fs/ocfs2/journal.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * journal.c
   5  *
   6  * Defines functions of journalling api
   7  *
   8  * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/fs.h>
  27 #include <linux/types.h>
  28 #include <linux/slab.h>
  29 #include <linux/highmem.h>
  30 #include <linux/kthread.h>
  31 #include <linux/time.h>
  32 #include <linux/random.h>
  33
  34 #include <cluster/masklog.h>
  35
  36 #include "ocfs2.h"
  37
  38 #include "alloc.h"
  39 #include "blockcheck.h"
  40 #include "dir.h"
  41 #include "dlmglue.h"
  42 #include "extent_map.h"
  43 #include "heartbeat.h"
  44 #include "inode.h"
  45 #include "journal.h"
  46 #include "localalloc.h"
  47 #include "slot_map.h"
  48 #include "super.h"
  49 #include "sysfile.h"
  50 #include "uptodate.h"
  51 #include "quota.h"
  52
  53 #include "buffer_head_io.h"
  54 #include "ocfs2_trace.h"
  55
  56 DEFINE_SPINLOCK(trans_inc_lock);
  57
  58 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
  59
  60 static int ocfs2_force_read_journal(struct inode *inode);
  61 static int ocfs2_recover_node(struct ocfs2_super *osb,
  62                               int node_num, int slot_num);
  63 static int __ocfs2_recovery_thread(void *arg);
  64 static int ocfs2_commit_cache(struct ocfs2_super *osb);
  65 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
  66 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
  67                                       int dirty, int replayed);
  68 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
  69                                  int slot_num);
  70 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
  71                                  int slot);
  72 static int ocfs2_commit_thread(void *arg);
  73 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
  74                                             int slot_num,
  75                                             struct ocfs2_dinode *la_dinode,
  76                                             struct ocfs2_dinode *tl_dinode,
  77                                             struct ocfs2_quota_recovery *qrec);
  78
  79 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
  80 {
  81         return __ocfs2_wait_on_mount(osb, 0);
  82 }
  83
  84 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
  85 {
  86         return __ocfs2_wait_on_mount(osb, 1);
  87 }
  88
  89 /*
  90  * This replay_map is to track online/offline slots, so we could recover
  91  * offline slots during recovery and mount
  92  */
  93
  94 enum ocfs2_replay_state {
  95         REPLAY_UNNEEDED = 0,    /* Replay is not needed, so ignore this map */
  96         REPLAY_NEEDED,          /* Replay slots marked in rm_replay_slots */
  97         REPLAY_DONE             /* Replay was already queued */
  98 };
  99
 100 struct ocfs2_replay_map {
 101         unsigned int rm_slots;
 102         enum ocfs2_replay_state rm_state;
 103         unsigned char rm_replay_slots[0];
 104 };
 105
 106 void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
 107 {
 108         if (!osb->replay_map)
 109                 return;
 110
 111         /* If we've already queued the replay, we don't have any more to do */
 112         if (osb->replay_map->rm_state == REPLAY_DONE)
 113                 return;
 114
 115         osb->replay_map->rm_state = state;
 116 }
 117
 118 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
 119 {
 120         struct ocfs2_replay_map *replay_map;
 121         int i, node_num;
 122
 123         /* If replay map is already set, we don't do it again */
 124         if (osb->replay_map)
 125                 return 0;
 126
 127         replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
 128                              (osb->max_slots * sizeof(char)), GFP_KERNEL);
 129
 130         if (!replay_map) {
 131                 mlog_errno(-ENOMEM);
 132                 return -ENOMEM;
 133         }
 134
 135         spin_lock(&osb->osb_lock);
 136
 137         replay_map->rm_slots = osb->max_slots;
 138         replay_map->rm_state = REPLAY_UNNEEDED;
 139
 140         /* set rm_replay_slots for offline slot(s) */
 141         for (i = 0; i < replay_map->rm_slots; i++) {
 142                 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
 143                         replay_map->rm_replay_slots[i] = 1;
 144         }
 145
 146         osb->replay_map = replay_map;
 147         spin_unlock(&osb->osb_lock);
 148         return 0;
 149 }
 150
 151 void ocfs2_queue_replay_slots(struct ocfs2_super *osb)
 152 {
 153         struct ocfs2_replay_map *replay_map = osb->replay_map;
 154         int i;
 155
 156         if (!replay_map)
 157                 return;
 158
 159         if (replay_map->rm_state != REPLAY_NEEDED)
 160                 return;
 161
 162         for (i = 0; i < replay_map->rm_slots; i++)
 163                 if (replay_map->rm_replay_slots[i])
 164                         ocfs2_queue_recovery_completion(osb->journal, i, NULL,
 165                                                         NULL, NULL);
 166         replay_map->rm_state = REPLAY_DONE;
 167 }
 168
 169 void ocfs2_free_replay_slots(struct ocfs2_super *osb)
 170 {
 171         struct ocfs2_replay_map *replay_map = osb->replay_map;
 172
 173         if (!osb->replay_map)
 174                 return;
 175
 176         kfree(replay_map);
 177         osb->replay_map = NULL;
 178 }
 179
 180 int ocfs2_recovery_init(struct ocfs2_super *osb)
 181 {
 182         struct ocfs2_recovery_map *rm;
 183
 184         mutex_init(&osb->recovery_lock);
 185         osb->disable_recovery = 0;
 186         osb->recovery_thread_task = NULL;
 187         init_waitqueue_head(&osb->recovery_event);
 188
 189         rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
 190                      osb->max_slots * sizeof(unsigned int),
 191                      GFP_KERNEL);
 192         if (!rm) {
 193                 mlog_errno(-ENOMEM);
 194                 return -ENOMEM;
 195         }
 196
 197         rm->rm_entries = (unsigned int *)((char *)rm +
 198                                           sizeof(struct ocfs2_recovery_map));
 199         osb->recovery_map = rm;
 200
 201         return 0;
 202 }
 203
 204 /* we can't grab the goofy sem lock from inside wait_event, so we use
 205  * memory barriers to make sure that we'll see the null task before
 206  * being woken up */
 207 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
 208 {
 209         mb();
 210         return osb->recovery_thread_task != NULL;
 211 }
 212
 213 void ocfs2_recovery_exit(struct ocfs2_super *osb)
 214 {
 215         struct ocfs2_recovery_map *rm;
 216
 217         /* disable any new recovery threads and wait for any currently
 218          * running ones to exit. Do this before setting the vol_state. */
 219         mutex_lock(&osb->recovery_lock);
 220         osb->disable_recovery = 1;
 221         mutex_unlock(&osb->recovery_lock);
 222         wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
 223
 224         /* At this point, we know that no more recovery threads can be
 225          * launched, so wait for any recovery completion work to
 226          * complete. */
 227         flush_workqueue(ocfs2_wq);
 228
 229         /*
 230          * Now that recovery is shut down, and the osb is about to be
 231          * freed,  the osb_lock is not taken here.
 232          */
 233         rm = osb->recovery_map;
 234         /* XXX: Should we bug if there are dirty entries? */
 235
 236         kfree(rm);
 237 }
 238
 239 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
 240                                      unsigned int node_num)
 241 {
 242         int i;
 243         struct ocfs2_recovery_map *rm = osb->recovery_map;
 244
 245         assert_spin_locked(&osb->osb_lock);
 246
 247         for (i = 0; i < rm->rm_used; i++) {
 248                 if (rm->rm_entries[i] == node_num)
 249                         return 1;
 250         }
 251
 252         return 0;
 253 }
 254
 255 /* Behaves like test-and-set.  Returns the previous value */
 256 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
 257                                   unsigned int node_num)
 258 {
 259         struct ocfs2_recovery_map *rm = osb->recovery_map;
 260
 261         spin_lock(&osb->osb_lock);
 262         if (__ocfs2_recovery_map_test(osb, node_num)) {
 263                 spin_unlock(&osb->osb_lock);
 264                 return 1;
 265         }
 266
 267         /* XXX: Can this be exploited? Not from o2dlm... */
 268         BUG_ON(rm->rm_used >= osb->max_slots);
 269
 270         rm->rm_entries[rm->rm_used] = node_num;
 271         rm->rm_used++;
 272         spin_unlock(&osb->osb_lock);
 273
 274         return 0;
 275 }
 276
 277 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
 278                                      unsigned int node_num)
 279 {
 280         int i;
 281         struct ocfs2_recovery_map *rm = osb->recovery_map;
 282
 283         spin_lock(&osb->osb_lock);
 284
 285         for (i = 0; i < rm->rm_used; i++) {
 286                 if (rm->rm_entries[i] == node_num)
 287                         break;
 288         }
 289
 290         if (i < rm->rm_used) {
 291                 /* XXX: be careful with the pointer math */
 292                 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
 293                         (rm->rm_used - i - 1) * sizeof(unsigned int));
 294                 rm->rm_used--;
 295         }
 296
 297         spin_unlock(&osb->osb_lock);
 298 }
 299
 300 static int ocfs2_commit_cache(struct ocfs2_super *osb)
 301 {
 302         int status = 0;
 303         unsigned int flushed;
 304         struct ocfs2_journal *journal = NULL;
 305
 306         journal = osb->journal;
 307
 308         /* Flush all pending commits and checkpoint the journal. */
 309         down_write(&journal->j_trans_barrier);
 310
 311         flushed = atomic_read(&journal->j_num_trans);
 312         trace_ocfs2_commit_cache_begin(flushed);
 313         if (flushed == 0) {
 314                 up_write(&journal->j_trans_barrier);
 315                 goto finally;
 316         }
 317
 318         jbd2_journal_lock_updates(journal->j_journal);
 319         status = jbd2_journal_flush(journal->j_journal);
 320         jbd2_journal_unlock_updates(journal->j_journal);
 321         if (status < 0) {
 322                 up_write(&journal->j_trans_barrier);
 323                 mlog_errno(status);
 324                 goto finally;
 325         }
 326
 327         ocfs2_inc_trans_id(journal);
 328
 329         flushed = atomic_read(&journal->j_num_trans);
 330         atomic_set(&journal->j_num_trans, 0);
 331         up_write(&journal->j_trans_barrier);
 332
 333         trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
 334
 335         ocfs2_wake_downconvert_thread(osb);
 336         wake_up(&journal->j_checkpointed);
 337 finally:
 338         return status;
 339 }
 340
 341 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
 342 {
 343         journal_t *journal = osb->journal->j_journal;
 344         handle_t *handle;
 345
 346         BUG_ON(!osb || !osb->journal->j_journal);
 347
 348         if (ocfs2_is_hard_readonly(osb))
 349                 return ERR_PTR(-EROFS);
 350
 351         BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
 352         BUG_ON(max_buffs <= 0);
 353
 354         /* Nested transaction? Just return the handle... */
 355         if (journal_current_handle())
 356                 return jbd2_journal_start(journal, max_buffs);
 357
 358         down_read(&osb->journal->j_trans_barrier);
 359
 360         handle = jbd2_journal_start(journal, max_buffs);
 361         if (IS_ERR(handle)) {
 362                 up_read(&osb->journal->j_trans_barrier);
 363
 364                 mlog_errno(PTR_ERR(handle));
 365
 366                 if (is_journal_aborted(journal)) {
 367                         ocfs2_abort(osb->sb, "Detected aborted journal");
 368                         handle = ERR_PTR(-EROFS);
 369                 }
 370         } else {
 371                 if (!ocfs2_mount_local(osb))
 372                         atomic_inc(&(osb->journal->j_num_trans));
 373         }
 374
 375         return handle;
 376 }
 377
 378 int ocfs2_commit_trans(struct ocfs2_super *osb,
 379                        handle_t *handle)
 380 {
 381         int ret, nested;
 382         struct ocfs2_journal *journal = osb->journal;
 383
 384         BUG_ON(!handle);
 385
 386         nested = handle->h_ref > 1;
 387         ret = jbd2_journal_stop(handle);
 388         if (ret < 0)
 389                 mlog_errno(ret);
 390
 391         if (!nested)
 392                 up_read(&journal->j_trans_barrier);
 393
 394         return ret;
 395 }
 396
 397 /*
 398  * 'nblocks' is what you want to add to the current transaction.
 399  *
 400  * This might call jbd2_journal_restart() which will commit dirty buffers
 401  * and then restart the transaction. Before calling
 402  * ocfs2_extend_trans(), any changed blocks should have been
 403  * dirtied. After calling it, all blocks which need to be changed must
 404  * go through another set of journal_access/journal_dirty calls.
 405  *
 406  * WARNING: This will not release any semaphores or disk locks taken
 407  * during the transaction, so make sure they were taken *before*
 408  * start_trans or we'll have ordering deadlocks.
 409  *
 410  * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
 411  * good because transaction ids haven't yet been recorded on the
 412  * cluster locks associated with this handle.
 413  */
 414 int ocfs2_extend_trans(handle_t *handle, int nblocks)
 415 {
 416         int status, old_nblocks;
 417
 418         BUG_ON(!handle);
 419         BUG_ON(nblocks < 0);
 420
 421         if (!nblocks)
 422                 return 0;
 423
 424         old_nblocks = handle->h_buffer_credits;
 425
 426         trace_ocfs2_extend_trans(old_nblocks, nblocks);
 427
 428 #ifdef CONFIG_OCFS2_DEBUG_FS
 429         status = 1;
 430 #else
 431         status = jbd2_journal_extend(handle, nblocks);
 432         if (status < 0) {
 433                 mlog_errno(status);
 434                 goto bail;
 435         }
 436 #endif
 437
 438         if (status > 0) {
 439                 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
 440                 status = jbd2_journal_restart(handle,
 441                                               old_nblocks + nblocks);
 442                 if (status < 0) {
 443                         mlog_errno(status);
 444                         goto bail;
 445                 }
 446         }
 447
 448         status = 0;
 449 bail:
 450         return status;
 451 }
 452
 453 struct ocfs2_triggers {
 454         struct jbd2_buffer_trigger_type ot_triggers;
 455         int                             ot_offset;
 456 };
 457
 458 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
 459 {
 460         return container_of(triggers, struct ocfs2_triggers, ot_triggers);
 461 }
 462
 463 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
 464                                  struct buffer_head *bh,
 465                                  void *data, size_t size)
 466 {
 467         struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
 468
 469         /*
 470          * We aren't guaranteed to have the superblock here, so we
 471          * must unconditionally compute the ecc data.
 472          * __ocfs2_journal_access() will only set the triggers if
 473          * metaecc is enabled.
 474          */
 475         ocfs2_block_check_compute(data, size, data + ot->ot_offset);
 476 }
 477
 478 /*
 479  * Quota blocks have their own trigger because the struct ocfs2_block_check
 480  * offset depends on the blocksize.
 481  */
 482 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
 483                                  struct buffer_head *bh,
 484                                  void *data, size_t size)
 485 {
 486         struct ocfs2_disk_dqtrailer *dqt =
 487                 ocfs2_block_dqtrailer(size, data);
 488
 489         /*
 490          * We aren't guaranteed to have the superblock here, so we
 491          * must unconditionally compute the ecc data.
 492          * __ocfs2_journal_access() will only set the triggers if
 493          * metaecc is enabled.
 494          */
 495         ocfs2_block_check_compute(data, size, &dqt->dq_check);
 496 }
 497
 498 /*
 499  * Directory blocks also have their own trigger because the
 500  * struct ocfs2_block_check offset depends on the blocksize.
 501  */
 502 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
 503                                  struct buffer_head *bh,
 504                                  void *data, size_t size)
 505 {
 506         struct ocfs2_dir_block_trailer *trailer =
 507                 ocfs2_dir_trailer_from_size(size, data);
 508
 509         /*
 510          * We aren't guaranteed to have the superblock here, so we
 511          * must unconditionally compute the ecc data.
 512          * __ocfs2_journal_access() will only set the triggers if
 513          * metaecc is enabled.
 514          */
 515         ocfs2_block_check_compute(data, size, &trailer->db_check);
 516 }
 517
 518 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
 519                                 struct buffer_head *bh)
 520 {
 521         mlog(ML_ERROR,
 522              "ocfs2_abort_trigger called by JBD2.  bh = 0x%lx, "
 523              "bh->b_blocknr = %llu\n",
 524              (unsigned long)bh,
 525              (unsigned long long)bh->b_blocknr);
 526
 527         /* We aren't guaranteed to have the superblock here - but if we
 528          * don't, it'll just crash. */
 529         ocfs2_error(bh->b_assoc_map->host->i_sb,
 530                     "JBD2 has aborted our journal, ocfs2 cannot continue\n");
 531 }
 532
 533 static struct ocfs2_triggers di_triggers = {
 534         .ot_triggers = {
 535                 .t_frozen = ocfs2_frozen_trigger,
 536                 .t_abort = ocfs2_abort_trigger,
 537         },
 538         .ot_offset      = offsetof(struct ocfs2_dinode, i_check),
 539 };
 540
 541 static struct ocfs2_triggers eb_triggers = {
 542         .ot_triggers = {
 543                 .t_frozen = ocfs2_frozen_trigger,
 544                 .t_abort = ocfs2_abort_trigger,
 545         },
 546         .ot_offset      = offsetof(struct ocfs2_extent_block, h_check),
 547 };
 548
 549 static struct ocfs2_triggers rb_triggers = {
 550         .ot_triggers = {
 551                 .t_frozen = ocfs2_frozen_trigger,
 552                 .t_abort = ocfs2_abort_trigger,
 553         },
 554         .ot_offset      = offsetof(struct ocfs2_refcount_block, rf_check),
 555 };
 556
 557 static struct ocfs2_triggers gd_triggers = {
 558         .ot_triggers = {
 559                 .t_frozen = ocfs2_frozen_trigger,
 560                 .t_abort = ocfs2_abort_trigger,
 561         },
 562         .ot_offset      = offsetof(struct ocfs2_group_desc, bg_check),
 563 };
 564
 565 static struct ocfs2_triggers db_triggers = {
 566         .ot_triggers = {
 567                 .t_frozen = ocfs2_db_frozen_trigger,
 568                 .t_abort = ocfs2_abort_trigger,
 569         },
 570 };
 571
 572 static struct ocfs2_triggers xb_triggers = {
 573         .ot_triggers = {
 574                 .t_frozen = ocfs2_frozen_trigger,
 575                 .t_abort = ocfs2_abort_trigger,
 576         },
 577         .ot_offset      = offsetof(struct ocfs2_xattr_block, xb_check),
 578 };
 579
 580 static struct ocfs2_triggers dq_triggers = {
 581         .ot_triggers = {
 582                 .t_frozen = ocfs2_dq_frozen_trigger,
 583                 .t_abort = ocfs2_abort_trigger,
 584         },
 585 };
 586
 587 static struct ocfs2_triggers dr_triggers = {
 588         .ot_triggers = {
 589                 .t_frozen = ocfs2_frozen_trigger,
 590                 .t_abort = ocfs2_abort_trigger,
 591         },
 592         .ot_offset      = offsetof(struct ocfs2_dx_root_block, dr_check),
 593 };
 594
 595 static struct ocfs2_triggers dl_triggers = {
 596         .ot_triggers = {
 597                 .t_frozen = ocfs2_frozen_trigger,
 598                 .t_abort = ocfs2_abort_trigger,
 599         },
 600         .ot_offset      = offsetof(struct ocfs2_dx_leaf, dl_check),
 601 };
 602
 603 static int __ocfs2_journal_access(handle_t *handle,
 604                                   struct ocfs2_caching_info *ci,
 605                                   struct buffer_head *bh,
 606                                   struct ocfs2_triggers *triggers,
 607                                   int type)
 608 {
 609         int status;
 610         struct ocfs2_super *osb =
 611                 OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
 612
 613         BUG_ON(!ci || !ci->ci_ops);
 614         BUG_ON(!handle);
 615         BUG_ON(!bh);
 616
 617         trace_ocfs2_journal_access(
 618                 (unsigned long long)ocfs2_metadata_cache_owner(ci),
 619                 (unsigned long long)bh->b_blocknr, type, bh->b_size);
 620
 621         /* we can safely remove this assertion after testing. */
 622         if (!buffer_uptodate(bh)) {
 623                 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
 624                 mlog(ML_ERROR, "b_blocknr=%llu\n",
 625                      (unsigned long long)bh->b_blocknr);
 626                 BUG();
 627         }
 628
 629         /* Set the current transaction information on the ci so
 630          * that the locking code knows whether it can drop it's locks
 631          * on this ci or not. We're protected from the commit
 632          * thread updating the current transaction id until
 633          * ocfs2_commit_trans() because ocfs2_start_trans() took
 634          * j_trans_barrier for us. */
 635         ocfs2_set_ci_lock_trans(osb->journal, ci);
 636
 637         ocfs2_metadata_cache_io_lock(ci);
 638         switch (type) {
 639         case OCFS2_JOURNAL_ACCESS_CREATE:
 640         case OCFS2_JOURNAL_ACCESS_WRITE:
 641                 status = jbd2_journal_get_write_access(handle, bh);
 642                 break;
 643
 644         case OCFS2_JOURNAL_ACCESS_UNDO:
 645                 status = jbd2_journal_get_undo_access(handle, bh);
 646                 break;
 647
 648         default:
 649                 status = -EINVAL;
 650                 mlog(ML_ERROR, "Unknown access type!\n");
 651         }
 652         if (!status && ocfs2_meta_ecc(osb) && triggers)
 653                 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
 654         ocfs2_metadata_cache_io_unlock(ci);
 655
 656         if (status < 0)
 657                 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
 658                      status, type);
 659
 660         return status;
 661 }
 662
 663 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
 664                             struct buffer_head *bh, int type)
 665 {
 666         return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
 667 }
 668
 669 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
 670                             struct buffer_head *bh, int type)
 671 {
 672         return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
 673 }
 674
 675 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
 676                             struct buffer_head *bh, int type)
 677 {
 678         return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
 679                                       type);
 680 }
 681
 682 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
 683                             struct buffer_head *bh, int type)
 684 {
 685         return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
 686 }
 687
 688 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
 689                             struct buffer_head *bh, int type)
 690 {
 691         return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
 692 }
 693
 694 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
 695                             struct buffer_head *bh, int type)
 696 {
 697         return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
 698 }
 699
 700 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
 701                             struct buffer_head *bh, int type)
 702 {
 703         return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
 704 }
 705
 706 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
 707                             struct buffer_head *bh, int type)
 708 {
 709         return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
 710 }
 711
 712 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
 713                             struct buffer_head *bh, int type)
 714 {
 715         return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
 716 }
 717
 718 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
 719                          struct buffer_head *bh, int type)
 720 {
 721         return __ocfs2_journal_access(handle, ci, bh, NULL, type);
 722 }
 723
 724 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
 725 {
 726         int status;
 727
 728         trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
 729
 730         status = jbd2_journal_dirty_metadata(handle, bh);
 731         BUG_ON(status);
 732 }
 733
 734 #define OCFS2_DEFAULT_COMMIT_INTERVAL   (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
 735
 736 void ocfs2_set_journal_params(struct ocfs2_super *osb)
 737 {
 738         journal_t *journal = osb->journal->j_journal;
 739         unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
 740
 741         if (osb->osb_commit_interval)
 742                 commit_interval = osb->osb_commit_interval;
 743
 744         write_lock(&journal->j_state_lock);
 745         journal->j_commit_interval = commit_interval;
 746         if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
 747                 journal->j_flags |= JBD2_BARRIER;
 748         else
 749                 journal->j_flags &= ~JBD2_BARRIER;
 750         write_unlock(&journal->j_state_lock);
 751 }
 752
 753 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
 754 {
 755         int status = -1;
 756         struct inode *inode = NULL; /* the journal inode */
 757         journal_t *j_journal = NULL;
 758         struct ocfs2_dinode *di = NULL;
 759         struct buffer_head *bh = NULL;
 760         struct ocfs2_super *osb;
 761         int inode_lock = 0;
 762
 763         BUG_ON(!journal);
 764
 765         osb = journal->j_osb;
 766
 767         /* already have the inode for our journal */
 768         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
 769                                             osb->slot_num);
 770         if (inode == NULL) {
 771                 status = -EACCES;
 772                 mlog_errno(status);
 773                 goto done;
 774         }
 775         if (is_bad_inode(inode)) {
 776                 mlog(ML_ERROR, "access error (bad inode)\n");
 777                 iput(inode);
 778                 inode = NULL;
 779                 status = -EACCES;
 780                 goto done;
 781         }
 782
 783         SET_INODE_JOURNAL(inode);
 784         OCFS2_I(inode)->ip_open_count++;
 785
 786         /* Skip recovery waits here - journal inode metadata never
 787          * changes in a live cluster so it can be considered an
 788          * exception to the rule. */
 789         status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
 790         if (status < 0) {
 791                 if (status != -ERESTARTSYS)
 792                         mlog(ML_ERROR, "Could not get lock on journal!\n");
 793                 goto done;
 794         }
 795
 796         inode_lock = 1;
 797         di = (struct ocfs2_dinode *)bh->b_data;
 798
 799         if (inode->i_size <  OCFS2_MIN_JOURNAL_SIZE) {
 800                 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
 801                      inode->i_size);
 802                 status = -EINVAL;
 803                 goto done;
 804         }
 805
 806         trace_ocfs2_journal_init(inode->i_size,
 807                                  (unsigned long long)inode->i_blocks,
 808                                  OCFS2_I(inode)->ip_clusters);
 809
 810         /* call the kernels journal init function now */
 811         j_journal = jbd2_journal_init_inode(inode);
 812         if (j_journal == NULL) {
 813                 mlog(ML_ERROR, "Linux journal layer error\n");
 814                 status = -EINVAL;
 815                 goto done;
 816         }
 817
 818         trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
 819
 820         *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
 821                   OCFS2_JOURNAL_DIRTY_FL);
 822
 823         journal->j_journal = j_journal;
 824         journal->j_inode = inode;
 825         journal->j_bh = bh;
 826
 827         ocfs2_set_journal_params(osb);
 828
 829         journal->j_state = OCFS2_JOURNAL_LOADED;
 830
 831         status = 0;
 832 done:
 833         if (status < 0) {
 834                 if (inode_lock)
 835                         ocfs2_inode_unlock(inode, 1);
 836                 brelse(bh);
 837                 if (inode) {
 838                         OCFS2_I(inode)->ip_open_count--;
 839                         iput(inode);
 840                 }
 841         }
 842
 843         return status;
 844 }
 845
 846 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
 847 {
 848         le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
 849 }
 850
 851 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
 852 {
 853         return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
 854 }
 855
 856 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
 857                                       int dirty, int replayed)
 858 {
 859         int status;
 860         unsigned int flags;
 861         struct ocfs2_journal *journal = osb->journal;
 862         struct buffer_head *bh = journal->j_bh;
 863         struct ocfs2_dinode *fe;
 864
 865         fe = (struct ocfs2_dinode *)bh->b_data;
 866
 867         /* The journal bh on the osb always comes from ocfs2_journal_init()
 868          * and was validated there inside ocfs2_inode_lock_full().  It's a
 869          * code bug if we mess it up. */
 870         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
 871
 872         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
 873         if (dirty)
 874                 flags |= OCFS2_JOURNAL_DIRTY_FL;
 875         else
 876                 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
 877         fe->id1.journal1.ij_flags = cpu_to_le32(flags);
 878
 879         if (replayed)
 880                 ocfs2_bump_recovery_generation(fe);
 881
 882         ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
 883         status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
 884         if (status < 0)
 885                 mlog_errno(status);
 886
 887         return status;
 888 }
 889
 890 /*
 891  * If the journal has been kmalloc'd it needs to be freed after this
 892  * call.
 893  */
 894 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
 895 {
 896         struct ocfs2_journal *journal = NULL;
 897         int status = 0;
 898         struct inode *inode = NULL;
 899         int num_running_trans = 0;
 900
 901         BUG_ON(!osb);
 902
 903         journal = osb->journal;
 904         if (!journal)
 905                 goto done;
 906
 907         inode = journal->j_inode;
 908
 909         if (journal->j_state != OCFS2_JOURNAL_LOADED)
 910                 goto done;
 911
 912         /* need to inc inode use count - jbd2_journal_destroy will iput. */
 913         if (!igrab(inode))
 914                 BUG();
 915
 916         num_running_trans = atomic_read(&(osb->journal->j_num_trans));
 917         trace_ocfs2_journal_shutdown(num_running_trans);
 918
 919         /* Do a commit_cache here. It will flush our journal, *and*
 920          * release any locks that are still held.
 921          * set the SHUTDOWN flag and release the trans lock.
 922          * the commit thread will take the trans lock for us below. */
 923         journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
 924
 925         /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
 926          * drop the trans_lock (which we want to hold until we
 927          * completely destroy the journal. */
 928         if (osb->commit_task) {
 929                 /* Wait for the commit thread */
 930                 trace_ocfs2_journal_shutdown_wait(osb->commit_task);
 931                 kthread_stop(osb->commit_task);
 932                 osb->commit_task = NULL;
 933         }
 934
 935         BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
 936
 937         if (ocfs2_mount_local(osb)) {
 938                 jbd2_journal_lock_updates(journal->j_journal);
 939                 status = jbd2_journal_flush(journal->j_journal);
 940                 jbd2_journal_unlock_updates(journal->j_journal);
 941                 if (status < 0)
 942                         mlog_errno(status);
 943         }
 944
 945         if (status == 0) {
 946                 /*
 947                  * Do not toggle if flush was unsuccessful otherwise
 948                  * will leave dirty metadata in a "clean" journal
 949                  */
 950                 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
 951                 if (status < 0)
 952                         mlog_errno(status);
 953         }
 954
 955         /* Shutdown the kernel journal system */
 956         jbd2_journal_destroy(journal->j_journal);
 957         journal->j_journal = NULL;
 958
 959         OCFS2_I(inode)->ip_open_count--;
 960
 961         /* unlock our journal */
 962         ocfs2_inode_unlock(inode, 1);
 963
 964         brelse(journal->j_bh);
 965         journal->j_bh = NULL;
 966
 967         journal->j_state = OCFS2_JOURNAL_FREE;
 968
 969 //      up_write(&journal->j_trans_barrier);
 970 done:
 971         if (inode)
 972                 iput(inode);
 973 }
 974
 975 static void ocfs2_clear_journal_error(struct super_block *sb,
 976                                       journal_t *journal,
 977                                       int slot)
 978 {
 979         int olderr;
 980
 981         olderr = jbd2_journal_errno(journal);
 982         if (olderr) {
 983                 mlog(ML_ERROR, "File system error %d recorded in "
 984                      "journal %u.\n", olderr, slot);
 985                 mlog(ML_ERROR, "File system on device %s needs checking.\n",
 986                      sb->s_id);
 987
 988                 jbd2_journal_ack_err(journal);
 989                 jbd2_journal_clear_err(journal);
 990         }
 991 }
 992
 993 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
 994 {
 995         int status = 0;
 996         struct ocfs2_super *osb;
 997
 998         BUG_ON(!journal);
 999
1000         osb = journal->j_osb;
1001
1002         status = jbd2_journal_load(journal->j_journal);
1003         if (status < 0) {
1004                 mlog(ML_ERROR, "Failed to load journal!\n");
1005                 goto done;
1006         }
1007
1008         ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1009
1010         status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1011         if (status < 0) {
1012                 mlog_errno(status);
1013                 goto done;
1014         }
1015
1016         /* Launch the commit thread */
1017         if (!local) {
1018                 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1019                                                "ocfs2cmt");
1020                 if (IS_ERR(osb->commit_task)) {
1021                         status = PTR_ERR(osb->commit_task);
1022                         osb->commit_task = NULL;
1023                         mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1024                              "error=%d", status);
1025                         goto done;
1026                 }
1027         } else
1028                 osb->commit_task = NULL;
1029
1030 done:
1031         return status;
1032 }
1033
1034
1035 /* 'full' flag tells us whether we clear out all blocks or if we just
1036  * mark the journal clean */
1037 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1038 {
1039         int status;
1040
1041         BUG_ON(!journal);
1042
1043         status = jbd2_journal_wipe(journal->j_journal, full);
1044         if (status < 0) {
1045                 mlog_errno(status);
1046                 goto bail;
1047         }
1048
1049         status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1050         if (status < 0)
1051                 mlog_errno(status);
1052
1053 bail:
1054         return status;
1055 }
1056
1057 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1058 {
1059         int empty;
1060         struct ocfs2_recovery_map *rm = osb->recovery_map;
1061
1062         spin_lock(&osb->osb_lock);
1063         empty = (rm->rm_used == 0);
1064         spin_unlock(&osb->osb_lock);
1065
1066         return empty;
1067 }
1068
1069 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1070 {
1071         wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1072 }
1073
1074 /*
1075  * JBD Might read a cached version of another nodes journal file. We
1076  * don't want this as this file changes often and we get no
1077  * notification on those changes. The only way to be sure that we've
1078  * got the most up to date version of those blocks then is to force
1079  * read them off disk. Just searching through the buffer cache won't
1080  * work as there may be pages backing this file which are still marked
1081  * up to date. We know things can't change on this file underneath us
1082  * as we have the lock by now :)
1083  */
1084 static int ocfs2_force_read_journal(struct inode *inode)
1085 {
1086         int status = 0;
1087         int i;
1088         u64 v_blkno, p_blkno, p_blocks, num_blocks;
1089 #define CONCURRENT_JOURNAL_FILL 32ULL
1090         struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1091
1092         memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1093
1094         num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
1095         v_blkno = 0;
1096         while (v_blkno < num_blocks) {
1097                 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1098                                                      &p_blkno, &p_blocks, NULL);
1099                 if (status < 0) {
1100                         mlog_errno(status);
1101                         goto bail;
1102                 }
1103
1104                 if (p_blocks > CONCURRENT_JOURNAL_FILL)
1105                         p_blocks = CONCURRENT_JOURNAL_FILL;
1106
1107                 /* We are reading journal data which should not
1108                  * be put in the uptodate cache */
1109                 status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1110                                                 p_blkno, p_blocks, bhs);
1111                 if (status < 0) {
1112                         mlog_errno(status);
1113                         goto bail;
1114                 }
1115
1116                 for(i = 0; i < p_blocks; i++) {
1117                         brelse(bhs[i]);
1118                         bhs[i] = NULL;
1119                 }
1120
1121                 v_blkno += p_blocks;
1122         }
1123
1124 bail:
1125         for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1126                 brelse(bhs[i]);
1127         return status;
1128 }
1129
1130 struct ocfs2_la_recovery_item {
1131         struct list_head        lri_list;
1132         int                     lri_slot;
1133         struct ocfs2_dinode     *lri_la_dinode;
1134         struct ocfs2_dinode     *lri_tl_dinode;
1135         struct ocfs2_quota_recovery *lri_qrec;
1136 };
1137
1138 /* Does the second half of the recovery process. By this point, the
1139  * node is marked clean and can actually be considered recovered,
1140  * hence it's no longer in the recovery map, but there's still some
1141  * cleanup we can do which shouldn't happen within the recovery thread
1142  * as locking in that context becomes very difficult if we are to take
1143  * recovering nodes into account.
1144  *
1145  * NOTE: This function can and will sleep on recovery of other nodes
1146  * during cluster locking, just like any other ocfs2 process.
1147  */
1148 void ocfs2_complete_recovery(struct work_struct *work)
1149 {
1150         int ret = 0;
1151         struct ocfs2_journal *journal =
1152                 container_of(work, struct ocfs2_journal, j_recovery_work);
1153         struct ocfs2_super *osb = journal->j_osb;
1154         struct ocfs2_dinode *la_dinode, *tl_dinode;
1155         struct ocfs2_la_recovery_item *item, *n;
1156         struct ocfs2_quota_recovery *qrec;
1157         LIST_HEAD(tmp_la_list);
1158
1159         trace_ocfs2_complete_recovery(
1160                 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1161
1162         spin_lock(&journal->j_lock);
1163         list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1164         spin_unlock(&journal->j_lock);
1165
1166         list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1167                 list_del_init(&item->lri_list);
1168
1169                 ocfs2_wait_on_quotas(osb);
1170
1171                 la_dinode = item->lri_la_dinode;
1172                 tl_dinode = item->lri_tl_dinode;
1173                 qrec = item->lri_qrec;
1174
1175                 trace_ocfs2_complete_recovery_slot(item->lri_slot,
1176                         la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1177                         tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1178                         qrec);
1179
1180                 if (la_dinode) {
1181                         ret = ocfs2_complete_local_alloc_recovery(osb,
1182                                                                   la_dinode);
1183                         if (ret < 0)
1184                                 mlog_errno(ret);
1185
1186                         kfree(la_dinode);
1187                 }
1188
1189                 if (tl_dinode) {
1190                         ret = ocfs2_complete_truncate_log_recovery(osb,
1191                                                                    tl_dinode);
1192                         if (ret < 0)
1193                                 mlog_errno(ret);
1194
1195                         kfree(tl_dinode);
1196                 }
1197
1198                 ret = ocfs2_recover_orphans(osb, item->lri_slot);
1199                 if (ret < 0)
1200                         mlog_errno(ret);
1201
1202                 if (qrec) {
1203                         ret = ocfs2_finish_quota_recovery(osb, qrec,
1204                                                           item->lri_slot);
1205                         if (ret < 0)
1206                                 mlog_errno(ret);
1207                         /* Recovery info is already freed now */
1208                 }
1209
1210                 kfree(item);
1211         }
1212
1213         trace_ocfs2_complete_recovery_end(ret);
1214 }
1215
1216 /* NOTE: This function always eats your references to la_dinode and
1217  * tl_dinode, either manually on error, or by passing them to
1218  * ocfs2_complete_recovery */
1219 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1220                                             int slot_num,
1221                                             struct ocfs2_dinode *la_dinode,
1222                                             struct ocfs2_dinode *tl_dinode,
1223                                             struct ocfs2_quota_recovery *qrec)
1224 {
1225         struct ocfs2_la_recovery_item *item;
1226
1227         item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1228         if (!item) {
1229                 /* Though we wish to avoid it, we are in fact safe in
1230                  * skipping local alloc cleanup as fsck.ocfs2 is more
1231                  * than capable of reclaiming unused space. */
1232                 if (la_dinode)
1233                         kfree(la_dinode);
1234
1235                 if (tl_dinode)
1236                         kfree(tl_dinode);
1237
1238                 if (qrec)
1239                         ocfs2_free_quota_recovery(qrec);
1240
1241                 mlog_errno(-ENOMEM);
1242                 return;
1243         }
1244
1245         INIT_LIST_HEAD(&item->lri_list);
1246         item->lri_la_dinode = la_dinode;
1247         item->lri_slot = slot_num;
1248         item->lri_tl_dinode = tl_dinode;
1249         item->lri_qrec = qrec;
1250
1251         spin_lock(&journal->j_lock);
1252         list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1253         queue_work(ocfs2_wq, &journal->j_recovery_work);
1254         spin_unlock(&journal->j_lock);
1255 }
1256
1257 /* Called by the mount code to queue recovery the last part of
1258  * recovery for it's own and offline slot(s). */
1259 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1260 {
1261         struct ocfs2_journal *journal = osb->journal;
1262
1263         /* No need to queue up our truncate_log as regular cleanup will catch
1264          * that */
1265         ocfs2_queue_recovery_completion(journal, osb->slot_num,
1266                                         osb->local_alloc_copy, NULL, NULL);
1267         ocfs2_schedule_truncate_log_flush(osb, 0);
1268
1269         osb->local_alloc_copy = NULL;
1270         osb->dirty = 0;
1271
1272         /* queue to recover orphan slots for all offline slots */
1273         ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1274         ocfs2_queue_replay_slots(osb);
1275         ocfs2_free_replay_slots(osb);
1276 }
1277
1278 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1279 {
1280         if (osb->quota_rec) {
1281                 ocfs2_queue_recovery_completion(osb->journal,
1282                                                 osb->slot_num,
1283                                                 NULL,
1284                                                 NULL,
1285                                                 osb->quota_rec);
1286                 osb->quota_rec = NULL;
1287         }
1288 }
1289
1290 static int __ocfs2_recovery_thread(void *arg)
1291 {
1292         int status, node_num, slot_num;
1293         struct ocfs2_super *osb = arg;
1294         struct ocfs2_recovery_map *rm = osb->recovery_map;
1295         int *rm_quota = NULL;
1296         int rm_quota_used = 0, i;
1297         struct ocfs2_quota_recovery *qrec;
1298
1299         status = ocfs2_wait_on_mount(osb);
1300         if (status < 0) {
1301                 goto bail;
1302         }
1303
1304         rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1305         if (!rm_quota) {
1306                 status = -ENOMEM;
1307                 goto bail;
1308         }
1309 restart:
1310         status = ocfs2_super_lock(osb, 1);
1311         if (status < 0) {
1312                 mlog_errno(status);
1313                 goto bail;
1314         }
1315
1316         status = ocfs2_compute_replay_slots(osb);
1317         if (status < 0)
1318                 mlog_errno(status);
1319
1320         /* queue recovery for our own slot */
1321         ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1322                                         NULL, NULL);
1323
1324         spin_lock(&osb->osb_lock);
1325         while (rm->rm_used) {
1326                 /* It's always safe to remove entry zero, as we won't
1327                  * clear it until ocfs2_recover_node() has succeeded. */
1328                 node_num = rm->rm_entries[0];
1329                 spin_unlock(&osb->osb_lock);
1330                 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1331                 trace_ocfs2_recovery_thread_node(node_num, slot_num);
1332                 if (slot_num == -ENOENT) {
1333                         status = 0;
1334                         goto skip_recovery;
1335                 }
1336
1337                 /* It is a bit subtle with quota recovery. We cannot do it
1338                  * immediately because we have to obtain cluster locks from
1339                  * quota files and we also don't want to just skip it because
1340                  * then quota usage would be out of sync until some node takes
1341                  * the slot. So we remember which nodes need quota recovery
1342                  * and when everything else is done, we recover quotas. */
1343                 for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1344                 if (i == rm_quota_used)
1345                         rm_quota[rm_quota_used++] = slot_num;
1346
1347                 status = ocfs2_recover_node(osb, node_num, slot_num);
1348 skip_recovery:
1349                 if (!status) {
1350                         ocfs2_recovery_map_clear(osb, node_num);
1351                 } else {
1352                         mlog(ML_ERROR,
1353                              "Error %d recovering node %d on device (%u,%u)!\n",
1354                              status, node_num,
1355                              MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1356                         mlog(ML_ERROR, "Volume requires unmount.\n");
1357                 }
1358
1359                 spin_lock(&osb->osb_lock);
1360         }
1361         spin_unlock(&osb->osb_lock);
1362         trace_ocfs2_recovery_thread_end(status);
1363
1364         /* Refresh all journal recovery generations from disk */
1365         status = ocfs2_check_journals_nolocks(osb);
1366         status = (status == -EROFS) ? 0 : status;
1367         if (status < 0)
1368                 mlog_errno(status);
1369
1370         /* Now it is right time to recover quotas... We have to do this under
1371          * superblock lock so that no one can start using the slot (and crash)
1372          * before we recover it */
1373         for (i = 0; i < rm_quota_used; i++) {
1374                 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1375                 if (IS_ERR(qrec)) {
1376                         status = PTR_ERR(qrec);
1377                         mlog_errno(status);
1378                         continue;
1379                 }
1380                 ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1381                                                 NULL, NULL, qrec);
1382         }
1383
1384         ocfs2_super_unlock(osb, 1);
1385
1386         /* queue recovery for offline slots */
1387         ocfs2_queue_replay_slots(osb);
1388
1389 bail:
1390         mutex_lock(&osb->recovery_lock);
1391         if (!status && !ocfs2_recovery_completed(osb)) {
1392                 mutex_unlock(&osb->recovery_lock);
1393                 goto restart;
1394         }
1395
1396         ocfs2_free_replay_slots(osb);
1397         osb->recovery_thread_task = NULL;
1398         mb(); /* sync with ocfs2_recovery_thread_running */
1399         wake_up(&osb->recovery_event);
1400
1401         mutex_unlock(&osb->recovery_lock);
1402
1403         if (rm_quota)
1404                 kfree(rm_quota);
1405
1406         /* no one is callint kthread_stop() for us so the kthread() api
1407          * requires that we call do_exit().  And it isn't exported, but
1408          * complete_and_exit() seems to be a minimal wrapper around it. */
1409         complete_and_exit(NULL, status);
1410         return status;
1411 }
1412
1413 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1414 {
1415         mutex_lock(&osb->recovery_lock);
1416
1417         trace_ocfs2_recovery_thread(node_num, osb->node_num,
1418                 osb->disable_recovery, osb->recovery_thread_task,
1419                 osb->disable_recovery ?
1420                 -1 : ocfs2_recovery_map_set(osb, node_num));
1421
1422         if (osb->disable_recovery)
1423                 goto out;
1424
1425         if (osb->recovery_thread_task)
1426                 goto out;
1427
1428         osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
1429                                                  "ocfs2rec");
1430         if (IS_ERR(osb->recovery_thread_task)) {
1431                 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1432                 osb->recovery_thread_task = NULL;
1433         }
1434
1435 out:
1436         mutex_unlock(&osb->recovery_lock);
1437         wake_up(&osb->recovery_event);
1438 }
1439
1440 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1441                                     int slot_num,
1442                                     struct buffer_head **bh,
1443                                     struct inode **ret_inode)
1444 {
1445         int status = -EACCES;
1446         struct inode *inode = NULL;
1447
1448         BUG_ON(slot_num >= osb->max_slots);
1449
1450         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1451                                             slot_num);
1452         if (!inode || is_bad_inode(inode)) {
1453                 mlog_errno(status);
1454                 goto bail;
1455         }
1456         SET_INODE_JOURNAL(inode);
1457
1458         status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1459         if (status < 0) {
1460                 mlog_errno(status);
1461                 goto bail;
1462         }
1463
1464         status = 0;
1465
1466 bail:
1467         if (inode) {
1468                 if (status || !ret_inode)
1469                         iput(inode);
1470                 else
1471                         *ret_inode = inode;
1472         }
1473         return status;
1474 }
1475
1476 /* Does the actual journal replay and marks the journal inode as
1477  * clean. Will only replay if the journal inode is marked dirty. */
1478 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1479                                 int node_num,
1480                                 int slot_num)
1481 {
1482         int status;
1483         int got_lock = 0;
1484         unsigned int flags;
1485         struct inode *inode = NULL;
1486         struct ocfs2_dinode *fe;
1487         journal_t *journal = NULL;
1488         struct buffer_head *bh = NULL;
1489         u32 slot_reco_gen;
1490
1491         status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1492         if (status) {
1493                 mlog_errno(status);
1494                 goto done;
1495         }
1496
1497         fe = (struct ocfs2_dinode *)bh->b_data;
1498         slot_reco_gen = ocfs2_get_recovery_generation(fe);
1499         brelse(bh);
1500         bh = NULL;
1501
1502         /*
1503          * As the fs recovery is asynchronous, there is a small chance that
1504          * another node mounted (and recovered) the slot before the recovery
1505          * thread could get the lock. To handle that, we dirty read the journal
1506          * inode for that slot to get the recovery generation. If it is
1507          * different than what we expected, the slot has been recovered.
1508          * If not, it needs recovery.
1509          */
1510         if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1511                 trace_ocfs2_replay_journal_recovered(slot_num,
1512                      osb->slot_recovery_generations[slot_num], slot_reco_gen);
1513                 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1514                 status = -EBUSY;
1515                 goto done;
1516         }
1517
1518         /* Continue with recovery as the journal has not yet been recovered */
1519
1520         status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1521         if (status < 0) {
1522                 trace_ocfs2_replay_journal_lock_err(status);
1523                 if (status != -ERESTARTSYS)
1524                         mlog(ML_ERROR, "Could not lock journal!\n");
1525                 goto done;
1526         }
1527         got_lock = 1;
1528
1529         fe = (struct ocfs2_dinode *) bh->b_data;
1530
1531         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1532         slot_reco_gen = ocfs2_get_recovery_generation(fe);
1533
1534         if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1535                 trace_ocfs2_replay_journal_skip(node_num);
1536                 /* Refresh recovery generation for the slot */
1537                 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1538                 goto done;
1539         }
1540
1541         /* we need to run complete recovery for offline orphan slots */
1542         ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1543
1544         mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
1545              node_num, slot_num,
1546              MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1547
1548         OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1549
1550         status = ocfs2_force_read_journal(inode);
1551         if (status < 0) {
1552                 mlog_errno(status);
1553                 goto done;
1554         }
1555
1556         journal = jbd2_journal_init_inode(inode);
1557         if (journal == NULL) {
1558                 mlog(ML_ERROR, "Linux journal layer error\n");
1559                 status = -EIO;
1560                 goto done;
1561         }
1562
1563         status = jbd2_journal_load(journal);
1564         if (status < 0) {
1565                 mlog_errno(status);
1566                 if (!igrab(inode))
1567                         BUG();
1568                 jbd2_journal_destroy(journal);
1569                 goto done;
1570         }
1571
1572         ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1573
1574         /* wipe the journal */
1575         jbd2_journal_lock_updates(journal);
1576         status = jbd2_journal_flush(journal);
1577         jbd2_journal_unlock_updates(journal);
1578         if (status < 0)
1579                 mlog_errno(status);
1580
1581         /* This will mark the node clean */
1582         flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1583         flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1584         fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1585
1586         /* Increment recovery generation to indicate successful recovery */
1587         ocfs2_bump_recovery_generation(fe);
1588         osb->slot_recovery_generations[slot_num] =
1589                                         ocfs2_get_recovery_generation(fe);
1590
1591         ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1592         status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1593         if (status < 0)
1594                 mlog_errno(status);
1595
1596         if (!igrab(inode))
1597                 BUG();
1598
1599         jbd2_journal_destroy(journal);
1600
1601 done:
1602         /* drop the lock on this nodes journal */
1603         if (got_lock)
1604                 ocfs2_inode_unlock(inode, 1);
1605
1606         if (inode)
1607                 iput(inode);
1608
1609         brelse(bh);
1610
1611         return status;
1612 }
1613
1614 /*
1615  * Do the most important parts of node recovery:
1616  *  - Replay it's journal
1617  *  - Stamp a clean local allocator file
1618  *  - Stamp a clean truncate log
1619  *  - Mark the node clean
1620  *
1621  * If this function completes without error, a node in OCFS2 can be
1622  * said to have been safely recovered. As a result, failure during the
1623  * second part of a nodes recovery process (local alloc recovery) is
1624  * far less concerning.
1625  */
1626 static int ocfs2_recover_node(struct ocfs2_super *osb,
1627                               int node_num, int slot_num)
1628 {
1629         int status = 0;
1630         struct ocfs2_dinode *la_copy = NULL;
1631         struct ocfs2_dinode *tl_copy = NULL;
1632
1633         trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1634
1635         /* Should not ever be called to recover ourselves -- in that
1636          * case we should've called ocfs2_journal_load instead. */
1637         BUG_ON(osb->node_num == node_num);
1638
1639         status = ocfs2_replay_journal(osb, node_num, slot_num);
1640         if (status < 0) {
1641                 if (status == -EBUSY) {
1642                         trace_ocfs2_recover_node_skip(slot_num, node_num);
1643                         status = 0;
1644                         goto done;
1645                 }
1646                 mlog_errno(status);
1647                 goto done;
1648         }
1649
1650         /* Stamp a clean local alloc file AFTER recovering the journal... */
1651         status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1652         if (status < 0) {
1653                 mlog_errno(status);
1654                 goto done;
1655         }
1656
1657         /* An error from begin_truncate_log_recovery is not
1658          * serious enough to warrant halting the rest of
1659          * recovery. */
1660         status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1661         if (status < 0)
1662                 mlog_errno(status);
1663
1664         /* Likewise, this would be a strange but ultimately not so
1665          * harmful place to get an error... */
1666         status = ocfs2_clear_slot(osb, slot_num);
1667         if (status < 0)
1668                 mlog_errno(status);
1669
1670         /* This will kfree the memory pointed to by la_copy and tl_copy */
1671         ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1672                                         tl_copy, NULL);
1673
1674         status = 0;
1675 done:
1676
1677         return status;
1678 }
1679
1680 /* Test node liveness by trylocking his journal. If we get the lock,
1681  * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1682  * still alive (we couldn't get the lock) and < 0 on error. */
1683 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1684                                  int slot_num)
1685 {
1686         int status, flags;
1687         struct inode *inode = NULL;
1688
1689         inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1690                                             slot_num);
1691         if (inode == NULL) {
1692                 mlog(ML_ERROR, "access error\n");
1693                 status = -EACCES;
1694                 goto bail;
1695         }
1696         if (is_bad_inode(inode)) {
1697                 mlog(ML_ERROR, "access error (bad inode)\n");
1698                 iput(inode);
1699                 inode = NULL;
1700                 status = -EACCES;
1701                 goto bail;
1702         }
1703         SET_INODE_JOURNAL(inode);
1704
1705         flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1706         status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1707         if (status < 0) {
1708                 if (status != -EAGAIN)
1709                         mlog_errno(status);
1710                 goto bail;
1711         }
1712
1713         ocfs2_inode_unlock(inode, 1);
1714 bail:
1715         if (inode)
1716                 iput(inode);
1717
1718         return status;
1719 }
1720
1721 /* Call this underneath ocfs2_super_lock. It also assumes that the
1722  * slot info struct has been updated from disk. */
1723 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1724 {
1725         unsigned int node_num;
1726         int status, i;
1727         u32 gen;
1728         struct buffer_head *bh = NULL;
1729         struct ocfs2_dinode *di;
1730
1731         /* This is called with the super block cluster lock, so we
1732          * know that the slot map can't change underneath us. */
1733
1734         for (i = 0; i < osb->max_slots; i++) {
1735                 /* Read journal inode to get the recovery generation */
1736                 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1737                 if (status) {
1738                         mlog_errno(status);
1739                         goto bail;
1740                 }
1741                 di = (struct ocfs2_dinode *)bh->b_data;
1742                 gen = ocfs2_get_recovery_generation(di);
1743                 brelse(bh);
1744                 bh = NULL;
1745
1746                 spin_lock(&osb->osb_lock);
1747                 osb->slot_recovery_generations[i] = gen;
1748
1749                 trace_ocfs2_mark_dead_nodes(i,
1750                                             osb->slot_recovery_generations[i]);
1751
1752                 if (i == osb->slot_num) {
1753                         spin_unlock(&osb->osb_lock);
1754                         continue;
1755                 }
1756
1757                 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1758                 if (status == -ENOENT) {
1759                         spin_unlock(&osb->osb_lock);
1760                         continue;
1761                 }
1762
1763                 if (__ocfs2_recovery_map_test(osb, node_num)) {
1764                         spin_unlock(&osb->osb_lock);
1765                         continue;
1766                 }
1767                 spin_unlock(&osb->osb_lock);
1768
1769                 /* Ok, we have a slot occupied by another node which
1770                  * is not in the recovery map. We trylock his journal
1771                  * file here to test if he's alive. */
1772                 status = ocfs2_trylock_journal(osb, i);
1773                 if (!status) {
1774                         /* Since we're called from mount, we know that
1775                          * the recovery thread can't race us on
1776                          * setting / checking the recovery bits. */
1777                         ocfs2_recovery_thread(osb, node_num);
1778                 } else if ((status < 0) && (status != -EAGAIN)) {
1779                         mlog_errno(status);
1780                         goto bail;
1781                 }
1782         }
1783
1784         status = 0;
1785 bail:
1786         return status;
1787 }
1788
1789 /*
1790  * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1791  * randomness to the timeout to minimize multple nodes firing the timer at the
1792  * same time.
1793  */
1794 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1795 {
1796         unsigned long time;
1797
1798         get_random_bytes(&time, sizeof(time));
1799         time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1800         return msecs_to_jiffies(time);
1801 }
1802
1803 /*
1804  * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1805  * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1806  * is done to catch any orphans that are left over in orphan directories.
1807  *
1808  * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1809  * seconds.  It gets an EX lock on os_lockres and checks sequence number
1810  * stored in LVB. If the sequence number has changed, it means some other
1811  * node has done the scan.  This node skips the scan and tracks the
1812  * sequence number.  If the sequence number didn't change, it means a scan
1813  * hasn't happened.  The node queues a scan and increments the
1814  * sequence number in the LVB.
1815  */
1816 void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1817 {
1818         struct ocfs2_orphan_scan *os;
1819         int status, i;
1820         u32 seqno = 0;
1821
1822         os = &osb->osb_orphan_scan;
1823
1824         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1825                 goto out;
1826
1827         trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1828                                             atomic_read(&os->os_state));
1829
1830         status = ocfs2_orphan_scan_lock(osb, &seqno);
1831         if (status < 0) {
1832                 if (status != -EAGAIN)
1833                         mlog_errno(status);
1834                 goto out;
1835         }
1836
1837         /* Do no queue the tasks if the volume is being umounted */
1838         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1839                 goto unlock;
1840
1841         if (os->os_seqno != seqno) {
1842                 os->os_seqno = seqno;
1843                 goto unlock;
1844         }
1845
1846         for (i = 0; i < osb->max_slots; i++)
1847                 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1848                                                 NULL);
1849         /*
1850          * We queued a recovery on orphan slots, increment the sequence
1851          * number and update LVB so other node will skip the scan for a while
1852          */
1853         seqno++;
1854         os->os_count++;
1855         os->os_scantime = CURRENT_TIME;
1856 unlock:
1857         ocfs2_orphan_scan_unlock(osb, seqno);
1858 out:
1859         trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1860                                           atomic_read(&os->os_state));
1861         return;
1862 }
1863
1864 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1865 void ocfs2_orphan_scan_work(struct work_struct *work)
1866 {
1867         struct ocfs2_orphan_scan *os;
1868         struct ocfs2_super *osb;
1869
1870         os = container_of(work, struct ocfs2_orphan_scan,
1871                           os_orphan_scan_work.work);
1872         osb = os->os_osb;
1873
1874         mutex_lock(&os->os_lock);
1875         ocfs2_queue_orphan_scan(osb);
1876         if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1877                 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1878                                       ocfs2_orphan_scan_timeout());
1879         mutex_unlock(&os->os_lock);
1880 }
1881
1882 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1883 {
1884         struct ocfs2_orphan_scan *os;
1885
1886         os = &osb->osb_orphan_scan;
1887         if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1888                 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1889                 mutex_lock(&os->os_lock);
1890                 cancel_delayed_work(&os->os_orphan_scan_work);
1891                 mutex_unlock(&os->os_lock);
1892         }
1893 }
1894
1895 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1896 {
1897         struct ocfs2_orphan_scan *os;
1898
1899         os = &osb->osb_orphan_scan;
1900         os->os_osb = osb;
1901         os->os_count = 0;
1902         os->os_seqno = 0;
1903         mutex_init(&os->os_lock);
1904         INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
1905 }
1906
1907 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
1908 {
1909         struct ocfs2_orphan_scan *os;
1910
1911         os = &osb->osb_orphan_scan;
1912         os->os_scantime = CURRENT_TIME;
1913         if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1914                 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1915         else {
1916                 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
1917                 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1918                                    ocfs2_orphan_scan_timeout());
1919         }
1920 }
1921
1922 struct ocfs2_orphan_filldir_priv {
1923         struct inode            *head;
1924         struct ocfs2_super      *osb;
1925 };
1926
1927 static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1928                                 loff_t pos, u64 ino, unsigned type)
1929 {
1930         struct ocfs2_orphan_filldir_priv *p = priv;
1931         struct inode *iter;
1932
1933         if (name_len == 1 && !strncmp(".", name, 1))
1934                 return 0;
1935         if (name_len == 2 && !strncmp("..", name, 2))
1936                 return 0;
1937
1938         /* Skip bad inodes so that recovery can continue */
1939         iter = ocfs2_iget(p->osb, ino,
1940                           OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
1941         if (IS_ERR(iter))
1942                 return 0;
1943
1944         trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
1945         /* No locking is required for the next_orphan queue as there
1946          * is only ever a single process doing orphan recovery. */
1947         OCFS2_I(iter)->ip_next_orphan = p->head;
1948         p->head = iter;
1949
1950         return 0;
1951 }
1952
1953 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1954                                int slot,
1955                                struct inode **head)
1956 {
1957         int status;
1958         struct inode *orphan_dir_inode = NULL;
1959         struct ocfs2_orphan_filldir_priv priv;
1960         loff_t pos = 0;
1961
1962         priv.osb = osb;
1963         priv.head = *head;
1964
1965         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1966                                                        ORPHAN_DIR_SYSTEM_INODE,
1967                                                        slot);
1968         if  (!orphan_dir_inode) {
1969                 status = -ENOENT;
1970                 mlog_errno(status);
1971                 return status;
1972         }
1973
1974         mutex_lock(&orphan_dir_inode->i_mutex);
1975         status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
1976         if (status < 0) {
1977                 mlog_errno(status);
1978                 goto out;
1979         }
1980
1981         status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
1982                                    ocfs2_orphan_filldir);
1983         if (status) {
1984                 mlog_errno(status);
1985                 goto out_cluster;
1986         }
1987
1988         *head = priv.head;
1989
1990 out_cluster:
1991         ocfs2_inode_unlock(orphan_dir_inode, 0);
1992 out:
1993         mutex_unlock(&orphan_dir_inode->i_mutex);
1994         iput(orphan_dir_inode);
1995         return status;
1996 }
1997
1998 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
1999                                               int slot)
2000 {
2001         int ret;
2002
2003         spin_lock(&osb->osb_lock);
2004         ret = !osb->osb_orphan_wipes[slot];
2005         spin_unlock(&osb->osb_lock);
2006         return ret;
2007 }
2008
2009 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2010                                              int slot)
2011 {
2012         spin_lock(&osb->osb_lock);
2013         /* Mark ourselves such that new processes in delete_inode()
2014          * know to quit early. */
2015         ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2016         while (osb->osb_orphan_wipes[slot]) {
2017                 /* If any processes are already in the middle of an
2018                  * orphan wipe on this dir, then we need to wait for
2019                  * them. */
2020                 spin_unlock(&osb->osb_lock);
2021                 wait_event_interruptible(osb->osb_wipe_event,
2022                                          ocfs2_orphan_recovery_can_continue(osb, slot));
2023                 spin_lock(&osb->osb_lock);
2024         }
2025         spin_unlock(&osb->osb_lock);
2026 }
2027
2028 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2029                                               int slot)
2030 {
2031         ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2032 }
2033
2034 /*
2035  * Orphan recovery. Each mounted node has it's own orphan dir which we
2036  * must run during recovery. Our strategy here is to build a list of
2037  * the inodes in the orphan dir and iget/iput them. The VFS does
2038  * (most) of the rest of the work.
2039  *
2040  * Orphan recovery can happen at any time, not just mount so we have a
2041  * couple of extra considerations.
2042  *
2043  * - We grab as many inodes as we can under the orphan dir lock -
2044  *   doing iget() outside the orphan dir risks getting a reference on
2045  *   an invalid inode.
2046  * - We must be sure not to deadlock with other processes on the
2047  *   system wanting to run delete_inode(). This can happen when they go
2048  *   to lock the orphan dir and the orphan recovery process attempts to
2049  *   iget() inside the orphan dir lock. This can be avoided by
2050  *   advertising our state to ocfs2_delete_inode().
2051  */
2052 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2053                                  int slot)
2054 {
2055         int ret = 0;
2056         struct inode *inode = NULL;
2057         struct inode *iter;
2058         struct ocfs2_inode_info *oi;
2059
2060         trace_ocfs2_recover_orphans(slot);
2061
2062         ocfs2_mark_recovering_orphan_dir(osb, slot);
2063         ret = ocfs2_queue_orphans(osb, slot, &inode);
2064         ocfs2_clear_recovering_orphan_dir(osb, slot);
2065
2066         /* Error here should be noted, but we want to continue with as
2067          * many queued inodes as we've got. */
2068         if (ret)
2069                 mlog_errno(ret);
2070
2071         while (inode) {
2072                 oi = OCFS2_I(inode);
2073                 trace_ocfs2_recover_orphans_iput(
2074                                         (unsigned long long)oi->ip_blkno);
2075
2076                 iter = oi->ip_next_orphan;
2077
2078                 spin_lock(&oi->ip_lock);
2079                 /* The remote delete code may have set these on the
2080                  * assumption that the other node would wipe them
2081                  * successfully.  If they are still in the node's
2082                  * orphan dir, we need to reset that state. */
2083                 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
2084
2085                 /* Set the proper information to get us going into
2086                  * ocfs2_delete_inode. */
2087                 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2088                 spin_unlock(&oi->ip_lock);
2089
2090                 iput(inode);
2091
2092                 inode = iter;
2093         }
2094
2095         return ret;
2096 }
2097
2098 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2099 {
2100         /* This check is good because ocfs2 will wait on our recovery
2101          * thread before changing it to something other than MOUNTED
2102          * or DISABLED. */
2103         wait_event(osb->osb_mount_event,
2104                   (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2105                    atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2106                    atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2107
2108         /* If there's an error on mount, then we may never get to the
2109          * MOUNTED flag, but this is set right before
2110          * dismount_volume() so we can trust it. */
2111         if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2112                 trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2113                 mlog(0, "mount error, exiting!\n");
2114                 return -EBUSY;
2115         }
2116
2117         return 0;
2118 }
2119
2120 static int ocfs2_commit_thread(void *arg)
2121 {
2122         int status;
2123         struct ocfs2_super *osb = arg;
2124         struct ocfs2_journal *journal = osb->journal;
2125
2126         /* we can trust j_num_trans here because _should_stop() is only set in
2127          * shutdown and nobody other than ourselves should be able to start
2128          * transactions.  committing on shutdown might take a few iterations
2129          * as final transactions put deleted inodes on the list */
2130         while (!(kthread_should_stop() &&
2131                  atomic_read(&journal->j_num_trans) == 0)) {
2132
2133                 wait_event_interruptible(osb->checkpoint_event,
2134                                          atomic_read(&journal->j_num_trans)
2135                                          || kthread_should_stop());
2136
2137                 status = ocfs2_commit_cache(osb);
2138                 if (status < 0)
2139                         mlog_errno(status);
2140
2141                 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2142                         mlog(ML_KTHREAD,
2143                              "commit_thread: %u transactions pending on "
2144                              "shutdown\n",
2145                              atomic_read(&journal->j_num_trans));
2146                 }
2147         }
2148
2149         return 0;
2150 }
2151
2152 /* Reads all the journal inodes without taking any cluster locks. Used
2153  * for hard readonly access to determine whether any journal requires
2154  * recovery. Also used to refresh the recovery generation numbers after
2155  * a journal has been recovered by another node.
2156  */
2157 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2158 {
2159         int ret = 0;
2160         unsigned int slot;
2161         struct buffer_head *di_bh = NULL;
2162         struct ocfs2_dinode *di;
2163         int journal_dirty = 0;
2164
2165         for(slot = 0; slot < osb->max_slots; slot++) {
2166                 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2167                 if (ret) {
2168                         mlog_errno(ret);
2169                         goto out;
2170                 }
2171
2172                 di = (struct ocfs2_dinode *) di_bh->b_data;
2173
2174                 osb->slot_recovery_generations[slot] =
2175                                         ocfs2_get_recovery_generation(di);
2176
2177                 if (le32_to_cpu(di->id1.journal1.ij_flags) &
2178                     OCFS2_JOURNAL_DIRTY_FL)
2179                         journal_dirty = 1;
2180
2181                 brelse(di_bh);
2182                 di_bh = NULL;
2183         }
2184
2185 out:
2186         if (journal_dirty)
2187                 ret = -EROFS;
2188         return ret;
2189 }