ceph/src/osd/PGBackend.cc

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3 /*
   4  * Ceph - scalable distributed file system
   5  *
   6  * Copyright (C) 2013,2014 Inktank Storage, Inc.
   7  * Copyright (C) 2013,2014 Cloudwatt <libre.licensing@cloudwatt.com>
   8  *
   9  * Author: Loic Dachary <loic@dachary.org>
  10  *
  11  * This is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License version 2.1, as published by the Free Software
  14  * Foundation.  See file COPYING.
  15  *
  16  */
  17
  18
  19 #include "common/errno.h"
  20 #include "common/scrub_types.h"
  21 #include "ReplicatedBackend.h"
  22 #include "ScrubStore.h"
  23 #include "ECBackend.h"
  24 #include "PGBackend.h"
  25 #include "OSD.h"
  26 #include "erasure-code/ErasureCodePlugin.h"
  27 #include "OSDMap.h"
  28 #include "PGLog.h"
  29 #include "common/LogClient.h"
  30 #include "messages/MOSDPGRecoveryDelete.h"
  31 #include "messages/MOSDPGRecoveryDeleteReply.h"
  32
  33 #define dout_context cct
  34 #define dout_subsys ceph_subsys_osd
  35 #define DOUT_PREFIX_ARGS this
  36 #undef dout_prefix
  37 #define dout_prefix _prefix(_dout, this)
  38 static ostream& _prefix(std::ostream *_dout, PGBackend *pgb) {
  39   return pgb->get_parent()->gen_dbg_prefix(*_dout);
  40 }
  41
  42 void PGBackend::recover_delete_object(const hobject_t &oid, eversion_t v,
  43                                       RecoveryHandle *h)
  44 {
  45   ceph_assert(get_parent()->get_acting_recovery_backfill_shards().size() > 0);
  46   for (const auto& shard : get_parent()->get_acting_recovery_backfill_shards()) {
  47     if (shard == get_parent()->whoami_shard())
  48       continue;
  49     if (get_parent()->get_shard_missing(shard).is_missing(oid)) {
  50       dout(20) << __func__ << " will remove " << oid << " " << v << " from "
  51                << shard << dendl;
  52       h->deletes[shard].push_back(make_pair(oid, v));
  53       get_parent()->begin_peer_recover(shard, oid);
  54     }
  55   }
  56 }
  57
  58 void PGBackend::send_recovery_deletes(int prio,
  59                                       const map<pg_shard_t, vector<pair<hobject_t, eversion_t> > > &deletes)
  60 {
  61   epoch_t min_epoch = get_parent()->get_last_peering_reset_epoch();
  62   for (const auto& p : deletes) {
  63     const auto& shard = p.first;
  64     const auto& objects = p.second;
  65     ConnectionRef con = get_parent()->get_con_osd_cluster(
  66       shard.osd,
  67       get_osdmap_epoch());
  68     if (!con)
  69       continue;
  70     auto it = objects.begin();
  71     while (it != objects.end()) {
  72       uint64_t cost = 0;
  73       uint64_t deletes = 0;
  74       spg_t target_pg = spg_t(get_parent()->get_info().pgid.pgid, shard.shard);
  75       MOSDPGRecoveryDelete *msg =
  76         new MOSDPGRecoveryDelete(get_parent()->whoami_shard(),
  77                                  target_pg,
  78                                  get_osdmap_epoch(),
  79                                  min_epoch);
  80       msg->set_priority(prio);
  81
  82       while (it != objects.end() &&
  83              cost < cct->_conf->osd_max_push_cost &&
  84              deletes < cct->_conf->osd_max_push_objects) {
  85         dout(20) << __func__ << ": sending recovery delete << " << it->first
  86                  << " " << it->second << " to osd." << shard << dendl;
  87         msg->objects.push_back(*it);
  88         cost += cct->_conf->osd_push_per_object_cost;
  89         ++deletes;
  90         ++it;
  91       }
  92
  93       msg->set_cost(cost);
  94       get_parent()->send_message_osd_cluster(msg, con);
  95     }
  96   }
  97 }
  98
  99 bool PGBackend::handle_message(OpRequestRef op)
 100 {
 101   switch (op->get_req()->get_type()) {
 102   case MSG_OSD_PG_RECOVERY_DELETE:
 103     handle_recovery_delete(op);
 104     return true;
 105
 106   case MSG_OSD_PG_RECOVERY_DELETE_REPLY:
 107     handle_recovery_delete_reply(op);
 108     return true;
 109
 110   default:
 111     break;
 112   }
 113
 114   return _handle_message(op);
 115 }
 116
 117 void PGBackend::handle_recovery_delete(OpRequestRef op)
 118 {
 119   auto m = op->get_req<MOSDPGRecoveryDelete>();
 120   ceph_assert(m->get_type() == MSG_OSD_PG_RECOVERY_DELETE);
 121   dout(20) << __func__ << " " << op << dendl;
 122
 123   op->mark_started();
 124
 125   C_GatherBuilder gather(cct);
 126   for (const auto &p : m->objects) {
 127     get_parent()->remove_missing_object(p.first, p.second, gather.new_sub());
 128   }
 129
 130   MOSDPGRecoveryDeleteReply *reply = new MOSDPGRecoveryDeleteReply;
 131   reply->from = get_parent()->whoami_shard();
 132   reply->set_priority(m->get_priority());
 133   reply->pgid = spg_t(get_parent()->get_info().pgid.pgid, m->from.shard);
 134   reply->map_epoch = m->map_epoch;
 135   reply->min_epoch = m->min_epoch;
 136   reply->objects = m->objects;
 137   ConnectionRef conn = m->get_connection();
 138
 139   gather.set_finisher(new LambdaContext(
 140     [=](int r) {
 141       if (r != -EAGAIN) {
 142         get_parent()->send_message_osd_cluster(reply, conn.get());
 143       } else {
 144         reply->put();
 145       }
 146     }));
 147   gather.activate();
 148 }
 149
 150 void PGBackend::handle_recovery_delete_reply(OpRequestRef op)
 151 {
 152   auto m = op->get_req<MOSDPGRecoveryDeleteReply>();
 153   ceph_assert(m->get_type() == MSG_OSD_PG_RECOVERY_DELETE_REPLY);
 154   dout(20) << __func__ << " " << op << dendl;
 155
 156   for (const auto &p : m->objects) {
 157     ObjectRecoveryInfo recovery_info;
 158     hobject_t oid = p.first;
 159     recovery_info.version = p.second;
 160     get_parent()->on_peer_recover(m->from, oid, recovery_info);
 161     bool peers_recovered = true;
 162     for (const auto& shard : get_parent()->get_acting_recovery_backfill_shards()) {
 163       if (shard == get_parent()->whoami_shard())
 164         continue;
 165       if (get_parent()->get_shard_missing(shard).is_missing(oid)) {
 166         dout(20) << __func__ << " " << oid << " still missing on at least "
 167                  << shard << dendl;
 168         peers_recovered = false;
 169         break;
 170       }
 171     }
 172     if (peers_recovered && !get_parent()->get_local_missing().is_missing(oid)) {
 173       dout(20) << __func__ << " completed recovery, local_missing = "
 174                << get_parent()->get_local_missing() << dendl;
 175       object_stat_sum_t stat_diff;
 176       stat_diff.num_objects_recovered = 1;
 177       get_parent()->on_global_recover(p.first, stat_diff, true);
 178     }
 179   }
 180 }
 181
 182 void PGBackend::rollback(
 183   const pg_log_entry_t &entry,
 184   ObjectStore::Transaction *t)
 185 {
 186
 187   struct RollbackVisitor : public ObjectModDesc::Visitor {
 188     const hobject_t &hoid;
 189     PGBackend *pg;
 190     ObjectStore::Transaction t;
 191     RollbackVisitor(
 192       const hobject_t &hoid,
 193       PGBackend *pg) : hoid(hoid), pg(pg) {}
 194     void append(uint64_t old_size) override {
 195       ObjectStore::Transaction temp;
 196       pg->rollback_append(hoid, old_size, &temp);
 197       temp.append(t);
 198       temp.swap(t);
 199     }
 200     void setattrs(map<string, std::optional<bufferlist> > &attrs) override {
 201       ObjectStore::Transaction temp;
 202       pg->rollback_setattrs(hoid, attrs, &temp);
 203       temp.append(t);
 204       temp.swap(t);
 205     }
 206     void rmobject(version_t old_version) override {
 207       ObjectStore::Transaction temp;
 208       pg->rollback_stash(hoid, old_version, &temp);
 209       temp.append(t);
 210       temp.swap(t);
 211     }
 212     void try_rmobject(version_t old_version) override {
 213       ObjectStore::Transaction temp;
 214       pg->rollback_try_stash(hoid, old_version, &temp);
 215       temp.append(t);
 216       temp.swap(t);
 217     }
 218     void create() override {
 219       ObjectStore::Transaction temp;
 220       pg->rollback_create(hoid, &temp);
 221       temp.append(t);
 222       temp.swap(t);
 223     }
 224     void update_snaps(const set<snapid_t> &snaps) override {
 225       ObjectStore::Transaction temp;
 226       pg->get_parent()->pgb_set_object_snap_mapping(hoid, snaps, &temp);
 227       temp.append(t);
 228       temp.swap(t);
 229     }
 230     void rollback_extents(
 231       version_t gen,
 232       const vector<pair<uint64_t, uint64_t> > &extents) override {
 233       ObjectStore::Transaction temp;
 234       pg->rollback_extents(gen, extents, hoid, &temp);
 235       temp.append(t);
 236       temp.swap(t);
 237     }
 238   };
 239
 240   ceph_assert(entry.mod_desc.can_rollback());
 241   RollbackVisitor vis(entry.soid, this);
 242   entry.mod_desc.visit(&vis);
 243   t->append(vis.t);
 244 }
 245
 246 struct Trimmer : public ObjectModDesc::Visitor {
 247   const hobject_t &soid;
 248   PGBackend *pg;
 249   ObjectStore::Transaction *t;
 250   Trimmer(
 251     const hobject_t &soid,
 252     PGBackend *pg,
 253     ObjectStore::Transaction *t)
 254     : soid(soid), pg(pg), t(t) {}
 255   void rmobject(version_t old_version) override {
 256     pg->trim_rollback_object(
 257       soid,
 258       old_version,
 259       t);
 260   }
 261   // try_rmobject defaults to rmobject
 262   void rollback_extents(
 263     version_t gen,
 264     const vector<pair<uint64_t, uint64_t> > &extents) override {
 265     pg->trim_rollback_object(
 266       soid,
 267       gen,
 268       t);
 269   }
 270 };
 271
 272 void PGBackend::rollforward(
 273   const pg_log_entry_t &entry,
 274   ObjectStore::Transaction *t)
 275 {
 276   auto dpp = get_parent()->get_dpp();
 277   ldpp_dout(dpp, 20) << __func__ << ": entry=" << entry << dendl;
 278   if (!entry.can_rollback())
 279     return;
 280   Trimmer trimmer(entry.soid, this, t);
 281   entry.mod_desc.visit(&trimmer);
 282 }
 283
 284 void PGBackend::trim(
 285   const pg_log_entry_t &entry,
 286   ObjectStore::Transaction *t)
 287 {
 288   if (!entry.can_rollback())
 289     return;
 290   Trimmer trimmer(entry.soid, this, t);
 291   entry.mod_desc.visit(&trimmer);
 292 }
 293
 294 void PGBackend::try_stash(
 295   const hobject_t &hoid,
 296   version_t v,
 297   ObjectStore::Transaction *t)
 298 {
 299   t->try_rename(
 300     coll,
 301     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 302     ghobject_t(hoid, v, get_parent()->whoami_shard().shard));
 303 }
 304
 305 void PGBackend::remove(
 306   const hobject_t &hoid,
 307   ObjectStore::Transaction *t) {
 308   ceph_assert(!hoid.is_temp());
 309   t->remove(
 310     coll,
 311     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard));
 312   get_parent()->pgb_clear_object_snap_mapping(hoid, t);
 313 }
 314
 315 void PGBackend::on_change_cleanup(ObjectStore::Transaction *t)
 316 {
 317   dout(10) << __func__ << dendl;
 318   // clear temp
 319   for (set<hobject_t>::iterator i = temp_contents.begin();
 320        i != temp_contents.end();
 321        ++i) {
 322     dout(10) << __func__ << ": Removing oid "
 323              << *i << " from the temp collection" << dendl;
 324     t->remove(
 325       coll,
 326       ghobject_t(*i, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard));
 327   }
 328   temp_contents.clear();
 329 }
 330
 331 int PGBackend::objects_list_partial(
 332   const hobject_t &begin,
 333   int min,
 334   int max,
 335   vector<hobject_t> *ls,
 336   hobject_t *next)
 337 {
 338   ceph_assert(ls);
 339   // Starts with the smallest generation to make sure the result list
 340   // has the marker object (it might have multiple generations
 341   // though, which would be filtered).
 342   ghobject_t _next;
 343   if (!begin.is_min())
 344     _next = ghobject_t(begin, 0, get_parent()->whoami_shard().shard);
 345   ls->reserve(max);
 346   int r = 0;
 347
 348   if (min > max)
 349     min = max;
 350
 351   while (!_next.is_max() && ls->size() < (unsigned)min) {
 352     vector<ghobject_t> objects;
 353     r = store->collection_list(
 354       ch,
 355       _next,
 356       ghobject_t::get_max(),
 357       max - ls->size(),
 358       &objects,
 359       &_next);
 360     if (r != 0) {
 361       derr << __func__ << " list collection " << ch << " got: " << cpp_strerror(r) << dendl;
 362       break;
 363     }
 364     for (vector<ghobject_t>::iterator i = objects.begin();
 365          i != objects.end();
 366          ++i) {
 367       if (i->is_pgmeta() || i->hobj.is_temp()) {
 368         continue;
 369       }
 370       if (i->is_no_gen()) {
 371         ls->push_back(i->hobj);
 372       }
 373     }
 374   }
 375   if (r == 0)
 376     *next = _next.hobj;
 377   return r;
 378 }
 379
 380 int PGBackend::objects_list_range(
 381   const hobject_t &start,
 382   const hobject_t &end,
 383   vector<hobject_t> *ls,
 384   vector<ghobject_t> *gen_obs)
 385 {
 386   ceph_assert(ls);
 387   vector<ghobject_t> objects;
 388   int r = store->collection_list(
 389     ch,
 390     ghobject_t(start, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 391     ghobject_t(end, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 392     INT_MAX,
 393     &objects,
 394     NULL);
 395   ls->reserve(objects.size());
 396   for (vector<ghobject_t>::iterator i = objects.begin();
 397        i != objects.end();
 398        ++i) {
 399     if (i->is_pgmeta() || i->hobj.is_temp()) {
 400       continue;
 401     }
 402     if (i->is_no_gen()) {
 403       ls->push_back(i->hobj);
 404     } else if (gen_obs) {
 405       gen_obs->push_back(*i);
 406     }
 407   }
 408   return r;
 409 }
 410
 411 int PGBackend::objects_get_attr(
 412   const hobject_t &hoid,
 413   const string &attr,
 414   bufferlist *out)
 415 {
 416   bufferptr bp;
 417   int r = store->getattr(
 418     ch,
 419     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 420     attr.c_str(),
 421     bp);
 422   if (r >= 0 && out) {
 423     out->clear();
 424     out->push_back(std::move(bp));
 425   }
 426   return r;
 427 }
 428
 429 int PGBackend::objects_get_attrs(
 430   const hobject_t &hoid,
 431   map<string, bufferlist> *out)
 432 {
 433   return store->getattrs(
 434     ch,
 435     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 436     *out);
 437 }
 438
 439 void PGBackend::rollback_setattrs(
 440   const hobject_t &hoid,
 441   map<string, std::optional<bufferlist> > &old_attrs,
 442   ObjectStore::Transaction *t) {
 443   map<string, bufferlist> to_set;
 444   ceph_assert(!hoid.is_temp());
 445   for (map<string, std::optional<bufferlist> >::iterator i = old_attrs.begin();
 446        i != old_attrs.end();
 447        ++i) {
 448     if (i->second) {
 449       to_set[i->first] = *(i->second);
 450     } else {
 451       t->rmattr(
 452         coll,
 453         ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 454         i->first);
 455     }
 456   }
 457   t->setattrs(
 458     coll,
 459     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 460     to_set);
 461 }
 462
 463 void PGBackend::rollback_append(
 464   const hobject_t &hoid,
 465   uint64_t old_size,
 466   ObjectStore::Transaction *t) {
 467   ceph_assert(!hoid.is_temp());
 468   t->truncate(
 469     coll,
 470     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 471     old_size);
 472 }
 473
 474 void PGBackend::rollback_stash(
 475   const hobject_t &hoid,
 476   version_t old_version,
 477   ObjectStore::Transaction *t) {
 478   ceph_assert(!hoid.is_temp());
 479   t->remove(
 480     coll,
 481     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard));
 482   t->collection_move_rename(
 483     coll,
 484     ghobject_t(hoid, old_version, get_parent()->whoami_shard().shard),
 485     coll,
 486     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard));
 487 }
 488
 489 void PGBackend::rollback_try_stash(
 490   const hobject_t &hoid,
 491   version_t old_version,
 492   ObjectStore::Transaction *t) {
 493   ceph_assert(!hoid.is_temp());
 494   t->remove(
 495     coll,
 496     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard));
 497   t->try_rename(
 498     coll,
 499     ghobject_t(hoid, old_version, get_parent()->whoami_shard().shard),
 500     ghobject_t(hoid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard));
 501 }
 502
 503 void PGBackend::rollback_extents(
 504   version_t gen,
 505   const vector<pair<uint64_t, uint64_t> > &extents,
 506   const hobject_t &hoid,
 507   ObjectStore::Transaction *t) {
 508   auto shard = get_parent()->whoami_shard().shard;
 509   for (auto &&extent: extents) {
 510     t->clone_range(
 511       coll,
 512       ghobject_t(hoid, gen, shard),
 513       ghobject_t(hoid, ghobject_t::NO_GEN, shard),
 514       extent.first,
 515       extent.second,
 516       extent.first);
 517   }
 518   t->remove(
 519     coll,
 520     ghobject_t(hoid, gen, shard));
 521 }
 522
 523 void PGBackend::trim_rollback_object(
 524   const hobject_t &hoid,
 525   version_t old_version,
 526   ObjectStore::Transaction *t) {
 527   ceph_assert(!hoid.is_temp());
 528   t->remove(
 529     coll, ghobject_t(hoid, old_version, get_parent()->whoami_shard().shard));
 530 }
 531
 532 PGBackend *PGBackend::build_pg_backend(
 533   const pg_pool_t &pool,
 534   const map<string,string>& profile,
 535   Listener *l,
 536   coll_t coll,
 537   ObjectStore::CollectionHandle &ch,
 538   ObjectStore *store,
 539   CephContext *cct)
 540 {
 541   ErasureCodeProfile ec_profile = profile;
 542   switch (pool.type) {
 543   case pg_pool_t::TYPE_REPLICATED: {
 544     return new ReplicatedBackend(l, coll, ch, store, cct);
 545   }
 546   case pg_pool_t::TYPE_ERASURE: {
 547     ErasureCodeInterfaceRef ec_impl;
 548     stringstream ss;
 549     ceph::ErasureCodePluginRegistry::instance().factory(
 550       profile.find("plugin")->second,
 551       cct->_conf.get_val<std::string>("erasure_code_dir"),
 552       ec_profile,
 553       &ec_impl,
 554       &ss);
 555     ceph_assert(ec_impl);
 556     return new ECBackend(
 557       l,
 558       coll,
 559       ch,
 560       store,
 561       cct,
 562       ec_impl,
 563       pool.stripe_width);
 564   }
 565   default:
 566     ceph_abort();
 567     return NULL;
 568   }
 569 }
 570
 571 int PGBackend::be_scan_list(
 572   ScrubMap &map,
 573   ScrubMapBuilder &pos)
 574 {
 575   dout(10) << __func__ << " " << pos << dendl;
 576   ceph_assert(!pos.done());
 577   ceph_assert(pos.pos < pos.ls.size());
 578   hobject_t& poid = pos.ls[pos.pos];
 579
 580   struct stat st;
 581   int r = store->stat(
 582     ch,
 583     ghobject_t(
 584       poid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 585     &st,
 586     true);
 587   if (r == 0) {
 588     ScrubMap::object &o = map.objects[poid];
 589     o.size = st.st_size;
 590     ceph_assert(!o.negative);
 591     store->getattrs(
 592       ch,
 593       ghobject_t(
 594         poid, ghobject_t::NO_GEN, get_parent()->whoami_shard().shard),
 595       o.attrs);
 596
 597     if (pos.deep) {
 598       r = be_deep_scrub(poid, map, pos, o);
 599     }
 600     dout(25) << __func__ << "  " << poid << dendl;
 601   } else if (r == -ENOENT) {
 602     dout(25) << __func__ << "  " << poid << " got " << r
 603              << ", skipping" << dendl;
 604   } else if (r == -EIO) {
 605     dout(25) << __func__ << "  " << poid << " got " << r
 606              << ", stat_error" << dendl;
 607     ScrubMap::object &o = map.objects[poid];
 608     o.stat_error = true;
 609   } else {
 610     derr << __func__ << " got: " << cpp_strerror(r) << dendl;
 611     ceph_abort();
 612   }
 613   if (r == -EINPROGRESS) {
 614     return -EINPROGRESS;
 615   }
 616   pos.next_object();
 617   return 0;
 618 }
 619
 620 bool PGBackend::be_compare_scrub_objects(
 621   pg_shard_t auth_shard,
 622   const ScrubMap::object &auth,
 623   const object_info_t& auth_oi,
 624   const ScrubMap::object &candidate,
 625   shard_info_wrapper &shard_result,
 626   inconsistent_obj_wrapper &obj_result,
 627   ostream &errorstream,
 628   bool has_snapset)
 629 {
 630   enum { CLEAN, FOUND_ERROR } error = CLEAN;
 631   if (auth.digest_present && candidate.digest_present) {
 632     if (auth.digest != candidate.digest) {
 633       if (error != CLEAN)
 634         errorstream << ", ";
 635       error = FOUND_ERROR;
 636       errorstream << "data_digest 0x" << std::hex << candidate.digest
 637                   << " != data_digest 0x" << auth.digest << std::dec
 638                   << " from shard " << auth_shard;
 639       obj_result.set_data_digest_mismatch();
 640     }
 641   }
 642   if (auth.omap_digest_present && candidate.omap_digest_present) {
 643     if (auth.omap_digest != candidate.omap_digest) {
 644       if (error != CLEAN)
 645         errorstream << ", ";
 646       error = FOUND_ERROR;
 647       errorstream << "omap_digest 0x" << std::hex << candidate.omap_digest
 648                   << " != omap_digest 0x" << auth.omap_digest << std::dec
 649                   << " from shard " << auth_shard;
 650       obj_result.set_omap_digest_mismatch();
 651     }
 652   }
 653   if (parent->get_pool().is_replicated()) {
 654     if (auth_oi.is_data_digest() && candidate.digest_present) {
 655       if (auth_oi.data_digest != candidate.digest) {
 656         if (error != CLEAN)
 657           errorstream << ", ";
 658         error = FOUND_ERROR;
 659         errorstream << "data_digest 0x" << std::hex << candidate.digest
 660                     << " != data_digest 0x" << auth_oi.data_digest << std::dec
 661                     << " from auth oi " << auth_oi;
 662         shard_result.set_data_digest_mismatch_info();
 663       }
 664     }
 665     if (auth_oi.is_omap_digest() && candidate.omap_digest_present) {
 666       if (auth_oi.omap_digest != candidate.omap_digest) {
 667         if (error != CLEAN)
 668           errorstream << ", ";
 669         error = FOUND_ERROR;
 670         errorstream << "omap_digest 0x" << std::hex << candidate.omap_digest
 671                     << " != omap_digest 0x" << auth_oi.omap_digest << std::dec
 672                     << " from auth oi " << auth_oi;
 673         shard_result.set_omap_digest_mismatch_info();
 674       }
 675     }
 676   }
 677   if (candidate.stat_error)
 678     return error == FOUND_ERROR;
 679   if (!shard_result.has_info_missing()
 680       && !shard_result.has_info_corrupted()) {
 681     bufferlist can_bl, auth_bl;
 682     auto can_attr = candidate.attrs.find(OI_ATTR);
 683     auto auth_attr = auth.attrs.find(OI_ATTR);
 684
 685     ceph_assert(auth_attr != auth.attrs.end());
 686     ceph_assert(can_attr != candidate.attrs.end());
 687
 688     can_bl.push_back(can_attr->second);
 689     auth_bl.push_back(auth_attr->second);
 690     if (!can_bl.contents_equal(auth_bl)) {
 691       if (error != CLEAN)
 692         errorstream << ", ";
 693       error = FOUND_ERROR;
 694       obj_result.set_object_info_inconsistency();
 695       errorstream << "object info inconsistent ";
 696     }
 697   }
 698   if (has_snapset) {
 699     if (!shard_result.has_snapset_missing()
 700         && !shard_result.has_snapset_corrupted()) {
 701       bufferlist can_bl, auth_bl;
 702       auto can_attr = candidate.attrs.find(SS_ATTR);
 703       auto auth_attr = auth.attrs.find(SS_ATTR);
 704
 705       ceph_assert(auth_attr != auth.attrs.end());
 706       ceph_assert(can_attr != candidate.attrs.end());
 707
 708       can_bl.push_back(can_attr->second);
 709       auth_bl.push_back(auth_attr->second);
 710       if (!can_bl.contents_equal(auth_bl)) {
 711           if (error != CLEAN)
 712             errorstream << ", ";
 713           error = FOUND_ERROR;
 714           obj_result.set_snapset_inconsistency();
 715           errorstream << "snapset inconsistent ";
 716       }
 717     }
 718   }
 719   if (parent->get_pool().is_erasure()) {
 720     if (!shard_result.has_hinfo_missing()
 721         && !shard_result.has_hinfo_corrupted()) {
 722       bufferlist can_bl, auth_bl;
 723       auto can_hi = candidate.attrs.find(ECUtil::get_hinfo_key());
 724       auto auth_hi = auth.attrs.find(ECUtil::get_hinfo_key());
 725
 726       ceph_assert(auth_hi != auth.attrs.end());
 727       ceph_assert(can_hi != candidate.attrs.end());
 728
 729       can_bl.push_back(can_hi->second);
 730       auth_bl.push_back(auth_hi->second);
 731       if (!can_bl.contents_equal(auth_bl)) {
 732         if (error != CLEAN)
 733           errorstream << ", ";
 734         error = FOUND_ERROR;
 735         obj_result.set_hinfo_inconsistency();
 736         errorstream << "hinfo inconsistent ";
 737       }
 738     }
 739   }
 740   uint64_t oi_size = be_get_ondisk_size(auth_oi.size);
 741   if (oi_size != candidate.size) {
 742     if (error != CLEAN)
 743       errorstream << ", ";
 744     error = FOUND_ERROR;
 745     errorstream << "size " << candidate.size
 746                 << " != size " << oi_size
 747                 << " from auth oi " << auth_oi;
 748     shard_result.set_size_mismatch_info();
 749   }
 750   if (auth.size != candidate.size) {
 751     if (error != CLEAN)
 752       errorstream << ", ";
 753     error = FOUND_ERROR;
 754     errorstream << "size " << candidate.size
 755                 << " != size " << auth.size
 756                 << " from shard " << auth_shard;
 757     obj_result.set_size_mismatch();
 758   }
 759   // If the replica is too large and we didn't already count it for this object
 760   //
 761   if (candidate.size > cct->_conf->osd_max_object_size
 762       && !obj_result.has_size_too_large()) {
 763     if (error != CLEAN)
 764       errorstream << ", ";
 765     error = FOUND_ERROR;
 766     errorstream << "size " << candidate.size
 767                 << " > " << cct->_conf->osd_max_object_size
 768                 << " is too large";
 769     obj_result.set_size_too_large();
 770   }
 771   for (map<string,bufferptr>::const_iterator i = auth.attrs.begin();
 772        i != auth.attrs.end();
 773        ++i) {
 774     // We check system keys seperately
 775     if (i->first == OI_ATTR || i->first[0] != '_')
 776       continue;
 777     if (!candidate.attrs.count(i->first)) {
 778       if (error != CLEAN)
 779         errorstream << ", ";
 780       error = FOUND_ERROR;
 781       errorstream << "attr name mismatch '" << i->first << "'";
 782       obj_result.set_attr_name_mismatch();
 783     } else if (candidate.attrs.find(i->first)->second.cmp(i->second)) {
 784       if (error != CLEAN)
 785         errorstream << ", ";
 786       error = FOUND_ERROR;
 787       errorstream << "attr value mismatch '" << i->first << "'";
 788       obj_result.set_attr_value_mismatch();
 789     }
 790   }
 791   for (map<string,bufferptr>::const_iterator i = candidate.attrs.begin();
 792        i != candidate.attrs.end();
 793        ++i) {
 794     // We check system keys seperately
 795     if (i->first == OI_ATTR || i->first[0] != '_')
 796       continue;
 797     if (!auth.attrs.count(i->first)) {
 798       if (error != CLEAN)
 799         errorstream << ", ";
 800       error = FOUND_ERROR;
 801       errorstream << "attr name mismatch '" << i->first << "'";
 802       obj_result.set_attr_name_mismatch();
 803     }
 804   }
 805   return error == FOUND_ERROR;
 806 }
 807
 808 static int dcount(const object_info_t &oi)
 809 {
 810   int count = 0;
 811   if (oi.is_data_digest())
 812     count++;
 813   if (oi.is_omap_digest())
 814     count++;
 815   return count;
 816 }
 817
 818 map<pg_shard_t, ScrubMap *>::const_iterator
 819   PGBackend::be_select_auth_object(
 820   const hobject_t &obj,
 821   const map<pg_shard_t,ScrubMap*> &maps,
 822   object_info_t *auth_oi,
 823   map<pg_shard_t, shard_info_wrapper> &shard_map,
 824   bool &digest_match,
 825   spg_t pgid,
 826   ostream &errorstream)
 827 {
 828   eversion_t auth_version;
 829
 830   // Create list of shards with primary first so it will be auth copy all
 831   // other things being equal.
 832   list<pg_shard_t> shards;
 833   for (map<pg_shard_t, ScrubMap *>::const_iterator j = maps.begin();
 834        j != maps.end();
 835        ++j) {
 836     if (j->first == get_parent()->whoami_shard())
 837       continue;
 838     shards.push_back(j->first);
 839   }
 840   shards.push_front(get_parent()->whoami_shard());
 841
 842   map<pg_shard_t, ScrubMap *>::const_iterator auth = maps.end();
 843   digest_match = true;
 844   for (auto &l : shards) {
 845     ostringstream shard_errorstream;
 846     bool error = false;
 847     map<pg_shard_t, ScrubMap *>::const_iterator j = maps.find(l);
 848     map<hobject_t, ScrubMap::object>::iterator i =
 849       j->second->objects.find(obj);
 850     if (i == j->second->objects.end()) {
 851       continue;
 852     }
 853     auto& shard_info = shard_map[j->first];
 854     if (j->first == get_parent()->whoami_shard())
 855       shard_info.primary = true;
 856     if (i->second.read_error) {
 857       shard_info.set_read_error();
 858       if (error)
 859         shard_errorstream << ", ";
 860       error = true;
 861       shard_errorstream << "candidate had a read error";
 862     }
 863     if (i->second.ec_hash_mismatch) {
 864       shard_info.set_ec_hash_mismatch();
 865       if (error)
 866         shard_errorstream << ", ";
 867       error = true;
 868       shard_errorstream << "candidate had an ec hash mismatch";
 869     }
 870     if (i->second.ec_size_mismatch) {
 871       shard_info.set_ec_size_mismatch();
 872       if (error)
 873         shard_errorstream << ", ";
 874       error = true;
 875       shard_errorstream << "candidate had an ec size mismatch";
 876     }
 877
 878     object_info_t oi;
 879     bufferlist bl;
 880     map<string, bufferptr>::iterator k;
 881     SnapSet ss;
 882     bufferlist ss_bl, hk_bl;
 883
 884     if (i->second.stat_error) {
 885       shard_info.set_stat_error();
 886       if (error)
 887         shard_errorstream << ", ";
 888       error = true;
 889       shard_errorstream << "candidate had a stat error";
 890       // With stat_error no further checking
 891       // We don't need to also see a missing_object_info_attr
 892       goto out;
 893     }
 894
 895     // We won't pick an auth copy if the snapset is missing or won't decode.
 896     ceph_assert(!obj.is_snapdir());
 897     if (obj.is_head()) {
 898       k = i->second.attrs.find(SS_ATTR);
 899       if (k == i->second.attrs.end()) {
 900         shard_info.set_snapset_missing();
 901         if (error)
 902           shard_errorstream << ", ";
 903         error = true;
 904         shard_errorstream << "candidate had a missing snapset key";
 905       } else {
 906         ss_bl.push_back(k->second);
 907         try {
 908           auto bliter = ss_bl.cbegin();
 909           decode(ss, bliter);
 910         } catch (...) {
 911           // invalid snapset, probably corrupt
 912           shard_info.set_snapset_corrupted();
 913           if (error)
 914             shard_errorstream << ", ";
 915           error = true;
 916           shard_errorstream << "candidate had a corrupt snapset";
 917         }
 918       }
 919     }
 920
 921     if (parent->get_pool().is_erasure()) {
 922       ECUtil::HashInfo hi;
 923       k = i->second.attrs.find(ECUtil::get_hinfo_key());
 924       if (k == i->second.attrs.end()) {
 925         shard_info.set_hinfo_missing();
 926         if (error)
 927           shard_errorstream << ", ";
 928         error = true;
 929         shard_errorstream << "candidate had a missing hinfo key";
 930       } else {
 931         hk_bl.push_back(k->second);
 932         try {
 933           auto bliter = hk_bl.cbegin();
 934           decode(hi, bliter);
 935         } catch (...) {
 936           // invalid snapset, probably corrupt
 937           shard_info.set_hinfo_corrupted();
 938           if (error)
 939             shard_errorstream << ", ";
 940           error = true;
 941           shard_errorstream << "candidate had a corrupt hinfo";
 942         }
 943       }
 944     }
 945
 946     k = i->second.attrs.find(OI_ATTR);
 947     if (k == i->second.attrs.end()) {
 948       // no object info on object, probably corrupt
 949       shard_info.set_info_missing();
 950       if (error)
 951         shard_errorstream << ", ";
 952       error = true;
 953       shard_errorstream << "candidate had a missing info key";
 954       goto out;
 955     }
 956     bl.push_back(k->second);
 957     try {
 958       auto bliter = bl.cbegin();
 959       decode(oi, bliter);
 960     } catch (...) {
 961       // invalid object info, probably corrupt
 962       shard_info.set_info_corrupted();
 963       if (error)
 964         shard_errorstream << ", ";
 965       error = true;
 966       shard_errorstream << "candidate had a corrupt info";
 967       goto out;
 968     }
 969
 970     // This is automatically corrected in PG::_repair_oinfo_oid()
 971     ceph_assert(oi.soid == obj);
 972
 973     if (i->second.size != be_get_ondisk_size(oi.size)) {
 974       shard_info.set_obj_size_info_mismatch();
 975       if (error)
 976         shard_errorstream << ", ";
 977       error = true;
 978       shard_errorstream << "candidate size " << i->second.size << " info size "
 979                         << oi.size << " mismatch";
 980     }
 981
 982     // digest_match will only be true if computed digests are the same
 983     if (auth_version != eversion_t()
 984         && auth->second->objects[obj].digest_present
 985         && i->second.digest_present
 986         && auth->second->objects[obj].digest != i->second.digest) {
 987       digest_match = false;
 988       dout(10) << __func__ << " digest_match = false, " << obj << " data_digest 0x" << std::hex << i->second.digest
 989                     << " != data_digest 0x" << auth->second->objects[obj].digest << std::dec
 990                     << dendl;
 991     }
 992
 993     // Don't use this particular shard due to previous errors
 994     // XXX: For now we can't pick one shard for repair and another's object info or snapset
 995     if (shard_info.errors)
 996       goto out;
 997
 998     if (auth_version == eversion_t() || oi.version > auth_version ||
 999         (oi.version == auth_version && dcount(oi) > dcount(*auth_oi))) {
1000       auth = j;
1001       *auth_oi = oi;
1002       auth_version = oi.version;
1003     }
1004
1005 out:
1006     if (error)
1007         errorstream << pgid.pgid << " shard " << l << " soid " << obj
1008                     << " : " << shard_errorstream.str() << "\n";
1009     // Keep scanning other shards
1010   }
1011   dout(10) << __func__ << ": selecting osd " << auth->first
1012            << " for obj " << obj
1013            << " with oi " << *auth_oi
1014            << dendl;
1015   return auth;
1016 }
1017
1018 void PGBackend::be_compare_scrubmaps(
1019   const map<pg_shard_t,ScrubMap*> &maps,
1020   const set<hobject_t> &master_set,
1021   bool repair,
1022   map<hobject_t, set<pg_shard_t>> &missing,
1023   map<hobject_t, set<pg_shard_t>> &inconsistent,
1024   map<hobject_t, list<pg_shard_t>> &authoritative,
1025   map<hobject_t, pair<std::optional<uint32_t>,
1026                       std::optional<uint32_t>>> &missing_digest,
1027   int &shallow_errors, int &deep_errors,
1028   Scrub::Store *store,
1029   const spg_t& pgid,
1030   const vector<int> &acting,
1031   ostream &errorstream)
1032 {
1033   utime_t now = ceph_clock_now();
1034
1035   // Check maps against master set and each other
1036   for (set<hobject_t>::const_iterator k = master_set.begin();
1037        k != master_set.end();
1038        ++k) {
1039     object_info_t auth_oi;
1040     map<pg_shard_t, shard_info_wrapper> shard_map;
1041
1042     inconsistent_obj_wrapper object_error{*k};
1043
1044     bool digest_match;
1045     map<pg_shard_t, ScrubMap *>::const_iterator auth =
1046       be_select_auth_object(*k, maps, &auth_oi, shard_map, digest_match,
1047                             pgid, errorstream);
1048
1049     list<pg_shard_t> auth_list;
1050     set<pg_shard_t> object_errors;
1051     if (auth == maps.end()) {
1052       object_error.set_version(0);
1053       object_error.set_auth_missing(*k, maps, shard_map, shallow_errors,
1054         deep_errors, get_parent()->whoami_shard());
1055       if (object_error.has_deep_errors())
1056         ++deep_errors;
1057       else if (object_error.has_shallow_errors())
1058         ++shallow_errors;
1059       store->add_object_error(k->pool, object_error);
1060       errorstream << pgid.pgid << " soid " << *k
1061                   << " : failed to pick suitable object info\n";
1062       continue;
1063     }
1064     object_error.set_version(auth_oi.user_version);
1065     ScrubMap::object& auth_object = auth->second->objects[*k];
1066     set<pg_shard_t> cur_missing;
1067     set<pg_shard_t> cur_inconsistent;
1068     bool fix_digest = false;
1069
1070     for (auto j = maps.cbegin(); j != maps.cend(); ++j) {
1071       if (j == auth)
1072         shard_map[auth->first].selected_oi = true;
1073       if (j->second->objects.count(*k)) {
1074         shard_map[j->first].set_object(j->second->objects[*k]);
1075         // Compare
1076         stringstream ss;
1077         bool found = be_compare_scrub_objects(auth->first,
1078                                    auth_object,
1079                                    auth_oi,
1080                                    j->second->objects[*k],
1081                                    shard_map[j->first],
1082                                    object_error,
1083                                    ss,
1084                                    k->has_snapset());
1085
1086         dout(20) << __func__ << (repair ? " repair " : " ") << (parent->get_pool().is_replicated() ? "replicated " : "")
1087          << (j == auth ? "auth" : "") << "shards " << shard_map.size() << (digest_match ? " digest_match " : " ")
1088          << (shard_map[j->first].only_data_digest_mismatch_info() ? "'info mismatch info'" : "")
1089          << dendl;
1090         // If all replicas match, but they don't match object_info we can
1091         // repair it by using missing_digest mechanism
1092         if (repair && parent->get_pool().is_replicated() && j == auth && shard_map.size() > 1
1093             && digest_match && shard_map[j->first].only_data_digest_mismatch_info()
1094             && auth_object.digest_present) {
1095           // Set in missing_digests
1096           fix_digest = true;
1097           // Clear the error
1098           shard_map[j->first].clear_data_digest_mismatch_info();
1099           errorstream << pgid << " soid " << *k << " : repairing object info data_digest" << "\n";
1100         }
1101         // Some errors might have already been set in be_select_auth_object()
1102         if (shard_map[j->first].errors != 0) {
1103           cur_inconsistent.insert(j->first);
1104           if (shard_map[j->first].has_deep_errors())
1105             ++deep_errors;
1106           else
1107             ++shallow_errors;
1108           // Only true if be_compare_scrub_objects() found errors and put something
1109           // in ss.
1110           if (found)
1111             errorstream << pgid << " shard " << j->first << " soid " << *k
1112                       << " : " << ss.str() << "\n";
1113         } else if (found) {
1114           // Track possible shard to use as authoritative, if needed
1115           // There are errors, without identifying the shard
1116           object_errors.insert(j->first);
1117           errorstream << pgid << " soid " << *k << " : " << ss.str() << "\n";
1118         } else {
1119           // XXX: The auth shard might get here that we don't know
1120           // that it has the "correct" data.
1121           auth_list.push_back(j->first);
1122         }
1123       } else {
1124         cur_missing.insert(j->first);
1125         shard_map[j->first].set_missing();
1126         shard_map[j->first].primary = (j->first == get_parent()->whoami_shard());
1127         // Can't have any other errors if there is no information available
1128         ++shallow_errors;
1129         errorstream << pgid << " shard " << j->first << " " << *k << " : missing\n";
1130       }
1131       object_error.add_shard(j->first, shard_map[j->first]);
1132     }
1133
1134     if (auth_list.empty()) {
1135       if (object_errors.empty()) {
1136         errorstream << pgid.pgid << " soid " << *k
1137                   << " : failed to pick suitable auth object\n";
1138         goto out;
1139       }
1140       // Object errors exist and nothing in auth_list
1141       // Prefer the auth shard otherwise take first from list.
1142       pg_shard_t shard;
1143       if (object_errors.count(auth->first)) {
1144         shard = auth->first;
1145       } else {
1146         shard = *(object_errors.begin());
1147       }
1148       auth_list.push_back(shard);
1149       object_errors.erase(shard);
1150     }
1151     // At this point auth_list is populated, so we add the object errors shards
1152     // as inconsistent.
1153     cur_inconsistent.insert(object_errors.begin(), object_errors.end());
1154     if (!cur_missing.empty()) {
1155       missing[*k] = cur_missing;
1156     }
1157     if (!cur_inconsistent.empty()) {
1158       inconsistent[*k] = cur_inconsistent;
1159     }
1160
1161     if (fix_digest) {
1162       std::optional<uint32_t> data_digest, omap_digest;
1163       ceph_assert(auth_object.digest_present);
1164       data_digest = auth_object.digest;
1165       if (auth_object.omap_digest_present) {
1166         omap_digest = auth_object.omap_digest;
1167       }
1168       missing_digest[*k] = make_pair(data_digest, omap_digest);
1169     }
1170     if (!cur_inconsistent.empty() || !cur_missing.empty()) {
1171       authoritative[*k] = auth_list;
1172     } else if (!fix_digest && parent->get_pool().is_replicated()) {
1173       enum {
1174         NO = 0,
1175         MAYBE = 1,
1176         FORCE = 2,
1177       } update = NO;
1178
1179       if (auth_object.digest_present && !auth_oi.is_data_digest()) {
1180         dout(20) << __func__ << " missing data digest on " << *k << dendl;
1181         update = MAYBE;
1182       }
1183       if (auth_object.omap_digest_present && !auth_oi.is_omap_digest()) {
1184         dout(20) << __func__ << " missing omap digest on " << *k << dendl;
1185         update = MAYBE;
1186       }
1187
1188       // recorded digest != actual digest?
1189       if (auth_oi.is_data_digest() && auth_object.digest_present &&
1190           auth_oi.data_digest != auth_object.digest) {
1191         ceph_assert(shard_map[auth->first].has_data_digest_mismatch_info());
1192         errorstream << pgid << " recorded data digest 0x"
1193                     << std::hex << auth_oi.data_digest << " != on disk 0x"
1194                     << auth_object.digest << std::dec << " on " << auth_oi.soid
1195                     << "\n";
1196         if (repair)
1197           update = FORCE;
1198       }
1199       if (auth_oi.is_omap_digest() && auth_object.omap_digest_present &&
1200           auth_oi.omap_digest != auth_object.omap_digest) {
1201         ceph_assert(shard_map[auth->first].has_omap_digest_mismatch_info());
1202         errorstream << pgid << " recorded omap digest 0x"
1203                     << std::hex << auth_oi.omap_digest << " != on disk 0x"
1204                     << auth_object.omap_digest << std::dec
1205                     << " on " << auth_oi.soid << "\n";
1206         if (repair)
1207           update = FORCE;
1208       }
1209
1210       if (update != NO) {
1211         utime_t age = now - auth_oi.local_mtime;
1212         if (update == FORCE ||
1213             age > cct->_conf->osd_deep_scrub_update_digest_min_age) {
1214           std::optional<uint32_t> data_digest, omap_digest;
1215           if (auth_object.digest_present) {
1216             data_digest = auth_object.digest;
1217             dout(20) << __func__ << " will update data digest on " << *k << dendl;
1218           }
1219           if (auth_object.omap_digest_present) {
1220             omap_digest = auth_object.omap_digest;
1221             dout(20) << __func__ << " will update omap digest on " << *k << dendl;
1222           }
1223           missing_digest[*k] = make_pair(data_digest, omap_digest);
1224         } else {
1225           dout(20) << __func__ << " missing digest but age " << age
1226                    << " < " << cct->_conf->osd_deep_scrub_update_digest_min_age
1227                    << " on " << *k << dendl;
1228         }
1229       }
1230     }
1231 out:
1232     if (object_error.has_deep_errors())
1233       ++deep_errors;
1234     else if (object_error.has_shallow_errors())
1235       ++shallow_errors;
1236     if (object_error.errors || object_error.union_shards.errors) {
1237       store->add_object_error(k->pool, object_error);
1238     }
1239   }
1240 }
1241
1242 void PGBackend::be_omap_checks(const map<pg_shard_t,ScrubMap*> &maps,
1243   const set<hobject_t> &master_set,
1244   omap_stat_t& omap_stats,
1245   ostream &warnstream) const
1246 {
1247   bool needs_omap_check = false;
1248   for (const auto& map : maps) {
1249     if (map.second->has_large_omap_object_errors || map.second->has_omap_keys) {
1250       needs_omap_check = true;
1251       break;
1252     }
1253   }
1254
1255   if (!needs_omap_check) {
1256     return; // Nothing to do
1257   }
1258
1259   // Iterate through objects and update omap stats
1260   for (const auto& k : master_set) {
1261     for (const auto& map : maps) {
1262       if (map.first != get_parent()->primary_shard()) {
1263         // Only set omap stats for the primary
1264         continue;
1265       }
1266       auto it = map.second->objects.find(k);
1267       if (it == map.second->objects.end())
1268         continue;
1269       ScrubMap::object& obj = it->second;
1270       omap_stats.omap_bytes += obj.object_omap_bytes;
1271       omap_stats.omap_keys += obj.object_omap_keys;
1272       if (obj.large_omap_object_found) {
1273         pg_t pg;
1274         auto osdmap = get_osdmap();
1275         osdmap->map_to_pg(k.pool, k.oid.name, k.get_key(), k.nspace, &pg);
1276         pg_t mpg = osdmap->raw_pg_to_pg(pg);
1277         omap_stats.large_omap_objects++;
1278         warnstream << "Large omap object found. Object: " << k
1279                    << " PG: " << pg << " (" << mpg << ")"
1280                    << " Key count: " << obj.large_omap_object_key_count
1281                    << " Size (bytes): " << obj.large_omap_object_value_size
1282                    << '\n';
1283         break;
1284       }
1285     }
1286   }
1287 }