fs/jbd2/journal.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * linux/fs/jbd2/journal.c
   4  *
   5  * Written by Stephen C. Tweedie <[email protected]>, 1998
   6  *
   7  * Copyright 1998 Red Hat corp --- All Rights Reserved
   8  *
   9  * Generic filesystem journal-writing code; part of the ext2fs
  10  * journaling system.
  11  *
  12  * This file manages journals: areas of disk reserved for logging
  13  * transactional updates.  This includes the kernel journaling thread
  14  * which is responsible for scheduling updates to the log.
  15  *
  16  * We do not actually manage the physical storage of the journal in this
  17  * file: that is left to a per-journal policy function, which allows us
  18  * to store the journal within a filesystem-specified area for ext2
  19  * journaling (ext2 can use a reserved inode for storing the log).
  20  */
  21
  22 #include <linux/module.h>
  23 #include <linux/time.h>
  24 #include <linux/fs.h>
  25 #include <linux/jbd2.h>
  26 #include <linux/errno.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/mm.h>
  30 #include <linux/freezer.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/kthread.h>
  33 #include <linux/poison.h>
  34 #include <linux/proc_fs.h>
  35 #include <linux/seq_file.h>
  36 #include <linux/math64.h>
  37 #include <linux/hash.h>
  38 #include <linux/log2.h>
  39 #include <linux/vmalloc.h>
  40 #include <linux/backing-dev.h>
  41 #include <linux/bitops.h>
  42 #include <linux/ratelimit.h>
  43 #include <linux/sched/mm.h>
  44
  45 #define CREATE_TRACE_POINTS
  46 #include <trace/events/jbd2.h>
  47
  48 #include <linux/uaccess.h>
  49 #include <asm/page.h>
  50
  51 #ifdef CONFIG_JBD2_DEBUG
  52 ushort jbd2_journal_enable_debug __read_mostly;
  53 EXPORT_SYMBOL(jbd2_journal_enable_debug);
  54
  55 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
  56 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
  57 #endif
  58
  59 EXPORT_SYMBOL(jbd2_journal_extend);
  60 EXPORT_SYMBOL(jbd2_journal_stop);
  61 EXPORT_SYMBOL(jbd2_journal_lock_updates);
  62 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
  63 EXPORT_SYMBOL(jbd2_journal_get_write_access);
  64 EXPORT_SYMBOL(jbd2_journal_get_create_access);
  65 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
  66 EXPORT_SYMBOL(jbd2_journal_set_triggers);
  67 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
  68 EXPORT_SYMBOL(jbd2_journal_forget);
  69 EXPORT_SYMBOL(jbd2_journal_flush);
  70 EXPORT_SYMBOL(jbd2_journal_revoke);
  71
  72 EXPORT_SYMBOL(jbd2_journal_init_dev);
  73 EXPORT_SYMBOL(jbd2_journal_init_inode);
  74 EXPORT_SYMBOL(jbd2_journal_check_used_features);
  75 EXPORT_SYMBOL(jbd2_journal_check_available_features);
  76 EXPORT_SYMBOL(jbd2_journal_set_features);
  77 EXPORT_SYMBOL(jbd2_journal_load);
  78 EXPORT_SYMBOL(jbd2_journal_destroy);
  79 EXPORT_SYMBOL(jbd2_journal_abort);
  80 EXPORT_SYMBOL(jbd2_journal_errno);
  81 EXPORT_SYMBOL(jbd2_journal_ack_err);
  82 EXPORT_SYMBOL(jbd2_journal_clear_err);
  83 EXPORT_SYMBOL(jbd2_log_wait_commit);
  84 EXPORT_SYMBOL(jbd2_log_start_commit);
  85 EXPORT_SYMBOL(jbd2_journal_start_commit);
  86 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
  87 EXPORT_SYMBOL(jbd2_journal_wipe);
  88 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
  89 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
  90 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
  91 EXPORT_SYMBOL(jbd2_journal_force_commit);
  92 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
  93 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
  94 EXPORT_SYMBOL(jbd2_journal_submit_inode_data_buffers);
  95 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers);
  96 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
  97 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
  98 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
  99 EXPORT_SYMBOL(jbd2_inode_cache);
 100
 101 static int jbd2_journal_create_slab(size_t slab_size);
 102
 103 #ifdef CONFIG_JBD2_DEBUG
 104 void __jbd2_debug(int level, const char *file, const char *func,
 105                   unsigned int line, const char *fmt, ...)
 106 {
 107         struct va_format vaf;
 108         va_list args;
 109
 110         if (level > jbd2_journal_enable_debug)
 111                 return;
 112         va_start(args, fmt);
 113         vaf.fmt = fmt;
 114         vaf.va = &args;
 115         printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
 116         va_end(args);
 117 }
 118 EXPORT_SYMBOL(__jbd2_debug);
 119 #endif
 120
 121 /* Checksumming functions */
 122 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
 123 {
 124         if (!jbd2_journal_has_csum_v2or3_feature(j))
 125                 return 1;
 126
 127         return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
 128 }
 129
 130 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
 131 {
 132         __u32 csum;
 133         __be32 old_csum;
 134
 135         old_csum = sb->s_checksum;
 136         sb->s_checksum = 0;
 137         csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
 138         sb->s_checksum = old_csum;
 139
 140         return cpu_to_be32(csum);
 141 }
 142
 143 /*
 144  * Helper function used to manage commit timeouts
 145  */
 146
 147 static void commit_timeout(struct timer_list *t)
 148 {
 149         journal_t *journal = from_timer(journal, t, j_commit_timer);
 150
 151         wake_up_process(journal->j_task);
 152 }
 153
 154 /*
 155  * kjournald2: The main thread function used to manage a logging device
 156  * journal.
 157  *
 158  * This kernel thread is responsible for two things:
 159  *
 160  * 1) COMMIT:  Every so often we need to commit the current state of the
 161  *    filesystem to disk.  The journal thread is responsible for writing
 162  *    all of the metadata buffers to disk. If a fast commit is ongoing
 163  *    journal thread waits until it's done and then continues from
 164  *    there on.
 165  *
 166  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
 167  *    of the data in that part of the log has been rewritten elsewhere on
 168  *    the disk.  Flushing these old buffers to reclaim space in the log is
 169  *    known as checkpointing, and this thread is responsible for that job.
 170  */
 171
 172 static int kjournald2(void *arg)
 173 {
 174         journal_t *journal = arg;
 175         transaction_t *transaction;
 176
 177         /*
 178          * Set up an interval timer which can be used to trigger a commit wakeup
 179          * after the commit interval expires
 180          */
 181         timer_setup(&journal->j_commit_timer, commit_timeout, 0);
 182
 183         set_freezable();
 184
 185         /* Record that the journal thread is running */
 186         journal->j_task = current;
 187         wake_up(&journal->j_wait_done_commit);
 188
 189         /*
 190          * Make sure that no allocations from this kernel thread will ever
 191          * recurse to the fs layer because we are responsible for the
 192          * transaction commit and any fs involvement might get stuck waiting for
 193          * the trasn. commit.
 194          */
 195         memalloc_nofs_save();
 196
 197         /*
 198          * And now, wait forever for commit wakeup events.
 199          */
 200         write_lock(&journal->j_state_lock);
 201
 202 loop:
 203         if (journal->j_flags & JBD2_UNMOUNT)
 204                 goto end_loop;
 205
 206         jbd_debug(1, "commit_sequence=%u, commit_request=%u\n",
 207                 journal->j_commit_sequence, journal->j_commit_request);
 208
 209         if (journal->j_commit_sequence != journal->j_commit_request) {
 210                 jbd_debug(1, "OK, requests differ\n");
 211                 write_unlock(&journal->j_state_lock);
 212                 del_timer_sync(&journal->j_commit_timer);
 213                 jbd2_journal_commit_transaction(journal);
 214                 write_lock(&journal->j_state_lock);
 215                 goto loop;
 216         }
 217
 218         wake_up(&journal->j_wait_done_commit);
 219         if (freezing(current)) {
 220                 /*
 221                  * The simpler the better. Flushing journal isn't a
 222                  * good idea, because that depends on threads that may
 223                  * be already stopped.
 224                  */
 225                 jbd_debug(1, "Now suspending kjournald2\n");
 226                 write_unlock(&journal->j_state_lock);
 227                 try_to_freeze();
 228                 write_lock(&journal->j_state_lock);
 229         } else {
 230                 /*
 231                  * We assume on resume that commits are already there,
 232                  * so we don't sleep
 233                  */
 234                 DEFINE_WAIT(wait);
 235                 int should_sleep = 1;
 236
 237                 prepare_to_wait(&journal->j_wait_commit, &wait,
 238                                 TASK_INTERRUPTIBLE);
 239                 if (journal->j_commit_sequence != journal->j_commit_request)
 240                         should_sleep = 0;
 241                 transaction = journal->j_running_transaction;
 242                 if (transaction && time_after_eq(jiffies,
 243                                                 transaction->t_expires))
 244                         should_sleep = 0;
 245                 if (journal->j_flags & JBD2_UNMOUNT)
 246                         should_sleep = 0;
 247                 if (should_sleep) {
 248                         write_unlock(&journal->j_state_lock);
 249                         schedule();
 250                         write_lock(&journal->j_state_lock);
 251                 }
 252                 finish_wait(&journal->j_wait_commit, &wait);
 253         }
 254
 255         jbd_debug(1, "kjournald2 wakes\n");
 256
 257         /*
 258          * Were we woken up by a commit wakeup event?
 259          */
 260         transaction = journal->j_running_transaction;
 261         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
 262                 journal->j_commit_request = transaction->t_tid;
 263                 jbd_debug(1, "woke because of timeout\n");
 264         }
 265         goto loop;
 266
 267 end_loop:
 268         del_timer_sync(&journal->j_commit_timer);
 269         journal->j_task = NULL;
 270         wake_up(&journal->j_wait_done_commit);
 271         jbd_debug(1, "Journal thread exiting.\n");
 272         write_unlock(&journal->j_state_lock);
 273         return 0;
 274 }
 275
 276 static int jbd2_journal_start_thread(journal_t *journal)
 277 {
 278         struct task_struct *t;
 279
 280         t = kthread_run(kjournald2, journal, "jbd2/%s",
 281                         journal->j_devname);
 282         if (IS_ERR(t))
 283                 return PTR_ERR(t);
 284
 285         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
 286         return 0;
 287 }
 288
 289 static void journal_kill_thread(journal_t *journal)
 290 {
 291         write_lock(&journal->j_state_lock);
 292         journal->j_flags |= JBD2_UNMOUNT;
 293
 294         while (journal->j_task) {
 295                 write_unlock(&journal->j_state_lock);
 296                 wake_up(&journal->j_wait_commit);
 297                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
 298                 write_lock(&journal->j_state_lock);
 299         }
 300         write_unlock(&journal->j_state_lock);
 301 }
 302
 303 /*
 304  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
 305  *
 306  * Writes a metadata buffer to a given disk block.  The actual IO is not
 307  * performed but a new buffer_head is constructed which labels the data
 308  * to be written with the correct destination disk block.
 309  *
 310  * Any magic-number escaping which needs to be done will cause a
 311  * copy-out here.  If the buffer happens to start with the
 312  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
 313  * magic number is only written to the log for descripter blocks.  In
 314  * this case, we copy the data and replace the first word with 0, and we
 315  * return a result code which indicates that this buffer needs to be
 316  * marked as an escaped buffer in the corresponding log descriptor
 317  * block.  The missing word can then be restored when the block is read
 318  * during recovery.
 319  *
 320  * If the source buffer has already been modified by a new transaction
 321  * since we took the last commit snapshot, we use the frozen copy of
 322  * that data for IO. If we end up using the existing buffer_head's data
 323  * for the write, then we have to make sure nobody modifies it while the
 324  * IO is in progress. do_get_write_access() handles this.
 325  *
 326  * The function returns a pointer to the buffer_head to be used for IO.
 327  *
 328  *
 329  * Return value:
 330  *  <0: Error
 331  * >=0: Finished OK
 332  *
 333  * On success:
 334  * Bit 0 set == escape performed on the data
 335  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
 336  */
 337
 338 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
 339                                   struct journal_head  *jh_in,
 340                                   struct buffer_head **bh_out,
 341                                   sector_t blocknr)
 342 {
 343         int need_copy_out = 0;
 344         int done_copy_out = 0;
 345         int do_escape = 0;
 346         char *mapped_data;
 347         struct buffer_head *new_bh;
 348         struct page *new_page;
 349         unsigned int new_offset;
 350         struct buffer_head *bh_in = jh2bh(jh_in);
 351         journal_t *journal = transaction->t_journal;
 352
 353         /*
 354          * The buffer really shouldn't be locked: only the current committing
 355          * transaction is allowed to write it, so nobody else is allowed
 356          * to do any IO.
 357          *
 358          * akpm: except if we're journalling data, and write() output is
 359          * also part of a shared mapping, and another thread has
 360          * decided to launch a writepage() against this buffer.
 361          */
 362         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
 363
 364         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
 365
 366         /* keep subsequent assertions sane */
 367         atomic_set(&new_bh->b_count, 1);
 368
 369         spin_lock(&jh_in->b_state_lock);
 370 repeat:
 371         /*
 372          * If a new transaction has already done a buffer copy-out, then
 373          * we use that version of the data for the commit.
 374          */
 375         if (jh_in->b_frozen_data) {
 376                 done_copy_out = 1;
 377                 new_page = virt_to_page(jh_in->b_frozen_data);
 378                 new_offset = offset_in_page(jh_in->b_frozen_data);
 379         } else {
 380                 new_page = jh2bh(jh_in)->b_page;
 381                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
 382         }
 383
 384         mapped_data = kmap_atomic(new_page);
 385         /*
 386          * Fire data frozen trigger if data already wasn't frozen.  Do this
 387          * before checking for escaping, as the trigger may modify the magic
 388          * offset.  If a copy-out happens afterwards, it will have the correct
 389          * data in the buffer.
 390          */
 391         if (!done_copy_out)
 392                 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
 393                                            jh_in->b_triggers);
 394
 395         /*
 396          * Check for escaping
 397          */
 398         if (*((__be32 *)(mapped_data + new_offset)) ==
 399                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
 400                 need_copy_out = 1;
 401                 do_escape = 1;
 402         }
 403         kunmap_atomic(mapped_data);
 404
 405         /*
 406          * Do we need to do a data copy?
 407          */
 408         if (need_copy_out && !done_copy_out) {
 409                 char *tmp;
 410
 411                 spin_unlock(&jh_in->b_state_lock);
 412                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
 413                 if (!tmp) {
 414                         brelse(new_bh);
 415                         return -ENOMEM;
 416                 }
 417                 spin_lock(&jh_in->b_state_lock);
 418                 if (jh_in->b_frozen_data) {
 419                         jbd2_free(tmp, bh_in->b_size);
 420                         goto repeat;
 421                 }
 422
 423                 jh_in->b_frozen_data = tmp;
 424                 mapped_data = kmap_atomic(new_page);
 425                 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
 426                 kunmap_atomic(mapped_data);
 427
 428                 new_page = virt_to_page(tmp);
 429                 new_offset = offset_in_page(tmp);
 430                 done_copy_out = 1;
 431
 432                 /*
 433                  * This isn't strictly necessary, as we're using frozen
 434                  * data for the escaping, but it keeps consistency with
 435                  * b_frozen_data usage.
 436                  */
 437                 jh_in->b_frozen_triggers = jh_in->b_triggers;
 438         }
 439
 440         /*
 441          * Did we need to do an escaping?  Now we've done all the
 442          * copying, we can finally do so.
 443          */
 444         if (do_escape) {
 445                 mapped_data = kmap_atomic(new_page);
 446                 *((unsigned int *)(mapped_data + new_offset)) = 0;
 447                 kunmap_atomic(mapped_data);
 448         }
 449
 450         set_bh_page(new_bh, new_page, new_offset);
 451         new_bh->b_size = bh_in->b_size;
 452         new_bh->b_bdev = journal->j_dev;
 453         new_bh->b_blocknr = blocknr;
 454         new_bh->b_private = bh_in;
 455         set_buffer_mapped(new_bh);
 456         set_buffer_dirty(new_bh);
 457
 458         *bh_out = new_bh;
 459
 460         /*
 461          * The to-be-written buffer needs to get moved to the io queue,
 462          * and the original buffer whose contents we are shadowing or
 463          * copying is moved to the transaction's shadow queue.
 464          */
 465         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
 466         spin_lock(&journal->j_list_lock);
 467         __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
 468         spin_unlock(&journal->j_list_lock);
 469         set_buffer_shadow(bh_in);
 470         spin_unlock(&jh_in->b_state_lock);
 471
 472         return do_escape | (done_copy_out << 1);
 473 }
 474
 475 /*
 476  * Allocation code for the journal file.  Manage the space left in the
 477  * journal, so that we can begin checkpointing when appropriate.
 478  */
 479
 480 /*
 481  * Called with j_state_lock locked for writing.
 482  * Returns true if a transaction commit was started.
 483  */
 484 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
 485 {
 486         /* Return if the txn has already requested to be committed */
 487         if (journal->j_commit_request == target)
 488                 return 0;
 489
 490         /*
 491          * The only transaction we can possibly wait upon is the
 492          * currently running transaction (if it exists).  Otherwise,
 493          * the target tid must be an old one.
 494          */
 495         if (journal->j_running_transaction &&
 496             journal->j_running_transaction->t_tid == target) {
 497                 /*
 498                  * We want a new commit: OK, mark the request and wakeup the
 499                  * commit thread.  We do _not_ do the commit ourselves.
 500                  */
 501
 502                 journal->j_commit_request = target;
 503                 jbd_debug(1, "JBD2: requesting commit %u/%u\n",
 504                           journal->j_commit_request,
 505                           journal->j_commit_sequence);
 506                 journal->j_running_transaction->t_requested = jiffies;
 507                 wake_up(&journal->j_wait_commit);
 508                 return 1;
 509         } else if (!tid_geq(journal->j_commit_request, target))
 510                 /* This should never happen, but if it does, preserve
 511                    the evidence before kjournald goes into a loop and
 512                    increments j_commit_sequence beyond all recognition. */
 513                 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
 514                           journal->j_commit_request,
 515                           journal->j_commit_sequence,
 516                           target, journal->j_running_transaction ?
 517                           journal->j_running_transaction->t_tid : 0);
 518         return 0;
 519 }
 520
 521 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
 522 {
 523         int ret;
 524
 525         write_lock(&journal->j_state_lock);
 526         ret = __jbd2_log_start_commit(journal, tid);
 527         write_unlock(&journal->j_state_lock);
 528         return ret;
 529 }
 530
 531 /*
 532  * Force and wait any uncommitted transactions.  We can only force the running
 533  * transaction if we don't have an active handle, otherwise, we will deadlock.
 534  * Returns: <0 in case of error,
 535  *           0 if nothing to commit,
 536  *           1 if transaction was successfully committed.
 537  */
 538 static int __jbd2_journal_force_commit(journal_t *journal)
 539 {
 540         transaction_t *transaction = NULL;
 541         tid_t tid;
 542         int need_to_start = 0, ret = 0;
 543
 544         read_lock(&journal->j_state_lock);
 545         if (journal->j_running_transaction && !current->journal_info) {
 546                 transaction = journal->j_running_transaction;
 547                 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
 548                         need_to_start = 1;
 549         } else if (journal->j_committing_transaction)
 550                 transaction = journal->j_committing_transaction;
 551
 552         if (!transaction) {
 553                 /* Nothing to commit */
 554                 read_unlock(&journal->j_state_lock);
 555                 return 0;
 556         }
 557         tid = transaction->t_tid;
 558         read_unlock(&journal->j_state_lock);
 559         if (need_to_start)
 560                 jbd2_log_start_commit(journal, tid);
 561         ret = jbd2_log_wait_commit(journal, tid);
 562         if (!ret)
 563                 ret = 1;
 564
 565         return ret;
 566 }
 567
 568 /**
 569  * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
 570  * calling process is not within transaction.
 571  *
 572  * @journal: journal to force
 573  * Returns true if progress was made.
 574  *
 575  * This is used for forcing out undo-protected data which contains
 576  * bitmaps, when the fs is running out of space.
 577  */
 578 int jbd2_journal_force_commit_nested(journal_t *journal)
 579 {
 580         int ret;
 581
 582         ret = __jbd2_journal_force_commit(journal);
 583         return ret > 0;
 584 }
 585
 586 /**
 587  * jbd2_journal_force_commit() - force any uncommitted transactions
 588  * @journal: journal to force
 589  *
 590  * Caller want unconditional commit. We can only force the running transaction
 591  * if we don't have an active handle, otherwise, we will deadlock.
 592  */
 593 int jbd2_journal_force_commit(journal_t *journal)
 594 {
 595         int ret;
 596
 597         J_ASSERT(!current->journal_info);
 598         ret = __jbd2_journal_force_commit(journal);
 599         if (ret > 0)
 600                 ret = 0;
 601         return ret;
 602 }
 603
 604 /*
 605  * Start a commit of the current running transaction (if any).  Returns true
 606  * if a transaction is going to be committed (or is currently already
 607  * committing), and fills its tid in at *ptid
 608  */
 609 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
 610 {
 611         int ret = 0;
 612
 613         write_lock(&journal->j_state_lock);
 614         if (journal->j_running_transaction) {
 615                 tid_t tid = journal->j_running_transaction->t_tid;
 616
 617                 __jbd2_log_start_commit(journal, tid);
 618                 /* There's a running transaction and we've just made sure
 619                  * it's commit has been scheduled. */
 620                 if (ptid)
 621                         *ptid = tid;
 622                 ret = 1;
 623         } else if (journal->j_committing_transaction) {
 624                 /*
 625                  * If commit has been started, then we have to wait for
 626                  * completion of that transaction.
 627                  */
 628                 if (ptid)
 629                         *ptid = journal->j_committing_transaction->t_tid;
 630                 ret = 1;
 631         }
 632         write_unlock(&journal->j_state_lock);
 633         return ret;
 634 }
 635
 636 /*
 637  * Return 1 if a given transaction has not yet sent barrier request
 638  * connected with a transaction commit. If 0 is returned, transaction
 639  * may or may not have sent the barrier. Used to avoid sending barrier
 640  * twice in common cases.
 641  */
 642 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
 643 {
 644         int ret = 0;
 645         transaction_t *commit_trans;
 646
 647         if (!(journal->j_flags & JBD2_BARRIER))
 648                 return 0;
 649         read_lock(&journal->j_state_lock);
 650         /* Transaction already committed? */
 651         if (tid_geq(journal->j_commit_sequence, tid))
 652                 goto out;
 653         commit_trans = journal->j_committing_transaction;
 654         if (!commit_trans || commit_trans->t_tid != tid) {
 655                 ret = 1;
 656                 goto out;
 657         }
 658         /*
 659          * Transaction is being committed and we already proceeded to
 660          * submitting a flush to fs partition?
 661          */
 662         if (journal->j_fs_dev != journal->j_dev) {
 663                 if (!commit_trans->t_need_data_flush ||
 664                     commit_trans->t_state >= T_COMMIT_DFLUSH)
 665                         goto out;
 666         } else {
 667                 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
 668                         goto out;
 669         }
 670         ret = 1;
 671 out:
 672         read_unlock(&journal->j_state_lock);
 673         return ret;
 674 }
 675 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
 676
 677 /*
 678  * Wait for a specified commit to complete.
 679  * The caller may not hold the journal lock.
 680  */
 681 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
 682 {
 683         int err = 0;
 684
 685         read_lock(&journal->j_state_lock);
 686 #ifdef CONFIG_PROVE_LOCKING
 687         /*
 688          * Some callers make sure transaction is already committing and in that
 689          * case we cannot block on open handles anymore. So don't warn in that
 690          * case.
 691          */
 692         if (tid_gt(tid, journal->j_commit_sequence) &&
 693             (!journal->j_committing_transaction ||
 694              journal->j_committing_transaction->t_tid != tid)) {
 695                 read_unlock(&journal->j_state_lock);
 696                 jbd2_might_wait_for_commit(journal);
 697                 read_lock(&journal->j_state_lock);
 698         }
 699 #endif
 700 #ifdef CONFIG_JBD2_DEBUG
 701         if (!tid_geq(journal->j_commit_request, tid)) {
 702                 printk(KERN_ERR
 703                        "%s: error: j_commit_request=%u, tid=%u\n",
 704                        __func__, journal->j_commit_request, tid);
 705         }
 706 #endif
 707         while (tid_gt(tid, journal->j_commit_sequence)) {
 708                 jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
 709                                   tid, journal->j_commit_sequence);
 710                 read_unlock(&journal->j_state_lock);
 711                 wake_up(&journal->j_wait_commit);
 712                 wait_event(journal->j_wait_done_commit,
 713                                 !tid_gt(tid, journal->j_commit_sequence));
 714                 read_lock(&journal->j_state_lock);
 715         }
 716         read_unlock(&journal->j_state_lock);
 717
 718         if (unlikely(is_journal_aborted(journal)))
 719                 err = -EIO;
 720         return err;
 721 }
 722
 723 /*
 724  * Start a fast commit. If there's an ongoing fast or full commit wait for
 725  * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
 726  * if a fast commit is not needed, either because there's an already a commit
 727  * going on or this tid has already been committed. Returns -EINVAL if no jbd2
 728  * commit has yet been performed.
 729  */
 730 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
 731 {
 732         if (unlikely(is_journal_aborted(journal)))
 733                 return -EIO;
 734         /*
 735          * Fast commits only allowed if at least one full commit has
 736          * been processed.
 737          */
 738         if (!journal->j_stats.ts_tid)
 739                 return -EINVAL;
 740
 741         write_lock(&journal->j_state_lock);
 742         if (tid <= journal->j_commit_sequence) {
 743                 write_unlock(&journal->j_state_lock);
 744                 return -EALREADY;
 745         }
 746
 747         if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
 748             (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
 749                 DEFINE_WAIT(wait);
 750
 751                 prepare_to_wait(&journal->j_fc_wait, &wait,
 752                                 TASK_UNINTERRUPTIBLE);
 753                 write_unlock(&journal->j_state_lock);
 754                 schedule();
 755                 finish_wait(&journal->j_fc_wait, &wait);
 756                 return -EALREADY;
 757         }
 758         journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
 759         write_unlock(&journal->j_state_lock);
 760         jbd2_journal_lock_updates(journal);
 761
 762         return 0;
 763 }
 764 EXPORT_SYMBOL(jbd2_fc_begin_commit);
 765
 766 /*
 767  * Stop a fast commit. If fallback is set, this function starts commit of
 768  * TID tid before any other fast commit can start.
 769  */
 770 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
 771 {
 772         jbd2_journal_unlock_updates(journal);
 773         if (journal->j_fc_cleanup_callback)
 774                 journal->j_fc_cleanup_callback(journal, 0);
 775         write_lock(&journal->j_state_lock);
 776         journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
 777         if (fallback)
 778                 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
 779         write_unlock(&journal->j_state_lock);
 780         wake_up(&journal->j_fc_wait);
 781         if (fallback)
 782                 return jbd2_complete_transaction(journal, tid);
 783         return 0;
 784 }
 785
 786 int jbd2_fc_end_commit(journal_t *journal)
 787 {
 788         return __jbd2_fc_end_commit(journal, 0, false);
 789 }
 790 EXPORT_SYMBOL(jbd2_fc_end_commit);
 791
 792 int jbd2_fc_end_commit_fallback(journal_t *journal)
 793 {
 794         tid_t tid;
 795
 796         read_lock(&journal->j_state_lock);
 797         tid = journal->j_running_transaction ?
 798                 journal->j_running_transaction->t_tid : 0;
 799         read_unlock(&journal->j_state_lock);
 800         return __jbd2_fc_end_commit(journal, tid, true);
 801 }
 802 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
 803
 804 /* Return 1 when transaction with given tid has already committed. */
 805 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
 806 {
 807         int ret = 1;
 808
 809         read_lock(&journal->j_state_lock);
 810         if (journal->j_running_transaction &&
 811             journal->j_running_transaction->t_tid == tid)
 812                 ret = 0;
 813         if (journal->j_committing_transaction &&
 814             journal->j_committing_transaction->t_tid == tid)
 815                 ret = 0;
 816         read_unlock(&journal->j_state_lock);
 817         return ret;
 818 }
 819 EXPORT_SYMBOL(jbd2_transaction_committed);
 820
 821 /*
 822  * When this function returns the transaction corresponding to tid
 823  * will be completed.  If the transaction has currently running, start
 824  * committing that transaction before waiting for it to complete.  If
 825  * the transaction id is stale, it is by definition already completed,
 826  * so just return SUCCESS.
 827  */
 828 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
 829 {
 830         int     need_to_wait = 1;
 831
 832         read_lock(&journal->j_state_lock);
 833         if (journal->j_running_transaction &&
 834             journal->j_running_transaction->t_tid == tid) {
 835                 if (journal->j_commit_request != tid) {
 836                         /* transaction not yet started, so request it */
 837                         read_unlock(&journal->j_state_lock);
 838                         jbd2_log_start_commit(journal, tid);
 839                         goto wait_commit;
 840                 }
 841         } else if (!(journal->j_committing_transaction &&
 842                      journal->j_committing_transaction->t_tid == tid))
 843                 need_to_wait = 0;
 844         read_unlock(&journal->j_state_lock);
 845         if (!need_to_wait)
 846                 return 0;
 847 wait_commit:
 848         return jbd2_log_wait_commit(journal, tid);
 849 }
 850 EXPORT_SYMBOL(jbd2_complete_transaction);
 851
 852 /*
 853  * Log buffer allocation routines:
 854  */
 855
 856 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
 857 {
 858         unsigned long blocknr;
 859
 860         write_lock(&journal->j_state_lock);
 861         J_ASSERT(journal->j_free > 1);
 862
 863         blocknr = journal->j_head;
 864         journal->j_head++;
 865         journal->j_free--;
 866         if (journal->j_head == journal->j_last)
 867                 journal->j_head = journal->j_first;
 868         write_unlock(&journal->j_state_lock);
 869         return jbd2_journal_bmap(journal, blocknr, retp);
 870 }
 871
 872 /* Map one fast commit buffer for use by the file system */
 873 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
 874 {
 875         unsigned long long pblock;
 876         unsigned long blocknr;
 877         int ret = 0;
 878         struct buffer_head *bh;
 879         int fc_off;
 880
 881         *bh_out = NULL;
 882
 883         if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
 884                 fc_off = journal->j_fc_off;
 885                 blocknr = journal->j_fc_first + fc_off;
 886                 journal->j_fc_off++;
 887         } else {
 888                 ret = -EINVAL;
 889         }
 890
 891         if (ret)
 892                 return ret;
 893
 894         ret = jbd2_journal_bmap(journal, blocknr, &pblock);
 895         if (ret)
 896                 return ret;
 897
 898         bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
 899         if (!bh)
 900                 return -ENOMEM;
 901
 902
 903         journal->j_fc_wbuf[fc_off] = bh;
 904
 905         *bh_out = bh;
 906
 907         return 0;
 908 }
 909 EXPORT_SYMBOL(jbd2_fc_get_buf);
 910
 911 /*
 912  * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
 913  * for completion.
 914  */
 915 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
 916 {
 917         struct buffer_head *bh;
 918         int i, j_fc_off;
 919
 920         j_fc_off = journal->j_fc_off;
 921
 922         /*
 923          * Wait in reverse order to minimize chances of us being woken up before
 924          * all IOs have completed
 925          */
 926         for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
 927                 bh = journal->j_fc_wbuf[i];
 928                 wait_on_buffer(bh);
 929                 put_bh(bh);
 930                 journal->j_fc_wbuf[i] = NULL;
 931                 if (unlikely(!buffer_uptodate(bh)))
 932                         return -EIO;
 933         }
 934
 935         return 0;
 936 }
 937 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
 938
 939 int jbd2_fc_release_bufs(journal_t *journal)
 940 {
 941         struct buffer_head *bh;
 942         int i, j_fc_off;
 943
 944         j_fc_off = journal->j_fc_off;
 945
 946         for (i = j_fc_off - 1; i >= 0; i--) {
 947                 bh = journal->j_fc_wbuf[i];
 948                 if (!bh)
 949                         break;
 950                 put_bh(bh);
 951                 journal->j_fc_wbuf[i] = NULL;
 952         }
 953
 954         return 0;
 955 }
 956 EXPORT_SYMBOL(jbd2_fc_release_bufs);
 957
 958 /*
 959  * Conversion of logical to physical block numbers for the journal
 960  *
 961  * On external journals the journal blocks are identity-mapped, so
 962  * this is a no-op.  If needed, we can use j_blk_offset - everything is
 963  * ready.
 964  */
 965 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
 966                  unsigned long long *retp)
 967 {
 968         int err = 0;
 969         unsigned long long ret;
 970         sector_t block = 0;
 971
 972         if (journal->j_inode) {
 973                 block = blocknr;
 974                 ret = bmap(journal->j_inode, &block);
 975
 976                 if (ret || !block) {
 977                         printk(KERN_ALERT "%s: journal block not found "
 978                                         "at offset %lu on %s\n",
 979                                __func__, blocknr, journal->j_devname);
 980                         err = -EIO;
 981                         jbd2_journal_abort(journal, err);
 982                 } else {
 983                         *retp = block;
 984                 }
 985
 986         } else {
 987                 *retp = blocknr; /* +journal->j_blk_offset */
 988         }
 989         return err;
 990 }
 991
 992 /*
 993  * We play buffer_head aliasing tricks to write data/metadata blocks to
 994  * the journal without copying their contents, but for journal
 995  * descriptor blocks we do need to generate bona fide buffers.
 996  *
 997  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
 998  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
 999  * But we don't bother doing that, so there will be coherency problems with
1000  * mmaps of blockdevs which hold live JBD-controlled filesystems.
1001  */
1002 struct buffer_head *
1003 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1004 {
1005         journal_t *journal = transaction->t_journal;
1006         struct buffer_head *bh;
1007         unsigned long long blocknr;
1008         journal_header_t *header;
1009         int err;
1010
1011         err = jbd2_journal_next_log_block(journal, &blocknr);
1012
1013         if (err)
1014                 return NULL;
1015
1016         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1017         if (!bh)
1018                 return NULL;
1019         atomic_dec(&transaction->t_outstanding_credits);
1020         lock_buffer(bh);
1021         memset(bh->b_data, 0, journal->j_blocksize);
1022         header = (journal_header_t *)bh->b_data;
1023         header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1024         header->h_blocktype = cpu_to_be32(type);
1025         header->h_sequence = cpu_to_be32(transaction->t_tid);
1026         set_buffer_uptodate(bh);
1027         unlock_buffer(bh);
1028         BUFFER_TRACE(bh, "return this buffer");
1029         return bh;
1030 }
1031
1032 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1033 {
1034         struct jbd2_journal_block_tail *tail;
1035         __u32 csum;
1036
1037         if (!jbd2_journal_has_csum_v2or3(j))
1038                 return;
1039
1040         tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1041                         sizeof(struct jbd2_journal_block_tail));
1042         tail->t_checksum = 0;
1043         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1044         tail->t_checksum = cpu_to_be32(csum);
1045 }
1046
1047 /*
1048  * Return tid of the oldest transaction in the journal and block in the journal
1049  * where the transaction starts.
1050  *
1051  * If the journal is now empty, return which will be the next transaction ID
1052  * we will write and where will that transaction start.
1053  *
1054  * The return value is 0 if journal tail cannot be pushed any further, 1 if
1055  * it can.
1056  */
1057 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1058                               unsigned long *block)
1059 {
1060         transaction_t *transaction;
1061         int ret;
1062
1063         read_lock(&journal->j_state_lock);
1064         spin_lock(&journal->j_list_lock);
1065         transaction = journal->j_checkpoint_transactions;
1066         if (transaction) {
1067                 *tid = transaction->t_tid;
1068                 *block = transaction->t_log_start;
1069         } else if ((transaction = journal->j_committing_transaction) != NULL) {
1070                 *tid = transaction->t_tid;
1071                 *block = transaction->t_log_start;
1072         } else if ((transaction = journal->j_running_transaction) != NULL) {
1073                 *tid = transaction->t_tid;
1074                 *block = journal->j_head;
1075         } else {
1076                 *tid = journal->j_transaction_sequence;
1077                 *block = journal->j_head;
1078         }
1079         ret = tid_gt(*tid, journal->j_tail_sequence);
1080         spin_unlock(&journal->j_list_lock);
1081         read_unlock(&journal->j_state_lock);
1082
1083         return ret;
1084 }
1085
1086 /*
1087  * Update information in journal structure and in on disk journal superblock
1088  * about log tail. This function does not check whether information passed in
1089  * really pushes log tail further. It's responsibility of the caller to make
1090  * sure provided log tail information is valid (e.g. by holding
1091  * j_checkpoint_mutex all the time between computing log tail and calling this
1092  * function as is the case with jbd2_cleanup_journal_tail()).
1093  *
1094  * Requires j_checkpoint_mutex
1095  */
1096 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1097 {
1098         unsigned long freed;
1099         int ret;
1100
1101         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1102
1103         /*
1104          * We cannot afford for write to remain in drive's caches since as
1105          * soon as we update j_tail, next transaction can start reusing journal
1106          * space and if we lose sb update during power failure we'd replay
1107          * old transaction with possibly newly overwritten data.
1108          */
1109         ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
1110                                               REQ_SYNC | REQ_FUA);
1111         if (ret)
1112                 goto out;
1113
1114         write_lock(&journal->j_state_lock);
1115         freed = block - journal->j_tail;
1116         if (block < journal->j_tail)
1117                 freed += journal->j_last - journal->j_first;
1118
1119         trace_jbd2_update_log_tail(journal, tid, block, freed);
1120         jbd_debug(1,
1121                   "Cleaning journal tail from %u to %u (offset %lu), "
1122                   "freeing %lu\n",
1123                   journal->j_tail_sequence, tid, block, freed);
1124
1125         journal->j_free += freed;
1126         journal->j_tail_sequence = tid;
1127         journal->j_tail = block;
1128         write_unlock(&journal->j_state_lock);
1129
1130 out:
1131         return ret;
1132 }
1133
1134 /*
1135  * This is a variation of __jbd2_update_log_tail which checks for validity of
1136  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1137  * with other threads updating log tail.
1138  */
1139 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1140 {
1141         mutex_lock_io(&journal->j_checkpoint_mutex);
1142         if (tid_gt(tid, journal->j_tail_sequence))
1143                 __jbd2_update_log_tail(journal, tid, block);
1144         mutex_unlock(&journal->j_checkpoint_mutex);
1145 }
1146
1147 struct jbd2_stats_proc_session {
1148         journal_t *journal;
1149         struct transaction_stats_s *stats;
1150         int start;
1151         int max;
1152 };
1153
1154 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1155 {
1156         return *pos ? NULL : SEQ_START_TOKEN;
1157 }
1158
1159 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1160 {
1161         (*pos)++;
1162         return NULL;
1163 }
1164
1165 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1166 {
1167         struct jbd2_stats_proc_session *s = seq->private;
1168
1169         if (v != SEQ_START_TOKEN)
1170                 return 0;
1171         seq_printf(seq, "%lu transactions (%lu requested), "
1172                    "each up to %u blocks\n",
1173                    s->stats->ts_tid, s->stats->ts_requested,
1174                    s->journal->j_max_transaction_buffers);
1175         if (s->stats->ts_tid == 0)
1176                 return 0;
1177         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
1178             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1179         seq_printf(seq, "  %ums request delay\n",
1180             (s->stats->ts_requested == 0) ? 0 :
1181             jiffies_to_msecs(s->stats->run.rs_request_delay /
1182                              s->stats->ts_requested));
1183         seq_printf(seq, "  %ums running transaction\n",
1184             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1185         seq_printf(seq, "  %ums transaction was being locked\n",
1186             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1187         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
1188             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1189         seq_printf(seq, "  %ums logging transaction\n",
1190             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1191         seq_printf(seq, "  %lluus average transaction commit time\n",
1192                    div_u64(s->journal->j_average_commit_time, 1000));
1193         seq_printf(seq, "  %lu handles per transaction\n",
1194             s->stats->run.rs_handle_count / s->stats->ts_tid);
1195         seq_printf(seq, "  %lu blocks per transaction\n",
1196             s->stats->run.rs_blocks / s->stats->ts_tid);
1197         seq_printf(seq, "  %lu logged blocks per transaction\n",
1198             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1199         return 0;
1200 }
1201
1202 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1203 {
1204 }
1205
1206 static const struct seq_operations jbd2_seq_info_ops = {
1207         .start  = jbd2_seq_info_start,
1208         .next   = jbd2_seq_info_next,
1209         .stop   = jbd2_seq_info_stop,
1210         .show   = jbd2_seq_info_show,
1211 };
1212
1213 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1214 {
1215         journal_t *journal = pde_data(inode);
1216         struct jbd2_stats_proc_session *s;
1217         int rc, size;
1218
1219         s = kmalloc(sizeof(*s), GFP_KERNEL);
1220         if (s == NULL)
1221                 return -ENOMEM;
1222         size = sizeof(struct transaction_stats_s);
1223         s->stats = kmalloc(size, GFP_KERNEL);
1224         if (s->stats == NULL) {
1225                 kfree(s);
1226                 return -ENOMEM;
1227         }
1228         spin_lock(&journal->j_history_lock);
1229         memcpy(s->stats, &journal->j_stats, size);
1230         s->journal = journal;
1231         spin_unlock(&journal->j_history_lock);
1232
1233         rc = seq_open(file, &jbd2_seq_info_ops);
1234         if (rc == 0) {
1235                 struct seq_file *m = file->private_data;
1236                 m->private = s;
1237         } else {
1238                 kfree(s->stats);
1239                 kfree(s);
1240         }
1241         return rc;
1242
1243 }
1244
1245 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1246 {
1247         struct seq_file *seq = file->private_data;
1248         struct jbd2_stats_proc_session *s = seq->private;
1249         kfree(s->stats);
1250         kfree(s);
1251         return seq_release(inode, file);
1252 }
1253
1254 static const struct proc_ops jbd2_info_proc_ops = {
1255         .proc_open      = jbd2_seq_info_open,
1256         .proc_read      = seq_read,
1257         .proc_lseek     = seq_lseek,
1258         .proc_release   = jbd2_seq_info_release,
1259 };
1260
1261 static struct proc_dir_entry *proc_jbd2_stats;
1262
1263 static void jbd2_stats_proc_init(journal_t *journal)
1264 {
1265         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1266         if (journal->j_proc_entry) {
1267                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1268                                  &jbd2_info_proc_ops, journal);
1269         }
1270 }
1271
1272 static void jbd2_stats_proc_exit(journal_t *journal)
1273 {
1274         remove_proc_entry("info", journal->j_proc_entry);
1275         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1276 }
1277
1278 /* Minimum size of descriptor tag */
1279 static int jbd2_min_tag_size(void)
1280 {
1281         /*
1282          * Tag with 32-bit block numbers does not use last four bytes of the
1283          * structure
1284          */
1285         return sizeof(journal_block_tag_t) - 4;
1286 }
1287
1288 /**
1289  * jbd2_journal_shrink_scan()
1290  *
1291  * Scan the checkpointed buffer on the checkpoint list and release the
1292  * journal_head.
1293  */
1294 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1295                                               struct shrink_control *sc)
1296 {
1297         journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1298         unsigned long nr_to_scan = sc->nr_to_scan;
1299         unsigned long nr_shrunk;
1300         unsigned long count;
1301
1302         count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1303         trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1304
1305         nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1306
1307         count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1308         trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1309
1310         return nr_shrunk;
1311 }
1312
1313 /**
1314  * jbd2_journal_shrink_count()
1315  *
1316  * Count the number of checkpoint buffers on the checkpoint list.
1317  */
1318 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1319                                                struct shrink_control *sc)
1320 {
1321         journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1322         unsigned long count;
1323
1324         count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1325         trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1326
1327         return count;
1328 }
1329
1330 /*
1331  * Management for journal control blocks: functions to create and
1332  * destroy journal_t structures, and to initialise and read existing
1333  * journal blocks from disk.  */
1334
1335 /* First: create and setup a journal_t object in memory.  We initialise
1336  * very few fields yet: that has to wait until we have created the
1337  * journal structures from from scratch, or loaded them from disk. */
1338
1339 static journal_t *journal_init_common(struct block_device *bdev,
1340                         struct block_device *fs_dev,
1341                         unsigned long long start, int len, int blocksize)
1342 {
1343         static struct lock_class_key jbd2_trans_commit_key;
1344         journal_t *journal;
1345         int err;
1346         struct buffer_head *bh;
1347         int n;
1348
1349         journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1350         if (!journal)
1351                 return NULL;
1352
1353         init_waitqueue_head(&journal->j_wait_transaction_locked);
1354         init_waitqueue_head(&journal->j_wait_done_commit);
1355         init_waitqueue_head(&journal->j_wait_commit);
1356         init_waitqueue_head(&journal->j_wait_updates);
1357         init_waitqueue_head(&journal->j_wait_reserved);
1358         init_waitqueue_head(&journal->j_fc_wait);
1359         mutex_init(&journal->j_abort_mutex);
1360         mutex_init(&journal->j_barrier);
1361         mutex_init(&journal->j_checkpoint_mutex);
1362         spin_lock_init(&journal->j_revoke_lock);
1363         spin_lock_init(&journal->j_list_lock);
1364         rwlock_init(&journal->j_state_lock);
1365
1366         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1367         journal->j_min_batch_time = 0;
1368         journal->j_max_batch_time = 15000; /* 15ms */
1369         atomic_set(&journal->j_reserved_credits, 0);
1370
1371         /* The journal is marked for error until we succeed with recovery! */
1372         journal->j_flags = JBD2_ABORT;
1373
1374         /* Set up a default-sized revoke table for the new mount. */
1375         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1376         if (err)
1377                 goto err_cleanup;
1378
1379         spin_lock_init(&journal->j_history_lock);
1380
1381         lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1382                          &jbd2_trans_commit_key, 0);
1383
1384         /* journal descriptor can store up to n blocks -bzzz */
1385         journal->j_blocksize = blocksize;
1386         journal->j_dev = bdev;
1387         journal->j_fs_dev = fs_dev;
1388         journal->j_blk_offset = start;
1389         journal->j_total_len = len;
1390         /* We need enough buffers to write out full descriptor block. */
1391         n = journal->j_blocksize / jbd2_min_tag_size();
1392         journal->j_wbufsize = n;
1393         journal->j_fc_wbuf = NULL;
1394         journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1395                                         GFP_KERNEL);
1396         if (!journal->j_wbuf)
1397                 goto err_cleanup;
1398
1399         bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1400         if (!bh) {
1401                 pr_err("%s: Cannot get buffer for journal superblock\n",
1402                         __func__);
1403                 goto err_cleanup;
1404         }
1405         journal->j_sb_buffer = bh;
1406         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1407
1408         journal->j_shrink_transaction = NULL;
1409         journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
1410         journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
1411         journal->j_shrinker.seeks = DEFAULT_SEEKS;
1412         journal->j_shrinker.batch = journal->j_max_transaction_buffers;
1413
1414         if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL))
1415                 goto err_cleanup;
1416
1417         if (register_shrinker(&journal->j_shrinker)) {
1418                 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1419                 goto err_cleanup;
1420         }
1421         return journal;
1422
1423 err_cleanup:
1424         brelse(journal->j_sb_buffer);
1425         kfree(journal->j_wbuf);
1426         jbd2_journal_destroy_revoke(journal);
1427         kfree(journal);
1428         return NULL;
1429 }
1430
1431 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1432  *
1433  * Create a journal structure assigned some fixed set of disk blocks to
1434  * the journal.  We don't actually touch those disk blocks yet, but we
1435  * need to set up all of the mapping information to tell the journaling
1436  * system where the journal blocks are.
1437  *
1438  */
1439
1440 /**
1441  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1442  *  @bdev: Block device on which to create the journal
1443  *  @fs_dev: Device which hold journalled filesystem for this journal.
1444  *  @start: Block nr Start of journal.
1445  *  @len:  Length of the journal in blocks.
1446  *  @blocksize: blocksize of journalling device
1447  *
1448  *  Returns: a newly created journal_t *
1449  *
1450  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1451  *  range of blocks on an arbitrary block device.
1452  *
1453  */
1454 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1455                         struct block_device *fs_dev,
1456                         unsigned long long start, int len, int blocksize)
1457 {
1458         journal_t *journal;
1459
1460         journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1461         if (!journal)
1462                 return NULL;
1463
1464         bdevname(journal->j_dev, journal->j_devname);
1465         strreplace(journal->j_devname, '/', '!');
1466         jbd2_stats_proc_init(journal);
1467
1468         return journal;
1469 }
1470
1471 /**
1472  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1473  *  @inode: An inode to create the journal in
1474  *
1475  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1476  * the journal.  The inode must exist already, must support bmap() and
1477  * must have all data blocks preallocated.
1478  */
1479 journal_t *jbd2_journal_init_inode(struct inode *inode)
1480 {
1481         journal_t *journal;
1482         sector_t blocknr;
1483         char *p;
1484         int err = 0;
1485
1486         blocknr = 0;
1487         err = bmap(inode, &blocknr);
1488
1489         if (err || !blocknr) {
1490                 pr_err("%s: Cannot locate journal superblock\n",
1491                         __func__);
1492                 return NULL;
1493         }
1494
1495         jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1496                   inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1497                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1498
1499         journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1500                         blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1501                         inode->i_sb->s_blocksize);
1502         if (!journal)
1503                 return NULL;
1504
1505         journal->j_inode = inode;
1506         bdevname(journal->j_dev, journal->j_devname);
1507         p = strreplace(journal->j_devname, '/', '!');
1508         sprintf(p, "-%lu", journal->j_inode->i_ino);
1509         jbd2_stats_proc_init(journal);
1510
1511         return journal;
1512 }
1513
1514 /*
1515  * If the journal init or create aborts, we need to mark the journal
1516  * superblock as being NULL to prevent the journal destroy from writing
1517  * back a bogus superblock.
1518  */
1519 static void journal_fail_superblock(journal_t *journal)
1520 {
1521         struct buffer_head *bh = journal->j_sb_buffer;
1522         brelse(bh);
1523         journal->j_sb_buffer = NULL;
1524 }
1525
1526 /*
1527  * Given a journal_t structure, initialise the various fields for
1528  * startup of a new journaling session.  We use this both when creating
1529  * a journal, and after recovering an old journal to reset it for
1530  * subsequent use.
1531  */
1532
1533 static int journal_reset(journal_t *journal)
1534 {
1535         journal_superblock_t *sb = journal->j_superblock;
1536         unsigned long long first, last;
1537
1538         first = be32_to_cpu(sb->s_first);
1539         last = be32_to_cpu(sb->s_maxlen);
1540         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1541                 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1542                        first, last);
1543                 journal_fail_superblock(journal);
1544                 return -EINVAL;
1545         }
1546
1547         journal->j_first = first;
1548         journal->j_last = last;
1549
1550         journal->j_head = journal->j_first;
1551         journal->j_tail = journal->j_first;
1552         journal->j_free = journal->j_last - journal->j_first;
1553
1554         journal->j_tail_sequence = journal->j_transaction_sequence;
1555         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1556         journal->j_commit_request = journal->j_commit_sequence;
1557
1558         journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1559
1560         /*
1561          * Now that journal recovery is done, turn fast commits off here. This
1562          * way, if fast commit was enabled before the crash but if now FS has
1563          * disabled it, we don't enable fast commits.
1564          */
1565         jbd2_clear_feature_fast_commit(journal);
1566
1567         /*
1568          * As a special case, if the on-disk copy is already marked as needing
1569          * no recovery (s_start == 0), then we can safely defer the superblock
1570          * update until the next commit by setting JBD2_FLUSHED.  This avoids
1571          * attempting a write to a potential-readonly device.
1572          */
1573         if (sb->s_start == 0) {
1574                 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1575                         "(start %ld, seq %u, errno %d)\n",
1576                         journal->j_tail, journal->j_tail_sequence,
1577                         journal->j_errno);
1578                 journal->j_flags |= JBD2_FLUSHED;
1579         } else {
1580                 /* Lock here to make assertions happy... */
1581                 mutex_lock_io(&journal->j_checkpoint_mutex);
1582                 /*
1583                  * Update log tail information. We use REQ_FUA since new
1584                  * transaction will start reusing journal space and so we
1585                  * must make sure information about current log tail is on
1586                  * disk before that.
1587                  */
1588                 jbd2_journal_update_sb_log_tail(journal,
1589                                                 journal->j_tail_sequence,
1590                                                 journal->j_tail,
1591                                                 REQ_SYNC | REQ_FUA);
1592                 mutex_unlock(&journal->j_checkpoint_mutex);
1593         }
1594         return jbd2_journal_start_thread(journal);
1595 }
1596
1597 /*
1598  * This function expects that the caller will have locked the journal
1599  * buffer head, and will return with it unlocked
1600  */
1601 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1602 {
1603         struct buffer_head *bh = journal->j_sb_buffer;
1604         journal_superblock_t *sb = journal->j_superblock;
1605         int ret;
1606
1607         /* Buffer got discarded which means block device got invalidated */
1608         if (!buffer_mapped(bh)) {
1609                 unlock_buffer(bh);
1610                 return -EIO;
1611         }
1612
1613         trace_jbd2_write_superblock(journal, write_flags);
1614         if (!(journal->j_flags & JBD2_BARRIER))
1615                 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1616         if (buffer_write_io_error(bh)) {
1617                 /*
1618                  * Oh, dear.  A previous attempt to write the journal
1619                  * superblock failed.  This could happen because the
1620                  * USB device was yanked out.  Or it could happen to
1621                  * be a transient write error and maybe the block will
1622                  * be remapped.  Nothing we can do but to retry the
1623                  * write and hope for the best.
1624                  */
1625                 printk(KERN_ERR "JBD2: previous I/O error detected "
1626                        "for journal superblock update for %s.\n",
1627                        journal->j_devname);
1628                 clear_buffer_write_io_error(bh);
1629                 set_buffer_uptodate(bh);
1630         }
1631         if (jbd2_journal_has_csum_v2or3(journal))
1632                 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1633         get_bh(bh);
1634         bh->b_end_io = end_buffer_write_sync;
1635         ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1636         wait_on_buffer(bh);
1637         if (buffer_write_io_error(bh)) {
1638                 clear_buffer_write_io_error(bh);
1639                 set_buffer_uptodate(bh);
1640                 ret = -EIO;
1641         }
1642         if (ret) {
1643                 printk(KERN_ERR "JBD2: Error %d detected when updating "
1644                        "journal superblock for %s.\n", ret,
1645                        journal->j_devname);
1646                 if (!is_journal_aborted(journal))
1647                         jbd2_journal_abort(journal, ret);
1648         }
1649
1650         return ret;
1651 }
1652
1653 /**
1654  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1655  * @journal: The journal to update.
1656  * @tail_tid: TID of the new transaction at the tail of the log
1657  * @tail_block: The first block of the transaction at the tail of the log
1658  * @write_op: With which operation should we write the journal sb
1659  *
1660  * Update a journal's superblock information about log tail and write it to
1661  * disk, waiting for the IO to complete.
1662  */
1663 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1664                                      unsigned long tail_block, int write_op)
1665 {
1666         journal_superblock_t *sb = journal->j_superblock;
1667         int ret;
1668
1669         if (is_journal_aborted(journal))
1670                 return -EIO;
1671         if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) {
1672                 jbd2_journal_abort(journal, -EIO);
1673                 return -EIO;
1674         }
1675
1676         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1677         jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1678                   tail_block, tail_tid);
1679
1680         lock_buffer(journal->j_sb_buffer);
1681         sb->s_sequence = cpu_to_be32(tail_tid);
1682         sb->s_start    = cpu_to_be32(tail_block);
1683
1684         ret = jbd2_write_superblock(journal, write_op);
1685         if (ret)
1686                 goto out;
1687
1688         /* Log is no longer empty */
1689         write_lock(&journal->j_state_lock);
1690         WARN_ON(!sb->s_sequence);
1691         journal->j_flags &= ~JBD2_FLUSHED;
1692         write_unlock(&journal->j_state_lock);
1693
1694 out:
1695         return ret;
1696 }
1697
1698 /**
1699  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1700  * @journal: The journal to update.
1701  * @write_op: With which operation should we write the journal sb
1702  *
1703  * Update a journal's dynamic superblock fields to show that journal is empty.
1704  * Write updated superblock to disk waiting for IO to complete.
1705  */
1706 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1707 {
1708         journal_superblock_t *sb = journal->j_superblock;
1709         bool had_fast_commit = false;
1710
1711         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1712         lock_buffer(journal->j_sb_buffer);
1713         if (sb->s_start == 0) {         /* Is it already empty? */
1714                 unlock_buffer(journal->j_sb_buffer);
1715                 return;
1716         }
1717
1718         jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1719                   journal->j_tail_sequence);
1720
1721         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1722         sb->s_start    = cpu_to_be32(0);
1723         if (jbd2_has_feature_fast_commit(journal)) {
1724                 /*
1725                  * When journal is clean, no need to commit fast commit flag and
1726                  * make file system incompatible with older kernels.
1727                  */
1728                 jbd2_clear_feature_fast_commit(journal);
1729                 had_fast_commit = true;
1730         }
1731
1732         jbd2_write_superblock(journal, write_op);
1733
1734         if (had_fast_commit)
1735                 jbd2_set_feature_fast_commit(journal);
1736
1737         /* Log is no longer empty */
1738         write_lock(&journal->j_state_lock);
1739         journal->j_flags |= JBD2_FLUSHED;
1740         write_unlock(&journal->j_state_lock);
1741 }
1742
1743 /**
1744  * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1745  * @journal: The journal to erase.
1746  * @flags: A discard/zeroout request is sent for each physically contigous
1747  *      region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1748  *      JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1749  *      to perform.
1750  *
1751  * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1752  * will be explicitly written if no hardware offload is available, see
1753  * blkdev_issue_zeroout for more details.
1754  */
1755 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1756 {
1757         int err = 0;
1758         unsigned long block, log_offset; /* logical */
1759         unsigned long long phys_block, block_start, block_stop; /* physical */
1760         loff_t byte_start, byte_stop, byte_count;
1761         struct request_queue *q = bdev_get_queue(journal->j_dev);
1762
1763         /* flags must be set to either discard or zeroout */
1764         if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1765                         ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1766                         (flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1767                 return -EINVAL;
1768
1769         if (!q)
1770                 return -ENXIO;
1771
1772         if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && !blk_queue_discard(q))
1773                 return -EOPNOTSUPP;
1774
1775         /*
1776          * lookup block mapping and issue discard/zeroout for each
1777          * contiguous region
1778          */
1779         log_offset = be32_to_cpu(journal->j_superblock->s_first);
1780         block_start =  ~0ULL;
1781         for (block = log_offset; block < journal->j_total_len; block++) {
1782                 err = jbd2_journal_bmap(journal, block, &phys_block);
1783                 if (err) {
1784                         pr_err("JBD2: bad block at offset %lu", block);
1785                         return err;
1786                 }
1787
1788                 if (block_start == ~0ULL) {
1789                         block_start = phys_block;
1790                         block_stop = block_start - 1;
1791                 }
1792
1793                 /*
1794                  * last block not contiguous with current block,
1795                  * process last contiguous region and return to this block on
1796                  * next loop
1797                  */
1798                 if (phys_block != block_stop + 1) {
1799                         block--;
1800                 } else {
1801                         block_stop++;
1802                         /*
1803                          * if this isn't the last block of journal,
1804                          * no need to process now because next block may also
1805                          * be part of this contiguous region
1806                          */
1807                         if (block != journal->j_total_len - 1)
1808                                 continue;
1809                 }
1810
1811                 /*
1812                  * end of contiguous region or this is last block of journal,
1813                  * take care of the region
1814                  */
1815                 byte_start = block_start * journal->j_blocksize;
1816                 byte_stop = block_stop * journal->j_blocksize;
1817                 byte_count = (block_stop - block_start + 1) *
1818                                 journal->j_blocksize;
1819
1820                 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
1821                                 byte_start, byte_stop);
1822
1823                 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
1824                         err = blkdev_issue_discard(journal->j_dev,
1825                                         byte_start >> SECTOR_SHIFT,
1826                                         byte_count >> SECTOR_SHIFT,
1827                                         GFP_NOFS, 0);
1828                 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
1829                         err = blkdev_issue_zeroout(journal->j_dev,
1830                                         byte_start >> SECTOR_SHIFT,
1831                                         byte_count >> SECTOR_SHIFT,
1832                                         GFP_NOFS, 0);
1833                 }
1834
1835                 if (unlikely(err != 0)) {
1836                         pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
1837                                         err, block_start, block_stop);
1838                         return err;
1839                 }
1840
1841                 /* reset start and stop after processing a region */
1842                 block_start = ~0ULL;
1843         }
1844
1845         return blkdev_issue_flush(journal->j_dev);
1846 }
1847
1848 /**
1849  * jbd2_journal_update_sb_errno() - Update error in the journal.
1850  * @journal: The journal to update.
1851  *
1852  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1853  * to complete.
1854  */
1855 void jbd2_journal_update_sb_errno(journal_t *journal)
1856 {
1857         journal_superblock_t *sb = journal->j_superblock;
1858         int errcode;
1859
1860         lock_buffer(journal->j_sb_buffer);
1861         errcode = journal->j_errno;
1862         if (errcode == -ESHUTDOWN)
1863                 errcode = 0;
1864         jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1865         sb->s_errno    = cpu_to_be32(errcode);
1866
1867         jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1868 }
1869 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1870
1871 static int journal_revoke_records_per_block(journal_t *journal)
1872 {
1873         int record_size;
1874         int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1875
1876         if (jbd2_has_feature_64bit(journal))
1877                 record_size = 8;
1878         else
1879                 record_size = 4;
1880
1881         if (jbd2_journal_has_csum_v2or3(journal))
1882                 space -= sizeof(struct jbd2_journal_block_tail);
1883         return space / record_size;
1884 }
1885
1886 /*
1887  * Read the superblock for a given journal, performing initial
1888  * validation of the format.
1889  */
1890 static int journal_get_superblock(journal_t *journal)
1891 {
1892         struct buffer_head *bh;
1893         journal_superblock_t *sb;
1894         int err = -EIO;
1895
1896         bh = journal->j_sb_buffer;
1897
1898         J_ASSERT(bh != NULL);
1899         if (!buffer_uptodate(bh)) {
1900                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1901                 wait_on_buffer(bh);
1902                 if (!buffer_uptodate(bh)) {
1903                         printk(KERN_ERR
1904                                 "JBD2: IO error reading journal superblock\n");
1905                         goto out;
1906                 }
1907         }
1908
1909         if (buffer_verified(bh))
1910                 return 0;
1911
1912         sb = journal->j_superblock;
1913
1914         err = -EINVAL;
1915
1916         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1917             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1918                 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1919                 goto out;
1920         }
1921
1922         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1923         case JBD2_SUPERBLOCK_V1:
1924                 journal->j_format_version = 1;
1925                 break;
1926         case JBD2_SUPERBLOCK_V2:
1927                 journal->j_format_version = 2;
1928                 break;
1929         default:
1930                 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1931                 goto out;
1932         }
1933
1934         if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1935                 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1936         else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1937                 printk(KERN_WARNING "JBD2: journal file too short\n");
1938                 goto out;
1939         }
1940
1941         if (be32_to_cpu(sb->s_first) == 0 ||
1942             be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1943                 printk(KERN_WARNING
1944                         "JBD2: Invalid start block of journal: %u\n",
1945                         be32_to_cpu(sb->s_first));
1946                 goto out;
1947         }
1948
1949         if (jbd2_has_feature_csum2(journal) &&
1950             jbd2_has_feature_csum3(journal)) {
1951                 /* Can't have checksum v2 and v3 at the same time! */
1952                 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1953                        "at the same time!\n");
1954                 goto out;
1955         }
1956
1957         if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1958             jbd2_has_feature_checksum(journal)) {
1959                 /* Can't have checksum v1 and v2 on at the same time! */
1960                 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1961                        "at the same time!\n");
1962                 goto out;
1963         }
1964
1965         if (!jbd2_verify_csum_type(journal, sb)) {
1966                 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1967                 goto out;
1968         }
1969
1970         /* Load the checksum driver */
1971         if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1972                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1973                 if (IS_ERR(journal->j_chksum_driver)) {
1974                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1975                         err = PTR_ERR(journal->j_chksum_driver);
1976                         journal->j_chksum_driver = NULL;
1977                         goto out;
1978                 }
1979         }
1980
1981         if (jbd2_journal_has_csum_v2or3(journal)) {
1982                 /* Check superblock checksum */
1983                 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1984                         printk(KERN_ERR "JBD2: journal checksum error\n");
1985                         err = -EFSBADCRC;
1986                         goto out;
1987                 }
1988
1989                 /* Precompute checksum seed for all metadata */
1990                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1991                                                    sizeof(sb->s_uuid));
1992         }
1993
1994         journal->j_revoke_records_per_block =
1995                                 journal_revoke_records_per_block(journal);
1996         set_buffer_verified(bh);
1997
1998         return 0;
1999
2000 out:
2001         journal_fail_superblock(journal);
2002         return err;
2003 }
2004
2005 /*
2006  * Load the on-disk journal superblock and read the key fields into the
2007  * journal_t.
2008  */
2009
2010 static int load_superblock(journal_t *journal)
2011 {
2012         int err;
2013         journal_superblock_t *sb;
2014         int num_fc_blocks;
2015
2016         err = journal_get_superblock(journal);
2017         if (err)
2018                 return err;
2019
2020         sb = journal->j_superblock;
2021
2022         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
2023         journal->j_tail = be32_to_cpu(sb->s_start);
2024         journal->j_first = be32_to_cpu(sb->s_first);
2025         journal->j_errno = be32_to_cpu(sb->s_errno);
2026         journal->j_last = be32_to_cpu(sb->s_maxlen);
2027
2028         if (jbd2_has_feature_fast_commit(journal)) {
2029                 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
2030                 num_fc_blocks = jbd2_journal_get_num_fc_blks(sb);
2031                 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
2032                         journal->j_last = journal->j_fc_last - num_fc_blocks;
2033                 journal->j_fc_first = journal->j_last + 1;
2034                 journal->j_fc_off = 0;
2035         }
2036
2037         return 0;
2038 }
2039
2040
2041 /**
2042  * jbd2_journal_load() - Read journal from disk.
2043  * @journal: Journal to act on.
2044  *
2045  * Given a journal_t structure which tells us which disk blocks contain
2046  * a journal, read the journal from disk to initialise the in-memory
2047  * structures.
2048  */
2049 int jbd2_journal_load(journal_t *journal)
2050 {
2051         int err;
2052         journal_superblock_t *sb;
2053
2054         err = load_superblock(journal);
2055         if (err)
2056                 return err;
2057
2058         sb = journal->j_superblock;
2059         /* If this is a V2 superblock, then we have to check the
2060          * features flags on it. */
2061
2062         if (journal->j_format_version >= 2) {
2063                 if ((sb->s_feature_ro_compat &
2064                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
2065                     (sb->s_feature_incompat &
2066                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
2067                         printk(KERN_WARNING
2068                                 "JBD2: Unrecognised features on journal\n");
2069                         return -EINVAL;
2070                 }
2071         }
2072
2073         /*
2074          * Create a slab for this blocksize
2075          */
2076         err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2077         if (err)
2078                 return err;
2079
2080         /* Let the recovery code check whether it needs to recover any
2081          * data from the journal. */
2082         if (jbd2_journal_recover(journal))
2083                 goto recovery_error;
2084
2085         if (journal->j_failed_commit) {
2086                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
2087                        "is corrupt.\n", journal->j_failed_commit,
2088                        journal->j_devname);
2089                 return -EFSCORRUPTED;
2090         }
2091         /*
2092          * clear JBD2_ABORT flag initialized in journal_init_common
2093          * here to update log tail information with the newest seq.
2094          */
2095         journal->j_flags &= ~JBD2_ABORT;
2096
2097         /* OK, we've finished with the dynamic journal bits:
2098          * reinitialise the dynamic contents of the superblock in memory
2099          * and reset them on disk. */
2100         if (journal_reset(journal))
2101                 goto recovery_error;
2102
2103         journal->j_flags |= JBD2_LOADED;
2104         return 0;
2105
2106 recovery_error:
2107         printk(KERN_WARNING "JBD2: recovery failed\n");
2108         return -EIO;
2109 }
2110
2111 /**
2112  * jbd2_journal_destroy() - Release a journal_t structure.
2113  * @journal: Journal to act on.
2114  *
2115  * Release a journal_t structure once it is no longer in use by the
2116  * journaled object.
2117  * Return <0 if we couldn't clean up the journal.
2118  */
2119 int jbd2_journal_destroy(journal_t *journal)
2120 {
2121         int err = 0;
2122
2123         /* Wait for the commit thread to wake up and die. */
2124         journal_kill_thread(journal);
2125
2126         /* Force a final log commit */
2127         if (journal->j_running_transaction)
2128                 jbd2_journal_commit_transaction(journal);
2129
2130         /* Force any old transactions to disk */
2131
2132         /* Totally anal locking here... */
2133         spin_lock(&journal->j_list_lock);
2134         while (journal->j_checkpoint_transactions != NULL) {
2135                 spin_unlock(&journal->j_list_lock);
2136                 mutex_lock_io(&journal->j_checkpoint_mutex);
2137                 err = jbd2_log_do_checkpoint(journal);
2138                 mutex_unlock(&journal->j_checkpoint_mutex);
2139                 /*
2140                  * If checkpointing failed, just free the buffers to avoid
2141                  * looping forever
2142                  */
2143                 if (err) {
2144                         jbd2_journal_destroy_checkpoint(journal);
2145                         spin_lock(&journal->j_list_lock);
2146                         break;
2147                 }
2148                 spin_lock(&journal->j_list_lock);
2149         }
2150
2151         J_ASSERT(journal->j_running_transaction == NULL);
2152         J_ASSERT(journal->j_committing_transaction == NULL);
2153         J_ASSERT(journal->j_checkpoint_transactions == NULL);
2154         spin_unlock(&journal->j_list_lock);
2155
2156         /*
2157          * OK, all checkpoint transactions have been checked, now check the
2158          * write out io error flag and abort the journal if some buffer failed
2159          * to write back to the original location, otherwise the filesystem
2160          * may become inconsistent.
2161          */
2162         if (!is_journal_aborted(journal) &&
2163             test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags))
2164                 jbd2_journal_abort(journal, -EIO);
2165
2166         if (journal->j_sb_buffer) {
2167                 if (!is_journal_aborted(journal)) {
2168                         mutex_lock_io(&journal->j_checkpoint_mutex);
2169
2170                         write_lock(&journal->j_state_lock);
2171                         journal->j_tail_sequence =
2172                                 ++journal->j_transaction_sequence;
2173                         write_unlock(&journal->j_state_lock);
2174
2175                         jbd2_mark_journal_empty(journal,
2176                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2177                         mutex_unlock(&journal->j_checkpoint_mutex);
2178                 } else
2179                         err = -EIO;
2180                 brelse(journal->j_sb_buffer);
2181         }
2182
2183         if (journal->j_shrinker.flags & SHRINKER_REGISTERED) {
2184                 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2185                 unregister_shrinker(&journal->j_shrinker);
2186         }
2187         if (journal->j_proc_entry)
2188                 jbd2_stats_proc_exit(journal);
2189         iput(journal->j_inode);
2190         if (journal->j_revoke)
2191                 jbd2_journal_destroy_revoke(journal);
2192         if (journal->j_chksum_driver)
2193                 crypto_free_shash(journal->j_chksum_driver);
2194         kfree(journal->j_fc_wbuf);
2195         kfree(journal->j_wbuf);
2196         kfree(journal);
2197
2198         return err;
2199 }
2200
2201
2202 /**
2203  * jbd2_journal_check_used_features() - Check if features specified are used.
2204  * @journal: Journal to check.
2205  * @compat: bitmask of compatible features
2206  * @ro: bitmask of features that force read-only mount
2207  * @incompat: bitmask of incompatible features
2208  *
2209  * Check whether the journal uses all of a given set of
2210  * features.  Return true (non-zero) if it does.
2211  **/
2212
2213 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2214                                  unsigned long ro, unsigned long incompat)
2215 {
2216         journal_superblock_t *sb;
2217
2218         if (!compat && !ro && !incompat)
2219                 return 1;
2220         /* Load journal superblock if it is not loaded yet. */
2221         if (journal->j_format_version == 0 &&
2222             journal_get_superblock(journal) != 0)
2223                 return 0;
2224         if (journal->j_format_version == 1)
2225                 return 0;
2226
2227         sb = journal->j_superblock;
2228
2229         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2230             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2231             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2232                 return 1;
2233
2234         return 0;
2235 }
2236
2237 /**
2238  * jbd2_journal_check_available_features() - Check feature set in journalling layer
2239  * @journal: Journal to check.
2240  * @compat: bitmask of compatible features
2241  * @ro: bitmask of features that force read-only mount
2242  * @incompat: bitmask of incompatible features
2243  *
2244  * Check whether the journaling code supports the use of
2245  * all of a given set of features on this journal.  Return true
2246  * (non-zero) if it can. */
2247
2248 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2249                                       unsigned long ro, unsigned long incompat)
2250 {
2251         if (!compat && !ro && !incompat)
2252                 return 1;
2253
2254         /* We can support any known requested features iff the
2255          * superblock is in version 2.  Otherwise we fail to support any
2256          * extended sb features. */
2257
2258         if (journal->j_format_version != 2)
2259                 return 0;
2260
2261         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2262             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2263             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2264                 return 1;
2265
2266         return 0;
2267 }
2268
2269 static int
2270 jbd2_journal_initialize_fast_commit(journal_t *journal)
2271 {
2272         journal_superblock_t *sb = journal->j_superblock;
2273         unsigned long long num_fc_blks;
2274
2275         num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2276         if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2277                 return -ENOSPC;
2278
2279         /* Are we called twice? */
2280         WARN_ON(journal->j_fc_wbuf != NULL);
2281         journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2282                                 sizeof(struct buffer_head *), GFP_KERNEL);
2283         if (!journal->j_fc_wbuf)
2284                 return -ENOMEM;
2285
2286         journal->j_fc_wbufsize = num_fc_blks;
2287         journal->j_fc_last = journal->j_last;
2288         journal->j_last = journal->j_fc_last - num_fc_blks;
2289         journal->j_fc_first = journal->j_last + 1;
2290         journal->j_fc_off = 0;
2291         journal->j_free = journal->j_last - journal->j_first;
2292         journal->j_max_transaction_buffers =
2293                 jbd2_journal_get_max_txn_bufs(journal);
2294
2295         return 0;
2296 }
2297
2298 /**
2299  * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2300  * @journal: Journal to act on.
2301  * @compat: bitmask of compatible features
2302  * @ro: bitmask of features that force read-only mount
2303  * @incompat: bitmask of incompatible features
2304  *
2305  * Mark a given journal feature as present on the
2306  * superblock.  Returns true if the requested features could be set.
2307  *
2308  */
2309
2310 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2311                           unsigned long ro, unsigned long incompat)
2312 {
2313 #define INCOMPAT_FEATURE_ON(f) \
2314                 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2315 #define COMPAT_FEATURE_ON(f) \
2316                 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2317         journal_superblock_t *sb;
2318
2319         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2320                 return 1;
2321
2322         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2323                 return 0;
2324
2325         /* If enabling v2 checksums, turn on v3 instead */
2326         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2327                 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2328                 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2329         }
2330
2331         /* Asking for checksumming v3 and v1?  Only give them v3. */
2332         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2333             compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2334                 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2335
2336         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2337                   compat, ro, incompat);
2338
2339         sb = journal->j_superblock;
2340
2341         if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2342                 if (jbd2_journal_initialize_fast_commit(journal)) {
2343                         pr_err("JBD2: Cannot enable fast commits.\n");
2344                         return 0;
2345                 }
2346         }
2347
2348         /* Load the checksum driver if necessary */
2349         if ((journal->j_chksum_driver == NULL) &&
2350             INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2351                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2352                 if (IS_ERR(journal->j_chksum_driver)) {
2353                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2354                         journal->j_chksum_driver = NULL;
2355                         return 0;
2356                 }
2357                 /* Precompute checksum seed for all metadata */
2358                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2359                                                    sizeof(sb->s_uuid));
2360         }
2361
2362         lock_buffer(journal->j_sb_buffer);
2363
2364         /* If enabling v3 checksums, update superblock */
2365         if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2366                 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2367                 sb->s_feature_compat &=
2368                         ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2369         }
2370
2371         /* If enabling v1 checksums, downgrade superblock */
2372         if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2373                 sb->s_feature_incompat &=
2374                         ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2375                                      JBD2_FEATURE_INCOMPAT_CSUM_V3);
2376
2377         sb->s_feature_compat    |= cpu_to_be32(compat);
2378         sb->s_feature_ro_compat |= cpu_to_be32(ro);
2379         sb->s_feature_incompat  |= cpu_to_be32(incompat);
2380         unlock_buffer(journal->j_sb_buffer);
2381         journal->j_revoke_records_per_block =
2382                                 journal_revoke_records_per_block(journal);
2383
2384         return 1;
2385 #undef COMPAT_FEATURE_ON
2386 #undef INCOMPAT_FEATURE_ON
2387 }
2388
2389 /*
2390  * jbd2_journal_clear_features() - Clear a given journal feature in the
2391  *                                  superblock
2392  * @journal: Journal to act on.
2393  * @compat: bitmask of compatible features
2394  * @ro: bitmask of features that force read-only mount
2395  * @incompat: bitmask of incompatible features
2396  *
2397  * Clear a given journal feature as present on the
2398  * superblock.
2399  */
2400 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2401                                 unsigned long ro, unsigned long incompat)
2402 {
2403         journal_superblock_t *sb;
2404
2405         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2406                   compat, ro, incompat);
2407
2408         sb = journal->j_superblock;
2409
2410         sb->s_feature_compat    &= ~cpu_to_be32(compat);
2411         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2412         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
2413         journal->j_revoke_records_per_block =
2414                                 journal_revoke_records_per_block(journal);
2415 }
2416 EXPORT_SYMBOL(jbd2_journal_clear_features);
2417
2418 /**
2419  * jbd2_journal_flush() - Flush journal
2420  * @journal: Journal to act on.
2421  * @flags: optional operation on the journal blocks after the flush (see below)
2422  *
2423  * Flush all data for a given journal to disk and empty the journal.
2424  * Filesystems can use this when remounting readonly to ensure that
2425  * recovery does not need to happen on remount. Optionally, a discard or zeroout
2426  * can be issued on the journal blocks after flushing.
2427  *
2428  * flags:
2429  *      JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2430  *      JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2431  */
2432 int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2433 {
2434         int err = 0;
2435         transaction_t *transaction = NULL;
2436
2437         write_lock(&journal->j_state_lock);
2438
2439         /* Force everything buffered to the log... */
2440         if (journal->j_running_transaction) {
2441                 transaction = journal->j_running_transaction;
2442                 __jbd2_log_start_commit(journal, transaction->t_tid);
2443         } else if (journal->j_committing_transaction)
2444                 transaction = journal->j_committing_transaction;
2445
2446         /* Wait for the log commit to complete... */
2447         if (transaction) {
2448                 tid_t tid = transaction->t_tid;
2449
2450                 write_unlock(&journal->j_state_lock);
2451                 jbd2_log_wait_commit(journal, tid);
2452         } else {
2453                 write_unlock(&journal->j_state_lock);
2454         }
2455
2456         /* ...and flush everything in the log out to disk. */
2457         spin_lock(&journal->j_list_lock);
2458         while (!err && journal->j_checkpoint_transactions != NULL) {
2459                 spin_unlock(&journal->j_list_lock);
2460                 mutex_lock_io(&journal->j_checkpoint_mutex);
2461                 err = jbd2_log_do_checkpoint(journal);
2462                 mutex_unlock(&journal->j_checkpoint_mutex);
2463                 spin_lock(&journal->j_list_lock);
2464         }
2465         spin_unlock(&journal->j_list_lock);
2466
2467         if (is_journal_aborted(journal))
2468                 return -EIO;
2469
2470         mutex_lock_io(&journal->j_checkpoint_mutex);
2471         if (!err) {
2472                 err = jbd2_cleanup_journal_tail(journal);
2473                 if (err < 0) {
2474                         mutex_unlock(&journal->j_checkpoint_mutex);
2475                         goto out;
2476                 }
2477                 err = 0;
2478         }
2479
2480         /* Finally, mark the journal as really needing no recovery.
2481          * This sets s_start==0 in the underlying superblock, which is
2482          * the magic code for a fully-recovered superblock.  Any future
2483          * commits of data to the journal will restore the current
2484          * s_start value. */
2485         jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2486
2487         if (flags)
2488                 err = __jbd2_journal_erase(journal, flags);
2489
2490         mutex_unlock(&journal->j_checkpoint_mutex);
2491         write_lock(&journal->j_state_lock);
2492         J_ASSERT(!journal->j_running_transaction);
2493         J_ASSERT(!journal->j_committing_transaction);
2494         J_ASSERT(!journal->j_checkpoint_transactions);
2495         J_ASSERT(journal->j_head == journal->j_tail);
2496         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2497         write_unlock(&journal->j_state_lock);
2498 out:
2499         return err;
2500 }
2501
2502 /**
2503  * jbd2_journal_wipe() - Wipe journal contents
2504  * @journal: Journal to act on.
2505  * @write: flag (see below)
2506  *
2507  * Wipe out all of the contents of a journal, safely.  This will produce
2508  * a warning if the journal contains any valid recovery information.
2509  * Must be called between journal_init_*() and jbd2_journal_load().
2510  *
2511  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2512  * we merely suppress recovery.
2513  */
2514
2515 int jbd2_journal_wipe(journal_t *journal, int write)
2516 {
2517         int err = 0;
2518
2519         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2520
2521         err = load_superblock(journal);
2522         if (err)
2523                 return err;
2524
2525         if (!journal->j_tail)
2526                 goto no_recovery;
2527
2528         printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2529                 write ? "Clearing" : "Ignoring");
2530
2531         err = jbd2_journal_skip_recovery(journal);
2532         if (write) {
2533                 /* Lock to make assertions happy... */
2534                 mutex_lock_io(&journal->j_checkpoint_mutex);
2535                 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2536                 mutex_unlock(&journal->j_checkpoint_mutex);
2537         }
2538
2539  no_recovery:
2540         return err;
2541 }
2542
2543 /**
2544  * jbd2_journal_abort () - Shutdown the journal immediately.
2545  * @journal: the journal to shutdown.
2546  * @errno:   an error number to record in the journal indicating
2547  *           the reason for the shutdown.
2548  *
2549  * Perform a complete, immediate shutdown of the ENTIRE
2550  * journal (not of a single transaction).  This operation cannot be
2551  * undone without closing and reopening the journal.
2552  *
2553  * The jbd2_journal_abort function is intended to support higher level error
2554  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2555  * mode.
2556  *
2557  * Journal abort has very specific semantics.  Any existing dirty,
2558  * unjournaled buffers in the main filesystem will still be written to
2559  * disk by bdflush, but the journaling mechanism will be suspended
2560  * immediately and no further transaction commits will be honoured.
2561  *
2562  * Any dirty, journaled buffers will be written back to disk without
2563  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2564  * filesystem, but we _do_ attempt to leave as much data as possible
2565  * behind for fsck to use for cleanup.
2566  *
2567  * Any attempt to get a new transaction handle on a journal which is in
2568  * ABORT state will just result in an -EROFS error return.  A
2569  * jbd2_journal_stop on an existing handle will return -EIO if we have
2570  * entered abort state during the update.
2571  *
2572  * Recursive transactions are not disturbed by journal abort until the
2573  * final jbd2_journal_stop, which will receive the -EIO error.
2574  *
2575  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2576  * which will be recorded (if possible) in the journal superblock.  This
2577  * allows a client to record failure conditions in the middle of a
2578  * transaction without having to complete the transaction to record the
2579  * failure to disk.  ext3_error, for example, now uses this
2580  * functionality.
2581  *
2582  */
2583
2584 void jbd2_journal_abort(journal_t *journal, int errno)
2585 {
2586         transaction_t *transaction;
2587
2588         /*
2589          * Lock the aborting procedure until everything is done, this avoid
2590          * races between filesystem's error handling flow (e.g. ext4_abort()),
2591          * ensure panic after the error info is written into journal's
2592          * superblock.
2593          */
2594         mutex_lock(&journal->j_abort_mutex);
2595         /*
2596          * ESHUTDOWN always takes precedence because a file system check
2597          * caused by any other journal abort error is not required after
2598          * a shutdown triggered.
2599          */
2600         write_lock(&journal->j_state_lock);
2601         if (journal->j_flags & JBD2_ABORT) {
2602                 int old_errno = journal->j_errno;
2603
2604                 write_unlock(&journal->j_state_lock);
2605                 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2606                         journal->j_errno = errno;
2607                         jbd2_journal_update_sb_errno(journal);
2608                 }
2609                 mutex_unlock(&journal->j_abort_mutex);
2610                 return;
2611         }
2612
2613         /*
2614          * Mark the abort as occurred and start current running transaction
2615          * to release all journaled buffer.
2616          */
2617         pr_err("Aborting journal on device %s.\n", journal->j_devname);
2618
2619         journal->j_flags |= JBD2_ABORT;
2620         journal->j_errno = errno;
2621         transaction = journal->j_running_transaction;
2622         if (transaction)
2623                 __jbd2_log_start_commit(journal, transaction->t_tid);
2624         write_unlock(&journal->j_state_lock);
2625
2626         /*
2627          * Record errno to the journal super block, so that fsck and jbd2
2628          * layer could realise that a filesystem check is needed.
2629          */
2630         jbd2_journal_update_sb_errno(journal);
2631         mutex_unlock(&journal->j_abort_mutex);
2632 }
2633
2634 /**
2635  * jbd2_journal_errno() - returns the journal's error state.
2636  * @journal: journal to examine.
2637  *
2638  * This is the errno number set with jbd2_journal_abort(), the last
2639  * time the journal was mounted - if the journal was stopped
2640  * without calling abort this will be 0.
2641  *
2642  * If the journal has been aborted on this mount time -EROFS will
2643  * be returned.
2644  */
2645 int jbd2_journal_errno(journal_t *journal)
2646 {
2647         int err;
2648
2649         read_lock(&journal->j_state_lock);
2650         if (journal->j_flags & JBD2_ABORT)
2651                 err = -EROFS;
2652         else
2653                 err = journal->j_errno;
2654         read_unlock(&journal->j_state_lock);
2655         return err;
2656 }
2657
2658 /**
2659  * jbd2_journal_clear_err() - clears the journal's error state
2660  * @journal: journal to act on.
2661  *
2662  * An error must be cleared or acked to take a FS out of readonly
2663  * mode.
2664  */
2665 int jbd2_journal_clear_err(journal_t *journal)
2666 {
2667         int err = 0;
2668
2669         write_lock(&journal->j_state_lock);
2670         if (journal->j_flags & JBD2_ABORT)
2671                 err = -EROFS;
2672         else
2673                 journal->j_errno = 0;
2674         write_unlock(&journal->j_state_lock);
2675         return err;
2676 }
2677
2678 /**
2679  * jbd2_journal_ack_err() - Ack journal err.
2680  * @journal: journal to act on.
2681  *
2682  * An error must be cleared or acked to take a FS out of readonly
2683  * mode.
2684  */
2685 void jbd2_journal_ack_err(journal_t *journal)
2686 {
2687         write_lock(&journal->j_state_lock);
2688         if (journal->j_errno)
2689                 journal->j_flags |= JBD2_ACK_ERR;
2690         write_unlock(&journal->j_state_lock);
2691 }
2692
2693 int jbd2_journal_blocks_per_page(struct inode *inode)
2694 {
2695         return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2696 }
2697
2698 /*
2699  * helper functions to deal with 32 or 64bit block numbers.
2700  */
2701 size_t journal_tag_bytes(journal_t *journal)
2702 {
2703         size_t sz;
2704
2705         if (jbd2_has_feature_csum3(journal))
2706                 return sizeof(journal_block_tag3_t);
2707
2708         sz = sizeof(journal_block_tag_t);
2709
2710         if (jbd2_has_feature_csum2(journal))
2711                 sz += sizeof(__u16);
2712
2713         if (jbd2_has_feature_64bit(journal))
2714                 return sz;
2715         else
2716                 return sz - sizeof(__u32);
2717 }
2718
2719 /*
2720  * JBD memory management
2721  *
2722  * These functions are used to allocate block-sized chunks of memory
2723  * used for making copies of buffer_head data.  Very often it will be
2724  * page-sized chunks of data, but sometimes it will be in
2725  * sub-page-size chunks.  (For example, 16k pages on Power systems
2726  * with a 4k block file system.)  For blocks smaller than a page, we
2727  * use a SLAB allocator.  There are slab caches for each block size,
2728  * which are allocated at mount time, if necessary, and we only free
2729  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2730  * this reason we don't need to a mutex to protect access to
2731  * jbd2_slab[] allocating or releasing memory; only in
2732  * jbd2_journal_create_slab().
2733  */
2734 #define JBD2_MAX_SLABS 8
2735 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2736
2737 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2738         "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2739         "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2740 };
2741
2742
2743 static void jbd2_journal_destroy_slabs(void)
2744 {
2745         int i;
2746
2747         for (i = 0; i < JBD2_MAX_SLABS; i++) {
2748                 kmem_cache_destroy(jbd2_slab[i]);
2749                 jbd2_slab[i] = NULL;
2750         }
2751 }
2752
2753 static int jbd2_journal_create_slab(size_t size)
2754 {
2755         static DEFINE_MUTEX(jbd2_slab_create_mutex);
2756         int i = order_base_2(size) - 10;
2757         size_t slab_size;
2758
2759         if (size == PAGE_SIZE)
2760                 return 0;
2761
2762         if (i >= JBD2_MAX_SLABS)
2763                 return -EINVAL;
2764
2765         if (unlikely(i < 0))
2766                 i = 0;
2767         mutex_lock(&jbd2_slab_create_mutex);
2768         if (jbd2_slab[i]) {
2769                 mutex_unlock(&jbd2_slab_create_mutex);
2770                 return 0;       /* Already created */
2771         }
2772
2773         slab_size = 1 << (i+10);
2774         jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2775                                          slab_size, 0, NULL);
2776         mutex_unlock(&jbd2_slab_create_mutex);
2777         if (!jbd2_slab[i]) {
2778                 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2779                 return -ENOMEM;
2780         }
2781         return 0;
2782 }
2783
2784 static struct kmem_cache *get_slab(size_t size)
2785 {
2786         int i = order_base_2(size) - 10;
2787
2788         BUG_ON(i >= JBD2_MAX_SLABS);
2789         if (unlikely(i < 0))
2790                 i = 0;
2791         BUG_ON(jbd2_slab[i] == NULL);
2792         return jbd2_slab[i];
2793 }
2794
2795 void *jbd2_alloc(size_t size, gfp_t flags)
2796 {
2797         void *ptr;
2798
2799         BUG_ON(size & (size-1)); /* Must be a power of 2 */
2800
2801         if (size < PAGE_SIZE)
2802                 ptr = kmem_cache_alloc(get_slab(size), flags);
2803         else
2804                 ptr = (void *)__get_free_pages(flags, get_order(size));
2805
2806         /* Check alignment; SLUB has gotten this wrong in the past,
2807          * and this can lead to user data corruption! */
2808         BUG_ON(((unsigned long) ptr) & (size-1));
2809
2810         return ptr;
2811 }
2812
2813 void jbd2_free(void *ptr, size_t size)
2814 {
2815         if (size < PAGE_SIZE)
2816                 kmem_cache_free(get_slab(size), ptr);
2817         else
2818                 free_pages((unsigned long)ptr, get_order(size));
2819 };
2820
2821 /*
2822  * Journal_head storage management
2823  */
2824 static struct kmem_cache *jbd2_journal_head_cache;
2825 #ifdef CONFIG_JBD2_DEBUG
2826 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2827 #endif
2828
2829 static int __init jbd2_journal_init_journal_head_cache(void)
2830 {
2831         J_ASSERT(!jbd2_journal_head_cache);
2832         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2833                                 sizeof(struct journal_head),
2834                                 0,              /* offset */
2835                                 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2836                                 NULL);          /* ctor */
2837         if (!jbd2_journal_head_cache) {
2838                 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2839                 return -ENOMEM;
2840         }
2841         return 0;
2842 }
2843
2844 static void jbd2_journal_destroy_journal_head_cache(void)
2845 {
2846         kmem_cache_destroy(jbd2_journal_head_cache);
2847         jbd2_journal_head_cache = NULL;
2848 }
2849
2850 /*
2851  * journal_head splicing and dicing
2852  */
2853 static struct journal_head *journal_alloc_journal_head(void)
2854 {
2855         struct journal_head *ret;
2856
2857 #ifdef CONFIG_JBD2_DEBUG
2858         atomic_inc(&nr_journal_heads);
2859 #endif
2860         ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2861         if (!ret) {
2862                 jbd_debug(1, "out of memory for journal_head\n");
2863                 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2864                 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2865                                 GFP_NOFS | __GFP_NOFAIL);
2866         }
2867         if (ret)
2868                 spin_lock_init(&ret->b_state_lock);
2869         return ret;
2870 }
2871
2872 static void journal_free_journal_head(struct journal_head *jh)
2873 {
2874 #ifdef CONFIG_JBD2_DEBUG
2875         atomic_dec(&nr_journal_heads);
2876         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2877 #endif
2878         kmem_cache_free(jbd2_journal_head_cache, jh);
2879 }
2880
2881 /*
2882  * A journal_head is attached to a buffer_head whenever JBD has an
2883  * interest in the buffer.
2884  *
2885  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2886  * is set.  This bit is tested in core kernel code where we need to take
2887  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2888  * there.
2889  *
2890  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2891  *
2892  * When a buffer has its BH_JBD bit set it is immune from being released by
2893  * core kernel code, mainly via ->b_count.
2894  *
2895  * A journal_head is detached from its buffer_head when the journal_head's
2896  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2897  * transaction (b_cp_transaction) hold their references to b_jcount.
2898  *
2899  * Various places in the kernel want to attach a journal_head to a buffer_head
2900  * _before_ attaching the journal_head to a transaction.  To protect the
2901  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2902  * journal_head's b_jcount refcount by one.  The caller must call
2903  * jbd2_journal_put_journal_head() to undo this.
2904  *
2905  * So the typical usage would be:
2906  *
2907  *      (Attach a journal_head if needed.  Increments b_jcount)
2908  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2909  *      ...
2910  *      (Get another reference for transaction)
2911  *      jbd2_journal_grab_journal_head(bh);
2912  *      jh->b_transaction = xxx;
2913  *      (Put original reference)
2914  *      jbd2_journal_put_journal_head(jh);
2915  */
2916
2917 /*
2918  * Give a buffer_head a journal_head.
2919  *
2920  * May sleep.
2921  */
2922 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2923 {
2924         struct journal_head *jh;
2925         struct journal_head *new_jh = NULL;
2926
2927 repeat:
2928         if (!buffer_jbd(bh))
2929                 new_jh = journal_alloc_journal_head();
2930
2931         jbd_lock_bh_journal_head(bh);
2932         if (buffer_jbd(bh)) {
2933                 jh = bh2jh(bh);
2934         } else {
2935                 J_ASSERT_BH(bh,
2936                         (atomic_read(&bh->b_count) > 0) ||
2937                         (bh->b_page && bh->b_page->mapping));
2938
2939                 if (!new_jh) {
2940                         jbd_unlock_bh_journal_head(bh);
2941                         goto repeat;
2942                 }
2943
2944                 jh = new_jh;
2945                 new_jh = NULL;          /* We consumed it */
2946                 set_buffer_jbd(bh);
2947                 bh->b_private = jh;
2948                 jh->b_bh = bh;
2949                 get_bh(bh);
2950                 BUFFER_TRACE(bh, "added journal_head");
2951         }
2952         jh->b_jcount++;
2953         jbd_unlock_bh_journal_head(bh);
2954         if (new_jh)
2955                 journal_free_journal_head(new_jh);
2956         return bh->b_private;
2957 }
2958
2959 /*
2960  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2961  * having a journal_head, return NULL
2962  */
2963 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2964 {
2965         struct journal_head *jh = NULL;
2966
2967         jbd_lock_bh_journal_head(bh);
2968         if (buffer_jbd(bh)) {
2969                 jh = bh2jh(bh);
2970                 jh->b_jcount++;
2971         }
2972         jbd_unlock_bh_journal_head(bh);
2973         return jh;
2974 }
2975
2976 static void __journal_remove_journal_head(struct buffer_head *bh)
2977 {
2978         struct journal_head *jh = bh2jh(bh);
2979
2980         J_ASSERT_JH(jh, jh->b_transaction == NULL);
2981         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2982         J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2983         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2984         J_ASSERT_BH(bh, buffer_jbd(bh));
2985         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2986         BUFFER_TRACE(bh, "remove journal_head");
2987
2988         /* Unlink before dropping the lock */
2989         bh->b_private = NULL;
2990         jh->b_bh = NULL;        /* debug, really */
2991         clear_buffer_jbd(bh);
2992 }
2993
2994 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2995 {
2996         if (jh->b_frozen_data) {
2997                 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2998                 jbd2_free(jh->b_frozen_data, b_size);
2999         }
3000         if (jh->b_committed_data) {
3001                 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
3002                 jbd2_free(jh->b_committed_data, b_size);
3003         }
3004         journal_free_journal_head(jh);
3005 }
3006
3007 /*
3008  * Drop a reference on the passed journal_head.  If it fell to zero then
3009  * release the journal_head from the buffer_head.
3010  */
3011 void jbd2_journal_put_journal_head(struct journal_head *jh)
3012 {
3013         struct buffer_head *bh = jh2bh(jh);
3014
3015         jbd_lock_bh_journal_head(bh);
3016         J_ASSERT_JH(jh, jh->b_jcount > 0);
3017         --jh->b_jcount;
3018         if (!jh->b_jcount) {
3019                 __journal_remove_journal_head(bh);
3020                 jbd_unlock_bh_journal_head(bh);
3021                 journal_release_journal_head(jh, bh->b_size);
3022                 __brelse(bh);
3023         } else {
3024                 jbd_unlock_bh_journal_head(bh);
3025         }
3026 }
3027
3028 /*
3029  * Initialize jbd inode head
3030  */
3031 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3032 {
3033         jinode->i_transaction = NULL;
3034         jinode->i_next_transaction = NULL;
3035         jinode->i_vfs_inode = inode;
3036         jinode->i_flags = 0;
3037         jinode->i_dirty_start = 0;
3038         jinode->i_dirty_end = 0;
3039         INIT_LIST_HEAD(&jinode->i_list);
3040 }
3041
3042 /*
3043  * Function to be called before we start removing inode from memory (i.e.,
3044  * clear_inode() is a fine place to be called from). It removes inode from
3045  * transaction's lists.
3046  */
3047 void jbd2_journal_release_jbd_inode(journal_t *journal,
3048                                     struct jbd2_inode *jinode)
3049 {
3050         if (!journal)
3051                 return;
3052 restart:
3053         spin_lock(&journal->j_list_lock);
3054         /* Is commit writing out inode - we have to wait */
3055         if (jinode->i_flags & JI_COMMIT_RUNNING) {
3056                 wait_queue_head_t *wq;
3057                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3058                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3059                 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3060                 spin_unlock(&journal->j_list_lock);
3061                 schedule();
3062                 finish_wait(wq, &wait.wq_entry);
3063                 goto restart;
3064         }
3065
3066         if (jinode->i_transaction) {
3067                 list_del(&jinode->i_list);
3068                 jinode->i_transaction = NULL;
3069         }
3070         spin_unlock(&journal->j_list_lock);
3071 }
3072
3073
3074 #ifdef CONFIG_PROC_FS
3075
3076 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3077
3078 static void __init jbd2_create_jbd_stats_proc_entry(void)
3079 {
3080         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3081 }
3082
3083 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3084 {
3085         if (proc_jbd2_stats)
3086                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3087 }
3088
3089 #else
3090
3091 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3092 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3093
3094 #endif
3095
3096 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3097
3098 static int __init jbd2_journal_init_inode_cache(void)
3099 {
3100         J_ASSERT(!jbd2_inode_cache);
3101         jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3102         if (!jbd2_inode_cache) {
3103                 pr_emerg("JBD2: failed to create inode cache\n");
3104                 return -ENOMEM;
3105         }
3106         return 0;
3107 }
3108
3109 static int __init jbd2_journal_init_handle_cache(void)
3110 {
3111         J_ASSERT(!jbd2_handle_cache);
3112         jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3113         if (!jbd2_handle_cache) {
3114                 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3115                 return -ENOMEM;
3116         }
3117         return 0;
3118 }
3119
3120 static void jbd2_journal_destroy_inode_cache(void)
3121 {
3122         kmem_cache_destroy(jbd2_inode_cache);
3123         jbd2_inode_cache = NULL;
3124 }
3125
3126 static void jbd2_journal_destroy_handle_cache(void)
3127 {
3128         kmem_cache_destroy(jbd2_handle_cache);
3129         jbd2_handle_cache = NULL;
3130 }
3131
3132 /*
3133  * Module startup and shutdown
3134  */
3135
3136 static int __init journal_init_caches(void)
3137 {
3138         int ret;
3139
3140         ret = jbd2_journal_init_revoke_record_cache();
3141         if (ret == 0)
3142                 ret = jbd2_journal_init_revoke_table_cache();
3143         if (ret == 0)
3144                 ret = jbd2_journal_init_journal_head_cache();
3145         if (ret == 0)
3146                 ret = jbd2_journal_init_handle_cache();
3147         if (ret == 0)
3148                 ret = jbd2_journal_init_inode_cache();
3149         if (ret == 0)
3150                 ret = jbd2_journal_init_transaction_cache();
3151         return ret;
3152 }
3153
3154 static void jbd2_journal_destroy_caches(void)
3155 {
3156         jbd2_journal_destroy_revoke_record_cache();
3157         jbd2_journal_destroy_revoke_table_cache();
3158         jbd2_journal_destroy_journal_head_cache();
3159         jbd2_journal_destroy_handle_cache();
3160         jbd2_journal_destroy_inode_cache();
3161         jbd2_journal_destroy_transaction_cache();
3162         jbd2_journal_destroy_slabs();
3163 }
3164
3165 static int __init journal_init(void)
3166 {
3167         int ret;
3168
3169         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3170
3171         ret = journal_init_caches();
3172         if (ret == 0) {
3173                 jbd2_create_jbd_stats_proc_entry();
3174         } else {
3175                 jbd2_journal_destroy_caches();
3176         }
3177         return ret;
3178 }
3179
3180 static void __exit journal_exit(void)
3181 {
3182 #ifdef CONFIG_JBD2_DEBUG
3183         int n = atomic_read(&nr_journal_heads);
3184         if (n)
3185                 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3186 #endif
3187         jbd2_remove_jbd_stats_proc_entry();
3188         jbd2_journal_destroy_caches();
3189 }
3190
3191 MODULE_LICENSE("GPL");
3192 module_init(journal_init);
3193 module_exit(journal_exit);
3194