Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
journal.c
Go to the documentation of this file.
1 /*
2  * linux/fs/jbd2/journal.c
3  *
4  * Written by Stephen C. Tweedie <[email protected]>, 1998
5  *
6  * Copyright 1998 Red Hat corp --- All Rights Reserved
7  *
8  * This file is part of the Linux kernel and is made available under
9  * the terms of the GNU General Public License, version 2, or at your
10  * option, any later version, incorporated herein by reference.
11  *
12  * Generic filesystem journal-writing code; part of the ext2fs
13  * journaling system.
14  *
15  * This file manages journals: areas of disk reserved for logging
16  * transactional updates. This includes the kernel journaling thread
17  * which is responsible for scheduling updates to the log.
18  *
19  * We do not actually manage the physical storage of the journal in this
20  * file: that is left to a per-journal policy function, which allows us
21  * to store the journal within a filesystem-specified area for ext2
22  * journaling (ext2 can use a reserved inode for storing the log).
23  */
24 
25 #include <linux/module.h>
26 #include <linux/time.h>
27 #include <linux/fs.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/mm.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39 #include <linux/seq_file.h>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42 #include <linux/log2.h>
43 #include <linux/vmalloc.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bitops.h>
46 #include <linux/ratelimit.h>
47 
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/jbd2.h>
50 
51 #include <asm/uaccess.h>
52 #include <asm/page.h>
53 
65 #if 0
66 EXPORT_SYMBOL(journal_sync_buffer);
67 #endif
69 EXPORT_SYMBOL(jbd2_journal_revoke);
70 
96 
97 static void __journal_abort_soft (journal_t *journal, int errno);
98 static int jbd2_journal_create_slab(size_t slab_size);
99 
100 /* Checksumming functions */
102 {
104  return 1;
105 
106  return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
107 }
108 
109 static __u32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
110 {
111  __u32 csum, old_csum;
112 
113  old_csum = sb->s_checksum;
114  sb->s_checksum = 0;
115  csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
116  sb->s_checksum = old_csum;
117 
118  return cpu_to_be32(csum);
119 }
120 
122 {
124  return 1;
125 
126  return sb->s_checksum == jbd2_superblock_csum(j, sb);
127 }
128 
130 {
132  return;
133 
134  sb->s_checksum = jbd2_superblock_csum(j, sb);
135 }
136 
137 /*
138  * Helper function used to manage commit timeouts
139  */
140 
141 static void commit_timeout(unsigned long __data)
142 {
143  struct task_struct * p = (struct task_struct *) __data;
144 
145  wake_up_process(p);
146 }
147 
148 /*
149  * kjournald2: The main thread function used to manage a logging device
150  * journal.
151  *
152  * This kernel thread is responsible for two things:
153  *
154  * 1) COMMIT: Every so often we need to commit the current state of the
155  * filesystem to disk. The journal thread is responsible for writing
156  * all of the metadata buffers to disk.
157  *
158  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
159  * of the data in that part of the log has been rewritten elsewhere on
160  * the disk. Flushing these old buffers to reclaim space in the log is
161  * known as checkpointing, and this thread is responsible for that job.
162  */
163 
164 static int kjournald2(void *arg)
165 {
166  journal_t *journal = arg;
168 
169  /*
170  * Set up an interval timer which can be used to trigger a commit wakeup
171  * after the commit interval expires
172  */
173  setup_timer(&journal->j_commit_timer, commit_timeout,
174  (unsigned long)current);
175 
176  set_freezable();
177 
178  /* Record that the journal thread is running */
179  journal->j_task = current;
180  wake_up(&journal->j_wait_done_commit);
181 
182  /*
183  * And now, wait forever for commit wakeup events.
184  */
185  write_lock(&journal->j_state_lock);
186 
187 loop:
188  if (journal->j_flags & JBD2_UNMOUNT)
189  goto end_loop;
190 
191  jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
192  journal->j_commit_sequence, journal->j_commit_request);
193 
194  if (journal->j_commit_sequence != journal->j_commit_request) {
195  jbd_debug(1, "OK, requests differ\n");
196  write_unlock(&journal->j_state_lock);
197  del_timer_sync(&journal->j_commit_timer);
199  write_lock(&journal->j_state_lock);
200  goto loop;
201  }
202 
203  wake_up(&journal->j_wait_done_commit);
204  if (freezing(current)) {
205  /*
206  * The simpler the better. Flushing journal isn't a
207  * good idea, because that depends on threads that may
208  * be already stopped.
209  */
210  jbd_debug(1, "Now suspending kjournald2\n");
211  write_unlock(&journal->j_state_lock);
212  try_to_freeze();
213  write_lock(&journal->j_state_lock);
214  } else {
215  /*
216  * We assume on resume that commits are already there,
217  * so we don't sleep
218  */
219  DEFINE_WAIT(wait);
220  int should_sleep = 1;
221 
222  prepare_to_wait(&journal->j_wait_commit, &wait,
224  if (journal->j_commit_sequence != journal->j_commit_request)
225  should_sleep = 0;
226  transaction = journal->j_running_transaction;
227  if (transaction && time_after_eq(jiffies,
228  transaction->t_expires))
229  should_sleep = 0;
230  if (journal->j_flags & JBD2_UNMOUNT)
231  should_sleep = 0;
232  if (should_sleep) {
233  write_unlock(&journal->j_state_lock);
234  schedule();
235  write_lock(&journal->j_state_lock);
236  }
237  finish_wait(&journal->j_wait_commit, &wait);
238  }
239 
240  jbd_debug(1, "kjournald2 wakes\n");
241 
242  /*
243  * Were we woken up by a commit wakeup event?
244  */
245  transaction = journal->j_running_transaction;
246  if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
247  journal->j_commit_request = transaction->t_tid;
248  jbd_debug(1, "woke because of timeout\n");
249  }
250  goto loop;
251 
252 end_loop:
253  write_unlock(&journal->j_state_lock);
254  del_timer_sync(&journal->j_commit_timer);
255  journal->j_task = NULL;
256  wake_up(&journal->j_wait_done_commit);
257  jbd_debug(1, "Journal thread exiting.\n");
258  return 0;
259 }
260 
261 static int jbd2_journal_start_thread(journal_t *journal)
262 {
263  struct task_struct *t;
264 
265  t = kthread_run(kjournald2, journal, "jbd2/%s",
266  journal->j_devname);
267  if (IS_ERR(t))
268  return PTR_ERR(t);
269 
270  wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
271  return 0;
272 }
273 
274 static void journal_kill_thread(journal_t *journal)
275 {
276  write_lock(&journal->j_state_lock);
277  journal->j_flags |= JBD2_UNMOUNT;
278 
279  while (journal->j_task) {
280  wake_up(&journal->j_wait_commit);
281  write_unlock(&journal->j_state_lock);
282  wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
283  write_lock(&journal->j_state_lock);
284  }
285  write_unlock(&journal->j_state_lock);
286 }
287 
288 /*
289  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
290  *
291  * Writes a metadata buffer to a given disk block. The actual IO is not
292  * performed but a new buffer_head is constructed which labels the data
293  * to be written with the correct destination disk block.
294  *
295  * Any magic-number escaping which needs to be done will cause a
296  * copy-out here. If the buffer happens to start with the
297  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
298  * magic number is only written to the log for descripter blocks. In
299  * this case, we copy the data and replace the first word with 0, and we
300  * return a result code which indicates that this buffer needs to be
301  * marked as an escaped buffer in the corresponding log descriptor
302  * block. The missing word can then be restored when the block is read
303  * during recovery.
304  *
305  * If the source buffer has already been modified by a new transaction
306  * since we took the last commit snapshot, we use the frozen copy of
307  * that data for IO. If we end up using the existing buffer_head's data
308  * for the write, then we *have* to lock the buffer to prevent anyone
309  * else from using and possibly modifying it while the IO is in
310  * progress.
311  *
312  * The function returns a pointer to the buffer_heads to be used for IO.
313  *
314  * We assume that the journal has already been locked in this function.
315  *
316  * Return value:
317  * <0: Error
318  * >=0: Finished OK
319  *
320  * On success:
321  * Bit 0 set == escape performed on the data
322  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
323  */
324 
326  struct journal_head *jh_in,
327  struct journal_head **jh_out,
328  unsigned long long blocknr)
329 {
330  int need_copy_out = 0;
331  int done_copy_out = 0;
332  int do_escape = 0;
333  char *mapped_data;
334  struct buffer_head *new_bh;
335  struct journal_head *new_jh;
336  struct page *new_page;
337  unsigned int new_offset;
338  struct buffer_head *bh_in = jh2bh(jh_in);
339  journal_t *journal = transaction->t_journal;
340 
341  /*
342  * The buffer really shouldn't be locked: only the current committing
343  * transaction is allowed to write it, so nobody else is allowed
344  * to do any IO.
345  *
346  * akpm: except if we're journalling data, and write() output is
347  * also part of a shared mapping, and another thread has
348  * decided to launch a writepage() against this buffer.
349  */
350  J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
351 
352 retry_alloc:
353  new_bh = alloc_buffer_head(GFP_NOFS);
354  if (!new_bh) {
355  /*
356  * Failure is not an option, but __GFP_NOFAIL is going
357  * away; so we retry ourselves here.
358  */
360  goto retry_alloc;
361  }
362 
363  /* keep subsequent assertions sane */
364  new_bh->b_state = 0;
365  init_buffer(new_bh, NULL, NULL);
366  atomic_set(&new_bh->b_count, 1);
367  new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
368 
369  /*
370  * If a new transaction has already done a buffer copy-out, then
371  * we use that version of the data for the commit.
372  */
373  jbd_lock_bh_state(bh_in);
374 repeat:
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)) ==
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  jbd_unlock_bh_state(bh_in);
412  tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
413  if (!tmp) {
415  return -ENOMEM;
416  }
417  jbd_lock_bh_state(bh_in);
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, jh2bh(jh_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_jh->b_transaction = NULL;
452  new_bh->b_size = jh2bh(jh_in)->b_size;
453  new_bh->b_bdev = transaction->t_journal->j_dev;
454  new_bh->b_blocknr = blocknr;
455  set_buffer_mapped(new_bh);
456  set_buffer_dirty(new_bh);
457 
458  *jh_out = new_jh;
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  jbd_unlock_bh_state(bh_in);
470 
471  JBUFFER_TRACE(new_jh, "file as BJ_IO");
472  jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
473 
474  return do_escape | (done_copy_out << 1);
475 }
476 
477 /*
478  * Allocation code for the journal file. Manage the space left in the
479  * journal, so that we can begin checkpointing when appropriate.
480  */
481 
482 /*
483  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
484  *
485  * Called with the journal already locked.
486  *
487  * Called under j_state_lock
488  */
489 
490 int __jbd2_log_space_left(journal_t *journal)
491 {
492  int left = journal->j_free;
493 
494  /* assert_spin_locked(&journal->j_state_lock); */
495 
496  /*
497  * Be pessimistic here about the number of those free blocks which
498  * might be required for log descriptor control blocks.
499  */
500 
501 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
502 
503  left -= MIN_LOG_RESERVED_BLOCKS;
504 
505  if (left <= 0)
506  return 0;
507  left -= (left >> 3);
508  return left;
509 }
510 
511 /*
512  * Called with j_state_lock locked for writing.
513  * Returns true if a transaction commit was started.
514  */
515 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
516 {
517  /*
518  * The only transaction we can possibly wait upon is the
519  * currently running transaction (if it exists). Otherwise,
520  * the target tid must be an old one.
521  */
522  if (journal->j_running_transaction &&
523  journal->j_running_transaction->t_tid == target) {
524  /*
525  * We want a new commit: OK, mark the request and wakeup the
526  * commit thread. We do _not_ do the commit ourselves.
527  */
528 
529  journal->j_commit_request = target;
530  jbd_debug(1, "JBD2: requesting commit %d/%d\n",
531  journal->j_commit_request,
532  journal->j_commit_sequence);
533  wake_up(&journal->j_wait_commit);
534  return 1;
535  } else if (!tid_geq(journal->j_commit_request, target))
536  /* This should never happen, but if it does, preserve
537  the evidence before kjournald goes into a loop and
538  increments j_commit_sequence beyond all recognition. */
539  WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
540  journal->j_commit_request,
541  journal->j_commit_sequence,
542  target, journal->j_running_transaction ?
543  journal->j_running_transaction->t_tid : 0);
544  return 0;
545 }
546 
547 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
548 {
549  int ret;
550 
551  write_lock(&journal->j_state_lock);
552  ret = __jbd2_log_start_commit(journal, tid);
553  write_unlock(&journal->j_state_lock);
554  return ret;
555 }
556 
557 /*
558  * Force and wait upon a commit if the calling process is not within
559  * transaction. This is used for forcing out undo-protected data which contains
560  * bitmaps, when the fs is running out of space.
561  *
562  * We can only force the running transaction if we don't have an active handle;
563  * otherwise, we will deadlock.
564  *
565  * Returns true if a transaction was started.
566  */
567 int jbd2_journal_force_commit_nested(journal_t *journal)
568 {
569  transaction_t *transaction = NULL;
570  tid_t tid;
571  int need_to_start = 0;
572 
573  read_lock(&journal->j_state_lock);
574  if (journal->j_running_transaction && !current->journal_info) {
575  transaction = journal->j_running_transaction;
576  if (!tid_geq(journal->j_commit_request, transaction->t_tid))
577  need_to_start = 1;
578  } else if (journal->j_committing_transaction)
579  transaction = journal->j_committing_transaction;
580 
581  if (!transaction) {
582  read_unlock(&journal->j_state_lock);
583  return 0; /* Nothing to retry */
584  }
585 
586  tid = transaction->t_tid;
587  read_unlock(&journal->j_state_lock);
588  if (need_to_start)
589  jbd2_log_start_commit(journal, tid);
590  jbd2_log_wait_commit(journal, tid);
591  return 1;
592 }
593 
594 /*
595  * Start a commit of the current running transaction (if any). Returns true
596  * if a transaction is going to be committed (or is currently already
597  * committing), and fills its tid in at *ptid
598  */
599 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
600 {
601  int ret = 0;
602 
603  write_lock(&journal->j_state_lock);
604  if (journal->j_running_transaction) {
605  tid_t tid = journal->j_running_transaction->t_tid;
606 
607  __jbd2_log_start_commit(journal, tid);
608  /* There's a running transaction and we've just made sure
609  * it's commit has been scheduled. */
610  if (ptid)
611  *ptid = tid;
612  ret = 1;
613  } else if (journal->j_committing_transaction) {
614  /*
615  * If commit has been started, then we have to wait for
616  * completion of that transaction.
617  */
618  if (ptid)
619  *ptid = journal->j_committing_transaction->t_tid;
620  ret = 1;
621  }
622  write_unlock(&journal->j_state_lock);
623  return ret;
624 }
625 
626 /*
627  * Return 1 if a given transaction has not yet sent barrier request
628  * connected with a transaction commit. If 0 is returned, transaction
629  * may or may not have sent the barrier. Used to avoid sending barrier
630  * twice in common cases.
631  */
632 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
633 {
634  int ret = 0;
635  transaction_t *commit_trans;
636 
637  if (!(journal->j_flags & JBD2_BARRIER))
638  return 0;
639  read_lock(&journal->j_state_lock);
640  /* Transaction already committed? */
641  if (tid_geq(journal->j_commit_sequence, tid))
642  goto out;
643  commit_trans = journal->j_committing_transaction;
644  if (!commit_trans || commit_trans->t_tid != tid) {
645  ret = 1;
646  goto out;
647  }
648  /*
649  * Transaction is being committed and we already proceeded to
650  * submitting a flush to fs partition?
651  */
652  if (journal->j_fs_dev != journal->j_dev) {
653  if (!commit_trans->t_need_data_flush ||
654  commit_trans->t_state >= T_COMMIT_DFLUSH)
655  goto out;
656  } else {
657  if (commit_trans->t_state >= T_COMMIT_JFLUSH)
658  goto out;
659  }
660  ret = 1;
661 out:
662  read_unlock(&journal->j_state_lock);
663  return ret;
664 }
666 
667 /*
668  * Wait for a specified commit to complete.
669  * The caller may not hold the journal lock.
670  */
671 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
672 {
673  int err = 0;
674 
675  read_lock(&journal->j_state_lock);
676 #ifdef CONFIG_JBD2_DEBUG
677  if (!tid_geq(journal->j_commit_request, tid)) {
679  "%s: error: j_commit_request=%d, tid=%d\n",
680  __func__, journal->j_commit_request, tid);
681  }
682 #endif
683  while (tid_gt(tid, journal->j_commit_sequence)) {
684  jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
685  tid, journal->j_commit_sequence);
686  wake_up(&journal->j_wait_commit);
687  read_unlock(&journal->j_state_lock);
688  wait_event(journal->j_wait_done_commit,
689  !tid_gt(tid, journal->j_commit_sequence));
690  read_lock(&journal->j_state_lock);
691  }
692  read_unlock(&journal->j_state_lock);
693 
694  if (unlikely(is_journal_aborted(journal))) {
695  printk(KERN_EMERG "journal commit I/O error\n");
696  err = -EIO;
697  }
698  return err;
699 }
700 
701 /*
702  * Log buffer allocation routines:
703  */
704 
705 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
706 {
707  unsigned long blocknr;
708 
709  write_lock(&journal->j_state_lock);
710  J_ASSERT(journal->j_free > 1);
711 
712  blocknr = journal->j_head;
713  journal->j_head++;
714  journal->j_free--;
715  if (journal->j_head == journal->j_last)
716  journal->j_head = journal->j_first;
717  write_unlock(&journal->j_state_lock);
718  return jbd2_journal_bmap(journal, blocknr, retp);
719 }
720 
721 /*
722  * Conversion of logical to physical block numbers for the journal
723  *
724  * On external journals the journal blocks are identity-mapped, so
725  * this is a no-op. If needed, we can use j_blk_offset - everything is
726  * ready.
727  */
728 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
729  unsigned long long *retp)
730 {
731  int err = 0;
732  unsigned long long ret;
733 
734  if (journal->j_inode) {
735  ret = bmap(journal->j_inode, blocknr);
736  if (ret)
737  *retp = ret;
738  else {
739  printk(KERN_ALERT "%s: journal block not found "
740  "at offset %lu on %s\n",
741  __func__, blocknr, journal->j_devname);
742  err = -EIO;
743  __journal_abort_soft(journal, err);
744  }
745  } else {
746  *retp = blocknr; /* +journal->j_blk_offset */
747  }
748  return err;
749 }
750 
751 /*
752  * We play buffer_head aliasing tricks to write data/metadata blocks to
753  * the journal without copying their contents, but for journal
754  * descriptor blocks we do need to generate bona fide buffers.
755  *
756  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
757  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
758  * But we don't bother doing that, so there will be coherency problems with
759  * mmaps of blockdevs which hold live JBD-controlled filesystems.
760  */
762 {
763  struct buffer_head *bh;
764  unsigned long long blocknr;
765  int err;
766 
767  err = jbd2_journal_next_log_block(journal, &blocknr);
768 
769  if (err)
770  return NULL;
771 
772  bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
773  if (!bh)
774  return NULL;
775  lock_buffer(bh);
776  memset(bh->b_data, 0, journal->j_blocksize);
777  set_buffer_uptodate(bh);
778  unlock_buffer(bh);
779  BUFFER_TRACE(bh, "return this buffer");
781 }
782 
783 /*
784  * Return tid of the oldest transaction in the journal and block in the journal
785  * where the transaction starts.
786  *
787  * If the journal is now empty, return which will be the next transaction ID
788  * we will write and where will that transaction start.
789  *
790  * The return value is 0 if journal tail cannot be pushed any further, 1 if
791  * it can.
792  */
793 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
794  unsigned long *block)
795 {
797  int ret;
798 
799  read_lock(&journal->j_state_lock);
800  spin_lock(&journal->j_list_lock);
801  transaction = journal->j_checkpoint_transactions;
802  if (transaction) {
803  *tid = transaction->t_tid;
804  *block = transaction->t_log_start;
805  } else if ((transaction = journal->j_committing_transaction) != NULL) {
806  *tid = transaction->t_tid;
807  *block = transaction->t_log_start;
808  } else if ((transaction = journal->j_running_transaction) != NULL) {
809  *tid = transaction->t_tid;
810  *block = journal->j_head;
811  } else {
812  *tid = journal->j_transaction_sequence;
813  *block = journal->j_head;
814  }
815  ret = tid_gt(*tid, journal->j_tail_sequence);
816  spin_unlock(&journal->j_list_lock);
817  read_unlock(&journal->j_state_lock);
818 
819  return ret;
820 }
821 
822 /*
823  * Update information in journal structure and in on disk journal superblock
824  * about log tail. This function does not check whether information passed in
825  * really pushes log tail further. It's responsibility of the caller to make
826  * sure provided log tail information is valid (e.g. by holding
827  * j_checkpoint_mutex all the time between computing log tail and calling this
828  * function as is the case with jbd2_cleanup_journal_tail()).
829  *
830  * Requires j_checkpoint_mutex
831  */
832 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
833 {
834  unsigned long freed;
835 
836  BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
837 
838  /*
839  * We cannot afford for write to remain in drive's caches since as
840  * soon as we update j_tail, next transaction can start reusing journal
841  * space and if we lose sb update during power failure we'd replay
842  * old transaction with possibly newly overwritten data.
843  */
844  jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
845  write_lock(&journal->j_state_lock);
846  freed = block - journal->j_tail;
847  if (block < journal->j_tail)
848  freed += journal->j_last - journal->j_first;
849 
850  trace_jbd2_update_log_tail(journal, tid, block, freed);
851  jbd_debug(1,
852  "Cleaning journal tail from %d to %d (offset %lu), "
853  "freeing %lu\n",
854  journal->j_tail_sequence, tid, block, freed);
855 
856  journal->j_free += freed;
857  journal->j_tail_sequence = tid;
858  journal->j_tail = block;
859  write_unlock(&journal->j_state_lock);
860 }
861 
862 /*
863  * This is a variaon of __jbd2_update_log_tail which checks for validity of
864  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
865  * with other threads updating log tail.
866  */
867 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
868 {
869  mutex_lock(&journal->j_checkpoint_mutex);
870  if (tid_gt(tid, journal->j_tail_sequence))
871  __jbd2_update_log_tail(journal, tid, block);
872  mutex_unlock(&journal->j_checkpoint_mutex);
873 }
874 
876  journal_t *journal;
877  struct transaction_stats_s *stats;
878  int start;
879  int max;
880 };
881 
882 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
883 {
884  return *pos ? NULL : SEQ_START_TOKEN;
885 }
886 
887 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
888 {
889  return NULL;
890 }
891 
892 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
893 {
894  struct jbd2_stats_proc_session *s = seq->private;
895 
896  if (v != SEQ_START_TOKEN)
897  return 0;
898  seq_printf(seq, "%lu transaction, each up to %u blocks\n",
899  s->stats->ts_tid,
900  s->journal->j_max_transaction_buffers);
901  if (s->stats->ts_tid == 0)
902  return 0;
903  seq_printf(seq, "average: \n %ums waiting for transaction\n",
904  jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
905  seq_printf(seq, " %ums running transaction\n",
906  jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
907  seq_printf(seq, " %ums transaction was being locked\n",
908  jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
909  seq_printf(seq, " %ums flushing data (in ordered mode)\n",
910  jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
911  seq_printf(seq, " %ums logging transaction\n",
912  jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
913  seq_printf(seq, " %lluus average transaction commit time\n",
914  div_u64(s->journal->j_average_commit_time, 1000));
915  seq_printf(seq, " %lu handles per transaction\n",
916  s->stats->run.rs_handle_count / s->stats->ts_tid);
917  seq_printf(seq, " %lu blocks per transaction\n",
918  s->stats->run.rs_blocks / s->stats->ts_tid);
919  seq_printf(seq, " %lu logged blocks per transaction\n",
920  s->stats->run.rs_blocks_logged / s->stats->ts_tid);
921  return 0;
922 }
923 
924 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
925 {
926 }
927 
928 static const struct seq_operations jbd2_seq_info_ops = {
929  .start = jbd2_seq_info_start,
930  .next = jbd2_seq_info_next,
931  .stop = jbd2_seq_info_stop,
932  .show = jbd2_seq_info_show,
933 };
934 
935 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
936 {
937  journal_t *journal = PDE(inode)->data;
938  struct jbd2_stats_proc_session *s;
939  int rc, size;
940 
941  s = kmalloc(sizeof(*s), GFP_KERNEL);
942  if (s == NULL)
943  return -ENOMEM;
944  size = sizeof(struct transaction_stats_s);
945  s->stats = kmalloc(size, GFP_KERNEL);
946  if (s->stats == NULL) {
947  kfree(s);
948  return -ENOMEM;
949  }
950  spin_lock(&journal->j_history_lock);
951  memcpy(s->stats, &journal->j_stats, size);
952  s->journal = journal;
953  spin_unlock(&journal->j_history_lock);
954 
955  rc = seq_open(file, &jbd2_seq_info_ops);
956  if (rc == 0) {
957  struct seq_file *m = file->private_data;
958  m->private = s;
959  } else {
960  kfree(s->stats);
961  kfree(s);
962  }
963  return rc;
964 
965 }
966 
967 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
968 {
969  struct seq_file *seq = file->private_data;
970  struct jbd2_stats_proc_session *s = seq->private;
971  kfree(s->stats);
972  kfree(s);
973  return seq_release(inode, file);
974 }
975 
976 static const struct file_operations jbd2_seq_info_fops = {
977  .owner = THIS_MODULE,
978  .open = jbd2_seq_info_open,
979  .read = seq_read,
980  .llseek = seq_lseek,
981  .release = jbd2_seq_info_release,
982 };
983 
984 static struct proc_dir_entry *proc_jbd2_stats;
985 
986 static void jbd2_stats_proc_init(journal_t *journal)
987 {
988  journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
989  if (journal->j_proc_entry) {
990  proc_create_data("info", S_IRUGO, journal->j_proc_entry,
991  &jbd2_seq_info_fops, journal);
992  }
993 }
994 
995 static void jbd2_stats_proc_exit(journal_t *journal)
996 {
997  remove_proc_entry("info", journal->j_proc_entry);
998  remove_proc_entry(journal->j_devname, proc_jbd2_stats);
999 }
1000 
1001 /*
1002  * Management for journal control blocks: functions to create and
1003  * destroy journal_t structures, and to initialise and read existing
1004  * journal blocks from disk. */
1005 
1006 /* First: create and setup a journal_t object in memory. We initialise
1007  * very few fields yet: that has to wait until we have created the
1008  * journal structures from from scratch, or loaded them from disk. */
1009 
1010 static journal_t * journal_init_common (void)
1011 {
1012  journal_t *journal;
1013  int err;
1014 
1015  journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1016  if (!journal)
1017  return NULL;
1018 
1019  init_waitqueue_head(&journal->j_wait_transaction_locked);
1020  init_waitqueue_head(&journal->j_wait_logspace);
1021  init_waitqueue_head(&journal->j_wait_done_commit);
1022  init_waitqueue_head(&journal->j_wait_checkpoint);
1023  init_waitqueue_head(&journal->j_wait_commit);
1024  init_waitqueue_head(&journal->j_wait_updates);
1025  mutex_init(&journal->j_barrier);
1026  mutex_init(&journal->j_checkpoint_mutex);
1027  spin_lock_init(&journal->j_revoke_lock);
1028  spin_lock_init(&journal->j_list_lock);
1029  rwlock_init(&journal->j_state_lock);
1030 
1031  journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1032  journal->j_min_batch_time = 0;
1033  journal->j_max_batch_time = 15000; /* 15ms */
1034 
1035  /* The journal is marked for error until we succeed with recovery! */
1036  journal->j_flags = JBD2_ABORT;
1037 
1038  /* Set up a default-sized revoke table for the new mount. */
1039  err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1040  if (err) {
1041  kfree(journal);
1042  return NULL;
1043  }
1044 
1045  spin_lock_init(&journal->j_history_lock);
1046 
1047  return journal;
1048 }
1049 
1050 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1051  *
1052  * Create a journal structure assigned some fixed set of disk blocks to
1053  * the journal. We don't actually touch those disk blocks yet, but we
1054  * need to set up all of the mapping information to tell the journaling
1055  * system where the journal blocks are.
1056  *
1057  */
1058 
1073 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1074  struct block_device *fs_dev,
1075  unsigned long long start, int len, int blocksize)
1076 {
1077  journal_t *journal = journal_init_common();
1078  struct buffer_head *bh;
1079  char *p;
1080  int n;
1081 
1082  if (!journal)
1083  return NULL;
1084 
1085  /* journal descriptor can store up to n blocks -bzzz */
1086  journal->j_blocksize = blocksize;
1087  journal->j_dev = bdev;
1088  journal->j_fs_dev = fs_dev;
1089  journal->j_blk_offset = start;
1090  journal->j_maxlen = len;
1091  bdevname(journal->j_dev, journal->j_devname);
1092  p = journal->j_devname;
1093  while ((p = strchr(p, '/')))
1094  *p = '!';
1095  jbd2_stats_proc_init(journal);
1096  n = journal->j_blocksize / sizeof(journal_block_tag_t);
1097  journal->j_wbufsize = n;
1098  journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1099  if (!journal->j_wbuf) {
1100  printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1101  __func__);
1102  goto out_err;
1103  }
1104 
1105  bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1106  if (!bh) {
1108  "%s: Cannot get buffer for journal superblock\n",
1109  __func__);
1110  goto out_err;
1111  }
1112  journal->j_sb_buffer = bh;
1113  journal->j_superblock = (journal_superblock_t *)bh->b_data;
1114 
1115  return journal;
1116 out_err:
1117  kfree(journal->j_wbuf);
1118  jbd2_stats_proc_exit(journal);
1119  kfree(journal);
1120  return NULL;
1121 }
1122 
1131 journal_t * jbd2_journal_init_inode (struct inode *inode)
1132 {
1133  struct buffer_head *bh;
1134  journal_t *journal = journal_init_common();
1135  char *p;
1136  int err;
1137  int n;
1138  unsigned long long blocknr;
1139 
1140  if (!journal)
1141  return NULL;
1142 
1143  journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1144  journal->j_inode = inode;
1145  bdevname(journal->j_dev, journal->j_devname);
1146  p = journal->j_devname;
1147  while ((p = strchr(p, '/')))
1148  *p = '!';
1149  p = journal->j_devname + strlen(journal->j_devname);
1150  sprintf(p, "-%lu", journal->j_inode->i_ino);
1151  jbd_debug(1,
1152  "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1153  journal, inode->i_sb->s_id, inode->i_ino,
1154  (long long) inode->i_size,
1155  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1156 
1157  journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1158  journal->j_blocksize = inode->i_sb->s_blocksize;
1159  jbd2_stats_proc_init(journal);
1160 
1161  /* journal descriptor can store up to n blocks -bzzz */
1162  n = journal->j_blocksize / sizeof(journal_block_tag_t);
1163  journal->j_wbufsize = n;
1164  journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1165  if (!journal->j_wbuf) {
1166  printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1167  __func__);
1168  goto out_err;
1169  }
1170 
1171  err = jbd2_journal_bmap(journal, 0, &blocknr);
1172  /* If that failed, give up */
1173  if (err) {
1174  printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1175  __func__);
1176  goto out_err;
1177  }
1178 
1179  bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1180  if (!bh) {
1182  "%s: Cannot get buffer for journal superblock\n",
1183  __func__);
1184  goto out_err;
1185  }
1186  journal->j_sb_buffer = bh;
1187  journal->j_superblock = (journal_superblock_t *)bh->b_data;
1188 
1189  return journal;
1190 out_err:
1191  kfree(journal->j_wbuf);
1192  jbd2_stats_proc_exit(journal);
1193  kfree(journal);
1194  return NULL;
1195 }
1196 
1197 /*
1198  * If the journal init or create aborts, we need to mark the journal
1199  * superblock as being NULL to prevent the journal destroy from writing
1200  * back a bogus superblock.
1201  */
1202 static void journal_fail_superblock (journal_t *journal)
1203 {
1204  struct buffer_head *bh = journal->j_sb_buffer;
1205  brelse(bh);
1206  journal->j_sb_buffer = NULL;
1207 }
1208 
1209 /*
1210  * Given a journal_t structure, initialise the various fields for
1211  * startup of a new journaling session. We use this both when creating
1212  * a journal, and after recovering an old journal to reset it for
1213  * subsequent use.
1214  */
1215 
1216 static int journal_reset(journal_t *journal)
1217 {
1218  journal_superblock_t *sb = journal->j_superblock;
1219  unsigned long long first, last;
1220 
1221  first = be32_to_cpu(sb->s_first);
1222  last = be32_to_cpu(sb->s_maxlen);
1223  if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1224  printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1225  first, last);
1226  journal_fail_superblock(journal);
1227  return -EINVAL;
1228  }
1229 
1230  journal->j_first = first;
1231  journal->j_last = last;
1232 
1233  journal->j_head = first;
1234  journal->j_tail = first;
1235  journal->j_free = last - first;
1236 
1237  journal->j_tail_sequence = journal->j_transaction_sequence;
1238  journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1239  journal->j_commit_request = journal->j_commit_sequence;
1240 
1241  journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1242 
1243  /*
1244  * As a special case, if the on-disk copy is already marked as needing
1245  * no recovery (s_start == 0), then we can safely defer the superblock
1246  * update until the next commit by setting JBD2_FLUSHED. This avoids
1247  * attempting a write to a potential-readonly device.
1248  */
1249  if (sb->s_start == 0) {
1250  jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1251  "(start %ld, seq %d, errno %d)\n",
1252  journal->j_tail, journal->j_tail_sequence,
1253  journal->j_errno);
1254  journal->j_flags |= JBD2_FLUSHED;
1255  } else {
1256  /* Lock here to make assertions happy... */
1257  mutex_lock(&journal->j_checkpoint_mutex);
1258  /*
1259  * Update log tail information. We use WRITE_FUA since new
1260  * transaction will start reusing journal space and so we
1261  * must make sure information about current log tail is on
1262  * disk before that.
1263  */
1265  journal->j_tail_sequence,
1266  journal->j_tail,
1267  WRITE_FUA);
1268  mutex_unlock(&journal->j_checkpoint_mutex);
1269  }
1270  return jbd2_journal_start_thread(journal);
1271 }
1272 
1273 static void jbd2_write_superblock(journal_t *journal, int write_op)
1274 {
1275  struct buffer_head *bh = journal->j_sb_buffer;
1276  int ret;
1277 
1278  trace_jbd2_write_superblock(journal, write_op);
1279  if (!(journal->j_flags & JBD2_BARRIER))
1280  write_op &= ~(REQ_FUA | REQ_FLUSH);
1281  lock_buffer(bh);
1282  if (buffer_write_io_error(bh)) {
1283  /*
1284  * Oh, dear. A previous attempt to write the journal
1285  * superblock failed. This could happen because the
1286  * USB device was yanked out. Or it could happen to
1287  * be a transient write error and maybe the block will
1288  * be remapped. Nothing we can do but to retry the
1289  * write and hope for the best.
1290  */
1291  printk(KERN_ERR "JBD2: previous I/O error detected "
1292  "for journal superblock update for %s.\n",
1293  journal->j_devname);
1294  clear_buffer_write_io_error(bh);
1295  set_buffer_uptodate(bh);
1296  }
1297  get_bh(bh);
1298  bh->b_end_io = end_buffer_write_sync;
1299  ret = submit_bh(write_op, bh);
1300  wait_on_buffer(bh);
1301  if (buffer_write_io_error(bh)) {
1302  clear_buffer_write_io_error(bh);
1303  set_buffer_uptodate(bh);
1304  ret = -EIO;
1305  }
1306  if (ret) {
1307  printk(KERN_ERR "JBD2: Error %d detected when updating "
1308  "journal superblock for %s.\n", ret,
1309  journal->j_devname);
1310  }
1311 }
1312 
1323 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1324  unsigned long tail_block, int write_op)
1325 {
1326  journal_superblock_t *sb = journal->j_superblock;
1327 
1328  BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1329  jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1330  tail_block, tail_tid);
1331 
1332  sb->s_sequence = cpu_to_be32(tail_tid);
1333  sb->s_start = cpu_to_be32(tail_block);
1334 
1335  jbd2_write_superblock(journal, write_op);
1336 
1337  /* Log is no longer empty */
1338  write_lock(&journal->j_state_lock);
1339  WARN_ON(!sb->s_sequence);
1340  journal->j_flags &= ~JBD2_FLUSHED;
1341  write_unlock(&journal->j_state_lock);
1342 }
1343 
1351 static void jbd2_mark_journal_empty(journal_t *journal)
1352 {
1353  journal_superblock_t *sb = journal->j_superblock;
1354 
1355  BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1356  read_lock(&journal->j_state_lock);
1357  /* Is it already empty? */
1358  if (sb->s_start == 0) {
1359  read_unlock(&journal->j_state_lock);
1360  return;
1361  }
1362  jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1363  journal->j_tail_sequence);
1364 
1365  sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1366  sb->s_start = cpu_to_be32(0);
1367  read_unlock(&journal->j_state_lock);
1368 
1369  jbd2_write_superblock(journal, WRITE_FUA);
1370 
1371  /* Log is no longer empty */
1372  write_lock(&journal->j_state_lock);
1373  journal->j_flags |= JBD2_FLUSHED;
1374  write_unlock(&journal->j_state_lock);
1375 }
1376 
1377 
1385 void jbd2_journal_update_sb_errno(journal_t *journal)
1386 {
1387  journal_superblock_t *sb = journal->j_superblock;
1388 
1389  read_lock(&journal->j_state_lock);
1390  jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1391  journal->j_errno);
1392  sb->s_errno = cpu_to_be32(journal->j_errno);
1393  jbd2_superblock_csum_set(journal, sb);
1394  read_unlock(&journal->j_state_lock);
1395 
1396  jbd2_write_superblock(journal, WRITE_SYNC);
1397 }
1399 
1400 /*
1401  * Read the superblock for a given journal, performing initial
1402  * validation of the format.
1403  */
1404 static int journal_get_superblock(journal_t *journal)
1405 {
1406  struct buffer_head *bh;
1408  int err = -EIO;
1409 
1410  bh = journal->j_sb_buffer;
1411 
1412  J_ASSERT(bh != NULL);
1413  if (!buffer_uptodate(bh)) {
1414  ll_rw_block(READ, 1, &bh);
1415  wait_on_buffer(bh);
1416  if (!buffer_uptodate(bh)) {
1418  "JBD2: IO error reading journal superblock\n");
1419  goto out;
1420  }
1421  }
1422 
1423  if (buffer_verified(bh))
1424  return 0;
1425 
1426  sb = journal->j_superblock;
1427 
1428  err = -EINVAL;
1429 
1431  sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1432  printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1433  goto out;
1434  }
1435 
1436  switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1437  case JBD2_SUPERBLOCK_V1:
1438  journal->j_format_version = 1;
1439  break;
1440  case JBD2_SUPERBLOCK_V2:
1441  journal->j_format_version = 2;
1442  break;
1443  default:
1444  printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1445  goto out;
1446  }
1447 
1448  if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1449  journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1450  else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1451  printk(KERN_WARNING "JBD2: journal file too short\n");
1452  goto out;
1453  }
1454 
1455  if (be32_to_cpu(sb->s_first) == 0 ||
1456  be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1458  "JBD2: Invalid start block of journal: %u\n",
1459  be32_to_cpu(sb->s_first));
1460  goto out;
1461  }
1462 
1465  /* Can't have checksum v1 and v2 on at the same time! */
1466  printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 "
1467  "at the same time!\n");
1468  goto out;
1469  }
1470 
1471  if (!jbd2_verify_csum_type(journal, sb)) {
1472  printk(KERN_ERR "JBD: Unknown checksum type\n");
1473  goto out;
1474  }
1475 
1476  /* Load the checksum driver */
1478  journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1479  if (IS_ERR(journal->j_chksum_driver)) {
1480  printk(KERN_ERR "JBD: Cannot load crc32c driver.\n");
1481  err = PTR_ERR(journal->j_chksum_driver);
1482  journal->j_chksum_driver = NULL;
1483  goto out;
1484  }
1485  }
1486 
1487  /* Check superblock checksum */
1488  if (!jbd2_superblock_csum_verify(journal, sb)) {
1489  printk(KERN_ERR "JBD: journal checksum error\n");
1490  goto out;
1491  }
1492 
1493  /* Precompute checksum seed for all metadata */
1495  journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1496  sizeof(sb->s_uuid));
1497 
1498  set_buffer_verified(bh);
1499 
1500  return 0;
1501 
1502 out:
1503  journal_fail_superblock(journal);
1504  return err;
1505 }
1506 
1507 /*
1508  * Load the on-disk journal superblock and read the key fields into the
1509  * journal_t.
1510  */
1511 
1512 static int load_superblock(journal_t *journal)
1513 {
1514  int err;
1516 
1517  err = journal_get_superblock(journal);
1518  if (err)
1519  return err;
1520 
1521  sb = journal->j_superblock;
1522 
1523  journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1524  journal->j_tail = be32_to_cpu(sb->s_start);
1525  journal->j_first = be32_to_cpu(sb->s_first);
1526  journal->j_last = be32_to_cpu(sb->s_maxlen);
1527  journal->j_errno = be32_to_cpu(sb->s_errno);
1528 
1529  return 0;
1530 }
1531 
1532 
1541 int jbd2_journal_load(journal_t *journal)
1542 {
1543  int err;
1545 
1546  err = load_superblock(journal);
1547  if (err)
1548  return err;
1549 
1550  sb = journal->j_superblock;
1551  /* If this is a V2 superblock, then we have to check the
1552  * features flags on it. */
1553 
1554  if (journal->j_format_version >= 2) {
1555  if ((sb->s_feature_ro_compat &
1557  (sb->s_feature_incompat &
1560  "JBD2: Unrecognised features on journal\n");
1561  return -EINVAL;
1562  }
1563  }
1564 
1565  /*
1566  * Create a slab for this blocksize
1567  */
1568  err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1569  if (err)
1570  return err;
1571 
1572  /* Let the recovery code check whether it needs to recover any
1573  * data from the journal. */
1574  if (jbd2_journal_recover(journal))
1575  goto recovery_error;
1576 
1577  if (journal->j_failed_commit) {
1578  printk(KERN_ERR "JBD2: journal transaction %u on %s "
1579  "is corrupt.\n", journal->j_failed_commit,
1580  journal->j_devname);
1581  return -EIO;
1582  }
1583 
1584  /* OK, we've finished with the dynamic journal bits:
1585  * reinitialise the dynamic contents of the superblock in memory
1586  * and reset them on disk. */
1587  if (journal_reset(journal))
1588  goto recovery_error;
1589 
1590  journal->j_flags &= ~JBD2_ABORT;
1591  journal->j_flags |= JBD2_LOADED;
1592  return 0;
1593 
1594 recovery_error:
1595  printk(KERN_WARNING "JBD2: recovery failed\n");
1596  return -EIO;
1597 }
1598 
1607 int jbd2_journal_destroy(journal_t *journal)
1608 {
1609  int err = 0;
1610 
1611  /* Wait for the commit thread to wake up and die. */
1612  journal_kill_thread(journal);
1613 
1614  /* Force a final log commit */
1615  if (journal->j_running_transaction)
1617 
1618  /* Force any old transactions to disk */
1619 
1620  /* Totally anal locking here... */
1621  spin_lock(&journal->j_list_lock);
1622  while (journal->j_checkpoint_transactions != NULL) {
1623  spin_unlock(&journal->j_list_lock);
1624  mutex_lock(&journal->j_checkpoint_mutex);
1625  jbd2_log_do_checkpoint(journal);
1626  mutex_unlock(&journal->j_checkpoint_mutex);
1627  spin_lock(&journal->j_list_lock);
1628  }
1629 
1630  J_ASSERT(journal->j_running_transaction == NULL);
1631  J_ASSERT(journal->j_committing_transaction == NULL);
1632  J_ASSERT(journal->j_checkpoint_transactions == NULL);
1633  spin_unlock(&journal->j_list_lock);
1634 
1635  if (journal->j_sb_buffer) {
1636  if (!is_journal_aborted(journal)) {
1637  mutex_lock(&journal->j_checkpoint_mutex);
1638  jbd2_mark_journal_empty(journal);
1639  mutex_unlock(&journal->j_checkpoint_mutex);
1640  } else
1641  err = -EIO;
1642  brelse(journal->j_sb_buffer);
1643  }
1644 
1645  if (journal->j_proc_entry)
1646  jbd2_stats_proc_exit(journal);
1647  if (journal->j_inode)
1648  iput(journal->j_inode);
1649  if (journal->j_revoke)
1650  jbd2_journal_destroy_revoke(journal);
1651  if (journal->j_chksum_driver)
1652  crypto_free_shash(journal->j_chksum_driver);
1653  kfree(journal->j_wbuf);
1654  kfree(journal);
1655 
1656  return err;
1657 }
1658 
1659 
1671 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1672  unsigned long ro, unsigned long incompat)
1673 {
1675 
1676  if (!compat && !ro && !incompat)
1677  return 1;
1678  /* Load journal superblock if it is not loaded yet. */
1679  if (journal->j_format_version == 0 &&
1680  journal_get_superblock(journal) != 0)
1681  return 0;
1682  if (journal->j_format_version == 1)
1683  return 0;
1684 
1685  sb = journal->j_superblock;
1686 
1687  if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1688  ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1689  ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1690  return 1;
1691 
1692  return 0;
1693 }
1694 
1706 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1707  unsigned long ro, unsigned long incompat)
1708 {
1709  if (!compat && !ro && !incompat)
1710  return 1;
1711 
1712  /* We can support any known requested features iff the
1713  * superblock is in version 2. Otherwise we fail to support any
1714  * extended sb features. */
1715 
1716  if (journal->j_format_version != 2)
1717  return 0;
1718 
1719  if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1720  (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1721  (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1722  return 1;
1723 
1724  return 0;
1725 }
1726 
1739 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1740  unsigned long ro, unsigned long incompat)
1741 {
1742 #define INCOMPAT_FEATURE_ON(f) \
1743  ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1744 #define COMPAT_FEATURE_ON(f) \
1745  ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1747 
1748  if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1749  return 1;
1750 
1751  if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1752  return 0;
1753 
1754  /* Asking for checksumming v2 and v1? Only give them v2. */
1755  if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 &&
1757  compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1758 
1759  jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1760  compat, ro, incompat);
1761 
1762  sb = journal->j_superblock;
1763 
1764  /* If enabling v2 checksums, update superblock */
1767  sb->s_feature_compat &=
1768  ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1769 
1770  /* Load the checksum driver */
1771  if (journal->j_chksum_driver == NULL) {
1772  journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1773  0, 0);
1774  if (IS_ERR(journal->j_chksum_driver)) {
1775  printk(KERN_ERR "JBD: Cannot load crc32c "
1776  "driver.\n");
1777  journal->j_chksum_driver = NULL;
1778  return 0;
1779  }
1780  }
1781 
1782  /* Precompute checksum seed for all metadata */
1783  if (JBD2_HAS_INCOMPAT_FEATURE(journal,
1785  journal->j_csum_seed = jbd2_chksum(journal, ~0,
1786  sb->s_uuid,
1787  sizeof(sb->s_uuid));
1788  }
1789 
1790  /* If enabling v1 checksums, downgrade superblock */
1791  if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1792  sb->s_feature_incompat &=
1794 
1795  sb->s_feature_compat |= cpu_to_be32(compat);
1796  sb->s_feature_ro_compat |= cpu_to_be32(ro);
1797  sb->s_feature_incompat |= cpu_to_be32(incompat);
1798 
1799  return 1;
1800 #undef COMPAT_FEATURE_ON
1801 #undef INCOMPAT_FEATURE_ON
1802 }
1803 
1804 /*
1805  * jbd2_journal_clear_features () - Clear a given journal feature in the
1806  * superblock
1807  * @journal: Journal to act on.
1808  * @compat: bitmask of compatible features
1809  * @ro: bitmask of features that force read-only mount
1810  * @incompat: bitmask of incompatible features
1811  *
1812  * Clear a given journal feature as present on the
1813  * superblock.
1814  */
1815 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1816  unsigned long ro, unsigned long incompat)
1817 {
1819 
1820  jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1821  compat, ro, incompat);
1822 
1823  sb = journal->j_superblock;
1824 
1825  sb->s_feature_compat &= ~cpu_to_be32(compat);
1826  sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1827  sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1828 }
1830 
1840 int jbd2_journal_flush(journal_t *journal)
1841 {
1842  int err = 0;
1843  transaction_t *transaction = NULL;
1844 
1845  write_lock(&journal->j_state_lock);
1846 
1847  /* Force everything buffered to the log... */
1848  if (journal->j_running_transaction) {
1849  transaction = journal->j_running_transaction;
1850  __jbd2_log_start_commit(journal, transaction->t_tid);
1851  } else if (journal->j_committing_transaction)
1852  transaction = journal->j_committing_transaction;
1853 
1854  /* Wait for the log commit to complete... */
1855  if (transaction) {
1856  tid_t tid = transaction->t_tid;
1857 
1858  write_unlock(&journal->j_state_lock);
1859  jbd2_log_wait_commit(journal, tid);
1860  } else {
1861  write_unlock(&journal->j_state_lock);
1862  }
1863 
1864  /* ...and flush everything in the log out to disk. */
1865  spin_lock(&journal->j_list_lock);
1866  while (!err && journal->j_checkpoint_transactions != NULL) {
1867  spin_unlock(&journal->j_list_lock);
1868  mutex_lock(&journal->j_checkpoint_mutex);
1869  err = jbd2_log_do_checkpoint(journal);
1870  mutex_unlock(&journal->j_checkpoint_mutex);
1871  spin_lock(&journal->j_list_lock);
1872  }
1873  spin_unlock(&journal->j_list_lock);
1874 
1875  if (is_journal_aborted(journal))
1876  return -EIO;
1877 
1878  mutex_lock(&journal->j_checkpoint_mutex);
1879  jbd2_cleanup_journal_tail(journal);
1880 
1881  /* Finally, mark the journal as really needing no recovery.
1882  * This sets s_start==0 in the underlying superblock, which is
1883  * the magic code for a fully-recovered superblock. Any future
1884  * commits of data to the journal will restore the current
1885  * s_start value. */
1886  jbd2_mark_journal_empty(journal);
1887  mutex_unlock(&journal->j_checkpoint_mutex);
1888  write_lock(&journal->j_state_lock);
1889  J_ASSERT(!journal->j_running_transaction);
1890  J_ASSERT(!journal->j_committing_transaction);
1891  J_ASSERT(!journal->j_checkpoint_transactions);
1892  J_ASSERT(journal->j_head == journal->j_tail);
1893  J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1894  write_unlock(&journal->j_state_lock);
1895  return 0;
1896 }
1897 
1911 int jbd2_journal_wipe(journal_t *journal, int write)
1912 {
1913  int err = 0;
1914 
1915  J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1916 
1917  err = load_superblock(journal);
1918  if (err)
1919  return err;
1920 
1921  if (!journal->j_tail)
1922  goto no_recovery;
1923 
1924  printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1925  write ? "Clearing" : "Ignoring");
1926 
1927  err = jbd2_journal_skip_recovery(journal);
1928  if (write) {
1929  /* Lock to make assertions happy... */
1930  mutex_lock(&journal->j_checkpoint_mutex);
1931  jbd2_mark_journal_empty(journal);
1932  mutex_unlock(&journal->j_checkpoint_mutex);
1933  }
1934 
1935  no_recovery:
1936  return err;
1937 }
1938 
1939 /*
1940  * Journal abort has very specific semantics, which we describe
1941  * for journal abort.
1942  *
1943  * Two internal functions, which provide abort to the jbd layer
1944  * itself are here.
1945  */
1946 
1947 /*
1948  * Quick version for internal journal use (doesn't lock the journal).
1949  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1950  * and don't attempt to make any other journal updates.
1951  */
1952 void __jbd2_journal_abort_hard(journal_t *journal)
1953 {
1955 
1956  if (journal->j_flags & JBD2_ABORT)
1957  return;
1958 
1959  printk(KERN_ERR "Aborting journal on device %s.\n",
1960  journal->j_devname);
1961 
1962  write_lock(&journal->j_state_lock);
1963  journal->j_flags |= JBD2_ABORT;
1964  transaction = journal->j_running_transaction;
1965  if (transaction)
1966  __jbd2_log_start_commit(journal, transaction->t_tid);
1967  write_unlock(&journal->j_state_lock);
1968 }
1969 
1970 /* Soft abort: record the abort error status in the journal superblock,
1971  * but don't do any other IO. */
1972 static void __journal_abort_soft (journal_t *journal, int errno)
1973 {
1974  if (journal->j_flags & JBD2_ABORT)
1975  return;
1976 
1977  if (!journal->j_errno)
1978  journal->j_errno = errno;
1979 
1980  __jbd2_journal_abort_hard(journal);
1981 
1982  if (errno)
1984 }
1985 
2032 void jbd2_journal_abort(journal_t *journal, int errno)
2033 {
2034  __journal_abort_soft(journal, errno);
2035 }
2036 
2048 int jbd2_journal_errno(journal_t *journal)
2049 {
2050  int err;
2051 
2052  read_lock(&journal->j_state_lock);
2053  if (journal->j_flags & JBD2_ABORT)
2054  err = -EROFS;
2055  else
2056  err = journal->j_errno;
2057  read_unlock(&journal->j_state_lock);
2058  return err;
2059 }
2060 
2068 int jbd2_journal_clear_err(journal_t *journal)
2069 {
2070  int err = 0;
2071 
2072  write_lock(&journal->j_state_lock);
2073  if (journal->j_flags & JBD2_ABORT)
2074  err = -EROFS;
2075  else
2076  journal->j_errno = 0;
2077  write_unlock(&journal->j_state_lock);
2078  return err;
2079 }
2080 
2088 void jbd2_journal_ack_err(journal_t *journal)
2089 {
2090  write_lock(&journal->j_state_lock);
2091  if (journal->j_errno)
2092  journal->j_flags |= JBD2_ACK_ERR;
2093  write_unlock(&journal->j_state_lock);
2094 }
2095 
2096 int jbd2_journal_blocks_per_page(struct inode *inode)
2097 {
2098  return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2099 }
2100 
2101 /*
2102  * helper functions to deal with 32 or 64bit block numbers.
2103  */
2104 size_t journal_tag_bytes(journal_t *journal)
2105 {
2107  size_t x = 0;
2108 
2110  x += sizeof(tag.t_checksum);
2111 
2113  return x + JBD2_TAG_SIZE64;
2114  else
2115  return x + JBD2_TAG_SIZE32;
2116 }
2117 
2118 /*
2119  * JBD memory management
2120  *
2121  * These functions are used to allocate block-sized chunks of memory
2122  * used for making copies of buffer_head data. Very often it will be
2123  * page-sized chunks of data, but sometimes it will be in
2124  * sub-page-size chunks. (For example, 16k pages on Power systems
2125  * with a 4k block file system.) For blocks smaller than a page, we
2126  * use a SLAB allocator. There are slab caches for each block size,
2127  * which are allocated at mount time, if necessary, and we only free
2128  * (all of) the slab caches when/if the jbd2 module is unloaded. For
2129  * this reason we don't need to a mutex to protect access to
2130  * jbd2_slab[] allocating or releasing memory; only in
2131  * jbd2_journal_create_slab().
2132  */
2133 #define JBD2_MAX_SLABS 8
2134 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2135 
2136 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2137  "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2138  "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2139 };
2140 
2141 
2142 static void jbd2_journal_destroy_slabs(void)
2143 {
2144  int i;
2145 
2146  for (i = 0; i < JBD2_MAX_SLABS; i++) {
2147  if (jbd2_slab[i])
2148  kmem_cache_destroy(jbd2_slab[i]);
2149  jbd2_slab[i] = NULL;
2150  }
2151 }
2152 
2153 static int jbd2_journal_create_slab(size_t size)
2154 {
2155  static DEFINE_MUTEX(jbd2_slab_create_mutex);
2156  int i = order_base_2(size) - 10;
2157  size_t slab_size;
2158 
2159  if (size == PAGE_SIZE)
2160  return 0;
2161 
2162  if (i >= JBD2_MAX_SLABS)
2163  return -EINVAL;
2164 
2165  if (unlikely(i < 0))
2166  i = 0;
2167  mutex_lock(&jbd2_slab_create_mutex);
2168  if (jbd2_slab[i]) {
2169  mutex_unlock(&jbd2_slab_create_mutex);
2170  return 0; /* Already created */
2171  }
2172 
2173  slab_size = 1 << (i+10);
2174  jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2175  slab_size, 0, NULL);
2176  mutex_unlock(&jbd2_slab_create_mutex);
2177  if (!jbd2_slab[i]) {
2178  printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2179  return -ENOMEM;
2180  }
2181  return 0;
2182 }
2183 
2184 static struct kmem_cache *get_slab(size_t size)
2185 {
2186  int i = order_base_2(size) - 10;
2187 
2188  BUG_ON(i >= JBD2_MAX_SLABS);
2189  if (unlikely(i < 0))
2190  i = 0;
2191  BUG_ON(jbd2_slab[i] == NULL);
2192  return jbd2_slab[i];
2193 }
2194 
2195 void *jbd2_alloc(size_t size, gfp_t flags)
2196 {
2197  void *ptr;
2198 
2199  BUG_ON(size & (size-1)); /* Must be a power of 2 */
2200 
2201  flags |= __GFP_REPEAT;
2202  if (size == PAGE_SIZE)
2203  ptr = (void *)__get_free_pages(flags, 0);
2204  else if (size > PAGE_SIZE) {
2205  int order = get_order(size);
2206 
2207  if (order < 3)
2208  ptr = (void *)__get_free_pages(flags, order);
2209  else
2210  ptr = vmalloc(size);
2211  } else
2212  ptr = kmem_cache_alloc(get_slab(size), flags);
2213 
2214  /* Check alignment; SLUB has gotten this wrong in the past,
2215  * and this can lead to user data corruption! */
2216  BUG_ON(((unsigned long) ptr) & (size-1));
2217 
2218  return ptr;
2219 }
2220 
2221 void jbd2_free(void *ptr, size_t size)
2222 {
2223  if (size == PAGE_SIZE) {
2224  free_pages((unsigned long)ptr, 0);
2225  return;
2226  }
2227  if (size > PAGE_SIZE) {
2228  int order = get_order(size);
2229 
2230  if (order < 3)
2231  free_pages((unsigned long)ptr, order);
2232  else
2233  vfree(ptr);
2234  return;
2235  }
2236  kmem_cache_free(get_slab(size), ptr);
2237 };
2238 
2239 /*
2240  * Journal_head storage management
2241  */
2242 static struct kmem_cache *jbd2_journal_head_cache;
2243 #ifdef CONFIG_JBD2_DEBUG
2244 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2245 #endif
2246 
2247 static int jbd2_journal_init_journal_head_cache(void)
2248 {
2249  int retval;
2250 
2251  J_ASSERT(jbd2_journal_head_cache == NULL);
2252  jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2253  sizeof(struct journal_head),
2254  0, /* offset */
2255  SLAB_TEMPORARY, /* flags */
2256  NULL); /* ctor */
2257  retval = 0;
2258  if (!jbd2_journal_head_cache) {
2259  retval = -ENOMEM;
2260  printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2261  }
2262  return retval;
2263 }
2264 
2265 static void jbd2_journal_destroy_journal_head_cache(void)
2266 {
2267  if (jbd2_journal_head_cache) {
2268  kmem_cache_destroy(jbd2_journal_head_cache);
2269  jbd2_journal_head_cache = NULL;
2270  }
2271 }
2272 
2273 /*
2274  * journal_head splicing and dicing
2275  */
2276 static struct journal_head *journal_alloc_journal_head(void)
2277 {
2278  struct journal_head *ret;
2279 
2280 #ifdef CONFIG_JBD2_DEBUG
2281  atomic_inc(&nr_journal_heads);
2282 #endif
2283  ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2284  if (!ret) {
2285  jbd_debug(1, "out of memory for journal_head\n");
2286  pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2287  while (!ret) {
2288  yield();
2289  ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2290  }
2291  }
2292  return ret;
2293 }
2294 
2295 static void journal_free_journal_head(struct journal_head *jh)
2296 {
2297 #ifdef CONFIG_JBD2_DEBUG
2298  atomic_dec(&nr_journal_heads);
2299  memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2300 #endif
2301  kmem_cache_free(jbd2_journal_head_cache, jh);
2302 }
2303 
2304 /*
2305  * A journal_head is attached to a buffer_head whenever JBD has an
2306  * interest in the buffer.
2307  *
2308  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2309  * is set. This bit is tested in core kernel code where we need to take
2310  * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2311  * there.
2312  *
2313  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2314  *
2315  * When a buffer has its BH_JBD bit set it is immune from being released by
2316  * core kernel code, mainly via ->b_count.
2317  *
2318  * A journal_head is detached from its buffer_head when the journal_head's
2319  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2320  * transaction (b_cp_transaction) hold their references to b_jcount.
2321  *
2322  * Various places in the kernel want to attach a journal_head to a buffer_head
2323  * _before_ attaching the journal_head to a transaction. To protect the
2324  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2325  * journal_head's b_jcount refcount by one. The caller must call
2326  * jbd2_journal_put_journal_head() to undo this.
2327  *
2328  * So the typical usage would be:
2329  *
2330  * (Attach a journal_head if needed. Increments b_jcount)
2331  * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2332  * ...
2333  * (Get another reference for transaction)
2334  * jbd2_journal_grab_journal_head(bh);
2335  * jh->b_transaction = xxx;
2336  * (Put original reference)
2337  * jbd2_journal_put_journal_head(jh);
2338  */
2339 
2340 /*
2341  * Give a buffer_head a journal_head.
2342  *
2343  * May sleep.
2344  */
2345 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2346 {
2347  struct journal_head *jh;
2348  struct journal_head *new_jh = NULL;
2349 
2350 repeat:
2351  if (!buffer_jbd(bh)) {
2352  new_jh = journal_alloc_journal_head();
2353  memset(new_jh, 0, sizeof(*new_jh));
2354  }
2355 
2356  jbd_lock_bh_journal_head(bh);
2357  if (buffer_jbd(bh)) {
2358  jh = bh2jh(bh);
2359  } else {
2360  J_ASSERT_BH(bh,
2361  (atomic_read(&bh->b_count) > 0) ||
2362  (bh->b_page && bh->b_page->mapping));
2363 
2364  if (!new_jh) {
2365  jbd_unlock_bh_journal_head(bh);
2366  goto repeat;
2367  }
2368 
2369  jh = new_jh;
2370  new_jh = NULL; /* We consumed it */
2371  set_buffer_jbd(bh);
2372  bh->b_private = jh;
2373  jh->b_bh = bh;
2374  get_bh(bh);
2375  BUFFER_TRACE(bh, "added journal_head");
2376  }
2377  jh->b_jcount++;
2378  jbd_unlock_bh_journal_head(bh);
2379  if (new_jh)
2380  journal_free_journal_head(new_jh);
2381  return bh->b_private;
2382 }
2383 
2384 /*
2385  * Grab a ref against this buffer_head's journal_head. If it ended up not
2386  * having a journal_head, return NULL
2387  */
2388 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2389 {
2390  struct journal_head *jh = NULL;
2391 
2392  jbd_lock_bh_journal_head(bh);
2393  if (buffer_jbd(bh)) {
2394  jh = bh2jh(bh);
2395  jh->b_jcount++;
2396  }
2397  jbd_unlock_bh_journal_head(bh);
2398  return jh;
2399 }
2400 
2401 static void __journal_remove_journal_head(struct buffer_head *bh)
2402 {
2403  struct journal_head *jh = bh2jh(bh);
2404 
2405  J_ASSERT_JH(jh, jh->b_jcount >= 0);
2406  J_ASSERT_JH(jh, jh->b_transaction == NULL);
2407  J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2408  J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2409  J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2410  J_ASSERT_BH(bh, buffer_jbd(bh));
2411  J_ASSERT_BH(bh, jh2bh(jh) == bh);
2412  BUFFER_TRACE(bh, "remove journal_head");
2413  if (jh->b_frozen_data) {
2414  printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2415  jbd2_free(jh->b_frozen_data, bh->b_size);
2416  }
2417  if (jh->b_committed_data) {
2418  printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2419  jbd2_free(jh->b_committed_data, bh->b_size);
2420  }
2421  bh->b_private = NULL;
2422  jh->b_bh = NULL; /* debug, really */
2423  clear_buffer_jbd(bh);
2424  journal_free_journal_head(jh);
2425 }
2426 
2427 /*
2428  * Drop a reference on the passed journal_head. If it fell to zero then
2429  * release the journal_head from the buffer_head.
2430  */
2432 {
2433  struct buffer_head *bh = jh2bh(jh);
2434 
2435  jbd_lock_bh_journal_head(bh);
2436  J_ASSERT_JH(jh, jh->b_jcount > 0);
2437  --jh->b_jcount;
2438  if (!jh->b_jcount) {
2439  __journal_remove_journal_head(bh);
2440  jbd_unlock_bh_journal_head(bh);
2441  __brelse(bh);
2442  } else
2443  jbd_unlock_bh_journal_head(bh);
2444 }
2445 
2446 /*
2447  * Initialize jbd inode head
2448  */
2449 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2450 {
2451  jinode->i_transaction = NULL;
2452  jinode->i_next_transaction = NULL;
2453  jinode->i_vfs_inode = inode;
2454  jinode->i_flags = 0;
2455  INIT_LIST_HEAD(&jinode->i_list);
2456 }
2457 
2458 /*
2459  * Function to be called before we start removing inode from memory (i.e.,
2460  * clear_inode() is a fine place to be called from). It removes inode from
2461  * transaction's lists.
2462  */
2463 void jbd2_journal_release_jbd_inode(journal_t *journal,
2464  struct jbd2_inode *jinode)
2465 {
2466  if (!journal)
2467  return;
2468 restart:
2469  spin_lock(&journal->j_list_lock);
2470  /* Is commit writing out inode - we have to wait */
2471  if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2472  wait_queue_head_t *wq;
2473  DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2474  wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2476  spin_unlock(&journal->j_list_lock);
2477  schedule();
2478  finish_wait(wq, &wait.wait);
2479  goto restart;
2480  }
2481 
2482  if (jinode->i_transaction) {
2483  list_del(&jinode->i_list);
2484  jinode->i_transaction = NULL;
2485  }
2486  spin_unlock(&journal->j_list_lock);
2487 }
2488 
2489 /*
2490  * debugfs tunables
2491  */
2492 #ifdef CONFIG_JBD2_DEBUG
2493 u8 jbd2_journal_enable_debug __read_mostly;
2494 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2495 
2496 #define JBD2_DEBUG_NAME "jbd2-debug"
2497 
2498 static struct dentry *jbd2_debugfs_dir;
2499 static struct dentry *jbd2_debug;
2500 
2501 static void __init jbd2_create_debugfs_entry(void)
2502 {
2503  jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2504  if (jbd2_debugfs_dir)
2505  jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2506  S_IRUGO | S_IWUSR,
2507  jbd2_debugfs_dir,
2508  &jbd2_journal_enable_debug);
2509 }
2510 
2511 static void __exit jbd2_remove_debugfs_entry(void)
2512 {
2513  debugfs_remove(jbd2_debug);
2514  debugfs_remove(jbd2_debugfs_dir);
2515 }
2516 
2517 #else
2518 
2519 static void __init jbd2_create_debugfs_entry(void)
2520 {
2521 }
2522 
2523 static void __exit jbd2_remove_debugfs_entry(void)
2524 {
2525 }
2526 
2527 #endif
2528 
2529 #ifdef CONFIG_PROC_FS
2530 
2531 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2532 
2533 static void __init jbd2_create_jbd_stats_proc_entry(void)
2534 {
2535  proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2536 }
2537 
2538 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2539 {
2540  if (proc_jbd2_stats)
2541  remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2542 }
2543 
2544 #else
2545 
2546 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2547 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2548 
2549 #endif
2550 
2552 
2553 static int __init jbd2_journal_init_handle_cache(void)
2554 {
2555  jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2556  if (jbd2_handle_cache == NULL) {
2557  printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2558  return -ENOMEM;
2559  }
2560  jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2561  if (jbd2_inode_cache == NULL) {
2562  printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2563  kmem_cache_destroy(jbd2_handle_cache);
2564  return -ENOMEM;
2565  }
2566  return 0;
2567 }
2568 
2569 static void jbd2_journal_destroy_handle_cache(void)
2570 {
2571  if (jbd2_handle_cache)
2572  kmem_cache_destroy(jbd2_handle_cache);
2573  if (jbd2_inode_cache)
2574  kmem_cache_destroy(jbd2_inode_cache);
2575 
2576 }
2577 
2578 /*
2579  * Module startup and shutdown
2580  */
2581 
2582 static int __init journal_init_caches(void)
2583 {
2584  int ret;
2585 
2587  if (ret == 0)
2588  ret = jbd2_journal_init_journal_head_cache();
2589  if (ret == 0)
2590  ret = jbd2_journal_init_handle_cache();
2591  if (ret == 0)
2593  return ret;
2594 }
2595 
2596 static void jbd2_journal_destroy_caches(void)
2597 {
2599  jbd2_journal_destroy_journal_head_cache();
2600  jbd2_journal_destroy_handle_cache();
2602  jbd2_journal_destroy_slabs();
2603 }
2604 
2605 static int __init journal_init(void)
2606 {
2607  int ret;
2608 
2609  BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2610 
2611  ret = journal_init_caches();
2612  if (ret == 0) {
2613  jbd2_create_debugfs_entry();
2615  } else {
2616  jbd2_journal_destroy_caches();
2617  }
2618  return ret;
2619 }
2620 
2621 static void __exit journal_exit(void)
2622 {
2623 #ifdef CONFIG_JBD2_DEBUG
2624  int n = atomic_read(&nr_journal_heads);
2625  if (n)
2626  printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
2627 #endif
2628  jbd2_remove_debugfs_entry();
2630  jbd2_journal_destroy_caches();
2631 }
2632 
2633 MODULE_LICENSE("GPL");
2634 module_init(journal_init);
2635 module_exit(journal_exit);
2636