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 <sct@redhat.com>, 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
 761         return 0;
 762 }
 763 EXPORT_SYMBOL(jbd2_fc_begin_commit);
 764
 765 /*
 766  * Stop a fast commit. If fallback is set, this function starts commit of
 767  * TID tid before any other fast commit can start.
 768  */
 769 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
 770 {
 771         if (journal->j_fc_cleanup_callback)
 772                 journal->j_fc_cleanup_callback(journal, 0);
 773         write_lock(&journal->j_state_lock);
 774         journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
 775         if (fallback)
 776                 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
 777         write_unlock(&journal->j_state_lock);
 778         wake_up(&journal->j_fc_wait);
 779         if (fallback)
 780                 return jbd2_complete_transaction(journal, tid);
 781         return 0;
 782 }
 783
 784 int jbd2_fc_end_commit(journal_t *journal)
 785 {
 786         return __jbd2_fc_end_commit(journal, 0, false);
 787 }
 788 EXPORT_SYMBOL(jbd2_fc_end_commit);
 789
 790 int jbd2_fc_end_commit_fallback(journal_t *journal)
 791 {
 792         tid_t tid;
 793
 794         read_lock(&journal->j_state_lock);
 795         tid = journal->j_running_transaction ?
 796                 journal->j_running_transaction->t_tid : 0;
 797         read_unlock(&journal->j_state_lock);
 798         return __jbd2_fc_end_commit(journal, tid, true);
 799 }
 800 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
 801
 802 /* Return 1 when transaction with given tid has already committed. */
 803 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
 804 {
 805         int ret = 1;
 806
 807         read_lock(&journal->j_state_lock);
 808         if (journal->j_running_transaction &&
 809             journal->j_running_transaction->t_tid == tid)
 810                 ret = 0;
 811         if (journal->j_committing_transaction &&
 812             journal->j_committing_transaction->t_tid == tid)
 813                 ret = 0;
 814         read_unlock(&journal->j_state_lock);
 815         return ret;
 816 }
 817 EXPORT_SYMBOL(jbd2_transaction_committed);
 818
 819 /*
 820  * When this function returns the transaction corresponding to tid
 821  * will be completed.  If the transaction has currently running, start
 822  * committing that transaction before waiting for it to complete.  If
 823  * the transaction id is stale, it is by definition already completed,
 824  * so just return SUCCESS.
 825  */
 826 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
 827 {
 828         int     need_to_wait = 1;
 829
 830         read_lock(&journal->j_state_lock);
 831         if (journal->j_running_transaction &&
 832             journal->j_running_transaction->t_tid == tid) {
 833                 if (journal->j_commit_request != tid) {
 834                         /* transaction not yet started, so request it */
 835                         read_unlock(&journal->j_state_lock);
 836                         jbd2_log_start_commit(journal, tid);
 837                         goto wait_commit;
 838                 }
 839         } else if (!(journal->j_committing_transaction &&
 840                      journal->j_committing_transaction->t_tid == tid))
 841                 need_to_wait = 0;
 842         read_unlock(&journal->j_state_lock);
 843         if (!need_to_wait)
 844                 return 0;
 845 wait_commit:
 846         return jbd2_log_wait_commit(journal, tid);
 847 }
 848 EXPORT_SYMBOL(jbd2_complete_transaction);
 849
 850 /*
 851  * Log buffer allocation routines:
 852  */
 853
 854 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
 855 {
 856         unsigned long blocknr;
 857
 858         write_lock(&journal->j_state_lock);
 859         J_ASSERT(journal->j_free > 1);
 860
 861         blocknr = journal->j_head;
 862         journal->j_head++;
 863         journal->j_free--;
 864         if (journal->j_head == journal->j_last)
 865                 journal->j_head = journal->j_first;
 866         write_unlock(&journal->j_state_lock);
 867         return jbd2_journal_bmap(journal, blocknr, retp);
 868 }
 869
 870 /* Map one fast commit buffer for use by the file system */
 871 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
 872 {
 873         unsigned long long pblock;
 874         unsigned long blocknr;
 875         int ret = 0;
 876         struct buffer_head *bh;
 877         int fc_off;
 878
 879         *bh_out = NULL;
 880
 881         if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
 882                 fc_off = journal->j_fc_off;
 883                 blocknr = journal->j_fc_first + fc_off;
 884                 journal->j_fc_off++;
 885         } else {
 886                 ret = -EINVAL;
 887         }
 888
 889         if (ret)
 890                 return ret;
 891
 892         ret = jbd2_journal_bmap(journal, blocknr, &pblock);
 893         if (ret)
 894                 return ret;
 895
 896         bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
 897         if (!bh)
 898                 return -ENOMEM;
 899
 900
 901         journal->j_fc_wbuf[fc_off] = bh;
 902
 903         *bh_out = bh;
 904
 905         return 0;
 906 }
 907 EXPORT_SYMBOL(jbd2_fc_get_buf);
 908
 909 /*
 910  * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
 911  * for completion.
 912  */
 913 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
 914 {
 915         struct buffer_head *bh;
 916         int i, j_fc_off;
 917
 918         j_fc_off = journal->j_fc_off;
 919
 920         /*
 921          * Wait in reverse order to minimize chances of us being woken up before
 922          * all IOs have completed
 923          */
 924         for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
 925                 bh = journal->j_fc_wbuf[i];
 926                 wait_on_buffer(bh);
 927                 put_bh(bh);
 928                 journal->j_fc_wbuf[i] = NULL;
 929                 if (unlikely(!buffer_uptodate(bh)))
 930                         return -EIO;
 931         }
 932
 933         return 0;
 934 }
 935 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
 936
 937 /*
 938  * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
 939  * for completion.
 940  */
 941 int jbd2_fc_release_bufs(journal_t *journal)
 942 {
 943         struct buffer_head *bh;
 944         int i, j_fc_off;
 945
 946         j_fc_off = journal->j_fc_off;
 947
 948         /*
 949          * Wait in reverse order to minimize chances of us being woken up before
 950          * all IOs have completed
 951          */
 952         for (i = j_fc_off - 1; i >= 0; i--) {
 953                 bh = journal->j_fc_wbuf[i];
 954                 if (!bh)
 955                         break;
 956                 put_bh(bh);
 957                 journal->j_fc_wbuf[i] = NULL;
 958         }
 959
 960         return 0;
 961 }
 962 EXPORT_SYMBOL(jbd2_fc_release_bufs);
 963
 964 /*
 965  * Conversion of logical to physical block numbers for the journal
 966  *
 967  * On external journals the journal blocks are identity-mapped, so
 968  * this is a no-op.  If needed, we can use j_blk_offset - everything is
 969  * ready.
 970  */
 971 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
 972                  unsigned long long *retp)
 973 {
 974         int err = 0;
 975         unsigned long long ret;
 976         sector_t block = 0;
 977
 978         if (journal->j_inode) {
 979                 block = blocknr;
 980                 ret = bmap(journal->j_inode, &block);
 981
 982                 if (ret || !block) {
 983                         printk(KERN_ALERT "%s: journal block not found "
 984                                         "at offset %lu on %s\n",
 985                                __func__, blocknr, journal->j_devname);
 986                         err = -EIO;
 987                         jbd2_journal_abort(journal, err);
 988                 } else {
 989                         *retp = block;
 990                 }
 991
 992         } else {
 993                 *retp = blocknr; /* +journal->j_blk_offset */
 994         }
 995         return err;
 996 }
 997
 998 /*
 999  * We play buffer_head aliasing tricks to write data/metadata blocks to
1000  * the journal without copying their contents, but for journal
1001  * descriptor blocks we do need to generate bona fide buffers.
1002  *
1003  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
1004  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
1005  * But we don't bother doing that, so there will be coherency problems with
1006  * mmaps of blockdevs which hold live JBD-controlled filesystems.
1007  */
1008 struct buffer_head *
1009 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1010 {
1011         journal_t *journal = transaction->t_journal;
1012         struct buffer_head *bh;
1013         unsigned long long blocknr;
1014         journal_header_t *header;
1015         int err;
1016
1017         err = jbd2_journal_next_log_block(journal, &blocknr);
1018
1019         if (err)
1020                 return NULL;
1021
1022         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1023         if (!bh)
1024                 return NULL;
1025         atomic_dec(&transaction->t_outstanding_credits);
1026         lock_buffer(bh);
1027         memset(bh->b_data, 0, journal->j_blocksize);
1028         header = (journal_header_t *)bh->b_data;
1029         header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1030         header->h_blocktype = cpu_to_be32(type);
1031         header->h_sequence = cpu_to_be32(transaction->t_tid);
1032         set_buffer_uptodate(bh);
1033         unlock_buffer(bh);
1034         BUFFER_TRACE(bh, "return this buffer");
1035         return bh;
1036 }
1037
1038 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1039 {
1040         struct jbd2_journal_block_tail *tail;
1041         __u32 csum;
1042
1043         if (!jbd2_journal_has_csum_v2or3(j))
1044                 return;
1045
1046         tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1047                         sizeof(struct jbd2_journal_block_tail));
1048         tail->t_checksum = 0;
1049         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1050         tail->t_checksum = cpu_to_be32(csum);
1051 }
1052
1053 /*
1054  * Return tid of the oldest transaction in the journal and block in the journal
1055  * where the transaction starts.
1056  *
1057  * If the journal is now empty, return which will be the next transaction ID
1058  * we will write and where will that transaction start.
1059  *
1060  * The return value is 0 if journal tail cannot be pushed any further, 1 if
1061  * it can.
1062  */
1063 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1064                               unsigned long *block)
1065 {
1066         transaction_t *transaction;
1067         int ret;
1068
1069         read_lock(&journal->j_state_lock);
1070         spin_lock(&journal->j_list_lock);
1071         transaction = journal->j_checkpoint_transactions;
1072         if (transaction) {
1073                 *tid = transaction->t_tid;
1074                 *block = transaction->t_log_start;
1075         } else if ((transaction = journal->j_committing_transaction) != NULL) {
1076                 *tid = transaction->t_tid;
1077                 *block = transaction->t_log_start;
1078         } else if ((transaction = journal->j_running_transaction) != NULL) {
1079                 *tid = transaction->t_tid;
1080                 *block = journal->j_head;
1081         } else {
1082                 *tid = journal->j_transaction_sequence;
1083                 *block = journal->j_head;
1084         }
1085         ret = tid_gt(*tid, journal->j_tail_sequence);
1086         spin_unlock(&journal->j_list_lock);
1087         read_unlock(&journal->j_state_lock);
1088
1089         return ret;
1090 }
1091
1092 /*
1093  * Update information in journal structure and in on disk journal superblock
1094  * about log tail. This function does not check whether information passed in
1095  * really pushes log tail further. It's responsibility of the caller to make
1096  * sure provided log tail information is valid (e.g. by holding
1097  * j_checkpoint_mutex all the time between computing log tail and calling this
1098  * function as is the case with jbd2_cleanup_journal_tail()).
1099  *
1100  * Requires j_checkpoint_mutex
1101  */
1102 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1103 {
1104         unsigned long freed;
1105         int ret;
1106
1107         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1108
1109         /*
1110          * We cannot afford for write to remain in drive's caches since as
1111          * soon as we update j_tail, next transaction can start reusing journal
1112          * space and if we lose sb update during power failure we'd replay
1113          * old transaction with possibly newly overwritten data.
1114          */
1115         ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
1116                                               REQ_SYNC | REQ_FUA);
1117         if (ret)
1118                 goto out;
1119
1120         write_lock(&journal->j_state_lock);
1121         freed = block - journal->j_tail;
1122         if (block < journal->j_tail)
1123                 freed += journal->j_last - journal->j_first;
1124
1125         trace_jbd2_update_log_tail(journal, tid, block, freed);
1126         jbd_debug(1,
1127                   "Cleaning journal tail from %u to %u (offset %lu), "
1128                   "freeing %lu\n",
1129                   journal->j_tail_sequence, tid, block, freed);
1130
1131         journal->j_free += freed;
1132         journal->j_tail_sequence = tid;
1133         journal->j_tail = block;
1134         write_unlock(&journal->j_state_lock);
1135
1136 out:
1137         return ret;
1138 }
1139
1140 /*
1141  * This is a variation of __jbd2_update_log_tail which checks for validity of
1142  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1143  * with other threads updating log tail.
1144  */
1145 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1146 {
1147         mutex_lock_io(&journal->j_checkpoint_mutex);
1148         if (tid_gt(tid, journal->j_tail_sequence))
1149                 __jbd2_update_log_tail(journal, tid, block);
1150         mutex_unlock(&journal->j_checkpoint_mutex);
1151 }
1152
1153 struct jbd2_stats_proc_session {
1154         journal_t *journal;
1155         struct transaction_stats_s *stats;
1156         int start;
1157         int max;
1158 };
1159
1160 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1161 {
1162         return *pos ? NULL : SEQ_START_TOKEN;
1163 }
1164
1165 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1166 {
1167         (*pos)++;
1168         return NULL;
1169 }
1170
1171 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1172 {
1173         struct jbd2_stats_proc_session *s = seq->private;
1174
1175         if (v != SEQ_START_TOKEN)
1176                 return 0;
1177         seq_printf(seq, "%lu transactions (%lu requested), "
1178                    "each up to %u blocks\n",
1179                    s->stats->ts_tid, s->stats->ts_requested,
1180                    s->journal->j_max_transaction_buffers);
1181         if (s->stats->ts_tid == 0)
1182                 return 0;
1183         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
1184             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1185         seq_printf(seq, "  %ums request delay\n",
1186             (s->stats->ts_requested == 0) ? 0 :
1187             jiffies_to_msecs(s->stats->run.rs_request_delay /
1188                              s->stats->ts_requested));
1189         seq_printf(seq, "  %ums running transaction\n",
1190             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1191         seq_printf(seq, "  %ums transaction was being locked\n",
1192             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1193         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
1194             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1195         seq_printf(seq, "  %ums logging transaction\n",
1196             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1197         seq_printf(seq, "  %lluus average transaction commit time\n",
1198                    div_u64(s->journal->j_average_commit_time, 1000));
1199         seq_printf(seq, "  %lu handles per transaction\n",
1200             s->stats->run.rs_handle_count / s->stats->ts_tid);
1201         seq_printf(seq, "  %lu blocks per transaction\n",
1202             s->stats->run.rs_blocks / s->stats->ts_tid);
1203         seq_printf(seq, "  %lu logged blocks per transaction\n",
1204             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1205         return 0;
1206 }
1207
1208 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1209 {
1210 }
1211
1212 static const struct seq_operations jbd2_seq_info_ops = {
1213         .start  = jbd2_seq_info_start,
1214         .next   = jbd2_seq_info_next,
1215         .stop   = jbd2_seq_info_stop,
1216         .show   = jbd2_seq_info_show,
1217 };
1218
1219 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1220 {
1221         journal_t *journal = PDE_DATA(inode);
1222         struct jbd2_stats_proc_session *s;
1223         int rc, size;
1224
1225         s = kmalloc(sizeof(*s), GFP_KERNEL);
1226         if (s == NULL)
1227                 return -ENOMEM;
1228         size = sizeof(struct transaction_stats_s);
1229         s->stats = kmalloc(size, GFP_KERNEL);
1230         if (s->stats == NULL) {
1231                 kfree(s);
1232                 return -ENOMEM;
1233         }
1234         spin_lock(&journal->j_history_lock);
1235         memcpy(s->stats, &journal->j_stats, size);
1236         s->journal = journal;
1237         spin_unlock(&journal->j_history_lock);
1238
1239         rc = seq_open(file, &jbd2_seq_info_ops);
1240         if (rc == 0) {
1241                 struct seq_file *m = file->private_data;
1242                 m->private = s;
1243         } else {
1244                 kfree(s->stats);
1245                 kfree(s);
1246         }
1247         return rc;
1248
1249 }
1250
1251 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1252 {
1253         struct seq_file *seq = file->private_data;
1254         struct jbd2_stats_proc_session *s = seq->private;
1255         kfree(s->stats);
1256         kfree(s);
1257         return seq_release(inode, file);
1258 }
1259
1260 static const struct proc_ops jbd2_info_proc_ops = {
1261         .proc_open      = jbd2_seq_info_open,
1262         .proc_read      = seq_read,
1263         .proc_lseek     = seq_lseek,
1264         .proc_release   = jbd2_seq_info_release,
1265 };
1266
1267 static struct proc_dir_entry *proc_jbd2_stats;
1268
1269 static void jbd2_stats_proc_init(journal_t *journal)
1270 {
1271         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1272         if (journal->j_proc_entry) {
1273                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1274                                  &jbd2_info_proc_ops, journal);
1275         }
1276 }
1277
1278 static void jbd2_stats_proc_exit(journal_t *journal)
1279 {
1280         remove_proc_entry("info", journal->j_proc_entry);
1281         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1282 }
1283
1284 /* Minimum size of descriptor tag */
1285 static int jbd2_min_tag_size(void)
1286 {
1287         /*
1288          * Tag with 32-bit block numbers does not use last four bytes of the
1289          * structure
1290          */
1291         return sizeof(journal_block_tag_t) - 4;
1292 }
1293
1294 /*
1295  * Management for journal control blocks: functions to create and
1296  * destroy journal_t structures, and to initialise and read existing
1297  * journal blocks from disk.  */
1298
1299 /* First: create and setup a journal_t object in memory.  We initialise
1300  * very few fields yet: that has to wait until we have created the
1301  * journal structures from from scratch, or loaded them from disk. */
1302
1303 static journal_t *journal_init_common(struct block_device *bdev,
1304                         struct block_device *fs_dev,
1305                         unsigned long long start, int len, int blocksize)
1306 {
1307         static struct lock_class_key jbd2_trans_commit_key;
1308         journal_t *journal;
1309         int err;
1310         struct buffer_head *bh;
1311         int n;
1312
1313         journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1314         if (!journal)
1315                 return NULL;
1316
1317         init_waitqueue_head(&journal->j_wait_transaction_locked);
1318         init_waitqueue_head(&journal->j_wait_done_commit);
1319         init_waitqueue_head(&journal->j_wait_commit);
1320         init_waitqueue_head(&journal->j_wait_updates);
1321         init_waitqueue_head(&journal->j_wait_reserved);
1322         init_waitqueue_head(&journal->j_fc_wait);
1323         mutex_init(&journal->j_abort_mutex);
1324         mutex_init(&journal->j_barrier);
1325         mutex_init(&journal->j_checkpoint_mutex);
1326         spin_lock_init(&journal->j_revoke_lock);
1327         spin_lock_init(&journal->j_list_lock);
1328         rwlock_init(&journal->j_state_lock);
1329
1330         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1331         journal->j_min_batch_time = 0;
1332         journal->j_max_batch_time = 15000; /* 15ms */
1333         atomic_set(&journal->j_reserved_credits, 0);
1334
1335         /* The journal is marked for error until we succeed with recovery! */
1336         journal->j_flags = JBD2_ABORT;
1337
1338         /* Set up a default-sized revoke table for the new mount. */
1339         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1340         if (err)
1341                 goto err_cleanup;
1342
1343         spin_lock_init(&journal->j_history_lock);
1344
1345         lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1346                          &jbd2_trans_commit_key, 0);
1347
1348         /* journal descriptor can store up to n blocks -bzzz */
1349         journal->j_blocksize = blocksize;
1350         journal->j_dev = bdev;
1351         journal->j_fs_dev = fs_dev;
1352         journal->j_blk_offset = start;
1353         journal->j_total_len = len;
1354         /* We need enough buffers to write out full descriptor block. */
1355         n = journal->j_blocksize / jbd2_min_tag_size();
1356         journal->j_wbufsize = n;
1357         journal->j_fc_wbuf = NULL;
1358         journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1359                                         GFP_KERNEL);
1360         if (!journal->j_wbuf)
1361                 goto err_cleanup;
1362
1363         bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1364         if (!bh) {
1365                 pr_err("%s: Cannot get buffer for journal superblock\n",
1366                         __func__);
1367                 goto err_cleanup;
1368         }
1369         journal->j_sb_buffer = bh;
1370         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1371
1372         return journal;
1373
1374 err_cleanup:
1375         kfree(journal->j_wbuf);
1376         jbd2_journal_destroy_revoke(journal);
1377         kfree(journal);
1378         return NULL;
1379 }
1380
1381 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1382  *
1383  * Create a journal structure assigned some fixed set of disk blocks to
1384  * the journal.  We don't actually touch those disk blocks yet, but we
1385  * need to set up all of the mapping information to tell the journaling
1386  * system where the journal blocks are.
1387  *
1388  */
1389
1390 /**
1391  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1392  *  @bdev: Block device on which to create the journal
1393  *  @fs_dev: Device which hold journalled filesystem for this journal.
1394  *  @start: Block nr Start of journal.
1395  *  @len:  Length of the journal in blocks.
1396  *  @blocksize: blocksize of journalling device
1397  *
1398  *  Returns: a newly created journal_t *
1399  *
1400  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1401  *  range of blocks on an arbitrary block device.
1402  *
1403  */
1404 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1405                         struct block_device *fs_dev,
1406                         unsigned long long start, int len, int blocksize)
1407 {
1408         journal_t *journal;
1409
1410         journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1411         if (!journal)
1412                 return NULL;
1413
1414         bdevname(journal->j_dev, journal->j_devname);
1415         strreplace(journal->j_devname, '/', '!');
1416         jbd2_stats_proc_init(journal);
1417
1418         return journal;
1419 }
1420
1421 /**
1422  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1423  *  @inode: An inode to create the journal in
1424  *
1425  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1426  * the journal.  The inode must exist already, must support bmap() and
1427  * must have all data blocks preallocated.
1428  */
1429 journal_t *jbd2_journal_init_inode(struct inode *inode)
1430 {
1431         journal_t *journal;
1432         sector_t blocknr;
1433         char *p;
1434         int err = 0;
1435
1436         blocknr = 0;
1437         err = bmap(inode, &blocknr);
1438
1439         if (err || !blocknr) {
1440                 pr_err("%s: Cannot locate journal superblock\n",
1441                         __func__);
1442                 return NULL;
1443         }
1444
1445         jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1446                   inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1447                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1448
1449         journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1450                         blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1451                         inode->i_sb->s_blocksize);
1452         if (!journal)
1453                 return NULL;
1454
1455         journal->j_inode = inode;
1456         bdevname(journal->j_dev, journal->j_devname);
1457         p = strreplace(journal->j_devname, '/', '!');
1458         sprintf(p, "-%lu", journal->j_inode->i_ino);
1459         jbd2_stats_proc_init(journal);
1460
1461         return journal;
1462 }
1463
1464 /*
1465  * If the journal init or create aborts, we need to mark the journal
1466  * superblock as being NULL to prevent the journal destroy from writing
1467  * back a bogus superblock.
1468  */
1469 static void journal_fail_superblock(journal_t *journal)
1470 {
1471         struct buffer_head *bh = journal->j_sb_buffer;
1472         brelse(bh);
1473         journal->j_sb_buffer = NULL;
1474 }
1475
1476 /*
1477  * Given a journal_t structure, initialise the various fields for
1478  * startup of a new journaling session.  We use this both when creating
1479  * a journal, and after recovering an old journal to reset it for
1480  * subsequent use.
1481  */
1482
1483 static int journal_reset(journal_t *journal)
1484 {
1485         journal_superblock_t *sb = journal->j_superblock;
1486         unsigned long long first, last;
1487
1488         first = be32_to_cpu(sb->s_first);
1489         last = be32_to_cpu(sb->s_maxlen);
1490         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1491                 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1492                        first, last);
1493                 journal_fail_superblock(journal);
1494                 return -EINVAL;
1495         }
1496
1497         journal->j_first = first;
1498         journal->j_last = last;
1499
1500         journal->j_head = journal->j_first;
1501         journal->j_tail = journal->j_first;
1502         journal->j_free = journal->j_last - journal->j_first;
1503
1504         journal->j_tail_sequence = journal->j_transaction_sequence;
1505         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1506         journal->j_commit_request = journal->j_commit_sequence;
1507
1508         journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1509
1510         /*
1511          * Now that journal recovery is done, turn fast commits off here. This
1512          * way, if fast commit was enabled before the crash but if now FS has
1513          * disabled it, we don't enable fast commits.
1514          */
1515         jbd2_clear_feature_fast_commit(journal);
1516
1517         /*
1518          * As a special case, if the on-disk copy is already marked as needing
1519          * no recovery (s_start == 0), then we can safely defer the superblock
1520          * update until the next commit by setting JBD2_FLUSHED.  This avoids
1521          * attempting a write to a potential-readonly device.
1522          */
1523         if (sb->s_start == 0) {
1524                 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1525                         "(start %ld, seq %u, errno %d)\n",
1526                         journal->j_tail, journal->j_tail_sequence,
1527                         journal->j_errno);
1528                 journal->j_flags |= JBD2_FLUSHED;
1529         } else {
1530                 /* Lock here to make assertions happy... */
1531                 mutex_lock_io(&journal->j_checkpoint_mutex);
1532                 /*
1533                  * Update log tail information. We use REQ_FUA since new
1534                  * transaction will start reusing journal space and so we
1535                  * must make sure information about current log tail is on
1536                  * disk before that.
1537                  */
1538                 jbd2_journal_update_sb_log_tail(journal,
1539                                                 journal->j_tail_sequence,
1540                                                 journal->j_tail,
1541                                                 REQ_SYNC | REQ_FUA);
1542                 mutex_unlock(&journal->j_checkpoint_mutex);
1543         }
1544         return jbd2_journal_start_thread(journal);
1545 }
1546
1547 /*
1548  * This function expects that the caller will have locked the journal
1549  * buffer head, and will return with it unlocked
1550  */
1551 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1552 {
1553         struct buffer_head *bh = journal->j_sb_buffer;
1554         journal_superblock_t *sb = journal->j_superblock;
1555         int ret;
1556
1557         /* Buffer got discarded which means block device got invalidated */
1558         if (!buffer_mapped(bh)) {
1559                 unlock_buffer(bh);
1560                 return -EIO;
1561         }
1562
1563         trace_jbd2_write_superblock(journal, write_flags);
1564         if (!(journal->j_flags & JBD2_BARRIER))
1565                 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1566         if (buffer_write_io_error(bh)) {
1567                 /*
1568                  * Oh, dear.  A previous attempt to write the journal
1569                  * superblock failed.  This could happen because the
1570                  * USB device was yanked out.  Or it could happen to
1571                  * be a transient write error and maybe the block will
1572                  * be remapped.  Nothing we can do but to retry the
1573                  * write and hope for the best.
1574                  */
1575                 printk(KERN_ERR "JBD2: previous I/O error detected "
1576                        "for journal superblock update for %s.\n",
1577                        journal->j_devname);
1578                 clear_buffer_write_io_error(bh);
1579                 set_buffer_uptodate(bh);
1580         }
1581         if (jbd2_journal_has_csum_v2or3(journal))
1582                 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1583         get_bh(bh);
1584         bh->b_end_io = end_buffer_write_sync;
1585         ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1586         wait_on_buffer(bh);
1587         if (buffer_write_io_error(bh)) {
1588                 clear_buffer_write_io_error(bh);
1589                 set_buffer_uptodate(bh);
1590                 ret = -EIO;
1591         }
1592         if (ret) {
1593                 printk(KERN_ERR "JBD2: Error %d detected when updating "
1594                        "journal superblock for %s.\n", ret,
1595                        journal->j_devname);
1596                 if (!is_journal_aborted(journal))
1597                         jbd2_journal_abort(journal, ret);
1598         }
1599
1600         return ret;
1601 }
1602
1603 /**
1604  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1605  * @journal: The journal to update.
1606  * @tail_tid: TID of the new transaction at the tail of the log
1607  * @tail_block: The first block of the transaction at the tail of the log
1608  * @write_op: With which operation should we write the journal sb
1609  *
1610  * Update a journal's superblock information about log tail and write it to
1611  * disk, waiting for the IO to complete.
1612  */
1613 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1614                                      unsigned long tail_block, int write_op)
1615 {
1616         journal_superblock_t *sb = journal->j_superblock;
1617         int ret;
1618
1619         if (is_journal_aborted(journal))
1620                 return -EIO;
1621
1622         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1623         jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1624                   tail_block, tail_tid);
1625
1626         lock_buffer(journal->j_sb_buffer);
1627         sb->s_sequence = cpu_to_be32(tail_tid);
1628         sb->s_start    = cpu_to_be32(tail_block);
1629
1630         ret = jbd2_write_superblock(journal, write_op);
1631         if (ret)
1632                 goto out;
1633
1634         /* Log is no longer empty */
1635         write_lock(&journal->j_state_lock);
1636         WARN_ON(!sb->s_sequence);
1637         journal->j_flags &= ~JBD2_FLUSHED;
1638         write_unlock(&journal->j_state_lock);
1639
1640 out:
1641         return ret;
1642 }
1643
1644 /**
1645  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1646  * @journal: The journal to update.
1647  * @write_op: With which operation should we write the journal sb
1648  *
1649  * Update a journal's dynamic superblock fields to show that journal is empty.
1650  * Write updated superblock to disk waiting for IO to complete.
1651  */
1652 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1653 {
1654         journal_superblock_t *sb = journal->j_superblock;
1655         bool had_fast_commit = false;
1656
1657         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1658         lock_buffer(journal->j_sb_buffer);
1659         if (sb->s_start == 0) {         /* Is it already empty? */
1660                 unlock_buffer(journal->j_sb_buffer);
1661                 return;
1662         }
1663
1664         jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1665                   journal->j_tail_sequence);
1666
1667         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1668         sb->s_start    = cpu_to_be32(0);
1669         if (jbd2_has_feature_fast_commit(journal)) {
1670                 /*
1671                  * When journal is clean, no need to commit fast commit flag and
1672                  * make file system incompatible with older kernels.
1673                  */
1674                 jbd2_clear_feature_fast_commit(journal);
1675                 had_fast_commit = true;
1676         }
1677
1678         jbd2_write_superblock(journal, write_op);
1679
1680         if (had_fast_commit)
1681                 jbd2_set_feature_fast_commit(journal);
1682
1683         /* Log is no longer empty */
1684         write_lock(&journal->j_state_lock);
1685         journal->j_flags |= JBD2_FLUSHED;
1686         write_unlock(&journal->j_state_lock);
1687 }
1688
1689
1690 /**
1691  * jbd2_journal_update_sb_errno() - Update error in the journal.
1692  * @journal: The journal to update.
1693  *
1694  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1695  * to complete.
1696  */
1697 void jbd2_journal_update_sb_errno(journal_t *journal)
1698 {
1699         journal_superblock_t *sb = journal->j_superblock;
1700         int errcode;
1701
1702         lock_buffer(journal->j_sb_buffer);
1703         errcode = journal->j_errno;
1704         if (errcode == -ESHUTDOWN)
1705                 errcode = 0;
1706         jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1707         sb->s_errno    = cpu_to_be32(errcode);
1708
1709         jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1710 }
1711 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1712
1713 static int journal_revoke_records_per_block(journal_t *journal)
1714 {
1715         int record_size;
1716         int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1717
1718         if (jbd2_has_feature_64bit(journal))
1719                 record_size = 8;
1720         else
1721                 record_size = 4;
1722
1723         if (jbd2_journal_has_csum_v2or3(journal))
1724                 space -= sizeof(struct jbd2_journal_block_tail);
1725         return space / record_size;
1726 }
1727
1728 /*
1729  * Read the superblock for a given journal, performing initial
1730  * validation of the format.
1731  */
1732 static int journal_get_superblock(journal_t *journal)
1733 {
1734         struct buffer_head *bh;
1735         journal_superblock_t *sb;
1736         int err = -EIO;
1737
1738         bh = journal->j_sb_buffer;
1739
1740         J_ASSERT(bh != NULL);
1741         if (!buffer_uptodate(bh)) {
1742                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1743                 wait_on_buffer(bh);
1744                 if (!buffer_uptodate(bh)) {
1745                         printk(KERN_ERR
1746                                 "JBD2: IO error reading journal superblock\n");
1747                         goto out;
1748                 }
1749         }
1750
1751         if (buffer_verified(bh))
1752                 return 0;
1753
1754         sb = journal->j_superblock;
1755
1756         err = -EINVAL;
1757
1758         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1759             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1760                 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1761                 goto out;
1762         }
1763
1764         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1765         case JBD2_SUPERBLOCK_V1:
1766                 journal->j_format_version = 1;
1767                 break;
1768         case JBD2_SUPERBLOCK_V2:
1769                 journal->j_format_version = 2;
1770                 break;
1771         default:
1772                 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1773                 goto out;
1774         }
1775
1776         if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1777                 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1778         else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1779                 printk(KERN_WARNING "JBD2: journal file too short\n");
1780                 goto out;
1781         }
1782
1783         if (be32_to_cpu(sb->s_first) == 0 ||
1784             be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1785                 printk(KERN_WARNING
1786                         "JBD2: Invalid start block of journal: %u\n",
1787                         be32_to_cpu(sb->s_first));
1788                 goto out;
1789         }
1790
1791         if (jbd2_has_feature_csum2(journal) &&
1792             jbd2_has_feature_csum3(journal)) {
1793                 /* Can't have checksum v2 and v3 at the same time! */
1794                 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1795                        "at the same time!\n");
1796                 goto out;
1797         }
1798
1799         if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1800             jbd2_has_feature_checksum(journal)) {
1801                 /* Can't have checksum v1 and v2 on at the same time! */
1802                 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1803                        "at the same time!\n");
1804                 goto out;
1805         }
1806
1807         if (!jbd2_verify_csum_type(journal, sb)) {
1808                 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1809                 goto out;
1810         }
1811
1812         /* Load the checksum driver */
1813         if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1814                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1815                 if (IS_ERR(journal->j_chksum_driver)) {
1816                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1817                         err = PTR_ERR(journal->j_chksum_driver);
1818                         journal->j_chksum_driver = NULL;
1819                         goto out;
1820                 }
1821         }
1822
1823         if (jbd2_journal_has_csum_v2or3(journal)) {
1824                 /* Check superblock checksum */
1825                 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1826                         printk(KERN_ERR "JBD2: journal checksum error\n");
1827                         err = -EFSBADCRC;
1828                         goto out;
1829                 }
1830
1831                 /* Precompute checksum seed for all metadata */
1832                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1833                                                    sizeof(sb->s_uuid));
1834         }
1835
1836         journal->j_revoke_records_per_block =
1837                                 journal_revoke_records_per_block(journal);
1838         set_buffer_verified(bh);
1839
1840         return 0;
1841
1842 out:
1843         journal_fail_superblock(journal);
1844         return err;
1845 }
1846
1847 /*
1848  * Load the on-disk journal superblock and read the key fields into the
1849  * journal_t.
1850  */
1851
1852 static int load_superblock(journal_t *journal)
1853 {
1854         int err;
1855         journal_superblock_t *sb;
1856         int num_fc_blocks;
1857
1858         err = journal_get_superblock(journal);
1859         if (err)
1860                 return err;
1861
1862         sb = journal->j_superblock;
1863
1864         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1865         journal->j_tail = be32_to_cpu(sb->s_start);
1866         journal->j_first = be32_to_cpu(sb->s_first);
1867         journal->j_errno = be32_to_cpu(sb->s_errno);
1868         journal->j_last = be32_to_cpu(sb->s_maxlen);
1869
1870         if (jbd2_has_feature_fast_commit(journal)) {
1871                 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
1872                 num_fc_blocks = be32_to_cpu(sb->s_num_fc_blks);
1873                 if (!num_fc_blocks)
1874                         num_fc_blocks = JBD2_MIN_FC_BLOCKS;
1875                 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
1876                         journal->j_last = journal->j_fc_last - num_fc_blocks;
1877                 journal->j_fc_first = journal->j_last + 1;
1878                 journal->j_fc_off = 0;
1879         }
1880
1881         return 0;
1882 }
1883
1884
1885 /**
1886  * jbd2_journal_load() - Read journal from disk.
1887  * @journal: Journal to act on.
1888  *
1889  * Given a journal_t structure which tells us which disk blocks contain
1890  * a journal, read the journal from disk to initialise the in-memory
1891  * structures.
1892  */
1893 int jbd2_journal_load(journal_t *journal)
1894 {
1895         int err;
1896         journal_superblock_t *sb;
1897
1898         err = load_superblock(journal);
1899         if (err)
1900                 return err;
1901
1902         sb = journal->j_superblock;
1903         /* If this is a V2 superblock, then we have to check the
1904          * features flags on it. */
1905
1906         if (journal->j_format_version >= 2) {
1907                 if ((sb->s_feature_ro_compat &
1908                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1909                     (sb->s_feature_incompat &
1910                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1911                         printk(KERN_WARNING
1912                                 "JBD2: Unrecognised features on journal\n");
1913                         return -EINVAL;
1914                 }
1915         }
1916
1917         /*
1918          * Create a slab for this blocksize
1919          */
1920         err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1921         if (err)
1922                 return err;
1923
1924         /* Let the recovery code check whether it needs to recover any
1925          * data from the journal. */
1926         if (jbd2_journal_recover(journal))
1927                 goto recovery_error;
1928
1929         if (journal->j_failed_commit) {
1930                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1931                        "is corrupt.\n", journal->j_failed_commit,
1932                        journal->j_devname);
1933                 return -EFSCORRUPTED;
1934         }
1935         /*
1936          * clear JBD2_ABORT flag initialized in journal_init_common
1937          * here to update log tail information with the newest seq.
1938          */
1939         journal->j_flags &= ~JBD2_ABORT;
1940
1941         /* OK, we've finished with the dynamic journal bits:
1942          * reinitialise the dynamic contents of the superblock in memory
1943          * and reset them on disk. */
1944         if (journal_reset(journal))
1945                 goto recovery_error;
1946
1947         journal->j_flags |= JBD2_LOADED;
1948         return 0;
1949
1950 recovery_error:
1951         printk(KERN_WARNING "JBD2: recovery failed\n");
1952         return -EIO;
1953 }
1954
1955 /**
1956  * jbd2_journal_destroy() - Release a journal_t structure.
1957  * @journal: Journal to act on.
1958  *
1959  * Release a journal_t structure once it is no longer in use by the
1960  * journaled object.
1961  * Return <0 if we couldn't clean up the journal.
1962  */
1963 int jbd2_journal_destroy(journal_t *journal)
1964 {
1965         int err = 0;
1966
1967         /* Wait for the commit thread to wake up and die. */
1968         journal_kill_thread(journal);
1969
1970         /* Force a final log commit */
1971         if (journal->j_running_transaction)
1972                 jbd2_journal_commit_transaction(journal);
1973
1974         /* Force any old transactions to disk */
1975
1976         /* Totally anal locking here... */
1977         spin_lock(&journal->j_list_lock);
1978         while (journal->j_checkpoint_transactions != NULL) {
1979                 spin_unlock(&journal->j_list_lock);
1980                 mutex_lock_io(&journal->j_checkpoint_mutex);
1981                 err = jbd2_log_do_checkpoint(journal);
1982                 mutex_unlock(&journal->j_checkpoint_mutex);
1983                 /*
1984                  * If checkpointing failed, just free the buffers to avoid
1985                  * looping forever
1986                  */
1987                 if (err) {
1988                         jbd2_journal_destroy_checkpoint(journal);
1989                         spin_lock(&journal->j_list_lock);
1990                         break;
1991                 }
1992                 spin_lock(&journal->j_list_lock);
1993         }
1994
1995         J_ASSERT(journal->j_running_transaction == NULL);
1996         J_ASSERT(journal->j_committing_transaction == NULL);
1997         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1998         spin_unlock(&journal->j_list_lock);
1999
2000         if (journal->j_sb_buffer) {
2001                 if (!is_journal_aborted(journal)) {
2002                         mutex_lock_io(&journal->j_checkpoint_mutex);
2003
2004                         write_lock(&journal->j_state_lock);
2005                         journal->j_tail_sequence =
2006                                 ++journal->j_transaction_sequence;
2007                         write_unlock(&journal->j_state_lock);
2008
2009                         jbd2_mark_journal_empty(journal,
2010                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2011                         mutex_unlock(&journal->j_checkpoint_mutex);
2012                 } else
2013                         err = -EIO;
2014                 brelse(journal->j_sb_buffer);
2015         }
2016
2017         if (journal->j_proc_entry)
2018                 jbd2_stats_proc_exit(journal);
2019         iput(journal->j_inode);
2020         if (journal->j_revoke)
2021                 jbd2_journal_destroy_revoke(journal);
2022         if (journal->j_chksum_driver)
2023                 crypto_free_shash(journal->j_chksum_driver);
2024         kfree(journal->j_fc_wbuf);
2025         kfree(journal->j_wbuf);
2026         kfree(journal);
2027
2028         return err;
2029 }
2030
2031
2032 /**
2033  * jbd2_journal_check_used_features() - Check if features specified are used.
2034  * @journal: Journal to check.
2035  * @compat: bitmask of compatible features
2036  * @ro: bitmask of features that force read-only mount
2037  * @incompat: bitmask of incompatible features
2038  *
2039  * Check whether the journal uses all of a given set of
2040  * features.  Return true (non-zero) if it does.
2041  **/
2042
2043 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2044                                  unsigned long ro, unsigned long incompat)
2045 {
2046         journal_superblock_t *sb;
2047
2048         if (!compat && !ro && !incompat)
2049                 return 1;
2050         /* Load journal superblock if it is not loaded yet. */
2051         if (journal->j_format_version == 0 &&
2052             journal_get_superblock(journal) != 0)
2053                 return 0;
2054         if (journal->j_format_version == 1)
2055                 return 0;
2056
2057         sb = journal->j_superblock;
2058
2059         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2060             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2061             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2062                 return 1;
2063
2064         return 0;
2065 }
2066
2067 /**
2068  * jbd2_journal_check_available_features() - Check feature set in journalling layer
2069  * @journal: Journal to check.
2070  * @compat: bitmask of compatible features
2071  * @ro: bitmask of features that force read-only mount
2072  * @incompat: bitmask of incompatible features
2073  *
2074  * Check whether the journaling code supports the use of
2075  * all of a given set of features on this journal.  Return true
2076  * (non-zero) if it can. */
2077
2078 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2079                                       unsigned long ro, unsigned long incompat)
2080 {
2081         if (!compat && !ro && !incompat)
2082                 return 1;
2083
2084         /* We can support any known requested features iff the
2085          * superblock is in version 2.  Otherwise we fail to support any
2086          * extended sb features. */
2087
2088         if (journal->j_format_version != 2)
2089                 return 0;
2090
2091         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2092             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2093             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2094                 return 1;
2095
2096         return 0;
2097 }
2098
2099 static int
2100 jbd2_journal_initialize_fast_commit(journal_t *journal)
2101 {
2102         journal_superblock_t *sb = journal->j_superblock;
2103         unsigned long long num_fc_blks;
2104
2105         num_fc_blks = be32_to_cpu(sb->s_num_fc_blks);
2106         if (num_fc_blks == 0)
2107                 num_fc_blks = JBD2_MIN_FC_BLOCKS;
2108         if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2109                 return -ENOSPC;
2110
2111         /* Are we called twice? */
2112         WARN_ON(journal->j_fc_wbuf != NULL);
2113         journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2114                                 sizeof(struct buffer_head *), GFP_KERNEL);
2115         if (!journal->j_fc_wbuf)
2116                 return -ENOMEM;
2117
2118         journal->j_fc_wbufsize = num_fc_blks;
2119         journal->j_fc_last = journal->j_last;
2120         journal->j_last = journal->j_fc_last - num_fc_blks;
2121         journal->j_fc_first = journal->j_last + 1;
2122         journal->j_fc_off = 0;
2123         journal->j_free = journal->j_last - journal->j_first;
2124         journal->j_max_transaction_buffers =
2125                 jbd2_journal_get_max_txn_bufs(journal);
2126
2127         return 0;
2128 }
2129
2130 /**
2131  * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2132  * @journal: Journal to act on.
2133  * @compat: bitmask of compatible features
2134  * @ro: bitmask of features that force read-only mount
2135  * @incompat: bitmask of incompatible features
2136  *
2137  * Mark a given journal feature as present on the
2138  * superblock.  Returns true if the requested features could be set.
2139  *
2140  */
2141
2142 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2143                           unsigned long ro, unsigned long incompat)
2144 {
2145 #define INCOMPAT_FEATURE_ON(f) \
2146                 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2147 #define COMPAT_FEATURE_ON(f) \
2148                 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2149         journal_superblock_t *sb;
2150
2151         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2152                 return 1;
2153
2154         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2155                 return 0;
2156
2157         /* If enabling v2 checksums, turn on v3 instead */
2158         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2159                 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2160                 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2161         }
2162
2163         /* Asking for checksumming v3 and v1?  Only give them v3. */
2164         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2165             compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2166                 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2167
2168         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2169                   compat, ro, incompat);
2170
2171         sb = journal->j_superblock;
2172
2173         if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2174                 if (jbd2_journal_initialize_fast_commit(journal)) {
2175                         pr_err("JBD2: Cannot enable fast commits.\n");
2176                         return 0;
2177                 }
2178         }
2179
2180         /* Load the checksum driver if necessary */
2181         if ((journal->j_chksum_driver == NULL) &&
2182             INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2183                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2184                 if (IS_ERR(journal->j_chksum_driver)) {
2185                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2186                         journal->j_chksum_driver = NULL;
2187                         return 0;
2188                 }
2189                 /* Precompute checksum seed for all metadata */
2190                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2191                                                    sizeof(sb->s_uuid));
2192         }
2193
2194         lock_buffer(journal->j_sb_buffer);
2195
2196         /* If enabling v3 checksums, update superblock */
2197         if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2198                 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2199                 sb->s_feature_compat &=
2200                         ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2201         }
2202
2203         /* If enabling v1 checksums, downgrade superblock */
2204         if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2205                 sb->s_feature_incompat &=
2206                         ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2207                                      JBD2_FEATURE_INCOMPAT_CSUM_V3);
2208
2209         sb->s_feature_compat    |= cpu_to_be32(compat);
2210         sb->s_feature_ro_compat |= cpu_to_be32(ro);
2211         sb->s_feature_incompat  |= cpu_to_be32(incompat);
2212         unlock_buffer(journal->j_sb_buffer);
2213         journal->j_revoke_records_per_block =
2214                                 journal_revoke_records_per_block(journal);
2215
2216         return 1;
2217 #undef COMPAT_FEATURE_ON
2218 #undef INCOMPAT_FEATURE_ON
2219 }
2220
2221 /*
2222  * jbd2_journal_clear_features() - Clear a given journal feature in the
2223  *                                  superblock
2224  * @journal: Journal to act on.
2225  * @compat: bitmask of compatible features
2226  * @ro: bitmask of features that force read-only mount
2227  * @incompat: bitmask of incompatible features
2228  *
2229  * Clear a given journal feature as present on the
2230  * superblock.
2231  */
2232 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2233                                 unsigned long ro, unsigned long incompat)
2234 {
2235         journal_superblock_t *sb;
2236
2237         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2238                   compat, ro, incompat);
2239
2240         sb = journal->j_superblock;
2241
2242         sb->s_feature_compat    &= ~cpu_to_be32(compat);
2243         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2244         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
2245         journal->j_revoke_records_per_block =
2246                                 journal_revoke_records_per_block(journal);
2247 }
2248 EXPORT_SYMBOL(jbd2_journal_clear_features);
2249
2250 /**
2251  * jbd2_journal_flush() - Flush journal
2252  * @journal: Journal to act on.
2253  *
2254  * Flush all data for a given journal to disk and empty the journal.
2255  * Filesystems can use this when remounting readonly to ensure that
2256  * recovery does not need to happen on remount.
2257  */
2258
2259 int jbd2_journal_flush(journal_t *journal)
2260 {
2261         int err = 0;
2262         transaction_t *transaction = NULL;
2263
2264         write_lock(&journal->j_state_lock);
2265
2266         /* Force everything buffered to the log... */
2267         if (journal->j_running_transaction) {
2268                 transaction = journal->j_running_transaction;
2269                 __jbd2_log_start_commit(journal, transaction->t_tid);
2270         } else if (journal->j_committing_transaction)
2271                 transaction = journal->j_committing_transaction;
2272
2273         /* Wait for the log commit to complete... */
2274         if (transaction) {
2275                 tid_t tid = transaction->t_tid;
2276
2277                 write_unlock(&journal->j_state_lock);
2278                 jbd2_log_wait_commit(journal, tid);
2279         } else {
2280                 write_unlock(&journal->j_state_lock);
2281         }
2282
2283         /* ...and flush everything in the log out to disk. */
2284         spin_lock(&journal->j_list_lock);
2285         while (!err && journal->j_checkpoint_transactions != NULL) {
2286                 spin_unlock(&journal->j_list_lock);
2287                 mutex_lock_io(&journal->j_checkpoint_mutex);
2288                 err = jbd2_log_do_checkpoint(journal);
2289                 mutex_unlock(&journal->j_checkpoint_mutex);
2290                 spin_lock(&journal->j_list_lock);
2291         }
2292         spin_unlock(&journal->j_list_lock);
2293
2294         if (is_journal_aborted(journal))
2295                 return -EIO;
2296
2297         mutex_lock_io(&journal->j_checkpoint_mutex);
2298         if (!err) {
2299                 err = jbd2_cleanup_journal_tail(journal);
2300                 if (err < 0) {
2301                         mutex_unlock(&journal->j_checkpoint_mutex);
2302                         goto out;
2303                 }
2304                 err = 0;
2305         }
2306
2307         /* Finally, mark the journal as really needing no recovery.
2308          * This sets s_start==0 in the underlying superblock, which is
2309          * the magic code for a fully-recovered superblock.  Any future
2310          * commits of data to the journal will restore the current
2311          * s_start value. */
2312         jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2313         mutex_unlock(&journal->j_checkpoint_mutex);
2314         write_lock(&journal->j_state_lock);
2315         J_ASSERT(!journal->j_running_transaction);
2316         J_ASSERT(!journal->j_committing_transaction);
2317         J_ASSERT(!journal->j_checkpoint_transactions);
2318         J_ASSERT(journal->j_head == journal->j_tail);
2319         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2320         write_unlock(&journal->j_state_lock);
2321 out:
2322         return err;
2323 }
2324
2325 /**
2326  * jbd2_journal_wipe() - Wipe journal contents
2327  * @journal: Journal to act on.
2328  * @write: flag (see below)
2329  *
2330  * Wipe out all of the contents of a journal, safely.  This will produce
2331  * a warning if the journal contains any valid recovery information.
2332  * Must be called between journal_init_*() and jbd2_journal_load().
2333  *
2334  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2335  * we merely suppress recovery.
2336  */
2337
2338 int jbd2_journal_wipe(journal_t *journal, int write)
2339 {
2340         int err = 0;
2341
2342         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2343
2344         err = load_superblock(journal);
2345         if (err)
2346                 return err;
2347
2348         if (!journal->j_tail)
2349                 goto no_recovery;
2350
2351         printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2352                 write ? "Clearing" : "Ignoring");
2353
2354         err = jbd2_journal_skip_recovery(journal);
2355         if (write) {
2356                 /* Lock to make assertions happy... */
2357                 mutex_lock_io(&journal->j_checkpoint_mutex);
2358                 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2359                 mutex_unlock(&journal->j_checkpoint_mutex);
2360         }
2361
2362  no_recovery:
2363         return err;
2364 }
2365
2366 /**
2367  * jbd2_journal_abort () - Shutdown the journal immediately.
2368  * @journal: the journal to shutdown.
2369  * @errno:   an error number to record in the journal indicating
2370  *           the reason for the shutdown.
2371  *
2372  * Perform a complete, immediate shutdown of the ENTIRE
2373  * journal (not of a single transaction).  This operation cannot be
2374  * undone without closing and reopening the journal.
2375  *
2376  * The jbd2_journal_abort function is intended to support higher level error
2377  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2378  * mode.
2379  *
2380  * Journal abort has very specific semantics.  Any existing dirty,
2381  * unjournaled buffers in the main filesystem will still be written to
2382  * disk by bdflush, but the journaling mechanism will be suspended
2383  * immediately and no further transaction commits will be honoured.
2384  *
2385  * Any dirty, journaled buffers will be written back to disk without
2386  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2387  * filesystem, but we _do_ attempt to leave as much data as possible
2388  * behind for fsck to use for cleanup.
2389  *
2390  * Any attempt to get a new transaction handle on a journal which is in
2391  * ABORT state will just result in an -EROFS error return.  A
2392  * jbd2_journal_stop on an existing handle will return -EIO if we have
2393  * entered abort state during the update.
2394  *
2395  * Recursive transactions are not disturbed by journal abort until the
2396  * final jbd2_journal_stop, which will receive the -EIO error.
2397  *
2398  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2399  * which will be recorded (if possible) in the journal superblock.  This
2400  * allows a client to record failure conditions in the middle of a
2401  * transaction without having to complete the transaction to record the
2402  * failure to disk.  ext3_error, for example, now uses this
2403  * functionality.
2404  *
2405  */
2406
2407 void jbd2_journal_abort(journal_t *journal, int errno)
2408 {
2409         transaction_t *transaction;
2410
2411         /*
2412          * Lock the aborting procedure until everything is done, this avoid
2413          * races between filesystem's error handling flow (e.g. ext4_abort()),
2414          * ensure panic after the error info is written into journal's
2415          * superblock.
2416          */
2417         mutex_lock(&journal->j_abort_mutex);
2418         /*
2419          * ESHUTDOWN always takes precedence because a file system check
2420          * caused by any other journal abort error is not required after
2421          * a shutdown triggered.
2422          */
2423         write_lock(&journal->j_state_lock);
2424         if (journal->j_flags & JBD2_ABORT) {
2425                 int old_errno = journal->j_errno;
2426
2427                 write_unlock(&journal->j_state_lock);
2428                 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2429                         journal->j_errno = errno;
2430                         jbd2_journal_update_sb_errno(journal);
2431                 }
2432                 mutex_unlock(&journal->j_abort_mutex);
2433                 return;
2434         }
2435
2436         /*
2437          * Mark the abort as occurred and start current running transaction
2438          * to release all journaled buffer.
2439          */
2440         pr_err("Aborting journal on device %s.\n", journal->j_devname);
2441
2442         journal->j_flags |= JBD2_ABORT;
2443         journal->j_errno = errno;
2444         transaction = journal->j_running_transaction;
2445         if (transaction)
2446                 __jbd2_log_start_commit(journal, transaction->t_tid);
2447         write_unlock(&journal->j_state_lock);
2448
2449         /*
2450          * Record errno to the journal super block, so that fsck and jbd2
2451          * layer could realise that a filesystem check is needed.
2452          */
2453         jbd2_journal_update_sb_errno(journal);
2454         mutex_unlock(&journal->j_abort_mutex);
2455 }
2456
2457 /**
2458  * jbd2_journal_errno() - returns the journal's error state.
2459  * @journal: journal to examine.
2460  *
2461  * This is the errno number set with jbd2_journal_abort(), the last
2462  * time the journal was mounted - if the journal was stopped
2463  * without calling abort this will be 0.
2464  *
2465  * If the journal has been aborted on this mount time -EROFS will
2466  * be returned.
2467  */
2468 int jbd2_journal_errno(journal_t *journal)
2469 {
2470         int err;
2471
2472         read_lock(&journal->j_state_lock);
2473         if (journal->j_flags & JBD2_ABORT)
2474                 err = -EROFS;
2475         else
2476                 err = journal->j_errno;
2477         read_unlock(&journal->j_state_lock);
2478         return err;
2479 }
2480
2481 /**
2482  * jbd2_journal_clear_err() - clears the journal's error state
2483  * @journal: journal to act on.
2484  *
2485  * An error must be cleared or acked to take a FS out of readonly
2486  * mode.
2487  */
2488 int jbd2_journal_clear_err(journal_t *journal)
2489 {
2490         int err = 0;
2491
2492         write_lock(&journal->j_state_lock);
2493         if (journal->j_flags & JBD2_ABORT)
2494                 err = -EROFS;
2495         else
2496                 journal->j_errno = 0;
2497         write_unlock(&journal->j_state_lock);
2498         return err;
2499 }
2500
2501 /**
2502  * jbd2_journal_ack_err() - Ack journal err.
2503  * @journal: journal to act on.
2504  *
2505  * An error must be cleared or acked to take a FS out of readonly
2506  * mode.
2507  */
2508 void jbd2_journal_ack_err(journal_t *journal)
2509 {
2510         write_lock(&journal->j_state_lock);
2511         if (journal->j_errno)
2512                 journal->j_flags |= JBD2_ACK_ERR;
2513         write_unlock(&journal->j_state_lock);
2514 }
2515
2516 int jbd2_journal_blocks_per_page(struct inode *inode)
2517 {
2518         return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2519 }
2520
2521 /*
2522  * helper functions to deal with 32 or 64bit block numbers.
2523  */
2524 size_t journal_tag_bytes(journal_t *journal)
2525 {
2526         size_t sz;
2527
2528         if (jbd2_has_feature_csum3(journal))
2529                 return sizeof(journal_block_tag3_t);
2530
2531         sz = sizeof(journal_block_tag_t);
2532
2533         if (jbd2_has_feature_csum2(journal))
2534                 sz += sizeof(__u16);
2535
2536         if (jbd2_has_feature_64bit(journal))
2537                 return sz;
2538         else
2539                 return sz - sizeof(__u32);
2540 }
2541
2542 /*
2543  * JBD memory management
2544  *
2545  * These functions are used to allocate block-sized chunks of memory
2546  * used for making copies of buffer_head data.  Very often it will be
2547  * page-sized chunks of data, but sometimes it will be in
2548  * sub-page-size chunks.  (For example, 16k pages on Power systems
2549  * with a 4k block file system.)  For blocks smaller than a page, we
2550  * use a SLAB allocator.  There are slab caches for each block size,
2551  * which are allocated at mount time, if necessary, and we only free
2552  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2553  * this reason we don't need to a mutex to protect access to
2554  * jbd2_slab[] allocating or releasing memory; only in
2555  * jbd2_journal_create_slab().
2556  */
2557 #define JBD2_MAX_SLABS 8
2558 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2559
2560 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2561         "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2562         "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2563 };
2564
2565
2566 static void jbd2_journal_destroy_slabs(void)
2567 {
2568         int i;
2569
2570         for (i = 0; i < JBD2_MAX_SLABS; i++) {
2571                 kmem_cache_destroy(jbd2_slab[i]);
2572                 jbd2_slab[i] = NULL;
2573         }
2574 }
2575
2576 static int jbd2_journal_create_slab(size_t size)
2577 {
2578         static DEFINE_MUTEX(jbd2_slab_create_mutex);
2579         int i = order_base_2(size) - 10;
2580         size_t slab_size;
2581
2582         if (size == PAGE_SIZE)
2583                 return 0;
2584
2585         if (i >= JBD2_MAX_SLABS)
2586                 return -EINVAL;
2587
2588         if (unlikely(i < 0))
2589                 i = 0;
2590         mutex_lock(&jbd2_slab_create_mutex);
2591         if (jbd2_slab[i]) {
2592                 mutex_unlock(&jbd2_slab_create_mutex);
2593                 return 0;       /* Already created */
2594         }
2595
2596         slab_size = 1 << (i+10);
2597         jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2598                                          slab_size, 0, NULL);
2599         mutex_unlock(&jbd2_slab_create_mutex);
2600         if (!jbd2_slab[i]) {
2601                 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2602                 return -ENOMEM;
2603         }
2604         return 0;
2605 }
2606
2607 static struct kmem_cache *get_slab(size_t size)
2608 {
2609         int i = order_base_2(size) - 10;
2610
2611         BUG_ON(i >= JBD2_MAX_SLABS);
2612         if (unlikely(i < 0))
2613                 i = 0;
2614         BUG_ON(jbd2_slab[i] == NULL);
2615         return jbd2_slab[i];
2616 }
2617
2618 void *jbd2_alloc(size_t size, gfp_t flags)
2619 {
2620         void *ptr;
2621
2622         BUG_ON(size & (size-1)); /* Must be a power of 2 */
2623
2624         if (size < PAGE_SIZE)
2625                 ptr = kmem_cache_alloc(get_slab(size), flags);
2626         else
2627                 ptr = (void *)__get_free_pages(flags, get_order(size));
2628
2629         /* Check alignment; SLUB has gotten this wrong in the past,
2630          * and this can lead to user data corruption! */
2631         BUG_ON(((unsigned long) ptr) & (size-1));
2632
2633         return ptr;
2634 }
2635
2636 void jbd2_free(void *ptr, size_t size)
2637 {
2638         if (size < PAGE_SIZE)
2639                 kmem_cache_free(get_slab(size), ptr);
2640         else
2641                 free_pages((unsigned long)ptr, get_order(size));
2642 };
2643
2644 /*
2645  * Journal_head storage management
2646  */
2647 static struct kmem_cache *jbd2_journal_head_cache;
2648 #ifdef CONFIG_JBD2_DEBUG
2649 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2650 #endif
2651
2652 static int __init jbd2_journal_init_journal_head_cache(void)
2653 {
2654         J_ASSERT(!jbd2_journal_head_cache);
2655         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2656                                 sizeof(struct journal_head),
2657                                 0,              /* offset */
2658                                 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2659                                 NULL);          /* ctor */
2660         if (!jbd2_journal_head_cache) {
2661                 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2662                 return -ENOMEM;
2663         }
2664         return 0;
2665 }
2666
2667 static void jbd2_journal_destroy_journal_head_cache(void)
2668 {
2669         kmem_cache_destroy(jbd2_journal_head_cache);
2670         jbd2_journal_head_cache = NULL;
2671 }
2672
2673 /*
2674  * journal_head splicing and dicing
2675  */
2676 static struct journal_head *journal_alloc_journal_head(void)
2677 {
2678         struct journal_head *ret;
2679
2680 #ifdef CONFIG_JBD2_DEBUG
2681         atomic_inc(&nr_journal_heads);
2682 #endif
2683         ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2684         if (!ret) {
2685                 jbd_debug(1, "out of memory for journal_head\n");
2686                 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2687                 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2688                                 GFP_NOFS | __GFP_NOFAIL);
2689         }
2690         if (ret)
2691                 spin_lock_init(&ret->b_state_lock);
2692         return ret;
2693 }
2694
2695 static void journal_free_journal_head(struct journal_head *jh)
2696 {
2697 #ifdef CONFIG_JBD2_DEBUG
2698         atomic_dec(&nr_journal_heads);
2699         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2700 #endif
2701         kmem_cache_free(jbd2_journal_head_cache, jh);
2702 }
2703
2704 /*
2705  * A journal_head is attached to a buffer_head whenever JBD has an
2706  * interest in the buffer.
2707  *
2708  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2709  * is set.  This bit is tested in core kernel code where we need to take
2710  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2711  * there.
2712  *
2713  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2714  *
2715  * When a buffer has its BH_JBD bit set it is immune from being released by
2716  * core kernel code, mainly via ->b_count.
2717  *
2718  * A journal_head is detached from its buffer_head when the journal_head's
2719  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2720  * transaction (b_cp_transaction) hold their references to b_jcount.
2721  *
2722  * Various places in the kernel want to attach a journal_head to a buffer_head
2723  * _before_ attaching the journal_head to a transaction.  To protect the
2724  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2725  * journal_head's b_jcount refcount by one.  The caller must call
2726  * jbd2_journal_put_journal_head() to undo this.
2727  *
2728  * So the typical usage would be:
2729  *
2730  *      (Attach a journal_head if needed.  Increments b_jcount)
2731  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2732  *      ...
2733  *      (Get another reference for transaction)
2734  *      jbd2_journal_grab_journal_head(bh);
2735  *      jh->b_transaction = xxx;
2736  *      (Put original reference)
2737  *      jbd2_journal_put_journal_head(jh);
2738  */
2739
2740 /*
2741  * Give a buffer_head a journal_head.
2742  *
2743  * May sleep.
2744  */
2745 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2746 {
2747         struct journal_head *jh;
2748         struct journal_head *new_jh = NULL;
2749
2750 repeat:
2751         if (!buffer_jbd(bh))
2752                 new_jh = journal_alloc_journal_head();
2753
2754         jbd_lock_bh_journal_head(bh);
2755         if (buffer_jbd(bh)) {
2756                 jh = bh2jh(bh);
2757         } else {
2758                 J_ASSERT_BH(bh,
2759                         (atomic_read(&bh->b_count) > 0) ||
2760                         (bh->b_page && bh->b_page->mapping));
2761
2762                 if (!new_jh) {
2763                         jbd_unlock_bh_journal_head(bh);
2764                         goto repeat;
2765                 }
2766
2767                 jh = new_jh;
2768                 new_jh = NULL;          /* We consumed it */
2769                 set_buffer_jbd(bh);
2770                 bh->b_private = jh;
2771                 jh->b_bh = bh;
2772                 get_bh(bh);
2773                 BUFFER_TRACE(bh, "added journal_head");
2774         }
2775         jh->b_jcount++;
2776         jbd_unlock_bh_journal_head(bh);
2777         if (new_jh)
2778                 journal_free_journal_head(new_jh);
2779         return bh->b_private;
2780 }
2781
2782 /*
2783  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2784  * having a journal_head, return NULL
2785  */
2786 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2787 {
2788         struct journal_head *jh = NULL;
2789
2790         jbd_lock_bh_journal_head(bh);
2791         if (buffer_jbd(bh)) {
2792                 jh = bh2jh(bh);
2793                 jh->b_jcount++;
2794         }
2795         jbd_unlock_bh_journal_head(bh);
2796         return jh;
2797 }
2798
2799 static void __journal_remove_journal_head(struct buffer_head *bh)
2800 {
2801         struct journal_head *jh = bh2jh(bh);
2802
2803         J_ASSERT_JH(jh, jh->b_transaction == NULL);
2804         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2805         J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2806         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2807         J_ASSERT_BH(bh, buffer_jbd(bh));
2808         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2809         BUFFER_TRACE(bh, "remove journal_head");
2810
2811         /* Unlink before dropping the lock */
2812         bh->b_private = NULL;
2813         jh->b_bh = NULL;        /* debug, really */
2814         clear_buffer_jbd(bh);
2815 }
2816
2817 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2818 {
2819         if (jh->b_frozen_data) {
2820                 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2821                 jbd2_free(jh->b_frozen_data, b_size);
2822         }
2823         if (jh->b_committed_data) {
2824                 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2825                 jbd2_free(jh->b_committed_data, b_size);
2826         }
2827         journal_free_journal_head(jh);
2828 }
2829
2830 /*
2831  * Drop a reference on the passed journal_head.  If it fell to zero then
2832  * release the journal_head from the buffer_head.
2833  */
2834 void jbd2_journal_put_journal_head(struct journal_head *jh)
2835 {
2836         struct buffer_head *bh = jh2bh(jh);
2837
2838         jbd_lock_bh_journal_head(bh);
2839         J_ASSERT_JH(jh, jh->b_jcount > 0);
2840         --jh->b_jcount;
2841         if (!jh->b_jcount) {
2842                 __journal_remove_journal_head(bh);
2843                 jbd_unlock_bh_journal_head(bh);
2844                 journal_release_journal_head(jh, bh->b_size);
2845                 __brelse(bh);
2846         } else {
2847                 jbd_unlock_bh_journal_head(bh);
2848         }
2849 }
2850
2851 /*
2852  * Initialize jbd inode head
2853  */
2854 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2855 {
2856         jinode->i_transaction = NULL;
2857         jinode->i_next_transaction = NULL;
2858         jinode->i_vfs_inode = inode;
2859         jinode->i_flags = 0;
2860         jinode->i_dirty_start = 0;
2861         jinode->i_dirty_end = 0;
2862         INIT_LIST_HEAD(&jinode->i_list);
2863 }
2864
2865 /*
2866  * Function to be called before we start removing inode from memory (i.e.,
2867  * clear_inode() is a fine place to be called from). It removes inode from
2868  * transaction's lists.
2869  */
2870 void jbd2_journal_release_jbd_inode(journal_t *journal,
2871                                     struct jbd2_inode *jinode)
2872 {
2873         if (!journal)
2874                 return;
2875 restart:
2876         spin_lock(&journal->j_list_lock);
2877         /* Is commit writing out inode - we have to wait */
2878         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2879                 wait_queue_head_t *wq;
2880                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2881                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2882                 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2883                 spin_unlock(&journal->j_list_lock);
2884                 schedule();
2885                 finish_wait(wq, &wait.wq_entry);
2886                 goto restart;
2887         }
2888
2889         if (jinode->i_transaction) {
2890                 list_del(&jinode->i_list);
2891                 jinode->i_transaction = NULL;
2892         }
2893         spin_unlock(&journal->j_list_lock);
2894 }
2895
2896
2897 #ifdef CONFIG_PROC_FS
2898
2899 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2900
2901 static void __init jbd2_create_jbd_stats_proc_entry(void)
2902 {
2903         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2904 }
2905
2906 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2907 {
2908         if (proc_jbd2_stats)
2909                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2910 }
2911
2912 #else
2913
2914 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2915 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2916
2917 #endif
2918
2919 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2920
2921 static int __init jbd2_journal_init_inode_cache(void)
2922 {
2923         J_ASSERT(!jbd2_inode_cache);
2924         jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2925         if (!jbd2_inode_cache) {
2926                 pr_emerg("JBD2: failed to create inode cache\n");
2927                 return -ENOMEM;
2928         }
2929         return 0;
2930 }
2931
2932 static int __init jbd2_journal_init_handle_cache(void)
2933 {
2934         J_ASSERT(!jbd2_handle_cache);
2935         jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2936         if (!jbd2_handle_cache) {
2937                 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2938                 return -ENOMEM;
2939         }
2940         return 0;
2941 }
2942
2943 static void jbd2_journal_destroy_inode_cache(void)
2944 {
2945         kmem_cache_destroy(jbd2_inode_cache);
2946         jbd2_inode_cache = NULL;
2947 }
2948
2949 static void jbd2_journal_destroy_handle_cache(void)
2950 {
2951         kmem_cache_destroy(jbd2_handle_cache);
2952         jbd2_handle_cache = NULL;
2953 }
2954
2955 /*
2956  * Module startup and shutdown
2957  */
2958
2959 static int __init journal_init_caches(void)
2960 {
2961         int ret;
2962
2963         ret = jbd2_journal_init_revoke_record_cache();
2964         if (ret == 0)
2965                 ret = jbd2_journal_init_revoke_table_cache();
2966         if (ret == 0)
2967                 ret = jbd2_journal_init_journal_head_cache();
2968         if (ret == 0)
2969                 ret = jbd2_journal_init_handle_cache();
2970         if (ret == 0)
2971                 ret = jbd2_journal_init_inode_cache();
2972         if (ret == 0)
2973                 ret = jbd2_journal_init_transaction_cache();
2974         return ret;
2975 }
2976
2977 static void jbd2_journal_destroy_caches(void)
2978 {
2979         jbd2_journal_destroy_revoke_record_cache();
2980         jbd2_journal_destroy_revoke_table_cache();
2981         jbd2_journal_destroy_journal_head_cache();
2982         jbd2_journal_destroy_handle_cache();
2983         jbd2_journal_destroy_inode_cache();
2984         jbd2_journal_destroy_transaction_cache();
2985         jbd2_journal_destroy_slabs();
2986 }
2987
2988 static int __init journal_init(void)
2989 {
2990         int ret;
2991
2992         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2993
2994         ret = journal_init_caches();
2995         if (ret == 0) {
2996                 jbd2_create_jbd_stats_proc_entry();
2997         } else {
2998                 jbd2_journal_destroy_caches();
2999         }
3000         return ret;
3001 }
3002
3003 static void __exit journal_exit(void)
3004 {
3005 #ifdef CONFIG_JBD2_DEBUG
3006         int n = atomic_read(&nr_journal_heads);
3007         if (n)
3008                 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3009 #endif
3010         jbd2_remove_jbd_stats_proc_entry();
3011         jbd2_journal_destroy_caches();
3012 }
3013
3014 MODULE_LICENSE("GPL");
3015 module_init(journal_init);
3016 module_exit(journal_exit);
3017