fs/jbd2/checkpoint.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * linux/fs/jbd2/checkpoint.c
   4  *
   5  * Written by Stephen C. Tweedie <[email protected]>, 1999
   6  *
   7  * Copyright 1999 Red Hat Software --- All Rights Reserved
   8  *
   9  * Checkpoint routines for the generic filesystem journaling code.
  10  * Part of the ext2fs journaling system.
  11  *
  12  * Checkpointing is the process of ensuring that a section of the log is
  13  * committed fully to disk, so that that portion of the log can be
  14  * reused.
  15  */
  16
  17 #include <linux/time.h>
  18 #include <linux/fs.h>
  19 #include <linux/jbd2.h>
  20 #include <linux/errno.h>
  21 #include <linux/slab.h>
  22 #include <linux/blkdev.h>
  23 #include <trace/events/jbd2.h>
  24
  25 /*
  26  * Unlink a buffer from a transaction checkpoint list.
  27  *
  28  * Called with j_list_lock held.
  29  */
  30 static inline void __buffer_unlink(struct journal_head *jh)
  31 {
  32         transaction_t *transaction = jh->b_cp_transaction;
  33
  34         jh->b_cpnext->b_cpprev = jh->b_cpprev;
  35         jh->b_cpprev->b_cpnext = jh->b_cpnext;
  36         if (transaction->t_checkpoint_list == jh) {
  37                 transaction->t_checkpoint_list = jh->b_cpnext;
  38                 if (transaction->t_checkpoint_list == jh)
  39                         transaction->t_checkpoint_list = NULL;
  40         }
  41 }
  42
  43 /*
  44  * __jbd2_log_wait_for_space: wait until there is space in the journal.
  45  *
  46  * Called under j-state_lock *only*.  It will be unlocked if we have to wait
  47  * for a checkpoint to free up some space in the log.
  48  */
  49 void __jbd2_log_wait_for_space(journal_t *journal)
  50 __acquires(&journal->j_state_lock)
  51 __releases(&journal->j_state_lock)
  52 {
  53         int nblocks, space_left;
  54         /* assert_spin_locked(&journal->j_state_lock); */
  55
  56         nblocks = journal->j_max_transaction_buffers;
  57         while (jbd2_log_space_left(journal) < nblocks) {
  58                 write_unlock(&journal->j_state_lock);
  59                 mutex_lock_io(&journal->j_checkpoint_mutex);
  60
  61                 /*
  62                  * Test again, another process may have checkpointed while we
  63                  * were waiting for the checkpoint lock. If there are no
  64                  * transactions ready to be checkpointed, try to recover
  65                  * journal space by calling cleanup_journal_tail(), and if
  66                  * that doesn't work, by waiting for the currently committing
  67                  * transaction to complete.  If there is absolutely no way
  68                  * to make progress, this is either a BUG or corrupted
  69                  * filesystem, so abort the journal and leave a stack
  70                  * trace for forensic evidence.
  71                  */
  72                 write_lock(&journal->j_state_lock);
  73                 if (journal->j_flags & JBD2_ABORT) {
  74                         mutex_unlock(&journal->j_checkpoint_mutex);
  75                         return;
  76                 }
  77                 spin_lock(&journal->j_list_lock);
  78                 space_left = jbd2_log_space_left(journal);
  79                 if (space_left < nblocks) {
  80                         int chkpt = journal->j_checkpoint_transactions != NULL;
  81                         tid_t tid = 0;
  82
  83                         if (journal->j_committing_transaction)
  84                                 tid = journal->j_committing_transaction->t_tid;
  85                         spin_unlock(&journal->j_list_lock);
  86                         write_unlock(&journal->j_state_lock);
  87                         if (chkpt) {
  88                                 jbd2_log_do_checkpoint(journal);
  89                         } else if (jbd2_cleanup_journal_tail(journal) == 0) {
  90                                 /* We were able to recover space; yay! */
  91                                 ;
  92                         } else if (tid) {
  93                                 /*
  94                                  * jbd2_journal_commit_transaction() may want
  95                                  * to take the checkpoint_mutex if JBD2_FLUSHED
  96                                  * is set.  So we need to temporarily drop it.
  97                                  */
  98                                 mutex_unlock(&journal->j_checkpoint_mutex);
  99                                 jbd2_log_wait_commit(journal, tid);
 100                                 write_lock(&journal->j_state_lock);
 101                                 continue;
 102                         } else {
 103                                 printk(KERN_ERR "%s: needed %d blocks and "
 104                                        "only had %d space available\n",
 105                                        __func__, nblocks, space_left);
 106                                 printk(KERN_ERR "%s: no way to get more "
 107                                        "journal space in %s\n", __func__,
 108                                        journal->j_devname);
 109                                 WARN_ON(1);
 110                                 jbd2_journal_abort(journal, -EIO);
 111                         }
 112                         write_lock(&journal->j_state_lock);
 113                 } else {
 114                         spin_unlock(&journal->j_list_lock);
 115                 }
 116                 mutex_unlock(&journal->j_checkpoint_mutex);
 117         }
 118 }
 119
 120 static void
 121 __flush_batch(journal_t *journal, int *batch_count)
 122 {
 123         int i;
 124         struct blk_plug plug;
 125
 126         blk_start_plug(&plug);
 127         for (i = 0; i < *batch_count; i++)
 128                 write_dirty_buffer(journal->j_chkpt_bhs[i], REQ_SYNC);
 129         blk_finish_plug(&plug);
 130
 131         for (i = 0; i < *batch_count; i++) {
 132                 struct buffer_head *bh = journal->j_chkpt_bhs[i];
 133                 BUFFER_TRACE(bh, "brelse");
 134                 __brelse(bh);
 135                 journal->j_chkpt_bhs[i] = NULL;
 136         }
 137         *batch_count = 0;
 138 }
 139
 140 /*
 141  * Perform an actual checkpoint. We take the first transaction on the
 142  * list of transactions to be checkpointed and send all its buffers
 143  * to disk. We submit larger chunks of data at once.
 144  *
 145  * The journal should be locked before calling this function.
 146  * Called with j_checkpoint_mutex held.
 147  */
 148 int jbd2_log_do_checkpoint(journal_t *journal)
 149 {
 150         struct journal_head     *jh;
 151         struct buffer_head      *bh;
 152         transaction_t           *transaction;
 153         tid_t                   this_tid;
 154         int                     result, batch_count = 0;
 155
 156         jbd2_debug(1, "Start checkpoint\n");
 157
 158         /*
 159          * First thing: if there are any transactions in the log which
 160          * don't need checkpointing, just eliminate them from the
 161          * journal straight away.
 162          */
 163         result = jbd2_cleanup_journal_tail(journal);
 164         trace_jbd2_checkpoint(journal, result);
 165         jbd2_debug(1, "cleanup_journal_tail returned %d\n", result);
 166         if (result <= 0)
 167                 return result;
 168
 169         /*
 170          * OK, we need to start writing disk blocks.  Take one transaction
 171          * and write it.
 172          */
 173         spin_lock(&journal->j_list_lock);
 174         if (!journal->j_checkpoint_transactions)
 175                 goto out;
 176         transaction = journal->j_checkpoint_transactions;
 177         if (transaction->t_chp_stats.cs_chp_time == 0)
 178                 transaction->t_chp_stats.cs_chp_time = jiffies;
 179         this_tid = transaction->t_tid;
 180 restart:
 181         /*
 182          * If someone cleaned up this transaction while we slept, we're
 183          * done (maybe it's a new transaction, but it fell at the same
 184          * address).
 185          */
 186         if (journal->j_checkpoint_transactions != transaction ||
 187             transaction->t_tid != this_tid)
 188                 goto out;
 189
 190         /* checkpoint all of the transaction's buffers */
 191         while (transaction->t_checkpoint_list) {
 192                 jh = transaction->t_checkpoint_list;
 193                 bh = jh2bh(jh);
 194
 195                 if (jh->b_transaction != NULL) {
 196                         transaction_t *t = jh->b_transaction;
 197                         tid_t tid = t->t_tid;
 198
 199                         transaction->t_chp_stats.cs_forced_to_close++;
 200                         spin_unlock(&journal->j_list_lock);
 201                         if (unlikely(journal->j_flags & JBD2_UNMOUNT))
 202                                 /*
 203                                  * The journal thread is dead; so
 204                                  * starting and waiting for a commit
 205                                  * to finish will cause us to wait for
 206                                  * a _very_ long time.
 207                                  */
 208                                 printk(KERN_ERR
 209                 "JBD2: %s: Waiting for Godot: block %llu\n",
 210                 journal->j_devname, (unsigned long long) bh->b_blocknr);
 211
 212                         if (batch_count)
 213                                 __flush_batch(journal, &batch_count);
 214                         jbd2_log_start_commit(journal, tid);
 215                         /*
 216                          * jbd2_journal_commit_transaction() may want
 217                          * to take the checkpoint_mutex if JBD2_FLUSHED
 218                          * is set, jbd2_update_log_tail() called by
 219                          * jbd2_journal_commit_transaction() may also take
 220                          * checkpoint_mutex.  So we need to temporarily
 221                          * drop it.
 222                          */
 223                         mutex_unlock(&journal->j_checkpoint_mutex);
 224                         jbd2_log_wait_commit(journal, tid);
 225                         mutex_lock_io(&journal->j_checkpoint_mutex);
 226                         spin_lock(&journal->j_list_lock);
 227                         goto restart;
 228                 }
 229                 if (!trylock_buffer(bh)) {
 230                         /*
 231                          * The buffer is locked, it may be writing back, or
 232                          * flushing out in the last couple of cycles, or
 233                          * re-adding into a new transaction, need to check
 234                          * it again until it's unlocked.
 235                          */
 236                         get_bh(bh);
 237                         spin_unlock(&journal->j_list_lock);
 238                         wait_on_buffer(bh);
 239                         /* the journal_head may have gone by now */
 240                         BUFFER_TRACE(bh, "brelse");
 241                         __brelse(bh);
 242                         goto retry;
 243                 } else if (!buffer_dirty(bh)) {
 244                         unlock_buffer(bh);
 245                         BUFFER_TRACE(bh, "remove from checkpoint");
 246                         /*
 247                          * If the transaction was released or the checkpoint
 248                          * list was empty, we're done.
 249                          */
 250                         if (__jbd2_journal_remove_checkpoint(jh) ||
 251                             !transaction->t_checkpoint_list)
 252                                 goto out;
 253                 } else {
 254                         unlock_buffer(bh);
 255                         /*
 256                          * We are about to write the buffer, it could be
 257                          * raced by some other transaction shrink or buffer
 258                          * re-log logic once we release the j_list_lock,
 259                          * leave it on the checkpoint list and check status
 260                          * again to make sure it's clean.
 261                          */
 262                         BUFFER_TRACE(bh, "queue");
 263                         get_bh(bh);
 264                         J_ASSERT_BH(bh, !buffer_jwrite(bh));
 265                         journal->j_chkpt_bhs[batch_count++] = bh;
 266                         transaction->t_chp_stats.cs_written++;
 267                         transaction->t_checkpoint_list = jh->b_cpnext;
 268                 }
 269
 270                 if ((batch_count == JBD2_NR_BATCH) ||
 271                     need_resched() || spin_needbreak(&journal->j_list_lock) ||
 272                     jh2bh(transaction->t_checkpoint_list) == journal->j_chkpt_bhs[0])
 273                         goto unlock_and_flush;
 274         }
 275
 276         if (batch_count) {
 277                 unlock_and_flush:
 278                         spin_unlock(&journal->j_list_lock);
 279                 retry:
 280                         if (batch_count)
 281                                 __flush_batch(journal, &batch_count);
 282                         spin_lock(&journal->j_list_lock);
 283                         goto restart;
 284         }
 285
 286 out:
 287         spin_unlock(&journal->j_list_lock);
 288         result = jbd2_cleanup_journal_tail(journal);
 289
 290         return (result < 0) ? result : 0;
 291 }
 292
 293 /*
 294  * Check the list of checkpoint transactions for the journal to see if
 295  * we have already got rid of any since the last update of the log tail
 296  * in the journal superblock.  If so, we can instantly roll the
 297  * superblock forward to remove those transactions from the log.
 298  *
 299  * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
 300  *
 301  * Called with the journal lock held.
 302  *
 303  * This is the only part of the journaling code which really needs to be
 304  * aware of transaction aborts.  Checkpointing involves writing to the
 305  * main filesystem area rather than to the journal, so it can proceed
 306  * even in abort state, but we must not update the super block if
 307  * checkpointing may have failed.  Otherwise, we would lose some metadata
 308  * buffers which should be written-back to the filesystem.
 309  */
 310
 311 int jbd2_cleanup_journal_tail(journal_t *journal)
 312 {
 313         tid_t           first_tid;
 314         unsigned long   blocknr;
 315
 316         if (is_journal_aborted(journal))
 317                 return -EIO;
 318
 319         if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr))
 320                 return 1;
 321         J_ASSERT(blocknr != 0);
 322
 323         /*
 324          * We need to make sure that any blocks that were recently written out
 325          * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before
 326          * we drop the transactions from the journal. It's unlikely this will
 327          * be necessary, especially with an appropriately sized journal, but we
 328          * need this to guarantee correctness.  Fortunately
 329          * jbd2_cleanup_journal_tail() doesn't get called all that often.
 330          */
 331         if (journal->j_flags & JBD2_BARRIER)
 332                 blkdev_issue_flush(journal->j_fs_dev);
 333
 334         return __jbd2_update_log_tail(journal, first_tid, blocknr);
 335 }
 336
 337
 338 /* Checkpoint list management */
 339
 340 enum shrink_type {SHRINK_DESTROY, SHRINK_BUSY_STOP, SHRINK_BUSY_SKIP};
 341
 342 /*
 343  * journal_shrink_one_cp_list
 344  *
 345  * Find all the written-back checkpoint buffers in the given list
 346  * and try to release them. If the whole transaction is released, set
 347  * the 'released' parameter. Return the number of released checkpointed
 348  * buffers.
 349  *
 350  * Called with j_list_lock held.
 351  */
 352 static unsigned long journal_shrink_one_cp_list(struct journal_head *jh,
 353                                                 enum shrink_type type,
 354                                                 bool *released)
 355 {
 356         struct journal_head *last_jh;
 357         struct journal_head *next_jh = jh;
 358         unsigned long nr_freed = 0;
 359         int ret;
 360
 361         *released = false;
 362         if (!jh)
 363                 return 0;
 364
 365         last_jh = jh->b_cpprev;
 366         do {
 367                 jh = next_jh;
 368                 next_jh = jh->b_cpnext;
 369
 370                 if (type == SHRINK_DESTROY) {
 371                         ret = __jbd2_journal_remove_checkpoint(jh);
 372                 } else {
 373                         ret = jbd2_journal_try_remove_checkpoint(jh);
 374                         if (ret < 0) {
 375                                 if (type == SHRINK_BUSY_SKIP)
 376                                         continue;
 377                                 break;
 378                         }
 379                 }
 380
 381                 nr_freed++;
 382                 if (ret) {
 383                         *released = true;
 384                         break;
 385                 }
 386
 387                 if (need_resched())
 388                         break;
 389         } while (jh != last_jh);
 390
 391         return nr_freed;
 392 }
 393
 394 /*
 395  * jbd2_journal_shrink_checkpoint_list
 396  *
 397  * Find 'nr_to_scan' written-back checkpoint buffers in the journal
 398  * and try to release them. Return the number of released checkpointed
 399  * buffers.
 400  *
 401  * Called with j_list_lock held.
 402  */
 403 unsigned long jbd2_journal_shrink_checkpoint_list(journal_t *journal,
 404                                                   unsigned long *nr_to_scan)
 405 {
 406         transaction_t *transaction, *last_transaction, *next_transaction;
 407         bool __maybe_unused released;
 408         tid_t first_tid = 0, last_tid = 0, next_tid = 0;
 409         tid_t tid = 0;
 410         unsigned long nr_freed = 0;
 411         unsigned long freed;
 412
 413 again:
 414         spin_lock(&journal->j_list_lock);
 415         if (!journal->j_checkpoint_transactions) {
 416                 spin_unlock(&journal->j_list_lock);
 417                 goto out;
 418         }
 419
 420         /*
 421          * Get next shrink transaction, resume previous scan or start
 422          * over again. If some others do checkpoint and drop transaction
 423          * from the checkpoint list, we ignore saved j_shrink_transaction
 424          * and start over unconditionally.
 425          */
 426         if (journal->j_shrink_transaction)
 427                 transaction = journal->j_shrink_transaction;
 428         else
 429                 transaction = journal->j_checkpoint_transactions;
 430
 431         if (!first_tid)
 432                 first_tid = transaction->t_tid;
 433         last_transaction = journal->j_checkpoint_transactions->t_cpprev;
 434         next_transaction = transaction;
 435         last_tid = last_transaction->t_tid;
 436         do {
 437                 transaction = next_transaction;
 438                 next_transaction = transaction->t_cpnext;
 439                 tid = transaction->t_tid;
 440
 441                 freed = journal_shrink_one_cp_list(transaction->t_checkpoint_list,
 442                                                    SHRINK_BUSY_SKIP, &released);
 443                 nr_freed += freed;
 444                 (*nr_to_scan) -= min(*nr_to_scan, freed);
 445                 if (*nr_to_scan == 0)
 446                         break;
 447                 if (need_resched() || spin_needbreak(&journal->j_list_lock))
 448                         break;
 449         } while (transaction != last_transaction);
 450
 451         if (transaction != last_transaction) {
 452                 journal->j_shrink_transaction = next_transaction;
 453                 next_tid = next_transaction->t_tid;
 454         } else {
 455                 journal->j_shrink_transaction = NULL;
 456                 next_tid = 0;
 457         }
 458
 459         spin_unlock(&journal->j_list_lock);
 460         cond_resched();
 461
 462         if (*nr_to_scan && next_tid)
 463                 goto again;
 464 out:
 465         trace_jbd2_shrink_checkpoint_list(journal, first_tid, tid, last_tid,
 466                                           nr_freed, next_tid);
 467
 468         return nr_freed;
 469 }
 470
 471 /*
 472  * journal_clean_checkpoint_list
 473  *
 474  * Find all the written-back checkpoint buffers in the journal and release them.
 475  * If 'destroy' is set, release all buffers unconditionally.
 476  *
 477  * Called with j_list_lock held.
 478  */
 479 void __jbd2_journal_clean_checkpoint_list(journal_t *journal, bool destroy)
 480 {
 481         transaction_t *transaction, *last_transaction, *next_transaction;
 482         enum shrink_type type;
 483         bool released;
 484
 485         transaction = journal->j_checkpoint_transactions;
 486         if (!transaction)
 487                 return;
 488
 489         type = destroy ? SHRINK_DESTROY : SHRINK_BUSY_STOP;
 490         last_transaction = transaction->t_cpprev;
 491         next_transaction = transaction;
 492         do {
 493                 transaction = next_transaction;
 494                 next_transaction = transaction->t_cpnext;
 495                 journal_shrink_one_cp_list(transaction->t_checkpoint_list,
 496                                            type, &released);
 497                 /*
 498                  * This function only frees up some memory if possible so we
 499                  * dont have an obligation to finish processing. Bail out if
 500                  * preemption requested:
 501                  */
 502                 if (need_resched())
 503                         return;
 504                 /*
 505                  * Stop scanning if we couldn't free the transaction. This
 506                  * avoids pointless scanning of transactions which still
 507                  * weren't checkpointed.
 508                  */
 509                 if (!released)
 510                         return;
 511         } while (transaction != last_transaction);
 512 }
 513
 514 /*
 515  * Remove buffers from all checkpoint lists as journal is aborted and we just
 516  * need to free memory
 517  */
 518 void jbd2_journal_destroy_checkpoint(journal_t *journal)
 519 {
 520         /*
 521          * We loop because __jbd2_journal_clean_checkpoint_list() may abort
 522          * early due to a need of rescheduling.
 523          */
 524         while (1) {
 525                 spin_lock(&journal->j_list_lock);
 526                 if (!journal->j_checkpoint_transactions) {
 527                         spin_unlock(&journal->j_list_lock);
 528                         break;
 529                 }
 530                 __jbd2_journal_clean_checkpoint_list(journal, true);
 531                 spin_unlock(&journal->j_list_lock);
 532                 cond_resched();
 533         }
 534 }
 535
 536 /*
 537  * journal_remove_checkpoint: called after a buffer has been committed
 538  * to disk (either by being write-back flushed to disk, or being
 539  * committed to the log).
 540  *
 541  * We cannot safely clean a transaction out of the log until all of the
 542  * buffer updates committed in that transaction have safely been stored
 543  * elsewhere on disk.  To achieve this, all of the buffers in a
 544  * transaction need to be maintained on the transaction's checkpoint
 545  * lists until they have been rewritten, at which point this function is
 546  * called to remove the buffer from the existing transaction's
 547  * checkpoint lists.
 548  *
 549  * The function returns 1 if it frees the transaction, 0 otherwise.
 550  * The function can free jh and bh.
 551  *
 552  * This function is called with j_list_lock held.
 553  */
 554 int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
 555 {
 556         struct transaction_chp_stats_s *stats;
 557         transaction_t *transaction;
 558         journal_t *journal;
 559         struct buffer_head *bh = jh2bh(jh);
 560
 561         JBUFFER_TRACE(jh, "entry");
 562
 563         transaction = jh->b_cp_transaction;
 564         if (!transaction) {
 565                 JBUFFER_TRACE(jh, "not on transaction");
 566                 return 0;
 567         }
 568         journal = transaction->t_journal;
 569
 570         JBUFFER_TRACE(jh, "removing from transaction");
 571
 572         /*
 573          * If we have failed to write the buffer out to disk, the filesystem
 574          * may become inconsistent. We cannot abort the journal here since
 575          * we hold j_list_lock and we have to be careful about races with
 576          * jbd2_journal_destroy(). So mark the writeback IO error in the
 577          * journal here and we abort the journal later from a better context.
 578          */
 579         if (buffer_write_io_error(bh))
 580                 set_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags);
 581
 582         __buffer_unlink(jh);
 583         jh->b_cp_transaction = NULL;
 584         percpu_counter_dec(&journal->j_checkpoint_jh_count);
 585         jbd2_journal_put_journal_head(jh);
 586
 587         /* Is this transaction empty? */
 588         if (transaction->t_checkpoint_list)
 589                 return 0;
 590
 591         /*
 592          * There is one special case to worry about: if we have just pulled the
 593          * buffer off a running or committing transaction's checkpoing list,
 594          * then even if the checkpoint list is empty, the transaction obviously
 595          * cannot be dropped!
 596          *
 597          * The locking here around t_state is a bit sleazy.
 598          * See the comment at the end of jbd2_journal_commit_transaction().
 599          */
 600         if (transaction->t_state != T_FINISHED)
 601                 return 0;
 602
 603         /*
 604          * OK, that was the last buffer for the transaction, we can now
 605          * safely remove this transaction from the log.
 606          */
 607         stats = &transaction->t_chp_stats;
 608         if (stats->cs_chp_time)
 609                 stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time,
 610                                                     jiffies);
 611         trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev,
 612                                     transaction->t_tid, stats);
 613
 614         __jbd2_journal_drop_transaction(journal, transaction);
 615         jbd2_journal_free_transaction(transaction);
 616         return 1;
 617 }
 618
 619 /*
 620  * Check the checkpoint buffer and try to remove it from the checkpoint
 621  * list if it's clean. Returns -EBUSY if it is not clean, returns 1 if
 622  * it frees the transaction, 0 otherwise.
 623  *
 624  * This function is called with j_list_lock held.
 625  */
 626 int jbd2_journal_try_remove_checkpoint(struct journal_head *jh)
 627 {
 628         struct buffer_head *bh = jh2bh(jh);
 629
 630         if (jh->b_transaction)
 631                 return -EBUSY;
 632         if (!trylock_buffer(bh))
 633                 return -EBUSY;
 634         if (buffer_dirty(bh)) {
 635                 unlock_buffer(bh);
 636                 return -EBUSY;
 637         }
 638         unlock_buffer(bh);
 639
 640         /*
 641          * Buffer is clean and the IO has finished (we held the buffer
 642          * lock) so the checkpoint is done. We can safely remove the
 643          * buffer from this transaction.
 644          */
 645         JBUFFER_TRACE(jh, "remove from checkpoint list");
 646         return __jbd2_journal_remove_checkpoint(jh);
 647 }
 648
 649 /*
 650  * journal_insert_checkpoint: put a committed buffer onto a checkpoint
 651  * list so that we know when it is safe to clean the transaction out of
 652  * the log.
 653  *
 654  * Called with the journal locked.
 655  * Called with j_list_lock held.
 656  */
 657 void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
 658                                transaction_t *transaction)
 659 {
 660         JBUFFER_TRACE(jh, "entry");
 661         J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
 662         J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
 663
 664         /* Get reference for checkpointing transaction */
 665         jbd2_journal_grab_journal_head(jh2bh(jh));
 666         jh->b_cp_transaction = transaction;
 667
 668         if (!transaction->t_checkpoint_list) {
 669                 jh->b_cpnext = jh->b_cpprev = jh;
 670         } else {
 671                 jh->b_cpnext = transaction->t_checkpoint_list;
 672                 jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
 673                 jh->b_cpprev->b_cpnext = jh;
 674                 jh->b_cpnext->b_cpprev = jh;
 675         }
 676         transaction->t_checkpoint_list = jh;
 677         percpu_counter_inc(&transaction->t_journal->j_checkpoint_jh_count);
 678 }
 679
 680 /*
 681  * We've finished with this transaction structure: adios...
 682  *
 683  * The transaction must have no links except for the checkpoint by this
 684  * point.
 685  *
 686  * Called with the journal locked.
 687  * Called with j_list_lock held.
 688  */
 689
 690 void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
 691 {
 692         assert_spin_locked(&journal->j_list_lock);
 693
 694         journal->j_shrink_transaction = NULL;
 695         if (transaction->t_cpnext) {
 696                 transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
 697                 transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
 698                 if (journal->j_checkpoint_transactions == transaction)
 699                         journal->j_checkpoint_transactions =
 700                                 transaction->t_cpnext;
 701                 if (journal->j_checkpoint_transactions == transaction)
 702                         journal->j_checkpoint_transactions = NULL;
 703         }
 704
 705         J_ASSERT(transaction->t_state == T_FINISHED);
 706         J_ASSERT(transaction->t_buffers == NULL);
 707         J_ASSERT(transaction->t_forget == NULL);
 708         J_ASSERT(transaction->t_shadow_list == NULL);
 709         J_ASSERT(transaction->t_checkpoint_list == NULL);
 710         J_ASSERT(atomic_read(&transaction->t_updates) == 0);
 711         J_ASSERT(journal->j_committing_transaction != transaction);
 712         J_ASSERT(journal->j_running_transaction != transaction);
 713
 714         trace_jbd2_drop_transaction(journal, transaction);
 715
 716         jbd2_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);
 717 }