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                 /*
 928                  * Update j_fc_off so jbd2_fc_release_bufs can release remain
 929                  * buffer head.
 930                  */
 931                 if (unlikely(!buffer_uptodate(bh))) {
 932                         journal->j_fc_off = i + 1;
 933                         return -EIO;
 934                 }
 935                 put_bh(bh);
 936                 journal->j_fc_wbuf[i] = NULL;
 937         }
 938
 939         return 0;
 940 }
 941 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
 942
 943 /*
 944  * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
 945  * for completion.
 946  */
 947 int jbd2_fc_release_bufs(journal_t *journal)
 948 {
 949         struct buffer_head *bh;
 950         int i, j_fc_off;
 951
 952         j_fc_off = journal->j_fc_off;
 953
 954         /*
 955          * Wait in reverse order to minimize chances of us being woken up before
 956          * all IOs have completed
 957          */
 958         for (i = j_fc_off - 1; i >= 0; i--) {
 959                 bh = journal->j_fc_wbuf[i];
 960                 if (!bh)
 961                         break;
 962                 put_bh(bh);
 963                 journal->j_fc_wbuf[i] = NULL;
 964         }
 965
 966         return 0;
 967 }
 968 EXPORT_SYMBOL(jbd2_fc_release_bufs);
 969
 970 /*
 971  * Conversion of logical to physical block numbers for the journal
 972  *
 973  * On external journals the journal blocks are identity-mapped, so
 974  * this is a no-op.  If needed, we can use j_blk_offset - everything is
 975  * ready.
 976  */
 977 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
 978                  unsigned long long *retp)
 979 {
 980         int err = 0;
 981         unsigned long long ret;
 982         sector_t block = 0;
 983
 984         if (journal->j_inode) {
 985                 block = blocknr;
 986                 ret = bmap(journal->j_inode, &block);
 987
 988                 if (ret || !block) {
 989                         printk(KERN_ALERT "%s: journal block not found "
 990                                         "at offset %lu on %s\n",
 991                                __func__, blocknr, journal->j_devname);
 992                         err = -EIO;
 993                         jbd2_journal_abort(journal, err);
 994                 } else {
 995                         *retp = block;
 996                 }
 997
 998         } else {
 999                 *retp = blocknr; /* +journal->j_blk_offset */
1000         }
1001         return err;
1002 }
1003
1004 /*
1005  * We play buffer_head aliasing tricks to write data/metadata blocks to
1006  * the journal without copying their contents, but for journal
1007  * descriptor blocks we do need to generate bona fide buffers.
1008  *
1009  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
1010  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
1011  * But we don't bother doing that, so there will be coherency problems with
1012  * mmaps of blockdevs which hold live JBD-controlled filesystems.
1013  */
1014 struct buffer_head *
1015 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1016 {
1017         journal_t *journal = transaction->t_journal;
1018         struct buffer_head *bh;
1019         unsigned long long blocknr;
1020         journal_header_t *header;
1021         int err;
1022
1023         err = jbd2_journal_next_log_block(journal, &blocknr);
1024
1025         if (err)
1026                 return NULL;
1027
1028         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1029         if (!bh)
1030                 return NULL;
1031         atomic_dec(&transaction->t_outstanding_credits);
1032         lock_buffer(bh);
1033         memset(bh->b_data, 0, journal->j_blocksize);
1034         header = (journal_header_t *)bh->b_data;
1035         header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1036         header->h_blocktype = cpu_to_be32(type);
1037         header->h_sequence = cpu_to_be32(transaction->t_tid);
1038         set_buffer_uptodate(bh);
1039         unlock_buffer(bh);
1040         BUFFER_TRACE(bh, "return this buffer");
1041         return bh;
1042 }
1043
1044 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1045 {
1046         struct jbd2_journal_block_tail *tail;
1047         __u32 csum;
1048
1049         if (!jbd2_journal_has_csum_v2or3(j))
1050                 return;
1051
1052         tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1053                         sizeof(struct jbd2_journal_block_tail));
1054         tail->t_checksum = 0;
1055         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1056         tail->t_checksum = cpu_to_be32(csum);
1057 }
1058
1059 /*
1060  * Return tid of the oldest transaction in the journal and block in the journal
1061  * where the transaction starts.
1062  *
1063  * If the journal is now empty, return which will be the next transaction ID
1064  * we will write and where will that transaction start.
1065  *
1066  * The return value is 0 if journal tail cannot be pushed any further, 1 if
1067  * it can.
1068  */
1069 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1070                               unsigned long *block)
1071 {
1072         transaction_t *transaction;
1073         int ret;
1074
1075         read_lock(&journal->j_state_lock);
1076         spin_lock(&journal->j_list_lock);
1077         transaction = journal->j_checkpoint_transactions;
1078         if (transaction) {
1079                 *tid = transaction->t_tid;
1080                 *block = transaction->t_log_start;
1081         } else if ((transaction = journal->j_committing_transaction) != NULL) {
1082                 *tid = transaction->t_tid;
1083                 *block = transaction->t_log_start;
1084         } else if ((transaction = journal->j_running_transaction) != NULL) {
1085                 *tid = transaction->t_tid;
1086                 *block = journal->j_head;
1087         } else {
1088                 *tid = journal->j_transaction_sequence;
1089                 *block = journal->j_head;
1090         }
1091         ret = tid_gt(*tid, journal->j_tail_sequence);
1092         spin_unlock(&journal->j_list_lock);
1093         read_unlock(&journal->j_state_lock);
1094
1095         return ret;
1096 }
1097
1098 /*
1099  * Update information in journal structure and in on disk journal superblock
1100  * about log tail. This function does not check whether information passed in
1101  * really pushes log tail further. It's responsibility of the caller to make
1102  * sure provided log tail information is valid (e.g. by holding
1103  * j_checkpoint_mutex all the time between computing log tail and calling this
1104  * function as is the case with jbd2_cleanup_journal_tail()).
1105  *
1106  * Requires j_checkpoint_mutex
1107  */
1108 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1109 {
1110         unsigned long freed;
1111         int ret;
1112
1113         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1114
1115         /*
1116          * We cannot afford for write to remain in drive's caches since as
1117          * soon as we update j_tail, next transaction can start reusing journal
1118          * space and if we lose sb update during power failure we'd replay
1119          * old transaction with possibly newly overwritten data.
1120          */
1121         ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
1122                                               REQ_SYNC | REQ_FUA);
1123         if (ret)
1124                 goto out;
1125
1126         write_lock(&journal->j_state_lock);
1127         freed = block - journal->j_tail;
1128         if (block < journal->j_tail)
1129                 freed += journal->j_last - journal->j_first;
1130
1131         trace_jbd2_update_log_tail(journal, tid, block, freed);
1132         jbd_debug(1,
1133                   "Cleaning journal tail from %u to %u (offset %lu), "
1134                   "freeing %lu\n",
1135                   journal->j_tail_sequence, tid, block, freed);
1136
1137         journal->j_free += freed;
1138         journal->j_tail_sequence = tid;
1139         journal->j_tail = block;
1140         write_unlock(&journal->j_state_lock);
1141
1142 out:
1143         return ret;
1144 }
1145
1146 /*
1147  * This is a variation of __jbd2_update_log_tail which checks for validity of
1148  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1149  * with other threads updating log tail.
1150  */
1151 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1152 {
1153         mutex_lock_io(&journal->j_checkpoint_mutex);
1154         if (tid_gt(tid, journal->j_tail_sequence))
1155                 __jbd2_update_log_tail(journal, tid, block);
1156         mutex_unlock(&journal->j_checkpoint_mutex);
1157 }
1158
1159 struct jbd2_stats_proc_session {
1160         journal_t *journal;
1161         struct transaction_stats_s *stats;
1162         int start;
1163         int max;
1164 };
1165
1166 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1167 {
1168         return *pos ? NULL : SEQ_START_TOKEN;
1169 }
1170
1171 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1172 {
1173         (*pos)++;
1174         return NULL;
1175 }
1176
1177 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1178 {
1179         struct jbd2_stats_proc_session *s = seq->private;
1180
1181         if (v != SEQ_START_TOKEN)
1182                 return 0;
1183         seq_printf(seq, "%lu transactions (%lu requested), "
1184                    "each up to %u blocks\n",
1185                    s->stats->ts_tid, s->stats->ts_requested,
1186                    s->journal->j_max_transaction_buffers);
1187         if (s->stats->ts_tid == 0)
1188                 return 0;
1189         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
1190             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1191         seq_printf(seq, "  %ums request delay\n",
1192             (s->stats->ts_requested == 0) ? 0 :
1193             jiffies_to_msecs(s->stats->run.rs_request_delay /
1194                              s->stats->ts_requested));
1195         seq_printf(seq, "  %ums running transaction\n",
1196             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1197         seq_printf(seq, "  %ums transaction was being locked\n",
1198             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1199         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
1200             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1201         seq_printf(seq, "  %ums logging transaction\n",
1202             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1203         seq_printf(seq, "  %lluus average transaction commit time\n",
1204                    div_u64(s->journal->j_average_commit_time, 1000));
1205         seq_printf(seq, "  %lu handles per transaction\n",
1206             s->stats->run.rs_handle_count / s->stats->ts_tid);
1207         seq_printf(seq, "  %lu blocks per transaction\n",
1208             s->stats->run.rs_blocks / s->stats->ts_tid);
1209         seq_printf(seq, "  %lu logged blocks per transaction\n",
1210             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1211         return 0;
1212 }
1213
1214 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1215 {
1216 }
1217
1218 static const struct seq_operations jbd2_seq_info_ops = {
1219         .start  = jbd2_seq_info_start,
1220         .next   = jbd2_seq_info_next,
1221         .stop   = jbd2_seq_info_stop,
1222         .show   = jbd2_seq_info_show,
1223 };
1224
1225 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1226 {
1227         journal_t *journal = PDE_DATA(inode);
1228         struct jbd2_stats_proc_session *s;
1229         int rc, size;
1230
1231         s = kmalloc(sizeof(*s), GFP_KERNEL);
1232         if (s == NULL)
1233                 return -ENOMEM;
1234         size = sizeof(struct transaction_stats_s);
1235         s->stats = kmalloc(size, GFP_KERNEL);
1236         if (s->stats == NULL) {
1237                 kfree(s);
1238                 return -ENOMEM;
1239         }
1240         spin_lock(&journal->j_history_lock);
1241         memcpy(s->stats, &journal->j_stats, size);
1242         s->journal = journal;
1243         spin_unlock(&journal->j_history_lock);
1244
1245         rc = seq_open(file, &jbd2_seq_info_ops);
1246         if (rc == 0) {
1247                 struct seq_file *m = file->private_data;
1248                 m->private = s;
1249         } else {
1250                 kfree(s->stats);
1251                 kfree(s);
1252         }
1253         return rc;
1254
1255 }
1256
1257 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1258 {
1259         struct seq_file *seq = file->private_data;
1260         struct jbd2_stats_proc_session *s = seq->private;
1261         kfree(s->stats);
1262         kfree(s);
1263         return seq_release(inode, file);
1264 }
1265
1266 static const struct proc_ops jbd2_info_proc_ops = {
1267         .proc_open      = jbd2_seq_info_open,
1268         .proc_read      = seq_read,
1269         .proc_lseek     = seq_lseek,
1270         .proc_release   = jbd2_seq_info_release,
1271 };
1272
1273 static struct proc_dir_entry *proc_jbd2_stats;
1274
1275 static void jbd2_stats_proc_init(journal_t *journal)
1276 {
1277         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1278         if (journal->j_proc_entry) {
1279                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1280                                  &jbd2_info_proc_ops, journal);
1281         }
1282 }
1283
1284 static void jbd2_stats_proc_exit(journal_t *journal)
1285 {
1286         remove_proc_entry("info", journal->j_proc_entry);
1287         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1288 }
1289
1290 /* Minimum size of descriptor tag */
1291 static int jbd2_min_tag_size(void)
1292 {
1293         /*
1294          * Tag with 32-bit block numbers does not use last four bytes of the
1295          * structure
1296          */
1297         return sizeof(journal_block_tag_t) - 4;
1298 }
1299
1300 /*
1301  * Management for journal control blocks: functions to create and
1302  * destroy journal_t structures, and to initialise and read existing
1303  * journal blocks from disk.  */
1304
1305 /* First: create and setup a journal_t object in memory.  We initialise
1306  * very few fields yet: that has to wait until we have created the
1307  * journal structures from from scratch, or loaded them from disk. */
1308
1309 static journal_t *journal_init_common(struct block_device *bdev,
1310                         struct block_device *fs_dev,
1311                         unsigned long long start, int len, int blocksize)
1312 {
1313         static struct lock_class_key jbd2_trans_commit_key;
1314         journal_t *journal;
1315         int err;
1316         struct buffer_head *bh;
1317         int n;
1318
1319         journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1320         if (!journal)
1321                 return NULL;
1322
1323         init_waitqueue_head(&journal->j_wait_transaction_locked);
1324         init_waitqueue_head(&journal->j_wait_done_commit);
1325         init_waitqueue_head(&journal->j_wait_commit);
1326         init_waitqueue_head(&journal->j_wait_updates);
1327         init_waitqueue_head(&journal->j_wait_reserved);
1328         init_waitqueue_head(&journal->j_fc_wait);
1329         mutex_init(&journal->j_abort_mutex);
1330         mutex_init(&journal->j_barrier);
1331         mutex_init(&journal->j_checkpoint_mutex);
1332         spin_lock_init(&journal->j_revoke_lock);
1333         spin_lock_init(&journal->j_list_lock);
1334         rwlock_init(&journal->j_state_lock);
1335
1336         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1337         journal->j_min_batch_time = 0;
1338         journal->j_max_batch_time = 15000; /* 15ms */
1339         atomic_set(&journal->j_reserved_credits, 0);
1340
1341         /* The journal is marked for error until we succeed with recovery! */
1342         journal->j_flags = JBD2_ABORT;
1343
1344         /* Set up a default-sized revoke table for the new mount. */
1345         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1346         if (err)
1347                 goto err_cleanup;
1348
1349         spin_lock_init(&journal->j_history_lock);
1350
1351         lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1352                          &jbd2_trans_commit_key, 0);
1353
1354         /* journal descriptor can store up to n blocks -bzzz */
1355         journal->j_blocksize = blocksize;
1356         journal->j_dev = bdev;
1357         journal->j_fs_dev = fs_dev;
1358         journal->j_blk_offset = start;
1359         journal->j_total_len = len;
1360         /* We need enough buffers to write out full descriptor block. */
1361         n = journal->j_blocksize / jbd2_min_tag_size();
1362         journal->j_wbufsize = n;
1363         journal->j_fc_wbuf = NULL;
1364         journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1365                                         GFP_KERNEL);
1366         if (!journal->j_wbuf)
1367                 goto err_cleanup;
1368
1369         bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1370         if (!bh) {
1371                 pr_err("%s: Cannot get buffer for journal superblock\n",
1372                         __func__);
1373                 goto err_cleanup;
1374         }
1375         journal->j_sb_buffer = bh;
1376         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1377
1378         return journal;
1379
1380 err_cleanup:
1381         kfree(journal->j_wbuf);
1382         jbd2_journal_destroy_revoke(journal);
1383         kfree(journal);
1384         return NULL;
1385 }
1386
1387 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1388  *
1389  * Create a journal structure assigned some fixed set of disk blocks to
1390  * the journal.  We don't actually touch those disk blocks yet, but we
1391  * need to set up all of the mapping information to tell the journaling
1392  * system where the journal blocks are.
1393  *
1394  */
1395
1396 /**
1397  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1398  *  @bdev: Block device on which to create the journal
1399  *  @fs_dev: Device which hold journalled filesystem for this journal.
1400  *  @start: Block nr Start of journal.
1401  *  @len:  Length of the journal in blocks.
1402  *  @blocksize: blocksize of journalling device
1403  *
1404  *  Returns: a newly created journal_t *
1405  *
1406  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1407  *  range of blocks on an arbitrary block device.
1408  *
1409  */
1410 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1411                         struct block_device *fs_dev,
1412                         unsigned long long start, int len, int blocksize)
1413 {
1414         journal_t *journal;
1415
1416         journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1417         if (!journal)
1418                 return NULL;
1419
1420         bdevname(journal->j_dev, journal->j_devname);
1421         strreplace(journal->j_devname, '/', '!');
1422         jbd2_stats_proc_init(journal);
1423
1424         return journal;
1425 }
1426
1427 /**
1428  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1429  *  @inode: An inode to create the journal in
1430  *
1431  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1432  * the journal.  The inode must exist already, must support bmap() and
1433  * must have all data blocks preallocated.
1434  */
1435 journal_t *jbd2_journal_init_inode(struct inode *inode)
1436 {
1437         journal_t *journal;
1438         sector_t blocknr;
1439         char *p;
1440         int err = 0;
1441
1442         blocknr = 0;
1443         err = bmap(inode, &blocknr);
1444
1445         if (err || !blocknr) {
1446                 pr_err("%s: Cannot locate journal superblock\n",
1447                         __func__);
1448                 return NULL;
1449         }
1450
1451         jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1452                   inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1453                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1454
1455         journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1456                         blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1457                         inode->i_sb->s_blocksize);
1458         if (!journal)
1459                 return NULL;
1460
1461         journal->j_inode = inode;
1462         bdevname(journal->j_dev, journal->j_devname);
1463         p = strreplace(journal->j_devname, '/', '!');
1464         sprintf(p, "-%lu", journal->j_inode->i_ino);
1465         jbd2_stats_proc_init(journal);
1466
1467         return journal;
1468 }
1469
1470 /*
1471  * If the journal init or create aborts, we need to mark the journal
1472  * superblock as being NULL to prevent the journal destroy from writing
1473  * back a bogus superblock.
1474  */
1475 static void journal_fail_superblock(journal_t *journal)
1476 {
1477         struct buffer_head *bh = journal->j_sb_buffer;
1478         brelse(bh);
1479         journal->j_sb_buffer = NULL;
1480 }
1481
1482 /*
1483  * Given a journal_t structure, initialise the various fields for
1484  * startup of a new journaling session.  We use this both when creating
1485  * a journal, and after recovering an old journal to reset it for
1486  * subsequent use.
1487  */
1488
1489 static int journal_reset(journal_t *journal)
1490 {
1491         journal_superblock_t *sb = journal->j_superblock;
1492         unsigned long long first, last;
1493
1494         first = be32_to_cpu(sb->s_first);
1495         last = be32_to_cpu(sb->s_maxlen);
1496         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1497                 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1498                        first, last);
1499                 journal_fail_superblock(journal);
1500                 return -EINVAL;
1501         }
1502
1503         journal->j_first = first;
1504         journal->j_last = last;
1505
1506         journal->j_head = journal->j_first;
1507         journal->j_tail = journal->j_first;
1508         journal->j_free = journal->j_last - journal->j_first;
1509
1510         journal->j_tail_sequence = journal->j_transaction_sequence;
1511         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1512         journal->j_commit_request = journal->j_commit_sequence;
1513
1514         journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1515
1516         /*
1517          * Now that journal recovery is done, turn fast commits off here. This
1518          * way, if fast commit was enabled before the crash but if now FS has
1519          * disabled it, we don't enable fast commits.
1520          */
1521         jbd2_clear_feature_fast_commit(journal);
1522
1523         /*
1524          * As a special case, if the on-disk copy is already marked as needing
1525          * no recovery (s_start == 0), then we can safely defer the superblock
1526          * update until the next commit by setting JBD2_FLUSHED.  This avoids
1527          * attempting a write to a potential-readonly device.
1528          */
1529         if (sb->s_start == 0) {
1530                 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1531                         "(start %ld, seq %u, errno %d)\n",
1532                         journal->j_tail, journal->j_tail_sequence,
1533                         journal->j_errno);
1534                 journal->j_flags |= JBD2_FLUSHED;
1535         } else {
1536                 /* Lock here to make assertions happy... */
1537                 mutex_lock_io(&journal->j_checkpoint_mutex);
1538                 /*
1539                  * Update log tail information. We use REQ_FUA since new
1540                  * transaction will start reusing journal space and so we
1541                  * must make sure information about current log tail is on
1542                  * disk before that.
1543                  */
1544                 jbd2_journal_update_sb_log_tail(journal,
1545                                                 journal->j_tail_sequence,
1546                                                 journal->j_tail,
1547                                                 REQ_SYNC | REQ_FUA);
1548                 mutex_unlock(&journal->j_checkpoint_mutex);
1549         }
1550         return jbd2_journal_start_thread(journal);
1551 }
1552
1553 /*
1554  * This function expects that the caller will have locked the journal
1555  * buffer head, and will return with it unlocked
1556  */
1557 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1558 {
1559         struct buffer_head *bh = journal->j_sb_buffer;
1560         journal_superblock_t *sb = journal->j_superblock;
1561         int ret;
1562
1563         /* Buffer got discarded which means block device got invalidated */
1564         if (!buffer_mapped(bh)) {
1565                 unlock_buffer(bh);
1566                 return -EIO;
1567         }
1568
1569         trace_jbd2_write_superblock(journal, write_flags);
1570         if (!(journal->j_flags & JBD2_BARRIER))
1571                 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1572         if (buffer_write_io_error(bh)) {
1573                 /*
1574                  * Oh, dear.  A previous attempt to write the journal
1575                  * superblock failed.  This could happen because the
1576                  * USB device was yanked out.  Or it could happen to
1577                  * be a transient write error and maybe the block will
1578                  * be remapped.  Nothing we can do but to retry the
1579                  * write and hope for the best.
1580                  */
1581                 printk(KERN_ERR "JBD2: previous I/O error detected "
1582                        "for journal superblock update for %s.\n",
1583                        journal->j_devname);
1584                 clear_buffer_write_io_error(bh);
1585                 set_buffer_uptodate(bh);
1586         }
1587         if (jbd2_journal_has_csum_v2or3(journal))
1588                 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1589         get_bh(bh);
1590         bh->b_end_io = end_buffer_write_sync;
1591         ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1592         wait_on_buffer(bh);
1593         if (buffer_write_io_error(bh)) {
1594                 clear_buffer_write_io_error(bh);
1595                 set_buffer_uptodate(bh);
1596                 ret = -EIO;
1597         }
1598         if (ret) {
1599                 printk(KERN_ERR "JBD2: Error %d detected when updating "
1600                        "journal superblock for %s.\n", ret,
1601                        journal->j_devname);
1602                 if (!is_journal_aborted(journal))
1603                         jbd2_journal_abort(journal, ret);
1604         }
1605
1606         return ret;
1607 }
1608
1609 /**
1610  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1611  * @journal: The journal to update.
1612  * @tail_tid: TID of the new transaction at the tail of the log
1613  * @tail_block: The first block of the transaction at the tail of the log
1614  * @write_op: With which operation should we write the journal sb
1615  *
1616  * Update a journal's superblock information about log tail and write it to
1617  * disk, waiting for the IO to complete.
1618  */
1619 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1620                                      unsigned long tail_block, int write_op)
1621 {
1622         journal_superblock_t *sb = journal->j_superblock;
1623         int ret;
1624
1625         if (is_journal_aborted(journal))
1626                 return -EIO;
1627
1628         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1629         jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1630                   tail_block, tail_tid);
1631
1632         lock_buffer(journal->j_sb_buffer);
1633         sb->s_sequence = cpu_to_be32(tail_tid);
1634         sb->s_start    = cpu_to_be32(tail_block);
1635
1636         ret = jbd2_write_superblock(journal, write_op);
1637         if (ret)
1638                 goto out;
1639
1640         /* Log is no longer empty */
1641         write_lock(&journal->j_state_lock);
1642         WARN_ON(!sb->s_sequence);
1643         journal->j_flags &= ~JBD2_FLUSHED;
1644         write_unlock(&journal->j_state_lock);
1645
1646 out:
1647         return ret;
1648 }
1649
1650 /**
1651  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1652  * @journal: The journal to update.
1653  * @write_op: With which operation should we write the journal sb
1654  *
1655  * Update a journal's dynamic superblock fields to show that journal is empty.
1656  * Write updated superblock to disk waiting for IO to complete.
1657  */
1658 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1659 {
1660         journal_superblock_t *sb = journal->j_superblock;
1661         bool had_fast_commit = false;
1662
1663         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1664         lock_buffer(journal->j_sb_buffer);
1665         if (sb->s_start == 0) {         /* Is it already empty? */
1666                 unlock_buffer(journal->j_sb_buffer);
1667                 return;
1668         }
1669
1670         jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1671                   journal->j_tail_sequence);
1672
1673         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1674         sb->s_start    = cpu_to_be32(0);
1675         if (jbd2_has_feature_fast_commit(journal)) {
1676                 /*
1677                  * When journal is clean, no need to commit fast commit flag and
1678                  * make file system incompatible with older kernels.
1679                  */
1680                 jbd2_clear_feature_fast_commit(journal);
1681                 had_fast_commit = true;
1682         }
1683
1684         jbd2_write_superblock(journal, write_op);
1685
1686         if (had_fast_commit)
1687                 jbd2_set_feature_fast_commit(journal);
1688
1689         /* Log is no longer empty */
1690         write_lock(&journal->j_state_lock);
1691         journal->j_flags |= JBD2_FLUSHED;
1692         write_unlock(&journal->j_state_lock);
1693 }
1694
1695
1696 /**
1697  * jbd2_journal_update_sb_errno() - Update error in the journal.
1698  * @journal: The journal to update.
1699  *
1700  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1701  * to complete.
1702  */
1703 void jbd2_journal_update_sb_errno(journal_t *journal)
1704 {
1705         journal_superblock_t *sb = journal->j_superblock;
1706         int errcode;
1707
1708         lock_buffer(journal->j_sb_buffer);
1709         errcode = journal->j_errno;
1710         if (errcode == -ESHUTDOWN)
1711                 errcode = 0;
1712         jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1713         sb->s_errno    = cpu_to_be32(errcode);
1714
1715         jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1716 }
1717 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1718
1719 static int journal_revoke_records_per_block(journal_t *journal)
1720 {
1721         int record_size;
1722         int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1723
1724         if (jbd2_has_feature_64bit(journal))
1725                 record_size = 8;
1726         else
1727                 record_size = 4;
1728
1729         if (jbd2_journal_has_csum_v2or3(journal))
1730                 space -= sizeof(struct jbd2_journal_block_tail);
1731         return space / record_size;
1732 }
1733
1734 /*
1735  * Read the superblock for a given journal, performing initial
1736  * validation of the format.
1737  */
1738 static int journal_get_superblock(journal_t *journal)
1739 {
1740         struct buffer_head *bh;
1741         journal_superblock_t *sb;
1742         int err = -EIO;
1743
1744         bh = journal->j_sb_buffer;
1745
1746         J_ASSERT(bh != NULL);
1747         if (!buffer_uptodate(bh)) {
1748                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1749                 wait_on_buffer(bh);
1750                 if (!buffer_uptodate(bh)) {
1751                         printk(KERN_ERR
1752                                 "JBD2: IO error reading journal superblock\n");
1753                         goto out;
1754                 }
1755         }
1756
1757         if (buffer_verified(bh))
1758                 return 0;
1759
1760         sb = journal->j_superblock;
1761
1762         err = -EINVAL;
1763
1764         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1765             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1766                 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1767                 goto out;
1768         }
1769
1770         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1771         case JBD2_SUPERBLOCK_V1:
1772                 journal->j_format_version = 1;
1773                 break;
1774         case JBD2_SUPERBLOCK_V2:
1775                 journal->j_format_version = 2;
1776                 break;
1777         default:
1778                 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1779                 goto out;
1780         }
1781
1782         if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1783                 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1784         else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1785                 printk(KERN_WARNING "JBD2: journal file too short\n");
1786                 goto out;
1787         }
1788
1789         if (be32_to_cpu(sb->s_first) == 0 ||
1790             be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1791                 printk(KERN_WARNING
1792                         "JBD2: Invalid start block of journal: %u\n",
1793                         be32_to_cpu(sb->s_first));
1794                 goto out;
1795         }
1796
1797         if (jbd2_has_feature_csum2(journal) &&
1798             jbd2_has_feature_csum3(journal)) {
1799                 /* Can't have checksum v2 and v3 at the same time! */
1800                 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1801                        "at the same time!\n");
1802                 goto out;
1803         }
1804
1805         if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1806             jbd2_has_feature_checksum(journal)) {
1807                 /* Can't have checksum v1 and v2 on at the same time! */
1808                 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1809                        "at the same time!\n");
1810                 goto out;
1811         }
1812
1813         if (!jbd2_verify_csum_type(journal, sb)) {
1814                 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1815                 goto out;
1816         }
1817
1818         /* Load the checksum driver */
1819         if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1820                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1821                 if (IS_ERR(journal->j_chksum_driver)) {
1822                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1823                         err = PTR_ERR(journal->j_chksum_driver);
1824                         journal->j_chksum_driver = NULL;
1825                         goto out;
1826                 }
1827         }
1828
1829         if (jbd2_journal_has_csum_v2or3(journal)) {
1830                 /* Check superblock checksum */
1831                 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1832                         printk(KERN_ERR "JBD2: journal checksum error\n");
1833                         err = -EFSBADCRC;
1834                         goto out;
1835                 }
1836
1837                 /* Precompute checksum seed for all metadata */
1838                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1839                                                    sizeof(sb->s_uuid));
1840         }
1841
1842         journal->j_revoke_records_per_block =
1843                                 journal_revoke_records_per_block(journal);
1844         set_buffer_verified(bh);
1845
1846         return 0;
1847
1848 out:
1849         journal_fail_superblock(journal);
1850         return err;
1851 }
1852
1853 /*
1854  * Load the on-disk journal superblock and read the key fields into the
1855  * journal_t.
1856  */
1857
1858 static int load_superblock(journal_t *journal)
1859 {
1860         int err;
1861         journal_superblock_t *sb;
1862         int num_fc_blocks;
1863
1864         err = journal_get_superblock(journal);
1865         if (err)
1866                 return err;
1867
1868         sb = journal->j_superblock;
1869
1870         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1871         journal->j_tail = be32_to_cpu(sb->s_start);
1872         journal->j_first = be32_to_cpu(sb->s_first);
1873         journal->j_errno = be32_to_cpu(sb->s_errno);
1874         journal->j_last = be32_to_cpu(sb->s_maxlen);
1875
1876         if (jbd2_has_feature_fast_commit(journal)) {
1877                 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
1878                 num_fc_blocks = be32_to_cpu(sb->s_num_fc_blks);
1879                 if (!num_fc_blocks)
1880                         num_fc_blocks = JBD2_MIN_FC_BLOCKS;
1881                 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
1882                         journal->j_last = journal->j_fc_last - num_fc_blocks;
1883                 journal->j_fc_first = journal->j_last + 1;
1884                 journal->j_fc_off = 0;
1885         }
1886
1887         return 0;
1888 }
1889
1890
1891 /**
1892  * jbd2_journal_load() - Read journal from disk.
1893  * @journal: Journal to act on.
1894  *
1895  * Given a journal_t structure which tells us which disk blocks contain
1896  * a journal, read the journal from disk to initialise the in-memory
1897  * structures.
1898  */
1899 int jbd2_journal_load(journal_t *journal)
1900 {
1901         int err;
1902         journal_superblock_t *sb;
1903
1904         err = load_superblock(journal);
1905         if (err)
1906                 return err;
1907
1908         sb = journal->j_superblock;
1909         /* If this is a V2 superblock, then we have to check the
1910          * features flags on it. */
1911
1912         if (journal->j_format_version >= 2) {
1913                 if ((sb->s_feature_ro_compat &
1914                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1915                     (sb->s_feature_incompat &
1916                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1917                         printk(KERN_WARNING
1918                                 "JBD2: Unrecognised features on journal\n");
1919                         return -EINVAL;
1920                 }
1921         }
1922
1923         /*
1924          * Create a slab for this blocksize
1925          */
1926         err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1927         if (err)
1928                 return err;
1929
1930         /* Let the recovery code check whether it needs to recover any
1931          * data from the journal. */
1932         if (jbd2_journal_recover(journal))
1933                 goto recovery_error;
1934
1935         if (journal->j_failed_commit) {
1936                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1937                        "is corrupt.\n", journal->j_failed_commit,
1938                        journal->j_devname);
1939                 return -EFSCORRUPTED;
1940         }
1941         /*
1942          * clear JBD2_ABORT flag initialized in journal_init_common
1943          * here to update log tail information with the newest seq.
1944          */
1945         journal->j_flags &= ~JBD2_ABORT;
1946
1947         /* OK, we've finished with the dynamic journal bits:
1948          * reinitialise the dynamic contents of the superblock in memory
1949          * and reset them on disk. */
1950         if (journal_reset(journal))
1951                 goto recovery_error;
1952
1953         journal->j_flags |= JBD2_LOADED;
1954         return 0;
1955
1956 recovery_error:
1957         printk(KERN_WARNING "JBD2: recovery failed\n");
1958         return -EIO;
1959 }
1960
1961 /**
1962  * jbd2_journal_destroy() - Release a journal_t structure.
1963  * @journal: Journal to act on.
1964  *
1965  * Release a journal_t structure once it is no longer in use by the
1966  * journaled object.
1967  * Return <0 if we couldn't clean up the journal.
1968  */
1969 int jbd2_journal_destroy(journal_t *journal)
1970 {
1971         int err = 0;
1972
1973         /* Wait for the commit thread to wake up and die. */
1974         journal_kill_thread(journal);
1975
1976         /* Force a final log commit */
1977         if (journal->j_running_transaction)
1978                 jbd2_journal_commit_transaction(journal);
1979
1980         /* Force any old transactions to disk */
1981
1982         /* Totally anal locking here... */
1983         spin_lock(&journal->j_list_lock);
1984         while (journal->j_checkpoint_transactions != NULL) {
1985                 spin_unlock(&journal->j_list_lock);
1986                 mutex_lock_io(&journal->j_checkpoint_mutex);
1987                 err = jbd2_log_do_checkpoint(journal);
1988                 mutex_unlock(&journal->j_checkpoint_mutex);
1989                 /*
1990                  * If checkpointing failed, just free the buffers to avoid
1991                  * looping forever
1992                  */
1993                 if (err) {
1994                         jbd2_journal_destroy_checkpoint(journal);
1995                         spin_lock(&journal->j_list_lock);
1996                         break;
1997                 }
1998                 spin_lock(&journal->j_list_lock);
1999         }
2000
2001         J_ASSERT(journal->j_running_transaction == NULL);
2002         J_ASSERT(journal->j_committing_transaction == NULL);
2003         J_ASSERT(journal->j_checkpoint_transactions == NULL);
2004         spin_unlock(&journal->j_list_lock);
2005
2006         if (journal->j_sb_buffer) {
2007                 if (!is_journal_aborted(journal)) {
2008                         mutex_lock_io(&journal->j_checkpoint_mutex);
2009
2010                         write_lock(&journal->j_state_lock);
2011                         journal->j_tail_sequence =
2012                                 ++journal->j_transaction_sequence;
2013                         write_unlock(&journal->j_state_lock);
2014
2015                         jbd2_mark_journal_empty(journal,
2016                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2017                         mutex_unlock(&journal->j_checkpoint_mutex);
2018                 } else
2019                         err = -EIO;
2020                 brelse(journal->j_sb_buffer);
2021         }
2022
2023         if (journal->j_proc_entry)
2024                 jbd2_stats_proc_exit(journal);
2025         iput(journal->j_inode);
2026         if (journal->j_revoke)
2027                 jbd2_journal_destroy_revoke(journal);
2028         if (journal->j_chksum_driver)
2029                 crypto_free_shash(journal->j_chksum_driver);
2030         kfree(journal->j_fc_wbuf);
2031         kfree(journal->j_wbuf);
2032         kfree(journal);
2033
2034         return err;
2035 }
2036
2037
2038 /**
2039  * jbd2_journal_check_used_features() - Check if features specified are used.
2040  * @journal: Journal to check.
2041  * @compat: bitmask of compatible features
2042  * @ro: bitmask of features that force read-only mount
2043  * @incompat: bitmask of incompatible features
2044  *
2045  * Check whether the journal uses all of a given set of
2046  * features.  Return true (non-zero) if it does.
2047  **/
2048
2049 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2050                                  unsigned long ro, unsigned long incompat)
2051 {
2052         journal_superblock_t *sb;
2053
2054         if (!compat && !ro && !incompat)
2055                 return 1;
2056         /* Load journal superblock if it is not loaded yet. */
2057         if (journal->j_format_version == 0 &&
2058             journal_get_superblock(journal) != 0)
2059                 return 0;
2060         if (journal->j_format_version == 1)
2061                 return 0;
2062
2063         sb = journal->j_superblock;
2064
2065         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2066             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2067             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2068                 return 1;
2069
2070         return 0;
2071 }
2072
2073 /**
2074  * jbd2_journal_check_available_features() - Check feature set in journalling layer
2075  * @journal: Journal to check.
2076  * @compat: bitmask of compatible features
2077  * @ro: bitmask of features that force read-only mount
2078  * @incompat: bitmask of incompatible features
2079  *
2080  * Check whether the journaling code supports the use of
2081  * all of a given set of features on this journal.  Return true
2082  * (non-zero) if it can. */
2083
2084 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2085                                       unsigned long ro, unsigned long incompat)
2086 {
2087         if (!compat && !ro && !incompat)
2088                 return 1;
2089
2090         /* We can support any known requested features iff the
2091          * superblock is in version 2.  Otherwise we fail to support any
2092          * extended sb features. */
2093
2094         if (journal->j_format_version != 2)
2095                 return 0;
2096
2097         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2098             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2099             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2100                 return 1;
2101
2102         return 0;
2103 }
2104
2105 static int
2106 jbd2_journal_initialize_fast_commit(journal_t *journal)
2107 {
2108         journal_superblock_t *sb = journal->j_superblock;
2109         unsigned long long num_fc_blks;
2110
2111         num_fc_blks = be32_to_cpu(sb->s_num_fc_blks);
2112         if (num_fc_blks == 0)
2113                 num_fc_blks = JBD2_MIN_FC_BLOCKS;
2114         if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2115                 return -ENOSPC;
2116
2117         /* Are we called twice? */
2118         WARN_ON(journal->j_fc_wbuf != NULL);
2119         journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2120                                 sizeof(struct buffer_head *), GFP_KERNEL);
2121         if (!journal->j_fc_wbuf)
2122                 return -ENOMEM;
2123
2124         journal->j_fc_wbufsize = num_fc_blks;
2125         journal->j_fc_last = journal->j_last;
2126         journal->j_last = journal->j_fc_last - num_fc_blks;
2127         journal->j_fc_first = journal->j_last + 1;
2128         journal->j_fc_off = 0;
2129         journal->j_free = journal->j_last - journal->j_first;
2130         journal->j_max_transaction_buffers =
2131                 jbd2_journal_get_max_txn_bufs(journal);
2132
2133         return 0;
2134 }
2135
2136 /**
2137  * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2138  * @journal: Journal to act on.
2139  * @compat: bitmask of compatible features
2140  * @ro: bitmask of features that force read-only mount
2141  * @incompat: bitmask of incompatible features
2142  *
2143  * Mark a given journal feature as present on the
2144  * superblock.  Returns true if the requested features could be set.
2145  *
2146  */
2147
2148 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2149                           unsigned long ro, unsigned long incompat)
2150 {
2151 #define INCOMPAT_FEATURE_ON(f) \
2152                 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2153 #define COMPAT_FEATURE_ON(f) \
2154                 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2155         journal_superblock_t *sb;
2156
2157         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2158                 return 1;
2159
2160         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2161                 return 0;
2162
2163         /* If enabling v2 checksums, turn on v3 instead */
2164         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2165                 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2166                 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2167         }
2168
2169         /* Asking for checksumming v3 and v1?  Only give them v3. */
2170         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2171             compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2172                 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2173
2174         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2175                   compat, ro, incompat);
2176
2177         sb = journal->j_superblock;
2178
2179         if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2180                 if (jbd2_journal_initialize_fast_commit(journal)) {
2181                         pr_err("JBD2: Cannot enable fast commits.\n");
2182                         return 0;
2183                 }
2184         }
2185
2186         /* Load the checksum driver if necessary */
2187         if ((journal->j_chksum_driver == NULL) &&
2188             INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2189                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2190                 if (IS_ERR(journal->j_chksum_driver)) {
2191                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2192                         journal->j_chksum_driver = NULL;
2193                         return 0;
2194                 }
2195                 /* Precompute checksum seed for all metadata */
2196                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2197                                                    sizeof(sb->s_uuid));
2198         }
2199
2200         lock_buffer(journal->j_sb_buffer);
2201
2202         /* If enabling v3 checksums, update superblock */
2203         if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2204                 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2205                 sb->s_feature_compat &=
2206                         ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2207         }
2208
2209         /* If enabling v1 checksums, downgrade superblock */
2210         if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2211                 sb->s_feature_incompat &=
2212                         ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2213                                      JBD2_FEATURE_INCOMPAT_CSUM_V3);
2214
2215         sb->s_feature_compat    |= cpu_to_be32(compat);
2216         sb->s_feature_ro_compat |= cpu_to_be32(ro);
2217         sb->s_feature_incompat  |= cpu_to_be32(incompat);
2218         unlock_buffer(journal->j_sb_buffer);
2219         journal->j_revoke_records_per_block =
2220                                 journal_revoke_records_per_block(journal);
2221
2222         return 1;
2223 #undef COMPAT_FEATURE_ON
2224 #undef INCOMPAT_FEATURE_ON
2225 }
2226
2227 /*
2228  * jbd2_journal_clear_features() - Clear a given journal feature in the
2229  *                                  superblock
2230  * @journal: Journal to act on.
2231  * @compat: bitmask of compatible features
2232  * @ro: bitmask of features that force read-only mount
2233  * @incompat: bitmask of incompatible features
2234  *
2235  * Clear a given journal feature as present on the
2236  * superblock.
2237  */
2238 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2239                                 unsigned long ro, unsigned long incompat)
2240 {
2241         journal_superblock_t *sb;
2242
2243         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2244                   compat, ro, incompat);
2245
2246         sb = journal->j_superblock;
2247
2248         sb->s_feature_compat    &= ~cpu_to_be32(compat);
2249         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2250         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
2251         journal->j_revoke_records_per_block =
2252                                 journal_revoke_records_per_block(journal);
2253 }
2254 EXPORT_SYMBOL(jbd2_journal_clear_features);
2255
2256 /**
2257  * jbd2_journal_flush() - Flush journal
2258  * @journal: Journal to act on.
2259  *
2260  * Flush all data for a given journal to disk and empty the journal.
2261  * Filesystems can use this when remounting readonly to ensure that
2262  * recovery does not need to happen on remount.
2263  */
2264
2265 int jbd2_journal_flush(journal_t *journal)
2266 {
2267         int err = 0;
2268         transaction_t *transaction = NULL;
2269
2270         write_lock(&journal->j_state_lock);
2271
2272         /* Force everything buffered to the log... */
2273         if (journal->j_running_transaction) {
2274                 transaction = journal->j_running_transaction;
2275                 __jbd2_log_start_commit(journal, transaction->t_tid);
2276         } else if (journal->j_committing_transaction)
2277                 transaction = journal->j_committing_transaction;
2278
2279         /* Wait for the log commit to complete... */
2280         if (transaction) {
2281                 tid_t tid = transaction->t_tid;
2282
2283                 write_unlock(&journal->j_state_lock);
2284                 jbd2_log_wait_commit(journal, tid);
2285         } else {
2286                 write_unlock(&journal->j_state_lock);
2287         }
2288
2289         /* ...and flush everything in the log out to disk. */
2290         spin_lock(&journal->j_list_lock);
2291         while (!err && journal->j_checkpoint_transactions != NULL) {
2292                 spin_unlock(&journal->j_list_lock);
2293                 mutex_lock_io(&journal->j_checkpoint_mutex);
2294                 err = jbd2_log_do_checkpoint(journal);
2295                 mutex_unlock(&journal->j_checkpoint_mutex);
2296                 spin_lock(&journal->j_list_lock);
2297         }
2298         spin_unlock(&journal->j_list_lock);
2299
2300         if (is_journal_aborted(journal))
2301                 return -EIO;
2302
2303         mutex_lock_io(&journal->j_checkpoint_mutex);
2304         if (!err) {
2305                 err = jbd2_cleanup_journal_tail(journal);
2306                 if (err < 0) {
2307                         mutex_unlock(&journal->j_checkpoint_mutex);
2308                         goto out;
2309                 }
2310                 err = 0;
2311         }
2312
2313         /* Finally, mark the journal as really needing no recovery.
2314          * This sets s_start==0 in the underlying superblock, which is
2315          * the magic code for a fully-recovered superblock.  Any future
2316          * commits of data to the journal will restore the current
2317          * s_start value. */
2318         jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2319         mutex_unlock(&journal->j_checkpoint_mutex);
2320         write_lock(&journal->j_state_lock);
2321         J_ASSERT(!journal->j_running_transaction);
2322         J_ASSERT(!journal->j_committing_transaction);
2323         J_ASSERT(!journal->j_checkpoint_transactions);
2324         J_ASSERT(journal->j_head == journal->j_tail);
2325         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2326         write_unlock(&journal->j_state_lock);
2327 out:
2328         return err;
2329 }
2330
2331 /**
2332  * jbd2_journal_wipe() - Wipe journal contents
2333  * @journal: Journal to act on.
2334  * @write: flag (see below)
2335  *
2336  * Wipe out all of the contents of a journal, safely.  This will produce
2337  * a warning if the journal contains any valid recovery information.
2338  * Must be called between journal_init_*() and jbd2_journal_load().
2339  *
2340  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2341  * we merely suppress recovery.
2342  */
2343
2344 int jbd2_journal_wipe(journal_t *journal, int write)
2345 {
2346         int err = 0;
2347
2348         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2349
2350         err = load_superblock(journal);
2351         if (err)
2352                 return err;
2353
2354         if (!journal->j_tail)
2355                 goto no_recovery;
2356
2357         printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2358                 write ? "Clearing" : "Ignoring");
2359
2360         err = jbd2_journal_skip_recovery(journal);
2361         if (write) {
2362                 /* Lock to make assertions happy... */
2363                 mutex_lock_io(&journal->j_checkpoint_mutex);
2364                 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2365                 mutex_unlock(&journal->j_checkpoint_mutex);
2366         }
2367
2368  no_recovery:
2369         return err;
2370 }
2371
2372 /**
2373  * jbd2_journal_abort () - Shutdown the journal immediately.
2374  * @journal: the journal to shutdown.
2375  * @errno:   an error number to record in the journal indicating
2376  *           the reason for the shutdown.
2377  *
2378  * Perform a complete, immediate shutdown of the ENTIRE
2379  * journal (not of a single transaction).  This operation cannot be
2380  * undone without closing and reopening the journal.
2381  *
2382  * The jbd2_journal_abort function is intended to support higher level error
2383  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2384  * mode.
2385  *
2386  * Journal abort has very specific semantics.  Any existing dirty,
2387  * unjournaled buffers in the main filesystem will still be written to
2388  * disk by bdflush, but the journaling mechanism will be suspended
2389  * immediately and no further transaction commits will be honoured.
2390  *
2391  * Any dirty, journaled buffers will be written back to disk without
2392  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2393  * filesystem, but we _do_ attempt to leave as much data as possible
2394  * behind for fsck to use for cleanup.
2395  *
2396  * Any attempt to get a new transaction handle on a journal which is in
2397  * ABORT state will just result in an -EROFS error return.  A
2398  * jbd2_journal_stop on an existing handle will return -EIO if we have
2399  * entered abort state during the update.
2400  *
2401  * Recursive transactions are not disturbed by journal abort until the
2402  * final jbd2_journal_stop, which will receive the -EIO error.
2403  *
2404  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2405  * which will be recorded (if possible) in the journal superblock.  This
2406  * allows a client to record failure conditions in the middle of a
2407  * transaction without having to complete the transaction to record the
2408  * failure to disk.  ext3_error, for example, now uses this
2409  * functionality.
2410  *
2411  */
2412
2413 void jbd2_journal_abort(journal_t *journal, int errno)
2414 {
2415         transaction_t *transaction;
2416
2417         /*
2418          * Lock the aborting procedure until everything is done, this avoid
2419          * races between filesystem's error handling flow (e.g. ext4_abort()),
2420          * ensure panic after the error info is written into journal's
2421          * superblock.
2422          */
2423         mutex_lock(&journal->j_abort_mutex);
2424         /*
2425          * ESHUTDOWN always takes precedence because a file system check
2426          * caused by any other journal abort error is not required after
2427          * a shutdown triggered.
2428          */
2429         write_lock(&journal->j_state_lock);
2430         if (journal->j_flags & JBD2_ABORT) {
2431                 int old_errno = journal->j_errno;
2432
2433                 write_unlock(&journal->j_state_lock);
2434                 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2435                         journal->j_errno = errno;
2436                         jbd2_journal_update_sb_errno(journal);
2437                 }
2438                 mutex_unlock(&journal->j_abort_mutex);
2439                 return;
2440         }
2441
2442         /*
2443          * Mark the abort as occurred and start current running transaction
2444          * to release all journaled buffer.
2445          */
2446         pr_err("Aborting journal on device %s.\n", journal->j_devname);
2447
2448         journal->j_flags |= JBD2_ABORT;
2449         journal->j_errno = errno;
2450         transaction = journal->j_running_transaction;
2451         if (transaction)
2452                 __jbd2_log_start_commit(journal, transaction->t_tid);
2453         write_unlock(&journal->j_state_lock);
2454
2455         /*
2456          * Record errno to the journal super block, so that fsck and jbd2
2457          * layer could realise that a filesystem check is needed.
2458          */
2459         jbd2_journal_update_sb_errno(journal);
2460         mutex_unlock(&journal->j_abort_mutex);
2461 }
2462
2463 /**
2464  * jbd2_journal_errno() - returns the journal's error state.
2465  * @journal: journal to examine.
2466  *
2467  * This is the errno number set with jbd2_journal_abort(), the last
2468  * time the journal was mounted - if the journal was stopped
2469  * without calling abort this will be 0.
2470  *
2471  * If the journal has been aborted on this mount time -EROFS will
2472  * be returned.
2473  */
2474 int jbd2_journal_errno(journal_t *journal)
2475 {
2476         int err;
2477
2478         read_lock(&journal->j_state_lock);
2479         if (journal->j_flags & JBD2_ABORT)
2480                 err = -EROFS;
2481         else
2482                 err = journal->j_errno;
2483         read_unlock(&journal->j_state_lock);
2484         return err;
2485 }
2486
2487 /**
2488  * jbd2_journal_clear_err() - clears the journal's error state
2489  * @journal: journal to act on.
2490  *
2491  * An error must be cleared or acked to take a FS out of readonly
2492  * mode.
2493  */
2494 int jbd2_journal_clear_err(journal_t *journal)
2495 {
2496         int err = 0;
2497
2498         write_lock(&journal->j_state_lock);
2499         if (journal->j_flags & JBD2_ABORT)
2500                 err = -EROFS;
2501         else
2502                 journal->j_errno = 0;
2503         write_unlock(&journal->j_state_lock);
2504         return err;
2505 }
2506
2507 /**
2508  * jbd2_journal_ack_err() - Ack journal err.
2509  * @journal: journal to act on.
2510  *
2511  * An error must be cleared or acked to take a FS out of readonly
2512  * mode.
2513  */
2514 void jbd2_journal_ack_err(journal_t *journal)
2515 {
2516         write_lock(&journal->j_state_lock);
2517         if (journal->j_errno)
2518                 journal->j_flags |= JBD2_ACK_ERR;
2519         write_unlock(&journal->j_state_lock);
2520 }
2521
2522 int jbd2_journal_blocks_per_page(struct inode *inode)
2523 {
2524         return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2525 }
2526
2527 /*
2528  * helper functions to deal with 32 or 64bit block numbers.
2529  */
2530 size_t journal_tag_bytes(journal_t *journal)
2531 {
2532         size_t sz;
2533
2534         if (jbd2_has_feature_csum3(journal))
2535                 return sizeof(journal_block_tag3_t);
2536
2537         sz = sizeof(journal_block_tag_t);
2538
2539         if (jbd2_has_feature_csum2(journal))
2540                 sz += sizeof(__u16);
2541
2542         if (jbd2_has_feature_64bit(journal))
2543                 return sz;
2544         else
2545                 return sz - sizeof(__u32);
2546 }
2547
2548 /*
2549  * JBD memory management
2550  *
2551  * These functions are used to allocate block-sized chunks of memory
2552  * used for making copies of buffer_head data.  Very often it will be
2553  * page-sized chunks of data, but sometimes it will be in
2554  * sub-page-size chunks.  (For example, 16k pages on Power systems
2555  * with a 4k block file system.)  For blocks smaller than a page, we
2556  * use a SLAB allocator.  There are slab caches for each block size,
2557  * which are allocated at mount time, if necessary, and we only free
2558  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2559  * this reason we don't need to a mutex to protect access to
2560  * jbd2_slab[] allocating or releasing memory; only in
2561  * jbd2_journal_create_slab().
2562  */
2563 #define JBD2_MAX_SLABS 8
2564 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2565
2566 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2567         "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2568         "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2569 };
2570
2571
2572 static void jbd2_journal_destroy_slabs(void)
2573 {
2574         int i;
2575
2576         for (i = 0; i < JBD2_MAX_SLABS; i++) {
2577                 kmem_cache_destroy(jbd2_slab[i]);
2578                 jbd2_slab[i] = NULL;
2579         }
2580 }
2581
2582 static int jbd2_journal_create_slab(size_t size)
2583 {
2584         static DEFINE_MUTEX(jbd2_slab_create_mutex);
2585         int i = order_base_2(size) - 10;
2586         size_t slab_size;
2587
2588         if (size == PAGE_SIZE)
2589                 return 0;
2590
2591         if (i >= JBD2_MAX_SLABS)
2592                 return -EINVAL;
2593
2594         if (unlikely(i < 0))
2595                 i = 0;
2596         mutex_lock(&jbd2_slab_create_mutex);
2597         if (jbd2_slab[i]) {
2598                 mutex_unlock(&jbd2_slab_create_mutex);
2599                 return 0;       /* Already created */
2600         }
2601
2602         slab_size = 1 << (i+10);
2603         jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2604                                          slab_size, 0, NULL);
2605         mutex_unlock(&jbd2_slab_create_mutex);
2606         if (!jbd2_slab[i]) {
2607                 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2608                 return -ENOMEM;
2609         }
2610         return 0;
2611 }
2612
2613 static struct kmem_cache *get_slab(size_t size)
2614 {
2615         int i = order_base_2(size) - 10;
2616
2617         BUG_ON(i >= JBD2_MAX_SLABS);
2618         if (unlikely(i < 0))
2619                 i = 0;
2620         BUG_ON(jbd2_slab[i] == NULL);
2621         return jbd2_slab[i];
2622 }
2623
2624 void *jbd2_alloc(size_t size, gfp_t flags)
2625 {
2626         void *ptr;
2627
2628         BUG_ON(size & (size-1)); /* Must be a power of 2 */
2629
2630         if (size < PAGE_SIZE)
2631                 ptr = kmem_cache_alloc(get_slab(size), flags);
2632         else
2633                 ptr = (void *)__get_free_pages(flags, get_order(size));
2634
2635         /* Check alignment; SLUB has gotten this wrong in the past,
2636          * and this can lead to user data corruption! */
2637         BUG_ON(((unsigned long) ptr) & (size-1));
2638
2639         return ptr;
2640 }
2641
2642 void jbd2_free(void *ptr, size_t size)
2643 {
2644         if (size < PAGE_SIZE)
2645                 kmem_cache_free(get_slab(size), ptr);
2646         else
2647                 free_pages((unsigned long)ptr, get_order(size));
2648 };
2649
2650 /*
2651  * Journal_head storage management
2652  */
2653 static struct kmem_cache *jbd2_journal_head_cache;
2654 #ifdef CONFIG_JBD2_DEBUG
2655 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2656 #endif
2657
2658 static int __init jbd2_journal_init_journal_head_cache(void)
2659 {
2660         J_ASSERT(!jbd2_journal_head_cache);
2661         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2662                                 sizeof(struct journal_head),
2663                                 0,              /* offset */
2664                                 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2665                                 NULL);          /* ctor */
2666         if (!jbd2_journal_head_cache) {
2667                 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2668                 return -ENOMEM;
2669         }
2670         return 0;
2671 }
2672
2673 static void jbd2_journal_destroy_journal_head_cache(void)
2674 {
2675         kmem_cache_destroy(jbd2_journal_head_cache);
2676         jbd2_journal_head_cache = NULL;
2677 }
2678
2679 /*
2680  * journal_head splicing and dicing
2681  */
2682 static struct journal_head *journal_alloc_journal_head(void)
2683 {
2684         struct journal_head *ret;
2685
2686 #ifdef CONFIG_JBD2_DEBUG
2687         atomic_inc(&nr_journal_heads);
2688 #endif
2689         ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2690         if (!ret) {
2691                 jbd_debug(1, "out of memory for journal_head\n");
2692                 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2693                 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2694                                 GFP_NOFS | __GFP_NOFAIL);
2695         }
2696         if (ret)
2697                 spin_lock_init(&ret->b_state_lock);
2698         return ret;
2699 }
2700
2701 static void journal_free_journal_head(struct journal_head *jh)
2702 {
2703 #ifdef CONFIG_JBD2_DEBUG
2704         atomic_dec(&nr_journal_heads);
2705         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2706 #endif
2707         kmem_cache_free(jbd2_journal_head_cache, jh);
2708 }
2709
2710 /*
2711  * A journal_head is attached to a buffer_head whenever JBD has an
2712  * interest in the buffer.
2713  *
2714  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2715  * is set.  This bit is tested in core kernel code where we need to take
2716  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2717  * there.
2718  *
2719  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2720  *
2721  * When a buffer has its BH_JBD bit set it is immune from being released by
2722  * core kernel code, mainly via ->b_count.
2723  *
2724  * A journal_head is detached from its buffer_head when the journal_head's
2725  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2726  * transaction (b_cp_transaction) hold their references to b_jcount.
2727  *
2728  * Various places in the kernel want to attach a journal_head to a buffer_head
2729  * _before_ attaching the journal_head to a transaction.  To protect the
2730  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2731  * journal_head's b_jcount refcount by one.  The caller must call
2732  * jbd2_journal_put_journal_head() to undo this.
2733  *
2734  * So the typical usage would be:
2735  *
2736  *      (Attach a journal_head if needed.  Increments b_jcount)
2737  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2738  *      ...
2739  *      (Get another reference for transaction)
2740  *      jbd2_journal_grab_journal_head(bh);
2741  *      jh->b_transaction = xxx;
2742  *      (Put original reference)
2743  *      jbd2_journal_put_journal_head(jh);
2744  */
2745
2746 /*
2747  * Give a buffer_head a journal_head.
2748  *
2749  * May sleep.
2750  */
2751 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2752 {
2753         struct journal_head *jh;
2754         struct journal_head *new_jh = NULL;
2755
2756 repeat:
2757         if (!buffer_jbd(bh))
2758                 new_jh = journal_alloc_journal_head();
2759
2760         jbd_lock_bh_journal_head(bh);
2761         if (buffer_jbd(bh)) {
2762                 jh = bh2jh(bh);
2763         } else {
2764                 J_ASSERT_BH(bh,
2765                         (atomic_read(&bh->b_count) > 0) ||
2766                         (bh->b_page && bh->b_page->mapping));
2767
2768                 if (!new_jh) {
2769                         jbd_unlock_bh_journal_head(bh);
2770                         goto repeat;
2771                 }
2772
2773                 jh = new_jh;
2774                 new_jh = NULL;          /* We consumed it */
2775                 set_buffer_jbd(bh);
2776                 bh->b_private = jh;
2777                 jh->b_bh = bh;
2778                 get_bh(bh);
2779                 BUFFER_TRACE(bh, "added journal_head");
2780         }
2781         jh->b_jcount++;
2782         jbd_unlock_bh_journal_head(bh);
2783         if (new_jh)
2784                 journal_free_journal_head(new_jh);
2785         return bh->b_private;
2786 }
2787
2788 /*
2789  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2790  * having a journal_head, return NULL
2791  */
2792 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2793 {
2794         struct journal_head *jh = NULL;
2795
2796         jbd_lock_bh_journal_head(bh);
2797         if (buffer_jbd(bh)) {
2798                 jh = bh2jh(bh);
2799                 jh->b_jcount++;
2800         }
2801         jbd_unlock_bh_journal_head(bh);
2802         return jh;
2803 }
2804 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2805
2806 static void __journal_remove_journal_head(struct buffer_head *bh)
2807 {
2808         struct journal_head *jh = bh2jh(bh);
2809
2810         J_ASSERT_JH(jh, jh->b_transaction == NULL);
2811         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2812         J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2813         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2814         J_ASSERT_BH(bh, buffer_jbd(bh));
2815         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2816         BUFFER_TRACE(bh, "remove journal_head");
2817
2818         /* Unlink before dropping the lock */
2819         bh->b_private = NULL;
2820         jh->b_bh = NULL;        /* debug, really */
2821         clear_buffer_jbd(bh);
2822 }
2823
2824 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2825 {
2826         if (jh->b_frozen_data) {
2827                 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2828                 jbd2_free(jh->b_frozen_data, b_size);
2829         }
2830         if (jh->b_committed_data) {
2831                 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2832                 jbd2_free(jh->b_committed_data, b_size);
2833         }
2834         journal_free_journal_head(jh);
2835 }
2836
2837 /*
2838  * Drop a reference on the passed journal_head.  If it fell to zero then
2839  * release the journal_head from the buffer_head.
2840  */
2841 void jbd2_journal_put_journal_head(struct journal_head *jh)
2842 {
2843         struct buffer_head *bh = jh2bh(jh);
2844
2845         jbd_lock_bh_journal_head(bh);
2846         J_ASSERT_JH(jh, jh->b_jcount > 0);
2847         --jh->b_jcount;
2848         if (!jh->b_jcount) {
2849                 __journal_remove_journal_head(bh);
2850                 jbd_unlock_bh_journal_head(bh);
2851                 journal_release_journal_head(jh, bh->b_size);
2852                 __brelse(bh);
2853         } else {
2854                 jbd_unlock_bh_journal_head(bh);
2855         }
2856 }
2857 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
2858
2859 /*
2860  * Initialize jbd inode head
2861  */
2862 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2863 {
2864         jinode->i_transaction = NULL;
2865         jinode->i_next_transaction = NULL;
2866         jinode->i_vfs_inode = inode;
2867         jinode->i_flags = 0;
2868         jinode->i_dirty_start = 0;
2869         jinode->i_dirty_end = 0;
2870         INIT_LIST_HEAD(&jinode->i_list);
2871 }
2872
2873 /*
2874  * Function to be called before we start removing inode from memory (i.e.,
2875  * clear_inode() is a fine place to be called from). It removes inode from
2876  * transaction's lists.
2877  */
2878 void jbd2_journal_release_jbd_inode(journal_t *journal,
2879                                     struct jbd2_inode *jinode)
2880 {
2881         if (!journal)
2882                 return;
2883 restart:
2884         spin_lock(&journal->j_list_lock);
2885         /* Is commit writing out inode - we have to wait */
2886         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2887                 wait_queue_head_t *wq;
2888                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2889                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2890                 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2891                 spin_unlock(&journal->j_list_lock);
2892                 schedule();
2893                 finish_wait(wq, &wait.wq_entry);
2894                 goto restart;
2895         }
2896
2897         if (jinode->i_transaction) {
2898                 list_del(&jinode->i_list);
2899                 jinode->i_transaction = NULL;
2900         }
2901         spin_unlock(&journal->j_list_lock);
2902 }
2903
2904
2905 #ifdef CONFIG_PROC_FS
2906
2907 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2908
2909 static void __init jbd2_create_jbd_stats_proc_entry(void)
2910 {
2911         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2912 }
2913
2914 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2915 {
2916         if (proc_jbd2_stats)
2917                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2918 }
2919
2920 #else
2921
2922 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2923 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2924
2925 #endif
2926
2927 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2928
2929 static int __init jbd2_journal_init_inode_cache(void)
2930 {
2931         J_ASSERT(!jbd2_inode_cache);
2932         jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2933         if (!jbd2_inode_cache) {
2934                 pr_emerg("JBD2: failed to create inode cache\n");
2935                 return -ENOMEM;
2936         }
2937         return 0;
2938 }
2939
2940 static int __init jbd2_journal_init_handle_cache(void)
2941 {
2942         J_ASSERT(!jbd2_handle_cache);
2943         jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2944         if (!jbd2_handle_cache) {
2945                 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2946                 return -ENOMEM;
2947         }
2948         return 0;
2949 }
2950
2951 static void jbd2_journal_destroy_inode_cache(void)
2952 {
2953         kmem_cache_destroy(jbd2_inode_cache);
2954         jbd2_inode_cache = NULL;
2955 }
2956
2957 static void jbd2_journal_destroy_handle_cache(void)
2958 {
2959         kmem_cache_destroy(jbd2_handle_cache);
2960         jbd2_handle_cache = NULL;
2961 }
2962
2963 /*
2964  * Module startup and shutdown
2965  */
2966
2967 static int __init journal_init_caches(void)
2968 {
2969         int ret;
2970
2971         ret = jbd2_journal_init_revoke_record_cache();
2972         if (ret == 0)
2973                 ret = jbd2_journal_init_revoke_table_cache();
2974         if (ret == 0)
2975                 ret = jbd2_journal_init_journal_head_cache();
2976         if (ret == 0)
2977                 ret = jbd2_journal_init_handle_cache();
2978         if (ret == 0)
2979                 ret = jbd2_journal_init_inode_cache();
2980         if (ret == 0)
2981                 ret = jbd2_journal_init_transaction_cache();
2982         return ret;
2983 }
2984
2985 static void jbd2_journal_destroy_caches(void)
2986 {
2987         jbd2_journal_destroy_revoke_record_cache();
2988         jbd2_journal_destroy_revoke_table_cache();
2989         jbd2_journal_destroy_journal_head_cache();
2990         jbd2_journal_destroy_handle_cache();
2991         jbd2_journal_destroy_inode_cache();
2992         jbd2_journal_destroy_transaction_cache();
2993         jbd2_journal_destroy_slabs();
2994 }
2995
2996 static int __init journal_init(void)
2997 {
2998         int ret;
2999
3000         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3001
3002         ret = journal_init_caches();
3003         if (ret == 0) {
3004                 jbd2_create_jbd_stats_proc_entry();
3005         } else {
3006                 jbd2_journal_destroy_caches();
3007         }
3008         return ret;
3009 }
3010
3011 static void __exit journal_exit(void)
3012 {
3013 #ifdef CONFIG_JBD2_DEBUG
3014         int n = atomic_read(&nr_journal_heads);
3015         if (n)
3016                 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3017 #endif
3018         jbd2_remove_jbd_stats_proc_entry();
3019         jbd2_journal_destroy_caches();
3020 }
3021
3022 MODULE_LICENSE("GPL");
3023 module_init(journal_init);
3024 module_exit(journal_exit);
3025