ceph/src/rgw/rgw_quota.cc

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3
   4 /*
   5  * Ceph - scalable distributed file system
   6  *
   7  * Copyright (C) 2013 Inktank, Inc
   8  *
   9  * This is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License version 2.1, as published by the Free Software
  12  * Foundation.  See file COPYING.
  13  *
  14  */
  15
  16
  17 #include "include/utime.h"
  18 #include "common/lru_map.h"
  19 #include "common/RefCountedObj.h"
  20 #include "common/Thread.h"
  21 #include "common/Mutex.h"
  22 #include "common/RWLock.h"
  23
  24 #include "rgw_common.h"
  25 #include "rgw_rados.h"
  26 #include "rgw_quota.h"
  27 #include "rgw_bucket.h"
  28 #include "rgw_user.h"
  29
  30 #include "services/svc_sys_obj.h"
  31
  32 #include <atomic>
  33
  34 #define dout_context g_ceph_context
  35 #define dout_subsys ceph_subsys_rgw
  36
  37
  38 struct RGWQuotaCacheStats {
  39   RGWStorageStats stats;
  40   utime_t expiration;
  41   utime_t async_refresh_time;
  42 };
  43
  44 template<class T>
  45 class RGWQuotaCache {
  46 protected:
  47   RGWRados *store;
  48   lru_map<T, RGWQuotaCacheStats> stats_map;
  49   RefCountedWaitObject *async_refcount;
  50
  51   class StatsAsyncTestSet : public lru_map<T, RGWQuotaCacheStats>::UpdateContext {
  52     int objs_delta;
  53     uint64_t added_bytes;
  54     uint64_t removed_bytes;
  55   public:
  56     StatsAsyncTestSet() : objs_delta(0), added_bytes(0), removed_bytes(0) {}
  57     bool update(RGWQuotaCacheStats *entry) override {
  58       if (entry->async_refresh_time.sec() == 0)
  59         return false;
  60
  61       entry->async_refresh_time = utime_t(0, 0);
  62
  63       return true;
  64     }
  65   };
  66
  67   virtual int fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats) = 0;
  68
  69   virtual bool map_find(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) = 0;
  70
  71   virtual bool map_find_and_update(const rgw_user& user, const rgw_bucket& bucket, typename lru_map<T, RGWQuotaCacheStats>::UpdateContext *ctx) = 0;
  72   virtual void map_add(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) = 0;
  73
  74   virtual void data_modified(const rgw_user& user, rgw_bucket& bucket) {}
  75 public:
  76   RGWQuotaCache(RGWRados *_store, int size) : store(_store), stats_map(size) {
  77     async_refcount = new RefCountedWaitObject;
  78   }
  79   virtual ~RGWQuotaCache() {
  80     async_refcount->put_wait(); /* wait for all pending async requests to complete */
  81   }
  82
  83   int get_stats(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats, RGWQuotaInfo& quota);
  84   void adjust_stats(const rgw_user& user, rgw_bucket& bucket, int objs_delta, uint64_t added_bytes, uint64_t removed_bytes);
  85
  86   virtual bool can_use_cached_stats(RGWQuotaInfo& quota, RGWStorageStats& stats);
  87
  88   void set_stats(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs, RGWStorageStats& stats);
  89   int async_refresh(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs);
  90   void async_refresh_response(const rgw_user& user, rgw_bucket& bucket, RGWStorageStats& stats);
  91   void async_refresh_fail(const rgw_user& user, rgw_bucket& bucket);
  92
  93   class AsyncRefreshHandler {
  94   protected:
  95     RGWRados *store;
  96     RGWQuotaCache<T> *cache;
  97   public:
  98     AsyncRefreshHandler(RGWRados *_store, RGWQuotaCache<T> *_cache) : store(_store), cache(_cache) {}
  99     virtual ~AsyncRefreshHandler() {}
 100
 101     virtual int init_fetch() = 0;
 102     virtual void drop_reference() = 0;
 103   };
 104
 105   virtual AsyncRefreshHandler *allocate_refresh_handler(const rgw_user& user, const rgw_bucket& bucket) = 0;
 106 };
 107
 108 template<class T>
 109 bool RGWQuotaCache<T>::can_use_cached_stats(RGWQuotaInfo& quota, RGWStorageStats& cached_stats)
 110 {
 111   if (quota.max_size >= 0) {
 112     if (quota.max_size_soft_threshold < 0) {
 113       quota.max_size_soft_threshold = quota.max_size * store->ctx()->_conf->rgw_bucket_quota_soft_threshold;
 114     }
 115
 116     if (cached_stats.size_rounded  >= (uint64_t)quota.max_size_soft_threshold) {
 117       ldout(store->ctx(), 20) << "quota: can't use cached stats, exceeded soft threshold (size): "
 118         << cached_stats.size_rounded << " >= " << quota.max_size_soft_threshold << dendl;
 119       return false;
 120     }
 121   }
 122
 123   if (quota.max_objects >= 0) {
 124     if (quota.max_objs_soft_threshold < 0) {
 125       quota.max_objs_soft_threshold = quota.max_objects * store->ctx()->_conf->rgw_bucket_quota_soft_threshold;
 126     }
 127
 128     if (cached_stats.num_objects >= (uint64_t)quota.max_objs_soft_threshold) {
 129       ldout(store->ctx(), 20) << "quota: can't use cached stats, exceeded soft threshold (num objs): "
 130         << cached_stats.num_objects << " >= " << quota.max_objs_soft_threshold << dendl;
 131       return false;
 132     }
 133   }
 134
 135   return true;
 136 }
 137
 138 template<class T>
 139 int RGWQuotaCache<T>::async_refresh(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs)
 140 {
 141   /* protect against multiple updates */
 142   StatsAsyncTestSet test_update;
 143   if (!map_find_and_update(user, bucket, &test_update)) {
 144     /* most likely we just raced with another update */
 145     return 0;
 146   }
 147
 148   async_refcount->get();
 149
 150
 151   AsyncRefreshHandler *handler = allocate_refresh_handler(user, bucket);
 152
 153   int ret = handler->init_fetch();
 154   if (ret < 0) {
 155     async_refcount->put();
 156     handler->drop_reference();
 157     return ret;
 158   }
 159
 160   return 0;
 161 }
 162
 163 template<class T>
 164 void RGWQuotaCache<T>::async_refresh_fail(const rgw_user& user, rgw_bucket& bucket)
 165 {
 166   ldout(store->ctx(), 20) << "async stats refresh response for bucket=" << bucket << dendl;
 167
 168   async_refcount->put();
 169 }
 170
 171 template<class T>
 172 void RGWQuotaCache<T>::async_refresh_response(const rgw_user& user, rgw_bucket& bucket, RGWStorageStats& stats)
 173 {
 174   ldout(store->ctx(), 20) << "async stats refresh response for bucket=" << bucket << dendl;
 175
 176   RGWQuotaCacheStats qs;
 177
 178   map_find(user, bucket, qs);
 179
 180   set_stats(user, bucket, qs, stats);
 181
 182   async_refcount->put();
 183 }
 184
 185 template<class T>
 186 void RGWQuotaCache<T>::set_stats(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs, RGWStorageStats& stats)
 187 {
 188   qs.stats = stats;
 189   qs.expiration = ceph_clock_now();
 190   qs.async_refresh_time = qs.expiration;
 191   qs.expiration += store->ctx()->_conf->rgw_bucket_quota_ttl;
 192   qs.async_refresh_time += store->ctx()->_conf->rgw_bucket_quota_ttl / 2;
 193
 194   map_add(user, bucket, qs);
 195 }
 196
 197 template<class T>
 198 int RGWQuotaCache<T>::get_stats(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats, RGWQuotaInfo& quota) {
 199   RGWQuotaCacheStats qs;
 200   utime_t now = ceph_clock_now();
 201   if (map_find(user, bucket, qs)) {
 202     if (qs.async_refresh_time.sec() > 0 && now >= qs.async_refresh_time) {
 203       int r = async_refresh(user, bucket, qs);
 204       if (r < 0) {
 205         ldout(store->ctx(), 0) << "ERROR: quota async refresh returned ret=" << r << dendl;
 206
 207         /* continue processing, might be a transient error, async refresh is just optimization */
 208       }
 209     }
 210
 211     if (can_use_cached_stats(quota, qs.stats) && qs.expiration >
 212         ceph_clock_now()) {
 213       stats = qs.stats;
 214       return 0;
 215     }
 216   }
 217
 218   int ret = fetch_stats_from_storage(user, bucket, stats);
 219   if (ret < 0 && ret != -ENOENT)
 220     return ret;
 221
 222   set_stats(user, bucket, qs, stats);
 223
 224   return 0;
 225 }
 226
 227
 228 template<class T>
 229 class RGWQuotaStatsUpdate : public lru_map<T, RGWQuotaCacheStats>::UpdateContext {
 230   const int objs_delta;
 231   const uint64_t added_bytes;
 232   const uint64_t removed_bytes;
 233 public:
 234   RGWQuotaStatsUpdate(const int objs_delta,
 235                       const uint64_t added_bytes,
 236                       const uint64_t removed_bytes)
 237     : objs_delta(objs_delta),
 238       added_bytes(added_bytes),
 239       removed_bytes(removed_bytes) {
 240   }
 241
 242   bool update(RGWQuotaCacheStats * const entry) override {
 243     const uint64_t rounded_added = rgw_rounded_objsize(added_bytes);
 244     const uint64_t rounded_removed = rgw_rounded_objsize(removed_bytes);
 245
 246     if (((int64_t)(entry->stats.size + added_bytes - removed_bytes)) >= 0) {
 247       entry->stats.size += added_bytes - removed_bytes;
 248     } else {
 249       entry->stats.size = 0;
 250     }
 251
 252     if (((int64_t)(entry->stats.size_rounded + rounded_added - rounded_removed)) >= 0) {
 253       entry->stats.size_rounded += rounded_added - rounded_removed;
 254     } else {
 255       entry->stats.size_rounded = 0;
 256     }
 257
 258     if (((int64_t)(entry->stats.num_objects + objs_delta)) >= 0) {
 259       entry->stats.num_objects += objs_delta;
 260     } else {
 261       entry->stats.num_objects = 0;
 262     }
 263
 264     return true;
 265   }
 266 };
 267
 268
 269 template<class T>
 270 void RGWQuotaCache<T>::adjust_stats(const rgw_user& user, rgw_bucket& bucket, int objs_delta,
 271                                  uint64_t added_bytes, uint64_t removed_bytes)
 272 {
 273   RGWQuotaStatsUpdate<T> update(objs_delta, added_bytes, removed_bytes);
 274   map_find_and_update(user, bucket, &update);
 275
 276   data_modified(user, bucket);
 277 }
 278
 279 class BucketAsyncRefreshHandler : public RGWQuotaCache<rgw_bucket>::AsyncRefreshHandler,
 280                                   public RGWGetBucketStats_CB {
 281   rgw_user user;
 282 public:
 283   BucketAsyncRefreshHandler(RGWRados *_store, RGWQuotaCache<rgw_bucket> *_cache,
 284                             const rgw_user& _user, const rgw_bucket& _bucket) :
 285                                       RGWQuotaCache<rgw_bucket>::AsyncRefreshHandler(_store, _cache),
 286                                       RGWGetBucketStats_CB(_bucket), user(_user) {}
 287
 288   void drop_reference() override { put(); }
 289   void handle_response(int r) override;
 290   int init_fetch() override;
 291 };
 292
 293 int BucketAsyncRefreshHandler::init_fetch()
 294 {
 295   RGWBucketInfo bucket_info;
 296
 297   auto obj_ctx = store->svc.sysobj->init_obj_ctx();
 298
 299   int r = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL);
 300   if (r < 0) {
 301     ldout(store->ctx(), 0) << "could not get bucket info for bucket=" << bucket << " r=" << r << dendl;
 302     return r;
 303   }
 304
 305   ldout(store->ctx(), 20) << "initiating async quota refresh for bucket=" << bucket << dendl;
 306
 307   r = store->get_bucket_stats_async(bucket_info, RGW_NO_SHARD, this);
 308   if (r < 0) {
 309     ldout(store->ctx(), 0) << "could not get bucket info for bucket=" << bucket.name << dendl;
 310
 311     /* get_bucket_stats_async() dropped our reference already */
 312     return r;
 313   }
 314
 315   return 0;
 316 }
 317
 318 void BucketAsyncRefreshHandler::handle_response(const int r)
 319 {
 320   if (r < 0) {
 321     ldout(store->ctx(), 20) << "AsyncRefreshHandler::handle_response() r=" << r << dendl;
 322     cache->async_refresh_fail(user, bucket);
 323     return;
 324   }
 325
 326   RGWStorageStats bs;
 327
 328   for (const auto& pair : *stats) {
 329     const RGWStorageStats& s = pair.second;
 330
 331     bs.size += s.size;
 332     bs.size_rounded += s.size_rounded;
 333     bs.num_objects += s.num_objects;
 334   }
 335
 336   cache->async_refresh_response(user, bucket, bs);
 337 }
 338
 339 class RGWBucketStatsCache : public RGWQuotaCache<rgw_bucket> {
 340 protected:
 341   bool map_find(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) override {
 342     return stats_map.find(bucket, qs);
 343   }
 344
 345   bool map_find_and_update(const rgw_user& user, const rgw_bucket& bucket, lru_map<rgw_bucket, RGWQuotaCacheStats>::UpdateContext *ctx) override {
 346     return stats_map.find_and_update(bucket, NULL, ctx);
 347   }
 348
 349   void map_add(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) override {
 350     stats_map.add(bucket, qs);
 351   }
 352
 353   int fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats) override;
 354
 355 public:
 356   explicit RGWBucketStatsCache(RGWRados *_store) : RGWQuotaCache<rgw_bucket>(_store, _store->ctx()->_conf->rgw_bucket_quota_cache_size) {
 357   }
 358
 359   AsyncRefreshHandler *allocate_refresh_handler(const rgw_user& user, const rgw_bucket& bucket) override {
 360     return new BucketAsyncRefreshHandler(store, this, user, bucket);
 361   }
 362 };
 363
 364 int RGWBucketStatsCache::fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats)
 365 {
 366   RGWBucketInfo bucket_info;
 367
 368   RGWSysObjectCtx obj_ctx = store->svc.sysobj->init_obj_ctx();
 369
 370   int r = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL);
 371   if (r < 0) {
 372     ldout(store->ctx(), 0) << "could not get bucket info for bucket=" << bucket << " r=" << r << dendl;
 373     return r;
 374   }
 375
 376   string bucket_ver;
 377   string master_ver;
 378
 379   map<RGWObjCategory, RGWStorageStats> bucket_stats;
 380   r = store->get_bucket_stats(bucket_info, RGW_NO_SHARD, &bucket_ver,
 381                                   &master_ver, bucket_stats, nullptr);
 382   if (r < 0) {
 383     ldout(store->ctx(), 0) << "could not get bucket stats for bucket="
 384                            << bucket.name << dendl;
 385     return r;
 386   }
 387
 388   stats = RGWStorageStats();
 389
 390   for (const auto& pair : bucket_stats) {
 391     const RGWStorageStats& s = pair.second;
 392
 393     stats.size += s.size;
 394     stats.size_rounded += s.size_rounded;
 395     stats.num_objects += s.num_objects;
 396   }
 397
 398   return 0;
 399 }
 400
 401 class UserAsyncRefreshHandler : public RGWQuotaCache<rgw_user>::AsyncRefreshHandler,
 402                                 public RGWGetUserStats_CB {
 403   rgw_bucket bucket;
 404 public:
 405   UserAsyncRefreshHandler(RGWRados *_store, RGWQuotaCache<rgw_user> *_cache,
 406                           const rgw_user& _user, const rgw_bucket& _bucket) :
 407                           RGWQuotaCache<rgw_user>::AsyncRefreshHandler(_store, _cache),
 408                           RGWGetUserStats_CB(_user),
 409                           bucket(_bucket) {}
 410
 411   void drop_reference() override { put(); }
 412   int init_fetch() override;
 413   void handle_response(int r) override;
 414 };
 415
 416 int UserAsyncRefreshHandler::init_fetch()
 417 {
 418   ldout(store->ctx(), 20) << "initiating async quota refresh for user=" << user << dendl;
 419   int r = store->get_user_stats_async(user, this);
 420   if (r < 0) {
 421     ldout(store->ctx(), 0) << "could not get bucket info for user=" << user << dendl;
 422
 423     /* get_bucket_stats_async() dropped our reference already */
 424     return r;
 425   }
 426
 427   return 0;
 428 }
 429
 430 void UserAsyncRefreshHandler::handle_response(int r)
 431 {
 432   if (r < 0) {
 433     ldout(store->ctx(), 20) << "AsyncRefreshHandler::handle_response() r=" << r << dendl;
 434     cache->async_refresh_fail(user, bucket);
 435     return;
 436   }
 437
 438   cache->async_refresh_response(user, bucket, stats);
 439 }
 440
 441 class RGWUserStatsCache : public RGWQuotaCache<rgw_user> {
 442   std::atomic<bool> down_flag = { false };
 443   RWLock rwlock;
 444   map<rgw_bucket, rgw_user> modified_buckets;
 445
 446   /* thread, sync recent modified buckets info */
 447   class BucketsSyncThread : public Thread {
 448     CephContext *cct;
 449     RGWUserStatsCache *stats;
 450
 451     Mutex lock;
 452     Cond cond;
 453   public:
 454
 455     BucketsSyncThread(CephContext *_cct, RGWUserStatsCache *_s) : cct(_cct), stats(_s), lock("RGWUserStatsCache::BucketsSyncThread") {}
 456
 457     void *entry() override {
 458       ldout(cct, 20) << "BucketsSyncThread: start" << dendl;
 459       do {
 460         map<rgw_bucket, rgw_user> buckets;
 461
 462         stats->swap_modified_buckets(buckets);
 463
 464         for (map<rgw_bucket, rgw_user>::iterator iter = buckets.begin(); iter != buckets.end(); ++iter) {
 465           rgw_bucket bucket = iter->first;
 466           rgw_user& user = iter->second;
 467           ldout(cct, 20) << "BucketsSyncThread: sync user=" << user << " bucket=" << bucket << dendl;
 468           int r = stats->sync_bucket(user, bucket);
 469           if (r < 0) {
 470             ldout(cct, 0) << "WARNING: sync_bucket() returned r=" << r << dendl;
 471           }
 472         }
 473
 474         if (stats->going_down())
 475           break;
 476
 477         lock.Lock();
 478         cond.WaitInterval(lock, utime_t(cct->_conf->rgw_user_quota_bucket_sync_interval, 0));
 479         lock.Unlock();
 480       } while (!stats->going_down());
 481       ldout(cct, 20) << "BucketsSyncThread: done" << dendl;
 482
 483       return NULL;
 484     }
 485
 486     void stop() {
 487       Mutex::Locker l(lock);
 488       cond.Signal();
 489     }
 490   };
 491
 492   /*
 493    * thread, full sync all users stats periodically
 494    *
 495    * only sync non idle users or ones that never got synced before, this is needed so that
 496    * users that didn't have quota turned on before (or existed before the user objclass
 497    * tracked stats) need to get their backend stats up to date.
 498    */
 499   class UserSyncThread : public Thread {
 500     CephContext *cct;
 501     RGWUserStatsCache *stats;
 502
 503     Mutex lock;
 504     Cond cond;
 505   public:
 506
 507     UserSyncThread(CephContext *_cct, RGWUserStatsCache *_s) : cct(_cct), stats(_s), lock("RGWUserStatsCache::UserSyncThread") {}
 508
 509     void *entry() override {
 510       ldout(cct, 20) << "UserSyncThread: start" << dendl;
 511       do {
 512         int ret = stats->sync_all_users();
 513         if (ret < 0) {
 514           ldout(cct, 5) << "ERROR: sync_all_users() returned ret=" << ret << dendl;
 515         }
 516
 517         if (stats->going_down())
 518           break;
 519
 520         lock.Lock();
 521         cond.WaitInterval(lock, utime_t(cct->_conf->rgw_user_quota_sync_interval, 0));
 522         lock.Unlock();
 523       } while (!stats->going_down());
 524       ldout(cct, 20) << "UserSyncThread: done" << dendl;
 525
 526       return NULL;
 527     }
 528
 529     void stop() {
 530       Mutex::Locker l(lock);
 531       cond.Signal();
 532     }
 533   };
 534
 535   BucketsSyncThread *buckets_sync_thread;
 536   UserSyncThread *user_sync_thread;
 537 protected:
 538   bool map_find(const rgw_user& user,const rgw_bucket& bucket, RGWQuotaCacheStats& qs) override {
 539     return stats_map.find(user, qs);
 540   }
 541
 542   bool map_find_and_update(const rgw_user& user, const rgw_bucket& bucket, lru_map<rgw_user, RGWQuotaCacheStats>::UpdateContext *ctx) override {
 543     return stats_map.find_and_update(user, NULL, ctx);
 544   }
 545
 546   void map_add(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) override {
 547     stats_map.add(user, qs);
 548   }
 549
 550   int fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats) override;
 551   int sync_bucket(const rgw_user& rgw_user, rgw_bucket& bucket);
 552   int sync_user(const rgw_user& user);
 553   int sync_all_users();
 554
 555   void data_modified(const rgw_user& user, rgw_bucket& bucket) override;
 556
 557   void swap_modified_buckets(map<rgw_bucket, rgw_user>& out) {
 558     rwlock.get_write();
 559     modified_buckets.swap(out);
 560     rwlock.unlock();
 561   }
 562
 563   template<class T> /* easier doing it as a template, Thread doesn't have ->stop() */
 564   void stop_thread(T **pthr) {
 565     T *thread = *pthr;
 566     if (!thread)
 567       return;
 568
 569     thread->stop();
 570     thread->join();
 571     delete thread;
 572     *pthr = NULL;
 573   }
 574
 575 public:
 576   RGWUserStatsCache(RGWRados *_store, bool quota_threads) : RGWQuotaCache<rgw_user>(_store, _store->ctx()->_conf->rgw_bucket_quota_cache_size),
 577                                         rwlock("RGWUserStatsCache::rwlock") {
 578     if (quota_threads) {
 579       buckets_sync_thread = new BucketsSyncThread(store->ctx(), this);
 580       buckets_sync_thread->create("rgw_buck_st_syn");
 581       user_sync_thread = new UserSyncThread(store->ctx(), this);
 582       user_sync_thread->create("rgw_user_st_syn");
 583     } else {
 584       buckets_sync_thread = NULL;
 585       user_sync_thread = NULL;
 586     }
 587   }
 588   ~RGWUserStatsCache() override {
 589     stop();
 590   }
 591
 592   AsyncRefreshHandler *allocate_refresh_handler(const rgw_user& user, const rgw_bucket& bucket) override {
 593     return new UserAsyncRefreshHandler(store, this, user, bucket);
 594   }
 595
 596   bool can_use_cached_stats(RGWQuotaInfo& quota, RGWStorageStats& stats) override {
 597     /* in the user case, the cached stats may contain a better estimation of the totals, as
 598      * the backend is only periodically getting updated.
 599      */
 600     return true;
 601   }
 602
 603   bool going_down() {
 604     return down_flag;
 605   }
 606
 607   void stop() {
 608     down_flag = true;
 609     rwlock.get_write();
 610     stop_thread(&buckets_sync_thread);
 611     rwlock.unlock();
 612     stop_thread(&user_sync_thread);
 613   }
 614 };
 615
 616 int RGWUserStatsCache::fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats)
 617 {
 618   int r = store->get_user_stats(user, stats);
 619   if (r < 0) {
 620     ldout(store->ctx(), 0) << "could not get user stats for user=" << user << dendl;
 621     return r;
 622   }
 623
 624   return 0;
 625 }
 626
 627 int RGWUserStatsCache::sync_bucket(const rgw_user& user, rgw_bucket& bucket)
 628 {
 629   RGWBucketInfo bucket_info;
 630
 631   RGWSysObjectCtx obj_ctx = store->svc.sysobj->init_obj_ctx();
 632
 633   int r = store->get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL);
 634   if (r < 0) {
 635     ldout(store->ctx(), 0) << "could not get bucket info for bucket=" << bucket << " r=" << r << dendl;
 636     return r;
 637   }
 638
 639   r = rgw_bucket_sync_user_stats(store, user, bucket_info);
 640   if (r < 0) {
 641     ldout(store->ctx(), 0) << "ERROR: rgw_bucket_sync_user_stats() for user=" << user << ", bucket=" << bucket << " returned " << r << dendl;
 642     return r;
 643   }
 644
 645   return 0;
 646 }
 647
 648 int RGWUserStatsCache::sync_user(const rgw_user& user)
 649 {
 650   cls_user_header header;
 651   string user_str = user.to_str();
 652   int ret = store->cls_user_get_header(user_str, &header);
 653   if (ret < 0) {
 654     ldout(store->ctx(), 5) << "ERROR: can't read user header: ret=" << ret << dendl;
 655     return ret;
 656   }
 657
 658   if (!store->ctx()->_conf->rgw_user_quota_sync_idle_users &&
 659       header.last_stats_update < header.last_stats_sync) {
 660     ldout(store->ctx(), 20) << "user is idle, not doing a full sync (user=" << user << ")" << dendl;
 661     return 0;
 662   }
 663
 664   real_time when_need_full_sync = header.last_stats_sync;
 665   when_need_full_sync += make_timespan(store->ctx()->_conf->rgw_user_quota_sync_wait_time);
 666
 667   // check if enough time passed since last full sync
 668   /* FIXME: missing check? */
 669
 670   ret = rgw_user_sync_all_stats(store, user);
 671   if (ret < 0) {
 672     ldout(store->ctx(), 0) << "ERROR: failed user stats sync, ret=" << ret << dendl;
 673     return ret;
 674   }
 675
 676   return 0;
 677 }
 678
 679 int RGWUserStatsCache::sync_all_users()
 680 {
 681   string key = "user";
 682   void *handle;
 683
 684   int ret = store->meta_mgr->list_keys_init(key, &handle);
 685   if (ret < 0) {
 686     ldout(store->ctx(), 10) << "ERROR: can't get key: ret=" << ret << dendl;
 687     return ret;
 688   }
 689
 690   bool truncated;
 691   int max = 1000;
 692
 693   do {
 694     list<string> keys;
 695     ret = store->meta_mgr->list_keys_next(handle, max, keys, &truncated);
 696     if (ret < 0) {
 697       ldout(store->ctx(), 0) << "ERROR: lists_keys_next(): ret=" << ret << dendl;
 698       goto done;
 699     }
 700     for (list<string>::iterator iter = keys.begin();
 701          iter != keys.end() && !going_down();
 702          ++iter) {
 703       rgw_user user(*iter);
 704       ldout(store->ctx(), 20) << "RGWUserStatsCache: sync user=" << user << dendl;
 705       int ret = sync_user(user);
 706       if (ret < 0) {
 707         ldout(store->ctx(), 5) << "ERROR: sync_user() failed, user=" << user << " ret=" << ret << dendl;
 708
 709         /* continuing to next user */
 710         continue;
 711       }
 712     }
 713   } while (truncated);
 714
 715   ret = 0;
 716 done:
 717   store->meta_mgr->list_keys_complete(handle);
 718   return ret;
 719 }
 720
 721 void RGWUserStatsCache::data_modified(const rgw_user& user, rgw_bucket& bucket)
 722 {
 723   /* racy, but it's ok */
 724   rwlock.get_read();
 725   bool need_update = modified_buckets.find(bucket) == modified_buckets.end();
 726   rwlock.unlock();
 727
 728   if (need_update) {
 729     rwlock.get_write();
 730     modified_buckets[bucket] = user;
 731     rwlock.unlock();
 732   }
 733 }
 734
 735
 736 class RGWQuotaInfoApplier {
 737   /* NOTE: no non-static field allowed as instances are supposed to live in
 738    * the static memory only. */
 739 protected:
 740   RGWQuotaInfoApplier() = default;
 741
 742 public:
 743   virtual ~RGWQuotaInfoApplier() {}
 744
 745   virtual bool is_size_exceeded(const char * const entity,
 746                                 const RGWQuotaInfo& qinfo,
 747                                 const RGWStorageStats& stats,
 748                                 const uint64_t size) const = 0;
 749
 750   virtual bool is_num_objs_exceeded(const char * const entity,
 751                                     const RGWQuotaInfo& qinfo,
 752                                     const RGWStorageStats& stats,
 753                                     const uint64_t num_objs) const = 0;
 754
 755   static const RGWQuotaInfoApplier& get_instance(const RGWQuotaInfo& qinfo);
 756 };
 757
 758 class RGWQuotaInfoDefApplier : public RGWQuotaInfoApplier {
 759 public:
 760   bool is_size_exceeded(const char * const entity,
 761                                 const RGWQuotaInfo& qinfo,
 762                                 const RGWStorageStats& stats,
 763                                 const uint64_t size) const override;
 764
 765   bool is_num_objs_exceeded(const char * const entity,
 766                                     const RGWQuotaInfo& qinfo,
 767                                     const RGWStorageStats& stats,
 768                                     const uint64_t num_objs) const override;
 769 };
 770
 771 class RGWQuotaInfoRawApplier : public RGWQuotaInfoApplier {
 772 public:
 773   bool is_size_exceeded(const char * const entity,
 774                                 const RGWQuotaInfo& qinfo,
 775                                 const RGWStorageStats& stats,
 776                                 const uint64_t size) const override;
 777
 778   bool is_num_objs_exceeded(const char * const entity,
 779                                     const RGWQuotaInfo& qinfo,
 780                                     const RGWStorageStats& stats,
 781                                     const uint64_t num_objs) const override;
 782 };
 783
 784
 785 bool RGWQuotaInfoDefApplier::is_size_exceeded(const char * const entity,
 786                                               const RGWQuotaInfo& qinfo,
 787                                               const RGWStorageStats& stats,
 788                                               const uint64_t size) const
 789 {
 790   if (qinfo.max_size < 0) {
 791     /* The limit is not enabled. */
 792     return false;
 793   }
 794
 795   const uint64_t cur_size = stats.size_rounded;
 796   const uint64_t new_size = rgw_rounded_objsize(size);
 797
 798   if (cur_size + new_size > static_cast<uint64_t>(qinfo.max_size)) {
 799     dout(10) << "quota exceeded: stats.size_rounded=" << stats.size_rounded
 800              << " size=" << new_size << " "
 801              << entity << "_quota.max_size=" << qinfo.max_size << dendl;
 802     return true;
 803   }
 804
 805   return false;
 806 }
 807
 808 bool RGWQuotaInfoDefApplier::is_num_objs_exceeded(const char * const entity,
 809                                                   const RGWQuotaInfo& qinfo,
 810                                                   const RGWStorageStats& stats,
 811                                                   const uint64_t num_objs) const
 812 {
 813   if (qinfo.max_objects < 0) {
 814     /* The limit is not enabled. */
 815     return false;
 816   }
 817
 818   if (stats.num_objects + num_objs > static_cast<uint64_t>(qinfo.max_objects)) {
 819     dout(10) << "quota exceeded: stats.num_objects=" << stats.num_objects
 820              << " " << entity << "_quota.max_objects=" << qinfo.max_objects
 821              << dendl;
 822     return true;
 823   }
 824
 825   return false;
 826 }
 827
 828 bool RGWQuotaInfoRawApplier::is_size_exceeded(const char * const entity,
 829                                               const RGWQuotaInfo& qinfo,
 830                                               const RGWStorageStats& stats,
 831                                               const uint64_t size) const
 832 {
 833   if (qinfo.max_size < 0) {
 834     /* The limit is not enabled. */
 835     return false;
 836   }
 837
 838   const uint64_t cur_size = stats.size;
 839
 840   if (cur_size + size > static_cast<uint64_t>(qinfo.max_size)) {
 841     dout(10) << "quota exceeded: stats.size=" << stats.size
 842              << " size=" << size << " "
 843              << entity << "_quota.max_size=" << qinfo.max_size << dendl;
 844     return true;
 845   }
 846
 847   return false;
 848 }
 849
 850 bool RGWQuotaInfoRawApplier::is_num_objs_exceeded(const char * const entity,
 851                                                   const RGWQuotaInfo& qinfo,
 852                                                   const RGWStorageStats& stats,
 853                                                   const uint64_t num_objs) const
 854 {
 855   if (qinfo.max_objects < 0) {
 856     /* The limit is not enabled. */
 857     return false;
 858   }
 859
 860   if (stats.num_objects + num_objs > static_cast<uint64_t>(qinfo.max_objects)) {
 861     dout(10) << "quota exceeded: stats.num_objects=" << stats.num_objects
 862              << " " << entity << "_quota.max_objects=" << qinfo.max_objects
 863              << dendl;
 864     return true;
 865   }
 866
 867   return false;
 868 }
 869
 870 const RGWQuotaInfoApplier& RGWQuotaInfoApplier::get_instance(
 871   const RGWQuotaInfo& qinfo)
 872 {
 873   static RGWQuotaInfoDefApplier default_qapplier;
 874   static RGWQuotaInfoRawApplier raw_qapplier;
 875
 876   if (qinfo.check_on_raw) {
 877     return raw_qapplier;
 878   } else {
 879     return default_qapplier;
 880   }
 881 }
 882
 883
 884 class RGWQuotaHandlerImpl : public RGWQuotaHandler {
 885   RGWRados *store;
 886   RGWBucketStatsCache bucket_stats_cache;
 887   RGWUserStatsCache user_stats_cache;
 888
 889   int check_quota(const char * const entity,
 890                   const RGWQuotaInfo& quota,
 891                   const RGWStorageStats& stats,
 892                   const uint64_t num_objs,
 893                   const uint64_t size) {
 894     if (!quota.enabled) {
 895       return 0;
 896     }
 897
 898     const auto& quota_applier = RGWQuotaInfoApplier::get_instance(quota);
 899
 900     ldout(store->ctx(), 20) << entity
 901                             << " quota: max_objects=" << quota.max_objects
 902                             << " max_size=" << quota.max_size << dendl;
 903
 904
 905     if (quota_applier.is_num_objs_exceeded(entity, quota, stats, num_objs)) {
 906       return -ERR_QUOTA_EXCEEDED;
 907     }
 908
 909     if (quota_applier.is_size_exceeded(entity, quota, stats, size)) {
 910       return -ERR_QUOTA_EXCEEDED;
 911     }
 912
 913     ldout(store->ctx(), 20) << entity << " quota OK:"
 914                             << " stats.num_objects=" << stats.num_objects
 915                             << " stats.size=" << stats.size << dendl;
 916     return 0;
 917   }
 918 public:
 919   RGWQuotaHandlerImpl(RGWRados *_store, bool quota_threads) : store(_store),
 920                                     bucket_stats_cache(_store),
 921                                     user_stats_cache(_store, quota_threads) {}
 922
 923   int check_quota(const rgw_user& user,
 924                           rgw_bucket& bucket,
 925                           RGWQuotaInfo& user_quota,
 926                           RGWQuotaInfo& bucket_quota,
 927                           uint64_t num_objs,
 928                           uint64_t size) override {
 929
 930     if (!bucket_quota.enabled && !user_quota.enabled) {
 931       return 0;
 932     }
 933
 934     /*
 935      * we need to fetch bucket stats if the user quota is enabled, because
 936      * the whole system relies on us periodically updating the user's bucket
 937      * stats in the user's header, this happens in get_stats() if we actually
 938      * fetch that info and not rely on cached data
 939      */
 940
 941     if (bucket_quota.enabled) {
 942       RGWStorageStats bucket_stats;
 943       int ret = bucket_stats_cache.get_stats(user, bucket, bucket_stats,
 944                                            bucket_quota);
 945       if (ret < 0) {
 946         return ret;
 947       }
 948       ret = check_quota("bucket", bucket_quota, bucket_stats, num_objs, size);
 949       if (ret < 0) {
 950         return ret;
 951       }
 952     }
 953
 954     if (user_quota.enabled) {
 955       RGWStorageStats user_stats;
 956       int ret = user_stats_cache.get_stats(user, bucket, user_stats, user_quota);
 957       if (ret < 0) {
 958         return ret;
 959       }
 960       ret = check_quota("user", user_quota, user_stats, num_objs, size);
 961       if (ret < 0) {
 962         return ret;
 963       }
 964     }
 965     return 0;
 966   }
 967
 968   void update_stats(const rgw_user& user, rgw_bucket& bucket, int obj_delta, uint64_t added_bytes, uint64_t removed_bytes) override {
 969     bucket_stats_cache.adjust_stats(user, bucket, obj_delta, added_bytes, removed_bytes);
 970     user_stats_cache.adjust_stats(user, bucket, obj_delta, added_bytes, removed_bytes);
 971   }
 972
 973   int check_bucket_shards(uint64_t max_objs_per_shard, uint64_t num_shards,
 974                           const rgw_user& user, const rgw_bucket& bucket, RGWQuotaInfo& bucket_quota,
 975                           uint64_t num_objs, bool& need_resharding, uint32_t *suggested_num_shards) override
 976   {
 977     RGWStorageStats bucket_stats;
 978     int ret = bucket_stats_cache.get_stats(user, bucket, bucket_stats,
 979                                            bucket_quota);
 980     if (ret < 0) {
 981       return ret;
 982     }
 983
 984     if (bucket_stats.num_objects  + num_objs > num_shards * max_objs_per_shard) {
 985       ldout(store->ctx(), 0) << __func__ << ": resharding needed: stats.num_objects=" << bucket_stats.num_objects
 986              << " shard max_objects=" <<  max_objs_per_shard * num_shards << dendl;
 987       need_resharding = true;
 988       if (suggested_num_shards) {
 989         *suggested_num_shards = (bucket_stats.num_objects  + num_objs) * 2 / max_objs_per_shard;
 990       }
 991     } else {
 992       need_resharding = false;
 993     }
 994
 995     return 0;
 996   }
 997
 998 };
 999
1000
1001 RGWQuotaHandler *RGWQuotaHandler::generate_handler(RGWRados *store, bool quota_threads)
1002 {
1003   return new RGWQuotaHandlerImpl(store, quota_threads);
1004 }
1005
1006 void RGWQuotaHandler::free_handler(RGWQuotaHandler *handler)
1007 {
1008   delete handler;
1009 }
1010
1011
1012 void rgw_apply_default_bucket_quota(RGWQuotaInfo& quota, const ConfigProxy& conf)
1013 {
1014   if (conf->rgw_bucket_default_quota_max_objects >= 0) {
1015     quota.max_objects = conf->rgw_bucket_default_quota_max_objects;
1016     quota.enabled = true;
1017   }
1018   if (conf->rgw_bucket_default_quota_max_size >= 0) {
1019     quota.max_size = conf->rgw_bucket_default_quota_max_size;
1020     quota.enabled = true;
1021   }
1022 }
1023
1024 void rgw_apply_default_user_quota(RGWQuotaInfo& quota, const ConfigProxy& conf)
1025 {
1026   if (conf->rgw_user_default_quota_max_objects >= 0) {
1027     quota.max_objects = conf->rgw_user_default_quota_max_objects;
1028     quota.enabled = true;
1029   }
1030   if (conf->rgw_user_default_quota_max_size >= 0) {
1031     quota.max_size = conf->rgw_user_default_quota_max_size;
1032     quota.enabled = true;
1033   }
1034 }