path: root/src/rgw/rgw_quota.cc

// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
// vim: ts=8 sw=2 smarttab ft=cpp

/*
 * Ceph - scalable distributed file system
 *
 * Copyright (C) 2013 Inktank, Inc
 *
 * This is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License version 2.1, as published by the Free Software
 * Foundation.  See file COPYING.
 *
 */


#include "include/utime.h"
#include "common/lru_map.h"
#include "common/RefCountedObj.h"
#include "common/Thread.h"
#include "common/ceph_mutex.h"

#include "rgw_common.h"
#include "rgw_sal.h"
#include "rgw_sal_rados.h"
#include "rgw_quota.h"
#include "rgw_bucket.h"
#include "rgw_user.h"

#include "services/svc_sys_obj.h"
#include "services/svc_meta.h"

#include <atomic>

#define dout_context g_ceph_context
#define dout_subsys ceph_subsys_rgw


struct RGWQuotaCacheStats {
  RGWStorageStats stats;
  utime_t expiration;
  utime_t async_refresh_time;
};

template<class T>
class RGWQuotaCache {
protected:
  rgw::sal::RGWRadosStore *store;
  lru_map<T, RGWQuotaCacheStats> stats_map;
  RefCountedWaitObject *async_refcount;

  class StatsAsyncTestSet : public lru_map<T, RGWQuotaCacheStats>::UpdateContext {
    int objs_delta;
    uint64_t added_bytes;
    uint64_t removed_bytes;
  public:
    StatsAsyncTestSet() : objs_delta(0), added_bytes(0), removed_bytes(0) {}
    bool update(RGWQuotaCacheStats *entry) override {
      if (entry->async_refresh_time.sec() == 0)
        return false;

      entry->async_refresh_time = utime_t(0, 0);

      return true;
    }
  };

  virtual int fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats, optional_yield y, const DoutPrefixProvider *dpp) = 0;

  virtual bool map_find(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) = 0;

  virtual bool map_find_and_update(const rgw_user& user, const rgw_bucket& bucket, typename lru_map<T, RGWQuotaCacheStats>::UpdateContext *ctx) = 0;
  virtual void map_add(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) = 0;

  virtual void data_modified(const rgw_user& user, rgw_bucket& bucket) {}
public:
  RGWQuotaCache(rgw::sal::RGWRadosStore *_store, int size) : store(_store), stats_map(size) {
    async_refcount = new RefCountedWaitObject;
  }
  virtual ~RGWQuotaCache() {
    async_refcount->put_wait(); /* wait for all pending async requests to complete */
  }

  int get_stats(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats, optional_yield y,
                const DoutPrefixProvider* dpp);
  void adjust_stats(const rgw_user& user, rgw_bucket& bucket, int objs_delta, uint64_t added_bytes, uint64_t removed_bytes);

  void set_stats(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs, RGWStorageStats& stats);
  int async_refresh(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs);
  void async_refresh_response(const rgw_user& user, rgw_bucket& bucket, RGWStorageStats& stats);
  void async_refresh_fail(const rgw_user& user, rgw_bucket& bucket);

  class AsyncRefreshHandler {
  protected:
    rgw::sal::RGWRadosStore *store;
    RGWQuotaCache<T> *cache;
  public:
    AsyncRefreshHandler(rgw::sal::RGWRadosStore *_store, RGWQuotaCache<T> *_cache) : store(_store), cache(_cache) {}
    virtual ~AsyncRefreshHandler() {}

    virtual int init_fetch() = 0;
    virtual void drop_reference() = 0;
  };

  virtual AsyncRefreshHandler *allocate_refresh_handler(const rgw_user& user, const rgw_bucket& bucket) = 0;
};

template<class T>
int RGWQuotaCache<T>::async_refresh(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs)
{
  /* protect against multiple updates */
  StatsAsyncTestSet test_update;
  if (!map_find_and_update(user, bucket, &test_update)) {
    /* most likely we just raced with another update */
    return 0;
  }

  async_refcount->get();


  AsyncRefreshHandler *handler = allocate_refresh_handler(user, bucket);

  int ret = handler->init_fetch();
  if (ret < 0) {
    async_refcount->put();
    handler->drop_reference();
    return ret;
  }

  return 0;
}

template<class T>
void RGWQuotaCache<T>::async_refresh_fail(const rgw_user& user, rgw_bucket& bucket)
{
  ldout(store->ctx(), 20) << "async stats refresh response for bucket=" << bucket << dendl;

  async_refcount->put();
}

template<class T>
void RGWQuotaCache<T>::async_refresh_response(const rgw_user& user, rgw_bucket& bucket, RGWStorageStats& stats)
{
  ldout(store->ctx(), 20) << "async stats refresh response for bucket=" << bucket << dendl;

  RGWQuotaCacheStats qs;

  map_find(user, bucket, qs);

  set_stats(user, bucket, qs, stats);

  async_refcount->put();
}

template<class T>
void RGWQuotaCache<T>::set_stats(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs, RGWStorageStats& stats)
{
  qs.stats = stats;
  qs.expiration = ceph_clock_now();
  qs.async_refresh_time = qs.expiration;
  qs.expiration += store->ctx()->_conf->rgw_bucket_quota_ttl;
  qs.async_refresh_time += store->ctx()->_conf->rgw_bucket_quota_ttl / 2;

  map_add(user, bucket, qs);
}

template<class T>
int RGWQuotaCache<T>::get_stats(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats, optional_yield y, const DoutPrefixProvider* dpp) {
  RGWQuotaCacheStats qs;
  utime_t now = ceph_clock_now();
  if (map_find(user, bucket, qs)) {
    if (qs.async_refresh_time.sec() > 0 && now >= qs.async_refresh_time) {
      int r = async_refresh(user, bucket, qs);
      if (r < 0) {
        ldout(store->ctx(), 0) << "ERROR: quota async refresh returned ret=" << r << dendl;

        /* continue processing, might be a transient error, async refresh is just optimization */
      }
    }

    if (qs.expiration > ceph_clock_now()) {
      stats = qs.stats;
      return 0;
    }
  }

  int ret = fetch_stats_from_storage(user, bucket, stats, y, dpp);
  if (ret < 0 && ret != -ENOENT)
    return ret;

  set_stats(user, bucket, qs, stats);

  return 0;
}


template<class T>
class RGWQuotaStatsUpdate : public lru_map<T, RGWQuotaCacheStats>::UpdateContext {
  const int objs_delta;
  const uint64_t added_bytes;
  const uint64_t removed_bytes;
public:
  RGWQuotaStatsUpdate(const int objs_delta,
                      const uint64_t added_bytes,
                      const uint64_t removed_bytes)
    : objs_delta(objs_delta),
      added_bytes(added_bytes),
      removed_bytes(removed_bytes) {
  }

  bool update(RGWQuotaCacheStats * const entry) override {
    const uint64_t rounded_added = rgw_rounded_objsize(added_bytes);
    const uint64_t rounded_removed = rgw_rounded_objsize(removed_bytes);

    if (((int64_t)(entry->stats.size + added_bytes - removed_bytes)) >= 0) {
      entry->stats.size += added_bytes - removed_bytes;
    } else {
      entry->stats.size = 0;
    }

    if (((int64_t)(entry->stats.size_rounded + rounded_added - rounded_removed)) >= 0) {
      entry->stats.size_rounded += rounded_added - rounded_removed;
    } else {
      entry->stats.size_rounded = 0;
    }

    if (((int64_t)(entry->stats.num_objects + objs_delta)) >= 0) {
      entry->stats.num_objects += objs_delta;
    } else {
      entry->stats.num_objects = 0;
    }

    return true;
  }
};


template<class T>
void RGWQuotaCache<T>::adjust_stats(const rgw_user& user, rgw_bucket& bucket, int objs_delta,
                                 uint64_t added_bytes, uint64_t removed_bytes)
{
  RGWQuotaStatsUpdate<T> update(objs_delta, added_bytes, removed_bytes);
  map_find_and_update(user, bucket, &update);

  data_modified(user, bucket);
}

class BucketAsyncRefreshHandler : public RGWQuotaCache<rgw_bucket>::AsyncRefreshHandler,
                                  public RGWGetBucketStats_CB {
  rgw_user user;
public:
  BucketAsyncRefreshHandler(rgw::sal::RGWRadosStore *_store, RGWQuotaCache<rgw_bucket> *_cache,
                            const rgw_user& _user, const rgw_bucket& _bucket) :
                                      RGWQuotaCache<rgw_bucket>::AsyncRefreshHandler(_store, _cache),
                                      RGWGetBucketStats_CB(_bucket), user(_user) {}

  void drop_reference() override { put(); }
  void handle_response(int r) override;
  int init_fetch() override;
};

int BucketAsyncRefreshHandler::init_fetch()
{
  RGWBucketInfo bucket_info;

  auto obj_ctx = store->svc()->sysobj->init_obj_ctx();

  const DoutPrefix dp(store->ctx(), dout_subsys, "rgw bucket async refresh handler: ");
  int r = store->getRados()->get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL, null_yield, &dp);
  if (r < 0) {
    ldpp_dout(&dp, 0) << "could not get bucket info for bucket=" << bucket << " r=" << r << dendl;
    return r;
  }

  ldpp_dout(&dp, 20) << "initiating async quota refresh for bucket=" << bucket << dendl;

  r = store->getRados()->get_bucket_stats_async(&dp, bucket_info, RGW_NO_SHARD, this);
  if (r < 0) {
    ldpp_dout(&dp, 0) << "could not get bucket info for bucket=" << bucket.name << dendl;

    /* get_bucket_stats_async() dropped our reference already */
    return r;
  }

  return 0;
}

void BucketAsyncRefreshHandler::handle_response(const int r)
{
  if (r < 0) {
    ldout(store->ctx(), 20) << "AsyncRefreshHandler::handle_response() r=" << r << dendl;
    cache->async_refresh_fail(user, bucket);
    return;
  }

  RGWStorageStats bs;

  for (const auto& pair : *stats) {
    const RGWStorageStats& s = pair.second;

    bs.size += s.size;
    bs.size_rounded += s.size_rounded;
    bs.num_objects += s.num_objects;
  }

  cache->async_refresh_response(user, bucket, bs);
}

class RGWBucketStatsCache : public RGWQuotaCache<rgw_bucket> {
protected:
  bool map_find(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) override {
    return stats_map.find(bucket, qs);
  }

  bool map_find_and_update(const rgw_user& user, const rgw_bucket& bucket, lru_map<rgw_bucket, RGWQuotaCacheStats>::UpdateContext *ctx) override {
    return stats_map.find_and_update(bucket, NULL, ctx);
  }

  void map_add(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) override {
    stats_map.add(bucket, qs);
  }

  int fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats, optional_yield y, const DoutPrefixProvider *dpp) override;

public:
  explicit RGWBucketStatsCache(rgw::sal::RGWRadosStore *_store) : RGWQuotaCache<rgw_bucket>(_store, _store->ctx()->_conf->rgw_bucket_quota_cache_size) {
  }

  AsyncRefreshHandler *allocate_refresh_handler(const rgw_user& user, const rgw_bucket& bucket) override {
    return new BucketAsyncRefreshHandler(store, this, user, bucket);
  }
};

int RGWBucketStatsCache::fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats, optional_yield y, const DoutPrefixProvider *dpp)
{
  RGWBucketInfo bucket_info;

  RGWSysObjectCtx obj_ctx = store->svc()->sysobj->init_obj_ctx();

  int r = store->getRados()->get_bucket_instance_info(obj_ctx, bucket, bucket_info, NULL, NULL, y, dpp);
  if (r < 0) {
    ldpp_dout(dpp, 0) << "could not get bucket info for bucket=" << bucket << " r=" << r << dendl;
    return r;
  }

  string bucket_ver;
  string master_ver;

  map<RGWObjCategory, RGWStorageStats> bucket_stats;
  r = store->getRados()->get_bucket_stats(dpp, bucket_info, RGW_NO_SHARD, &bucket_ver,
                                  &master_ver, bucket_stats, nullptr);
  if (r < 0) {
    ldpp_dout(dpp, 0) << "could not get bucket stats for bucket="
                           << bucket.name << dendl;
    return r;
  }

  stats = RGWStorageStats();

  for (const auto& pair : bucket_stats) {
    const RGWStorageStats& s = pair.second;

    stats.size += s.size;
    stats.size_rounded += s.size_rounded;
    stats.num_objects += s.num_objects;
  }

  return 0;
}

class UserAsyncRefreshHandler : public RGWQuotaCache<rgw_user>::AsyncRefreshHandler,
                                public RGWGetUserStats_CB {
  const DoutPrefixProvider *dpp;
  rgw_bucket bucket;
public:
  UserAsyncRefreshHandler(const DoutPrefixProvider *_dpp, rgw::sal::RGWRadosStore *_store, RGWQuotaCache<rgw_user> *_cache,
                          const rgw_user& _user, const rgw_bucket& _bucket) :
                          RGWQuotaCache<rgw_user>::AsyncRefreshHandler(_store, _cache),
                          RGWGetUserStats_CB(_user),
                          dpp(_dpp),
                          bucket(_bucket) {}

  void drop_reference() override { put(); }
  int init_fetch() override;
  void handle_response(int r) override;
};

int UserAsyncRefreshHandler::init_fetch()
{
  ldpp_dout(dpp, 20) << "initiating async quota refresh for user=" << user << dendl;
  int r = store->ctl()->user->read_stats_async(dpp, user, this);
  if (r < 0) {
    ldpp_dout(dpp, 0) << "could not get bucket info for user=" << user << dendl;

    /* get_bucket_stats_async() dropped our reference already */
    return r;
  }

  return 0;
}

void UserAsyncRefreshHandler::handle_response(int r)
{
  if (r < 0) {
    ldout(store->ctx(), 20) << "AsyncRefreshHandler::handle_response() r=" << r << dendl;
    cache->async_refresh_fail(user, bucket);
    return;
  }

  cache->async_refresh_response(user, bucket, stats);
}

class RGWUserStatsCache : public RGWQuotaCache<rgw_user> {
  const DoutPrefixProvider *dpp;
  std::atomic<bool> down_flag = { false };
  ceph::shared_mutex mutex = ceph::make_shared_mutex("RGWUserStatsCache");
  map<rgw_bucket, rgw_user> modified_buckets;

  /* thread, sync recent modified buckets info */
  class BucketsSyncThread : public Thread {
    CephContext *cct;
    RGWUserStatsCache *stats;

    ceph::mutex lock = ceph::make_mutex("RGWUserStatsCache::BucketsSyncThread");
    ceph::condition_variable cond;
  public:

    BucketsSyncThread(CephContext *_cct, RGWUserStatsCache *_s) : cct(_cct), stats(_s) {}

    void *entry() override {
      ldout(cct, 20) << "BucketsSyncThread: start" << dendl;
      do {
        map<rgw_bucket, rgw_user> buckets;

        stats->swap_modified_buckets(buckets);

        for (map<rgw_bucket, rgw_user>::iterator iter = buckets.begin(); iter != buckets.end(); ++iter) {
          rgw_bucket bucket = iter->first;
          rgw_user& user = iter->second;
          ldout(cct, 20) << "BucketsSyncThread: sync user=" << user << " bucket=" << bucket << dendl;
          const DoutPrefix dp(cct, dout_subsys, "rgw bucket sync thread: ");
          int r = stats->sync_bucket(user, bucket, null_yield, &dp);
          if (r < 0) {
            ldout(cct, 0) << "WARNING: sync_bucket() returned r=" << r << dendl;
          }
        }

        if (stats->going_down())
          break;

	std::unique_lock locker{lock};
	cond.wait_for(
          locker,
          std::chrono::seconds(cct->_conf->rgw_user_quota_bucket_sync_interval));
      } while (!stats->going_down());
      ldout(cct, 20) << "BucketsSyncThread: done" << dendl;

      return NULL;
    }

    void stop() {
      std::lock_guard l{lock};
      cond.notify_all();
    }
  };

  /*
   * thread, full sync all users stats periodically
   *
   * only sync non idle users or ones that never got synced before, this is needed so that
   * users that didn't have quota turned on before (or existed before the user objclass
   * tracked stats) need to get their backend stats up to date.
   */
  class UserSyncThread : public Thread {
    CephContext *cct;
    RGWUserStatsCache *stats;

    ceph::mutex lock = ceph::make_mutex("RGWUserStatsCache::UserSyncThread");
    ceph::condition_variable cond;
  public:

    UserSyncThread(CephContext *_cct, RGWUserStatsCache *_s) : cct(_cct), stats(_s) {}

    void *entry() override {
      ldout(cct, 20) << "UserSyncThread: start" << dendl;
      do {
        const DoutPrefix dp(cct, dout_subsys, "rgw user sync thread: ");
        int ret = stats->sync_all_users(&dp, null_yield);
        if (ret < 0) {
          ldout(cct, 5) << "ERROR: sync_all_users() returned ret=" << ret << dendl;
        }

        if (stats->going_down())
          break;

	std::unique_lock l{lock};
        cond.wait_for(l, std::chrono::seconds(cct->_conf->rgw_user_quota_sync_interval));
      } while (!stats->going_down());
      ldout(cct, 20) << "UserSyncThread: done" << dendl;

      return NULL;
    }

    void stop() {
      std::lock_guard l{lock};
      cond.notify_all();
    }
  };

  BucketsSyncThread *buckets_sync_thread;
  UserSyncThread *user_sync_thread;
protected:
  bool map_find(const rgw_user& user,const rgw_bucket& bucket, RGWQuotaCacheStats& qs) override {
    return stats_map.find(user, qs);
  }

  bool map_find_and_update(const rgw_user& user, const rgw_bucket& bucket, lru_map<rgw_user, RGWQuotaCacheStats>::UpdateContext *ctx) override {
    return stats_map.find_and_update(user, NULL, ctx);
  }

  void map_add(const rgw_user& user, const rgw_bucket& bucket, RGWQuotaCacheStats& qs) override {
    stats_map.add(user, qs);
  }

  int fetch_stats_from_storage(const rgw_user& user, const rgw_bucket& bucket, RGWStorageStats& stats, optional_yield y, const DoutPrefixProvider *dpp) override;
  int sync_bucket(const rgw_user& rgw_user, rgw_bucket& bucket, optional_yield y, const DoutPrefixProvider *dpp);
  int sync_user(const DoutPrefixProvider *dpp, const rgw_user& user, optional_yield y);
  int sync_all_users(const DoutPrefixProvider *dpp, optional_yield y);

  void data_modified(const rgw_user& user, rgw_bucket& bucket) override;

  void swap_modified_buckets(map<rgw_bucket, rgw_user>& out) {
    std::unique_lock lock{mutex};
    modified_buckets.swap(out);
  }

  template<class T> /* easier doing it as a template, Thread doesn't have ->stop() */
  void stop_thread(T **pthr) {
    T *thread = *pthr;
    if (!thread)
      return;

    thread->stop();
    thread->join();
    delete thread;
    *pthr = NULL;
  }

public:
  RGWUserStatsCache(const DoutPrefixProvider *dpp, rgw::sal::RGWRadosStore *_store, bool quota_threads)
    : RGWQuotaCache<rgw_user>(_store, _store->ctx()->_conf->rgw_bucket_quota_cache_size), dpp(dpp)
  {
    if (quota_threads) {
      buckets_sync_thread = new BucketsSyncThread(store->ctx(), this);
      buckets_sync_thread->create("rgw_buck_st_syn");
      user_sync_thread = new UserSyncThread(store->ctx(), this);
      user_sync_thread->create("rgw_user_st_syn");
    } else {
      buckets_sync_thread = NULL;
      user_sync_thread = NULL;
    }
  }
  ~RGWUserStatsCache() override {
    stop();
  }

  AsyncRefreshHandler *allocate_refresh_handler(const rgw_user& user, const rgw_bucket& bucket) override {
    return new UserAsyncRefreshHandler(dpp, store, this, user, bucket);
  }

  bool going_down() {
    return down_flag;
  }

  void stop() {
    down_flag = true;
    {
      std::unique_lock lock{mutex};
      stop_thread(&buckets_sync_thread);
    }
    stop_thread(&user_sync_thread);
  }
};

int RGWUserStatsCache::fetch_stats_from_storage(const rgw_user& user,
						const rgw_bucket& bucket,
						RGWStorageStats& stats,
						optional_yield y,
                                                const DoutPrefixProvider *dpp)
{
  int r = store->ctl()->user->read_stats(dpp, user, &stats, y);
  if (r < 0) {
    ldout(store->ctx(), 0) << "could not get user stats for user=" << user << dendl;
    return r;
  }

  return 0;
}

int RGWUserStatsCache::sync_bucket(const rgw_user& user, rgw_bucket& bucket, optional_yield y, const DoutPrefixProvider *dpp)
{
  RGWBucketInfo bucket_info;

  int r = store->ctl()->bucket->read_bucket_instance_info(bucket, &bucket_info, y, dpp);
  if (r < 0) {
    ldout(store->ctx(), 0) << "could not get bucket info for bucket=" << bucket << " r=" << r << dendl;
    return r;
  }

  RGWBucketEnt ent;
  r = store->ctl()->bucket->sync_user_stats(dpp, user, bucket_info, y, &ent);
  if (r < 0) {
    ldout(store->ctx(), 0) << "ERROR: sync_user_stats() for user=" << user << ", bucket=" << bucket << " returned " << r << dendl;
    return r;
  }

  return store->getRados()->check_bucket_shards(bucket_info, bucket, ent.count, dpp);
}

int RGWUserStatsCache::sync_user(const DoutPrefixProvider *dpp, const rgw_user& user, optional_yield y)
{
  string user_str = user.to_str();
  RGWStorageStats stats;
  ceph::real_time last_stats_sync;
  ceph::real_time last_stats_update;

  int ret = store->ctl()->user->read_stats(dpp, rgw_user(user_str), &stats, y, &last_stats_sync, &last_stats_update);
  if (ret < 0) {
    ldout(store->ctx(), 5) << "ERROR: can't read user header: ret=" << ret << dendl;
    return ret;
  }

  if (!store->ctx()->_conf->rgw_user_quota_sync_idle_users &&
      last_stats_update < last_stats_sync) {
    ldout(store->ctx(), 20) << "user is idle, not doing a full sync (user=" << user << ")" << dendl;
    return 0;
  }

  real_time when_need_full_sync = last_stats_sync;
  when_need_full_sync += make_timespan(store->ctx()->_conf->rgw_user_quota_sync_wait_time);
  
  // check if enough time passed since last full sync
  /* FIXME: missing check? */

  ret = rgw_user_sync_all_stats(dpp, store, user, y);
  if (ret < 0) {
    ldout(store->ctx(), 0) << "ERROR: failed user stats sync, ret=" << ret << dendl;
    return ret;
  }

  return 0;
}

int RGWUserStatsCache::sync_all_users(const DoutPrefixProvider *dpp, optional_yield y)
{
  string key = "user";
  void *handle;

  int ret = store->ctl()->meta.mgr->list_keys_init(dpp, key, &handle);
  if (ret < 0) {
    ldpp_dout(dpp, 10) << "ERROR: can't get key: ret=" << ret << dendl;
    return ret;
  }

  bool truncated;
  int max = 1000;

  do {
    list<string> keys;
    ret = store->ctl()->meta.mgr->list_keys_next(handle, max, keys, &truncated);
    if (ret < 0) {
      ldpp_dout(dpp, 0) << "ERROR: lists_keys_next(): ret=" << ret << dendl;
      goto done;
    }
    for (list<string>::iterator iter = keys.begin();
         iter != keys.end() && !going_down(); 
         ++iter) {
      rgw_user user(*iter);
      ldpp_dout(dpp, 20) << "RGWUserStatsCache: sync user=" << user << dendl;
      int ret = sync_user(dpp, user, y);
      if (ret < 0) {
        ldpp_dout(dpp, 5) << "ERROR: sync_user() failed, user=" << user << " ret=" << ret << dendl;

        /* continuing to next user */
        continue;
      }
    }
  } while (truncated);

  ret = 0;
done:
  store->ctl()->meta.mgr->list_keys_complete(handle);
  return ret;
}

void RGWUserStatsCache::data_modified(const rgw_user& user, rgw_bucket& bucket)
{
  /* racy, but it's ok */
  mutex.lock_shared();
  bool need_update = modified_buckets.find(bucket) == modified_buckets.end();
  mutex.unlock_shared();

  if (need_update) {
    std::unique_lock lock{mutex};
    modified_buckets[bucket] = user;
  }
}


class RGWQuotaInfoApplier {
  /* NOTE: no non-static field allowed as instances are supposed to live in
   * the static memory only. */
protected:
  RGWQuotaInfoApplier() = default;

public:
  virtual ~RGWQuotaInfoApplier() {}

  virtual bool is_size_exceeded(const char * const entity,
                                const RGWQuotaInfo& qinfo,
                                const RGWStorageStats& stats,
                                const uint64_t size) const = 0;

  virtual bool is_num_objs_exceeded(const char * const entity,
                                    const RGWQuotaInfo& qinfo,
                                    const RGWStorageStats& stats,
                                    const uint64_t num_objs) const = 0;

  static const RGWQuotaInfoApplier& get_instance(const RGWQuotaInfo& qinfo);
};

class RGWQuotaInfoDefApplier : public RGWQuotaInfoApplier {
public:
  bool is_size_exceeded(const char * const entity,
                                const RGWQuotaInfo& qinfo,
                                const RGWStorageStats& stats,
                                const uint64_t size) const override;

  bool is_num_objs_exceeded(const char * const entity,
                                    const RGWQuotaInfo& qinfo,
                                    const RGWStorageStats& stats,
                                    const uint64_t num_objs) const override;
};

class RGWQuotaInfoRawApplier : public RGWQuotaInfoApplier {
public:
  bool is_size_exceeded(const char * const entity,
                                const RGWQuotaInfo& qinfo,
                                const RGWStorageStats& stats,
                                const uint64_t size) const override;

  bool is_num_objs_exceeded(const char * const entity,
                                    const RGWQuotaInfo& qinfo,
                                    const RGWStorageStats& stats,
                                    const uint64_t num_objs) const override;
};


bool RGWQuotaInfoDefApplier::is_size_exceeded(const char * const entity,
                                              const RGWQuotaInfo& qinfo,
                                              const RGWStorageStats& stats,
                                              const uint64_t size) const
{
  if (qinfo.max_size < 0) {
    /* The limit is not enabled. */
    return false;
  }

  const uint64_t cur_size = stats.size_rounded;
  const uint64_t new_size = rgw_rounded_objsize(size);

  if (cur_size + new_size > static_cast<uint64_t>(qinfo.max_size)) {
    dout(10) << "quota exceeded: stats.size_rounded=" << stats.size_rounded
             << " size=" << new_size << " "
             << entity << "_quota.max_size=" << qinfo.max_size << dendl;
    return true;
  }

  return false;
}

bool RGWQuotaInfoDefApplier::is_num_objs_exceeded(const char * const entity,
                                                  const RGWQuotaInfo& qinfo,
                                                  const RGWStorageStats& stats,
                                                  const uint64_t num_objs) const
{
  if (qinfo.max_objects < 0) {
    /* The limit is not enabled. */
    return false;
  }

  if (stats.num_objects + num_objs > static_cast<uint64_t>(qinfo.max_objects)) {
    dout(10) << "quota exceeded: stats.num_objects=" << stats.num_objects
             << " " << entity << "_quota.max_objects=" << qinfo.max_objects
             << dendl;
    return true;
  }

  return false;
}

bool RGWQuotaInfoRawApplier::is_size_exceeded(const char * const entity,
                                              const RGWQuotaInfo& qinfo,
                                              const RGWStorageStats& stats,
                                              const uint64_t size) const
{
  if (qinfo.max_size < 0) {
    /* The limit is not enabled. */
    return false;
  }

  const uint64_t cur_size = stats.size;

  if (cur_size + size > static_cast<uint64_t>(qinfo.max_size)) {
    dout(10) << "quota exceeded: stats.size=" << stats.size
             << " size=" << size << " "
             << entity << "_quota.max_size=" << qinfo.max_size << dendl;
    return true;
  }

  return false;
}

bool RGWQuotaInfoRawApplier::is_num_objs_exceeded(const char * const entity,
                                                  const RGWQuotaInfo& qinfo,
                                                  const RGWStorageStats& stats,
                                                  const uint64_t num_objs) const
{
  if (qinfo.max_objects < 0) {
    /* The limit is not enabled. */
    return false;
  }

  if (stats.num_objects + num_objs > static_cast<uint64_t>(qinfo.max_objects)) {
    dout(10) << "quota exceeded: stats.num_objects=" << stats.num_objects
             << " " << entity << "_quota.max_objects=" << qinfo.max_objects
             << dendl;
    return true;
  }

  return false;
}

const RGWQuotaInfoApplier& RGWQuotaInfoApplier::get_instance(
  const RGWQuotaInfo& qinfo)
{
  static RGWQuotaInfoDefApplier default_qapplier;
  static RGWQuotaInfoRawApplier raw_qapplier;

  if (qinfo.check_on_raw) {
    return raw_qapplier;
  } else {
    return default_qapplier;
  }
}


class RGWQuotaHandlerImpl : public RGWQuotaHandler {
  rgw::sal::RGWRadosStore *store;
  RGWBucketStatsCache bucket_stats_cache;
  RGWUserStatsCache user_stats_cache;

  int check_quota(const char * const entity,
                  const RGWQuotaInfo& quota,
                  const RGWStorageStats& stats,
                  const uint64_t num_objs,
                  const uint64_t size) {
    if (!quota.enabled) {
      return 0;
    }

    const auto& quota_applier = RGWQuotaInfoApplier::get_instance(quota);

    ldout(store->ctx(), 20) << entity
                            << " quota: max_objects=" << quota.max_objects
                            << " max_size=" << quota.max_size << dendl;


    if (quota_applier.is_num_objs_exceeded(entity, quota, stats, num_objs)) {
      return -ERR_QUOTA_EXCEEDED;
    }

    if (quota_applier.is_size_exceeded(entity, quota, stats, size)) {
      return -ERR_QUOTA_EXCEEDED;
    }

    ldout(store->ctx(), 20) << entity << " quota OK:"
                            << " stats.num_objects=" << stats.num_objects
                            << " stats.size=" << stats.size << dendl;
    return 0;
  }
public:
  RGWQuotaHandlerImpl(const DoutPrefixProvider *dpp, rgw::sal::RGWRadosStore *_store, bool quota_threads) : store(_store),
                                    bucket_stats_cache(_store),
                                    user_stats_cache(dpp, _store, quota_threads) {}

  int check_quota(const rgw_user& user,
		  rgw_bucket& bucket,
		  RGWQuotaInfo& user_quota,
		  RGWQuotaInfo& bucket_quota,
		  uint64_t num_objs,
		  uint64_t size, optional_yield y) override {

    if (!bucket_quota.enabled && !user_quota.enabled) {
      return 0;
    }

    /*
     * we need to fetch bucket stats if the user quota is enabled, because
     * the whole system relies on us periodically updating the user's bucket
     * stats in the user's header, this happens in get_stats() if we actually
     * fetch that info and not rely on cached data
     */

    const DoutPrefix dp(store->ctx(), dout_subsys, "rgw quota handler: ");
    if (bucket_quota.enabled) {
      RGWStorageStats bucket_stats;
      int ret = bucket_stats_cache.get_stats(user, bucket, bucket_stats, y, &dp);
      if (ret < 0) {
        return ret;
      }
      ret = check_quota("bucket", bucket_quota, bucket_stats, num_objs, size);
      if (ret < 0) {
        return ret;
      }
    }

    if (user_quota.enabled) {
      RGWStorageStats user_stats;
      int ret = user_stats_cache.get_stats(user, bucket, user_stats, y, &dp);
      if (ret < 0) {
        return ret;
      }
      ret = check_quota("user", user_quota, user_stats, num_objs, size);
      if (ret < 0) {
        return ret;
      }
    }
    return 0;
  }

  void update_stats(const rgw_user& user, rgw_bucket& bucket, int obj_delta, uint64_t added_bytes, uint64_t removed_bytes) override {
    bucket_stats_cache.adjust_stats(user, bucket, obj_delta, added_bytes, removed_bytes);
    user_stats_cache.adjust_stats(user, bucket, obj_delta, added_bytes, removed_bytes);
  }

  void check_bucket_shards(uint64_t max_objs_per_shard, uint64_t num_shards,
			   uint64_t num_objs, bool& need_resharding, uint32_t *suggested_num_shards) override
  {
    if (num_objs > num_shards * max_objs_per_shard) {
      ldout(store->ctx(), 0) << __func__ << ": resharding needed: stats.num_objects=" << num_objs
             << " shard max_objects=" <<  max_objs_per_shard * num_shards << dendl;
      need_resharding = true;
      if (suggested_num_shards) {
        *suggested_num_shards = num_objs * 2 / max_objs_per_shard;
      }
    } else {
      need_resharding = false;
    }
  }
};


RGWQuotaHandler *RGWQuotaHandler::generate_handler(const DoutPrefixProvider *dpp, rgw::sal::RGWRadosStore *store, bool quota_threads)
{
  return new RGWQuotaHandlerImpl(dpp, store, quota_threads);
}

void RGWQuotaHandler::free_handler(RGWQuotaHandler *handler)
{
  delete handler;
}


void rgw_apply_default_bucket_quota(RGWQuotaInfo& quota, const ConfigProxy& conf)
{
  if (conf->rgw_bucket_default_quota_max_objects >= 0) {
    quota.max_objects = conf->rgw_bucket_default_quota_max_objects;
    quota.enabled = true;
  }
  if (conf->rgw_bucket_default_quota_max_size >= 0) {
    quota.max_size = conf->rgw_bucket_default_quota_max_size;
    quota.enabled = true;
  }
}

void rgw_apply_default_user_quota(RGWQuotaInfo& quota, const ConfigProxy& conf)
{
  if (conf->rgw_user_default_quota_max_objects >= 0) {
    quota.max_objects = conf->rgw_user_default_quota_max_objects;
    quota.enabled = true;
  }
  if (conf->rgw_user_default_quota_max_size >= 0) {
    quota.max_size = conf->rgw_user_default_quota_max_size;
    quota.enabled = true;
  }
}