ceph/src/mds/journal.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) 2004-2006 Sage Weil <sage@newdream.net>
   7  *
   8  * This is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License version 2.1, as published by the Free Software
  11  * Foundation.  See file COPYING.
  12  *
  13  */
  14
  15 #include "common/config.h"
  16 #include "osdc/Journaler.h"
  17 #include "events/ESubtreeMap.h"
  18 #include "events/ESession.h"
  19 #include "events/ESessions.h"
  20
  21 #include "events/EMetaBlob.h"
  22 #include "events/EResetJournal.h"
  23 #include "events/ENoOp.h"
  24
  25 #include "events/EUpdate.h"
  26 #include "events/ESlaveUpdate.h"
  27 #include "events/EOpen.h"
  28 #include "events/ECommitted.h"
  29
  30 #include "events/EExport.h"
  31 #include "events/EImportStart.h"
  32 #include "events/EImportFinish.h"
  33 #include "events/EFragment.h"
  34
  35 #include "events/ETableClient.h"
  36 #include "events/ETableServer.h"
  37
  38 #include "include/stringify.h"
  39
  40 #include "LogSegment.h"
  41
  42 #include "MDSRank.h"
  43 #include "MDLog.h"
  44 #include "MDCache.h"
  45 #include "Server.h"
  46 #include "Migrator.h"
  47 #include "Mutation.h"
  48
  49 #include "InoTable.h"
  50 #include "MDSTableClient.h"
  51 #include "MDSTableServer.h"
  52
  53 #include "Locker.h"
  54
  55 #define dout_context g_ceph_context
  56 #define dout_subsys ceph_subsys_mds
  57 #undef dout_prefix
  58 #define dout_prefix *_dout << "mds." << mds->get_nodeid() << ".journal "
  59
  60
  61 // -----------------------
  62 // LogSegment
  63
  64 void LogSegment::try_to_expire(MDSRank *mds, MDSGatherBuilder &gather_bld, int op_prio)
  65 {
  66   set<CDir*> commit;
  67
  68   dout(6) << "LogSegment(" << seq << "/" << offset << ").try_to_expire" << dendl;
  69
  70   assert(g_conf->mds_kill_journal_expire_at != 1);
  71
  72   // commit dirs
  73   for (elist<CDir*>::iterator p = new_dirfrags.begin(); !p.end(); ++p) {
  74     dout(20) << " new_dirfrag " << **p << dendl;
  75     assert((*p)->is_auth());
  76     commit.insert(*p);
  77   }
  78   for (elist<CDir*>::iterator p = dirty_dirfrags.begin(); !p.end(); ++p) {
  79     dout(20) << " dirty_dirfrag " << **p << dendl;
  80     assert((*p)->is_auth());
  81     commit.insert(*p);
  82   }
  83   for (elist<CDentry*>::iterator p = dirty_dentries.begin(); !p.end(); ++p) {
  84     dout(20) << " dirty_dentry " << **p << dendl;
  85     assert((*p)->is_auth());
  86     commit.insert((*p)->get_dir());
  87   }
  88   for (elist<CInode*>::iterator p = dirty_inodes.begin(); !p.end(); ++p) {
  89     dout(20) << " dirty_inode " << **p << dendl;
  90     assert((*p)->is_auth());
  91     if ((*p)->is_base()) {
  92       (*p)->store(gather_bld.new_sub());
  93     } else
  94       commit.insert((*p)->get_parent_dn()->get_dir());
  95   }
  96
  97   if (!commit.empty()) {
  98     for (set<CDir*>::iterator p = commit.begin();
  99          p != commit.end();
 100          ++p) {
 101       CDir *dir = *p;
 102       assert(dir->is_auth());
 103       if (dir->can_auth_pin()) {
 104         dout(15) << "try_to_expire committing " << *dir << dendl;
 105         dir->commit(0, gather_bld.new_sub(), false, op_prio);
 106       } else {
 107         dout(15) << "try_to_expire waiting for unfreeze on " << *dir << dendl;
 108         dir->add_waiter(CDir::WAIT_UNFREEZE, gather_bld.new_sub());
 109       }
 110     }
 111   }
 112
 113   // master ops with possibly uncommitted slaves
 114   for (set<metareqid_t>::iterator p = uncommitted_masters.begin();
 115        p != uncommitted_masters.end();
 116        ++p) {
 117     dout(10) << "try_to_expire waiting for slaves to ack commit on " << *p << dendl;
 118     mds->mdcache->wait_for_uncommitted_master(*p, gather_bld.new_sub());
 119   }
 120
 121   // uncommitted fragments
 122   for (set<dirfrag_t>::iterator p = uncommitted_fragments.begin();
 123        p != uncommitted_fragments.end();
 124        ++p) {
 125     dout(10) << "try_to_expire waiting for uncommitted fragment " << *p << dendl;
 126     mds->mdcache->wait_for_uncommitted_fragment(*p, gather_bld.new_sub());
 127   }
 128
 129   // nudge scatterlocks
 130   for (elist<CInode*>::iterator p = dirty_dirfrag_dir.begin(); !p.end(); ++p) {
 131     CInode *in = *p;
 132     dout(10) << "try_to_expire waiting for dirlock flush on " << *in << dendl;
 133     mds->locker->scatter_nudge(&in->filelock, gather_bld.new_sub());
 134   }
 135   for (elist<CInode*>::iterator p = dirty_dirfrag_dirfragtree.begin(); !p.end(); ++p) {
 136     CInode *in = *p;
 137     dout(10) << "try_to_expire waiting for dirfragtreelock flush on " << *in << dendl;
 138     mds->locker->scatter_nudge(&in->dirfragtreelock, gather_bld.new_sub());
 139   }
 140   for (elist<CInode*>::iterator p = dirty_dirfrag_nest.begin(); !p.end(); ++p) {
 141     CInode *in = *p;
 142     dout(10) << "try_to_expire waiting for nest flush on " << *in << dendl;
 143     mds->locker->scatter_nudge(&in->nestlock, gather_bld.new_sub());
 144   }
 145
 146   assert(g_conf->mds_kill_journal_expire_at != 2);
 147
 148   // open files and snap inodes
 149   if (!open_files.empty()) {
 150     assert(!mds->mdlog->is_capped()); // hmm FIXME
 151     EOpen *le = 0;
 152     LogSegment *ls = mds->mdlog->get_current_segment();
 153     assert(ls != this);
 154     elist<CInode*>::iterator p = open_files.begin(member_offset(CInode, item_open_file));
 155     while (!p.end()) {
 156       CInode *in = *p;
 157       ++p;
 158       if (in->last == CEPH_NOSNAP && in->is_auth() &&
 159           !in->is_ambiguous_auth() && in->is_any_caps()) {
 160         if (in->is_any_caps_wanted()) {
 161           dout(20) << "try_to_expire requeueing open file " << *in << dendl;
 162           if (!le) {
 163             le = new EOpen(mds->mdlog);
 164             mds->mdlog->start_entry(le);
 165           }
 166           le->add_clean_inode(in);
 167           ls->open_files.push_back(&in->item_open_file);
 168         } else {
 169           // drop inodes that aren't wanted
 170           dout(20) << "try_to_expire not requeueing and delisting unwanted file " << *in << dendl;
 171           in->item_open_file.remove_myself();
 172         }
 173       } else if (in->last != CEPH_NOSNAP && !in->client_snap_caps.empty()) {
 174         // journal snap inodes that need flush. This simplify the mds failover hanlding
 175         dout(20) << "try_to_expire requeueing snap needflush inode " << *in << dendl;
 176         if (!le) {
 177           le = new EOpen(mds->mdlog);
 178           mds->mdlog->start_entry(le);
 179         }
 180         le->add_clean_inode(in);
 181         ls->open_files.push_back(&in->item_open_file);
 182       } else {
 183         /*
 184          * we can get a capless inode here if we replay an open file, the client fails to
 185          * reconnect it, but does REPLAY an open request (that adds it to the logseg).  AFAICS
 186          * it's ok for the client to replay an open on a file it doesn't have in it's cache
 187          * anymore.
 188          *
 189          * this makes the mds less sensitive to strict open_file consistency, although it does
 190          * make it easier to miss subtle problems.
 191          */
 192         dout(20) << "try_to_expire not requeueing and delisting capless file " << *in << dendl;
 193         in->item_open_file.remove_myself();
 194       }
 195     }
 196     if (le) {
 197       mds->mdlog->submit_entry(le);
 198       mds->mdlog->wait_for_safe(gather_bld.new_sub());
 199       dout(10) << "try_to_expire waiting for open files to rejournal" << dendl;
 200     }
 201   }
 202
 203   assert(g_conf->mds_kill_journal_expire_at != 3);
 204
 205   // backtraces to be stored/updated
 206   for (elist<CInode*>::iterator p = dirty_parent_inodes.begin(); !p.end(); ++p) {
 207     CInode *in = *p;
 208     assert(in->is_auth());
 209     if (in->can_auth_pin()) {
 210       dout(15) << "try_to_expire waiting for storing backtrace on " << *in << dendl;
 211       in->store_backtrace(gather_bld.new_sub(), op_prio);
 212     } else {
 213       dout(15) << "try_to_expire waiting for unfreeze on " << *in << dendl;
 214       in->add_waiter(CInode::WAIT_UNFREEZE, gather_bld.new_sub());
 215     }
 216   }
 217
 218   assert(g_conf->mds_kill_journal_expire_at != 4);
 219
 220   // slave updates
 221   for (elist<MDSlaveUpdate*>::iterator p = slave_updates.begin(member_offset(MDSlaveUpdate,
 222                                                                              item));
 223        !p.end(); ++p) {
 224     MDSlaveUpdate *su = *p;
 225     dout(10) << "try_to_expire waiting on slave update " << su << dendl;
 226     assert(su->waiter == 0);
 227     su->waiter = gather_bld.new_sub();
 228   }
 229
 230   // idalloc
 231   if (inotablev > mds->inotable->get_committed_version()) {
 232     dout(10) << "try_to_expire saving inotable table, need " << inotablev
 233               << ", committed is " << mds->inotable->get_committed_version()
 234               << " (" << mds->inotable->get_committing_version() << ")"
 235               << dendl;
 236     mds->inotable->save(gather_bld.new_sub(), inotablev);
 237   }
 238
 239   // sessionmap
 240   if (sessionmapv > mds->sessionmap.get_committed()) {
 241     dout(10) << "try_to_expire saving sessionmap, need " << sessionmapv
 242               << ", committed is " << mds->sessionmap.get_committed()
 243               << " (" << mds->sessionmap.get_committing() << ")"
 244               << dendl;
 245     mds->sessionmap.save(gather_bld.new_sub(), sessionmapv);
 246   }
 247
 248   // updates to sessions for completed_requests
 249   mds->sessionmap.save_if_dirty(touched_sessions, &gather_bld);
 250   touched_sessions.clear();
 251
 252   // pending commit atids
 253   for (map<int, ceph::unordered_set<version_t> >::iterator p = pending_commit_tids.begin();
 254        p != pending_commit_tids.end();
 255        ++p) {
 256     MDSTableClient *client = mds->get_table_client(p->first);
 257     assert(client);
 258     for (ceph::unordered_set<version_t>::iterator q = p->second.begin();
 259          q != p->second.end();
 260          ++q) {
 261       dout(10) << "try_to_expire " << get_mdstable_name(p->first) << " transaction " << *q
 262                << " pending commit (not yet acked), waiting" << dendl;
 263       assert(!client->has_committed(*q));
 264       client->wait_for_ack(*q, gather_bld.new_sub());
 265     }
 266   }
 267
 268   // table servers
 269   for (map<int, version_t>::iterator p = tablev.begin();
 270        p != tablev.end();
 271        ++p) {
 272     MDSTableServer *server = mds->get_table_server(p->first);
 273     assert(server);
 274     if (p->second > server->get_committed_version()) {
 275       dout(10) << "try_to_expire waiting for " << get_mdstable_name(p->first)
 276                << " to save, need " << p->second << dendl;
 277       server->save(gather_bld.new_sub());
 278     }
 279   }
 280
 281   // truncating
 282   for (set<CInode*>::iterator p = truncating_inodes.begin();
 283        p != truncating_inodes.end();
 284        ++p) {
 285     dout(10) << "try_to_expire waiting for truncate of " << **p << dendl;
 286     (*p)->add_waiter(CInode::WAIT_TRUNC, gather_bld.new_sub());
 287   }
 288
 289   if (gather_bld.has_subs()) {
 290     dout(6) << "LogSegment(" << seq << "/" << offset << ").try_to_expire waiting" << dendl;
 291     mds->mdlog->flush();
 292   } else {
 293     assert(g_conf->mds_kill_journal_expire_at != 5);
 294     dout(6) << "LogSegment(" << seq << "/" << offset << ").try_to_expire success" << dendl;
 295   }
 296 }
 297
 298
 299 // -----------------------
 300 // EMetaBlob
 301
 302 EMetaBlob::EMetaBlob(MDLog *mdlog) : opened_ino(0), renamed_dirino(0),
 303                                      inotablev(0), sessionmapv(0), allocated_ino(0),
 304                                      last_subtree_map(0), event_seq(0)
 305 { }
 306
 307 void EMetaBlob::add_dir_context(CDir *dir, int mode)
 308 {
 309   MDSRank *mds = dir->cache->mds;
 310
 311   list<CDentry*> parents;
 312
 313   // it may be okay not to include the maybe items, if
 314   //  - we journaled the maybe child inode in this segment
 315   //  - that subtree turns out to be unambiguously auth
 316   list<CDentry*> maybe;
 317   bool maybenot = false;
 318
 319   while (true) {
 320     // already have this dir?  (we must always add in order)
 321     if (lump_map.count(dir->dirfrag())) {
 322       dout(20) << "EMetaBlob::add_dir_context(" << dir << ") have lump " << dir->dirfrag() << dendl;
 323       break;
 324     }
 325
 326     // stop at root/stray
 327     CInode *diri = dir->get_inode();
 328     CDentry *parent = diri->get_projected_parent_dn();
 329
 330     if (mode == TO_AUTH_SUBTREE_ROOT) {
 331       // subtree root?
 332       if (dir->is_subtree_root() &&
 333           !dir->state_test(CDir::STATE_EXPORTBOUND)) {
 334         if (dir->is_auth() && !dir->is_ambiguous_auth() ) {
 335           if (dir->state_test(CDir::STATE_AUXSUBTREE) &&
 336               dir->get_dir_auth().first == diri->authority().first) {
 337             // auxiliary subtree. treat it as normal dirfrag
 338             dout(20) << "EMetaBlob::add_dir_context(" << dir << ") auxiliary subtree " << dendl;
 339           } else {
 340             // it's an auth subtree, we don't need maybe (if any), and we're done.
 341             dout(20) << "EMetaBlob::add_dir_context(" << dir << ") reached unambig auth subtree, don't need " << maybe
 342                      << " at " << *dir << dendl;
 343             maybe.clear();
 344             break;
 345           }
 346         } else {
 347           dout(20) << "EMetaBlob::add_dir_context(" << dir << ") reached ambig or !auth subtree, need " << maybe
 348                    << " at " << *dir << dendl;
 349           // we need the maybe list after all!
 350           parents.splice(parents.begin(), maybe);
 351           maybenot = false;
 352         }
 353       }
 354
 355       // was the inode journaled in this blob?
 356       if (event_seq && diri->last_journaled == event_seq) {
 357         dout(20) << "EMetaBlob::add_dir_context(" << dir << ") already have diri this blob " << *diri << dendl;
 358         break;
 359       }
 360
 361       // have we journaled this inode since the last subtree map?
 362       if (!maybenot && last_subtree_map && diri->last_journaled >= last_subtree_map) {
 363         dout(20) << "EMetaBlob::add_dir_context(" << dir << ") already have diri in this segment ("
 364                  << diri->last_journaled << " >= " << last_subtree_map << "), setting maybenot flag "
 365                  << *diri << dendl;
 366         maybenot = true;
 367       }
 368     }
 369
 370     if (!parent)
 371       break;
 372
 373     if (maybenot) {
 374       dout(25) << "EMetaBlob::add_dir_context(" << dir << ")      maybe " << *parent << dendl;
 375       maybe.push_front(parent);
 376     } else {
 377       dout(25) << "EMetaBlob::add_dir_context(" << dir << ") definitely " << *parent << dendl;
 378       parents.push_front(parent);
 379     }
 380
 381     dir = parent->get_dir();
 382   }
 383
 384   parents.splice(parents.begin(), maybe);
 385
 386   dout(20) << "EMetaBlob::add_dir_context final: " << parents << dendl;
 387   for (list<CDentry*>::iterator p = parents.begin(); p != parents.end(); ++p) {
 388     assert((*p)->get_projected_linkage()->is_primary());
 389     add_dentry(*p, false);
 390   }
 391 }
 392
 393 void EMetaBlob::update_segment(LogSegment *ls)
 394 {
 395   // dirty inode mtimes
 396   // -> handled directly by Server.cc, replay()
 397
 398   // alloc table update?
 399   if (inotablev)
 400     ls->inotablev = inotablev;
 401   if (sessionmapv)
 402     ls->sessionmapv = sessionmapv;
 403
 404   // truncated inodes
 405   // -> handled directly by Server.cc
 406
 407   // client requests
 408   //  note the newest request per client
 409   //if (!client_reqs.empty())
 410     //    ls->last_client_tid[client_reqs.rbegin()->client] = client_reqs.rbegin()->tid);
 411 }
 412
 413 // EMetaBlob::fullbit
 414
 415 void EMetaBlob::fullbit::encode(bufferlist& bl, uint64_t features) const {
 416   ENCODE_START(8, 5, bl);
 417   ::encode(dn, bl);
 418   ::encode(dnfirst, bl);
 419   ::encode(dnlast, bl);
 420   ::encode(dnv, bl);
 421   ::encode(inode, bl, features);
 422   ::encode(xattrs, bl);
 423   if (inode.is_symlink())
 424     ::encode(symlink, bl);
 425   if (inode.is_dir()) {
 426     ::encode(dirfragtree, bl);
 427     ::encode(snapbl, bl);
 428   }
 429   ::encode(state, bl);
 430   if (old_inodes.empty()) {
 431     ::encode(false, bl);
 432   } else {
 433     ::encode(true, bl);
 434     ::encode(old_inodes, bl, features);
 435   }
 436   if (!inode.is_dir())
 437     ::encode(snapbl, bl);
 438   ::encode(oldest_snap, bl);
 439   ENCODE_FINISH(bl);
 440 }
 441
 442 void EMetaBlob::fullbit::decode(bufferlist::iterator &bl) {
 443   DECODE_START_LEGACY_COMPAT_LEN(7, 5, 5, bl);
 444   ::decode(dn, bl);
 445   ::decode(dnfirst, bl);
 446   ::decode(dnlast, bl);
 447   ::decode(dnv, bl);
 448   ::decode(inode, bl);
 449   ::decode(xattrs, bl);
 450   if (inode.is_symlink())
 451     ::decode(symlink, bl);
 452   if (inode.is_dir()) {
 453     ::decode(dirfragtree, bl);
 454     ::decode(snapbl, bl);
 455     if ((struct_v == 2) || (struct_v == 3)) {
 456       bool dir_layout_exists;
 457       ::decode(dir_layout_exists, bl);
 458       if (dir_layout_exists) {
 459         __u8 dir_struct_v;
 460         ::decode(dir_struct_v, bl); // default_file_layout version
 461         ::decode(inode.layout, bl); // and actual layout, that we care about
 462       }
 463     }
 464   }
 465   if (struct_v >= 6) {
 466     ::decode(state, bl);
 467   } else {
 468     bool dirty;
 469     ::decode(dirty, bl);
 470     state = dirty ? EMetaBlob::fullbit::STATE_DIRTY : 0;
 471   }
 472
 473   if (struct_v >= 3) {
 474     bool old_inodes_present;
 475     ::decode(old_inodes_present, bl);
 476     if (old_inodes_present) {
 477       ::decode(old_inodes, bl);
 478     }
 479   }
 480   if (!inode.is_dir()) {
 481     if (struct_v >= 7)
 482       ::decode(snapbl, bl);
 483   }
 484   if (struct_v >= 8)
 485     ::decode(oldest_snap, bl);
 486   else
 487     oldest_snap = CEPH_NOSNAP;
 488
 489   DECODE_FINISH(bl);
 490 }
 491
 492 void EMetaBlob::fullbit::dump(Formatter *f) const
 493 {
 494   f->dump_string("dentry", dn);
 495   f->dump_stream("snapid.first") << dnfirst;
 496   f->dump_stream("snapid.last") << dnlast;
 497   f->dump_int("dentry version", dnv);
 498   f->open_object_section("inode");
 499   inode.dump(f);
 500   f->close_section(); // inode
 501   f->open_object_section("xattrs");
 502   for (map<string, bufferptr>::const_iterator iter = xattrs.begin();
 503       iter != xattrs.end(); ++iter) {
 504     string s(iter->second.c_str(), iter->second.length());
 505     f->dump_string(iter->first.c_str(), s);
 506   }
 507   f->close_section(); // xattrs
 508   if (inode.is_symlink()) {
 509     f->dump_string("symlink", symlink);
 510   }
 511   if (inode.is_dir()) {
 512     f->dump_stream("frag tree") << dirfragtree;
 513     f->dump_string("has_snapbl", snapbl.length() ? "true" : "false");
 514     if (inode.has_layout()) {
 515       f->open_object_section("file layout policy");
 516       // FIXME
 517       f->dump_string("layout", "the layout exists");
 518       f->close_section(); // file layout policy
 519     }
 520   }
 521   f->dump_string("state", state_string());
 522   if (!old_inodes.empty()) {
 523     f->open_array_section("old inodes");
 524     for (old_inodes_t::const_iterator iter = old_inodes.begin();
 525          iter != old_inodes.end();
 526          ++iter) {
 527       f->open_object_section("inode");
 528       f->dump_int("snapid", iter->first);
 529       iter->second.dump(f);
 530       f->close_section(); // inode
 531     }
 532     f->close_section(); // old inodes
 533   }
 534 }
 535
 536 void EMetaBlob::fullbit::generate_test_instances(list<EMetaBlob::fullbit*>& ls)
 537 {
 538   inode_t inode;
 539   fragtree_t fragtree;
 540   map<string,bufferptr> empty_xattrs;
 541   bufferlist empty_snapbl;
 542   fullbit *sample = new fullbit("/testdn", 0, 0, 0,
 543                                 inode, fragtree, empty_xattrs, "", 0, empty_snapbl,
 544                                 false, NULL);
 545   ls.push_back(sample);
 546 }
 547
 548 void EMetaBlob::fullbit::update_inode(MDSRank *mds, CInode *in)
 549 {
 550   in->inode = inode;
 551   in->xattrs = xattrs;
 552   if (in->inode.is_dir()) {
 553     if (!(in->dirfragtree == dirfragtree)) {
 554       dout(10) << "EMetaBlob::fullbit::update_inode dft " << in->dirfragtree << " -> "
 555                << dirfragtree << " on " << *in << dendl;
 556       in->dirfragtree = dirfragtree;
 557       in->force_dirfrags();
 558       if (in->has_dirfrags() && in->authority() == CDIR_AUTH_UNDEF) {
 559         list<CDir*> ls;
 560         in->get_nested_dirfrags(ls);
 561         for (list<CDir*>::iterator p = ls.begin(); p != ls.end(); ++p) {
 562           CDir *dir = *p;
 563           if (dir->get_num_any() == 0 &&
 564               mds->mdcache->can_trim_non_auth_dirfrag(dir)) {
 565             dout(10) << " closing empty non-auth dirfrag " << *dir << dendl;
 566             in->close_dirfrag(dir->get_frag());
 567           }
 568         }
 569       }
 570     }
 571   } else if (in->inode.is_symlink()) {
 572     in->symlink = symlink;
 573   }
 574   in->old_inodes = old_inodes;
 575   if (!in->old_inodes.empty()) {
 576     snapid_t min_first = in->old_inodes.rbegin()->first + 1;
 577     if (min_first > in->first)
 578       in->first = min_first;
 579   }
 580
 581   /*
 582    * we can do this before linking hte inode bc the split_at would
 583    * be a no-op.. we have no children (namely open snaprealms) to
 584    * divy up
 585    */
 586   in->oldest_snap = oldest_snap;
 587   in->decode_snap_blob(snapbl);
 588
 589   /*
 590    * In case there was anything malformed in the journal that we are
 591    * replaying, do sanity checks on the inodes we're replaying and
 592    * go damaged instead of letting any trash into a live cache
 593    */
 594   if (in->is_file()) {
 595     // Files must have valid layouts with a pool set
 596     if (in->inode.layout.pool_id == -1 || !in->inode.layout.is_valid()) {
 597       dout(0) << "EMetaBlob.replay invalid layout on ino " << *in
 598               << ": " << in->inode.layout << dendl;
 599       std::ostringstream oss;
 600       oss << "Invalid layout for inode 0x" << std::hex << in->inode.ino
 601           << std::dec << " in journal";
 602       mds->clog->error() << oss.str();
 603       mds->damaged();
 604       ceph_abort();  // Should be unreachable because damaged() calls respawn()
 605     }
 606   }
 607 }
 608
 609 // EMetaBlob::remotebit
 610
 611 void EMetaBlob::remotebit::encode(bufferlist& bl) const
 612 {
 613   ENCODE_START(2, 2, bl);
 614   ::encode(dn, bl);
 615   ::encode(dnfirst, bl);
 616   ::encode(dnlast, bl);
 617   ::encode(dnv, bl);
 618   ::encode(ino, bl);
 619   ::encode(d_type, bl);
 620   ::encode(dirty, bl);
 621   ENCODE_FINISH(bl);
 622 }
 623
 624 void EMetaBlob::remotebit::decode(bufferlist::iterator &bl)
 625 {
 626   DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
 627   ::decode(dn, bl);
 628   ::decode(dnfirst, bl);
 629   ::decode(dnlast, bl);
 630   ::decode(dnv, bl);
 631   ::decode(ino, bl);
 632   ::decode(d_type, bl);
 633   ::decode(dirty, bl);
 634   DECODE_FINISH(bl);
 635 }
 636
 637 void EMetaBlob::remotebit::dump(Formatter *f) const
 638 {
 639   f->dump_string("dentry", dn);
 640   f->dump_int("snapid.first", dnfirst);
 641   f->dump_int("snapid.last", dnlast);
 642   f->dump_int("dentry version", dnv);
 643   f->dump_int("inodeno", ino);
 644   uint32_t type = DTTOIF(d_type) & S_IFMT; // convert to type entries
 645   string type_string;
 646   switch(type) {
 647   case S_IFREG:
 648     type_string = "file"; break;
 649   case S_IFLNK:
 650     type_string = "symlink"; break;
 651   case S_IFDIR:
 652     type_string = "directory"; break;
 653   case S_IFIFO:
 654     type_string = "fifo"; break;
 655   case S_IFCHR:
 656     type_string = "chr"; break;
 657   case S_IFBLK:
 658     type_string = "blk"; break;
 659   case S_IFSOCK:
 660     type_string = "sock"; break;
 661   default:
 662     assert (0 == "unknown d_type!");
 663   }
 664   f->dump_string("d_type", type_string);
 665   f->dump_string("dirty", dirty ? "true" : "false");
 666 }
 667
 668 void EMetaBlob::remotebit::
 669 generate_test_instances(list<EMetaBlob::remotebit*>& ls)
 670 {
 671   remotebit *remote = new remotebit("/test/dn", 0, 10, 15, 1, IFTODT(S_IFREG), false);
 672   ls.push_back(remote);
 673 }
 674
 675 // EMetaBlob::nullbit
 676
 677 void EMetaBlob::nullbit::encode(bufferlist& bl) const
 678 {
 679   ENCODE_START(2, 2, bl);
 680   ::encode(dn, bl);
 681   ::encode(dnfirst, bl);
 682   ::encode(dnlast, bl);
 683   ::encode(dnv, bl);
 684   ::encode(dirty, bl);
 685   ENCODE_FINISH(bl);
 686 }
 687
 688 void EMetaBlob::nullbit::decode(bufferlist::iterator &bl)
 689 {
 690   DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
 691   ::decode(dn, bl);
 692   ::decode(dnfirst, bl);
 693   ::decode(dnlast, bl);
 694   ::decode(dnv, bl);
 695   ::decode(dirty, bl);
 696   DECODE_FINISH(bl);
 697 }
 698
 699 void EMetaBlob::nullbit::dump(Formatter *f) const
 700 {
 701   f->dump_string("dentry", dn);
 702   f->dump_int("snapid.first", dnfirst);
 703   f->dump_int("snapid.last", dnlast);
 704   f->dump_int("dentry version", dnv);
 705   f->dump_string("dirty", dirty ? "true" : "false");
 706 }
 707
 708 void EMetaBlob::nullbit::generate_test_instances(list<nullbit*>& ls)
 709 {
 710   nullbit *sample = new nullbit("/test/dentry", 0, 10, 15, false);
 711   nullbit *sample2 = new nullbit("/test/dirty", 10, 20, 25, true);
 712   ls.push_back(sample);
 713   ls.push_back(sample2);
 714 }
 715
 716 // EMetaBlob::dirlump
 717
 718 void EMetaBlob::dirlump::encode(bufferlist& bl, uint64_t features) const
 719 {
 720   ENCODE_START(2, 2, bl);
 721   ::encode(fnode, bl);
 722   ::encode(state, bl);
 723   ::encode(nfull, bl);
 724   ::encode(nremote, bl);
 725   ::encode(nnull, bl);
 726   _encode_bits(features);
 727   ::encode(dnbl, bl);
 728   ENCODE_FINISH(bl);
 729 }
 730
 731 void EMetaBlob::dirlump::decode(bufferlist::iterator &bl)
 732 {
 733   DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl)
 734   ::decode(fnode, bl);
 735   ::decode(state, bl);
 736   ::decode(nfull, bl);
 737   ::decode(nremote, bl);
 738   ::decode(nnull, bl);
 739   ::decode(dnbl, bl);
 740   dn_decoded = false;      // don't decode bits unless we need them.
 741   DECODE_FINISH(bl);
 742 }
 743
 744 void EMetaBlob::dirlump::dump(Formatter *f) const
 745 {
 746   if (!dn_decoded) {
 747     dirlump *me = const_cast<dirlump*>(this);
 748     me->_decode_bits();
 749   }
 750   f->open_object_section("fnode");
 751   fnode.dump(f);
 752   f->close_section(); // fnode
 753   f->dump_string("state", state_string());
 754   f->dump_int("nfull", nfull);
 755   f->dump_int("nremote", nremote);
 756   f->dump_int("nnull", nnull);
 757
 758   f->open_array_section("full bits");
 759   for (list<ceph::shared_ptr<fullbit> >::const_iterator
 760       iter = dfull.begin(); iter != dfull.end(); ++iter) {
 761     f->open_object_section("fullbit");
 762     (*iter)->dump(f);
 763     f->close_section(); // fullbit
 764   }
 765   f->close_section(); // full bits
 766   f->open_array_section("remote bits");
 767   for (list<remotebit>::const_iterator
 768       iter = dremote.begin(); iter != dremote.end(); ++iter) {
 769     f->open_object_section("remotebit");
 770     (*iter).dump(f);
 771     f->close_section(); // remotebit
 772   }
 773   f->close_section(); // remote bits
 774   f->open_array_section("null bits");
 775   for (list<nullbit>::const_iterator
 776       iter = dnull.begin(); iter != dnull.end(); ++iter) {
 777     f->open_object_section("null bit");
 778     (*iter).dump(f);
 779     f->close_section(); // null bit
 780   }
 781   f->close_section(); // null bits
 782 }
 783
 784 void EMetaBlob::dirlump::generate_test_instances(list<dirlump*>& ls)
 785 {
 786   ls.push_back(new dirlump());
 787 }
 788
 789 /**
 790  * EMetaBlob proper
 791  */
 792 void EMetaBlob::encode(bufferlist& bl, uint64_t features) const
 793 {
 794   ENCODE_START(8, 5, bl);
 795   ::encode(lump_order, bl);
 796   ::encode(lump_map, bl, features);
 797   ::encode(roots, bl, features);
 798   ::encode(table_tids, bl);
 799   ::encode(opened_ino, bl);
 800   ::encode(allocated_ino, bl);
 801   ::encode(used_preallocated_ino, bl);
 802   ::encode(preallocated_inos, bl);
 803   ::encode(client_name, bl);
 804   ::encode(inotablev, bl);
 805   ::encode(sessionmapv, bl);
 806   ::encode(truncate_start, bl);
 807   ::encode(truncate_finish, bl);
 808   ::encode(destroyed_inodes, bl);
 809   ::encode(client_reqs, bl);
 810   ::encode(renamed_dirino, bl);
 811   ::encode(renamed_dir_frags, bl);
 812   {
 813     // make MDSRank use v6 format happy
 814     int64_t i = -1;
 815     bool b = false;
 816     ::encode(i, bl);
 817     ::encode(b, bl);
 818   }
 819   ::encode(client_flushes, bl);
 820   ENCODE_FINISH(bl);
 821 }
 822 void EMetaBlob::decode(bufferlist::iterator &bl)
 823 {
 824   DECODE_START_LEGACY_COMPAT_LEN(7, 5, 5, bl);
 825   ::decode(lump_order, bl);
 826   ::decode(lump_map, bl);
 827   if (struct_v >= 4) {
 828     ::decode(roots, bl);
 829   } else {
 830     bufferlist rootbl;
 831     ::decode(rootbl, bl);
 832     if (rootbl.length()) {
 833       bufferlist::iterator p = rootbl.begin();
 834       roots.push_back(ceph::shared_ptr<fullbit>(new fullbit(p)));
 835     }
 836   }
 837   ::decode(table_tids, bl);
 838   ::decode(opened_ino, bl);
 839   ::decode(allocated_ino, bl);
 840   ::decode(used_preallocated_ino, bl);
 841   ::decode(preallocated_inos, bl);
 842   ::decode(client_name, bl);
 843   ::decode(inotablev, bl);
 844   ::decode(sessionmapv, bl);
 845   ::decode(truncate_start, bl);
 846   ::decode(truncate_finish, bl);
 847   ::decode(destroyed_inodes, bl);
 848   if (struct_v >= 2) {
 849     ::decode(client_reqs, bl);
 850   } else {
 851     list<metareqid_t> r;
 852     ::decode(r, bl);
 853     while (!r.empty()) {
 854         client_reqs.push_back(pair<metareqid_t,uint64_t>(r.front(), 0));
 855         r.pop_front();
 856     }
 857   }
 858   if (struct_v >= 3) {
 859     ::decode(renamed_dirino, bl);
 860     ::decode(renamed_dir_frags, bl);
 861   }
 862   if (struct_v >= 6) {
 863     // ignore
 864     int64_t i;
 865     bool b;
 866     ::decode(i, bl);
 867     ::decode(b, bl);
 868   }
 869   if (struct_v >= 8) {
 870     ::decode(client_flushes, bl);
 871   }
 872   DECODE_FINISH(bl);
 873 }
 874
 875
 876 /**
 877  * Get all inodes touched by this metablob.  Includes the 'bits' within
 878  * dirlumps, and the inodes of the dirs themselves.
 879  */
 880 void EMetaBlob::get_inodes(
 881     std::set<inodeno_t> &inodes) const
 882 {
 883   // For all dirlumps in this metablob
 884   for (std::map<dirfrag_t, dirlump>::const_iterator i = lump_map.begin(); i != lump_map.end(); ++i) {
 885     // Record inode of dirlump
 886     inodeno_t const dir_ino = i->first.ino;
 887     inodes.insert(dir_ino);
 888
 889     // Decode dirlump bits
 890     dirlump const &dl = i->second;
 891     dl._decode_bits();
 892
 893     // Record inodes of fullbits
 894     list<ceph::shared_ptr<fullbit> > const &fb_list = dl.get_dfull();
 895     for (list<ceph::shared_ptr<fullbit> >::const_iterator
 896         iter = fb_list.begin(); iter != fb_list.end(); ++iter) {
 897       inodes.insert((*iter)->inode.ino);
 898     }
 899
 900     // Record inodes of remotebits
 901     list<remotebit> const &rb_list = dl.get_dremote();
 902     for (list<remotebit>::const_iterator
 903         iter = rb_list.begin(); iter != rb_list.end(); ++iter) {
 904       inodes.insert(iter->ino);
 905     }
 906   }
 907 }
 908
 909
 910 /**
 911  * Get a map of dirfrag to set of dentries in that dirfrag which are
 912  * touched in this operation.
 913  */
 914 void EMetaBlob::get_dentries(std::map<dirfrag_t, std::set<std::string> > &dentries) const
 915 {
 916   for (std::map<dirfrag_t, dirlump>::const_iterator i = lump_map.begin(); i != lump_map.end(); ++i) {
 917     dirlump const &dl = i->second;
 918     dirfrag_t const &df = i->first;
 919
 920     // Get all bits
 921     dl._decode_bits();
 922     list<ceph::shared_ptr<fullbit> > const &fb_list = dl.get_dfull();
 923     list<nullbit> const &nb_list = dl.get_dnull();
 924     list<remotebit> const &rb_list = dl.get_dremote();
 925
 926     // For all bits, store dentry
 927     for (list<ceph::shared_ptr<fullbit> >::const_iterator
 928         iter = fb_list.begin(); iter != fb_list.end(); ++iter) {
 929       dentries[df].insert((*iter)->dn);
 930
 931     }
 932     for (list<nullbit>::const_iterator
 933         iter = nb_list.begin(); iter != nb_list.end(); ++iter) {
 934       dentries[df].insert(iter->dn);
 935     }
 936     for (list<remotebit>::const_iterator
 937         iter = rb_list.begin(); iter != rb_list.end(); ++iter) {
 938       dentries[df].insert(iter->dn);
 939     }
 940   }
 941 }
 942
 943
 944
 945 /**
 946  * Calculate all paths that we can infer are touched by this metablob.  Only uses
 947  * information local to this metablob so it may only be the path within the
 948  * subtree.
 949  */
 950 void EMetaBlob::get_paths(
 951     std::vector<std::string> &paths) const
 952 {
 953   // Each dentry has a 'location' which is a 2-tuple of parent inode and dentry name
 954   typedef std::pair<inodeno_t, std::string> Location;
 955
 956   // Whenever we see a dentry within a dirlump, we remember it as a child of
 957   // the dirlump's inode
 958   std::map<inodeno_t, std::list<std::string> > children;
 959
 960   // Whenever we see a location for an inode, remember it: this allows us to
 961   // build a path given an inode
 962   std::map<inodeno_t, Location> ino_locations;
 963
 964   // Special case: operations on root inode populate roots but not dirlumps
 965   if (lump_map.empty() && !roots.empty()) {
 966     paths.push_back("/");
 967     return;
 968   }
 969
 970   // First pass
 971   // ==========
 972   // Build a tiny local metadata cache for the path structure in this metablob
 973   for (std::map<dirfrag_t, dirlump>::const_iterator i = lump_map.begin(); i != lump_map.end(); ++i) {
 974     inodeno_t const dir_ino = i->first.ino;
 975     dirlump const &dl = i->second;
 976     dl._decode_bits();
 977
 978     list<ceph::shared_ptr<fullbit> > const &fb_list = dl.get_dfull();
 979     list<nullbit> const &nb_list = dl.get_dnull();
 980     list<remotebit> const &rb_list = dl.get_dremote();
 981
 982     for (list<ceph::shared_ptr<fullbit> >::const_iterator
 983         iter = fb_list.begin(); iter != fb_list.end(); ++iter) {
 984       std::string const &dentry = (*iter)->dn;
 985       children[dir_ino].push_back(dentry);
 986       ino_locations[(*iter)->inode.ino] = Location(dir_ino, dentry);
 987     }
 988
 989     for (list<nullbit>::const_iterator
 990         iter = nb_list.begin(); iter != nb_list.end(); ++iter) {
 991       std::string const &dentry = iter->dn;
 992       children[dir_ino].push_back(dentry);
 993     }
 994
 995     for (list<remotebit>::const_iterator
 996         iter = rb_list.begin(); iter != rb_list.end(); ++iter) {
 997       std::string const &dentry = iter->dn;
 998       children[dir_ino].push_back(dentry);
 999     }
1000   }
1001
1002   std::vector<Location> leaf_locations;
1003
1004   // Second pass
1005   // ===========
1006   // Output paths for all childless nodes in the metablob
1007   for (std::map<dirfrag_t, dirlump>::const_iterator i = lump_map.begin(); i != lump_map.end(); ++i) {
1008     inodeno_t const dir_ino = i->first.ino;
1009     dirlump const &dl = i->second;
1010     dl._decode_bits();
1011
1012     list<ceph::shared_ptr<fullbit> > const &fb_list = dl.get_dfull();
1013     for (list<ceph::shared_ptr<fullbit> >::const_iterator
1014         iter = fb_list.begin(); iter != fb_list.end(); ++iter) {
1015       std::string const &dentry = (*iter)->dn;
1016       children[dir_ino].push_back(dentry);
1017       ino_locations[(*iter)->inode.ino] = Location(dir_ino, dentry);
1018       if (children.find((*iter)->inode.ino) == children.end()) {
1019         leaf_locations.push_back(Location(dir_ino, dentry));
1020
1021       }
1022     }
1023
1024     list<nullbit> const &nb_list = dl.get_dnull();
1025     for (list<nullbit>::const_iterator
1026         iter = nb_list.begin(); iter != nb_list.end(); ++iter) {
1027       std::string const &dentry = iter->dn;
1028       leaf_locations.push_back(Location(dir_ino, dentry));
1029     }
1030
1031     list<remotebit> const &rb_list = dl.get_dremote();
1032     for (list<remotebit>::const_iterator
1033         iter = rb_list.begin(); iter != rb_list.end(); ++iter) {
1034       std::string const &dentry = iter->dn;
1035       leaf_locations.push_back(Location(dir_ino, dentry));
1036     }
1037   }
1038
1039   // For all the leaf locations identified, generate paths
1040   for (std::vector<Location>::iterator i = leaf_locations.begin(); i != leaf_locations.end(); ++i) {
1041     Location const &loc = *i;
1042     std::string path = loc.second;
1043     inodeno_t ino = loc.first;
1044     while(ino_locations.find(ino) != ino_locations.end()) {
1045       Location const &loc = ino_locations[ino];
1046       if (!path.empty()) {
1047         path = loc.second + "/" + path;
1048       } else {
1049         path = loc.second + path;
1050       }
1051       ino = loc.first;
1052     }
1053
1054     paths.push_back(path);
1055   }
1056 }
1057
1058
1059 void EMetaBlob::dump(Formatter *f) const
1060 {
1061   f->open_array_section("lumps");
1062   for (list<dirfrag_t>::const_iterator i = lump_order.begin();
1063        i != lump_order.end(); ++i) {
1064     f->open_object_section("lump");
1065     f->open_object_section("dirfrag");
1066     f->dump_stream("dirfrag") << *i;
1067     f->close_section(); // dirfrag
1068     f->open_object_section("dirlump");
1069     lump_map.at(*i).dump(f);
1070     f->close_section(); // dirlump
1071     f->close_section(); // lump
1072   }
1073   f->close_section(); // lumps
1074
1075   f->open_array_section("roots");
1076   for (list<ceph::shared_ptr<fullbit> >::const_iterator i = roots.begin();
1077        i != roots.end(); ++i) {
1078     f->open_object_section("root");
1079     (*i)->dump(f);
1080     f->close_section(); // root
1081   }
1082   f->close_section(); // roots
1083
1084   f->open_array_section("tableclient tranactions");
1085   for (list<pair<__u8,version_t> >::const_iterator i = table_tids.begin();
1086        i != table_tids.end(); ++i) {
1087     f->open_object_section("transaction");
1088     f->dump_int("tid", i->first);
1089     f->dump_int("version", i->second);
1090     f->close_section(); // transaction
1091   }
1092   f->close_section(); // tableclient transactions
1093
1094   f->dump_int("renamed directory inodeno", renamed_dirino);
1095
1096   f->open_array_section("renamed directory fragments");
1097   for (list<frag_t>::const_iterator i = renamed_dir_frags.begin();
1098        i != renamed_dir_frags.end(); ++i) {
1099     f->dump_int("frag", *i);
1100   }
1101   f->close_section(); // renamed directory fragments
1102
1103   f->dump_int("inotable version", inotablev);
1104   f->dump_int("SessionMap version", sessionmapv);
1105   f->dump_int("allocated ino", allocated_ino);
1106
1107   f->dump_stream("preallocated inos") << preallocated_inos;
1108   f->dump_int("used preallocated ino", used_preallocated_ino);
1109
1110   f->open_object_section("client name");
1111   client_name.dump(f);
1112   f->close_section(); // client name
1113
1114   f->open_array_section("inodes starting a truncate");
1115   for(list<inodeno_t>::const_iterator i = truncate_start.begin();
1116       i != truncate_start.end(); ++i) {
1117     f->dump_int("inodeno", *i);
1118   }
1119   f->close_section(); // truncate inodes
1120   f->open_array_section("inodes finishing a truncated");
1121   for(map<inodeno_t,uint64_t>::const_iterator i = truncate_finish.begin();
1122       i != truncate_finish.end(); ++i) {
1123     f->open_object_section("inode+segment");
1124     f->dump_int("inodeno", i->first);
1125     f->dump_int("truncate starting segment", i->second);
1126     f->close_section(); // truncated inode
1127   }
1128   f->close_section(); // truncate finish inodes
1129
1130   f->open_array_section("destroyed inodes");
1131   for(vector<inodeno_t>::const_iterator i = destroyed_inodes.begin();
1132       i != destroyed_inodes.end(); ++i) {
1133     f->dump_int("inodeno", *i);
1134   }
1135   f->close_section(); // destroyed inodes
1136
1137   f->open_array_section("client requests");
1138   for(list<pair<metareqid_t,uint64_t> >::const_iterator i = client_reqs.begin();
1139       i != client_reqs.end(); ++i) {
1140     f->open_object_section("Client request");
1141     f->dump_stream("request ID") << i->first;
1142     f->dump_int("oldest request on client", i->second);
1143     f->close_section(); // request
1144   }
1145   f->close_section(); // client requests
1146 }
1147
1148 void EMetaBlob::generate_test_instances(list<EMetaBlob*>& ls)
1149 {
1150   ls.push_back(new EMetaBlob());
1151 }
1152
1153 void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
1154 {
1155   dout(10) << "EMetaBlob.replay " << lump_map.size() << " dirlumps by " << client_name << dendl;
1156
1157   assert(logseg);
1158
1159   assert(g_conf->mds_kill_journal_replay_at != 1);
1160
1161   for (list<ceph::shared_ptr<fullbit> >::iterator p = roots.begin(); p != roots.end(); ++p) {
1162     CInode *in = mds->mdcache->get_inode((*p)->inode.ino);
1163     bool isnew = in ? false:true;
1164     if (!in)
1165       in = new CInode(mds->mdcache, false);
1166     (*p)->update_inode(mds, in);
1167
1168     if (isnew)
1169       mds->mdcache->add_inode(in);
1170     if ((*p)->is_dirty()) in->_mark_dirty(logseg);
1171     dout(10) << "EMetaBlob.replay " << (isnew ? " added root ":" updated root ") << *in << dendl;
1172   }
1173
1174   CInode *renamed_diri = 0;
1175   CDir *olddir = 0;
1176   if (renamed_dirino) {
1177     renamed_diri = mds->mdcache->get_inode(renamed_dirino);
1178     if (renamed_diri)
1179       dout(10) << "EMetaBlob.replay renamed inode is " << *renamed_diri << dendl;
1180     else
1181       dout(10) << "EMetaBlob.replay don't have renamed ino " << renamed_dirino << dendl;
1182
1183     int nnull = 0;
1184     for (list<dirfrag_t>::iterator lp = lump_order.begin(); lp != lump_order.end(); ++lp) {
1185       dirlump &lump = lump_map[*lp];
1186       if (lump.nnull) {
1187         dout(10) << "EMetaBlob.replay found null dentry in dir " << *lp << dendl;
1188         nnull += lump.nnull;
1189       }
1190     }
1191     assert(nnull <= 1);
1192   }
1193
1194   // keep track of any inodes we unlink and don't relink elsewhere
1195   map<CInode*, CDir*> unlinked;
1196   set<CInode*> linked;
1197
1198   // walk through my dirs (in order!)
1199   for (list<dirfrag_t>::iterator lp = lump_order.begin();
1200        lp != lump_order.end();
1201        ++lp) {
1202     dout(10) << "EMetaBlob.replay dir " << *lp << dendl;
1203     dirlump &lump = lump_map[*lp];
1204
1205     // the dir
1206     CDir *dir = mds->mdcache->get_force_dirfrag(*lp, true);
1207     if (!dir) {
1208       // hmm.  do i have the inode?
1209       CInode *diri = mds->mdcache->get_inode((*lp).ino);
1210       if (!diri) {
1211         if (MDS_INO_IS_MDSDIR(lp->ino)) {
1212           assert(MDS_INO_MDSDIR(mds->get_nodeid()) != lp->ino);
1213           diri = mds->mdcache->create_system_inode(lp->ino, S_IFDIR|0755);
1214           diri->state_clear(CInode::STATE_AUTH);
1215           dout(10) << "EMetaBlob.replay created base " << *diri << dendl;
1216         } else {
1217           dout(0) << "EMetaBlob.replay missing dir ino  " << (*lp).ino << dendl;
1218           mds->clog->error() << "failure replaying journal (EMetaBlob)";
1219           mds->damaged();
1220           ceph_abort();  // Should be unreachable because damaged() calls respawn()
1221         }
1222       }
1223
1224       // create the dirfrag
1225       dir = diri->get_or_open_dirfrag(mds->mdcache, (*lp).frag);
1226
1227       if (MDS_INO_IS_BASE(lp->ino))
1228         mds->mdcache->adjust_subtree_auth(dir, CDIR_AUTH_UNDEF);
1229
1230       dout(10) << "EMetaBlob.replay added dir " << *dir << dendl;
1231     }
1232     dir->set_version( lump.fnode.version );
1233     dir->fnode = lump.fnode;
1234
1235     if (lump.is_importing()) {
1236       dir->state_set(CDir::STATE_AUTH);
1237       dir->state_clear(CDir::STATE_COMPLETE);
1238     }
1239     if (lump.is_dirty()) {
1240       dir->_mark_dirty(logseg);
1241
1242       if (!(dir->fnode.rstat == dir->fnode.accounted_rstat)) {
1243         dout(10) << "EMetaBlob.replay      dirty nestinfo on " << *dir << dendl;
1244         mds->locker->mark_updated_scatterlock(&dir->inode->nestlock);
1245         logseg->dirty_dirfrag_nest.push_back(&dir->inode->item_dirty_dirfrag_nest);
1246       } else {
1247         dout(10) << "EMetaBlob.replay      clean nestinfo on " << *dir << dendl;
1248       }
1249       if (!(dir->fnode.fragstat == dir->fnode.accounted_fragstat)) {
1250         dout(10) << "EMetaBlob.replay      dirty fragstat on " << *dir << dendl;
1251         mds->locker->mark_updated_scatterlock(&dir->inode->filelock);
1252         logseg->dirty_dirfrag_dir.push_back(&dir->inode->item_dirty_dirfrag_dir);
1253       } else {
1254         dout(10) << "EMetaBlob.replay      clean fragstat on " << *dir << dendl;
1255       }
1256     }
1257     if (lump.is_dirty_dft()) {
1258       dout(10) << "EMetaBlob.replay      dirty dirfragtree on " << *dir << dendl;
1259       dir->state_set(CDir::STATE_DIRTYDFT);
1260       mds->locker->mark_updated_scatterlock(&dir->inode->dirfragtreelock);
1261       logseg->dirty_dirfrag_dirfragtree.push_back(&dir->inode->item_dirty_dirfrag_dirfragtree);
1262     }
1263     if (lump.is_new())
1264       dir->mark_new(logseg);
1265     if (lump.is_complete())
1266       dir->mark_complete();
1267
1268     dout(10) << "EMetaBlob.replay updated dir " << *dir << dendl;
1269
1270     // decode bits
1271     lump._decode_bits();
1272
1273     // full dentry+inode pairs
1274     for (list<ceph::shared_ptr<fullbit> >::const_iterator pp = lump.get_dfull().begin();
1275          pp != lump.get_dfull().end();
1276          ++pp) {
1277       ceph::shared_ptr<fullbit> p = *pp;
1278       CDentry *dn = dir->lookup_exact_snap(p->dn, p->dnlast);
1279       if (!dn) {
1280         dn = dir->add_null_dentry(p->dn, p->dnfirst, p->dnlast);
1281         dn->set_version(p->dnv);
1282         if (p->is_dirty()) dn->_mark_dirty(logseg);
1283         dout(10) << "EMetaBlob.replay added (full) " << *dn << dendl;
1284       } else {
1285         dn->set_version(p->dnv);
1286         if (p->is_dirty()) dn->_mark_dirty(logseg);
1287         dout(10) << "EMetaBlob.replay for [" << p->dnfirst << "," << p->dnlast << "] had " << *dn << dendl;
1288         dn->first = p->dnfirst;
1289         assert(dn->last == p->dnlast);
1290       }
1291       if (lump.is_importing())
1292         dn->state_set(CDentry::STATE_AUTH);
1293
1294       CInode *in = mds->mdcache->get_inode(p->inode.ino, p->dnlast);
1295       if (!in) {
1296         in = new CInode(mds->mdcache, dn->is_auth(), p->dnfirst, p->dnlast);
1297         p->update_inode(mds, in);
1298         mds->mdcache->add_inode(in);
1299         if (!dn->get_linkage()->is_null()) {
1300           if (dn->get_linkage()->is_primary()) {
1301             unlinked[dn->get_linkage()->get_inode()] = dir;
1302             stringstream ss;
1303             ss << "EMetaBlob.replay FIXME had dentry linked to wrong inode " << *dn
1304                << " " << *dn->get_linkage()->get_inode() << " should be " << p->inode.ino;
1305             dout(0) << ss.str() << dendl;
1306             mds->clog->warn(ss);
1307           }
1308           dir->unlink_inode(dn);
1309           mds->mdcache->touch_dentry_bottom(dn);
1310         }
1311         if (unlinked.count(in))
1312           linked.insert(in);
1313         dir->link_primary_inode(dn, in);
1314         dout(10) << "EMetaBlob.replay added " << *in << dendl;
1315       } else {
1316         in->first = p->dnfirst;
1317         p->update_inode(mds, in);
1318         if (dn->get_linkage()->get_inode() != in && in->get_parent_dn()) {
1319           dout(10) << "EMetaBlob.replay unlinking " << *in << dendl;
1320           unlinked[in] = in->get_parent_dir();
1321           CDentry *unlinked_dn = in->get_parent_dn();
1322           in->get_parent_dir()->unlink_inode(in->get_parent_dn());
1323           mds->mdcache->touch_dentry_bottom(unlinked_dn);
1324         }
1325         if (dn->get_linkage()->get_inode() != in) {
1326           if (!dn->get_linkage()->is_null()) { // note: might be remote.  as with stray reintegration.
1327             if (dn->get_linkage()->is_primary()) {
1328               unlinked[dn->get_linkage()->get_inode()] = dir;
1329               stringstream ss;
1330               ss << "EMetaBlob.replay FIXME had dentry linked to wrong inode " << *dn
1331                  << " " << *dn->get_linkage()->get_inode() << " should be " << p->inode.ino;
1332               dout(0) << ss.str() << dendl;
1333               mds->clog->warn(ss);
1334             }
1335             dir->unlink_inode(dn);
1336             mds->mdcache->touch_dentry_bottom(dn);
1337           }
1338           if (unlinked.count(in))
1339             linked.insert(in);
1340           dir->link_primary_inode(dn, in);
1341           dout(10) << "EMetaBlob.replay linked " << *in << dendl;
1342         } else {
1343           dout(10) << "EMetaBlob.replay for [" << p->dnfirst << "," << p->dnlast << "] had " << *in << dendl;
1344         }
1345         assert(in->first == p->dnfirst ||
1346                (in->is_multiversion() && in->first > p->dnfirst));
1347       }
1348       if (p->is_dirty())
1349         in->_mark_dirty(logseg);
1350       if (p->is_dirty_parent())
1351         in->_mark_dirty_parent(logseg, p->is_dirty_pool());
1352       if (p->need_snapflush())
1353         logseg->open_files.push_back(&in->item_open_file);
1354       if (dn->is_auth())
1355         in->state_set(CInode::STATE_AUTH);
1356       else
1357         in->state_clear(CInode::STATE_AUTH);
1358       assert(g_conf->mds_kill_journal_replay_at != 2);
1359     }
1360
1361     // remote dentries
1362     for (list<remotebit>::const_iterator p = lump.get_dremote().begin();
1363          p != lump.get_dremote().end();
1364          ++p) {
1365       CDentry *dn = dir->lookup_exact_snap(p->dn, p->dnlast);
1366       if (!dn) {
1367         dn = dir->add_remote_dentry(p->dn, p->ino, p->d_type, p->dnfirst, p->dnlast);
1368         dn->set_version(p->dnv);
1369         if (p->dirty) dn->_mark_dirty(logseg);
1370         dout(10) << "EMetaBlob.replay added " << *dn << dendl;
1371       } else {
1372         if (!dn->get_linkage()->is_null()) {
1373           dout(10) << "EMetaBlob.replay unlinking " << *dn << dendl;
1374           if (dn->get_linkage()->is_primary()) {
1375             unlinked[dn->get_linkage()->get_inode()] = dir;
1376             stringstream ss;
1377             ss << "EMetaBlob.replay FIXME had dentry linked to wrong inode " << *dn
1378                << " " << *dn->get_linkage()->get_inode() << " should be remote " << p->ino;
1379             dout(0) << ss.str() << dendl;
1380           }
1381           dir->unlink_inode(dn);
1382           mds->mdcache->touch_dentry_bottom(dn);
1383         }
1384         dir->link_remote_inode(dn, p->ino, p->d_type);
1385         dn->set_version(p->dnv);
1386         if (p->dirty) dn->_mark_dirty(logseg);
1387         dout(10) << "EMetaBlob.replay for [" << p->dnfirst << "," << p->dnlast << "] had " << *dn << dendl;
1388         dn->first = p->dnfirst;
1389         assert(dn->last == p->dnlast);
1390       }
1391       if (lump.is_importing())
1392         dn->state_set(CDentry::STATE_AUTH);
1393     }
1394
1395     // null dentries
1396     for (list<nullbit>::const_iterator p = lump.get_dnull().begin();
1397          p != lump.get_dnull().end();
1398          ++p) {
1399       CDentry *dn = dir->lookup_exact_snap(p->dn, p->dnlast);
1400       if (!dn) {
1401         dn = dir->add_null_dentry(p->dn, p->dnfirst, p->dnlast);
1402         dn->set_version(p->dnv);
1403         if (p->dirty) dn->_mark_dirty(logseg);
1404         dout(10) << "EMetaBlob.replay added (nullbit) " << *dn << dendl;
1405       } else {
1406         dn->first = p->dnfirst;
1407         if (!dn->get_linkage()->is_null()) {
1408           dout(10) << "EMetaBlob.replay unlinking " << *dn << dendl;
1409           CInode *in = dn->get_linkage()->get_inode();
1410           // For renamed inode, We may call CInode::force_dirfrag() later.
1411           // CInode::force_dirfrag() doesn't work well when inode is detached
1412           // from the hierarchy.
1413           if (!renamed_diri || renamed_diri != in) {
1414             if (dn->get_linkage()->is_primary())
1415               unlinked[in] = dir;
1416             dir->unlink_inode(dn);
1417             mds->mdcache->touch_dentry_bottom(dn);
1418           }
1419         }
1420         dn->set_version(p->dnv);
1421         if (p->dirty) dn->_mark_dirty(logseg);
1422         dout(10) << "EMetaBlob.replay had " << *dn << dendl;
1423         assert(dn->last == p->dnlast);
1424       }
1425       olddir = dir;
1426       if (lump.is_importing())
1427         dn->state_set(CDentry::STATE_AUTH);
1428
1429       // Make null dentries the first things we trim
1430       dout(10) << "EMetaBlob.replay pushing to bottom of lru " << *dn << dendl;
1431       mds->mdcache->touch_dentry_bottom(dn);
1432     }
1433   }
1434
1435   assert(g_conf->mds_kill_journal_replay_at != 3);
1436
1437   if (renamed_dirino) {
1438     if (renamed_diri) {
1439       assert(unlinked.count(renamed_diri));
1440       assert(linked.count(renamed_diri));
1441       olddir = unlinked[renamed_diri];
1442     } else {
1443       // we imported a diri we haven't seen before
1444       renamed_diri = mds->mdcache->get_inode(renamed_dirino);
1445       assert(renamed_diri);  // it was in the metablob
1446     }
1447
1448     if (olddir) {
1449       if (olddir->authority() != CDIR_AUTH_UNDEF &&
1450           renamed_diri->authority() == CDIR_AUTH_UNDEF) {
1451         assert(slaveup); // auth to non-auth, must be slave prepare
1452         list<frag_t> leaves;
1453         renamed_diri->dirfragtree.get_leaves(leaves);
1454         for (list<frag_t>::iterator p = leaves.begin(); p != leaves.end(); ++p) {
1455           CDir *dir = renamed_diri->get_dirfrag(*p);
1456           assert(dir);
1457           if (dir->get_dir_auth() == CDIR_AUTH_UNDEF)
1458             // preserve subtree bound until slave commit
1459             slaveup->olddirs.insert(dir->inode);
1460           else
1461             dir->state_set(CDir::STATE_AUTH);
1462         }
1463       }
1464
1465       mds->mdcache->adjust_subtree_after_rename(renamed_diri, olddir, false);
1466
1467       // see if we can discard the subtree we renamed out of
1468       CDir *root = mds->mdcache->get_subtree_root(olddir);
1469       if (root->get_dir_auth() == CDIR_AUTH_UNDEF) {
1470         if (slaveup) // preserve the old dir until slave commit
1471           slaveup->olddirs.insert(olddir->inode);
1472         else
1473           mds->mdcache->try_trim_non_auth_subtree(root);
1474       }
1475     }
1476
1477     // if we are the srci importer, we'll also have some dirfrags we have to open up...
1478     if (renamed_diri->authority() != CDIR_AUTH_UNDEF) {
1479       for (list<frag_t>::iterator p = renamed_dir_frags.begin(); p != renamed_dir_frags.end(); ++p) {
1480         CDir *dir = renamed_diri->get_dirfrag(*p);
1481         if (dir) {
1482           // we already had the inode before, and we already adjusted this subtree accordingly.
1483           dout(10) << " already had+adjusted rename import bound " << *dir << dendl;
1484           assert(olddir);
1485           continue;
1486         }
1487         dir = renamed_diri->get_or_open_dirfrag(mds->mdcache, *p);
1488         dout(10) << " creating new rename import bound " << *dir << dendl;
1489         dir->state_clear(CDir::STATE_AUTH);
1490         mds->mdcache->adjust_subtree_auth(dir, CDIR_AUTH_UNDEF, false);
1491       }
1492     }
1493
1494     // rename may overwrite an empty directory and move it into stray dir.
1495     unlinked.erase(renamed_diri);
1496     for (map<CInode*, CDir*>::iterator p = unlinked.begin(); p != unlinked.end(); ++p) {
1497       if (!linked.count(p->first))
1498         continue;
1499       assert(p->first->is_dir());
1500       mds->mdcache->adjust_subtree_after_rename(p->first, p->second, false);
1501     }
1502   }
1503
1504   if (!unlinked.empty()) {
1505     for (set<CInode*>::iterator p = linked.begin(); p != linked.end(); ++p)
1506       unlinked.erase(*p);
1507     dout(10) << " unlinked set contains " << unlinked << dendl;
1508     for (map<CInode*, CDir*>::iterator p = unlinked.begin(); p != unlinked.end(); ++p) {
1509       if (slaveup) // preserve unlinked inodes until slave commit
1510         slaveup->unlinked.insert(p->first);
1511       else
1512         mds->mdcache->remove_inode_recursive(p->first);
1513     }
1514   }
1515
1516   // table client transactions
1517   for (list<pair<__u8,version_t> >::iterator p = table_tids.begin();
1518        p != table_tids.end();
1519        ++p) {
1520     dout(10) << "EMetaBlob.replay noting " << get_mdstable_name(p->first)
1521              << " transaction " << p->second << dendl;
1522     MDSTableClient *client = mds->get_table_client(p->first);
1523     if (client)
1524       client->got_journaled_agree(p->second, logseg);
1525   }
1526
1527   // opened ino?
1528   if (opened_ino) {
1529     CInode *in = mds->mdcache->get_inode(opened_ino);
1530     assert(in);
1531     dout(10) << "EMetaBlob.replay noting opened inode " << *in << dendl;
1532     logseg->open_files.push_back(&in->item_open_file);
1533   }
1534
1535   // allocated_inos
1536   if (inotablev) {
1537     if (mds->inotable->get_version() >= inotablev) {
1538       dout(10) << "EMetaBlob.replay inotable tablev " << inotablev
1539                << " <= table " << mds->inotable->get_version() << dendl;
1540     } else {
1541       dout(10) << "EMetaBlob.replay inotable v " << inotablev
1542                << " - 1 == table " << mds->inotable->get_version()
1543                << " allocated+used " << allocated_ino
1544                << " prealloc " << preallocated_inos
1545                << dendl;
1546       if (allocated_ino)
1547         mds->inotable->replay_alloc_id(allocated_ino);
1548       if (preallocated_inos.size())
1549         mds->inotable->replay_alloc_ids(preallocated_inos);
1550
1551       // [repair bad inotable updates]
1552       if (inotablev > mds->inotable->get_version()) {
1553         mds->clog->error() << "journal replay inotablev mismatch "
1554             << mds->inotable->get_version() << " -> " << inotablev;
1555         mds->inotable->force_replay_version(inotablev);
1556       }
1557
1558       assert(inotablev == mds->inotable->get_version());
1559     }
1560   }
1561   if (sessionmapv) {
1562     if (mds->sessionmap.get_version() >= sessionmapv) {
1563       dout(10) << "EMetaBlob.replay sessionmap v " << sessionmapv
1564                << " <= table " << mds->sessionmap.get_version() << dendl;
1565     } else if (mds->sessionmap.get_version() + 2 >= sessionmapv) {
1566       dout(10) << "EMetaBlob.replay sessionmap v " << sessionmapv
1567                << " -(1|2) == table " << mds->sessionmap.get_version()
1568                << " prealloc " << preallocated_inos
1569                << " used " << used_preallocated_ino
1570                << dendl;
1571       Session *session = mds->sessionmap.get_session(client_name);
1572       if (session) {
1573         dout(20) << " (session prealloc " << session->info.prealloc_inos << ")" << dendl;
1574         if (used_preallocated_ino) {
1575           if (!session->info.prealloc_inos.empty()) {
1576             inodeno_t next = session->next_ino();
1577             inodeno_t i = session->take_ino(used_preallocated_ino);
1578             if (next != i)
1579               mds->clog->warn() << " replayed op " << client_reqs << " used ino " << i
1580                                << " but session next is " << next;
1581             assert(i == used_preallocated_ino);
1582             session->info.used_inos.clear();
1583           }
1584           mds->sessionmap.replay_dirty_session(session);
1585         }
1586         if (!preallocated_inos.empty()) {
1587           session->info.prealloc_inos.insert(preallocated_inos);
1588           mds->sessionmap.replay_dirty_session(session);
1589         }
1590
1591       } else {
1592         dout(10) << "EMetaBlob.replay no session for " << client_name << dendl;
1593         if (used_preallocated_ino) {
1594           mds->sessionmap.replay_advance_version();
1595         }
1596         if (!preallocated_inos.empty())
1597           mds->sessionmap.replay_advance_version();
1598       }
1599       assert(sessionmapv == mds->sessionmap.get_version());
1600     } else {
1601       mds->clog->error() << "journal replay sessionmap v " << sessionmapv
1602                         << " -(1|2) > table " << mds->sessionmap.get_version();
1603       assert(g_conf->mds_wipe_sessions);
1604       mds->sessionmap.wipe();
1605       mds->sessionmap.set_version(sessionmapv);
1606     }
1607   }
1608
1609   // truncating inodes
1610   for (list<inodeno_t>::iterator p = truncate_start.begin();
1611        p != truncate_start.end();
1612        ++p) {
1613     CInode *in = mds->mdcache->get_inode(*p);
1614     assert(in);
1615     mds->mdcache->add_recovered_truncate(in, logseg);
1616   }
1617   for (map<inodeno_t,uint64_t>::iterator p = truncate_finish.begin();
1618        p != truncate_finish.end();
1619        ++p) {
1620     LogSegment *ls = mds->mdlog->get_segment(p->second);
1621     if (ls) {
1622       CInode *in = mds->mdcache->get_inode(p->first);
1623       assert(in);
1624       mds->mdcache->remove_recovered_truncate(in, ls);
1625     }
1626   }
1627
1628   // destroyed inodes
1629   for (vector<inodeno_t>::iterator p = destroyed_inodes.begin();
1630        p != destroyed_inodes.end();
1631        ++p) {
1632     CInode *in = mds->mdcache->get_inode(*p);
1633     if (in) {
1634       dout(10) << "EMetaBlob.replay destroyed " << *p << ", dropping " << *in << dendl;
1635       CDentry *parent = in->get_parent_dn();
1636       mds->mdcache->remove_inode(in);
1637       if (parent) {
1638         dout(10) << "EMetaBlob.replay unlinked from dentry " << *parent << dendl;
1639         assert(parent->get_linkage()->is_null());
1640         mds->mdcache->touch_dentry_bottom(parent);
1641       }
1642     } else {
1643       dout(10) << "EMetaBlob.replay destroyed " << *p << ", not in cache" << dendl;
1644     }
1645   }
1646
1647   // client requests
1648   for (list<pair<metareqid_t, uint64_t> >::iterator p = client_reqs.begin();
1649        p != client_reqs.end();
1650        ++p) {
1651     if (p->first.name.is_client()) {
1652       dout(10) << "EMetaBlob.replay request " << p->first << " trim_to " << p->second << dendl;
1653       inodeno_t created = allocated_ino ? allocated_ino : used_preallocated_ino;
1654       // if we allocated an inode, there should be exactly one client request id.
1655       assert(created == inodeno_t() || client_reqs.size() == 1);
1656
1657       Session *session = mds->sessionmap.get_session(p->first.name);
1658       if (session) {
1659         session->add_completed_request(p->first.tid, created);
1660         if (p->second)
1661           session->trim_completed_requests(p->second);
1662       }
1663     }
1664   }
1665
1666   // client flushes
1667   for (list<pair<metareqid_t, uint64_t> >::iterator p = client_flushes.begin();
1668        p != client_flushes.end();
1669        ++p) {
1670     if (p->first.name.is_client()) {
1671       dout(10) << "EMetaBlob.replay flush " << p->first << " trim_to " << p->second << dendl;
1672       Session *session = mds->sessionmap.get_session(p->first.name);
1673       if (session) {
1674         session->add_completed_flush(p->first.tid);
1675         if (p->second)
1676           session->trim_completed_flushes(p->second);
1677       }
1678     }
1679   }
1680
1681   // update segment
1682   update_segment(logseg);
1683
1684   assert(g_conf->mds_kill_journal_replay_at != 4);
1685 }
1686
1687 // -----------------------
1688 // ESession
1689
1690 void ESession::update_segment()
1691 {
1692   _segment->sessionmapv = cmapv;
1693   if (inos.size() && inotablev)
1694     _segment->inotablev = inotablev;
1695 }
1696
1697 void ESession::replay(MDSRank *mds)
1698 {
1699   if (mds->sessionmap.get_version() >= cmapv) {
1700     dout(10) << "ESession.replay sessionmap " << mds->sessionmap.get_version()
1701              << " >= " << cmapv << ", noop" << dendl;
1702   } else {
1703     dout(10) << "ESession.replay sessionmap " << mds->sessionmap.get_version()
1704              << " < " << cmapv << " " << (open ? "open":"close") << " " << client_inst << dendl;
1705     Session *session;
1706     if (open) {
1707       session = mds->sessionmap.get_or_add_session(client_inst);
1708       mds->sessionmap.set_state(session, Session::STATE_OPEN);
1709       session->set_client_metadata(client_metadata);
1710       dout(10) << " opened session " << session->info.inst << dendl;
1711     } else {
1712       session = mds->sessionmap.get_session(client_inst.name);
1713       if (session) { // there always should be a session, but there's a bug
1714         if (session->connection == NULL) {
1715           dout(10) << " removed session " << session->info.inst << dendl;
1716           mds->sessionmap.remove_session(session);
1717           session = NULL;
1718         } else {
1719           session->clear();    // the client has reconnected; keep the Session, but reset
1720           dout(10) << " reset session " << session->info.inst << " (they reconnected)" << dendl;
1721         }
1722       } else {
1723         mds->clog->error() << "replayed stray Session close event for " << client_inst
1724                           << " from time " << stamp << ", ignoring";
1725       }
1726     }
1727     if (session) {
1728       mds->sessionmap.replay_dirty_session(session);
1729     } else {
1730       mds->sessionmap.replay_advance_version();
1731     }
1732     assert(mds->sessionmap.get_version() == cmapv);
1733   }
1734
1735   if (inos.size() && inotablev) {
1736     if (mds->inotable->get_version() >= inotablev) {
1737       dout(10) << "ESession.replay inotable " << mds->inotable->get_version()
1738                << " >= " << inotablev << ", noop" << dendl;
1739     } else {
1740       dout(10) << "ESession.replay inotable " << mds->inotable->get_version()
1741                << " < " << inotablev << " " << (open ? "add":"remove") << dendl;
1742       assert(!open);  // for now
1743       mds->inotable->replay_release_ids(inos);
1744       assert(mds->inotable->get_version() == inotablev);
1745     }
1746   }
1747
1748   update_segment();
1749 }
1750
1751 void ESession::encode(bufferlist &bl, uint64_t features) const
1752 {
1753   ENCODE_START(4, 3, bl);
1754   ::encode(stamp, bl);
1755   ::encode(client_inst, bl, features);
1756   ::encode(open, bl);
1757   ::encode(cmapv, bl);
1758   ::encode(inos, bl);
1759   ::encode(inotablev, bl);
1760   ::encode(client_metadata, bl);
1761   ENCODE_FINISH(bl);
1762 }
1763
1764 void ESession::decode(bufferlist::iterator &bl)
1765 {
1766   DECODE_START_LEGACY_COMPAT_LEN(4, 3, 3, bl);
1767   if (struct_v >= 2)
1768     ::decode(stamp, bl);
1769   ::decode(client_inst, bl);
1770   ::decode(open, bl);
1771   ::decode(cmapv, bl);
1772   ::decode(inos, bl);
1773   ::decode(inotablev, bl);
1774   if (struct_v >= 4) {
1775     ::decode(client_metadata, bl);
1776   }
1777   DECODE_FINISH(bl);
1778 }
1779
1780 void ESession::dump(Formatter *f) const
1781 {
1782   f->dump_stream("client instance") << client_inst;
1783   f->dump_string("open", open ? "true" : "false");
1784   f->dump_int("client map version", cmapv);
1785   f->dump_stream("inos") << inos;
1786   f->dump_int("inotable version", inotablev);
1787   f->open_object_section("client_metadata");
1788   for (map<string, string>::const_iterator i = client_metadata.begin();
1789       i != client_metadata.end(); ++i) {
1790     f->dump_string(i->first.c_str(), i->second);
1791   }
1792   f->close_section();  // client_metadata
1793 }
1794
1795 void ESession::generate_test_instances(list<ESession*>& ls)
1796 {
1797   ls.push_back(new ESession);
1798 }
1799
1800 // -----------------------
1801 // ESessions
1802
1803 void ESessions::encode(bufferlist &bl, uint64_t features) const
1804 {
1805   ENCODE_START(1, 1, bl);
1806   ::encode(client_map, bl, features);
1807   ::encode(cmapv, bl);
1808   ::encode(stamp, bl);
1809   ENCODE_FINISH(bl);
1810 }
1811
1812 void ESessions::decode_old(bufferlist::iterator &bl)
1813 {
1814   ::decode(client_map, bl);
1815   ::decode(cmapv, bl);
1816   if (!bl.end())
1817     ::decode(stamp, bl);
1818 }
1819
1820 void ESessions::decode_new(bufferlist::iterator &bl)
1821 {
1822   DECODE_START(1, bl);
1823   ::decode(client_map, bl);
1824   ::decode(cmapv, bl);
1825   if (!bl.end())
1826     ::decode(stamp, bl);
1827   DECODE_FINISH(bl);
1828 }
1829
1830 void ESessions::dump(Formatter *f) const
1831 {
1832   f->dump_int("client map version", cmapv);
1833
1834   f->open_array_section("client map");
1835   for (map<client_t,entity_inst_t>::const_iterator i = client_map.begin();
1836        i != client_map.end(); ++i) {
1837     f->open_object_section("client");
1838     f->dump_int("client id", i->first.v);
1839     f->dump_stream("client entity") << i->second;
1840     f->close_section(); // client
1841   }
1842   f->close_section(); // client map
1843 }
1844
1845 void ESessions::generate_test_instances(list<ESessions*>& ls)
1846 {
1847   ls.push_back(new ESessions());
1848 }
1849
1850 void ESessions::update_segment()
1851 {
1852   _segment->sessionmapv = cmapv;
1853 }
1854
1855 void ESessions::replay(MDSRank *mds)
1856 {
1857   if (mds->sessionmap.get_version() >= cmapv) {
1858     dout(10) << "ESessions.replay sessionmap " << mds->sessionmap.get_version()
1859              << " >= " << cmapv << ", noop" << dendl;
1860   } else {
1861     dout(10) << "ESessions.replay sessionmap " << mds->sessionmap.get_version()
1862              << " < " << cmapv << dendl;
1863     mds->sessionmap.open_sessions(client_map);
1864     assert(mds->sessionmap.get_version() == cmapv);
1865     mds->sessionmap.set_projected(mds->sessionmap.get_version());
1866   }
1867   update_segment();
1868 }
1869
1870
1871 // -----------------------
1872 // ETableServer
1873
1874 void ETableServer::encode(bufferlist& bl, uint64_t features) const
1875 {
1876   ENCODE_START(3, 3, bl);
1877   ::encode(stamp, bl);
1878   ::encode(table, bl);
1879   ::encode(op, bl);
1880   ::encode(reqid, bl);
1881   ::encode(bymds, bl);
1882   ::encode(mutation, bl);
1883   ::encode(tid, bl);
1884   ::encode(version, bl);
1885   ENCODE_FINISH(bl);
1886 }
1887
1888 void ETableServer::decode(bufferlist::iterator &bl)
1889 {
1890   DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
1891   if (struct_v >= 2)
1892     ::decode(stamp, bl);
1893   ::decode(table, bl);
1894   ::decode(op, bl);
1895   ::decode(reqid, bl);
1896   ::decode(bymds, bl);
1897   ::decode(mutation, bl);
1898   ::decode(tid, bl);
1899   ::decode(version, bl);
1900   DECODE_FINISH(bl);
1901 }
1902
1903 void ETableServer::dump(Formatter *f) const
1904 {
1905   f->dump_int("table id", table);
1906   f->dump_int("op", op);
1907   f->dump_int("request id", reqid);
1908   f->dump_int("by mds", bymds);
1909   f->dump_int("tid", tid);
1910   f->dump_int("version", version);
1911 }
1912
1913 void ETableServer::generate_test_instances(list<ETableServer*>& ls)
1914 {
1915   ls.push_back(new ETableServer());
1916 }
1917
1918
1919 void ETableServer::update_segment()
1920 {
1921   _segment->tablev[table] = version;
1922 }
1923
1924 void ETableServer::replay(MDSRank *mds)
1925 {
1926   MDSTableServer *server = mds->get_table_server(table);
1927   if (!server)
1928     return;
1929
1930   if (server->get_version() >= version) {
1931     dout(10) << "ETableServer.replay " << get_mdstable_name(table)
1932              << " " << get_mdstableserver_opname(op)
1933              << " event " << version
1934              << " <= table " << server->get_version() << dendl;
1935     return;
1936   }
1937
1938   dout(10) << " ETableServer.replay " << get_mdstable_name(table)
1939            << " " << get_mdstableserver_opname(op)
1940            << " event " << version << " - 1 == table " << server->get_version() << dendl;
1941   assert(version-1 == server->get_version());
1942
1943   switch (op) {
1944   case TABLESERVER_OP_PREPARE:
1945     server->_prepare(mutation, reqid, bymds);
1946     server->_note_prepare(bymds, reqid);
1947     break;
1948   case TABLESERVER_OP_COMMIT:
1949     server->_commit(tid);
1950     server->_note_commit(tid);
1951     break;
1952   case TABLESERVER_OP_ROLLBACK:
1953     server->_rollback(tid);
1954     server->_note_rollback(tid);
1955     break;
1956   case TABLESERVER_OP_SERVER_UPDATE:
1957     server->_server_update(mutation);
1958     break;
1959   default:
1960     mds->clog->error() << "invalid tableserver op in ETableServer";
1961     mds->damaged();
1962     ceph_abort();  // Should be unreachable because damaged() calls respawn()
1963   }
1964
1965   assert(version == server->get_version());
1966   update_segment();
1967 }
1968
1969
1970 // ---------------------
1971 // ETableClient
1972
1973 void ETableClient::encode(bufferlist& bl, uint64_t features) const
1974 {
1975   ENCODE_START(3, 3, bl);
1976   ::encode(stamp, bl);
1977   ::encode(table, bl);
1978   ::encode(op, bl);
1979   ::encode(tid, bl);
1980   ENCODE_FINISH(bl);
1981 }
1982
1983 void ETableClient::decode(bufferlist::iterator &bl)
1984 {
1985   DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
1986   if (struct_v >= 2)
1987     ::decode(stamp, bl);
1988   ::decode(table, bl);
1989   ::decode(op, bl);
1990   ::decode(tid, bl);
1991   DECODE_FINISH(bl);
1992 }
1993
1994 void ETableClient::dump(Formatter *f) const
1995 {
1996   f->dump_int("table", table);
1997   f->dump_int("op", op);
1998   f->dump_int("tid", tid);
1999 }
2000
2001 void ETableClient::generate_test_instances(list<ETableClient*>& ls)
2002 {
2003   ls.push_back(new ETableClient());
2004 }
2005
2006 void ETableClient::replay(MDSRank *mds)
2007 {
2008   dout(10) << " ETableClient.replay " << get_mdstable_name(table)
2009            << " op " << get_mdstableserver_opname(op)
2010            << " tid " << tid << dendl;
2011
2012   MDSTableClient *client = mds->get_table_client(table);
2013   if (!client)
2014     return;
2015
2016   assert(op == TABLESERVER_OP_ACK);
2017   client->got_journaled_ack(tid);
2018 }
2019
2020
2021 // -----------------------
2022 // ESnap
2023 /*
2024 void ESnap::update_segment()
2025 {
2026   _segment->tablev[TABLE_SNAP] = version;
2027 }
2028
2029 void ESnap::replay(MDSRank *mds)
2030 {
2031   if (mds->snaptable->get_version() >= version) {
2032     dout(10) << "ESnap.replay event " << version
2033              << " <= table " << mds->snaptable->get_version() << dendl;
2034     return;
2035   }
2036
2037   dout(10) << " ESnap.replay event " << version
2038            << " - 1 == table " << mds->snaptable->get_version() << dendl;
2039   assert(version-1 == mds->snaptable->get_version());
2040
2041   if (create) {
2042     version_t v;
2043     snapid_t s = mds->snaptable->create(snap.dirino, snap.name, snap.stamp, &v);
2044     assert(s == snap.snapid);
2045   } else {
2046     mds->snaptable->remove(snap.snapid);
2047   }
2048
2049   assert(version == mds->snaptable->get_version());
2050 }
2051 */
2052
2053
2054
2055 // -----------------------
2056 // EUpdate
2057
2058 void EUpdate::encode(bufferlist &bl, uint64_t features) const
2059 {
2060   ENCODE_START(4, 4, bl);
2061   ::encode(stamp, bl);
2062   ::encode(type, bl);
2063   ::encode(metablob, bl, features);
2064   ::encode(client_map, bl);
2065   ::encode(cmapv, bl);
2066   ::encode(reqid, bl);
2067   ::encode(had_slaves, bl);
2068   ENCODE_FINISH(bl);
2069 }
2070
2071 void EUpdate::decode(bufferlist::iterator &bl)
2072 {
2073   DECODE_START_LEGACY_COMPAT_LEN(4, 4, 4, bl);
2074   if (struct_v >= 2)
2075     ::decode(stamp, bl);
2076   ::decode(type, bl);
2077   ::decode(metablob, bl);
2078   ::decode(client_map, bl);
2079   if (struct_v >= 3)
2080     ::decode(cmapv, bl);
2081   ::decode(reqid, bl);
2082   ::decode(had_slaves, bl);
2083   DECODE_FINISH(bl);
2084 }
2085
2086 void EUpdate::dump(Formatter *f) const
2087 {
2088   f->open_object_section("metablob");
2089   metablob.dump(f);
2090   f->close_section(); // metablob
2091
2092   f->dump_string("type", type);
2093   f->dump_int("client map length", client_map.length());
2094   f->dump_int("client map version", cmapv);
2095   f->dump_stream("reqid") << reqid;
2096   f->dump_string("had slaves", had_slaves ? "true" : "false");
2097 }
2098
2099 void EUpdate::generate_test_instances(list<EUpdate*>& ls)
2100 {
2101   ls.push_back(new EUpdate());
2102 }
2103
2104
2105 void EUpdate::update_segment()
2106 {
2107   metablob.update_segment(_segment);
2108
2109   if (client_map.length())
2110     _segment->sessionmapv = cmapv;
2111
2112   if (had_slaves)
2113     _segment->uncommitted_masters.insert(reqid);
2114 }
2115
2116 void EUpdate::replay(MDSRank *mds)
2117 {
2118   metablob.replay(mds, _segment);
2119
2120   if (had_slaves) {
2121     dout(10) << "EUpdate.replay " << reqid << " had slaves, expecting a matching ECommitted" << dendl;
2122     _segment->uncommitted_masters.insert(reqid);
2123     set<mds_rank_t> slaves;
2124     mds->mdcache->add_uncommitted_master(reqid, _segment, slaves, true);
2125   }
2126
2127   if (client_map.length()) {
2128     if (mds->sessionmap.get_version() >= cmapv) {
2129       dout(10) << "EUpdate.replay sessionmap v " << cmapv
2130                << " <= table " << mds->sessionmap.get_version() << dendl;
2131     } else {
2132       dout(10) << "EUpdate.replay sessionmap " << mds->sessionmap.get_version()
2133                << " < " << cmapv << dendl;
2134       // open client sessions?
2135       map<client_t,entity_inst_t> cm;
2136       bufferlist::iterator blp = client_map.begin();
2137       ::decode(cm, blp);
2138       mds->sessionmap.open_sessions(cm);
2139
2140       assert(mds->sessionmap.get_version() == cmapv);
2141       mds->sessionmap.set_projected(mds->sessionmap.get_version());
2142     }
2143   }
2144   update_segment();
2145 }
2146
2147
2148 // ------------------------
2149 // EOpen
2150
2151 void EOpen::encode(bufferlist &bl, uint64_t features) const {
2152   ENCODE_START(4, 3, bl);
2153   ::encode(stamp, bl);
2154   ::encode(metablob, bl, features);
2155   ::encode(inos, bl);
2156   ::encode(snap_inos, bl);
2157   ENCODE_FINISH(bl);
2158 }
2159
2160 void EOpen::decode(bufferlist::iterator &bl) {
2161   DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
2162   if (struct_v >= 2)
2163     ::decode(stamp, bl);
2164   ::decode(metablob, bl);
2165   ::decode(inos, bl);
2166   if (struct_v >= 4)
2167     ::decode(snap_inos, bl);
2168   DECODE_FINISH(bl);
2169 }
2170
2171 void EOpen::dump(Formatter *f) const
2172 {
2173   f->open_object_section("metablob");
2174   metablob.dump(f);
2175   f->close_section(); // metablob
2176   f->open_array_section("inos involved");
2177   for (vector<inodeno_t>::const_iterator i = inos.begin();
2178        i != inos.end(); ++i) {
2179     f->dump_int("ino", *i);
2180   }
2181   f->close_section(); // inos
2182 }
2183
2184 void EOpen::generate_test_instances(list<EOpen*>& ls)
2185 {
2186   ls.push_back(new EOpen());
2187   ls.push_back(new EOpen());
2188   ls.back()->add_ino(0);
2189 }
2190
2191 void EOpen::update_segment()
2192 {
2193   // ??
2194 }
2195
2196 void EOpen::replay(MDSRank *mds)
2197 {
2198   dout(10) << "EOpen.replay " << dendl;
2199   metablob.replay(mds, _segment);
2200
2201   // note which segments inodes belong to, so we don't have to start rejournaling them
2202   for (const auto &ino : inos) {
2203     CInode *in = mds->mdcache->get_inode(ino);
2204     if (!in) {
2205       dout(0) << "EOpen.replay ino " << ino << " not in metablob" << dendl;
2206       assert(in);
2207     }
2208     _segment->open_files.push_back(&in->item_open_file);
2209   }
2210   for (const auto &vino : snap_inos) {
2211     CInode *in = mds->mdcache->get_inode(vino);
2212     if (!in) {
2213       dout(0) << "EOpen.replay ino " << vino << " not in metablob" << dendl;
2214       assert(in);
2215     }
2216     _segment->open_files.push_back(&in->item_open_file);
2217   }
2218 }
2219
2220
2221 // -----------------------
2222 // ECommitted
2223
2224 void ECommitted::replay(MDSRank *mds)
2225 {
2226   if (mds->mdcache->uncommitted_masters.count(reqid)) {
2227     dout(10) << "ECommitted.replay " << reqid << dendl;
2228     mds->mdcache->uncommitted_masters[reqid].ls->uncommitted_masters.erase(reqid);
2229     mds->mdcache->uncommitted_masters.erase(reqid);
2230   } else {
2231     dout(10) << "ECommitted.replay " << reqid << " -- didn't see original op" << dendl;
2232   }
2233 }
2234
2235 void ECommitted::encode(bufferlist& bl, uint64_t features) const
2236 {
2237   ENCODE_START(3, 3, bl);
2238   ::encode(stamp, bl);
2239   ::encode(reqid, bl);
2240   ENCODE_FINISH(bl);
2241 }
2242
2243 void ECommitted::decode(bufferlist::iterator& bl)
2244 {
2245   DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
2246   if (struct_v >= 2)
2247     ::decode(stamp, bl);
2248   ::decode(reqid, bl);
2249   DECODE_FINISH(bl);
2250 }
2251
2252 void ECommitted::dump(Formatter *f) const {
2253   f->dump_stream("stamp") << stamp;
2254   f->dump_stream("reqid") << reqid;
2255 }
2256
2257 void ECommitted::generate_test_instances(list<ECommitted*>& ls)
2258 {
2259   ls.push_back(new ECommitted);
2260   ls.push_back(new ECommitted);
2261   ls.back()->stamp = utime_t(1, 2);
2262   ls.back()->reqid = metareqid_t(entity_name_t::CLIENT(123), 456);
2263 }
2264
2265 // -----------------------
2266 // ESlaveUpdate
2267
2268 void link_rollback::encode(bufferlist &bl) const
2269 {
2270   ENCODE_START(2, 2, bl);
2271   ::encode(reqid, bl);
2272   ::encode(ino, bl);
2273   ::encode(was_inc, bl);
2274   ::encode(old_ctime, bl);
2275   ::encode(old_dir_mtime, bl);
2276   ::encode(old_dir_rctime, bl);
2277   ENCODE_FINISH(bl);
2278 }
2279
2280 void link_rollback::decode(bufferlist::iterator &bl)
2281 {
2282   DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
2283   ::decode(reqid, bl);
2284   ::decode(ino, bl);
2285   ::decode(was_inc, bl);
2286   ::decode(old_ctime, bl);
2287   ::decode(old_dir_mtime, bl);
2288   ::decode(old_dir_rctime, bl);
2289   DECODE_FINISH(bl);
2290 }
2291
2292 void link_rollback::dump(Formatter *f) const
2293 {
2294   f->dump_stream("metareqid") << reqid;
2295   f->dump_int("ino", ino);
2296   f->dump_string("was incremented", was_inc ? "true" : "false");
2297   f->dump_stream("old_ctime") << old_ctime;
2298   f->dump_stream("old_dir_mtime") << old_dir_mtime;
2299   f->dump_stream("old_dir_rctime") << old_dir_rctime;
2300 }
2301
2302 void link_rollback::generate_test_instances(list<link_rollback*>& ls)
2303 {
2304   ls.push_back(new link_rollback());
2305 }
2306
2307 void rmdir_rollback::encode(bufferlist& bl) const
2308 {
2309   ENCODE_START(2, 2, bl);
2310   ::encode(reqid, bl);
2311   ::encode(src_dir, bl);
2312   ::encode(src_dname, bl);
2313   ::encode(dest_dir, bl);
2314   ::encode(dest_dname, bl);
2315   ENCODE_FINISH(bl);
2316 }
2317
2318 void rmdir_rollback::decode(bufferlist::iterator& bl)
2319 {
2320   DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
2321   ::decode(reqid, bl);
2322   ::decode(src_dir, bl);
2323   ::decode(src_dname, bl);
2324   ::decode(dest_dir, bl);
2325   ::decode(dest_dname, bl);
2326   DECODE_FINISH(bl);
2327 }
2328
2329 void rmdir_rollback::dump(Formatter *f) const
2330 {
2331   f->dump_stream("metareqid") << reqid;
2332   f->dump_stream("source directory") << src_dir;
2333   f->dump_string("source dname", src_dname);
2334   f->dump_stream("destination directory") << dest_dir;
2335   f->dump_string("destination dname", dest_dname);
2336 }
2337
2338 void rmdir_rollback::generate_test_instances(list<rmdir_rollback*>& ls)
2339 {
2340   ls.push_back(new rmdir_rollback());
2341 }
2342
2343 void rename_rollback::drec::encode(bufferlist &bl) const
2344 {
2345   ENCODE_START(2, 2, bl);
2346   ::encode(dirfrag, bl);
2347   ::encode(dirfrag_old_mtime, bl);
2348   ::encode(dirfrag_old_rctime, bl);
2349   ::encode(ino, bl);
2350   ::encode(remote_ino, bl);
2351   ::encode(dname, bl);
2352   ::encode(remote_d_type, bl);
2353   ::encode(old_ctime, bl);
2354   ENCODE_FINISH(bl);
2355 }
2356
2357 void rename_rollback::drec::decode(bufferlist::iterator &bl)
2358 {
2359   DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
2360   ::decode(dirfrag, bl);
2361   ::decode(dirfrag_old_mtime, bl);
2362   ::decode(dirfrag_old_rctime, bl);
2363   ::decode(ino, bl);
2364   ::decode(remote_ino, bl);
2365   ::decode(dname, bl);
2366   ::decode(remote_d_type, bl);
2367   ::decode(old_ctime, bl);
2368   DECODE_FINISH(bl);
2369 }
2370
2371 void rename_rollback::drec::dump(Formatter *f) const
2372 {
2373   f->dump_stream("directory fragment") << dirfrag;
2374   f->dump_stream("directory old mtime") << dirfrag_old_mtime;
2375   f->dump_stream("directory old rctime") << dirfrag_old_rctime;
2376   f->dump_int("ino", ino);
2377   f->dump_int("remote ino", remote_ino);
2378   f->dump_string("dname", dname);
2379   uint32_t type = DTTOIF(remote_d_type) & S_IFMT; // convert to type entries
2380   string type_string;
2381   switch(type) {
2382   case S_IFREG:
2383     type_string = "file"; break;
2384   case S_IFLNK:
2385     type_string = "symlink"; break;
2386   case S_IFDIR:
2387     type_string = "directory"; break;
2388   default:
2389     type_string = "UNKNOWN-" + stringify((int)type); break;
2390   }
2391   f->dump_string("remote dtype", type_string);
2392   f->dump_stream("old ctime") << old_ctime;
2393 }
2394
2395 void rename_rollback::drec::generate_test_instances(list<drec*>& ls)
2396 {
2397   ls.push_back(new drec());
2398   ls.back()->remote_d_type = IFTODT(S_IFREG);
2399 }
2400
2401 void rename_rollback::encode(bufferlist &bl) const
2402 {
2403   ENCODE_START(2, 2, bl);
2404   ::encode(reqid, bl);
2405   encode(orig_src, bl);
2406   encode(orig_dest, bl);
2407   encode(stray, bl);
2408   ::encode(ctime, bl);
2409   ENCODE_FINISH(bl);
2410 }
2411
2412 void rename_rollback::decode(bufferlist::iterator &bl)
2413 {
2414   DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
2415   ::decode(reqid, bl);
2416   decode(orig_src, bl);
2417   decode(orig_dest, bl);
2418   decode(stray, bl);
2419   ::decode(ctime, bl);
2420   DECODE_FINISH(bl);
2421 }
2422
2423 void rename_rollback::dump(Formatter *f) const
2424 {
2425   f->dump_stream("request id") << reqid;
2426   f->open_object_section("original src drec");
2427   orig_src.dump(f);
2428   f->close_section(); // original src drec
2429   f->open_object_section("original dest drec");
2430   orig_dest.dump(f);
2431   f->close_section(); // original dest drec
2432   f->open_object_section("stray drec");
2433   stray.dump(f);
2434   f->close_section(); // stray drec
2435   f->dump_stream("ctime") << ctime;
2436 }
2437
2438 void rename_rollback::generate_test_instances(list<rename_rollback*>& ls)
2439 {
2440   ls.push_back(new rename_rollback());
2441   ls.back()->orig_src.remote_d_type = IFTODT(S_IFREG);
2442   ls.back()->orig_dest.remote_d_type = IFTODT(S_IFREG);
2443   ls.back()->stray.remote_d_type = IFTODT(S_IFREG);
2444 }
2445
2446 void ESlaveUpdate::encode(bufferlist &bl, uint64_t features) const
2447 {
2448   ENCODE_START(3, 3, bl);
2449   ::encode(stamp, bl);
2450   ::encode(type, bl);
2451   ::encode(reqid, bl);
2452   ::encode(master, bl);
2453   ::encode(op, bl);
2454   ::encode(origop, bl);
2455   ::encode(commit, bl, features);
2456   ::encode(rollback, bl);
2457   ENCODE_FINISH(bl);
2458 }
2459
2460 void ESlaveUpdate::decode(bufferlist::iterator &bl)
2461 {
2462   DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
2463   if (struct_v >= 2)
2464     ::decode(stamp, bl);
2465   ::decode(type, bl);
2466   ::decode(reqid, bl);
2467   ::decode(master, bl);
2468   ::decode(op, bl);
2469   ::decode(origop, bl);
2470   ::decode(commit, bl);
2471   ::decode(rollback, bl);
2472   DECODE_FINISH(bl);
2473 }
2474
2475 void ESlaveUpdate::dump(Formatter *f) const
2476 {
2477   f->open_object_section("metablob");
2478   commit.dump(f);
2479   f->close_section(); // metablob
2480
2481   f->dump_int("rollback length", rollback.length());
2482   f->dump_string("type", type);
2483   f->dump_stream("metareqid") << reqid;
2484   f->dump_int("master", master);
2485   f->dump_int("op", op);
2486   f->dump_int("original op", origop);
2487 }
2488
2489 void ESlaveUpdate::generate_test_instances(list<ESlaveUpdate*>& ls)
2490 {
2491   ls.push_back(new ESlaveUpdate());
2492 }
2493
2494
2495 void ESlaveUpdate::replay(MDSRank *mds)
2496 {
2497   MDSlaveUpdate *su;
2498   switch (op) {
2499   case ESlaveUpdate::OP_PREPARE:
2500     dout(10) << "ESlaveUpdate.replay prepare " << reqid << " for mds." << master
2501              << ": applying commit, saving rollback info" << dendl;
2502     su = new MDSlaveUpdate(origop, rollback, _segment->slave_updates);
2503     commit.replay(mds, _segment, su);
2504     mds->mdcache->add_uncommitted_slave_update(reqid, master, su);
2505     break;
2506
2507   case ESlaveUpdate::OP_COMMIT:
2508     su = mds->mdcache->get_uncommitted_slave_update(reqid, master);
2509     if (su) {
2510       dout(10) << "ESlaveUpdate.replay commit " << reqid << " for mds." << master << dendl;
2511       mds->mdcache->finish_uncommitted_slave_update(reqid, master);
2512     } else {
2513       dout(10) << "ESlaveUpdate.replay commit " << reqid << " for mds." << master
2514                << ": ignoring, no previously saved prepare" << dendl;
2515     }
2516     break;
2517
2518   case ESlaveUpdate::OP_ROLLBACK:
2519     dout(10) << "ESlaveUpdate.replay abort " << reqid << " for mds." << master
2520              << ": applying rollback commit blob" << dendl;
2521     commit.replay(mds, _segment);
2522     su = mds->mdcache->get_uncommitted_slave_update(reqid, master);
2523     if (su)
2524       mds->mdcache->finish_uncommitted_slave_update(reqid, master);
2525     break;
2526
2527   default:
2528     mds->clog->error() << "invalid op in ESlaveUpdate";
2529     mds->damaged();
2530     ceph_abort();  // Should be unreachable because damaged() calls respawn()
2531   }
2532 }
2533
2534
2535 // -----------------------
2536 // ESubtreeMap
2537
2538 void ESubtreeMap::encode(bufferlist& bl, uint64_t features) const
2539 {
2540   ENCODE_START(6, 5, bl);
2541   ::encode(stamp, bl);
2542   ::encode(metablob, bl, features);
2543   ::encode(subtrees, bl);
2544   ::encode(ambiguous_subtrees, bl);
2545   ::encode(expire_pos, bl);
2546   ::encode(event_seq, bl);
2547   ENCODE_FINISH(bl);
2548 }
2549
2550 void ESubtreeMap::decode(bufferlist::iterator &bl)
2551 {
2552   DECODE_START_LEGACY_COMPAT_LEN(6, 5, 5, bl);
2553   if (struct_v >= 2)
2554     ::decode(stamp, bl);
2555   ::decode(metablob, bl);
2556   ::decode(subtrees, bl);
2557   if (struct_v >= 4)
2558     ::decode(ambiguous_subtrees, bl);
2559   if (struct_v >= 3)
2560     ::decode(expire_pos, bl);
2561   if (struct_v >= 6)
2562     ::decode(event_seq, bl);
2563   DECODE_FINISH(bl);
2564 }
2565
2566 void ESubtreeMap::dump(Formatter *f) const
2567 {
2568   f->open_object_section("metablob");
2569   metablob.dump(f);
2570   f->close_section(); // metablob
2571
2572   f->open_array_section("subtrees");
2573   for(map<dirfrag_t,vector<dirfrag_t> >::const_iterator i = subtrees.begin();
2574       i != subtrees.end(); ++i) {
2575     f->open_object_section("tree");
2576     f->dump_stream("root dirfrag") << i->first;
2577     for (vector<dirfrag_t>::const_iterator j = i->second.begin();
2578          j != i->second.end(); ++j) {
2579       f->dump_stream("bound dirfrag") << *j;
2580     }
2581     f->close_section(); // tree
2582   }
2583   f->close_section(); // subtrees
2584
2585   f->open_array_section("ambiguous subtrees");
2586   for(set<dirfrag_t>::const_iterator i = ambiguous_subtrees.begin();
2587       i != ambiguous_subtrees.end(); ++i) {
2588     f->dump_stream("dirfrag") << *i;
2589   }
2590   f->close_section(); // ambiguous subtrees
2591
2592   f->dump_int("expire position", expire_pos);
2593 }
2594
2595 void ESubtreeMap::generate_test_instances(list<ESubtreeMap*>& ls)
2596 {
2597   ls.push_back(new ESubtreeMap());
2598 }
2599
2600 void ESubtreeMap::replay(MDSRank *mds)
2601 {
2602   if (expire_pos && expire_pos > mds->mdlog->journaler->get_expire_pos())
2603     mds->mdlog->journaler->set_expire_pos(expire_pos);
2604
2605   // suck up the subtree map?
2606   if (mds->mdcache->is_subtrees()) {
2607     dout(10) << "ESubtreeMap.replay -- i already have import map; verifying" << dendl;
2608     int errors = 0;
2609
2610     for (map<dirfrag_t, vector<dirfrag_t> >::iterator p = subtrees.begin();
2611          p != subtrees.end();
2612          ++p) {
2613       CDir *dir = mds->mdcache->get_dirfrag(p->first);
2614       if (!dir) {
2615         mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2616                           << " subtree root " << p->first << " not in cache";
2617         ++errors;
2618         continue;
2619       }
2620
2621       if (!mds->mdcache->is_subtree(dir)) {
2622         mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2623                           << " subtree root " << p->first << " not a subtree in cache";
2624         ++errors;
2625         continue;
2626       }
2627       if (dir->get_dir_auth().first != mds->get_nodeid()) {
2628         mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2629                           << " subtree root " << p->first
2630                           << " is not mine in cache (it's " << dir->get_dir_auth() << ")";
2631         ++errors;
2632         continue;
2633       }
2634
2635       for (vector<dirfrag_t>::iterator q = p->second.begin(); q != p->second.end(); ++q)
2636         mds->mdcache->get_force_dirfrag(*q, true);
2637
2638       set<CDir*> bounds;
2639       mds->mdcache->get_subtree_bounds(dir, bounds);
2640       for (vector<dirfrag_t>::iterator q = p->second.begin(); q != p->second.end(); ++q) {
2641         CDir *b = mds->mdcache->get_dirfrag(*q);
2642         if (!b) {
2643           mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2644                             << " subtree " << p->first << " bound " << *q << " not in cache";
2645         ++errors;
2646           continue;
2647         }
2648         if (bounds.count(b) == 0) {
2649           mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2650                             << " subtree " << p->first << " bound " << *q << " not a bound in cache";
2651         ++errors;
2652           continue;
2653         }
2654         bounds.erase(b);
2655       }
2656       for (set<CDir*>::iterator q = bounds.begin(); q != bounds.end(); ++q) {
2657         mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2658                           << " subtree " << p->first << " has extra bound in cache " << (*q)->dirfrag();
2659         ++errors;
2660       }
2661
2662       if (ambiguous_subtrees.count(p->first)) {
2663         if (!mds->mdcache->have_ambiguous_import(p->first)) {
2664           mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2665                             << " subtree " << p->first << " is ambiguous but is not in our cache";
2666           ++errors;
2667         }
2668       } else {
2669         if (mds->mdcache->have_ambiguous_import(p->first)) {
2670           mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2671                             << " subtree " << p->first << " is not ambiguous but is in our cache";
2672           ++errors;
2673         }
2674       }
2675     }
2676
2677     list<CDir*> subs;
2678     mds->mdcache->list_subtrees(subs);
2679     for (list<CDir*>::iterator p = subs.begin(); p != subs.end(); ++p) {
2680       CDir *dir = *p;
2681       if (dir->get_dir_auth().first != mds->get_nodeid())
2682         continue;
2683       if (subtrees.count(dir->dirfrag()) == 0) {
2684         mds->clog->error() << " replayed ESubtreeMap at " << get_start_off()
2685                           << " does not include cache subtree " << dir->dirfrag();
2686         ++errors;
2687       }
2688     }
2689
2690     if (errors) {
2691       dout(0) << "journal subtrees: " << subtrees << dendl;
2692       dout(0) << "journal ambig_subtrees: " << ambiguous_subtrees << dendl;
2693       mds->mdcache->show_subtrees();
2694       assert(!g_conf->mds_debug_subtrees || errors == 0);
2695     }
2696     return;
2697   }
2698
2699   dout(10) << "ESubtreeMap.replay -- reconstructing (auth) subtree spanning tree" << dendl;
2700
2701   // first, stick the spanning tree in my cache
2702   //metablob.print(*_dout);
2703   metablob.replay(mds, _segment);
2704
2705   // restore import/export maps
2706   for (map<dirfrag_t, vector<dirfrag_t> >::iterator p = subtrees.begin();
2707        p != subtrees.end();
2708        ++p) {
2709     CDir *dir = mds->mdcache->get_dirfrag(p->first);
2710     assert(dir);
2711     if (ambiguous_subtrees.count(p->first)) {
2712       // ambiguous!
2713       mds->mdcache->add_ambiguous_import(p->first, p->second);
2714       mds->mdcache->adjust_bounded_subtree_auth(dir, p->second,
2715                                                 mds_authority_t(mds->get_nodeid(), mds->get_nodeid()));
2716     } else {
2717       // not ambiguous
2718       mds->mdcache->adjust_bounded_subtree_auth(dir, p->second, mds->get_nodeid());
2719     }
2720   }
2721
2722   mds->mdcache->recalc_auth_bits(true);
2723
2724   mds->mdcache->show_subtrees();
2725 }
2726
2727
2728
2729 // -----------------------
2730 // EFragment
2731
2732 void EFragment::replay(MDSRank *mds)
2733 {
2734   dout(10) << "EFragment.replay " << op_name(op) << " " << ino << " " << basefrag << " by " << bits << dendl;
2735
2736   list<CDir*> resultfrags;
2737   list<MDSInternalContextBase*> waiters;
2738   list<frag_t> old_frags;
2739
2740   // in may be NULL if it wasn't in our cache yet.  if it's a prepare
2741   // it will be once we replay the metablob , but first we need to
2742   // refragment anything we already have in the cache.
2743   CInode *in = mds->mdcache->get_inode(ino);
2744
2745   switch (op) {
2746   case OP_PREPARE:
2747     mds->mdcache->add_uncommitted_fragment(dirfrag_t(ino, basefrag), bits, orig_frags, _segment, &rollback);
2748
2749     if (in)
2750       mds->mdcache->adjust_dir_fragments(in, basefrag, bits, resultfrags, waiters, true);
2751     break;
2752
2753   case OP_ROLLBACK:
2754     if (in) {
2755       in->dirfragtree.get_leaves_under(basefrag, old_frags);
2756       if (orig_frags.empty()) {
2757         // old format EFragment
2758         mds->mdcache->adjust_dir_fragments(in, basefrag, -bits, resultfrags, waiters, true);
2759       } else {
2760         for (list<frag_t>::iterator p = orig_frags.begin(); p != orig_frags.end(); ++p)
2761           mds->mdcache->force_dir_fragment(in, *p);
2762       }
2763     }
2764     mds->mdcache->rollback_uncommitted_fragment(dirfrag_t(ino, basefrag), old_frags);
2765     break;
2766
2767   case OP_COMMIT:
2768   case OP_FINISH:
2769     mds->mdcache->finish_uncommitted_fragment(dirfrag_t(ino, basefrag), op);
2770     break;
2771
2772   default:
2773     ceph_abort();
2774   }
2775
2776   metablob.replay(mds, _segment);
2777   if (in && g_conf->mds_debug_frag)
2778     in->verify_dirfrags();
2779 }
2780
2781 void EFragment::encode(bufferlist &bl, uint64_t features) const {
2782   ENCODE_START(5, 4, bl);
2783   ::encode(stamp, bl);
2784   ::encode(op, bl);
2785   ::encode(ino, bl);
2786   ::encode(basefrag, bl);
2787   ::encode(bits, bl);
2788   ::encode(metablob, bl, features);
2789   ::encode(orig_frags, bl);
2790   ::encode(rollback, bl);
2791   ENCODE_FINISH(bl);
2792 }
2793
2794 void EFragment::decode(bufferlist::iterator &bl) {
2795   DECODE_START_LEGACY_COMPAT_LEN(5, 4, 4, bl);
2796   if (struct_v >= 2)
2797     ::decode(stamp, bl);
2798   if (struct_v >= 3)
2799     ::decode(op, bl);
2800   ::decode(ino, bl);
2801   ::decode(basefrag, bl);
2802   ::decode(bits, bl);
2803   ::decode(metablob, bl);
2804   if (struct_v >= 5) {
2805     ::decode(orig_frags, bl);
2806     ::decode(rollback, bl);
2807   }
2808   DECODE_FINISH(bl);
2809 }
2810
2811 void EFragment::dump(Formatter *f) const
2812 {
2813   /*f->open_object_section("Metablob");
2814   metablob.dump(f); // sadly we don't have this; dunno if we'll get it
2815   f->close_section();*/
2816   f->dump_string("op", op_name(op));
2817   f->dump_stream("ino") << ino;
2818   f->dump_stream("base frag") << basefrag;
2819   f->dump_int("bits", bits);
2820 }
2821
2822 void EFragment::generate_test_instances(list<EFragment*>& ls)
2823 {
2824   ls.push_back(new EFragment);
2825   ls.push_back(new EFragment);
2826   ls.back()->op = OP_PREPARE;
2827   ls.back()->ino = 1;
2828   ls.back()->bits = 5;
2829 }
2830
2831 void dirfrag_rollback::encode(bufferlist &bl) const
2832 {
2833   ENCODE_START(1, 1, bl);
2834   ::encode(fnode, bl);
2835   ENCODE_FINISH(bl);
2836 }
2837
2838 void dirfrag_rollback::decode(bufferlist::iterator &bl)
2839 {
2840   DECODE_START(1, bl);
2841   ::decode(fnode, bl);
2842   DECODE_FINISH(bl);
2843 }
2844
2845
2846
2847 // =========================================================================
2848
2849 // -----------------------
2850 // EExport
2851
2852 void EExport::replay(MDSRank *mds)
2853 {
2854   dout(10) << "EExport.replay " << base << dendl;
2855   metablob.replay(mds, _segment);
2856
2857   CDir *dir = mds->mdcache->get_dirfrag(base);
2858   assert(dir);
2859
2860   set<CDir*> realbounds;
2861   for (set<dirfrag_t>::iterator p = bounds.begin();
2862        p != bounds.end();
2863        ++p) {
2864     CDir *bd = mds->mdcache->get_dirfrag(*p);
2865     assert(bd);
2866     realbounds.insert(bd);
2867   }
2868
2869   // adjust auth away
2870   mds->mdcache->adjust_bounded_subtree_auth(dir, realbounds, CDIR_AUTH_UNDEF);
2871
2872   mds->mdcache->try_trim_non_auth_subtree(dir);
2873 }
2874
2875 void EExport::encode(bufferlist& bl, uint64_t features) const
2876 {
2877   ENCODE_START(3, 3, bl);
2878   ::encode(stamp, bl);
2879   ::encode(metablob, bl, features);
2880   ::encode(base, bl);
2881   ::encode(bounds, bl);
2882   ENCODE_FINISH(bl);
2883 }
2884
2885 void EExport::decode(bufferlist::iterator &bl)
2886 {
2887   DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
2888   if (struct_v >= 2)
2889     ::decode(stamp, bl);
2890   ::decode(metablob, bl);
2891   ::decode(base, bl);
2892   ::decode(bounds, bl);
2893   DECODE_FINISH(bl);
2894 }
2895
2896 void EExport::dump(Formatter *f) const
2897 {
2898   f->dump_float("stamp", (double)stamp);
2899   /*f->open_object_section("Metablob");
2900   metablob.dump(f); // sadly we don't have this; dunno if we'll get it
2901   f->close_section();*/
2902   f->dump_stream("base dirfrag") << base;
2903   f->open_array_section("bounds dirfrags");
2904   for (set<dirfrag_t>::const_iterator i = bounds.begin();
2905       i != bounds.end(); ++i) {
2906     f->dump_stream("dirfrag") << *i;
2907   }
2908   f->close_section(); // bounds dirfrags
2909 }
2910
2911 void EExport::generate_test_instances(list<EExport*>& ls)
2912 {
2913   EExport *sample = new EExport();
2914   ls.push_back(sample);
2915 }
2916
2917
2918 // -----------------------
2919 // EImportStart
2920
2921 void EImportStart::update_segment()
2922 {
2923   _segment->sessionmapv = cmapv;
2924 }
2925
2926 void EImportStart::replay(MDSRank *mds)
2927 {
2928   dout(10) << "EImportStart.replay " << base << " bounds " << bounds << dendl;
2929   //metablob.print(*_dout);
2930   metablob.replay(mds, _segment);
2931
2932   // put in ambiguous import list
2933   mds->mdcache->add_ambiguous_import(base, bounds);
2934
2935   // set auth partially to us so we don't trim it
2936   CDir *dir = mds->mdcache->get_dirfrag(base);
2937   assert(dir);
2938
2939   set<CDir*> realbounds;
2940   for (vector<dirfrag_t>::iterator p = bounds.begin();
2941        p != bounds.end();
2942        ++p) {
2943     CDir *bd = mds->mdcache->get_dirfrag(*p);
2944     assert(bd);
2945     if (!bd->is_subtree_root())
2946       bd->state_clear(CDir::STATE_AUTH);
2947     realbounds.insert(bd);
2948   }
2949
2950   mds->mdcache->adjust_bounded_subtree_auth(dir, realbounds,
2951                                             mds_authority_t(mds->get_nodeid(), mds->get_nodeid()));
2952
2953   // open client sessions?
2954   if (mds->sessionmap.get_version() >= cmapv) {
2955     dout(10) << "EImportStart.replay sessionmap " << mds->sessionmap.get_version()
2956              << " >= " << cmapv << ", noop" << dendl;
2957   } else {
2958     dout(10) << "EImportStart.replay sessionmap " << mds->sessionmap.get_version()
2959              << " < " << cmapv << dendl;
2960     map<client_t,entity_inst_t> cm;
2961     bufferlist::iterator blp = client_map.begin();
2962     ::decode(cm, blp);
2963     mds->sessionmap.open_sessions(cm);
2964     assert(mds->sessionmap.get_version() == cmapv);
2965     mds->sessionmap.set_projected(mds->sessionmap.get_version());
2966   }
2967   update_segment();
2968 }
2969
2970 void EImportStart::encode(bufferlist &bl, uint64_t features) const {
2971   ENCODE_START(3, 3, bl);
2972   ::encode(stamp, bl);
2973   ::encode(base, bl);
2974   ::encode(metablob, bl, features);
2975   ::encode(bounds, bl);
2976   ::encode(cmapv, bl);
2977   ::encode(client_map, bl);
2978   ENCODE_FINISH(bl);
2979 }
2980
2981 void EImportStart::decode(bufferlist::iterator &bl) {
2982   DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
2983   if (struct_v >= 2)
2984     ::decode(stamp, bl);
2985   ::decode(base, bl);
2986   ::decode(metablob, bl);
2987   ::decode(bounds, bl);
2988   ::decode(cmapv, bl);
2989   ::decode(client_map, bl);
2990   DECODE_FINISH(bl);
2991 }
2992
2993 void EImportStart::dump(Formatter *f) const
2994 {
2995   f->dump_stream("base dirfrag") << base;
2996   f->open_array_section("boundary dirfrags");
2997   for (vector<dirfrag_t>::const_iterator iter = bounds.begin();
2998       iter != bounds.end(); ++iter) {
2999     f->dump_stream("frag") << *iter;
3000   }
3001   f->close_section();
3002 }
3003
3004 void EImportStart::generate_test_instances(list<EImportStart*>& ls)
3005 {
3006   ls.push_back(new EImportStart);
3007 }
3008
3009 // -----------------------
3010 // EImportFinish
3011
3012 void EImportFinish::replay(MDSRank *mds)
3013 {
3014   if (mds->mdcache->have_ambiguous_import(base)) {
3015     dout(10) << "EImportFinish.replay " << base << " success=" << success << dendl;
3016     if (success) {
3017       mds->mdcache->finish_ambiguous_import(base);
3018     } else {
3019       CDir *dir = mds->mdcache->get_dirfrag(base);
3020       assert(dir);
3021       vector<dirfrag_t> bounds;
3022       mds->mdcache->get_ambiguous_import_bounds(base, bounds);
3023       mds->mdcache->adjust_bounded_subtree_auth(dir, bounds, CDIR_AUTH_UNDEF);
3024       mds->mdcache->cancel_ambiguous_import(dir);
3025       mds->mdcache->try_trim_non_auth_subtree(dir);
3026    }
3027   } else {
3028     // this shouldn't happen unless this is an old journal
3029     dout(10) << "EImportFinish.replay " << base << " success=" << success
3030              << " on subtree not marked as ambiguous"
3031              << dendl;
3032     mds->clog->error() << "failure replaying journal (EImportFinish)";
3033     mds->damaged();
3034     ceph_abort();  // Should be unreachable because damaged() calls respawn()
3035   }
3036 }
3037
3038 void EImportFinish::encode(bufferlist& bl, uint64_t features) const
3039 {
3040   ENCODE_START(3, 3, bl);
3041   ::encode(stamp, bl);
3042   ::encode(base, bl);
3043   ::encode(success, bl);
3044   ENCODE_FINISH(bl);
3045 }
3046
3047 void EImportFinish::decode(bufferlist::iterator &bl)
3048 {
3049   DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
3050   if (struct_v >= 2)
3051     ::decode(stamp, bl);
3052   ::decode(base, bl);
3053   ::decode(success, bl);
3054   DECODE_FINISH(bl);
3055 }
3056
3057 void EImportFinish::dump(Formatter *f) const
3058 {
3059   f->dump_stream("base dirfrag") << base;
3060   f->dump_string("success", success ? "true" : "false");
3061 }
3062 void EImportFinish::generate_test_instances(list<EImportFinish*>& ls)
3063 {
3064   ls.push_back(new EImportFinish);
3065   ls.push_back(new EImportFinish);
3066   ls.back()->success = true;
3067 }
3068
3069
3070 // ------------------------
3071 // EResetJournal
3072
3073 void EResetJournal::encode(bufferlist& bl, uint64_t features) const
3074 {
3075   ENCODE_START(2, 2, bl);
3076   ::encode(stamp, bl);
3077   ENCODE_FINISH(bl);
3078 }
3079
3080 void EResetJournal::decode(bufferlist::iterator &bl)
3081 {
3082   DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
3083   ::decode(stamp, bl);
3084   DECODE_FINISH(bl);
3085 }
3086
3087 void EResetJournal::dump(Formatter *f) const
3088 {
3089   f->dump_stream("timestamp") << stamp;
3090 }
3091
3092 void EResetJournal::generate_test_instances(list<EResetJournal*>& ls)
3093 {
3094   ls.push_back(new EResetJournal());
3095 }
3096
3097 void EResetJournal::replay(MDSRank *mds)
3098 {
3099   dout(1) << "EResetJournal" << dendl;
3100
3101   mds->sessionmap.wipe();
3102   mds->inotable->replay_reset();
3103
3104   if (mds->mdsmap->get_root() == mds->get_nodeid()) {
3105     CDir *rootdir = mds->mdcache->get_root()->get_or_open_dirfrag(mds->mdcache, frag_t());
3106     mds->mdcache->adjust_subtree_auth(rootdir, mds->get_nodeid());
3107   }
3108
3109   CDir *mydir = mds->mdcache->get_myin()->get_or_open_dirfrag(mds->mdcache, frag_t());
3110   mds->mdcache->adjust_subtree_auth(mydir, mds->get_nodeid());
3111
3112   mds->mdcache->recalc_auth_bits(true);
3113
3114   mds->mdcache->show_subtrees();
3115 }
3116
3117
3118 void ENoOp::encode(bufferlist &bl, uint64_t features) const
3119 {
3120   ENCODE_START(2, 2, bl);
3121   ::encode(pad_size, bl);
3122   uint8_t const pad = 0xff;
3123   for (unsigned int i = 0; i < pad_size; ++i) {
3124     ::encode(pad, bl);
3125   }
3126   ENCODE_FINISH(bl);
3127 }
3128
3129
3130 void ENoOp::decode(bufferlist::iterator &bl)
3131 {
3132   DECODE_START(2, bl);
3133   ::decode(pad_size, bl);
3134   if (bl.get_remaining() != pad_size) {
3135     // This is spiritually an assertion, but expressing in a way that will let
3136     // journal debug tools catch it and recognise a malformed entry.
3137     throw buffer::end_of_buffer();
3138   } else {
3139     bl.advance(pad_size);
3140   }
3141   DECODE_FINISH(bl);
3142 }
3143
3144
3145 void ENoOp::replay(MDSRank *mds)
3146 {
3147   dout(4) << "ENoOp::replay, " << pad_size << " bytes skipped in journal" << dendl;
3148 }
3149
3150 /**
3151  * If re-formatting an old journal that used absolute log position
3152  * references as segment sequence numbers, use this function to update
3153  * it.
3154  *
3155  * @param mds
3156  * MDSRank instance, just used for logging
3157  * @param old_to_new
3158  * Map of old journal segment sequence numbers to new journal segment sequence numbers
3159  *
3160  * @return
3161  * True if the event was modified.
3162  */
3163 bool EMetaBlob::rewrite_truncate_finish(MDSRank const *mds,
3164     std::map<log_segment_seq_t, log_segment_seq_t> const &old_to_new)
3165 {
3166   bool modified = false;
3167   map<inodeno_t, log_segment_seq_t> new_trunc_finish;
3168   for (std::map<inodeno_t, log_segment_seq_t>::iterator i = truncate_finish.begin();
3169       i != truncate_finish.end(); ++i) {
3170     if (old_to_new.count(i->second)) {
3171       dout(20) << __func__ << " applying segment seq mapping "
3172         << i->second << " -> " << old_to_new.find(i->second)->second << dendl;
3173       new_trunc_finish[i->first] = old_to_new.find(i->second)->second;
3174       modified = true;
3175     } else {
3176       dout(20) << __func__ << " no segment seq mapping found for "
3177         << i->second << dendl;
3178       new_trunc_finish[i->first] = i->second;
3179     }
3180   }
3181   truncate_finish = new_trunc_finish;
3182
3183   return modified;
3184 }