | plugin(plugin), | |
3284 | processor(p), | |
7c673cae FG |
3285 | progress_cb(_progress_cb), |
3286 | progress_data(_progress_data), | |
11fdf7f2 | 3287 | attrs_handler(_attrs_handler) {} |
7c673cae FG |
3288 | |
3289 | int process_attrs(void) { | |
3290 | if (extra_data_bl.length()) { | |
3291 | JSONParser jp; | |
3292 | if (!jp.parse(extra_data_bl.c_str(), extra_data_bl.length())) { | |
3293 | ldout(cct, 0) << "failed to parse response extra data. len=" << extra_data_bl.length() << " data=" << extra_data_bl.c_str() << dendl; | |
3294 | return -EIO; | |
3295 | } | |
3296 | ||
3297 | JSONDecoder::decode_json("attrs", src_attrs, &jp); | |
3298 | ||
3299 | src_attrs.erase(RGW_ATTR_COMPRESSION); | |
3300 | src_attrs.erase(RGW_ATTR_MANIFEST); // not interested in original object layout | |
a8e16298 TL |
3301 | |
3302 | // filter out olh attributes | |
3303 | auto iter = src_attrs.lower_bound(RGW_ATTR_OLH_PREFIX); | |
3304 | while (iter != src_attrs.end()) { | |
3305 | if (!boost::algorithm::starts_with(iter->first, RGW_ATTR_OLH_PREFIX)) { | |
3306 | break; | |
3307 | } | |
3308 | iter = src_attrs.erase(iter); | |
3309 | } | |
7c673cae FG |
3310 | } |
3311 | ||
11fdf7f2 TL |
3312 | int ret = attrs_handler(src_attrs); |
3313 | if (ret < 0) { | |
3314 | return ret; | |
3315 | } | |
3316 | ||
7c673cae FG |
3317 | if (plugin && src_attrs.find(RGW_ATTR_CRYPT_MODE) == src_attrs.end()) { |
3318 | //do not compress if object is encrypted | |
3319 | compressor = boost::in_place(cct, plugin, filter); | |
11fdf7f2 TL |
3320 | // add a filter that buffers data so we don't try to compress tiny blocks. |
3321 | // libcurl reads in 16k at a time, and we need at least 64k to get a good | |
3322 | // compression ratio | |
28e407b8 AA |
3323 | constexpr unsigned buffer_size = 512 * 1024; |
3324 | buffering = boost::in_place(&*compressor, buffer_size); | |
3325 | filter = &*buffering; | |
7c673cae | 3326 | } |
11fdf7f2 TL |
3327 | |
3328 | need_to_process_attrs = false; | |
3329 | ||
7c673cae FG |
3330 | return 0; |
3331 | } | |
3332 | ||
11fdf7f2 | 3333 | int handle_data(bufferlist& bl, bool *pause) override { |
7c673cae | 3334 | if (progress_cb) { |
11fdf7f2 | 3335 | progress_cb(data_len, progress_data); |
7c673cae | 3336 | } |
b32b8144 | 3337 | if (extra_data_left) { |
11fdf7f2 | 3338 | uint64_t extra_len = bl.length(); |
b32b8144 FG |
3339 | if (extra_len > extra_data_left) |
3340 | extra_len = extra_data_left; | |
7c673cae FG |
3341 | |
3342 | bufferlist extra; | |
3343 | bl.splice(0, extra_len, &extra); | |
3344 | extra_data_bl.append(extra); | |
3345 | ||
b32b8144 FG |
3346 | extra_data_left -= extra_len; |
3347 | if (extra_data_left == 0) { | |
7c673cae FG |
3348 | int res = process_attrs(); |
3349 | if (res < 0) | |
3350 | return res; | |
3351 | } | |
11fdf7f2 | 3352 | ofs += extra_len; |
7c673cae FG |
3353 | if (bl.length() == 0) { |
3354 | return 0; | |
3355 | } | |
3356 | } | |
11fdf7f2 TL |
3357 | if (need_to_process_attrs) { |
3358 | /* need to call process_attrs() even if we don't get any attrs, | |
3359 | * need it to call attrs_handler(). | |
3360 | */ | |
3361 | int res = process_attrs(); | |
3362 | if (res < 0) { | |
3363 | return res; | |
3364 | } | |
3365 | } | |
7c673cae | 3366 | |
11fdf7f2 | 3367 | ceph_assert(uint64_t(ofs) >= extra_data_len); |
7c673cae | 3368 | |
11fdf7f2 TL |
3369 | uint64_t size = bl.length(); |
3370 | ofs += size; | |
7c673cae | 3371 | |
11fdf7f2 TL |
3372 | const uint64_t lofs = data_len; |
3373 | data_len += size; | |
7c673cae | 3374 | |
11fdf7f2 | 3375 | return filter->process(std::move(bl), lofs); |
7c673cae FG |
3376 | } |
3377 | ||
28e407b8 | 3378 | int flush() { |
11fdf7f2 | 3379 | return filter->process({}, data_len); |
28e407b8 AA |
3380 | } |
3381 | ||
7c673cae FG |
3382 | bufferlist& get_extra_data() { return extra_data_bl; } |
3383 | ||
3384 | map<string, bufferlist>& get_attrs() { return src_attrs; } | |
3385 | ||
3386 | void set_extra_data_len(uint64_t len) override { | |
b32b8144 | 3387 | extra_data_left = len; |
11fdf7f2 | 3388 | RGWHTTPStreamRWRequest::ReceiveCB::set_extra_data_len(len); |
7c673cae FG |
3389 | } |
3390 | ||
3391 | uint64_t get_data_len() { | |
3392 | return data_len; | |
3393 | } | |
7c673cae FG |
3394 | }; |
3395 | ||
3396 | /* | |
3397 | * prepare attrset depending on attrs_mod. | |
3398 | */ | |
3399 | static void set_copy_attrs(map<string, bufferlist>& src_attrs, | |
3400 | map<string, bufferlist>& attrs, | |
3401 | RGWRados::AttrsMod attrs_mod) | |
3402 | { | |
3403 | switch (attrs_mod) { | |
3404 | case RGWRados::ATTRSMOD_NONE: | |
3405 | attrs = src_attrs; | |
3406 | break; | |
3407 | case RGWRados::ATTRSMOD_REPLACE: | |
3408 | if (!attrs[RGW_ATTR_ETAG].length()) { | |
3409 | attrs[RGW_ATTR_ETAG] = src_attrs[RGW_ATTR_ETAG]; | |
3410 | } | |
181888fb FG |
3411 | if (!attrs[RGW_ATTR_TAIL_TAG].length()) { |
3412 | auto ttiter = src_attrs.find(RGW_ATTR_TAIL_TAG); | |
3413 | if (ttiter != src_attrs.end()) { | |
3414 | attrs[RGW_ATTR_TAIL_TAG] = src_attrs[RGW_ATTR_TAIL_TAG]; | |
3415 | } | |
3416 | } | |
7c673cae FG |
3417 | break; |
3418 | case RGWRados::ATTRSMOD_MERGE: | |
3419 | for (map<string, bufferlist>::iterator it = src_attrs.begin(); it != src_attrs.end(); ++it) { | |
3420 | if (attrs.find(it->first) == attrs.end()) { | |
3421 | attrs[it->first] = it->second; | |
3422 | } | |
3423 | } | |
3424 | break; | |
3425 | } | |
3426 | } | |
3427 | ||
9f95a23c | 3428 | int RGWRados::rewrite_obj(RGWBucketInfo& dest_bucket_info, const rgw_obj& obj, const DoutPrefixProvider *dpp, optional_yield y) |
7c673cae FG |
3429 | { |
3430 | map<string, bufferlist> attrset; | |
3431 | ||
3432 | real_time mtime; | |
3433 | uint64_t obj_size; | |
9f95a23c | 3434 | RGWObjectCtx rctx(this->store); |
7c673cae FG |
3435 | |
3436 | RGWRados::Object op_target(this, dest_bucket_info, rctx, obj); | |
3437 | RGWRados::Object::Read read_op(&op_target); | |
3438 | ||
3439 | read_op.params.attrs = &attrset; | |
3440 | read_op.params.lastmod = &mtime; | |
3441 | read_op.params.obj_size = &obj_size; | |
3442 | ||
9f95a23c | 3443 | int ret = read_op.prepare(y); |
7c673cae FG |
3444 | if (ret < 0) |
3445 | return ret; | |
3446 | ||
3447 | attrset.erase(RGW_ATTR_ID_TAG); | |
181888fb | 3448 | attrset.erase(RGW_ATTR_TAIL_TAG); |
7c673cae | 3449 | |
11fdf7f2 TL |
3450 | return copy_obj_data(rctx, dest_bucket_info, dest_bucket_info.placement_rule, |
3451 | read_op, obj_size - 1, obj, NULL, mtime, attrset, | |
9f95a23c | 3452 | 0, real_time(), NULL, dpp, y); |
7c673cae FG |
3453 | } |
3454 | ||
3455 | struct obj_time_weight { | |
3456 | real_time mtime; | |
3457 | uint32_t zone_short_id; | |
3458 | uint64_t pg_ver; | |
3459 | bool high_precision; | |
3460 | ||
3461 | obj_time_weight() : zone_short_id(0), pg_ver(0), high_precision(false) {} | |
3462 | ||
3463 | bool compare_low_precision(const obj_time_weight& rhs) { | |
3464 | struct timespec l = ceph::real_clock::to_timespec(mtime); | |
3465 | struct timespec r = ceph::real_clock::to_timespec(rhs.mtime); | |
3466 | l.tv_nsec = 0; | |
3467 | r.tv_nsec = 0; | |
3468 | if (l > r) { | |
3469 | return false; | |
3470 | } | |
3471 | if (l < r) { | |
3472 | return true; | |
3473 | } | |
11fdf7f2 TL |
3474 | if (!zone_short_id || !rhs.zone_short_id) { |
3475 | /* don't compare zone ids, if one wasn't provided */ | |
3476 | return false; | |
3477 | } | |
7c673cae FG |
3478 | if (zone_short_id != rhs.zone_short_id) { |
3479 | return (zone_short_id < rhs.zone_short_id); | |
3480 | } | |
3481 | return (pg_ver < rhs.pg_ver); | |
3482 | ||
3483 | } | |
3484 | ||
3485 | bool operator<(const obj_time_weight& rhs) { | |
3486 | if (!high_precision || !rhs.high_precision) { | |
3487 | return compare_low_precision(rhs); | |
3488 | } | |
3489 | if (mtime > rhs.mtime) { | |
3490 | return false; | |
3491 | } | |
3492 | if (mtime < rhs.mtime) { | |
3493 | return true; | |
3494 | } | |
11fdf7f2 TL |
3495 | if (!zone_short_id || !rhs.zone_short_id) { |
3496 | /* don't compare zone ids, if one wasn't provided */ | |
3497 | return false; | |
3498 | } | |
7c673cae FG |
3499 | if (zone_short_id != rhs.zone_short_id) { |
3500 | return (zone_short_id < rhs.zone_short_id); | |
3501 | } | |
3502 | return (pg_ver < rhs.pg_ver); | |
3503 | } | |
3504 | ||
3505 | void init(const real_time& _mtime, uint32_t _short_id, uint64_t _pg_ver) { | |
3506 | mtime = _mtime; | |
3507 | zone_short_id = _short_id; | |
3508 | pg_ver = _pg_ver; | |
3509 | } | |
3510 | ||
3511 | void init(RGWObjState *state) { | |
3512 | mtime = state->mtime; | |
3513 | zone_short_id = state->zone_short_id; | |
3514 | pg_ver = state->pg_ver; | |
3515 | } | |
3516 | }; | |
3517 | ||
3518 | inline ostream& operator<<(ostream& out, const obj_time_weight &o) { | |
3519 | out << o.mtime; | |
3520 | ||
3521 | if (o.zone_short_id != 0 || o.pg_ver != 0) { | |
3522 | out << "[zid=" << o.zone_short_id << ", pgv=" << o.pg_ver << "]"; | |
3523 | } | |
3524 | ||
3525 | return out; | |
3526 | } | |
3527 | ||
11fdf7f2 | 3528 | class RGWGetExtraDataCB : public RGWHTTPStreamRWRequest::ReceiveCB { |
7c673cae FG |
3529 | bufferlist extra_data; |
3530 | public: | |
3531 | RGWGetExtraDataCB() {} | |
11fdf7f2 TL |
3532 | int handle_data(bufferlist& bl, bool *pause) override { |
3533 | int bl_len = (int)bl.length(); | |
7c673cae FG |
3534 | if (extra_data.length() < extra_data_len) { |
3535 | off_t max = extra_data_len - extra_data.length(); | |
3536 | if (max > bl_len) { | |
3537 | max = bl_len; | |
3538 | } | |
3539 | bl.splice(0, max, &extra_data); | |
3540 | } | |
3541 | return bl_len; | |
3542 | } | |
3543 | ||
3544 | bufferlist& get_extra_data() { | |
3545 | return extra_data; | |
3546 | } | |
3547 | }; | |
3548 | ||
3549 | int RGWRados::stat_remote_obj(RGWObjectCtx& obj_ctx, | |
3550 | const rgw_user& user_id, | |
7c673cae | 3551 | req_info *info, |
9f95a23c | 3552 | const rgw_zone_id& source_zone, |
7c673cae | 3553 | rgw_obj& src_obj, |
9f95a23c | 3554 | const RGWBucketInfo *src_bucket_info, |
7c673cae FG |
3555 | real_time *src_mtime, |
3556 | uint64_t *psize, | |
3557 | const real_time *mod_ptr, | |
3558 | const real_time *unmod_ptr, | |
3559 | bool high_precision_time, | |
3560 | const char *if_match, | |
3561 | const char *if_nomatch, | |
3562 | map<string, bufferlist> *pattrs, | |
11fdf7f2 | 3563 | map<string, string> *pheaders, |
7c673cae FG |
3564 | string *version_id, |
3565 | string *ptag, | |
3566 | string *petag) | |
3567 | { | |
3568 | /* source is in a different zonegroup, copy from there */ | |
3569 | ||
3570 | RGWRESTStreamRWRequest *in_stream_req; | |
3571 | string tag; | |
3572 | map<string, bufferlist> src_attrs; | |
3573 | append_rand_alpha(cct, tag, tag, 32); | |
3574 | obj_time_weight set_mtime_weight; | |
3575 | set_mtime_weight.high_precision = high_precision_time; | |
3576 | ||
3577 | RGWRESTConn *conn; | |
3578 | if (source_zone.empty()) { | |
9f95a23c | 3579 | if (!src_bucket_info || src_bucket_info->zonegroup.empty()) { |
7c673cae | 3580 | /* source is in the master zonegroup */ |
11fdf7f2 | 3581 | conn = svc.zone->get_master_conn(); |
7c673cae | 3582 | } else { |
11fdf7f2 | 3583 | auto& zonegroup_conn_map = svc.zone->get_zonegroup_conn_map(); |
9f95a23c | 3584 | map<string, RGWRESTConn *>::iterator iter = zonegroup_conn_map.find(src_bucket_info->zonegroup); |
7c673cae FG |
3585 | if (iter == zonegroup_conn_map.end()) { |
3586 | ldout(cct, 0) << "could not find zonegroup connection to zonegroup: " << source_zone << dendl; | |
3587 | return -ENOENT; | |
3588 | } | |
3589 | conn = iter->second; | |
3590 | } | |
3591 | } else { | |
11fdf7f2 | 3592 | auto& zone_conn_map = svc.zone->get_zone_conn_map(); |
9f95a23c | 3593 | auto iter = zone_conn_map.find(source_zone); |
7c673cae FG |
3594 | if (iter == zone_conn_map.end()) { |
3595 | ldout(cct, 0) << "could not find zone connection to zone: " << source_zone << dendl; | |
3596 | return -ENOENT; | |
3597 | } | |
3598 | conn = iter->second; | |
3599 | } | |
3600 | ||
3601 | RGWGetExtraDataCB cb; | |
7c673cae FG |
3602 | map<string, string> req_headers; |
3603 | real_time set_mtime; | |
3604 | ||
3605 | const real_time *pmod = mod_ptr; | |
3606 | ||
3607 | obj_time_weight dest_mtime_weight; | |
3608 | ||
181888fb FG |
3609 | constexpr bool prepend_meta = true; |
3610 | constexpr bool get_op = true; | |
3611 | constexpr bool rgwx_stat = true; | |
3612 | constexpr bool sync_manifest = true; | |
3613 | constexpr bool skip_decrypt = true; | |
7c673cae FG |
3614 | int ret = conn->get_obj(user_id, info, src_obj, pmod, unmod_ptr, |
3615 | dest_mtime_weight.zone_short_id, dest_mtime_weight.pg_ver, | |
181888fb | 3616 | prepend_meta, get_op, rgwx_stat, |
11fdf7f2 TL |
3617 | sync_manifest, skip_decrypt, |
3618 | true, &cb, &in_stream_req); | |
7c673cae FG |
3619 | if (ret < 0) { |
3620 | return ret; | |
3621 | } | |
3622 | ||
11fdf7f2 | 3623 | ret = conn->complete_request(in_stream_req, nullptr, &set_mtime, psize, nullptr, pheaders); |
7c673cae FG |
3624 | if (ret < 0) { |
3625 | return ret; | |
3626 | } | |
3627 | ||
3628 | bufferlist& extra_data_bl = cb.get_extra_data(); | |
3629 | if (extra_data_bl.length()) { | |
3630 | JSONParser jp; | |
3631 | if (!jp.parse(extra_data_bl.c_str(), extra_data_bl.length())) { | |
3632 | ldout(cct, 0) << "failed to parse response extra data. len=" << extra_data_bl.length() << " data=" << extra_data_bl.c_str() << dendl; | |
3633 | return -EIO; | |
3634 | } | |
3635 | ||
3636 | JSONDecoder::decode_json("attrs", src_attrs, &jp); | |
3637 | ||
3638 | src_attrs.erase(RGW_ATTR_MANIFEST); // not interested in original object layout | |
3639 | } | |
3640 | ||
3641 | if (src_mtime) { | |
3642 | *src_mtime = set_mtime; | |
3643 | } | |
3644 | ||
3645 | if (petag) { | |
3646 | map<string, bufferlist>::iterator iter = src_attrs.find(RGW_ATTR_ETAG); | |
3647 | if (iter != src_attrs.end()) { | |
3648 | bufferlist& etagbl = iter->second; | |
3649 | *petag = etagbl.to_str(); | |
11fdf7f2 TL |
3650 | while (petag->size() > 0 && (*petag)[petag->size() - 1] == '\0') { |
3651 | *petag = petag->substr(0, petag->size() - 1); | |
3652 | } | |
7c673cae FG |
3653 | } |
3654 | } | |
3655 | ||
3656 | if (pattrs) { | |
11fdf7f2 | 3657 | *pattrs = std::move(src_attrs); |
7c673cae FG |
3658 | } |
3659 | ||
3660 | return 0; | |
3661 | } | |
3662 | ||
9f95a23c TL |
3663 | int RGWFetchObjFilter_Default::filter(CephContext *cct, |
3664 | const rgw_obj_key& source_key, | |
3665 | const RGWBucketInfo& dest_bucket_info, | |
3666 | std::optional<rgw_placement_rule> dest_placement_rule, | |
3667 | const map<string, bufferlist>& obj_attrs, | |
3668 | std::optional<rgw_user> *poverride_owner, | |
3669 | const rgw_placement_rule **prule) | |
3670 | { | |
3671 | const rgw_placement_rule *ptail_rule = (dest_placement_rule ? &(*dest_placement_rule) : nullptr); | |
3672 | if (!ptail_rule) { | |
3673 | auto iter = obj_attrs.find(RGW_ATTR_STORAGE_CLASS); | |
3674 | if (iter != obj_attrs.end()) { | |
3675 | dest_rule.storage_class = iter->second.to_str(); | |
3676 | dest_rule.inherit_from(dest_bucket_info.placement_rule); | |
3677 | ptail_rule = &dest_rule; | |
3678 | } else { | |
3679 | ptail_rule = &dest_bucket_info.placement_rule; | |
3680 | } | |
3681 | } | |
3682 | *prule = ptail_rule; | |
3683 | return 0; | |
3684 | } | |
3685 | ||
7c673cae FG |
3686 | int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx, |
3687 | const rgw_user& user_id, | |
7c673cae | 3688 | req_info *info, |
9f95a23c | 3689 | const rgw_zone_id& source_zone, |
11fdf7f2 TL |
3690 | const rgw_obj& dest_obj, |
3691 | const rgw_obj& src_obj, | |
9f95a23c TL |
3692 | const RGWBucketInfo& dest_bucket_info, |
3693 | const RGWBucketInfo *src_bucket_info, | |
11fdf7f2 | 3694 | std::optional<rgw_placement_rule> dest_placement_rule, |
7c673cae FG |
3695 | real_time *src_mtime, |
3696 | real_time *mtime, | |
3697 | const real_time *mod_ptr, | |
3698 | const real_time *unmod_ptr, | |
3699 | bool high_precision_time, | |
3700 | const char *if_match, | |
3701 | const char *if_nomatch, | |
3702 | AttrsMod attrs_mod, | |
3703 | bool copy_if_newer, | |
3704 | map<string, bufferlist>& attrs, | |
3705 | RGWObjCategory category, | |
11fdf7f2 | 3706 | std::optional<uint64_t> olh_epoch, |
7c673cae | 3707 | real_time delete_at, |
7c673cae | 3708 | string *ptag, |
11fdf7f2 | 3709 | string *petag, |
7c673cae | 3710 | void (*progress_cb)(off_t, void *), |
31f18b77 | 3711 | void *progress_data, |
9f95a23c TL |
3712 | const DoutPrefixProvider *dpp, |
3713 | RGWFetchObjFilter *filter, | |
81eedcae TL |
3714 | rgw_zone_set *zones_trace, |
3715 | std::optional<uint64_t>* bytes_transferred) | |
7c673cae FG |
3716 | { |
3717 | /* source is in a different zonegroup, copy from there */ | |
3718 | ||
3719 | RGWRESTStreamRWRequest *in_stream_req; | |
3720 | string tag; | |
3721 | int i; | |
3722 | append_rand_alpha(cct, tag, tag, 32); | |
3723 | obj_time_weight set_mtime_weight; | |
3724 | set_mtime_weight.high_precision = high_precision_time; | |
11fdf7f2 | 3725 | int ret; |
7c673cae | 3726 | |
9f95a23c | 3727 | rgw::BlockingAioThrottle aio(cct->_conf->rgw_put_obj_min_window_size); |
11fdf7f2 | 3728 | using namespace rgw::putobj; |
9f95a23c TL |
3729 | AtomicObjectProcessor processor(&aio, this->store, dest_bucket_info, nullptr, user_id, |
3730 | obj_ctx, dest_obj, olh_epoch, tag, dpp, null_yield); | |
7c673cae | 3731 | RGWRESTConn *conn; |
11fdf7f2 TL |
3732 | auto& zone_conn_map = svc.zone->get_zone_conn_map(); |
3733 | auto& zonegroup_conn_map = svc.zone->get_zonegroup_conn_map(); | |
7c673cae | 3734 | if (source_zone.empty()) { |
9f95a23c | 3735 | if (!src_bucket_info || src_bucket_info->zonegroup.empty()) { |
7c673cae | 3736 | /* source is in the master zonegroup */ |
11fdf7f2 | 3737 | conn = svc.zone->get_master_conn(); |
7c673cae | 3738 | } else { |
9f95a23c | 3739 | map<string, RGWRESTConn *>::iterator iter = zonegroup_conn_map.find(src_bucket_info->zonegroup); |
7c673cae FG |
3740 | if (iter == zonegroup_conn_map.end()) { |
3741 | ldout(cct, 0) << "could not find zonegroup connection to zonegroup: " << source_zone << dendl; | |
3742 | return -ENOENT; | |
3743 | } | |
3744 | conn = iter->second; | |
3745 | } | |
3746 | } else { | |
9f95a23c | 3747 | auto iter = zone_conn_map.find(source_zone); |
7c673cae FG |
3748 | if (iter == zone_conn_map.end()) { |
3749 | ldout(cct, 0) << "could not find zone connection to zone: " << source_zone << dendl; | |
11fdf7f2 | 3750 | return -ENOENT; |
7c673cae | 3751 | } |
11fdf7f2 | 3752 | conn = iter->second; |
7c673cae FG |
3753 | } |
3754 | ||
3755 | boost::optional<RGWPutObj_Compress> compressor; | |
3756 | CompressorRef plugin; | |
3757 | ||
9f95a23c TL |
3758 | RGWFetchObjFilter_Default source_filter; |
3759 | if (!filter) { | |
3760 | filter = &source_filter; | |
3761 | } | |
3762 | ||
3763 | std::optional<rgw_user> override_owner; | |
3764 | ||
11fdf7f2 | 3765 | RGWRadosPutObj cb(cct, plugin, compressor, &processor, progress_cb, progress_data, |
9f95a23c TL |
3766 | [&](map<string, bufferlist>& obj_attrs) { |
3767 | const rgw_placement_rule *ptail_rule; | |
3768 | ||
3769 | int ret = filter->filter(cct, | |
3770 | src_obj.key, | |
3771 | dest_bucket_info, | |
3772 | dest_placement_rule, | |
3773 | obj_attrs, | |
3774 | &override_owner, | |
3775 | &ptail_rule); | |
3776 | if (ret < 0) { | |
3777 | ldout(cct, 5) << "Aborting fetch: source object filter returned ret=" << ret << dendl; | |
3778 | return ret; | |
11fdf7f2 | 3779 | } |
9f95a23c TL |
3780 | |
3781 | processor.set_tail_placement(*ptail_rule); | |
3782 | ||
11fdf7f2 TL |
3783 | const auto& compression_type = svc.zone->get_zone_params().get_compression_type(*ptail_rule); |
3784 | if (compression_type != "none") { | |
3785 | plugin = Compressor::create(cct, compression_type); | |
3786 | if (!plugin) { | |
3787 | ldout(cct, 1) << "Cannot load plugin for compression type " | |
3788 | << compression_type << dendl; | |
3789 | } | |
3790 | } | |
3791 | ||
9f95a23c | 3792 | ret = processor.prepare(null_yield); |
11fdf7f2 TL |
3793 | if (ret < 0) { |
3794 | return ret; | |
3795 | } | |
3796 | return 0; | |
3797 | }); | |
7c673cae FG |
3798 | |
3799 | string etag; | |
7c673cae | 3800 | real_time set_mtime; |
81eedcae | 3801 | uint64_t expected_size = 0; |
7c673cae FG |
3802 | |
3803 | RGWObjState *dest_state = NULL; | |
3804 | ||
3805 | const real_time *pmod = mod_ptr; | |
3806 | ||
3807 | obj_time_weight dest_mtime_weight; | |
3808 | ||
3809 | if (copy_if_newer) { | |
3810 | /* need to get mtime for destination */ | |
9f95a23c | 3811 | ret = get_obj_state(&obj_ctx, dest_bucket_info, dest_obj, &dest_state, false, null_yield); |
7c673cae FG |
3812 | if (ret < 0) |
3813 | goto set_err_state; | |
3814 | ||
3815 | if (!real_clock::is_zero(dest_state->mtime)) { | |
3816 | dest_mtime_weight.init(dest_state); | |
3817 | pmod = &dest_mtime_weight.mtime; | |
3818 | } | |
3819 | } | |
3820 | ||
181888fb FG |
3821 | static constexpr bool prepend_meta = true; |
3822 | static constexpr bool get_op = true; | |
3823 | static constexpr bool rgwx_stat = false; | |
3824 | static constexpr bool sync_manifest = true; | |
3825 | static constexpr bool skip_decrypt = true; | |
7c673cae FG |
3826 | ret = conn->get_obj(user_id, info, src_obj, pmod, unmod_ptr, |
3827 | dest_mtime_weight.zone_short_id, dest_mtime_weight.pg_ver, | |
181888fb | 3828 | prepend_meta, get_op, rgwx_stat, |
11fdf7f2 TL |
3829 | sync_manifest, skip_decrypt, |
3830 | true, | |
3831 | &cb, &in_stream_req); | |
7c673cae FG |
3832 | if (ret < 0) { |
3833 | goto set_err_state; | |
3834 | } | |
3835 | ||
81eedcae TL |
3836 | ret = conn->complete_request(in_stream_req, &etag, &set_mtime, |
3837 | &expected_size, nullptr, nullptr); | |
7c673cae FG |
3838 | if (ret < 0) { |
3839 | goto set_err_state; | |
3840 | } | |
28e407b8 AA |
3841 | ret = cb.flush(); |
3842 | if (ret < 0) { | |
3843 | goto set_err_state; | |
3844 | } | |
81eedcae TL |
3845 | if (cb.get_data_len() != expected_size) { |
3846 | ret = -EIO; | |
3847 | ldout(cct, 0) << "ERROR: object truncated during fetching, expected " | |
3848 | << expected_size << " bytes but received " << cb.get_data_len() << dendl; | |
3849 | goto set_err_state; | |
3850 | } | |
7c673cae FG |
3851 | if (compressor && compressor->is_compressed()) { |
3852 | bufferlist tmp; | |
3853 | RGWCompressionInfo cs_info; | |
3854 | cs_info.compression_type = plugin->get_type_name(); | |
3855 | cs_info.orig_size = cb.get_data_len(); | |
3856 | cs_info.blocks = move(compressor->get_compression_blocks()); | |
11fdf7f2 | 3857 | encode(cs_info, tmp); |
7c673cae FG |
3858 | cb.get_attrs()[RGW_ATTR_COMPRESSION] = tmp; |
3859 | } | |
3860 | ||
9f95a23c TL |
3861 | if (override_owner) { |
3862 | processor.set_owner(*override_owner); | |
3863 | ||
3864 | auto& obj_attrs = cb.get_attrs(); | |
3865 | ||
3866 | RGWUserInfo owner_info; | |
3867 | if (ctl.user->get_info_by_uid(*override_owner, &owner_info, null_yield) < 0) { | |
3868 | ldout(cct, 10) << "owner info does not exist" << dendl; | |
3869 | return -EINVAL; | |
3870 | } | |
3871 | ||
3872 | RGWAccessControlPolicy acl; | |
3873 | ||
3874 | auto aiter = obj_attrs.find(RGW_ATTR_ACL); | |
3875 | if (aiter == obj_attrs.end()) { | |
3876 | ldout(cct, 0) << "WARNING: " << __func__ << "(): object doesn't have ACL attribute, setting default ACLs" << dendl; | |
3877 | acl.create_default(owner_info.user_id, owner_info.display_name); | |
3878 | } else { | |
3879 | auto iter = aiter->second.cbegin(); | |
3880 | try { | |
3881 | acl.decode(iter); | |
3882 | } catch (buffer::error& err) { | |
3883 | ldout(cct, 0) << "ERROR: " << __func__ << "(): could not decode policy, caught buffer::error" << dendl; | |
3884 | return -EIO; | |
3885 | } | |
3886 | } | |
3887 | ||
3888 | ACLOwner new_owner; | |
3889 | new_owner.set_id(*override_owner); | |
3890 | new_owner.set_name(owner_info.display_name); | |
3891 | ||
3892 | acl.set_owner(new_owner); | |
3893 | ||
3894 | bufferlist bl; | |
3895 | acl.encode(bl); | |
3896 | obj_attrs[RGW_ATTR_ACL] = std::move(bl); | |
3897 | } | |
3898 | ||
7c673cae FG |
3899 | if (source_zone.empty()) { /* need to preserve expiration if copy in the same zonegroup */ |
3900 | cb.get_attrs().erase(RGW_ATTR_DELETE_AT); | |
3901 | } else { | |
3902 | map<string, bufferlist>::iterator iter = cb.get_attrs().find(RGW_ATTR_DELETE_AT); | |
3903 | if (iter != cb.get_attrs().end()) { | |
3904 | try { | |
11fdf7f2 | 3905 | decode(delete_at, iter->second); |
7c673cae FG |
3906 | } catch (buffer::error& err) { |
3907 | ldout(cct, 0) << "ERROR: failed to decode delete_at field in intra zone copy" << dendl; | |
3908 | } | |
3909 | } | |
3910 | } | |
3911 | ||
3912 | if (src_mtime) { | |
3913 | *src_mtime = set_mtime; | |
3914 | } | |
3915 | ||
3916 | if (petag) { | |
3917 | const auto iter = cb.get_attrs().find(RGW_ATTR_ETAG); | |
3918 | if (iter != cb.get_attrs().end()) { | |
11fdf7f2 | 3919 | *petag = iter->second.to_str(); |
7c673cae FG |
3920 | } |
3921 | } | |
3922 | ||
11fdf7f2 TL |
3923 | //erase the append attr |
3924 | cb.get_attrs().erase(RGW_ATTR_APPEND_PART_NUM); | |
3925 | ||
7c673cae FG |
3926 | if (source_zone.empty()) { |
3927 | set_copy_attrs(cb.get_attrs(), attrs, attrs_mod); | |
3928 | } else { | |
3929 | attrs = cb.get_attrs(); | |
3930 | } | |
3931 | ||
3932 | if (copy_if_newer) { | |
3933 | uint64_t pg_ver = 0; | |
3934 | auto i = attrs.find(RGW_ATTR_PG_VER); | |
3935 | if (i != attrs.end() && i->second.length() > 0) { | |
11fdf7f2 | 3936 | auto iter = i->second.cbegin(); |
7c673cae | 3937 | try { |
11fdf7f2 | 3938 | decode(pg_ver, iter); |
7c673cae FG |
3939 | } catch (buffer::error& err) { |
3940 | ldout(ctx(), 0) << "ERROR: failed to decode pg ver attribute, ignoring" << dendl; | |
3941 | /* non critical error */ | |
3942 | } | |
3943 | } | |
11fdf7f2 | 3944 | set_mtime_weight.init(set_mtime, svc.zone->get_zone_short_id(), pg_ver); |
7c673cae FG |
3945 | } |
3946 | ||
3947 | #define MAX_COMPLETE_RETRY 100 | |
3948 | for (i = 0; i < MAX_COMPLETE_RETRY; i++) { | |
11fdf7f2 TL |
3949 | bool canceled = false; |
3950 | ret = processor.complete(cb.get_data_len(), etag, mtime, set_mtime, | |
3951 | attrs, delete_at, nullptr, nullptr, nullptr, | |
9f95a23c | 3952 | zones_trace, &canceled, null_yield); |
7c673cae FG |
3953 | if (ret < 0) { |
3954 | goto set_err_state; | |
3955 | } | |
11fdf7f2 | 3956 | if (copy_if_newer && canceled) { |
7c673cae | 3957 | ldout(cct, 20) << "raced with another write of obj: " << dest_obj << dendl; |
11fdf7f2 | 3958 | obj_ctx.invalidate(dest_obj); /* object was overwritten */ |
9f95a23c | 3959 | ret = get_obj_state(&obj_ctx, dest_bucket_info, dest_obj, &dest_state, false, null_yield); |
7c673cae FG |
3960 | if (ret < 0) { |
3961 | ldout(cct, 0) << "ERROR: " << __func__ << ": get_err_state() returned ret=" << ret << dendl; | |
3962 | goto set_err_state; | |
3963 | } | |
3964 | dest_mtime_weight.init(dest_state); | |
3965 | dest_mtime_weight.high_precision = high_precision_time; | |
3966 | if (!dest_state->exists || | |
3967 | dest_mtime_weight < set_mtime_weight) { | |
3968 | ldout(cct, 20) << "retrying writing object mtime=" << set_mtime << " dest_state->mtime=" << dest_state->mtime << " dest_state->exists=" << dest_state->exists << dendl; | |
3969 | continue; | |
3970 | } else { | |
3971 | ldout(cct, 20) << "not retrying writing object mtime=" << set_mtime << " dest_state->mtime=" << dest_state->mtime << " dest_state->exists=" << dest_state->exists << dendl; | |
3972 | } | |
3973 | } | |
3974 | break; | |
3975 | } | |
3976 | ||
3977 | if (i == MAX_COMPLETE_RETRY) { | |
3978 | ldout(cct, 0) << "ERROR: retried object completion too many times, something is wrong!" << dendl; | |
3979 | ret = -EIO; | |
3980 | goto set_err_state; | |
3981 | } | |
3982 | ||
81eedcae TL |
3983 | if (bytes_transferred) { |
3984 | *bytes_transferred = cb.get_data_len(); | |
3985 | } | |
7c673cae FG |
3986 | return 0; |
3987 | set_err_state: | |
3988 | if (copy_if_newer && ret == -ERR_NOT_MODIFIED) { | |
91327a77 AA |
3989 | // we may have already fetched during sync of OP_ADD, but were waiting |
3990 | // for OP_LINK_OLH to call set_olh() with a real olh_epoch | |
3991 | if (olh_epoch && *olh_epoch > 0) { | |
3992 | constexpr bool log_data_change = true; | |
3993 | ret = set_olh(obj_ctx, dest_bucket_info, dest_obj, false, nullptr, | |
9f95a23c | 3994 | *olh_epoch, real_time(), false, null_yield, zones_trace, log_data_change); |
91327a77 AA |
3995 | } else { |
3996 | // we already have the latest copy | |
3997 | ret = 0; | |
3998 | } | |
7c673cae | 3999 | } |
7c673cae FG |
4000 | return ret; |
4001 | } | |
4002 | ||
4003 | ||
4004 | int RGWRados::copy_obj_to_remote_dest(RGWObjState *astate, | |
4005 | map<string, bufferlist>& src_attrs, | |
4006 | RGWRados::Object::Read& read_op, | |
4007 | const rgw_user& user_id, | |
4008 | rgw_obj& dest_obj, | |
4009 | real_time *mtime) | |
4010 | { | |
4011 | string etag; | |
4012 | ||
11fdf7f2 | 4013 | RGWRESTStreamS3PutObj *out_stream_req; |
7c673cae | 4014 | |
11fdf7f2 TL |
4015 | auto rest_master_conn = svc.zone->get_master_conn(); |
4016 | ||
4017 | int ret = rest_master_conn->put_obj_async(user_id, dest_obj, astate->size, src_attrs, true, &out_stream_req); | |
7c673cae | 4018 | if (ret < 0) { |
7c673cae FG |
4019 | return ret; |
4020 | } | |
4021 | ||
9f95a23c | 4022 | ret = read_op.iterate(0, astate->size - 1, out_stream_req->get_out_cb(), null_yield); |
224ce89b WB |
4023 | if (ret < 0) { |
4024 | delete out_stream_req; | |
7c673cae | 4025 | return ret; |
224ce89b | 4026 | } |
7c673cae FG |
4027 | |
4028 | ret = rest_master_conn->complete_request(out_stream_req, etag, mtime); | |
4029 | if (ret < 0) | |
4030 | return ret; | |
4031 | ||
4032 | return 0; | |
4033 | } | |
4034 | ||
4035 | /** | |
4036 | * Copy an object. | |
4037 | * dest_obj: the object to copy into | |
4038 | * src_obj: the object to copy from | |
4039 | * attrs: usage depends on attrs_mod parameter | |
4040 | * attrs_mod: the modification mode of the attrs, may have the following values: | |
4041 | * ATTRSMOD_NONE - the attributes of the source object will be | |
4042 | * copied without modifications, attrs parameter is ignored; | |
4043 | * ATTRSMOD_REPLACE - new object will have the attributes provided by attrs | |
4044 | * parameter, source object attributes are not copied; | |
4045 | * ATTRSMOD_MERGE - any conflicting meta keys on the source object's attributes | |
4046 | * are overwritten by values contained in attrs parameter. | |
4047 | * err: stores any errors resulting from the get of the original object | |
4048 | * Returns: 0 on success, -ERR# otherwise. | |
4049 | */ | |
4050 | int RGWRados::copy_obj(RGWObjectCtx& obj_ctx, | |
4051 | const rgw_user& user_id, | |
7c673cae | 4052 | req_info *info, |
9f95a23c | 4053 | const rgw_zone_id& source_zone, |
7c673cae FG |
4054 | rgw_obj& dest_obj, |
4055 | rgw_obj& src_obj, | |
4056 | RGWBucketInfo& dest_bucket_info, | |
4057 | RGWBucketInfo& src_bucket_info, | |
11fdf7f2 | 4058 | const rgw_placement_rule& dest_placement, |
7c673cae FG |
4059 | real_time *src_mtime, |
4060 | real_time *mtime, | |
4061 | const real_time *mod_ptr, | |
4062 | const real_time *unmod_ptr, | |
4063 | bool high_precision_time, | |
4064 | const char *if_match, | |
4065 | const char *if_nomatch, | |
4066 | AttrsMod attrs_mod, | |
4067 | bool copy_if_newer, | |
4068 | map<string, bufferlist>& attrs, | |
4069 | RGWObjCategory category, | |
4070 | uint64_t olh_epoch, | |
4071 | real_time delete_at, | |
4072 | string *version_id, | |
4073 | string *ptag, | |
11fdf7f2 | 4074 | string *petag, |
7c673cae | 4075 | void (*progress_cb)(off_t, void *), |
9f95a23c TL |
4076 | void *progress_data, |
4077 | const DoutPrefixProvider *dpp, | |
4078 | optional_yield y) | |
7c673cae FG |
4079 | { |
4080 | int ret; | |
4081 | uint64_t obj_size; | |
4082 | rgw_obj shadow_obj = dest_obj; | |
4083 | string shadow_oid; | |
4084 | ||
4085 | bool remote_src; | |
4086 | bool remote_dest; | |
4087 | ||
4088 | append_rand_alpha(cct, dest_obj.get_oid(), shadow_oid, 32); | |
4089 | shadow_obj.init_ns(dest_obj.bucket, shadow_oid, shadow_ns); | |
4090 | ||
11fdf7f2 TL |
4091 | auto& zonegroup = svc.zone->get_zonegroup(); |
4092 | ||
4093 | remote_dest = !zonegroup.equals(dest_bucket_info.zonegroup); | |
4094 | remote_src = !zonegroup.equals(src_bucket_info.zonegroup); | |
7c673cae FG |
4095 | |
4096 | if (remote_src && remote_dest) { | |
9f95a23c | 4097 | ldpp_dout(dpp, 0) << "ERROR: can't copy object when both src and dest buckets are remote" << dendl; |
7c673cae FG |
4098 | return -EINVAL; |
4099 | } | |
4100 | ||
9f95a23c | 4101 | ldpp_dout(dpp, 5) << "Copy object " << src_obj.bucket << ":" << src_obj.get_oid() << " => " << dest_obj.bucket << ":" << dest_obj.get_oid() << dendl; |
7c673cae FG |
4102 | |
4103 | if (remote_src || !source_zone.empty()) { | |
11fdf7f2 | 4104 | return fetch_remote_obj(obj_ctx, user_id, info, source_zone, |
9f95a23c | 4105 | dest_obj, src_obj, dest_bucket_info, &src_bucket_info, |
11fdf7f2 | 4106 | dest_placement, src_mtime, mtime, mod_ptr, |
7c673cae FG |
4107 | unmod_ptr, high_precision_time, |
4108 | if_match, if_nomatch, attrs_mod, copy_if_newer, attrs, category, | |
9f95a23c TL |
4109 | olh_epoch, delete_at, ptag, petag, progress_cb, progress_data, dpp, |
4110 | nullptr /* filter */); | |
7c673cae FG |
4111 | } |
4112 | ||
4113 | map<string, bufferlist> src_attrs; | |
4114 | RGWRados::Object src_op_target(this, src_bucket_info, obj_ctx, src_obj); | |
4115 | RGWRados::Object::Read read_op(&src_op_target); | |
4116 | ||
4117 | read_op.conds.mod_ptr = mod_ptr; | |
4118 | read_op.conds.unmod_ptr = unmod_ptr; | |
4119 | read_op.conds.high_precision_time = high_precision_time; | |
4120 | read_op.conds.if_match = if_match; | |
4121 | read_op.conds.if_nomatch = if_nomatch; | |
4122 | read_op.params.attrs = &src_attrs; | |
4123 | read_op.params.lastmod = src_mtime; | |
4124 | read_op.params.obj_size = &obj_size; | |
7c673cae | 4125 | |
9f95a23c | 4126 | ret = read_op.prepare(y); |
7c673cae FG |
4127 | if (ret < 0) { |
4128 | return ret; | |
4129 | } | |
94b18763 FG |
4130 | if (src_attrs.count(RGW_ATTR_CRYPT_MODE)) { |
4131 | // Current implementation does not follow S3 spec and even | |
4132 | // may result in data corruption silently when copying | |
4133 | // multipart objects acorss pools. So reject COPY operations | |
4134 | //on encrypted objects before it is fully functional. | |
9f95a23c | 4135 | ldpp_dout(dpp, 0) << "ERROR: copy op for encrypted object " << src_obj |
94b18763 FG |
4136 | << " has not been implemented." << dendl; |
4137 | return -ERR_NOT_IMPLEMENTED; | |
4138 | } | |
7c673cae FG |
4139 | |
4140 | src_attrs[RGW_ATTR_ACL] = attrs[RGW_ATTR_ACL]; | |
4141 | src_attrs.erase(RGW_ATTR_DELETE_AT); | |
4142 | ||
4143 | set_copy_attrs(src_attrs, attrs, attrs_mod); | |
4144 | attrs.erase(RGW_ATTR_ID_TAG); | |
4145 | attrs.erase(RGW_ATTR_PG_VER); | |
4146 | attrs.erase(RGW_ATTR_SOURCE_ZONE); | |
4147 | map<string, bufferlist>::iterator cmp = src_attrs.find(RGW_ATTR_COMPRESSION); | |
4148 | if (cmp != src_attrs.end()) | |
4149 | attrs[RGW_ATTR_COMPRESSION] = cmp->second; | |
4150 | ||
4151 | RGWObjManifest manifest; | |
4152 | RGWObjState *astate = NULL; | |
4153 | ||
9f95a23c | 4154 | ret = get_obj_state(&obj_ctx, src_bucket_info, src_obj, &astate, y); |
7c673cae FG |
4155 | if (ret < 0) { |
4156 | return ret; | |
4157 | } | |
4158 | ||
4159 | vector<rgw_raw_obj> ref_objs; | |
4160 | ||
4161 | if (remote_dest) { | |
4162 | /* dest is in a different zonegroup, copy it there */ | |
4163 | return copy_obj_to_remote_dest(astate, attrs, read_op, user_id, dest_obj, mtime); | |
4164 | } | |
4165 | uint64_t max_chunk_size; | |
4166 | ||
4167 | ret = get_max_chunk_size(dest_bucket_info.placement_rule, dest_obj, &max_chunk_size); | |
4168 | if (ret < 0) { | |
9f95a23c | 4169 | ldpp_dout(dpp, 0) << "ERROR: failed to get max_chunk_size() for bucket " << dest_obj.bucket << dendl; |
7c673cae FG |
4170 | return ret; |
4171 | } | |
4172 | ||
4173 | rgw_pool src_pool; | |
4174 | rgw_pool dest_pool; | |
11fdf7f2 TL |
4175 | |
4176 | const rgw_placement_rule *src_rule{nullptr}; | |
4177 | ||
9f95a23c TL |
4178 | if (astate->manifest) { |
4179 | src_rule = &astate->manifest->get_tail_placement().placement_rule; | |
4180 | ldpp_dout(dpp, 20) << __func__ << "(): manifest src_rule=" << src_rule->to_str() << dendl; | |
11fdf7f2 TL |
4181 | } |
4182 | ||
4183 | if (!src_rule || src_rule->empty()) { | |
4184 | src_rule = &src_bucket_info.placement_rule; | |
4185 | } | |
4186 | ||
4187 | if (!get_obj_data_pool(*src_rule, src_obj, &src_pool)) { | |
9f95a23c | 4188 | ldpp_dout(dpp, 0) << "ERROR: failed to locate data pool for " << src_obj << dendl; |
7c673cae FG |
4189 | return -EIO; |
4190 | } | |
11fdf7f2 TL |
4191 | |
4192 | if (!get_obj_data_pool(dest_placement, dest_obj, &dest_pool)) { | |
9f95a23c | 4193 | ldpp_dout(dpp, 0) << "ERROR: failed to locate data pool for " << dest_obj << dendl; |
7c673cae FG |
4194 | return -EIO; |
4195 | } | |
4196 | ||
9f95a23c | 4197 | ldpp_dout(dpp, 20) << __func__ << "(): src_rule=" << src_rule->to_str() << " src_pool=" << src_pool |
11fdf7f2 TL |
4198 | << " dest_rule=" << dest_placement.to_str() << " dest_pool=" << dest_pool << dendl; |
4199 | ||
9f95a23c | 4200 | bool copy_data = (!astate->manifest) || |
11fdf7f2 TL |
4201 | (*src_rule != dest_placement) || |
4202 | (src_pool != dest_pool); | |
7c673cae | 4203 | |
7c673cae | 4204 | bool copy_first = false; |
9f95a23c TL |
4205 | if (astate->manifest) { |
4206 | if (!astate->manifest->has_tail()) { | |
7c673cae FG |
4207 | copy_data = true; |
4208 | } else { | |
9f95a23c | 4209 | uint64_t head_size = astate->manifest->get_head_size(); |
7c673cae FG |
4210 | |
4211 | if (head_size > 0) { | |
4212 | if (head_size > max_chunk_size) { | |
4213 | copy_data = true; | |
4214 | } else { | |
4215 | copy_first = true; | |
4216 | } | |
4217 | } | |
4218 | } | |
4219 | } | |
4220 | ||
4221 | if (petag) { | |
4222 | const auto iter = attrs.find(RGW_ATTR_ETAG); | |
4223 | if (iter != attrs.end()) { | |
11fdf7f2 | 4224 | *petag = iter->second.to_str(); |
7c673cae FG |
4225 | } |
4226 | } | |
4227 | ||
4228 | if (copy_data) { /* refcounting tail wouldn't work here, just copy the data */ | |
11fdf7f2 TL |
4229 | attrs.erase(RGW_ATTR_TAIL_TAG); |
4230 | return copy_obj_data(obj_ctx, dest_bucket_info, dest_placement, read_op, obj_size - 1, dest_obj, | |
9f95a23c | 4231 | mtime, real_time(), attrs, olh_epoch, delete_at, petag, dpp, y); |
7c673cae FG |
4232 | } |
4233 | ||
9f95a23c | 4234 | RGWObjManifest::obj_iterator miter = astate->manifest->obj_begin(); |
7c673cae FG |
4235 | |
4236 | if (copy_first) { // we need to copy first chunk, not increase refcount | |
4237 | ++miter; | |
4238 | } | |
4239 | ||
4240 | rgw_rados_ref ref; | |
4241 | ret = get_raw_obj_ref(miter.get_location().get_raw_obj(this), &ref); | |
4242 | if (ret < 0) { | |
4243 | return ret; | |
4244 | } | |
4245 | ||
7c673cae FG |
4246 | bufferlist first_chunk; |
4247 | ||
4248 | bool copy_itself = (dest_obj == src_obj); | |
4249 | RGWObjManifest *pmanifest; | |
9f95a23c | 4250 | ldpp_dout(dpp, 20) << "dest_obj=" << dest_obj << " src_obj=" << src_obj << " copy_itself=" << (int)copy_itself << dendl; |
7c673cae FG |
4251 | |
4252 | RGWRados::Object dest_op_target(this, dest_bucket_info, obj_ctx, dest_obj); | |
4253 | RGWRados::Object::Write write_op(&dest_op_target); | |
4254 | ||
4255 | string tag; | |
4256 | ||
4257 | if (ptag) { | |
4258 | tag = *ptag; | |
4259 | } | |
4260 | ||
4261 | if (tag.empty()) { | |
4262 | append_rand_alpha(cct, tag, tag, 32); | |
4263 | } | |
4264 | ||
4265 | if (!copy_itself) { | |
181888fb | 4266 | attrs.erase(RGW_ATTR_TAIL_TAG); |
9f95a23c | 4267 | manifest = *astate->manifest; |
7c673cae FG |
4268 | const rgw_bucket_placement& tail_placement = manifest.get_tail_placement(); |
4269 | if (tail_placement.bucket.name.empty()) { | |
4270 | manifest.set_tail_placement(tail_placement.placement_rule, src_obj.bucket); | |
4271 | } | |
3efd9988 | 4272 | string ref_tag; |
9f95a23c | 4273 | for (; miter != astate->manifest->obj_end(); ++miter) { |
7c673cae | 4274 | ObjectWriteOperation op; |
3efd9988 FG |
4275 | ref_tag = tag + '\0'; |
4276 | cls_refcount_get(op, ref_tag, true); | |
7c673cae | 4277 | const rgw_raw_obj& loc = miter.get_location().get_raw_obj(this); |
7c673cae | 4278 | |
9f95a23c TL |
4279 | auto& ioctx = ref.pool.ioctx(); |
4280 | ioctx.locator_set_key(loc.loc); | |
4281 | ||
4282 | ret = rgw_rados_operate(ioctx, loc.oid, &op, null_yield); | |
7c673cae FG |
4283 | if (ret < 0) { |
4284 | goto done_ret; | |
4285 | } | |
4286 | ||
4287 | ref_objs.push_back(loc); | |
4288 | } | |
4289 | ||
4290 | pmanifest = &manifest; | |
4291 | } else { | |
9f95a23c | 4292 | pmanifest = &(*astate->manifest); |
7c673cae FG |
4293 | /* don't send the object's tail for garbage collection */ |
4294 | astate->keep_tail = true; | |
4295 | } | |
4296 | ||
4297 | if (copy_first) { | |
9f95a23c | 4298 | ret = read_op.read(0, max_chunk_size, first_chunk, y); |
7c673cae FG |
4299 | if (ret < 0) { |
4300 | goto done_ret; | |
4301 | } | |
4302 | ||
4303 | pmanifest->set_head(dest_bucket_info.placement_rule, dest_obj, first_chunk.length()); | |
4304 | } else { | |
4305 | pmanifest->set_head(dest_bucket_info.placement_rule, dest_obj, 0); | |
4306 | } | |
4307 | ||
4308 | write_op.meta.data = &first_chunk; | |
4309 | write_op.meta.manifest = pmanifest; | |
4310 | write_op.meta.ptag = &tag; | |
4311 | write_op.meta.owner = dest_bucket_info.owner; | |
4312 | write_op.meta.mtime = mtime; | |
4313 | write_op.meta.flags = PUT_OBJ_CREATE; | |
4314 | write_op.meta.category = category; | |
4315 | write_op.meta.olh_epoch = olh_epoch; | |
4316 | write_op.meta.delete_at = delete_at; | |
181888fb | 4317 | write_op.meta.modify_tail = !copy_itself; |
7c673cae | 4318 | |
9f95a23c | 4319 | ret = write_op.write_meta(obj_size, astate->accounted_size, attrs, y); |
7c673cae FG |
4320 | if (ret < 0) { |
4321 | goto done_ret; | |
4322 | } | |
4323 | ||
4324 | return 0; | |
4325 | ||
4326 | done_ret: | |
4327 | if (!copy_itself) { | |
4328 | vector<rgw_raw_obj>::iterator riter; | |
4329 | ||
7c673cae | 4330 | /* rollback reference */ |
92f5a8d4 | 4331 | string ref_tag = tag + '\0'; |
7c673cae FG |
4332 | for (riter = ref_objs.begin(); riter != ref_objs.end(); ++riter) { |
4333 | ObjectWriteOperation op; | |
92f5a8d4 | 4334 | cls_refcount_put(op, ref_tag, true); |
7c673cae | 4335 | |
9f95a23c | 4336 | ref.pool.ioctx().locator_set_key(riter->loc); |
7c673cae | 4337 | |
9f95a23c | 4338 | int r = rgw_rados_operate(ref.pool.ioctx(), riter->oid, &op, null_yield); |
7c673cae | 4339 | if (r < 0) { |
9f95a23c | 4340 | ldpp_dout(dpp, 0) << "ERROR: cleanup after error failed to drop reference on obj=" << *riter << dendl; |
7c673cae FG |
4341 | } |
4342 | } | |
4343 | } | |
4344 | return ret; | |
4345 | } | |
4346 | ||
4347 | ||
4348 | int RGWRados::copy_obj_data(RGWObjectCtx& obj_ctx, | |
4349 | RGWBucketInfo& dest_bucket_info, | |
11fdf7f2 | 4350 | const rgw_placement_rule& dest_placement, |
7c673cae | 4351 | RGWRados::Object::Read& read_op, off_t end, |
11fdf7f2 | 4352 | const rgw_obj& dest_obj, |
7c673cae FG |
4353 | real_time *mtime, |
4354 | real_time set_mtime, | |
4355 | map<string, bufferlist>& attrs, | |
7c673cae FG |
4356 | uint64_t olh_epoch, |
4357 | real_time delete_at, | |
9f95a23c TL |
4358 | string *petag, |
4359 | const DoutPrefixProvider *dpp, | |
4360 | optional_yield y) | |
7c673cae | 4361 | { |
7c673cae FG |
4362 | string tag; |
4363 | append_rand_alpha(cct, tag, tag, 32); | |
4364 | ||
9f95a23c | 4365 | rgw::BlockingAioThrottle aio(cct->_conf->rgw_put_obj_min_window_size); |
11fdf7f2 | 4366 | using namespace rgw::putobj; |
9f95a23c TL |
4367 | // do not change the null_yield in the initialization of this AtomicObjectProcessor |
4368 | // it causes crashes in the ragweed tests | |
4369 | AtomicObjectProcessor processor(&aio, this->store, dest_bucket_info, &dest_placement, | |
11fdf7f2 | 4370 | dest_bucket_info.owner, obj_ctx, |
9f95a23c TL |
4371 | dest_obj, olh_epoch, tag, dpp, null_yield); |
4372 | int ret = processor.prepare(y); | |
7c673cae FG |
4373 | if (ret < 0) |
4374 | return ret; | |
4375 | ||
4376 | off_t ofs = 0; | |
4377 | ||
4378 | do { | |
4379 | bufferlist bl; | |
9f95a23c | 4380 | ret = read_op.read(ofs, end, bl, y); |
11fdf7f2 | 4381 | if (ret < 0) { |
9f95a23c | 4382 | ldpp_dout(dpp, 0) << "ERROR: fail to read object data, ret = " << ret << dendl; |
11fdf7f2 TL |
4383 | return ret; |
4384 | } | |
7c673cae FG |
4385 | |
4386 | uint64_t read_len = ret; | |
11fdf7f2 TL |
4387 | ret = processor.process(std::move(bl), ofs); |
4388 | if (ret < 0) { | |
4389 | return ret; | |
4390 | } | |
7c673cae FG |
4391 | |
4392 | ofs += read_len; | |
4393 | } while (ofs <= end); | |
4394 | ||
11fdf7f2 TL |
4395 | // flush |
4396 | ret = processor.process({}, ofs); | |
4397 | if (ret < 0) { | |
4398 | return ret; | |
4399 | } | |
4400 | ||
7c673cae FG |
4401 | string etag; |
4402 | auto iter = attrs.find(RGW_ATTR_ETAG); | |
4403 | if (iter != attrs.end()) { | |
4404 | bufferlist& bl = iter->second; | |
11fdf7f2 | 4405 | etag = bl.to_str(); |
7c673cae | 4406 | if (petag) { |
11fdf7f2 | 4407 | *petag = etag; |
7c673cae FG |
4408 | } |
4409 | } | |
4410 | ||
4411 | uint64_t accounted_size; | |
4412 | { | |
4413 | bool compressed{false}; | |
4414 | RGWCompressionInfo cs_info; | |
4415 | ret = rgw_compression_info_from_attrset(attrs, compressed, cs_info); | |
4416 | if (ret < 0) { | |
9f95a23c | 4417 | ldpp_dout(dpp, 0) << "ERROR: failed to read compression info" << dendl; |
7c673cae FG |
4418 | return ret; |
4419 | } | |
4420 | // pass original size if compressed | |
4421 | accounted_size = compressed ? cs_info.orig_size : ofs; | |
4422 | } | |
4423 | ||
11fdf7f2 | 4424 | return processor.complete(accounted_size, etag, mtime, set_mtime, attrs, delete_at, |
9f95a23c | 4425 | nullptr, nullptr, nullptr, nullptr, nullptr, y); |
7c673cae FG |
4426 | } |
4427 | ||
11fdf7f2 TL |
4428 | int RGWRados::transition_obj(RGWObjectCtx& obj_ctx, |
4429 | RGWBucketInfo& bucket_info, | |
4430 | rgw_obj& obj, | |
4431 | const rgw_placement_rule& placement_rule, | |
4432 | const real_time& mtime, | |
9f95a23c TL |
4433 | uint64_t olh_epoch, |
4434 | const DoutPrefixProvider *dpp, | |
4435 | optional_yield y) | |
7c673cae | 4436 | { |
11fdf7f2 TL |
4437 | map<string, bufferlist> attrs; |
4438 | real_time read_mtime; | |
4439 | uint64_t obj_size; | |
7c673cae | 4440 | |
9f95a23c TL |
4441 | obj_ctx.set_atomic(obj); |
4442 | ||
11fdf7f2 TL |
4443 | RGWRados::Object op_target(this, bucket_info, obj_ctx, obj); |
4444 | RGWRados::Object::Read read_op(&op_target); | |
7c673cae | 4445 | |
11fdf7f2 TL |
4446 | read_op.params.attrs = &attrs; |
4447 | read_op.params.lastmod = &read_mtime; | |
4448 | read_op.params.obj_size = &obj_size; | |
7c673cae | 4449 | |
9f95a23c | 4450 | int ret = read_op.prepare(y); |
11fdf7f2 TL |
4451 | if (ret < 0) { |
4452 | return ret; | |
7c673cae FG |
4453 | } |
4454 | ||
11fdf7f2 TL |
4455 | if (read_mtime != mtime) { |
4456 | /* raced */ | |
4457 | return -ECANCELED; | |
7c673cae FG |
4458 | } |
4459 | ||
9f95a23c TL |
4460 | attrs.erase(RGW_ATTR_ID_TAG); |
4461 | attrs.erase(RGW_ATTR_TAIL_TAG); | |
4462 | ||
11fdf7f2 TL |
4463 | ret = copy_obj_data(obj_ctx, |
4464 | bucket_info, | |
4465 | placement_rule, | |
4466 | read_op, | |
4467 | obj_size - 1, | |
4468 | obj, | |
4469 | nullptr /* pmtime */, | |
4470 | mtime, | |
4471 | attrs, | |
4472 | olh_epoch, | |
4473 | real_time(), | |
9f95a23c TL |
4474 | nullptr /* petag */, |
4475 | dpp, | |
4476 | y); | |
11fdf7f2 TL |
4477 | if (ret < 0) { |
4478 | return ret; | |
7c673cae FG |
4479 | } |
4480 | ||
11fdf7f2 | 4481 | return 0; |
7c673cae FG |
4482 | } |
4483 | ||
9f95a23c | 4484 | int RGWRados::check_bucket_empty(RGWBucketInfo& bucket_info, optional_yield y) |
7c673cae | 4485 | { |
9f95a23c TL |
4486 | constexpr uint NUM_ENTRIES = 1000u; |
4487 | ||
7c673cae FG |
4488 | rgw_obj_index_key marker; |
4489 | string prefix; | |
4490 | bool is_truncated; | |
4491 | ||
4492 | do { | |
9f95a23c TL |
4493 | std::vector<rgw_bucket_dir_entry> ent_list; |
4494 | ent_list.reserve(NUM_ENTRIES); | |
4495 | ||
1adf2230 AA |
4496 | int r = cls_bucket_list_unordered(bucket_info, |
4497 | RGW_NO_SHARD, | |
4498 | marker, | |
4499 | prefix, | |
4500 | NUM_ENTRIES, | |
4501 | true, | |
4502 | ent_list, | |
4503 | &is_truncated, | |
9f95a23c TL |
4504 | &marker, |
4505 | y); | |
4506 | if (r < 0) { | |
7c673cae | 4507 | return r; |
9f95a23c | 4508 | } |
7c673cae FG |
4509 | |
4510 | string ns; | |
1adf2230 | 4511 | for (auto const& dirent : ent_list) { |
7c673cae FG |
4512 | rgw_obj_key obj; |
4513 | ||
9f95a23c | 4514 | if (rgw_obj_key::oid_to_key_in_ns(dirent.key.name, &obj, ns)) { |
7c673cae | 4515 | return -ENOTEMPTY; |
9f95a23c | 4516 | } |
7c673cae FG |
4517 | } |
4518 | } while (is_truncated); | |
1adf2230 | 4519 | |
7c673cae FG |
4520 | return 0; |
4521 | } | |
4522 | ||
4523 | /** | |
4524 | * Delete a bucket. | |
4525 | * bucket: the name of the bucket to delete | |
4526 | * Returns 0 on success, -ERR# otherwise. | |
4527 | */ | |
9f95a23c | 4528 | int RGWRados::delete_bucket(RGWBucketInfo& bucket_info, RGWObjVersionTracker& objv_tracker, optional_yield y, bool check_empty) |
7c673cae FG |
4529 | { |
4530 | const rgw_bucket& bucket = bucket_info.bucket; | |
9f95a23c | 4531 | RGWSI_RADOS::Pool index_pool; |
7c673cae | 4532 | map<int, string> bucket_objs; |
9f95a23c | 4533 | int r = svc.bi_rados->open_bucket_index(bucket_info, std::nullopt, &index_pool, &bucket_objs, nullptr); |
7c673cae FG |
4534 | if (r < 0) |
4535 | return r; | |
4536 | ||
4537 | if (check_empty) { | |
9f95a23c | 4538 | r = check_bucket_empty(bucket_info, y); |
7c673cae FG |
4539 | if (r < 0) { |
4540 | return r; | |
4541 | } | |
4542 | } | |
9f95a23c TL |
4543 | |
4544 | bool remove_ep = true; | |
4545 | ||
4546 | if (objv_tracker.read_version.empty()) { | |
4547 | RGWBucketEntryPoint ep; | |
4548 | r = ctl.bucket->read_bucket_entrypoint_info(bucket_info.bucket, | |
4549 | &ep, | |
4550 | null_yield, | |
4551 | RGWBucketCtl::Bucket::GetParams() | |
4552 | .set_objv_tracker(&objv_tracker)); | |
4553 | if (r < 0 || | |
4554 | (!bucket_info.bucket.bucket_id.empty() && | |
4555 | ep.bucket.bucket_id != bucket_info.bucket.bucket_id)) { | |
4556 | if (r != -ENOENT) { | |
4557 | ldout(cct, 0) << "ERROR: read_bucket_entrypoint_info() bucket=" << bucket_info.bucket << " returned error: r=" << r << dendl; | |
4558 | /* we have no idea what caused the error, will not try to remove it */ | |
4559 | } | |
4560 | /* | |
4561 | * either failed to read bucket entrypoint, or it points to a different bucket instance than | |
4562 | * requested | |
4563 | */ | |
4564 | remove_ep = false; | |
4565 | } | |
4566 | } | |
4567 | ||
4568 | if (remove_ep) { | |
4569 | r = ctl.bucket->remove_bucket_entrypoint_info(bucket_info.bucket, null_yield, | |
4570 | RGWBucketCtl::Bucket::RemoveParams() | |
4571 | .set_objv_tracker(&objv_tracker)); | |
4572 | if (r < 0) | |
4573 | return r; | |
4574 | } | |
7c673cae FG |
4575 | |
4576 | /* if the bucket is not synced we can remove the meta file */ | |
11fdf7f2 | 4577 | if (!svc.zone->is_syncing_bucket_meta(bucket)) { |
7c673cae | 4578 | RGWObjVersionTracker objv_tracker; |
9f95a23c | 4579 | r = ctl.bucket->remove_bucket_instance_info(bucket, bucket_info, null_yield); |
7c673cae FG |
4580 | if (r < 0) { |
4581 | return r; | |
4582 | } | |
f64942e4 AA |
4583 | |
4584 | /* remove bucket index objects asynchronously by best effort */ | |
9f95a23c | 4585 | (void) CLSRGWIssueBucketIndexClean(index_pool.ioctx(), |
f64942e4 AA |
4586 | bucket_objs, |
4587 | cct->_conf->rgw_bucket_index_max_aio)(); | |
7c673cae | 4588 | } |
f64942e4 | 4589 | |
7c673cae FG |
4590 | return 0; |
4591 | } | |
4592 | ||
4593 | int RGWRados::set_bucket_owner(rgw_bucket& bucket, ACLOwner& owner) | |
4594 | { | |
4595 | RGWBucketInfo info; | |
4596 | map<string, bufferlist> attrs; | |
31f18b77 | 4597 | int r; |
9f95a23c TL |
4598 | auto obj_ctx = svc.sysobj->init_obj_ctx(); |
4599 | ||
31f18b77 | 4600 | if (bucket.bucket_id.empty()) { |
9f95a23c | 4601 | r = get_bucket_info(&svc, bucket.tenant, bucket.name, info, NULL, null_yield, &attrs); |
31f18b77 | 4602 | } else { |
9f95a23c | 4603 | r = get_bucket_instance_info(obj_ctx, bucket, info, nullptr, &attrs, null_yield); |
31f18b77 | 4604 | } |
7c673cae FG |
4605 | if (r < 0) { |
4606 | ldout(cct, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl; | |
4607 | return r; | |
4608 | } | |
4609 | ||
4610 | info.owner = owner.get_id(); | |
4611 | ||
4612 | r = put_bucket_instance_info(info, false, real_time(), &attrs); | |
4613 | if (r < 0) { | |
4614 | ldout(cct, 0) << "NOTICE: put_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl; | |
4615 | return r; | |
4616 | } | |
4617 | ||
4618 | return 0; | |
4619 | } | |
4620 | ||
4621 | ||
4622 | int RGWRados::set_buckets_enabled(vector<rgw_bucket>& buckets, bool enabled) | |
4623 | { | |
4624 | int ret = 0; | |
4625 | ||
4626 | vector<rgw_bucket>::iterator iter; | |
4627 | ||
4628 | for (iter = buckets.begin(); iter != buckets.end(); ++iter) { | |
4629 | rgw_bucket& bucket = *iter; | |
4630 | if (enabled) | |
4631 | ldout(cct, 20) << "enabling bucket name=" << bucket.name << dendl; | |
4632 | else | |
4633 | ldout(cct, 20) << "disabling bucket name=" << bucket.name << dendl; | |
4634 | ||
4635 | RGWBucketInfo info; | |
4636 | map<string, bufferlist> attrs; | |
9f95a23c | 4637 | int r = get_bucket_info(&svc, bucket.tenant, bucket.name, info, NULL, null_yield, &attrs); |
7c673cae FG |
4638 | if (r < 0) { |
4639 | ldout(cct, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl; | |
4640 | ret = r; | |
4641 | continue; | |
4642 | } | |
4643 | if (enabled) { | |
4644 | info.flags &= ~BUCKET_SUSPENDED; | |
4645 | } else { | |
4646 | info.flags |= BUCKET_SUSPENDED; | |
4647 | } | |
4648 | ||
4649 | r = put_bucket_instance_info(info, false, real_time(), &attrs); | |
4650 | if (r < 0) { | |
4651 | ldout(cct, 0) << "NOTICE: put_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl; | |
4652 | ret = r; | |
4653 | continue; | |
4654 | } | |
4655 | } | |
4656 | return ret; | |
4657 | } | |
4658 | ||
4659 | int RGWRados::bucket_suspended(rgw_bucket& bucket, bool *suspended) | |
4660 | { | |
4661 | RGWBucketInfo bucket_info; | |
9f95a23c | 4662 | int ret = get_bucket_info(&svc, bucket.tenant, bucket.name, bucket_info, NULL, null_yield); |
7c673cae FG |
4663 | if (ret < 0) { |
4664 | return ret; | |
4665 | } | |
4666 | ||
4667 | *suspended = ((bucket_info.flags & BUCKET_SUSPENDED) != 0); | |
4668 | return 0; | |
4669 | } | |
4670 | ||
4671 | int RGWRados::Object::complete_atomic_modification() | |
4672 | { | |
9f95a23c | 4673 | if ((!state->manifest)|| state->keep_tail) |
7c673cae FG |
4674 | return 0; |
4675 | ||
4676 | cls_rgw_obj_chain chain; | |
9f95a23c | 4677 | store->update_gc_chain(obj, *state->manifest, &chain); |
7c673cae FG |
4678 | |
4679 | if (chain.empty()) { | |
4680 | return 0; | |
4681 | } | |
4682 | ||
181888fb | 4683 | string tag = (state->tail_tag.length() > 0 ? state->tail_tag.to_str() : state->obj_tag.to_str()); |
9f95a23c TL |
4684 | auto ret = store->gc->send_chain(chain, tag); // do it synchronously |
4685 | if (ret < 0) { | |
4686 | //Delete objects inline if send chain to gc fails | |
4687 | store->delete_objs_inline(chain, tag); | |
4688 | } | |
4689 | return 0; | |
7c673cae FG |
4690 | } |
4691 | ||
4692 | void RGWRados::update_gc_chain(rgw_obj& head_obj, RGWObjManifest& manifest, cls_rgw_obj_chain *chain) | |
4693 | { | |
4694 | RGWObjManifest::obj_iterator iter; | |
4695 | rgw_raw_obj raw_head; | |
4696 | obj_to_raw(manifest.get_head_placement_rule(), head_obj, &raw_head); | |
4697 | for (iter = manifest.obj_begin(); iter != manifest.obj_end(); ++iter) { | |
4698 | const rgw_raw_obj& mobj = iter.get_location().get_raw_obj(this); | |
4699 | if (mobj == raw_head) | |
4700 | continue; | |
4701 | cls_rgw_obj_key key(mobj.oid); | |
4702 | chain->push_obj(mobj.pool.to_str(), key, mobj.loc); | |
4703 | } | |
4704 | } | |
4705 | ||
9f95a23c | 4706 | int RGWRados::send_chain_to_gc(cls_rgw_obj_chain& chain, const string& tag) |
7c673cae | 4707 | { |
9f95a23c | 4708 | return gc->send_chain(chain, tag); |
7c673cae FG |
4709 | } |
4710 | ||
9f95a23c | 4711 | void RGWRados::delete_objs_inline(cls_rgw_obj_chain& chain, const string& tag) |
7c673cae | 4712 | { |
9f95a23c TL |
4713 | string last_pool; |
4714 | std::unique_ptr<IoCtx> ctx(new IoCtx); | |
4715 | int ret = 0; | |
4716 | for (auto liter = chain.objs.begin(); liter != chain.objs.end(); ++liter) { | |
4717 | cls_rgw_obj& obj = *liter; | |
4718 | if (obj.pool != last_pool) { | |
4719 | ctx.reset(new IoCtx); | |
4720 | ret = rgw_init_ioctx(get_rados_handle(), obj.pool, *ctx); | |
4721 | if (ret < 0) { | |
4722 | last_pool = ""; | |
4723 | ldout(cct, 0) << "ERROR: failed to create ioctx pool=" << | |
4724 | obj.pool << dendl; | |
4725 | continue; | |
4726 | } | |
4727 | last_pool = obj.pool; | |
4728 | } | |
4729 | ctx->locator_set_key(obj.loc); | |
4730 | const string& oid = obj.key.name; /* just stored raw oid there */ | |
4731 | ldout(cct, 5) << "delete_objs_inline: removing " << obj.pool << | |
4732 | ":" << obj.key.name << dendl; | |
4733 | ObjectWriteOperation op; | |
4734 | cls_refcount_put(op, tag, true); | |
4735 | ret = ctx->operate(oid, &op); | |
4736 | if (ret < 0) { | |
4737 | ldout(cct, 5) << "delete_objs_inline: refcount put returned error " << ret << dendl; | |
4738 | } | |
7c673cae | 4739 | } |
7c673cae FG |
4740 | } |
4741 | ||
4742 | static void accumulate_raw_stats(const rgw_bucket_dir_header& header, | |
4743 | map<RGWObjCategory, RGWStorageStats>& stats) | |
4744 | { | |
4745 | for (const auto& pair : header.stats) { | |
4746 | const RGWObjCategory category = static_cast<RGWObjCategory>(pair.first); | |
4747 | const rgw_bucket_category_stats& header_stats = pair.second; | |
4748 | ||
4749 | RGWStorageStats& s = stats[category]; | |
4750 | ||
4751 | s.category = category; | |
4752 | s.size += header_stats.total_size; | |
4753 | s.size_rounded += header_stats.total_size_rounded; | |
4754 | s.size_utilized += header_stats.actual_size; | |
4755 | s.num_objects += header_stats.num_entries; | |
4756 | } | |
4757 | } | |
4758 | ||
4759 | int RGWRados::bucket_check_index(RGWBucketInfo& bucket_info, | |
4760 | map<RGWObjCategory, RGWStorageStats> *existing_stats, | |
4761 | map<RGWObjCategory, RGWStorageStats> *calculated_stats) | |
4762 | { | |
9f95a23c | 4763 | RGWSI_RADOS::Pool index_pool; |
7c673cae FG |
4764 | // key - bucket index object id |
4765 | // value - bucket index check OP returned result with the given bucket index object (shard) | |
4766 | map<int, string> oids; | |
4767 | map<int, struct rgw_cls_check_index_ret> bucket_objs_ret; | |
31f18b77 | 4768 | |
9f95a23c | 4769 | int ret = svc.bi_rados->open_bucket_index(bucket_info, std::nullopt, &index_pool, &oids, nullptr); |
31f18b77 FG |
4770 | if (ret < 0) { |
4771 | return ret; | |
4772 | } | |
7c673cae | 4773 | |
9f95a23c TL |
4774 | for (auto& iter : oids) { |
4775 | bucket_objs_ret[iter.first] = rgw_cls_check_index_ret(); | |
4776 | } | |
4777 | ||
4778 | ret = CLSRGWIssueBucketCheck(index_pool.ioctx(), oids, bucket_objs_ret, cct->_conf->rgw_bucket_index_max_aio)(); | |
31f18b77 FG |
4779 | if (ret < 0) { |
4780 | return ret; | |
4781 | } | |
7c673cae FG |
4782 | |
4783 | // Aggregate results (from different shards if there is any) | |
4784 | map<int, struct rgw_cls_check_index_ret>::iterator iter; | |
4785 | for (iter = bucket_objs_ret.begin(); iter != bucket_objs_ret.end(); ++iter) { | |
4786 | accumulate_raw_stats(iter->second.existing_header, *existing_stats); | |
4787 | accumulate_raw_stats(iter->second.calculated_header, *calculated_stats); | |
4788 | } | |
4789 | ||
4790 | return 0; | |
4791 | } | |
4792 | ||
4793 | int RGWRados::bucket_rebuild_index(RGWBucketInfo& bucket_info) | |
4794 | { | |
9f95a23c | 4795 | RGWSI_RADOS::Pool index_pool; |
7c673cae | 4796 | map<int, string> bucket_objs; |
31f18b77 | 4797 | |
9f95a23c | 4798 | int r = svc.bi_rados->open_bucket_index(bucket_info, std::nullopt, &index_pool, &bucket_objs, nullptr); |
31f18b77 | 4799 | if (r < 0) { |
7c673cae | 4800 | return r; |
31f18b77 | 4801 | } |
7c673cae | 4802 | |
9f95a23c | 4803 | return CLSRGWIssueBucketRebuild(index_pool.ioctx(), bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); |
7c673cae FG |
4804 | } |
4805 | ||
f64942e4 | 4806 | int RGWRados::bucket_set_reshard(const RGWBucketInfo& bucket_info, const cls_rgw_bucket_instance_entry& entry) |
31f18b77 | 4807 | { |
9f95a23c | 4808 | RGWSI_RADOS::Pool index_pool; |
31f18b77 FG |
4809 | map<int, string> bucket_objs; |
4810 | ||
9f95a23c | 4811 | int r = svc.bi_rados->open_bucket_index(bucket_info, std::nullopt, &index_pool, &bucket_objs, nullptr); |
31f18b77 FG |
4812 | if (r < 0) { |
4813 | return r; | |
4814 | } | |
4815 | ||
9f95a23c | 4816 | return CLSRGWIssueSetBucketResharding(index_pool.ioctx(), bucket_objs, entry, cct->_conf->rgw_bucket_index_max_aio)(); |
31f18b77 | 4817 | } |
7c673cae | 4818 | |
9f95a23c | 4819 | int RGWRados::defer_gc(void *ctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, optional_yield y) |
7c673cae FG |
4820 | { |
4821 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
4822 | std::string oid, key; | |
4823 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
4824 | if (!rctx) | |
4825 | return 0; | |
4826 | ||
4827 | RGWObjState *state = NULL; | |
4828 | ||
9f95a23c | 4829 | int r = get_obj_state(rctx, bucket_info, obj, &state, false, y); |
7c673cae FG |
4830 | if (r < 0) |
4831 | return r; | |
4832 | ||
4833 | if (!state->is_atomic) { | |
4834 | ldout(cct, 20) << "state for obj=" << obj << " is not atomic, not deferring gc operation" << dendl; | |
4835 | return -EINVAL; | |
4836 | } | |
4837 | ||
181888fb FG |
4838 | string tag; |
4839 | ||
4840 | if (state->tail_tag.length() > 0) { | |
4841 | tag = state->tail_tag.c_str(); | |
4842 | } else if (state->obj_tag.length() > 0) { | |
4843 | tag = state->obj_tag.c_str(); | |
4844 | } else { | |
7c673cae FG |
4845 | ldout(cct, 20) << "state->obj_tag is empty, not deferring gc operation" << dendl; |
4846 | return -EINVAL; | |
4847 | } | |
4848 | ||
7c673cae FG |
4849 | ldout(cct, 0) << "defer chain tag=" << tag << dendl; |
4850 | ||
9f95a23c TL |
4851 | cls_rgw_obj_chain chain; |
4852 | update_gc_chain(state->obj, *state->manifest, &chain); | |
4853 | return gc->async_defer_chain(tag, chain); | |
7c673cae FG |
4854 | } |
4855 | ||
4856 | void RGWRados::remove_rgw_head_obj(ObjectWriteOperation& op) | |
4857 | { | |
4858 | list<string> prefixes; | |
4859 | prefixes.push_back(RGW_ATTR_OLH_PREFIX); | |
4860 | cls_rgw_remove_obj(op, prefixes); | |
4861 | } | |
4862 | ||
4863 | void RGWRados::cls_obj_check_prefix_exist(ObjectOperation& op, const string& prefix, bool fail_if_exist) | |
4864 | { | |
4865 | cls_rgw_obj_check_attrs_prefix(op, prefix, fail_if_exist); | |
4866 | } | |
4867 | ||
4868 | void RGWRados::cls_obj_check_mtime(ObjectOperation& op, const real_time& mtime, bool high_precision_time, RGWCheckMTimeType type) | |
4869 | { | |
4870 | cls_rgw_obj_check_mtime(op, mtime, high_precision_time, type); | |
4871 | } | |
4872 | ||
9f95a23c TL |
4873 | struct tombstone_entry { |
4874 | ceph::real_time mtime; | |
4875 | uint32_t zone_short_id; | |
4876 | uint64_t pg_ver; | |
4877 | ||
4878 | tombstone_entry() = default; | |
4879 | explicit tombstone_entry(const RGWObjState& state) | |
4880 | : mtime(state.mtime), zone_short_id(state.zone_short_id), | |
4881 | pg_ver(state.pg_ver) {} | |
4882 | }; | |
7c673cae FG |
4883 | |
4884 | /** | |
4885 | * Delete an object. | |
4886 | * bucket: name of the bucket storing the object | |
4887 | * obj: name of the object to delete | |
4888 | * Returns: 0 on success, -ERR# otherwise. | |
4889 | */ | |
9f95a23c | 4890 | int RGWRados::Object::Delete::delete_obj(optional_yield y) |
7c673cae FG |
4891 | { |
4892 | RGWRados *store = target->get_store(); | |
4893 | rgw_obj& src_obj = target->get_obj(); | |
4894 | const string& instance = src_obj.key.instance; | |
4895 | rgw_obj obj = src_obj; | |
4896 | ||
4897 | if (instance == "null") { | |
4898 | obj.key.instance.clear(); | |
4899 | } | |
4900 | ||
4901 | bool explicit_marker_version = (!params.marker_version_id.empty()); | |
4902 | ||
4903 | if (params.versioning_status & BUCKET_VERSIONED || explicit_marker_version) { | |
4904 | if (instance.empty() || explicit_marker_version) { | |
4905 | rgw_obj marker = obj; | |
4906 | ||
4907 | if (!params.marker_version_id.empty()) { | |
4908 | if (params.marker_version_id != "null") { | |
4909 | marker.key.set_instance(params.marker_version_id); | |
4910 | } | |
4911 | } else if ((params.versioning_status & BUCKET_VERSIONS_SUSPENDED) == 0) { | |
4912 | store->gen_rand_obj_instance_name(&marker); | |
4913 | } | |
4914 | ||
4915 | result.version_id = marker.key.instance; | |
91327a77 AA |
4916 | if (result.version_id.empty()) |
4917 | result.version_id = "null"; | |
7c673cae FG |
4918 | result.delete_marker = true; |
4919 | ||
4920 | struct rgw_bucket_dir_entry_meta meta; | |
4921 | ||
4922 | meta.owner = params.obj_owner.get_id().to_str(); | |
4923 | meta.owner_display_name = params.obj_owner.get_display_name(); | |
4924 | ||
4925 | if (real_clock::is_zero(params.mtime)) { | |
4926 | meta.mtime = real_clock::now(); | |
4927 | } else { | |
4928 | meta.mtime = params.mtime; | |
4929 | } | |
4930 | ||
9f95a23c | 4931 | 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, y, params.zones_trace); |
7c673cae FG |
4932 | if (r < 0) { |
4933 | return r; | |
4934 | } | |
4935 | } else { | |
4936 | rgw_bucket_dir_entry dirent; | |
4937 | ||
4938 | int r = store->bi_get_instance(target->get_bucket_info(), obj, &dirent); | |
4939 | if (r < 0) { | |
4940 | return r; | |
4941 | } | |
4942 | result.delete_marker = dirent.is_delete_marker(); | |
9f95a23c | 4943 | r = store->unlink_obj_instance(target->get_ctx(), target->get_bucket_info(), obj, params.olh_epoch, y, params.zones_trace); |
7c673cae FG |
4944 | if (r < 0) { |
4945 | return r; | |
4946 | } | |
4947 | result.version_id = instance; | |
4948 | } | |
4949 | ||
4950 | BucketShard *bs; | |
4951 | int r = target->get_bucket_shard(&bs); | |
4952 | if (r < 0) { | |
4953 | ldout(store->ctx(), 5) << "failed to get BucketShard object: r=" << r << dendl; | |
4954 | return r; | |
4955 | } | |
4956 | ||
9f95a23c TL |
4957 | r = store->svc.datalog_rados->add_entry(target->bucket_info, bs->shard_id); |
4958 | if (r < 0) { | |
4959 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
4960 | return r; | |
7c673cae FG |
4961 | } |
4962 | ||
4963 | return 0; | |
4964 | } | |
4965 | ||
4966 | rgw_rados_ref ref; | |
4967 | int r = store->get_obj_head_ref(target->get_bucket_info(), obj, &ref); | |
4968 | if (r < 0) { | |
4969 | return r; | |
4970 | } | |
4971 | ||
4972 | RGWObjState *state; | |
9f95a23c | 4973 | r = target->get_state(&state, false, y); |
7c673cae FG |
4974 | if (r < 0) |
4975 | return r; | |
4976 | ||
4977 | ObjectWriteOperation op; | |
4978 | ||
4979 | if (!real_clock::is_zero(params.unmod_since)) { | |
4980 | struct timespec ctime = ceph::real_clock::to_timespec(state->mtime); | |
4981 | struct timespec unmod = ceph::real_clock::to_timespec(params.unmod_since); | |
4982 | if (!params.high_precision_time) { | |
4983 | ctime.tv_nsec = 0; | |
4984 | unmod.tv_nsec = 0; | |
4985 | } | |
4986 | ||
4987 | ldout(store->ctx(), 10) << "If-UnModified-Since: " << params.unmod_since << " Last-Modified: " << ctime << dendl; | |
4988 | if (ctime > unmod) { | |
4989 | return -ERR_PRECONDITION_FAILED; | |
4990 | } | |
4991 | ||
4992 | /* only delete object if mtime is less than or equal to params.unmod_since */ | |
4993 | store->cls_obj_check_mtime(op, params.unmod_since, params.high_precision_time, CLS_RGW_CHECK_TIME_MTIME_LE); | |
4994 | } | |
11fdf7f2 | 4995 | uint64_t obj_accounted_size = state->accounted_size; |
7c673cae | 4996 | |
9f95a23c TL |
4997 | if(params.abortmp) { |
4998 | obj_accounted_size = params.parts_accounted_size; | |
4999 | } | |
5000 | ||
7c673cae FG |
5001 | if (!real_clock::is_zero(params.expiration_time)) { |
5002 | bufferlist bl; | |
5003 | real_time delete_at; | |
5004 | ||
5005 | if (state->get_attr(RGW_ATTR_DELETE_AT, bl)) { | |
5006 | try { | |
11fdf7f2 TL |
5007 | auto iter = bl.cbegin(); |
5008 | decode(delete_at, iter); | |
7c673cae FG |
5009 | } catch (buffer::error& err) { |
5010 | ldout(store->ctx(), 0) << "ERROR: couldn't decode RGW_ATTR_DELETE_AT" << dendl; | |
5011 | return -EIO; | |
5012 | } | |
5013 | ||
5014 | if (params.expiration_time != delete_at) { | |
5015 | return -ERR_PRECONDITION_FAILED; | |
5016 | } | |
5017 | } else { | |
5018 | return -ERR_PRECONDITION_FAILED; | |
5019 | } | |
5020 | } | |
5021 | ||
5022 | if (!state->exists) { | |
5023 | target->invalidate_state(); | |
5024 | return -ENOENT; | |
5025 | } | |
5026 | ||
9f95a23c | 5027 | r = target->prepare_atomic_modification(op, false, NULL, NULL, NULL, true, false, y); |
7c673cae FG |
5028 | if (r < 0) |
5029 | return r; | |
5030 | ||
5031 | RGWBucketInfo& bucket_info = target->get_bucket_info(); | |
5032 | ||
5033 | RGWRados::Bucket bop(store, bucket_info); | |
5034 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
31f18b77 FG |
5035 | |
5036 | index_op.set_zones_trace(params.zones_trace); | |
7c673cae FG |
5037 | index_op.set_bilog_flags(params.bilog_flags); |
5038 | ||
9f95a23c | 5039 | r = index_op.prepare(CLS_RGW_OP_DEL, &state->write_tag, y); |
7c673cae FG |
5040 | if (r < 0) |
5041 | return r; | |
5042 | ||
5043 | store->remove_rgw_head_obj(op); | |
9f95a23c TL |
5044 | |
5045 | auto& ioctx = ref.pool.ioctx(); | |
5046 | r = rgw_rados_operate(ioctx, ref.obj.oid, &op, null_yield); | |
94b18763 FG |
5047 | |
5048 | /* raced with another operation, object state is indeterminate */ | |
5049 | const bool need_invalidate = (r == -ECANCELED); | |
7c673cae | 5050 | |
9f95a23c | 5051 | int64_t poolid = ioctx.get_id(); |
7c673cae FG |
5052 | if (r >= 0) { |
5053 | tombstone_cache_t *obj_tombstone_cache = store->get_tombstone_cache(); | |
5054 | if (obj_tombstone_cache) { | |
5055 | tombstone_entry entry{*state}; | |
5056 | obj_tombstone_cache->add(obj, entry); | |
5057 | } | |
9f95a23c | 5058 | r = index_op.complete_del(poolid, ioctx.get_last_version(), state->mtime, params.remove_objs); |
224ce89b | 5059 | |
7c673cae FG |
5060 | int ret = target->complete_atomic_modification(); |
5061 | if (ret < 0) { | |
5062 | ldout(store->ctx(), 0) << "ERROR: complete_atomic_modification returned ret=" << ret << dendl; | |
5063 | } | |
5064 | /* other than that, no need to propagate error */ | |
224ce89b WB |
5065 | } else { |
5066 | int ret = index_op.cancel(); | |
5067 | if (ret < 0) { | |
5068 | ldout(store->ctx(), 0) << "ERROR: index_op.cancel() returned ret=" << ret << dendl; | |
5069 | } | |
7c673cae FG |
5070 | } |
5071 | ||
5072 | if (need_invalidate) { | |
5073 | target->invalidate_state(); | |
5074 | } | |
5075 | ||
5076 | if (r < 0) | |
5077 | return r; | |
5078 | ||
5079 | /* update quota cache */ | |
11fdf7f2 | 5080 | store->quota_handler->update_stats(params.bucket_owner, obj.bucket, -1, 0, obj_accounted_size); |
7c673cae FG |
5081 | |
5082 | return 0; | |
5083 | } | |
5084 | ||
5085 | int RGWRados::delete_obj(RGWObjectCtx& obj_ctx, | |
5086 | const RGWBucketInfo& bucket_info, | |
5087 | const rgw_obj& obj, | |
5088 | int versioning_status, | |
5089 | uint16_t bilog_flags, | |
31f18b77 FG |
5090 | const real_time& expiration_time, |
5091 | rgw_zone_set *zones_trace) | |
7c673cae FG |
5092 | { |
5093 | RGWRados::Object del_target(this, bucket_info, obj_ctx, obj); | |
5094 | RGWRados::Object::Delete del_op(&del_target); | |
5095 | ||
5096 | del_op.params.bucket_owner = bucket_info.owner; | |
5097 | del_op.params.versioning_status = versioning_status; | |
5098 | del_op.params.bilog_flags = bilog_flags; | |
5099 | del_op.params.expiration_time = expiration_time; | |
31f18b77 | 5100 | del_op.params.zones_trace = zones_trace; |
7c673cae | 5101 | |
9f95a23c | 5102 | return del_op.delete_obj(null_yield); |
7c673cae FG |
5103 | } |
5104 | ||
5105 | int RGWRados::delete_raw_obj(const rgw_raw_obj& obj) | |
5106 | { | |
5107 | rgw_rados_ref ref; | |
224ce89b | 5108 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
5109 | if (r < 0) { |
5110 | return r; | |
5111 | } | |
5112 | ||
5113 | ObjectWriteOperation op; | |
5114 | ||
5115 | op.remove(); | |
9f95a23c | 5116 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); |
7c673cae FG |
5117 | if (r < 0) |
5118 | return r; | |
5119 | ||
5120 | return 0; | |
5121 | } | |
5122 | ||
494da23a | 5123 | int RGWRados::delete_obj_index(const rgw_obj& obj, ceph::real_time mtime) |
7c673cae FG |
5124 | { |
5125 | std::string oid, key; | |
5126 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
5127 | ||
11fdf7f2 | 5128 | auto obj_ctx = svc.sysobj->init_obj_ctx(); |
7c673cae FG |
5129 | |
5130 | RGWBucketInfo bucket_info; | |
9f95a23c | 5131 | int ret = get_bucket_instance_info(obj_ctx, obj.bucket, bucket_info, NULL, NULL, null_yield); |
7c673cae FG |
5132 | if (ret < 0) { |
5133 | ldout(cct, 0) << "ERROR: " << __func__ << "() get_bucket_instance_info(bucket=" << obj.bucket << ") returned ret=" << ret << dendl; | |
5134 | return ret; | |
5135 | } | |
5136 | ||
5137 | RGWRados::Bucket bop(this, bucket_info); | |
5138 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
5139 | ||
494da23a | 5140 | return index_op.complete_del(-1 /* pool */, 0, mtime, NULL); |
7c673cae FG |
5141 | } |
5142 | ||
5143 | static void generate_fake_tag(RGWRados *store, map<string, bufferlist>& attrset, RGWObjManifest& manifest, bufferlist& manifest_bl, bufferlist& tag_bl) | |
5144 | { | |
5145 | string tag; | |
5146 | ||
5147 | RGWObjManifest::obj_iterator mi = manifest.obj_begin(); | |
5148 | if (mi != manifest.obj_end()) { | |
5149 | if (manifest.has_tail()) // first object usually points at the head, let's skip to a more unique part | |
5150 | ++mi; | |
5151 | tag = mi.get_location().get_raw_obj(store).oid; | |
5152 | tag.append("_"); | |
5153 | } | |
5154 | ||
5155 | unsigned char md5[CEPH_CRYPTO_MD5_DIGESTSIZE]; | |
5156 | char md5_str[CEPH_CRYPTO_MD5_DIGESTSIZE * 2 + 1]; | |
5157 | MD5 hash; | |
11fdf7f2 | 5158 | hash.Update((const unsigned char *)manifest_bl.c_str(), manifest_bl.length()); |
7c673cae FG |
5159 | |
5160 | map<string, bufferlist>::iterator iter = attrset.find(RGW_ATTR_ETAG); | |
5161 | if (iter != attrset.end()) { | |
5162 | bufferlist& bl = iter->second; | |
11fdf7f2 | 5163 | hash.Update((const unsigned char *)bl.c_str(), bl.length()); |
7c673cae FG |
5164 | } |
5165 | ||
5166 | hash.Final(md5); | |
5167 | buf_to_hex(md5, CEPH_CRYPTO_MD5_DIGESTSIZE, md5_str); | |
5168 | tag.append(md5_str); | |
5169 | ||
5170 | ldout(store->ctx(), 10) << "generate_fake_tag new tag=" << tag << dendl; | |
5171 | ||
5172 | tag_bl.append(tag.c_str(), tag.size() + 1); | |
5173 | } | |
5174 | ||
5175 | static bool is_olh(map<string, bufferlist>& attrs) | |
5176 | { | |
5177 | map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_OLH_INFO); | |
5178 | return (iter != attrs.end()); | |
5179 | } | |
5180 | ||
5181 | static bool has_olh_tag(map<string, bufferlist>& attrs) | |
5182 | { | |
5183 | map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_OLH_ID_TAG); | |
5184 | return (iter != attrs.end()); | |
5185 | } | |
5186 | ||
5187 | int RGWRados::get_olh_target_state(RGWObjectCtx& obj_ctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
9f95a23c | 5188 | RGWObjState *olh_state, RGWObjState **target_state, optional_yield y) |
7c673cae | 5189 | { |
11fdf7f2 | 5190 | ceph_assert(olh_state->is_olh); |
7c673cae FG |
5191 | |
5192 | rgw_obj target; | |
5193 | int r = RGWRados::follow_olh(bucket_info, obj_ctx, olh_state, obj, &target); /* might return -EAGAIN */ | |
5194 | if (r < 0) { | |
5195 | return r; | |
5196 | } | |
9f95a23c | 5197 | r = get_obj_state(&obj_ctx, bucket_info, target, target_state, false, y); |
7c673cae FG |
5198 | if (r < 0) { |
5199 | return r; | |
5200 | } | |
5201 | ||
5202 | return 0; | |
5203 | } | |
5204 | ||
7c673cae | 5205 | int RGWRados::get_obj_state_impl(RGWObjectCtx *rctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, |
9f95a23c | 5206 | RGWObjState **state, bool follow_olh, optional_yield y, bool assume_noent) |
7c673cae FG |
5207 | { |
5208 | if (obj.empty()) { | |
5209 | return -EINVAL; | |
5210 | } | |
5211 | ||
5212 | bool need_follow_olh = follow_olh && obj.key.instance.empty(); | |
5213 | ||
11fdf7f2 | 5214 | RGWObjState *s = rctx->get_state(obj); |
7c673cae FG |
5215 | ldout(cct, 20) << "get_obj_state: rctx=" << (void *)rctx << " obj=" << obj << " state=" << (void *)s << " s->prefetch_data=" << s->prefetch_data << dendl; |
5216 | *state = s; | |
5217 | if (s->has_attrs) { | |
5218 | if (s->is_olh && need_follow_olh) { | |
9f95a23c | 5219 | return get_olh_target_state(*rctx, bucket_info, obj, s, state, y); |
7c673cae FG |
5220 | } |
5221 | return 0; | |
5222 | } | |
5223 | ||
5224 | s->obj = obj; | |
5225 | ||
5226 | rgw_raw_obj raw_obj; | |
5227 | obj_to_raw(bucket_info.placement_rule, obj, &raw_obj); | |
5228 | ||
5229 | int r = -ENOENT; | |
5230 | ||
5231 | if (!assume_noent) { | |
9f95a23c | 5232 | r = RGWRados::raw_obj_stat(raw_obj, &s->size, &s->mtime, &s->epoch, &s->attrset, (s->prefetch_data ? &s->data : NULL), NULL, y); |
7c673cae FG |
5233 | } |
5234 | ||
5235 | if (r == -ENOENT) { | |
5236 | s->exists = false; | |
5237 | s->has_attrs = true; | |
5238 | tombstone_entry entry; | |
5239 | if (obj_tombstone_cache && obj_tombstone_cache->find(obj, entry)) { | |
5240 | s->mtime = entry.mtime; | |
5241 | s->zone_short_id = entry.zone_short_id; | |
5242 | s->pg_ver = entry.pg_ver; | |
5243 | ldout(cct, 20) << __func__ << "(): found obj in tombstone cache: obj=" << obj | |
5244 | << " mtime=" << s->mtime << " pgv=" << s->pg_ver << dendl; | |
5245 | } else { | |
5246 | s->mtime = real_time(); | |
5247 | } | |
5248 | return 0; | |
5249 | } | |
5250 | if (r < 0) | |
5251 | return r; | |
5252 | ||
5253 | s->exists = true; | |
5254 | s->has_attrs = true; | |
5255 | s->accounted_size = s->size; | |
5256 | ||
11fdf7f2 TL |
5257 | auto iter = s->attrset.find(RGW_ATTR_ETAG); |
5258 | if (iter != s->attrset.end()) { | |
5259 | /* get rid of extra null character at the end of the etag, as we used to store it like that */ | |
5260 | bufferlist& bletag = iter->second; | |
5261 | if (bletag.length() > 0 && bletag[bletag.length() - 1] == '\0') { | |
5262 | bufferlist newbl; | |
5263 | bletag.splice(0, bletag.length() - 1, &newbl); | |
5264 | bletag.claim(newbl); | |
5265 | } | |
5266 | } | |
5267 | ||
5268 | iter = s->attrset.find(RGW_ATTR_COMPRESSION); | |
31f18b77 FG |
5269 | const bool compressed = (iter != s->attrset.end()); |
5270 | if (compressed) { | |
7c673cae FG |
5271 | // use uncompressed size for accounted_size |
5272 | try { | |
5273 | RGWCompressionInfo info; | |
11fdf7f2 TL |
5274 | auto p = iter->second.cbegin(); |
5275 | decode(info, p); | |
31f18b77 | 5276 | s->accounted_size = info.orig_size; |
7c673cae FG |
5277 | } catch (buffer::error&) { |
5278 | dout(0) << "ERROR: could not decode compression info for object: " << obj << dendl; | |
5279 | return -EIO; | |
5280 | } | |
5281 | } | |
5282 | ||
5283 | iter = s->attrset.find(RGW_ATTR_SHADOW_OBJ); | |
5284 | if (iter != s->attrset.end()) { | |
5285 | bufferlist bl = iter->second; | |
5286 | bufferlist::iterator it = bl.begin(); | |
5287 | it.copy(bl.length(), s->shadow_obj); | |
5288 | s->shadow_obj[bl.length()] = '\0'; | |
5289 | } | |
5290 | s->obj_tag = s->attrset[RGW_ATTR_ID_TAG]; | |
181888fb FG |
5291 | auto ttiter = s->attrset.find(RGW_ATTR_TAIL_TAG); |
5292 | if (ttiter != s->attrset.end()) { | |
5293 | s->tail_tag = s->attrset[RGW_ATTR_TAIL_TAG]; | |
5294 | } | |
7c673cae FG |
5295 | |
5296 | bufferlist manifest_bl = s->attrset[RGW_ATTR_MANIFEST]; | |
5297 | if (manifest_bl.length()) { | |
11fdf7f2 | 5298 | auto miter = manifest_bl.cbegin(); |
7c673cae | 5299 | try { |
9f95a23c TL |
5300 | s->manifest.emplace(); |
5301 | decode(*s->manifest, miter); | |
5302 | s->manifest->set_head(bucket_info.placement_rule, obj, s->size); /* patch manifest to reflect the head we just read, some manifests might be | |
7c673cae | 5303 | broken due to old bugs */ |
9f95a23c | 5304 | s->size = s->manifest->get_obj_size(); |
31f18b77 FG |
5305 | if (!compressed) |
5306 | s->accounted_size = s->size; | |
7c673cae FG |
5307 | } catch (buffer::error& err) { |
5308 | ldout(cct, 0) << "ERROR: couldn't decode manifest" << dendl; | |
5309 | return -EIO; | |
5310 | } | |
9f95a23c | 5311 | ldout(cct, 10) << "manifest: total_size = " << s->manifest->get_obj_size() << dendl; |
11fdf7f2 | 5312 | if (cct->_conf->subsys.should_gather<ceph_subsys_rgw, 20>() && \ |
9f95a23c | 5313 | s->manifest->has_explicit_objs()) { |
7c673cae | 5314 | RGWObjManifest::obj_iterator mi; |
9f95a23c | 5315 | for (mi = s->manifest->obj_begin(); mi != s->manifest->obj_end(); ++mi) { |
7c673cae FG |
5316 | ldout(cct, 20) << "manifest: ofs=" << mi.get_ofs() << " loc=" << mi.get_location().get_raw_obj(this) << dendl; |
5317 | } | |
5318 | } | |
5319 | ||
5320 | if (!s->obj_tag.length()) { | |
5321 | /* | |
5322 | * Uh oh, something's wrong, object with manifest should have tag. Let's | |
5323 | * create one out of the manifest, would be unique | |
5324 | */ | |
9f95a23c | 5325 | generate_fake_tag(this, s->attrset, *s->manifest, manifest_bl, s->obj_tag); |
7c673cae FG |
5326 | s->fake_tag = true; |
5327 | } | |
5328 | } | |
5329 | map<string, bufferlist>::iterator aiter = s->attrset.find(RGW_ATTR_PG_VER); | |
5330 | if (aiter != s->attrset.end()) { | |
5331 | bufferlist& pg_ver_bl = aiter->second; | |
5332 | if (pg_ver_bl.length()) { | |
11fdf7f2 | 5333 | auto pgbl = pg_ver_bl.cbegin(); |
7c673cae | 5334 | try { |
11fdf7f2 | 5335 | decode(s->pg_ver, pgbl); |
7c673cae FG |
5336 | } catch (buffer::error& err) { |
5337 | ldout(cct, 0) << "ERROR: couldn't decode pg ver attr for object " << s->obj << ", non-critical error, ignoring" << dendl; | |
5338 | } | |
5339 | } | |
5340 | } | |
5341 | aiter = s->attrset.find(RGW_ATTR_SOURCE_ZONE); | |
5342 | if (aiter != s->attrset.end()) { | |
5343 | bufferlist& zone_short_id_bl = aiter->second; | |
5344 | if (zone_short_id_bl.length()) { | |
11fdf7f2 | 5345 | auto zbl = zone_short_id_bl.cbegin(); |
7c673cae | 5346 | try { |
11fdf7f2 | 5347 | decode(s->zone_short_id, zbl); |
7c673cae FG |
5348 | } catch (buffer::error& err) { |
5349 | ldout(cct, 0) << "ERROR: couldn't decode zone short id attr for object " << s->obj << ", non-critical error, ignoring" << dendl; | |
5350 | } | |
5351 | } | |
5352 | } | |
5353 | if (s->obj_tag.length()) | |
31f18b77 | 5354 | ldout(cct, 20) << "get_obj_state: setting s->obj_tag to " << s->obj_tag.c_str() << dendl; |
7c673cae FG |
5355 | else |
5356 | ldout(cct, 20) << "get_obj_state: s->obj_tag was set empty" << dendl; | |
5357 | ||
5358 | /* an object might not be olh yet, but could have olh id tag, so we should set it anyway if | |
5359 | * it exist, and not only if is_olh() returns true | |
5360 | */ | |
5361 | iter = s->attrset.find(RGW_ATTR_OLH_ID_TAG); | |
5362 | if (iter != s->attrset.end()) { | |
5363 | s->olh_tag = iter->second; | |
5364 | } | |
5365 | ||
5366 | if (is_olh(s->attrset)) { | |
5367 | s->is_olh = true; | |
5368 | ||
5369 | ldout(cct, 20) << __func__ << ": setting s->olh_tag to " << string(s->olh_tag.c_str(), s->olh_tag.length()) << dendl; | |
5370 | ||
5371 | if (need_follow_olh) { | |
9f95a23c TL |
5372 | return get_olh_target_state(*rctx, bucket_info, obj, s, state, y); |
5373 | } else if (obj.key.have_null_instance() && !s->manifest) { | |
11fdf7f2 TL |
5374 | // read null version, and the head object only have olh info |
5375 | s->exists = false; | |
5376 | return -ENOENT; | |
7c673cae FG |
5377 | } |
5378 | } | |
5379 | ||
5380 | return 0; | |
5381 | } | |
5382 | ||
5383 | int RGWRados::get_obj_state(RGWObjectCtx *rctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, RGWObjState **state, | |
9f95a23c | 5384 | bool follow_olh, optional_yield y, bool assume_noent) |
7c673cae FG |
5385 | { |
5386 | int ret; | |
5387 | ||
5388 | do { | |
9f95a23c | 5389 | ret = get_obj_state_impl(rctx, bucket_info, obj, state, follow_olh, y, assume_noent); |
7c673cae FG |
5390 | } while (ret == -EAGAIN); |
5391 | ||
5392 | return ret; | |
5393 | } | |
5394 | ||
9f95a23c | 5395 | int RGWRados::Object::get_manifest(RGWObjManifest **pmanifest, optional_yield y) |
7c673cae FG |
5396 | { |
5397 | RGWObjState *astate; | |
9f95a23c | 5398 | int r = get_state(&astate, true, y); |
7c673cae FG |
5399 | if (r < 0) { |
5400 | return r; | |
5401 | } | |
5402 | ||
9f95a23c | 5403 | *pmanifest = &(*astate->manifest); |
7c673cae FG |
5404 | |
5405 | return 0; | |
5406 | } | |
5407 | ||
9f95a23c | 5408 | int RGWRados::Object::Read::get_attr(const char *name, bufferlist& dest, optional_yield y) |
7c673cae FG |
5409 | { |
5410 | RGWObjState *state; | |
9f95a23c | 5411 | int r = source->get_state(&state, true, y); |
7c673cae FG |
5412 | if (r < 0) |
5413 | return r; | |
5414 | if (!state->exists) | |
5415 | return -ENOENT; | |
5416 | if (!state->get_attr(name, dest)) | |
5417 | return -ENODATA; | |
5418 | ||
5419 | return 0; | |
5420 | } | |
5421 | ||
7c673cae FG |
5422 | int RGWRados::Object::Stat::stat_async() |
5423 | { | |
5424 | RGWObjectCtx& ctx = source->get_ctx(); | |
5425 | rgw_obj& obj = source->get_obj(); | |
5426 | RGWRados *store = source->get_store(); | |
5427 | ||
11fdf7f2 | 5428 | RGWObjState *s = ctx.get_state(obj); /* calling this one directly because otherwise a sync request will be sent */ |
7c673cae FG |
5429 | result.obj = obj; |
5430 | if (s->has_attrs) { | |
5431 | state.ret = 0; | |
5432 | result.size = s->size; | |
5433 | result.mtime = ceph::real_clock::to_timespec(s->mtime); | |
5434 | result.attrs = s->attrset; | |
7c673cae FG |
5435 | result.manifest = s->manifest; |
5436 | return 0; | |
5437 | } | |
5438 | ||
5439 | string oid; | |
5440 | string loc; | |
5441 | get_obj_bucket_and_oid_loc(obj, oid, loc); | |
5442 | ||
5443 | int r = store->get_obj_head_ioctx(source->get_bucket_info(), obj, &state.io_ctx); | |
5444 | if (r < 0) { | |
5445 | return r; | |
5446 | } | |
5447 | ||
5448 | librados::ObjectReadOperation op; | |
5449 | op.stat2(&result.size, &result.mtime, NULL); | |
5450 | op.getxattrs(&result.attrs, NULL); | |
9f95a23c | 5451 | state.completion = librados::Rados::aio_create_completion(nullptr, nullptr); |
7c673cae FG |
5452 | state.io_ctx.locator_set_key(loc); |
5453 | r = state.io_ctx.aio_operate(oid, state.completion, &op, NULL); | |
5454 | if (r < 0) { | |
5455 | ldout(store->ctx(), 5) << __func__ | |
5456 | << ": ERROR: aio_operate() returned ret=" << r | |
5457 | << dendl; | |
5458 | return r; | |
5459 | } | |
5460 | ||
5461 | return 0; | |
5462 | } | |
5463 | ||
5464 | ||
5465 | int RGWRados::Object::Stat::wait() | |
5466 | { | |
5467 | if (!state.completion) { | |
5468 | return state.ret; | |
5469 | } | |
5470 | ||
9f95a23c | 5471 | state.completion->wait_for_complete(); |
7c673cae FG |
5472 | state.ret = state.completion->get_return_value(); |
5473 | state.completion->release(); | |
5474 | ||
5475 | if (state.ret != 0) { | |
5476 | return state.ret; | |
5477 | } | |
5478 | ||
5479 | return finish(); | |
5480 | } | |
5481 | ||
5482 | int RGWRados::Object::Stat::finish() | |
5483 | { | |
5484 | map<string, bufferlist>::iterator iter = result.attrs.find(RGW_ATTR_MANIFEST); | |
5485 | if (iter != result.attrs.end()) { | |
5486 | bufferlist& bl = iter->second; | |
11fdf7f2 | 5487 | auto biter = bl.cbegin(); |
7c673cae | 5488 | try { |
9f95a23c TL |
5489 | result.manifest.emplace(); |
5490 | decode(*result.manifest, biter); | |
7c673cae FG |
5491 | } catch (buffer::error& err) { |
5492 | RGWRados *store = source->get_store(); | |
5493 | ldout(store->ctx(), 0) << "ERROR: " << __func__ << ": failed to decode manifest" << dendl; | |
5494 | return -EIO; | |
5495 | } | |
7c673cae FG |
5496 | } |
5497 | ||
5498 | return 0; | |
5499 | } | |
5500 | ||
7c673cae FG |
5501 | int RGWRados::append_atomic_test(RGWObjectCtx *rctx, |
5502 | const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
9f95a23c | 5503 | ObjectOperation& op, RGWObjState **pstate, optional_yield y) |
7c673cae FG |
5504 | { |
5505 | if (!rctx) | |
5506 | return 0; | |
5507 | ||
9f95a23c | 5508 | int r = get_obj_state(rctx, bucket_info, obj, pstate, false, y); |
7c673cae FG |
5509 | if (r < 0) |
5510 | return r; | |
5511 | ||
11fdf7f2 TL |
5512 | return append_atomic_test(*pstate, op); |
5513 | } | |
7c673cae | 5514 | |
11fdf7f2 TL |
5515 | int RGWRados::append_atomic_test(const RGWObjState* state, |
5516 | librados::ObjectOperation& op) | |
5517 | { | |
7c673cae | 5518 | if (!state->is_atomic) { |
11fdf7f2 | 5519 | ldout(cct, 20) << "state for obj=" << state->obj << " is not atomic, not appending atomic test" << dendl; |
7c673cae FG |
5520 | return 0; |
5521 | } | |
5522 | ||
5523 | if (state->obj_tag.length() > 0 && !state->fake_tag) {// check for backward compatibility | |
5524 | op.cmpxattr(RGW_ATTR_ID_TAG, LIBRADOS_CMPXATTR_OP_EQ, state->obj_tag); | |
5525 | } else { | |
5526 | ldout(cct, 20) << "state->obj_tag is empty, not appending atomic test" << dendl; | |
5527 | } | |
5528 | return 0; | |
5529 | } | |
5530 | ||
9f95a23c | 5531 | int RGWRados::Object::get_state(RGWObjState **pstate, bool follow_olh, optional_yield y, bool assume_noent) |
7c673cae | 5532 | { |
9f95a23c | 5533 | return store->get_obj_state(&ctx, bucket_info, obj, pstate, follow_olh, y, assume_noent); |
7c673cae FG |
5534 | } |
5535 | ||
5536 | void RGWRados::Object::invalidate_state() | |
5537 | { | |
11fdf7f2 | 5538 | ctx.invalidate(obj); |
7c673cae FG |
5539 | } |
5540 | ||
5541 | int RGWRados::Object::prepare_atomic_modification(ObjectWriteOperation& op, bool reset_obj, const string *ptag, | |
181888fb | 5542 | const char *if_match, const char *if_nomatch, bool removal_op, |
9f95a23c | 5543 | bool modify_tail, optional_yield y) |
7c673cae | 5544 | { |
9f95a23c | 5545 | int r = get_state(&state, false, y); |
7c673cae FG |
5546 | if (r < 0) |
5547 | return r; | |
5548 | ||
9f95a23c | 5549 | bool need_guard = ((state->manifest) || (state->obj_tag.length() != 0) || |
7c673cae FG |
5550 | if_match != NULL || if_nomatch != NULL) && |
5551 | (!state->fake_tag); | |
5552 | ||
5553 | if (!state->is_atomic) { | |
5554 | ldout(store->ctx(), 20) << "prepare_atomic_modification: state is not atomic. state=" << (void *)state << dendl; | |
5555 | ||
5556 | if (reset_obj) { | |
5557 | op.create(false); | |
5558 | store->remove_rgw_head_obj(op); // we're not dropping reference here, actually removing object | |
5559 | } | |
5560 | ||
5561 | return 0; | |
5562 | } | |
5563 | ||
5564 | if (need_guard) { | |
5565 | /* first verify that the object wasn't replaced under */ | |
5566 | if (if_nomatch == NULL || strcmp(if_nomatch, "*") != 0) { | |
5567 | op.cmpxattr(RGW_ATTR_ID_TAG, LIBRADOS_CMPXATTR_OP_EQ, state->obj_tag); | |
5568 | // FIXME: need to add FAIL_NOTEXIST_OK for racing deletion | |
5569 | } | |
5570 | ||
5571 | if (if_match) { | |
5572 | if (strcmp(if_match, "*") == 0) { | |
5573 | // test the object is existing | |
5574 | if (!state->exists) { | |
5575 | return -ERR_PRECONDITION_FAILED; | |
5576 | } | |
5577 | } else { | |
5578 | bufferlist bl; | |
5579 | if (!state->get_attr(RGW_ATTR_ETAG, bl) || | |
5580 | strncmp(if_match, bl.c_str(), bl.length()) != 0) { | |
5581 | return -ERR_PRECONDITION_FAILED; | |
5582 | } | |
5583 | } | |
5584 | } | |
5585 | ||
5586 | if (if_nomatch) { | |
5587 | if (strcmp(if_nomatch, "*") == 0) { | |
5588 | // test the object is NOT existing | |
5589 | if (state->exists) { | |
5590 | return -ERR_PRECONDITION_FAILED; | |
5591 | } | |
5592 | } else { | |
5593 | bufferlist bl; | |
5594 | if (!state->get_attr(RGW_ATTR_ETAG, bl) || | |
5595 | strncmp(if_nomatch, bl.c_str(), bl.length()) == 0) { | |
5596 | return -ERR_PRECONDITION_FAILED; | |
5597 | } | |
5598 | } | |
5599 | } | |
5600 | } | |
5601 | ||
5602 | if (reset_obj) { | |
5603 | if (state->exists) { | |
5604 | op.create(false); | |
5605 | store->remove_rgw_head_obj(op); | |
5606 | } else { | |
5607 | op.create(true); | |
5608 | } | |
5609 | } | |
5610 | ||
5611 | if (removal_op) { | |
5612 | /* the object is being removed, no need to update its tag */ | |
5613 | return 0; | |
5614 | } | |
5615 | ||
5616 | if (ptag) { | |
5617 | state->write_tag = *ptag; | |
5618 | } else { | |
5619 | append_rand_alpha(store->ctx(), state->write_tag, state->write_tag, 32); | |
5620 | } | |
5621 | bufferlist bl; | |
5622 | bl.append(state->write_tag.c_str(), state->write_tag.size() + 1); | |
5623 | ||
5624 | ldout(store->ctx(), 10) << "setting object write_tag=" << state->write_tag << dendl; | |
5625 | ||
5626 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
181888fb FG |
5627 | if (modify_tail) { |
5628 | op.setxattr(RGW_ATTR_TAIL_TAG, bl); | |
5629 | } | |
7c673cae FG |
5630 | |
5631 | return 0; | |
5632 | } | |
5633 | ||
7c673cae FG |
5634 | /** |
5635 | * Set an attr on an object. | |
5636 | * bucket: name of the bucket holding the object | |
5637 | * obj: name of the object to set the attr on | |
5638 | * name: the attr to set | |
5639 | * bl: the contents of the attr | |
5640 | * Returns: 0 on success, -ERR# otherwise. | |
5641 | */ | |
5642 | int RGWRados::set_attr(void *ctx, const RGWBucketInfo& bucket_info, rgw_obj& obj, const char *name, bufferlist& bl) | |
5643 | { | |
5644 | map<string, bufferlist> attrs; | |
5645 | attrs[name] = bl; | |
9f95a23c | 5646 | return set_attrs(ctx, bucket_info, obj, attrs, NULL, null_yield); |
7c673cae FG |
5647 | } |
5648 | ||
494da23a | 5649 | int RGWRados::set_attrs(void *ctx, const RGWBucketInfo& bucket_info, rgw_obj& src_obj, |
7c673cae | 5650 | map<string, bufferlist>& attrs, |
9f95a23c TL |
5651 | map<string, bufferlist>* rmattrs, |
5652 | optional_yield y) | |
7c673cae | 5653 | { |
494da23a TL |
5654 | rgw_obj obj = src_obj; |
5655 | if (obj.key.instance == "null") { | |
5656 | obj.key.instance.clear(); | |
5657 | } | |
5658 | ||
7c673cae FG |
5659 | rgw_rados_ref ref; |
5660 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
5661 | if (r < 0) { | |
5662 | return r; | |
5663 | } | |
5664 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
5665 | ||
5666 | ObjectWriteOperation op; | |
5667 | RGWObjState *state = NULL; | |
5668 | ||
9f95a23c | 5669 | r = append_atomic_test(rctx, bucket_info, obj, op, &state, y); |
7c673cae FG |
5670 | if (r < 0) |
5671 | return r; | |
5672 | ||
494da23a | 5673 | // ensure null version object exist |
9f95a23c | 5674 | if (src_obj.key.instance == "null" && !state->manifest) { |
494da23a TL |
5675 | return -ENOENT; |
5676 | } | |
5677 | ||
7c673cae FG |
5678 | map<string, bufferlist>::iterator iter; |
5679 | if (rmattrs) { | |
5680 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
5681 | const string& name = iter->first; | |
5682 | op.rmxattr(name.c_str()); | |
5683 | } | |
5684 | } | |
5685 | ||
5686 | const rgw_bucket& bucket = obj.bucket; | |
5687 | ||
5688 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
5689 | const string& name = iter->first; | |
5690 | bufferlist& bl = iter->second; | |
5691 | ||
5692 | if (!bl.length()) | |
5693 | continue; | |
5694 | ||
5695 | op.setxattr(name.c_str(), bl); | |
5696 | ||
5697 | if (name.compare(RGW_ATTR_DELETE_AT) == 0) { | |
5698 | real_time ts; | |
5699 | try { | |
11fdf7f2 | 5700 | decode(ts, bl); |
7c673cae FG |
5701 | |
5702 | rgw_obj_index_key obj_key; | |
5703 | obj.key.get_index_key(&obj_key); | |
5704 | ||
9f95a23c | 5705 | obj_expirer->hint_add(ts, bucket.tenant, bucket.name, bucket.bucket_id, obj_key); |
7c673cae FG |
5706 | } catch (buffer::error& err) { |
5707 | ldout(cct, 0) << "ERROR: failed to decode " RGW_ATTR_DELETE_AT << " attr" << dendl; | |
5708 | } | |
5709 | } | |
5710 | } | |
5711 | ||
5712 | if (!op.size()) | |
5713 | return 0; | |
5714 | ||
9f95a23c | 5715 | RGWObjectCtx obj_ctx(this->store); |
7c673cae FG |
5716 | |
5717 | bufferlist bl; | |
5718 | RGWRados::Bucket bop(this, bucket_info); | |
5719 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
5720 | ||
5721 | if (state) { | |
5722 | string tag; | |
5723 | append_rand_alpha(cct, tag, tag, 32); | |
5724 | state->write_tag = tag; | |
9f95a23c | 5725 | r = index_op.prepare(CLS_RGW_OP_ADD, &state->write_tag, y); |
7c673cae FG |
5726 | |
5727 | if (r < 0) | |
5728 | return r; | |
5729 | ||
5730 | bl.append(tag.c_str(), tag.size() + 1); | |
7c673cae FG |
5731 | op.setxattr(RGW_ATTR_ID_TAG, bl); |
5732 | } | |
5733 | ||
3efd9988 FG |
5734 | |
5735 | real_time mtime = real_clock::now(); | |
5736 | struct timespec mtime_ts = real_clock::to_timespec(mtime); | |
5737 | op.mtime2(&mtime_ts); | |
9f95a23c TL |
5738 | auto& ioctx = ref.pool.ioctx(); |
5739 | r = rgw_rados_operate(ioctx, ref.obj.oid, &op, null_yield); | |
7c673cae FG |
5740 | if (state) { |
5741 | if (r >= 0) { | |
5742 | bufferlist acl_bl = attrs[RGW_ATTR_ACL]; | |
5743 | bufferlist etag_bl = attrs[RGW_ATTR_ETAG]; | |
5744 | bufferlist content_type_bl = attrs[RGW_ATTR_CONTENT_TYPE]; | |
11fdf7f2 TL |
5745 | string etag = rgw_bl_str(etag_bl); |
5746 | string content_type = rgw_bl_str(content_type_bl); | |
5747 | string storage_class; | |
5748 | auto iter = attrs.find(RGW_ATTR_STORAGE_CLASS); | |
5749 | if (iter != attrs.end()) { | |
5750 | storage_class = rgw_bl_str(iter->second); | |
5751 | } | |
9f95a23c TL |
5752 | uint64_t epoch = ioctx.get_last_version(); |
5753 | int64_t poolid = ioctx.get_id(); | |
7c673cae | 5754 | r = index_op.complete(poolid, epoch, state->size, state->accounted_size, |
11fdf7f2 TL |
5755 | mtime, etag, content_type, storage_class, &acl_bl, |
5756 | RGWObjCategory::Main, NULL); | |
7c673cae FG |
5757 | } else { |
5758 | int ret = index_op.cancel(); | |
5759 | if (ret < 0) { | |
5760 | ldout(cct, 0) << "ERROR: complete_update_index_cancel() returned ret=" << ret << dendl; | |
5761 | } | |
5762 | } | |
5763 | } | |
5764 | if (r < 0) | |
5765 | return r; | |
5766 | ||
5767 | if (state) { | |
5768 | state->obj_tag.swap(bl); | |
5769 | if (rmattrs) { | |
5770 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
5771 | state->attrset.erase(iter->first); | |
5772 | } | |
5773 | } | |
92f5a8d4 | 5774 | |
7c673cae FG |
5775 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { |
5776 | state->attrset[iter->first] = iter->second; | |
5777 | } | |
92f5a8d4 TL |
5778 | |
5779 | auto iter = state->attrset.find(RGW_ATTR_ID_TAG); | |
5780 | if (iter != state->attrset.end()) { | |
5781 | iter->second = state->obj_tag; | |
5782 | } | |
7c673cae FG |
5783 | } |
5784 | ||
5785 | return 0; | |
5786 | } | |
5787 | ||
9f95a23c | 5788 | int RGWRados::Object::Read::prepare(optional_yield y) |
7c673cae FG |
5789 | { |
5790 | RGWRados *store = source->get_store(); | |
5791 | CephContext *cct = store->ctx(); | |
5792 | ||
5793 | bufferlist etag; | |
5794 | ||
5795 | map<string, bufferlist>::iterator iter; | |
5796 | ||
5797 | RGWObjState *astate; | |
9f95a23c | 5798 | int r = source->get_state(&astate, true, y); |
7c673cae FG |
5799 | if (r < 0) |
5800 | return r; | |
5801 | ||
5802 | if (!astate->exists) { | |
5803 | return -ENOENT; | |
5804 | } | |
5805 | ||
5806 | const RGWBucketInfo& bucket_info = source->get_bucket_info(); | |
5807 | ||
5808 | state.obj = astate->obj; | |
5809 | store->obj_to_raw(bucket_info.placement_rule, state.obj, &state.head_obj); | |
5810 | ||
11fdf7f2 TL |
5811 | state.cur_pool = state.head_obj.pool; |
5812 | state.cur_ioctx = &state.io_ctxs[state.cur_pool]; | |
5813 | ||
5814 | r = store->get_obj_head_ioctx(bucket_info, state.obj, state.cur_ioctx); | |
7c673cae FG |
5815 | if (r < 0) { |
5816 | return r; | |
5817 | } | |
eafe8130 TL |
5818 | if (params.target_obj) { |
5819 | *params.target_obj = state.obj; | |
5820 | } | |
7c673cae FG |
5821 | if (params.attrs) { |
5822 | *params.attrs = astate->attrset; | |
11fdf7f2 | 5823 | if (cct->_conf->subsys.should_gather<ceph_subsys_rgw, 20>()) { |
7c673cae FG |
5824 | for (iter = params.attrs->begin(); iter != params.attrs->end(); ++iter) { |
5825 | ldout(cct, 20) << "Read xattr: " << iter->first << dendl; | |
5826 | } | |
5827 | } | |
5828 | } | |
5829 | ||
5830 | /* Convert all times go GMT to make them compatible */ | |
5831 | if (conds.mod_ptr || conds.unmod_ptr) { | |
5832 | obj_time_weight src_weight; | |
5833 | src_weight.init(astate); | |
5834 | src_weight.high_precision = conds.high_precision_time; | |
5835 | ||
5836 | obj_time_weight dest_weight; | |
5837 | dest_weight.high_precision = conds.high_precision_time; | |
5838 | ||
9f95a23c | 5839 | if (conds.mod_ptr && !conds.if_nomatch) { |
7c673cae FG |
5840 | dest_weight.init(*conds.mod_ptr, conds.mod_zone_id, conds.mod_pg_ver); |
5841 | ldout(cct, 10) << "If-Modified-Since: " << dest_weight << " Last-Modified: " << src_weight << dendl; | |
5842 | if (!(dest_weight < src_weight)) { | |
5843 | return -ERR_NOT_MODIFIED; | |
5844 | } | |
5845 | } | |
5846 | ||
9f95a23c | 5847 | if (conds.unmod_ptr && !conds.if_match) { |
7c673cae FG |
5848 | dest_weight.init(*conds.unmod_ptr, conds.mod_zone_id, conds.mod_pg_ver); |
5849 | ldout(cct, 10) << "If-UnModified-Since: " << dest_weight << " Last-Modified: " << src_weight << dendl; | |
5850 | if (dest_weight < src_weight) { | |
5851 | return -ERR_PRECONDITION_FAILED; | |
5852 | } | |
5853 | } | |
5854 | } | |
5855 | if (conds.if_match || conds.if_nomatch) { | |
9f95a23c | 5856 | r = get_attr(RGW_ATTR_ETAG, etag, y); |
7c673cae FG |
5857 | if (r < 0) |
5858 | return r; | |
5859 | ||
5860 | if (conds.if_match) { | |
5861 | string if_match_str = rgw_string_unquote(conds.if_match); | |
11fdf7f2 TL |
5862 | ldout(cct, 10) << "ETag: " << string(etag.c_str(), etag.length()) << " " << " If-Match: " << if_match_str << dendl; |
5863 | if (if_match_str.compare(0, etag.length(), etag.c_str(), etag.length()) != 0) { | |
7c673cae FG |
5864 | return -ERR_PRECONDITION_FAILED; |
5865 | } | |
5866 | } | |
5867 | ||
5868 | if (conds.if_nomatch) { | |
5869 | string if_nomatch_str = rgw_string_unquote(conds.if_nomatch); | |
11fdf7f2 TL |
5870 | ldout(cct, 10) << "ETag: " << string(etag.c_str(), etag.length()) << " " << " If-NoMatch: " << if_nomatch_str << dendl; |
5871 | if (if_nomatch_str.compare(0, etag.length(), etag.c_str(), etag.length()) == 0) { | |
7c673cae FG |
5872 | return -ERR_NOT_MODIFIED; |
5873 | } | |
5874 | } | |
5875 | } | |
5876 | ||
5877 | if (params.obj_size) | |
5878 | *params.obj_size = astate->size; | |
5879 | if (params.lastmod) | |
5880 | *params.lastmod = astate->mtime; | |
5881 | ||
5882 | return 0; | |
5883 | } | |
5884 | ||
5885 | int RGWRados::Object::Read::range_to_ofs(uint64_t obj_size, int64_t &ofs, int64_t &end) | |
5886 | { | |
5887 | if (ofs < 0) { | |
5888 | ofs += obj_size; | |
11fdf7f2 TL |
5889 | if (ofs < 0) |
5890 | ofs = 0; | |
5891 | end = obj_size - 1; | |
5892 | } else if (end < 0) { | |
5893 | end = obj_size - 1; | |
7c673cae FG |
5894 | } |
5895 | ||
11fdf7f2 TL |
5896 | if (obj_size > 0) { |
5897 | if (ofs >= (off_t)obj_size) { | |
5898 | return -ERANGE; | |
5899 | } | |
5900 | if (end >= (off_t)obj_size) { | |
5901 | end = obj_size - 1; | |
7c673cae FG |
5902 | } |
5903 | } | |
7c673cae FG |
5904 | return 0; |
5905 | } | |
5906 | ||
31f18b77 FG |
5907 | int RGWRados::Bucket::UpdateIndex::guard_reshard(BucketShard **pbs, std::function<int(BucketShard *)> call) |
5908 | { | |
5909 | RGWRados *store = target->get_store(); | |
5910 | BucketShard *bs; | |
5911 | int r; | |
5912 | ||
5913 | #define NUM_RESHARD_RETRIES 10 | |
5914 | for (int i = 0; i < NUM_RESHARD_RETRIES; ++i) { | |
5915 | int ret = get_bucket_shard(&bs); | |
5916 | if (ret < 0) { | |
5917 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
5918 | return ret; | |
5919 | } | |
5920 | r = call(bs); | |
5921 | if (r != -ERR_BUSY_RESHARDING) { | |
5922 | break; | |
5923 | } | |
5924 | ldout(store->ctx(), 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl; | |
5925 | string new_bucket_id; | |
11fdf7f2 TL |
5926 | r = store->block_while_resharding(bs, &new_bucket_id, |
5927 | target->bucket_info, null_yield); | |
31f18b77 FG |
5928 | if (r == -ERR_BUSY_RESHARDING) { |
5929 | continue; | |
5930 | } | |
5931 | if (r < 0) { | |
5932 | return r; | |
5933 | } | |
5934 | ldout(store->ctx(), 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id << dendl; | |
5935 | i = 0; /* resharding is finished, make sure we can retry */ | |
5936 | r = target->update_bucket_id(new_bucket_id); | |
5937 | if (r < 0) { | |
5938 | ldout(store->ctx(), 0) << "ERROR: update_bucket_id() new_bucket_id=" << new_bucket_id << " returned r=" << r << dendl; | |
5939 | return r; | |
5940 | } | |
5941 | invalidate_bs(); | |
81eedcae | 5942 | } // for loop |
31f18b77 FG |
5943 | |
5944 | if (r < 0) { | |
5945 | return r; | |
5946 | } | |
5947 | ||
5948 | if (pbs) { | |
5949 | *pbs = bs; | |
5950 | } | |
5951 | ||
5952 | return 0; | |
5953 | } | |
5954 | ||
9f95a23c | 5955 | int RGWRados::Bucket::UpdateIndex::prepare(RGWModifyOp op, const string *write_tag, optional_yield y) |
7c673cae FG |
5956 | { |
5957 | if (blind) { | |
5958 | return 0; | |
5959 | } | |
5960 | RGWRados *store = target->get_store(); | |
7c673cae FG |
5961 | |
5962 | if (write_tag && write_tag->length()) { | |
5963 | optag = string(write_tag->c_str(), write_tag->length()); | |
5964 | } else { | |
5965 | if (optag.empty()) { | |
5966 | append_rand_alpha(store->ctx(), optag, optag, 32); | |
5967 | } | |
5968 | } | |
5969 | ||
f64942e4 | 5970 | int r = guard_reshard(nullptr, [&](BucketShard *bs) -> int { |
9f95a23c | 5971 | return store->cls_obj_prepare_op(*bs, op, optag, obj, bilog_flags, y, zones_trace); |
f64942e4 | 5972 | }); |
31f18b77 | 5973 | |
7c673cae FG |
5974 | if (r < 0) { |
5975 | return r; | |
5976 | } | |
5977 | prepared = true; | |
31f18b77 | 5978 | |
7c673cae FG |
5979 | return 0; |
5980 | } | |
5981 | ||
5982 | int RGWRados::Bucket::UpdateIndex::complete(int64_t poolid, uint64_t epoch, | |
5983 | uint64_t size, uint64_t accounted_size, | |
5984 | ceph::real_time& ut, const string& etag, | |
11fdf7f2 | 5985 | const string& content_type, const string& storage_class, |
7c673cae FG |
5986 | bufferlist *acl_bl, |
5987 | RGWObjCategory category, | |
11fdf7f2 TL |
5988 | list<rgw_obj_index_key> *remove_objs, const string *user_data, |
5989 | bool appendable) | |
7c673cae FG |
5990 | { |
5991 | if (blind) { | |
5992 | return 0; | |
5993 | } | |
5994 | RGWRados *store = target->get_store(); | |
5995 | BucketShard *bs; | |
31f18b77 | 5996 | |
7c673cae FG |
5997 | int ret = get_bucket_shard(&bs); |
5998 | if (ret < 0) { | |
5999 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
6000 | return ret; | |
6001 | } | |
6002 | ||
6003 | rgw_bucket_dir_entry ent; | |
6004 | obj.key.get_index_key(&ent.key); | |
6005 | ent.meta.size = size; | |
6006 | ent.meta.accounted_size = accounted_size; | |
6007 | ent.meta.mtime = ut; | |
6008 | ent.meta.etag = etag; | |
11fdf7f2 | 6009 | ent.meta.storage_class = storage_class; |
7c673cae FG |
6010 | if (user_data) |
6011 | ent.meta.user_data = *user_data; | |
6012 | ||
6013 | ACLOwner owner; | |
6014 | if (acl_bl && acl_bl->length()) { | |
6015 | int ret = store->decode_policy(*acl_bl, &owner); | |
6016 | if (ret < 0) { | |
6017 | ldout(store->ctx(), 0) << "WARNING: could not decode policy ret=" << ret << dendl; | |
6018 | } | |
6019 | } | |
6020 | ent.meta.owner = owner.get_id().to_str(); | |
6021 | ent.meta.owner_display_name = owner.get_display_name(); | |
6022 | ent.meta.content_type = content_type; | |
11fdf7f2 | 6023 | ent.meta.appendable = appendable; |
7c673cae | 6024 | |
31f18b77 | 6025 | ret = store->cls_obj_complete_add(*bs, obj, optag, poolid, epoch, ent, category, remove_objs, bilog_flags, zones_trace); |
7c673cae | 6026 | |
9f95a23c TL |
6027 | int r = store->svc.datalog_rados->add_entry(target->bucket_info, bs->shard_id); |
6028 | if (r < 0) { | |
6029 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
7c673cae FG |
6030 | } |
6031 | ||
6032 | return ret; | |
6033 | } | |
6034 | ||
6035 | int RGWRados::Bucket::UpdateIndex::complete_del(int64_t poolid, uint64_t epoch, | |
6036 | real_time& removed_mtime, | |
6037 | list<rgw_obj_index_key> *remove_objs) | |
6038 | { | |
6039 | if (blind) { | |
6040 | return 0; | |
6041 | } | |
6042 | RGWRados *store = target->get_store(); | |
6043 | BucketShard *bs; | |
31f18b77 | 6044 | |
7c673cae FG |
6045 | int ret = get_bucket_shard(&bs); |
6046 | if (ret < 0) { | |
6047 | ldout(store->ctx(), 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
6048 | return ret; | |
6049 | } | |
6050 | ||
31f18b77 | 6051 | ret = store->cls_obj_complete_del(*bs, optag, poolid, epoch, obj, removed_mtime, remove_objs, bilog_flags, zones_trace); |
7c673cae | 6052 | |
9f95a23c TL |
6053 | int r = store->svc.datalog_rados->add_entry(target->bucket_info, bs->shard_id); |
6054 | if (r < 0) { | |
6055 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
7c673cae FG |
6056 | } |
6057 | ||
6058 | return ret; | |
6059 | } | |
6060 | ||
6061 | ||
6062 | int RGWRados::Bucket::UpdateIndex::cancel() | |
6063 | { | |
6064 | if (blind) { | |
6065 | return 0; | |
6066 | } | |
6067 | RGWRados *store = target->get_store(); | |
6068 | BucketShard *bs; | |
7c673cae | 6069 | |
f64942e4 AA |
6070 | int ret = guard_reshard(&bs, [&](BucketShard *bs) -> int { |
6071 | return store->cls_obj_complete_cancel(*bs, optag, obj, bilog_flags, zones_trace); | |
6072 | }); | |
7c673cae FG |
6073 | |
6074 | /* | |
6075 | * need to update data log anyhow, so that whoever follows needs to update its internal markers | |
6076 | * for following the specific bucket shard log. Otherwise they end up staying behind, and users | |
6077 | * have no way to tell that they're all caught up | |
6078 | */ | |
9f95a23c TL |
6079 | int r = store->svc.datalog_rados->add_entry(target->bucket_info, bs->shard_id); |
6080 | if (r < 0) { | |
6081 | lderr(store->ctx()) << "ERROR: failed writing data log" << dendl; | |
7c673cae FG |
6082 | } |
6083 | ||
6084 | return ret; | |
6085 | } | |
6086 | ||
9f95a23c | 6087 | int RGWRados::Object::Read::read(int64_t ofs, int64_t end, bufferlist& bl, optional_yield y) |
7c673cae FG |
6088 | { |
6089 | RGWRados *store = source->get_store(); | |
6090 | CephContext *cct = store->ctx(); | |
6091 | ||
7c673cae FG |
6092 | rgw_raw_obj read_obj; |
6093 | uint64_t read_ofs = ofs; | |
6094 | uint64_t len, read_len; | |
6095 | bool reading_from_head = true; | |
6096 | ObjectReadOperation op; | |
6097 | ||
6098 | bool merge_bl = false; | |
6099 | bufferlist *pbl = &bl; | |
6100 | bufferlist read_bl; | |
6101 | uint64_t max_chunk_size; | |
6102 | ||
6103 | RGWObjState *astate; | |
9f95a23c | 6104 | int r = source->get_state(&astate, true, y); |
7c673cae FG |
6105 | if (r < 0) |
6106 | return r; | |
6107 | ||
11fdf7f2 TL |
6108 | if (astate->size == 0) { |
6109 | end = 0; | |
6110 | } else if (end >= (int64_t)astate->size) { | |
6111 | end = astate->size - 1; | |
6112 | } | |
6113 | ||
7c673cae FG |
6114 | if (end < 0) |
6115 | len = 0; | |
6116 | else | |
6117 | len = end - ofs + 1; | |
6118 | ||
9f95a23c | 6119 | if (astate->manifest && astate->manifest->has_tail()) { |
7c673cae | 6120 | /* now get the relevant object part */ |
9f95a23c | 6121 | RGWObjManifest::obj_iterator iter = astate->manifest->obj_find(ofs); |
7c673cae FG |
6122 | |
6123 | uint64_t stripe_ofs = iter.get_stripe_ofs(); | |
6124 | read_obj = iter.get_location().get_raw_obj(store); | |
11fdf7f2 | 6125 | len = std::min(len, iter.get_stripe_size() - (ofs - stripe_ofs)); |
7c673cae FG |
6126 | read_ofs = iter.location_ofs() + (ofs - stripe_ofs); |
6127 | reading_from_head = (read_obj == state.head_obj); | |
6128 | } else { | |
6129 | read_obj = state.head_obj; | |
6130 | } | |
6131 | ||
6132 | r = store->get_max_chunk_size(read_obj.pool, &max_chunk_size); | |
6133 | if (r < 0) { | |
6134 | ldout(cct, 0) << "ERROR: failed to get max_chunk_size() for pool " << read_obj.pool << dendl; | |
6135 | return r; | |
6136 | } | |
6137 | ||
6138 | if (len > max_chunk_size) | |
6139 | len = max_chunk_size; | |
6140 | ||
6141 | ||
7c673cae FG |
6142 | read_len = len; |
6143 | ||
6144 | if (reading_from_head) { | |
6145 | /* only when reading from the head object do we need to do the atomic test */ | |
9f95a23c | 6146 | r = store->append_atomic_test(&source->get_ctx(), source->get_bucket_info(), state.obj, op, &astate, y); |
7c673cae FG |
6147 | if (r < 0) |
6148 | return r; | |
6149 | ||
6150 | if (astate && astate->prefetch_data) { | |
6151 | if (!ofs && astate->data.length() >= len) { | |
6152 | bl = astate->data; | |
6153 | return bl.length(); | |
6154 | } | |
6155 | ||
6156 | if (ofs < astate->data.length()) { | |
11fdf7f2 | 6157 | unsigned copy_len = std::min((uint64_t)astate->data.length() - ofs, len); |
9f95a23c | 6158 | astate->data.begin(ofs).copy(copy_len, bl); |
7c673cae FG |
6159 | read_len -= copy_len; |
6160 | read_ofs += copy_len; | |
6161 | if (!read_len) | |
6162 | return bl.length(); | |
6163 | ||
6164 | merge_bl = true; | |
6165 | pbl = &read_bl; | |
6166 | } | |
6167 | } | |
6168 | } | |
6169 | ||
6170 | ldout(cct, 20) << "rados->read obj-ofs=" << ofs << " read_ofs=" << read_ofs << " read_len=" << read_len << dendl; | |
6171 | op.read(read_ofs, read_len, pbl, NULL); | |
6172 | ||
11fdf7f2 TL |
6173 | if (state.cur_pool != read_obj.pool) { |
6174 | auto iter = state.io_ctxs.find(read_obj.pool); | |
6175 | if (iter == state.io_ctxs.end()) { | |
6176 | state.cur_ioctx = &state.io_ctxs[read_obj.pool]; | |
494da23a | 6177 | r = store->open_pool_ctx(read_obj.pool, *state.cur_ioctx, false); |
11fdf7f2 TL |
6178 | if (r < 0) { |
6179 | ldout(cct, 20) << "ERROR: failed to open pool context for pool=" << read_obj.pool << " r=" << r << dendl; | |
6180 | return r; | |
6181 | } | |
6182 | } else { | |
6183 | state.cur_ioctx = &iter->second; | |
7c673cae | 6184 | } |
11fdf7f2 | 6185 | state.cur_pool = read_obj.pool; |
7c673cae FG |
6186 | } |
6187 | ||
11fdf7f2 | 6188 | state.cur_ioctx->locator_set_key(read_obj.loc); |
7c673cae | 6189 | |
11fdf7f2 TL |
6190 | r = state.cur_ioctx->operate(read_obj.oid, &op, NULL); |
6191 | ldout(cct, 20) << "rados->read r=" << r << " bl.length=" << bl.length() << dendl; | |
7c673cae | 6192 | |
7c673cae | 6193 | if (r < 0) { |
7c673cae FG |
6194 | return r; |
6195 | } | |
7c673cae | 6196 | |
11fdf7f2 TL |
6197 | if (merge_bl) { |
6198 | bl.append(read_bl); | |
7c673cae FG |
6199 | } |
6200 | ||
7c673cae FG |
6201 | return bl.length(); |
6202 | } | |
6203 | ||
11fdf7f2 TL |
6204 | struct get_obj_data { |
6205 | RGWRados* store; | |
6206 | RGWGetDataCB* client_cb; | |
6207 | rgw::Aio* aio; | |
6208 | uint64_t offset; // next offset to write to client | |
6209 | rgw::AioResultList completed; // completed read results, sorted by offset | |
9f95a23c | 6210 | optional_yield yield; |
7c673cae | 6211 | |
9f95a23c TL |
6212 | get_obj_data(RGWRados* store, RGWGetDataCB* cb, rgw::Aio* aio, |
6213 | uint64_t offset, optional_yield yield) | |
6214 | : store(store), client_cb(cb), aio(aio), offset(offset), yield(yield) {} | |
7c673cae | 6215 | |
11fdf7f2 TL |
6216 | int flush(rgw::AioResultList&& results) { |
6217 | int r = rgw::check_for_errors(results); | |
6218 | if (r < 0) { | |
6219 | return r; | |
7c673cae | 6220 | } |
7c673cae | 6221 | |
11fdf7f2 TL |
6222 | auto cmp = [](const auto& lhs, const auto& rhs) { return lhs.id < rhs.id; }; |
6223 | results.sort(cmp); // merge() requires results to be sorted first | |
6224 | completed.merge(results, cmp); // merge results in sorted order | |
7c673cae | 6225 | |
11fdf7f2 TL |
6226 | while (!completed.empty() && completed.front().id == offset) { |
6227 | auto bl = std::move(completed.front().data); | |
6228 | completed.pop_front_and_dispose(std::default_delete<rgw::AioResultEntry>{}); | |
7c673cae | 6229 | |
11fdf7f2 TL |
6230 | offset += bl.length(); |
6231 | int r = client_cb->handle_data(bl, 0, bl.length()); | |
6232 | if (r < 0) { | |
6233 | return r; | |
6234 | } | |
7c673cae | 6235 | } |
11fdf7f2 | 6236 | return 0; |
7c673cae FG |
6237 | } |
6238 | ||
11fdf7f2 TL |
6239 | void cancel() { |
6240 | // wait for all completions to drain and ignore the results | |
6241 | aio->drain(); | |
7c673cae FG |
6242 | } |
6243 | ||
11fdf7f2 TL |
6244 | int drain() { |
6245 | auto c = aio->wait(); | |
6246 | while (!c.empty()) { | |
6247 | int r = flush(std::move(c)); | |
7c673cae | 6248 | if (r < 0) { |
11fdf7f2 | 6249 | cancel(); |
7c673cae FG |
6250 | return r; |
6251 | } | |
11fdf7f2 | 6252 | c = aio->wait(); |
7c673cae | 6253 | } |
11fdf7f2 | 6254 | return flush(std::move(c)); |
7c673cae FG |
6255 | } |
6256 | }; | |
6257 | ||
11fdf7f2 TL |
6258 | static int _get_obj_iterate_cb(const rgw_raw_obj& read_obj, off_t obj_ofs, |
6259 | off_t read_ofs, off_t len, bool is_head_obj, | |
6260 | RGWObjState *astate, void *arg) | |
7c673cae FG |
6261 | { |
6262 | struct get_obj_data *d = (struct get_obj_data *)arg; | |
6263 | ||
11fdf7f2 TL |
6264 | return d->store->get_obj_iterate_cb(read_obj, obj_ofs, read_ofs, len, |
6265 | is_head_obj, astate, arg); | |
7c673cae FG |
6266 | } |
6267 | ||
11fdf7f2 TL |
6268 | int RGWRados::get_obj_iterate_cb(const rgw_raw_obj& read_obj, off_t obj_ofs, |
6269 | off_t read_ofs, off_t len, bool is_head_obj, | |
6270 | RGWObjState *astate, void *arg) | |
7c673cae | 6271 | { |
7c673cae FG |
6272 | ObjectReadOperation op; |
6273 | struct get_obj_data *d = (struct get_obj_data *)arg; | |
6274 | string oid, key; | |
7c673cae FG |
6275 | |
6276 | if (is_head_obj) { | |
6277 | /* only when reading from the head object do we need to do the atomic test */ | |
11fdf7f2 | 6278 | int r = append_atomic_test(astate, op); |
7c673cae FG |
6279 | if (r < 0) |
6280 | return r; | |
6281 | ||
6282 | if (astate && | |
6283 | obj_ofs < astate->data.length()) { | |
11fdf7f2 | 6284 | unsigned chunk_len = std::min((uint64_t)astate->data.length() - obj_ofs, (uint64_t)len); |
7c673cae | 6285 | |
7c673cae | 6286 | r = d->client_cb->handle_data(astate->data, obj_ofs, chunk_len); |
7c673cae FG |
6287 | if (r < 0) |
6288 | return r; | |
6289 | ||
7c673cae | 6290 | len -= chunk_len; |
11fdf7f2 | 6291 | d->offset += chunk_len; |
7c673cae FG |
6292 | read_ofs += chunk_len; |
6293 | obj_ofs += chunk_len; | |
6294 | if (!len) | |
6295 | return 0; | |
6296 | } | |
6297 | } | |
6298 | ||
11fdf7f2 TL |
6299 | auto obj = d->store->svc.rados->obj(read_obj); |
6300 | int r = obj.open(); | |
7c673cae | 6301 | if (r < 0) { |
11fdf7f2 TL |
6302 | ldout(cct, 4) << "failed to open rados context for " << read_obj << dendl; |
6303 | return r; | |
7c673cae FG |
6304 | } |
6305 | ||
11fdf7f2 TL |
6306 | ldout(cct, 20) << "rados->get_obj_iterate_cb oid=" << read_obj.oid << " obj-ofs=" << obj_ofs << " read_ofs=" << read_ofs << " len=" << len << dendl; |
6307 | op.read(read_ofs, len, nullptr, nullptr); | |
7c673cae | 6308 | |
11fdf7f2 TL |
6309 | const uint64_t cost = len; |
6310 | const uint64_t id = obj_ofs; // use logical object offset for sorting replies | |
7c673cae | 6311 | |
9f95a23c | 6312 | auto completed = d->aio->get(obj, rgw::Aio::librados_op(std::move(op), d->yield), cost, id); |
7c673cae | 6313 | |
11fdf7f2 | 6314 | return d->flush(std::move(completed)); |
7c673cae FG |
6315 | } |
6316 | ||
9f95a23c TL |
6317 | int RGWRados::Object::Read::iterate(int64_t ofs, int64_t end, RGWGetDataCB *cb, |
6318 | optional_yield y) | |
7c673cae FG |
6319 | { |
6320 | RGWRados *store = source->get_store(); | |
6321 | CephContext *cct = store->ctx(); | |
7c673cae | 6322 | RGWObjectCtx& obj_ctx = source->get_ctx(); |
11fdf7f2 TL |
6323 | const uint64_t chunk_size = cct->_conf->rgw_get_obj_max_req_size; |
6324 | const uint64_t window_size = cct->_conf->rgw_get_obj_window_size; | |
7c673cae | 6325 | |
9f95a23c TL |
6326 | auto aio = rgw::make_throttle(window_size, y); |
6327 | get_obj_data data(store, cb, &*aio, ofs, y); | |
7c673cae | 6328 | |
11fdf7f2 | 6329 | int r = store->iterate_obj(obj_ctx, source->get_bucket_info(), state.obj, |
9f95a23c | 6330 | ofs, end, chunk_size, _get_obj_iterate_cb, &data, y); |
7c673cae | 6331 | if (r < 0) { |
11fdf7f2 TL |
6332 | ldout(cct, 0) << "iterate_obj() failed with " << r << dendl; |
6333 | data.cancel(); // drain completions without writing back to client | |
6334 | return r; | |
7c673cae FG |
6335 | } |
6336 | ||
11fdf7f2 | 6337 | return data.drain(); |
7c673cae FG |
6338 | } |
6339 | ||
6340 | int RGWRados::iterate_obj(RGWObjectCtx& obj_ctx, | |
6341 | const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
11fdf7f2 | 6342 | off_t ofs, off_t end, uint64_t max_chunk_size, |
9f95a23c | 6343 | iterate_obj_cb cb, void *arg, optional_yield y) |
7c673cae FG |
6344 | { |
6345 | rgw_raw_obj head_obj; | |
6346 | rgw_raw_obj read_obj; | |
6347 | uint64_t read_ofs = ofs; | |
6348 | uint64_t len; | |
6349 | bool reading_from_head = true; | |
6350 | RGWObjState *astate = NULL; | |
6351 | ||
6352 | obj_to_raw(bucket_info.placement_rule, obj, &head_obj); | |
6353 | ||
9f95a23c | 6354 | int r = get_obj_state(&obj_ctx, bucket_info, obj, &astate, false, y); |
7c673cae FG |
6355 | if (r < 0) { |
6356 | return r; | |
6357 | } | |
6358 | ||
6359 | if (end < 0) | |
6360 | len = 0; | |
6361 | else | |
6362 | len = end - ofs + 1; | |
6363 | ||
9f95a23c | 6364 | if (astate->manifest) { |
7c673cae | 6365 | /* now get the relevant object stripe */ |
9f95a23c | 6366 | RGWObjManifest::obj_iterator iter = astate->manifest->obj_find(ofs); |
7c673cae | 6367 | |
9f95a23c | 6368 | RGWObjManifest::obj_iterator obj_end = astate->manifest->obj_end(); |
7c673cae FG |
6369 | |
6370 | for (; iter != obj_end && ofs <= end; ++iter) { | |
6371 | off_t stripe_ofs = iter.get_stripe_ofs(); | |
6372 | off_t next_stripe_ofs = stripe_ofs + iter.get_stripe_size(); | |
6373 | ||
6374 | while (ofs < next_stripe_ofs && ofs <= end) { | |
6375 | read_obj = iter.get_location().get_raw_obj(this); | |
11fdf7f2 | 6376 | uint64_t read_len = std::min(len, iter.get_stripe_size() - (ofs - stripe_ofs)); |
7c673cae FG |
6377 | read_ofs = iter.location_ofs() + (ofs - stripe_ofs); |
6378 | ||
6379 | if (read_len > max_chunk_size) { | |
6380 | read_len = max_chunk_size; | |
6381 | } | |
6382 | ||
6383 | reading_from_head = (read_obj == head_obj); | |
11fdf7f2 | 6384 | r = cb(read_obj, ofs, read_ofs, read_len, reading_from_head, astate, arg); |
7c673cae FG |
6385 | if (r < 0) { |
6386 | return r; | |
6387 | } | |
6388 | ||
6389 | len -= read_len; | |
6390 | ofs += read_len; | |
6391 | } | |
6392 | } | |
6393 | } else { | |
6394 | while (ofs <= end) { | |
6395 | read_obj = head_obj; | |
11fdf7f2 | 6396 | uint64_t read_len = std::min(len, max_chunk_size); |
7c673cae | 6397 | |
11fdf7f2 | 6398 | r = cb(read_obj, ofs, ofs, read_len, reading_from_head, astate, arg); |
7c673cae FG |
6399 | if (r < 0) { |
6400 | return r; | |
6401 | } | |
6402 | ||
6403 | len -= read_len; | |
6404 | ofs += read_len; | |
6405 | } | |
6406 | } | |
6407 | ||
6408 | return 0; | |
6409 | } | |
6410 | ||
6411 | int RGWRados::obj_operate(const RGWBucketInfo& bucket_info, const rgw_obj& obj, ObjectWriteOperation *op) | |
6412 | { | |
6413 | rgw_rados_ref ref; | |
6414 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
6415 | if (r < 0) { | |
6416 | return r; | |
6417 | } | |
6418 | ||
9f95a23c | 6419 | return rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, op, null_yield); |
7c673cae FG |
6420 | } |
6421 | ||
6422 | int RGWRados::obj_operate(const RGWBucketInfo& bucket_info, const rgw_obj& obj, ObjectReadOperation *op) | |
6423 | { | |
6424 | rgw_rados_ref ref; | |
6425 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
6426 | if (r < 0) { | |
6427 | return r; | |
6428 | } | |
6429 | ||
6430 | bufferlist outbl; | |
6431 | ||
9f95a23c | 6432 | return rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, op, &outbl, null_yield); |
7c673cae FG |
6433 | } |
6434 | ||
6435 | int RGWRados::olh_init_modification_impl(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& olh_obj, string *op_tag) | |
6436 | { | |
6437 | ObjectWriteOperation op; | |
6438 | ||
11fdf7f2 | 6439 | ceph_assert(olh_obj.key.instance.empty()); |
7c673cae FG |
6440 | |
6441 | bool has_tag = (state.exists && has_olh_tag(state.attrset)); | |
6442 | ||
6443 | if (!state.exists) { | |
6444 | op.create(true); | |
6445 | } else { | |
6446 | op.assert_exists(); | |
b32b8144 FG |
6447 | struct timespec mtime_ts = real_clock::to_timespec(state.mtime); |
6448 | op.mtime2(&mtime_ts); | |
7c673cae FG |
6449 | } |
6450 | ||
6451 | /* | |
6452 | * 3 possible cases: olh object doesn't exist, it exists as an olh, it exists as a regular object. | |
6453 | * If it exists as a regular object we'll need to transform it into an olh. We'll do it in two | |
6454 | * steps, first change its tag and set the olh pending attrs. Once write is done we'll need to | |
6455 | * truncate it, remove extra attrs, and send it to the garbage collection. The bucket index olh | |
6456 | * log will reflect that. | |
6457 | * | |
6458 | * Need to generate separate olh and obj tags, as olh can be colocated with object data. obj_tag | |
6459 | * is used for object data instance, olh_tag for olh instance. | |
6460 | */ | |
6461 | if (has_tag) { | |
6462 | /* guard against racing writes */ | |
6463 | bucket_index_guard_olh_op(state, op); | |
6464 | } | |
6465 | ||
6466 | if (!has_tag) { | |
6467 | /* obj tag */ | |
9f95a23c | 6468 | string obj_tag = gen_rand_alphanumeric_lower(cct, 32); |
11fdf7f2 | 6469 | |
7c673cae FG |
6470 | bufferlist bl; |
6471 | bl.append(obj_tag.c_str(), obj_tag.size()); | |
6472 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
6473 | ||
6474 | state.attrset[RGW_ATTR_ID_TAG] = bl; | |
6475 | state.obj_tag = bl; | |
6476 | ||
6477 | /* olh tag */ | |
9f95a23c | 6478 | string olh_tag = gen_rand_alphanumeric_lower(cct, 32); |
11fdf7f2 | 6479 | |
7c673cae FG |
6480 | bufferlist olh_bl; |
6481 | olh_bl.append(olh_tag.c_str(), olh_tag.size()); | |
6482 | op.setxattr(RGW_ATTR_OLH_ID_TAG, olh_bl); | |
6483 | ||
6484 | state.attrset[RGW_ATTR_OLH_ID_TAG] = olh_bl; | |
6485 | state.olh_tag = olh_bl; | |
6486 | state.is_olh = true; | |
6487 | ||
6488 | bufferlist verbl; | |
6489 | op.setxattr(RGW_ATTR_OLH_VER, verbl); | |
6490 | } | |
6491 | ||
6492 | bufferlist bl; | |
6493 | RGWOLHPendingInfo pending_info; | |
6494 | pending_info.time = real_clock::now(); | |
11fdf7f2 | 6495 | encode(pending_info, bl); |
7c673cae FG |
6496 | |
6497 | #define OLH_PENDING_TAG_LEN 32 | |
6498 | /* tag will start with current time epoch, this so that entries are sorted by time */ | |
6499 | char buf[32]; | |
6500 | utime_t ut(pending_info.time); | |
6501 | snprintf(buf, sizeof(buf), "%016llx", (unsigned long long)ut.sec()); | |
6502 | *op_tag = buf; | |
6503 | ||
9f95a23c | 6504 | string s = gen_rand_alphanumeric_lower(cct, OLH_PENDING_TAG_LEN - op_tag->size()); |
11fdf7f2 | 6505 | |
7c673cae FG |
6506 | op_tag->append(s); |
6507 | ||
6508 | string attr_name = RGW_ATTR_OLH_PENDING_PREFIX; | |
6509 | attr_name.append(*op_tag); | |
6510 | ||
6511 | op.setxattr(attr_name.c_str(), bl); | |
6512 | ||
11fdf7f2 | 6513 | int ret = obj_operate(bucket_info, olh_obj, &op); |
7c673cae FG |
6514 | if (ret < 0) { |
6515 | return ret; | |
6516 | } | |
6517 | ||
6518 | state.exists = true; | |
6519 | state.attrset[attr_name] = bl; | |
6520 | ||
6521 | return 0; | |
6522 | } | |
6523 | ||
6524 | int RGWRados::olh_init_modification(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj, string *op_tag) | |
6525 | { | |
6526 | int ret; | |
6527 | ||
6528 | ret = olh_init_modification_impl(bucket_info, state, obj, op_tag); | |
6529 | if (ret == -EEXIST) { | |
6530 | ret = -ECANCELED; | |
6531 | } | |
6532 | ||
6533 | return ret; | |
6534 | } | |
6535 | ||
f64942e4 AA |
6536 | int RGWRados::guard_reshard(BucketShard *bs, |
6537 | const rgw_obj& obj_instance, | |
6538 | const RGWBucketInfo& bucket_info, | |
6539 | std::function<int(BucketShard *)> call) | |
31f18b77 FG |
6540 | { |
6541 | rgw_obj obj; | |
6542 | const rgw_obj *pobj = &obj_instance; | |
6543 | int r; | |
6544 | ||
6545 | for (int i = 0; i < NUM_RESHARD_RETRIES; ++i) { | |
f64942e4 | 6546 | r = bs->init(pobj->bucket, *pobj, nullptr /* no RGWBucketInfo */); |
31f18b77 FG |
6547 | if (r < 0) { |
6548 | ldout(cct, 5) << "bs.init() returned ret=" << r << dendl; | |
6549 | return r; | |
6550 | } | |
6551 | r = call(bs); | |
6552 | if (r != -ERR_BUSY_RESHARDING) { | |
6553 | break; | |
6554 | } | |
6555 | ldout(cct, 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl; | |
6556 | string new_bucket_id; | |
11fdf7f2 | 6557 | r = block_while_resharding(bs, &new_bucket_id, bucket_info, null_yield); |
31f18b77 FG |
6558 | if (r == -ERR_BUSY_RESHARDING) { |
6559 | continue; | |
6560 | } | |
6561 | if (r < 0) { | |
6562 | return r; | |
6563 | } | |
6564 | ldout(cct, 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id << dendl; | |
6565 | i = 0; /* resharding is finished, make sure we can retry */ | |
6566 | ||
6567 | obj = *pobj; | |
6568 | obj.bucket.update_bucket_id(new_bucket_id); | |
6569 | pobj = &obj; | |
81eedcae | 6570 | } // for loop |
31f18b77 FG |
6571 | |
6572 | if (r < 0) { | |
6573 | return r; | |
6574 | } | |
6575 | ||
6576 | return 0; | |
6577 | } | |
6578 | ||
f64942e4 AA |
6579 | int RGWRados::block_while_resharding(RGWRados::BucketShard *bs, |
6580 | string *new_bucket_id, | |
11fdf7f2 TL |
6581 | const RGWBucketInfo& bucket_info, |
6582 | optional_yield y) | |
31f18b77 | 6583 | { |
11fdf7f2 TL |
6584 | int ret = 0; |
6585 | cls_rgw_bucket_instance_entry entry; | |
6586 | ||
81eedcae TL |
6587 | // since we want to run this recovery code from two distinct places, |
6588 | // let's just put it in a lambda so we can easily re-use; if the | |
6589 | // lambda successfully fetches a new bucket id, it sets | |
6590 | // new_bucket_id and returns 0, otherwise it returns a negative | |
6591 | // error code | |
6592 | auto fetch_new_bucket_id = | |
9f95a23c TL |
6593 | [this, &bucket_info](const std::string& log_tag, |
6594 | std::string* new_bucket_id) -> int { | |
81eedcae TL |
6595 | RGWBucketInfo fresh_bucket_info = bucket_info; |
6596 | int ret = try_refresh_bucket_info(fresh_bucket_info, nullptr); | |
6597 | if (ret < 0) { | |
6598 | ldout(cct, 0) << __func__ << | |
6599 | " ERROR: failed to refresh bucket info after reshard at " << | |
6600 | log_tag << ": " << cpp_strerror(-ret) << dendl; | |
6601 | return ret; | |
6602 | } | |
6603 | *new_bucket_id = fresh_bucket_info.bucket.bucket_id; | |
6604 | return 0; | |
6605 | }; | |
6606 | ||
6607 | constexpr int num_retries = 10; | |
6608 | for (int i = 1; i <= num_retries; i++) { // nb: 1-based for loop | |
9f95a23c TL |
6609 | auto& ref = bs->bucket_obj.get_ref(); |
6610 | ret = cls_rgw_get_bucket_resharding(ref.pool.ioctx(), ref.obj.oid, &entry); | |
81eedcae TL |
6611 | if (ret == -ENOENT) { |
6612 | return fetch_new_bucket_id("get_bucket_resharding_failed", new_bucket_id); | |
6613 | } else if (ret < 0) { | |
6614 | ldout(cct, 0) << __func__ << | |
6615 | " ERROR: failed to get bucket resharding : " << cpp_strerror(-ret) << | |
6616 | dendl; | |
11fdf7f2 TL |
6617 | return ret; |
6618 | } | |
81eedcae | 6619 | |
11fdf7f2 | 6620 | if (!entry.resharding_in_progress()) { |
81eedcae TL |
6621 | return fetch_new_bucket_id("get_bucket_resharding_succeeded", |
6622 | new_bucket_id); | |
11fdf7f2 | 6623 | } |
31f18b77 | 6624 | |
81eedcae TL |
6625 | ldout(cct, 20) << "NOTICE: reshard still in progress; " << |
6626 | (i < num_retries ? "retrying" : "too many retries") << dendl; | |
6627 | ||
6628 | if (i == num_retries) { | |
11fdf7f2 TL |
6629 | break; |
6630 | } | |
6631 | ||
6632 | // If bucket is erroneously marked as resharding (e.g., crash or | |
6633 | // other error) then fix it. If we can take the bucket reshard | |
6634 | // lock then it means no other resharding should be taking place, | |
6635 | // and we're free to clear the flags. | |
6636 | { | |
6637 | // since we expect to do this rarely, we'll do our work in a | |
6638 | // block and erase our work after each try | |
6639 | ||
9f95a23c | 6640 | RGWObjectCtx obj_ctx(this->store); |
11fdf7f2 TL |
6641 | const rgw_bucket& b = bs->bucket; |
6642 | std::string bucket_id = b.get_key(); | |
9f95a23c | 6643 | RGWBucketReshardLock reshard_lock(this->store, bucket_info, true); |
11fdf7f2 TL |
6644 | ret = reshard_lock.lock(); |
6645 | if (ret < 0) { | |
6646 | ldout(cct, 20) << __func__ << | |
6647 | " INFO: failed to take reshard lock for bucket " << | |
6648 | bucket_id << "; expected if resharding underway" << dendl; | |
6649 | } else { | |
6650 | ldout(cct, 10) << __func__ << | |
6651 | " INFO: was able to take reshard lock for bucket " << | |
6652 | bucket_id << dendl; | |
9f95a23c | 6653 | ret = RGWBucketReshard::clear_resharding(this->store, bucket_info); |
11fdf7f2 TL |
6654 | if (ret < 0) { |
6655 | reshard_lock.unlock(); | |
6656 | ldout(cct, 0) << __func__ << | |
6657 | " ERROR: failed to clear resharding flags for bucket " << | |
6658 | bucket_id << dendl; | |
6659 | } else { | |
6660 | reshard_lock.unlock(); | |
6661 | ldout(cct, 5) << __func__ << | |
6662 | " INFO: apparently successfully cleared resharding flags for " | |
6663 | "bucket " << bucket_id << dendl; | |
6664 | continue; // if we apparently succeed immediately test again | |
6665 | } // if clear resharding succeeded | |
6666 | } // if taking of lock succeeded | |
6667 | } // block to encapsulate recovery from incomplete reshard | |
6668 | ||
6669 | ret = reshard_wait->wait(y); | |
6670 | if (ret < 0) { | |
81eedcae TL |
6671 | ldout(cct, 0) << __func__ << |
6672 | " ERROR: bucket is still resharding, please retry" << dendl; | |
11fdf7f2 TL |
6673 | return ret; |
6674 | } | |
81eedcae TL |
6675 | } // for loop |
6676 | ||
6677 | ldout(cct, 0) << __func__ << | |
6678 | " ERROR: bucket is still resharding, please retry" << dendl; | |
11fdf7f2 | 6679 | return -ERR_BUSY_RESHARDING; |
31f18b77 FG |
6680 | } |
6681 | ||
7c673cae FG |
6682 | int RGWRados::bucket_index_link_olh(const RGWBucketInfo& bucket_info, RGWObjState& olh_state, const rgw_obj& obj_instance, |
6683 | bool delete_marker, | |
6684 | const string& op_tag, | |
6685 | struct rgw_bucket_dir_entry_meta *meta, | |
6686 | uint64_t olh_epoch, | |
91327a77 AA |
6687 | real_time unmod_since, bool high_precision_time, |
6688 | rgw_zone_set *_zones_trace, bool log_data_change) | |
7c673cae FG |
6689 | { |
6690 | rgw_rados_ref ref; | |
6691 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
6692 | if (r < 0) { | |
6693 | return r; | |
6694 | } | |
6695 | ||
31f18b77 FG |
6696 | rgw_zone_set zones_trace; |
6697 | if (_zones_trace) { | |
6698 | zones_trace = *_zones_trace; | |
7c673cae | 6699 | } |
9f95a23c | 6700 | zones_trace.insert(svc.zone->get_zone().id, bucket_info.bucket.get_key()); |
7c673cae | 6701 | |
31f18b77 FG |
6702 | BucketShard bs(this); |
6703 | ||
f64942e4 AA |
6704 | r = guard_reshard(&bs, obj_instance, bucket_info, |
6705 | [&](BucketShard *bs) -> int { | |
9f95a23c TL |
6706 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), obj_instance.key.instance); |
6707 | auto& ref = bs->bucket_obj.get_ref(); | |
f64942e4 AA |
6708 | librados::ObjectWriteOperation op; |
6709 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
9f95a23c TL |
6710 | cls_rgw_bucket_link_olh(op, key, olh_state.olh_tag, |
6711 | delete_marker, op_tag, meta, olh_epoch, | |
6712 | unmod_since, high_precision_time, | |
6713 | svc.zone->get_zone().log_data, zones_trace); | |
6714 | return rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
31f18b77 FG |
6715 | }); |
6716 | if (r < 0) { | |
9f95a23c | 6717 | ldout(cct, 20) << "rgw_rados_operate() after cls_rgw_bucket_link_olh() returned r=" << r << dendl; |
31f18b77 | 6718 | return r; |
7c673cae FG |
6719 | } |
6720 | ||
9f95a23c TL |
6721 | r = svc.datalog_rados->add_entry(bucket_info, bs.shard_id); |
6722 | if (r < 0) { | |
6723 | ldout(cct, 0) << "ERROR: failed writing data log" << dendl; | |
91327a77 AA |
6724 | } |
6725 | ||
7c673cae FG |
6726 | return 0; |
6727 | } | |
6728 | ||
6729 | void RGWRados::bucket_index_guard_olh_op(RGWObjState& olh_state, ObjectOperation& op) | |
6730 | { | |
6731 | ldout(cct, 20) << __func__ << "(): olh_state.olh_tag=" << string(olh_state.olh_tag.c_str(), olh_state.olh_tag.length()) << dendl; | |
6732 | op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_state.olh_tag); | |
6733 | } | |
6734 | ||
6735 | int RGWRados::bucket_index_unlink_instance(const RGWBucketInfo& bucket_info, const rgw_obj& obj_instance, | |
31f18b77 | 6736 | const string& op_tag, const string& olh_tag, uint64_t olh_epoch, rgw_zone_set *_zones_trace) |
7c673cae FG |
6737 | { |
6738 | rgw_rados_ref ref; | |
6739 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
6740 | if (r < 0) { | |
6741 | return r; | |
6742 | } | |
6743 | ||
31f18b77 FG |
6744 | rgw_zone_set zones_trace; |
6745 | if (_zones_trace) { | |
6746 | zones_trace = *_zones_trace; | |
7c673cae | 6747 | } |
9f95a23c | 6748 | zones_trace.insert(svc.zone->get_zone().id, bucket_info.bucket.get_key()); |
31f18b77 FG |
6749 | |
6750 | BucketShard bs(this); | |
7c673cae FG |
6751 | |
6752 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), obj_instance.key.instance); | |
f64942e4 AA |
6753 | r = guard_reshard(&bs, obj_instance, bucket_info, |
6754 | [&](BucketShard *bs) -> int { | |
9f95a23c | 6755 | auto& ref = bs->bucket_obj.get_ref(); |
f64942e4 AA |
6756 | librados::ObjectWriteOperation op; |
6757 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
9f95a23c TL |
6758 | cls_rgw_bucket_unlink_instance(op, key, op_tag, |
6759 | olh_tag, olh_epoch, svc.zone->get_zone().log_data, zones_trace); | |
6760 | return rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
31f18b77 FG |
6761 | }); |
6762 | if (r < 0) { | |
9f95a23c | 6763 | ldout(cct, 20) << "rgw_rados_operate() after cls_rgw_bucket_link_instance() returned r=" << r << dendl; |
31f18b77 | 6764 | return r; |
7c673cae FG |
6765 | } |
6766 | ||
6767 | return 0; | |
6768 | } | |
6769 | ||
6770 | int RGWRados::bucket_index_read_olh_log(const RGWBucketInfo& bucket_info, RGWObjState& state, | |
6771 | const rgw_obj& obj_instance, uint64_t ver_marker, | |
6772 | map<uint64_t, vector<rgw_bucket_olh_log_entry> > *log, | |
6773 | bool *is_truncated) | |
6774 | { | |
6775 | rgw_rados_ref ref; | |
6776 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
6777 | if (r < 0) { | |
6778 | return r; | |
6779 | } | |
6780 | ||
6781 | BucketShard bs(this); | |
f64942e4 AA |
6782 | int ret = |
6783 | bs.init(obj_instance.bucket, obj_instance, nullptr /* no RGWBucketInfo */); | |
7c673cae FG |
6784 | if (ret < 0) { |
6785 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
6786 | return ret; | |
6787 | } | |
6788 | ||
6789 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
6790 | ||
6791 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
6792 | ||
f64942e4 AA |
6793 | ret = guard_reshard(&bs, obj_instance, bucket_info, |
6794 | [&](BucketShard *bs) -> int { | |
9f95a23c | 6795 | auto& ref = bs->bucket_obj.get_ref(); |
f64942e4 AA |
6796 | ObjectReadOperation op; |
6797 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
9f95a23c TL |
6798 | |
6799 | rgw_cls_read_olh_log_ret log_ret; | |
6800 | int op_ret = 0; | |
6801 | cls_rgw_get_olh_log(op, key, ver_marker, olh_tag, log_ret, op_ret); | |
6802 | bufferlist outbl; | |
6803 | int r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, &outbl, null_yield); | |
6804 | if (r < 0) { | |
6805 | return r; | |
6806 | } | |
6807 | if (op_ret < 0) { | |
6808 | return op_ret; | |
6809 | } | |
6810 | ||
6811 | *log = std::move(log_ret.log); | |
6812 | *is_truncated = log_ret.is_truncated; | |
6813 | return r; | |
f64942e4 | 6814 | }); |
31f18b77 FG |
6815 | if (ret < 0) { |
6816 | ldout(cct, 20) << "cls_rgw_get_olh_log() returned r=" << r << dendl; | |
7c673cae | 6817 | return ret; |
31f18b77 | 6818 | } |
7c673cae FG |
6819 | |
6820 | return 0; | |
6821 | } | |
6822 | ||
a8e16298 TL |
6823 | // a multisite sync bug resulted in the OLH head attributes being overwritten by |
6824 | // the attributes from another zone, causing link_olh() to fail endlessly due to | |
6825 | // olh_tag mismatch. this attempts to detect this case and reconstruct the OLH | |
6826 | // attributes from the bucket index. see http://tracker.ceph.com/issues/37792 | |
6827 | int RGWRados::repair_olh(RGWObjState* state, const RGWBucketInfo& bucket_info, | |
6828 | const rgw_obj& obj) | |
6829 | { | |
6830 | // fetch the current olh entry from the bucket index | |
6831 | rgw_bucket_olh_entry olh; | |
6832 | int r = bi_get_olh(bucket_info, obj, &olh); | |
6833 | if (r < 0) { | |
6834 | ldout(cct, 0) << "repair_olh failed to read olh entry for " << obj << dendl; | |
6835 | return r; | |
6836 | } | |
11fdf7f2 | 6837 | if (olh.tag == rgw_bl_str(state->olh_tag)) { // mismatch already resolved? |
a8e16298 TL |
6838 | return 0; |
6839 | } | |
6840 | ||
6841 | ldout(cct, 4) << "repair_olh setting olh_tag=" << olh.tag | |
6842 | << " key=" << olh.key << " delete_marker=" << olh.delete_marker << dendl; | |
6843 | ||
6844 | // rewrite OLH_ID_TAG and OLH_INFO from current olh | |
6845 | ObjectWriteOperation op; | |
6846 | // assert this is the same olh tag we think we're fixing | |
6847 | bucket_index_guard_olh_op(*state, op); | |
6848 | // preserve existing mtime | |
6849 | struct timespec mtime_ts = ceph::real_clock::to_timespec(state->mtime); | |
6850 | op.mtime2(&mtime_ts); | |
6851 | { | |
6852 | bufferlist bl; | |
6853 | bl.append(olh.tag.c_str(), olh.tag.size()); | |
6854 | op.setxattr(RGW_ATTR_OLH_ID_TAG, bl); | |
6855 | } | |
6856 | { | |
6857 | RGWOLHInfo info; | |
6858 | info.target = rgw_obj(bucket_info.bucket, olh.key); | |
6859 | info.removed = olh.delete_marker; | |
6860 | bufferlist bl; | |
6861 | encode(info, bl); | |
6862 | op.setxattr(RGW_ATTR_OLH_INFO, bl); | |
6863 | } | |
6864 | rgw_rados_ref ref; | |
6865 | r = get_obj_head_ref(bucket_info, obj, &ref); | |
6866 | if (r < 0) { | |
6867 | return r; | |
6868 | } | |
9f95a23c | 6869 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); |
a8e16298 TL |
6870 | if (r < 0) { |
6871 | ldout(cct, 0) << "repair_olh failed to write olh attributes with " | |
6872 | << cpp_strerror(r) << dendl; | |
6873 | return r; | |
6874 | } | |
6875 | return 0; | |
6876 | } | |
6877 | ||
7c673cae FG |
6878 | int RGWRados::bucket_index_trim_olh_log(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj_instance, uint64_t ver) |
6879 | { | |
6880 | rgw_rados_ref ref; | |
6881 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
6882 | if (r < 0) { | |
6883 | return r; | |
6884 | } | |
6885 | ||
6886 | BucketShard bs(this); | |
f64942e4 AA |
6887 | int ret = |
6888 | bs.init(obj_instance.bucket, obj_instance, nullptr /* no RGWBucketInfo */); | |
7c673cae FG |
6889 | if (ret < 0) { |
6890 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
6891 | return ret; | |
6892 | } | |
6893 | ||
6894 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
6895 | ||
6896 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
6897 | ||
f64942e4 AA |
6898 | ret = guard_reshard(&bs, obj_instance, bucket_info, |
6899 | [&](BucketShard *pbs) -> int { | |
6900 | ObjectWriteOperation op; | |
6901 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
6902 | cls_rgw_trim_olh_log(op, key, ver, olh_tag); | |
9f95a23c | 6903 | return pbs->bucket_obj.operate(&op, null_yield); |
31f18b77 FG |
6904 | }); |
6905 | if (ret < 0) { | |
6906 | ldout(cct, 20) << "cls_rgw_trim_olh_log() returned r=" << ret << dendl; | |
7c673cae | 6907 | return ret; |
31f18b77 | 6908 | } |
7c673cae FG |
6909 | |
6910 | return 0; | |
6911 | } | |
6912 | ||
6913 | int RGWRados::bucket_index_clear_olh(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj_instance) | |
6914 | { | |
6915 | rgw_rados_ref ref; | |
6916 | int r = get_obj_head_ref(bucket_info, obj_instance, &ref); | |
6917 | if (r < 0) { | |
6918 | return r; | |
6919 | } | |
6920 | ||
6921 | BucketShard bs(this); | |
7c673cae FG |
6922 | |
6923 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
6924 | ||
6925 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
6926 | ||
f64942e4 AA |
6927 | int ret = guard_reshard(&bs, obj_instance, bucket_info, |
6928 | [&](BucketShard *pbs) -> int { | |
6929 | ObjectWriteOperation op; | |
9f95a23c | 6930 | auto& ref = pbs->bucket_obj.get_ref(); |
f64942e4 | 6931 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); |
9f95a23c TL |
6932 | cls_rgw_clear_olh(op, key, olh_tag); |
6933 | return rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
31f18b77 | 6934 | }); |
7c673cae | 6935 | if (ret < 0) { |
9f95a23c | 6936 | ldout(cct, 5) << "rgw_rados_operate() after cls_rgw_clear_olh() returned ret=" << ret << dendl; |
7c673cae FG |
6937 | return ret; |
6938 | } | |
6939 | ||
6940 | return 0; | |
6941 | } | |
6942 | ||
92f5a8d4 TL |
6943 | static int decode_olh_info(CephContext* cct, const bufferlist& bl, RGWOLHInfo *olh) |
6944 | { | |
6945 | try { | |
6946 | auto biter = bl.cbegin(); | |
6947 | decode(*olh, biter); | |
6948 | return 0; | |
6949 | } catch (buffer::error& err) { | |
6950 | ldout(cct, 0) << "ERROR: failed to decode olh info" << dendl; | |
6951 | return -EIO; | |
6952 | } | |
6953 | } | |
6954 | ||
7c673cae FG |
6955 | int RGWRados::apply_olh_log(RGWObjectCtx& obj_ctx, RGWObjState& state, const RGWBucketInfo& bucket_info, const rgw_obj& obj, |
6956 | bufferlist& olh_tag, map<uint64_t, vector<rgw_bucket_olh_log_entry> >& log, | |
31f18b77 | 6957 | uint64_t *plast_ver, rgw_zone_set* zones_trace) |
7c673cae FG |
6958 | { |
6959 | if (log.empty()) { | |
6960 | return 0; | |
6961 | } | |
6962 | ||
6963 | librados::ObjectWriteOperation op; | |
6964 | ||
6965 | uint64_t last_ver = log.rbegin()->first; | |
6966 | *plast_ver = last_ver; | |
6967 | ||
6968 | map<uint64_t, vector<rgw_bucket_olh_log_entry> >::iterator iter = log.begin(); | |
6969 | ||
6970 | op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_tag); | |
92f5a8d4 | 6971 | op.cmpxattr(RGW_ATTR_OLH_VER, CEPH_OSD_CMPXATTR_OP_GTE, last_ver); |
7c673cae | 6972 | |
a8e16298 TL |
6973 | bufferlist ver_bl; |
6974 | string last_ver_s = to_string(last_ver); | |
6975 | ver_bl.append(last_ver_s.c_str(), last_ver_s.size()); | |
6976 | op.setxattr(RGW_ATTR_OLH_VER, ver_bl); | |
6977 | ||
b32b8144 FG |
6978 | struct timespec mtime_ts = real_clock::to_timespec(state.mtime); |
6979 | op.mtime2(&mtime_ts); | |
6980 | ||
7c673cae | 6981 | bool need_to_link = false; |
92f5a8d4 | 6982 | uint64_t link_epoch = 0; |
7c673cae FG |
6983 | cls_rgw_obj_key key; |
6984 | bool delete_marker = false; | |
6985 | list<cls_rgw_obj_key> remove_instances; | |
6986 | bool need_to_remove = false; | |
6987 | ||
92f5a8d4 TL |
6988 | // decode current epoch and instance |
6989 | auto olh_ver = state.attrset.find(RGW_ATTR_OLH_VER); | |
6990 | if (olh_ver != state.attrset.end()) { | |
6991 | std::string str = olh_ver->second.to_str(); | |
6992 | std::string err; | |
6993 | link_epoch = strict_strtoll(str.c_str(), 10, &err); | |
6994 | } | |
6995 | auto olh_info = state.attrset.find(RGW_ATTR_OLH_INFO); | |
6996 | if (olh_info != state.attrset.end()) { | |
6997 | RGWOLHInfo info; | |
6998 | int r = decode_olh_info(cct, olh_info->second, &info); | |
6999 | if (r < 0) { | |
7000 | return r; | |
7001 | } | |
7002 | info.target.key.get_index_key(&key); | |
7003 | delete_marker = info.removed; | |
7004 | } | |
7005 | ||
7c673cae FG |
7006 | for (iter = log.begin(); iter != log.end(); ++iter) { |
7007 | vector<rgw_bucket_olh_log_entry>::iterator viter = iter->second.begin(); | |
7008 | for (; viter != iter->second.end(); ++viter) { | |
7009 | rgw_bucket_olh_log_entry& entry = *viter; | |
7010 | ||
92f5a8d4 | 7011 | ldout(cct, 20) << "olh_log_entry: epoch=" << iter->first << " op=" << (int)entry.op |
7c673cae FG |
7012 | << " key=" << entry.key.name << "[" << entry.key.instance << "] " |
7013 | << (entry.delete_marker ? "(delete)" : "") << dendl; | |
7014 | switch (entry.op) { | |
7015 | case CLS_RGW_OLH_OP_REMOVE_INSTANCE: | |
7016 | remove_instances.push_back(entry.key); | |
7017 | break; | |
7018 | case CLS_RGW_OLH_OP_LINK_OLH: | |
92f5a8d4 TL |
7019 | // only overwrite a link of the same epoch if its key sorts before |
7020 | if (link_epoch < iter->first || key.instance.empty() || | |
7021 | key.instance > entry.key.instance) { | |
7022 | ldout(cct, 20) << "apply_olh_log applying key=" << entry.key << " epoch=" << iter->first << " delete_marker=" << entry.delete_marker | |
7023 | << " over current=" << key << " epoch=" << link_epoch << " delete_marker=" << delete_marker << dendl; | |
7024 | need_to_link = true; | |
7025 | need_to_remove = false; | |
7026 | key = entry.key; | |
7027 | delete_marker = entry.delete_marker; | |
7028 | } else { | |
7029 | ldout(cct, 20) << "apply_olh skipping key=" << entry.key<< " epoch=" << iter->first << " delete_marker=" << entry.delete_marker | |
7030 | << " before current=" << key << " epoch=" << link_epoch << " delete_marker=" << delete_marker << dendl; | |
7031 | } | |
7c673cae FG |
7032 | break; |
7033 | case CLS_RGW_OLH_OP_UNLINK_OLH: | |
7034 | need_to_remove = true; | |
7035 | need_to_link = false; | |
7036 | break; | |
7037 | default: | |
7038 | ldout(cct, 0) << "ERROR: apply_olh_log: invalid op: " << (int)entry.op << dendl; | |
7039 | return -EIO; | |
7040 | } | |
7041 | string attr_name = RGW_ATTR_OLH_PENDING_PREFIX; | |
7042 | attr_name.append(entry.op_tag); | |
7043 | op.rmxattr(attr_name.c_str()); | |
7044 | } | |
7045 | } | |
7046 | ||
7047 | rgw_rados_ref ref; | |
7048 | int r = get_obj_head_ref(bucket_info, obj, &ref); | |
7049 | if (r < 0) { | |
7050 | return r; | |
7051 | } | |
7052 | ||
7053 | const rgw_bucket& bucket = obj.bucket; | |
7054 | ||
7055 | if (need_to_link) { | |
7056 | rgw_obj target(bucket, key); | |
7057 | RGWOLHInfo info; | |
7058 | info.target = target; | |
7059 | info.removed = delete_marker; | |
7060 | bufferlist bl; | |
11fdf7f2 | 7061 | encode(info, bl); |
7c673cae FG |
7062 | op.setxattr(RGW_ATTR_OLH_INFO, bl); |
7063 | } | |
7064 | ||
7065 | /* first remove object instances */ | |
7066 | for (list<cls_rgw_obj_key>::iterator liter = remove_instances.begin(); | |
7067 | liter != remove_instances.end(); ++liter) { | |
7068 | cls_rgw_obj_key& key = *liter; | |
7069 | rgw_obj obj_instance(bucket, key); | |
31f18b77 | 7070 | int ret = delete_obj(obj_ctx, bucket_info, obj_instance, 0, RGW_BILOG_FLAG_VERSIONED_OP, ceph::real_time(), zones_trace); |
7c673cae FG |
7071 | if (ret < 0 && ret != -ENOENT) { |
7072 | ldout(cct, 0) << "ERROR: delete_obj() returned " << ret << " obj_instance=" << obj_instance << dendl; | |
7073 | return ret; | |
7074 | } | |
7075 | } | |
7076 | ||
7077 | /* update olh object */ | |
9f95a23c | 7078 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); |
7c673cae FG |
7079 | if (r == -ECANCELED) { |
7080 | r = 0; | |
7081 | } | |
7082 | if (r < 0) { | |
7083 | ldout(cct, 0) << "ERROR: could not apply olh update, r=" << r << dendl; | |
7084 | return r; | |
7085 | } | |
7086 | ||
7087 | r = bucket_index_trim_olh_log(bucket_info, state, obj, last_ver); | |
7088 | if (r < 0) { | |
7089 | ldout(cct, 0) << "ERROR: could not trim olh log, r=" << r << dendl; | |
7090 | return r; | |
7091 | } | |
7092 | ||
7093 | if (need_to_remove) { | |
7094 | ObjectWriteOperation rm_op; | |
7095 | ||
7096 | rm_op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_tag); | |
a8e16298 | 7097 | rm_op.cmpxattr(RGW_ATTR_OLH_VER, CEPH_OSD_CMPXATTR_OP_EQ, last_ver); |
7c673cae FG |
7098 | cls_obj_check_prefix_exist(rm_op, RGW_ATTR_OLH_PENDING_PREFIX, true); /* fail if found one of these, pending modification */ |
7099 | rm_op.remove(); | |
7100 | ||
9f95a23c | 7101 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &rm_op, null_yield); |
7c673cae FG |
7102 | if (r == -ECANCELED) { |
7103 | return 0; /* someone else won this race */ | |
7104 | } else { | |
7105 | /* | |
7106 | * only clear if was successful, otherwise we might clobber pending operations on this object | |
7107 | */ | |
7108 | r = bucket_index_clear_olh(bucket_info, state, obj); | |
7109 | if (r < 0) { | |
7110 | ldout(cct, 0) << "ERROR: could not clear bucket index olh entries r=" << r << dendl; | |
7111 | return r; | |
7112 | } | |
7113 | } | |
7114 | } | |
7115 | ||
7116 | return 0; | |
7117 | } | |
7118 | ||
7119 | /* | |
7120 | * read olh log and apply it | |
7121 | */ | |
31f18b77 | 7122 | int RGWRados::update_olh(RGWObjectCtx& obj_ctx, RGWObjState *state, const RGWBucketInfo& bucket_info, const rgw_obj& obj, rgw_zone_set *zones_trace) |
7c673cae FG |
7123 | { |
7124 | map<uint64_t, vector<rgw_bucket_olh_log_entry> > log; | |
7125 | bool is_truncated; | |
7126 | uint64_t ver_marker = 0; | |
7127 | ||
7128 | do { | |
7129 | int ret = bucket_index_read_olh_log(bucket_info, *state, obj, ver_marker, &log, &is_truncated); | |
7130 | if (ret < 0) { | |
7131 | return ret; | |
7132 | } | |
31f18b77 | 7133 | ret = apply_olh_log(obj_ctx, *state, bucket_info, obj, state->olh_tag, log, &ver_marker, zones_trace); |
7c673cae FG |
7134 | if (ret < 0) { |
7135 | return ret; | |
7136 | } | |
7137 | } while (is_truncated); | |
7138 | ||
7139 | return 0; | |
7140 | } | |
7141 | ||
9f95a23c | 7142 | int RGWRados::set_olh(RGWObjectCtx& obj_ctx, const RGWBucketInfo& bucket_info, const rgw_obj& target_obj, bool delete_marker, rgw_bucket_dir_entry_meta *meta, |
91327a77 | 7143 | uint64_t olh_epoch, real_time unmod_since, bool high_precision_time, |
9f95a23c | 7144 | optional_yield y, rgw_zone_set *zones_trace, bool log_data_change) |
7c673cae FG |
7145 | { |
7146 | string op_tag; | |
7147 | ||
7148 | rgw_obj olh_obj = target_obj; | |
7149 | olh_obj.key.instance.clear(); | |
7150 | ||
7151 | RGWObjState *state = NULL; | |
7152 | ||
7153 | int ret = 0; | |
7154 | int i; | |
31f18b77 | 7155 | |
7c673cae FG |
7156 | #define MAX_ECANCELED_RETRY 100 |
7157 | for (i = 0; i < MAX_ECANCELED_RETRY; i++) { | |
7158 | if (ret == -ECANCELED) { | |
11fdf7f2 | 7159 | obj_ctx.invalidate(olh_obj); |
7c673cae FG |
7160 | } |
7161 | ||
9f95a23c | 7162 | ret = get_obj_state(&obj_ctx, bucket_info, olh_obj, &state, false, y); /* don't follow olh */ |
7c673cae FG |
7163 | if (ret < 0) { |
7164 | return ret; | |
7165 | } | |
7166 | ||
7167 | ret = olh_init_modification(bucket_info, *state, olh_obj, &op_tag); | |
7168 | if (ret < 0) { | |
7169 | ldout(cct, 20) << "olh_init_modification() target_obj=" << target_obj << " delete_marker=" << (int)delete_marker << " returned " << ret << dendl; | |
7170 | if (ret == -ECANCELED) { | |
7171 | continue; | |
7172 | } | |
7173 | return ret; | |
7174 | } | |
91327a77 AA |
7175 | ret = bucket_index_link_olh(bucket_info, *state, target_obj, delete_marker, |
7176 | op_tag, meta, olh_epoch, unmod_since, high_precision_time, | |
7177 | zones_trace, log_data_change); | |
7c673cae FG |
7178 | if (ret < 0) { |
7179 | ldout(cct, 20) << "bucket_index_link_olh() target_obj=" << target_obj << " delete_marker=" << (int)delete_marker << " returned " << ret << dendl; | |
7180 | if (ret == -ECANCELED) { | |
a8e16298 TL |
7181 | // the bucket index rejected the link_olh() due to olh tag mismatch; |
7182 | // attempt to reconstruct olh head attributes based on the bucket index | |
7183 | int r2 = repair_olh(state, bucket_info, olh_obj); | |
7184 | if (r2 < 0 && r2 != -ECANCELED) { | |
7185 | return r2; | |
7186 | } | |
7c673cae FG |
7187 | continue; |
7188 | } | |
7189 | return ret; | |
7190 | } | |
7191 | break; | |
7192 | } | |
7193 | ||
7194 | if (i == MAX_ECANCELED_RETRY) { | |
7195 | ldout(cct, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl; | |
7196 | return -EIO; | |
7197 | } | |
7198 | ||
7199 | ret = update_olh(obj_ctx, state, bucket_info, olh_obj); | |
7200 | if (ret == -ECANCELED) { /* already did what we needed, no need to retry, raced with another user */ | |
7201 | ret = 0; | |
7202 | } | |
7203 | if (ret < 0) { | |
7204 | ldout(cct, 20) << "update_olh() target_obj=" << target_obj << " returned " << ret << dendl; | |
7205 | return ret; | |
7206 | } | |
7207 | ||
7208 | return 0; | |
7209 | } | |
7210 | ||
7211 | int RGWRados::unlink_obj_instance(RGWObjectCtx& obj_ctx, RGWBucketInfo& bucket_info, const rgw_obj& target_obj, | |
9f95a23c | 7212 | uint64_t olh_epoch, optional_yield y, rgw_zone_set *zones_trace) |
7c673cae FG |
7213 | { |
7214 | string op_tag; | |
7215 | ||
7216 | rgw_obj olh_obj = target_obj; | |
7217 | olh_obj.key.instance.clear(); | |
7218 | ||
7219 | RGWObjState *state = NULL; | |
7220 | ||
7221 | int ret = 0; | |
7222 | int i; | |
7223 | ||
7224 | for (i = 0; i < MAX_ECANCELED_RETRY; i++) { | |
7225 | if (ret == -ECANCELED) { | |
11fdf7f2 | 7226 | obj_ctx.invalidate(olh_obj); |
7c673cae FG |
7227 | } |
7228 | ||
9f95a23c | 7229 | ret = get_obj_state(&obj_ctx, bucket_info, olh_obj, &state, false, y); /* don't follow olh */ |
7c673cae FG |
7230 | if (ret < 0) |
7231 | return ret; | |
7232 | ||
7233 | ret = olh_init_modification(bucket_info, *state, olh_obj, &op_tag); | |
7234 | if (ret < 0) { | |
7235 | ldout(cct, 20) << "olh_init_modification() target_obj=" << target_obj << " returned " << ret << dendl; | |
7236 | if (ret == -ECANCELED) { | |
7237 | continue; | |
7238 | } | |
7239 | return ret; | |
7240 | } | |
7241 | ||
7242 | string olh_tag(state->olh_tag.c_str(), state->olh_tag.length()); | |
7243 | ||
31f18b77 | 7244 | ret = bucket_index_unlink_instance(bucket_info, target_obj, op_tag, olh_tag, olh_epoch, zones_trace); |
7c673cae FG |
7245 | if (ret < 0) { |
7246 | ldout(cct, 20) << "bucket_index_unlink_instance() target_obj=" << target_obj << " returned " << ret << dendl; | |
7247 | if (ret == -ECANCELED) { | |
7248 | continue; | |
7249 | } | |
7250 | return ret; | |
7251 | } | |
7252 | break; | |
7253 | } | |
7254 | ||
7255 | if (i == MAX_ECANCELED_RETRY) { | |
7256 | ldout(cct, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl; | |
7257 | return -EIO; | |
7258 | } | |
7259 | ||
31f18b77 | 7260 | ret = update_olh(obj_ctx, state, bucket_info, olh_obj, zones_trace); |
7c673cae FG |
7261 | if (ret == -ECANCELED) { /* already did what we needed, no need to retry, raced with another user */ |
7262 | return 0; | |
7263 | } | |
7264 | if (ret < 0) { | |
7265 | ldout(cct, 20) << "update_olh() target_obj=" << target_obj << " returned " << ret << dendl; | |
7266 | return ret; | |
7267 | } | |
7268 | ||
7269 | return 0; | |
7270 | } | |
7271 | ||
11fdf7f2 | 7272 | void RGWRados::gen_rand_obj_instance_name(rgw_obj_key *target_key) |
7c673cae FG |
7273 | { |
7274 | #define OBJ_INSTANCE_LEN 32 | |
7275 | char buf[OBJ_INSTANCE_LEN + 1]; | |
7276 | ||
7277 | gen_rand_alphanumeric_no_underscore(cct, buf, OBJ_INSTANCE_LEN); /* don't want it to get url escaped, | |
7278 | no underscore for instance name due to the way we encode the raw keys */ | |
7279 | ||
11fdf7f2 | 7280 | target_key->set_instance(buf); |
7c673cae FG |
7281 | } |
7282 | ||
11fdf7f2 | 7283 | void RGWRados::gen_rand_obj_instance_name(rgw_obj *target_obj) |
7c673cae | 7284 | { |
11fdf7f2 | 7285 | gen_rand_obj_instance_name(&target_obj->key); |
7c673cae FG |
7286 | } |
7287 | ||
7288 | int RGWRados::get_olh(const RGWBucketInfo& bucket_info, const rgw_obj& obj, RGWOLHInfo *olh) | |
7289 | { | |
92f5a8d4 | 7290 | map<string, bufferlist> attrset; |
7c673cae FG |
7291 | |
7292 | ObjectReadOperation op; | |
92f5a8d4 | 7293 | op.getxattrs(&attrset, NULL); |
7c673cae | 7294 | |
7c673cae | 7295 | int r = obj_operate(bucket_info, obj, &op); |
7c673cae FG |
7296 | if (r < 0) { |
7297 | return r; | |
7298 | } | |
7c673cae | 7299 | |
92f5a8d4 | 7300 | auto iter = attrset.find(RGW_ATTR_OLH_INFO); |
7c673cae FG |
7301 | if (iter == attrset.end()) { /* not an olh */ |
7302 | return -EINVAL; | |
7303 | } | |
7304 | ||
92f5a8d4 | 7305 | return decode_olh_info(cct, iter->second, olh); |
7c673cae FG |
7306 | } |
7307 | ||
7308 | void RGWRados::check_pending_olh_entries(map<string, bufferlist>& pending_entries, | |
7309 | map<string, bufferlist> *rm_pending_entries) | |
7310 | { | |
7311 | map<string, bufferlist>::iterator iter = pending_entries.begin(); | |
7312 | ||
7313 | real_time now = real_clock::now(); | |
7314 | ||
7315 | while (iter != pending_entries.end()) { | |
11fdf7f2 | 7316 | auto biter = iter->second.cbegin(); |
7c673cae FG |
7317 | RGWOLHPendingInfo pending_info; |
7318 | try { | |
11fdf7f2 | 7319 | decode(pending_info, biter); |
7c673cae FG |
7320 | } catch (buffer::error& err) { |
7321 | /* skipping bad entry, we could remove it but it might hide a bug */ | |
7322 | ldout(cct, 0) << "ERROR: failed to decode pending entry " << iter->first << dendl; | |
7323 | ++iter; | |
7324 | continue; | |
7325 | } | |
7326 | ||
7327 | map<string, bufferlist>::iterator cur_iter = iter; | |
7328 | ++iter; | |
7329 | if (now - pending_info.time >= make_timespan(cct->_conf->rgw_olh_pending_timeout_sec)) { | |
7330 | (*rm_pending_entries)[cur_iter->first] = cur_iter->second; | |
7331 | pending_entries.erase(cur_iter); | |
7332 | } else { | |
7333 | /* entries names are sorted by time (rounded to a second) */ | |
7334 | break; | |
7335 | } | |
7336 | } | |
7337 | } | |
7338 | ||
7339 | int RGWRados::remove_olh_pending_entries(const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& olh_obj, map<string, bufferlist>& pending_attrs) | |
7340 | { | |
7c673cae FG |
7341 | rgw_rados_ref ref; |
7342 | int r = get_obj_head_ref(bucket_info, olh_obj, &ref); | |
7343 | if (r < 0) { | |
7344 | return r; | |
7345 | } | |
7346 | ||
81eedcae TL |
7347 | // trim no more than 1000 entries per osd op |
7348 | constexpr int max_entries = 1000; | |
7c673cae | 7349 | |
81eedcae TL |
7350 | auto i = pending_attrs.begin(); |
7351 | while (i != pending_attrs.end()) { | |
7352 | ObjectWriteOperation op; | |
7353 | bucket_index_guard_olh_op(state, op); | |
7354 | ||
7355 | for (int n = 0; n < max_entries && i != pending_attrs.end(); ++n, ++i) { | |
7356 | op.rmxattr(i->first.c_str()); | |
7357 | } | |
7358 | ||
9f95a23c | 7359 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); |
81eedcae TL |
7360 | if (r == -ENOENT || r == -ECANCELED) { |
7361 | /* raced with some other change, shouldn't sweat about it */ | |
7362 | return 0; | |
7363 | } | |
7364 | if (r < 0) { | |
7365 | ldout(cct, 0) << "ERROR: could not apply olh update, r=" << r << dendl; | |
7366 | return r; | |
7367 | } | |
7368 | } | |
7c673cae FG |
7369 | return 0; |
7370 | } | |
7371 | ||
7372 | int RGWRados::follow_olh(const RGWBucketInfo& bucket_info, RGWObjectCtx& obj_ctx, RGWObjState *state, const rgw_obj& olh_obj, rgw_obj *target) | |
7373 | { | |
7374 | map<string, bufferlist> pending_entries; | |
11fdf7f2 | 7375 | rgw_filter_attrset(state->attrset, RGW_ATTR_OLH_PENDING_PREFIX, &pending_entries); |
7c673cae FG |
7376 | |
7377 | map<string, bufferlist> rm_pending_entries; | |
7378 | check_pending_olh_entries(pending_entries, &rm_pending_entries); | |
7379 | ||
7380 | if (!rm_pending_entries.empty()) { | |
7381 | int ret = remove_olh_pending_entries(bucket_info, *state, olh_obj, rm_pending_entries); | |
7382 | if (ret < 0) { | |
7383 | ldout(cct, 20) << "ERROR: rm_pending_entries returned ret=" << ret << dendl; | |
7384 | return ret; | |
7385 | } | |
7386 | } | |
7387 | if (!pending_entries.empty()) { | |
7388 | ldout(cct, 20) << __func__ << "(): found pending entries, need to update_olh() on bucket=" << olh_obj.bucket << dendl; | |
7389 | ||
7390 | int ret = update_olh(obj_ctx, state, bucket_info, olh_obj); | |
7391 | if (ret < 0) { | |
7392 | return ret; | |
7393 | } | |
7394 | } | |
7395 | ||
92f5a8d4 TL |
7396 | auto iter = state->attrset.find(RGW_ATTR_OLH_INFO); |
7397 | if (iter == state->attrset.end()) { | |
7398 | return -EINVAL; | |
7399 | } | |
7400 | ||
7c673cae | 7401 | RGWOLHInfo olh; |
92f5a8d4 TL |
7402 | int ret = decode_olh_info(cct, iter->second, &olh); |
7403 | if (ret < 0) { | |
7404 | return ret; | |
7c673cae FG |
7405 | } |
7406 | ||
7407 | if (olh.removed) { | |
7408 | return -ENOENT; | |
7409 | } | |
7410 | ||
7411 | *target = olh.target; | |
7412 | ||
7413 | return 0; | |
7414 | } | |
7415 | ||
7416 | int RGWRados::raw_obj_stat(rgw_raw_obj& obj, uint64_t *psize, real_time *pmtime, uint64_t *epoch, | |
7417 | map<string, bufferlist> *attrs, bufferlist *first_chunk, | |
9f95a23c | 7418 | RGWObjVersionTracker *objv_tracker, optional_yield y) |
7c673cae FG |
7419 | { |
7420 | rgw_rados_ref ref; | |
7421 | int r = get_raw_obj_ref(obj, &ref); | |
7422 | if (r < 0) { | |
7423 | return r; | |
7424 | } | |
7425 | ||
7426 | map<string, bufferlist> unfiltered_attrset; | |
7427 | uint64_t size = 0; | |
7428 | struct timespec mtime_ts; | |
7429 | ||
7430 | ObjectReadOperation op; | |
7431 | if (objv_tracker) { | |
7432 | objv_tracker->prepare_op_for_read(&op); | |
7433 | } | |
7434 | if (attrs) { | |
7435 | op.getxattrs(&unfiltered_attrset, NULL); | |
7436 | } | |
7437 | if (psize || pmtime) { | |
7438 | op.stat2(&size, &mtime_ts, NULL); | |
7439 | } | |
7440 | if (first_chunk) { | |
7441 | op.read(0, cct->_conf->rgw_max_chunk_size, first_chunk, NULL); | |
7442 | } | |
7443 | bufferlist outbl; | |
9f95a23c | 7444 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, &outbl, null_yield); |
7c673cae FG |
7445 | |
7446 | if (epoch) { | |
9f95a23c | 7447 | *epoch = ref.pool.ioctx().get_last_version(); |
7c673cae FG |
7448 | } |
7449 | ||
7450 | if (r < 0) | |
7451 | return r; | |
7452 | ||
7453 | if (psize) | |
7454 | *psize = size; | |
7455 | if (pmtime) | |
7456 | *pmtime = ceph::real_clock::from_timespec(mtime_ts); | |
7457 | if (attrs) { | |
11fdf7f2 | 7458 | rgw_filter_attrset(unfiltered_attrset, RGW_ATTR_PREFIX, attrs); |
7c673cae FG |
7459 | } |
7460 | ||
7461 | return 0; | |
7462 | } | |
7463 | ||
7464 | int RGWRados::get_bucket_stats(RGWBucketInfo& bucket_info, int shard_id, string *bucket_ver, string *master_ver, | |
c07f9fc5 | 7465 | map<RGWObjCategory, RGWStorageStats>& stats, string *max_marker, bool *syncstopped) |
7c673cae | 7466 | { |
a8e16298 | 7467 | vector<rgw_bucket_dir_header> headers; |
7c673cae FG |
7468 | map<int, string> bucket_instance_ids; |
7469 | int r = cls_bucket_head(bucket_info, shard_id, headers, &bucket_instance_ids); | |
7470 | if (r < 0) { | |
7471 | return r; | |
7472 | } | |
7473 | ||
11fdf7f2 | 7474 | ceph_assert(headers.size() == bucket_instance_ids.size()); |
7c673cae | 7475 | |
a8e16298 | 7476 | auto iter = headers.begin(); |
7c673cae FG |
7477 | map<int, string>::iterator viter = bucket_instance_ids.begin(); |
7478 | BucketIndexShardsManager ver_mgr; | |
7479 | BucketIndexShardsManager master_ver_mgr; | |
7480 | BucketIndexShardsManager marker_mgr; | |
7c673cae FG |
7481 | char buf[64]; |
7482 | for(; iter != headers.end(); ++iter, ++viter) { | |
a8e16298 TL |
7483 | accumulate_raw_stats(*iter, stats); |
7484 | snprintf(buf, sizeof(buf), "%lu", (unsigned long)iter->ver); | |
7c673cae | 7485 | ver_mgr.add(viter->first, string(buf)); |
a8e16298 | 7486 | snprintf(buf, sizeof(buf), "%lu", (unsigned long)iter->master_ver); |
7c673cae FG |
7487 | master_ver_mgr.add(viter->first, string(buf)); |
7488 | if (shard_id >= 0) { | |
a8e16298 | 7489 | *max_marker = iter->max_marker; |
7c673cae | 7490 | } else { |
a8e16298 | 7491 | marker_mgr.add(viter->first, iter->max_marker); |
7c673cae | 7492 | } |
c07f9fc5 | 7493 | if (syncstopped != NULL) |
a8e16298 | 7494 | *syncstopped = iter->syncstopped; |
7c673cae FG |
7495 | } |
7496 | ver_mgr.to_string(bucket_ver); | |
7497 | master_ver_mgr.to_string(master_ver); | |
7498 | if (shard_id < 0) { | |
7499 | marker_mgr.to_string(max_marker); | |
7500 | } | |
7501 | return 0; | |
7502 | } | |
7503 | ||
7c673cae FG |
7504 | class RGWGetBucketStatsContext : public RGWGetDirHeader_CB { |
7505 | RGWGetBucketStats_CB *cb; | |
7506 | uint32_t pendings; | |
7507 | map<RGWObjCategory, RGWStorageStats> stats; | |
7508 | int ret_code; | |
7509 | bool should_cb; | |
9f95a23c | 7510 | ceph::mutex lock = ceph::make_mutex("RGWGetBucketStatsContext"); |
7c673cae FG |
7511 | |
7512 | public: | |
7513 | RGWGetBucketStatsContext(RGWGetBucketStats_CB *_cb, uint32_t _pendings) | |
9f95a23c TL |
7514 | : cb(_cb), pendings(_pendings), stats(), ret_code(0), should_cb(true) |
7515 | {} | |
7c673cae FG |
7516 | |
7517 | void handle_response(int r, rgw_bucket_dir_header& header) override { | |
9f95a23c | 7518 | std::lock_guard l{lock}; |
7c673cae FG |
7519 | if (should_cb) { |
7520 | if ( r >= 0) { | |
7521 | accumulate_raw_stats(header, stats); | |
7522 | } else { | |
7523 | ret_code = r; | |
7524 | } | |
7525 | ||
7526 | // Are we all done? | |
7527 | if (--pendings == 0) { | |
7528 | if (!ret_code) { | |
7529 | cb->set_response(&stats); | |
7530 | } | |
7531 | cb->handle_response(ret_code); | |
7532 | cb->put(); | |
7533 | } | |
7534 | } | |
7535 | } | |
7536 | ||
7537 | void unset_cb() { | |
9f95a23c | 7538 | std::lock_guard l{lock}; |
7c673cae FG |
7539 | should_cb = false; |
7540 | } | |
7541 | }; | |
7542 | ||
7543 | int RGWRados::get_bucket_stats_async(RGWBucketInfo& bucket_info, int shard_id, RGWGetBucketStats_CB *ctx) | |
7544 | { | |
7545 | int num_aio = 0; | |
c07f9fc5 | 7546 | RGWGetBucketStatsContext *get_ctx = new RGWGetBucketStatsContext(ctx, bucket_info.num_shards ? : 1); |
11fdf7f2 | 7547 | ceph_assert(get_ctx); |
7c673cae | 7548 | int r = cls_bucket_head_async(bucket_info, shard_id, get_ctx, &num_aio); |
7c673cae FG |
7549 | if (r < 0) { |
7550 | ctx->put(); | |
7551 | if (num_aio) { | |
7552 | get_ctx->unset_cb(); | |
7553 | } | |
7554 | } | |
c07f9fc5 | 7555 | get_ctx->put(); |
7c673cae FG |
7556 | return r; |
7557 | } | |
7558 | ||
e306af50 TL |
7559 | int RGWRados::get_bucket_instance_info(RGWSysObjectCtx& obj_ctx, |
7560 | const string& meta_key, | |
7561 | RGWBucketInfo& info, | |
7562 | real_time *pmtime, | |
7563 | map<string, bufferlist> *pattrs, | |
7564 | optional_yield y) | |
9f95a23c TL |
7565 | { |
7566 | rgw_bucket bucket; | |
7567 | rgw_bucket_parse_bucket_key(cct, meta_key, &bucket, nullptr); | |
7c673cae | 7568 | |
9f95a23c TL |
7569 | return get_bucket_instance_info(obj_ctx, bucket, info, pmtime, pattrs, y); |
7570 | } | |
7c673cae | 7571 | |
11fdf7f2 | 7572 | int RGWRados::get_bucket_instance_info(RGWSysObjectCtx& obj_ctx, const rgw_bucket& bucket, RGWBucketInfo& info, |
9f95a23c | 7573 | real_time *pmtime, map<string, bufferlist> *pattrs, optional_yield y) |
7c673cae | 7574 | { |
9f95a23c TL |
7575 | RGWSI_MetaBackend_CtxParams bectx_params = RGWSI_MetaBackend_CtxParams_SObj(&obj_ctx); |
7576 | return ctl.bucket->read_bucket_instance_info(bucket, &info, | |
7577 | y, | |
7578 | RGWBucketCtl::BucketInstance::GetParams() | |
7579 | .set_mtime(pmtime) | |
7580 | .set_attrs(pattrs) | |
7581 | .set_bectx_params(bectx_params)); | |
7c673cae FG |
7582 | } |
7583 | ||
9f95a23c | 7584 | int RGWRados::get_bucket_info(RGWServices *svc, |
b32b8144 FG |
7585 | const string& tenant, const string& bucket_name, |
7586 | RGWBucketInfo& info, | |
9f95a23c TL |
7587 | real_time *pmtime, |
7588 | optional_yield y, map<string, bufferlist> *pattrs) | |
b32b8144 | 7589 | { |
9f95a23c TL |
7590 | auto obj_ctx = svc->sysobj->init_obj_ctx(); |
7591 | RGWSI_MetaBackend_CtxParams bectx_params = RGWSI_MetaBackend_CtxParams_SObj(&obj_ctx); | |
7592 | rgw_bucket bucket; | |
7593 | bucket.tenant = tenant; | |
7594 | bucket.name = bucket_name; | |
7595 | return ctl.bucket->read_bucket_info(bucket, &info, y, | |
7596 | RGWBucketCtl::BucketInstance::GetParams() | |
7597 | .set_mtime(pmtime) | |
7598 | .set_attrs(pattrs) | |
7599 | .set_bectx_params(bectx_params)); | |
b32b8144 FG |
7600 | } |
7601 | ||
7602 | int RGWRados::try_refresh_bucket_info(RGWBucketInfo& info, | |
7603 | ceph::real_time *pmtime, | |
7604 | map<string, bufferlist> *pattrs) | |
7605 | { | |
9f95a23c TL |
7606 | rgw_bucket bucket = info.bucket; |
7607 | bucket.bucket_id.clear(); | |
b32b8144 | 7608 | |
9f95a23c | 7609 | auto rv = info.objv_tracker.read_version; |
b32b8144 | 7610 | |
9f95a23c TL |
7611 | return ctl.bucket->read_bucket_info(bucket, &info, null_yield, |
7612 | RGWBucketCtl::BucketInstance::GetParams() | |
7613 | .set_mtime(pmtime) | |
7614 | .set_attrs(pattrs) | |
7615 | .set_refresh_version(rv)); | |
7c673cae FG |
7616 | } |
7617 | ||
7618 | int RGWRados::put_bucket_instance_info(RGWBucketInfo& info, bool exclusive, | |
7619 | real_time mtime, map<string, bufferlist> *pattrs) | |
7620 | { | |
9f95a23c TL |
7621 | return ctl.bucket->store_bucket_instance_info(info.bucket, info, null_yield, |
7622 | RGWBucketCtl::BucketInstance::PutParams() | |
7623 | .set_exclusive(exclusive) | |
7624 | .set_mtime(mtime) | |
7625 | .set_attrs(pattrs)); | |
7c673cae FG |
7626 | } |
7627 | ||
7628 | int RGWRados::put_linked_bucket_info(RGWBucketInfo& info, bool exclusive, real_time mtime, obj_version *pep_objv, | |
7629 | map<string, bufferlist> *pattrs, bool create_entry_point) | |
7630 | { | |
7631 | bool create_head = !info.has_instance_obj || create_entry_point; | |
7632 | ||
7633 | int ret = put_bucket_instance_info(info, exclusive, mtime, pattrs); | |
7634 | if (ret < 0) { | |
7635 | return ret; | |
7636 | } | |
7637 | ||
7638 | if (!create_head) | |
7639 | return 0; /* done! */ | |
7640 | ||
7641 | RGWBucketEntryPoint entry_point; | |
7642 | entry_point.bucket = info.bucket; | |
7643 | entry_point.owner = info.owner; | |
7644 | entry_point.creation_time = info.creation_time; | |
7645 | entry_point.linked = true; | |
7646 | RGWObjVersionTracker ot; | |
7647 | if (pep_objv && !pep_objv->tag.empty()) { | |
7648 | ot.write_version = *pep_objv; | |
7649 | } else { | |
7650 | ot.generate_new_write_ver(cct); | |
7651 | if (pep_objv) { | |
7652 | *pep_objv = ot.write_version; | |
7653 | } | |
7654 | } | |
9f95a23c TL |
7655 | ret = ctl.bucket->store_bucket_entrypoint_info(info.bucket, entry_point, null_yield, RGWBucketCtl::Bucket::PutParams() |
7656 | .set_exclusive(exclusive) | |
7657 | .set_objv_tracker(&ot) | |
7658 | .set_mtime(mtime)); | |
7c673cae FG |
7659 | if (ret < 0) |
7660 | return ret; | |
7661 | ||
7662 | return 0; | |
7663 | } | |
7664 | ||
7c673cae FG |
7665 | int RGWRados::update_containers_stats(map<string, RGWBucketEnt>& m) |
7666 | { | |
11fdf7f2 | 7667 | auto obj_ctx = svc.sysobj->init_obj_ctx(); |
7c673cae FG |
7668 | |
7669 | map<string, RGWBucketEnt>::iterator iter; | |
7670 | for (iter = m.begin(); iter != m.end(); ++iter) { | |
7671 | RGWBucketEnt& ent = iter->second; | |
7672 | rgw_bucket& bucket = ent.bucket; | |
7673 | ent.count = 0; | |
7674 | ent.size = 0; | |
7675 | ent.size_rounded = 0; | |
7676 | ||
a8e16298 | 7677 | vector<rgw_bucket_dir_header> headers; |
7c673cae FG |
7678 | |
7679 | RGWBucketInfo bucket_info; | |
9f95a23c | 7680 | int ret = get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL, null_yield); |
7c673cae FG |
7681 | if (ret < 0) { |
7682 | return ret; | |
7683 | } | |
7684 | ||
7685 | int r = cls_bucket_head(bucket_info, RGW_NO_SHARD, headers); | |
7686 | if (r < 0) | |
7687 | return r; | |
7688 | ||
a8e16298 | 7689 | auto hiter = headers.begin(); |
7c673cae FG |
7690 | for (; hiter != headers.end(); ++hiter) { |
7691 | RGWObjCategory category = main_category; | |
11fdf7f2 | 7692 | auto iter = (hiter->stats).find(category); |
a8e16298 | 7693 | if (iter != hiter->stats.end()) { |
7c673cae FG |
7694 | struct rgw_bucket_category_stats& stats = iter->second; |
7695 | ent.count += stats.num_entries; | |
7696 | ent.size += stats.total_size; | |
7697 | ent.size_rounded += stats.total_size_rounded; | |
7698 | } | |
7699 | } | |
3efd9988 FG |
7700 | |
7701 | // fill in placement_rule from the bucket instance for use in swift's | |
7702 | // per-storage policy statistics | |
7703 | ent.placement_rule = std::move(bucket_info.placement_rule); | |
7c673cae FG |
7704 | } |
7705 | ||
7706 | return m.size(); | |
7707 | } | |
7708 | ||
7709 | int RGWRados::append_async(rgw_raw_obj& obj, size_t size, bufferlist& bl) | |
7710 | { | |
7711 | rgw_rados_ref ref; | |
7712 | int r = get_raw_obj_ref(obj, &ref); | |
7713 | if (r < 0) { | |
7714 | return r; | |
7715 | } | |
7716 | librados::Rados *rad = get_rados_handle(); | |
9f95a23c | 7717 | librados::AioCompletion *completion = rad->aio_create_completion(nullptr, nullptr); |
7c673cae | 7718 | |
9f95a23c | 7719 | r = ref.pool.ioctx().aio_append(ref.obj.oid, completion, bl, size); |
7c673cae FG |
7720 | completion->release(); |
7721 | return r; | |
7722 | } | |
7723 | ||
7c673cae FG |
7724 | int RGWRados::pool_iterate_begin(const rgw_pool& pool, RGWPoolIterCtx& ctx) |
7725 | { | |
7726 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
7727 | librados::NObjectIterator& iter = ctx.iter; | |
7728 | ||
494da23a | 7729 | int r = open_pool_ctx(pool, io_ctx, false); |
7c673cae FG |
7730 | if (r < 0) |
7731 | return r; | |
7732 | ||
7733 | iter = io_ctx.nobjects_begin(); | |
7734 | ||
7735 | return 0; | |
7736 | } | |
7737 | ||
181888fb FG |
7738 | int RGWRados::pool_iterate_begin(const rgw_pool& pool, const string& cursor, RGWPoolIterCtx& ctx) |
7739 | { | |
7740 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
7741 | librados::NObjectIterator& iter = ctx.iter; | |
7742 | ||
494da23a | 7743 | int r = open_pool_ctx(pool, io_ctx, false); |
181888fb FG |
7744 | if (r < 0) |
7745 | return r; | |
7746 | ||
7747 | librados::ObjectCursor oc; | |
7748 | if (!oc.from_str(cursor)) { | |
7749 | ldout(cct, 10) << "failed to parse cursor: " << cursor << dendl; | |
7750 | return -EINVAL; | |
7751 | } | |
7752 | ||
f64942e4 AA |
7753 | try { |
7754 | iter = io_ctx.nobjects_begin(oc); | |
7755 | return 0; | |
7756 | } catch (const std::system_error& e) { | |
7757 | r = -e.code().value(); | |
7758 | ldout(cct, 10) << "nobjects_begin threw " << e.what() | |
7759 | << ", returning " << r << dendl; | |
7760 | return r; | |
7761 | } catch (const std::exception& e) { | |
7762 | ldout(cct, 10) << "nobjects_begin threw " << e.what() | |
7763 | << ", returning -5" << dendl; | |
7764 | return -EIO; | |
7765 | } | |
181888fb FG |
7766 | } |
7767 | ||
7768 | string RGWRados::pool_iterate_get_cursor(RGWPoolIterCtx& ctx) | |
7769 | { | |
7770 | return ctx.iter.get_cursor().to_str(); | |
7771 | } | |
7772 | ||
f64942e4 AA |
7773 | static int do_pool_iterate(CephContext* cct, RGWPoolIterCtx& ctx, uint32_t num, |
7774 | vector<rgw_bucket_dir_entry>& objs, | |
7c673cae FG |
7775 | bool *is_truncated, RGWAccessListFilter *filter) |
7776 | { | |
7777 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
7778 | librados::NObjectIterator& iter = ctx.iter; | |
7779 | ||
7780 | if (iter == io_ctx.nobjects_end()) | |
7781 | return -ENOENT; | |
7782 | ||
7783 | uint32_t i; | |
7784 | ||
7785 | for (i = 0; i < num && iter != io_ctx.nobjects_end(); ++i, ++iter) { | |
7786 | rgw_bucket_dir_entry e; | |
7787 | ||
7788 | string oid = iter->get_oid(); | |
7789 | ldout(cct, 20) << "RGWRados::pool_iterate: got " << oid << dendl; | |
7790 | ||
7791 | // fill it in with initial values; we may correct later | |
7792 | if (filter && !filter->filter(oid, oid)) | |
7793 | continue; | |
7794 | ||
7795 | e.key = oid; | |
7796 | objs.push_back(e); | |
7797 | } | |
7798 | ||
7799 | if (is_truncated) | |
7800 | *is_truncated = (iter != io_ctx.nobjects_end()); | |
7801 | ||
7802 | return objs.size(); | |
7803 | } | |
7c673cae | 7804 | |
f64942e4 AA |
7805 | int RGWRados::pool_iterate(RGWPoolIterCtx& ctx, uint32_t num, vector<rgw_bucket_dir_entry>& objs, |
7806 | bool *is_truncated, RGWAccessListFilter *filter) | |
7807 | { | |
7808 | // catch exceptions from NObjectIterator::operator++() | |
7809 | try { | |
7810 | return do_pool_iterate(cct, ctx, num, objs, is_truncated, filter); | |
7811 | } catch (const std::system_error& e) { | |
7812 | int r = -e.code().value(); | |
7813 | ldout(cct, 10) << "NObjectIterator threw exception " << e.what() | |
7814 | << ", returning " << r << dendl; | |
7815 | return r; | |
7816 | } catch (const std::exception& e) { | |
7817 | ldout(cct, 10) << "NObjectIterator threw exception " << e.what() | |
7818 | << ", returning -5" << dendl; | |
7819 | return -EIO; | |
7820 | } | |
7821 | } | |
7822 | ||
181888fb | 7823 | int RGWRados::list_raw_objects_init(const rgw_pool& pool, const string& marker, RGWListRawObjsCtx *ctx) |
7c673cae | 7824 | { |
181888fb FG |
7825 | if (!ctx->initialized) { |
7826 | int r = pool_iterate_begin(pool, marker, ctx->iter_ctx); | |
7c673cae FG |
7827 | if (r < 0) { |
7828 | ldout(cct, 10) << "failed to list objects pool_iterate_begin() returned r=" << r << dendl; | |
7829 | return r; | |
7830 | } | |
181888fb | 7831 | ctx->initialized = true; |
7c673cae | 7832 | } |
181888fb FG |
7833 | return 0; |
7834 | } | |
7c673cae | 7835 | |
181888fb FG |
7836 | int RGWRados::list_raw_objects_next(const string& prefix_filter, int max, |
7837 | RGWListRawObjsCtx& ctx, list<string>& oids, | |
7838 | bool *is_truncated) | |
7839 | { | |
7840 | if (!ctx.initialized) { | |
7841 | return -EINVAL; | |
7842 | } | |
7843 | RGWAccessListFilterPrefix filter(prefix_filter); | |
7c673cae FG |
7844 | vector<rgw_bucket_dir_entry> objs; |
7845 | int r = pool_iterate(ctx.iter_ctx, max, objs, is_truncated, &filter); | |
7846 | if (r < 0) { | |
7847 | if(r != -ENOENT) | |
7848 | ldout(cct, 10) << "failed to list objects pool_iterate returned r=" << r << dendl; | |
7849 | return r; | |
7850 | } | |
7851 | ||
7852 | vector<rgw_bucket_dir_entry>::iterator iter; | |
7853 | for (iter = objs.begin(); iter != objs.end(); ++iter) { | |
7854 | oids.push_back(iter->key.name); | |
7855 | } | |
7856 | ||
7857 | return oids.size(); | |
7858 | } | |
7859 | ||
181888fb FG |
7860 | int RGWRados::list_raw_objects(const rgw_pool& pool, const string& prefix_filter, |
7861 | int max, RGWListRawObjsCtx& ctx, list<string>& oids, | |
7862 | bool *is_truncated) | |
7863 | { | |
7864 | if (!ctx.initialized) { | |
7865 | int r = list_raw_objects_init(pool, string(), &ctx); | |
7866 | if (r < 0) { | |
7867 | return r; | |
7868 | } | |
7869 | } | |
7870 | ||
7871 | return list_raw_objects_next(prefix_filter, max, ctx, oids, is_truncated); | |
7872 | } | |
7873 | ||
7874 | string RGWRados::list_raw_objs_get_cursor(RGWListRawObjsCtx& ctx) | |
7875 | { | |
7876 | return pool_iterate_get_cursor(ctx.iter_ctx); | |
7877 | } | |
7878 | ||
a8e16298 TL |
7879 | int RGWRados::bi_get_instance(const RGWBucketInfo& bucket_info, const rgw_obj& obj, |
7880 | rgw_bucket_dir_entry *dirent) | |
7c673cae | 7881 | { |
a8e16298 | 7882 | rgw_cls_bi_entry bi_entry; |
11fdf7f2 | 7883 | int r = bi_get(bucket_info, obj, BIIndexType::Instance, &bi_entry); |
a8e16298 TL |
7884 | if (r < 0 && r != -ENOENT) { |
7885 | ldout(cct, 0) << "ERROR: bi_get() returned r=" << r << dendl; | |
7886 | } | |
7c673cae FG |
7887 | if (r < 0) { |
7888 | return r; | |
7889 | } | |
11fdf7f2 | 7890 | auto iter = bi_entry.data.cbegin(); |
a8e16298 | 7891 | try { |
11fdf7f2 | 7892 | decode(*dirent, iter); |
a8e16298 TL |
7893 | } catch (buffer::error& err) { |
7894 | ldout(cct, 0) << "ERROR: failed to decode bi_entry()" << dendl; | |
7895 | return -EIO; | |
7896 | } | |
7897 | ||
7898 | return 0; | |
7899 | } | |
7c673cae | 7900 | |
a8e16298 TL |
7901 | int RGWRados::bi_get_olh(const RGWBucketInfo& bucket_info, const rgw_obj& obj, |
7902 | rgw_bucket_olh_entry *olh) | |
7903 | { | |
7c673cae | 7904 | rgw_cls_bi_entry bi_entry; |
11fdf7f2 | 7905 | int r = bi_get(bucket_info, obj, BIIndexType::OLH, &bi_entry); |
7c673cae FG |
7906 | if (r < 0 && r != -ENOENT) { |
7907 | ldout(cct, 0) << "ERROR: bi_get() returned r=" << r << dendl; | |
7908 | } | |
7909 | if (r < 0) { | |
7910 | return r; | |
7911 | } | |
11fdf7f2 | 7912 | auto iter = bi_entry.data.cbegin(); |
7c673cae | 7913 | try { |
a8e16298 | 7914 | decode(*olh, iter); |
7c673cae FG |
7915 | } catch (buffer::error& err) { |
7916 | ldout(cct, 0) << "ERROR: failed to decode bi_entry()" << dendl; | |
7917 | return -EIO; | |
7918 | } | |
7919 | ||
7920 | return 0; | |
7921 | } | |
7922 | ||
a8e16298 TL |
7923 | int RGWRados::bi_get(const RGWBucketInfo& bucket_info, const rgw_obj& obj, |
7924 | BIIndexType index_type, rgw_cls_bi_entry *entry) | |
7c673cae FG |
7925 | { |
7926 | BucketShard bs(this); | |
a8e16298 | 7927 | int ret = bs.init(bucket_info, obj); |
7c673cae FG |
7928 | if (ret < 0) { |
7929 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
7930 | return ret; | |
7931 | } | |
7932 | ||
7933 | cls_rgw_obj_key key(obj.key.get_index_key_name(), obj.key.instance); | |
9f95a23c TL |
7934 | |
7935 | auto& ref = bs.bucket_obj.get_ref(); | |
7c673cae | 7936 | |
9f95a23c | 7937 | return cls_rgw_bi_get(ref.pool.ioctx(), ref.obj.oid, index_type, key, entry); |
7c673cae FG |
7938 | } |
7939 | ||
7940 | void RGWRados::bi_put(ObjectWriteOperation& op, BucketShard& bs, rgw_cls_bi_entry& entry) | |
7941 | { | |
9f95a23c TL |
7942 | auto& ref = bs.bucket_obj.get_ref(); |
7943 | cls_rgw_bi_put(op, ref.obj.oid, entry); | |
7c673cae FG |
7944 | } |
7945 | ||
7946 | int RGWRados::bi_put(BucketShard& bs, rgw_cls_bi_entry& entry) | |
7947 | { | |
9f95a23c TL |
7948 | auto& ref = bs.bucket_obj.get_ref(); |
7949 | int ret = cls_rgw_bi_put(ref.pool.ioctx(), ref.obj.oid, entry); | |
7c673cae FG |
7950 | if (ret < 0) |
7951 | return ret; | |
7952 | ||
7953 | return 0; | |
7954 | } | |
7955 | ||
7956 | int RGWRados::bi_put(rgw_bucket& bucket, rgw_obj& obj, rgw_cls_bi_entry& entry) | |
7957 | { | |
7958 | BucketShard bs(this); | |
f64942e4 | 7959 | int ret = bs.init(bucket, obj, nullptr /* no RGWBucketInfo */); |
7c673cae FG |
7960 | if (ret < 0) { |
7961 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
7962 | return ret; | |
7963 | } | |
7964 | ||
7965 | return bi_put(bs, entry); | |
7966 | } | |
7967 | ||
7968 | 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) | |
7969 | { | |
7970 | rgw_obj obj(bucket, obj_name); | |
7971 | BucketShard bs(this); | |
f64942e4 | 7972 | int ret = bs.init(bucket, obj, nullptr /* no RGWBucketInfo */); |
7c673cae FG |
7973 | if (ret < 0) { |
7974 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
7975 | return ret; | |
7976 | } | |
7977 | ||
9f95a23c TL |
7978 | auto& ref = bs.bucket_obj.get_ref(); |
7979 | ret = cls_rgw_bi_list(ref.pool.ioctx(), ref.obj.oid, obj_name, marker, max, entries, is_truncated); | |
31f18b77 FG |
7980 | if (ret == -ENOENT) { |
7981 | *is_truncated = false; | |
7982 | } | |
7c673cae FG |
7983 | if (ret < 0) |
7984 | return ret; | |
7985 | ||
7986 | return 0; | |
7987 | } | |
7988 | ||
7989 | 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) | |
7990 | { | |
9f95a23c TL |
7991 | auto& ref = bs.bucket_obj.get_ref(); |
7992 | int ret = cls_rgw_bi_list(ref.pool.ioctx(), ref.obj.oid, filter_obj, marker, max, entries, is_truncated); | |
7c673cae FG |
7993 | if (ret < 0) |
7994 | return ret; | |
7995 | ||
7996 | return 0; | |
7997 | } | |
7998 | ||
7999 | int RGWRados::bi_remove(BucketShard& bs) | |
8000 | { | |
9f95a23c TL |
8001 | auto& ref = bs.bucket_obj.get_ref(); |
8002 | int ret = ref.pool.ioctx().remove(ref.obj.oid); | |
7c673cae FG |
8003 | if (ret == -ENOENT) { |
8004 | ret = 0; | |
8005 | } | |
8006 | if (ret < 0) { | |
8007 | ldout(cct, 5) << "bs.index_ctx.remove(" << bs.bucket_obj << ") returned ret=" << ret << dendl; | |
8008 | return ret; | |
8009 | } | |
8010 | ||
8011 | return 0; | |
8012 | } | |
8013 | ||
8014 | 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) | |
8015 | { | |
8016 | BucketShard bs(this); | |
f64942e4 | 8017 | int ret = bs.init(bucket, shard_id, nullptr /* no RGWBucketInfo */); |
7c673cae FG |
8018 | if (ret < 0) { |
8019 | ldout(cct, 5) << "bs.init() returned ret=" << ret << dendl; | |
8020 | return ret; | |
8021 | } | |
8022 | ||
8023 | return bi_list(bs, filter_obj, marker, max, entries, is_truncated); | |
8024 | } | |
8025 | ||
8026 | int RGWRados::gc_operate(string& oid, librados::ObjectWriteOperation *op) | |
8027 | { | |
9f95a23c | 8028 | return rgw_rados_operate(gc_pool_ctx, oid, op, null_yield); |
7c673cae FG |
8029 | } |
8030 | ||
9f95a23c TL |
8031 | int RGWRados::gc_aio_operate(const string& oid, librados::AioCompletion *c, |
8032 | librados::ObjectWriteOperation *op) | |
7c673cae | 8033 | { |
9f95a23c | 8034 | return gc_pool_ctx.aio_operate(oid, c, op); |
7c673cae FG |
8035 | } |
8036 | ||
8037 | int RGWRados::gc_operate(string& oid, librados::ObjectReadOperation *op, bufferlist *pbl) | |
8038 | { | |
9f95a23c | 8039 | return rgw_rados_operate(gc_pool_ctx, oid, op, pbl, null_yield); |
7c673cae FG |
8040 | } |
8041 | ||
9f95a23c | 8042 | 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, bool& processing_queue) |
7c673cae | 8043 | { |
9f95a23c | 8044 | return gc->list(index, marker, max, expired_only, result, truncated, processing_queue); |
7c673cae FG |
8045 | } |
8046 | ||
11fdf7f2 | 8047 | int RGWRados::process_gc(bool expired_only) |
7c673cae | 8048 | { |
11fdf7f2 | 8049 | return gc->process(expired_only); |
7c673cae FG |
8050 | } |
8051 | ||
f6b5b4d7 TL |
8052 | int RGWRados::list_lc_progress(string& marker, uint32_t max_entries, |
8053 | vector<cls_rgw_lc_entry>& progress_map, | |
8054 | int& index) | |
7c673cae | 8055 | { |
f6b5b4d7 | 8056 | return lc->list_lc_progress(marker, max_entries, progress_map, index); |
7c673cae FG |
8057 | } |
8058 | ||
8059 | int RGWRados::process_lc() | |
8060 | { | |
f6b5b4d7 TL |
8061 | RGWLC lc; |
8062 | lc.initialize(cct, this->store); | |
8063 | RGWLC::LCWorker worker(&lc, cct, &lc, 0); | |
8064 | auto ret = lc.process(&worker, true /* once */); | |
8065 | lc.stop_processor(); // sets down_flag, but returns immediately | |
8066 | return ret; | |
7c673cae FG |
8067 | } |
8068 | ||
1adf2230 | 8069 | bool RGWRados::process_expire_objects() |
7c673cae | 8070 | { |
1adf2230 | 8071 | return obj_expirer->inspect_all_shards(utime_t(), ceph_clock_now()); |
7c673cae FG |
8072 | } |
8073 | ||
7c673cae | 8074 | int RGWRados::cls_obj_prepare_op(BucketShard& bs, RGWModifyOp op, string& tag, |
9f95a23c | 8075 | rgw_obj& obj, uint16_t bilog_flags, optional_yield y, rgw_zone_set *_zones_trace) |
7c673cae | 8076 | { |
31f18b77 FG |
8077 | rgw_zone_set zones_trace; |
8078 | if (_zones_trace) { | |
8079 | zones_trace = *_zones_trace; | |
8080 | } | |
9f95a23c | 8081 | zones_trace.insert(svc.zone->get_zone().id, bs.bucket.get_key()); |
1adf2230 | 8082 | |
7c673cae FG |
8083 | ObjectWriteOperation o; |
8084 | cls_rgw_obj_key key(obj.key.get_index_key_name(), obj.key.instance); | |
31f18b77 | 8085 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); |
11fdf7f2 | 8086 | cls_rgw_bucket_prepare_op(o, op, tag, key, obj.key.get_loc(), svc.zone->get_zone().log_data, bilog_flags, zones_trace); |
9f95a23c | 8087 | return bs.bucket_obj.operate(&o, y); |
7c673cae FG |
8088 | } |
8089 | ||
31f18b77 | 8090 | int RGWRados::cls_obj_complete_op(BucketShard& bs, const rgw_obj& obj, RGWModifyOp op, string& tag, |
7c673cae FG |
8091 | int64_t pool, uint64_t epoch, |
8092 | rgw_bucket_dir_entry& ent, RGWObjCategory category, | |
31f18b77 | 8093 | list<rgw_obj_index_key> *remove_objs, uint16_t bilog_flags, rgw_zone_set *_zones_trace) |
7c673cae | 8094 | { |
7c673cae FG |
8095 | ObjectWriteOperation o; |
8096 | rgw_bucket_dir_entry_meta dir_meta; | |
8097 | dir_meta = ent.meta; | |
8098 | dir_meta.category = category; | |
8099 | ||
1adf2230 AA |
8100 | rgw_zone_set zones_trace; |
8101 | if (_zones_trace) { | |
8102 | zones_trace = *_zones_trace; | |
8103 | } | |
9f95a23c | 8104 | zones_trace.insert(svc.zone->get_zone().id, bs.bucket.get_key()); |
1adf2230 | 8105 | |
7c673cae FG |
8106 | rgw_bucket_entry_ver ver; |
8107 | ver.pool = pool; | |
8108 | ver.epoch = epoch; | |
8109 | cls_rgw_obj_key key(ent.key.name, ent.key.instance); | |
31f18b77 FG |
8110 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); |
8111 | cls_rgw_bucket_complete_op(o, op, tag, ver, key, dir_meta, remove_objs, | |
11fdf7f2 | 8112 | svc.zone->get_zone().log_data, bilog_flags, &zones_trace); |
31f18b77 FG |
8113 | complete_op_data *arg; |
8114 | index_completion_manager->create_completion(obj, op, tag, ver, key, dir_meta, remove_objs, | |
11fdf7f2 | 8115 | svc.zone->get_zone().log_data, bilog_flags, &zones_trace, &arg); |
31f18b77 | 8116 | librados::AioCompletion *completion = arg->rados_completion; |
9f95a23c | 8117 | int ret = bs.bucket_obj.aio_operate(arg->rados_completion, &o); |
31f18b77 | 8118 | completion->release(); /* can't reference arg here, as it might have already been released */ |
7c673cae FG |
8119 | return ret; |
8120 | } | |
8121 | ||
31f18b77 | 8122 | int RGWRados::cls_obj_complete_add(BucketShard& bs, const rgw_obj& obj, string& tag, |
7c673cae FG |
8123 | int64_t pool, uint64_t epoch, |
8124 | rgw_bucket_dir_entry& ent, RGWObjCategory category, | |
31f18b77 | 8125 | list<rgw_obj_index_key> *remove_objs, uint16_t bilog_flags, rgw_zone_set *zones_trace) |
7c673cae | 8126 | { |
31f18b77 | 8127 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_ADD, tag, pool, epoch, ent, category, remove_objs, bilog_flags, zones_trace); |
7c673cae FG |
8128 | } |
8129 | ||
8130 | int RGWRados::cls_obj_complete_del(BucketShard& bs, string& tag, | |
8131 | int64_t pool, uint64_t epoch, | |
8132 | rgw_obj& obj, | |
8133 | real_time& removed_mtime, | |
8134 | list<rgw_obj_index_key> *remove_objs, | |
31f18b77 FG |
8135 | uint16_t bilog_flags, |
8136 | rgw_zone_set *zones_trace) | |
7c673cae FG |
8137 | { |
8138 | rgw_bucket_dir_entry ent; | |
8139 | ent.meta.mtime = removed_mtime; | |
8140 | obj.key.get_index_key(&ent.key); | |
11fdf7f2 TL |
8141 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_DEL, tag, pool, epoch, |
8142 | ent, RGWObjCategory::None, remove_objs, | |
8143 | bilog_flags, zones_trace); | |
7c673cae FG |
8144 | } |
8145 | ||
31f18b77 | 8146 | int RGWRados::cls_obj_complete_cancel(BucketShard& bs, string& tag, rgw_obj& obj, uint16_t bilog_flags, rgw_zone_set *zones_trace) |
7c673cae FG |
8147 | { |
8148 | rgw_bucket_dir_entry ent; | |
8149 | obj.key.get_index_key(&ent.key); | |
11fdf7f2 TL |
8150 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_CANCEL, tag, |
8151 | -1 /* pool id */, 0, ent, | |
8152 | RGWObjCategory::None, NULL, bilog_flags, | |
8153 | zones_trace); | |
7c673cae FG |
8154 | } |
8155 | ||
8156 | int RGWRados::cls_obj_set_bucket_tag_timeout(RGWBucketInfo& bucket_info, uint64_t timeout) | |
8157 | { | |
9f95a23c | 8158 | RGWSI_RADOS::Pool index_pool; |
7c673cae | 8159 | map<int, string> bucket_objs; |
9f95a23c | 8160 | int r = svc.bi_rados->open_bucket_index(bucket_info, std::nullopt, &index_pool, &bucket_objs, nullptr); |
7c673cae FG |
8161 | if (r < 0) |
8162 | return r; | |
8163 | ||
9f95a23c TL |
8164 | return CLSRGWIssueSetTagTimeout(index_pool.ioctx(), bucket_objs, cct->_conf->rgw_bucket_index_max_aio, timeout)(); |
8165 | } | |
8166 | ||
8167 | ||
8168 | uint32_t RGWRados::calc_ordered_bucket_list_per_shard(uint32_t num_entries, | |
8169 | uint32_t num_shards) | |
8170 | { | |
8171 | // We want to minimize the chances that when num_shards >> | |
8172 | // num_entries that we return much fewer than num_entries to the | |
8173 | // client. Given all the overhead of making a cls call to the osd, | |
8174 | // returning a few entries is not much more work than returning one | |
8175 | // entry. This minimum might be better tuned based on future | |
8176 | // experiments where num_shards >> num_entries. (Note: ">>" should | |
8177 | // be interpreted as "much greater than".) | |
8178 | constexpr uint32_t min_read = 8; | |
8179 | ||
8180 | // The following is based on _"Balls into Bins" -- A Simple and | |
8181 | // Tight Analysis_ by Raab and Steger. We add 1 as a way to handle | |
8182 | // cases when num_shards >> num_entries (it almost serves as a | |
8183 | // ceiling calculation). We also assume alpha is 1.0 and extract it | |
8184 | // from the calculation. Future work could involve memoizing some of | |
8185 | // the transcendental functions to minimize repeatedly re-calling | |
8186 | // them with the same parameters, which we expect to be the case the | |
8187 | // majority of the time. | |
8188 | uint32_t calc_read = | |
8189 | 1 + | |
8190 | static_cast<uint32_t>((num_entries / num_shards) + | |
8191 | sqrt((2 * num_entries) * | |
8192 | log(num_shards) / num_shards)); | |
8193 | ||
8194 | return std::max(min_read, calc_read); | |
7c673cae FG |
8195 | } |
8196 | ||
1adf2230 AA |
8197 | |
8198 | int RGWRados::cls_bucket_list_ordered(RGWBucketInfo& bucket_info, | |
9f95a23c TL |
8199 | const int shard_id, |
8200 | const rgw_obj_index_key& start_after, | |
1adf2230 | 8201 | const string& prefix, |
9f95a23c TL |
8202 | const string& delimiter, |
8203 | const uint32_t num_entries, | |
8204 | const bool list_versions, | |
8205 | const uint16_t expansion_factor, | |
8206 | ent_map_t& m, | |
8207 | bool* is_truncated, | |
8208 | bool* cls_filtered, | |
1adf2230 | 8209 | rgw_obj_index_key *last_entry, |
9f95a23c TL |
8210 | optional_yield y, |
8211 | check_filter_t force_check_filter) | |
7c673cae | 8212 | { |
9f95a23c TL |
8213 | /* expansion_factor allows the number of entries to read to grow |
8214 | * exponentially; this is used when earlier reads are producing too | |
8215 | * few results, perhaps due to filtering or to a series of | |
8216 | * namespaced entries */ | |
8217 | ||
8218 | ldout(cct, 10) << "RGWRados::" << __func__ << ": " << bucket_info.bucket << | |
8219 | " start_after=\"" << start_after.name << | |
8220 | "[" << start_after.instance << | |
8221 | "]\", prefix=\"" << prefix << | |
8222 | "\" num_entries=" << num_entries << | |
8223 | ", list_versions=" << list_versions << | |
8224 | ", expansion_factor=" << expansion_factor << dendl; | |
7c673cae | 8225 | |
9f95a23c TL |
8226 | m.clear(); |
8227 | ||
8228 | RGWSI_RADOS::Pool index_pool; | |
7c673cae | 8229 | // key - oid (for different shards if there is any) |
1adf2230 AA |
8230 | // value - list result for the corresponding oid (shard), it is filled by |
8231 | // the AIO callback | |
9f95a23c TL |
8232 | map<int, string> shard_oids; |
8233 | int r = svc.bi_rados->open_bucket_index(bucket_info, shard_id, | |
8234 | &index_pool, &shard_oids, | |
8235 | nullptr); | |
8236 | if (r < 0) { | |
7c673cae | 8237 | return r; |
9f95a23c TL |
8238 | } |
8239 | ||
8240 | const uint32_t shard_count = shard_oids.size(); | |
8241 | uint32_t num_entries_per_shard; | |
8242 | if (expansion_factor == 0) { | |
8243 | num_entries_per_shard = | |
8244 | calc_ordered_bucket_list_per_shard(num_entries, shard_count); | |
8245 | } else if (expansion_factor <= 11) { | |
8246 | // we'll max out the exponential multiplication factor at 1024 (2<<10) | |
8247 | num_entries_per_shard = | |
8248 | std::min(num_entries, | |
8249 | (uint32_t(1 << (expansion_factor - 1)) * | |
8250 | calc_ordered_bucket_list_per_shard(num_entries, shard_count))); | |
8251 | } else { | |
8252 | num_entries_per_shard = num_entries; | |
8253 | } | |
8254 | ||
8255 | ldout(cct, 10) << "RGWRados::" << __func__ << | |
8256 | " request from each of " << shard_count << | |
8257 | " shard(s) for " << num_entries_per_shard << " entries to get " << | |
8258 | num_entries << " total entries" << dendl; | |
7c673cae | 8259 | |
9f95a23c TL |
8260 | auto& ioctx = index_pool.ioctx(); |
8261 | map<int, rgw_cls_list_ret> shard_list_results; | |
8262 | cls_rgw_obj_key start_after_key(start_after.name, start_after.instance); | |
8263 | r = CLSRGWIssueBucketList(ioctx, start_after_key, prefix, delimiter, | |
8264 | num_entries_per_shard, | |
8265 | list_versions, shard_oids, shard_list_results, | |
1adf2230 | 8266 | cct->_conf->rgw_bucket_index_max_aio)(); |
9f95a23c | 8267 | if (r < 0) { |
7c673cae | 8268 | return r; |
9f95a23c | 8269 | } |
7c673cae | 8270 | |
9f95a23c TL |
8271 | // to manage the iterators through each shard's list results |
8272 | struct ShardTracker { | |
8273 | const size_t shard_idx; | |
8274 | rgw_cls_list_ret& result; | |
8275 | const std::string& oid_name; | |
8276 | RGWRados::ent_map_t::iterator cursor; | |
8277 | RGWRados::ent_map_t::iterator end; | |
8278 | ||
8279 | // manages an iterator through a shard and provides other | |
8280 | // accessors | |
8281 | ShardTracker(size_t _shard_idx, | |
8282 | rgw_cls_list_ret& _result, | |
8283 | const std::string& _oid_name): | |
8284 | shard_idx(_shard_idx), | |
8285 | result(_result), | |
8286 | oid_name(_oid_name), | |
8287 | cursor(_result.dir.m.begin()), | |
8288 | end(_result.dir.m.end()) | |
8289 | {} | |
8290 | ||
8291 | inline const std::string& entry_name() const { | |
8292 | return cursor->first; | |
8293 | } | |
8294 | rgw_bucket_dir_entry& dir_entry() const { | |
8295 | return cursor->second; | |
8296 | } | |
8297 | inline bool is_truncated() const { | |
8298 | return result.is_truncated; | |
8299 | } | |
8300 | inline ShardTracker& advance() { | |
8301 | ++cursor; | |
8302 | // return a self-reference to allow for chaining of calls, such | |
8303 | // as x.advance().at_end() | |
8304 | return *this; | |
8305 | } | |
8306 | inline bool at_end() const { | |
8307 | return cursor == end; | |
8308 | } | |
8309 | }; // ShardTracker | |
8310 | ||
8311 | // add the next unique candidate, or return false if we reach the end | |
8312 | auto next_candidate = [] (ShardTracker& t, | |
8313 | std::map<std::string, size_t>& candidates, | |
8314 | size_t tracker_idx) { | |
8315 | while (!t.at_end()) { | |
8316 | if (candidates.emplace(t.entry_name(), tracker_idx).second) { | |
8317 | return; | |
8318 | } | |
8319 | t.advance(); // skip duplicate common prefixes | |
8320 | } | |
8321 | }; | |
8322 | ||
8323 | // one tracker per shard requested (may not be all shards) | |
8324 | std::vector<ShardTracker> results_trackers; | |
8325 | results_trackers.reserve(shard_list_results.size()); | |
8326 | for (auto& r : shard_list_results) { | |
8327 | results_trackers.emplace_back(r.first, r.second, shard_oids[r.first]); | |
8328 | ||
8329 | // if any *one* shard's result is trucated, the entire result is | |
8330 | // truncated | |
8331 | *is_truncated = *is_truncated || r.second.is_truncated; | |
8332 | ||
8333 | // unless *all* are shards are cls_filtered, the entire result is | |
8334 | // not filtered | |
8335 | *cls_filtered = *cls_filtered && r.second.cls_filtered; | |
7c673cae FG |
8336 | } |
8337 | ||
9f95a23c TL |
8338 | // create a map to track the next candidate entry from ShardTracker |
8339 | // (key=candidate, value=index into results_trackers); as we consume | |
8340 | // entries from shards, we replace them with the next entries in the | |
8341 | // shards until we run out | |
7c673cae | 8342 | map<string, size_t> candidates; |
9f95a23c TL |
8343 | size_t tracker_idx = 0; |
8344 | for (auto& t : results_trackers) { | |
8345 | // it's important that the values in the map refer to the index | |
8346 | // into the results_trackers vector, which may not be the same | |
8347 | // as the shard number (i.e., when not all shards are requested) | |
8348 | next_candidate(t, candidates, tracker_idx); | |
8349 | ++tracker_idx; | |
7c673cae FG |
8350 | } |
8351 | ||
9f95a23c TL |
8352 | rgw_bucket_dir_entry* |
8353 | last_entry_visited = nullptr; // to set last_entry (marker) | |
7c673cae FG |
8354 | map<string, bufferlist> updates; |
8355 | uint32_t count = 0; | |
8356 | while (count < num_entries && !candidates.empty()) { | |
8357 | r = 0; | |
9f95a23c TL |
8358 | // select the next entry in lexical order (first key in map); |
8359 | // again tracker_idx is not necessarily shard number, but is index | |
8360 | // into results_trackers vector | |
8361 | tracker_idx = candidates.begin()->second; | |
8362 | auto& tracker = results_trackers.at(tracker_idx); | |
e306af50 | 8363 | |
9f95a23c TL |
8364 | const string& name = tracker.entry_name(); |
8365 | rgw_bucket_dir_entry& dirent = tracker.dir_entry(); | |
8366 | ||
8367 | ldout(cct, 20) << "RGWRados::" << __func__ << " currently processing " << | |
8368 | dirent.key << " from shard " << tracker.shard_idx << dendl; | |
8369 | ||
8370 | const bool force_check = | |
8371 | force_check_filter && force_check_filter(dirent.key.name); | |
8372 | ||
8373 | if ((!dirent.exists && | |
8374 | !dirent.is_delete_marker() && | |
8375 | !dirent.is_common_prefix()) || | |
3efd9988 FG |
8376 | !dirent.pending_map.empty() || |
8377 | force_check) { | |
9f95a23c TL |
8378 | /* there are uncommitted ops. We need to check the current |
8379 | * state, and if the tags are old we need to do clean-up as | |
8380 | * well. */ | |
7c673cae | 8381 | librados::IoCtx sub_ctx; |
9f95a23c | 8382 | sub_ctx.dup(ioctx); |
1adf2230 | 8383 | r = check_disk_state(sub_ctx, bucket_info, dirent, dirent, |
9f95a23c | 8384 | updates[tracker.oid_name], y); |
7c673cae | 8385 | if (r < 0 && r != -ENOENT) { |
9f95a23c | 8386 | return r; |
7c673cae | 8387 | } |
eafe8130 | 8388 | } else { |
9f95a23c | 8389 | r = 0; |
7c673cae | 8390 | } |
9f95a23c | 8391 | |
7c673cae | 8392 | if (r >= 0) { |
9f95a23c | 8393 | ldout(cct, 10) << "RGWRados::" << __func__ << ": got " << |
1adf2230 | 8394 | dirent.key.name << "[" << dirent.key.instance << "]" << dendl; |
7c673cae | 8395 | m[name] = std::move(dirent); |
e306af50 | 8396 | last_entry_visited = &(m[name]); |
7c673cae | 8397 | ++count; |
9f95a23c TL |
8398 | } else { |
8399 | ldout(cct, 10) << "RGWRados::" << __func__ << ": skipping " << | |
8400 | dirent.key.name << "[" << dirent.key.instance << "]" << dendl; | |
e306af50 | 8401 | last_entry_visited = &tracker.dir_entry(); |
7c673cae FG |
8402 | } |
8403 | ||
9f95a23c | 8404 | // refresh the candidates map |
7c673cae | 8405 | candidates.erase(candidates.begin()); |
9f95a23c TL |
8406 | tracker.advance(); |
8407 | ||
8408 | next_candidate(tracker, candidates, tracker_idx); | |
8409 | ||
8410 | if (tracker.at_end() && tracker.is_truncated()) { | |
8411 | // once we exhaust one shard that is truncated, we need to stop, | |
8412 | // as we cannot be certain that one of the next entries needs to | |
8413 | // come from that shard; S3 and swift protocols allow returning | |
8414 | // fewer than what was requested | |
8415 | break; | |
7c673cae | 8416 | } |
9f95a23c | 8417 | } // while we haven't provided requested # of result entries |
7c673cae | 8418 | |
9f95a23c TL |
8419 | // suggest updates if there are any |
8420 | for (auto& miter : updates) { | |
8421 | if (miter.second.length()) { | |
7c673cae | 8422 | ObjectWriteOperation o; |
9f95a23c | 8423 | cls_rgw_suggest_changes(o, miter.second); |
7c673cae | 8424 | // we don't care if we lose suggested updates, send them off blindly |
9f95a23c TL |
8425 | AioCompletion *c = |
8426 | librados::Rados::aio_create_completion(nullptr, nullptr); | |
8427 | ioctx.aio_operate(miter.first, c, &o); | |
1adf2230 | 8428 | c->release(); |
7c673cae | 8429 | } |
9f95a23c | 8430 | } // updates loop |
7c673cae | 8431 | |
9f95a23c TL |
8432 | // determine truncation by checking if all the returned entries are |
8433 | // consumed or not | |
8434 | *is_truncated = false; | |
8435 | for (const auto& t : results_trackers) { | |
8436 | if (!t.at_end() || t.is_truncated()) { | |
7c673cae | 8437 | *is_truncated = true; |
1adf2230 AA |
8438 | break; |
8439 | } | |
7c673cae | 8440 | } |
92f5a8d4 | 8441 | |
9f95a23c TL |
8442 | ldout(cct, 20) << "RGWRados::" << __func__ << |
8443 | ": returning, count=" << count << ", is_truncated=" << *is_truncated << | |
8444 | dendl; | |
8445 | ||
8446 | if (*is_truncated && count < num_entries) { | |
8447 | ldout(cct, 10) << "RGWRados::" << __func__ << | |
8448 | ": INFO requested " << num_entries << " entries but returning " << | |
8449 | count << ", which is truncated" << dendl; | |
8450 | } | |
8451 | ||
8452 | if (last_entry_visited != nullptr && last_entry) { | |
e306af50 | 8453 | *last_entry = last_entry_visited->key; |
9f95a23c TL |
8454 | ldout(cct, 20) << "RGWRados::" << __func__ << |
8455 | ": returning, last_entry=" << *last_entry << dendl; | |
8456 | } else { | |
8457 | ldout(cct, 20) << "RGWRados::" << __func__ << | |
8458 | ": returning, last_entry NOT SET" << dendl; | |
8459 | } | |
7c673cae FG |
8460 | |
8461 | return 0; | |
8462 | } | |
8463 | ||
1adf2230 AA |
8464 | |
8465 | int RGWRados::cls_bucket_list_unordered(RGWBucketInfo& bucket_info, | |
8466 | int shard_id, | |
9f95a23c | 8467 | const rgw_obj_index_key& start_after, |
1adf2230 AA |
8468 | const string& prefix, |
8469 | uint32_t num_entries, | |
8470 | bool list_versions, | |
8471 | std::vector<rgw_bucket_dir_entry>& ent_list, | |
8472 | bool *is_truncated, | |
8473 | rgw_obj_index_key *last_entry, | |
9f95a23c TL |
8474 | optional_yield y, |
8475 | check_filter_t force_check_filter) { | |
1adf2230 | 8476 | ldout(cct, 10) << "cls_bucket_list_unordered " << bucket_info.bucket << |
9f95a23c | 8477 | " start_after " << start_after.name << "[" << start_after.instance << |
1adf2230 AA |
8478 | "] num_entries " << num_entries << dendl; |
8479 | ||
9f95a23c | 8480 | ent_list.clear(); |
11fdf7f2 TL |
8481 | static MultipartMetaFilter multipart_meta_filter; |
8482 | ||
1adf2230 | 8483 | *is_truncated = false; |
9f95a23c | 8484 | RGWSI_RADOS::Pool index_pool; |
1adf2230 | 8485 | |
1adf2230 | 8486 | map<int, string> oids; |
9f95a23c | 8487 | int r = svc.bi_rados->open_bucket_index(bucket_info, shard_id, &index_pool, &oids, nullptr); |
1adf2230 AA |
8488 | if (r < 0) |
8489 | return r; | |
9f95a23c TL |
8490 | |
8491 | auto& ioctx = index_pool.ioctx(); | |
8492 | ||
1adf2230 AA |
8493 | const uint32_t num_shards = oids.size(); |
8494 | ||
9f95a23c | 8495 | rgw_obj_index_key marker = start_after; |
1adf2230 AA |
8496 | uint32_t current_shard; |
8497 | if (shard_id >= 0) { | |
8498 | current_shard = shard_id; | |
9f95a23c | 8499 | } else if (start_after.empty()) { |
1adf2230 AA |
8500 | current_shard = 0u; |
8501 | } else { | |
9f95a23c TL |
8502 | // at this point we have a marker (start_after) that has something |
8503 | // in it, so we need to get to the bucket shard index, so we can | |
11fdf7f2 TL |
8504 | // start reading from there |
8505 | ||
8506 | std::string key; | |
8507 | // test whether object name is a multipart meta name | |
9f95a23c | 8508 | if(! multipart_meta_filter.filter(start_after.name, key)) { |
11fdf7f2 TL |
8509 | // if multipart_meta_filter fails, must be "regular" (i.e., |
8510 | // unadorned) and the name is the key | |
9f95a23c | 8511 | key = start_after.name; |
11fdf7f2 TL |
8512 | } |
8513 | ||
8514 | // now convert the key (oid) to an rgw_obj_key since that will | |
8515 | // separate out the namespace, name, and instance | |
8516 | rgw_obj_key obj_key; | |
8517 | bool parsed = rgw_obj_key::parse_raw_oid(key, &obj_key); | |
8518 | if (!parsed) { | |
8519 | ldout(cct, 0) << | |
8520 | "ERROR: RGWRados::cls_bucket_list_unordered received an invalid " | |
9f95a23c | 8521 | "start marker: '" << start_after << "'" << dendl; |
11fdf7f2 TL |
8522 | return -EINVAL; |
8523 | } else if (obj_key.name.empty()) { | |
8524 | // if the name is empty that means the object name came in with | |
8525 | // a namespace only, and therefore we need to start our scan at | |
8526 | // the first bucket index shard | |
8527 | current_shard = 0u; | |
8528 | } else { | |
8529 | // so now we have the key used to compute the bucket index shard | |
8530 | // and can extract the specific shard from it | |
9f95a23c | 8531 | current_shard = svc.bi_rados->bucket_shard_index(obj_key.name, num_shards); |
11fdf7f2 | 8532 | } |
1adf2230 AA |
8533 | } |
8534 | ||
8535 | uint32_t count = 0u; | |
8536 | map<string, bufferlist> updates; | |
11fdf7f2 | 8537 | rgw_obj_index_key last_added_entry; |
1adf2230 AA |
8538 | while (count <= num_entries && |
8539 | ((shard_id >= 0 && current_shard == uint32_t(shard_id)) || | |
8540 | current_shard < num_shards)) { | |
81eedcae TL |
8541 | const std::string& oid = oids[current_shard]; |
8542 | rgw_cls_list_ret result; | |
8543 | ||
8544 | librados::ObjectReadOperation op; | |
9f95a23c TL |
8545 | string empty_delimiter; |
8546 | cls_rgw_bucket_list_op(op, marker, prefix, empty_delimiter, | |
8547 | num_entries, | |
81eedcae | 8548 | list_versions, &result); |
9f95a23c | 8549 | r = rgw_rados_operate(ioctx, oid, &op, nullptr, null_yield); |
1adf2230 AA |
8550 | if (r < 0) |
8551 | return r; | |
8552 | ||
1adf2230 AA |
8553 | for (auto& entry : result.dir.m) { |
8554 | rgw_bucket_dir_entry& dirent = entry.second; | |
8555 | ||
8556 | bool force_check = force_check_filter && | |
8557 | force_check_filter(dirent.key.name); | |
8558 | if ((!dirent.exists && !dirent.is_delete_marker()) || | |
8559 | !dirent.pending_map.empty() || | |
8560 | force_check) { | |
8561 | /* there are uncommitted ops. We need to check the current state, | |
8562 | * and if the tags are old we need to do cleanup as well. */ | |
8563 | librados::IoCtx sub_ctx; | |
9f95a23c TL |
8564 | sub_ctx.dup(ioctx); |
8565 | r = check_disk_state(sub_ctx, bucket_info, dirent, dirent, updates[oid], y); | |
1adf2230 AA |
8566 | if (r < 0 && r != -ENOENT) { |
8567 | return r; | |
8568 | } | |
eafe8130 TL |
8569 | } else { |
8570 | r = 0; | |
1adf2230 AA |
8571 | } |
8572 | ||
8573 | // at this point either r >=0 or r == -ENOENT | |
8574 | if (r >= 0) { // i.e., if r != -ENOENT | |
8575 | ldout(cct, 10) << "RGWRados::cls_bucket_list_unordered: got " << | |
8576 | dirent.key.name << "[" << dirent.key.instance << "]" << dendl; | |
8577 | ||
8578 | if (count < num_entries) { | |
11fdf7f2 | 8579 | marker = last_added_entry = dirent.key; // double assign |
1adf2230 AA |
8580 | ent_list.emplace_back(std::move(dirent)); |
8581 | ++count; | |
8582 | } else { | |
8583 | *is_truncated = true; | |
8584 | goto check_updates; | |
8585 | } | |
8586 | } else { // r == -ENOENT | |
8587 | // in the case of -ENOENT, make sure we're advancing marker | |
8588 | // for possible next call to CLSRGWIssueBucketList | |
11fdf7f2 | 8589 | marker = dirent.key; |
1adf2230 AA |
8590 | } |
8591 | } // entry for loop | |
8592 | ||
8593 | if (!result.is_truncated) { | |
8594 | // if we reached the end of the shard read next shard | |
8595 | ++current_shard; | |
11fdf7f2 | 8596 | marker = rgw_obj_index_key(); |
1adf2230 AA |
8597 | } |
8598 | } // shard loop | |
8599 | ||
8600 | check_updates: | |
11fdf7f2 | 8601 | |
1adf2230 AA |
8602 | // suggest updates if there is any |
8603 | map<string, bufferlist>::iterator miter = updates.begin(); | |
8604 | for (; miter != updates.end(); ++miter) { | |
8605 | if (miter->second.length()) { | |
8606 | ObjectWriteOperation o; | |
8607 | cls_rgw_suggest_changes(o, miter->second); | |
8608 | // we don't care if we lose suggested updates, send them off blindly | |
9f95a23c TL |
8609 | AioCompletion *c = librados::Rados::aio_create_completion(nullptr, nullptr); |
8610 | ioctx.aio_operate(miter->first, c, &o); | |
1adf2230 AA |
8611 | c->release(); |
8612 | } | |
8613 | } | |
8614 | ||
8615 | if (last_entry && !ent_list.empty()) { | |
8616 | *last_entry = last_added_entry; | |
8617 | } | |
8618 | ||
8619 | return 0; | |
11fdf7f2 | 8620 | } // RGWRados::cls_bucket_list_unordered |
1adf2230 AA |
8621 | |
8622 | ||
8623 | int RGWRados::cls_obj_usage_log_add(const string& oid, | |
8624 | rgw_usage_log_info& info) | |
7c673cae | 8625 | { |
11fdf7f2 | 8626 | rgw_raw_obj obj(svc.zone->get_zone_params().usage_log_pool, oid); |
7c673cae FG |
8627 | |
8628 | rgw_rados_ref ref; | |
224ce89b | 8629 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
8630 | if (r < 0) { |
8631 | return r; | |
8632 | } | |
8633 | ||
8634 | ObjectWriteOperation op; | |
8635 | cls_rgw_usage_log_add(op, info); | |
8636 | ||
9f95a23c | 8637 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); |
7c673cae FG |
8638 | return r; |
8639 | } | |
8640 | ||
11fdf7f2 TL |
8641 | int RGWRados::cls_obj_usage_log_read(const string& oid, const string& user, const string& bucket, |
8642 | uint64_t start_epoch, uint64_t end_epoch, uint32_t max_entries, | |
8643 | string& read_iter, map<rgw_user_bucket, rgw_usage_log_entry>& usage, | |
8644 | bool *is_truncated) | |
7c673cae | 8645 | { |
11fdf7f2 | 8646 | rgw_raw_obj obj(svc.zone->get_zone_params().usage_log_pool, oid); |
7c673cae FG |
8647 | |
8648 | rgw_rados_ref ref; | |
224ce89b | 8649 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
8650 | if (r < 0) { |
8651 | return r; | |
8652 | } | |
8653 | ||
8654 | *is_truncated = false; | |
8655 | ||
9f95a23c | 8656 | r = cls_rgw_usage_log_read(ref.pool.ioctx(), ref.obj.oid, user, bucket, start_epoch, end_epoch, |
7c673cae FG |
8657 | max_entries, read_iter, usage, is_truncated); |
8658 | ||
8659 | return r; | |
8660 | } | |
8661 | ||
9f95a23c TL |
8662 | static int cls_rgw_usage_log_trim_repeat(rgw_rados_ref ref, const string& user, const string& bucket, uint64_t start_epoch, uint64_t end_epoch) |
8663 | { | |
8664 | bool done = false; | |
8665 | do { | |
8666 | librados::ObjectWriteOperation op; | |
8667 | cls_rgw_usage_log_trim(op, user, bucket, start_epoch, end_epoch); | |
8668 | int r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
8669 | if (r == -ENODATA) | |
8670 | done = true; | |
8671 | else if (r < 0) | |
8672 | return r; | |
8673 | } while (!done); | |
8674 | ||
8675 | return 0; | |
8676 | } | |
8677 | ||
11fdf7f2 TL |
8678 | int RGWRados::cls_obj_usage_log_trim(const string& oid, const string& user, const string& bucket, |
8679 | uint64_t start_epoch, uint64_t end_epoch) | |
7c673cae | 8680 | { |
11fdf7f2 | 8681 | rgw_raw_obj obj(svc.zone->get_zone_params().usage_log_pool, oid); |
7c673cae FG |
8682 | |
8683 | rgw_rados_ref ref; | |
224ce89b | 8684 | int r = get_raw_obj_ref(obj, &ref); |
7c673cae FG |
8685 | if (r < 0) { |
8686 | return r; | |
8687 | } | |
8688 | ||
9f95a23c | 8689 | r = cls_rgw_usage_log_trim_repeat(ref, user, bucket, start_epoch, end_epoch); |
11fdf7f2 TL |
8690 | return r; |
8691 | } | |
8692 | ||
8693 | int RGWRados::cls_obj_usage_log_clear(string& oid) | |
8694 | { | |
8695 | rgw_raw_obj obj(svc.zone->get_zone_params().usage_log_pool, oid); | |
8696 | ||
8697 | rgw_rados_ref ref; | |
8698 | int r = get_raw_obj_ref(obj, &ref); | |
8699 | if (r < 0) { | |
8700 | return r; | |
8701 | } | |
8702 | librados::ObjectWriteOperation op; | |
8703 | cls_rgw_usage_log_clear(op); | |
9f95a23c | 8704 | r = rgw_rados_operate(ref.pool.ioctx(), ref.obj.oid, &op, null_yield); |
7c673cae FG |
8705 | return r; |
8706 | } | |
8707 | ||
11fdf7f2 | 8708 | |
7c673cae FG |
8709 | int RGWRados::remove_objs_from_index(RGWBucketInfo& bucket_info, list<rgw_obj_index_key>& oid_list) |
8710 | { | |
9f95a23c | 8711 | RGWSI_RADOS::Pool index_pool; |
7c673cae FG |
8712 | string dir_oid; |
8713 | ||
11fdf7f2 | 8714 | uint8_t suggest_flag = (svc.zone->get_zone().log_data ? CEPH_RGW_DIR_SUGGEST_LOG_OP : 0); |
7c673cae | 8715 | |
9f95a23c | 8716 | int r = svc.bi_rados->open_bucket_index(bucket_info, &index_pool, &dir_oid); |
7c673cae FG |
8717 | if (r < 0) |
8718 | return r; | |
8719 | ||
8720 | bufferlist updates; | |
8721 | ||
8722 | for (auto iter = oid_list.begin(); iter != oid_list.end(); ++iter) { | |
8723 | rgw_bucket_dir_entry entry; | |
8724 | entry.key = *iter; | |
8725 | dout(2) << "RGWRados::remove_objs_from_index bucket=" << bucket_info.bucket << " obj=" << entry.key.name << ":" << entry.key.instance << dendl; | |
8726 | entry.ver.epoch = (uint64_t)-1; // ULLONG_MAX, needed to that objclass doesn't skip out request | |
8727 | updates.append(CEPH_RGW_REMOVE | suggest_flag); | |
11fdf7f2 | 8728 | encode(entry, updates); |
7c673cae FG |
8729 | } |
8730 | ||
8731 | bufferlist out; | |
8732 | ||
9f95a23c | 8733 | r = index_pool.ioctx().exec(dir_oid, RGW_CLASS, RGW_DIR_SUGGEST_CHANGES, updates, out); |
7c673cae FG |
8734 | |
8735 | return r; | |
8736 | } | |
8737 | ||
8738 | int RGWRados::check_disk_state(librados::IoCtx io_ctx, | |
8739 | const RGWBucketInfo& bucket_info, | |
8740 | rgw_bucket_dir_entry& list_state, | |
8741 | rgw_bucket_dir_entry& object, | |
9f95a23c TL |
8742 | bufferlist& suggested_updates, |
8743 | optional_yield y) | |
7c673cae FG |
8744 | { |
8745 | const rgw_bucket& bucket = bucket_info.bucket; | |
11fdf7f2 | 8746 | uint8_t suggest_flag = (svc.zone->get_zone().log_data ? CEPH_RGW_DIR_SUGGEST_LOG_OP : 0); |
7c673cae FG |
8747 | |
8748 | std::string loc; | |
8749 | ||
8750 | rgw_obj obj(bucket, list_state.key); | |
8751 | ||
8752 | string oid; | |
8753 | get_obj_bucket_and_oid_loc(obj, oid, loc); | |
8754 | ||
8755 | if (loc != list_state.locator) { | |
8756 | ldout(cct, 0) << "WARNING: generated locator (" << loc << ") is different from listed locator (" << list_state.locator << ")" << dendl; | |
8757 | } | |
8758 | ||
8759 | io_ctx.locator_set_key(list_state.locator); | |
8760 | ||
8761 | RGWObjState *astate = NULL; | |
9f95a23c TL |
8762 | RGWObjectCtx rctx(this->store); |
8763 | int r = get_obj_state(&rctx, bucket_info, obj, &astate, false, y); | |
7c673cae FG |
8764 | if (r < 0) |
8765 | return r; | |
8766 | ||
8767 | list_state.pending_map.clear(); // we don't need this and it inflates size | |
9f95a23c | 8768 | if (!list_state.is_delete_marker() && !astate->exists) { |
7c673cae FG |
8769 | /* object doesn't exist right now -- hopefully because it's |
8770 | * marked as !exists and got deleted */ | |
8771 | if (list_state.exists) { | |
8772 | /* FIXME: what should happen now? Work out if there are any | |
8773 | * non-bad ways this could happen (there probably are, but annoying | |
8774 | * to handle!) */ | |
8775 | } | |
8776 | // encode a suggested removal of that key | |
8777 | list_state.ver.epoch = io_ctx.get_last_version(); | |
8778 | list_state.ver.pool = io_ctx.get_id(); | |
8779 | cls_rgw_encode_suggestion(CEPH_RGW_REMOVE, list_state, suggested_updates); | |
8780 | return -ENOENT; | |
8781 | } | |
8782 | ||
8783 | string etag; | |
8784 | string content_type; | |
8785 | ACLOwner owner; | |
8786 | ||
8787 | object.meta.size = astate->size; | |
8788 | object.meta.accounted_size = astate->accounted_size; | |
8789 | object.meta.mtime = astate->mtime; | |
8790 | ||
8791 | map<string, bufferlist>::iterator iter = astate->attrset.find(RGW_ATTR_ETAG); | |
8792 | if (iter != astate->attrset.end()) { | |
11fdf7f2 | 8793 | etag = rgw_bl_str(iter->second); |
7c673cae FG |
8794 | } |
8795 | iter = astate->attrset.find(RGW_ATTR_CONTENT_TYPE); | |
8796 | if (iter != astate->attrset.end()) { | |
11fdf7f2 | 8797 | content_type = rgw_bl_str(iter->second); |
7c673cae FG |
8798 | } |
8799 | iter = astate->attrset.find(RGW_ATTR_ACL); | |
8800 | if (iter != astate->attrset.end()) { | |
8801 | r = decode_policy(iter->second, &owner); | |
8802 | if (r < 0) { | |
8803 | dout(0) << "WARNING: could not decode policy for object: " << obj << dendl; | |
8804 | } | |
8805 | } | |
8806 | ||
9f95a23c | 8807 | if (astate->manifest) { |
7c673cae | 8808 | RGWObjManifest::obj_iterator miter; |
9f95a23c | 8809 | RGWObjManifest& manifest = *astate->manifest; |
7c673cae FG |
8810 | for (miter = manifest.obj_begin(); miter != manifest.obj_end(); ++miter) { |
8811 | const rgw_raw_obj& raw_loc = miter.get_location().get_raw_obj(this); | |
8812 | rgw_obj loc; | |
9f95a23c | 8813 | RGWSI_Tier_RADOS::raw_obj_to_obj(manifest.get_obj().bucket, raw_loc, &loc); |
7c673cae FG |
8814 | |
8815 | if (loc.key.ns == RGW_OBJ_NS_MULTIPART) { | |
8816 | dout(10) << "check_disk_state(): removing manifest part from index: " << loc << dendl; | |
494da23a | 8817 | r = delete_obj_index(loc, astate->mtime); |
7c673cae FG |
8818 | if (r < 0) { |
8819 | dout(0) << "WARNING: delete_obj_index() returned r=" << r << dendl; | |
8820 | } | |
8821 | } | |
8822 | } | |
8823 | } | |
8824 | ||
8825 | object.meta.etag = etag; | |
8826 | object.meta.content_type = content_type; | |
8827 | object.meta.owner = owner.get_id().to_str(); | |
8828 | object.meta.owner_display_name = owner.get_display_name(); | |
8829 | ||
8830 | // encode suggested updates | |
8831 | list_state.ver.pool = io_ctx.get_id(); | |
8832 | list_state.ver.epoch = astate->epoch; | |
8833 | list_state.meta.size = object.meta.size; | |
8834 | list_state.meta.accounted_size = object.meta.accounted_size; | |
8835 | list_state.meta.mtime = object.meta.mtime; | |
8836 | list_state.meta.category = main_category; | |
8837 | list_state.meta.etag = etag; | |
8838 | list_state.meta.content_type = content_type; | |
8839 | if (astate->obj_tag.length() > 0) | |
8840 | list_state.tag = astate->obj_tag.c_str(); | |
8841 | list_state.meta.owner = owner.get_id().to_str(); | |
8842 | list_state.meta.owner_display_name = owner.get_display_name(); | |
8843 | ||
8844 | list_state.exists = true; | |
8845 | cls_rgw_encode_suggestion(CEPH_RGW_UPDATE | suggest_flag, list_state, suggested_updates); | |
8846 | return 0; | |
8847 | } | |
8848 | ||
a8e16298 | 8849 | int RGWRados::cls_bucket_head(const RGWBucketInfo& bucket_info, int shard_id, vector<rgw_bucket_dir_header>& headers, map<int, string> *bucket_instance_ids) |
7c673cae | 8850 | { |
9f95a23c | 8851 | RGWSI_RADOS::Pool index_pool; |
7c673cae FG |
8852 | map<int, string> oids; |
8853 | map<int, struct rgw_cls_list_ret> list_results; | |
9f95a23c TL |
8854 | int r = svc.bi_rados->open_bucket_index(bucket_info, shard_id, &index_pool, &oids, bucket_instance_ids); |
8855 | if (r < 0) { | |
8856 | ldout(cct, 20) << "cls_bucket_head: open_bucket_index() returned " | |
8857 | << r << dendl; | |
7c673cae | 8858 | return r; |
9f95a23c | 8859 | } |
7c673cae | 8860 | |
9f95a23c TL |
8861 | r = CLSRGWIssueGetDirHeader(index_pool.ioctx(), oids, list_results, cct->_conf->rgw_bucket_index_max_aio)(); |
8862 | if (r < 0) { | |
8863 | ldout(cct, 20) << "cls_bucket_head: CLSRGWIssueGetDirHeader() returned " | |
8864 | << r << dendl; | |
7c673cae | 8865 | return r; |
9f95a23c | 8866 | } |
7c673cae FG |
8867 | |
8868 | map<int, struct rgw_cls_list_ret>::iterator iter = list_results.begin(); | |
8869 | for(; iter != list_results.end(); ++iter) { | |
a8e16298 | 8870 | headers.push_back(std::move(iter->second.dir.header)); |
7c673cae FG |
8871 | } |
8872 | return 0; | |
8873 | } | |
8874 | ||
8875 | int RGWRados::cls_bucket_head_async(const RGWBucketInfo& bucket_info, int shard_id, RGWGetDirHeader_CB *ctx, int *num_aio) | |
8876 | { | |
9f95a23c | 8877 | RGWSI_RADOS::Pool index_pool; |
7c673cae | 8878 | map<int, string> bucket_objs; |
9f95a23c | 8879 | int r = svc.bi_rados->open_bucket_index(bucket_info, shard_id, &index_pool, &bucket_objs, nullptr); |
7c673cae FG |
8880 | if (r < 0) |
8881 | return r; | |
8882 | ||
8883 | map<int, string>::iterator iter = bucket_objs.begin(); | |
8884 | for (; iter != bucket_objs.end(); ++iter) { | |
9f95a23c | 8885 | r = cls_rgw_get_dir_header_async(index_pool.ioctx(), iter->second, static_cast<RGWGetDirHeader_CB*>(ctx->get())); |
7c673cae FG |
8886 | if (r < 0) { |
8887 | ctx->put(); | |
8888 | break; | |
8889 | } else { | |
8890 | (*num_aio)++; | |
8891 | } | |
8892 | } | |
8893 | return r; | |
8894 | } | |
8895 | ||
9f95a23c TL |
8896 | int RGWRados::check_bucket_shards(const RGWBucketInfo& bucket_info, |
8897 | const rgw_bucket& bucket, | |
8898 | uint64_t num_objs) | |
31f18b77 | 8899 | { |
11fdf7f2 | 8900 | if (! cct->_conf.get_val<bool>("rgw_dynamic_resharding")) { |
31f18b77 FG |
8901 | return 0; |
8902 | } | |
8903 | ||
8904 | bool need_resharding = false; | |
9f95a23c TL |
8905 | uint32_t num_source_shards = |
8906 | (bucket_info.num_shards > 0 ? bucket_info.num_shards : 1); | |
8907 | const uint32_t max_dynamic_shards = | |
8908 | uint32_t(cct->_conf.get_val<uint64_t>("rgw_max_dynamic_shards")); | |
8909 | ||
8910 | if (num_source_shards >= max_dynamic_shards) { | |
8911 | return 0; | |
8912 | } | |
31f18b77 | 8913 | |
9f95a23c | 8914 | uint32_t suggested_num_shards = 0; |
11fdf7f2 TL |
8915 | const uint64_t max_objs_per_shard = |
8916 | cct->_conf.get_val<uint64_t>("rgw_max_objs_per_shard"); | |
9f95a23c TL |
8917 | |
8918 | quota_handler->check_bucket_shards(max_objs_per_shard, num_source_shards, | |
8919 | num_objs, need_resharding, &suggested_num_shards); | |
8920 | if (! need_resharding) { | |
8921 | return 0; | |
31f18b77 FG |
8922 | } |
8923 | ||
9f95a23c TL |
8924 | const uint32_t final_num_shards = |
8925 | RGWBucketReshard::get_preferred_shards(suggested_num_shards, | |
8926 | max_dynamic_shards); | |
8927 | // final verification, so we don't reduce number of shards | |
8928 | if (final_num_shards <= num_source_shards) { | |
8929 | return 0; | |
31f18b77 FG |
8930 | } |
8931 | ||
f91f0fd5 | 8932 | ldout(cct, 1) << "RGWRados::" << __func__ << " bucket " << bucket.name << |
9f95a23c TL |
8933 | " needs resharding; current num shards " << bucket_info.num_shards << |
8934 | "; new num shards " << final_num_shards << " (suggested " << | |
8935 | suggested_num_shards << ")" << dendl; | |
8936 | ||
8937 | return add_bucket_to_reshard(bucket_info, final_num_shards); | |
31f18b77 FG |
8938 | } |
8939 | ||
8940 | int RGWRados::add_bucket_to_reshard(const RGWBucketInfo& bucket_info, uint32_t new_num_shards) | |
8941 | { | |
9f95a23c | 8942 | RGWReshard reshard(this->store); |
31f18b77 FG |
8943 | |
8944 | uint32_t num_source_shards = (bucket_info.num_shards > 0 ? bucket_info.num_shards : 1); | |
8945 | ||
11fdf7f2 | 8946 | new_num_shards = std::min(new_num_shards, get_max_bucket_shards()); |
31f18b77 FG |
8947 | if (new_num_shards <= num_source_shards) { |
8948 | ldout(cct, 20) << "not resharding bucket name=" << bucket_info.bucket.name << ", orig_num=" << num_source_shards << ", new_num_shards=" << new_num_shards << dendl; | |
8949 | return 0; | |
8950 | } | |
8951 | ||
8952 | cls_rgw_reshard_entry entry; | |
8953 | entry.time = real_clock::now(); | |
8954 | entry.tenant = bucket_info.owner.tenant; | |
8955 | entry.bucket_name = bucket_info.bucket.name; | |
8956 | entry.bucket_id = bucket_info.bucket.bucket_id; | |
8957 | entry.old_num_shards = num_source_shards; | |
8958 | entry.new_num_shards = new_num_shards; | |
8959 | ||
8960 | return reshard.add(entry); | |
8961 | } | |
8962 | ||
7c673cae | 8963 | int RGWRados::check_quota(const rgw_user& bucket_owner, rgw_bucket& bucket, |
11fdf7f2 | 8964 | RGWQuotaInfo& user_quota, RGWQuotaInfo& bucket_quota, uint64_t obj_size, bool check_size_only) |
7c673cae | 8965 | { |
11fdf7f2 TL |
8966 | // if we only check size, then num_objs will set to 0 |
8967 | if(check_size_only) | |
8968 | return quota_handler->check_quota(bucket_owner, bucket, user_quota, bucket_quota, 0, obj_size); | |
8969 | ||
7c673cae FG |
8970 | return quota_handler->check_quota(bucket_owner, bucket, user_quota, bucket_quota, 1, obj_size); |
8971 | } | |
8972 | ||
11fdf7f2 TL |
8973 | int RGWRados::get_target_shard_id(const RGWBucketInfo& bucket_info, const string& obj_key, |
8974 | int *shard_id) | |
7c673cae | 8975 | { |
11fdf7f2 TL |
8976 | int r = 0; |
8977 | switch (bucket_info.bucket_index_shard_hash_type) { | |
8978 | case RGWBucketInfo::MOD: | |
8979 | if (!bucket_info.num_shards) { | |
8980 | if (shard_id) { | |
8981 | *shard_id = -1; | |
8982 | } | |
8983 | } else { | |
9f95a23c | 8984 | uint32_t sid = svc.bi_rados->bucket_shard_index(obj_key, bucket_info.num_shards); |
11fdf7f2 TL |
8985 | if (shard_id) { |
8986 | *shard_id = (int)sid; | |
8987 | } | |
8988 | } | |
8989 | break; | |
8990 | default: | |
8991 | r = -ENOTSUP; | |
7c673cae | 8992 | } |
11fdf7f2 | 8993 | return r; |
7c673cae FG |
8994 | } |
8995 | ||
7c673cae FG |
8996 | uint64_t RGWRados::instance_id() |
8997 | { | |
8998 | return get_rados_handle()->get_instance_id(); | |
8999 | } | |
9000 | ||
9001 | uint64_t RGWRados::next_bucket_id() | |
9002 | { | |
9f95a23c | 9003 | std::lock_guard l{bucket_id_lock}; |
7c673cae FG |
9004 | return ++max_bucket_id; |
9005 | } | |
9006 | ||
7c673cae FG |
9007 | librados::Rados* RGWRados::get_rados_handle() |
9008 | { | |
494da23a | 9009 | return &rados; |
7c673cae FG |
9010 | } |
9011 | ||
9012 | int RGWRados::delete_raw_obj_aio(const rgw_raw_obj& obj, list<librados::AioCompletion *>& handles) | |
9013 | { | |
9014 | rgw_rados_ref ref; | |
9015 | int ret = get_raw_obj_ref(obj, &ref); | |
9016 | if (ret < 0) { | |
9017 | lderr(cct) << "ERROR: failed to get obj ref with ret=" << ret << dendl; | |
9018 | return ret; | |
9019 | } | |
9020 | ||
9021 | ObjectWriteOperation op; | |
9022 | list<string> prefixes; | |
9023 | cls_rgw_remove_obj(op, prefixes); | |
9024 | ||
9f95a23c TL |
9025 | AioCompletion *c = librados::Rados::aio_create_completion(nullptr, nullptr); |
9026 | ret = ref.pool.ioctx().aio_operate(ref.obj.oid, c, &op); | |
7c673cae FG |
9027 | if (ret < 0) { |
9028 | lderr(cct) << "ERROR: AioOperate failed with ret=" << ret << dendl; | |
9029 | c->release(); | |
9030 | return ret; | |
9031 | } | |
9032 | ||
9033 | handles.push_back(c); | |
9034 | ||
9035 | return 0; | |
9036 | } | |
9037 | ||
9038 | int RGWRados::delete_obj_aio(const rgw_obj& obj, | |
9039 | RGWBucketInfo& bucket_info, RGWObjState *astate, | |
9f95a23c TL |
9040 | list<librados::AioCompletion *>& handles, bool keep_index_consistent, |
9041 | optional_yield y) | |
7c673cae FG |
9042 | { |
9043 | rgw_rados_ref ref; | |
9044 | int ret = get_obj_head_ref(bucket_info, obj, &ref); | |
9045 | if (ret < 0) { | |
9046 | lderr(cct) << "ERROR: failed to get obj ref with ret=" << ret << dendl; | |
9047 | return ret; | |
9048 | } | |
9049 | ||
9050 | if (keep_index_consistent) { | |
9051 | RGWRados::Bucket bop(this, bucket_info); | |
9052 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
9053 | ||
9f95a23c | 9054 | ret = index_op.prepare(CLS_RGW_OP_DEL, &astate->write_tag, y); |
7c673cae FG |
9055 | if (ret < 0) { |
9056 | lderr(cct) << "ERROR: failed to prepare index op with ret=" << ret << dendl; | |
9057 | return ret; | |
9058 | } | |
9059 | } | |
9060 | ||
9061 | ObjectWriteOperation op; | |
9062 | list<string> prefixes; | |
9063 | cls_rgw_remove_obj(op, prefixes); | |
9064 | ||
9f95a23c TL |
9065 | AioCompletion *c = librados::Rados::aio_create_completion(nullptr, nullptr); |
9066 | ret = ref.pool.ioctx().aio_operate(ref.obj.oid, c, &op); | |
7c673cae FG |
9067 | if (ret < 0) { |
9068 | lderr(cct) << "ERROR: AioOperate failed with ret=" << ret << dendl; | |
9069 | c->release(); | |
9070 | return ret; | |
9071 | } | |
9072 | ||
9073 | handles.push_back(c); | |
9074 | ||
9075 | if (keep_index_consistent) { | |
494da23a | 9076 | ret = delete_obj_index(obj, astate->mtime); |
7c673cae FG |
9077 | if (ret < 0) { |
9078 | lderr(cct) << "ERROR: failed to delete obj index with ret=" << ret << dendl; | |
9079 | return ret; | |
9080 | } | |
9081 | } | |
9082 | return ret; | |
9083 | } |