~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/jbd2/journal.c

Version: ~ [ linux-5.8 ] ~ [ linux-5.7.14 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.57 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.138 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.193 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.232 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.232 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * linux/fs/jbd2/journal.c
  3  *
  4  * Written by Stephen C. Tweedie <sct@redhat.com>, 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/seq_file.h>
 39 #include <linux/math64.h>
 40 #include <linux/hash.h>
 41 #include <linux/log2.h>
 42 #include <linux/vmalloc.h>
 43 #include <linux/backing-dev.h>
 44 #include <linux/bitops.h>
 45 #include <linux/ratelimit.h>
 46 
 47 #define CREATE_TRACE_POINTS
 48 #include <trace/events/jbd2.h>
 49 
 50 #include <asm/uaccess.h>
 51 #include <asm/page.h>
 52 
 53 #ifdef CONFIG_JBD2_DEBUG
 54 ushort jbd2_journal_enable_debug __read_mostly;
 55 EXPORT_SYMBOL(jbd2_journal_enable_debug);
 56 
 57 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
 58 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
 59 #endif
 60 
 61 EXPORT_SYMBOL(jbd2_journal_extend);
 62 EXPORT_SYMBOL(jbd2_journal_stop);
 63 EXPORT_SYMBOL(jbd2_journal_lock_updates);
 64 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
 65 EXPORT_SYMBOL(jbd2_journal_get_write_access);
 66 EXPORT_SYMBOL(jbd2_journal_get_create_access);
 67 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
 68 EXPORT_SYMBOL(jbd2_journal_set_triggers);
 69 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
 70 EXPORT_SYMBOL(jbd2_journal_forget);
 71 #if 0
 72 EXPORT_SYMBOL(journal_sync_buffer);
 73 #endif
 74 EXPORT_SYMBOL(jbd2_journal_flush);
 75 EXPORT_SYMBOL(jbd2_journal_revoke);
 76 
 77 EXPORT_SYMBOL(jbd2_journal_init_dev);
 78 EXPORT_SYMBOL(jbd2_journal_init_inode);
 79 EXPORT_SYMBOL(jbd2_journal_check_used_features);
 80 EXPORT_SYMBOL(jbd2_journal_check_available_features);
 81 EXPORT_SYMBOL(jbd2_journal_set_features);
 82 EXPORT_SYMBOL(jbd2_journal_load);
 83 EXPORT_SYMBOL(jbd2_journal_destroy);
 84 EXPORT_SYMBOL(jbd2_journal_abort);
 85 EXPORT_SYMBOL(jbd2_journal_errno);
 86 EXPORT_SYMBOL(jbd2_journal_ack_err);
 87 EXPORT_SYMBOL(jbd2_journal_clear_err);
 88 EXPORT_SYMBOL(jbd2_log_wait_commit);
 89 EXPORT_SYMBOL(jbd2_log_start_commit);
 90 EXPORT_SYMBOL(jbd2_journal_start_commit);
 91 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
 92 EXPORT_SYMBOL(jbd2_journal_wipe);
 93 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
 94 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
 95 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
 96 EXPORT_SYMBOL(jbd2_journal_force_commit);
 97 EXPORT_SYMBOL(jbd2_journal_file_inode);
 98 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
 99 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
101 EXPORT_SYMBOL(jbd2_inode_cache);
102 
103 static void __journal_abort_soft (journal_t *journal, int errno);
104 static int jbd2_journal_create_slab(size_t slab_size);
105 
106 #ifdef CONFIG_JBD2_DEBUG
107 void __jbd2_debug(int level, const char *file, const char *func,
108                   unsigned int line, const char *fmt, ...)
109 {
110         struct va_format vaf;
111         va_list args;
112 
113         if (level > jbd2_journal_enable_debug)
114                 return;
115         va_start(args, fmt);
116         vaf.fmt = fmt;
117         vaf.va = &args;
118         printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
119         va_end(args);
120 }
121 EXPORT_SYMBOL(__jbd2_debug);
122 #endif
123 
124 /* Checksumming functions */
125 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
126 {
127         if (!jbd2_journal_has_csum_v2or3(j))
128                 return 1;
129 
130         return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
131 }
132 
133 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
134 {
135         __u32 csum;
136         __be32 old_csum;
137 
138         old_csum = sb->s_checksum;
139         sb->s_checksum = 0;
140         csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
141         sb->s_checksum = old_csum;
142 
143         return cpu_to_be32(csum);
144 }
145 
146 static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
147 {
148         if (!jbd2_journal_has_csum_v2or3(j))
149                 return 1;
150 
151         return sb->s_checksum == jbd2_superblock_csum(j, sb);
152 }
153 
154 static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
155 {
156         if (!jbd2_journal_has_csum_v2or3(j))
157                 return;
158 
159         sb->s_checksum = jbd2_superblock_csum(j, sb);
160 }
161 
162 /*
163  * Helper function used to manage commit timeouts
164  */
165 
166 static void commit_timeout(unsigned long __data)
167 {
168         struct task_struct * p = (struct task_struct *) __data;
169 
170         wake_up_process(p);
171 }
172 
173 /*
174  * kjournald2: The main thread function used to manage a logging device
175  * journal.
176  *
177  * This kernel thread is responsible for two things:
178  *
179  * 1) COMMIT:  Every so often we need to commit the current state of the
180  *    filesystem to disk.  The journal thread is responsible for writing
181  *    all of the metadata buffers to disk.
182  *
183  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
184  *    of the data in that part of the log has been rewritten elsewhere on
185  *    the disk.  Flushing these old buffers to reclaim space in the log is
186  *    known as checkpointing, and this thread is responsible for that job.
187  */
188 
189 static int kjournald2(void *arg)
190 {
191         journal_t *journal = arg;
192         transaction_t *transaction;
193 
194         /*
195          * Set up an interval timer which can be used to trigger a commit wakeup
196          * after the commit interval expires
197          */
198         setup_timer(&journal->j_commit_timer, commit_timeout,
199                         (unsigned long)current);
200 
201         set_freezable();
202 
203         /* Record that the journal thread is running */
204         journal->j_task = current;
205         wake_up(&journal->j_wait_done_commit);
206 
207         /*
208          * And now, wait forever for commit wakeup events.
209          */
210         write_lock(&journal->j_state_lock);
211 
212 loop:
213         if (journal->j_flags & JBD2_UNMOUNT)
214                 goto end_loop;
215 
216         jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
217                 journal->j_commit_sequence, journal->j_commit_request);
218 
219         if (journal->j_commit_sequence != journal->j_commit_request) {
220                 jbd_debug(1, "OK, requests differ\n");
221                 write_unlock(&journal->j_state_lock);
222                 del_timer_sync(&journal->j_commit_timer);
223                 jbd2_journal_commit_transaction(journal);
224                 write_lock(&journal->j_state_lock);
225                 goto loop;
226         }
227 
228         wake_up(&journal->j_wait_done_commit);
229         if (freezing(current)) {
230                 /*
231                  * The simpler the better. Flushing journal isn't a
232                  * good idea, because that depends on threads that may
233                  * be already stopped.
234                  */
235                 jbd_debug(1, "Now suspending kjournald2\n");
236                 write_unlock(&journal->j_state_lock);
237                 try_to_freeze();
238                 write_lock(&journal->j_state_lock);
239         } else {
240                 /*
241                  * We assume on resume that commits are already there,
242                  * so we don't sleep
243                  */
244                 DEFINE_WAIT(wait);
245                 int should_sleep = 1;
246 
247                 prepare_to_wait(&journal->j_wait_commit, &wait,
248                                 TASK_INTERRUPTIBLE);
249                 if (journal->j_commit_sequence != journal->j_commit_request)
250                         should_sleep = 0;
251                 transaction = journal->j_running_transaction;
252                 if (transaction && time_after_eq(jiffies,
253                                                 transaction->t_expires))
254                         should_sleep = 0;
255                 if (journal->j_flags & JBD2_UNMOUNT)
256                         should_sleep = 0;
257                 if (should_sleep) {
258                         write_unlock(&journal->j_state_lock);
259                         schedule();
260                         write_lock(&journal->j_state_lock);
261                 }
262                 finish_wait(&journal->j_wait_commit, &wait);
263         }
264 
265         jbd_debug(1, "kjournald2 wakes\n");
266 
267         /*
268          * Were we woken up by a commit wakeup event?
269          */
270         transaction = journal->j_running_transaction;
271         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
272                 journal->j_commit_request = transaction->t_tid;
273                 jbd_debug(1, "woke because of timeout\n");
274         }
275         goto loop;
276 
277 end_loop:
278         del_timer_sync(&journal->j_commit_timer);
279         journal->j_task = NULL;
280         wake_up(&journal->j_wait_done_commit);
281         jbd_debug(1, "Journal thread exiting.\n");
282         write_unlock(&journal->j_state_lock);
283         return 0;
284 }
285 
286 static int jbd2_journal_start_thread(journal_t *journal)
287 {
288         struct task_struct *t;
289 
290         t = kthread_run(kjournald2, journal, "jbd2/%s",
291                         journal->j_devname);
292         if (IS_ERR(t))
293                 return PTR_ERR(t);
294 
295         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
296         return 0;
297 }
298 
299 static void journal_kill_thread(journal_t *journal)
300 {
301         write_lock(&journal->j_state_lock);
302         journal->j_flags |= JBD2_UNMOUNT;
303 
304         while (journal->j_task) {
305                 write_unlock(&journal->j_state_lock);
306                 wake_up(&journal->j_wait_commit);
307                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
308                 write_lock(&journal->j_state_lock);
309         }
310         write_unlock(&journal->j_state_lock);
311 }
312 
313 /*
314  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
315  *
316  * Writes a metadata buffer to a given disk block.  The actual IO is not
317  * performed but a new buffer_head is constructed which labels the data
318  * to be written with the correct destination disk block.
319  *
320  * Any magic-number escaping which needs to be done will cause a
321  * copy-out here.  If the buffer happens to start with the
322  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
323  * magic number is only written to the log for descripter blocks.  In
324  * this case, we copy the data and replace the first word with 0, and we
325  * return a result code which indicates that this buffer needs to be
326  * marked as an escaped buffer in the corresponding log descriptor
327  * block.  The missing word can then be restored when the block is read
328  * during recovery.
329  *
330  * If the source buffer has already been modified by a new transaction
331  * since we took the last commit snapshot, we use the frozen copy of
332  * that data for IO. If we end up using the existing buffer_head's data
333  * for the write, then we have to make sure nobody modifies it while the
334  * IO is in progress. do_get_write_access() handles this.
335  *
336  * The function returns a pointer to the buffer_head to be used for IO.
337  * 
338  *
339  * Return value:
340  *  <0: Error
341  * >=0: Finished OK
342  *
343  * On success:
344  * Bit 0 set == escape performed on the data
345  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
346  */
347 
348 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
349                                   struct journal_head  *jh_in,
350                                   struct buffer_head **bh_out,
351                                   sector_t blocknr)
352 {
353         int need_copy_out = 0;
354         int done_copy_out = 0;
355         int do_escape = 0;
356         char *mapped_data;
357         struct buffer_head *new_bh;
358         struct page *new_page;
359         unsigned int new_offset;
360         struct buffer_head *bh_in = jh2bh(jh_in);
361         journal_t *journal = transaction->t_journal;
362 
363         /*
364          * The buffer really shouldn't be locked: only the current committing
365          * transaction is allowed to write it, so nobody else is allowed
366          * to do any IO.
367          *
368          * akpm: except if we're journalling data, and write() output is
369          * also part of a shared mapping, and another thread has
370          * decided to launch a writepage() against this buffer.
371          */
372         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
373 
374 retry_alloc:
375         new_bh = alloc_buffer_head(GFP_NOFS);
376         if (!new_bh) {
377                 /*
378                  * Failure is not an option, but __GFP_NOFAIL is going
379                  * away; so we retry ourselves here.
380                  */
381                 congestion_wait(BLK_RW_ASYNC, HZ/50);
382                 goto retry_alloc;
383         }
384 
385         /* keep subsequent assertions sane */
386         atomic_set(&new_bh->b_count, 1);
387 
388         jbd_lock_bh_state(bh_in);
389 repeat:
390         /*
391          * If a new transaction has already done a buffer copy-out, then
392          * we use that version of the data for the commit.
393          */
394         if (jh_in->b_frozen_data) {
395                 done_copy_out = 1;
396                 new_page = virt_to_page(jh_in->b_frozen_data);
397                 new_offset = offset_in_page(jh_in->b_frozen_data);
398         } else {
399                 new_page = jh2bh(jh_in)->b_page;
400                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
401         }
402 
403         mapped_data = kmap_atomic(new_page);
404         /*
405          * Fire data frozen trigger if data already wasn't frozen.  Do this
406          * before checking for escaping, as the trigger may modify the magic
407          * offset.  If a copy-out happens afterwards, it will have the correct
408          * data in the buffer.
409          */
410         if (!done_copy_out)
411                 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
412                                            jh_in->b_triggers);
413 
414         /*
415          * Check for escaping
416          */
417         if (*((__be32 *)(mapped_data + new_offset)) ==
418                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
419                 need_copy_out = 1;
420                 do_escape = 1;
421         }
422         kunmap_atomic(mapped_data);
423 
424         /*
425          * Do we need to do a data copy?
426          */
427         if (need_copy_out && !done_copy_out) {
428                 char *tmp;
429 
430                 jbd_unlock_bh_state(bh_in);
431                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
432                 if (!tmp) {
433                         brelse(new_bh);
434                         return -ENOMEM;
435                 }
436                 jbd_lock_bh_state(bh_in);
437                 if (jh_in->b_frozen_data) {
438                         jbd2_free(tmp, bh_in->b_size);
439                         goto repeat;
440                 }
441 
442                 jh_in->b_frozen_data = tmp;
443                 mapped_data = kmap_atomic(new_page);
444                 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
445                 kunmap_atomic(mapped_data);
446 
447                 new_page = virt_to_page(tmp);
448                 new_offset = offset_in_page(tmp);
449                 done_copy_out = 1;
450 
451                 /*
452                  * This isn't strictly necessary, as we're using frozen
453                  * data for the escaping, but it keeps consistency with
454                  * b_frozen_data usage.
455                  */
456                 jh_in->b_frozen_triggers = jh_in->b_triggers;
457         }
458 
459         /*
460          * Did we need to do an escaping?  Now we've done all the
461          * copying, we can finally do so.
462          */
463         if (do_escape) {
464                 mapped_data = kmap_atomic(new_page);
465                 *((unsigned int *)(mapped_data + new_offset)) = 0;
466                 kunmap_atomic(mapped_data);
467         }
468 
469         set_bh_page(new_bh, new_page, new_offset);
470         new_bh->b_size = bh_in->b_size;
471         new_bh->b_bdev = journal->j_dev;
472         new_bh->b_blocknr = blocknr;
473         new_bh->b_private = bh_in;
474         set_buffer_mapped(new_bh);
475         set_buffer_dirty(new_bh);
476 
477         *bh_out = new_bh;
478 
479         /*
480          * The to-be-written buffer needs to get moved to the io queue,
481          * and the original buffer whose contents we are shadowing or
482          * copying is moved to the transaction's shadow queue.
483          */
484         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
485         spin_lock(&journal->j_list_lock);
486         __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
487         spin_unlock(&journal->j_list_lock);
488         set_buffer_shadow(bh_in);
489         jbd_unlock_bh_state(bh_in);
490 
491         return do_escape | (done_copy_out << 1);
492 }
493 
494 /*
495  * Allocation code for the journal file.  Manage the space left in the
496  * journal, so that we can begin checkpointing when appropriate.
497  */
498 
499 /*
500  * Called with j_state_lock locked for writing.
501  * Returns true if a transaction commit was started.
502  */
503 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
504 {
505         /* Return if the txn has already requested to be committed */
506         if (journal->j_commit_request == target)
507                 return 0;
508 
509         /*
510          * The only transaction we can possibly wait upon is the
511          * currently running transaction (if it exists).  Otherwise,
512          * the target tid must be an old one.
513          */
514         if (journal->j_running_transaction &&
515             journal->j_running_transaction->t_tid == target) {
516                 /*
517                  * We want a new commit: OK, mark the request and wakeup the
518                  * commit thread.  We do _not_ do the commit ourselves.
519                  */
520 
521                 journal->j_commit_request = target;
522                 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
523                           journal->j_commit_request,
524                           journal->j_commit_sequence);
525                 journal->j_running_transaction->t_requested = jiffies;
526                 wake_up(&journal->j_wait_commit);
527                 return 1;
528         } else if (!tid_geq(journal->j_commit_request, target))
529                 /* This should never happen, but if it does, preserve
530                    the evidence before kjournald goes into a loop and
531                    increments j_commit_sequence beyond all recognition. */
532                 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
533                           journal->j_commit_request,
534                           journal->j_commit_sequence,
535                           target, journal->j_running_transaction ? 
536                           journal->j_running_transaction->t_tid : 0);
537         return 0;
538 }
539 
540 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
541 {
542         int ret;
543 
544         write_lock(&journal->j_state_lock);
545         ret = __jbd2_log_start_commit(journal, tid);
546         write_unlock(&journal->j_state_lock);
547         return ret;
548 }
549 
550 /*
551  * Force and wait any uncommitted transactions.  We can only force the running
552  * transaction if we don't have an active handle, otherwise, we will deadlock.
553  * Returns: <0 in case of error,
554  *           0 if nothing to commit,
555  *           1 if transaction was successfully committed.
556  */
557 static int __jbd2_journal_force_commit(journal_t *journal)
558 {
559         transaction_t *transaction = NULL;
560         tid_t tid;
561         int need_to_start = 0, ret = 0;
562 
563         read_lock(&journal->j_state_lock);
564         if (journal->j_running_transaction && !current->journal_info) {
565                 transaction = journal->j_running_transaction;
566                 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
567                         need_to_start = 1;
568         } else if (journal->j_committing_transaction)
569                 transaction = journal->j_committing_transaction;
570 
571         if (!transaction) {
572                 /* Nothing to commit */
573                 read_unlock(&journal->j_state_lock);
574                 return 0;
575         }
576         tid = transaction->t_tid;
577         read_unlock(&journal->j_state_lock);
578         if (need_to_start)
579                 jbd2_log_start_commit(journal, tid);
580         ret = jbd2_log_wait_commit(journal, tid);
581         if (!ret)
582                 ret = 1;
583 
584         return ret;
585 }
586 
587 /**
588  * Force and wait upon a commit if the calling process is not within
589  * transaction.  This is used for forcing out undo-protected data which contains
590  * bitmaps, when the fs is running out of space.
591  *
592  * @journal: journal to force
593  * Returns true if progress was made.
594  */
595 int jbd2_journal_force_commit_nested(journal_t *journal)
596 {
597         int ret;
598 
599         ret = __jbd2_journal_force_commit(journal);
600         return ret > 0;
601 }
602 
603 /**
604  * int journal_force_commit() - force any uncommitted transactions
605  * @journal: journal to force
606  *
607  * Caller want unconditional commit. We can only force the running transaction
608  * if we don't have an active handle, otherwise, we will deadlock.
609  */
610 int jbd2_journal_force_commit(journal_t *journal)
611 {
612         int ret;
613 
614         J_ASSERT(!current->journal_info);
615         ret = __jbd2_journal_force_commit(journal);
616         if (ret > 0)
617                 ret = 0;
618         return ret;
619 }
620 
621 /*
622  * Start a commit of the current running transaction (if any).  Returns true
623  * if a transaction is going to be committed (or is currently already
624  * committing), and fills its tid in at *ptid
625  */
626 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
627 {
628         int ret = 0;
629 
630         write_lock(&journal->j_state_lock);
631         if (journal->j_running_transaction) {
632                 tid_t tid = journal->j_running_transaction->t_tid;
633 
634                 __jbd2_log_start_commit(journal, tid);
635                 /* There's a running transaction and we've just made sure
636                  * it's commit has been scheduled. */
637                 if (ptid)
638                         *ptid = tid;
639                 ret = 1;
640         } else if (journal->j_committing_transaction) {
641                 /*
642                  * If commit has been started, then we have to wait for
643                  * completion of that transaction.
644                  */
645                 if (ptid)
646                         *ptid = journal->j_committing_transaction->t_tid;
647                 ret = 1;
648         }
649         write_unlock(&journal->j_state_lock);
650         return ret;
651 }
652 
653 /*
654  * Return 1 if a given transaction has not yet sent barrier request
655  * connected with a transaction commit. If 0 is returned, transaction
656  * may or may not have sent the barrier. Used to avoid sending barrier
657  * twice in common cases.
658  */
659 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
660 {
661         int ret = 0;
662         transaction_t *commit_trans;
663 
664         if (!(journal->j_flags & JBD2_BARRIER))
665                 return 0;
666         read_lock(&journal->j_state_lock);
667         /* Transaction already committed? */
668         if (tid_geq(journal->j_commit_sequence, tid))
669                 goto out;
670         commit_trans = journal->j_committing_transaction;
671         if (!commit_trans || commit_trans->t_tid != tid) {
672                 ret = 1;
673                 goto out;
674         }
675         /*
676          * Transaction is being committed and we already proceeded to
677          * submitting a flush to fs partition?
678          */
679         if (journal->j_fs_dev != journal->j_dev) {
680                 if (!commit_trans->t_need_data_flush ||
681                     commit_trans->t_state >= T_COMMIT_DFLUSH)
682                         goto out;
683         } else {
684                 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
685                         goto out;
686         }
687         ret = 1;
688 out:
689         read_unlock(&journal->j_state_lock);
690         return ret;
691 }
692 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
693 
694 /*
695  * Wait for a specified commit to complete.
696  * The caller may not hold the journal lock.
697  */
698 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
699 {
700         int err = 0;
701 
702         read_lock(&journal->j_state_lock);
703 #ifdef CONFIG_JBD2_DEBUG
704         if (!tid_geq(journal->j_commit_request, tid)) {
705                 printk(KERN_ERR
706                        "%s: error: j_commit_request=%d, tid=%d\n",
707                        __func__, journal->j_commit_request, tid);
708         }
709 #endif
710         while (tid_gt(tid, journal->j_commit_sequence)) {
711                 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
712                                   tid, journal->j_commit_sequence);
713                 read_unlock(&journal->j_state_lock);
714                 wake_up(&journal->j_wait_commit);
715                 wait_event(journal->j_wait_done_commit,
716                                 !tid_gt(tid, journal->j_commit_sequence));
717                 read_lock(&journal->j_state_lock);
718         }
719         read_unlock(&journal->j_state_lock);
720 
721         if (unlikely(is_journal_aborted(journal)))
722                 err = -EIO;
723         return err;
724 }
725 
726 /*
727  * When this function returns the transaction corresponding to tid
728  * will be completed.  If the transaction has currently running, start
729  * committing that transaction before waiting for it to complete.  If
730  * the transaction id is stale, it is by definition already completed,
731  * so just return SUCCESS.
732  */
733 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
734 {
735         int     need_to_wait = 1;
736 
737         read_lock(&journal->j_state_lock);
738         if (journal->j_running_transaction &&
739             journal->j_running_transaction->t_tid == tid) {
740                 if (journal->j_commit_request != tid) {
741                         /* transaction not yet started, so request it */
742                         read_unlock(&journal->j_state_lock);
743                         jbd2_log_start_commit(journal, tid);
744                         goto wait_commit;
745                 }
746         } else if (!(journal->j_committing_transaction &&
747                      journal->j_committing_transaction->t_tid == tid))
748                 need_to_wait = 0;
749         read_unlock(&journal->j_state_lock);
750         if (!need_to_wait)
751                 return 0;
752 wait_commit:
753         return jbd2_log_wait_commit(journal, tid);
754 }
755 EXPORT_SYMBOL(jbd2_complete_transaction);
756 
757 /*
758  * Log buffer allocation routines:
759  */
760 
761 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
762 {
763         unsigned long blocknr;
764 
765         write_lock(&journal->j_state_lock);
766         J_ASSERT(journal->j_free > 1);
767 
768         blocknr = journal->j_head;
769         journal->j_head++;
770         journal->j_free--;
771         if (journal->j_head == journal->j_last)
772                 journal->j_head = journal->j_first;
773         write_unlock(&journal->j_state_lock);
774         return jbd2_journal_bmap(journal, blocknr, retp);
775 }
776 
777 /*
778  * Conversion of logical to physical block numbers for the journal
779  *
780  * On external journals the journal blocks are identity-mapped, so
781  * this is a no-op.  If needed, we can use j_blk_offset - everything is
782  * ready.
783  */
784 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
785                  unsigned long long *retp)
786 {
787         int err = 0;
788         unsigned long long ret;
789 
790         if (journal->j_inode) {
791                 ret = bmap(journal->j_inode, blocknr);
792                 if (ret)
793                         *retp = ret;
794                 else {
795                         printk(KERN_ALERT "%s: journal block not found "
796                                         "at offset %lu on %s\n",
797                                __func__, blocknr, journal->j_devname);
798                         err = -EIO;
799                         __journal_abort_soft(journal, err);
800                 }
801         } else {
802                 *retp = blocknr; /* +journal->j_blk_offset */
803         }
804         return err;
805 }
806 
807 /*
808  * We play buffer_head aliasing tricks to write data/metadata blocks to
809  * the journal without copying their contents, but for journal
810  * descriptor blocks we do need to generate bona fide buffers.
811  *
812  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
813  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
814  * But we don't bother doing that, so there will be coherency problems with
815  * mmaps of blockdevs which hold live JBD-controlled filesystems.
816  */
817 struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
818 {
819         struct buffer_head *bh;
820         unsigned long long blocknr;
821         int err;
822 
823         err = jbd2_journal_next_log_block(journal, &blocknr);
824 
825         if (err)
826                 return NULL;
827 
828         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
829         if (!bh)
830                 return NULL;
831         lock_buffer(bh);
832         memset(bh->b_data, 0, journal->j_blocksize);
833         set_buffer_uptodate(bh);
834         unlock_buffer(bh);
835         BUFFER_TRACE(bh, "return this buffer");
836         return bh;
837 }
838 
839 /*
840  * Return tid of the oldest transaction in the journal and block in the journal
841  * where the transaction starts.
842  *
843  * If the journal is now empty, return which will be the next transaction ID
844  * we will write and where will that transaction start.
845  *
846  * The return value is 0 if journal tail cannot be pushed any further, 1 if
847  * it can.
848  */
849 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
850                               unsigned long *block)
851 {
852         transaction_t *transaction;
853         int ret;
854 
855         read_lock(&journal->j_state_lock);
856         spin_lock(&journal->j_list_lock);
857         transaction = journal->j_checkpoint_transactions;
858         if (transaction) {
859                 *tid = transaction->t_tid;
860                 *block = transaction->t_log_start;
861         } else if ((transaction = journal->j_committing_transaction) != NULL) {
862                 *tid = transaction->t_tid;
863                 *block = transaction->t_log_start;
864         } else if ((transaction = journal->j_running_transaction) != NULL) {
865                 *tid = transaction->t_tid;
866                 *block = journal->j_head;
867         } else {
868                 *tid = journal->j_transaction_sequence;
869                 *block = journal->j_head;
870         }
871         ret = tid_gt(*tid, journal->j_tail_sequence);
872         spin_unlock(&journal->j_list_lock);
873         read_unlock(&journal->j_state_lock);
874 
875         return ret;
876 }
877 
878 /*
879  * Update information in journal structure and in on disk journal superblock
880  * about log tail. This function does not check whether information passed in
881  * really pushes log tail further. It's responsibility of the caller to make
882  * sure provided log tail information is valid (e.g. by holding
883  * j_checkpoint_mutex all the time between computing log tail and calling this
884  * function as is the case with jbd2_cleanup_journal_tail()).
885  *
886  * Requires j_checkpoint_mutex
887  */
888 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
889 {
890         unsigned long freed;
891         int ret;
892 
893         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
894 
895         /*
896          * We cannot afford for write to remain in drive's caches since as
897          * soon as we update j_tail, next transaction can start reusing journal
898          * space and if we lose sb update during power failure we'd replay
899          * old transaction with possibly newly overwritten data.
900          */
901         ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
902         if (ret)
903                 goto out;
904 
905         write_lock(&journal->j_state_lock);
906         freed = block - journal->j_tail;
907         if (block < journal->j_tail)
908                 freed += journal->j_last - journal->j_first;
909 
910         trace_jbd2_update_log_tail(journal, tid, block, freed);
911         jbd_debug(1,
912                   "Cleaning journal tail from %d to %d (offset %lu), "
913                   "freeing %lu\n",
914                   journal->j_tail_sequence, tid, block, freed);
915 
916         journal->j_free += freed;
917         journal->j_tail_sequence = tid;
918         journal->j_tail = block;
919         write_unlock(&journal->j_state_lock);
920 
921 out:
922         return ret;
923 }
924 
925 /*
926  * This is a variation of __jbd2_update_log_tail which checks for validity of
927  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
928  * with other threads updating log tail.
929  */
930 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
931 {
932         mutex_lock(&journal->j_checkpoint_mutex);
933         if (tid_gt(tid, journal->j_tail_sequence))
934                 __jbd2_update_log_tail(journal, tid, block);
935         mutex_unlock(&journal->j_checkpoint_mutex);
936 }
937 
938 struct jbd2_stats_proc_session {
939         journal_t *journal;
940         struct transaction_stats_s *stats;
941         int start;
942         int max;
943 };
944 
945 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
946 {
947         return *pos ? NULL : SEQ_START_TOKEN;
948 }
949 
950 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
951 {
952         return NULL;
953 }
954 
955 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
956 {
957         struct jbd2_stats_proc_session *s = seq->private;
958 
959         if (v != SEQ_START_TOKEN)
960                 return 0;
961         seq_printf(seq, "%lu transactions (%lu requested), "
962                    "each up to %u blocks\n",
963                    s->stats->ts_tid, s->stats->ts_requested,
964                    s->journal->j_max_transaction_buffers);
965         if (s->stats->ts_tid == 0)
966                 return 0;
967         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
968             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
969         seq_printf(seq, "  %ums request delay\n",
970             (s->stats->ts_requested == 0) ? 0 :
971             jiffies_to_msecs(s->stats->run.rs_request_delay /
972                              s->stats->ts_requested));
973         seq_printf(seq, "  %ums running transaction\n",
974             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
975         seq_printf(seq, "  %ums transaction was being locked\n",
976             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
977         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
978             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
979         seq_printf(seq, "  %ums logging transaction\n",
980             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
981         seq_printf(seq, "  %lluus average transaction commit time\n",
982                    div_u64(s->journal->j_average_commit_time, 1000));
983         seq_printf(seq, "  %lu handles per transaction\n",
984             s->stats->run.rs_handle_count / s->stats->ts_tid);
985         seq_printf(seq, "  %lu blocks per transaction\n",
986             s->stats->run.rs_blocks / s->stats->ts_tid);
987         seq_printf(seq, "  %lu logged blocks per transaction\n",
988             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
989         return 0;
990 }
991 
992 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
993 {
994 }
995 
996 static const struct seq_operations jbd2_seq_info_ops = {
997         .start  = jbd2_seq_info_start,
998         .next   = jbd2_seq_info_next,
999         .stop   = jbd2_seq_info_stop,
1000         .show   = jbd2_seq_info_show,
1001 };
1002 
1003 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1004 {
1005         journal_t *journal = PDE_DATA(inode);
1006         struct jbd2_stats_proc_session *s;
1007         int rc, size;
1008 
1009         s = kmalloc(sizeof(*s), GFP_KERNEL);
1010         if (s == NULL)
1011                 return -ENOMEM;
1012         size = sizeof(struct transaction_stats_s);
1013         s->stats = kmalloc(size, GFP_KERNEL);
1014         if (s->stats == NULL) {
1015                 kfree(s);
1016                 return -ENOMEM;
1017         }
1018         spin_lock(&journal->j_history_lock);
1019         memcpy(s->stats, &journal->j_stats, size);
1020         s->journal = journal;
1021         spin_unlock(&journal->j_history_lock);
1022 
1023         rc = seq_open(file, &jbd2_seq_info_ops);
1024         if (rc == 0) {
1025                 struct seq_file *m = file->private_data;
1026                 m->private = s;
1027         } else {
1028                 kfree(s->stats);
1029                 kfree(s);
1030         }
1031         return rc;
1032 
1033 }
1034 
1035 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1036 {
1037         struct seq_file *seq = file->private_data;
1038         struct jbd2_stats_proc_session *s = seq->private;
1039         kfree(s->stats);
1040         kfree(s);
1041         return seq_release(inode, file);
1042 }
1043 
1044 static const struct file_operations jbd2_seq_info_fops = {
1045         .owner          = THIS_MODULE,
1046         .open           = jbd2_seq_info_open,
1047         .read           = seq_read,
1048         .llseek         = seq_lseek,
1049         .release        = jbd2_seq_info_release,
1050 };
1051 
1052 static struct proc_dir_entry *proc_jbd2_stats;
1053 
1054 static void jbd2_stats_proc_init(journal_t *journal)
1055 {
1056         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1057         if (journal->j_proc_entry) {
1058                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1059                                  &jbd2_seq_info_fops, journal);
1060         }
1061 }
1062 
1063 static void jbd2_stats_proc_exit(journal_t *journal)
1064 {
1065         remove_proc_entry("info", journal->j_proc_entry);
1066         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1067 }
1068 
1069 /*
1070  * Management for journal control blocks: functions to create and
1071  * destroy journal_t structures, and to initialise and read existing
1072  * journal blocks from disk.  */
1073 
1074 /* First: create and setup a journal_t object in memory.  We initialise
1075  * very few fields yet: that has to wait until we have created the
1076  * journal structures from from scratch, or loaded them from disk. */
1077 
1078 static journal_t * journal_init_common (void)
1079 {
1080         journal_t *journal;
1081         int err;
1082 
1083         journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1084         if (!journal)
1085                 return NULL;
1086 
1087         init_waitqueue_head(&journal->j_wait_transaction_locked);
1088         init_waitqueue_head(&journal->j_wait_done_commit);
1089         init_waitqueue_head(&journal->j_wait_commit);
1090         init_waitqueue_head(&journal->j_wait_updates);
1091         init_waitqueue_head(&journal->j_wait_reserved);
1092         mutex_init(&journal->j_barrier);
1093         mutex_init(&journal->j_checkpoint_mutex);
1094         spin_lock_init(&journal->j_revoke_lock);
1095         spin_lock_init(&journal->j_list_lock);
1096         rwlock_init(&journal->j_state_lock);
1097 
1098         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1099         journal->j_min_batch_time = 0;
1100         journal->j_max_batch_time = 15000; /* 15ms */
1101         atomic_set(&journal->j_reserved_credits, 0);
1102 
1103         /* The journal is marked for error until we succeed with recovery! */
1104         journal->j_flags = JBD2_ABORT;
1105 
1106         /* Set up a default-sized revoke table for the new mount. */
1107         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1108         if (err) {
1109                 kfree(journal);
1110                 return NULL;
1111         }
1112 
1113         spin_lock_init(&journal->j_history_lock);
1114 
1115         return journal;
1116 }
1117 
1118 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1119  *
1120  * Create a journal structure assigned some fixed set of disk blocks to
1121  * the journal.  We don't actually touch those disk blocks yet, but we
1122  * need to set up all of the mapping information to tell the journaling
1123  * system where the journal blocks are.
1124  *
1125  */
1126 
1127 /**
1128  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1129  *  @bdev: Block device on which to create the journal
1130  *  @fs_dev: Device which hold journalled filesystem for this journal.
1131  *  @start: Block nr Start of journal.
1132  *  @len:  Length of the journal in blocks.
1133  *  @blocksize: blocksize of journalling device
1134  *
1135  *  Returns: a newly created journal_t *
1136  *
1137  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1138  *  range of blocks on an arbitrary block device.
1139  *
1140  */
1141 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1142                         struct block_device *fs_dev,
1143                         unsigned long long start, int len, int blocksize)
1144 {
1145         journal_t *journal = journal_init_common();
1146         struct buffer_head *bh;
1147         char *p;
1148         int n;
1149 
1150         if (!journal)
1151                 return NULL;
1152 
1153         /* journal descriptor can store up to n blocks -bzzz */
1154         journal->j_blocksize = blocksize;
1155         journal->j_dev = bdev;
1156         journal->j_fs_dev = fs_dev;
1157         journal->j_blk_offset = start;
1158         journal->j_maxlen = len;
1159         bdevname(journal->j_dev, journal->j_devname);
1160         p = journal->j_devname;
1161         while ((p = strchr(p, '/')))
1162                 *p = '!';
1163         jbd2_stats_proc_init(journal);
1164         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1165         journal->j_wbufsize = n;
1166         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1167         if (!journal->j_wbuf) {
1168                 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1169                         __func__);
1170                 goto out_err;
1171         }
1172 
1173         bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1174         if (!bh) {
1175                 printk(KERN_ERR
1176                        "%s: Cannot get buffer for journal superblock\n",
1177                        __func__);
1178                 goto out_err;
1179         }
1180         journal->j_sb_buffer = bh;
1181         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1182 
1183         return journal;
1184 out_err:
1185         kfree(journal->j_wbuf);
1186         jbd2_stats_proc_exit(journal);
1187         kfree(journal);
1188         return NULL;
1189 }
1190 
1191 /**
1192  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1193  *  @inode: An inode to create the journal in
1194  *
1195  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1196  * the journal.  The inode must exist already, must support bmap() and
1197  * must have all data blocks preallocated.
1198  */
1199 journal_t * jbd2_journal_init_inode (struct inode *inode)
1200 {
1201         struct buffer_head *bh;
1202         journal_t *journal = journal_init_common();
1203         char *p;
1204         int err;
1205         int n;
1206         unsigned long long blocknr;
1207 
1208         if (!journal)
1209                 return NULL;
1210 
1211         journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1212         journal->j_inode = inode;
1213         bdevname(journal->j_dev, journal->j_devname);
1214         p = journal->j_devname;
1215         while ((p = strchr(p, '/')))
1216                 *p = '!';
1217         p = journal->j_devname + strlen(journal->j_devname);
1218         sprintf(p, "-%lu", journal->j_inode->i_ino);
1219         jbd_debug(1,
1220                   "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1221                   journal, inode->i_sb->s_id, inode->i_ino,
1222                   (long long) inode->i_size,
1223                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1224 
1225         journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1226         journal->j_blocksize = inode->i_sb->s_blocksize;
1227         jbd2_stats_proc_init(journal);
1228 
1229         /* journal descriptor can store up to n blocks -bzzz */
1230         n = journal->j_blocksize / sizeof(journal_block_tag_t);
1231         journal->j_wbufsize = n;
1232         journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1233         if (!journal->j_wbuf) {
1234                 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1235                         __func__);
1236                 goto out_err;
1237         }
1238 
1239         err = jbd2_journal_bmap(journal, 0, &blocknr);
1240         /* If that failed, give up */
1241         if (err) {
1242                 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1243                        __func__);
1244                 goto out_err;
1245         }
1246 
1247         bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1248         if (!bh) {
1249                 printk(KERN_ERR
1250                        "%s: Cannot get buffer for journal superblock\n",
1251                        __func__);
1252                 goto out_err;
1253         }
1254         journal->j_sb_buffer = bh;
1255         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1256 
1257         return journal;
1258 out_err:
1259         kfree(journal->j_wbuf);
1260         jbd2_stats_proc_exit(journal);
1261         kfree(journal);
1262         return NULL;
1263 }
1264 
1265 /*
1266  * If the journal init or create aborts, we need to mark the journal
1267  * superblock as being NULL to prevent the journal destroy from writing
1268  * back a bogus superblock.
1269  */
1270 static void journal_fail_superblock (journal_t *journal)
1271 {
1272         struct buffer_head *bh = journal->j_sb_buffer;
1273         brelse(bh);
1274         journal->j_sb_buffer = NULL;
1275 }
1276 
1277 /*
1278  * Given a journal_t structure, initialise the various fields for
1279  * startup of a new journaling session.  We use this both when creating
1280  * a journal, and after recovering an old journal to reset it for
1281  * subsequent use.
1282  */
1283 
1284 static int journal_reset(journal_t *journal)
1285 {
1286         journal_superblock_t *sb = journal->j_superblock;
1287         unsigned long long first, last;
1288 
1289         first = be32_to_cpu(sb->s_first);
1290         last = be32_to_cpu(sb->s_maxlen);
1291         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1292                 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1293                        first, last);
1294                 journal_fail_superblock(journal);
1295                 return -EINVAL;
1296         }
1297 
1298         journal->j_first = first;
1299         journal->j_last = last;
1300 
1301         journal->j_head = first;
1302         journal->j_tail = first;
1303         journal->j_free = last - first;
1304 
1305         journal->j_tail_sequence = journal->j_transaction_sequence;
1306         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1307         journal->j_commit_request = journal->j_commit_sequence;
1308 
1309         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1310 
1311         /*
1312          * As a special case, if the on-disk copy is already marked as needing
1313          * no recovery (s_start == 0), then we can safely defer the superblock
1314          * update until the next commit by setting JBD2_FLUSHED.  This avoids
1315          * attempting a write to a potential-readonly device.
1316          */
1317         if (sb->s_start == 0) {
1318                 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1319                         "(start %ld, seq %d, errno %d)\n",
1320                         journal->j_tail, journal->j_tail_sequence,
1321                         journal->j_errno);
1322                 journal->j_flags |= JBD2_FLUSHED;
1323         } else {
1324                 /* Lock here to make assertions happy... */
1325                 mutex_lock(&journal->j_checkpoint_mutex);
1326                 /*
1327                  * Update log tail information. We use WRITE_FUA since new
1328                  * transaction will start reusing journal space and so we
1329                  * must make sure information about current log tail is on
1330                  * disk before that.
1331                  */
1332                 jbd2_journal_update_sb_log_tail(journal,
1333                                                 journal->j_tail_sequence,
1334                                                 journal->j_tail,
1335                                                 WRITE_FUA);
1336                 mutex_unlock(&journal->j_checkpoint_mutex);
1337         }
1338         return jbd2_journal_start_thread(journal);
1339 }
1340 
1341 static int jbd2_write_superblock(journal_t *journal, int write_op)
1342 {
1343         struct buffer_head *bh = journal->j_sb_buffer;
1344         journal_superblock_t *sb = journal->j_superblock;
1345         int ret;
1346 
1347         trace_jbd2_write_superblock(journal, write_op);
1348         if (!(journal->j_flags & JBD2_BARRIER))
1349                 write_op &= ~(REQ_FUA | REQ_FLUSH);
1350         lock_buffer(bh);
1351         if (buffer_write_io_error(bh)) {
1352                 /*
1353                  * Oh, dear.  A previous attempt to write the journal
1354                  * superblock failed.  This could happen because the
1355                  * USB device was yanked out.  Or it could happen to
1356                  * be a transient write error and maybe the block will
1357                  * be remapped.  Nothing we can do but to retry the
1358                  * write and hope for the best.
1359                  */
1360                 printk(KERN_ERR "JBD2: previous I/O error detected "
1361                        "for journal superblock update for %s.\n",
1362                        journal->j_devname);
1363                 clear_buffer_write_io_error(bh);
1364                 set_buffer_uptodate(bh);
1365         }
1366         jbd2_superblock_csum_set(journal, sb);
1367         get_bh(bh);
1368         bh->b_end_io = end_buffer_write_sync;
1369         ret = submit_bh(write_op, bh);
1370         wait_on_buffer(bh);
1371         if (buffer_write_io_error(bh)) {
1372                 clear_buffer_write_io_error(bh);
1373                 set_buffer_uptodate(bh);
1374                 ret = -EIO;
1375         }
1376         if (ret) {
1377                 printk(KERN_ERR "JBD2: Error %d detected when updating "
1378                        "journal superblock for %s.\n", ret,
1379                        journal->j_devname);
1380                 jbd2_journal_abort(journal, ret);
1381         }
1382 
1383         return ret;
1384 }
1385 
1386 /**
1387  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1388  * @journal: The journal to update.
1389  * @tail_tid: TID of the new transaction at the tail of the log
1390  * @tail_block: The first block of the transaction at the tail of the log
1391  * @write_op: With which operation should we write the journal sb
1392  *
1393  * Update a journal's superblock information about log tail and write it to
1394  * disk, waiting for the IO to complete.
1395  */
1396 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1397                                      unsigned long tail_block, int write_op)
1398 {
1399         journal_superblock_t *sb = journal->j_superblock;
1400         int ret;
1401 
1402         if (is_journal_aborted(journal))
1403                 return -EIO;
1404 
1405         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1406         jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1407                   tail_block, tail_tid);
1408 
1409         sb->s_sequence = cpu_to_be32(tail_tid);
1410         sb->s_start    = cpu_to_be32(tail_block);
1411 
1412         ret = jbd2_write_superblock(journal, write_op);
1413         if (ret)
1414                 goto out;
1415 
1416         /* Log is no longer empty */
1417         write_lock(&journal->j_state_lock);
1418         WARN_ON(!sb->s_sequence);
1419         journal->j_flags &= ~JBD2_FLUSHED;
1420         write_unlock(&journal->j_state_lock);
1421 
1422 out:
1423         return ret;
1424 }
1425 
1426 /**
1427  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1428  * @journal: The journal to update.
1429  * @write_op: With which operation should we write the journal sb
1430  *
1431  * Update a journal's dynamic superblock fields to show that journal is empty.
1432  * Write updated superblock to disk waiting for IO to complete.
1433  */
1434 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1435 {
1436         journal_superblock_t *sb = journal->j_superblock;
1437 
1438         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1439         read_lock(&journal->j_state_lock);
1440         /* Is it already empty? */
1441         if (sb->s_start == 0) {
1442                 read_unlock(&journal->j_state_lock);
1443                 return;
1444         }
1445         jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1446                   journal->j_tail_sequence);
1447 
1448         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1449         sb->s_start    = cpu_to_be32(0);
1450         read_unlock(&journal->j_state_lock);
1451 
1452         jbd2_write_superblock(journal, write_op);
1453 
1454         /* Log is no longer empty */
1455         write_lock(&journal->j_state_lock);
1456         journal->j_flags |= JBD2_FLUSHED;
1457         write_unlock(&journal->j_state_lock);
1458 }
1459 
1460 
1461 /**
1462  * jbd2_journal_update_sb_errno() - Update error in the journal.
1463  * @journal: The journal to update.
1464  *
1465  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1466  * to complete.
1467  */
1468 void jbd2_journal_update_sb_errno(journal_t *journal)
1469 {
1470         journal_superblock_t *sb = journal->j_superblock;
1471 
1472         read_lock(&journal->j_state_lock);
1473         jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1474                   journal->j_errno);
1475         sb->s_errno    = cpu_to_be32(journal->j_errno);
1476         read_unlock(&journal->j_state_lock);
1477 
1478         jbd2_write_superblock(journal, WRITE_SYNC);
1479 }
1480 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1481 
1482 /*
1483  * Read the superblock for a given journal, performing initial
1484  * validation of the format.
1485  */
1486 static int journal_get_superblock(journal_t *journal)
1487 {
1488         struct buffer_head *bh;
1489         journal_superblock_t *sb;
1490         int err = -EIO;
1491 
1492         bh = journal->j_sb_buffer;
1493 
1494         J_ASSERT(bh != NULL);
1495         if (!buffer_uptodate(bh)) {
1496                 ll_rw_block(READ, 1, &bh);
1497                 wait_on_buffer(bh);
1498                 if (!buffer_uptodate(bh)) {
1499                         printk(KERN_ERR
1500                                 "JBD2: IO error reading journal superblock\n");
1501                         goto out;
1502                 }
1503         }
1504 
1505         if (buffer_verified(bh))
1506                 return 0;
1507 
1508         sb = journal->j_superblock;
1509 
1510         err = -EINVAL;
1511 
1512         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1513             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1514                 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1515                 goto out;
1516         }
1517 
1518         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1519         case JBD2_SUPERBLOCK_V1:
1520                 journal->j_format_version = 1;
1521                 break;
1522         case JBD2_SUPERBLOCK_V2:
1523                 journal->j_format_version = 2;
1524                 break;
1525         default:
1526                 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1527                 goto out;
1528         }
1529 
1530         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1531                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1532         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1533                 printk(KERN_WARNING "JBD2: journal file too short\n");
1534                 goto out;
1535         }
1536 
1537         if (be32_to_cpu(sb->s_first) == 0 ||
1538             be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1539                 printk(KERN_WARNING
1540                         "JBD2: Invalid start block of journal: %u\n",
1541                         be32_to_cpu(sb->s_first));
1542                 goto out;
1543         }
1544 
1545         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2) &&
1546             JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1547                 /* Can't have checksum v2 and v3 at the same time! */
1548                 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1549                        "at the same time!\n");
1550                 goto out;
1551         }
1552 
1553         if (jbd2_journal_has_csum_v2or3(journal) &&
1554             JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM)) {
1555                 /* Can't have checksum v1 and v2 on at the same time! */
1556                 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1557                        "at the same time!\n");
1558                 goto out;
1559         }
1560 
1561         if (!jbd2_verify_csum_type(journal, sb)) {
1562                 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1563                 goto out;
1564         }
1565 
1566         /* Load the checksum driver */
1567         if (jbd2_journal_has_csum_v2or3(journal)) {
1568                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1569                 if (IS_ERR(journal->j_chksum_driver)) {
1570                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1571                         err = PTR_ERR(journal->j_chksum_driver);
1572                         journal->j_chksum_driver = NULL;
1573                         goto out;
1574                 }
1575         }
1576 
1577         /* Check superblock checksum */
1578         if (!jbd2_superblock_csum_verify(journal, sb)) {
1579                 printk(KERN_ERR "JBD2: journal checksum error\n");
1580                 goto out;
1581         }
1582 
1583         /* Precompute checksum seed for all metadata */
1584         if (jbd2_journal_has_csum_v2or3(journal))
1585                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1586                                                    sizeof(sb->s_uuid));
1587 
1588         set_buffer_verified(bh);
1589 
1590         return 0;
1591 
1592 out:
1593         journal_fail_superblock(journal);
1594         return err;
1595 }
1596 
1597 /*
1598  * Load the on-disk journal superblock and read the key fields into the
1599  * journal_t.
1600  */
1601 
1602 static int load_superblock(journal_t *journal)
1603 {
1604         int err;
1605         journal_superblock_t *sb;
1606 
1607         err = journal_get_superblock(journal);
1608         if (err)
1609                 return err;
1610 
1611         sb = journal->j_superblock;
1612 
1613         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1614         journal->j_tail = be32_to_cpu(sb->s_start);
1615         journal->j_first = be32_to_cpu(sb->s_first);
1616         journal->j_last = be32_to_cpu(sb->s_maxlen);
1617         journal->j_errno = be32_to_cpu(sb->s_errno);
1618 
1619         return 0;
1620 }
1621 
1622 
1623 /**
1624  * int jbd2_journal_load() - Read journal from disk.
1625  * @journal: Journal to act on.
1626  *
1627  * Given a journal_t structure which tells us which disk blocks contain
1628  * a journal, read the journal from disk to initialise the in-memory
1629  * structures.
1630  */
1631 int jbd2_journal_load(journal_t *journal)
1632 {
1633         int err;
1634         journal_superblock_t *sb;
1635 
1636         err = load_superblock(journal);
1637         if (err)
1638                 return err;
1639 
1640         sb = journal->j_superblock;
1641         /* If this is a V2 superblock, then we have to check the
1642          * features flags on it. */
1643 
1644         if (journal->j_format_version >= 2) {
1645                 if ((sb->s_feature_ro_compat &
1646                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1647                     (sb->s_feature_incompat &
1648                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1649                         printk(KERN_WARNING
1650                                 "JBD2: Unrecognised features on journal\n");
1651                         return -EINVAL;
1652                 }
1653         }
1654 
1655         /*
1656          * Create a slab for this blocksize
1657          */
1658         err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1659         if (err)
1660                 return err;
1661 
1662         /* Let the recovery code check whether it needs to recover any
1663          * data from the journal. */
1664         if (jbd2_journal_recover(journal))
1665                 goto recovery_error;
1666 
1667         if (journal->j_failed_commit) {
1668                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1669                        "is corrupt.\n", journal->j_failed_commit,
1670                        journal->j_devname);
1671                 return -EIO;
1672         }
1673 
1674         /* OK, we've finished with the dynamic journal bits:
1675          * reinitialise the dynamic contents of the superblock in memory
1676          * and reset them on disk. */
1677         if (journal_reset(journal))
1678                 goto recovery_error;
1679 
1680         journal->j_flags &= ~JBD2_ABORT;
1681         journal->j_flags |= JBD2_LOADED;
1682         return 0;
1683 
1684 recovery_error:
1685         printk(KERN_WARNING "JBD2: recovery failed\n");
1686         return -EIO;
1687 }
1688 
1689 /**
1690  * void jbd2_journal_destroy() - Release a journal_t structure.
1691  * @journal: Journal to act on.
1692  *
1693  * Release a journal_t structure once it is no longer in use by the
1694  * journaled object.
1695  * Return <0 if we couldn't clean up the journal.
1696  */
1697 int jbd2_journal_destroy(journal_t *journal)
1698 {
1699         int err = 0;
1700 
1701         /* Wait for the commit thread to wake up and die. */
1702         journal_kill_thread(journal);
1703 
1704         /* Force a final log commit */
1705         if (journal->j_running_transaction)
1706                 jbd2_journal_commit_transaction(journal);
1707 
1708         /* Force any old transactions to disk */
1709 
1710         /* Totally anal locking here... */
1711         spin_lock(&journal->j_list_lock);
1712         while (journal->j_checkpoint_transactions != NULL) {
1713                 spin_unlock(&journal->j_list_lock);
1714                 mutex_lock(&journal->j_checkpoint_mutex);
1715                 err = jbd2_log_do_checkpoint(journal);
1716                 mutex_unlock(&journal->j_checkpoint_mutex);
1717                 /*
1718                  * If checkpointing failed, just free the buffers to avoid
1719                  * looping forever
1720                  */
1721                 if (err) {
1722                         jbd2_journal_destroy_checkpoint(journal);
1723                         spin_lock(&journal->j_list_lock);
1724                         break;
1725                 }
1726                 spin_lock(&journal->j_list_lock);
1727         }
1728 
1729         J_ASSERT(journal->j_running_transaction == NULL);
1730         J_ASSERT(journal->j_committing_transaction == NULL);
1731         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1732         spin_unlock(&journal->j_list_lock);
1733 
1734         if (journal->j_sb_buffer) {
1735                 if (!is_journal_aborted(journal)) {
1736                         mutex_lock(&journal->j_checkpoint_mutex);
1737 
1738                         write_lock(&journal->j_state_lock);
1739                         journal->j_tail_sequence =
1740                                 ++journal->j_transaction_sequence;
1741                         write_unlock(&journal->j_state_lock);
1742 
1743                         jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA);
1744                         mutex_unlock(&journal->j_checkpoint_mutex);
1745                 } else
1746                         err = -EIO;
1747                 brelse(journal->j_sb_buffer);
1748         }
1749 
1750         if (journal->j_proc_entry)
1751                 jbd2_stats_proc_exit(journal);
1752         if (journal->j_inode)
1753                 iput(journal->j_inode);
1754         if (journal->j_revoke)
1755                 jbd2_journal_destroy_revoke(journal);
1756         if (journal->j_chksum_driver)
1757                 crypto_free_shash(journal->j_chksum_driver);
1758         kfree(journal->j_wbuf);
1759         kfree(journal);
1760 
1761         return err;
1762 }
1763 
1764 
1765 /**
1766  *int jbd2_journal_check_used_features () - Check if features specified are used.
1767  * @journal: Journal to check.
1768  * @compat: bitmask of compatible features
1769  * @ro: bitmask of features that force read-only mount
1770  * @incompat: bitmask of incompatible features
1771  *
1772  * Check whether the journal uses all of a given set of
1773  * features.  Return true (non-zero) if it does.
1774  **/
1775 
1776 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1777                                  unsigned long ro, unsigned long incompat)
1778 {
1779         journal_superblock_t *sb;
1780 
1781         if (!compat && !ro && !incompat)
1782                 return 1;
1783         /* Load journal superblock if it is not loaded yet. */
1784         if (journal->j_format_version == 0 &&
1785             journal_get_superblock(journal) != 0)
1786                 return 0;
1787         if (journal->j_format_version == 1)
1788                 return 0;
1789 
1790         sb = journal->j_superblock;
1791 
1792         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1793             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1794             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1795                 return 1;
1796 
1797         return 0;
1798 }
1799 
1800 /**
1801  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1802  * @journal: Journal to check.
1803  * @compat: bitmask of compatible features
1804  * @ro: bitmask of features that force read-only mount
1805  * @incompat: bitmask of incompatible features
1806  *
1807  * Check whether the journaling code supports the use of
1808  * all of a given set of features on this journal.  Return true
1809  * (non-zero) if it can. */
1810 
1811 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1812                                       unsigned long ro, unsigned long incompat)
1813 {
1814         if (!compat && !ro && !incompat)
1815                 return 1;
1816 
1817         /* We can support any known requested features iff the
1818          * superblock is in version 2.  Otherwise we fail to support any
1819          * extended sb features. */
1820 
1821         if (journal->j_format_version != 2)
1822                 return 0;
1823 
1824         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1825             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1826             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1827                 return 1;
1828 
1829         return 0;
1830 }
1831 
1832 /**
1833  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1834  * @journal: Journal to act on.
1835  * @compat: bitmask of compatible features
1836  * @ro: bitmask of features that force read-only mount
1837  * @incompat: bitmask of incompatible features
1838  *
1839  * Mark a given journal feature as present on the
1840  * superblock.  Returns true if the requested features could be set.
1841  *
1842  */
1843 
1844 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1845                           unsigned long ro, unsigned long incompat)
1846 {
1847 #define INCOMPAT_FEATURE_ON(f) \
1848                 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1849 #define COMPAT_FEATURE_ON(f) \
1850                 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1851         journal_superblock_t *sb;
1852 
1853         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1854                 return 1;
1855 
1856         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1857                 return 0;
1858 
1859         /* If enabling v2 checksums, turn on v3 instead */
1860         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1861                 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1862                 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1863         }
1864 
1865         /* Asking for checksumming v3 and v1?  Only give them v3. */
1866         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1867             compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1868                 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1869 
1870         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1871                   compat, ro, incompat);
1872 
1873         sb = journal->j_superblock;
1874 
1875         /* If enabling v3 checksums, update superblock */
1876         if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1877                 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1878                 sb->s_feature_compat &=
1879                         ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1880 
1881                 /* Load the checksum driver */
1882                 if (journal->j_chksum_driver == NULL) {
1883                         journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1884                                                                       0, 0);
1885                         if (IS_ERR(journal->j_chksum_driver)) {
1886                                 printk(KERN_ERR "JBD2: Cannot load crc32c "
1887                                        "driver.\n");
1888                                 journal->j_chksum_driver = NULL;
1889                                 return 0;
1890                         }
1891 
1892                         /* Precompute checksum seed for all metadata */
1893                         journal->j_csum_seed = jbd2_chksum(journal, ~0,
1894                                                            sb->s_uuid,
1895                                                            sizeof(sb->s_uuid));
1896                 }
1897         }
1898 
1899         /* If enabling v1 checksums, downgrade superblock */
1900         if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1901                 sb->s_feature_incompat &=
1902                         ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1903                                      JBD2_FEATURE_INCOMPAT_CSUM_V3);
1904 
1905         sb->s_feature_compat    |= cpu_to_be32(compat);
1906         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1907         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1908 
1909         return 1;
1910 #undef COMPAT_FEATURE_ON
1911 #undef INCOMPAT_FEATURE_ON
1912 }
1913 
1914 /*
1915  * jbd2_journal_clear_features () - Clear a given journal feature in the
1916  *                                  superblock
1917  * @journal: Journal to act on.
1918  * @compat: bitmask of compatible features
1919  * @ro: bitmask of features that force read-only mount
1920  * @incompat: bitmask of incompatible features
1921  *
1922  * Clear a given journal feature as present on the
1923  * superblock.
1924  */
1925 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1926                                 unsigned long ro, unsigned long incompat)
1927 {
1928         journal_superblock_t *sb;
1929 
1930         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1931                   compat, ro, incompat);
1932 
1933         sb = journal->j_superblock;
1934 
1935         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1936         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1937         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1938 }
1939 EXPORT_SYMBOL(jbd2_journal_clear_features);
1940 
1941 /**
1942  * int jbd2_journal_flush () - Flush journal
1943  * @journal: Journal to act on.
1944  *
1945  * Flush all data for a given journal to disk and empty the journal.
1946  * Filesystems can use this when remounting readonly to ensure that
1947  * recovery does not need to happen on remount.
1948  */
1949 
1950 int jbd2_journal_flush(journal_t *journal)
1951 {
1952         int err = 0;
1953         transaction_t *transaction = NULL;
1954 
1955         write_lock(&journal->j_state_lock);
1956 
1957         /* Force everything buffered to the log... */
1958         if (journal->j_running_transaction) {
1959                 transaction = journal->j_running_transaction;
1960                 __jbd2_log_start_commit(journal, transaction->t_tid);
1961         } else if (journal->j_committing_transaction)
1962                 transaction = journal->j_committing_transaction;
1963 
1964         /* Wait for the log commit to complete... */
1965         if (transaction) {
1966                 tid_t tid = transaction->t_tid;
1967 
1968                 write_unlock(&journal->j_state_lock);
1969                 jbd2_log_wait_commit(journal, tid);
1970         } else {
1971                 write_unlock(&journal->j_state_lock);
1972         }
1973 
1974         /* ...and flush everything in the log out to disk. */
1975         spin_lock(&journal->j_list_lock);
1976         while (!err && journal->j_checkpoint_transactions != NULL) {
1977                 spin_unlock(&journal->j_list_lock);
1978                 mutex_lock(&journal->j_checkpoint_mutex);
1979                 err = jbd2_log_do_checkpoint(journal);
1980                 mutex_unlock(&journal->j_checkpoint_mutex);
1981                 spin_lock(&journal->j_list_lock);
1982         }
1983         spin_unlock(&journal->j_list_lock);
1984 
1985         if (is_journal_aborted(journal))
1986                 return -EIO;
1987 
1988         mutex_lock(&journal->j_checkpoint_mutex);
1989         if (!err) {
1990                 err = jbd2_cleanup_journal_tail(journal);
1991                 if (err < 0) {
1992                         mutex_unlock(&journal->j_checkpoint_mutex);
1993                         goto out;
1994                 }
1995                 err = 0;
1996         }
1997 
1998         /* Finally, mark the journal as really needing no recovery.
1999          * This sets s_start==0 in the underlying superblock, which is
2000          * the magic code for a fully-recovered superblock.  Any future
2001          * commits of data to the journal will restore the current
2002          * s_start value. */
2003         jbd2_mark_journal_empty(journal, WRITE_FUA);
2004         mutex_unlock(&journal->j_checkpoint_mutex);
2005         write_lock(&journal->j_state_lock);
2006         J_ASSERT(!journal->j_running_transaction);
2007         J_ASSERT(!journal->j_committing_transaction);
2008         J_ASSERT(!journal->j_checkpoint_transactions);
2009         J_ASSERT(journal->j_head == journal->j_tail);
2010         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2011         write_unlock(&journal->j_state_lock);
2012 out:
2013         return err;
2014 }
2015 
2016 /**
2017  * int jbd2_journal_wipe() - Wipe journal contents
2018  * @journal: Journal to act on.
2019  * @write: flag (see below)
2020  *
2021  * Wipe out all of the contents of a journal, safely.  This will produce
2022  * a warning if the journal contains any valid recovery information.
2023  * Must be called between journal_init_*() and jbd2_journal_load().
2024  *
2025  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2026  * we merely suppress recovery.
2027  */
2028 
2029 int jbd2_journal_wipe(journal_t *journal, int write)
2030 {
2031         int err = 0;
2032 
2033         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2034 
2035         err = load_superblock(journal);
2036         if (err)
2037                 return err;
2038 
2039         if (!journal->j_tail)
2040                 goto no_recovery;
2041 
2042         printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2043                 write ? "Clearing" : "Ignoring");
2044 
2045         err = jbd2_journal_skip_recovery(journal);
2046         if (write) {
2047                 /* Lock to make assertions happy... */
2048                 mutex_lock(&journal->j_checkpoint_mutex);
2049                 jbd2_mark_journal_empty(journal, WRITE_FUA);
2050                 mutex_unlock(&journal->j_checkpoint_mutex);
2051         }
2052 
2053  no_recovery:
2054         return err;
2055 }
2056 
2057 /*
2058  * Journal abort has very specific semantics, which we describe
2059  * for journal abort.
2060  *
2061  * Two internal functions, which provide abort to the jbd layer
2062  * itself are here.
2063  */
2064 
2065 /*
2066  * Quick version for internal journal use (doesn't lock the journal).
2067  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2068  * and don't attempt to make any other journal updates.
2069  */
2070 void __jbd2_journal_abort_hard(journal_t *journal)
2071 {
2072         transaction_t *transaction;
2073 
2074         if (journal->j_flags & JBD2_ABORT)
2075                 return;
2076 
2077         printk(KERN_ERR "Aborting journal on device %s.\n",
2078                journal->j_devname);
2079 
2080         write_lock(&journal->j_state_lock);
2081         journal->j_flags |= JBD2_ABORT;
2082         transaction = journal->j_running_transaction;
2083         if (transaction)
2084                 __jbd2_log_start_commit(journal, transaction->t_tid);
2085         write_unlock(&journal->j_state_lock);
2086 }
2087 
2088 /* Soft abort: record the abort error status in the journal superblock,
2089  * but don't do any other IO. */
2090 static void __journal_abort_soft (journal_t *journal, int errno)
2091 {
2092         if (journal->j_flags & JBD2_ABORT)
2093                 return;
2094 
2095         if (!journal->j_errno)
2096                 journal->j_errno = errno;
2097 
2098         __jbd2_journal_abort_hard(journal);
2099 
2100         if (errno) {
2101                 jbd2_journal_update_sb_errno(journal);
2102                 write_lock(&journal->j_state_lock);
2103                 journal->j_flags |= JBD2_REC_ERR;
2104                 write_unlock(&journal->j_state_lock);
2105         }
2106 }
2107 
2108 /**
2109  * void jbd2_journal_abort () - Shutdown the journal immediately.
2110  * @journal: the journal to shutdown.
2111  * @errno:   an error number to record in the journal indicating
2112  *           the reason for the shutdown.
2113  *
2114  * Perform a complete, immediate shutdown of the ENTIRE
2115  * journal (not of a single transaction).  This operation cannot be
2116  * undone without closing and reopening the journal.
2117  *
2118  * The jbd2_journal_abort function is intended to support higher level error
2119  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2120  * mode.
2121  *
2122  * Journal abort has very specific semantics.  Any existing dirty,
2123  * unjournaled buffers in the main filesystem will still be written to
2124  * disk by bdflush, but the journaling mechanism will be suspended
2125  * immediately and no further transaction commits will be honoured.
2126  *
2127  * Any dirty, journaled buffers will be written back to disk without
2128  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2129  * filesystem, but we _do_ attempt to leave as much data as possible
2130  * behind for fsck to use for cleanup.
2131  *
2132  * Any attempt to get a new transaction handle on a journal which is in
2133  * ABORT state will just result in an -EROFS error return.  A
2134  * jbd2_journal_stop on an existing handle will return -EIO if we have
2135  * entered abort state during the update.
2136  *
2137  * Recursive transactions are not disturbed by journal abort until the
2138  * final jbd2_journal_stop, which will receive the -EIO error.
2139  *
2140  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2141  * which will be recorded (if possible) in the journal superblock.  This
2142  * allows a client to record failure conditions in the middle of a
2143  * transaction without having to complete the transaction to record the
2144  * failure to disk.  ext3_error, for example, now uses this
2145  * functionality.
2146  *
2147  * Errors which originate from within the journaling layer will NOT
2148  * supply an errno; a null errno implies that absolutely no further
2149  * writes are done to the journal (unless there are any already in
2150  * progress).
2151  *
2152  */
2153 
2154 void jbd2_journal_abort(journal_t *journal, int errno)
2155 {
2156         __journal_abort_soft(journal, errno);
2157 }
2158 
2159 /**
2160  * int jbd2_journal_errno () - returns the journal's error state.
2161  * @journal: journal to examine.
2162  *
2163  * This is the errno number set with jbd2_journal_abort(), the last
2164  * time the journal was mounted - if the journal was stopped
2165  * without calling abort this will be 0.
2166  *
2167  * If the journal has been aborted on this mount time -EROFS will
2168  * be returned.
2169  */
2170 int jbd2_journal_errno(journal_t *journal)
2171 {
2172         int err;
2173 
2174         read_lock(&journal->j_state_lock);
2175         if (journal->j_flags & JBD2_ABORT)
2176                 err = -EROFS;
2177         else
2178                 err = journal->j_errno;
2179         read_unlock(&journal->j_state_lock);
2180         return err;
2181 }
2182 
2183 /**
2184  * int jbd2_journal_clear_err () - clears the journal's error state
2185  * @journal: journal to act on.
2186  *
2187  * An error must be cleared or acked to take a FS out of readonly
2188  * mode.
2189  */
2190 int jbd2_journal_clear_err(journal_t *journal)
2191 {
2192         int err = 0;
2193 
2194         write_lock(&journal->j_state_lock);
2195         if (journal->j_flags & JBD2_ABORT)
2196                 err = -EROFS;
2197         else
2198                 journal->j_errno = 0;
2199         write_unlock(&journal->j_state_lock);
2200         return err;
2201 }
2202 
2203 /**
2204  * void jbd2_journal_ack_err() - Ack journal err.
2205  * @journal: journal to act on.
2206  *
2207  * An error must be cleared or acked to take a FS out of readonly
2208  * mode.
2209  */
2210 void jbd2_journal_ack_err(journal_t *journal)
2211 {
2212         write_lock(&journal->j_state_lock);
2213         if (journal->j_errno)
2214                 journal->j_flags |= JBD2_ACK_ERR;
2215         write_unlock(&journal->j_state_lock);
2216 }
2217 
2218 int jbd2_journal_blocks_per_page(struct inode *inode)
2219 {
2220         return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2221 }
2222 
2223 /*
2224  * helper functions to deal with 32 or 64bit block numbers.
2225  */
2226 size_t journal_tag_bytes(journal_t *journal)
2227 {
2228         size_t sz;
2229 
2230         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3))
2231                 return sizeof(journal_block_tag3_t);
2232 
2233         sz = sizeof(journal_block_tag_t);
2234 
2235         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
2236                 sz += sizeof(__u16);
2237 
2238         if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2239                 return sz;
2240         else
2241                 return sz - sizeof(__u32);
2242 }
2243 
2244 /*
2245  * JBD memory management
2246  *
2247  * These functions are used to allocate block-sized chunks of memory
2248  * used for making copies of buffer_head data.  Very often it will be
2249  * page-sized chunks of data, but sometimes it will be in
2250  * sub-page-size chunks.  (For example, 16k pages on Power systems
2251  * with a 4k block file system.)  For blocks smaller than a page, we
2252  * use a SLAB allocator.  There are slab caches for each block size,
2253  * which are allocated at mount time, if necessary, and we only free
2254  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2255  * this reason we don't need to a mutex to protect access to
2256  * jbd2_slab[] allocating or releasing memory; only in
2257  * jbd2_journal_create_slab().
2258  */
2259 #define JBD2_MAX_SLABS 8
2260 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2261 
2262 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2263         "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2264         "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2265 };
2266 
2267 
2268 static void jbd2_journal_destroy_slabs(void)
2269 {
2270         int i;
2271 
2272         for (i = 0; i < JBD2_MAX_SLABS; i++) {
2273                 if (jbd2_slab[i])
2274                         kmem_cache_destroy(jbd2_slab[i]);
2275                 jbd2_slab[i] = NULL;
2276         }
2277 }
2278 
2279 static int jbd2_journal_create_slab(size_t size)
2280 {
2281         static DEFINE_MUTEX(jbd2_slab_create_mutex);
2282         int i = order_base_2(size) - 10;
2283         size_t slab_size;
2284 
2285         if (size == PAGE_SIZE)
2286                 return 0;
2287 
2288         if (i >= JBD2_MAX_SLABS)
2289                 return -EINVAL;
2290 
2291         if (unlikely(i < 0))
2292                 i = 0;
2293         mutex_lock(&jbd2_slab_create_mutex);
2294         if (jbd2_slab[i]) {
2295                 mutex_unlock(&jbd2_slab_create_mutex);
2296                 return 0;       /* Already created */
2297         }
2298 
2299         slab_size = 1 << (i+10);
2300         jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2301                                          slab_size, 0, NULL);
2302         mutex_unlock(&jbd2_slab_create_mutex);
2303         if (!jbd2_slab[i]) {
2304                 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2305                 return -ENOMEM;
2306         }
2307         return 0;
2308 }
2309 
2310 static struct kmem_cache *get_slab(size_t size)
2311 {
2312         int i = order_base_2(size) - 10;
2313 
2314         BUG_ON(i >= JBD2_MAX_SLABS);
2315         if (unlikely(i < 0))
2316                 i = 0;
2317         BUG_ON(jbd2_slab[i] == NULL);
2318         return jbd2_slab[i];
2319 }
2320 
2321 void *jbd2_alloc(size_t size, gfp_t flags)
2322 {
2323         void *ptr;
2324 
2325         BUG_ON(size & (size-1)); /* Must be a power of 2 */
2326 
2327         flags |= __GFP_REPEAT;
2328         if (size == PAGE_SIZE)
2329                 ptr = (void *)__get_free_pages(flags, 0);
2330         else if (size > PAGE_SIZE) {
2331                 int order = get_order(size);
2332 
2333                 if (order < 3)
2334                         ptr = (void *)__get_free_pages(flags, order);
2335                 else
2336                         ptr = vmalloc(size);
2337         } else
2338                 ptr = kmem_cache_alloc(get_slab(size), flags);
2339 
2340         /* Check alignment; SLUB has gotten this wrong in the past,
2341          * and this can lead to user data corruption! */
2342         BUG_ON(((unsigned long) ptr) & (size-1));
2343 
2344         return ptr;
2345 }
2346 
2347 void jbd2_free(void *ptr, size_t size)
2348 {
2349         if (size == PAGE_SIZE) {
2350                 free_pages((unsigned long)ptr, 0);
2351                 return;
2352         }
2353         if (size > PAGE_SIZE) {
2354                 int order = get_order(size);
2355 
2356                 if (order < 3)
2357                         free_pages((unsigned long)ptr, order);
2358                 else
2359                         vfree(ptr);
2360                 return;
2361         }
2362         kmem_cache_free(get_slab(size), ptr);
2363 };
2364 
2365 /*
2366  * Journal_head storage management
2367  */
2368 static struct kmem_cache *jbd2_journal_head_cache;
2369 #ifdef CONFIG_JBD2_DEBUG
2370 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2371 #endif
2372 
2373 static int jbd2_journal_init_journal_head_cache(void)
2374 {
2375         int retval;
2376 
2377         J_ASSERT(jbd2_journal_head_cache == NULL);
2378         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2379                                 sizeof(struct journal_head),
2380                                 0,              /* offset */
2381                                 SLAB_TEMPORARY, /* flags */
2382                                 NULL);          /* ctor */
2383         retval = 0;
2384         if (!jbd2_journal_head_cache) {
2385                 retval = -ENOMEM;
2386                 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2387         }
2388         return retval;
2389 }
2390 
2391 static void jbd2_journal_destroy_journal_head_cache(void)
2392 {
2393         if (jbd2_journal_head_cache) {
2394                 kmem_cache_destroy(jbd2_journal_head_cache);
2395                 jbd2_journal_head_cache = NULL;
2396         }
2397 }
2398 
2399 /*
2400  * journal_head splicing and dicing
2401  */
2402 static struct journal_head *journal_alloc_journal_head(void)
2403 {
2404         struct journal_head *ret;
2405 
2406 #ifdef CONFIG_JBD2_DEBUG
2407         atomic_inc(&nr_journal_heads);
2408 #endif
2409         ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2410         if (!ret) {
2411                 jbd_debug(1, "out of memory for journal_head\n");
2412                 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2413                 while (!ret) {
2414                         yield();
2415                         ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2416                 }
2417         }
2418         return ret;
2419 }
2420 
2421 static void journal_free_journal_head(struct journal_head *jh)
2422 {
2423 #ifdef CONFIG_JBD2_DEBUG
2424         atomic_dec(&nr_journal_heads);
2425         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2426 #endif
2427         kmem_cache_free(jbd2_journal_head_cache, jh);
2428 }
2429 
2430 /*
2431  * A journal_head is attached to a buffer_head whenever JBD has an
2432  * interest in the buffer.
2433  *
2434  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2435  * is set.  This bit is tested in core kernel code where we need to take
2436  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2437  * there.
2438  *
2439  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2440  *
2441  * When a buffer has its BH_JBD bit set it is immune from being released by
2442  * core kernel code, mainly via ->b_count.
2443  *
2444  * A journal_head is detached from its buffer_head when the journal_head's
2445  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2446  * transaction (b_cp_transaction) hold their references to b_jcount.
2447  *
2448  * Various places in the kernel want to attach a journal_head to a buffer_head
2449  * _before_ attaching the journal_head to a transaction.  To protect the
2450  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2451  * journal_head's b_jcount refcount by one.  The caller must call
2452  * jbd2_journal_put_journal_head() to undo this.
2453  *
2454  * So the typical usage would be:
2455  *
2456  *      (Attach a journal_head if needed.  Increments b_jcount)
2457  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2458  *      ...
2459  *      (Get another reference for transaction)
2460  *      jbd2_journal_grab_journal_head(bh);
2461  *      jh->b_transaction = xxx;
2462  *      (Put original reference)
2463  *      jbd2_journal_put_journal_head(jh);
2464  */
2465 
2466 /*
2467  * Give a buffer_head a journal_head.
2468  *
2469  * May sleep.
2470  */
2471 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2472 {
2473         struct journal_head *jh;
2474         struct journal_head *new_jh = NULL;
2475 
2476 repeat:
2477         if (!buffer_jbd(bh))
2478                 new_jh = journal_alloc_journal_head();
2479 
2480         jbd_lock_bh_journal_head(bh);
2481         if (buffer_jbd(bh)) {
2482                 jh = bh2jh(bh);
2483         } else {
2484                 J_ASSERT_BH(bh,
2485                         (atomic_read(&bh->b_count) > 0) ||
2486                         (bh->b_page && bh->b_page->mapping));
2487 
2488                 if (!new_jh) {
2489                         jbd_unlock_bh_journal_head(bh);
2490                         goto repeat;
2491                 }
2492 
2493                 jh = new_jh;
2494                 new_jh = NULL;          /* We consumed it */
2495                 set_buffer_jbd(bh);
2496                 bh->b_private = jh;
2497                 jh->b_bh = bh;
2498                 get_bh(bh);
2499                 BUFFER_TRACE(bh, "added journal_head");
2500         }
2501         jh->b_jcount++;
2502         jbd_unlock_bh_journal_head(bh);
2503         if (new_jh)
2504                 journal_free_journal_head(new_jh);
2505         return bh->b_private;
2506 }
2507 
2508 /*
2509  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2510  * having a journal_head, return NULL
2511  */
2512 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2513 {
2514         struct journal_head *jh = NULL;
2515 
2516         jbd_lock_bh_journal_head(bh);
2517         if (buffer_jbd(bh)) {
2518                 jh = bh2jh(bh);
2519                 jh->b_jcount++;
2520         }
2521         jbd_unlock_bh_journal_head(bh);
2522         return jh;
2523 }
2524 
2525 static void __journal_remove_journal_head(struct buffer_head *bh)
2526 {
2527         struct journal_head *jh = bh2jh(bh);
2528 
2529         J_ASSERT_JH(jh, jh->b_jcount >= 0);
2530         J_ASSERT_JH(jh, jh->b_transaction == NULL);
2531         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2532         J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2533         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2534         J_ASSERT_BH(bh, buffer_jbd(bh));
2535         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2536         BUFFER_TRACE(bh, "remove journal_head");
2537         if (jh->b_frozen_data) {
2538                 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2539                 jbd2_free(jh->b_frozen_data, bh->b_size);
2540         }
2541         if (jh->b_committed_data) {
2542                 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2543                 jbd2_free(jh->b_committed_data, bh->b_size);
2544         }
2545         bh->b_private = NULL;
2546         jh->b_bh = NULL;        /* debug, really */
2547         clear_buffer_jbd(bh);
2548         journal_free_journal_head(jh);
2549 }
2550 
2551 /*
2552  * Drop a reference on the passed journal_head.  If it fell to zero then
2553  * release the journal_head from the buffer_head.
2554  */
2555 void jbd2_journal_put_journal_head(struct journal_head *jh)
2556 {
2557         struct buffer_head *bh = jh2bh(jh);
2558 
2559         jbd_lock_bh_journal_head(bh);
2560         J_ASSERT_JH(jh, jh->b_jcount > 0);
2561         --jh->b_jcount;
2562         if (!jh->b_jcount) {
2563                 __journal_remove_journal_head(bh);
2564                 jbd_unlock_bh_journal_head(bh);
2565                 __brelse(bh);
2566         } else
2567                 jbd_unlock_bh_journal_head(bh);
2568 }
2569 
2570 /*
2571  * Initialize jbd inode head
2572  */
2573 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2574 {
2575         jinode->i_transaction = NULL;
2576         jinode->i_next_transaction = NULL;
2577         jinode->i_vfs_inode = inode;
2578         jinode->i_flags = 0;
2579         INIT_LIST_HEAD(&jinode->i_list);
2580 }
2581 
2582 /*
2583  * Function to be called before we start removing inode from memory (i.e.,
2584  * clear_inode() is a fine place to be called from). It removes inode from
2585  * transaction's lists.
2586  */
2587 void jbd2_journal_release_jbd_inode(journal_t *journal,
2588                                     struct jbd2_inode *jinode)
2589 {
2590         if (!journal)
2591                 return;
2592 restart:
2593         spin_lock(&journal->j_list_lock);
2594         /* Is commit writing out inode - we have to wait */
2595         if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2596                 wait_queue_head_t *wq;
2597                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2598                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2599                 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2600                 spin_unlock(&journal->j_list_lock);
2601                 schedule();
2602                 finish_wait(wq, &wait.wait);
2603                 goto restart;
2604         }
2605 
2606         if (jinode->i_transaction) {
2607                 list_del(&jinode->i_list);
2608                 jinode->i_transaction = NULL;
2609         }
2610         spin_unlock(&journal->j_list_lock);
2611 }
2612 
2613 
2614 #ifdef CONFIG_PROC_FS
2615 
2616 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2617 
2618 static void __init jbd2_create_jbd_stats_proc_entry(void)
2619 {
2620         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2621 }
2622 
2623 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2624 {
2625         if (proc_jbd2_stats)
2626                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2627 }
2628 
2629 #else
2630 
2631 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2632 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2633 
2634 #endif
2635 
2636 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2637 
2638 static int __init jbd2_journal_init_handle_cache(void)
2639 {
2640         jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2641         if (jbd2_handle_cache == NULL) {
2642                 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2643                 return -ENOMEM;
2644         }
2645         jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2646         if (jbd2_inode_cache == NULL) {
2647                 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2648                 kmem_cache_destroy(jbd2_handle_cache);
2649                 return -ENOMEM;
2650         }
2651         return 0;
2652 }
2653 
2654 static void jbd2_journal_destroy_handle_cache(void)
2655 {
2656         if (jbd2_handle_cache)
2657                 kmem_cache_destroy(jbd2_handle_cache);
2658         if (jbd2_inode_cache)
2659                 kmem_cache_destroy(jbd2_inode_cache);
2660 
2661 }
2662 
2663 /*
2664  * Module startup and shutdown
2665  */
2666 
2667 static int __init journal_init_caches(void)
2668 {
2669         int ret;
2670 
2671         ret = jbd2_journal_init_revoke_caches();
2672         if (ret == 0)
2673                 ret = jbd2_journal_init_journal_head_cache();
2674         if (ret == 0)
2675                 ret = jbd2_journal_init_handle_cache();
2676         if (ret == 0)
2677                 ret = jbd2_journal_init_transaction_cache();
2678         return ret;
2679 }
2680 
2681 static void jbd2_journal_destroy_caches(void)
2682 {
2683         jbd2_journal_destroy_revoke_caches();
2684         jbd2_journal_destroy_journal_head_cache();
2685         jbd2_journal_destroy_handle_cache();
2686         jbd2_journal_destroy_transaction_cache();
2687         jbd2_journal_destroy_slabs();
2688 }
2689 
2690 static int __init journal_init(void)
2691 {
2692         int ret;
2693 
2694         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2695 
2696         ret = journal_init_caches();
2697         if (ret == 0) {
2698                 jbd2_create_jbd_stats_proc_entry();
2699         } else {
2700                 jbd2_journal_destroy_caches();
2701         }
2702         return ret;
2703 }
2704 
2705 static void __exit journal_exit(void)
2706 {
2707 #ifdef CONFIG_JBD2_DEBUG
2708         int n = atomic_read(&nr_journal_heads);
2709         if (n)
2710                 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2711 #endif
2712         jbd2_remove_jbd_stats_proc_entry();
2713         jbd2_journal_destroy_caches();
2714 }
2715 
2716 MODULE_LICENSE("GPL");
2717 module_init(journal_init);
2718 module_exit(journal_exit);
2719 
2720 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp