]>
Commit | Line | Data |
---|---|---|
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- | |
2 | // vim: ts=8 sw=2 smarttab ft=cpp | |
3 | ||
4 | #include "include/compat.h" | |
5 | #include <errno.h> | |
6 | #include <stdlib.h> | |
7 | #include <sys/types.h> | |
8 | #include <sstream> | |
9 | ||
10 | #include <boost/algorithm/string.hpp> | |
11 | #include <string_view> | |
12 | ||
13 | #include <boost/container/flat_set.hpp> | |
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" | |
19 | ||
20 | #include "common/errno.h" | |
21 | #include "common/Formatter.h" | |
22 | #include "common/Throttle.h" | |
23 | ||
24 | #include "rgw_sal.h" | |
25 | #include "rgw_zone.h" | |
26 | #include "rgw_cache.h" | |
27 | #include "rgw_acl.h" | |
28 | #include "rgw_acl_s3.h" /* for dumping s3policy in debug log */ | |
29 | #include "rgw_aio_throttle.h" | |
30 | #include "rgw_bucket.h" | |
31 | #include "rgw_rest_conn.h" | |
32 | #include "rgw_cr_rados.h" | |
33 | #include "rgw_cr_rest.h" | |
34 | #include "rgw_datalog.h" | |
35 | #include "rgw_putobj_processor.h" | |
36 | ||
37 | #include "cls/rgw/cls_rgw_ops.h" | |
38 | #include "cls/rgw/cls_rgw_client.h" | |
39 | #include "cls/rgw/cls_rgw_const.h" | |
40 | #include "cls/refcount/cls_refcount_client.h" | |
41 | #include "cls/version/cls_version_client.h" | |
42 | #include "osd/osd_types.h" | |
43 | ||
44 | #include "rgw_tools.h" | |
45 | #include "rgw_coroutine.h" | |
46 | #include "rgw_compression.h" | |
47 | #include "rgw_etag_verifier.h" | |
48 | #include "rgw_worker.h" | |
49 | #include "rgw_notify.h" | |
50 | ||
51 | #undef fork // fails to compile RGWPeriod::fork() below | |
52 | ||
53 | #include "common/Clock.h" | |
54 | ||
55 | using namespace librados; | |
56 | ||
57 | #include <string> | |
58 | #include <iostream> | |
59 | #include <vector> | |
60 | #include <atomic> | |
61 | #include <list> | |
62 | #include <map> | |
63 | #include "include/random.h" | |
64 | ||
65 | #include "rgw_gc.h" | |
66 | #include "rgw_lc.h" | |
67 | ||
68 | #include "rgw_object_expirer_core.h" | |
69 | #include "rgw_sync.h" | |
70 | #include "rgw_sync_counters.h" | |
71 | #include "rgw_sync_trace.h" | |
72 | #include "rgw_trim_datalog.h" | |
73 | #include "rgw_trim_mdlog.h" | |
74 | #include "rgw_data_sync.h" | |
75 | #include "rgw_realm_watcher.h" | |
76 | #include "rgw_reshard.h" | |
77 | ||
78 | #include "services/svc_zone.h" | |
79 | #include "services/svc_zone_utils.h" | |
80 | #include "services/svc_quota.h" | |
81 | #include "services/svc_sync_modules.h" | |
82 | #include "services/svc_sys_obj.h" | |
83 | #include "services/svc_sys_obj_cache.h" | |
84 | #include "services/svc_bucket.h" | |
85 | #include "services/svc_mdlog.h" | |
86 | ||
87 | #include "compressor/Compressor.h" | |
88 | ||
89 | #ifdef WITH_LTTNG | |
90 | #define TRACEPOINT_DEFINE | |
91 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
92 | #include "tracing/rgw_rados.h" | |
93 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
94 | #undef TRACEPOINT_DEFINE | |
95 | #else | |
96 | #define tracepoint(...) | |
97 | #endif | |
98 | ||
99 | #define dout_context g_ceph_context | |
100 | #define dout_subsys ceph_subsys_rgw | |
101 | ||
102 | ||
103 | static string shadow_ns = "shadow"; | |
104 | static string default_bucket_index_pool_suffix = "rgw.buckets.index"; | |
105 | static string default_storage_extra_pool_suffix = "rgw.buckets.non-ec"; | |
106 | ||
107 | static RGWObjCategory main_category = RGWObjCategory::Main; | |
108 | #define RGW_USAGE_OBJ_PREFIX "usage." | |
109 | ||
110 | #define dout_subsys ceph_subsys_rgw | |
111 | ||
112 | ||
113 | static bool rgw_get_obj_data_pool(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params, | |
114 | const rgw_placement_rule& head_placement_rule, | |
115 | const rgw_obj& obj, rgw_pool *pool) | |
116 | { | |
117 | if (!zone_params.get_head_data_pool(head_placement_rule, obj, pool)) { | |
118 | RGWZonePlacementInfo placement; | |
119 | if (!zone_params.get_placement(zonegroup.default_placement.name, &placement)) { | |
120 | return false; | |
121 | } | |
122 | ||
123 | if (!obj.in_extra_data) { | |
124 | *pool = placement.get_data_pool(zonegroup.default_placement.storage_class); | |
125 | } else { | |
126 | *pool = placement.get_data_extra_pool(); | |
127 | } | |
128 | } | |
129 | ||
130 | return true; | |
131 | } | |
132 | ||
133 | static bool rgw_obj_to_raw(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params, | |
134 | const rgw_placement_rule& head_placement_rule, | |
135 | const rgw_obj& obj, rgw_raw_obj *raw_obj) | |
136 | { | |
137 | get_obj_bucket_and_oid_loc(obj, raw_obj->oid, raw_obj->loc); | |
138 | ||
139 | return rgw_get_obj_data_pool(zonegroup, zone_params, head_placement_rule, obj, &raw_obj->pool); | |
140 | } | |
141 | ||
142 | rgw_raw_obj rgw_obj_select::get_raw_obj(const RGWZoneGroup& zonegroup, const RGWZoneParams& zone_params) const | |
143 | { | |
144 | if (!is_raw) { | |
145 | rgw_raw_obj r; | |
146 | rgw_obj_to_raw(zonegroup, zone_params, placement_rule, obj, &r); | |
147 | return r; | |
148 | } | |
149 | return raw_obj; | |
150 | } | |
151 | ||
152 | rgw_raw_obj rgw_obj_select::get_raw_obj(rgw::sal::RGWStore* store) const | |
153 | { | |
154 | if (!is_raw) { | |
155 | rgw_raw_obj r; | |
156 | store->get_raw_obj(placement_rule, obj, &r); | |
157 | return r; | |
158 | } | |
159 | return raw_obj; | |
160 | } | |
161 | ||
162 | void RGWObjVersionTracker::prepare_op_for_read(ObjectReadOperation *op) | |
163 | { | |
164 | obj_version *check_objv = version_for_check(); | |
165 | ||
166 | if (check_objv) { | |
167 | cls_version_check(*op, *check_objv, VER_COND_EQ); | |
168 | } | |
169 | ||
170 | cls_version_read(*op, &read_version); | |
171 | } | |
172 | ||
173 | void RGWObjVersionTracker::prepare_op_for_write(ObjectWriteOperation *op) | |
174 | { | |
175 | obj_version *check_objv = version_for_check(); | |
176 | obj_version *modify_version = version_for_write(); | |
177 | ||
178 | if (check_objv) { | |
179 | cls_version_check(*op, *check_objv, VER_COND_EQ); | |
180 | } | |
181 | ||
182 | if (modify_version) { | |
183 | cls_version_set(*op, *modify_version); | |
184 | } else { | |
185 | cls_version_inc(*op); | |
186 | } | |
187 | } | |
188 | ||
189 | void RGWObjVersionTracker::apply_write() | |
190 | { | |
191 | const bool checked = (read_version.ver != 0); | |
192 | const bool incremented = (write_version.ver == 0); | |
193 | ||
194 | if (checked && incremented) { | |
195 | // apply cls_version_inc() so our next operation can recheck it | |
196 | ++read_version.ver; | |
197 | } else { | |
198 | read_version = write_version; | |
199 | } | |
200 | write_version = obj_version(); | |
201 | } | |
202 | ||
203 | RGWObjState::RGWObjState() { | |
204 | } | |
205 | ||
206 | RGWObjState::~RGWObjState() { | |
207 | } | |
208 | ||
209 | RGWObjState::RGWObjState(const RGWObjState& rhs) : obj (rhs.obj) { | |
210 | is_atomic = rhs.is_atomic; | |
211 | has_attrs = rhs.has_attrs; | |
212 | exists = rhs.exists; | |
213 | size = rhs.size; | |
214 | accounted_size = rhs.accounted_size; | |
215 | mtime = rhs.mtime; | |
216 | epoch = rhs.epoch; | |
217 | if (rhs.obj_tag.length()) { | |
218 | obj_tag = rhs.obj_tag; | |
219 | } | |
220 | if (rhs.tail_tag.length()) { | |
221 | tail_tag = rhs.tail_tag; | |
222 | } | |
223 | write_tag = rhs.write_tag; | |
224 | fake_tag = rhs.fake_tag; | |
225 | manifest = rhs.manifest; | |
226 | shadow_obj = rhs.shadow_obj; | |
227 | has_data = rhs.has_data; | |
228 | if (rhs.data.length()) { | |
229 | data = rhs.data; | |
230 | } | |
231 | prefetch_data = rhs.prefetch_data; | |
232 | keep_tail = rhs.keep_tail; | |
233 | is_olh = rhs.is_olh; | |
234 | objv_tracker = rhs.objv_tracker; | |
235 | pg_ver = rhs.pg_ver; | |
236 | } | |
237 | ||
238 | RGWObjState *RGWObjectCtx::get_state(const rgw_obj& obj) { | |
239 | RGWObjState *result; | |
240 | typename std::map<rgw_obj, RGWObjState>::iterator iter; | |
241 | lock.lock_shared(); | |
242 | assert (!obj.empty()); | |
243 | iter = objs_state.find(obj); | |
244 | if (iter != objs_state.end()) { | |
245 | result = &iter->second; | |
246 | lock.unlock_shared(); | |
247 | } else { | |
248 | lock.unlock_shared(); | |
249 | lock.lock(); | |
250 | result = &objs_state[obj]; | |
251 | lock.unlock(); | |
252 | } | |
253 | return result; | |
254 | } | |
255 | ||
256 | void RGWObjectCtx::set_atomic(rgw_obj& obj) { | |
257 | std::unique_lock wl{lock}; | |
258 | assert (!obj.empty()); | |
259 | objs_state[obj].is_atomic = true; | |
260 | } | |
261 | void RGWObjectCtx::set_prefetch_data(const rgw_obj& obj) { | |
262 | std::unique_lock wl{lock}; | |
263 | assert (!obj.empty()); | |
264 | objs_state[obj].prefetch_data = true; | |
265 | } | |
266 | ||
267 | void RGWObjectCtx::invalidate(const rgw_obj& obj) { | |
268 | std::unique_lock wl{lock}; | |
269 | auto iter = objs_state.find(obj); | |
270 | if (iter == objs_state.end()) { | |
271 | return; | |
272 | } | |
273 | bool is_atomic = iter->second.is_atomic; | |
274 | bool prefetch_data = iter->second.prefetch_data; | |
275 | ||
276 | objs_state.erase(iter); | |
277 | ||
278 | if (is_atomic || prefetch_data) { | |
279 | auto& state = objs_state[obj]; | |
280 | state.is_atomic = is_atomic; | |
281 | state.prefetch_data = prefetch_data; | |
282 | } | |
283 | } | |
284 | ||
285 | void RGWObjVersionTracker::generate_new_write_ver(CephContext *cct) | |
286 | { | |
287 | write_version.ver = 1; | |
288 | #define TAG_LEN 24 | |
289 | ||
290 | write_version.tag.clear(); | |
291 | append_rand_alpha(cct, write_version.tag, write_version.tag, TAG_LEN); | |
292 | } | |
293 | ||
294 | class RGWMetaNotifierManager : public RGWCoroutinesManager { | |
295 | RGWRados *store; | |
296 | RGWHTTPManager http_manager; | |
297 | ||
298 | public: | |
299 | RGWMetaNotifierManager(RGWRados *_store) : RGWCoroutinesManager(_store->ctx(), _store->get_cr_registry()), store(_store), | |
300 | http_manager(store->ctx(), completion_mgr) { | |
301 | http_manager.start(); | |
302 | } | |
303 | ||
304 | int notify_all(const DoutPrefixProvider *dpp, map<rgw_zone_id, RGWRESTConn *>& conn_map, set<int>& shards) { | |
305 | rgw_http_param_pair pairs[] = { { "type", "metadata" }, | |
306 | { "notify", NULL }, | |
307 | { NULL, NULL } }; | |
308 | ||
309 | list<RGWCoroutinesStack *> stacks; | |
310 | for (auto iter = conn_map.begin(); iter != conn_map.end(); ++iter) { | |
311 | RGWRESTConn *conn = iter->second; | |
312 | RGWCoroutinesStack *stack = new RGWCoroutinesStack(store->ctx(), this); | |
313 | stack->call(new RGWPostRESTResourceCR<set<int>, int>(store->ctx(), conn, &http_manager, "/admin/log", pairs, shards, NULL)); | |
314 | ||
315 | stacks.push_back(stack); | |
316 | } | |
317 | return run(dpp, stacks); | |
318 | } | |
319 | }; | |
320 | ||
321 | class RGWDataNotifierManager : public RGWCoroutinesManager { | |
322 | RGWRados *store; | |
323 | RGWHTTPManager http_manager; | |
324 | ||
325 | public: | |
326 | RGWDataNotifierManager(RGWRados *_store) : RGWCoroutinesManager(_store->ctx(), _store->get_cr_registry()), store(_store), | |
327 | http_manager(store->ctx(), completion_mgr) { | |
328 | http_manager.start(); | |
329 | } | |
330 | ||
331 | int notify_all(const DoutPrefixProvider *dpp, map<rgw_zone_id, RGWRESTConn *>& conn_map, | |
332 | bc::flat_map<int, bc::flat_set<string> >& shards) { | |
333 | rgw_http_param_pair pairs[] = { { "type", "data" }, | |
334 | { "notify", NULL }, | |
335 | { "source-zone", store->svc.zone->get_zone_params().get_id().c_str() }, | |
336 | { NULL, NULL } }; | |
337 | ||
338 | list<RGWCoroutinesStack *> stacks; | |
339 | for (auto iter = conn_map.begin(); iter != conn_map.end(); ++iter) { | |
340 | RGWRESTConn *conn = iter->second; | |
341 | RGWCoroutinesStack *stack = new RGWCoroutinesStack(store->ctx(), this); | |
342 | stack->call(new RGWPostRESTResourceCR<bc::flat_map<int, bc::flat_set<string> >, int>(store->ctx(), conn, &http_manager, "/admin/log", pairs, shards, NULL)); | |
343 | ||
344 | stacks.push_back(stack); | |
345 | } | |
346 | return run(dpp, stacks); | |
347 | } | |
348 | }; | |
349 | ||
350 | /* class RGWRadosThread */ | |
351 | ||
352 | void RGWRadosThread::start() | |
353 | { | |
354 | worker = new Worker(cct, this); | |
355 | worker->create(thread_name.c_str()); | |
356 | } | |
357 | ||
358 | void RGWRadosThread::stop() | |
359 | { | |
360 | down_flag = true; | |
361 | stop_process(); | |
362 | if (worker) { | |
363 | worker->signal(); | |
364 | worker->join(); | |
365 | } | |
366 | delete worker; | |
367 | worker = NULL; | |
368 | } | |
369 | ||
370 | void *RGWRadosThread::Worker::entry() { | |
371 | uint64_t msec = processor->interval_msec(); | |
372 | auto interval = std::chrono::milliseconds(msec); | |
373 | ||
374 | do { | |
375 | auto start = ceph::real_clock::now(); | |
376 | int r = processor->process(this); | |
377 | if (r < 0) { | |
378 | ldpp_dout(this, 0) << "ERROR: processor->process() returned error r=" << r << dendl; | |
379 | } | |
380 | ||
381 | if (processor->going_down()) | |
382 | break; | |
383 | ||
384 | auto end = ceph::real_clock::now() - start; | |
385 | ||
386 | uint64_t cur_msec = processor->interval_msec(); | |
387 | if (cur_msec != msec) { /* was it reconfigured? */ | |
388 | msec = cur_msec; | |
389 | interval = std::chrono::milliseconds(msec); | |
390 | } | |
391 | ||
392 | if (cur_msec > 0) { | |
393 | if (interval <= end) | |
394 | continue; // next round | |
395 | ||
396 | auto wait_time = interval - end; | |
397 | wait_interval(wait_time); | |
398 | } else { | |
399 | wait(); | |
400 | } | |
401 | } while (!processor->going_down()); | |
402 | ||
403 | return NULL; | |
404 | } | |
405 | ||
406 | class RGWMetaNotifier : public RGWRadosThread { | |
407 | RGWMetaNotifierManager notify_mgr; | |
408 | RGWMetadataLog *const log; | |
409 | ||
410 | uint64_t interval_msec() override { | |
411 | return cct->_conf->rgw_md_notify_interval_msec; | |
412 | } | |
413 | void stop_process() override { | |
414 | notify_mgr.stop(); | |
415 | } | |
416 | public: | |
417 | RGWMetaNotifier(RGWRados *_store, RGWMetadataLog* log) | |
418 | : RGWRadosThread(_store, "meta-notifier"), notify_mgr(_store), log(log) {} | |
419 | ||
420 | int process(const DoutPrefixProvider *dpp) override; | |
421 | }; | |
422 | ||
423 | int RGWMetaNotifier::process(const DoutPrefixProvider *dpp) | |
424 | { | |
425 | set<int> shards; | |
426 | ||
427 | log->read_clear_modified(shards); | |
428 | ||
429 | if (shards.empty()) { | |
430 | return 0; | |
431 | } | |
432 | ||
433 | for (set<int>::iterator iter = shards.begin(); iter != shards.end(); ++iter) { | |
434 | ldpp_dout(dpp, 20) << __func__ << "(): notifying mdlog change, shard_id=" << *iter << dendl; | |
435 | } | |
436 | ||
437 | notify_mgr.notify_all(dpp, store->svc.zone->get_zone_conn_map(), shards); | |
438 | ||
439 | return 0; | |
440 | } | |
441 | ||
442 | class RGWDataNotifier : public RGWRadosThread { | |
443 | RGWDataNotifierManager notify_mgr; | |
444 | ||
445 | uint64_t interval_msec() override { | |
446 | return cct->_conf.get_val<int64_t>("rgw_data_notify_interval_msec"); | |
447 | } | |
448 | void stop_process() override { | |
449 | notify_mgr.stop(); | |
450 | } | |
451 | public: | |
452 | RGWDataNotifier(RGWRados *_store) : RGWRadosThread(_store, "data-notifier"), notify_mgr(_store) {} | |
453 | ||
454 | int process(const DoutPrefixProvider *dpp) override; | |
455 | }; | |
456 | ||
457 | int RGWDataNotifier::process(const DoutPrefixProvider *dpp) | |
458 | { | |
459 | auto data_log = store->svc.datalog_rados; | |
460 | if (!data_log) { | |
461 | return 0; | |
462 | } | |
463 | ||
464 | auto shards = data_log->read_clear_modified(); | |
465 | ||
466 | if (shards.empty()) { | |
467 | return 0; | |
468 | } | |
469 | ||
470 | for (const auto& [shard_id, keys] : shards) { | |
471 | ldpp_dout(dpp, 20) << __func__ << "(): notifying datalog change, shard_id=" | |
472 | << shard_id << ": " << keys << dendl; | |
473 | } | |
474 | ||
475 | notify_mgr.notify_all(dpp, store->svc.zone->get_zone_data_notify_to_map(), shards); | |
476 | ||
477 | return 0; | |
478 | } | |
479 | ||
480 | class RGWSyncProcessorThread : public RGWRadosThread { | |
481 | public: | |
482 | RGWSyncProcessorThread(RGWRados *_store, const string& thread_name = "radosgw") : RGWRadosThread(_store, thread_name) {} | |
483 | RGWSyncProcessorThread(RGWRados *_store) : RGWRadosThread(_store) {} | |
484 | ~RGWSyncProcessorThread() override {} | |
485 | int init(const DoutPrefixProvider *dpp) override = 0 ; | |
486 | int process(const DoutPrefixProvider *dpp) override = 0; | |
487 | }; | |
488 | ||
489 | class RGWMetaSyncProcessorThread : public RGWSyncProcessorThread | |
490 | { | |
491 | RGWMetaSyncStatusManager sync; | |
492 | ||
493 | uint64_t interval_msec() override { | |
494 | return 0; /* no interval associated, it'll run once until stopped */ | |
495 | } | |
496 | void stop_process() override { | |
497 | sync.stop(); | |
498 | } | |
499 | public: | |
500 | RGWMetaSyncProcessorThread(rgw::sal::RGWRadosStore *_store, RGWAsyncRadosProcessor *async_rados) | |
501 | : RGWSyncProcessorThread(_store->getRados(), "meta-sync"), sync(_store, async_rados) {} | |
502 | ||
503 | void wakeup_sync_shards(set<int>& shard_ids) { | |
504 | for (set<int>::iterator iter = shard_ids.begin(); iter != shard_ids.end(); ++iter) { | |
505 | sync.wakeup(*iter); | |
506 | } | |
507 | } | |
508 | RGWMetaSyncStatusManager* get_manager() { return &sync; } | |
509 | ||
510 | int init(const DoutPrefixProvider *dpp) override { | |
511 | int ret = sync.init(dpp); | |
512 | if (ret < 0) { | |
513 | ldpp_dout(dpp, 0) << "ERROR: sync.init() returned " << ret << dendl; | |
514 | return ret; | |
515 | } | |
516 | return 0; | |
517 | } | |
518 | ||
519 | int process(const DoutPrefixProvider *dpp) override { | |
520 | sync.run(dpp, null_yield); | |
521 | return 0; | |
522 | } | |
523 | }; | |
524 | ||
525 | class RGWDataSyncProcessorThread : public RGWSyncProcessorThread | |
526 | { | |
527 | PerfCountersRef counters; | |
528 | RGWDataSyncStatusManager sync; | |
529 | bool initialized; | |
530 | ||
531 | uint64_t interval_msec() override { | |
532 | if (initialized) { | |
533 | return 0; /* no interval associated, it'll run once until stopped */ | |
534 | } else { | |
535 | #define DATA_SYNC_INIT_WAIT_SEC 20 | |
536 | return DATA_SYNC_INIT_WAIT_SEC * 1000; | |
537 | } | |
538 | } | |
539 | void stop_process() override { | |
540 | sync.stop(); | |
541 | } | |
542 | public: | |
543 | RGWDataSyncProcessorThread(rgw::sal::RGWRadosStore *_store, RGWAsyncRadosProcessor *async_rados, | |
544 | const RGWZone* source_zone) | |
545 | : RGWSyncProcessorThread(_store->getRados(), "data-sync"), | |
546 | counters(sync_counters::build(store->ctx(), std::string("data-sync-from-") + source_zone->name)), | |
547 | sync(_store, async_rados, source_zone->id, counters.get()), | |
548 | initialized(false) {} | |
549 | ||
550 | void wakeup_sync_shards(map<int, set<string> >& shard_ids) { | |
551 | for (map<int, set<string> >::iterator iter = shard_ids.begin(); iter != shard_ids.end(); ++iter) { | |
552 | sync.wakeup(iter->first, iter->second); | |
553 | } | |
554 | } | |
555 | RGWDataSyncStatusManager* get_manager() { return &sync; } | |
556 | ||
557 | int init(const DoutPrefixProvider *dpp) override { | |
558 | return 0; | |
559 | } | |
560 | ||
561 | int process(const DoutPrefixProvider *dpp) override { | |
562 | while (!initialized) { | |
563 | if (going_down()) { | |
564 | return 0; | |
565 | } | |
566 | int ret = sync.init(dpp); | |
567 | if (ret >= 0) { | |
568 | initialized = true; | |
569 | break; | |
570 | } | |
571 | /* we'll be back! */ | |
572 | return 0; | |
573 | } | |
574 | sync.run(dpp); | |
575 | return 0; | |
576 | } | |
577 | }; | |
578 | ||
579 | class RGWSyncLogTrimThread : public RGWSyncProcessorThread, DoutPrefixProvider | |
580 | { | |
581 | RGWCoroutinesManager crs; | |
582 | rgw::sal::RGWRadosStore *store; | |
583 | rgw::BucketTrimManager *bucket_trim; | |
584 | RGWHTTPManager http; | |
585 | const utime_t trim_interval; | |
586 | ||
587 | uint64_t interval_msec() override { return 0; } | |
588 | void stop_process() override { crs.stop(); } | |
589 | public: | |
590 | RGWSyncLogTrimThread(rgw::sal::RGWRadosStore *store, rgw::BucketTrimManager *bucket_trim, | |
591 | int interval) | |
592 | : RGWSyncProcessorThread(store->getRados(), "sync-log-trim"), | |
593 | crs(store->ctx(), store->getRados()->get_cr_registry()), store(store), | |
594 | bucket_trim(bucket_trim), | |
595 | http(store->ctx(), crs.get_completion_mgr()), | |
596 | trim_interval(interval, 0) | |
597 | {} | |
598 | ||
599 | int init(const DoutPrefixProvider *dpp) override { | |
600 | return http.start(); | |
601 | } | |
602 | int process(const DoutPrefixProvider *dpp) override { | |
603 | list<RGWCoroutinesStack*> stacks; | |
604 | auto meta = new RGWCoroutinesStack(store->ctx(), &crs); | |
605 | meta->call(create_meta_log_trim_cr(this, store, &http, | |
606 | cct->_conf->rgw_md_log_max_shards, | |
607 | trim_interval)); | |
608 | stacks.push_back(meta); | |
609 | ||
610 | if (store->svc()->zone->sync_module_exports_data()) { | |
611 | auto data = new RGWCoroutinesStack(store->ctx(), &crs); | |
612 | data->call(create_data_log_trim_cr(this, store, &http, | |
613 | cct->_conf->rgw_data_log_num_shards, | |
614 | trim_interval)); | |
615 | stacks.push_back(data); | |
616 | ||
617 | auto bucket = new RGWCoroutinesStack(store->ctx(), &crs); | |
618 | bucket->call(bucket_trim->create_bucket_trim_cr(&http)); | |
619 | stacks.push_back(bucket); | |
620 | } | |
621 | ||
622 | crs.run(dpp, stacks); | |
623 | return 0; | |
624 | } | |
625 | ||
626 | // implements DoutPrefixProvider | |
627 | CephContext *get_cct() const override { return store->ctx(); } | |
628 | unsigned get_subsys() const override | |
629 | { | |
630 | return dout_subsys; | |
631 | } | |
632 | ||
633 | std::ostream& gen_prefix(std::ostream& out) const override | |
634 | { | |
635 | return out << "sync log trim: "; | |
636 | } | |
637 | ||
638 | }; | |
639 | ||
640 | void RGWRados::wakeup_meta_sync_shards(set<int>& shard_ids) | |
641 | { | |
642 | std::lock_guard l{meta_sync_thread_lock}; | |
643 | if (meta_sync_processor_thread) { | |
644 | meta_sync_processor_thread->wakeup_sync_shards(shard_ids); | |
645 | } | |
646 | } | |
647 | ||
648 | void RGWRados::wakeup_data_sync_shards(const rgw_zone_id& source_zone, map<int, set<string> >& shard_ids) | |
649 | { | |
650 | ldout(ctx(), 20) << __func__ << ": source_zone=" << source_zone << ", shard_ids=" << shard_ids << dendl; | |
651 | std::lock_guard l{data_sync_thread_lock}; | |
652 | auto iter = data_sync_processor_threads.find(source_zone); | |
653 | if (iter == data_sync_processor_threads.end()) { | |
654 | ldout(ctx(), 10) << __func__ << ": couldn't find sync thread for zone " << source_zone << ", skipping async data sync processing" << dendl; | |
655 | return; | |
656 | } | |
657 | ||
658 | RGWDataSyncProcessorThread *thread = iter->second; | |
659 | ceph_assert(thread); | |
660 | thread->wakeup_sync_shards(shard_ids); | |
661 | } | |
662 | ||
663 | RGWMetaSyncStatusManager* RGWRados::get_meta_sync_manager() | |
664 | { | |
665 | std::lock_guard l{meta_sync_thread_lock}; | |
666 | if (meta_sync_processor_thread) { | |
667 | return meta_sync_processor_thread->get_manager(); | |
668 | } | |
669 | return nullptr; | |
670 | } | |
671 | ||
672 | RGWDataSyncStatusManager* RGWRados::get_data_sync_manager(const rgw_zone_id& source_zone) | |
673 | { | |
674 | std::lock_guard l{data_sync_thread_lock}; | |
675 | auto thread = data_sync_processor_threads.find(source_zone); | |
676 | if (thread == data_sync_processor_threads.end()) { | |
677 | return nullptr; | |
678 | } | |
679 | return thread->second->get_manager(); | |
680 | } | |
681 | ||
682 | int RGWRados::get_required_alignment(const DoutPrefixProvider *dpp, const rgw_pool& pool, uint64_t *alignment) | |
683 | { | |
684 | IoCtx ioctx; | |
685 | int r = open_pool_ctx(dpp, pool, ioctx, false); | |
686 | if (r < 0) { | |
687 | ldout(cct, 0) << "ERROR: open_pool_ctx() returned " << r << dendl; | |
688 | return r; | |
689 | } | |
690 | ||
691 | bool requires; | |
692 | r = ioctx.pool_requires_alignment2(&requires); | |
693 | if (r < 0) { | |
694 | ldout(cct, 0) << "ERROR: ioctx.pool_requires_alignment2() returned " | |
695 | << r << dendl; | |
696 | return r; | |
697 | } | |
698 | ||
699 | if (!requires) { | |
700 | *alignment = 0; | |
701 | return 0; | |
702 | } | |
703 | ||
704 | uint64_t align; | |
705 | r = ioctx.pool_required_alignment2(&align); | |
706 | if (r < 0) { | |
707 | ldout(cct, 0) << "ERROR: ioctx.pool_required_alignment2() returned " | |
708 | << r << dendl; | |
709 | return r; | |
710 | } | |
711 | if (align != 0) { | |
712 | ldout(cct, 20) << "required alignment=" << align << dendl; | |
713 | } | |
714 | *alignment = align; | |
715 | return 0; | |
716 | } | |
717 | ||
718 | void RGWRados::get_max_aligned_size(uint64_t size, uint64_t alignment, uint64_t *max_size) | |
719 | { | |
720 | if (alignment == 0) { | |
721 | *max_size = size; | |
722 | return; | |
723 | } | |
724 | ||
725 | if (size <= alignment) { | |
726 | *max_size = alignment; | |
727 | return; | |
728 | } | |
729 | ||
730 | *max_size = size - (size % alignment); | |
731 | } | |
732 | ||
733 | int RGWRados::get_max_chunk_size(const rgw_pool& pool, uint64_t *max_chunk_size, const DoutPrefixProvider *dpp, uint64_t *palignment) | |
734 | { | |
735 | uint64_t alignment; | |
736 | int r = get_required_alignment(dpp, pool, &alignment); | |
737 | if (r < 0) { | |
738 | return r; | |
739 | } | |
740 | ||
741 | if (palignment) { | |
742 | *palignment = alignment; | |
743 | } | |
744 | ||
745 | uint64_t config_chunk_size = cct->_conf->rgw_max_chunk_size; | |
746 | ||
747 | get_max_aligned_size(config_chunk_size, alignment, max_chunk_size); | |
748 | ||
749 | ldpp_dout(dpp, 20) << "max_chunk_size=" << *max_chunk_size << dendl; | |
750 | ||
751 | return 0; | |
752 | } | |
753 | ||
754 | int RGWRados::get_max_chunk_size(const rgw_placement_rule& placement_rule, const rgw_obj& obj, | |
755 | uint64_t *max_chunk_size, const DoutPrefixProvider *dpp, uint64_t *palignment) | |
756 | { | |
757 | rgw_pool pool; | |
758 | if (!get_obj_data_pool(placement_rule, obj, &pool)) { | |
759 | ldpp_dout(dpp, 0) << "ERROR: failed to get data pool for object " << obj << dendl; | |
760 | return -EIO; | |
761 | } | |
762 | return get_max_chunk_size(pool, max_chunk_size, dpp, palignment); | |
763 | } | |
764 | ||
765 | class RGWIndexCompletionManager; | |
766 | ||
767 | struct complete_op_data { | |
768 | ceph::mutex lock = ceph::make_mutex("complete_op_data"); | |
769 | AioCompletion *rados_completion{nullptr}; | |
770 | int manager_shard_id{-1}; | |
771 | RGWIndexCompletionManager *manager{nullptr}; | |
772 | rgw_obj obj; | |
773 | RGWModifyOp op; | |
774 | string tag; | |
775 | rgw_bucket_entry_ver ver; | |
776 | cls_rgw_obj_key key; | |
777 | rgw_bucket_dir_entry_meta dir_meta; | |
778 | list<cls_rgw_obj_key> remove_objs; | |
779 | bool log_op; | |
780 | uint16_t bilog_op; | |
781 | rgw_zone_set zones_trace; | |
782 | ||
783 | bool stopped{false}; | |
784 | ||
785 | void stop() { | |
786 | std::lock_guard l{lock}; | |
787 | stopped = true; | |
788 | } | |
789 | }; | |
790 | ||
791 | class RGWIndexCompletionThread : public RGWRadosThread, public DoutPrefixProvider { | |
792 | RGWRados *store; | |
793 | ||
794 | uint64_t interval_msec() override { | |
795 | return 0; | |
796 | } | |
797 | ||
798 | list<complete_op_data *> completions; | |
799 | ||
800 | ceph::mutex completions_lock = | |
801 | ceph::make_mutex("RGWIndexCompletionThread::completions_lock"); | |
802 | public: | |
803 | RGWIndexCompletionThread(RGWRados *_store) | |
804 | : RGWRadosThread(_store, "index-complete"), store(_store) {} | |
805 | ||
806 | int process(const DoutPrefixProvider *dpp) override; | |
807 | ||
808 | void add_completion(complete_op_data *completion) { | |
809 | { | |
810 | std::lock_guard l{completions_lock}; | |
811 | completions.push_back(completion); | |
812 | } | |
813 | ||
814 | signal(); | |
815 | } | |
816 | ||
817 | CephContext *get_cct() const override { return store->ctx(); } | |
818 | unsigned get_subsys() const { return dout_subsys; } | |
819 | std::ostream& gen_prefix(std::ostream& out) const { return out << "rgw index completion thread: "; } | |
820 | }; | |
821 | ||
822 | int RGWIndexCompletionThread::process(const DoutPrefixProvider *dpp) | |
823 | { | |
824 | list<complete_op_data *> comps; | |
825 | ||
826 | { | |
827 | std::lock_guard l{completions_lock}; | |
828 | completions.swap(comps); | |
829 | } | |
830 | ||
831 | for (auto c : comps) { | |
832 | std::unique_ptr<complete_op_data> up{c}; | |
833 | ||
834 | if (going_down()) { | |
835 | continue; | |
836 | } | |
837 | ldpp_dout(this, 20) << __func__ << "(): handling completion for key=" << c->key << dendl; | |
838 | ||
839 | RGWRados::BucketShard bs(store); | |
840 | RGWBucketInfo bucket_info; | |
841 | ||
842 | int r = bs.init(c->obj.bucket, c->obj, &bucket_info, this); | |
843 | if (r < 0) { | |
844 | ldpp_dout(this, 0) << "ERROR: " << __func__ << "(): failed to initialize BucketShard, obj=" << c->obj << " r=" << r << dendl; | |
845 | /* not much to do */ | |
846 | continue; | |
847 | } | |
848 | ||
849 | r = store->guard_reshard(this, &bs, c->obj, bucket_info, | |
850 | [&](RGWRados::BucketShard *bs) -> int { | |
851 | librados::ObjectWriteOperation o; | |
852 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); | |
853 | cls_rgw_bucket_complete_op(o, c->op, c->tag, c->ver, c->key, c->dir_meta, &c->remove_objs, | |
854 | c->log_op, c->bilog_op, &c->zones_trace); | |
855 | return bs->bucket_obj.operate(this, &o, null_yield); | |
856 | }); | |
857 | if (r < 0) { | |
858 | ldpp_dout(this, 0) << "ERROR: " << __func__ << "(): bucket index completion failed, obj=" << c->obj << " r=" << r << dendl; | |
859 | /* ignoring error, can't do anything about it */ | |
860 | continue; | |
861 | } | |
862 | r = store->svc.datalog_rados->add_entry(this, bucket_info, bs.shard_id); | |
863 | if (r < 0) { | |
864 | ldpp_dout(this, -1) << "ERROR: failed writing data log" << dendl; | |
865 | } | |
866 | } | |
867 | ||
868 | return 0; | |
869 | } | |
870 | ||
871 | class RGWIndexCompletionManager { | |
872 | RGWRados *store{nullptr}; | |
873 | ceph::containers::tiny_vector<ceph::mutex> locks; | |
874 | vector<set<complete_op_data *> > completions; | |
875 | ||
876 | RGWIndexCompletionThread *completion_thread{nullptr}; | |
877 | ||
878 | int num_shards; | |
879 | ||
880 | std::atomic<int> cur_shard {0}; | |
881 | ||
882 | ||
883 | public: | |
884 | RGWIndexCompletionManager(RGWRados *_store) : | |
885 | store(_store), | |
886 | locks{ceph::make_lock_container<ceph::mutex>( | |
887 | store->ctx()->_conf->rgw_thread_pool_size, | |
888 | [](const size_t i) { | |
889 | return ceph::make_mutex("RGWIndexCompletionManager::lock::" + | |
890 | std::to_string(i)); | |
891 | })} | |
892 | { | |
893 | num_shards = store->ctx()->_conf->rgw_thread_pool_size; | |
894 | completions.resize(num_shards); | |
895 | } | |
896 | ~RGWIndexCompletionManager() { | |
897 | stop(); | |
898 | } | |
899 | ||
900 | int next_shard() { | |
901 | int result = cur_shard % num_shards; | |
902 | cur_shard++; | |
903 | return result; | |
904 | } | |
905 | ||
906 | void create_completion(const rgw_obj& obj, | |
907 | RGWModifyOp op, string& tag, | |
908 | rgw_bucket_entry_ver& ver, | |
909 | const cls_rgw_obj_key& key, | |
910 | rgw_bucket_dir_entry_meta& dir_meta, | |
911 | list<cls_rgw_obj_key> *remove_objs, bool log_op, | |
912 | uint16_t bilog_op, | |
913 | rgw_zone_set *zones_trace, | |
914 | complete_op_data **result); | |
915 | bool handle_completion(completion_t cb, complete_op_data *arg); | |
916 | ||
917 | int start(const DoutPrefixProvider *dpp) { | |
918 | completion_thread = new RGWIndexCompletionThread(store); | |
919 | int ret = completion_thread->init(dpp); | |
920 | if (ret < 0) { | |
921 | return ret; | |
922 | } | |
923 | completion_thread->start(); | |
924 | return 0; | |
925 | } | |
926 | void stop() { | |
927 | if (completion_thread) { | |
928 | completion_thread->stop(); | |
929 | delete completion_thread; | |
930 | } | |
931 | ||
932 | for (int i = 0; i < num_shards; ++i) { | |
933 | std::lock_guard l{locks[i]}; | |
934 | for (auto c : completions[i]) { | |
935 | c->stop(); | |
936 | } | |
937 | } | |
938 | completions.clear(); | |
939 | } | |
940 | }; | |
941 | ||
942 | static void obj_complete_cb(completion_t cb, void *arg) | |
943 | { | |
944 | complete_op_data *completion = (complete_op_data *)arg; | |
945 | completion->lock.lock(); | |
946 | if (completion->stopped) { | |
947 | completion->lock.unlock(); /* can drop lock, no one else is referencing us */ | |
948 | delete completion; | |
949 | return; | |
950 | } | |
951 | bool need_delete = completion->manager->handle_completion(cb, completion); | |
952 | completion->lock.unlock(); | |
953 | if (need_delete) { | |
954 | delete completion; | |
955 | } | |
956 | } | |
957 | ||
958 | ||
959 | void RGWIndexCompletionManager::create_completion(const rgw_obj& obj, | |
960 | RGWModifyOp op, string& tag, | |
961 | rgw_bucket_entry_ver& ver, | |
962 | const cls_rgw_obj_key& key, | |
963 | rgw_bucket_dir_entry_meta& dir_meta, | |
964 | list<cls_rgw_obj_key> *remove_objs, bool log_op, | |
965 | uint16_t bilog_op, | |
966 | rgw_zone_set *zones_trace, | |
967 | complete_op_data **result) | |
968 | { | |
969 | complete_op_data *entry = new complete_op_data; | |
970 | ||
971 | int shard_id = next_shard(); | |
972 | ||
973 | entry->manager_shard_id = shard_id; | |
974 | entry->manager = this; | |
975 | entry->obj = obj; | |
976 | entry->op = op; | |
977 | entry->tag = tag; | |
978 | entry->ver = ver; | |
979 | entry->key = key; | |
980 | entry->dir_meta = dir_meta; | |
981 | entry->log_op = log_op; | |
982 | entry->bilog_op = bilog_op; | |
983 | ||
984 | if (remove_objs) { | |
985 | for (auto iter = remove_objs->begin(); iter != remove_objs->end(); ++iter) { | |
986 | entry->remove_objs.push_back(*iter); | |
987 | } | |
988 | } | |
989 | ||
990 | if (zones_trace) { | |
991 | entry->zones_trace = *zones_trace; | |
992 | } else { | |
993 | entry->zones_trace.insert(store->svc.zone->get_zone().id, obj.bucket.get_key()); | |
994 | } | |
995 | ||
996 | *result = entry; | |
997 | ||
998 | entry->rados_completion = librados::Rados::aio_create_completion(entry, obj_complete_cb); | |
999 | ||
1000 | std::lock_guard l{locks[shard_id]}; | |
1001 | completions[shard_id].insert(entry); | |
1002 | } | |
1003 | ||
1004 | bool RGWIndexCompletionManager::handle_completion(completion_t cb, complete_op_data *arg) | |
1005 | { | |
1006 | int shard_id = arg->manager_shard_id; | |
1007 | { | |
1008 | std::lock_guard l{locks[shard_id]}; | |
1009 | ||
1010 | auto& comps = completions[shard_id]; | |
1011 | ||
1012 | auto iter = comps.find(arg); | |
1013 | if (iter == comps.end()) { | |
1014 | return true; | |
1015 | } | |
1016 | ||
1017 | comps.erase(iter); | |
1018 | } | |
1019 | ||
1020 | int r = rados_aio_get_return_value(cb); | |
1021 | if (r != -ERR_BUSY_RESHARDING) { | |
1022 | return true; | |
1023 | } | |
1024 | completion_thread->add_completion(arg); | |
1025 | return false; | |
1026 | } | |
1027 | ||
1028 | void RGWRados::finalize() | |
1029 | { | |
1030 | if (run_sync_thread) { | |
1031 | std::lock_guard l{meta_sync_thread_lock}; | |
1032 | meta_sync_processor_thread->stop(); | |
1033 | ||
1034 | std::lock_guard dl{data_sync_thread_lock}; | |
1035 | for (auto iter : data_sync_processor_threads) { | |
1036 | RGWDataSyncProcessorThread *thread = iter.second; | |
1037 | thread->stop(); | |
1038 | } | |
1039 | if (sync_log_trimmer) { | |
1040 | sync_log_trimmer->stop(); | |
1041 | } | |
1042 | } | |
1043 | if (run_sync_thread) { | |
1044 | delete meta_sync_processor_thread; | |
1045 | meta_sync_processor_thread = NULL; | |
1046 | std::lock_guard dl{data_sync_thread_lock}; | |
1047 | for (auto iter : data_sync_processor_threads) { | |
1048 | RGWDataSyncProcessorThread *thread = iter.second; | |
1049 | delete thread; | |
1050 | } | |
1051 | data_sync_processor_threads.clear(); | |
1052 | delete sync_log_trimmer; | |
1053 | sync_log_trimmer = nullptr; | |
1054 | bucket_trim = boost::none; | |
1055 | } | |
1056 | if (meta_notifier) { | |
1057 | meta_notifier->stop(); | |
1058 | delete meta_notifier; | |
1059 | } | |
1060 | if (data_notifier) { | |
1061 | data_notifier->stop(); | |
1062 | delete data_notifier; | |
1063 | } | |
1064 | delete sync_tracer; | |
1065 | ||
1066 | delete lc; | |
1067 | lc = NULL; | |
1068 | ||
1069 | delete gc; | |
1070 | gc = NULL; | |
1071 | ||
1072 | delete obj_expirer; | |
1073 | obj_expirer = NULL; | |
1074 | ||
1075 | RGWQuotaHandler::free_handler(quota_handler); | |
1076 | if (cr_registry) { | |
1077 | cr_registry->put(); | |
1078 | } | |
1079 | ||
1080 | svc.shutdown(); | |
1081 | ||
1082 | delete binfo_cache; | |
1083 | delete obj_tombstone_cache; | |
1084 | ||
1085 | if (reshard_wait.get()) { | |
1086 | reshard_wait->stop(); | |
1087 | reshard_wait.reset(); | |
1088 | } | |
1089 | ||
1090 | if (run_reshard_thread) { | |
1091 | reshard->stop_processor(); | |
1092 | } | |
1093 | delete reshard; | |
1094 | delete index_completion_manager; | |
1095 | ||
1096 | rgw::notify::shutdown(); | |
1097 | } | |
1098 | ||
1099 | /** | |
1100 | * Initialize the RADOS instance and prepare to do other ops | |
1101 | * Returns 0 on success, -ERR# on failure. | |
1102 | */ | |
1103 | int RGWRados::init_rados() | |
1104 | { | |
1105 | int ret = 0; | |
1106 | ||
1107 | ret = rados.init_with_context(cct); | |
1108 | if (ret < 0) { | |
1109 | return ret; | |
1110 | } | |
1111 | ret = rados.connect(); | |
1112 | if (ret < 0) { | |
1113 | return ret; | |
1114 | } | |
1115 | ||
1116 | auto crs = std::unique_ptr<RGWCoroutinesManagerRegistry>{ | |
1117 | new RGWCoroutinesManagerRegistry(cct)}; | |
1118 | ret = crs->hook_to_admin_command("cr dump"); | |
1119 | if (ret < 0) { | |
1120 | return ret; | |
1121 | } | |
1122 | ||
1123 | cr_registry = crs.release(); | |
1124 | return ret; | |
1125 | } | |
1126 | ||
1127 | int RGWRados::register_to_service_map(const string& daemon_type, const map<string, string>& meta) | |
1128 | { | |
1129 | string name = cct->_conf->name.get_id(); | |
1130 | if (name.compare(0, 4, "rgw.") == 0) { | |
1131 | name = name.substr(4); | |
1132 | } | |
1133 | map<string,string> metadata = meta; | |
1134 | metadata["num_handles"] = "1"s; | |
1135 | metadata["zonegroup_id"] = svc.zone->get_zonegroup().get_id(); | |
1136 | metadata["zonegroup_name"] = svc.zone->get_zonegroup().get_name(); | |
1137 | metadata["zone_name"] = svc.zone->zone_name(); | |
1138 | metadata["zone_id"] = svc.zone->zone_id().id; | |
1139 | metadata["id"] = name; | |
1140 | int ret = rados.service_daemon_register( | |
1141 | daemon_type, | |
1142 | stringify(rados.get_instance_id()), | |
1143 | metadata); | |
1144 | if (ret < 0) { | |
1145 | ldout(cct, 0) << "ERROR: service_daemon_register() returned ret=" << ret << ": " << cpp_strerror(-ret) << dendl; | |
1146 | return ret; | |
1147 | } | |
1148 | ||
1149 | return 0; | |
1150 | } | |
1151 | ||
1152 | int RGWRados::update_service_map(std::map<std::string, std::string>&& status) | |
1153 | { | |
1154 | int ret = rados.service_daemon_update_status(move(status)); | |
1155 | if (ret < 0) { | |
1156 | ldout(cct, 0) << "ERROR: service_daemon_update_status() returned ret=" << ret << ": " << cpp_strerror(-ret) << dendl; | |
1157 | return ret; | |
1158 | } | |
1159 | ||
1160 | return 0; | |
1161 | } | |
1162 | ||
1163 | /** | |
1164 | * Initialize the RADOS instance and prepare to do other ops | |
1165 | * Returns 0 on success, -ERR# on failure. | |
1166 | */ | |
1167 | int RGWRados::init_complete(const DoutPrefixProvider *dpp) | |
1168 | { | |
1169 | int ret; | |
1170 | ||
1171 | /* | |
1172 | * create sync module instance even if we don't run sync thread, might need it for radosgw-admin | |
1173 | */ | |
1174 | sync_module = svc.sync_modules->get_sync_module(); | |
1175 | ||
1176 | ret = open_root_pool_ctx(dpp); | |
1177 | if (ret < 0) | |
1178 | return ret; | |
1179 | ||
1180 | ret = open_gc_pool_ctx(dpp); | |
1181 | if (ret < 0) | |
1182 | return ret; | |
1183 | ||
1184 | ret = open_lc_pool_ctx(dpp); | |
1185 | if (ret < 0) | |
1186 | return ret; | |
1187 | ||
1188 | ret = open_objexp_pool_ctx(dpp); | |
1189 | if (ret < 0) | |
1190 | return ret; | |
1191 | ||
1192 | ret = open_reshard_pool_ctx(dpp); | |
1193 | if (ret < 0) | |
1194 | return ret; | |
1195 | ||
1196 | ret = open_notif_pool_ctx(dpp); | |
1197 | if (ret < 0) | |
1198 | return ret; | |
1199 | ||
1200 | pools_initialized = true; | |
1201 | ||
1202 | gc = new RGWGC(); | |
1203 | gc->initialize(cct, this); | |
1204 | ||
1205 | obj_expirer = new RGWObjectExpirer(this->store); | |
1206 | ||
1207 | if (use_gc_thread) { | |
1208 | gc->start_processor(); | |
1209 | obj_expirer->start_processor(); | |
1210 | } | |
1211 | ||
1212 | auto& current_period = svc.zone->get_current_period(); | |
1213 | auto& zonegroup = svc.zone->get_zonegroup(); | |
1214 | auto& zone_params = svc.zone->get_zone_params(); | |
1215 | auto& zone = svc.zone->get_zone(); | |
1216 | ||
1217 | /* no point of running sync thread if we don't have a master zone configured | |
1218 | or there is no rest_master_conn */ | |
1219 | if (!svc.zone->need_to_sync()) { | |
1220 | run_sync_thread = false; | |
1221 | } | |
1222 | ||
1223 | if (svc.zone->is_meta_master()) { | |
1224 | auto md_log = svc.mdlog->get_log(current_period.get_id()); | |
1225 | meta_notifier = new RGWMetaNotifier(this, md_log); | |
1226 | meta_notifier->start(); | |
1227 | } | |
1228 | ||
1229 | /* init it anyway, might run sync through radosgw-admin explicitly */ | |
1230 | sync_tracer = new RGWSyncTraceManager(cct, cct->_conf->rgw_sync_trace_history_size); | |
1231 | sync_tracer->init(this); | |
1232 | ret = sync_tracer->hook_to_admin_command(); | |
1233 | if (ret < 0) { | |
1234 | return ret; | |
1235 | } | |
1236 | ||
1237 | if (run_sync_thread) { | |
1238 | for (const auto &pt: zonegroup.placement_targets) { | |
1239 | if (zone_params.placement_pools.find(pt.second.name) | |
1240 | == zone_params.placement_pools.end()){ | |
1241 | ldpp_dout(dpp, 0) << "WARNING: This zone does not contain the placement target " | |
1242 | << pt.second.name << " present in zonegroup" << dendl; | |
1243 | } | |
1244 | } | |
1245 | auto async_processor = svc.rados->get_async_processor(); | |
1246 | std::lock_guard l{meta_sync_thread_lock}; | |
1247 | meta_sync_processor_thread = new RGWMetaSyncProcessorThread(this->store, async_processor); | |
1248 | ret = meta_sync_processor_thread->init(dpp); | |
1249 | if (ret < 0) { | |
1250 | ldpp_dout(dpp, 0) << "ERROR: failed to initialize meta sync thread" << dendl; | |
1251 | return ret; | |
1252 | } | |
1253 | meta_sync_processor_thread->start(); | |
1254 | ||
1255 | // configure the bucket trim manager | |
1256 | rgw::BucketTrimConfig config; | |
1257 | rgw::configure_bucket_trim(cct, config); | |
1258 | ||
1259 | bucket_trim.emplace(this->store, config); | |
1260 | ret = bucket_trim->init(); | |
1261 | if (ret < 0) { | |
1262 | ldpp_dout(dpp, 0) << "ERROR: failed to start bucket trim manager" << dendl; | |
1263 | return ret; | |
1264 | } | |
1265 | svc.datalog_rados->set_observer(&*bucket_trim); | |
1266 | ||
1267 | std::lock_guard dl{data_sync_thread_lock}; | |
1268 | for (auto source_zone : svc.zone->get_data_sync_source_zones()) { | |
1269 | ldpp_dout(dpp, 5) << "starting data sync thread for zone " << source_zone->name << dendl; | |
1270 | auto *thread = new RGWDataSyncProcessorThread(this->store, svc.rados->get_async_processor(), source_zone); | |
1271 | ret = thread->init(dpp); | |
1272 | if (ret < 0) { | |
1273 | ldpp_dout(dpp, 0) << "ERROR: failed to initialize data sync thread" << dendl; | |
1274 | return ret; | |
1275 | } | |
1276 | thread->start(); | |
1277 | data_sync_processor_threads[rgw_zone_id(source_zone->id)] = thread; | |
1278 | } | |
1279 | auto interval = cct->_conf->rgw_sync_log_trim_interval; | |
1280 | if (interval > 0) { | |
1281 | sync_log_trimmer = new RGWSyncLogTrimThread(this->store, &*bucket_trim, interval); | |
1282 | ret = sync_log_trimmer->init(dpp); | |
1283 | if (ret < 0) { | |
1284 | ldpp_dout(dpp, 0) << "ERROR: failed to initialize sync log trim thread" << dendl; | |
1285 | return ret; | |
1286 | } | |
1287 | sync_log_trimmer->start(); | |
1288 | } | |
1289 | } | |
1290 | data_notifier = new RGWDataNotifier(this); | |
1291 | data_notifier->start(); | |
1292 | ||
1293 | binfo_cache = new RGWChainedCacheImpl<bucket_info_entry>; | |
1294 | binfo_cache->init(svc.cache); | |
1295 | ||
1296 | lc = new RGWLC(); | |
1297 | lc->initialize(cct, this->store); | |
1298 | ||
1299 | if (use_lc_thread) | |
1300 | lc->start_processor(); | |
1301 | ||
1302 | quota_handler = RGWQuotaHandler::generate_handler(dpp, this->store, quota_threads); | |
1303 | ||
1304 | bucket_index_max_shards = (cct->_conf->rgw_override_bucket_index_max_shards ? cct->_conf->rgw_override_bucket_index_max_shards : | |
1305 | zone.bucket_index_max_shards); | |
1306 | if (bucket_index_max_shards > get_max_bucket_shards()) { | |
1307 | bucket_index_max_shards = get_max_bucket_shards(); | |
1308 | ldpp_dout(dpp, 1) << __func__ << " bucket index max shards is too large, reset to value: " | |
1309 | << get_max_bucket_shards() << dendl; | |
1310 | } | |
1311 | ldpp_dout(dpp, 20) << __func__ << " bucket index max shards: " << bucket_index_max_shards << dendl; | |
1312 | ||
1313 | bool need_tombstone_cache = !svc.zone->get_zone_data_notify_to_map().empty(); /* have zones syncing from us */ | |
1314 | ||
1315 | if (need_tombstone_cache) { | |
1316 | obj_tombstone_cache = new tombstone_cache_t(cct->_conf->rgw_obj_tombstone_cache_size); | |
1317 | } | |
1318 | ||
1319 | reshard_wait = std::make_shared<RGWReshardWait>(); | |
1320 | ||
1321 | reshard = new RGWReshard(this->store); | |
1322 | ||
1323 | /* only the master zone in the zonegroup reshards buckets */ | |
1324 | run_reshard_thread = run_reshard_thread && (zonegroup.master_zone == zone.id); | |
1325 | if (run_reshard_thread) { | |
1326 | reshard->start_processor(); | |
1327 | } | |
1328 | ||
1329 | index_completion_manager = new RGWIndexCompletionManager(this); | |
1330 | ret = index_completion_manager->start(dpp); | |
1331 | if (ret < 0) { | |
1332 | return ret; | |
1333 | } | |
1334 | ret = rgw::notify::init(cct, store, dpp); | |
1335 | if (ret < 0 ) { | |
1336 | ldpp_dout(dpp, 1) << "ERROR: failed to initialize notification manager" << dendl; | |
1337 | } | |
1338 | ||
1339 | return ret; | |
1340 | } | |
1341 | ||
1342 | int RGWRados::init_svc(bool raw, const DoutPrefixProvider *dpp) | |
1343 | { | |
1344 | if (raw) { | |
1345 | return svc.init_raw(cct, use_cache, null_yield, dpp); | |
1346 | } | |
1347 | ||
1348 | return svc.init(cct, use_cache, run_sync_thread, null_yield, dpp); | |
1349 | } | |
1350 | ||
1351 | int RGWRados::init_ctl(const DoutPrefixProvider *dpp) | |
1352 | { | |
1353 | return ctl.init(&svc, dpp); | |
1354 | } | |
1355 | ||
1356 | /** | |
1357 | * Initialize the RADOS instance and prepare to do other ops | |
1358 | * Returns 0 on success, -ERR# on failure. | |
1359 | */ | |
1360 | int RGWRados::initialize(const DoutPrefixProvider *dpp) | |
1361 | { | |
1362 | int ret; | |
1363 | ||
1364 | inject_notify_timeout_probability = | |
1365 | cct->_conf.get_val<double>("rgw_inject_notify_timeout_probability"); | |
1366 | max_notify_retries = cct->_conf.get_val<uint64_t>("rgw_max_notify_retries"); | |
1367 | ||
1368 | ret = init_svc(false, dpp); | |
1369 | if (ret < 0) { | |
1370 | ldpp_dout(dpp, 0) << "ERROR: failed to init services (ret=" << cpp_strerror(-ret) << ")" << dendl; | |
1371 | return ret; | |
1372 | } | |
1373 | ||
1374 | ret = init_ctl(dpp); | |
1375 | if (ret < 0) { | |
1376 | ldpp_dout(dpp, 0) << "ERROR: failed to init ctls (ret=" << cpp_strerror(-ret) << ")" << dendl; | |
1377 | return ret; | |
1378 | } | |
1379 | ||
1380 | host_id = svc.zone_utils->gen_host_id(); | |
1381 | ||
1382 | ret = init_rados(); | |
1383 | if (ret < 0) | |
1384 | return ret; | |
1385 | ||
1386 | return init_complete(dpp); | |
1387 | } | |
1388 | ||
1389 | /** | |
1390 | * Open the pool used as root for this gateway | |
1391 | * Returns: 0 on success, -ERR# otherwise. | |
1392 | */ | |
1393 | int RGWRados::open_root_pool_ctx(const DoutPrefixProvider *dpp) | |
1394 | { | |
1395 | return rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().domain_root, root_pool_ctx, true, true); | |
1396 | } | |
1397 | ||
1398 | int RGWRados::open_gc_pool_ctx(const DoutPrefixProvider *dpp) | |
1399 | { | |
1400 | return rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().gc_pool, gc_pool_ctx, true, true); | |
1401 | } | |
1402 | ||
1403 | int RGWRados::open_lc_pool_ctx(const DoutPrefixProvider *dpp) | |
1404 | { | |
1405 | return rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().lc_pool, lc_pool_ctx, true, true); | |
1406 | } | |
1407 | ||
1408 | int RGWRados::open_objexp_pool_ctx(const DoutPrefixProvider *dpp) | |
1409 | { | |
1410 | return rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().log_pool, objexp_pool_ctx, true, true); | |
1411 | } | |
1412 | ||
1413 | int RGWRados::open_reshard_pool_ctx(const DoutPrefixProvider *dpp) | |
1414 | { | |
1415 | return rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().reshard_pool, reshard_pool_ctx, true, true); | |
1416 | } | |
1417 | ||
1418 | int RGWRados::open_notif_pool_ctx(const DoutPrefixProvider *dpp) | |
1419 | { | |
1420 | return rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().notif_pool, notif_pool_ctx, true, true); | |
1421 | } | |
1422 | ||
1423 | int RGWRados::open_pool_ctx(const DoutPrefixProvider *dpp, const rgw_pool& pool, librados::IoCtx& io_ctx, | |
1424 | bool mostly_omap) | |
1425 | { | |
1426 | constexpr bool create = true; // create the pool if it doesn't exist | |
1427 | return rgw_init_ioctx(dpp, get_rados_handle(), pool, io_ctx, create, mostly_omap); | |
1428 | } | |
1429 | ||
1430 | /**** logs ****/ | |
1431 | ||
1432 | struct log_list_state { | |
1433 | string prefix; | |
1434 | librados::IoCtx io_ctx; | |
1435 | librados::NObjectIterator obit; | |
1436 | }; | |
1437 | ||
1438 | int RGWRados::log_list_init(const DoutPrefixProvider *dpp, const string& prefix, RGWAccessHandle *handle) | |
1439 | { | |
1440 | log_list_state *state = new log_list_state; | |
1441 | int r = rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().log_pool, state->io_ctx); | |
1442 | if (r < 0) { | |
1443 | delete state; | |
1444 | return r; | |
1445 | } | |
1446 | state->prefix = prefix; | |
1447 | state->obit = state->io_ctx.nobjects_begin(); | |
1448 | *handle = (RGWAccessHandle)state; | |
1449 | return 0; | |
1450 | } | |
1451 | ||
1452 | int RGWRados::log_list_next(RGWAccessHandle handle, string *name) | |
1453 | { | |
1454 | log_list_state *state = static_cast<log_list_state *>(handle); | |
1455 | while (true) { | |
1456 | if (state->obit == state->io_ctx.nobjects_end()) { | |
1457 | delete state; | |
1458 | return -ENOENT; | |
1459 | } | |
1460 | if (state->prefix.length() && | |
1461 | state->obit->get_oid().find(state->prefix) != 0) { | |
1462 | state->obit++; | |
1463 | continue; | |
1464 | } | |
1465 | *name = state->obit->get_oid(); | |
1466 | state->obit++; | |
1467 | break; | |
1468 | } | |
1469 | return 0; | |
1470 | } | |
1471 | ||
1472 | int RGWRados::log_remove(const DoutPrefixProvider *dpp, const string& name) | |
1473 | { | |
1474 | librados::IoCtx io_ctx; | |
1475 | int r = rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().log_pool, io_ctx); | |
1476 | if (r < 0) | |
1477 | return r; | |
1478 | return io_ctx.remove(name); | |
1479 | } | |
1480 | ||
1481 | struct log_show_state { | |
1482 | librados::IoCtx io_ctx; | |
1483 | bufferlist bl; | |
1484 | bufferlist::const_iterator p; | |
1485 | string name; | |
1486 | uint64_t pos; | |
1487 | bool eof; | |
1488 | log_show_state() : pos(0), eof(false) {} | |
1489 | }; | |
1490 | ||
1491 | int RGWRados::log_show_init(const DoutPrefixProvider *dpp, const string& name, RGWAccessHandle *handle) | |
1492 | { | |
1493 | log_show_state *state = new log_show_state; | |
1494 | int r = rgw_init_ioctx(dpp, get_rados_handle(), svc.zone->get_zone_params().log_pool, state->io_ctx); | |
1495 | if (r < 0) { | |
1496 | delete state; | |
1497 | return r; | |
1498 | } | |
1499 | state->name = name; | |
1500 | *handle = (RGWAccessHandle)state; | |
1501 | return 0; | |
1502 | } | |
1503 | ||
1504 | int RGWRados::log_show_next(RGWAccessHandle handle, rgw_log_entry *entry) | |
1505 | { | |
1506 | log_show_state *state = static_cast<log_show_state *>(handle); | |
1507 | off_t off = state->p.get_off(); | |
1508 | ||
1509 | ldout(cct, 10) << "log_show_next pos " << state->pos << " bl " << state->bl.length() | |
1510 | << " off " << off | |
1511 | << " eof " << (int)state->eof | |
1512 | << dendl; | |
1513 | // read some? | |
1514 | unsigned chunk = 1024*1024; | |
1515 | if ((state->bl.length() - off) < chunk/2 && !state->eof) { | |
1516 | bufferlist more; | |
1517 | int r = state->io_ctx.read(state->name, more, chunk, state->pos); | |
1518 | if (r < 0) | |
1519 | return r; | |
1520 | state->pos += r; | |
1521 | bufferlist old; | |
1522 | try { | |
1523 | old.substr_of(state->bl, off, state->bl.length() - off); | |
1524 | } catch (buffer::error& err) { | |
1525 | return -EINVAL; | |
1526 | } | |
1527 | state->bl = std::move(old); | |
1528 | state->bl.claim_append(more); | |
1529 | state->p = state->bl.cbegin(); | |
1530 | if ((unsigned)r < chunk) | |
1531 | state->eof = true; | |
1532 | ldout(cct, 10) << " read " << r << dendl; | |
1533 | } | |
1534 | ||
1535 | if (state->p.end()) | |
1536 | return 0; // end of file | |
1537 | try { | |
1538 | decode(*entry, state->p); | |
1539 | } | |
1540 | catch (const buffer::error &e) { | |
1541 | return -EINVAL; | |
1542 | } | |
1543 | return 1; | |
1544 | } | |
1545 | ||
1546 | /** | |
1547 | * usage_log_hash: get usage log key hash, based on name and index | |
1548 | * | |
1549 | * Get the usage object name. Since a user may have more than 1 | |
1550 | * object holding that info (multiple shards), we use index to | |
1551 | * specify that shard number. Once index exceeds max shards it | |
1552 | * wraps. | |
1553 | * If name is not being set, results for all users will be returned | |
1554 | * and index will wrap only after total shards number. | |
1555 | * | |
1556 | * @param cct [in] ceph context | |
1557 | * @param name [in] user name | |
1558 | * @param hash [out] hash value | |
1559 | * @param index [in] shard index number | |
1560 | */ | |
1561 | static void usage_log_hash(CephContext *cct, const string& name, string& hash, uint32_t index) | |
1562 | { | |
1563 | uint32_t val = index; | |
1564 | ||
1565 | if (!name.empty()) { | |
1566 | int max_user_shards = cct->_conf->rgw_usage_max_user_shards; | |
1567 | val %= max_user_shards; | |
1568 | val += ceph_str_hash_linux(name.c_str(), name.size()); | |
1569 | } | |
1570 | char buf[17]; | |
1571 | int max_shards = cct->_conf->rgw_usage_max_shards; | |
1572 | snprintf(buf, sizeof(buf), RGW_USAGE_OBJ_PREFIX "%u", (unsigned)(val % max_shards)); | |
1573 | hash = buf; | |
1574 | } | |
1575 | ||
1576 | int RGWRados::log_usage(const DoutPrefixProvider *dpp, map<rgw_user_bucket, RGWUsageBatch>& usage_info) | |
1577 | { | |
1578 | uint32_t index = 0; | |
1579 | ||
1580 | map<string, rgw_usage_log_info> log_objs; | |
1581 | ||
1582 | string hash; | |
1583 | string last_user; | |
1584 | ||
1585 | /* restructure usage map, zone by object hash */ | |
1586 | map<rgw_user_bucket, RGWUsageBatch>::iterator iter; | |
1587 | for (iter = usage_info.begin(); iter != usage_info.end(); ++iter) { | |
1588 | const rgw_user_bucket& ub = iter->first; | |
1589 | RGWUsageBatch& info = iter->second; | |
1590 | ||
1591 | if (ub.user.empty()) { | |
1592 | ldpp_dout(dpp, 0) << "WARNING: RGWRados::log_usage(): user name empty (bucket=" << ub.bucket << "), skipping" << dendl; | |
1593 | continue; | |
1594 | } | |
1595 | ||
1596 | if (ub.user != last_user) { | |
1597 | /* index *should* be random, but why waste extra cycles | |
1598 | in most cases max user shards is not going to exceed 1, | |
1599 | so just incrementing it */ | |
1600 | usage_log_hash(cct, ub.user, hash, index++); | |
1601 | } | |
1602 | last_user = ub.user; | |
1603 | vector<rgw_usage_log_entry>& v = log_objs[hash].entries; | |
1604 | ||
1605 | for (auto miter = info.m.begin(); miter != info.m.end(); ++miter) { | |
1606 | v.push_back(miter->second); | |
1607 | } | |
1608 | } | |
1609 | ||
1610 | map<string, rgw_usage_log_info>::iterator liter; | |
1611 | ||
1612 | for (liter = log_objs.begin(); liter != log_objs.end(); ++liter) { | |
1613 | int r = cls_obj_usage_log_add(dpp, liter->first, liter->second); | |
1614 | if (r < 0) | |
1615 | return r; | |
1616 | } | |
1617 | return 0; | |
1618 | } | |
1619 | ||
1620 | int RGWRados::read_usage(const DoutPrefixProvider *dpp, const rgw_user& user, const string& bucket_name, uint64_t start_epoch, uint64_t end_epoch, | |
1621 | uint32_t max_entries, bool *is_truncated, RGWUsageIter& usage_iter, map<rgw_user_bucket, | |
1622 | rgw_usage_log_entry>& usage) | |
1623 | { | |
1624 | uint32_t num = max_entries; | |
1625 | string hash, first_hash; | |
1626 | string user_str = user.to_str(); | |
1627 | usage_log_hash(cct, user_str, first_hash, 0); | |
1628 | ||
1629 | if (usage_iter.index) { | |
1630 | usage_log_hash(cct, user_str, hash, usage_iter.index); | |
1631 | } else { | |
1632 | hash = first_hash; | |
1633 | } | |
1634 | ||
1635 | usage.clear(); | |
1636 | ||
1637 | do { | |
1638 | map<rgw_user_bucket, rgw_usage_log_entry> ret_usage; | |
1639 | map<rgw_user_bucket, rgw_usage_log_entry>::iterator iter; | |
1640 | ||
1641 | int ret = cls_obj_usage_log_read(dpp, hash, user_str, bucket_name, start_epoch, end_epoch, num, | |
1642 | usage_iter.read_iter, ret_usage, is_truncated); | |
1643 | if (ret == -ENOENT) | |
1644 | goto next; | |
1645 | ||
1646 | if (ret < 0) | |
1647 | return ret; | |
1648 | ||
1649 | num -= ret_usage.size(); | |
1650 | ||
1651 | for (iter = ret_usage.begin(); iter != ret_usage.end(); ++iter) { | |
1652 | usage[iter->first].aggregate(iter->second); | |
1653 | } | |
1654 | ||
1655 | next: | |
1656 | if (!*is_truncated) { | |
1657 | usage_iter.read_iter.clear(); | |
1658 | usage_log_hash(cct, user_str, hash, ++usage_iter.index); | |
1659 | } | |
1660 | } while (num && !*is_truncated && hash != first_hash); | |
1661 | return 0; | |
1662 | } | |
1663 | ||
1664 | int RGWRados::trim_usage(const DoutPrefixProvider *dpp, const rgw_user& user, const string& bucket_name, uint64_t start_epoch, uint64_t end_epoch) | |
1665 | { | |
1666 | uint32_t index = 0; | |
1667 | string hash, first_hash; | |
1668 | string user_str = user.to_str(); | |
1669 | usage_log_hash(cct, user_str, first_hash, index); | |
1670 | ||
1671 | hash = first_hash; | |
1672 | do { | |
1673 | int ret = cls_obj_usage_log_trim(dpp, hash, user_str, bucket_name, start_epoch, end_epoch); | |
1674 | ||
1675 | if (ret < 0 && ret != -ENOENT) | |
1676 | return ret; | |
1677 | ||
1678 | usage_log_hash(cct, user_str, hash, ++index); | |
1679 | } while (hash != first_hash); | |
1680 | ||
1681 | return 0; | |
1682 | } | |
1683 | ||
1684 | ||
1685 | int RGWRados::clear_usage(const DoutPrefixProvider *dpp) | |
1686 | { | |
1687 | auto max_shards = cct->_conf->rgw_usage_max_shards; | |
1688 | int ret=0; | |
1689 | for (unsigned i=0; i < max_shards; i++){ | |
1690 | string oid = RGW_USAGE_OBJ_PREFIX + to_string(i); | |
1691 | ret = cls_obj_usage_log_clear(dpp, oid); | |
1692 | if (ret < 0){ | |
1693 | ldpp_dout(dpp,0) << "usage clear on oid="<< oid << "failed with ret=" << ret << dendl; | |
1694 | return ret; | |
1695 | } | |
1696 | } | |
1697 | return ret; | |
1698 | } | |
1699 | ||
1700 | int RGWRados::decode_policy(bufferlist& bl, ACLOwner *owner) | |
1701 | { | |
1702 | auto i = bl.cbegin(); | |
1703 | RGWAccessControlPolicy policy(cct); | |
1704 | try { | |
1705 | policy.decode_owner(i); | |
1706 | } catch (buffer::error& err) { | |
1707 | ldout(cct, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl; | |
1708 | return -EIO; | |
1709 | } | |
1710 | *owner = policy.get_owner(); | |
1711 | return 0; | |
1712 | } | |
1713 | ||
1714 | int rgw_policy_from_attrset(const DoutPrefixProvider *dpp, CephContext *cct, map<string, bufferlist>& attrset, RGWAccessControlPolicy *policy) | |
1715 | { | |
1716 | map<string, bufferlist>::iterator aiter = attrset.find(RGW_ATTR_ACL); | |
1717 | if (aiter == attrset.end()) | |
1718 | return -EIO; | |
1719 | ||
1720 | bufferlist& bl = aiter->second; | |
1721 | auto iter = bl.cbegin(); | |
1722 | try { | |
1723 | policy->decode(iter); | |
1724 | } catch (buffer::error& err) { | |
1725 | ldpp_dout(dpp, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl; | |
1726 | return -EIO; | |
1727 | } | |
1728 | if (cct->_conf->subsys.should_gather<ceph_subsys_rgw, 15>()) { | |
1729 | RGWAccessControlPolicy_S3 *s3policy = static_cast<RGWAccessControlPolicy_S3 *>(policy); | |
1730 | ldpp_dout(dpp, 15) << __func__ << " Read AccessControlPolicy"; | |
1731 | s3policy->to_xml(*_dout); | |
1732 | *_dout << dendl; | |
1733 | } | |
1734 | return 0; | |
1735 | } | |
1736 | ||
1737 | ||
1738 | int RGWRados::Bucket::update_bucket_id(const string& new_bucket_id, const DoutPrefixProvider *dpp) | |
1739 | { | |
1740 | rgw_bucket bucket = bucket_info.bucket; | |
1741 | bucket.update_bucket_id(new_bucket_id); | |
1742 | ||
1743 | auto obj_ctx = store->svc.sysobj->init_obj_ctx(); | |
1744 | ||
1745 | bucket_info.objv_tracker.clear(); | |
1746 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, nullptr, nullptr, null_yield, dpp); | |
1747 | if (ret < 0) { | |
1748 | return ret; | |
1749 | } | |
1750 | ||
1751 | return 0; | |
1752 | } | |
1753 | ||
1754 | ||
1755 | /** | |
1756 | * Get ordered listing of the objects in a bucket. | |
1757 | * | |
1758 | * max_p: maximum number of results to return | |
1759 | * bucket: bucket to list contents of | |
1760 | * prefix: only return results that match this prefix | |
1761 | * delim: do not include results that match this string. | |
1762 | * Any skipped results will have the matching portion of their name | |
1763 | * inserted in common_prefixes with a "true" mark. | |
1764 | * marker: if filled in, begin the listing with this object. | |
1765 | * end_marker: if filled in, end the listing with this object. | |
1766 | * result: the objects are put in here. | |
1767 | * common_prefixes: if delim is filled in, any matching prefixes are | |
1768 | * placed here. | |
1769 | * is_truncated: if number of objects in the bucket is bigger than | |
1770 | * max, then truncated. | |
1771 | */ | |
1772 | int RGWRados::Bucket::List::list_objects_ordered( | |
1773 | const DoutPrefixProvider *dpp, | |
1774 | int64_t max_p, | |
1775 | vector<rgw_bucket_dir_entry> *result, | |
1776 | map<string, bool> *common_prefixes, | |
1777 | bool *is_truncated, | |
1778 | optional_yield y) | |
1779 | { | |
1780 | RGWRados *store = target->get_store(); | |
1781 | CephContext *cct = store->ctx(); | |
1782 | int shard_id = target->get_shard_id(); | |
1783 | ||
1784 | int count = 0; | |
1785 | bool truncated = true; | |
1786 | bool cls_filtered = false; | |
1787 | const int64_t max = // protect against memory issues and negative vals | |
1788 | std::min(bucket_list_objects_absolute_max, std::max(int64_t(0), max_p)); | |
1789 | int read_ahead = std::max(cct->_conf->rgw_list_bucket_min_readahead, max); | |
1790 | ||
1791 | result->clear(); | |
1792 | ||
1793 | // use a local marker; either the marker will have a previous entry | |
1794 | // or it will be empty; either way it's OK to copy | |
1795 | rgw_obj_key marker_obj(params.marker.name, | |
1796 | params.marker.instance, | |
1797 | params.ns.empty() ? params.marker.ns : params.ns); | |
1798 | rgw_obj_index_key cur_marker; | |
1799 | marker_obj.get_index_key(&cur_marker); | |
1800 | ||
1801 | rgw_obj_key end_marker_obj(params.end_marker.name, | |
1802 | params.end_marker.instance, | |
1803 | params.ns.empty() ? params.end_marker.ns : params.ns); | |
1804 | rgw_obj_index_key cur_end_marker; | |
1805 | end_marker_obj.get_index_key(&cur_end_marker); | |
1806 | const bool cur_end_marker_valid = !params.end_marker.empty(); | |
1807 | ||
1808 | rgw_obj_key prefix_obj(params.prefix); | |
1809 | prefix_obj.set_ns(params.ns); | |
1810 | string cur_prefix = prefix_obj.get_index_key_name(); | |
1811 | string after_delim_s; /* needed in !params.delim.empty() AND later */ | |
1812 | ||
1813 | if (!params.delim.empty()) { | |
1814 | after_delim_s = cls_rgw_after_delim(params.delim); | |
1815 | /* if marker points at a common prefix, fast forward it into its | |
1816 | * upper bound string */ | |
1817 | int delim_pos = cur_marker.name.find(params.delim, cur_prefix.size()); | |
1818 | if (delim_pos >= 0) { | |
1819 | string s = cur_marker.name.substr(0, delim_pos); | |
1820 | s.append(after_delim_s); | |
1821 | cur_marker = s; | |
1822 | } | |
1823 | } | |
1824 | ||
1825 | rgw_obj_index_key prev_marker; | |
1826 | for (uint16_t attempt = 1; /* empty */; ++attempt) { | |
1827 | ldpp_dout(dpp, 20) << "RGWRados::Bucket::List::" << __func__ << | |
1828 | " starting attempt " << attempt << dendl; | |
1829 | ||
1830 | if (attempt > 1 && !(prev_marker < cur_marker)) { | |
1831 | // we've failed to make forward progress | |
1832 | ldpp_dout(dpp, 0) << "RGWRados::Bucket::List::" << __func__ << | |
1833 | ": ERROR marker failed to make forward progress; attempt=" << attempt << | |
1834 | ", prev_marker=" << prev_marker << | |
1835 | ", cur_marker=" << cur_marker << dendl; | |
1836 | break; | |
1837 | } | |
1838 | prev_marker = cur_marker; | |
1839 | ||
1840 | ent_map_t ent_map; | |
1841 | ent_map.reserve(read_ahead); | |
1842 | int r = store->cls_bucket_list_ordered(dpp, | |
1843 | target->get_bucket_info(), | |
1844 | shard_id, | |
1845 | cur_marker, | |
1846 | cur_prefix, | |
1847 | params.delim, | |
1848 | read_ahead + 1 - count, | |
1849 | params.list_versions, | |
1850 | attempt, | |
1851 | ent_map, | |
1852 | &truncated, | |
1853 | &cls_filtered, | |
1854 | &cur_marker, | |
1855 | y); | |
1856 | if (r < 0) { | |
1857 | return r; | |
1858 | } | |
1859 | ||
1860 | for (auto eiter = ent_map.begin(); eiter != ent_map.end(); ++eiter) { | |
1861 | rgw_bucket_dir_entry& entry = eiter->second; | |
1862 | rgw_obj_index_key index_key = entry.key; | |
1863 | rgw_obj_key obj(index_key); | |
1864 | ||
1865 | ldpp_dout(dpp, 20) << "RGWRados::Bucket::List::" << __func__ << | |
1866 | " considering entry " << entry.key << dendl; | |
1867 | ||
1868 | /* note that parse_raw_oid() here will not set the correct | |
1869 | * object's instance, as rgw_obj_index_key encodes that | |
1870 | * separately. We don't need to set the instance because it's | |
1871 | * not needed for the checks here and we end up using the raw | |
1872 | * entry for the return vector | |
1873 | */ | |
1874 | bool valid = rgw_obj_key::parse_raw_oid(index_key.name, &obj); | |
1875 | if (!valid) { | |
1876 | ldpp_dout(dpp, 0) << "ERROR: could not parse object name: " << | |
1877 | obj.name << dendl; | |
1878 | continue; | |
1879 | } | |
1880 | ||
1881 | bool matched_ns = (obj.ns == params.ns); | |
1882 | if (!params.list_versions && !entry.is_visible()) { | |
1883 | continue; | |
1884 | } | |
1885 | ||
1886 | if (params.enforce_ns && !matched_ns) { | |
1887 | if (!params.ns.empty()) { | |
1888 | /* we've iterated past the namespace we're searching -- done now */ | |
1889 | truncated = false; | |
1890 | goto done; | |
1891 | } | |
1892 | ||
1893 | /* we're not looking at the namespace this object is in, next! */ | |
1894 | continue; | |
1895 | } | |
1896 | ||
1897 | if (cur_end_marker_valid && cur_end_marker <= index_key) { | |
1898 | truncated = false; | |
1899 | goto done; | |
1900 | } | |
1901 | ||
1902 | if (count < max) { | |
1903 | params.marker = index_key; | |
1904 | next_marker = index_key; | |
1905 | } | |
1906 | ||
1907 | if (params.filter && | |
1908 | ! params.filter->filter(obj.name, index_key.name)) { | |
1909 | continue; | |
1910 | } | |
1911 | ||
1912 | if (params.prefix.size() && | |
1913 | 0 != obj.name.compare(0, params.prefix.size(), params.prefix)) { | |
1914 | continue; | |
1915 | } | |
1916 | ||
1917 | if (!params.delim.empty()) { | |
1918 | const int delim_pos = obj.name.find(params.delim, params.prefix.size()); | |
1919 | if (delim_pos >= 0) { | |
1920 | // run either the code where delimiter filtering is done a) | |
1921 | // in the OSD/CLS or b) here. | |
1922 | if (cls_filtered) { | |
1923 | // NOTE: this condition is for the newer versions of the | |
1924 | // OSD that does filtering on the CLS side | |
1925 | ||
1926 | // should only find one delimiter at the end if it finds any | |
1927 | // after the prefix | |
1928 | if (delim_pos != | |
1929 | int(obj.name.length() - params.delim.length())) { | |
1930 | ldpp_dout(dpp, 0) << | |
1931 | "WARNING: found delimiter in place other than the end of " | |
1932 | "the prefix; obj.name=" << obj.name << | |
1933 | ", prefix=" << params.prefix << dendl; | |
1934 | } | |
1935 | if (common_prefixes) { | |
1936 | if (count >= max) { | |
1937 | truncated = true; | |
1938 | goto done; | |
1939 | } | |
1940 | ||
1941 | (*common_prefixes)[obj.name] = true; | |
1942 | count++; | |
1943 | } | |
1944 | ||
1945 | continue; | |
1946 | } else { | |
1947 | // NOTE: this condition is for older versions of the OSD | |
1948 | // that do not filter on the CLS side, so the following code | |
1949 | // must do the filtering; once we reach version 16 of ceph, | |
1950 | // this code can be removed along with the conditional that | |
1951 | // can lead this way | |
1952 | ||
1953 | /* extract key -with trailing delimiter- for CommonPrefix */ | |
1954 | string prefix_key = | |
1955 | obj.name.substr(0, delim_pos + params.delim.length()); | |
1956 | ||
1957 | if (common_prefixes && | |
1958 | common_prefixes->find(prefix_key) == common_prefixes->end()) { | |
1959 | if (count >= max) { | |
1960 | truncated = true; | |
1961 | goto done; | |
1962 | } | |
1963 | next_marker = prefix_key; | |
1964 | (*common_prefixes)[prefix_key] = true; | |
1965 | ||
1966 | count++; | |
1967 | } | |
1968 | ||
1969 | continue; | |
1970 | } // if we're running an older OSD version | |
1971 | } // if a delimiter was found after prefix | |
1972 | } // if a delimiter was passed in | |
1973 | ||
1974 | if (count >= max) { | |
1975 | truncated = true; | |
1976 | goto done; | |
1977 | } | |
1978 | ||
1979 | ldpp_dout(dpp, 20) << "RGWRados::Bucket::List::" << __func__ << | |
1980 | " adding entry " << entry.key << " to result" << dendl; | |
1981 | ||
1982 | result->emplace_back(std::move(entry)); | |
1983 | count++; | |
1984 | } // eiter for loop | |
1985 | ||
1986 | // NOTE: the following conditional is needed by older versions of | |
1987 | // the OSD that don't do delimiter filtering on the CLS side; once | |
1988 | // we reach version 16 of ceph, the following conditional and the | |
1989 | // code within can be removed | |
1990 | if (!cls_filtered && !params.delim.empty()) { | |
1991 | int marker_delim_pos = | |
1992 | cur_marker.name.find(params.delim, cur_prefix.size()); | |
1993 | if (marker_delim_pos >= 0) { | |
1994 | std::string skip_after_delim = | |
1995 | cur_marker.name.substr(0, marker_delim_pos); | |
1996 | skip_after_delim.append(after_delim_s); | |
1997 | ||
1998 | ldpp_dout(dpp, 20) << "skip_after_delim=" << skip_after_delim << dendl; | |
1999 | ||
2000 | if (skip_after_delim > cur_marker.name) { | |
2001 | cur_marker = skip_after_delim; | |
2002 | ldpp_dout(dpp, 20) << "setting cur_marker=" | |
2003 | << cur_marker.name | |
2004 | << "[" << cur_marker.instance << "]" | |
2005 | << dendl; | |
2006 | } | |
2007 | } | |
2008 | } // if older osd didn't do delimiter filtering | |
2009 | ||
2010 | ldpp_dout(dpp, 20) << "RGWRados::Bucket::List::" << __func__ << | |
2011 | " INFO end of outer loop, truncated=" << truncated << | |
2012 | ", count=" << count << ", attempt=" << attempt << dendl; | |
2013 | ||
2014 | if (!truncated || count >= (max + 1) / 2) { | |
2015 | // if we finished listing, or if we're returning at least half the | |
2016 | // requested entries, that's enough; S3 and swift protocols allow | |
2017 | // returning fewer than max entries | |
2018 | break; | |
2019 | } else if (attempt > 8 && count >= 1) { | |
2020 | // if we've made at least 8 attempts and we have some, but very | |
2021 | // few, results, return with what we have | |
2022 | break; | |
2023 | } | |
2024 | } // for (uint16_t attempt... | |
2025 | ||
2026 | done: | |
2027 | ||
2028 | if (is_truncated) { | |
2029 | *is_truncated = truncated; | |
2030 | } | |
2031 | ||
2032 | return 0; | |
2033 | } // list_objects_ordered | |
2034 | ||
2035 | ||
2036 | /** | |
2037 | * Get listing of the objects in a bucket and allow the results to be out | |
2038 | * of order. | |
2039 | * | |
2040 | * Even though there are key differences with the ordered counterpart, | |
2041 | * the parameters are the same to maintain some compatability. | |
2042 | * | |
2043 | * max: maximum number of results to return | |
2044 | * bucket: bucket to list contents of | |
2045 | * prefix: only return results that match this prefix | |
2046 | * delim: should not be set; if it is we should have indicated an error | |
2047 | * marker: if filled in, begin the listing with this object. | |
2048 | * end_marker: if filled in, end the listing with this object. | |
2049 | * result: the objects are put in here. | |
2050 | * common_prefixes: this is never filled with an unordered list; the param | |
2051 | * is maintained for compatibility | |
2052 | * is_truncated: if number of objects in the bucket is bigger than max, then | |
2053 | * truncated. | |
2054 | */ | |
2055 | int RGWRados::Bucket::List::list_objects_unordered(const DoutPrefixProvider *dpp, | |
2056 | int64_t max_p, | |
2057 | vector<rgw_bucket_dir_entry> *result, | |
2058 | map<string, bool> *common_prefixes, | |
2059 | bool *is_truncated, | |
2060 | optional_yield y) | |
2061 | { | |
2062 | RGWRados *store = target->get_store(); | |
2063 | int shard_id = target->get_shard_id(); | |
2064 | ||
2065 | int count = 0; | |
2066 | bool truncated = true; | |
2067 | ||
2068 | const int64_t max = // protect against memory issues and negative vals | |
2069 | std::min(bucket_list_objects_absolute_max, std::max(int64_t(0), max_p)); | |
2070 | ||
2071 | // read a few extra in each call to cls_bucket_list_unordered in | |
2072 | // case some are filtered out due to namespace matching, versioning, | |
2073 | // filtering, etc. | |
2074 | const int64_t max_read_ahead = 100; | |
2075 | const uint32_t read_ahead = uint32_t(max + std::min(max, max_read_ahead)); | |
2076 | ||
2077 | result->clear(); | |
2078 | ||
2079 | // use a local marker; either the marker will have a previous entry | |
2080 | // or it will be empty; either way it's OK to copy | |
2081 | rgw_obj_key marker_obj(params.marker.name, | |
2082 | params.marker.instance, | |
2083 | params.ns.empty() ? params.marker.ns : params.ns); | |
2084 | rgw_obj_index_key cur_marker; | |
2085 | marker_obj.get_index_key(&cur_marker); | |
2086 | ||
2087 | rgw_obj_key end_marker_obj(params.end_marker.name, | |
2088 | params.end_marker.instance, | |
2089 | params.ns.empty() ? params.end_marker.ns : params.ns); | |
2090 | rgw_obj_index_key cur_end_marker; | |
2091 | end_marker_obj.get_index_key(&cur_end_marker); | |
2092 | const bool cur_end_marker_valid = !params.end_marker.empty(); | |
2093 | ||
2094 | rgw_obj_key prefix_obj(params.prefix); | |
2095 | prefix_obj.set_ns(params.ns); | |
2096 | string cur_prefix = prefix_obj.get_index_key_name(); | |
2097 | ||
2098 | while (truncated && count <= max) { | |
2099 | std::vector<rgw_bucket_dir_entry> ent_list; | |
2100 | ent_list.reserve(read_ahead); | |
2101 | ||
2102 | int r = store->cls_bucket_list_unordered(dpp, | |
2103 | target->get_bucket_info(), | |
2104 | shard_id, | |
2105 | cur_marker, | |
2106 | cur_prefix, | |
2107 | read_ahead, | |
2108 | params.list_versions, | |
2109 | ent_list, | |
2110 | &truncated, | |
2111 | &cur_marker, | |
2112 | y); | |
2113 | if (r < 0) | |
2114 | return r; | |
2115 | ||
2116 | // NB: while regions of ent_list will be sorted, we have no | |
2117 | // guarantee that all items will be sorted since they can cross | |
2118 | // shard boundaries | |
2119 | ||
2120 | for (auto& entry : ent_list) { | |
2121 | rgw_obj_index_key index_key = entry.key; | |
2122 | rgw_obj_key obj(index_key); | |
2123 | ||
2124 | if (count < max) { | |
2125 | params.marker.set(index_key); | |
2126 | next_marker.set(index_key); | |
2127 | } | |
2128 | ||
2129 | /* note that parse_raw_oid() here will not set the correct | |
2130 | * object's instance, as rgw_obj_index_key encodes that | |
2131 | * separately. We don't need to set the instance because it's | |
2132 | * not needed for the checks here and we end up using the raw | |
2133 | * entry for the return vector | |
2134 | */ | |
2135 | bool valid = rgw_obj_key::parse_raw_oid(index_key.name, &obj); | |
2136 | if (!valid) { | |
2137 | ldpp_dout(dpp, 0) << "ERROR: could not parse object name: " << | |
2138 | obj.name << dendl; | |
2139 | continue; | |
2140 | } | |
2141 | ||
2142 | if (!params.list_versions && !entry.is_visible()) { | |
2143 | continue; | |
2144 | } | |
2145 | ||
2146 | if (params.enforce_ns && obj.ns != params.ns) { | |
2147 | continue; | |
2148 | } | |
2149 | ||
2150 | if (cur_end_marker_valid && cur_end_marker <= index_key) { | |
2151 | // we're not guaranteed items will come in order, so we have | |
2152 | // to loop through all | |
2153 | continue; | |
2154 | } | |
2155 | ||
2156 | if (params.filter && !params.filter->filter(obj.name, index_key.name)) | |
2157 | continue; | |
2158 | ||
2159 | if (params.prefix.size() && | |
2160 | (0 != obj.name.compare(0, params.prefix.size(), params.prefix))) | |
2161 | continue; | |
2162 | ||
2163 | if (count >= max) { | |
2164 | truncated = true; | |
2165 | goto done; | |
2166 | } | |
2167 | ||
2168 | result->emplace_back(std::move(entry)); | |
2169 | count++; | |
2170 | } // for (auto& entry : ent_list) | |
2171 | } // while (truncated && count <= max) | |
2172 | ||
2173 | done: | |
2174 | if (is_truncated) | |
2175 | *is_truncated = truncated; | |
2176 | ||
2177 | return 0; | |
2178 | } // list_objects_unordered | |
2179 | ||
2180 | ||
2181 | /** | |
2182 | * create a rados pool, associated meta info | |
2183 | * returns 0 on success, -ERR# otherwise. | |
2184 | */ | |
2185 | int RGWRados::create_pool(const DoutPrefixProvider *dpp, const rgw_pool& pool) | |
2186 | { | |
2187 | librados::IoCtx io_ctx; | |
2188 | constexpr bool create = true; | |
2189 | return rgw_init_ioctx(dpp, get_rados_handle(), pool, io_ctx, create); | |
2190 | } | |
2191 | ||
2192 | void RGWRados::create_bucket_id(string *bucket_id) | |
2193 | { | |
2194 | uint64_t iid = instance_id(); | |
2195 | uint64_t bid = next_bucket_id(); | |
2196 | char buf[svc.zone->get_zone_params().get_id().size() + 48]; | |
2197 | snprintf(buf, sizeof(buf), "%s.%" PRIu64 ".%" PRIu64, | |
2198 | svc.zone->get_zone_params().get_id().c_str(), iid, bid); | |
2199 | *bucket_id = buf; | |
2200 | } | |
2201 | ||
2202 | int RGWRados::create_bucket(const RGWUserInfo& owner, rgw_bucket& bucket, | |
2203 | const string& zonegroup_id, | |
2204 | const rgw_placement_rule& placement_rule, | |
2205 | const string& swift_ver_location, | |
2206 | const RGWQuotaInfo * pquota_info, | |
2207 | map<std::string, bufferlist>& attrs, | |
2208 | RGWBucketInfo& info, | |
2209 | obj_version *pobjv, | |
2210 | obj_version *pep_objv, | |
2211 | real_time creation_time, | |
2212 | rgw_bucket *pmaster_bucket, | |
2213 | uint32_t *pmaster_num_shards, | |
2214 | optional_yield y, | |
2215 | const DoutPrefixProvider *dpp, | |
2216 | bool exclusive) | |
2217 | { | |
2218 | #define MAX_CREATE_RETRIES 20 /* need to bound retries */ | |
2219 | rgw_placement_rule selected_placement_rule; | |
2220 | RGWZonePlacementInfo rule_info; | |
2221 | ||
2222 | for (int i = 0; i < MAX_CREATE_RETRIES; i++) { | |
2223 | int ret = 0; | |
2224 | ret = svc.zone->select_bucket_placement(dpp, owner, zonegroup_id, placement_rule, | |
2225 | &selected_placement_rule, &rule_info, y); | |
2226 | if (ret < 0) | |
2227 | return ret; | |
2228 | ||
2229 | if (!pmaster_bucket) { | |
2230 | create_bucket_id(&bucket.marker); | |
2231 | bucket.bucket_id = bucket.marker; | |
2232 | } else { | |
2233 | bucket.marker = pmaster_bucket->marker; | |
2234 | bucket.bucket_id = pmaster_bucket->bucket_id; | |
2235 | } | |
2236 | ||
2237 | RGWObjVersionTracker& objv_tracker = info.objv_tracker; | |
2238 | ||
2239 | objv_tracker.read_version.clear(); | |
2240 | ||
2241 | if (pobjv) { | |
2242 | objv_tracker.write_version = *pobjv; | |
2243 | } else { | |
2244 | objv_tracker.generate_new_write_ver(cct); | |
2245 | } | |
2246 | ||
2247 | info.bucket = bucket; | |
2248 | info.owner = owner.user_id; | |
2249 | info.zonegroup = zonegroup_id; | |
2250 | info.placement_rule = selected_placement_rule; | |
2251 | info.swift_ver_location = swift_ver_location; | |
2252 | info.swift_versioning = (!swift_ver_location.empty()); | |
2253 | ||
2254 | init_default_bucket_layout(cct, info.layout, svc.zone->get_zone(), | |
2255 | pmaster_num_shards ? | |
2256 | std::optional{*pmaster_num_shards} : | |
2257 | std::nullopt, | |
2258 | rule_info.index_type); | |
2259 | ||
2260 | info.requester_pays = false; | |
2261 | if (real_clock::is_zero(creation_time)) { | |
2262 | info.creation_time = ceph::real_clock::now(); | |
2263 | } else { | |
2264 | info.creation_time = creation_time; | |
2265 | } | |
2266 | if (pquota_info) { | |
2267 | info.quota = *pquota_info; | |
2268 | } | |
2269 | ||
2270 | int r = svc.bi->init_index(dpp, info); | |
2271 | if (r < 0) { | |
2272 | return r; | |
2273 | } | |
2274 | ||
2275 | ret = put_linked_bucket_info(info, exclusive, ceph::real_time(), pep_objv, &attrs, true, dpp); | |
2276 | if (ret == -ECANCELED) { | |
2277 | ret = -EEXIST; | |
2278 | } | |
2279 | if (ret == -EEXIST) { | |
2280 | /* we need to reread the info and return it, caller will have a use for it */ | |
2281 | RGWBucketInfo orig_info; | |
2282 | r = get_bucket_info(&svc, bucket.tenant, bucket.name, orig_info, NULL, null_yield, NULL); | |
2283 | if (r < 0) { | |
2284 | if (r == -ENOENT) { | |
2285 | continue; | |
2286 | } | |
2287 | ldpp_dout(dpp, 0) << "get_bucket_info returned " << r << dendl; | |
2288 | return r; | |
2289 | } | |
2290 | ||
2291 | /* only remove it if it's a different bucket instance */ | |
2292 | if (orig_info.bucket.bucket_id != bucket.bucket_id) { | |
2293 | int r = svc.bi->clean_index(dpp, info); | |
2294 | if (r < 0) { | |
2295 | ldpp_dout(dpp, 0) << "WARNING: could not remove bucket index (r=" << r << ")" << dendl; | |
2296 | } | |
2297 | r = ctl.bucket->remove_bucket_instance_info(info.bucket, info, null_yield, dpp); | |
2298 | if (r < 0) { | |
2299 | ldpp_dout(dpp, 0) << "WARNING: " << __func__ << "(): failed to remove bucket instance info: bucket instance=" << info.bucket.get_key() << ": r=" << r << dendl; | |
2300 | /* continue anyway */ | |
2301 | } | |
2302 | } | |
2303 | ||
2304 | info = std::move(orig_info); | |
2305 | /* ret == -EEXIST here */ | |
2306 | } | |
2307 | return ret; | |
2308 | } | |
2309 | ||
2310 | /* this is highly unlikely */ | |
2311 | ldpp_dout(dpp, 0) << "ERROR: could not create bucket, continuously raced with bucket creation and removal" << dendl; | |
2312 | return -ENOENT; | |
2313 | } | |
2314 | ||
2315 | bool RGWRados::get_obj_data_pool(const rgw_placement_rule& placement_rule, const rgw_obj& obj, rgw_pool *pool) | |
2316 | { | |
2317 | return rgw_get_obj_data_pool(svc.zone->get_zonegroup(), svc.zone->get_zone_params(), placement_rule, obj, pool); | |
2318 | } | |
2319 | ||
2320 | bool RGWRados::obj_to_raw(const rgw_placement_rule& placement_rule, const rgw_obj& obj, rgw_raw_obj *raw_obj) | |
2321 | { | |
2322 | get_obj_bucket_and_oid_loc(obj, raw_obj->oid, raw_obj->loc); | |
2323 | ||
2324 | return get_obj_data_pool(placement_rule, obj, &raw_obj->pool); | |
2325 | } | |
2326 | ||
2327 | int RGWRados::get_obj_head_ioctx(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj, librados::IoCtx *ioctx) | |
2328 | { | |
2329 | string oid, key; | |
2330 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
2331 | ||
2332 | rgw_pool pool; | |
2333 | if (!get_obj_data_pool(bucket_info.placement_rule, obj, &pool)) { | |
2334 | ldpp_dout(dpp, 0) << "ERROR: cannot get data pool for obj=" << obj << ", probably misconfiguration" << dendl; | |
2335 | return -EIO; | |
2336 | } | |
2337 | ||
2338 | int r = open_pool_ctx(dpp, pool, *ioctx, false); | |
2339 | if (r < 0) { | |
2340 | return r; | |
2341 | } | |
2342 | ||
2343 | ioctx->locator_set_key(key); | |
2344 | ||
2345 | return 0; | |
2346 | } | |
2347 | ||
2348 | int RGWRados::get_obj_head_ref(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj, rgw_rados_ref *ref) | |
2349 | { | |
2350 | get_obj_bucket_and_oid_loc(obj, ref->obj.oid, ref->obj.loc); | |
2351 | ||
2352 | rgw_pool pool; | |
2353 | if (!get_obj_data_pool(bucket_info.placement_rule, obj, &pool)) { | |
2354 | ldpp_dout(dpp, 0) << "ERROR: cannot get data pool for obj=" << obj << ", probably misconfiguration" << dendl; | |
2355 | return -EIO; | |
2356 | } | |
2357 | ||
2358 | ref->pool = svc.rados->pool(pool); | |
2359 | ||
2360 | int r = ref->pool.open(dpp, RGWSI_RADOS::OpenParams() | |
2361 | .set_mostly_omap(false)); | |
2362 | if (r < 0) { | |
2363 | ldpp_dout(dpp, 0) << "ERROR: failed opening data pool (pool=" << pool << "); r=" << r << dendl; | |
2364 | return r; | |
2365 | } | |
2366 | ||
2367 | ref->pool.ioctx().locator_set_key(ref->obj.loc); | |
2368 | ||
2369 | return 0; | |
2370 | } | |
2371 | ||
2372 | int RGWRados::get_raw_obj_ref(const DoutPrefixProvider *dpp, const rgw_raw_obj& obj, rgw_rados_ref *ref) | |
2373 | { | |
2374 | ref->obj = obj; | |
2375 | ||
2376 | if (ref->obj.oid.empty()) { | |
2377 | ref->obj.oid = obj.pool.to_str(); | |
2378 | ref->obj.pool = svc.zone->get_zone_params().domain_root; | |
2379 | } | |
2380 | ref->pool = svc.rados->pool(obj.pool); | |
2381 | int r = ref->pool.open(dpp, RGWSI_RADOS::OpenParams() | |
2382 | .set_mostly_omap(false)); | |
2383 | if (r < 0) { | |
2384 | ldpp_dout(dpp, 0) << "ERROR: failed opening pool (pool=" << obj.pool << "); r=" << r << dendl; | |
2385 | return r; | |
2386 | } | |
2387 | ||
2388 | ref->pool.ioctx().locator_set_key(ref->obj.loc); | |
2389 | ||
2390 | return 0; | |
2391 | } | |
2392 | ||
2393 | int RGWRados::get_system_obj_ref(const DoutPrefixProvider *dpp, const rgw_raw_obj& obj, rgw_rados_ref *ref) | |
2394 | { | |
2395 | return get_raw_obj_ref(dpp, obj, ref); | |
2396 | } | |
2397 | ||
2398 | /* | |
2399 | * fixes an issue where head objects were supposed to have a locator created, but ended | |
2400 | * up without one | |
2401 | */ | |
2402 | int RGWRados::fix_head_obj_locator(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, bool copy_obj, bool remove_bad, rgw_obj_key& key) | |
2403 | { | |
2404 | const rgw_bucket& bucket = bucket_info.bucket; | |
2405 | string oid; | |
2406 | string locator; | |
2407 | ||
2408 | rgw_obj obj(bucket, key); | |
2409 | ||
2410 | get_obj_bucket_and_oid_loc(obj, oid, locator); | |
2411 | ||
2412 | if (locator.empty()) { | |
2413 | ldpp_dout(dpp, 20) << "object does not have a locator, nothing to fix" << dendl; | |
2414 | return 0; | |
2415 | } | |
2416 | ||
2417 | librados::IoCtx ioctx; | |
2418 | ||
2419 | int ret = get_obj_head_ioctx(dpp, bucket_info, obj, &ioctx); | |
2420 | if (ret < 0) { | |
2421 | cerr << "ERROR: get_obj_head_ioctx() returned ret=" << ret << std::endl; | |
2422 | return ret; | |
2423 | } | |
2424 | ioctx.locator_set_key(string()); /* override locator for this object, use empty locator */ | |
2425 | ||
2426 | uint64_t size; | |
2427 | bufferlist data; | |
2428 | ||
2429 | struct timespec mtime_ts; | |
2430 | map<string, bufferlist> attrs; | |
2431 | librados::ObjectReadOperation op; | |
2432 | op.getxattrs(&attrs, NULL); | |
2433 | op.stat2(&size, &mtime_ts, NULL); | |
2434 | #define HEAD_SIZE 512 * 1024 | |
2435 | op.read(0, HEAD_SIZE, &data, NULL); | |
2436 | ||
2437 | ret = rgw_rados_operate(dpp, ioctx, oid, &op, &data, null_yield); | |
2438 | if (ret < 0) { | |
2439 | ldpp_dout(dpp, -1) << "ERROR: rgw_rados_operate(oid=" << oid << ") returned ret=" << ret << dendl; | |
2440 | return ret; | |
2441 | } | |
2442 | ||
2443 | if (size > HEAD_SIZE) { | |
2444 | ldpp_dout(dpp, -1) << "ERROR: returned object size (" << size << ") > HEAD_SIZE (" << HEAD_SIZE << ")" << dendl; | |
2445 | return -EIO; | |
2446 | } | |
2447 | ||
2448 | if (size != data.length()) { | |
2449 | ldpp_dout(dpp, -1) << "ERROR: returned object size (" << size << ") != data.length() (" << data.length() << ")" << dendl; | |
2450 | return -EIO; | |
2451 | } | |
2452 | ||
2453 | if (copy_obj) { | |
2454 | librados::ObjectWriteOperation wop; | |
2455 | ||
2456 | wop.mtime2(&mtime_ts); | |
2457 | ||
2458 | map<string, bufferlist>::iterator iter; | |
2459 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
2460 | wop.setxattr(iter->first.c_str(), iter->second); | |
2461 | } | |
2462 | ||
2463 | wop.write(0, data); | |
2464 | ||
2465 | ioctx.locator_set_key(locator); | |
2466 | rgw_rados_operate(dpp, ioctx, oid, &wop, null_yield); | |
2467 | } | |
2468 | ||
2469 | if (remove_bad) { | |
2470 | ioctx.locator_set_key(string()); | |
2471 | ||
2472 | ret = ioctx.remove(oid); | |
2473 | if (ret < 0) { | |
2474 | ldpp_dout(dpp, -1) << "ERROR: failed to remove original bad object" << dendl; | |
2475 | return ret; | |
2476 | } | |
2477 | } | |
2478 | ||
2479 | return 0; | |
2480 | } | |
2481 | ||
2482 | int RGWRados::move_rados_obj(const DoutPrefixProvider *dpp, | |
2483 | librados::IoCtx& src_ioctx, | |
2484 | const string& src_oid, const string& src_locator, | |
2485 | librados::IoCtx& dst_ioctx, | |
2486 | const string& dst_oid, const string& dst_locator) | |
2487 | { | |
2488 | ||
2489 | #define COPY_BUF_SIZE (4 * 1024 * 1024) | |
2490 | bool done = false; | |
2491 | uint64_t chunk_size = COPY_BUF_SIZE; | |
2492 | uint64_t ofs = 0; | |
2493 | int ret = 0; | |
2494 | real_time mtime; | |
2495 | struct timespec mtime_ts; | |
2496 | uint64_t size; | |
2497 | ||
2498 | if (src_oid == dst_oid && src_locator == dst_locator) { | |
2499 | return 0; | |
2500 | } | |
2501 | ||
2502 | src_ioctx.locator_set_key(src_locator); | |
2503 | dst_ioctx.locator_set_key(dst_locator); | |
2504 | ||
2505 | do { | |
2506 | bufferlist data; | |
2507 | ObjectReadOperation rop; | |
2508 | ObjectWriteOperation wop; | |
2509 | ||
2510 | if (ofs == 0) { | |
2511 | rop.stat2(&size, &mtime_ts, NULL); | |
2512 | mtime = real_clock::from_timespec(mtime_ts); | |
2513 | } | |
2514 | rop.read(ofs, chunk_size, &data, NULL); | |
2515 | ret = rgw_rados_operate(dpp, src_ioctx, src_oid, &rop, &data, null_yield); | |
2516 | if (ret < 0) { | |
2517 | goto done_err; | |
2518 | } | |
2519 | ||
2520 | if (data.length() == 0) { | |
2521 | break; | |
2522 | } | |
2523 | ||
2524 | if (ofs == 0) { | |
2525 | wop.create(true); /* make it exclusive */ | |
2526 | wop.mtime2(&mtime_ts); | |
2527 | mtime = real_clock::from_timespec(mtime_ts); | |
2528 | } | |
2529 | wop.write(ofs, data); | |
2530 | ret = rgw_rados_operate(dpp, dst_ioctx, dst_oid, &wop, null_yield); | |
2531 | if (ret < 0) { | |
2532 | goto done_err; | |
2533 | } | |
2534 | ofs += data.length(); | |
2535 | done = data.length() != chunk_size; | |
2536 | } while (!done); | |
2537 | ||
2538 | if (ofs != size) { | |
2539 | ldpp_dout(dpp, -1) << "ERROR: " << __func__ << ": copying " << src_oid << " -> " << dst_oid | |
2540 | << ": expected " << size << " bytes to copy, ended up with " << ofs << dendl; | |
2541 | ret = -EIO; | |
2542 | goto done_err; | |
2543 | } | |
2544 | ||
2545 | src_ioctx.remove(src_oid); | |
2546 | ||
2547 | return 0; | |
2548 | ||
2549 | done_err: | |
2550 | // TODO: clean up dst_oid if we created it | |
2551 | ldpp_dout(dpp, -1) << "ERROR: failed to copy " << src_oid << " -> " << dst_oid << dendl; | |
2552 | return ret; | |
2553 | } | |
2554 | ||
2555 | /* | |
2556 | * fixes an issue where head objects were supposed to have a locator created, but ended | |
2557 | * up without one | |
2558 | */ | |
2559 | int RGWRados::fix_tail_obj_locator(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, rgw_obj_key& key, bool fix, bool *need_fix, optional_yield y) | |
2560 | { | |
2561 | const rgw_bucket& bucket = bucket_info.bucket; | |
2562 | rgw_obj obj(bucket, key); | |
2563 | ||
2564 | if (need_fix) { | |
2565 | *need_fix = false; | |
2566 | } | |
2567 | ||
2568 | rgw_rados_ref ref; | |
2569 | int r = get_obj_head_ref(dpp, bucket_info, obj, &ref); | |
2570 | if (r < 0) { | |
2571 | return r; | |
2572 | } | |
2573 | ||
2574 | RGWObjState *astate = NULL; | |
2575 | RGWObjectCtx rctx(this->store); | |
2576 | r = get_obj_state(dpp, &rctx, bucket_info, obj, &astate, false, y); | |
2577 | if (r < 0) | |
2578 | return r; | |
2579 | ||
2580 | if (astate->manifest) { | |
2581 | RGWObjManifest::obj_iterator miter; | |
2582 | RGWObjManifest& manifest = *astate->manifest; | |
2583 | for (miter = manifest.obj_begin(dpp); miter != manifest.obj_end(dpp); ++miter) { | |
2584 | rgw_raw_obj raw_loc = miter.get_location().get_raw_obj(store); | |
2585 | rgw_obj loc; | |
2586 | string oid; | |
2587 | string locator; | |
2588 | ||
2589 | RGWSI_Tier_RADOS::raw_obj_to_obj(manifest.get_tail_placement().bucket, raw_loc, &loc); | |
2590 | ||
2591 | if (loc.key.ns.empty()) { | |
2592 | /* continue, we're only interested in tail objects */ | |
2593 | continue; | |
2594 | } | |
2595 | ||
2596 | auto& ioctx = ref.pool.ioctx(); | |
2597 | ||
2598 | get_obj_bucket_and_oid_loc(loc, oid, locator); | |
2599 | ref.pool.ioctx().locator_set_key(locator); | |
2600 | ||
2601 | ldpp_dout(dpp, 20) << __func__ << ": key=" << key << " oid=" << oid << " locator=" << locator << dendl; | |
2602 | ||
2603 | r = ioctx.stat(oid, NULL, NULL); | |
2604 | if (r != -ENOENT) { | |
2605 | continue; | |
2606 | } | |
2607 | ||
2608 | string bad_loc; | |
2609 | prepend_bucket_marker(bucket, loc.key.name, bad_loc); | |
2610 | ||
2611 | /* create a new ioctx with the bad locator */ | |
2612 | librados::IoCtx src_ioctx; | |
2613 | src_ioctx.dup(ioctx); | |
2614 | src_ioctx.locator_set_key(bad_loc); | |
2615 | ||
2616 | r = src_ioctx.stat(oid, NULL, NULL); | |
2617 | if (r != 0) { | |
2618 | /* cannot find a broken part */ | |
2619 | continue; | |
2620 | } | |
2621 | ldpp_dout(dpp, 20) << __func__ << ": found bad object part: " << loc << dendl; | |
2622 | if (need_fix) { | |
2623 | *need_fix = true; | |
2624 | } | |
2625 | if (fix) { | |
2626 | r = move_rados_obj(dpp, src_ioctx, oid, bad_loc, ioctx, oid, locator); | |
2627 | if (r < 0) { | |
2628 | ldpp_dout(dpp, -1) << "ERROR: copy_rados_obj() on oid=" << oid << " returned r=" << r << dendl; | |
2629 | } | |
2630 | } | |
2631 | } | |
2632 | } | |
2633 | ||
2634 | return 0; | |
2635 | } | |
2636 | ||
2637 | int RGWRados::BucketShard::init(const rgw_bucket& _bucket, | |
2638 | const rgw_obj& obj, | |
2639 | RGWBucketInfo* bucket_info_out, | |
2640 | const DoutPrefixProvider *dpp) | |
2641 | { | |
2642 | bucket = _bucket; | |
2643 | ||
2644 | auto obj_ctx = store->svc.sysobj->init_obj_ctx(); | |
2645 | ||
2646 | RGWBucketInfo bucket_info; | |
2647 | RGWBucketInfo* bucket_info_p = | |
2648 | bucket_info_out ? bucket_info_out : &bucket_info; | |
2649 | ||
2650 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, *bucket_info_p, NULL, NULL, null_yield, dpp); | |
2651 | if (ret < 0) { | |
2652 | return ret; | |
2653 | } | |
2654 | ||
2655 | string oid; | |
2656 | ||
2657 | ret = store->svc.bi_rados->open_bucket_index_shard(dpp, *bucket_info_p, obj.get_hash_object(), &bucket_obj, &shard_id); | |
2658 | if (ret < 0) { | |
2659 | ldpp_dout(dpp, 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
2660 | return ret; | |
2661 | } | |
2662 | ldpp_dout(dpp, 20) << " bucket index object: " << bucket_obj.get_raw_obj() << dendl; | |
2663 | ||
2664 | return 0; | |
2665 | } | |
2666 | ||
2667 | int RGWRados::BucketShard::init(const rgw_bucket& _bucket, | |
2668 | int sid, const rgw::bucket_index_layout_generation& idx_layout, | |
2669 | RGWBucketInfo* bucket_info_out, | |
2670 | const DoutPrefixProvider *dpp) | |
2671 | { | |
2672 | bucket = _bucket; | |
2673 | shard_id = sid; | |
2674 | ||
2675 | auto obj_ctx = store->svc.sysobj->init_obj_ctx(); | |
2676 | ||
2677 | ||
2678 | RGWBucketInfo bucket_info; | |
2679 | RGWBucketInfo* bucket_info_p = | |
2680 | bucket_info_out ? bucket_info_out : &bucket_info; | |
2681 | int ret = store->get_bucket_instance_info(obj_ctx, bucket, *bucket_info_p, NULL, NULL, null_yield, dpp); | |
2682 | if (ret < 0) { | |
2683 | return ret; | |
2684 | } | |
2685 | ||
2686 | string oid; | |
2687 | ||
2688 | ret = store->svc.bi_rados->open_bucket_index_shard(dpp, *bucket_info_p, shard_id, idx_layout, &bucket_obj); | |
2689 | if (ret < 0) { | |
2690 | ldpp_dout(dpp, 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
2691 | return ret; | |
2692 | } | |
2693 | ldpp_dout(dpp, 20) << " bucket index oid: " << bucket_obj.get_raw_obj() << dendl; | |
2694 | ||
2695 | return 0; | |
2696 | } | |
2697 | ||
2698 | int RGWRados::BucketShard::init(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, | |
2699 | const rgw_obj& obj) | |
2700 | { | |
2701 | bucket = bucket_info.bucket; | |
2702 | ||
2703 | int ret = store->svc.bi_rados->open_bucket_index_shard(dpp, bucket_info, | |
2704 | obj.get_hash_object(), | |
2705 | &bucket_obj, | |
2706 | &shard_id); | |
2707 | if (ret < 0) { | |
2708 | ldpp_dout(dpp, 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
2709 | return ret; | |
2710 | } | |
2711 | ldpp_dout(dpp, 20) << " bucket index object: " << bucket_obj << dendl; | |
2712 | ||
2713 | return 0; | |
2714 | } | |
2715 | ||
2716 | int RGWRados::BucketShard::init(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw::bucket_index_layout_generation& idx_layout, int sid) | |
2717 | { | |
2718 | bucket = bucket_info.bucket; | |
2719 | shard_id = sid; | |
2720 | ||
2721 | int ret = store->svc.bi_rados->open_bucket_index_shard(dpp, bucket_info, shard_id, idx_layout, &bucket_obj); | |
2722 | if (ret < 0) { | |
2723 | ldpp_dout(dpp, 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret << dendl; | |
2724 | return ret; | |
2725 | } | |
2726 | ldpp_dout(dpp, 20) << " bucket index object: " << bucket_obj << dendl; | |
2727 | ||
2728 | return 0; | |
2729 | } | |
2730 | ||
2731 | ||
2732 | /* Execute @handler on last item in bucket listing for bucket specified | |
2733 | * in @bucket_info. @obj_prefix and @obj_delim narrow down the listing | |
2734 | * to objects matching these criterias. */ | |
2735 | int RGWRados::on_last_entry_in_listing(const DoutPrefixProvider *dpp, | |
2736 | RGWBucketInfo& bucket_info, | |
2737 | const std::string& obj_prefix, | |
2738 | const std::string& obj_delim, | |
2739 | std::function<int(const rgw_bucket_dir_entry&)> handler) | |
2740 | { | |
2741 | RGWRados::Bucket target(this, bucket_info); | |
2742 | RGWRados::Bucket::List list_op(&target); | |
2743 | ||
2744 | list_op.params.prefix = obj_prefix; | |
2745 | list_op.params.delim = obj_delim; | |
2746 | ||
2747 | ldpp_dout(dpp, 20) << "iterating listing for bucket=" << bucket_info.bucket.name | |
2748 | << ", obj_prefix=" << obj_prefix | |
2749 | << ", obj_delim=" << obj_delim | |
2750 | << dendl; | |
2751 | ||
2752 | bool is_truncated = false; | |
2753 | ||
2754 | boost::optional<rgw_bucket_dir_entry> last_entry; | |
2755 | /* We need to rewind to the last object in a listing. */ | |
2756 | do { | |
2757 | /* List bucket entries in chunks. */ | |
2758 | static constexpr int MAX_LIST_OBJS = 100; | |
2759 | std::vector<rgw_bucket_dir_entry> entries(MAX_LIST_OBJS); | |
2760 | ||
2761 | int ret = list_op.list_objects(dpp, MAX_LIST_OBJS, &entries, nullptr, | |
2762 | &is_truncated, null_yield); | |
2763 | if (ret < 0) { | |
2764 | return ret; | |
2765 | } else if (!entries.empty()) { | |
2766 | last_entry = entries.back(); | |
2767 | } | |
2768 | } while (is_truncated); | |
2769 | ||
2770 | if (last_entry) { | |
2771 | return handler(*last_entry); | |
2772 | } | |
2773 | ||
2774 | /* Empty listing - no items we can run handler on. */ | |
2775 | return 0; | |
2776 | } | |
2777 | ||
2778 | bool RGWRados::swift_versioning_enabled(rgw::sal::RGWBucket* bucket) const | |
2779 | { | |
2780 | return bucket->get_info().has_swift_versioning() && | |
2781 | bucket->get_info().swift_ver_location.size(); | |
2782 | } | |
2783 | ||
2784 | int RGWRados::swift_versioning_copy(RGWObjectCtx& obj_ctx, | |
2785 | const rgw_user& user, | |
2786 | rgw::sal::RGWBucket* bucket, | |
2787 | rgw::sal::RGWObject* obj, | |
2788 | const DoutPrefixProvider *dpp, | |
2789 | optional_yield y) | |
2790 | { | |
2791 | if (! swift_versioning_enabled(bucket)) { | |
2792 | return 0; | |
2793 | } | |
2794 | ||
2795 | obj->set_atomic(&obj_ctx); | |
2796 | ||
2797 | RGWObjState * state = nullptr; | |
2798 | int r = get_obj_state(dpp, &obj_ctx, bucket->get_info(), obj->get_obj(), &state, false, y); | |
2799 | if (r < 0) { | |
2800 | return r; | |
2801 | } | |
2802 | ||
2803 | if (!state->exists) { | |
2804 | return 0; | |
2805 | } | |
2806 | ||
2807 | const string& src_name = obj->get_oid(); | |
2808 | char buf[src_name.size() + 32]; | |
2809 | struct timespec ts = ceph::real_clock::to_timespec(state->mtime); | |
2810 | snprintf(buf, sizeof(buf), "%03x%s/%lld.%06ld", (int)src_name.size(), | |
2811 | src_name.c_str(), (long long)ts.tv_sec, ts.tv_nsec / 1000); | |
2812 | ||
2813 | RGWBucketInfo dest_bucket_info; | |
2814 | ||
2815 | r = get_bucket_info(&svc, bucket->get_tenant(), bucket->get_info().swift_ver_location, dest_bucket_info, NULL, null_yield, NULL); | |
2816 | if (r < 0) { | |
2817 | ldpp_dout(dpp, 10) << "failed to read dest bucket info: r=" << r << dendl; | |
2818 | if (r == -ENOENT) { | |
2819 | return -ERR_PRECONDITION_FAILED; | |
2820 | } | |
2821 | return r; | |
2822 | } | |
2823 | ||
2824 | if (dest_bucket_info.owner != bucket->get_info().owner) { | |
2825 | return -ERR_PRECONDITION_FAILED; | |
2826 | } | |
2827 | ||
2828 | rgw::sal::RGWRadosBucket dest_bucket(store, dest_bucket_info); | |
2829 | rgw::sal::RGWRadosObject dest_obj(store, rgw_obj_key(buf), &dest_bucket); | |
2830 | ||
2831 | if (dest_bucket_info.versioning_enabled()){ | |
2832 | dest_obj.gen_rand_obj_instance_name(); | |
2833 | } | |
2834 | ||
2835 | dest_obj.set_atomic(&obj_ctx); | |
2836 | ||
2837 | rgw_zone_id no_zone; | |
2838 | ||
2839 | r = copy_obj(obj_ctx, | |
2840 | user, | |
2841 | NULL, /* req_info *info */ | |
2842 | no_zone, | |
2843 | &dest_obj, | |
2844 | obj, | |
2845 | &dest_bucket, | |
2846 | bucket, | |
2847 | bucket->get_placement_rule(), | |
2848 | NULL, /* time_t *src_mtime */ | |
2849 | NULL, /* time_t *mtime */ | |
2850 | NULL, /* const time_t *mod_ptr */ | |
2851 | NULL, /* const time_t *unmod_ptr */ | |
2852 | false, /* bool high_precision_time */ | |
2853 | NULL, /* const char *if_match */ | |
2854 | NULL, /* const char *if_nomatch */ | |
2855 | RGWRados::ATTRSMOD_NONE, | |
2856 | true, /* bool copy_if_newer */ | |
2857 | state->attrset, | |
2858 | RGWObjCategory::Main, | |
2859 | 0, /* uint64_t olh_epoch */ | |
2860 | real_time(), /* time_t delete_at */ | |
2861 | NULL, /* string *version_id */ | |
2862 | NULL, /* string *ptag */ | |
2863 | NULL, /* string *petag */ | |
2864 | NULL, /* void (*progress_cb)(off_t, void *) */ | |
2865 | NULL, /* void *progress_data */ | |
2866 | dpp, | |
2867 | null_yield); | |
2868 | if (r == -ECANCELED || r == -ENOENT) { | |
2869 | /* Has already been overwritten, meaning another rgw process already | |
2870 | * copied it out */ | |
2871 | return 0; | |
2872 | } | |
2873 | ||
2874 | return r; | |
2875 | } | |
2876 | ||
2877 | int RGWRados::swift_versioning_restore(RGWObjectCtx& obj_ctx, | |
2878 | const rgw_user& user, | |
2879 | rgw::sal::RGWBucket* bucket, | |
2880 | rgw::sal::RGWObject* obj, | |
2881 | bool& restored, /* out */ | |
2882 | const DoutPrefixProvider *dpp) | |
2883 | { | |
2884 | if (! swift_versioning_enabled(bucket)) { | |
2885 | return 0; | |
2886 | } | |
2887 | ||
2888 | /* Bucket info of the bucket that stores previous versions of our object. */ | |
2889 | RGWBucketInfo archive_binfo; | |
2890 | ||
2891 | int ret = get_bucket_info(&svc, bucket->get_tenant(), | |
2892 | bucket->get_info().swift_ver_location, | |
2893 | archive_binfo, nullptr, null_yield, nullptr); | |
2894 | if (ret < 0) { | |
2895 | return ret; | |
2896 | } | |
2897 | ||
2898 | /* Abort the operation if the bucket storing our archive belongs to someone | |
2899 | * else. This is a limitation in comparison to Swift as we aren't taking ACLs | |
2900 | * into consideration. For we can live with that. | |
2901 | * | |
2902 | * TODO: delegate this check to un upper layer and compare with ACLs. */ | |
2903 | if (bucket->get_info().owner != archive_binfo.owner) { | |
2904 | return -EPERM; | |
2905 | } | |
2906 | ||
2907 | /* This code will be executed on latest version of the object. */ | |
2908 | const auto handler = [&](const rgw_bucket_dir_entry& entry) -> int { | |
2909 | rgw_zone_id no_zone; | |
2910 | ||
2911 | /* We don't support object versioning of Swift API on those buckets that | |
2912 | * are already versioned using the S3 mechanism. This affects also bucket | |
2913 | * storing archived objects. Otherwise the delete operation would create | |
2914 | * a deletion marker. */ | |
2915 | if (archive_binfo.versioned()) { | |
2916 | restored = false; | |
2917 | return -ERR_PRECONDITION_FAILED; | |
2918 | } | |
2919 | ||
2920 | /* We are requesting ATTRSMOD_NONE so the attr attribute is perfectly | |
2921 | * irrelevant and may be safely skipped. */ | |
2922 | std::map<std::string, ceph::bufferlist> no_attrs; | |
2923 | ||
2924 | rgw::sal::RGWRadosBucket archive_bucket(store, archive_binfo); | |
2925 | rgw::sal::RGWRadosObject archive_obj(store, entry.key, &archive_bucket); | |
2926 | ||
2927 | if (bucket->versioning_enabled()){ | |
2928 | obj->gen_rand_obj_instance_name(); | |
2929 | } | |
2930 | ||
2931 | archive_obj.set_atomic(&obj_ctx); | |
2932 | obj->set_atomic(&obj_ctx); | |
2933 | ||
2934 | int ret = copy_obj(obj_ctx, | |
2935 | user, | |
2936 | nullptr, /* req_info *info */ | |
2937 | no_zone, | |
2938 | obj, /* dest obj */ | |
2939 | &archive_obj, /* src obj */ | |
2940 | bucket, /* dest bucket info */ | |
2941 | &archive_bucket, /* src bucket info */ | |
2942 | bucket->get_placement_rule(), /* placement_rule */ | |
2943 | nullptr, /* time_t *src_mtime */ | |
2944 | nullptr, /* time_t *mtime */ | |
2945 | nullptr, /* const time_t *mod_ptr */ | |
2946 | nullptr, /* const time_t *unmod_ptr */ | |
2947 | false, /* bool high_precision_time */ | |
2948 | nullptr, /* const char *if_match */ | |
2949 | nullptr, /* const char *if_nomatch */ | |
2950 | RGWRados::ATTRSMOD_NONE, | |
2951 | true, /* bool copy_if_newer */ | |
2952 | no_attrs, | |
2953 | RGWObjCategory::Main, | |
2954 | 0, /* uint64_t olh_epoch */ | |
2955 | real_time(), /* time_t delete_at */ | |
2956 | nullptr, /* string *version_id */ | |
2957 | nullptr, /* string *ptag */ | |
2958 | nullptr, /* string *petag */ | |
2959 | nullptr, /* void (*progress_cb)(off_t, void *) */ | |
2960 | nullptr, /* void *progress_data */ | |
2961 | dpp, | |
2962 | null_yield); | |
2963 | if (ret == -ECANCELED || ret == -ENOENT) { | |
2964 | /* Has already been overwritten, meaning another rgw process already | |
2965 | * copied it out */ | |
2966 | return 0; | |
2967 | } else if (ret < 0) { | |
2968 | return ret; | |
2969 | } else { | |
2970 | restored = true; | |
2971 | } | |
2972 | ||
2973 | /* Need to remove the archived copy. */ | |
2974 | ret = delete_obj(dpp, obj_ctx, archive_binfo, archive_obj.get_obj(), | |
2975 | archive_binfo.versioning_status()); | |
2976 | ||
2977 | return ret; | |
2978 | }; | |
2979 | ||
2980 | const std::string& obj_name = obj->get_oid(); | |
2981 | const auto prefix = boost::str(boost::format("%03x%s") % obj_name.size() | |
2982 | % obj_name); | |
2983 | ||
2984 | return on_last_entry_in_listing(dpp, archive_binfo, prefix, std::string(), | |
2985 | handler); | |
2986 | } | |
2987 | ||
2988 | int RGWRados::Object::Write::_do_write_meta(const DoutPrefixProvider *dpp, | |
2989 | uint64_t size, uint64_t accounted_size, | |
2990 | map<string, bufferlist>& attrs, | |
2991 | bool assume_noent, bool modify_tail, | |
2992 | void *_index_op, optional_yield y) | |
2993 | { | |
2994 | RGWRados::Bucket::UpdateIndex *index_op = static_cast<RGWRados::Bucket::UpdateIndex *>(_index_op); | |
2995 | RGWRados *store = target->get_store(); | |
2996 | ||
2997 | ObjectWriteOperation op; | |
2998 | #ifdef WITH_LTTNG | |
2999 | const struct req_state* s = get_req_state(); | |
3000 | string req_id; | |
3001 | if (!s) { | |
3002 | // fake req_id | |
3003 | req_id = store->svc.zone_utils->unique_id(store->get_new_req_id()); | |
3004 | } else { | |
3005 | req_id = s->req_id; | |
3006 | } | |
3007 | #endif | |
3008 | ||
3009 | RGWObjState *state; | |
3010 | int r = target->get_state(dpp, &state, false, y, assume_noent); | |
3011 | if (r < 0) | |
3012 | return r; | |
3013 | ||
3014 | rgw_obj& obj = target->get_obj(); | |
3015 | ||
3016 | if (obj.get_oid().empty()) { | |
3017 | ldpp_dout(dpp, 0) << "ERROR: " << __func__ << "(): cannot write object with empty name" << dendl; | |
3018 | return -EIO; | |
3019 | } | |
3020 | ||
3021 | rgw_rados_ref ref; | |
3022 | r = store->get_obj_head_ref(dpp, target->get_bucket_info(), obj, &ref); | |
3023 | if (r < 0) | |
3024 | return r; | |
3025 | ||
3026 | bool is_olh = state->is_olh; | |
3027 | ||
3028 | bool reset_obj = (meta.flags & PUT_OBJ_CREATE) != 0; | |
3029 | ||
3030 | const string *ptag = meta.ptag; | |
3031 | if (!ptag && !index_op->get_optag()->empty()) { | |
3032 | ptag = index_op->get_optag(); | |
3033 | } | |
3034 | r = target->prepare_atomic_modification(dpp, op, reset_obj, ptag, meta.if_match, meta.if_nomatch, false, modify_tail, y); | |
3035 | if (r < 0) | |
3036 | return r; | |
3037 | ||
3038 | if (real_clock::is_zero(meta.set_mtime)) { | |
3039 | meta.set_mtime = real_clock::now(); | |
3040 | } | |
3041 | ||
3042 | if (target->bucket_info.obj_lock_enabled() && target->bucket_info.obj_lock.has_rule() && meta.flags == PUT_OBJ_CREATE) { | |
3043 | auto iter = attrs.find(RGW_ATTR_OBJECT_RETENTION); | |
3044 | if (iter == attrs.end()) { | |
3045 | real_time lock_until_date = target->bucket_info.obj_lock.get_lock_until_date(meta.set_mtime); | |
3046 | string mode = target->bucket_info.obj_lock.get_mode(); | |
3047 | RGWObjectRetention obj_retention(mode, lock_until_date); | |
3048 | bufferlist bl; | |
3049 | obj_retention.encode(bl); | |
3050 | op.setxattr(RGW_ATTR_OBJECT_RETENTION, bl); | |
3051 | } | |
3052 | } | |
3053 | ||
3054 | if (state->is_olh) { | |
3055 | op.setxattr(RGW_ATTR_OLH_ID_TAG, state->olh_tag); | |
3056 | } | |
3057 | ||
3058 | struct timespec mtime_ts = real_clock::to_timespec(meta.set_mtime); | |
3059 | op.mtime2(&mtime_ts); | |
3060 | ||
3061 | if (meta.data) { | |
3062 | /* if we want to overwrite the data, we also want to overwrite the | |
3063 | xattrs, so just remove the object */ | |
3064 | op.write_full(*meta.data); | |
3065 | } | |
3066 | ||
3067 | string etag; | |
3068 | string content_type; | |
3069 | bufferlist acl_bl; | |
3070 | string storage_class; | |
3071 | ||
3072 | map<string, bufferlist>::iterator iter; | |
3073 | if (meta.rmattrs) { | |
3074 | for (iter = meta.rmattrs->begin(); iter != meta.rmattrs->end(); ++iter) { | |
3075 | const string& name = iter->first; | |
3076 | op.rmxattr(name.c_str()); | |
3077 | } | |
3078 | } | |
3079 | ||
3080 | if (meta.manifest) { | |
3081 | storage_class = meta.manifest->get_tail_placement().placement_rule.storage_class; | |
3082 | ||
3083 | /* remove existing manifest attr */ | |
3084 | iter = attrs.find(RGW_ATTR_MANIFEST); | |
3085 | if (iter != attrs.end()) | |
3086 | attrs.erase(iter); | |
3087 | ||
3088 | bufferlist bl; | |
3089 | encode(*meta.manifest, bl); | |
3090 | op.setxattr(RGW_ATTR_MANIFEST, bl); | |
3091 | } | |
3092 | ||
3093 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
3094 | const string& name = iter->first; | |
3095 | bufferlist& bl = iter->second; | |
3096 | ||
3097 | if (!bl.length()) | |
3098 | continue; | |
3099 | ||
3100 | op.setxattr(name.c_str(), bl); | |
3101 | ||
3102 | if (name.compare(RGW_ATTR_ETAG) == 0) { | |
3103 | etag = rgw_bl_str(bl); | |
3104 | } else if (name.compare(RGW_ATTR_CONTENT_TYPE) == 0) { | |
3105 | content_type = rgw_bl_str(bl); | |
3106 | } else if (name.compare(RGW_ATTR_ACL) == 0) { | |
3107 | acl_bl = bl; | |
3108 | } | |
3109 | } | |
3110 | if (attrs.find(RGW_ATTR_PG_VER) == attrs.end()) { | |
3111 | cls_rgw_obj_store_pg_ver(op, RGW_ATTR_PG_VER); | |
3112 | } | |
3113 | ||
3114 | if (attrs.find(RGW_ATTR_SOURCE_ZONE) == attrs.end()) { | |
3115 | bufferlist bl; | |
3116 | encode(store->svc.zone->get_zone_short_id(), bl); | |
3117 | op.setxattr(RGW_ATTR_SOURCE_ZONE, bl); | |
3118 | } | |
3119 | ||
3120 | if (!storage_class.empty()) { | |
3121 | bufferlist bl; | |
3122 | bl.append(storage_class); | |
3123 | op.setxattr(RGW_ATTR_STORAGE_CLASS, bl); | |
3124 | } | |
3125 | ||
3126 | if (!op.size()) | |
3127 | return 0; | |
3128 | ||
3129 | uint64_t epoch; | |
3130 | int64_t poolid; | |
3131 | bool orig_exists; | |
3132 | uint64_t orig_size; | |
3133 | ||
3134 | if (!reset_obj) { //Multipart upload, it has immutable head. | |
3135 | orig_exists = false; | |
3136 | orig_size = 0; | |
3137 | } else { | |
3138 | orig_exists = state->exists; | |
3139 | orig_size = state->accounted_size; | |
3140 | } | |
3141 | ||
3142 | bool versioned_target = (meta.olh_epoch && *meta.olh_epoch > 0) || | |
3143 | !obj.key.instance.empty(); | |
3144 | ||
3145 | bool versioned_op = (target->versioning_enabled() || is_olh || versioned_target); | |
3146 | ||
3147 | if (versioned_op) { | |
3148 | index_op->set_bilog_flags(RGW_BILOG_FLAG_VERSIONED_OP); | |
3149 | } | |
3150 | ||
3151 | if (!index_op->is_prepared()) { | |
3152 | tracepoint(rgw_rados, prepare_enter, req_id.c_str()); | |
3153 | r = index_op->prepare(dpp, CLS_RGW_OP_ADD, &state->write_tag, y); | |
3154 | tracepoint(rgw_rados, prepare_exit, req_id.c_str()); | |
3155 | if (r < 0) | |
3156 | return r; | |
3157 | } | |
3158 | ||
3159 | auto& ioctx = ref.pool.ioctx(); | |
3160 | ||
3161 | tracepoint(rgw_rados, operate_enter, req_id.c_str()); | |
3162 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
3163 | tracepoint(rgw_rados, operate_exit, req_id.c_str()); | |
3164 | if (r < 0) { /* we can expect to get -ECANCELED if object was replaced under, | |
3165 | or -ENOENT if was removed, or -EEXIST if it did not exist | |
3166 | before and now it does */ | |
3167 | if (r == -EEXIST && assume_noent) { | |
3168 | target->invalidate_state(); | |
3169 | return r; | |
3170 | } | |
3171 | goto done_cancel; | |
3172 | } | |
3173 | ||
3174 | epoch = ioctx.get_last_version(); | |
3175 | poolid = ioctx.get_id(); | |
3176 | ||
3177 | r = target->complete_atomic_modification(dpp); | |
3178 | if (r < 0) { | |
3179 | ldpp_dout(dpp, 0) << "ERROR: complete_atomic_modification returned r=" << r << dendl; | |
3180 | } | |
3181 | ||
3182 | tracepoint(rgw_rados, complete_enter, req_id.c_str()); | |
3183 | r = index_op->complete(dpp, poolid, epoch, size, accounted_size, | |
3184 | meta.set_mtime, etag, content_type, | |
3185 | storage_class, &acl_bl, | |
3186 | meta.category, meta.remove_objs, meta.user_data, meta.appendable); | |
3187 | tracepoint(rgw_rados, complete_exit, req_id.c_str()); | |
3188 | if (r < 0) | |
3189 | goto done_cancel; | |
3190 | ||
3191 | if (meta.mtime) { | |
3192 | *meta.mtime = meta.set_mtime; | |
3193 | } | |
3194 | ||
3195 | /* note that index_op was using state so we couldn't invalidate it earlier */ | |
3196 | target->invalidate_state(); | |
3197 | state = NULL; | |
3198 | ||
3199 | if (versioned_op && meta.olh_epoch) { | |
3200 | r = store->set_olh(dpp, target->get_ctx(), target->get_bucket_info(), obj, false, NULL, *meta.olh_epoch, real_time(), false, y, meta.zones_trace); | |
3201 | if (r < 0) { | |
3202 | return r; | |
3203 | } | |
3204 | } | |
3205 | ||
3206 | if (!real_clock::is_zero(meta.delete_at)) { | |
3207 | rgw_obj_index_key obj_key; | |
3208 | obj.key.get_index_key(&obj_key); | |
3209 | ||
3210 | r = store->obj_expirer->hint_add(dpp, meta.delete_at, obj.bucket.tenant, obj.bucket.name, | |
3211 | obj.bucket.bucket_id, obj_key); | |
3212 | if (r < 0) { | |
3213 | ldpp_dout(dpp, 0) << "ERROR: objexp_hint_add() returned r=" << r << ", object will not get removed" << dendl; | |
3214 | /* ignoring error, nothing we can do at this point */ | |
3215 | } | |
3216 | } | |
3217 | meta.canceled = false; | |
3218 | ||
3219 | /* update quota cache */ | |
3220 | if (meta.completeMultipart){ | |
3221 | store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1), | |
3222 | 0, orig_size); | |
3223 | } | |
3224 | else { | |
3225 | store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1), | |
3226 | accounted_size, orig_size); | |
3227 | } | |
3228 | return 0; | |
3229 | ||
3230 | done_cancel: | |
3231 | int ret = index_op->cancel(dpp); | |
3232 | if (ret < 0) { | |
3233 | ldpp_dout(dpp, 0) << "ERROR: index_op.cancel()() returned ret=" << ret << dendl; | |
3234 | } | |
3235 | ||
3236 | meta.canceled = true; | |
3237 | ||
3238 | /* we lost in a race. There are a few options: | |
3239 | * - existing object was rewritten (ECANCELED) | |
3240 | * - non existing object was created (EEXIST) | |
3241 | * - object was removed (ENOENT) | |
3242 | * should treat it as a success | |
3243 | */ | |
3244 | if (meta.if_match == NULL && meta.if_nomatch == NULL) { | |
3245 | if (r == -ECANCELED || r == -ENOENT || r == -EEXIST) { | |
3246 | r = 0; | |
3247 | } | |
3248 | } else { | |
3249 | if (meta.if_match != NULL) { | |
3250 | // only overwrite existing object | |
3251 | if (strcmp(meta.if_match, "*") == 0) { | |
3252 | if (r == -ENOENT) { | |
3253 | r = -ERR_PRECONDITION_FAILED; | |
3254 | } else if (r == -ECANCELED) { | |
3255 | r = 0; | |
3256 | } | |
3257 | } | |
3258 | } | |
3259 | ||
3260 | if (meta.if_nomatch != NULL) { | |
3261 | // only create a new object | |
3262 | if (strcmp(meta.if_nomatch, "*") == 0) { | |
3263 | if (r == -EEXIST) { | |
3264 | r = -ERR_PRECONDITION_FAILED; | |
3265 | } else if (r == -ENOENT) { | |
3266 | r = 0; | |
3267 | } | |
3268 | } | |
3269 | } | |
3270 | } | |
3271 | ||
3272 | return r; | |
3273 | } | |
3274 | ||
3275 | int RGWRados::Object::Write::write_meta(const DoutPrefixProvider *dpp, uint64_t size, uint64_t accounted_size, | |
3276 | map<string, bufferlist>& attrs, optional_yield y) | |
3277 | { | |
3278 | RGWBucketInfo& bucket_info = target->get_bucket_info(); | |
3279 | ||
3280 | RGWRados::Bucket bop(target->get_store(), bucket_info); | |
3281 | RGWRados::Bucket::UpdateIndex index_op(&bop, target->get_obj()); | |
3282 | index_op.set_zones_trace(meta.zones_trace); | |
3283 | ||
3284 | bool assume_noent = (meta.if_match == NULL && meta.if_nomatch == NULL); | |
3285 | int r; | |
3286 | if (assume_noent) { | |
3287 | r = _do_write_meta(dpp, size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op, y); | |
3288 | if (r == -EEXIST) { | |
3289 | assume_noent = false; | |
3290 | } | |
3291 | } | |
3292 | if (!assume_noent) { | |
3293 | r = _do_write_meta(dpp, size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op, y); | |
3294 | } | |
3295 | return r; | |
3296 | } | |
3297 | ||
3298 | class RGWRadosPutObj : public RGWHTTPStreamRWRequest::ReceiveCB | |
3299 | { | |
3300 | const DoutPrefixProvider *dpp; | |
3301 | CephContext* cct; | |
3302 | rgw_obj obj; | |
3303 | rgw::putobj::DataProcessor *filter; | |
3304 | boost::optional<RGWPutObj_Compress>& compressor; | |
3305 | bool try_etag_verify; | |
3306 | rgw::putobj::etag_verifier_ptr etag_verifier; | |
3307 | boost::optional<rgw::putobj::ChunkProcessor> buffering; | |
3308 | CompressorRef& plugin; | |
3309 | rgw::putobj::ObjectProcessor *processor; | |
3310 | void (*progress_cb)(off_t, void *); | |
3311 | void *progress_data; | |
3312 | bufferlist extra_data_bl, manifest_bl; | |
3313 | std::optional<RGWCompressionInfo> compression_info; | |
3314 | uint64_t extra_data_left{0}; | |
3315 | bool need_to_process_attrs{true}; | |
3316 | uint64_t data_len{0}; | |
3317 | map<string, bufferlist> src_attrs; | |
3318 | uint64_t ofs{0}; | |
3319 | uint64_t lofs{0}; /* logical ofs */ | |
3320 | std::function<int(map<string, bufferlist>&)> attrs_handler; | |
3321 | public: | |
3322 | RGWRadosPutObj(const DoutPrefixProvider *dpp, | |
3323 | CephContext* cct, | |
3324 | CompressorRef& plugin, | |
3325 | boost::optional<RGWPutObj_Compress>& compressor, | |
3326 | rgw::putobj::ObjectProcessor *p, | |
3327 | void (*_progress_cb)(off_t, void *), | |
3328 | void *_progress_data, | |
3329 | std::function<int(map<string, bufferlist>&)> _attrs_handler) : | |
3330 | dpp(dpp), | |
3331 | cct(cct), | |
3332 | filter(p), | |
3333 | compressor(compressor), | |
3334 | try_etag_verify(cct->_conf->rgw_sync_obj_etag_verify), | |
3335 | plugin(plugin), | |
3336 | processor(p), | |
3337 | progress_cb(_progress_cb), | |
3338 | progress_data(_progress_data), | |
3339 | attrs_handler(_attrs_handler) {} | |
3340 | ||
3341 | int process_attrs(void) { | |
3342 | if (extra_data_bl.length()) { | |
3343 | JSONParser jp; | |
3344 | if (!jp.parse(extra_data_bl.c_str(), extra_data_bl.length())) { | |
3345 | ldpp_dout(dpp, 0) << "failed to parse response extra data. len=" << extra_data_bl.length() << " data=" << extra_data_bl.c_str() << dendl; | |
3346 | return -EIO; | |
3347 | } | |
3348 | ||
3349 | JSONDecoder::decode_json("attrs", src_attrs, &jp); | |
3350 | ||
3351 | auto iter = src_attrs.find(RGW_ATTR_COMPRESSION); | |
3352 | if (iter != src_attrs.end()) { | |
3353 | const bufferlist bl = std::move(iter->second); | |
3354 | src_attrs.erase(iter); // don't preserve source compression info | |
3355 | ||
3356 | if (try_etag_verify) { | |
3357 | // if we're trying to verify etags, we need to convert compressed | |
3358 | // ranges in the manifest back into logical multipart part offsets | |
3359 | RGWCompressionInfo info; | |
3360 | bool compressed = false; | |
3361 | int r = rgw_compression_info_from_attr(bl, compressed, info); | |
3362 | if (r < 0) { | |
3363 | ldpp_dout(dpp, 4) << "failed to decode compression info, " | |
3364 | "disabling etag verification" << dendl; | |
3365 | try_etag_verify = false; | |
3366 | } else if (compressed) { | |
3367 | compression_info = std::move(info); | |
3368 | } | |
3369 | } | |
3370 | } | |
3371 | /* We need the manifest to recompute the ETag for verification */ | |
3372 | iter = src_attrs.find(RGW_ATTR_MANIFEST); | |
3373 | if (iter != src_attrs.end()) { | |
3374 | manifest_bl = std::move(iter->second); | |
3375 | src_attrs.erase(iter); | |
3376 | } | |
3377 | ||
3378 | // filter out olh attributes | |
3379 | iter = src_attrs.lower_bound(RGW_ATTR_OLH_PREFIX); | |
3380 | while (iter != src_attrs.end()) { | |
3381 | if (!boost::algorithm::starts_with(iter->first, RGW_ATTR_OLH_PREFIX)) { | |
3382 | break; | |
3383 | } | |
3384 | iter = src_attrs.erase(iter); | |
3385 | } | |
3386 | } | |
3387 | ||
3388 | int ret = attrs_handler(src_attrs); | |
3389 | if (ret < 0) { | |
3390 | return ret; | |
3391 | } | |
3392 | ||
3393 | if (plugin && src_attrs.find(RGW_ATTR_CRYPT_MODE) == src_attrs.end()) { | |
3394 | //do not compress if object is encrypted | |
3395 | compressor = boost::in_place(cct, plugin, filter); | |
3396 | // add a filter that buffers data so we don't try to compress tiny blocks. | |
3397 | // libcurl reads in 16k at a time, and we need at least 64k to get a good | |
3398 | // compression ratio | |
3399 | constexpr unsigned buffer_size = 512 * 1024; | |
3400 | buffering = boost::in_place(&*compressor, buffer_size); | |
3401 | filter = &*buffering; | |
3402 | } | |
3403 | ||
3404 | /* | |
3405 | * Presently we don't support ETag based verification if encryption is | |
3406 | * requested. We can enable simultaneous support once we have a mechanism | |
3407 | * to know the sequence in which the filters must be applied. | |
3408 | */ | |
3409 | if (try_etag_verify && src_attrs.find(RGW_ATTR_CRYPT_MODE) == src_attrs.end()) { | |
3410 | ret = rgw::putobj::create_etag_verifier(dpp, cct, filter, manifest_bl, | |
3411 | compression_info, | |
3412 | etag_verifier); | |
3413 | if (ret < 0) { | |
3414 | ldpp_dout(dpp, 4) << "failed to initial etag verifier, " | |
3415 | "disabling etag verification" << dendl; | |
3416 | } else { | |
3417 | filter = etag_verifier.get(); | |
3418 | } | |
3419 | } | |
3420 | ||
3421 | need_to_process_attrs = false; | |
3422 | ||
3423 | return 0; | |
3424 | } | |
3425 | ||
3426 | int handle_data(bufferlist& bl, bool *pause) override { | |
3427 | if (progress_cb) { | |
3428 | progress_cb(data_len, progress_data); | |
3429 | } | |
3430 | if (extra_data_left) { | |
3431 | uint64_t extra_len = bl.length(); | |
3432 | if (extra_len > extra_data_left) | |
3433 | extra_len = extra_data_left; | |
3434 | ||
3435 | bufferlist extra; | |
3436 | bl.splice(0, extra_len, &extra); | |
3437 | extra_data_bl.append(extra); | |
3438 | ||
3439 | extra_data_left -= extra_len; | |
3440 | if (extra_data_left == 0) { | |
3441 | int res = process_attrs(); | |
3442 | if (res < 0) | |
3443 | return res; | |
3444 | } | |
3445 | ofs += extra_len; | |
3446 | if (bl.length() == 0) { | |
3447 | return 0; | |
3448 | } | |
3449 | } | |
3450 | if (need_to_process_attrs) { | |
3451 | /* need to call process_attrs() even if we don't get any attrs, | |
3452 | * need it to call attrs_handler(). | |
3453 | */ | |
3454 | int res = process_attrs(); | |
3455 | if (res < 0) { | |
3456 | return res; | |
3457 | } | |
3458 | } | |
3459 | ||
3460 | ceph_assert(uint64_t(ofs) >= extra_data_len); | |
3461 | ||
3462 | uint64_t size = bl.length(); | |
3463 | ofs += size; | |
3464 | ||
3465 | const uint64_t lofs = data_len; | |
3466 | data_len += size; | |
3467 | ||
3468 | return filter->process(std::move(bl), lofs); | |
3469 | } | |
3470 | ||
3471 | int flush() { | |
3472 | return filter->process({}, data_len); | |
3473 | } | |
3474 | ||
3475 | bufferlist& get_extra_data() { return extra_data_bl; } | |
3476 | ||
3477 | map<string, bufferlist>& get_attrs() { return src_attrs; } | |
3478 | ||
3479 | void set_extra_data_len(uint64_t len) override { | |
3480 | extra_data_left = len; | |
3481 | RGWHTTPStreamRWRequest::ReceiveCB::set_extra_data_len(len); | |
3482 | } | |
3483 | ||
3484 | uint64_t get_data_len() { | |
3485 | return data_len; | |
3486 | } | |
3487 | ||
3488 | std::string get_verifier_etag() { | |
3489 | if (etag_verifier) { | |
3490 | etag_verifier->calculate_etag(); | |
3491 | return etag_verifier->get_calculated_etag(); | |
3492 | } else { | |
3493 | return ""; | |
3494 | } | |
3495 | } | |
3496 | }; | |
3497 | ||
3498 | /* | |
3499 | * prepare attrset depending on attrs_mod. | |
3500 | */ | |
3501 | static void set_copy_attrs(map<string, bufferlist>& src_attrs, | |
3502 | map<string, bufferlist>& attrs, | |
3503 | RGWRados::AttrsMod attrs_mod) | |
3504 | { | |
3505 | switch (attrs_mod) { | |
3506 | case RGWRados::ATTRSMOD_NONE: | |
3507 | attrs = src_attrs; | |
3508 | break; | |
3509 | case RGWRados::ATTRSMOD_REPLACE: | |
3510 | if (!attrs[RGW_ATTR_ETAG].length()) { | |
3511 | attrs[RGW_ATTR_ETAG] = src_attrs[RGW_ATTR_ETAG]; | |
3512 | } | |
3513 | if (!attrs[RGW_ATTR_TAIL_TAG].length()) { | |
3514 | auto ttiter = src_attrs.find(RGW_ATTR_TAIL_TAG); | |
3515 | if (ttiter != src_attrs.end()) { | |
3516 | attrs[RGW_ATTR_TAIL_TAG] = src_attrs[RGW_ATTR_TAIL_TAG]; | |
3517 | } | |
3518 | } | |
3519 | break; | |
3520 | case RGWRados::ATTRSMOD_MERGE: | |
3521 | for (map<string, bufferlist>::iterator it = src_attrs.begin(); it != src_attrs.end(); ++it) { | |
3522 | if (attrs.find(it->first) == attrs.end()) { | |
3523 | attrs[it->first] = it->second; | |
3524 | } | |
3525 | } | |
3526 | break; | |
3527 | } | |
3528 | } | |
3529 | ||
3530 | int RGWRados::rewrite_obj(RGWBucketInfo& dest_bucket_info, rgw::sal::RGWObject* obj, const DoutPrefixProvider *dpp, optional_yield y) | |
3531 | { | |
3532 | RGWObjectCtx rctx(this->store); | |
3533 | rgw::sal::RGWRadosBucket bucket(store, dest_bucket_info); | |
3534 | ||
3535 | return obj->copy_obj_data(rctx, &bucket, obj, 0, NULL, dpp, y); | |
3536 | } | |
3537 | ||
3538 | struct obj_time_weight { | |
3539 | real_time mtime; | |
3540 | uint32_t zone_short_id; | |
3541 | uint64_t pg_ver; | |
3542 | bool high_precision; | |
3543 | ||
3544 | obj_time_weight() : zone_short_id(0), pg_ver(0), high_precision(false) {} | |
3545 | ||
3546 | bool compare_low_precision(const obj_time_weight& rhs) { | |
3547 | struct timespec l = ceph::real_clock::to_timespec(mtime); | |
3548 | struct timespec r = ceph::real_clock::to_timespec(rhs.mtime); | |
3549 | l.tv_nsec = 0; | |
3550 | r.tv_nsec = 0; | |
3551 | if (l > r) { | |
3552 | return false; | |
3553 | } | |
3554 | if (l < r) { | |
3555 | return true; | |
3556 | } | |
3557 | if (!zone_short_id || !rhs.zone_short_id) { | |
3558 | /* don't compare zone ids, if one wasn't provided */ | |
3559 | return false; | |
3560 | } | |
3561 | if (zone_short_id != rhs.zone_short_id) { | |
3562 | return (zone_short_id < rhs.zone_short_id); | |
3563 | } | |
3564 | return (pg_ver < rhs.pg_ver); | |
3565 | ||
3566 | } | |
3567 | ||
3568 | bool operator<(const obj_time_weight& rhs) { | |
3569 | if (!high_precision || !rhs.high_precision) { | |
3570 | return compare_low_precision(rhs); | |
3571 | } | |
3572 | if (mtime > rhs.mtime) { | |
3573 | return false; | |
3574 | } | |
3575 | if (mtime < rhs.mtime) { | |
3576 | return true; | |
3577 | } | |
3578 | if (!zone_short_id || !rhs.zone_short_id) { | |
3579 | /* don't compare zone ids, if one wasn't provided */ | |
3580 | return false; | |
3581 | } | |
3582 | if (zone_short_id != rhs.zone_short_id) { | |
3583 | return (zone_short_id < rhs.zone_short_id); | |
3584 | } | |
3585 | return (pg_ver < rhs.pg_ver); | |
3586 | } | |
3587 | ||
3588 | void init(const real_time& _mtime, uint32_t _short_id, uint64_t _pg_ver) { | |
3589 | mtime = _mtime; | |
3590 | zone_short_id = _short_id; | |
3591 | pg_ver = _pg_ver; | |
3592 | } | |
3593 | ||
3594 | void init(RGWObjState *state) { | |
3595 | mtime = state->mtime; | |
3596 | zone_short_id = state->zone_short_id; | |
3597 | pg_ver = state->pg_ver; | |
3598 | } | |
3599 | }; | |
3600 | ||
3601 | inline ostream& operator<<(ostream& out, const obj_time_weight &o) { | |
3602 | out << o.mtime; | |
3603 | ||
3604 | if (o.zone_short_id != 0 || o.pg_ver != 0) { | |
3605 | out << "[zid=" << o.zone_short_id << ", pgv=" << o.pg_ver << "]"; | |
3606 | } | |
3607 | ||
3608 | return out; | |
3609 | } | |
3610 | ||
3611 | class RGWGetExtraDataCB : public RGWHTTPStreamRWRequest::ReceiveCB { | |
3612 | bufferlist extra_data; | |
3613 | public: | |
3614 | RGWGetExtraDataCB() {} | |
3615 | int handle_data(bufferlist& bl, bool *pause) override { | |
3616 | int bl_len = (int)bl.length(); | |
3617 | if (extra_data.length() < extra_data_len) { | |
3618 | off_t max = extra_data_len - extra_data.length(); | |
3619 | if (max > bl_len) { | |
3620 | max = bl_len; | |
3621 | } | |
3622 | bl.splice(0, max, &extra_data); | |
3623 | } | |
3624 | return bl_len; | |
3625 | } | |
3626 | ||
3627 | bufferlist& get_extra_data() { | |
3628 | return extra_data; | |
3629 | } | |
3630 | }; | |
3631 | ||
3632 | int RGWRados::stat_remote_obj(const DoutPrefixProvider *dpp, | |
3633 | RGWObjectCtx& obj_ctx, | |
3634 | const rgw_user& user_id, | |
3635 | req_info *info, | |
3636 | const rgw_zone_id& source_zone, | |
3637 | rgw::sal::RGWObject* src_obj, | |
3638 | const RGWBucketInfo *src_bucket_info, | |
3639 | real_time *src_mtime, | |
3640 | uint64_t *psize, | |
3641 | const real_time *mod_ptr, | |
3642 | const real_time *unmod_ptr, | |
3643 | bool high_precision_time, | |
3644 | const char *if_match, | |
3645 | const char *if_nomatch, | |
3646 | map<string, bufferlist> *pattrs, | |
3647 | map<string, string> *pheaders, | |
3648 | string *version_id, | |
3649 | string *ptag, | |
3650 | string *petag) | |
3651 | { | |
3652 | /* source is in a different zonegroup, copy from there */ | |
3653 | ||
3654 | RGWRESTStreamRWRequest *in_stream_req; | |
3655 | string tag; | |
3656 | map<string, bufferlist> src_attrs; | |
3657 | append_rand_alpha(cct, tag, tag, 32); | |
3658 | obj_time_weight set_mtime_weight; | |
3659 | set_mtime_weight.high_precision = high_precision_time; | |
3660 | ||
3661 | RGWRESTConn *conn; | |
3662 | if (source_zone.empty()) { | |
3663 | if (!src_bucket_info || src_bucket_info->zonegroup.empty()) { | |
3664 | /* source is in the master zonegroup */ | |
3665 | conn = svc.zone->get_master_conn(); | |
3666 | } else { | |
3667 | auto& zonegroup_conn_map = svc.zone->get_zonegroup_conn_map(); | |
3668 | map<string, RGWRESTConn *>::iterator iter = zonegroup_conn_map.find(src_bucket_info->zonegroup); | |
3669 | if (iter == zonegroup_conn_map.end()) { | |
3670 | ldout(cct, 0) << "could not find zonegroup connection to zonegroup: " << source_zone << dendl; | |
3671 | return -ENOENT; | |
3672 | } | |
3673 | conn = iter->second; | |
3674 | } | |
3675 | } else { | |
3676 | auto& zone_conn_map = svc.zone->get_zone_conn_map(); | |
3677 | auto iter = zone_conn_map.find(source_zone); | |
3678 | if (iter == zone_conn_map.end()) { | |
3679 | ldout(cct, 0) << "could not find zone connection to zone: " << source_zone << dendl; | |
3680 | return -ENOENT; | |
3681 | } | |
3682 | conn = iter->second; | |
3683 | } | |
3684 | ||
3685 | RGWGetExtraDataCB cb; | |
3686 | map<string, string> req_headers; | |
3687 | real_time set_mtime; | |
3688 | ||
3689 | const real_time *pmod = mod_ptr; | |
3690 | ||
3691 | obj_time_weight dest_mtime_weight; | |
3692 | ||
3693 | constexpr bool prepend_meta = true; | |
3694 | constexpr bool get_op = true; | |
3695 | constexpr bool rgwx_stat = true; | |
3696 | constexpr bool sync_manifest = true; | |
3697 | constexpr bool skip_decrypt = true; | |
3698 | int ret = conn->get_obj(dpp, user_id, info, src_obj, pmod, unmod_ptr, | |
3699 | dest_mtime_weight.zone_short_id, dest_mtime_weight.pg_ver, | |
3700 | prepend_meta, get_op, rgwx_stat, | |
3701 | sync_manifest, skip_decrypt, | |
3702 | true, &cb, &in_stream_req); | |
3703 | if (ret < 0) { | |
3704 | return ret; | |
3705 | } | |
3706 | ||
3707 | ret = conn->complete_request(in_stream_req, nullptr, &set_mtime, psize, | |
3708 | nullptr, pheaders, null_yield); | |
3709 | if (ret < 0) { | |
3710 | return ret; | |
3711 | } | |
3712 | ||
3713 | bufferlist& extra_data_bl = cb.get_extra_data(); | |
3714 | if (extra_data_bl.length()) { | |
3715 | JSONParser jp; | |
3716 | if (!jp.parse(extra_data_bl.c_str(), extra_data_bl.length())) { | |
3717 | ldout(cct, 0) << "failed to parse response extra data. len=" << extra_data_bl.length() << " data=" << extra_data_bl.c_str() << dendl; | |
3718 | return -EIO; | |
3719 | } | |
3720 | ||
3721 | JSONDecoder::decode_json("attrs", src_attrs, &jp); | |
3722 | ||
3723 | src_attrs.erase(RGW_ATTR_MANIFEST); // not interested in original object layout | |
3724 | } | |
3725 | ||
3726 | if (src_mtime) { | |
3727 | *src_mtime = set_mtime; | |
3728 | } | |
3729 | ||
3730 | if (petag) { | |
3731 | map<string, bufferlist>::iterator iter = src_attrs.find(RGW_ATTR_ETAG); | |
3732 | if (iter != src_attrs.end()) { | |
3733 | bufferlist& etagbl = iter->second; | |
3734 | *petag = etagbl.to_str(); | |
3735 | while (petag->size() > 0 && (*petag)[petag->size() - 1] == '\0') { | |
3736 | *petag = petag->substr(0, petag->size() - 1); | |
3737 | } | |
3738 | } | |
3739 | } | |
3740 | ||
3741 | if (pattrs) { | |
3742 | *pattrs = std::move(src_attrs); | |
3743 | } | |
3744 | ||
3745 | return 0; | |
3746 | } | |
3747 | ||
3748 | int RGWFetchObjFilter_Default::filter(CephContext *cct, | |
3749 | const rgw_obj_key& source_key, | |
3750 | const RGWBucketInfo& dest_bucket_info, | |
3751 | std::optional<rgw_placement_rule> dest_placement_rule, | |
3752 | const map<string, bufferlist>& obj_attrs, | |
3753 | std::optional<rgw_user> *poverride_owner, | |
3754 | const rgw_placement_rule **prule) | |
3755 | { | |
3756 | const rgw_placement_rule *ptail_rule = (dest_placement_rule ? &(*dest_placement_rule) : nullptr); | |
3757 | if (!ptail_rule) { | |
3758 | auto iter = obj_attrs.find(RGW_ATTR_STORAGE_CLASS); | |
3759 | if (iter != obj_attrs.end()) { | |
3760 | dest_rule.storage_class = iter->second.to_str(); | |
3761 | dest_rule.inherit_from(dest_bucket_info.placement_rule); | |
3762 | ptail_rule = &dest_rule; | |
3763 | } else { | |
3764 | ptail_rule = &dest_bucket_info.placement_rule; | |
3765 | } | |
3766 | } | |
3767 | *prule = ptail_rule; | |
3768 | return 0; | |
3769 | } | |
3770 | ||
3771 | int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx, | |
3772 | const rgw_user& user_id, | |
3773 | req_info *info, | |
3774 | const rgw_zone_id& source_zone, | |
3775 | rgw::sal::RGWObject* dest_obj, | |
3776 | rgw::sal::RGWObject* src_obj, | |
3777 | rgw::sal::RGWBucket* dest_bucket, | |
3778 | rgw::sal::RGWBucket* src_bucket, | |
3779 | std::optional<rgw_placement_rule> dest_placement_rule, | |
3780 | real_time *src_mtime, | |
3781 | real_time *mtime, | |
3782 | const real_time *mod_ptr, | |
3783 | const real_time *unmod_ptr, | |
3784 | bool high_precision_time, | |
3785 | const char *if_match, | |
3786 | const char *if_nomatch, | |
3787 | AttrsMod attrs_mod, | |
3788 | bool copy_if_newer, | |
3789 | rgw::sal::RGWAttrs& attrs, | |
3790 | RGWObjCategory category, | |
3791 | std::optional<uint64_t> olh_epoch, | |
3792 | real_time delete_at, | |
3793 | string *ptag, | |
3794 | string *petag, | |
3795 | void (*progress_cb)(off_t, void *), | |
3796 | void *progress_data, | |
3797 | const DoutPrefixProvider *dpp, | |
3798 | RGWFetchObjFilter *filter, | |
3799 | rgw_zone_set *zones_trace, | |
3800 | std::optional<uint64_t>* bytes_transferred) | |
3801 | { | |
3802 | /* source is in a different zonegroup, copy from there */ | |
3803 | ||
3804 | RGWRESTStreamRWRequest *in_stream_req; | |
3805 | string tag; | |
3806 | int i; | |
3807 | append_rand_alpha(cct, tag, tag, 32); | |
3808 | obj_time_weight set_mtime_weight; | |
3809 | set_mtime_weight.high_precision = high_precision_time; | |
3810 | int ret; | |
3811 | ||
3812 | rgw::BlockingAioThrottle aio(cct->_conf->rgw_put_obj_min_window_size); | |
3813 | using namespace rgw::putobj; | |
3814 | AtomicObjectProcessor processor(&aio, this->store, dest_bucket, nullptr, user_id, | |
3815 | obj_ctx, dest_obj->clone(), olh_epoch, | |
3816 | tag, dpp, null_yield); | |
3817 | RGWRESTConn *conn; | |
3818 | auto& zone_conn_map = svc.zone->get_zone_conn_map(); | |
3819 | auto& zonegroup_conn_map = svc.zone->get_zonegroup_conn_map(); | |
3820 | if (source_zone.empty()) { | |
3821 | if (!src_bucket || src_bucket->get_info().zonegroup.empty()) { | |
3822 | /* source is in the master zonegroup */ | |
3823 | conn = svc.zone->get_master_conn(); | |
3824 | } else { | |
3825 | map<string, RGWRESTConn *>::iterator iter = zonegroup_conn_map.find(src_bucket->get_info().zonegroup); | |
3826 | if (iter == zonegroup_conn_map.end()) { | |
3827 | ldpp_dout(dpp, 0) << "could not find zonegroup connection to zonegroup: " << source_zone << dendl; | |
3828 | return -ENOENT; | |
3829 | } | |
3830 | conn = iter->second; | |
3831 | } | |
3832 | } else { | |
3833 | auto iter = zone_conn_map.find(source_zone); | |
3834 | if (iter == zone_conn_map.end()) { | |
3835 | ldpp_dout(dpp, 0) << "could not find zone connection to zone: " << source_zone << dendl; | |
3836 | return -ENOENT; | |
3837 | } | |
3838 | conn = iter->second; | |
3839 | } | |
3840 | ||
3841 | boost::optional<RGWPutObj_Compress> compressor; | |
3842 | CompressorRef plugin; | |
3843 | ||
3844 | RGWFetchObjFilter_Default source_filter; | |
3845 | if (!filter) { | |
3846 | filter = &source_filter; | |
3847 | } | |
3848 | ||
3849 | std::optional<rgw_user> override_owner; | |
3850 | ||
3851 | RGWRadosPutObj cb(dpp, cct, plugin, compressor, &processor, progress_cb, progress_data, | |
3852 | [&](map<string, bufferlist>& obj_attrs) { | |
3853 | const rgw_placement_rule *ptail_rule; | |
3854 | ||
3855 | int ret = filter->filter(cct, | |
3856 | src_obj->get_key(), | |
3857 | dest_bucket->get_info(), | |
3858 | dest_placement_rule, | |
3859 | obj_attrs, | |
3860 | &override_owner, | |
3861 | &ptail_rule); | |
3862 | if (ret < 0) { | |
3863 | ldpp_dout(dpp, 5) << "Aborting fetch: source object filter returned ret=" << ret << dendl; | |
3864 | return ret; | |
3865 | } | |
3866 | ||
3867 | processor.set_tail_placement(*ptail_rule); | |
3868 | ||
3869 | const auto& compression_type = svc.zone->get_zone_params().get_compression_type(*ptail_rule); | |
3870 | if (compression_type != "none") { | |
3871 | plugin = Compressor::create(cct, compression_type); | |
3872 | if (!plugin) { | |
3873 | ldpp_dout(dpp, 1) << "Cannot load plugin for compression type " | |
3874 | << compression_type << dendl; | |
3875 | } | |
3876 | } | |
3877 | ||
3878 | ret = processor.prepare(null_yield); | |
3879 | if (ret < 0) { | |
3880 | return ret; | |
3881 | } | |
3882 | return 0; | |
3883 | }); | |
3884 | ||
3885 | string etag; | |
3886 | real_time set_mtime; | |
3887 | uint64_t expected_size = 0; | |
3888 | ||
3889 | RGWObjState *dest_state = NULL; | |
3890 | ||
3891 | const real_time *pmod = mod_ptr; | |
3892 | ||
3893 | obj_time_weight dest_mtime_weight; | |
3894 | ||
3895 | if (copy_if_newer) { | |
3896 | /* need to get mtime for destination */ | |
3897 | ret = get_obj_state(dpp, &obj_ctx, dest_bucket->get_info(), dest_obj->get_obj(), &dest_state, false, null_yield); | |
3898 | if (ret < 0) | |
3899 | goto set_err_state; | |
3900 | ||
3901 | if (!real_clock::is_zero(dest_state->mtime)) { | |
3902 | dest_mtime_weight.init(dest_state); | |
3903 | pmod = &dest_mtime_weight.mtime; | |
3904 | } | |
3905 | } | |
3906 | ||
3907 | static constexpr bool prepend_meta = true; | |
3908 | static constexpr bool get_op = true; | |
3909 | static constexpr bool rgwx_stat = false; | |
3910 | static constexpr bool sync_manifest = true; | |
3911 | static constexpr bool skip_decrypt = true; | |
3912 | ret = conn->get_obj(dpp, user_id, info, src_obj, pmod, unmod_ptr, | |
3913 | dest_mtime_weight.zone_short_id, dest_mtime_weight.pg_ver, | |
3914 | prepend_meta, get_op, rgwx_stat, | |
3915 | sync_manifest, skip_decrypt, | |
3916 | true, | |
3917 | &cb, &in_stream_req); | |
3918 | if (ret < 0) { | |
3919 | goto set_err_state; | |
3920 | } | |
3921 | ||
3922 | ret = conn->complete_request(in_stream_req, &etag, &set_mtime, | |
3923 | &expected_size, nullptr, nullptr, null_yield); | |
3924 | if (ret < 0) { | |
3925 | goto set_err_state; | |
3926 | } | |
3927 | ret = cb.flush(); | |
3928 | if (ret < 0) { | |
3929 | goto set_err_state; | |
3930 | } | |
3931 | if (cb.get_data_len() != expected_size) { | |
3932 | ret = -EIO; | |
3933 | ldpp_dout(dpp, 0) << "ERROR: object truncated during fetching, expected " | |
3934 | << expected_size << " bytes but received " << cb.get_data_len() << dendl; | |
3935 | goto set_err_state; | |
3936 | } | |
3937 | if (compressor && compressor->is_compressed()) { | |
3938 | bufferlist tmp; | |
3939 | RGWCompressionInfo cs_info; | |
3940 | cs_info.compression_type = plugin->get_type_name(); | |
3941 | cs_info.orig_size = cb.get_data_len(); | |
3942 | cs_info.compressor_message = compressor->get_compressor_message(); | |
3943 | cs_info.blocks = move(compressor->get_compression_blocks()); | |
3944 | encode(cs_info, tmp); | |
3945 | cb.get_attrs()[RGW_ATTR_COMPRESSION] = tmp; | |
3946 | } | |
3947 | ||
3948 | if (override_owner) { | |
3949 | processor.set_owner(*override_owner); | |
3950 | ||
3951 | auto& obj_attrs = cb.get_attrs(); | |
3952 | ||
3953 | RGWUserInfo owner_info; | |
3954 | if (ctl.user->get_info_by_uid(dpp, *override_owner, &owner_info, null_yield) < 0) { | |
3955 | ldpp_dout(dpp, 10) << "owner info does not exist" << dendl; | |
3956 | return -EINVAL; | |
3957 | } | |
3958 | ||
3959 | RGWAccessControlPolicy acl; | |
3960 | ||
3961 | auto aiter = obj_attrs.find(RGW_ATTR_ACL); | |
3962 | if (aiter == obj_attrs.end()) { | |
3963 | ldpp_dout(dpp, 0) << "WARNING: " << __func__ << "(): object doesn't have ACL attribute, setting default ACLs" << dendl; | |
3964 | acl.create_default(owner_info.user_id, owner_info.display_name); | |
3965 | } else { | |
3966 | auto iter = aiter->second.cbegin(); | |
3967 | try { | |
3968 | acl.decode(iter); | |
3969 | } catch (buffer::error& err) { | |
3970 | ldpp_dout(dpp, 0) << "ERROR: " << __func__ << "(): could not decode policy, caught buffer::error" << dendl; | |
3971 | return -EIO; | |
3972 | } | |
3973 | } | |
3974 | ||
3975 | ACLOwner new_owner; | |
3976 | new_owner.set_id(*override_owner); | |
3977 | new_owner.set_name(owner_info.display_name); | |
3978 | ||
3979 | acl.set_owner(new_owner); | |
3980 | ||
3981 | bufferlist bl; | |
3982 | acl.encode(bl); | |
3983 | obj_attrs[RGW_ATTR_ACL] = std::move(bl); | |
3984 | } | |
3985 | ||
3986 | if (source_zone.empty()) { /* need to preserve expiration if copy in the same zonegroup */ | |
3987 | cb.get_attrs().erase(RGW_ATTR_DELETE_AT); | |
3988 | } else { | |
3989 | map<string, bufferlist>::iterator iter = cb.get_attrs().find(RGW_ATTR_DELETE_AT); | |
3990 | if (iter != cb.get_attrs().end()) { | |
3991 | try { | |
3992 | decode(delete_at, iter->second); | |
3993 | } catch (buffer::error& err) { | |
3994 | ldpp_dout(dpp, 0) << "ERROR: failed to decode delete_at field in intra zone copy" << dendl; | |
3995 | } | |
3996 | } | |
3997 | } | |
3998 | ||
3999 | if (src_mtime) { | |
4000 | *src_mtime = set_mtime; | |
4001 | } | |
4002 | ||
4003 | if (petag) { | |
4004 | const auto iter = cb.get_attrs().find(RGW_ATTR_ETAG); | |
4005 | if (iter != cb.get_attrs().end()) { | |
4006 | *petag = iter->second.to_str(); | |
4007 | } | |
4008 | } | |
4009 | ||
4010 | //erase the append attr | |
4011 | cb.get_attrs().erase(RGW_ATTR_APPEND_PART_NUM); | |
4012 | ||
4013 | if (source_zone.empty()) { | |
4014 | set_copy_attrs(cb.get_attrs(), attrs, attrs_mod); | |
4015 | } else { | |
4016 | attrs = cb.get_attrs(); | |
4017 | } | |
4018 | ||
4019 | if (copy_if_newer) { | |
4020 | uint64_t pg_ver = 0; | |
4021 | auto i = attrs.find(RGW_ATTR_PG_VER); | |
4022 | if (i != attrs.end() && i->second.length() > 0) { | |
4023 | auto iter = i->second.cbegin(); | |
4024 | try { | |
4025 | decode(pg_ver, iter); | |
4026 | } catch (buffer::error& err) { | |
4027 | ldpp_dout(dpp, 0) << "ERROR: failed to decode pg ver attribute, ignoring" << dendl; | |
4028 | /* non critical error */ | |
4029 | } | |
4030 | } | |
4031 | set_mtime_weight.init(set_mtime, svc.zone->get_zone_short_id(), pg_ver); | |
4032 | } | |
4033 | ||
4034 | /* Perform ETag verification is we have computed the object's MD5 sum at our end */ | |
4035 | if (const auto& verifier_etag = cb.get_verifier_etag(); | |
4036 | !verifier_etag.empty()) { | |
4037 | string trimmed_etag = etag; | |
4038 | ||
4039 | /* Remove the leading and trailing double quotes from etag */ | |
4040 | trimmed_etag.erase(std::remove(trimmed_etag.begin(), trimmed_etag.end(),'\"'), | |
4041 | trimmed_etag.end()); | |
4042 | ||
4043 | if (verifier_etag != trimmed_etag) { | |
4044 | ret = -EIO; | |
4045 | ldpp_dout(dpp, 0) << "ERROR: source and destination objects don't match. Expected etag:" | |
4046 | << trimmed_etag << " Computed etag:" << verifier_etag << dendl; | |
4047 | goto set_err_state; | |
4048 | } | |
4049 | } | |
4050 | ||
4051 | #define MAX_COMPLETE_RETRY 100 | |
4052 | for (i = 0; i < MAX_COMPLETE_RETRY; i++) { | |
4053 | bool canceled = false; | |
4054 | ret = processor.complete(cb.get_data_len(), etag, mtime, set_mtime, | |
4055 | attrs, delete_at, nullptr, nullptr, nullptr, | |
4056 | zones_trace, &canceled, null_yield); | |
4057 | if (ret < 0) { | |
4058 | goto set_err_state; | |
4059 | } | |
4060 | ||
4061 | if (copy_if_newer && canceled) { | |
4062 | ldpp_dout(dpp, 20) << "raced with another write of obj: " << dest_obj << dendl; | |
4063 | obj_ctx.invalidate(dest_obj->get_obj()); /* object was overwritten */ | |
4064 | ret = get_obj_state(dpp, &obj_ctx, dest_bucket->get_info(), dest_obj->get_obj(), &dest_state, false, null_yield); | |
4065 | if (ret < 0) { | |
4066 | ldpp_dout(dpp, 0) << "ERROR: " << __func__ << ": get_err_state() returned ret=" << ret << dendl; | |
4067 | goto set_err_state; | |
4068 | } | |
4069 | dest_mtime_weight.init(dest_state); | |
4070 | dest_mtime_weight.high_precision = high_precision_time; | |
4071 | if (!dest_state->exists || | |
4072 | dest_mtime_weight < set_mtime_weight) { | |
4073 | ldpp_dout(dpp, 20) << "retrying writing object mtime=" << set_mtime << " dest_state->mtime=" << dest_state->mtime << " dest_state->exists=" << dest_state->exists << dendl; | |
4074 | continue; | |
4075 | } else { | |
4076 | ldpp_dout(dpp, 20) << "not retrying writing object mtime=" << set_mtime << " dest_state->mtime=" << dest_state->mtime << " dest_state->exists=" << dest_state->exists << dendl; | |
4077 | } | |
4078 | } | |
4079 | break; | |
4080 | } | |
4081 | ||
4082 | if (i == MAX_COMPLETE_RETRY) { | |
4083 | ldpp_dout(dpp, 0) << "ERROR: retried object completion too many times, something is wrong!" << dendl; | |
4084 | ret = -EIO; | |
4085 | goto set_err_state; | |
4086 | } | |
4087 | ||
4088 | if (bytes_transferred) { | |
4089 | *bytes_transferred = cb.get_data_len(); | |
4090 | } | |
4091 | return 0; | |
4092 | set_err_state: | |
4093 | if (copy_if_newer && ret == -ERR_NOT_MODIFIED) { | |
4094 | // we may have already fetched during sync of OP_ADD, but were waiting | |
4095 | // for OP_LINK_OLH to call set_olh() with a real olh_epoch | |
4096 | if (olh_epoch && *olh_epoch > 0) { | |
4097 | constexpr bool log_data_change = true; | |
4098 | ret = set_olh(dpp, obj_ctx, dest_bucket->get_info(), dest_obj->get_obj(), false, nullptr, | |
4099 | *olh_epoch, real_time(), false, null_yield, zones_trace, log_data_change); | |
4100 | } else { | |
4101 | // we already have the latest copy | |
4102 | ret = 0; | |
4103 | } | |
4104 | } | |
4105 | return ret; | |
4106 | } | |
4107 | ||
4108 | ||
4109 | int RGWRados::copy_obj_to_remote_dest(const DoutPrefixProvider *dpp, | |
4110 | RGWObjState *astate, | |
4111 | map<string, bufferlist>& src_attrs, | |
4112 | RGWRados::Object::Read& read_op, | |
4113 | const rgw_user& user_id, | |
4114 | rgw::sal::RGWObject* dest_obj, | |
4115 | real_time *mtime) | |
4116 | { | |
4117 | string etag; | |
4118 | ||
4119 | RGWRESTStreamS3PutObj *out_stream_req; | |
4120 | ||
4121 | auto rest_master_conn = svc.zone->get_master_conn(); | |
4122 | ||
4123 | int ret = rest_master_conn->put_obj_async(dpp, user_id, dest_obj, astate->size, src_attrs, true, &out_stream_req); | |
4124 | if (ret < 0) { | |
4125 | return ret; | |
4126 | } | |
4127 | ||
4128 | ret = read_op.iterate(dpp, 0, astate->size - 1, out_stream_req->get_out_cb(), null_yield); | |
4129 | if (ret < 0) { | |
4130 | delete out_stream_req; | |
4131 | return ret; | |
4132 | } | |
4133 | ||
4134 | ret = rest_master_conn->complete_request(out_stream_req, etag, mtime, null_yield); | |
4135 | if (ret < 0) | |
4136 | return ret; | |
4137 | ||
4138 | return 0; | |
4139 | } | |
4140 | ||
4141 | /** | |
4142 | * Copy an object. | |
4143 | * dest_obj: the object to copy into | |
4144 | * src_obj: the object to copy from | |
4145 | * attrs: usage depends on attrs_mod parameter | |
4146 | * attrs_mod: the modification mode of the attrs, may have the following values: | |
4147 | * ATTRSMOD_NONE - the attributes of the source object will be | |
4148 | * copied without modifications, attrs parameter is ignored; | |
4149 | * ATTRSMOD_REPLACE - new object will have the attributes provided by attrs | |
4150 | * parameter, source object attributes are not copied; | |
4151 | * ATTRSMOD_MERGE - any conflicting meta keys on the source object's attributes | |
4152 | * are overwritten by values contained in attrs parameter. | |
4153 | * err: stores any errors resulting from the get of the original object | |
4154 | * Returns: 0 on success, -ERR# otherwise. | |
4155 | */ | |
4156 | int RGWRados::copy_obj(RGWObjectCtx& obj_ctx, | |
4157 | const rgw_user& user_id, | |
4158 | req_info *info, | |
4159 | const rgw_zone_id& source_zone, | |
4160 | rgw::sal::RGWObject* dest_obj, | |
4161 | rgw::sal::RGWObject* src_obj, | |
4162 | rgw::sal::RGWBucket* dest_bucket, | |
4163 | rgw::sal::RGWBucket* src_bucket, | |
4164 | const rgw_placement_rule& dest_placement, | |
4165 | real_time *src_mtime, | |
4166 | real_time *mtime, | |
4167 | const real_time *mod_ptr, | |
4168 | const real_time *unmod_ptr, | |
4169 | bool high_precision_time, | |
4170 | const char *if_match, | |
4171 | const char *if_nomatch, | |
4172 | AttrsMod attrs_mod, | |
4173 | bool copy_if_newer, | |
4174 | rgw::sal::RGWAttrs& attrs, | |
4175 | RGWObjCategory category, | |
4176 | uint64_t olh_epoch, | |
4177 | real_time delete_at, | |
4178 | string *version_id, | |
4179 | string *ptag, | |
4180 | string *petag, | |
4181 | void (*progress_cb)(off_t, void *), | |
4182 | void *progress_data, | |
4183 | const DoutPrefixProvider *dpp, | |
4184 | optional_yield y) | |
4185 | { | |
4186 | int ret; | |
4187 | uint64_t obj_size; | |
4188 | rgw_obj shadow_obj = dest_obj->get_obj(); | |
4189 | string shadow_oid; | |
4190 | ||
4191 | bool remote_src; | |
4192 | bool remote_dest; | |
4193 | ||
4194 | append_rand_alpha(cct, dest_obj->get_oid(), shadow_oid, 32); | |
4195 | shadow_obj.init_ns(dest_obj->get_bucket()->get_key(), shadow_oid, shadow_ns); | |
4196 | ||
4197 | auto& zonegroup = svc.zone->get_zonegroup(); | |
4198 | ||
4199 | remote_dest = !zonegroup.equals(dest_bucket->get_info().zonegroup); | |
4200 | remote_src = !zonegroup.equals(src_bucket->get_info().zonegroup); | |
4201 | ||
4202 | if (remote_src && remote_dest) { | |
4203 | ldpp_dout(dpp, 0) << "ERROR: can't copy object when both src and dest buckets are remote" << dendl; | |
4204 | return -EINVAL; | |
4205 | } | |
4206 | ||
4207 | ldpp_dout(dpp, 5) << "Copy object " << src_obj->get_bucket() << ":" << src_obj->get_oid() << " => " << dest_obj->get_bucket() << ":" << dest_obj->get_oid() << dendl; | |
4208 | ||
4209 | if (remote_src || !source_zone.empty()) { | |
4210 | return fetch_remote_obj(obj_ctx, user_id, info, source_zone, | |
4211 | dest_obj, src_obj, dest_bucket, src_bucket, | |
4212 | dest_placement, src_mtime, mtime, mod_ptr, | |
4213 | unmod_ptr, high_precision_time, | |
4214 | if_match, if_nomatch, attrs_mod, copy_if_newer, attrs, category, | |
4215 | olh_epoch, delete_at, ptag, petag, progress_cb, progress_data, dpp, | |
4216 | nullptr /* filter */); | |
4217 | } | |
4218 | ||
4219 | map<string, bufferlist> src_attrs; | |
4220 | RGWRados::Object src_op_target(this, src_bucket->get_info(), obj_ctx, src_obj->get_obj()); | |
4221 | RGWRados::Object::Read read_op(&src_op_target); | |
4222 | ||
4223 | read_op.conds.mod_ptr = mod_ptr; | |
4224 | read_op.conds.unmod_ptr = unmod_ptr; | |
4225 | read_op.conds.high_precision_time = high_precision_time; | |
4226 | read_op.conds.if_match = if_match; | |
4227 | read_op.conds.if_nomatch = if_nomatch; | |
4228 | read_op.params.attrs = &src_attrs; | |
4229 | read_op.params.lastmod = src_mtime; | |
4230 | read_op.params.obj_size = &obj_size; | |
4231 | ||
4232 | ret = read_op.prepare(y, dpp); | |
4233 | if (ret < 0) { | |
4234 | return ret; | |
4235 | } | |
4236 | if (src_attrs.count(RGW_ATTR_CRYPT_MODE)) { | |
4237 | // Current implementation does not follow S3 spec and even | |
4238 | // may result in data corruption silently when copying | |
4239 | // multipart objects acorss pools. So reject COPY operations | |
4240 | //on encrypted objects before it is fully functional. | |
4241 | ldpp_dout(dpp, 0) << "ERROR: copy op for encrypted object " << src_obj | |
4242 | << " has not been implemented." << dendl; | |
4243 | return -ERR_NOT_IMPLEMENTED; | |
4244 | } | |
4245 | ||
4246 | src_attrs[RGW_ATTR_ACL] = attrs[RGW_ATTR_ACL]; | |
4247 | src_attrs.erase(RGW_ATTR_DELETE_AT); | |
4248 | ||
4249 | set_copy_attrs(src_attrs, attrs, attrs_mod); | |
4250 | attrs.erase(RGW_ATTR_ID_TAG); | |
4251 | attrs.erase(RGW_ATTR_PG_VER); | |
4252 | attrs.erase(RGW_ATTR_SOURCE_ZONE); | |
4253 | map<string, bufferlist>::iterator cmp = src_attrs.find(RGW_ATTR_COMPRESSION); | |
4254 | if (cmp != src_attrs.end()) | |
4255 | attrs[RGW_ATTR_COMPRESSION] = cmp->second; | |
4256 | ||
4257 | RGWObjManifest manifest; | |
4258 | RGWObjState *astate = NULL; | |
4259 | ||
4260 | ret = get_obj_state(dpp, &obj_ctx, src_bucket->get_info(), src_obj->get_obj(), &astate, y); | |
4261 | if (ret < 0) { | |
4262 | return ret; | |
4263 | } | |
4264 | ||
4265 | vector<rgw_raw_obj> ref_objs; | |
4266 | ||
4267 | if (remote_dest) { | |
4268 | /* dest is in a different zonegroup, copy it there */ | |
4269 | return copy_obj_to_remote_dest(dpp, astate, attrs, read_op, user_id, dest_obj, mtime); | |
4270 | } | |
4271 | uint64_t max_chunk_size; | |
4272 | ||
4273 | ret = get_max_chunk_size(dest_bucket->get_placement_rule(), dest_obj->get_obj(), &max_chunk_size, dpp); | |
4274 | if (ret < 0) { | |
4275 | ldpp_dout(dpp, 0) << "ERROR: failed to get max_chunk_size() for bucket " << dest_obj->get_bucket() << dendl; | |
4276 | return ret; | |
4277 | } | |
4278 | ||
4279 | rgw_pool src_pool; | |
4280 | rgw_pool dest_pool; | |
4281 | ||
4282 | const rgw_placement_rule *src_rule{nullptr}; | |
4283 | ||
4284 | if (astate->manifest) { | |
4285 | src_rule = &astate->manifest->get_tail_placement().placement_rule; | |
4286 | ldpp_dout(dpp, 20) << __func__ << "(): manifest src_rule=" << src_rule->to_str() << dendl; | |
4287 | } | |
4288 | ||
4289 | if (!src_rule || src_rule->empty()) { | |
4290 | src_rule = &src_bucket->get_placement_rule(); | |
4291 | } | |
4292 | ||
4293 | if (!get_obj_data_pool(*src_rule, src_obj->get_obj(), &src_pool)) { | |
4294 | ldpp_dout(dpp, 0) << "ERROR: failed to locate data pool for " << src_obj << dendl; | |
4295 | return -EIO; | |
4296 | } | |
4297 | ||
4298 | if (!get_obj_data_pool(dest_placement, dest_obj->get_obj(), &dest_pool)) { | |
4299 | ldpp_dout(dpp, 0) << "ERROR: failed to locate data pool for " << dest_obj << dendl; | |
4300 | return -EIO; | |
4301 | } | |
4302 | ||
4303 | ldpp_dout(dpp, 20) << __func__ << "(): src_rule=" << src_rule->to_str() << " src_pool=" << src_pool | |
4304 | << " dest_rule=" << dest_placement.to_str() << " dest_pool=" << dest_pool << dendl; | |
4305 | ||
4306 | bool copy_data = (!astate->manifest) || | |
4307 | (*src_rule != dest_placement) || | |
4308 | (src_pool != dest_pool); | |
4309 | ||
4310 | bool copy_first = false; | |
4311 | if (astate->manifest) { | |
4312 | if (!astate->manifest->has_tail()) { | |
4313 | copy_data = true; | |
4314 | } else { | |
4315 | uint64_t head_size = astate->manifest->get_head_size(); | |
4316 | ||
4317 | if (head_size > 0) { | |
4318 | if (head_size > max_chunk_size) { | |
4319 | copy_data = true; | |
4320 | } else { | |
4321 | copy_first = true; | |
4322 | } | |
4323 | } | |
4324 | } | |
4325 | } | |
4326 | ||
4327 | if (petag) { | |
4328 | const auto iter = attrs.find(RGW_ATTR_ETAG); | |
4329 | if (iter != attrs.end()) { | |
4330 | *petag = iter->second.to_str(); | |
4331 | } | |
4332 | } | |
4333 | ||
4334 | if (copy_data) { /* refcounting tail wouldn't work here, just copy the data */ | |
4335 | attrs.erase(RGW_ATTR_TAIL_TAG); | |
4336 | return copy_obj_data(obj_ctx, dest_bucket, dest_placement, read_op, obj_size - 1, dest_obj, | |
4337 | mtime, real_time(), attrs, olh_epoch, delete_at, petag, dpp, y); | |
4338 | } | |
4339 | ||
4340 | RGWObjManifest::obj_iterator miter = astate->manifest->obj_begin(dpp); | |
4341 | ||
4342 | if (copy_first) { // we need to copy first chunk, not increase refcount | |
4343 | ++miter; | |
4344 | } | |
4345 | ||
4346 | rgw_rados_ref ref; | |
4347 | ret = get_raw_obj_ref(dpp, miter.get_location().get_raw_obj(store), &ref); | |
4348 | if (ret < 0) { | |
4349 | return ret; | |
4350 | } | |
4351 | ||
4352 | bufferlist first_chunk; | |
4353 | ||
4354 | bool copy_itself = (dest_obj == src_obj); | |
4355 | RGWObjManifest *pmanifest; | |
4356 | ldpp_dout(dpp, 20) << "dest_obj=" << dest_obj << " src_obj=" << src_obj << " copy_itself=" << (int)copy_itself << dendl; | |
4357 | ||
4358 | RGWRados::Object dest_op_target(this, dest_bucket->get_info(), obj_ctx, dest_obj->get_obj()); | |
4359 | RGWRados::Object::Write write_op(&dest_op_target); | |
4360 | ||
4361 | string tag; | |
4362 | ||
4363 | if (ptag) { | |
4364 | tag = *ptag; | |
4365 | } | |
4366 | ||
4367 | if (tag.empty()) { | |
4368 | append_rand_alpha(cct, tag, tag, 32); | |
4369 | } | |
4370 | ||
4371 | if (!copy_itself) { | |
4372 | attrs.erase(RGW_ATTR_TAIL_TAG); | |
4373 | manifest = *astate->manifest; | |
4374 | const rgw_bucket_placement& tail_placement = manifest.get_tail_placement(); | |
4375 | if (tail_placement.bucket.name.empty()) { | |
4376 | manifest.set_tail_placement(tail_placement.placement_rule, src_obj->get_bucket()->get_key()); | |
4377 | } | |
4378 | string ref_tag; | |
4379 | for (; miter != astate->manifest->obj_end(dpp); ++miter) { | |
4380 | ObjectWriteOperation op; | |
4381 | ref_tag = tag + '\0'; | |
4382 | cls_refcount_get(op, ref_tag, true); | |
4383 | const rgw_raw_obj& loc = miter.get_location().get_raw_obj(store); | |
4384 | ||
4385 | auto& ioctx = ref.pool.ioctx(); | |
4386 | ioctx.locator_set_key(loc.loc); | |
4387 | ||
4388 | ret = rgw_rados_operate(dpp, ioctx, loc.oid, &op, null_yield); | |
4389 | if (ret < 0) { | |
4390 | goto done_ret; | |
4391 | } | |
4392 | ||
4393 | ref_objs.push_back(loc); | |
4394 | } | |
4395 | ||
4396 | pmanifest = &manifest; | |
4397 | } else { | |
4398 | pmanifest = &(*astate->manifest); | |
4399 | /* don't send the object's tail for garbage collection */ | |
4400 | astate->keep_tail = true; | |
4401 | } | |
4402 | ||
4403 | if (copy_first) { | |
4404 | ret = read_op.read(0, max_chunk_size, first_chunk, y, dpp); | |
4405 | if (ret < 0) { | |
4406 | goto done_ret; | |
4407 | } | |
4408 | ||
4409 | pmanifest->set_head(dest_bucket->get_placement_rule(), dest_obj->get_obj(), first_chunk.length()); | |
4410 | } else { | |
4411 | pmanifest->set_head(dest_bucket->get_placement_rule(), dest_obj->get_obj(), 0); | |
4412 | } | |
4413 | ||
4414 | write_op.meta.data = &first_chunk; | |
4415 | write_op.meta.manifest = pmanifest; | |
4416 | write_op.meta.ptag = &tag; | |
4417 | write_op.meta.owner = dest_bucket->get_info().owner; | |
4418 | write_op.meta.mtime = mtime; | |
4419 | write_op.meta.flags = PUT_OBJ_CREATE; | |
4420 | write_op.meta.category = category; | |
4421 | write_op.meta.olh_epoch = olh_epoch; | |
4422 | write_op.meta.delete_at = delete_at; | |
4423 | write_op.meta.modify_tail = !copy_itself; | |
4424 | ||
4425 | ret = write_op.write_meta(dpp, obj_size, astate->accounted_size, attrs, y); | |
4426 | if (ret < 0) { | |
4427 | goto done_ret; | |
4428 | } | |
4429 | ||
4430 | return 0; | |
4431 | ||
4432 | done_ret: | |
4433 | if (!copy_itself) { | |
4434 | vector<rgw_raw_obj>::iterator riter; | |
4435 | ||
4436 | /* rollback reference */ | |
4437 | string ref_tag = tag + '\0'; | |
4438 | for (riter = ref_objs.begin(); riter != ref_objs.end(); ++riter) { | |
4439 | ObjectWriteOperation op; | |
4440 | cls_refcount_put(op, ref_tag, true); | |
4441 | ||
4442 | ref.pool.ioctx().locator_set_key(riter->loc); | |
4443 | ||
4444 | int r = rgw_rados_operate(dpp, ref.pool.ioctx(), riter->oid, &op, null_yield); | |
4445 | if (r < 0) { | |
4446 | ldpp_dout(dpp, 0) << "ERROR: cleanup after error failed to drop reference on obj=" << *riter << dendl; | |
4447 | } | |
4448 | } | |
4449 | } | |
4450 | return ret; | |
4451 | } | |
4452 | ||
4453 | ||
4454 | int RGWRados::copy_obj_data(RGWObjectCtx& obj_ctx, | |
4455 | rgw::sal::RGWBucket* bucket, | |
4456 | const rgw_placement_rule& dest_placement, | |
4457 | RGWRados::Object::Read& read_op, off_t end, | |
4458 | rgw::sal::RGWObject* dest_obj, | |
4459 | real_time *mtime, | |
4460 | real_time set_mtime, | |
4461 | rgw::sal::RGWAttrs& attrs, | |
4462 | uint64_t olh_epoch, | |
4463 | real_time delete_at, | |
4464 | string *petag, | |
4465 | const DoutPrefixProvider *dpp, | |
4466 | optional_yield y) | |
4467 | { | |
4468 | string tag; | |
4469 | append_rand_alpha(cct, tag, tag, 32); | |
4470 | ||
4471 | rgw::BlockingAioThrottle aio(cct->_conf->rgw_put_obj_min_window_size); | |
4472 | using namespace rgw::putobj; | |
4473 | // do not change the null_yield in the initialization of this AtomicObjectProcessor | |
4474 | // it causes crashes in the ragweed tests | |
4475 | AtomicObjectProcessor processor(&aio, this->store, bucket, &dest_placement, | |
4476 | bucket->get_info().owner, obj_ctx, | |
4477 | dest_obj->clone(), olh_epoch, tag, | |
4478 | dpp, null_yield); | |
4479 | int ret = processor.prepare(y); | |
4480 | if (ret < 0) | |
4481 | return ret; | |
4482 | ||
4483 | off_t ofs = 0; | |
4484 | ||
4485 | do { | |
4486 | bufferlist bl; | |
4487 | ret = read_op.read(ofs, end, bl, y, dpp); | |
4488 | if (ret < 0) { | |
4489 | ldpp_dout(dpp, 0) << "ERROR: fail to read object data, ret = " << ret << dendl; | |
4490 | return ret; | |
4491 | } | |
4492 | ||
4493 | uint64_t read_len = ret; | |
4494 | ret = processor.process(std::move(bl), ofs); | |
4495 | if (ret < 0) { | |
4496 | return ret; | |
4497 | } | |
4498 | ||
4499 | ofs += read_len; | |
4500 | } while (ofs <= end); | |
4501 | ||
4502 | // flush | |
4503 | ret = processor.process({}, ofs); | |
4504 | if (ret < 0) { | |
4505 | return ret; | |
4506 | } | |
4507 | ||
4508 | string etag; | |
4509 | auto iter = attrs.find(RGW_ATTR_ETAG); | |
4510 | if (iter != attrs.end()) { | |
4511 | bufferlist& bl = iter->second; | |
4512 | etag = bl.to_str(); | |
4513 | if (petag) { | |
4514 | *petag = etag; | |
4515 | } | |
4516 | } | |
4517 | ||
4518 | uint64_t accounted_size; | |
4519 | { | |
4520 | bool compressed{false}; | |
4521 | RGWCompressionInfo cs_info; | |
4522 | ret = rgw_compression_info_from_attrset(attrs, compressed, cs_info); | |
4523 | if (ret < 0) { | |
4524 | ldpp_dout(dpp, 0) << "ERROR: failed to read compression info" << dendl; | |
4525 | return ret; | |
4526 | } | |
4527 | // pass original size if compressed | |
4528 | accounted_size = compressed ? cs_info.orig_size : ofs; | |
4529 | } | |
4530 | ||
4531 | return processor.complete(accounted_size, etag, mtime, set_mtime, attrs, delete_at, | |
4532 | nullptr, nullptr, nullptr, nullptr, nullptr, y); | |
4533 | } | |
4534 | ||
4535 | int RGWRados::transition_obj(RGWObjectCtx& obj_ctx, | |
4536 | rgw::sal::RGWBucket* bucket, | |
4537 | rgw::sal::RGWObject& obj, | |
4538 | const rgw_placement_rule& placement_rule, | |
4539 | const real_time& mtime, | |
4540 | uint64_t olh_epoch, | |
4541 | const DoutPrefixProvider *dpp, | |
4542 | optional_yield y) | |
4543 | { | |
4544 | rgw::sal::RGWAttrs attrs; | |
4545 | real_time read_mtime; | |
4546 | uint64_t obj_size; | |
4547 | ||
4548 | obj.set_atomic(&obj_ctx); | |
4549 | RGWRados::Object op_target(this, bucket->get_info(), obj_ctx, obj.get_obj()); | |
4550 | RGWRados::Object::Read read_op(&op_target); | |
4551 | ||
4552 | read_op.params.attrs = &attrs; | |
4553 | read_op.params.lastmod = &read_mtime; | |
4554 | read_op.params.obj_size = &obj_size; | |
4555 | ||
4556 | int ret = read_op.prepare(y, dpp); | |
4557 | if (ret < 0) { | |
4558 | return ret; | |
4559 | } | |
4560 | ||
4561 | if (read_mtime != mtime) { | |
4562 | /* raced */ | |
4563 | return -ECANCELED; | |
4564 | } | |
4565 | ||
4566 | attrs.erase(RGW_ATTR_ID_TAG); | |
4567 | attrs.erase(RGW_ATTR_TAIL_TAG); | |
4568 | ||
4569 | ret = copy_obj_data(obj_ctx, | |
4570 | bucket, | |
4571 | placement_rule, | |
4572 | read_op, | |
4573 | obj_size - 1, | |
4574 | &obj, | |
4575 | nullptr /* pmtime */, | |
4576 | mtime, | |
4577 | attrs, | |
4578 | olh_epoch, | |
4579 | real_time(), | |
4580 | nullptr /* petag */, | |
4581 | dpp, | |
4582 | y); | |
4583 | if (ret < 0) { | |
4584 | return ret; | |
4585 | } | |
4586 | ||
4587 | return 0; | |
4588 | } | |
4589 | ||
4590 | int RGWRados::check_bucket_empty(const DoutPrefixProvider *dpp, RGWBucketInfo& bucket_info, optional_yield y) | |
4591 | { | |
4592 | constexpr uint NUM_ENTRIES = 1000u; | |
4593 | ||
4594 | rgw_obj_index_key marker; | |
4595 | string prefix; | |
4596 | bool is_truncated; | |
4597 | ||
4598 | do { | |
4599 | std::vector<rgw_bucket_dir_entry> ent_list; | |
4600 | ent_list.reserve(NUM_ENTRIES); | |
4601 | ||
4602 | int r = cls_bucket_list_unordered(dpp, | |
4603 | bucket_info, | |
4604 | RGW_NO_SHARD, | |
4605 | marker, | |
4606 | prefix, | |
4607 | NUM_ENTRIES, | |
4608 | true, | |
4609 | ent_list, | |
4610 | &is_truncated, | |
4611 | &marker, | |
4612 | y); | |
4613 | if (r < 0) { | |
4614 | return r; | |
4615 | } | |
4616 | ||
4617 | string ns; | |
4618 | for (auto const& dirent : ent_list) { | |
4619 | rgw_obj_key obj; | |
4620 | ||
4621 | if (rgw_obj_key::oid_to_key_in_ns(dirent.key.name, &obj, ns)) { | |
4622 | return -ENOTEMPTY; | |
4623 | } | |
4624 | } | |
4625 | } while (is_truncated); | |
4626 | ||
4627 | return 0; | |
4628 | } | |
4629 | ||
4630 | /** | |
4631 | * Delete a bucket. | |
4632 | * bucket: the name of the bucket to delete | |
4633 | * Returns 0 on success, -ERR# otherwise. | |
4634 | */ | |
4635 | int RGWRados::delete_bucket(RGWBucketInfo& bucket_info, RGWObjVersionTracker& objv_tracker, optional_yield y, const DoutPrefixProvider *dpp, bool check_empty) | |
4636 | { | |
4637 | const rgw_bucket& bucket = bucket_info.bucket; | |
4638 | RGWSI_RADOS::Pool index_pool; | |
4639 | map<int, string> bucket_objs; | |
4640 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, std::nullopt, &index_pool, &bucket_objs, nullptr); | |
4641 | if (r < 0) | |
4642 | return r; | |
4643 | ||
4644 | if (check_empty) { | |
4645 | r = check_bucket_empty(dpp, bucket_info, y); | |
4646 | if (r < 0) { | |
4647 | return r; | |
4648 | } | |
4649 | } | |
4650 | ||
4651 | bool remove_ep = true; | |
4652 | ||
4653 | if (objv_tracker.read_version.empty()) { | |
4654 | RGWBucketEntryPoint ep; | |
4655 | r = ctl.bucket->read_bucket_entrypoint_info(bucket_info.bucket, | |
4656 | &ep, | |
4657 | null_yield, | |
4658 | dpp, | |
4659 | RGWBucketCtl::Bucket::GetParams() | |
4660 | .set_objv_tracker(&objv_tracker)); | |
4661 | if (r < 0 || | |
4662 | (!bucket_info.bucket.bucket_id.empty() && | |
4663 | ep.bucket.bucket_id != bucket_info.bucket.bucket_id)) { | |
4664 | if (r != -ENOENT) { | |
4665 | ldpp_dout(dpp, 0) << "ERROR: read_bucket_entrypoint_info() bucket=" << bucket_info.bucket << " returned error: r=" << r << dendl; | |
4666 | /* we have no idea what caused the error, will not try to remove it */ | |
4667 | } | |
4668 | /* | |
4669 | * either failed to read bucket entrypoint, or it points to a different bucket instance than | |
4670 | * requested | |
4671 | */ | |
4672 | remove_ep = false; | |
4673 | } | |
4674 | } | |
4675 | ||
4676 | if (remove_ep) { | |
4677 | r = ctl.bucket->remove_bucket_entrypoint_info(bucket_info.bucket, null_yield, dpp, | |
4678 | RGWBucketCtl::Bucket::RemoveParams() | |
4679 | .set_objv_tracker(&objv_tracker)); | |
4680 | if (r < 0) | |
4681 | return r; | |
4682 | } | |
4683 | ||
4684 | /* if the bucket is not synced we can remove the meta file */ | |
4685 | if (!svc.zone->is_syncing_bucket_meta(bucket)) { | |
4686 | RGWObjVersionTracker objv_tracker; | |
4687 | r = ctl.bucket->remove_bucket_instance_info(bucket, bucket_info, null_yield, dpp); | |
4688 | if (r < 0) { | |
4689 | return r; | |
4690 | } | |
4691 | ||
4692 | /* remove bucket index objects asynchronously by best effort */ | |
4693 | (void) CLSRGWIssueBucketIndexClean(index_pool.ioctx(), | |
4694 | bucket_objs, | |
4695 | cct->_conf->rgw_bucket_index_max_aio)(); | |
4696 | } | |
4697 | ||
4698 | return 0; | |
4699 | } | |
4700 | ||
4701 | int RGWRados::set_bucket_owner(rgw_bucket& bucket, ACLOwner& owner, const DoutPrefixProvider *dpp) | |
4702 | { | |
4703 | RGWBucketInfo info; | |
4704 | map<string, bufferlist> attrs; | |
4705 | int r; | |
4706 | auto obj_ctx = svc.sysobj->init_obj_ctx(); | |
4707 | ||
4708 | if (bucket.bucket_id.empty()) { | |
4709 | r = get_bucket_info(&svc, bucket.tenant, bucket.name, info, NULL, null_yield, dpp, &attrs); | |
4710 | } else { | |
4711 | r = get_bucket_instance_info(obj_ctx, bucket, info, nullptr, &attrs, null_yield, dpp); | |
4712 | } | |
4713 | if (r < 0) { | |
4714 | ldpp_dout(dpp, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl; | |
4715 | return r; | |
4716 | } | |
4717 | ||
4718 | info.owner = owner.get_id(); | |
4719 | ||
4720 | r = put_bucket_instance_info(info, false, real_time(), &attrs, dpp); | |
4721 | if (r < 0) { | |
4722 | ldpp_dout(dpp, 0) << "NOTICE: put_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl; | |
4723 | return r; | |
4724 | } | |
4725 | ||
4726 | return 0; | |
4727 | } | |
4728 | ||
4729 | ||
4730 | int RGWRados::set_buckets_enabled(vector<rgw_bucket>& buckets, bool enabled, const DoutPrefixProvider *dpp) | |
4731 | { | |
4732 | int ret = 0; | |
4733 | ||
4734 | vector<rgw_bucket>::iterator iter; | |
4735 | ||
4736 | for (iter = buckets.begin(); iter != buckets.end(); ++iter) { | |
4737 | rgw_bucket& bucket = *iter; | |
4738 | if (enabled) { | |
4739 | ldpp_dout(dpp, 20) << "enabling bucket name=" << bucket.name << dendl; | |
4740 | } else { | |
4741 | ldpp_dout(dpp, 20) << "disabling bucket name=" << bucket.name << dendl; | |
4742 | } | |
4743 | ||
4744 | RGWBucketInfo info; | |
4745 | map<string, bufferlist> attrs; | |
4746 | int r = get_bucket_info(&svc, bucket.tenant, bucket.name, info, NULL, null_yield, dpp, &attrs); | |
4747 | if (r < 0) { | |
4748 | ldpp_dout(dpp, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl; | |
4749 | ret = r; | |
4750 | continue; | |
4751 | } | |
4752 | if (enabled) { | |
4753 | info.flags &= ~BUCKET_SUSPENDED; | |
4754 | } else { | |
4755 | info.flags |= BUCKET_SUSPENDED; | |
4756 | } | |
4757 | ||
4758 | r = put_bucket_instance_info(info, false, real_time(), &attrs, dpp); | |
4759 | if (r < 0) { | |
4760 | ldpp_dout(dpp, 0) << "NOTICE: put_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl; | |
4761 | ret = r; | |
4762 | continue; | |
4763 | } | |
4764 | } | |
4765 | return ret; | |
4766 | } | |
4767 | ||
4768 | int RGWRados::bucket_suspended(const DoutPrefixProvider *dpp, rgw_bucket& bucket, bool *suspended) | |
4769 | { | |
4770 | RGWBucketInfo bucket_info; | |
4771 | int ret = get_bucket_info(&svc, bucket.tenant, bucket.name, bucket_info, NULL, null_yield, dpp); | |
4772 | if (ret < 0) { | |
4773 | return ret; | |
4774 | } | |
4775 | ||
4776 | *suspended = ((bucket_info.flags & BUCKET_SUSPENDED) != 0); | |
4777 | return 0; | |
4778 | } | |
4779 | ||
4780 | int RGWRados::Object::complete_atomic_modification(const DoutPrefixProvider *dpp) | |
4781 | { | |
4782 | if ((!state->manifest)|| state->keep_tail) | |
4783 | return 0; | |
4784 | ||
4785 | cls_rgw_obj_chain chain; | |
4786 | store->update_gc_chain(dpp, obj, *state->manifest, &chain); | |
4787 | ||
4788 | if (chain.empty()) { | |
4789 | return 0; | |
4790 | } | |
4791 | ||
4792 | string tag = (state->tail_tag.length() > 0 ? state->tail_tag.to_str() : state->obj_tag.to_str()); | |
4793 | auto ret = store->gc->send_chain(chain, tag); // do it synchronously | |
4794 | if (ret < 0) { | |
4795 | //Delete objects inline if send chain to gc fails | |
4796 | store->delete_objs_inline(dpp, chain, tag); | |
4797 | } | |
4798 | return 0; | |
4799 | } | |
4800 | ||
4801 | void RGWRados::update_gc_chain(const DoutPrefixProvider *dpp, rgw_obj& head_obj, RGWObjManifest& manifest, cls_rgw_obj_chain *chain) | |
4802 | { | |
4803 | RGWObjManifest::obj_iterator iter; | |
4804 | rgw_raw_obj raw_head; | |
4805 | obj_to_raw(manifest.get_head_placement_rule(), head_obj, &raw_head); | |
4806 | for (iter = manifest.obj_begin(dpp); iter != manifest.obj_end(dpp); ++iter) { | |
4807 | const rgw_raw_obj& mobj = iter.get_location().get_raw_obj(store); | |
4808 | if (mobj == raw_head) | |
4809 | continue; | |
4810 | cls_rgw_obj_key key(mobj.oid); | |
4811 | chain->push_obj(mobj.pool.to_str(), key, mobj.loc); | |
4812 | } | |
4813 | } | |
4814 | ||
4815 | int RGWRados::send_chain_to_gc(cls_rgw_obj_chain& chain, const string& tag) | |
4816 | { | |
4817 | if (chain.empty()) { | |
4818 | return 0; | |
4819 | } | |
4820 | ||
4821 | return gc->send_chain(chain, tag); | |
4822 | } | |
4823 | ||
4824 | void RGWRados::delete_objs_inline(const DoutPrefixProvider *dpp, cls_rgw_obj_chain& chain, const string& tag) | |
4825 | { | |
4826 | string last_pool; | |
4827 | std::unique_ptr<IoCtx> ctx(new IoCtx); | |
4828 | int ret = 0; | |
4829 | for (auto liter = chain.objs.begin(); liter != chain.objs.end(); ++liter) { | |
4830 | cls_rgw_obj& obj = *liter; | |
4831 | if (obj.pool != last_pool) { | |
4832 | ctx.reset(new IoCtx); | |
4833 | ret = rgw_init_ioctx(dpp, get_rados_handle(), obj.pool, *ctx); | |
4834 | if (ret < 0) { | |
4835 | last_pool = ""; | |
4836 | ldpp_dout(dpp, 0) << "ERROR: failed to create ioctx pool=" << | |
4837 | obj.pool << dendl; | |
4838 | continue; | |
4839 | } | |
4840 | last_pool = obj.pool; | |
4841 | } | |
4842 | ctx->locator_set_key(obj.loc); | |
4843 | const string& oid = obj.key.name; /* just stored raw oid there */ | |
4844 | ldpp_dout(dpp, 5) << "delete_objs_inline: removing " << obj.pool << | |
4845 | ":" << obj.key.name << dendl; | |
4846 | ObjectWriteOperation op; | |
4847 | cls_refcount_put(op, tag, true); | |
4848 | ret = ctx->operate(oid, &op); | |
4849 | if (ret < 0) { | |
4850 | ldpp_dout(dpp, 5) << "delete_objs_inline: refcount put returned error " << ret << dendl; | |
4851 | } | |
4852 | } | |
4853 | } | |
4854 | ||
4855 | static void accumulate_raw_stats(const rgw_bucket_dir_header& header, | |
4856 | map<RGWObjCategory, RGWStorageStats>& stats) | |
4857 | { | |
4858 | for (const auto& pair : header.stats) { | |
4859 | const RGWObjCategory category = static_cast<RGWObjCategory>(pair.first); | |
4860 | const rgw_bucket_category_stats& header_stats = pair.second; | |
4861 | ||
4862 | RGWStorageStats& s = stats[category]; | |
4863 | ||
4864 | s.category = category; | |
4865 | s.size += header_stats.total_size; | |
4866 | s.size_rounded += header_stats.total_size_rounded; | |
4867 | s.size_utilized += header_stats.actual_size; | |
4868 | s.num_objects += header_stats.num_entries; | |
4869 | } | |
4870 | } | |
4871 | ||
4872 | int RGWRados::bucket_check_index(const DoutPrefixProvider *dpp, RGWBucketInfo& bucket_info, | |
4873 | map<RGWObjCategory, RGWStorageStats> *existing_stats, | |
4874 | map<RGWObjCategory, RGWStorageStats> *calculated_stats) | |
4875 | { | |
4876 | RGWSI_RADOS::Pool index_pool; | |
4877 | // key - bucket index object id | |
4878 | // value - bucket index check OP returned result with the given bucket index object (shard) | |
4879 | map<int, string> oids; | |
4880 | map<int, struct rgw_cls_check_index_ret> bucket_objs_ret; | |
4881 | ||
4882 | int ret = svc.bi_rados->open_bucket_index(dpp, bucket_info, std::nullopt, &index_pool, &oids, nullptr); | |
4883 | if (ret < 0) { | |
4884 | return ret; | |
4885 | } | |
4886 | ||
4887 | for (auto& iter : oids) { | |
4888 | bucket_objs_ret[iter.first] = rgw_cls_check_index_ret(); | |
4889 | } | |
4890 | ||
4891 | ret = CLSRGWIssueBucketCheck(index_pool.ioctx(), oids, bucket_objs_ret, cct->_conf->rgw_bucket_index_max_aio)(); | |
4892 | if (ret < 0) { | |
4893 | return ret; | |
4894 | } | |
4895 | ||
4896 | // Aggregate results (from different shards if there is any) | |
4897 | map<int, struct rgw_cls_check_index_ret>::iterator iter; | |
4898 | for (iter = bucket_objs_ret.begin(); iter != bucket_objs_ret.end(); ++iter) { | |
4899 | accumulate_raw_stats(iter->second.existing_header, *existing_stats); | |
4900 | accumulate_raw_stats(iter->second.calculated_header, *calculated_stats); | |
4901 | } | |
4902 | ||
4903 | return 0; | |
4904 | } | |
4905 | ||
4906 | int RGWRados::bucket_rebuild_index(const DoutPrefixProvider *dpp, RGWBucketInfo& bucket_info) | |
4907 | { | |
4908 | RGWSI_RADOS::Pool index_pool; | |
4909 | map<int, string> bucket_objs; | |
4910 | ||
4911 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, std::nullopt, &index_pool, &bucket_objs, nullptr); | |
4912 | if (r < 0) { | |
4913 | return r; | |
4914 | } | |
4915 | ||
4916 | return CLSRGWIssueBucketRebuild(index_pool.ioctx(), bucket_objs, cct->_conf->rgw_bucket_index_max_aio)(); | |
4917 | } | |
4918 | ||
4919 | int RGWRados::bucket_set_reshard(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const cls_rgw_bucket_instance_entry& entry) | |
4920 | { | |
4921 | RGWSI_RADOS::Pool index_pool; | |
4922 | map<int, string> bucket_objs; | |
4923 | ||
4924 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, std::nullopt, &index_pool, &bucket_objs, nullptr); | |
4925 | if (r < 0) { | |
4926 | return r; | |
4927 | } | |
4928 | ||
4929 | return CLSRGWIssueSetBucketResharding(index_pool.ioctx(), bucket_objs, entry, cct->_conf->rgw_bucket_index_max_aio)(); | |
4930 | } | |
4931 | ||
4932 | int RGWRados::defer_gc(const DoutPrefixProvider *dpp, void *ctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, optional_yield y) | |
4933 | { | |
4934 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
4935 | std::string oid, key; | |
4936 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
4937 | if (!rctx) | |
4938 | return 0; | |
4939 | ||
4940 | RGWObjState *state = NULL; | |
4941 | ||
4942 | int r = get_obj_state(dpp, rctx, bucket_info, obj, &state, false, y); | |
4943 | if (r < 0) | |
4944 | return r; | |
4945 | ||
4946 | if (!state->is_atomic) { | |
4947 | ldpp_dout(dpp, 20) << "state for obj=" << obj << " is not atomic, not deferring gc operation" << dendl; | |
4948 | return -EINVAL; | |
4949 | } | |
4950 | ||
4951 | string tag; | |
4952 | ||
4953 | if (state->tail_tag.length() > 0) { | |
4954 | tag = state->tail_tag.c_str(); | |
4955 | } else if (state->obj_tag.length() > 0) { | |
4956 | tag = state->obj_tag.c_str(); | |
4957 | } else { | |
4958 | ldpp_dout(dpp, 20) << "state->obj_tag is empty, not deferring gc operation" << dendl; | |
4959 | return -EINVAL; | |
4960 | } | |
4961 | ||
4962 | ldpp_dout(dpp, 0) << "defer chain tag=" << tag << dendl; | |
4963 | ||
4964 | cls_rgw_obj_chain chain; | |
4965 | update_gc_chain(dpp, state->obj, *state->manifest, &chain); | |
4966 | return gc->async_defer_chain(tag, chain); | |
4967 | } | |
4968 | ||
4969 | void RGWRados::remove_rgw_head_obj(ObjectWriteOperation& op) | |
4970 | { | |
4971 | list<string> prefixes; | |
4972 | prefixes.push_back(RGW_ATTR_OLH_PREFIX); | |
4973 | cls_rgw_remove_obj(op, prefixes); | |
4974 | } | |
4975 | ||
4976 | void RGWRados::cls_obj_check_prefix_exist(ObjectOperation& op, const string& prefix, bool fail_if_exist) | |
4977 | { | |
4978 | cls_rgw_obj_check_attrs_prefix(op, prefix, fail_if_exist); | |
4979 | } | |
4980 | ||
4981 | void RGWRados::cls_obj_check_mtime(ObjectOperation& op, const real_time& mtime, bool high_precision_time, RGWCheckMTimeType type) | |
4982 | { | |
4983 | cls_rgw_obj_check_mtime(op, mtime, high_precision_time, type); | |
4984 | } | |
4985 | ||
4986 | struct tombstone_entry { | |
4987 | ceph::real_time mtime; | |
4988 | uint32_t zone_short_id; | |
4989 | uint64_t pg_ver; | |
4990 | ||
4991 | tombstone_entry() = default; | |
4992 | explicit tombstone_entry(const RGWObjState& state) | |
4993 | : mtime(state.mtime), zone_short_id(state.zone_short_id), | |
4994 | pg_ver(state.pg_ver) {} | |
4995 | }; | |
4996 | ||
4997 | /** | |
4998 | * Delete an object. | |
4999 | * bucket: name of the bucket storing the object | |
5000 | * obj: name of the object to delete | |
5001 | * Returns: 0 on success, -ERR# otherwise. | |
5002 | */ | |
5003 | int RGWRados::Object::Delete::delete_obj(optional_yield y, const DoutPrefixProvider *dpp) | |
5004 | { | |
5005 | RGWRados *store = target->get_store(); | |
5006 | rgw_obj& src_obj = target->get_obj(); | |
5007 | const string& instance = src_obj.key.instance; | |
5008 | rgw_obj obj = src_obj; | |
5009 | ||
5010 | if (instance == "null") { | |
5011 | obj.key.instance.clear(); | |
5012 | } | |
5013 | ||
5014 | bool explicit_marker_version = (!params.marker_version_id.empty()); | |
5015 | ||
5016 | if (params.versioning_status & BUCKET_VERSIONED || explicit_marker_version) { | |
5017 | if (instance.empty() || explicit_marker_version) { | |
5018 | rgw_obj marker = obj; | |
5019 | ||
5020 | if (!params.marker_version_id.empty()) { | |
5021 | if (params.marker_version_id != "null") { | |
5022 | marker.key.set_instance(params.marker_version_id); | |
5023 | } | |
5024 | } else if ((params.versioning_status & BUCKET_VERSIONS_SUSPENDED) == 0) { | |
5025 | store->gen_rand_obj_instance_name(&marker); | |
5026 | } | |
5027 | ||
5028 | result.version_id = marker.key.instance; | |
5029 | if (result.version_id.empty()) | |
5030 | result.version_id = "null"; | |
5031 | result.delete_marker = true; | |
5032 | ||
5033 | struct rgw_bucket_dir_entry_meta meta; | |
5034 | ||
5035 | meta.owner = params.obj_owner.get_id().to_str(); | |
5036 | meta.owner_display_name = params.obj_owner.get_display_name(); | |
5037 | ||
5038 | if (real_clock::is_zero(params.mtime)) { | |
5039 | meta.mtime = real_clock::now(); | |
5040 | } else { | |
5041 | meta.mtime = params.mtime; | |
5042 | } | |
5043 | ||
5044 | int r = store->set_olh(dpp, target->get_ctx(), target->get_bucket_info(), marker, true, &meta, params.olh_epoch, params.unmod_since, params.high_precision_time, y, params.zones_trace); | |
5045 | if (r < 0) { | |
5046 | return r; | |
5047 | } | |
5048 | } else { | |
5049 | rgw_bucket_dir_entry dirent; | |
5050 | ||
5051 | int r = store->bi_get_instance(dpp, target->get_bucket_info(), obj, &dirent); | |
5052 | if (r < 0) { | |
5053 | return r; | |
5054 | } | |
5055 | result.delete_marker = dirent.is_delete_marker(); | |
5056 | r = store->unlink_obj_instance(dpp, target->get_ctx(), target->get_bucket_info(), obj, params.olh_epoch, y, params.zones_trace); | |
5057 | if (r < 0) { | |
5058 | return r; | |
5059 | } | |
5060 | result.version_id = instance; | |
5061 | } | |
5062 | ||
5063 | BucketShard *bs; | |
5064 | int r = target->get_bucket_shard(&bs, dpp); | |
5065 | if (r < 0) { | |
5066 | ldpp_dout(dpp, 5) << "failed to get BucketShard object: r=" << r << dendl; | |
5067 | return r; | |
5068 | } | |
5069 | ||
5070 | r = store->svc.datalog_rados->add_entry(dpp, target->bucket_info, bs->shard_id); | |
5071 | if (r < 0) { | |
5072 | ldpp_dout(dpp, -1) << "ERROR: failed writing data log" << dendl; | |
5073 | return r; | |
5074 | } | |
5075 | ||
5076 | return 0; | |
5077 | } | |
5078 | ||
5079 | rgw_rados_ref ref; | |
5080 | int r = store->get_obj_head_ref(dpp, target->get_bucket_info(), obj, &ref); | |
5081 | if (r < 0) { | |
5082 | return r; | |
5083 | } | |
5084 | ||
5085 | RGWObjState *state; | |
5086 | r = target->get_state(dpp, &state, false, y); | |
5087 | if (r < 0) | |
5088 | return r; | |
5089 | ||
5090 | ObjectWriteOperation op; | |
5091 | ||
5092 | if (!real_clock::is_zero(params.unmod_since)) { | |
5093 | struct timespec ctime = ceph::real_clock::to_timespec(state->mtime); | |
5094 | struct timespec unmod = ceph::real_clock::to_timespec(params.unmod_since); | |
5095 | if (!params.high_precision_time) { | |
5096 | ctime.tv_nsec = 0; | |
5097 | unmod.tv_nsec = 0; | |
5098 | } | |
5099 | ||
5100 | ldpp_dout(dpp, 10) << "If-UnModified-Since: " << params.unmod_since << " Last-Modified: " << ctime << dendl; | |
5101 | if (ctime > unmod) { | |
5102 | return -ERR_PRECONDITION_FAILED; | |
5103 | } | |
5104 | ||
5105 | /* only delete object if mtime is less than or equal to params.unmod_since */ | |
5106 | store->cls_obj_check_mtime(op, params.unmod_since, params.high_precision_time, CLS_RGW_CHECK_TIME_MTIME_LE); | |
5107 | } | |
5108 | uint64_t obj_accounted_size = state->accounted_size; | |
5109 | ||
5110 | if(params.abortmp) { | |
5111 | obj_accounted_size = params.parts_accounted_size; | |
5112 | } | |
5113 | ||
5114 | if (!real_clock::is_zero(params.expiration_time)) { | |
5115 | bufferlist bl; | |
5116 | real_time delete_at; | |
5117 | ||
5118 | if (state->get_attr(RGW_ATTR_DELETE_AT, bl)) { | |
5119 | try { | |
5120 | auto iter = bl.cbegin(); | |
5121 | decode(delete_at, iter); | |
5122 | } catch (buffer::error& err) { | |
5123 | ldpp_dout(dpp, 0) << "ERROR: couldn't decode RGW_ATTR_DELETE_AT" << dendl; | |
5124 | return -EIO; | |
5125 | } | |
5126 | ||
5127 | if (params.expiration_time != delete_at) { | |
5128 | return -ERR_PRECONDITION_FAILED; | |
5129 | } | |
5130 | } else { | |
5131 | return -ERR_PRECONDITION_FAILED; | |
5132 | } | |
5133 | } | |
5134 | ||
5135 | if (!state->exists) { | |
5136 | target->invalidate_state(); | |
5137 | return -ENOENT; | |
5138 | } | |
5139 | ||
5140 | r = target->prepare_atomic_modification(dpp, op, false, NULL, NULL, NULL, true, false, y); | |
5141 | if (r < 0) | |
5142 | return r; | |
5143 | ||
5144 | RGWBucketInfo& bucket_info = target->get_bucket_info(); | |
5145 | ||
5146 | RGWRados::Bucket bop(store, bucket_info); | |
5147 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
5148 | ||
5149 | index_op.set_zones_trace(params.zones_trace); | |
5150 | index_op.set_bilog_flags(params.bilog_flags); | |
5151 | ||
5152 | r = index_op.prepare(dpp, CLS_RGW_OP_DEL, &state->write_tag, y); | |
5153 | if (r < 0) | |
5154 | return r; | |
5155 | ||
5156 | store->remove_rgw_head_obj(op); | |
5157 | ||
5158 | auto& ioctx = ref.pool.ioctx(); | |
5159 | r = rgw_rados_operate(dpp, ioctx, ref.obj.oid, &op, null_yield); | |
5160 | ||
5161 | /* raced with another operation, object state is indeterminate */ | |
5162 | const bool need_invalidate = (r == -ECANCELED); | |
5163 | ||
5164 | int64_t poolid = ioctx.get_id(); | |
5165 | if (r >= 0) { | |
5166 | tombstone_cache_t *obj_tombstone_cache = store->get_tombstone_cache(); | |
5167 | if (obj_tombstone_cache) { | |
5168 | tombstone_entry entry{*state}; | |
5169 | obj_tombstone_cache->add(obj, entry); | |
5170 | } | |
5171 | r = index_op.complete_del(dpp, poolid, ioctx.get_last_version(), state->mtime, params.remove_objs); | |
5172 | ||
5173 | int ret = target->complete_atomic_modification(dpp); | |
5174 | if (ret < 0) { | |
5175 | ldpp_dout(dpp, 0) << "ERROR: complete_atomic_modification returned ret=" << ret << dendl; | |
5176 | } | |
5177 | /* other than that, no need to propagate error */ | |
5178 | } else { | |
5179 | int ret = index_op.cancel(dpp); | |
5180 | if (ret < 0) { | |
5181 | ldpp_dout(dpp, 0) << "ERROR: index_op.cancel() returned ret=" << ret << dendl; | |
5182 | } | |
5183 | } | |
5184 | ||
5185 | if (need_invalidate) { | |
5186 | target->invalidate_state(); | |
5187 | } | |
5188 | ||
5189 | if (r < 0) | |
5190 | return r; | |
5191 | ||
5192 | /* update quota cache */ | |
5193 | store->quota_handler->update_stats(params.bucket_owner, obj.bucket, -1, 0, obj_accounted_size); | |
5194 | ||
5195 | return 0; | |
5196 | } | |
5197 | ||
5198 | int RGWRados::delete_obj(const DoutPrefixProvider *dpp, | |
5199 | RGWObjectCtx& obj_ctx, | |
5200 | const RGWBucketInfo& bucket_info, | |
5201 | const rgw_obj& obj, | |
5202 | int versioning_status, // versioning flags in enum RGWBucketFlags | |
5203 | uint16_t bilog_flags, | |
5204 | const real_time& expiration_time, | |
5205 | rgw_zone_set *zones_trace) | |
5206 | { | |
5207 | RGWRados::Object del_target(this, bucket_info, obj_ctx, obj); | |
5208 | RGWRados::Object::Delete del_op(&del_target); | |
5209 | ||
5210 | del_op.params.bucket_owner = bucket_info.owner; | |
5211 | del_op.params.versioning_status = versioning_status; | |
5212 | del_op.params.bilog_flags = bilog_flags; | |
5213 | del_op.params.expiration_time = expiration_time; | |
5214 | del_op.params.zones_trace = zones_trace; | |
5215 | ||
5216 | return del_op.delete_obj(null_yield, dpp); | |
5217 | } | |
5218 | ||
5219 | int RGWRados::delete_raw_obj(const DoutPrefixProvider *dpp, const rgw_raw_obj& obj) | |
5220 | { | |
5221 | rgw_rados_ref ref; | |
5222 | int r = get_raw_obj_ref(dpp, obj, &ref); | |
5223 | if (r < 0) { | |
5224 | return r; | |
5225 | } | |
5226 | ||
5227 | ObjectWriteOperation op; | |
5228 | ||
5229 | op.remove(); | |
5230 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
5231 | if (r < 0) | |
5232 | return r; | |
5233 | ||
5234 | return 0; | |
5235 | } | |
5236 | ||
5237 | int RGWRados::delete_obj_index(const rgw_obj& obj, ceph::real_time mtime, const DoutPrefixProvider *dpp) | |
5238 | { | |
5239 | std::string oid, key; | |
5240 | get_obj_bucket_and_oid_loc(obj, oid, key); | |
5241 | ||
5242 | auto obj_ctx = svc.sysobj->init_obj_ctx(); | |
5243 | ||
5244 | RGWBucketInfo bucket_info; | |
5245 | int ret = get_bucket_instance_info(obj_ctx, obj.bucket, bucket_info, NULL, NULL, null_yield, dpp); | |
5246 | if (ret < 0) { | |
5247 | ldpp_dout(dpp, 0) << "ERROR: " << __func__ << "() get_bucket_instance_info(bucket=" << obj.bucket << ") returned ret=" << ret << dendl; | |
5248 | return ret; | |
5249 | } | |
5250 | ||
5251 | RGWRados::Bucket bop(this, bucket_info); | |
5252 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
5253 | ||
5254 | return index_op.complete_del(dpp, -1 /* pool */, 0, mtime, NULL); | |
5255 | } | |
5256 | ||
5257 | static void generate_fake_tag(const DoutPrefixProvider *dpp, rgw::sal::RGWStore* store, map<string, bufferlist>& attrset, RGWObjManifest& manifest, bufferlist& manifest_bl, bufferlist& tag_bl) | |
5258 | { | |
5259 | string tag; | |
5260 | ||
5261 | RGWObjManifest::obj_iterator mi = manifest.obj_begin(dpp); | |
5262 | if (mi != manifest.obj_end(dpp)) { | |
5263 | if (manifest.has_tail()) // first object usually points at the head, let's skip to a more unique part | |
5264 | ++mi; | |
5265 | tag = mi.get_location().get_raw_obj(store).oid; | |
5266 | tag.append("_"); | |
5267 | } | |
5268 | ||
5269 | unsigned char md5[CEPH_CRYPTO_MD5_DIGESTSIZE]; | |
5270 | char md5_str[CEPH_CRYPTO_MD5_DIGESTSIZE * 2 + 1]; | |
5271 | MD5 hash; | |
5272 | hash.Update((const unsigned char *)manifest_bl.c_str(), manifest_bl.length()); | |
5273 | ||
5274 | map<string, bufferlist>::iterator iter = attrset.find(RGW_ATTR_ETAG); | |
5275 | if (iter != attrset.end()) { | |
5276 | bufferlist& bl = iter->second; | |
5277 | hash.Update((const unsigned char *)bl.c_str(), bl.length()); | |
5278 | } | |
5279 | ||
5280 | hash.Final(md5); | |
5281 | buf_to_hex(md5, CEPH_CRYPTO_MD5_DIGESTSIZE, md5_str); | |
5282 | tag.append(md5_str); | |
5283 | ||
5284 | ldout(store->ctx(), 10) << "generate_fake_tag new tag=" << tag << dendl; | |
5285 | ||
5286 | tag_bl.append(tag.c_str(), tag.size() + 1); | |
5287 | } | |
5288 | ||
5289 | static bool is_olh(map<string, bufferlist>& attrs) | |
5290 | { | |
5291 | map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_OLH_INFO); | |
5292 | return (iter != attrs.end()); | |
5293 | } | |
5294 | ||
5295 | static bool has_olh_tag(map<string, bufferlist>& attrs) | |
5296 | { | |
5297 | map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_OLH_ID_TAG); | |
5298 | return (iter != attrs.end()); | |
5299 | } | |
5300 | ||
5301 | int RGWRados::get_olh_target_state(const DoutPrefixProvider *dpp, RGWObjectCtx& obj_ctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
5302 | RGWObjState *olh_state, RGWObjState **target_state, optional_yield y) | |
5303 | { | |
5304 | ceph_assert(olh_state->is_olh); | |
5305 | ||
5306 | rgw_obj target; | |
5307 | int r = RGWRados::follow_olh(dpp, bucket_info, obj_ctx, olh_state, obj, &target); /* might return -EAGAIN */ | |
5308 | if (r < 0) { | |
5309 | return r; | |
5310 | } | |
5311 | r = get_obj_state(dpp, &obj_ctx, bucket_info, target, target_state, false, y); | |
5312 | if (r < 0) { | |
5313 | return r; | |
5314 | } | |
5315 | ||
5316 | return 0; | |
5317 | } | |
5318 | ||
5319 | int RGWRados::get_obj_state_impl(const DoutPrefixProvider *dpp, RGWObjectCtx *rctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
5320 | RGWObjState **state, bool follow_olh, optional_yield y, bool assume_noent) | |
5321 | { | |
5322 | if (obj.empty()) { | |
5323 | return -EINVAL; | |
5324 | } | |
5325 | ||
5326 | bool need_follow_olh = follow_olh && obj.key.instance.empty(); | |
5327 | ||
5328 | RGWObjState *s = rctx->get_state(obj); | |
5329 | ldpp_dout(dpp, 20) << "get_obj_state: rctx=" << (void *)rctx << " obj=" << obj << " state=" << (void *)s << " s->prefetch_data=" << s->prefetch_data << dendl; | |
5330 | *state = s; | |
5331 | if (s->has_attrs) { | |
5332 | if (s->is_olh && need_follow_olh) { | |
5333 | return get_olh_target_state(dpp, *rctx, bucket_info, obj, s, state, y); | |
5334 | } | |
5335 | return 0; | |
5336 | } | |
5337 | ||
5338 | s->obj = obj; | |
5339 | ||
5340 | rgw_raw_obj raw_obj; | |
5341 | obj_to_raw(bucket_info.placement_rule, obj, &raw_obj); | |
5342 | ||
5343 | int r = -ENOENT; | |
5344 | ||
5345 | if (!assume_noent) { | |
5346 | r = RGWRados::raw_obj_stat(dpp, raw_obj, &s->size, &s->mtime, &s->epoch, &s->attrset, (s->prefetch_data ? &s->data : NULL), NULL, y); | |
5347 | } | |
5348 | ||
5349 | if (r == -ENOENT) { | |
5350 | s->exists = false; | |
5351 | s->has_attrs = true; | |
5352 | tombstone_entry entry; | |
5353 | if (obj_tombstone_cache && obj_tombstone_cache->find(obj, entry)) { | |
5354 | s->mtime = entry.mtime; | |
5355 | s->zone_short_id = entry.zone_short_id; | |
5356 | s->pg_ver = entry.pg_ver; | |
5357 | ldpp_dout(dpp, 20) << __func__ << "(): found obj in tombstone cache: obj=" << obj | |
5358 | << " mtime=" << s->mtime << " pgv=" << s->pg_ver << dendl; | |
5359 | } else { | |
5360 | s->mtime = real_time(); | |
5361 | } | |
5362 | return 0; | |
5363 | } | |
5364 | if (r < 0) | |
5365 | return r; | |
5366 | ||
5367 | s->exists = true; | |
5368 | s->has_attrs = true; | |
5369 | s->accounted_size = s->size; | |
5370 | ||
5371 | auto iter = s->attrset.find(RGW_ATTR_ETAG); | |
5372 | if (iter != s->attrset.end()) { | |
5373 | /* get rid of extra null character at the end of the etag, as we used to store it like that */ | |
5374 | bufferlist& bletag = iter->second; | |
5375 | if (bletag.length() > 0 && bletag[bletag.length() - 1] == '\0') { | |
5376 | bufferlist newbl; | |
5377 | bletag.splice(0, bletag.length() - 1, &newbl); | |
5378 | bletag = std::move(newbl); | |
5379 | } | |
5380 | } | |
5381 | ||
5382 | iter = s->attrset.find(RGW_ATTR_COMPRESSION); | |
5383 | const bool compressed = (iter != s->attrset.end()); | |
5384 | if (compressed) { | |
5385 | // use uncompressed size for accounted_size | |
5386 | try { | |
5387 | RGWCompressionInfo info; | |
5388 | auto p = iter->second.cbegin(); | |
5389 | decode(info, p); | |
5390 | s->accounted_size = info.orig_size; | |
5391 | } catch (buffer::error&) { | |
5392 | ldpp_dout(dpp, 0) << "ERROR: could not decode compression info for object: " << obj << dendl; | |
5393 | return -EIO; | |
5394 | } | |
5395 | } | |
5396 | ||
5397 | iter = s->attrset.find(RGW_ATTR_SHADOW_OBJ); | |
5398 | if (iter != s->attrset.end()) { | |
5399 | bufferlist bl = iter->second; | |
5400 | bufferlist::iterator it = bl.begin(); | |
5401 | it.copy(bl.length(), s->shadow_obj); | |
5402 | s->shadow_obj[bl.length()] = '\0'; | |
5403 | } | |
5404 | s->obj_tag = s->attrset[RGW_ATTR_ID_TAG]; | |
5405 | auto ttiter = s->attrset.find(RGW_ATTR_TAIL_TAG); | |
5406 | if (ttiter != s->attrset.end()) { | |
5407 | s->tail_tag = s->attrset[RGW_ATTR_TAIL_TAG]; | |
5408 | } | |
5409 | ||
5410 | bufferlist manifest_bl = s->attrset[RGW_ATTR_MANIFEST]; | |
5411 | if (manifest_bl.length()) { | |
5412 | auto miter = manifest_bl.cbegin(); | |
5413 | try { | |
5414 | s->manifest.emplace(); | |
5415 | decode(*s->manifest, miter); | |
5416 | s->manifest->set_head(bucket_info.placement_rule, obj, s->size); /* patch manifest to reflect the head we just read, some manifests might be | |
5417 | broken due to old bugs */ | |
5418 | s->size = s->manifest->get_obj_size(); | |
5419 | if (!compressed) | |
5420 | s->accounted_size = s->size; | |
5421 | } catch (buffer::error& err) { | |
5422 | ldpp_dout(dpp, 0) << "ERROR: couldn't decode manifest" << dendl; | |
5423 | return -EIO; | |
5424 | } | |
5425 | ldpp_dout(dpp, 10) << "manifest: total_size = " << s->manifest->get_obj_size() << dendl; | |
5426 | if (cct->_conf->subsys.should_gather<ceph_subsys_rgw, 20>() && \ | |
5427 | s->manifest->has_explicit_objs()) { | |
5428 | RGWObjManifest::obj_iterator mi; | |
5429 | for (mi = s->manifest->obj_begin(dpp); mi != s->manifest->obj_end(dpp); ++mi) { | |
5430 | ldpp_dout(dpp, 20) << "manifest: ofs=" << mi.get_ofs() << " loc=" << mi.get_location().get_raw_obj(store) << dendl; | |
5431 | } | |
5432 | } | |
5433 | ||
5434 | if (!s->obj_tag.length()) { | |
5435 | /* | |
5436 | * Uh oh, something's wrong, object with manifest should have tag. Let's | |
5437 | * create one out of the manifest, would be unique | |
5438 | */ | |
5439 | generate_fake_tag(dpp, store, s->attrset, *s->manifest, manifest_bl, s->obj_tag); | |
5440 | s->fake_tag = true; | |
5441 | } | |
5442 | } | |
5443 | map<string, bufferlist>::iterator aiter = s->attrset.find(RGW_ATTR_PG_VER); | |
5444 | if (aiter != s->attrset.end()) { | |
5445 | bufferlist& pg_ver_bl = aiter->second; | |
5446 | if (pg_ver_bl.length()) { | |
5447 | auto pgbl = pg_ver_bl.cbegin(); | |
5448 | try { | |
5449 | decode(s->pg_ver, pgbl); | |
5450 | } catch (buffer::error& err) { | |
5451 | ldpp_dout(dpp, 0) << "ERROR: couldn't decode pg ver attr for object " << s->obj << ", non-critical error, ignoring" << dendl; | |
5452 | } | |
5453 | } | |
5454 | } | |
5455 | aiter = s->attrset.find(RGW_ATTR_SOURCE_ZONE); | |
5456 | if (aiter != s->attrset.end()) { | |
5457 | bufferlist& zone_short_id_bl = aiter->second; | |
5458 | if (zone_short_id_bl.length()) { | |
5459 | auto zbl = zone_short_id_bl.cbegin(); | |
5460 | try { | |
5461 | decode(s->zone_short_id, zbl); | |
5462 | } catch (buffer::error& err) { | |
5463 | ldpp_dout(dpp, 0) << "ERROR: couldn't decode zone short id attr for object " << s->obj << ", non-critical error, ignoring" << dendl; | |
5464 | } | |
5465 | } | |
5466 | } | |
5467 | if (s->obj_tag.length()) { | |
5468 | ldpp_dout(dpp, 20) << "get_obj_state: setting s->obj_tag to " << s->obj_tag.c_str() << dendl; | |
5469 | } else { | |
5470 | ldpp_dout(dpp, 20) << "get_obj_state: s->obj_tag was set empty" << dendl; | |
5471 | } | |
5472 | ||
5473 | /* an object might not be olh yet, but could have olh id tag, so we should set it anyway if | |
5474 | * it exist, and not only if is_olh() returns true | |
5475 | */ | |
5476 | iter = s->attrset.find(RGW_ATTR_OLH_ID_TAG); | |
5477 | if (iter != s->attrset.end()) { | |
5478 | s->olh_tag = iter->second; | |
5479 | } | |
5480 | ||
5481 | if (is_olh(s->attrset)) { | |
5482 | s->is_olh = true; | |
5483 | ||
5484 | ldpp_dout(dpp, 20) << __func__ << ": setting s->olh_tag to " << string(s->olh_tag.c_str(), s->olh_tag.length()) << dendl; | |
5485 | ||
5486 | if (need_follow_olh) { | |
5487 | return get_olh_target_state(dpp, *rctx, bucket_info, obj, s, state, y); | |
5488 | } else if (obj.key.have_null_instance() && !s->manifest) { | |
5489 | // read null version, and the head object only have olh info | |
5490 | s->exists = false; | |
5491 | return -ENOENT; | |
5492 | } | |
5493 | } | |
5494 | ||
5495 | return 0; | |
5496 | } | |
5497 | ||
5498 | int RGWRados::get_obj_state(const DoutPrefixProvider *dpp, RGWObjectCtx *rctx, const RGWBucketInfo& bucket_info, const rgw_obj& obj, RGWObjState **state, | |
5499 | bool follow_olh, optional_yield y, bool assume_noent) | |
5500 | { | |
5501 | int ret; | |
5502 | ||
5503 | do { | |
5504 | ret = get_obj_state_impl(dpp, rctx, bucket_info, obj, state, follow_olh, y, assume_noent); | |
5505 | } while (ret == -EAGAIN); | |
5506 | ||
5507 | return ret; | |
5508 | } | |
5509 | ||
5510 | int RGWRados::Object::get_manifest(const DoutPrefixProvider *dpp, RGWObjManifest **pmanifest, optional_yield y) | |
5511 | { | |
5512 | RGWObjState *astate; | |
5513 | int r = get_state(dpp, &astate, true, y); | |
5514 | if (r < 0) { | |
5515 | return r; | |
5516 | } | |
5517 | ||
5518 | *pmanifest = &(*astate->manifest); | |
5519 | ||
5520 | return 0; | |
5521 | } | |
5522 | ||
5523 | int RGWRados::Object::Read::get_attr(const DoutPrefixProvider *dpp, const char *name, bufferlist& dest, optional_yield y) | |
5524 | { | |
5525 | RGWObjState *state; | |
5526 | int r = source->get_state(dpp, &state, true, y); | |
5527 | if (r < 0) | |
5528 | return r; | |
5529 | if (!state->exists) | |
5530 | return -ENOENT; | |
5531 | if (!state->get_attr(name, dest)) | |
5532 | return -ENODATA; | |
5533 | ||
5534 | return 0; | |
5535 | } | |
5536 | ||
5537 | int RGWRados::Object::Stat::stat_async(const DoutPrefixProvider *dpp) | |
5538 | { | |
5539 | RGWObjectCtx& ctx = source->get_ctx(); | |
5540 | rgw_obj& obj = source->get_obj(); | |
5541 | RGWRados *store = source->get_store(); | |
5542 | ||
5543 | RGWObjState *s = ctx.get_state(obj); /* calling this one directly because otherwise a sync request will be sent */ | |
5544 | result.obj = obj; | |
5545 | if (s->has_attrs) { | |
5546 | state.ret = 0; | |
5547 | result.size = s->size; | |
5548 | result.mtime = ceph::real_clock::to_timespec(s->mtime); | |
5549 | result.attrs = s->attrset; | |
5550 | result.manifest = s->manifest; | |
5551 | return 0; | |
5552 | } | |
5553 | ||
5554 | string oid; | |
5555 | string loc; | |
5556 | get_obj_bucket_and_oid_loc(obj, oid, loc); | |
5557 | ||
5558 | int r = store->get_obj_head_ioctx(dpp, source->get_bucket_info(), obj, &state.io_ctx); | |
5559 | if (r < 0) { | |
5560 | return r; | |
5561 | } | |
5562 | ||
5563 | librados::ObjectReadOperation op; | |
5564 | op.stat2(&result.size, &result.mtime, NULL); | |
5565 | op.getxattrs(&result.attrs, NULL); | |
5566 | state.completion = librados::Rados::aio_create_completion(nullptr, nullptr); | |
5567 | state.io_ctx.locator_set_key(loc); | |
5568 | r = state.io_ctx.aio_operate(oid, state.completion, &op, NULL); | |
5569 | if (r < 0) { | |
5570 | ldpp_dout(dpp, 5) << __func__ | |
5571 | << ": ERROR: aio_operate() returned ret=" << r | |
5572 | << dendl; | |
5573 | return r; | |
5574 | } | |
5575 | ||
5576 | return 0; | |
5577 | } | |
5578 | ||
5579 | ||
5580 | int RGWRados::Object::Stat::wait() | |
5581 | { | |
5582 | if (!state.completion) { | |
5583 | return state.ret; | |
5584 | } | |
5585 | ||
5586 | state.completion->wait_for_complete(); | |
5587 | state.ret = state.completion->get_return_value(); | |
5588 | state.completion->release(); | |
5589 | ||
5590 | if (state.ret != 0) { | |
5591 | return state.ret; | |
5592 | } | |
5593 | ||
5594 | return finish(); | |
5595 | } | |
5596 | ||
5597 | int RGWRados::Object::Stat::finish() | |
5598 | { | |
5599 | map<string, bufferlist>::iterator iter = result.attrs.find(RGW_ATTR_MANIFEST); | |
5600 | if (iter != result.attrs.end()) { | |
5601 | bufferlist& bl = iter->second; | |
5602 | auto biter = bl.cbegin(); | |
5603 | try { | |
5604 | result.manifest.emplace(); | |
5605 | decode(*result.manifest, biter); | |
5606 | } catch (buffer::error& err) { | |
5607 | RGWRados *store = source->get_store(); | |
5608 | ldout(store->ctx(), 0) << "ERROR: " << __func__ << ": failed to decode manifest" << dendl; | |
5609 | return -EIO; | |
5610 | } | |
5611 | } | |
5612 | ||
5613 | return 0; | |
5614 | } | |
5615 | ||
5616 | int RGWRados::append_atomic_test(const DoutPrefixProvider *dpp, RGWObjectCtx *rctx, | |
5617 | const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
5618 | ObjectOperation& op, RGWObjState **pstate, optional_yield y) | |
5619 | { | |
5620 | if (!rctx) | |
5621 | return 0; | |
5622 | ||
5623 | int r = get_obj_state(dpp, rctx, bucket_info, obj, pstate, false, y); | |
5624 | if (r < 0) | |
5625 | return r; | |
5626 | ||
5627 | return append_atomic_test(dpp, *pstate, op); | |
5628 | } | |
5629 | ||
5630 | int RGWRados::append_atomic_test(const DoutPrefixProvider *dpp, | |
5631 | const RGWObjState* state, | |
5632 | librados::ObjectOperation& op) | |
5633 | { | |
5634 | if (!state->is_atomic) { | |
5635 | ldpp_dout(dpp, 20) << "state for obj=" << state->obj << " is not atomic, not appending atomic test" << dendl; | |
5636 | return 0; | |
5637 | } | |
5638 | ||
5639 | if (state->obj_tag.length() > 0 && !state->fake_tag) {// check for backward compatibility | |
5640 | op.cmpxattr(RGW_ATTR_ID_TAG, LIBRADOS_CMPXATTR_OP_EQ, state->obj_tag); | |
5641 | } else { | |
5642 | ldpp_dout(dpp, 20) << "state->obj_tag is empty, not appending atomic test" << dendl; | |
5643 | } | |
5644 | return 0; | |
5645 | } | |
5646 | ||
5647 | int RGWRados::Object::get_state(const DoutPrefixProvider *dpp, RGWObjState **pstate, bool follow_olh, optional_yield y, bool assume_noent) | |
5648 | { | |
5649 | return store->get_obj_state(dpp, &ctx, bucket_info, obj, pstate, follow_olh, y, assume_noent); | |
5650 | } | |
5651 | ||
5652 | void RGWRados::Object::invalidate_state() | |
5653 | { | |
5654 | ctx.invalidate(obj); | |
5655 | } | |
5656 | ||
5657 | int RGWRados::Object::prepare_atomic_modification(const DoutPrefixProvider *dpp, | |
5658 | ObjectWriteOperation& op, bool reset_obj, const string *ptag, | |
5659 | const char *if_match, const char *if_nomatch, bool removal_op, | |
5660 | bool modify_tail, optional_yield y) | |
5661 | { | |
5662 | int r = get_state(dpp, &state, false, y); | |
5663 | if (r < 0) | |
5664 | return r; | |
5665 | ||
5666 | bool need_guard = ((state->manifest) || (state->obj_tag.length() != 0) || | |
5667 | if_match != NULL || if_nomatch != NULL) && | |
5668 | (!state->fake_tag); | |
5669 | ||
5670 | if (!state->is_atomic) { | |
5671 | ldpp_dout(dpp, 20) << "prepare_atomic_modification: state is not atomic. state=" << (void *)state << dendl; | |
5672 | ||
5673 | if (reset_obj) { | |
5674 | op.create(false); | |
5675 | store->remove_rgw_head_obj(op); // we're not dropping reference here, actually removing object | |
5676 | } | |
5677 | ||
5678 | return 0; | |
5679 | } | |
5680 | ||
5681 | if (need_guard) { | |
5682 | /* first verify that the object wasn't replaced under */ | |
5683 | if (if_nomatch == NULL || strcmp(if_nomatch, "*") != 0) { | |
5684 | op.cmpxattr(RGW_ATTR_ID_TAG, LIBRADOS_CMPXATTR_OP_EQ, state->obj_tag); | |
5685 | // FIXME: need to add FAIL_NOTEXIST_OK for racing deletion | |
5686 | } | |
5687 | ||
5688 | if (if_match) { | |
5689 | if (strcmp(if_match, "*") == 0) { | |
5690 | // test the object is existing | |
5691 | if (!state->exists) { | |
5692 | return -ERR_PRECONDITION_FAILED; | |
5693 | } | |
5694 | } else { | |
5695 | bufferlist bl; | |
5696 | if (!state->get_attr(RGW_ATTR_ETAG, bl) || | |
5697 | strncmp(if_match, bl.c_str(), bl.length()) != 0) { | |
5698 | return -ERR_PRECONDITION_FAILED; | |
5699 | } | |
5700 | } | |
5701 | } | |
5702 | ||
5703 | if (if_nomatch) { | |
5704 | if (strcmp(if_nomatch, "*") == 0) { | |
5705 | // test the object is NOT existing | |
5706 | if (state->exists) { | |
5707 | return -ERR_PRECONDITION_FAILED; | |
5708 | } | |
5709 | } else { | |
5710 | bufferlist bl; | |
5711 | if (!state->get_attr(RGW_ATTR_ETAG, bl) || | |
5712 | strncmp(if_nomatch, bl.c_str(), bl.length()) == 0) { | |
5713 | return -ERR_PRECONDITION_FAILED; | |
5714 | } | |
5715 | } | |
5716 | } | |
5717 | } | |
5718 | ||
5719 | if (reset_obj) { | |
5720 | if (state->exists) { | |
5721 | op.create(false); | |
5722 | store->remove_rgw_head_obj(op); | |
5723 | } else { | |
5724 | op.create(true); | |
5725 | } | |
5726 | } | |
5727 | ||
5728 | if (removal_op) { | |
5729 | /* the object is being removed, no need to update its tag */ | |
5730 | return 0; | |
5731 | } | |
5732 | ||
5733 | if (ptag) { | |
5734 | state->write_tag = *ptag; | |
5735 | } else { | |
5736 | append_rand_alpha(store->ctx(), state->write_tag, state->write_tag, 32); | |
5737 | } | |
5738 | bufferlist bl; | |
5739 | bl.append(state->write_tag.c_str(), state->write_tag.size() + 1); | |
5740 | ||
5741 | ldpp_dout(dpp, 10) << "setting object write_tag=" << state->write_tag << dendl; | |
5742 | ||
5743 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
5744 | if (modify_tail) { | |
5745 | op.setxattr(RGW_ATTR_TAIL_TAG, bl); | |
5746 | } | |
5747 | ||
5748 | return 0; | |
5749 | } | |
5750 | ||
5751 | /** | |
5752 | * Set an attr on an object. | |
5753 | * bucket: name of the bucket holding the object | |
5754 | * obj: name of the object to set the attr on | |
5755 | * name: the attr to set | |
5756 | * bl: the contents of the attr | |
5757 | * Returns: 0 on success, -ERR# otherwise. | |
5758 | */ | |
5759 | int RGWRados::set_attr(const DoutPrefixProvider *dpp, void *ctx, const RGWBucketInfo& bucket_info, rgw_obj& obj, const char *name, bufferlist& bl) | |
5760 | { | |
5761 | map<string, bufferlist> attrs; | |
5762 | attrs[name] = bl; | |
5763 | return set_attrs(dpp, ctx, bucket_info, obj, attrs, NULL, null_yield); | |
5764 | } | |
5765 | ||
5766 | int RGWRados::set_attrs(const DoutPrefixProvider *dpp, void *ctx, const RGWBucketInfo& bucket_info, rgw_obj& src_obj, | |
5767 | map<string, bufferlist>& attrs, | |
5768 | map<string, bufferlist>* rmattrs, | |
5769 | optional_yield y) | |
5770 | { | |
5771 | rgw_obj obj = src_obj; | |
5772 | if (obj.key.instance == "null") { | |
5773 | obj.key.instance.clear(); | |
5774 | } | |
5775 | ||
5776 | rgw_rados_ref ref; | |
5777 | int r = get_obj_head_ref(dpp, bucket_info, obj, &ref); | |
5778 | if (r < 0) { | |
5779 | return r; | |
5780 | } | |
5781 | RGWObjectCtx *rctx = static_cast<RGWObjectCtx *>(ctx); | |
5782 | ||
5783 | ObjectWriteOperation op; | |
5784 | RGWObjState *state = NULL; | |
5785 | ||
5786 | r = append_atomic_test(dpp, rctx, bucket_info, obj, op, &state, y); | |
5787 | if (r < 0) | |
5788 | return r; | |
5789 | ||
5790 | // ensure null version object exist | |
5791 | if (src_obj.key.instance == "null" && !state->manifest) { | |
5792 | return -ENOENT; | |
5793 | } | |
5794 | ||
5795 | map<string, bufferlist>::iterator iter; | |
5796 | if (rmattrs) { | |
5797 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
5798 | const string& name = iter->first; | |
5799 | op.rmxattr(name.c_str()); | |
5800 | } | |
5801 | } | |
5802 | ||
5803 | const rgw_bucket& bucket = obj.bucket; | |
5804 | ||
5805 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
5806 | const string& name = iter->first; | |
5807 | bufferlist& bl = iter->second; | |
5808 | ||
5809 | if (!bl.length()) | |
5810 | continue; | |
5811 | ||
5812 | op.setxattr(name.c_str(), bl); | |
5813 | ||
5814 | if (name.compare(RGW_ATTR_DELETE_AT) == 0) { | |
5815 | real_time ts; | |
5816 | try { | |
5817 | decode(ts, bl); | |
5818 | ||
5819 | rgw_obj_index_key obj_key; | |
5820 | obj.key.get_index_key(&obj_key); | |
5821 | ||
5822 | obj_expirer->hint_add(dpp, ts, bucket.tenant, bucket.name, bucket.bucket_id, obj_key); | |
5823 | } catch (buffer::error& err) { | |
5824 | ldpp_dout(dpp, 0) << "ERROR: failed to decode " RGW_ATTR_DELETE_AT << " attr" << dendl; | |
5825 | } | |
5826 | } | |
5827 | } | |
5828 | ||
5829 | if (!op.size()) | |
5830 | return 0; | |
5831 | ||
5832 | RGWObjectCtx obj_ctx(this->store); | |
5833 | ||
5834 | bufferlist bl; | |
5835 | RGWRados::Bucket bop(this, bucket_info); | |
5836 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
5837 | ||
5838 | if (state) { | |
5839 | string tag; | |
5840 | append_rand_alpha(cct, tag, tag, 32); | |
5841 | state->write_tag = tag; | |
5842 | r = index_op.prepare(dpp, CLS_RGW_OP_ADD, &state->write_tag, y); | |
5843 | ||
5844 | if (r < 0) | |
5845 | return r; | |
5846 | ||
5847 | bl.append(tag.c_str(), tag.size() + 1); | |
5848 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
5849 | } | |
5850 | ||
5851 | ||
5852 | real_time mtime = real_clock::now(); | |
5853 | struct timespec mtime_ts = real_clock::to_timespec(mtime); | |
5854 | op.mtime2(&mtime_ts); | |
5855 | auto& ioctx = ref.pool.ioctx(); | |
5856 | r = rgw_rados_operate(dpp, ioctx, ref.obj.oid, &op, null_yield); | |
5857 | if (state) { | |
5858 | if (r >= 0) { | |
5859 | bufferlist acl_bl = attrs[RGW_ATTR_ACL]; | |
5860 | bufferlist etag_bl = attrs[RGW_ATTR_ETAG]; | |
5861 | bufferlist content_type_bl = attrs[RGW_ATTR_CONTENT_TYPE]; | |
5862 | string etag = rgw_bl_str(etag_bl); | |
5863 | string content_type = rgw_bl_str(content_type_bl); | |
5864 | string storage_class; | |
5865 | auto iter = attrs.find(RGW_ATTR_STORAGE_CLASS); | |
5866 | if (iter != attrs.end()) { | |
5867 | storage_class = rgw_bl_str(iter->second); | |
5868 | } | |
5869 | uint64_t epoch = ioctx.get_last_version(); | |
5870 | int64_t poolid = ioctx.get_id(); | |
5871 | r = index_op.complete(dpp, poolid, epoch, state->size, state->accounted_size, | |
5872 | mtime, etag, content_type, storage_class, &acl_bl, | |
5873 | RGWObjCategory::Main, NULL); | |
5874 | } else { | |
5875 | int ret = index_op.cancel(dpp); | |
5876 | if (ret < 0) { | |
5877 | ldpp_dout(dpp, 0) << "ERROR: complete_update_index_cancel() returned ret=" << ret << dendl; | |
5878 | } | |
5879 | } | |
5880 | } | |
5881 | if (r < 0) | |
5882 | return r; | |
5883 | ||
5884 | if (state) { | |
5885 | state->obj_tag.swap(bl); | |
5886 | if (rmattrs) { | |
5887 | for (iter = rmattrs->begin(); iter != rmattrs->end(); ++iter) { | |
5888 | state->attrset.erase(iter->first); | |
5889 | } | |
5890 | } | |
5891 | ||
5892 | for (iter = attrs.begin(); iter != attrs.end(); ++iter) { | |
5893 | state->attrset[iter->first] = iter->second; | |
5894 | } | |
5895 | ||
5896 | auto iter = state->attrset.find(RGW_ATTR_ID_TAG); | |
5897 | if (iter != state->attrset.end()) { | |
5898 | iter->second = state->obj_tag; | |
5899 | } | |
5900 | } | |
5901 | ||
5902 | return 0; | |
5903 | } | |
5904 | ||
5905 | int RGWRados::Object::Read::prepare(optional_yield y, const DoutPrefixProvider *dpp) | |
5906 | { | |
5907 | RGWRados *store = source->get_store(); | |
5908 | CephContext *cct = store->ctx(); | |
5909 | ||
5910 | bufferlist etag; | |
5911 | ||
5912 | map<string, bufferlist>::iterator iter; | |
5913 | ||
5914 | RGWObjState *astate; | |
5915 | int r = source->get_state(dpp, &astate, true, y); | |
5916 | if (r < 0) | |
5917 | return r; | |
5918 | ||
5919 | if (!astate->exists) { | |
5920 | return -ENOENT; | |
5921 | } | |
5922 | ||
5923 | const RGWBucketInfo& bucket_info = source->get_bucket_info(); | |
5924 | ||
5925 | state.obj = astate->obj; | |
5926 | store->obj_to_raw(bucket_info.placement_rule, state.obj, &state.head_obj); | |
5927 | ||
5928 | state.cur_pool = state.head_obj.pool; | |
5929 | state.cur_ioctx = &state.io_ctxs[state.cur_pool]; | |
5930 | ||
5931 | r = store->get_obj_head_ioctx(dpp, bucket_info, state.obj, state.cur_ioctx); | |
5932 | if (r < 0) { | |
5933 | return r; | |
5934 | } | |
5935 | if (params.target_obj) { | |
5936 | *params.target_obj = state.obj; | |
5937 | } | |
5938 | if (params.attrs) { | |
5939 | *params.attrs = astate->attrset; | |
5940 | if (cct->_conf->subsys.should_gather<ceph_subsys_rgw, 20>()) { | |
5941 | for (iter = params.attrs->begin(); iter != params.attrs->end(); ++iter) { | |
5942 | ldpp_dout(dpp, 20) << "Read xattr rgw_rados: " << iter->first << dendl; | |
5943 | } | |
5944 | } | |
5945 | } | |
5946 | ||
5947 | /* Convert all times go GMT to make them compatible */ | |
5948 | if (conds.mod_ptr || conds.unmod_ptr) { | |
5949 | obj_time_weight src_weight; | |
5950 | src_weight.init(astate); | |
5951 | src_weight.high_precision = conds.high_precision_time; | |
5952 | ||
5953 | obj_time_weight dest_weight; | |
5954 | dest_weight.high_precision = conds.high_precision_time; | |
5955 | ||
5956 | if (conds.mod_ptr && !conds.if_nomatch) { | |
5957 | dest_weight.init(*conds.mod_ptr, conds.mod_zone_id, conds.mod_pg_ver); | |
5958 | ldpp_dout(dpp, 10) << "If-Modified-Since: " << dest_weight << " Last-Modified: " << src_weight << dendl; | |
5959 | if (!(dest_weight < src_weight)) { | |
5960 | return -ERR_NOT_MODIFIED; | |
5961 | } | |
5962 | } | |
5963 | ||
5964 | if (conds.unmod_ptr && !conds.if_match) { | |
5965 | dest_weight.init(*conds.unmod_ptr, conds.mod_zone_id, conds.mod_pg_ver); | |
5966 | ldpp_dout(dpp, 10) << "If-UnModified-Since: " << dest_weight << " Last-Modified: " << src_weight << dendl; | |
5967 | if (dest_weight < src_weight) { | |
5968 | return -ERR_PRECONDITION_FAILED; | |
5969 | } | |
5970 | } | |
5971 | } | |
5972 | if (conds.if_match || conds.if_nomatch) { | |
5973 | r = get_attr(dpp, RGW_ATTR_ETAG, etag, y); | |
5974 | if (r < 0) | |
5975 | return r; | |
5976 | ||
5977 | if (conds.if_match) { | |
5978 | string if_match_str = rgw_string_unquote(conds.if_match); | |
5979 | ldpp_dout(dpp, 10) << "ETag: " << string(etag.c_str(), etag.length()) << " " << " If-Match: " << if_match_str << dendl; | |
5980 | if (if_match_str.compare(0, etag.length(), etag.c_str(), etag.length()) != 0) { | |
5981 | return -ERR_PRECONDITION_FAILED; | |
5982 | } | |
5983 | } | |
5984 | ||
5985 | if (conds.if_nomatch) { | |
5986 | string if_nomatch_str = rgw_string_unquote(conds.if_nomatch); | |
5987 | ldpp_dout(dpp, 10) << "ETag: " << string(etag.c_str(), etag.length()) << " " << " If-NoMatch: " << if_nomatch_str << dendl; | |
5988 | if (if_nomatch_str.compare(0, etag.length(), etag.c_str(), etag.length()) == 0) { | |
5989 | return -ERR_NOT_MODIFIED; | |
5990 | } | |
5991 | } | |
5992 | } | |
5993 | ||
5994 | if (params.obj_size) | |
5995 | *params.obj_size = astate->size; | |
5996 | if (params.lastmod) | |
5997 | *params.lastmod = astate->mtime; | |
5998 | ||
5999 | return 0; | |
6000 | } | |
6001 | ||
6002 | int RGWRados::Object::Read::range_to_ofs(uint64_t obj_size, int64_t &ofs, int64_t &end) | |
6003 | { | |
6004 | if (ofs < 0) { | |
6005 | ofs += obj_size; | |
6006 | if (ofs < 0) | |
6007 | ofs = 0; | |
6008 | end = obj_size - 1; | |
6009 | } else if (end < 0) { | |
6010 | end = obj_size - 1; | |
6011 | } | |
6012 | ||
6013 | if (obj_size > 0) { | |
6014 | if (ofs >= (off_t)obj_size) { | |
6015 | return -ERANGE; | |
6016 | } | |
6017 | if (end >= (off_t)obj_size) { | |
6018 | end = obj_size - 1; | |
6019 | } | |
6020 | } | |
6021 | return 0; | |
6022 | } | |
6023 | ||
6024 | int RGWRados::Bucket::UpdateIndex::guard_reshard(const DoutPrefixProvider *dpp, BucketShard **pbs, std::function<int(BucketShard *)> call) | |
6025 | { | |
6026 | RGWRados *store = target->get_store(); | |
6027 | BucketShard *bs; | |
6028 | int r; | |
6029 | ||
6030 | #define NUM_RESHARD_RETRIES 10 | |
6031 | for (int i = 0; i < NUM_RESHARD_RETRIES; ++i) { | |
6032 | int ret = get_bucket_shard(&bs, dpp); | |
6033 | if (ret < 0) { | |
6034 | ldpp_dout(dpp, 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
6035 | return ret; | |
6036 | } | |
6037 | r = call(bs); | |
6038 | if (r != -ERR_BUSY_RESHARDING) { | |
6039 | break; | |
6040 | } | |
6041 | ldpp_dout(dpp, 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl; | |
6042 | string new_bucket_id; | |
6043 | r = store->block_while_resharding(bs, &new_bucket_id, | |
6044 | target->bucket_info, null_yield, dpp); | |
6045 | if (r == -ERR_BUSY_RESHARDING) { | |
6046 | continue; | |
6047 | } | |
6048 | if (r < 0) { | |
6049 | return r; | |
6050 | } | |
6051 | ldpp_dout(dpp, 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id << dendl; | |
6052 | i = 0; /* resharding is finished, make sure we can retry */ | |
6053 | r = target->update_bucket_id(new_bucket_id, dpp); | |
6054 | if (r < 0) { | |
6055 | ldpp_dout(dpp, 0) << "ERROR: update_bucket_id() new_bucket_id=" << new_bucket_id << " returned r=" << r << dendl; | |
6056 | return r; | |
6057 | } | |
6058 | invalidate_bs(); | |
6059 | } // for loop | |
6060 | ||
6061 | if (r < 0) { | |
6062 | return r; | |
6063 | } | |
6064 | ||
6065 | if (pbs) { | |
6066 | *pbs = bs; | |
6067 | } | |
6068 | ||
6069 | return 0; | |
6070 | } | |
6071 | ||
6072 | int RGWRados::Bucket::UpdateIndex::prepare(const DoutPrefixProvider *dpp, RGWModifyOp op, const string *write_tag, optional_yield y) | |
6073 | { | |
6074 | if (blind) { | |
6075 | return 0; | |
6076 | } | |
6077 | RGWRados *store = target->get_store(); | |
6078 | ||
6079 | if (write_tag && write_tag->length()) { | |
6080 | optag = string(write_tag->c_str(), write_tag->length()); | |
6081 | } else { | |
6082 | if (optag.empty()) { | |
6083 | append_rand_alpha(store->ctx(), optag, optag, 32); | |
6084 | } | |
6085 | } | |
6086 | ||
6087 | int r = guard_reshard(dpp, nullptr, [&](BucketShard *bs) -> int { | |
6088 | return store->cls_obj_prepare_op(dpp, *bs, op, optag, obj, bilog_flags, y, zones_trace); | |
6089 | }); | |
6090 | ||
6091 | if (r < 0) { | |
6092 | return r; | |
6093 | } | |
6094 | prepared = true; | |
6095 | ||
6096 | return 0; | |
6097 | } | |
6098 | ||
6099 | int RGWRados::Bucket::UpdateIndex::complete(const DoutPrefixProvider *dpp, int64_t poolid, uint64_t epoch, | |
6100 | uint64_t size, uint64_t accounted_size, | |
6101 | ceph::real_time& ut, const string& etag, | |
6102 | const string& content_type, const string& storage_class, | |
6103 | bufferlist *acl_bl, | |
6104 | RGWObjCategory category, | |
6105 | list<rgw_obj_index_key> *remove_objs, const string *user_data, | |
6106 | bool appendable) | |
6107 | { | |
6108 | if (blind) { | |
6109 | return 0; | |
6110 | } | |
6111 | RGWRados *store = target->get_store(); | |
6112 | BucketShard *bs; | |
6113 | ||
6114 | int ret = get_bucket_shard(&bs, dpp); | |
6115 | if (ret < 0) { | |
6116 | ldpp_dout(dpp, 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
6117 | return ret; | |
6118 | } | |
6119 | ||
6120 | rgw_bucket_dir_entry ent; | |
6121 | obj.key.get_index_key(&ent.key); | |
6122 | ent.meta.size = size; | |
6123 | ent.meta.accounted_size = accounted_size; | |
6124 | ent.meta.mtime = ut; | |
6125 | ent.meta.etag = etag; | |
6126 | ent.meta.storage_class = storage_class; | |
6127 | if (user_data) | |
6128 | ent.meta.user_data = *user_data; | |
6129 | ||
6130 | ACLOwner owner; | |
6131 | if (acl_bl && acl_bl->length()) { | |
6132 | int ret = store->decode_policy(*acl_bl, &owner); | |
6133 | if (ret < 0) { | |
6134 | ldpp_dout(dpp, 0) << "WARNING: could not decode policy ret=" << ret << dendl; | |
6135 | } | |
6136 | } | |
6137 | ent.meta.owner = owner.get_id().to_str(); | |
6138 | ent.meta.owner_display_name = owner.get_display_name(); | |
6139 | ent.meta.content_type = content_type; | |
6140 | ent.meta.appendable = appendable; | |
6141 | ||
6142 | ret = store->cls_obj_complete_add(*bs, obj, optag, poolid, epoch, ent, category, remove_objs, bilog_flags, zones_trace); | |
6143 | ||
6144 | int r = store->svc.datalog_rados->add_entry(dpp, target->bucket_info, bs->shard_id); | |
6145 | if (r < 0) { | |
6146 | ldpp_dout(dpp, -1) << "ERROR: failed writing data log" << dendl; | |
6147 | } | |
6148 | ||
6149 | return ret; | |
6150 | } | |
6151 | ||
6152 | int RGWRados::Bucket::UpdateIndex::complete_del(const DoutPrefixProvider *dpp, | |
6153 | int64_t poolid, uint64_t epoch, | |
6154 | real_time& removed_mtime, | |
6155 | list<rgw_obj_index_key> *remove_objs) | |
6156 | { | |
6157 | if (blind) { | |
6158 | return 0; | |
6159 | } | |
6160 | RGWRados *store = target->get_store(); | |
6161 | BucketShard *bs; | |
6162 | ||
6163 | int ret = get_bucket_shard(&bs, dpp); | |
6164 | if (ret < 0) { | |
6165 | ldpp_dout(dpp, 5) << "failed to get BucketShard object: ret=" << ret << dendl; | |
6166 | return ret; | |
6167 | } | |
6168 | ||
6169 | ret = store->cls_obj_complete_del(*bs, optag, poolid, epoch, obj, removed_mtime, remove_objs, bilog_flags, zones_trace); | |
6170 | ||
6171 | int r = store->svc.datalog_rados->add_entry(dpp, target->bucket_info, bs->shard_id); | |
6172 | if (r < 0) { | |
6173 | ldpp_dout(dpp, -1) << "ERROR: failed writing data log" << dendl; | |
6174 | } | |
6175 | ||
6176 | return ret; | |
6177 | } | |
6178 | ||
6179 | ||
6180 | int RGWRados::Bucket::UpdateIndex::cancel(const DoutPrefixProvider *dpp) | |
6181 | { | |
6182 | if (blind) { | |
6183 | return 0; | |
6184 | } | |
6185 | RGWRados *store = target->get_store(); | |
6186 | BucketShard *bs; | |
6187 | ||
6188 | int ret = guard_reshard(dpp, &bs, [&](BucketShard *bs) -> int { | |
6189 | return store->cls_obj_complete_cancel(*bs, optag, obj, bilog_flags, zones_trace); | |
6190 | }); | |
6191 | ||
6192 | /* | |
6193 | * need to update data log anyhow, so that whoever follows needs to update its internal markers | |
6194 | * for following the specific bucket shard log. Otherwise they end up staying behind, and users | |
6195 | * have no way to tell that they're all caught up | |
6196 | */ | |
6197 | int r = store->svc.datalog_rados->add_entry(dpp, target->bucket_info, bs->shard_id); | |
6198 | if (r < 0) { | |
6199 | ldpp_dout(dpp, -1) << "ERROR: failed writing data log" << dendl; | |
6200 | } | |
6201 | ||
6202 | return ret; | |
6203 | } | |
6204 | ||
6205 | int RGWRados::Object::Read::read(int64_t ofs, int64_t end, bufferlist& bl, optional_yield y, const DoutPrefixProvider *dpp) | |
6206 | { | |
6207 | RGWRados *store = source->get_store(); | |
6208 | ||
6209 | rgw_raw_obj read_obj; | |
6210 | uint64_t read_ofs = ofs; | |
6211 | uint64_t len, read_len; | |
6212 | bool reading_from_head = true; | |
6213 | ObjectReadOperation op; | |
6214 | ||
6215 | bool merge_bl = false; | |
6216 | bufferlist *pbl = &bl; | |
6217 | bufferlist read_bl; | |
6218 | uint64_t max_chunk_size; | |
6219 | ||
6220 | RGWObjState *astate; | |
6221 | int r = source->get_state(dpp, &astate, true, y); | |
6222 | if (r < 0) | |
6223 | return r; | |
6224 | ||
6225 | if (astate->size == 0) { | |
6226 | end = 0; | |
6227 | } else if (end >= (int64_t)astate->size) { | |
6228 | end = astate->size - 1; | |
6229 | } | |
6230 | ||
6231 | if (end < 0) | |
6232 | len = 0; | |
6233 | else | |
6234 | len = end - ofs + 1; | |
6235 | ||
6236 | if (astate->manifest && astate->manifest->has_tail()) { | |
6237 | /* now get the relevant object part */ | |
6238 | RGWObjManifest::obj_iterator iter = astate->manifest->obj_find(dpp, ofs); | |
6239 | ||
6240 | uint64_t stripe_ofs = iter.get_stripe_ofs(); | |
6241 | read_obj = iter.get_location().get_raw_obj(store->store); | |
6242 | len = std::min(len, iter.get_stripe_size() - (ofs - stripe_ofs)); | |
6243 | read_ofs = iter.location_ofs() + (ofs - stripe_ofs); | |
6244 | reading_from_head = (read_obj == state.head_obj); | |
6245 | } else { | |
6246 | read_obj = state.head_obj; | |
6247 | } | |
6248 | ||
6249 | r = store->get_max_chunk_size(read_obj.pool, &max_chunk_size, dpp); | |
6250 | if (r < 0) { | |
6251 | ldpp_dout(dpp, 0) << "ERROR: failed to get max_chunk_size() for pool " << read_obj.pool << dendl; | |
6252 | return r; | |
6253 | } | |
6254 | ||
6255 | if (len > max_chunk_size) | |
6256 | len = max_chunk_size; | |
6257 | ||
6258 | ||
6259 | read_len = len; | |
6260 | ||
6261 | if (reading_from_head) { | |
6262 | /* only when reading from the head object do we need to do the atomic test */ | |
6263 | r = store->append_atomic_test(dpp, &source->get_ctx(), source->get_bucket_info(), state.obj, op, &astate, y); | |
6264 | if (r < 0) | |
6265 | return r; | |
6266 | ||
6267 | if (astate && astate->prefetch_data) { | |
6268 | if (!ofs && astate->data.length() >= len) { | |
6269 | bl = astate->data; | |
6270 | return bl.length(); | |
6271 | } | |
6272 | ||
6273 | if (ofs < astate->data.length()) { | |
6274 | unsigned copy_len = std::min((uint64_t)astate->data.length() - ofs, len); | |
6275 | astate->data.begin(ofs).copy(copy_len, bl); | |
6276 | read_len -= copy_len; | |
6277 | read_ofs += copy_len; | |
6278 | if (!read_len) | |
6279 | return bl.length(); | |
6280 | ||
6281 | merge_bl = true; | |
6282 | pbl = &read_bl; | |
6283 | } | |
6284 | } | |
6285 | } | |
6286 | ||
6287 | ldpp_dout(dpp, 20) << "rados->read obj-ofs=" << ofs << " read_ofs=" << read_ofs << " read_len=" << read_len << dendl; | |
6288 | op.read(read_ofs, read_len, pbl, NULL); | |
6289 | ||
6290 | if (state.cur_pool != read_obj.pool) { | |
6291 | auto iter = state.io_ctxs.find(read_obj.pool); | |
6292 | if (iter == state.io_ctxs.end()) { | |
6293 | state.cur_ioctx = &state.io_ctxs[read_obj.pool]; | |
6294 | r = store->open_pool_ctx(dpp, read_obj.pool, *state.cur_ioctx, false); | |
6295 | if (r < 0) { | |
6296 | ldpp_dout(dpp, 20) << "ERROR: failed to open pool context for pool=" << read_obj.pool << " r=" << r << dendl; | |
6297 | return r; | |
6298 | } | |
6299 | } else { | |
6300 | state.cur_ioctx = &iter->second; | |
6301 | } | |
6302 | state.cur_pool = read_obj.pool; | |
6303 | } | |
6304 | ||
6305 | state.cur_ioctx->locator_set_key(read_obj.loc); | |
6306 | ||
6307 | r = state.cur_ioctx->operate(read_obj.oid, &op, NULL); | |
6308 | ldpp_dout(dpp, 20) << "rados->read r=" << r << " bl.length=" << bl.length() << dendl; | |
6309 | ||
6310 | if (r < 0) { | |
6311 | return r; | |
6312 | } | |
6313 | ||
6314 | if (merge_bl) { | |
6315 | bl.append(read_bl); | |
6316 | } | |
6317 | ||
6318 | return bl.length(); | |
6319 | } | |
6320 | ||
6321 | struct get_obj_data { | |
6322 | RGWRados* store; | |
6323 | RGWGetDataCB* client_cb; | |
6324 | rgw::Aio* aio; | |
6325 | uint64_t offset; // next offset to write to client | |
6326 | rgw::AioResultList completed; // completed read results, sorted by offset | |
6327 | optional_yield yield; | |
6328 | ||
6329 | get_obj_data(RGWRados* store, RGWGetDataCB* cb, rgw::Aio* aio, | |
6330 | uint64_t offset, optional_yield yield) | |
6331 | : store(store), client_cb(cb), aio(aio), offset(offset), yield(yield) {} | |
6332 | ||
6333 | int flush(rgw::AioResultList&& results) { | |
6334 | int r = rgw::check_for_errors(results); | |
6335 | if (r < 0) { | |
6336 | return r; | |
6337 | } | |
6338 | ||
6339 | auto cmp = [](const auto& lhs, const auto& rhs) { return lhs.id < rhs.id; }; | |
6340 | results.sort(cmp); // merge() requires results to be sorted first | |
6341 | completed.merge(results, cmp); // merge results in sorted order | |
6342 | ||
6343 | while (!completed.empty() && completed.front().id == offset) { | |
6344 | auto bl = std::move(completed.front().data); | |
6345 | completed.pop_front_and_dispose(std::default_delete<rgw::AioResultEntry>{}); | |
6346 | ||
6347 | offset += bl.length(); | |
6348 | int r = client_cb->handle_data(bl, 0, bl.length()); | |
6349 | if (r < 0) { | |
6350 | return r; | |
6351 | } | |
6352 | } | |
6353 | return 0; | |
6354 | } | |
6355 | ||
6356 | void cancel() { | |
6357 | // wait for all completions to drain and ignore the results | |
6358 | aio->drain(); | |
6359 | } | |
6360 | ||
6361 | int drain() { | |
6362 | auto c = aio->wait(); | |
6363 | while (!c.empty()) { | |
6364 | int r = flush(std::move(c)); | |
6365 | if (r < 0) { | |
6366 | cancel(); | |
6367 | return r; | |
6368 | } | |
6369 | c = aio->wait(); | |
6370 | } | |
6371 | return flush(std::move(c)); | |
6372 | } | |
6373 | }; | |
6374 | ||
6375 | static int _get_obj_iterate_cb(const DoutPrefixProvider *dpp, | |
6376 | const rgw_raw_obj& read_obj, off_t obj_ofs, | |
6377 | off_t read_ofs, off_t len, bool is_head_obj, | |
6378 | RGWObjState *astate, void *arg) | |
6379 | { | |
6380 | struct get_obj_data *d = (struct get_obj_data *)arg; | |
6381 | ||
6382 | return d->store->get_obj_iterate_cb(dpp, read_obj, obj_ofs, read_ofs, len, | |
6383 | is_head_obj, astate, arg); | |
6384 | } | |
6385 | ||
6386 | int RGWRados::get_obj_iterate_cb(const DoutPrefixProvider *dpp, | |
6387 | const rgw_raw_obj& read_obj, off_t obj_ofs, | |
6388 | off_t read_ofs, off_t len, bool is_head_obj, | |
6389 | RGWObjState *astate, void *arg) | |
6390 | { | |
6391 | ObjectReadOperation op; | |
6392 | struct get_obj_data *d = (struct get_obj_data *)arg; | |
6393 | string oid, key; | |
6394 | ||
6395 | if (is_head_obj) { | |
6396 | /* only when reading from the head object do we need to do the atomic test */ | |
6397 | int r = append_atomic_test(dpp, astate, op); | |
6398 | if (r < 0) | |
6399 | return r; | |
6400 | ||
6401 | if (astate && | |
6402 | obj_ofs < astate->data.length()) { | |
6403 | unsigned chunk_len = std::min((uint64_t)astate->data.length() - obj_ofs, (uint64_t)len); | |
6404 | ||
6405 | r = d->client_cb->handle_data(astate->data, obj_ofs, chunk_len); | |
6406 | if (r < 0) | |
6407 | return r; | |
6408 | ||
6409 | len -= chunk_len; | |
6410 | d->offset += chunk_len; | |
6411 | read_ofs += chunk_len; | |
6412 | obj_ofs += chunk_len; | |
6413 | if (!len) | |
6414 | return 0; | |
6415 | } | |
6416 | } | |
6417 | ||
6418 | auto obj = d->store->svc.rados->obj(read_obj); | |
6419 | int r = obj.open(dpp); | |
6420 | if (r < 0) { | |
6421 | ldpp_dout(dpp, 4) << "failed to open rados context for " << read_obj << dendl; | |
6422 | return r; | |
6423 | } | |
6424 | ||
6425 | ldpp_dout(dpp, 20) << "rados->get_obj_iterate_cb oid=" << read_obj.oid << " obj-ofs=" << obj_ofs << " read_ofs=" << read_ofs << " len=" << len << dendl; | |
6426 | op.read(read_ofs, len, nullptr, nullptr); | |
6427 | ||
6428 | const uint64_t cost = len; | |
6429 | const uint64_t id = obj_ofs; // use logical object offset for sorting replies | |
6430 | ||
6431 | auto completed = d->aio->get(obj, rgw::Aio::librados_op(std::move(op), d->yield), cost, id); | |
6432 | ||
6433 | return d->flush(std::move(completed)); | |
6434 | } | |
6435 | ||
6436 | int RGWRados::Object::Read::iterate(const DoutPrefixProvider *dpp, int64_t ofs, int64_t end, RGWGetDataCB *cb, | |
6437 | optional_yield y) | |
6438 | { | |
6439 | RGWRados *store = source->get_store(); | |
6440 | CephContext *cct = store->ctx(); | |
6441 | RGWObjectCtx& obj_ctx = source->get_ctx(); | |
6442 | const uint64_t chunk_size = cct->_conf->rgw_get_obj_max_req_size; | |
6443 | const uint64_t window_size = cct->_conf->rgw_get_obj_window_size; | |
6444 | ||
6445 | auto aio = rgw::make_throttle(window_size, y); | |
6446 | get_obj_data data(store, cb, &*aio, ofs, y); | |
6447 | ||
6448 | int r = store->iterate_obj(dpp, obj_ctx, source->get_bucket_info(), state.obj, | |
6449 | ofs, end, chunk_size, _get_obj_iterate_cb, &data, y); | |
6450 | if (r < 0) { | |
6451 | ldpp_dout(dpp, 0) << "iterate_obj() failed with " << r << dendl; | |
6452 | data.cancel(); // drain completions without writing back to client | |
6453 | return r; | |
6454 | } | |
6455 | ||
6456 | return data.drain(); | |
6457 | } | |
6458 | ||
6459 | int RGWRados::iterate_obj(const DoutPrefixProvider *dpp, RGWObjectCtx& obj_ctx, | |
6460 | const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
6461 | off_t ofs, off_t end, uint64_t max_chunk_size, | |
6462 | iterate_obj_cb cb, void *arg, optional_yield y) | |
6463 | { | |
6464 | rgw_raw_obj head_obj; | |
6465 | rgw_raw_obj read_obj; | |
6466 | uint64_t read_ofs = ofs; | |
6467 | uint64_t len; | |
6468 | bool reading_from_head = true; | |
6469 | RGWObjState *astate = NULL; | |
6470 | ||
6471 | obj_to_raw(bucket_info.placement_rule, obj, &head_obj); | |
6472 | ||
6473 | int r = get_obj_state(dpp, &obj_ctx, bucket_info, obj, &astate, false, y); | |
6474 | if (r < 0) { | |
6475 | return r; | |
6476 | } | |
6477 | ||
6478 | if (end < 0) | |
6479 | len = 0; | |
6480 | else | |
6481 | len = end - ofs + 1; | |
6482 | ||
6483 | if (astate->manifest) { | |
6484 | /* now get the relevant object stripe */ | |
6485 | RGWObjManifest::obj_iterator iter = astate->manifest->obj_find(dpp, ofs); | |
6486 | ||
6487 | RGWObjManifest::obj_iterator obj_end = astate->manifest->obj_end(dpp); | |
6488 | ||
6489 | for (; iter != obj_end && ofs <= end; ++iter) { | |
6490 | off_t stripe_ofs = iter.get_stripe_ofs(); | |
6491 | off_t next_stripe_ofs = stripe_ofs + iter.get_stripe_size(); | |
6492 | ||
6493 | while (ofs < next_stripe_ofs && ofs <= end) { | |
6494 | read_obj = iter.get_location().get_raw_obj(store); | |
6495 | uint64_t read_len = std::min(len, iter.get_stripe_size() - (ofs - stripe_ofs)); | |
6496 | read_ofs = iter.location_ofs() + (ofs - stripe_ofs); | |
6497 | ||
6498 | if (read_len > max_chunk_size) { | |
6499 | read_len = max_chunk_size; | |
6500 | } | |
6501 | ||
6502 | reading_from_head = (read_obj == head_obj); | |
6503 | r = cb(dpp, read_obj, ofs, read_ofs, read_len, reading_from_head, astate, arg); | |
6504 | if (r < 0) { | |
6505 | return r; | |
6506 | } | |
6507 | ||
6508 | len -= read_len; | |
6509 | ofs += read_len; | |
6510 | } | |
6511 | } | |
6512 | } else { | |
6513 | while (ofs <= end) { | |
6514 | read_obj = head_obj; | |
6515 | uint64_t read_len = std::min(len, max_chunk_size); | |
6516 | ||
6517 | r = cb(dpp, read_obj, ofs, ofs, read_len, reading_from_head, astate, arg); | |
6518 | if (r < 0) { | |
6519 | return r; | |
6520 | } | |
6521 | ||
6522 | len -= read_len; | |
6523 | ofs += read_len; | |
6524 | } | |
6525 | } | |
6526 | ||
6527 | return 0; | |
6528 | } | |
6529 | ||
6530 | int RGWRados::obj_operate(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj, ObjectWriteOperation *op) | |
6531 | { | |
6532 | rgw_rados_ref ref; | |
6533 | int r = get_obj_head_ref(dpp, bucket_info, obj, &ref); | |
6534 | if (r < 0) { | |
6535 | return r; | |
6536 | } | |
6537 | ||
6538 | return rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, op, null_yield); | |
6539 | } | |
6540 | ||
6541 | int RGWRados::obj_operate(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj, ObjectReadOperation *op) | |
6542 | { | |
6543 | rgw_rados_ref ref; | |
6544 | int r = get_obj_head_ref(dpp, bucket_info, obj, &ref); | |
6545 | if (r < 0) { | |
6546 | return r; | |
6547 | } | |
6548 | ||
6549 | bufferlist outbl; | |
6550 | ||
6551 | return rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, op, &outbl, null_yield); | |
6552 | } | |
6553 | ||
6554 | int RGWRados::olh_init_modification_impl(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& olh_obj, string *op_tag) | |
6555 | { | |
6556 | ObjectWriteOperation op; | |
6557 | ||
6558 | ceph_assert(olh_obj.key.instance.empty()); | |
6559 | ||
6560 | bool has_tag = (state.exists && has_olh_tag(state.attrset)); | |
6561 | ||
6562 | if (!state.exists) { | |
6563 | op.create(true); | |
6564 | } else { | |
6565 | op.assert_exists(); | |
6566 | struct timespec mtime_ts = real_clock::to_timespec(state.mtime); | |
6567 | op.mtime2(&mtime_ts); | |
6568 | } | |
6569 | ||
6570 | /* | |
6571 | * 3 possible cases: olh object doesn't exist, it exists as an olh, it exists as a regular object. | |
6572 | * If it exists as a regular object we'll need to transform it into an olh. We'll do it in two | |
6573 | * steps, first change its tag and set the olh pending attrs. Once write is done we'll need to | |
6574 | * truncate it, remove extra attrs, and send it to the garbage collection. The bucket index olh | |
6575 | * log will reflect that. | |
6576 | * | |
6577 | * Need to generate separate olh and obj tags, as olh can be colocated with object data. obj_tag | |
6578 | * is used for object data instance, olh_tag for olh instance. | |
6579 | */ | |
6580 | if (has_tag) { | |
6581 | /* guard against racing writes */ | |
6582 | bucket_index_guard_olh_op(dpp, state, op); | |
6583 | } | |
6584 | ||
6585 | if (!has_tag) { | |
6586 | /* obj tag */ | |
6587 | string obj_tag = gen_rand_alphanumeric_lower(cct, 32); | |
6588 | ||
6589 | bufferlist bl; | |
6590 | bl.append(obj_tag.c_str(), obj_tag.size()); | |
6591 | op.setxattr(RGW_ATTR_ID_TAG, bl); | |
6592 | ||
6593 | state.attrset[RGW_ATTR_ID_TAG] = bl; | |
6594 | state.obj_tag = bl; | |
6595 | ||
6596 | /* olh tag */ | |
6597 | string olh_tag = gen_rand_alphanumeric_lower(cct, 32); | |
6598 | ||
6599 | bufferlist olh_bl; | |
6600 | olh_bl.append(olh_tag.c_str(), olh_tag.size()); | |
6601 | op.setxattr(RGW_ATTR_OLH_ID_TAG, olh_bl); | |
6602 | ||
6603 | state.attrset[RGW_ATTR_OLH_ID_TAG] = olh_bl; | |
6604 | state.olh_tag = olh_bl; | |
6605 | state.is_olh = true; | |
6606 | ||
6607 | bufferlist verbl; | |
6608 | op.setxattr(RGW_ATTR_OLH_VER, verbl); | |
6609 | } | |
6610 | ||
6611 | bufferlist bl; | |
6612 | RGWOLHPendingInfo pending_info; | |
6613 | pending_info.time = real_clock::now(); | |
6614 | encode(pending_info, bl); | |
6615 | ||
6616 | #define OLH_PENDING_TAG_LEN 32 | |
6617 | /* tag will start with current time epoch, this so that entries are sorted by time */ | |
6618 | char buf[32]; | |
6619 | utime_t ut(pending_info.time); | |
6620 | snprintf(buf, sizeof(buf), "%016llx", (unsigned long long)ut.sec()); | |
6621 | *op_tag = buf; | |
6622 | ||
6623 | string s = gen_rand_alphanumeric_lower(cct, OLH_PENDING_TAG_LEN - op_tag->size()); | |
6624 | ||
6625 | op_tag->append(s); | |
6626 | ||
6627 | string attr_name = RGW_ATTR_OLH_PENDING_PREFIX; | |
6628 | attr_name.append(*op_tag); | |
6629 | ||
6630 | op.setxattr(attr_name.c_str(), bl); | |
6631 | ||
6632 | int ret = obj_operate(dpp, bucket_info, olh_obj, &op); | |
6633 | if (ret < 0) { | |
6634 | return ret; | |
6635 | } | |
6636 | ||
6637 | state.exists = true; | |
6638 | state.attrset[attr_name] = bl; | |
6639 | ||
6640 | return 0; | |
6641 | } | |
6642 | ||
6643 | int RGWRados::olh_init_modification(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj, string *op_tag) | |
6644 | { | |
6645 | int ret; | |
6646 | ||
6647 | ret = olh_init_modification_impl(dpp, bucket_info, state, obj, op_tag); | |
6648 | if (ret == -EEXIST) { | |
6649 | ret = -ECANCELED; | |
6650 | } | |
6651 | ||
6652 | return ret; | |
6653 | } | |
6654 | ||
6655 | int RGWRados::guard_reshard(const DoutPrefixProvider *dpp, | |
6656 | BucketShard *bs, | |
6657 | const rgw_obj& obj_instance, | |
6658 | const RGWBucketInfo& bucket_info, | |
6659 | std::function<int(BucketShard *)> call) | |
6660 | { | |
6661 | rgw_obj obj; | |
6662 | const rgw_obj *pobj = &obj_instance; | |
6663 | int r; | |
6664 | ||
6665 | for (int i = 0; i < NUM_RESHARD_RETRIES; ++i) { | |
6666 | r = bs->init(pobj->bucket, *pobj, nullptr /* no RGWBucketInfo */, dpp); | |
6667 | if (r < 0) { | |
6668 | ldpp_dout(dpp, 5) << "bs.init() returned ret=" << r << dendl; | |
6669 | return r; | |
6670 | } | |
6671 | r = call(bs); | |
6672 | if (r != -ERR_BUSY_RESHARDING) { | |
6673 | break; | |
6674 | } | |
6675 | ldpp_dout(dpp, 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl; | |
6676 | string new_bucket_id; | |
6677 | r = block_while_resharding(bs, &new_bucket_id, bucket_info, null_yield, dpp); | |
6678 | if (r == -ERR_BUSY_RESHARDING) { | |
6679 | continue; | |
6680 | } | |
6681 | if (r < 0) { | |
6682 | return r; | |
6683 | } | |
6684 | ldpp_dout(dpp, 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id << dendl; | |
6685 | i = 0; /* resharding is finished, make sure we can retry */ | |
6686 | ||
6687 | obj = *pobj; | |
6688 | obj.bucket.update_bucket_id(new_bucket_id); | |
6689 | pobj = &obj; | |
6690 | } // for loop | |
6691 | ||
6692 | if (r < 0) { | |
6693 | return r; | |
6694 | } | |
6695 | ||
6696 | return 0; | |
6697 | } | |
6698 | ||
6699 | int RGWRados::block_while_resharding(RGWRados::BucketShard *bs, | |
6700 | string *new_bucket_id, | |
6701 | const RGWBucketInfo& bucket_info, | |
6702 | optional_yield y, | |
6703 | const DoutPrefixProvider *dpp) | |
6704 | { | |
6705 | int ret = 0; | |
6706 | cls_rgw_bucket_instance_entry entry; | |
6707 | ||
6708 | // since we want to run this recovery code from two distinct places, | |
6709 | // let's just put it in a lambda so we can easily re-use; if the | |
6710 | // lambda successfully fetches a new bucket id, it sets | |
6711 | // new_bucket_id and returns 0, otherwise it returns a negative | |
6712 | // error code | |
6713 | auto fetch_new_bucket_id = | |
6714 | [this, &bucket_info, dpp](const std::string& log_tag, | |
6715 | std::string* new_bucket_id) -> int { | |
6716 | RGWBucketInfo fresh_bucket_info = bucket_info; | |
6717 | int ret = try_refresh_bucket_info(fresh_bucket_info, nullptr, dpp); | |
6718 | if (ret < 0) { | |
6719 | ldpp_dout(dpp, 0) << __func__ << | |
6720 | " ERROR: failed to refresh bucket info after reshard at " << | |
6721 | log_tag << ": " << cpp_strerror(-ret) << dendl; | |
6722 | return ret; | |
6723 | } | |
6724 | *new_bucket_id = fresh_bucket_info.bucket.bucket_id; | |
6725 | return 0; | |
6726 | }; | |
6727 | ||
6728 | constexpr int num_retries = 10; | |
6729 | for (int i = 1; i <= num_retries; i++) { // nb: 1-based for loop | |
6730 | auto& ref = bs->bucket_obj.get_ref(); | |
6731 | ret = cls_rgw_get_bucket_resharding(ref.pool.ioctx(), ref.obj.oid, &entry); | |
6732 | if (ret == -ENOENT) { | |
6733 | return fetch_new_bucket_id("get_bucket_resharding_failed", new_bucket_id); | |
6734 | } else if (ret < 0) { | |
6735 | ldpp_dout(dpp, 0) << __func__ << | |
6736 | " ERROR: failed to get bucket resharding : " << cpp_strerror(-ret) << | |
6737 | dendl; | |
6738 | return ret; | |
6739 | } | |
6740 | ||
6741 | if (!entry.resharding_in_progress()) { | |
6742 | return fetch_new_bucket_id("get_bucket_resharding_succeeded", | |
6743 | new_bucket_id); | |
6744 | } | |
6745 | ||
6746 | ldpp_dout(dpp, 20) << "NOTICE: reshard still in progress; " << | |
6747 | (i < num_retries ? "retrying" : "too many retries") << dendl; | |
6748 | ||
6749 | if (i == num_retries) { | |
6750 | break; | |
6751 | } | |
6752 | ||
6753 | // If bucket is erroneously marked as resharding (e.g., crash or | |
6754 | // other error) then fix it. If we can take the bucket reshard | |
6755 | // lock then it means no other resharding should be taking place, | |
6756 | // and we're free to clear the flags. | |
6757 | { | |
6758 | // since we expect to do this rarely, we'll do our work in a | |
6759 | // block and erase our work after each try | |
6760 | ||
6761 | RGWObjectCtx obj_ctx(this->store); | |
6762 | const rgw_bucket& b = bs->bucket; | |
6763 | std::string bucket_id = b.get_key(); | |
6764 | RGWBucketReshardLock reshard_lock(this->store, bucket_info, true); | |
6765 | ret = reshard_lock.lock(); | |
6766 | if (ret < 0) { | |
6767 | ldpp_dout(dpp, 20) << __func__ << | |
6768 | " INFO: failed to take reshard lock for bucket " << | |
6769 | bucket_id << "; expected if resharding underway" << dendl; | |
6770 | } else { | |
6771 | ldpp_dout(dpp, 10) << __func__ << | |
6772 | " INFO: was able to take reshard lock for bucket " << | |
6773 | bucket_id << dendl; | |
6774 | ret = RGWBucketReshard::clear_resharding(dpp, this->store, bucket_info); | |
6775 | if (ret < 0) { | |
6776 | reshard_lock.unlock(); | |
6777 | ldpp_dout(dpp, 0) << __func__ << | |
6778 | " ERROR: failed to clear resharding flags for bucket " << | |
6779 | bucket_id << dendl; | |
6780 | } else { | |
6781 | reshard_lock.unlock(); | |
6782 | ldpp_dout(dpp, 5) << __func__ << | |
6783 | " INFO: apparently successfully cleared resharding flags for " | |
6784 | "bucket " << bucket_id << dendl; | |
6785 | continue; // if we apparently succeed immediately test again | |
6786 | } // if clear resharding succeeded | |
6787 | } // if taking of lock succeeded | |
6788 | } // block to encapsulate recovery from incomplete reshard | |
6789 | ||
6790 | ret = reshard_wait->wait(y); | |
6791 | if (ret < 0) { | |
6792 | ldpp_dout(dpp, 0) << __func__ << | |
6793 | " ERROR: bucket is still resharding, please retry" << dendl; | |
6794 | return ret; | |
6795 | } | |
6796 | } // for loop | |
6797 | ||
6798 | ldpp_dout(dpp, 0) << __func__ << | |
6799 | " ERROR: bucket is still resharding, please retry" << dendl; | |
6800 | return -ERR_BUSY_RESHARDING; | |
6801 | } | |
6802 | ||
6803 | int RGWRados::bucket_index_link_olh(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, RGWObjState& olh_state, const rgw_obj& obj_instance, | |
6804 | bool delete_marker, | |
6805 | const string& op_tag, | |
6806 | struct rgw_bucket_dir_entry_meta *meta, | |
6807 | uint64_t olh_epoch, | |
6808 | real_time unmod_since, bool high_precision_time, | |
6809 | rgw_zone_set *_zones_trace, bool log_data_change) | |
6810 | { | |
6811 | rgw_rados_ref ref; | |
6812 | int r = get_obj_head_ref(dpp, bucket_info, obj_instance, &ref); | |
6813 | if (r < 0) { | |
6814 | return r; | |
6815 | } | |
6816 | ||
6817 | rgw_zone_set zones_trace; | |
6818 | if (_zones_trace) { | |
6819 | zones_trace = *_zones_trace; | |
6820 | } | |
6821 | zones_trace.insert(svc.zone->get_zone().id, bucket_info.bucket.get_key()); | |
6822 | ||
6823 | BucketShard bs(this); | |
6824 | ||
6825 | r = guard_reshard(dpp, &bs, obj_instance, bucket_info, | |
6826 | [&](BucketShard *bs) -> int { | |
6827 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), obj_instance.key.instance); | |
6828 | auto& ref = bs->bucket_obj.get_ref(); | |
6829 | librados::ObjectWriteOperation op; | |
6830 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
6831 | cls_rgw_bucket_link_olh(op, key, olh_state.olh_tag, | |
6832 | delete_marker, op_tag, meta, olh_epoch, | |
6833 | unmod_since, high_precision_time, | |
6834 | svc.zone->get_zone().log_data, zones_trace); | |
6835 | return rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
6836 | }); | |
6837 | if (r < 0) { | |
6838 | ldpp_dout(dpp, 20) << "rgw_rados_operate() after cls_rgw_bucket_link_olh() returned r=" << r << dendl; | |
6839 | return r; | |
6840 | } | |
6841 | ||
6842 | r = svc.datalog_rados->add_entry(dpp, bucket_info, bs.shard_id); | |
6843 | if (r < 0) { | |
6844 | ldpp_dout(dpp, 0) << "ERROR: failed writing data log" << dendl; | |
6845 | } | |
6846 | ||
6847 | return 0; | |
6848 | } | |
6849 | ||
6850 | void RGWRados::bucket_index_guard_olh_op(const DoutPrefixProvider *dpp, RGWObjState& olh_state, ObjectOperation& op) | |
6851 | { | |
6852 | ldpp_dout(dpp, 20) << __func__ << "(): olh_state.olh_tag=" << string(olh_state.olh_tag.c_str(), olh_state.olh_tag.length()) << dendl; | |
6853 | op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_state.olh_tag); | |
6854 | } | |
6855 | ||
6856 | int RGWRados::bucket_index_unlink_instance(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj_instance, | |
6857 | const string& op_tag, const string& olh_tag, uint64_t olh_epoch, rgw_zone_set *_zones_trace) | |
6858 | { | |
6859 | rgw_rados_ref ref; | |
6860 | int r = get_obj_head_ref(dpp, bucket_info, obj_instance, &ref); | |
6861 | if (r < 0) { | |
6862 | return r; | |
6863 | } | |
6864 | ||
6865 | rgw_zone_set zones_trace; | |
6866 | if (_zones_trace) { | |
6867 | zones_trace = *_zones_trace; | |
6868 | } | |
6869 | zones_trace.insert(svc.zone->get_zone().id, bucket_info.bucket.get_key()); | |
6870 | ||
6871 | BucketShard bs(this); | |
6872 | ||
6873 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), obj_instance.key.instance); | |
6874 | r = guard_reshard(dpp, &bs, obj_instance, bucket_info, | |
6875 | [&](BucketShard *bs) -> int { | |
6876 | auto& ref = bs->bucket_obj.get_ref(); | |
6877 | librados::ObjectWriteOperation op; | |
6878 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
6879 | cls_rgw_bucket_unlink_instance(op, key, op_tag, | |
6880 | olh_tag, olh_epoch, svc.zone->get_zone().log_data, zones_trace); | |
6881 | return rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
6882 | }); | |
6883 | if (r < 0) { | |
6884 | ldpp_dout(dpp, 20) << "rgw_rados_operate() after cls_rgw_bucket_link_instance() returned r=" << r << dendl; | |
6885 | return r; | |
6886 | } | |
6887 | ||
6888 | return 0; | |
6889 | } | |
6890 | ||
6891 | int RGWRados::bucket_index_read_olh_log(const DoutPrefixProvider *dpp, | |
6892 | const RGWBucketInfo& bucket_info, RGWObjState& state, | |
6893 | const rgw_obj& obj_instance, uint64_t ver_marker, | |
6894 | map<uint64_t, vector<rgw_bucket_olh_log_entry> > *log, | |
6895 | bool *is_truncated) | |
6896 | { | |
6897 | rgw_rados_ref ref; | |
6898 | int r = get_obj_head_ref(dpp, bucket_info, obj_instance, &ref); | |
6899 | if (r < 0) { | |
6900 | return r; | |
6901 | } | |
6902 | ||
6903 | BucketShard bs(this); | |
6904 | int ret = | |
6905 | bs.init(obj_instance.bucket, obj_instance, nullptr /* no RGWBucketInfo */, dpp); | |
6906 | if (ret < 0) { | |
6907 | ldpp_dout(dpp, 5) << "bs.init() returned ret=" << ret << dendl; | |
6908 | return ret; | |
6909 | } | |
6910 | ||
6911 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
6912 | ||
6913 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
6914 | ||
6915 | ret = guard_reshard(dpp, &bs, obj_instance, bucket_info, | |
6916 | [&](BucketShard *bs) -> int { | |
6917 | auto& ref = bs->bucket_obj.get_ref(); | |
6918 | ObjectReadOperation op; | |
6919 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
6920 | ||
6921 | rgw_cls_read_olh_log_ret log_ret; | |
6922 | int op_ret = 0; | |
6923 | cls_rgw_get_olh_log(op, key, ver_marker, olh_tag, log_ret, op_ret); | |
6924 | bufferlist outbl; | |
6925 | int r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, &outbl, null_yield); | |
6926 | if (r < 0) { | |
6927 | return r; | |
6928 | } | |
6929 | if (op_ret < 0) { | |
6930 | return op_ret; | |
6931 | } | |
6932 | ||
6933 | *log = std::move(log_ret.log); | |
6934 | *is_truncated = log_ret.is_truncated; | |
6935 | return r; | |
6936 | }); | |
6937 | if (ret < 0) { | |
6938 | ldpp_dout(dpp, 20) << "cls_rgw_get_olh_log() returned r=" << r << dendl; | |
6939 | return ret; | |
6940 | } | |
6941 | ||
6942 | return 0; | |
6943 | } | |
6944 | ||
6945 | // a multisite sync bug resulted in the OLH head attributes being overwritten by | |
6946 | // the attributes from another zone, causing link_olh() to fail endlessly due to | |
6947 | // olh_tag mismatch. this attempts to detect this case and reconstruct the OLH | |
6948 | // attributes from the bucket index. see http://tracker.ceph.com/issues/37792 | |
6949 | int RGWRados::repair_olh(const DoutPrefixProvider *dpp, RGWObjState* state, const RGWBucketInfo& bucket_info, | |
6950 | const rgw_obj& obj) | |
6951 | { | |
6952 | // fetch the current olh entry from the bucket index | |
6953 | rgw_bucket_olh_entry olh; | |
6954 | int r = bi_get_olh(dpp, bucket_info, obj, &olh); | |
6955 | if (r < 0) { | |
6956 | ldpp_dout(dpp, 0) << "repair_olh failed to read olh entry for " << obj << dendl; | |
6957 | return r; | |
6958 | } | |
6959 | if (olh.tag == rgw_bl_str(state->olh_tag)) { // mismatch already resolved? | |
6960 | return 0; | |
6961 | } | |
6962 | ||
6963 | ldpp_dout(dpp, 4) << "repair_olh setting olh_tag=" << olh.tag | |
6964 | << " key=" << olh.key << " delete_marker=" << olh.delete_marker << dendl; | |
6965 | ||
6966 | // rewrite OLH_ID_TAG and OLH_INFO from current olh | |
6967 | ObjectWriteOperation op; | |
6968 | // assert this is the same olh tag we think we're fixing | |
6969 | bucket_index_guard_olh_op(dpp, *state, op); | |
6970 | // preserve existing mtime | |
6971 | struct timespec mtime_ts = ceph::real_clock::to_timespec(state->mtime); | |
6972 | op.mtime2(&mtime_ts); | |
6973 | { | |
6974 | bufferlist bl; | |
6975 | bl.append(olh.tag.c_str(), olh.tag.size()); | |
6976 | op.setxattr(RGW_ATTR_OLH_ID_TAG, bl); | |
6977 | } | |
6978 | { | |
6979 | RGWOLHInfo info; | |
6980 | info.target = rgw_obj(bucket_info.bucket, olh.key); | |
6981 | info.removed = olh.delete_marker; | |
6982 | bufferlist bl; | |
6983 | encode(info, bl); | |
6984 | op.setxattr(RGW_ATTR_OLH_INFO, bl); | |
6985 | } | |
6986 | rgw_rados_ref ref; | |
6987 | r = get_obj_head_ref(dpp, bucket_info, obj, &ref); | |
6988 | if (r < 0) { | |
6989 | return r; | |
6990 | } | |
6991 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
6992 | if (r < 0) { | |
6993 | ldpp_dout(dpp, 0) << "repair_olh failed to write olh attributes with " | |
6994 | << cpp_strerror(r) << dendl; | |
6995 | return r; | |
6996 | } | |
6997 | return 0; | |
6998 | } | |
6999 | ||
7000 | int RGWRados::bucket_index_trim_olh_log(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj_instance, uint64_t ver) | |
7001 | { | |
7002 | rgw_rados_ref ref; | |
7003 | int r = get_obj_head_ref(dpp, bucket_info, obj_instance, &ref); | |
7004 | if (r < 0) { | |
7005 | return r; | |
7006 | } | |
7007 | ||
7008 | BucketShard bs(this); | |
7009 | int ret = | |
7010 | bs.init(obj_instance.bucket, obj_instance, nullptr /* no RGWBucketInfo */, dpp); | |
7011 | if (ret < 0) { | |
7012 | ldpp_dout(dpp, 5) << "bs.init() returned ret=" << ret << dendl; | |
7013 | return ret; | |
7014 | } | |
7015 | ||
7016 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
7017 | ||
7018 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
7019 | ||
7020 | ret = guard_reshard(dpp, &bs, obj_instance, bucket_info, | |
7021 | [&](BucketShard *pbs) -> int { | |
7022 | ObjectWriteOperation op; | |
7023 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
7024 | cls_rgw_trim_olh_log(op, key, ver, olh_tag); | |
7025 | return pbs->bucket_obj.operate(dpp, &op, null_yield); | |
7026 | }); | |
7027 | if (ret < 0) { | |
7028 | ldpp_dout(dpp, 20) << "cls_rgw_trim_olh_log() returned r=" << ret << dendl; | |
7029 | return ret; | |
7030 | } | |
7031 | ||
7032 | return 0; | |
7033 | } | |
7034 | ||
7035 | int RGWRados::bucket_index_clear_olh(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& obj_instance) | |
7036 | { | |
7037 | rgw_rados_ref ref; | |
7038 | int r = get_obj_head_ref(dpp, bucket_info, obj_instance, &ref); | |
7039 | if (r < 0) { | |
7040 | return r; | |
7041 | } | |
7042 | ||
7043 | BucketShard bs(this); | |
7044 | ||
7045 | string olh_tag(state.olh_tag.c_str(), state.olh_tag.length()); | |
7046 | ||
7047 | cls_rgw_obj_key key(obj_instance.key.get_index_key_name(), string()); | |
7048 | ||
7049 | int ret = guard_reshard(dpp, &bs, obj_instance, bucket_info, | |
7050 | [&](BucketShard *pbs) -> int { | |
7051 | ObjectWriteOperation op; | |
7052 | auto& ref = pbs->bucket_obj.get_ref(); | |
7053 | cls_rgw_guard_bucket_resharding(op, -ERR_BUSY_RESHARDING); | |
7054 | cls_rgw_clear_olh(op, key, olh_tag); | |
7055 | return rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
7056 | }); | |
7057 | if (ret < 0) { | |
7058 | ldpp_dout(dpp, 5) << "rgw_rados_operate() after cls_rgw_clear_olh() returned ret=" << ret << dendl; | |
7059 | return ret; | |
7060 | } | |
7061 | ||
7062 | return 0; | |
7063 | } | |
7064 | ||
7065 | static int decode_olh_info(CephContext* cct, const bufferlist& bl, RGWOLHInfo *olh) | |
7066 | { | |
7067 | try { | |
7068 | auto biter = bl.cbegin(); | |
7069 | decode(*olh, biter); | |
7070 | return 0; | |
7071 | } catch (buffer::error& err) { | |
7072 | ldout(cct, 0) << "ERROR: failed to decode olh info" << dendl; | |
7073 | return -EIO; | |
7074 | } | |
7075 | } | |
7076 | ||
7077 | int RGWRados::apply_olh_log(const DoutPrefixProvider *dpp, RGWObjectCtx& obj_ctx, RGWObjState& state, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
7078 | bufferlist& olh_tag, map<uint64_t, vector<rgw_bucket_olh_log_entry> >& log, | |
7079 | uint64_t *plast_ver, rgw_zone_set* zones_trace) | |
7080 | { | |
7081 | if (log.empty()) { | |
7082 | return 0; | |
7083 | } | |
7084 | ||
7085 | librados::ObjectWriteOperation op; | |
7086 | ||
7087 | uint64_t last_ver = log.rbegin()->first; | |
7088 | *plast_ver = last_ver; | |
7089 | ||
7090 | map<uint64_t, vector<rgw_bucket_olh_log_entry> >::iterator iter = log.begin(); | |
7091 | ||
7092 | op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_tag); | |
7093 | op.cmpxattr(RGW_ATTR_OLH_VER, CEPH_OSD_CMPXATTR_OP_GTE, last_ver); | |
7094 | ||
7095 | bufferlist ver_bl; | |
7096 | string last_ver_s = to_string(last_ver); | |
7097 | ver_bl.append(last_ver_s.c_str(), last_ver_s.size()); | |
7098 | op.setxattr(RGW_ATTR_OLH_VER, ver_bl); | |
7099 | ||
7100 | struct timespec mtime_ts = real_clock::to_timespec(state.mtime); | |
7101 | op.mtime2(&mtime_ts); | |
7102 | ||
7103 | bool need_to_link = false; | |
7104 | uint64_t link_epoch = 0; | |
7105 | cls_rgw_obj_key key; | |
7106 | bool delete_marker = false; | |
7107 | list<cls_rgw_obj_key> remove_instances; | |
7108 | bool need_to_remove = false; | |
7109 | ||
7110 | // decode current epoch and instance | |
7111 | auto olh_ver = state.attrset.find(RGW_ATTR_OLH_VER); | |
7112 | if (olh_ver != state.attrset.end()) { | |
7113 | std::string str = olh_ver->second.to_str(); | |
7114 | std::string err; | |
7115 | link_epoch = strict_strtoll(str.c_str(), 10, &err); | |
7116 | } | |
7117 | auto olh_info = state.attrset.find(RGW_ATTR_OLH_INFO); | |
7118 | if (olh_info != state.attrset.end()) { | |
7119 | RGWOLHInfo info; | |
7120 | int r = decode_olh_info(cct, olh_info->second, &info); | |
7121 | if (r < 0) { | |
7122 | return r; | |
7123 | } | |
7124 | info.target.key.get_index_key(&key); | |
7125 | delete_marker = info.removed; | |
7126 | } | |
7127 | ||
7128 | for (iter = log.begin(); iter != log.end(); ++iter) { | |
7129 | vector<rgw_bucket_olh_log_entry>::iterator viter = iter->second.begin(); | |
7130 | for (; viter != iter->second.end(); ++viter) { | |
7131 | rgw_bucket_olh_log_entry& entry = *viter; | |
7132 | ||
7133 | ldpp_dout(dpp, 20) << "olh_log_entry: epoch=" << iter->first << " op=" << (int)entry.op | |
7134 | << " key=" << entry.key.name << "[" << entry.key.instance << "] " | |
7135 | << (entry.delete_marker ? "(delete)" : "") << dendl; | |
7136 | switch (entry.op) { | |
7137 | case CLS_RGW_OLH_OP_REMOVE_INSTANCE: | |
7138 | remove_instances.push_back(entry.key); | |
7139 | break; | |
7140 | case CLS_RGW_OLH_OP_LINK_OLH: | |
7141 | // only overwrite a link of the same epoch if its key sorts before | |
7142 | if (link_epoch < iter->first || key.instance.empty() || | |
7143 | key.instance > entry.key.instance) { | |
7144 | ldpp_dout(dpp, 20) << "apply_olh_log applying key=" << entry.key << " epoch=" << iter->first << " delete_marker=" << entry.delete_marker | |
7145 | << " over current=" << key << " epoch=" << link_epoch << " delete_marker=" << delete_marker << dendl; | |
7146 | need_to_link = true; | |
7147 | need_to_remove = false; | |
7148 | key = entry.key; | |
7149 | delete_marker = entry.delete_marker; | |
7150 | } else { | |
7151 | ldpp_dout(dpp, 20) << "apply_olh skipping key=" << entry.key<< " epoch=" << iter->first << " delete_marker=" << entry.delete_marker | |
7152 | << " before current=" << key << " epoch=" << link_epoch << " delete_marker=" << delete_marker << dendl; | |
7153 | } | |
7154 | break; | |
7155 | case CLS_RGW_OLH_OP_UNLINK_OLH: | |
7156 | need_to_remove = true; | |
7157 | need_to_link = false; | |
7158 | break; | |
7159 | default: | |
7160 | ldpp_dout(dpp, 0) << "ERROR: apply_olh_log: invalid op: " << (int)entry.op << dendl; | |
7161 | return -EIO; | |
7162 | } | |
7163 | string attr_name = RGW_ATTR_OLH_PENDING_PREFIX; | |
7164 | attr_name.append(entry.op_tag); | |
7165 | op.rmxattr(attr_name.c_str()); | |
7166 | } | |
7167 | } | |
7168 | ||
7169 | rgw_rados_ref ref; | |
7170 | int r = get_obj_head_ref(dpp, bucket_info, obj, &ref); | |
7171 | if (r < 0) { | |
7172 | return r; | |
7173 | } | |
7174 | ||
7175 | const rgw_bucket& bucket = obj.bucket; | |
7176 | ||
7177 | if (need_to_link) { | |
7178 | rgw_obj target(bucket, key); | |
7179 | RGWOLHInfo info; | |
7180 | info.target = target; | |
7181 | info.removed = delete_marker; | |
7182 | bufferlist bl; | |
7183 | encode(info, bl); | |
7184 | op.setxattr(RGW_ATTR_OLH_INFO, bl); | |
7185 | } | |
7186 | ||
7187 | /* first remove object instances */ | |
7188 | for (list<cls_rgw_obj_key>::iterator liter = remove_instances.begin(); | |
7189 | liter != remove_instances.end(); ++liter) { | |
7190 | cls_rgw_obj_key& key = *liter; | |
7191 | rgw_obj obj_instance(bucket, key); | |
7192 | int ret = delete_obj(dpp, obj_ctx, bucket_info, obj_instance, 0, RGW_BILOG_FLAG_VERSIONED_OP, ceph::real_time(), zones_trace); | |
7193 | if (ret < 0 && ret != -ENOENT) { | |
7194 | ldpp_dout(dpp, 0) << "ERROR: delete_obj() returned " << ret << " obj_instance=" << obj_instance << dendl; | |
7195 | return ret; | |
7196 | } | |
7197 | } | |
7198 | ||
7199 | /* update olh object */ | |
7200 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
7201 | if (r == -ECANCELED) { | |
7202 | r = 0; | |
7203 | } | |
7204 | if (r < 0) { | |
7205 | ldpp_dout(dpp, 0) << "ERROR: could not apply olh update, r=" << r << dendl; | |
7206 | return r; | |
7207 | } | |
7208 | ||
7209 | r = bucket_index_trim_olh_log(dpp, bucket_info, state, obj, last_ver); | |
7210 | if (r < 0) { | |
7211 | ldpp_dout(dpp, 0) << "ERROR: could not trim olh log, r=" << r << dendl; | |
7212 | return r; | |
7213 | } | |
7214 | ||
7215 | if (need_to_remove) { | |
7216 | ObjectWriteOperation rm_op; | |
7217 | ||
7218 | rm_op.cmpxattr(RGW_ATTR_OLH_ID_TAG, CEPH_OSD_CMPXATTR_OP_EQ, olh_tag); | |
7219 | rm_op.cmpxattr(RGW_ATTR_OLH_VER, CEPH_OSD_CMPXATTR_OP_EQ, last_ver); | |
7220 | cls_obj_check_prefix_exist(rm_op, RGW_ATTR_OLH_PENDING_PREFIX, true); /* fail if found one of these, pending modification */ | |
7221 | rm_op.remove(); | |
7222 | ||
7223 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &rm_op, null_yield); | |
7224 | if (r == -ECANCELED) { | |
7225 | return 0; /* someone else won this race */ | |
7226 | } else { | |
7227 | /* | |
7228 | * only clear if was successful, otherwise we might clobber pending operations on this object | |
7229 | */ | |
7230 | r = bucket_index_clear_olh(dpp, bucket_info, state, obj); | |
7231 | if (r < 0) { | |
7232 | ldpp_dout(dpp, 0) << "ERROR: could not clear bucket index olh entries r=" << r << dendl; | |
7233 | return r; | |
7234 | } | |
7235 | } | |
7236 | } | |
7237 | ||
7238 | return 0; | |
7239 | } | |
7240 | ||
7241 | /* | |
7242 | * read olh log and apply it | |
7243 | */ | |
7244 | int RGWRados::update_olh(const DoutPrefixProvider *dpp, RGWObjectCtx& obj_ctx, RGWObjState *state, const RGWBucketInfo& bucket_info, const rgw_obj& obj, rgw_zone_set *zones_trace) | |
7245 | { | |
7246 | map<uint64_t, vector<rgw_bucket_olh_log_entry> > log; | |
7247 | bool is_truncated; | |
7248 | uint64_t ver_marker = 0; | |
7249 | ||
7250 | do { | |
7251 | int ret = bucket_index_read_olh_log(dpp, bucket_info, *state, obj, ver_marker, &log, &is_truncated); | |
7252 | if (ret < 0) { | |
7253 | return ret; | |
7254 | } | |
7255 | ret = apply_olh_log(dpp, obj_ctx, *state, bucket_info, obj, state->olh_tag, log, &ver_marker, zones_trace); | |
7256 | if (ret < 0) { | |
7257 | return ret; | |
7258 | } | |
7259 | } while (is_truncated); | |
7260 | ||
7261 | return 0; | |
7262 | } | |
7263 | ||
7264 | int RGWRados::set_olh(const DoutPrefixProvider *dpp, RGWObjectCtx& obj_ctx, const RGWBucketInfo& bucket_info, const rgw_obj& target_obj, bool delete_marker, rgw_bucket_dir_entry_meta *meta, | |
7265 | uint64_t olh_epoch, real_time unmod_since, bool high_precision_time, | |
7266 | optional_yield y, rgw_zone_set *zones_trace, bool log_data_change) | |
7267 | { | |
7268 | string op_tag; | |
7269 | ||
7270 | rgw_obj olh_obj = target_obj; | |
7271 | olh_obj.key.instance.clear(); | |
7272 | ||
7273 | RGWObjState *state = NULL; | |
7274 | ||
7275 | int ret = 0; | |
7276 | int i; | |
7277 | ||
7278 | #define MAX_ECANCELED_RETRY 100 | |
7279 | for (i = 0; i < MAX_ECANCELED_RETRY; i++) { | |
7280 | if (ret == -ECANCELED) { | |
7281 | obj_ctx.invalidate(olh_obj); | |
7282 | } | |
7283 | ||
7284 | ret = get_obj_state(dpp, &obj_ctx, bucket_info, olh_obj, &state, false, y); /* don't follow olh */ | |
7285 | if (ret < 0) { | |
7286 | return ret; | |
7287 | } | |
7288 | ||
7289 | ret = olh_init_modification(dpp, bucket_info, *state, olh_obj, &op_tag); | |
7290 | if (ret < 0) { | |
7291 | ldpp_dout(dpp, 20) << "olh_init_modification() target_obj=" << target_obj << " delete_marker=" << (int)delete_marker << " returned " << ret << dendl; | |
7292 | if (ret == -ECANCELED) { | |
7293 | continue; | |
7294 | } | |
7295 | return ret; | |
7296 | } | |
7297 | ret = bucket_index_link_olh(dpp, bucket_info, *state, target_obj, delete_marker, | |
7298 | op_tag, meta, olh_epoch, unmod_since, high_precision_time, | |
7299 | zones_trace, log_data_change); | |
7300 | if (ret < 0) { | |
7301 | ldpp_dout(dpp, 20) << "bucket_index_link_olh() target_obj=" << target_obj << " delete_marker=" << (int)delete_marker << " returned " << ret << dendl; | |
7302 | if (ret == -ECANCELED) { | |
7303 | // the bucket index rejected the link_olh() due to olh tag mismatch; | |
7304 | // attempt to reconstruct olh head attributes based on the bucket index | |
7305 | int r2 = repair_olh(dpp, state, bucket_info, olh_obj); | |
7306 | if (r2 < 0 && r2 != -ECANCELED) { | |
7307 | return r2; | |
7308 | } | |
7309 | continue; | |
7310 | } | |
7311 | return ret; | |
7312 | } | |
7313 | break; | |
7314 | } | |
7315 | ||
7316 | if (i == MAX_ECANCELED_RETRY) { | |
7317 | ldpp_dout(dpp, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl; | |
7318 | return -EIO; | |
7319 | } | |
7320 | ||
7321 | ret = update_olh(dpp, obj_ctx, state, bucket_info, olh_obj); | |
7322 | if (ret == -ECANCELED) { /* already did what we needed, no need to retry, raced with another user */ | |
7323 | ret = 0; | |
7324 | } | |
7325 | if (ret < 0) { | |
7326 | ldpp_dout(dpp, 20) << "update_olh() target_obj=" << target_obj << " returned " << ret << dendl; | |
7327 | return ret; | |
7328 | } | |
7329 | ||
7330 | return 0; | |
7331 | } | |
7332 | ||
7333 | int RGWRados::unlink_obj_instance(const DoutPrefixProvider *dpp, RGWObjectCtx& obj_ctx, RGWBucketInfo& bucket_info, const rgw_obj& target_obj, | |
7334 | uint64_t olh_epoch, optional_yield y, rgw_zone_set *zones_trace) | |
7335 | { | |
7336 | string op_tag; | |
7337 | ||
7338 | rgw_obj olh_obj = target_obj; | |
7339 | olh_obj.key.instance.clear(); | |
7340 | ||
7341 | RGWObjState *state = NULL; | |
7342 | ||
7343 | int ret = 0; | |
7344 | int i; | |
7345 | ||
7346 | for (i = 0; i < MAX_ECANCELED_RETRY; i++) { | |
7347 | if (ret == -ECANCELED) { | |
7348 | obj_ctx.invalidate(olh_obj); | |
7349 | } | |
7350 | ||
7351 | ret = get_obj_state(dpp, &obj_ctx, bucket_info, olh_obj, &state, false, y); /* don't follow olh */ | |
7352 | if (ret < 0) | |
7353 | return ret; | |
7354 | ||
7355 | ret = olh_init_modification(dpp, bucket_info, *state, olh_obj, &op_tag); | |
7356 | if (ret < 0) { | |
7357 | ldpp_dout(dpp, 20) << "olh_init_modification() target_obj=" << target_obj << " returned " << ret << dendl; | |
7358 | if (ret == -ECANCELED) { | |
7359 | continue; | |
7360 | } | |
7361 | return ret; | |
7362 | } | |
7363 | ||
7364 | string olh_tag(state->olh_tag.c_str(), state->olh_tag.length()); | |
7365 | ||
7366 | ret = bucket_index_unlink_instance(dpp, bucket_info, target_obj, op_tag, olh_tag, olh_epoch, zones_trace); | |
7367 | if (ret < 0) { | |
7368 | ldpp_dout(dpp, 20) << "bucket_index_unlink_instance() target_obj=" << target_obj << " returned " << ret << dendl; | |
7369 | if (ret == -ECANCELED) { | |
7370 | continue; | |
7371 | } | |
7372 | return ret; | |
7373 | } | |
7374 | break; | |
7375 | } | |
7376 | ||
7377 | if (i == MAX_ECANCELED_RETRY) { | |
7378 | ldpp_dout(dpp, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl; | |
7379 | return -EIO; | |
7380 | } | |
7381 | ||
7382 | ret = update_olh(dpp, obj_ctx, state, bucket_info, olh_obj, zones_trace); | |
7383 | if (ret == -ECANCELED) { /* already did what we needed, no need to retry, raced with another user */ | |
7384 | return 0; | |
7385 | } | |
7386 | if (ret < 0) { | |
7387 | ldpp_dout(dpp, 20) << "update_olh() target_obj=" << target_obj << " returned " << ret << dendl; | |
7388 | return ret; | |
7389 | } | |
7390 | ||
7391 | return 0; | |
7392 | } | |
7393 | ||
7394 | void RGWRados::gen_rand_obj_instance_name(rgw_obj_key *target_key) | |
7395 | { | |
7396 | #define OBJ_INSTANCE_LEN 32 | |
7397 | char buf[OBJ_INSTANCE_LEN + 1]; | |
7398 | ||
7399 | gen_rand_alphanumeric_no_underscore(cct, buf, OBJ_INSTANCE_LEN); /* don't want it to get url escaped, | |
7400 | no underscore for instance name due to the way we encode the raw keys */ | |
7401 | ||
7402 | target_key->set_instance(buf); | |
7403 | } | |
7404 | ||
7405 | void RGWRados::gen_rand_obj_instance_name(rgw_obj *target_obj) | |
7406 | { | |
7407 | gen_rand_obj_instance_name(&target_obj->key); | |
7408 | } | |
7409 | ||
7410 | int RGWRados::get_olh(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj, RGWOLHInfo *olh) | |
7411 | { | |
7412 | map<string, bufferlist> attrset; | |
7413 | ||
7414 | ObjectReadOperation op; | |
7415 | op.getxattrs(&attrset, NULL); | |
7416 | ||
7417 | int r = obj_operate(dpp, bucket_info, obj, &op); | |
7418 | if (r < 0) { | |
7419 | return r; | |
7420 | } | |
7421 | ||
7422 | auto iter = attrset.find(RGW_ATTR_OLH_INFO); | |
7423 | if (iter == attrset.end()) { /* not an olh */ | |
7424 | return -EINVAL; | |
7425 | } | |
7426 | ||
7427 | return decode_olh_info(cct, iter->second, olh); | |
7428 | } | |
7429 | ||
7430 | void RGWRados::check_pending_olh_entries(map<string, bufferlist>& pending_entries, | |
7431 | map<string, bufferlist> *rm_pending_entries) | |
7432 | { | |
7433 | map<string, bufferlist>::iterator iter = pending_entries.begin(); | |
7434 | ||
7435 | real_time now = real_clock::now(); | |
7436 | ||
7437 | while (iter != pending_entries.end()) { | |
7438 | auto biter = iter->second.cbegin(); | |
7439 | RGWOLHPendingInfo pending_info; | |
7440 | try { | |
7441 | decode(pending_info, biter); | |
7442 | } catch (buffer::error& err) { | |
7443 | /* skipping bad entry, we could remove it but it might hide a bug */ | |
7444 | ldout(cct, 0) << "ERROR: failed to decode pending entry " << iter->first << dendl; | |
7445 | ++iter; | |
7446 | continue; | |
7447 | } | |
7448 | ||
7449 | map<string, bufferlist>::iterator cur_iter = iter; | |
7450 | ++iter; | |
7451 | if (now - pending_info.time >= make_timespan(cct->_conf->rgw_olh_pending_timeout_sec)) { | |
7452 | (*rm_pending_entries)[cur_iter->first] = cur_iter->second; | |
7453 | pending_entries.erase(cur_iter); | |
7454 | } else { | |
7455 | /* entries names are sorted by time (rounded to a second) */ | |
7456 | break; | |
7457 | } | |
7458 | } | |
7459 | } | |
7460 | ||
7461 | int RGWRados::remove_olh_pending_entries(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, RGWObjState& state, const rgw_obj& olh_obj, map<string, bufferlist>& pending_attrs) | |
7462 | { | |
7463 | rgw_rados_ref ref; | |
7464 | int r = get_obj_head_ref(dpp, bucket_info, olh_obj, &ref); | |
7465 | if (r < 0) { | |
7466 | return r; | |
7467 | } | |
7468 | ||
7469 | // trim no more than 1000 entries per osd op | |
7470 | constexpr int max_entries = 1000; | |
7471 | ||
7472 | auto i = pending_attrs.begin(); | |
7473 | while (i != pending_attrs.end()) { | |
7474 | ObjectWriteOperation op; | |
7475 | bucket_index_guard_olh_op(dpp, state, op); | |
7476 | ||
7477 | for (int n = 0; n < max_entries && i != pending_attrs.end(); ++n, ++i) { | |
7478 | op.rmxattr(i->first.c_str()); | |
7479 | } | |
7480 | ||
7481 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
7482 | if (r == -ENOENT || r == -ECANCELED) { | |
7483 | /* raced with some other change, shouldn't sweat about it */ | |
7484 | return 0; | |
7485 | } | |
7486 | if (r < 0) { | |
7487 | ldpp_dout(dpp, 0) << "ERROR: could not apply olh update, r=" << r << dendl; | |
7488 | return r; | |
7489 | } | |
7490 | } | |
7491 | return 0; | |
7492 | } | |
7493 | ||
7494 | int RGWRados::follow_olh(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, RGWObjectCtx& obj_ctx, RGWObjState *state, const rgw_obj& olh_obj, rgw_obj *target) | |
7495 | { | |
7496 | map<string, bufferlist> pending_entries; | |
7497 | rgw_filter_attrset(state->attrset, RGW_ATTR_OLH_PENDING_PREFIX, &pending_entries); | |
7498 | ||
7499 | map<string, bufferlist> rm_pending_entries; | |
7500 | check_pending_olh_entries(pending_entries, &rm_pending_entries); | |
7501 | ||
7502 | if (!rm_pending_entries.empty()) { | |
7503 | int ret = remove_olh_pending_entries(dpp, bucket_info, *state, olh_obj, rm_pending_entries); | |
7504 | if (ret < 0) { | |
7505 | ldpp_dout(dpp, 20) << "ERROR: rm_pending_entries returned ret=" << ret << dendl; | |
7506 | return ret; | |
7507 | } | |
7508 | } | |
7509 | if (!pending_entries.empty()) { | |
7510 | ldpp_dout(dpp, 20) << __func__ << "(): found pending entries, need to update_olh() on bucket=" << olh_obj.bucket << dendl; | |
7511 | ||
7512 | int ret = update_olh(dpp, obj_ctx, state, bucket_info, olh_obj); | |
7513 | if (ret < 0) { | |
7514 | return ret; | |
7515 | } | |
7516 | } | |
7517 | ||
7518 | auto iter = state->attrset.find(RGW_ATTR_OLH_INFO); | |
7519 | if (iter == state->attrset.end()) { | |
7520 | return -EINVAL; | |
7521 | } | |
7522 | ||
7523 | RGWOLHInfo olh; | |
7524 | int ret = decode_olh_info(cct, iter->second, &olh); | |
7525 | if (ret < 0) { | |
7526 | return ret; | |
7527 | } | |
7528 | ||
7529 | if (olh.removed) { | |
7530 | return -ENOENT; | |
7531 | } | |
7532 | ||
7533 | *target = olh.target; | |
7534 | ||
7535 | return 0; | |
7536 | } | |
7537 | ||
7538 | int RGWRados::raw_obj_stat(const DoutPrefixProvider *dpp, | |
7539 | rgw_raw_obj& obj, uint64_t *psize, real_time *pmtime, uint64_t *epoch, | |
7540 | map<string, bufferlist> *attrs, bufferlist *first_chunk, | |
7541 | RGWObjVersionTracker *objv_tracker, optional_yield y) | |
7542 | { | |
7543 | rgw_rados_ref ref; | |
7544 | int r = get_raw_obj_ref(dpp, obj, &ref); | |
7545 | if (r < 0) { | |
7546 | return r; | |
7547 | } | |
7548 | ||
7549 | map<string, bufferlist> unfiltered_attrset; | |
7550 | uint64_t size = 0; | |
7551 | struct timespec mtime_ts; | |
7552 | ||
7553 | ObjectReadOperation op; | |
7554 | if (objv_tracker) { | |
7555 | objv_tracker->prepare_op_for_read(&op); | |
7556 | } | |
7557 | if (attrs) { | |
7558 | op.getxattrs(&unfiltered_attrset, NULL); | |
7559 | } | |
7560 | if (psize || pmtime) { | |
7561 | op.stat2(&size, &mtime_ts, NULL); | |
7562 | } | |
7563 | if (first_chunk) { | |
7564 | op.read(0, cct->_conf->rgw_max_chunk_size, first_chunk, NULL); | |
7565 | } | |
7566 | bufferlist outbl; | |
7567 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, &outbl, null_yield); | |
7568 | ||
7569 | if (epoch) { | |
7570 | *epoch = ref.pool.ioctx().get_last_version(); | |
7571 | } | |
7572 | ||
7573 | if (r < 0) | |
7574 | return r; | |
7575 | ||
7576 | if (psize) | |
7577 | *psize = size; | |
7578 | if (pmtime) | |
7579 | *pmtime = ceph::real_clock::from_timespec(mtime_ts); | |
7580 | if (attrs) { | |
7581 | rgw_filter_attrset(unfiltered_attrset, RGW_ATTR_PREFIX, attrs); | |
7582 | } | |
7583 | ||
7584 | return 0; | |
7585 | } | |
7586 | ||
7587 | int RGWRados::get_bucket_stats(const DoutPrefixProvider *dpp, RGWBucketInfo& bucket_info, int shard_id, string *bucket_ver, string *master_ver, | |
7588 | map<RGWObjCategory, RGWStorageStats>& stats, string *max_marker, bool *syncstopped) | |
7589 | { | |
7590 | vector<rgw_bucket_dir_header> headers; | |
7591 | map<int, string> bucket_instance_ids; | |
7592 | int r = cls_bucket_head(dpp, bucket_info, shard_id, headers, &bucket_instance_ids); | |
7593 | if (r < 0) { | |
7594 | return r; | |
7595 | } | |
7596 | ||
7597 | ceph_assert(headers.size() == bucket_instance_ids.size()); | |
7598 | ||
7599 | auto iter = headers.begin(); | |
7600 | map<int, string>::iterator viter = bucket_instance_ids.begin(); | |
7601 | BucketIndexShardsManager ver_mgr; | |
7602 | BucketIndexShardsManager master_ver_mgr; | |
7603 | BucketIndexShardsManager marker_mgr; | |
7604 | char buf[64]; | |
7605 | for(; iter != headers.end(); ++iter, ++viter) { | |
7606 | accumulate_raw_stats(*iter, stats); | |
7607 | snprintf(buf, sizeof(buf), "%lu", (unsigned long)iter->ver); | |
7608 | ver_mgr.add(viter->first, string(buf)); | |
7609 | snprintf(buf, sizeof(buf), "%lu", (unsigned long)iter->master_ver); | |
7610 | master_ver_mgr.add(viter->first, string(buf)); | |
7611 | if (shard_id >= 0) { | |
7612 | *max_marker = iter->max_marker; | |
7613 | } else { | |
7614 | marker_mgr.add(viter->first, iter->max_marker); | |
7615 | } | |
7616 | if (syncstopped != NULL) | |
7617 | *syncstopped = iter->syncstopped; | |
7618 | } | |
7619 | ver_mgr.to_string(bucket_ver); | |
7620 | master_ver_mgr.to_string(master_ver); | |
7621 | if (shard_id < 0) { | |
7622 | marker_mgr.to_string(max_marker); | |
7623 | } | |
7624 | return 0; | |
7625 | } | |
7626 | ||
7627 | class RGWGetBucketStatsContext : public RGWGetDirHeader_CB { | |
7628 | RGWGetBucketStats_CB *cb; | |
7629 | uint32_t pendings; | |
7630 | map<RGWObjCategory, RGWStorageStats> stats; | |
7631 | int ret_code; | |
7632 | bool should_cb; | |
7633 | ceph::mutex lock = ceph::make_mutex("RGWGetBucketStatsContext"); | |
7634 | ||
7635 | public: | |
7636 | RGWGetBucketStatsContext(RGWGetBucketStats_CB *_cb, uint32_t _pendings) | |
7637 | : cb(_cb), pendings(_pendings), stats(), ret_code(0), should_cb(true) | |
7638 | {} | |
7639 | ||
7640 | void handle_response(int r, rgw_bucket_dir_header& header) override { | |
7641 | std::lock_guard l{lock}; | |
7642 | if (should_cb) { | |
7643 | if ( r >= 0) { | |
7644 | accumulate_raw_stats(header, stats); | |
7645 | } else { | |
7646 | ret_code = r; | |
7647 | } | |
7648 | ||
7649 | // Are we all done? | |
7650 | if (--pendings == 0) { | |
7651 | if (!ret_code) { | |
7652 | cb->set_response(&stats); | |
7653 | } | |
7654 | cb->handle_response(ret_code); | |
7655 | cb->put(); | |
7656 | } | |
7657 | } | |
7658 | } | |
7659 | ||
7660 | void unset_cb() { | |
7661 | std::lock_guard l{lock}; | |
7662 | should_cb = false; | |
7663 | } | |
7664 | }; | |
7665 | ||
7666 | int RGWRados::get_bucket_stats_async(const DoutPrefixProvider *dpp, RGWBucketInfo& bucket_info, int shard_id, RGWGetBucketStats_CB *ctx) | |
7667 | { | |
7668 | int num_aio = 0; | |
7669 | RGWGetBucketStatsContext *get_ctx = new RGWGetBucketStatsContext(ctx, bucket_info.layout.current_index.layout.normal.num_shards ? : 1); | |
7670 | ceph_assert(get_ctx); | |
7671 | int r = cls_bucket_head_async(dpp, bucket_info, shard_id, get_ctx, &num_aio); | |
7672 | if (r < 0) { | |
7673 | ctx->put(); | |
7674 | if (num_aio) { | |
7675 | get_ctx->unset_cb(); | |
7676 | } | |
7677 | } | |
7678 | get_ctx->put(); | |
7679 | return r; | |
7680 | } | |
7681 | ||
7682 | int RGWRados::get_bucket_instance_info(RGWSysObjectCtx& obj_ctx, | |
7683 | const string& meta_key, | |
7684 | RGWBucketInfo& info, | |
7685 | real_time *pmtime, | |
7686 | map<string, bufferlist> *pattrs, | |
7687 | optional_yield y, | |
7688 | const DoutPrefixProvider *dpp) | |
7689 | { | |
7690 | rgw_bucket bucket; | |
7691 | rgw_bucket_parse_bucket_key(cct, meta_key, &bucket, nullptr); | |
7692 | ||
7693 | return get_bucket_instance_info(obj_ctx, bucket, info, pmtime, pattrs, y, dpp); | |
7694 | } | |
7695 | ||
7696 | int RGWRados::get_bucket_instance_info(RGWSysObjectCtx& obj_ctx, const rgw_bucket& bucket, RGWBucketInfo& info, | |
7697 | real_time *pmtime, map<string, bufferlist> *pattrs, optional_yield y, | |
7698 | const DoutPrefixProvider *dpp) | |
7699 | { | |
7700 | RGWSI_MetaBackend_CtxParams bectx_params = RGWSI_MetaBackend_CtxParams_SObj(&obj_ctx); | |
7701 | return ctl.bucket->read_bucket_instance_info(bucket, &info, | |
7702 | y, | |
7703 | dpp, | |
7704 | RGWBucketCtl::BucketInstance::GetParams() | |
7705 | .set_mtime(pmtime) | |
7706 | .set_attrs(pattrs) | |
7707 | .set_bectx_params(bectx_params)); | |
7708 | } | |
7709 | ||
7710 | int RGWRados::get_bucket_info(RGWServices *svc, | |
7711 | const string& tenant, const string& bucket_name, | |
7712 | RGWBucketInfo& info, | |
7713 | real_time *pmtime, | |
7714 | optional_yield y, | |
7715 | const DoutPrefixProvider *dpp, map<string, bufferlist> *pattrs) | |
7716 | { | |
7717 | auto obj_ctx = svc->sysobj->init_obj_ctx(); | |
7718 | RGWSI_MetaBackend_CtxParams bectx_params = RGWSI_MetaBackend_CtxParams_SObj(&obj_ctx); | |
7719 | rgw_bucket bucket; | |
7720 | bucket.tenant = tenant; | |
7721 | bucket.name = bucket_name; | |
7722 | return ctl.bucket->read_bucket_info(bucket, &info, y, dpp, | |
7723 | RGWBucketCtl::BucketInstance::GetParams() | |
7724 | .set_mtime(pmtime) | |
7725 | .set_attrs(pattrs) | |
7726 | .set_bectx_params(bectx_params)); | |
7727 | } | |
7728 | ||
7729 | int RGWRados::try_refresh_bucket_info(RGWBucketInfo& info, | |
7730 | ceph::real_time *pmtime, | |
7731 | const DoutPrefixProvider *dpp, | |
7732 | map<string, bufferlist> *pattrs) | |
7733 | { | |
7734 | rgw_bucket bucket = info.bucket; | |
7735 | bucket.bucket_id.clear(); | |
7736 | ||
7737 | auto rv = info.objv_tracker.read_version; | |
7738 | ||
7739 | return ctl.bucket->read_bucket_info(bucket, &info, null_yield, dpp, | |
7740 | RGWBucketCtl::BucketInstance::GetParams() | |
7741 | .set_mtime(pmtime) | |
7742 | .set_attrs(pattrs) | |
7743 | .set_refresh_version(rv)); | |
7744 | } | |
7745 | ||
7746 | int RGWRados::put_bucket_instance_info(RGWBucketInfo& info, bool exclusive, | |
7747 | real_time mtime, map<string, bufferlist> *pattrs, | |
7748 | const DoutPrefixProvider *dpp) | |
7749 | { | |
7750 | return ctl.bucket->store_bucket_instance_info(info.bucket, info, null_yield, dpp, | |
7751 | RGWBucketCtl::BucketInstance::PutParams() | |
7752 | .set_exclusive(exclusive) | |
7753 | .set_mtime(mtime) | |
7754 | .set_attrs(pattrs)); | |
7755 | } | |
7756 | ||
7757 | int RGWRados::put_linked_bucket_info(RGWBucketInfo& info, bool exclusive, real_time mtime, obj_version *pep_objv, | |
7758 | map<string, bufferlist> *pattrs, bool create_entry_point, | |
7759 | const DoutPrefixProvider *dpp) | |
7760 | { | |
7761 | bool create_head = !info.has_instance_obj || create_entry_point; | |
7762 | ||
7763 | int ret = put_bucket_instance_info(info, exclusive, mtime, pattrs, dpp); | |
7764 | if (ret < 0) { | |
7765 | return ret; | |
7766 | } | |
7767 | ||
7768 | if (!create_head) | |
7769 | return 0; /* done! */ | |
7770 | ||
7771 | RGWBucketEntryPoint entry_point; | |
7772 | entry_point.bucket = info.bucket; | |
7773 | entry_point.owner = info.owner; | |
7774 | entry_point.creation_time = info.creation_time; | |
7775 | entry_point.linked = true; | |
7776 | RGWObjVersionTracker ot; | |
7777 | if (pep_objv && !pep_objv->tag.empty()) { | |
7778 | ot.write_version = *pep_objv; | |
7779 | } else { | |
7780 | ot.generate_new_write_ver(cct); | |
7781 | if (pep_objv) { | |
7782 | *pep_objv = ot.write_version; | |
7783 | } | |
7784 | } | |
7785 | ret = ctl.bucket->store_bucket_entrypoint_info(info.bucket, entry_point, null_yield, dpp, RGWBucketCtl::Bucket::PutParams() | |
7786 | .set_exclusive(exclusive) | |
7787 | .set_objv_tracker(&ot) | |
7788 | .set_mtime(mtime)); | |
7789 | if (ret < 0) | |
7790 | return ret; | |
7791 | ||
7792 | return 0; | |
7793 | } | |
7794 | ||
7795 | int RGWRados::update_containers_stats(map<string, RGWBucketEnt>& m, const DoutPrefixProvider *dpp) | |
7796 | { | |
7797 | auto obj_ctx = svc.sysobj->init_obj_ctx(); | |
7798 | ||
7799 | map<string, RGWBucketEnt>::iterator iter; | |
7800 | for (iter = m.begin(); iter != m.end(); ++iter) { | |
7801 | RGWBucketEnt& ent = iter->second; | |
7802 | rgw_bucket& bucket = ent.bucket; | |
7803 | ent.count = 0; | |
7804 | ent.size = 0; | |
7805 | ent.size_rounded = 0; | |
7806 | ||
7807 | vector<rgw_bucket_dir_header> headers; | |
7808 | ||
7809 | RGWBucketInfo bucket_info; | |
7810 | int ret = get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL, null_yield, dpp); | |
7811 | if (ret < 0) { | |
7812 | return ret; | |
7813 | } | |
7814 | ||
7815 | int r = cls_bucket_head(dpp, bucket_info, RGW_NO_SHARD, headers); | |
7816 | if (r < 0) | |
7817 | return r; | |
7818 | ||
7819 | auto hiter = headers.begin(); | |
7820 | for (; hiter != headers.end(); ++hiter) { | |
7821 | RGWObjCategory category = main_category; | |
7822 | auto iter = (hiter->stats).find(category); | |
7823 | if (iter != hiter->stats.end()) { | |
7824 | struct rgw_bucket_category_stats& stats = iter->second; | |
7825 | ent.count += stats.num_entries; | |
7826 | ent.size += stats.total_size; | |
7827 | ent.size_rounded += stats.total_size_rounded; | |
7828 | } | |
7829 | } | |
7830 | ||
7831 | // fill in placement_rule from the bucket instance for use in swift's | |
7832 | // per-storage policy statistics | |
7833 | ent.placement_rule = std::move(bucket_info.placement_rule); | |
7834 | } | |
7835 | ||
7836 | return m.size(); | |
7837 | } | |
7838 | ||
7839 | int RGWRados::append_async(const DoutPrefixProvider *dpp, rgw_raw_obj& obj, size_t size, bufferlist& bl) | |
7840 | { | |
7841 | rgw_rados_ref ref; | |
7842 | int r = get_raw_obj_ref(dpp, obj, &ref); | |
7843 | if (r < 0) { | |
7844 | return r; | |
7845 | } | |
7846 | librados::Rados *rad = get_rados_handle(); | |
7847 | librados::AioCompletion *completion = rad->aio_create_completion(nullptr, nullptr); | |
7848 | ||
7849 | r = ref.pool.ioctx().aio_append(ref.obj.oid, completion, bl, size); | |
7850 | completion->release(); | |
7851 | return r; | |
7852 | } | |
7853 | ||
7854 | int RGWRados::pool_iterate_begin(const DoutPrefixProvider *dpp, const rgw_pool& pool, RGWPoolIterCtx& ctx) | |
7855 | { | |
7856 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
7857 | librados::NObjectIterator& iter = ctx.iter; | |
7858 | ||
7859 | int r = open_pool_ctx(dpp, pool, io_ctx, false); | |
7860 | if (r < 0) | |
7861 | return r; | |
7862 | ||
7863 | iter = io_ctx.nobjects_begin(); | |
7864 | ||
7865 | return 0; | |
7866 | } | |
7867 | ||
7868 | int RGWRados::pool_iterate_begin(const DoutPrefixProvider *dpp, const rgw_pool& pool, const string& cursor, RGWPoolIterCtx& ctx) | |
7869 | { | |
7870 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
7871 | librados::NObjectIterator& iter = ctx.iter; | |
7872 | ||
7873 | int r = open_pool_ctx(dpp, pool, io_ctx, false); | |
7874 | if (r < 0) | |
7875 | return r; | |
7876 | ||
7877 | librados::ObjectCursor oc; | |
7878 | if (!oc.from_str(cursor)) { | |
7879 | ldpp_dout(dpp, 10) << "failed to parse cursor: " << cursor << dendl; | |
7880 | return -EINVAL; | |
7881 | } | |
7882 | ||
7883 | try { | |
7884 | iter = io_ctx.nobjects_begin(oc); | |
7885 | return 0; | |
7886 | } catch (const std::system_error& e) { | |
7887 | r = -e.code().value(); | |
7888 | ldpp_dout(dpp, 10) << "nobjects_begin threw " << e.what() | |
7889 | << ", returning " << r << dendl; | |
7890 | return r; | |
7891 | } catch (const std::exception& e) { | |
7892 | ldpp_dout(dpp, 10) << "nobjects_begin threw " << e.what() | |
7893 | << ", returning -5" << dendl; | |
7894 | return -EIO; | |
7895 | } | |
7896 | } | |
7897 | ||
7898 | string RGWRados::pool_iterate_get_cursor(RGWPoolIterCtx& ctx) | |
7899 | { | |
7900 | return ctx.iter.get_cursor().to_str(); | |
7901 | } | |
7902 | ||
7903 | static int do_pool_iterate(CephContext* cct, RGWPoolIterCtx& ctx, uint32_t num, | |
7904 | vector<rgw_bucket_dir_entry>& objs, | |
7905 | bool *is_truncated, RGWAccessListFilter *filter) | |
7906 | { | |
7907 | librados::IoCtx& io_ctx = ctx.io_ctx; | |
7908 | librados::NObjectIterator& iter = ctx.iter; | |
7909 | ||
7910 | if (iter == io_ctx.nobjects_end()) | |
7911 | return -ENOENT; | |
7912 | ||
7913 | uint32_t i; | |
7914 | ||
7915 | for (i = 0; i < num && iter != io_ctx.nobjects_end(); ++i, ++iter) { | |
7916 | rgw_bucket_dir_entry e; | |
7917 | ||
7918 | string oid = iter->get_oid(); | |
7919 | ldout(cct, 20) << "RGWRados::pool_iterate: got " << oid << dendl; | |
7920 | ||
7921 | // fill it in with initial values; we may correct later | |
7922 | if (filter && !filter->filter(oid, oid)) | |
7923 | continue; | |
7924 | ||
7925 | e.key = oid; | |
7926 | objs.push_back(e); | |
7927 | } | |
7928 | ||
7929 | if (is_truncated) | |
7930 | *is_truncated = (iter != io_ctx.nobjects_end()); | |
7931 | ||
7932 | return objs.size(); | |
7933 | } | |
7934 | ||
7935 | int RGWRados::pool_iterate(RGWPoolIterCtx& ctx, uint32_t num, vector<rgw_bucket_dir_entry>& objs, | |
7936 | bool *is_truncated, RGWAccessListFilter *filter) | |
7937 | { | |
7938 | // catch exceptions from NObjectIterator::operator++() | |
7939 | try { | |
7940 | return do_pool_iterate(cct, ctx, num, objs, is_truncated, filter); | |
7941 | } catch (const std::system_error& e) { | |
7942 | int r = -e.code().value(); | |
7943 | ldout(cct, 10) << "NObjectIterator threw exception " << e.what() | |
7944 | << ", returning " << r << dendl; | |
7945 | return r; | |
7946 | } catch (const std::exception& e) { | |
7947 | ldout(cct, 10) << "NObjectIterator threw exception " << e.what() | |
7948 | << ", returning -5" << dendl; | |
7949 | return -EIO; | |
7950 | } | |
7951 | } | |
7952 | ||
7953 | int RGWRados::list_raw_objects_init(const DoutPrefixProvider *dpp, const rgw_pool& pool, const string& marker, RGWListRawObjsCtx *ctx) | |
7954 | { | |
7955 | if (!ctx->initialized) { | |
7956 | int r = pool_iterate_begin(dpp, pool, marker, ctx->iter_ctx); | |
7957 | if (r < 0) { | |
7958 | ldpp_dout(dpp, 10) << "failed to list objects pool_iterate_begin() returned r=" << r << dendl; | |
7959 | return r; | |
7960 | } | |
7961 | ctx->initialized = true; | |
7962 | } | |
7963 | return 0; | |
7964 | } | |
7965 | ||
7966 | int RGWRados::list_raw_objects_next(const DoutPrefixProvider *dpp, const string& prefix_filter, int max, | |
7967 | RGWListRawObjsCtx& ctx, list<string>& oids, | |
7968 | bool *is_truncated) | |
7969 | { | |
7970 | if (!ctx.initialized) { | |
7971 | return -EINVAL; | |
7972 | } | |
7973 | RGWAccessListFilterPrefix filter(prefix_filter); | |
7974 | vector<rgw_bucket_dir_entry> objs; | |
7975 | int r = pool_iterate(ctx.iter_ctx, max, objs, is_truncated, &filter); | |
7976 | if (r < 0) { | |
7977 | if(r != -ENOENT) | |
7978 | ldpp_dout(dpp, 10) << "failed to list objects pool_iterate returned r=" << r << dendl; | |
7979 | return r; | |
7980 | } | |
7981 | ||
7982 | vector<rgw_bucket_dir_entry>::iterator iter; | |
7983 | for (iter = objs.begin(); iter != objs.end(); ++iter) { | |
7984 | oids.push_back(iter->key.name); | |
7985 | } | |
7986 | ||
7987 | return oids.size(); | |
7988 | } | |
7989 | ||
7990 | int RGWRados::list_raw_objects(const DoutPrefixProvider *dpp, const rgw_pool& pool, const string& prefix_filter, | |
7991 | int max, RGWListRawObjsCtx& ctx, list<string>& oids, | |
7992 | bool *is_truncated) | |
7993 | { | |
7994 | if (!ctx.initialized) { | |
7995 | int r = list_raw_objects_init(dpp, pool, string(), &ctx); | |
7996 | if (r < 0) { | |
7997 | return r; | |
7998 | } | |
7999 | } | |
8000 | ||
8001 | return list_raw_objects_next(dpp, prefix_filter, max, ctx, oids, is_truncated); | |
8002 | } | |
8003 | ||
8004 | string RGWRados::list_raw_objs_get_cursor(RGWListRawObjsCtx& ctx) | |
8005 | { | |
8006 | return pool_iterate_get_cursor(ctx.iter_ctx); | |
8007 | } | |
8008 | ||
8009 | int RGWRados::bi_get_instance(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
8010 | rgw_bucket_dir_entry *dirent) | |
8011 | { | |
8012 | rgw_cls_bi_entry bi_entry; | |
8013 | int r = bi_get(dpp, bucket_info, obj, BIIndexType::Instance, &bi_entry); | |
8014 | if (r < 0 && r != -ENOENT) { | |
8015 | ldpp_dout(dpp, 0) << "ERROR: bi_get() returned r=" << r << dendl; | |
8016 | } | |
8017 | if (r < 0) { | |
8018 | return r; | |
8019 | } | |
8020 | auto iter = bi_entry.data.cbegin(); | |
8021 | try { | |
8022 | decode(*dirent, iter); | |
8023 | } catch (buffer::error& err) { | |
8024 | ldpp_dout(dpp, 0) << "ERROR: failed to decode bi_entry()" << dendl; | |
8025 | return -EIO; | |
8026 | } | |
8027 | ||
8028 | return 0; | |
8029 | } | |
8030 | ||
8031 | int RGWRados::bi_get_olh(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
8032 | rgw_bucket_olh_entry *olh) | |
8033 | { | |
8034 | rgw_cls_bi_entry bi_entry; | |
8035 | int r = bi_get(dpp, bucket_info, obj, BIIndexType::OLH, &bi_entry); | |
8036 | if (r < 0 && r != -ENOENT) { | |
8037 | ldpp_dout(dpp, 0) << "ERROR: bi_get() returned r=" << r << dendl; | |
8038 | } | |
8039 | if (r < 0) { | |
8040 | return r; | |
8041 | } | |
8042 | auto iter = bi_entry.data.cbegin(); | |
8043 | try { | |
8044 | decode(*olh, iter); | |
8045 | } catch (buffer::error& err) { | |
8046 | ldpp_dout(dpp, 0) << "ERROR: failed to decode bi_entry()" << dendl; | |
8047 | return -EIO; | |
8048 | } | |
8049 | ||
8050 | return 0; | |
8051 | } | |
8052 | ||
8053 | int RGWRados::bi_get(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, const rgw_obj& obj, | |
8054 | BIIndexType index_type, rgw_cls_bi_entry *entry) | |
8055 | { | |
8056 | BucketShard bs(this); | |
8057 | int ret = bs.init(dpp, bucket_info, obj); | |
8058 | if (ret < 0) { | |
8059 | ldpp_dout(dpp, 5) << "bs.init() returned ret=" << ret << dendl; | |
8060 | return ret; | |
8061 | } | |
8062 | ||
8063 | cls_rgw_obj_key key(obj.key.get_index_key_name(), obj.key.instance); | |
8064 | ||
8065 | auto& ref = bs.bucket_obj.get_ref(); | |
8066 | ||
8067 | return cls_rgw_bi_get(ref.pool.ioctx(), ref.obj.oid, index_type, key, entry); | |
8068 | } | |
8069 | ||
8070 | void RGWRados::bi_put(ObjectWriteOperation& op, BucketShard& bs, rgw_cls_bi_entry& entry) | |
8071 | { | |
8072 | auto& ref = bs.bucket_obj.get_ref(); | |
8073 | cls_rgw_bi_put(op, ref.obj.oid, entry); | |
8074 | } | |
8075 | ||
8076 | int RGWRados::bi_put(BucketShard& bs, rgw_cls_bi_entry& entry) | |
8077 | { | |
8078 | auto& ref = bs.bucket_obj.get_ref(); | |
8079 | int ret = cls_rgw_bi_put(ref.pool.ioctx(), ref.obj.oid, entry); | |
8080 | if (ret < 0) | |
8081 | return ret; | |
8082 | ||
8083 | return 0; | |
8084 | } | |
8085 | ||
8086 | int RGWRados::bi_put(const DoutPrefixProvider *dpp, rgw_bucket& bucket, rgw_obj& obj, rgw_cls_bi_entry& entry) | |
8087 | { | |
8088 | BucketShard bs(this); | |
8089 | int ret = bs.init(bucket, obj, nullptr /* no RGWBucketInfo */, dpp); | |
8090 | if (ret < 0) { | |
8091 | ldpp_dout(dpp, 5) << "bs.init() returned ret=" << ret << dendl; | |
8092 | return ret; | |
8093 | } | |
8094 | ||
8095 | return bi_put(bs, entry); | |
8096 | } | |
8097 | ||
8098 | int RGWRados::bi_list(const DoutPrefixProvider *dpp, rgw_bucket& bucket, const string& obj_name, const string& marker, uint32_t max, list<rgw_cls_bi_entry> *entries, bool *is_truncated) | |
8099 | { | |
8100 | rgw_obj obj(bucket, obj_name); | |
8101 | BucketShard bs(this); | |
8102 | int ret = bs.init(bucket, obj, nullptr /* no RGWBucketInfo */, dpp); | |
8103 | if (ret < 0) { | |
8104 | ldpp_dout(dpp, 5) << "bs.init() returned ret=" << ret << dendl; | |
8105 | return ret; | |
8106 | } | |
8107 | ||
8108 | auto& ref = bs.bucket_obj.get_ref(); | |
8109 | ret = cls_rgw_bi_list(ref.pool.ioctx(), ref.obj.oid, obj_name, marker, max, entries, is_truncated); | |
8110 | if (ret == -ENOENT) { | |
8111 | *is_truncated = false; | |
8112 | } | |
8113 | if (ret < 0) | |
8114 | return ret; | |
8115 | ||
8116 | return 0; | |
8117 | } | |
8118 | ||
8119 | 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) | |
8120 | { | |
8121 | auto& ref = bs.bucket_obj.get_ref(); | |
8122 | int ret = cls_rgw_bi_list(ref.pool.ioctx(), ref.obj.oid, filter_obj, marker, max, entries, is_truncated); | |
8123 | if (ret < 0) | |
8124 | return ret; | |
8125 | ||
8126 | return 0; | |
8127 | } | |
8128 | ||
8129 | int RGWRados::bi_remove(BucketShard& bs) | |
8130 | { | |
8131 | auto& ref = bs.bucket_obj.get_ref(); | |
8132 | int ret = ref.pool.ioctx().remove(ref.obj.oid); | |
8133 | if (ret == -ENOENT) { | |
8134 | ret = 0; | |
8135 | } | |
8136 | if (ret < 0) { | |
8137 | ldout(cct, 5) << "bs.index_ctx.remove(" << bs.bucket_obj << ") returned ret=" << ret << dendl; | |
8138 | return ret; | |
8139 | } | |
8140 | ||
8141 | return 0; | |
8142 | } | |
8143 | ||
8144 | int RGWRados::bi_list(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, int shard_id, const string& filter_obj, const string& marker, uint32_t max, list<rgw_cls_bi_entry> *entries, bool *is_truncated) | |
8145 | { | |
8146 | BucketShard bs(this); | |
8147 | int ret = bs.init(bucket_info.bucket, shard_id, bucket_info.layout.current_index, nullptr /* no RGWBucketInfo */, dpp); | |
8148 | if (ret < 0) { | |
8149 | ldpp_dout(dpp, 5) << "bs.init() returned ret=" << ret << dendl; | |
8150 | return ret; | |
8151 | } | |
8152 | ||
8153 | return bi_list(bs, filter_obj, marker, max, entries, is_truncated); | |
8154 | } | |
8155 | ||
8156 | int RGWRados::gc_operate(const DoutPrefixProvider *dpp, string& oid, librados::ObjectWriteOperation *op) | |
8157 | { | |
8158 | return rgw_rados_operate(dpp, gc_pool_ctx, oid, op, null_yield); | |
8159 | } | |
8160 | ||
8161 | int RGWRados::gc_aio_operate(const string& oid, librados::AioCompletion *c, | |
8162 | librados::ObjectWriteOperation *op) | |
8163 | { | |
8164 | return gc_pool_ctx.aio_operate(oid, c, op); | |
8165 | } | |
8166 | ||
8167 | int RGWRados::gc_operate(const DoutPrefixProvider *dpp, string& oid, librados::ObjectReadOperation *op, bufferlist *pbl) | |
8168 | { | |
8169 | return rgw_rados_operate(dpp, gc_pool_ctx, oid, op, pbl, null_yield); | |
8170 | } | |
8171 | ||
8172 | 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) | |
8173 | { | |
8174 | return gc->list(index, marker, max, expired_only, result, truncated, processing_queue); | |
8175 | } | |
8176 | ||
8177 | int RGWRados::process_gc(bool expired_only) | |
8178 | { | |
8179 | return gc->process(expired_only); | |
8180 | } | |
8181 | ||
8182 | int RGWRados::list_lc_progress(string& marker, uint32_t max_entries, | |
8183 | vector<rgw::sal::Lifecycle::LCEntry>& progress_map, | |
8184 | int& index) | |
8185 | { | |
8186 | return lc->list_lc_progress(marker, max_entries, progress_map, index); | |
8187 | } | |
8188 | ||
8189 | int RGWRados::process_lc() | |
8190 | { | |
8191 | RGWLC lc; | |
8192 | lc.initialize(cct, this->store); | |
8193 | RGWLC::LCWorker worker(&lc, cct, &lc, 0); | |
8194 | auto ret = lc.process(&worker, true /* once */); | |
8195 | lc.stop_processor(); // sets down_flag, but returns immediately | |
8196 | return ret; | |
8197 | } | |
8198 | ||
8199 | bool RGWRados::process_expire_objects(const DoutPrefixProvider *dpp) | |
8200 | { | |
8201 | return obj_expirer->inspect_all_shards(dpp, utime_t(), ceph_clock_now()); | |
8202 | } | |
8203 | ||
8204 | int RGWRados::cls_obj_prepare_op(const DoutPrefixProvider *dpp, BucketShard& bs, RGWModifyOp op, string& tag, | |
8205 | rgw_obj& obj, uint16_t bilog_flags, optional_yield y, rgw_zone_set *_zones_trace) | |
8206 | { | |
8207 | rgw_zone_set zones_trace; | |
8208 | if (_zones_trace) { | |
8209 | zones_trace = *_zones_trace; | |
8210 | } | |
8211 | zones_trace.insert(svc.zone->get_zone().id, bs.bucket.get_key()); | |
8212 | ||
8213 | ObjectWriteOperation o; | |
8214 | cls_rgw_obj_key key(obj.key.get_index_key_name(), obj.key.instance); | |
8215 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); | |
8216 | cls_rgw_bucket_prepare_op(o, op, tag, key, obj.key.get_loc(), svc.zone->get_zone().log_data, bilog_flags, zones_trace); | |
8217 | return bs.bucket_obj.operate(dpp, &o, y); | |
8218 | } | |
8219 | ||
8220 | int RGWRados::cls_obj_complete_op(BucketShard& bs, const rgw_obj& obj, RGWModifyOp op, string& tag, | |
8221 | int64_t pool, uint64_t epoch, | |
8222 | rgw_bucket_dir_entry& ent, RGWObjCategory category, | |
8223 | list<rgw_obj_index_key> *remove_objs, uint16_t bilog_flags, rgw_zone_set *_zones_trace) | |
8224 | { | |
8225 | ObjectWriteOperation o; | |
8226 | rgw_bucket_dir_entry_meta dir_meta; | |
8227 | dir_meta = ent.meta; | |
8228 | dir_meta.category = category; | |
8229 | ||
8230 | rgw_zone_set zones_trace; | |
8231 | if (_zones_trace) { | |
8232 | zones_trace = *_zones_trace; | |
8233 | } | |
8234 | zones_trace.insert(svc.zone->get_zone().id, bs.bucket.get_key()); | |
8235 | ||
8236 | rgw_bucket_entry_ver ver; | |
8237 | ver.pool = pool; | |
8238 | ver.epoch = epoch; | |
8239 | cls_rgw_obj_key key(ent.key.name, ent.key.instance); | |
8240 | cls_rgw_guard_bucket_resharding(o, -ERR_BUSY_RESHARDING); | |
8241 | cls_rgw_bucket_complete_op(o, op, tag, ver, key, dir_meta, remove_objs, | |
8242 | svc.zone->get_zone().log_data, bilog_flags, &zones_trace); | |
8243 | complete_op_data *arg; | |
8244 | index_completion_manager->create_completion(obj, op, tag, ver, key, dir_meta, remove_objs, | |
8245 | svc.zone->get_zone().log_data, bilog_flags, &zones_trace, &arg); | |
8246 | librados::AioCompletion *completion = arg->rados_completion; | |
8247 | int ret = bs.bucket_obj.aio_operate(arg->rados_completion, &o); | |
8248 | completion->release(); /* can't reference arg here, as it might have already been released */ | |
8249 | return ret; | |
8250 | } | |
8251 | ||
8252 | int RGWRados::cls_obj_complete_add(BucketShard& bs, const rgw_obj& obj, string& tag, | |
8253 | int64_t pool, uint64_t epoch, | |
8254 | rgw_bucket_dir_entry& ent, RGWObjCategory category, | |
8255 | list<rgw_obj_index_key> *remove_objs, uint16_t bilog_flags, rgw_zone_set *zones_trace) | |
8256 | { | |
8257 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_ADD, tag, pool, epoch, ent, category, remove_objs, bilog_flags, zones_trace); | |
8258 | } | |
8259 | ||
8260 | int RGWRados::cls_obj_complete_del(BucketShard& bs, string& tag, | |
8261 | int64_t pool, uint64_t epoch, | |
8262 | rgw_obj& obj, | |
8263 | real_time& removed_mtime, | |
8264 | list<rgw_obj_index_key> *remove_objs, | |
8265 | uint16_t bilog_flags, | |
8266 | rgw_zone_set *zones_trace) | |
8267 | { | |
8268 | rgw_bucket_dir_entry ent; | |
8269 | ent.meta.mtime = removed_mtime; | |
8270 | obj.key.get_index_key(&ent.key); | |
8271 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_DEL, tag, pool, epoch, | |
8272 | ent, RGWObjCategory::None, remove_objs, | |
8273 | bilog_flags, zones_trace); | |
8274 | } | |
8275 | ||
8276 | int RGWRados::cls_obj_complete_cancel(BucketShard& bs, string& tag, rgw_obj& obj, uint16_t bilog_flags, rgw_zone_set *zones_trace) | |
8277 | { | |
8278 | rgw_bucket_dir_entry ent; | |
8279 | obj.key.get_index_key(&ent.key); | |
8280 | return cls_obj_complete_op(bs, obj, CLS_RGW_OP_CANCEL, tag, | |
8281 | -1 /* pool id */, 0, ent, | |
8282 | RGWObjCategory::None, NULL, bilog_flags, | |
8283 | zones_trace); | |
8284 | } | |
8285 | ||
8286 | int RGWRados::cls_obj_set_bucket_tag_timeout(const DoutPrefixProvider *dpp, RGWBucketInfo& bucket_info, uint64_t timeout) | |
8287 | { | |
8288 | RGWSI_RADOS::Pool index_pool; | |
8289 | map<int, string> bucket_objs; | |
8290 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, std::nullopt, &index_pool, &bucket_objs, nullptr); | |
8291 | if (r < 0) | |
8292 | return r; | |
8293 | ||
8294 | return CLSRGWIssueSetTagTimeout(index_pool.ioctx(), bucket_objs, cct->_conf->rgw_bucket_index_max_aio, timeout)(); | |
8295 | } | |
8296 | ||
8297 | ||
8298 | uint32_t RGWRados::calc_ordered_bucket_list_per_shard(uint32_t num_entries, | |
8299 | uint32_t num_shards) | |
8300 | { | |
8301 | // We want to minimize the chances that when num_shards >> | |
8302 | // num_entries that we return much fewer than num_entries to the | |
8303 | // client. Given all the overhead of making a cls call to the osd, | |
8304 | // returning a few entries is not much more work than returning one | |
8305 | // entry. This minimum might be better tuned based on future | |
8306 | // experiments where num_shards >> num_entries. (Note: ">>" should | |
8307 | // be interpreted as "much greater than".) | |
8308 | constexpr uint32_t min_read = 8; | |
8309 | ||
8310 | // The following is based on _"Balls into Bins" -- A Simple and | |
8311 | // Tight Analysis_ by Raab and Steger. We add 1 as a way to handle | |
8312 | // cases when num_shards >> num_entries (it almost serves as a | |
8313 | // ceiling calculation). We also assume alpha is 1.0 and extract it | |
8314 | // from the calculation. Future work could involve memoizing some of | |
8315 | // the transcendental functions to minimize repeatedly re-calling | |
8316 | // them with the same parameters, which we expect to be the case the | |
8317 | // majority of the time. | |
8318 | uint32_t calc_read = | |
8319 | 1 + | |
8320 | static_cast<uint32_t>((num_entries / num_shards) + | |
8321 | sqrt((2 * num_entries) * | |
8322 | log(num_shards) / num_shards)); | |
8323 | ||
8324 | return std::max(min_read, calc_read); | |
8325 | } | |
8326 | ||
8327 | ||
8328 | int RGWRados::cls_bucket_list_ordered(const DoutPrefixProvider *dpp, | |
8329 | RGWBucketInfo& bucket_info, | |
8330 | const int shard_id, | |
8331 | const rgw_obj_index_key& start_after, | |
8332 | const string& prefix, | |
8333 | const string& delimiter, | |
8334 | const uint32_t num_entries, | |
8335 | const bool list_versions, | |
8336 | const uint16_t expansion_factor, | |
8337 | ent_map_t& m, | |
8338 | bool* is_truncated, | |
8339 | bool* cls_filtered, | |
8340 | rgw_obj_index_key *last_entry, | |
8341 | optional_yield y, | |
8342 | check_filter_t force_check_filter) | |
8343 | { | |
8344 | /* expansion_factor allows the number of entries to read to grow | |
8345 | * exponentially; this is used when earlier reads are producing too | |
8346 | * few results, perhaps due to filtering or to a series of | |
8347 | * namespaced entries */ | |
8348 | ||
8349 | ldpp_dout(dpp, 10) << "RGWRados::" << __func__ << ": " << bucket_info.bucket << | |
8350 | " start_after=\"" << start_after.name << | |
8351 | "[" << start_after.instance << | |
8352 | "]\", prefix=\"" << prefix << | |
8353 | "\" num_entries=" << num_entries << | |
8354 | ", list_versions=" << list_versions << | |
8355 | ", expansion_factor=" << expansion_factor << dendl; | |
8356 | ||
8357 | m.clear(); | |
8358 | ||
8359 | RGWSI_RADOS::Pool index_pool; | |
8360 | // key - oid (for different shards if there is any) | |
8361 | // value - list result for the corresponding oid (shard), it is filled by | |
8362 | // the AIO callback | |
8363 | map<int, string> shard_oids; | |
8364 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, shard_id, | |
8365 | &index_pool, &shard_oids, | |
8366 | nullptr); | |
8367 | if (r < 0) { | |
8368 | return r; | |
8369 | } | |
8370 | ||
8371 | const uint32_t shard_count = shard_oids.size(); | |
8372 | uint32_t num_entries_per_shard; | |
8373 | if (expansion_factor == 0) { | |
8374 | num_entries_per_shard = | |
8375 | calc_ordered_bucket_list_per_shard(num_entries, shard_count); | |
8376 | } else if (expansion_factor <= 11) { | |
8377 | // we'll max out the exponential multiplication factor at 1024 (2<<10) | |
8378 | num_entries_per_shard = | |
8379 | std::min(num_entries, | |
8380 | (uint32_t(1 << (expansion_factor - 1)) * | |
8381 | calc_ordered_bucket_list_per_shard(num_entries, shard_count))); | |
8382 | } else { | |
8383 | num_entries_per_shard = num_entries; | |
8384 | } | |
8385 | ||
8386 | ldpp_dout(dpp, 10) << "RGWRados::" << __func__ << | |
8387 | " request from each of " << shard_count << | |
8388 | " shard(s) for " << num_entries_per_shard << " entries to get " << | |
8389 | num_entries << " total entries" << dendl; | |
8390 | ||
8391 | auto& ioctx = index_pool.ioctx(); | |
8392 | map<int, rgw_cls_list_ret> shard_list_results; | |
8393 | cls_rgw_obj_key start_after_key(start_after.name, start_after.instance); | |
8394 | r = CLSRGWIssueBucketList(ioctx, start_after_key, prefix, delimiter, | |
8395 | num_entries_per_shard, | |
8396 | list_versions, shard_oids, shard_list_results, | |
8397 | cct->_conf->rgw_bucket_index_max_aio)(); | |
8398 | if (r < 0) { | |
8399 | return r; | |
8400 | } | |
8401 | ||
8402 | // to manage the iterators through each shard's list results | |
8403 | struct ShardTracker { | |
8404 | const size_t shard_idx; | |
8405 | rgw_cls_list_ret& result; | |
8406 | const std::string& oid_name; | |
8407 | RGWRados::ent_map_t::iterator cursor; | |
8408 | RGWRados::ent_map_t::iterator end; | |
8409 | ||
8410 | // manages an iterator through a shard and provides other | |
8411 | // accessors | |
8412 | ShardTracker(size_t _shard_idx, | |
8413 | rgw_cls_list_ret& _result, | |
8414 | const std::string& _oid_name): | |
8415 | shard_idx(_shard_idx), | |
8416 | result(_result), | |
8417 | oid_name(_oid_name), | |
8418 | cursor(_result.dir.m.begin()), | |
8419 | end(_result.dir.m.end()) | |
8420 | {} | |
8421 | ||
8422 | inline const std::string& entry_name() const { | |
8423 | return cursor->first; | |
8424 | } | |
8425 | rgw_bucket_dir_entry& dir_entry() const { | |
8426 | return cursor->second; | |
8427 | } | |
8428 | inline bool is_truncated() const { | |
8429 | return result.is_truncated; | |
8430 | } | |
8431 | inline ShardTracker& advance() { | |
8432 | ++cursor; | |
8433 | // return a self-reference to allow for chaining of calls, such | |
8434 | // as x.advance().at_end() | |
8435 | return *this; | |
8436 | } | |
8437 | inline bool at_end() const { | |
8438 | return cursor == end; | |
8439 | } | |
8440 | }; // ShardTracker | |
8441 | ||
8442 | // add the next unique candidate, or return false if we reach the end | |
8443 | auto next_candidate = [] (CephContext *cct, ShardTracker& t, | |
8444 | std::map<std::string, size_t>& candidates, | |
8445 | size_t tracker_idx) { | |
8446 | while (!t.at_end()) { | |
8447 | if (candidates.emplace(t.entry_name(), tracker_idx).second) { | |
8448 | return; | |
8449 | } | |
8450 | t.advance(); // skip duplicate common prefixes | |
8451 | } | |
8452 | }; | |
8453 | ||
8454 | // one tracker per shard requested (may not be all shards) | |
8455 | std::vector<ShardTracker> results_trackers; | |
8456 | results_trackers.reserve(shard_list_results.size()); | |
8457 | for (auto& r : shard_list_results) { | |
8458 | results_trackers.emplace_back(r.first, r.second, shard_oids[r.first]); | |
8459 | ||
8460 | // if any *one* shard's result is trucated, the entire result is | |
8461 | // truncated | |
8462 | *is_truncated = *is_truncated || r.second.is_truncated; | |
8463 | ||
8464 | // unless *all* are shards are cls_filtered, the entire result is | |
8465 | // not filtered | |
8466 | *cls_filtered = *cls_filtered && r.second.cls_filtered; | |
8467 | } | |
8468 | ||
8469 | // create a map to track the next candidate entry from ShardTracker | |
8470 | // (key=candidate, value=index into results_trackers); as we consume | |
8471 | // entries from shards, we replace them with the next entries in the | |
8472 | // shards until we run out | |
8473 | map<string, size_t> candidates; | |
8474 | size_t tracker_idx = 0; | |
8475 | for (auto& t : results_trackers) { | |
8476 | // it's important that the values in the map refer to the index | |
8477 | // into the results_trackers vector, which may not be the same | |
8478 | // as the shard number (i.e., when not all shards are requested) | |
8479 | next_candidate(cct, t, candidates, tracker_idx); | |
8480 | ++tracker_idx; | |
8481 | } | |
8482 | ||
8483 | rgw_bucket_dir_entry* | |
8484 | last_entry_visited = nullptr; // to set last_entry (marker) | |
8485 | map<string, bufferlist> updates; | |
8486 | uint32_t count = 0; | |
8487 | while (count < num_entries && !candidates.empty()) { | |
8488 | r = 0; | |
8489 | // select the next entry in lexical order (first key in map); | |
8490 | // again tracker_idx is not necessarily shard number, but is index | |
8491 | // into results_trackers vector | |
8492 | tracker_idx = candidates.begin()->second; | |
8493 | auto& tracker = results_trackers.at(tracker_idx); | |
8494 | ||
8495 | const string& name = tracker.entry_name(); | |
8496 | rgw_bucket_dir_entry& dirent = tracker.dir_entry(); | |
8497 | ||
8498 | ldpp_dout(dpp, 20) << "RGWRados::" << __func__ << " currently processing " << | |
8499 | dirent.key << " from shard " << tracker.shard_idx << dendl; | |
8500 | ||
8501 | const bool force_check = | |
8502 | force_check_filter && force_check_filter(dirent.key.name); | |
8503 | ||
8504 | if ((!dirent.exists && | |
8505 | !dirent.is_delete_marker() && | |
8506 | !dirent.is_common_prefix()) || | |
8507 | !dirent.pending_map.empty() || | |
8508 | force_check) { | |
8509 | /* there are uncommitted ops. We need to check the current | |
8510 | * state, and if the tags are old we need to do clean-up as | |
8511 | * well. */ | |
8512 | librados::IoCtx sub_ctx; | |
8513 | sub_ctx.dup(ioctx); | |
8514 | r = check_disk_state(dpp, sub_ctx, bucket_info, dirent, dirent, | |
8515 | updates[tracker.oid_name], y); | |
8516 | if (r < 0 && r != -ENOENT) { | |
8517 | return r; | |
8518 | } | |
8519 | } else { | |
8520 | r = 0; | |
8521 | } | |
8522 | ||
8523 | if (r >= 0) { | |
8524 | ldpp_dout(dpp, 10) << "RGWRados::" << __func__ << ": got " << | |
8525 | dirent.key.name << "[" << dirent.key.instance << "]" << dendl; | |
8526 | m[name] = std::move(dirent); | |
8527 | last_entry_visited = &(m[name]); | |
8528 | ++count; | |
8529 | } else { | |
8530 | ldpp_dout(dpp, 10) << "RGWRados::" << __func__ << ": skipping " << | |
8531 | dirent.key.name << "[" << dirent.key.instance << "]" << dendl; | |
8532 | last_entry_visited = &tracker.dir_entry(); | |
8533 | } | |
8534 | ||
8535 | // refresh the candidates map | |
8536 | candidates.erase(candidates.begin()); | |
8537 | tracker.advance(); | |
8538 | ||
8539 | next_candidate(cct, tracker, candidates, tracker_idx); | |
8540 | ||
8541 | if (tracker.at_end() && tracker.is_truncated()) { | |
8542 | // once we exhaust one shard that is truncated, we need to stop, | |
8543 | // as we cannot be certain that one of the next entries needs to | |
8544 | // come from that shard; S3 and swift protocols allow returning | |
8545 | // fewer than what was requested | |
8546 | break; | |
8547 | } | |
8548 | } // while we haven't provided requested # of result entries | |
8549 | ||
8550 | // suggest updates if there are any | |
8551 | for (auto& miter : updates) { | |
8552 | if (miter.second.length()) { | |
8553 | ObjectWriteOperation o; | |
8554 | cls_rgw_suggest_changes(o, miter.second); | |
8555 | // we don't care if we lose suggested updates, send them off blindly | |
8556 | AioCompletion *c = | |
8557 | librados::Rados::aio_create_completion(nullptr, nullptr); | |
8558 | ioctx.aio_operate(miter.first, c, &o); | |
8559 | c->release(); | |
8560 | } | |
8561 | } // updates loop | |
8562 | ||
8563 | // determine truncation by checking if all the returned entries are | |
8564 | // consumed or not | |
8565 | *is_truncated = false; | |
8566 | for (const auto& t : results_trackers) { | |
8567 | if (!t.at_end() || t.is_truncated()) { | |
8568 | *is_truncated = true; | |
8569 | break; | |
8570 | } | |
8571 | } | |
8572 | ||
8573 | ldpp_dout(dpp, 20) << "RGWRados::" << __func__ << | |
8574 | ": returning, count=" << count << ", is_truncated=" << *is_truncated << | |
8575 | dendl; | |
8576 | ||
8577 | if (*is_truncated && count < num_entries) { | |
8578 | ldpp_dout(dpp, 10) << "RGWRados::" << __func__ << | |
8579 | ": INFO requested " << num_entries << " entries but returning " << | |
8580 | count << ", which is truncated" << dendl; | |
8581 | } | |
8582 | ||
8583 | if (last_entry_visited != nullptr && last_entry) { | |
8584 | *last_entry = last_entry_visited->key; | |
8585 | ldpp_dout(dpp, 20) << "RGWRados::" << __func__ << | |
8586 | ": returning, last_entry=" << *last_entry << dendl; | |
8587 | } else { | |
8588 | ldpp_dout(dpp, 20) << "RGWRados::" << __func__ << | |
8589 | ": returning, last_entry NOT SET" << dendl; | |
8590 | } | |
8591 | ||
8592 | return 0; | |
8593 | } | |
8594 | ||
8595 | ||
8596 | int RGWRados::cls_bucket_list_unordered(const DoutPrefixProvider *dpp, | |
8597 | RGWBucketInfo& bucket_info, | |
8598 | int shard_id, | |
8599 | const rgw_obj_index_key& start_after, | |
8600 | const string& prefix, | |
8601 | uint32_t num_entries, | |
8602 | bool list_versions, | |
8603 | std::vector<rgw_bucket_dir_entry>& ent_list, | |
8604 | bool *is_truncated, | |
8605 | rgw_obj_index_key *last_entry, | |
8606 | optional_yield y, | |
8607 | check_filter_t force_check_filter) { | |
8608 | ldpp_dout(dpp, 10) << "cls_bucket_list_unordered " << bucket_info.bucket << | |
8609 | " start_after " << start_after.name << "[" << start_after.instance << | |
8610 | "] num_entries " << num_entries << dendl; | |
8611 | ||
8612 | ent_list.clear(); | |
8613 | static MultipartMetaFilter multipart_meta_filter; | |
8614 | ||
8615 | *is_truncated = false; | |
8616 | RGWSI_RADOS::Pool index_pool; | |
8617 | ||
8618 | map<int, string> oids; | |
8619 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, shard_id, &index_pool, &oids, nullptr); | |
8620 | if (r < 0) | |
8621 | return r; | |
8622 | ||
8623 | auto& ioctx = index_pool.ioctx(); | |
8624 | ||
8625 | const uint32_t num_shards = oids.size(); | |
8626 | ||
8627 | rgw_obj_index_key marker = start_after; | |
8628 | uint32_t current_shard; | |
8629 | if (shard_id >= 0) { | |
8630 | current_shard = shard_id; | |
8631 | } else if (start_after.empty()) { | |
8632 | current_shard = 0u; | |
8633 | } else { | |
8634 | // at this point we have a marker (start_after) that has something | |
8635 | // in it, so we need to get to the bucket shard index, so we can | |
8636 | // start reading from there | |
8637 | ||
8638 | std::string key; | |
8639 | // test whether object name is a multipart meta name | |
8640 | if(! multipart_meta_filter.filter(start_after.name, key)) { | |
8641 | // if multipart_meta_filter fails, must be "regular" (i.e., | |
8642 | // unadorned) and the name is the key | |
8643 | key = start_after.name; | |
8644 | } | |
8645 | ||
8646 | // now convert the key (oid) to an rgw_obj_key since that will | |
8647 | // separate out the namespace, name, and instance | |
8648 | rgw_obj_key obj_key; | |
8649 | bool parsed = rgw_obj_key::parse_raw_oid(key, &obj_key); | |
8650 | if (!parsed) { | |
8651 | ldpp_dout(dpp, 0) << | |
8652 | "ERROR: RGWRados::cls_bucket_list_unordered received an invalid " | |
8653 | "start marker: '" << start_after << "'" << dendl; | |
8654 | return -EINVAL; | |
8655 | } else if (obj_key.name.empty()) { | |
8656 | // if the name is empty that means the object name came in with | |
8657 | // a namespace only, and therefore we need to start our scan at | |
8658 | // the first bucket index shard | |
8659 | current_shard = 0u; | |
8660 | } else { | |
8661 | // so now we have the key used to compute the bucket index shard | |
8662 | // and can extract the specific shard from it | |
8663 | current_shard = svc.bi_rados->bucket_shard_index(obj_key.name, num_shards); | |
8664 | } | |
8665 | } | |
8666 | ||
8667 | uint32_t count = 0u; | |
8668 | map<string, bufferlist> updates; | |
8669 | rgw_obj_index_key last_added_entry; | |
8670 | while (count <= num_entries && | |
8671 | ((shard_id >= 0 && current_shard == uint32_t(shard_id)) || | |
8672 | current_shard < num_shards)) { | |
8673 | const std::string& oid = oids[current_shard]; | |
8674 | rgw_cls_list_ret result; | |
8675 | ||
8676 | librados::ObjectReadOperation op; | |
8677 | string empty_delimiter; | |
8678 | cls_rgw_bucket_list_op(op, marker, prefix, empty_delimiter, | |
8679 | num_entries, | |
8680 | list_versions, &result); | |
8681 | r = rgw_rados_operate(dpp, ioctx, oid, &op, nullptr, null_yield); | |
8682 | if (r < 0) | |
8683 | return r; | |
8684 | ||
8685 | for (auto& entry : result.dir.m) { | |
8686 | rgw_bucket_dir_entry& dirent = entry.second; | |
8687 | ||
8688 | bool force_check = force_check_filter && | |
8689 | force_check_filter(dirent.key.name); | |
8690 | if ((!dirent.exists && !dirent.is_delete_marker()) || | |
8691 | !dirent.pending_map.empty() || | |
8692 | force_check) { | |
8693 | /* there are uncommitted ops. We need to check the current state, | |
8694 | * and if the tags are old we need to do cleanup as well. */ | |
8695 | librados::IoCtx sub_ctx; | |
8696 | sub_ctx.dup(ioctx); | |
8697 | r = check_disk_state(dpp, sub_ctx, bucket_info, dirent, dirent, updates[oid], y); | |
8698 | if (r < 0 && r != -ENOENT) { | |
8699 | return r; | |
8700 | } | |
8701 | } else { | |
8702 | r = 0; | |
8703 | } | |
8704 | ||
8705 | // at this point either r >=0 or r == -ENOENT | |
8706 | if (r >= 0) { // i.e., if r != -ENOENT | |
8707 | ldpp_dout(dpp, 10) << "RGWRados::cls_bucket_list_unordered: got " << | |
8708 | dirent.key.name << "[" << dirent.key.instance << "]" << dendl; | |
8709 | ||
8710 | if (count < num_entries) { | |
8711 | marker = last_added_entry = dirent.key; // double assign | |
8712 | ent_list.emplace_back(std::move(dirent)); | |
8713 | ++count; | |
8714 | } else { | |
8715 | *is_truncated = true; | |
8716 | goto check_updates; | |
8717 | } | |
8718 | } else { // r == -ENOENT | |
8719 | // in the case of -ENOENT, make sure we're advancing marker | |
8720 | // for possible next call to CLSRGWIssueBucketList | |
8721 | marker = dirent.key; | |
8722 | } | |
8723 | } // entry for loop | |
8724 | ||
8725 | if (!result.is_truncated) { | |
8726 | // if we reached the end of the shard read next shard | |
8727 | ++current_shard; | |
8728 | marker = rgw_obj_index_key(); | |
8729 | } | |
8730 | } // shard loop | |
8731 | ||
8732 | check_updates: | |
8733 | ||
8734 | // suggest updates if there is any | |
8735 | map<string, bufferlist>::iterator miter = updates.begin(); | |
8736 | for (; miter != updates.end(); ++miter) { | |
8737 | if (miter->second.length()) { | |
8738 | ObjectWriteOperation o; | |
8739 | cls_rgw_suggest_changes(o, miter->second); | |
8740 | // we don't care if we lose suggested updates, send them off blindly | |
8741 | AioCompletion *c = librados::Rados::aio_create_completion(nullptr, nullptr); | |
8742 | ioctx.aio_operate(miter->first, c, &o); | |
8743 | c->release(); | |
8744 | } | |
8745 | } | |
8746 | ||
8747 | if (last_entry && !ent_list.empty()) { | |
8748 | *last_entry = last_added_entry; | |
8749 | } | |
8750 | ||
8751 | return 0; | |
8752 | } // RGWRados::cls_bucket_list_unordered | |
8753 | ||
8754 | ||
8755 | int RGWRados::cls_obj_usage_log_add(const DoutPrefixProvider *dpp, const string& oid, | |
8756 | rgw_usage_log_info& info) | |
8757 | { | |
8758 | rgw_raw_obj obj(svc.zone->get_zone_params().usage_log_pool, oid); | |
8759 | ||
8760 | rgw_rados_ref ref; | |
8761 | int r = get_raw_obj_ref(dpp, obj, &ref); | |
8762 | if (r < 0) { | |
8763 | return r; | |
8764 | } | |
8765 | ||
8766 | ObjectWriteOperation op; | |
8767 | cls_rgw_usage_log_add(op, info); | |
8768 | ||
8769 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
8770 | return r; | |
8771 | } | |
8772 | ||
8773 | int RGWRados::cls_obj_usage_log_read(const DoutPrefixProvider *dpp, const string& oid, const string& user, const string& bucket, | |
8774 | uint64_t start_epoch, uint64_t end_epoch, uint32_t max_entries, | |
8775 | string& read_iter, map<rgw_user_bucket, rgw_usage_log_entry>& usage, | |
8776 | bool *is_truncated) | |
8777 | { | |
8778 | rgw_raw_obj obj(svc.zone->get_zone_params().usage_log_pool, oid); | |
8779 | ||
8780 | rgw_rados_ref ref; | |
8781 | int r = get_raw_obj_ref(dpp, obj, &ref); | |
8782 | if (r < 0) { | |
8783 | return r; | |
8784 | } | |
8785 | ||
8786 | *is_truncated = false; | |
8787 | ||
8788 | r = cls_rgw_usage_log_read(ref.pool.ioctx(), ref.obj.oid, user, bucket, start_epoch, end_epoch, | |
8789 | max_entries, read_iter, usage, is_truncated); | |
8790 | ||
8791 | return r; | |
8792 | } | |
8793 | ||
8794 | static int cls_rgw_usage_log_trim_repeat(const DoutPrefixProvider *dpp, rgw_rados_ref ref, const string& user, const string& bucket, uint64_t start_epoch, uint64_t end_epoch) | |
8795 | { | |
8796 | bool done = false; | |
8797 | do { | |
8798 | librados::ObjectWriteOperation op; | |
8799 | cls_rgw_usage_log_trim(op, user, bucket, start_epoch, end_epoch); | |
8800 | int r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
8801 | if (r == -ENODATA) | |
8802 | done = true; | |
8803 | else if (r < 0) | |
8804 | return r; | |
8805 | } while (!done); | |
8806 | ||
8807 | return 0; | |
8808 | } | |
8809 | ||
8810 | int RGWRados::cls_obj_usage_log_trim(const DoutPrefixProvider *dpp, const string& oid, const string& user, const string& bucket, | |
8811 | uint64_t start_epoch, uint64_t end_epoch) | |
8812 | { | |
8813 | rgw_raw_obj obj(svc.zone->get_zone_params().usage_log_pool, oid); | |
8814 | ||
8815 | rgw_rados_ref ref; | |
8816 | int r = get_raw_obj_ref(dpp, obj, &ref); | |
8817 | if (r < 0) { | |
8818 | return r; | |
8819 | } | |
8820 | ||
8821 | r = cls_rgw_usage_log_trim_repeat(dpp, ref, user, bucket, start_epoch, end_epoch); | |
8822 | return r; | |
8823 | } | |
8824 | ||
8825 | int RGWRados::cls_obj_usage_log_clear(const DoutPrefixProvider *dpp, string& oid) | |
8826 | { | |
8827 | rgw_raw_obj obj(svc.zone->get_zone_params().usage_log_pool, oid); | |
8828 | ||
8829 | rgw_rados_ref ref; | |
8830 | int r = get_raw_obj_ref(dpp, obj, &ref); | |
8831 | if (r < 0) { | |
8832 | return r; | |
8833 | } | |
8834 | librados::ObjectWriteOperation op; | |
8835 | cls_rgw_usage_log_clear(op); | |
8836 | r = rgw_rados_operate(dpp, ref.pool.ioctx(), ref.obj.oid, &op, null_yield); | |
8837 | return r; | |
8838 | } | |
8839 | ||
8840 | ||
8841 | int RGWRados::remove_objs_from_index(const DoutPrefixProvider *dpp, RGWBucketInfo& bucket_info, list<rgw_obj_index_key>& oid_list) | |
8842 | { | |
8843 | RGWSI_RADOS::Pool index_pool; | |
8844 | string dir_oid; | |
8845 | ||
8846 | uint8_t suggest_flag = (svc.zone->get_zone().log_data ? CEPH_RGW_DIR_SUGGEST_LOG_OP : 0); | |
8847 | ||
8848 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, &index_pool, &dir_oid); | |
8849 | if (r < 0) | |
8850 | return r; | |
8851 | ||
8852 | bufferlist updates; | |
8853 | ||
8854 | for (auto iter = oid_list.begin(); iter != oid_list.end(); ++iter) { | |
8855 | rgw_bucket_dir_entry entry; | |
8856 | entry.key = *iter; | |
8857 | ldpp_dout(dpp, 2) << "RGWRados::remove_objs_from_index bucket=" << bucket_info.bucket << " obj=" << entry.key.name << ":" << entry.key.instance << dendl; | |
8858 | entry.ver.epoch = (uint64_t)-1; // ULLONG_MAX, needed to that objclass doesn't skip out request | |
8859 | updates.append(CEPH_RGW_REMOVE | suggest_flag); | |
8860 | encode(entry, updates); | |
8861 | } | |
8862 | ||
8863 | bufferlist out; | |
8864 | ||
8865 | r = index_pool.ioctx().exec(dir_oid, RGW_CLASS, RGW_DIR_SUGGEST_CHANGES, updates, out); | |
8866 | ||
8867 | return r; | |
8868 | } | |
8869 | ||
8870 | int RGWRados::check_disk_state(const DoutPrefixProvider *dpp, | |
8871 | librados::IoCtx io_ctx, | |
8872 | const RGWBucketInfo& bucket_info, | |
8873 | rgw_bucket_dir_entry& list_state, | |
8874 | rgw_bucket_dir_entry& object, | |
8875 | bufferlist& suggested_updates, | |
8876 | optional_yield y) | |
8877 | { | |
8878 | const rgw_bucket& bucket = bucket_info.bucket; | |
8879 | uint8_t suggest_flag = (svc.zone->get_zone().log_data ? CEPH_RGW_DIR_SUGGEST_LOG_OP : 0); | |
8880 | ||
8881 | std::string loc; | |
8882 | ||
8883 | rgw_obj obj(bucket, list_state.key); | |
8884 | ||
8885 | string oid; | |
8886 | get_obj_bucket_and_oid_loc(obj, oid, loc); | |
8887 | ||
8888 | if (loc != list_state.locator) { | |
8889 | ldpp_dout(dpp, 0) << "WARNING: generated locator (" << loc << ") is different from listed locator (" << list_state.locator << ")" << dendl; | |
8890 | } | |
8891 | ||
8892 | io_ctx.locator_set_key(list_state.locator); | |
8893 | ||
8894 | RGWObjState *astate = NULL; | |
8895 | RGWObjectCtx rctx(this->store); | |
8896 | int r = get_obj_state(dpp, &rctx, bucket_info, obj, &astate, false, y); | |
8897 | if (r < 0) | |
8898 | return r; | |
8899 | ||
8900 | list_state.pending_map.clear(); // we don't need this and it inflates size | |
8901 | if (!list_state.is_delete_marker() && !astate->exists) { | |
8902 | /* object doesn't exist right now -- hopefully because it's | |
8903 | * marked as !exists and got deleted */ | |
8904 | if (list_state.exists) { | |
8905 | /* FIXME: what should happen now? Work out if there are any | |
8906 | * non-bad ways this could happen (there probably are, but annoying | |
8907 | * to handle!) */ | |
8908 | } | |
8909 | // encode a suggested removal of that key | |
8910 | list_state.ver.epoch = io_ctx.get_last_version(); | |
8911 | list_state.ver.pool = io_ctx.get_id(); | |
8912 | cls_rgw_encode_suggestion(CEPH_RGW_REMOVE, list_state, suggested_updates); | |
8913 | return -ENOENT; | |
8914 | } | |
8915 | ||
8916 | string etag; | |
8917 | string content_type; | |
8918 | ACLOwner owner; | |
8919 | ||
8920 | object.meta.size = astate->size; | |
8921 | object.meta.accounted_size = astate->accounted_size; | |
8922 | object.meta.mtime = astate->mtime; | |
8923 | ||
8924 | map<string, bufferlist>::iterator iter = astate->attrset.find(RGW_ATTR_ETAG); | |
8925 | if (iter != astate->attrset.end()) { | |
8926 | etag = rgw_bl_str(iter->second); | |
8927 | } | |
8928 | iter = astate->attrset.find(RGW_ATTR_CONTENT_TYPE); | |
8929 | if (iter != astate->attrset.end()) { | |
8930 | content_type = rgw_bl_str(iter->second); | |
8931 | } | |
8932 | iter = astate->attrset.find(RGW_ATTR_ACL); | |
8933 | if (iter != astate->attrset.end()) { | |
8934 | r = decode_policy(iter->second, &owner); | |
8935 | if (r < 0) { | |
8936 | ldpp_dout(dpp, 0) << "WARNING: could not decode policy for object: " << obj << dendl; | |
8937 | } | |
8938 | } | |
8939 | ||
8940 | if (astate->manifest) { | |
8941 | RGWObjManifest::obj_iterator miter; | |
8942 | RGWObjManifest& manifest = *astate->manifest; | |
8943 | for (miter = manifest.obj_begin(dpp); miter != manifest.obj_end(dpp); ++miter) { | |
8944 | const rgw_raw_obj& raw_loc = miter.get_location().get_raw_obj(store); | |
8945 | rgw_obj loc; | |
8946 | RGWSI_Tier_RADOS::raw_obj_to_obj(manifest.get_obj().bucket, raw_loc, &loc); | |
8947 | ||
8948 | if (loc.key.ns == RGW_OBJ_NS_MULTIPART) { | |
8949 | ldpp_dout(dpp, 0) << "check_disk_state(): removing manifest part from index: " << loc << dendl; | |
8950 | r = delete_obj_index(loc, astate->mtime, dpp); | |
8951 | if (r < 0) { | |
8952 | ldpp_dout(dpp, 0) << "WARNING: delete_obj_index() returned r=" << r << dendl; | |
8953 | } | |
8954 | } | |
8955 | } | |
8956 | } | |
8957 | ||
8958 | object.meta.etag = etag; | |
8959 | object.meta.content_type = content_type; | |
8960 | object.meta.owner = owner.get_id().to_str(); | |
8961 | object.meta.owner_display_name = owner.get_display_name(); | |
8962 | ||
8963 | // encode suggested updates | |
8964 | list_state.ver.pool = io_ctx.get_id(); | |
8965 | list_state.ver.epoch = astate->epoch; | |
8966 | list_state.meta.size = object.meta.size; | |
8967 | list_state.meta.accounted_size = object.meta.accounted_size; | |
8968 | list_state.meta.mtime = object.meta.mtime; | |
8969 | list_state.meta.category = main_category; | |
8970 | list_state.meta.etag = etag; | |
8971 | list_state.meta.content_type = content_type; | |
8972 | if (astate->obj_tag.length() > 0) | |
8973 | list_state.tag = astate->obj_tag.c_str(); | |
8974 | list_state.meta.owner = owner.get_id().to_str(); | |
8975 | list_state.meta.owner_display_name = owner.get_display_name(); | |
8976 | ||
8977 | list_state.exists = true; | |
8978 | cls_rgw_encode_suggestion(CEPH_RGW_UPDATE | suggest_flag, list_state, suggested_updates); | |
8979 | return 0; | |
8980 | } | |
8981 | ||
8982 | int RGWRados::cls_bucket_head(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, int shard_id, vector<rgw_bucket_dir_header>& headers, map<int, string> *bucket_instance_ids) | |
8983 | { | |
8984 | RGWSI_RADOS::Pool index_pool; | |
8985 | map<int, string> oids; | |
8986 | map<int, struct rgw_cls_list_ret> list_results; | |
8987 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, shard_id, &index_pool, &oids, bucket_instance_ids); | |
8988 | if (r < 0) { | |
8989 | ldpp_dout(dpp, 20) << "cls_bucket_head: open_bucket_index() returned " | |
8990 | << r << dendl; | |
8991 | return r; | |
8992 | } | |
8993 | ||
8994 | r = CLSRGWIssueGetDirHeader(index_pool.ioctx(), oids, list_results, cct->_conf->rgw_bucket_index_max_aio)(); | |
8995 | if (r < 0) { | |
8996 | ldpp_dout(dpp, 20) << "cls_bucket_head: CLSRGWIssueGetDirHeader() returned " | |
8997 | << r << dendl; | |
8998 | return r; | |
8999 | } | |
9000 | ||
9001 | map<int, struct rgw_cls_list_ret>::iterator iter = list_results.begin(); | |
9002 | for(; iter != list_results.end(); ++iter) { | |
9003 | headers.push_back(std::move(iter->second.dir.header)); | |
9004 | } | |
9005 | return 0; | |
9006 | } | |
9007 | ||
9008 | int RGWRados::cls_bucket_head_async(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, int shard_id, RGWGetDirHeader_CB *ctx, int *num_aio) | |
9009 | { | |
9010 | RGWSI_RADOS::Pool index_pool; | |
9011 | map<int, string> bucket_objs; | |
9012 | int r = svc.bi_rados->open_bucket_index(dpp, bucket_info, shard_id, &index_pool, &bucket_objs, nullptr); | |
9013 | if (r < 0) | |
9014 | return r; | |
9015 | ||
9016 | map<int, string>::iterator iter = bucket_objs.begin(); | |
9017 | for (; iter != bucket_objs.end(); ++iter) { | |
9018 | r = cls_rgw_get_dir_header_async(index_pool.ioctx(), iter->second, static_cast<RGWGetDirHeader_CB*>(ctx->get())); | |
9019 | if (r < 0) { | |
9020 | ctx->put(); | |
9021 | break; | |
9022 | } else { | |
9023 | (*num_aio)++; | |
9024 | } | |
9025 | } | |
9026 | return r; | |
9027 | } | |
9028 | ||
9029 | int RGWRados::check_bucket_shards(const RGWBucketInfo& bucket_info, | |
9030 | const rgw_bucket& bucket, | |
9031 | uint64_t num_objs, | |
9032 | const DoutPrefixProvider *dpp) | |
9033 | { | |
9034 | if (! cct->_conf.get_val<bool>("rgw_dynamic_resharding")) { | |
9035 | return 0; | |
9036 | } | |
9037 | ||
9038 | bool need_resharding = false; | |
9039 | uint32_t num_source_shards = | |
9040 | (bucket_info.layout.current_index.layout.normal.num_shards > 0 ? bucket_info.layout.current_index.layout.normal.num_shards : 1); | |
9041 | const uint32_t max_dynamic_shards = | |
9042 | uint32_t(cct->_conf.get_val<uint64_t>("rgw_max_dynamic_shards")); | |
9043 | ||
9044 | if (num_source_shards >= max_dynamic_shards) { | |
9045 | return 0; | |
9046 | } | |
9047 | ||
9048 | uint32_t suggested_num_shards = 0; | |
9049 | const uint64_t max_objs_per_shard = | |
9050 | cct->_conf.get_val<uint64_t>("rgw_max_objs_per_shard"); | |
9051 | ||
9052 | quota_handler->check_bucket_shards(max_objs_per_shard, num_source_shards, | |
9053 | num_objs, need_resharding, &suggested_num_shards); | |
9054 | if (! need_resharding) { | |
9055 | return 0; | |
9056 | } | |
9057 | ||
9058 | const uint32_t final_num_shards = | |
9059 | RGWBucketReshard::get_preferred_shards(suggested_num_shards, | |
9060 | max_dynamic_shards); | |
9061 | // final verification, so we don't reduce number of shards | |
9062 | if (final_num_shards <= num_source_shards) { | |
9063 | return 0; | |
9064 | } | |
9065 | ||
9066 | ldpp_dout(dpp, 1) << "RGWRados::" << __func__ << " bucket " << bucket.name << | |
9067 | " needs resharding; current num shards " << bucket_info.layout.current_index.layout.normal.num_shards << | |
9068 | "; new num shards " << final_num_shards << " (suggested " << | |
9069 | suggested_num_shards << ")" << dendl; | |
9070 | ||
9071 | return add_bucket_to_reshard(dpp, bucket_info, final_num_shards); | |
9072 | } | |
9073 | ||
9074 | int RGWRados::add_bucket_to_reshard(const DoutPrefixProvider *dpp, const RGWBucketInfo& bucket_info, uint32_t new_num_shards) | |
9075 | { | |
9076 | RGWReshard reshard(this->store, dpp); | |
9077 | ||
9078 | uint32_t num_source_shards = (bucket_info.layout.current_index.layout.normal.num_shards > 0 ? bucket_info.layout.current_index.layout.normal.num_shards : 1); | |
9079 | ||
9080 | new_num_shards = std::min(new_num_shards, get_max_bucket_shards()); | |
9081 | if (new_num_shards <= num_source_shards) { | |
9082 | ldpp_dout(dpp, 20) << "not resharding bucket name=" << bucket_info.bucket.name << ", orig_num=" << num_source_shards << ", new_num_shards=" << new_num_shards << dendl; | |
9083 | return 0; | |
9084 | } | |
9085 | ||
9086 | cls_rgw_reshard_entry entry; | |
9087 | entry.time = real_clock::now(); | |
9088 | entry.tenant = bucket_info.owner.tenant; | |
9089 | entry.bucket_name = bucket_info.bucket.name; | |
9090 | entry.bucket_id = bucket_info.bucket.bucket_id; | |
9091 | entry.old_num_shards = num_source_shards; | |
9092 | entry.new_num_shards = new_num_shards; | |
9093 | ||
9094 | return reshard.add(dpp, entry); | |
9095 | } | |
9096 | ||
9097 | int RGWRados::check_quota(const rgw_user& bucket_owner, rgw_bucket& bucket, | |
9098 | RGWQuotaInfo& user_quota, RGWQuotaInfo& bucket_quota, | |
9099 | uint64_t obj_size, optional_yield y, | |
9100 | bool check_size_only) | |
9101 | { | |
9102 | // if we only check size, then num_objs will set to 0 | |
9103 | if(check_size_only) | |
9104 | return quota_handler->check_quota(bucket_owner, bucket, user_quota, bucket_quota, 0, obj_size, y); | |
9105 | ||
9106 | return quota_handler->check_quota(bucket_owner, bucket, user_quota, bucket_quota, 1, obj_size, y); | |
9107 | } | |
9108 | ||
9109 | int RGWRados::get_target_shard_id(const rgw::bucket_index_normal_layout& layout, const string& obj_key, | |
9110 | int *shard_id) | |
9111 | { | |
9112 | int r = 0; | |
9113 | switch (layout.hash_type) { | |
9114 | case rgw::BucketHashType::Mod: | |
9115 | if (!layout.num_shards) { | |
9116 | if (shard_id) { | |
9117 | *shard_id = -1; | |
9118 | } | |
9119 | } else { | |
9120 | uint32_t sid = svc.bi_rados->bucket_shard_index(obj_key, layout.num_shards); | |
9121 | if (shard_id) { | |
9122 | *shard_id = (int)sid; | |
9123 | } | |
9124 | } | |
9125 | break; | |
9126 | default: | |
9127 | r = -ENOTSUP; | |
9128 | } | |
9129 | return r; | |
9130 | } | |
9131 | ||
9132 | uint64_t RGWRados::instance_id() | |
9133 | { | |
9134 | return get_rados_handle()->get_instance_id(); | |
9135 | } | |
9136 | ||
9137 | uint64_t RGWRados::next_bucket_id() | |
9138 | { | |
9139 | std::lock_guard l{bucket_id_lock}; | |
9140 | return ++max_bucket_id; | |
9141 | } | |
9142 | ||
9143 | librados::Rados* RGWRados::get_rados_handle() | |
9144 | { | |
9145 | return &rados; | |
9146 | } | |
9147 | ||
9148 | int RGWRados::delete_raw_obj_aio(const DoutPrefixProvider *dpp, const rgw_raw_obj& obj, list<librados::AioCompletion *>& handles) | |
9149 | { | |
9150 | rgw_rados_ref ref; | |
9151 | int ret = get_raw_obj_ref(dpp, obj, &ref); | |
9152 | if (ret < 0) { | |
9153 | ldpp_dout(dpp, -1) << "ERROR: failed to get obj ref with ret=" << ret << dendl; | |
9154 | return ret; | |
9155 | } | |
9156 | ||
9157 | ObjectWriteOperation op; | |
9158 | list<string> prefixes; | |
9159 | cls_rgw_remove_obj(op, prefixes); | |
9160 | ||
9161 | AioCompletion *c = librados::Rados::aio_create_completion(nullptr, nullptr); | |
9162 | ret = ref.pool.ioctx().aio_operate(ref.obj.oid, c, &op); | |
9163 | if (ret < 0) { | |
9164 | ldpp_dout(dpp, -1) << "ERROR: AioOperate failed with ret=" << ret << dendl; | |
9165 | c->release(); | |
9166 | return ret; | |
9167 | } | |
9168 | ||
9169 | handles.push_back(c); | |
9170 | ||
9171 | return 0; | |
9172 | } | |
9173 | ||
9174 | int RGWRados::delete_obj_aio(const DoutPrefixProvider *dpp, const rgw_obj& obj, | |
9175 | RGWBucketInfo& bucket_info, RGWObjState *astate, | |
9176 | list<librados::AioCompletion *>& handles, bool keep_index_consistent, | |
9177 | optional_yield y) | |
9178 | { | |
9179 | rgw_rados_ref ref; | |
9180 | int ret = get_obj_head_ref(dpp, bucket_info, obj, &ref); | |
9181 | if (ret < 0) { | |
9182 | ldpp_dout(dpp, -1) << "ERROR: failed to get obj ref with ret=" << ret << dendl; | |
9183 | return ret; | |
9184 | } | |
9185 | ||
9186 | if (keep_index_consistent) { | |
9187 | RGWRados::Bucket bop(this, bucket_info); | |
9188 | RGWRados::Bucket::UpdateIndex index_op(&bop, obj); | |
9189 | ||
9190 | ret = index_op.prepare(dpp, CLS_RGW_OP_DEL, &astate->write_tag, y); | |
9191 | if (ret < 0) { | |
9192 | ldpp_dout(dpp, -1) << "ERROR: failed to prepare index op with ret=" << ret << dendl; | |
9193 | return ret; | |
9194 | } | |
9195 | } | |
9196 | ||
9197 | ObjectWriteOperation op; | |
9198 | list<string> prefixes; | |
9199 | cls_rgw_remove_obj(op, prefixes); | |
9200 | ||
9201 | AioCompletion *c = librados::Rados::aio_create_completion(nullptr, nullptr); | |
9202 | ret = ref.pool.ioctx().aio_operate(ref.obj.oid, c, &op); | |
9203 | if (ret < 0) { | |
9204 | ldpp_dout(dpp, -1) << "ERROR: AioOperate failed with ret=" << ret << dendl; | |
9205 | c->release(); | |
9206 | return ret; | |
9207 | } | |
9208 | ||
9209 | handles.push_back(c); | |
9210 | ||
9211 | if (keep_index_consistent) { | |
9212 | ret = delete_obj_index(obj, astate->mtime, dpp); | |
9213 | if (ret < 0) { | |
9214 | ldpp_dout(dpp, -1) << "ERROR: failed to delete obj index with ret=" << ret << dendl; | |
9215 | return ret; | |
9216 | } | |
9217 | } | |
9218 | return ret; | |
9219 | } |