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

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

Version: ~ [ linux-6.1-rc7 ] ~ [ linux-6.0.10 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.80 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.156 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.225 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.267 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.300 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.334 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0+
  2 /*
  3  * linux/fs/jbd2/journal.c
  4  *
  5  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
  6  *
  7  * Copyright 1998 Red Hat corp --- All Rights Reserved
  8  *
  9  * Generic filesystem journal-writing code; part of the ext2fs
 10  * journaling system.
 11  *
 12  * This file manages journals: areas of disk reserved for logging
 13  * transactional updates.  This includes the kernel journaling thread
 14  * which is responsible for scheduling updates to the log.
 15  *
 16  * We do not actually manage the physical storage of the journal in this
 17  * file: that is left to a per-journal policy function, which allows us
 18  * to store the journal within a filesystem-specified area for ext2
 19  * journaling (ext2 can use a reserved inode for storing the log).
 20  */
 21 
 22 #include <linux/module.h>
 23 #include <linux/time.h>
 24 #include <linux/fs.h>
 25 #include <linux/jbd2.h>
 26 #include <linux/errno.h>
 27 #include <linux/slab.h>
 28 #include <linux/init.h>
 29 #include <linux/mm.h>
 30 #include <linux/freezer.h>
 31 #include <linux/pagemap.h>
 32 #include <linux/kthread.h>
 33 #include <linux/poison.h>
 34 #include <linux/proc_fs.h>
 35 #include <linux/seq_file.h>
 36 #include <linux/math64.h>
 37 #include <linux/hash.h>
 38 #include <linux/log2.h>
 39 #include <linux/vmalloc.h>
 40 #include <linux/backing-dev.h>
 41 #include <linux/bitops.h>
 42 #include <linux/ratelimit.h>
 43 #include <linux/sched/mm.h>
 44 
 45 #define CREATE_TRACE_POINTS
 46 #include <trace/events/jbd2.h>
 47 
 48 #include <linux/uaccess.h>
 49 #include <asm/page.h>
 50 
 51 #ifdef CONFIG_JBD2_DEBUG
 52 ushort jbd2_journal_enable_debug __read_mostly;
 53 EXPORT_SYMBOL(jbd2_journal_enable_debug);
 54 
 55 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
 56 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
 57 #endif
 58 
 59 EXPORT_SYMBOL(jbd2_journal_extend);
 60 EXPORT_SYMBOL(jbd2_journal_stop);
 61 EXPORT_SYMBOL(jbd2_journal_lock_updates);
 62 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
 63 EXPORT_SYMBOL(jbd2_journal_get_write_access);
 64 EXPORT_SYMBOL(jbd2_journal_get_create_access);
 65 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
 66 EXPORT_SYMBOL(jbd2_journal_set_triggers);
 67 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
 68 EXPORT_SYMBOL(jbd2_journal_forget);
 69 EXPORT_SYMBOL(jbd2_journal_flush);
 70 EXPORT_SYMBOL(jbd2_journal_revoke);
 71 
 72 EXPORT_SYMBOL(jbd2_journal_init_dev);
 73 EXPORT_SYMBOL(jbd2_journal_init_inode);
 74 EXPORT_SYMBOL(jbd2_journal_check_used_features);
 75 EXPORT_SYMBOL(jbd2_journal_check_available_features);
 76 EXPORT_SYMBOL(jbd2_journal_set_features);
 77 EXPORT_SYMBOL(jbd2_journal_load);
 78 EXPORT_SYMBOL(jbd2_journal_destroy);
 79 EXPORT_SYMBOL(jbd2_journal_abort);
 80 EXPORT_SYMBOL(jbd2_journal_errno);
 81 EXPORT_SYMBOL(jbd2_journal_ack_err);
 82 EXPORT_SYMBOL(jbd2_journal_clear_err);
 83 EXPORT_SYMBOL(jbd2_log_wait_commit);
 84 EXPORT_SYMBOL(jbd2_log_start_commit);
 85 EXPORT_SYMBOL(jbd2_journal_start_commit);
 86 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
 87 EXPORT_SYMBOL(jbd2_journal_wipe);
 88 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
 89 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
 90 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
 91 EXPORT_SYMBOL(jbd2_journal_force_commit);
 92 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
 93 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
 94 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
 95 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
 96 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
 97 EXPORT_SYMBOL(jbd2_inode_cache);
 98 
 99 static int jbd2_journal_create_slab(size_t slab_size);
100 
101 #ifdef CONFIG_JBD2_DEBUG
102 void __jbd2_debug(int level, const char *file, const char *func,
103                   unsigned int line, const char *fmt, ...)
104 {
105         struct va_format vaf;
106         va_list args;
107 
108         if (level > jbd2_journal_enable_debug)
109                 return;
110         va_start(args, fmt);
111         vaf.fmt = fmt;
112         vaf.va = &args;
113         printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
114         va_end(args);
115 }
116 EXPORT_SYMBOL(__jbd2_debug);
117 #endif
118 
119 /* Checksumming functions */
120 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
121 {
122         if (!jbd2_journal_has_csum_v2or3_feature(j))
123                 return 1;
124 
125         return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
126 }
127 
128 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
129 {
130         __u32 csum;
131         __be32 old_csum;
132 
133         old_csum = sb->s_checksum;
134         sb->s_checksum = 0;
135         csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
136         sb->s_checksum = old_csum;
137 
138         return cpu_to_be32(csum);
139 }
140 
141 /*
142  * Helper function used to manage commit timeouts
143  */
144 
145 static void commit_timeout(struct timer_list *t)
146 {
147         journal_t *journal = from_timer(journal, t, j_commit_timer);
148 
149         wake_up_process(journal->j_task);
150 }
151 
152 /*
153  * kjournald2: The main thread function used to manage a logging device
154  * journal.
155  *
156  * This kernel thread is responsible for two things:
157  *
158  * 1) COMMIT:  Every so often we need to commit the current state of the
159  *    filesystem to disk.  The journal thread is responsible for writing
160  *    all of the metadata buffers to disk.
161  *
162  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
163  *    of the data in that part of the log has been rewritten elsewhere on
164  *    the disk.  Flushing these old buffers to reclaim space in the log is
165  *    known as checkpointing, and this thread is responsible for that job.
166  */
167 
168 static int kjournald2(void *arg)
169 {
170         journal_t *journal = arg;
171         transaction_t *transaction;
172 
173         /*
174          * Set up an interval timer which can be used to trigger a commit wakeup
175          * after the commit interval expires
176          */
177         timer_setup(&journal->j_commit_timer, commit_timeout, 0);
178 
179         set_freezable();
180 
181         /* Record that the journal thread is running */
182         journal->j_task = current;
183         wake_up(&journal->j_wait_done_commit);
184 
185         /*
186          * Make sure that no allocations from this kernel thread will ever
187          * recurse to the fs layer because we are responsible for the
188          * transaction commit and any fs involvement might get stuck waiting for
189          * the trasn. commit.
190          */
191         memalloc_nofs_save();
192 
193         /*
194          * And now, wait forever for commit wakeup events.
195          */
196         write_lock(&journal->j_state_lock);
197 
198 loop:
199         if (journal->j_flags & JBD2_UNMOUNT)
200                 goto end_loop;
201 
202         jbd_debug(1, "commit_sequence=%u, commit_request=%u\n",
203                 journal->j_commit_sequence, journal->j_commit_request);
204 
205         if (journal->j_commit_sequence != journal->j_commit_request) {
206                 jbd_debug(1, "OK, requests differ\n");
207                 write_unlock(&journal->j_state_lock);
208                 del_timer_sync(&journal->j_commit_timer);
209                 jbd2_journal_commit_transaction(journal);
210                 write_lock(&journal->j_state_lock);
211                 goto loop;
212         }
213 
214         wake_up(&journal->j_wait_done_commit);
215         if (freezing(current)) {
216                 /*
217                  * The simpler the better. Flushing journal isn't a
218                  * good idea, because that depends on threads that may
219                  * be already stopped.
220                  */
221                 jbd_debug(1, "Now suspending kjournald2\n");
222                 write_unlock(&journal->j_state_lock);
223                 try_to_freeze();
224                 write_lock(&journal->j_state_lock);
225         } else {
226                 /*
227                  * We assume on resume that commits are already there,
228                  * so we don't sleep
229                  */
230                 DEFINE_WAIT(wait);
231                 int should_sleep = 1;
232 
233                 prepare_to_wait(&journal->j_wait_commit, &wait,
234                                 TASK_INTERRUPTIBLE);
235                 if (journal->j_commit_sequence != journal->j_commit_request)
236                         should_sleep = 0;
237                 transaction = journal->j_running_transaction;
238                 if (transaction && time_after_eq(jiffies,
239                                                 transaction->t_expires))
240                         should_sleep = 0;
241                 if (journal->j_flags & JBD2_UNMOUNT)
242                         should_sleep = 0;
243                 if (should_sleep) {
244                         write_unlock(&journal->j_state_lock);
245                         schedule();
246                         write_lock(&journal->j_state_lock);
247                 }
248                 finish_wait(&journal->j_wait_commit, &wait);
249         }
250 
251         jbd_debug(1, "kjournald2 wakes\n");
252 
253         /*
254          * Were we woken up by a commit wakeup event?
255          */
256         transaction = journal->j_running_transaction;
257         if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
258                 journal->j_commit_request = transaction->t_tid;
259                 jbd_debug(1, "woke because of timeout\n");
260         }
261         goto loop;
262 
263 end_loop:
264         del_timer_sync(&journal->j_commit_timer);
265         journal->j_task = NULL;
266         wake_up(&journal->j_wait_done_commit);
267         jbd_debug(1, "Journal thread exiting.\n");
268         write_unlock(&journal->j_state_lock);
269         return 0;
270 }
271 
272 static int jbd2_journal_start_thread(journal_t *journal)
273 {
274         struct task_struct *t;
275 
276         t = kthread_run(kjournald2, journal, "jbd2/%s",
277                         journal->j_devname);
278         if (IS_ERR(t))
279                 return PTR_ERR(t);
280 
281         wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
282         return 0;
283 }
284 
285 static void journal_kill_thread(journal_t *journal)
286 {
287         write_lock(&journal->j_state_lock);
288         journal->j_flags |= JBD2_UNMOUNT;
289 
290         while (journal->j_task) {
291                 write_unlock(&journal->j_state_lock);
292                 wake_up(&journal->j_wait_commit);
293                 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
294                 write_lock(&journal->j_state_lock);
295         }
296         write_unlock(&journal->j_state_lock);
297 }
298 
299 /*
300  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
301  *
302  * Writes a metadata buffer to a given disk block.  The actual IO is not
303  * performed but a new buffer_head is constructed which labels the data
304  * to be written with the correct destination disk block.
305  *
306  * Any magic-number escaping which needs to be done will cause a
307  * copy-out here.  If the buffer happens to start with the
308  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
309  * magic number is only written to the log for descripter blocks.  In
310  * this case, we copy the data and replace the first word with 0, and we
311  * return a result code which indicates that this buffer needs to be
312  * marked as an escaped buffer in the corresponding log descriptor
313  * block.  The missing word can then be restored when the block is read
314  * during recovery.
315  *
316  * If the source buffer has already been modified by a new transaction
317  * since we took the last commit snapshot, we use the frozen copy of
318  * that data for IO. If we end up using the existing buffer_head's data
319  * for the write, then we have to make sure nobody modifies it while the
320  * IO is in progress. do_get_write_access() handles this.
321  *
322  * The function returns a pointer to the buffer_head to be used for IO.
323  *
324  *
325  * Return value:
326  *  <0: Error
327  * >=0: Finished OK
328  *
329  * On success:
330  * Bit 0 set == escape performed on the data
331  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
332  */
333 
334 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
335                                   struct journal_head  *jh_in,
336                                   struct buffer_head **bh_out,
337                                   sector_t blocknr)
338 {
339         int need_copy_out = 0;
340         int done_copy_out = 0;
341         int do_escape = 0;
342         char *mapped_data;
343         struct buffer_head *new_bh;
344         struct page *new_page;
345         unsigned int new_offset;
346         struct buffer_head *bh_in = jh2bh(jh_in);
347         journal_t *journal = transaction->t_journal;
348 
349         /*
350          * The buffer really shouldn't be locked: only the current committing
351          * transaction is allowed to write it, so nobody else is allowed
352          * to do any IO.
353          *
354          * akpm: except if we're journalling data, and write() output is
355          * also part of a shared mapping, and another thread has
356          * decided to launch a writepage() against this buffer.
357          */
358         J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
359 
360         new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
361 
362         /* keep subsequent assertions sane */
363         atomic_set(&new_bh->b_count, 1);
364 
365         spin_lock(&jh_in->b_state_lock);
366 repeat:
367         /*
368          * If a new transaction has already done a buffer copy-out, then
369          * we use that version of the data for the commit.
370          */
371         if (jh_in->b_frozen_data) {
372                 done_copy_out = 1;
373                 new_page = virt_to_page(jh_in->b_frozen_data);
374                 new_offset = offset_in_page(jh_in->b_frozen_data);
375         } else {
376                 new_page = jh2bh(jh_in)->b_page;
377                 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
378         }
379 
380         mapped_data = kmap_atomic(new_page);
381         /*
382          * Fire data frozen trigger if data already wasn't frozen.  Do this
383          * before checking for escaping, as the trigger may modify the magic
384          * offset.  If a copy-out happens afterwards, it will have the correct
385          * data in the buffer.
386          */
387         if (!done_copy_out)
388                 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
389                                            jh_in->b_triggers);
390 
391         /*
392          * Check for escaping
393          */
394         if (*((__be32 *)(mapped_data + new_offset)) ==
395                                 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
396                 need_copy_out = 1;
397                 do_escape = 1;
398         }
399         kunmap_atomic(mapped_data);
400 
401         /*
402          * Do we need to do a data copy?
403          */
404         if (need_copy_out && !done_copy_out) {
405                 char *tmp;
406 
407                 spin_unlock(&jh_in->b_state_lock);
408                 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
409                 if (!tmp) {
410                         brelse(new_bh);
411                         return -ENOMEM;
412                 }
413                 spin_lock(&jh_in->b_state_lock);
414                 if (jh_in->b_frozen_data) {
415                         jbd2_free(tmp, bh_in->b_size);
416                         goto repeat;
417                 }
418 
419                 jh_in->b_frozen_data = tmp;
420                 mapped_data = kmap_atomic(new_page);
421                 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
422                 kunmap_atomic(mapped_data);
423 
424                 new_page = virt_to_page(tmp);
425                 new_offset = offset_in_page(tmp);
426                 done_copy_out = 1;
427 
428                 /*
429                  * This isn't strictly necessary, as we're using frozen
430                  * data for the escaping, but it keeps consistency with
431                  * b_frozen_data usage.
432                  */
433                 jh_in->b_frozen_triggers = jh_in->b_triggers;
434         }
435 
436         /*
437          * Did we need to do an escaping?  Now we've done all the
438          * copying, we can finally do so.
439          */
440         if (do_escape) {
441                 mapped_data = kmap_atomic(new_page);
442                 *((unsigned int *)(mapped_data + new_offset)) = 0;
443                 kunmap_atomic(mapped_data);
444         }
445 
446         set_bh_page(new_bh, new_page, new_offset);
447         new_bh->b_size = bh_in->b_size;
448         new_bh->b_bdev = journal->j_dev;
449         new_bh->b_blocknr = blocknr;
450         new_bh->b_private = bh_in;
451         set_buffer_mapped(new_bh);
452         set_buffer_dirty(new_bh);
453 
454         *bh_out = new_bh;
455 
456         /*
457          * The to-be-written buffer needs to get moved to the io queue,
458          * and the original buffer whose contents we are shadowing or
459          * copying is moved to the transaction's shadow queue.
460          */
461         JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
462         spin_lock(&journal->j_list_lock);
463         __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
464         spin_unlock(&journal->j_list_lock);
465         set_buffer_shadow(bh_in);
466         spin_unlock(&jh_in->b_state_lock);
467 
468         return do_escape | (done_copy_out << 1);
469 }
470 
471 /*
472  * Allocation code for the journal file.  Manage the space left in the
473  * journal, so that we can begin checkpointing when appropriate.
474  */
475 
476 /*
477  * Called with j_state_lock locked for writing.
478  * Returns true if a transaction commit was started.
479  */
480 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
481 {
482         /* Return if the txn has already requested to be committed */
483         if (journal->j_commit_request == target)
484                 return 0;
485 
486         /*
487          * The only transaction we can possibly wait upon is the
488          * currently running transaction (if it exists).  Otherwise,
489          * the target tid must be an old one.
490          */
491         if (journal->j_running_transaction &&
492             journal->j_running_transaction->t_tid == target) {
493                 /*
494                  * We want a new commit: OK, mark the request and wakeup the
495                  * commit thread.  We do _not_ do the commit ourselves.
496                  */
497 
498                 journal->j_commit_request = target;
499                 jbd_debug(1, "JBD2: requesting commit %u/%u\n",
500                           journal->j_commit_request,
501                           journal->j_commit_sequence);
502                 journal->j_running_transaction->t_requested = jiffies;
503                 wake_up(&journal->j_wait_commit);
504                 return 1;
505         } else if (!tid_geq(journal->j_commit_request, target))
506                 /* This should never happen, but if it does, preserve
507                    the evidence before kjournald goes into a loop and
508                    increments j_commit_sequence beyond all recognition. */
509                 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
510                           journal->j_commit_request,
511                           journal->j_commit_sequence,
512                           target, journal->j_running_transaction ?
513                           journal->j_running_transaction->t_tid : 0);
514         return 0;
515 }
516 
517 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
518 {
519         int ret;
520 
521         write_lock(&journal->j_state_lock);
522         ret = __jbd2_log_start_commit(journal, tid);
523         write_unlock(&journal->j_state_lock);
524         return ret;
525 }
526 
527 /*
528  * Force and wait any uncommitted transactions.  We can only force the running
529  * transaction if we don't have an active handle, otherwise, we will deadlock.
530  * Returns: <0 in case of error,
531  *           0 if nothing to commit,
532  *           1 if transaction was successfully committed.
533  */
534 static int __jbd2_journal_force_commit(journal_t *journal)
535 {
536         transaction_t *transaction = NULL;
537         tid_t tid;
538         int need_to_start = 0, ret = 0;
539 
540         read_lock(&journal->j_state_lock);
541         if (journal->j_running_transaction && !current->journal_info) {
542                 transaction = journal->j_running_transaction;
543                 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
544                         need_to_start = 1;
545         } else if (journal->j_committing_transaction)
546                 transaction = journal->j_committing_transaction;
547 
548         if (!transaction) {
549                 /* Nothing to commit */
550                 read_unlock(&journal->j_state_lock);
551                 return 0;
552         }
553         tid = transaction->t_tid;
554         read_unlock(&journal->j_state_lock);
555         if (need_to_start)
556                 jbd2_log_start_commit(journal, tid);
557         ret = jbd2_log_wait_commit(journal, tid);
558         if (!ret)
559                 ret = 1;
560 
561         return ret;
562 }
563 
564 /**
565  * Force and wait upon a commit if the calling process is not within
566  * transaction.  This is used for forcing out undo-protected data which contains
567  * bitmaps, when the fs is running out of space.
568  *
569  * @journal: journal to force
570  * Returns true if progress was made.
571  */
572 int jbd2_journal_force_commit_nested(journal_t *journal)
573 {
574         int ret;
575 
576         ret = __jbd2_journal_force_commit(journal);
577         return ret > 0;
578 }
579 
580 /**
581  * int journal_force_commit() - force any uncommitted transactions
582  * @journal: journal to force
583  *
584  * Caller want unconditional commit. We can only force the running transaction
585  * if we don't have an active handle, otherwise, we will deadlock.
586  */
587 int jbd2_journal_force_commit(journal_t *journal)
588 {
589         int ret;
590 
591         J_ASSERT(!current->journal_info);
592         ret = __jbd2_journal_force_commit(journal);
593         if (ret > 0)
594                 ret = 0;
595         return ret;
596 }
597 
598 /*
599  * Start a commit of the current running transaction (if any).  Returns true
600  * if a transaction is going to be committed (or is currently already
601  * committing), and fills its tid in at *ptid
602  */
603 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
604 {
605         int ret = 0;
606 
607         write_lock(&journal->j_state_lock);
608         if (journal->j_running_transaction) {
609                 tid_t tid = journal->j_running_transaction->t_tid;
610 
611                 __jbd2_log_start_commit(journal, tid);
612                 /* There's a running transaction and we've just made sure
613                  * it's commit has been scheduled. */
614                 if (ptid)
615                         *ptid = tid;
616                 ret = 1;
617         } else if (journal->j_committing_transaction) {
618                 /*
619                  * If commit has been started, then we have to wait for
620                  * completion of that transaction.
621                  */
622                 if (ptid)
623                         *ptid = journal->j_committing_transaction->t_tid;
624                 ret = 1;
625         }
626         write_unlock(&journal->j_state_lock);
627         return ret;
628 }
629 
630 /*
631  * Return 1 if a given transaction has not yet sent barrier request
632  * connected with a transaction commit. If 0 is returned, transaction
633  * may or may not have sent the barrier. Used to avoid sending barrier
634  * twice in common cases.
635  */
636 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
637 {
638         int ret = 0;
639         transaction_t *commit_trans;
640 
641         if (!(journal->j_flags & JBD2_BARRIER))
642                 return 0;
643         read_lock(&journal->j_state_lock);
644         /* Transaction already committed? */
645         if (tid_geq(journal->j_commit_sequence, tid))
646                 goto out;
647         commit_trans = journal->j_committing_transaction;
648         if (!commit_trans || commit_trans->t_tid != tid) {
649                 ret = 1;
650                 goto out;
651         }
652         /*
653          * Transaction is being committed and we already proceeded to
654          * submitting a flush to fs partition?
655          */
656         if (journal->j_fs_dev != journal->j_dev) {
657                 if (!commit_trans->t_need_data_flush ||
658                     commit_trans->t_state >= T_COMMIT_DFLUSH)
659                         goto out;
660         } else {
661                 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
662                         goto out;
663         }
664         ret = 1;
665 out:
666         read_unlock(&journal->j_state_lock);
667         return ret;
668 }
669 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
670 
671 /*
672  * Wait for a specified commit to complete.
673  * The caller may not hold the journal lock.
674  */
675 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
676 {
677         int err = 0;
678 
679         read_lock(&journal->j_state_lock);
680 #ifdef CONFIG_PROVE_LOCKING
681         /*
682          * Some callers make sure transaction is already committing and in that
683          * case we cannot block on open handles anymore. So don't warn in that
684          * case.
685          */
686         if (tid_gt(tid, journal->j_commit_sequence) &&
687             (!journal->j_committing_transaction ||
688              journal->j_committing_transaction->t_tid != tid)) {
689                 read_unlock(&journal->j_state_lock);
690                 jbd2_might_wait_for_commit(journal);
691                 read_lock(&journal->j_state_lock);
692         }
693 #endif
694 #ifdef CONFIG_JBD2_DEBUG
695         if (!tid_geq(journal->j_commit_request, tid)) {
696                 printk(KERN_ERR
697                        "%s: error: j_commit_request=%u, tid=%u\n",
698                        __func__, journal->j_commit_request, tid);
699         }
700 #endif
701         while (tid_gt(tid, journal->j_commit_sequence)) {
702                 jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
703                                   tid, journal->j_commit_sequence);
704                 read_unlock(&journal->j_state_lock);
705                 wake_up(&journal->j_wait_commit);
706                 wait_event(journal->j_wait_done_commit,
707                                 !tid_gt(tid, journal->j_commit_sequence));
708                 read_lock(&journal->j_state_lock);
709         }
710         read_unlock(&journal->j_state_lock);
711 
712         if (unlikely(is_journal_aborted(journal)))
713                 err = -EIO;
714         return err;
715 }
716 
717 /* Return 1 when transaction with given tid has already committed. */
718 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
719 {
720         int ret = 1;
721 
722         read_lock(&journal->j_state_lock);
723         if (journal->j_running_transaction &&
724             journal->j_running_transaction->t_tid == tid)
725                 ret = 0;
726         if (journal->j_committing_transaction &&
727             journal->j_committing_transaction->t_tid == tid)
728                 ret = 0;
729         read_unlock(&journal->j_state_lock);
730         return ret;
731 }
732 EXPORT_SYMBOL(jbd2_transaction_committed);
733 
734 /*
735  * When this function returns the transaction corresponding to tid
736  * will be completed.  If the transaction has currently running, start
737  * committing that transaction before waiting for it to complete.  If
738  * the transaction id is stale, it is by definition already completed,
739  * so just return SUCCESS.
740  */
741 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
742 {
743         int     need_to_wait = 1;
744 
745         read_lock(&journal->j_state_lock);
746         if (journal->j_running_transaction &&
747             journal->j_running_transaction->t_tid == tid) {
748                 if (journal->j_commit_request != tid) {
749                         /* transaction not yet started, so request it */
750                         read_unlock(&journal->j_state_lock);
751                         jbd2_log_start_commit(journal, tid);
752                         goto wait_commit;
753                 }
754         } else if (!(journal->j_committing_transaction &&
755                      journal->j_committing_transaction->t_tid == tid))
756                 need_to_wait = 0;
757         read_unlock(&journal->j_state_lock);
758         if (!need_to_wait)
759                 return 0;
760 wait_commit:
761         return jbd2_log_wait_commit(journal, tid);
762 }
763 EXPORT_SYMBOL(jbd2_complete_transaction);
764 
765 /*
766  * Log buffer allocation routines:
767  */
768 
769 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
770 {
771         unsigned long blocknr;
772 
773         write_lock(&journal->j_state_lock);
774         J_ASSERT(journal->j_free > 1);
775 
776         blocknr = journal->j_head;
777         journal->j_head++;
778         journal->j_free--;
779         if (journal->j_head == journal->j_last)
780                 journal->j_head = journal->j_first;
781         write_unlock(&journal->j_state_lock);
782         return jbd2_journal_bmap(journal, blocknr, retp);
783 }
784 
785 /*
786  * Conversion of logical to physical block numbers for the journal
787  *
788  * On external journals the journal blocks are identity-mapped, so
789  * this is a no-op.  If needed, we can use j_blk_offset - everything is
790  * ready.
791  */
792 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
793                  unsigned long long *retp)
794 {
795         int err = 0;
796         unsigned long long ret;
797         sector_t block = 0;
798 
799         if (journal->j_inode) {
800                 block = blocknr;
801                 ret = bmap(journal->j_inode, &block);
802 
803                 if (ret || !block) {
804                         printk(KERN_ALERT "%s: journal block not found "
805                                         "at offset %lu on %s\n",
806                                __func__, blocknr, journal->j_devname);
807                         err = -EIO;
808                         jbd2_journal_abort(journal, err);
809                 } else {
810                         *retp = block;
811                 }
812 
813         } else {
814                 *retp = blocknr; /* +journal->j_blk_offset */
815         }
816         return err;
817 }
818 
819 /*
820  * We play buffer_head aliasing tricks to write data/metadata blocks to
821  * the journal without copying their contents, but for journal
822  * descriptor blocks we do need to generate bona fide buffers.
823  *
824  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
825  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
826  * But we don't bother doing that, so there will be coherency problems with
827  * mmaps of blockdevs which hold live JBD-controlled filesystems.
828  */
829 struct buffer_head *
830 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
831 {
832         journal_t *journal = transaction->t_journal;
833         struct buffer_head *bh;
834         unsigned long long blocknr;
835         journal_header_t *header;
836         int err;
837 
838         err = jbd2_journal_next_log_block(journal, &blocknr);
839 
840         if (err)
841                 return NULL;
842 
843         bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
844         if (!bh)
845                 return NULL;
846         atomic_dec(&transaction->t_outstanding_credits);
847         lock_buffer(bh);
848         memset(bh->b_data, 0, journal->j_blocksize);
849         header = (journal_header_t *)bh->b_data;
850         header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
851         header->h_blocktype = cpu_to_be32(type);
852         header->h_sequence = cpu_to_be32(transaction->t_tid);
853         set_buffer_uptodate(bh);
854         unlock_buffer(bh);
855         BUFFER_TRACE(bh, "return this buffer");
856         return bh;
857 }
858 
859 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
860 {
861         struct jbd2_journal_block_tail *tail;
862         __u32 csum;
863 
864         if (!jbd2_journal_has_csum_v2or3(j))
865                 return;
866 
867         tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
868                         sizeof(struct jbd2_journal_block_tail));
869         tail->t_checksum = 0;
870         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
871         tail->t_checksum = cpu_to_be32(csum);
872 }
873 
874 /*
875  * Return tid of the oldest transaction in the journal and block in the journal
876  * where the transaction starts.
877  *
878  * If the journal is now empty, return which will be the next transaction ID
879  * we will write and where will that transaction start.
880  *
881  * The return value is 0 if journal tail cannot be pushed any further, 1 if
882  * it can.
883  */
884 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
885                               unsigned long *block)
886 {
887         transaction_t *transaction;
888         int ret;
889 
890         read_lock(&journal->j_state_lock);
891         spin_lock(&journal->j_list_lock);
892         transaction = journal->j_checkpoint_transactions;
893         if (transaction) {
894                 *tid = transaction->t_tid;
895                 *block = transaction->t_log_start;
896         } else if ((transaction = journal->j_committing_transaction) != NULL) {
897                 *tid = transaction->t_tid;
898                 *block = transaction->t_log_start;
899         } else if ((transaction = journal->j_running_transaction) != NULL) {
900                 *tid = transaction->t_tid;
901                 *block = journal->j_head;
902         } else {
903                 *tid = journal->j_transaction_sequence;
904                 *block = journal->j_head;
905         }
906         ret = tid_gt(*tid, journal->j_tail_sequence);
907         spin_unlock(&journal->j_list_lock);
908         read_unlock(&journal->j_state_lock);
909 
910         return ret;
911 }
912 
913 /*
914  * Update information in journal structure and in on disk journal superblock
915  * about log tail. This function does not check whether information passed in
916  * really pushes log tail further. It's responsibility of the caller to make
917  * sure provided log tail information is valid (e.g. by holding
918  * j_checkpoint_mutex all the time between computing log tail and calling this
919  * function as is the case with jbd2_cleanup_journal_tail()).
920  *
921  * Requires j_checkpoint_mutex
922  */
923 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
924 {
925         unsigned long freed;
926         int ret;
927 
928         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
929 
930         /*
931          * We cannot afford for write to remain in drive's caches since as
932          * soon as we update j_tail, next transaction can start reusing journal
933          * space and if we lose sb update during power failure we'd replay
934          * old transaction with possibly newly overwritten data.
935          */
936         ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
937                                               REQ_SYNC | REQ_FUA);
938         if (ret)
939                 goto out;
940 
941         write_lock(&journal->j_state_lock);
942         freed = block - journal->j_tail;
943         if (block < journal->j_tail)
944                 freed += journal->j_last - journal->j_first;
945 
946         trace_jbd2_update_log_tail(journal, tid, block, freed);
947         jbd_debug(1,
948                   "Cleaning journal tail from %u to %u (offset %lu), "
949                   "freeing %lu\n",
950                   journal->j_tail_sequence, tid, block, freed);
951 
952         journal->j_free += freed;
953         journal->j_tail_sequence = tid;
954         journal->j_tail = block;
955         write_unlock(&journal->j_state_lock);
956 
957 out:
958         return ret;
959 }
960 
961 /*
962  * This is a variation of __jbd2_update_log_tail which checks for validity of
963  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
964  * with other threads updating log tail.
965  */
966 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
967 {
968         mutex_lock_io(&journal->j_checkpoint_mutex);
969         if (tid_gt(tid, journal->j_tail_sequence))
970                 __jbd2_update_log_tail(journal, tid, block);
971         mutex_unlock(&journal->j_checkpoint_mutex);
972 }
973 
974 struct jbd2_stats_proc_session {
975         journal_t *journal;
976         struct transaction_stats_s *stats;
977         int start;
978         int max;
979 };
980 
981 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
982 {
983         return *pos ? NULL : SEQ_START_TOKEN;
984 }
985 
986 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
987 {
988         (*pos)++;
989         return NULL;
990 }
991 
992 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
993 {
994         struct jbd2_stats_proc_session *s = seq->private;
995 
996         if (v != SEQ_START_TOKEN)
997                 return 0;
998         seq_printf(seq, "%lu transactions (%lu requested), "
999                    "each up to %u blocks\n",
1000                    s->stats->ts_tid, s->stats->ts_requested,
1001                    s->journal->j_max_transaction_buffers);
1002         if (s->stats->ts_tid == 0)
1003                 return 0;
1004         seq_printf(seq, "average: \n  %ums waiting for transaction\n",
1005             jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1006         seq_printf(seq, "  %ums request delay\n",
1007             (s->stats->ts_requested == 0) ? 0 :
1008             jiffies_to_msecs(s->stats->run.rs_request_delay /
1009                              s->stats->ts_requested));
1010         seq_printf(seq, "  %ums running transaction\n",
1011             jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1012         seq_printf(seq, "  %ums transaction was being locked\n",
1013             jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1014         seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
1015             jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1016         seq_printf(seq, "  %ums logging transaction\n",
1017             jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1018         seq_printf(seq, "  %lluus average transaction commit time\n",
1019                    div_u64(s->journal->j_average_commit_time, 1000));
1020         seq_printf(seq, "  %lu handles per transaction\n",
1021             s->stats->run.rs_handle_count / s->stats->ts_tid);
1022         seq_printf(seq, "  %lu blocks per transaction\n",
1023             s->stats->run.rs_blocks / s->stats->ts_tid);
1024         seq_printf(seq, "  %lu logged blocks per transaction\n",
1025             s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1026         return 0;
1027 }
1028 
1029 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1030 {
1031 }
1032 
1033 static const struct seq_operations jbd2_seq_info_ops = {
1034         .start  = jbd2_seq_info_start,
1035         .next   = jbd2_seq_info_next,
1036         .stop   = jbd2_seq_info_stop,
1037         .show   = jbd2_seq_info_show,
1038 };
1039 
1040 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1041 {
1042         journal_t *journal = PDE_DATA(inode);
1043         struct jbd2_stats_proc_session *s;
1044         int rc, size;
1045 
1046         s = kmalloc(sizeof(*s), GFP_KERNEL);
1047         if (s == NULL)
1048                 return -ENOMEM;
1049         size = sizeof(struct transaction_stats_s);
1050         s->stats = kmalloc(size, GFP_KERNEL);
1051         if (s->stats == NULL) {
1052                 kfree(s);
1053                 return -ENOMEM;
1054         }
1055         spin_lock(&journal->j_history_lock);
1056         memcpy(s->stats, &journal->j_stats, size);
1057         s->journal = journal;
1058         spin_unlock(&journal->j_history_lock);
1059 
1060         rc = seq_open(file, &jbd2_seq_info_ops);
1061         if (rc == 0) {
1062                 struct seq_file *m = file->private_data;
1063                 m->private = s;
1064         } else {
1065                 kfree(s->stats);
1066                 kfree(s);
1067         }
1068         return rc;
1069 
1070 }
1071 
1072 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1073 {
1074         struct seq_file *seq = file->private_data;
1075         struct jbd2_stats_proc_session *s = seq->private;
1076         kfree(s->stats);
1077         kfree(s);
1078         return seq_release(inode, file);
1079 }
1080 
1081 static const struct proc_ops jbd2_info_proc_ops = {
1082         .proc_open      = jbd2_seq_info_open,
1083         .proc_read      = seq_read,
1084         .proc_lseek     = seq_lseek,
1085         .proc_release   = jbd2_seq_info_release,
1086 };
1087 
1088 static struct proc_dir_entry *proc_jbd2_stats;
1089 
1090 static void jbd2_stats_proc_init(journal_t *journal)
1091 {
1092         journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1093         if (journal->j_proc_entry) {
1094                 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1095                                  &jbd2_info_proc_ops, journal);
1096         }
1097 }
1098 
1099 static void jbd2_stats_proc_exit(journal_t *journal)
1100 {
1101         remove_proc_entry("info", journal->j_proc_entry);
1102         remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1103 }
1104 
1105 /* Minimum size of descriptor tag */
1106 static int jbd2_min_tag_size(void)
1107 {
1108         /*
1109          * Tag with 32-bit block numbers does not use last four bytes of the
1110          * structure
1111          */
1112         return sizeof(journal_block_tag_t) - 4;
1113 }
1114 
1115 /*
1116  * Management for journal control blocks: functions to create and
1117  * destroy journal_t structures, and to initialise and read existing
1118  * journal blocks from disk.  */
1119 
1120 /* First: create and setup a journal_t object in memory.  We initialise
1121  * very few fields yet: that has to wait until we have created the
1122  * journal structures from from scratch, or loaded them from disk. */
1123 
1124 static journal_t *journal_init_common(struct block_device *bdev,
1125                         struct block_device *fs_dev,
1126                         unsigned long long start, int len, int blocksize)
1127 {
1128         static struct lock_class_key jbd2_trans_commit_key;
1129         journal_t *journal;
1130         int err;
1131         struct buffer_head *bh;
1132         int n;
1133 
1134         journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1135         if (!journal)
1136                 return NULL;
1137 
1138         init_waitqueue_head(&journal->j_wait_transaction_locked);
1139         init_waitqueue_head(&journal->j_wait_done_commit);
1140         init_waitqueue_head(&journal->j_wait_commit);
1141         init_waitqueue_head(&journal->j_wait_updates);
1142         init_waitqueue_head(&journal->j_wait_reserved);
1143         mutex_init(&journal->j_barrier);
1144         mutex_init(&journal->j_checkpoint_mutex);
1145         spin_lock_init(&journal->j_revoke_lock);
1146         spin_lock_init(&journal->j_list_lock);
1147         rwlock_init(&journal->j_state_lock);
1148 
1149         journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1150         journal->j_min_batch_time = 0;
1151         journal->j_max_batch_time = 15000; /* 15ms */
1152         atomic_set(&journal->j_reserved_credits, 0);
1153 
1154         /* The journal is marked for error until we succeed with recovery! */
1155         journal->j_flags = JBD2_ABORT;
1156 
1157         /* Set up a default-sized revoke table for the new mount. */
1158         err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1159         if (err)
1160                 goto err_cleanup;
1161 
1162         spin_lock_init(&journal->j_history_lock);
1163 
1164         lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1165                          &jbd2_trans_commit_key, 0);
1166 
1167         /* journal descriptor can store up to n blocks -bzzz */
1168         journal->j_blocksize = blocksize;
1169         journal->j_dev = bdev;
1170         journal->j_fs_dev = fs_dev;
1171         journal->j_blk_offset = start;
1172         journal->j_maxlen = len;
1173         /* We need enough buffers to write out full descriptor block. */
1174         n = journal->j_blocksize / jbd2_min_tag_size();
1175         journal->j_wbufsize = n;
1176         journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1177                                         GFP_KERNEL);
1178         if (!journal->j_wbuf)
1179                 goto err_cleanup;
1180 
1181         bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1182         if (!bh) {
1183                 pr_err("%s: Cannot get buffer for journal superblock\n",
1184                         __func__);
1185                 goto err_cleanup;
1186         }
1187         journal->j_sb_buffer = bh;
1188         journal->j_superblock = (journal_superblock_t *)bh->b_data;
1189 
1190         return journal;
1191 
1192 err_cleanup:
1193         kfree(journal->j_wbuf);
1194         jbd2_journal_destroy_revoke(journal);
1195         kfree(journal);
1196         return NULL;
1197 }
1198 
1199 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1200  *
1201  * Create a journal structure assigned some fixed set of disk blocks to
1202  * the journal.  We don't actually touch those disk blocks yet, but we
1203  * need to set up all of the mapping information to tell the journaling
1204  * system where the journal blocks are.
1205  *
1206  */
1207 
1208 /**
1209  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1210  *  @bdev: Block device on which to create the journal
1211  *  @fs_dev: Device which hold journalled filesystem for this journal.
1212  *  @start: Block nr Start of journal.
1213  *  @len:  Length of the journal in blocks.
1214  *  @blocksize: blocksize of journalling device
1215  *
1216  *  Returns: a newly created journal_t *
1217  *
1218  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1219  *  range of blocks on an arbitrary block device.
1220  *
1221  */
1222 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1223                         struct block_device *fs_dev,
1224                         unsigned long long start, int len, int blocksize)
1225 {
1226         journal_t *journal;
1227 
1228         journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1229         if (!journal)
1230                 return NULL;
1231 
1232         bdevname(journal->j_dev, journal->j_devname);
1233         strreplace(journal->j_devname, '/', '!');
1234         jbd2_stats_proc_init(journal);
1235 
1236         return journal;
1237 }
1238 
1239 /**
1240  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1241  *  @inode: An inode to create the journal in
1242  *
1243  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1244  * the journal.  The inode must exist already, must support bmap() and
1245  * must have all data blocks preallocated.
1246  */
1247 journal_t *jbd2_journal_init_inode(struct inode *inode)
1248 {
1249         journal_t *journal;
1250         sector_t blocknr;
1251         char *p;
1252         int err = 0;
1253 
1254         blocknr = 0;
1255         err = bmap(inode, &blocknr);
1256 
1257         if (err || !blocknr) {
1258                 pr_err("%s: Cannot locate journal superblock\n",
1259                         __func__);
1260                 return NULL;
1261         }
1262 
1263         jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1264                   inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1265                   inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1266 
1267         journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1268                         blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1269                         inode->i_sb->s_blocksize);
1270         if (!journal)
1271                 return NULL;
1272 
1273         journal->j_inode = inode;
1274         bdevname(journal->j_dev, journal->j_devname);
1275         p = strreplace(journal->j_devname, '/', '!');
1276         sprintf(p, "-%lu", journal->j_inode->i_ino);
1277         jbd2_stats_proc_init(journal);
1278 
1279         return journal;
1280 }
1281 
1282 /*
1283  * If the journal init or create aborts, we need to mark the journal
1284  * superblock as being NULL to prevent the journal destroy from writing
1285  * back a bogus superblock.
1286  */
1287 static void journal_fail_superblock (journal_t *journal)
1288 {
1289         struct buffer_head *bh = journal->j_sb_buffer;
1290         brelse(bh);
1291         journal->j_sb_buffer = NULL;
1292 }
1293 
1294 /*
1295  * Given a journal_t structure, initialise the various fields for
1296  * startup of a new journaling session.  We use this both when creating
1297  * a journal, and after recovering an old journal to reset it for
1298  * subsequent use.
1299  */
1300 
1301 static int journal_reset(journal_t *journal)
1302 {
1303         journal_superblock_t *sb = journal->j_superblock;
1304         unsigned long long first, last;
1305 
1306         first = be32_to_cpu(sb->s_first);
1307         last = be32_to_cpu(sb->s_maxlen);
1308         if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1309                 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1310                        first, last);
1311                 journal_fail_superblock(journal);
1312                 return -EINVAL;
1313         }
1314 
1315         journal->j_first = first;
1316         journal->j_last = last;
1317 
1318         journal->j_head = first;
1319         journal->j_tail = first;
1320         journal->j_free = last - first;
1321 
1322         journal->j_tail_sequence = journal->j_transaction_sequence;
1323         journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1324         journal->j_commit_request = journal->j_commit_sequence;
1325 
1326         journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1327 
1328         /*
1329          * As a special case, if the on-disk copy is already marked as needing
1330          * no recovery (s_start == 0), then we can safely defer the superblock
1331          * update until the next commit by setting JBD2_FLUSHED.  This avoids
1332          * attempting a write to a potential-readonly device.
1333          */
1334         if (sb->s_start == 0) {
1335                 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1336                         "(start %ld, seq %u, errno %d)\n",
1337                         journal->j_tail, journal->j_tail_sequence,
1338                         journal->j_errno);
1339                 journal->j_flags |= JBD2_FLUSHED;
1340         } else {
1341                 /* Lock here to make assertions happy... */
1342                 mutex_lock_io(&journal->j_checkpoint_mutex);
1343                 /*
1344                  * Update log tail information. We use REQ_FUA since new
1345                  * transaction will start reusing journal space and so we
1346                  * must make sure information about current log tail is on
1347                  * disk before that.
1348                  */
1349                 jbd2_journal_update_sb_log_tail(journal,
1350                                                 journal->j_tail_sequence,
1351                                                 journal->j_tail,
1352                                                 REQ_SYNC | REQ_FUA);
1353                 mutex_unlock(&journal->j_checkpoint_mutex);
1354         }
1355         return jbd2_journal_start_thread(journal);
1356 }
1357 
1358 /*
1359  * This function expects that the caller will have locked the journal
1360  * buffer head, and will return with it unlocked
1361  */
1362 static int jbd2_write_superblock(journal_t *journal, int write_flags)
1363 {
1364         struct buffer_head *bh = journal->j_sb_buffer;
1365         journal_superblock_t *sb = journal->j_superblock;
1366         int ret;
1367 
1368         /* Buffer got discarded which means block device got invalidated */
1369         if (!buffer_mapped(bh))
1370                 return -EIO;
1371 
1372         trace_jbd2_write_superblock(journal, write_flags);
1373         if (!(journal->j_flags & JBD2_BARRIER))
1374                 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1375         if (buffer_write_io_error(bh)) {
1376                 /*
1377                  * Oh, dear.  A previous attempt to write the journal
1378                  * superblock failed.  This could happen because the
1379                  * USB device was yanked out.  Or it could happen to
1380                  * be a transient write error and maybe the block will
1381                  * be remapped.  Nothing we can do but to retry the
1382                  * write and hope for the best.
1383                  */
1384                 printk(KERN_ERR "JBD2: previous I/O error detected "
1385                        "for journal superblock update for %s.\n",
1386                        journal->j_devname);
1387                 clear_buffer_write_io_error(bh);
1388                 set_buffer_uptodate(bh);
1389         }
1390         if (jbd2_journal_has_csum_v2or3(journal))
1391                 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1392         get_bh(bh);
1393         bh->b_end_io = end_buffer_write_sync;
1394         ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1395         wait_on_buffer(bh);
1396         if (buffer_write_io_error(bh)) {
1397                 clear_buffer_write_io_error(bh);
1398                 set_buffer_uptodate(bh);
1399                 ret = -EIO;
1400         }
1401         if (ret) {
1402                 printk(KERN_ERR "JBD2: Error %d detected when updating "
1403                        "journal superblock for %s.\n", ret,
1404                        journal->j_devname);
1405                 jbd2_journal_abort(journal, ret);
1406         }
1407 
1408         return ret;
1409 }
1410 
1411 /**
1412  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1413  * @journal: The journal to update.
1414  * @tail_tid: TID of the new transaction at the tail of the log
1415  * @tail_block: The first block of the transaction at the tail of the log
1416  * @write_op: With which operation should we write the journal sb
1417  *
1418  * Update a journal's superblock information about log tail and write it to
1419  * disk, waiting for the IO to complete.
1420  */
1421 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1422                                      unsigned long tail_block, int write_op)
1423 {
1424         journal_superblock_t *sb = journal->j_superblock;
1425         int ret;
1426 
1427         if (is_journal_aborted(journal))
1428                 return -EIO;
1429 
1430         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1431         jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1432                   tail_block, tail_tid);
1433 
1434         lock_buffer(journal->j_sb_buffer);
1435         sb->s_sequence = cpu_to_be32(tail_tid);
1436         sb->s_start    = cpu_to_be32(tail_block);
1437 
1438         ret = jbd2_write_superblock(journal, write_op);
1439         if (ret)
1440                 goto out;
1441 
1442         /* Log is no longer empty */
1443         write_lock(&journal->j_state_lock);
1444         WARN_ON(!sb->s_sequence);
1445         journal->j_flags &= ~JBD2_FLUSHED;
1446         write_unlock(&journal->j_state_lock);
1447 
1448 out:
1449         return ret;
1450 }
1451 
1452 /**
1453  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1454  * @journal: The journal to update.
1455  * @write_op: With which operation should we write the journal sb
1456  *
1457  * Update a journal's dynamic superblock fields to show that journal is empty.
1458  * Write updated superblock to disk waiting for IO to complete.
1459  */
1460 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1461 {
1462         journal_superblock_t *sb = journal->j_superblock;
1463 
1464         BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1465         lock_buffer(journal->j_sb_buffer);
1466         if (sb->s_start == 0) {         /* Is it already empty? */
1467                 unlock_buffer(journal->j_sb_buffer);
1468                 return;
1469         }
1470 
1471         jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1472                   journal->j_tail_sequence);
1473 
1474         sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1475         sb->s_start    = cpu_to_be32(0);
1476 
1477         jbd2_write_superblock(journal, write_op);
1478 
1479         /* Log is no longer empty */
1480         write_lock(&journal->j_state_lock);
1481         journal->j_flags |= JBD2_FLUSHED;
1482         write_unlock(&journal->j_state_lock);
1483 }
1484 
1485 
1486 /**
1487  * jbd2_journal_update_sb_errno() - Update error in the journal.
1488  * @journal: The journal to update.
1489  *
1490  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1491  * to complete.
1492  */
1493 void jbd2_journal_update_sb_errno(journal_t *journal)
1494 {
1495         journal_superblock_t *sb = journal->j_superblock;
1496         int errcode;
1497 
1498         lock_buffer(journal->j_sb_buffer);
1499         errcode = journal->j_errno;
1500         if (errcode == -ESHUTDOWN)
1501                 errcode = 0;
1502         jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1503         sb->s_errno    = cpu_to_be32(errcode);
1504 
1505         jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1506 }
1507 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1508 
1509 static int journal_revoke_records_per_block(journal_t *journal)
1510 {
1511         int record_size;
1512         int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1513 
1514         if (jbd2_has_feature_64bit(journal))
1515                 record_size = 8;
1516         else
1517                 record_size = 4;
1518 
1519         if (jbd2_journal_has_csum_v2or3(journal))
1520                 space -= sizeof(struct jbd2_journal_block_tail);
1521         return space / record_size;
1522 }
1523 
1524 /*
1525  * Read the superblock for a given journal, performing initial
1526  * validation of the format.
1527  */
1528 static int journal_get_superblock(journal_t *journal)
1529 {
1530         struct buffer_head *bh;
1531         journal_superblock_t *sb;
1532         int err = -EIO;
1533 
1534         bh = journal->j_sb_buffer;
1535 
1536         J_ASSERT(bh != NULL);
1537         if (!buffer_uptodate(bh)) {
1538                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1539                 wait_on_buffer(bh);
1540                 if (!buffer_uptodate(bh)) {
1541                         printk(KERN_ERR
1542                                 "JBD2: IO error reading journal superblock\n");
1543                         goto out;
1544                 }
1545         }
1546 
1547         if (buffer_verified(bh))
1548                 return 0;
1549 
1550         sb = journal->j_superblock;
1551 
1552         err = -EINVAL;
1553 
1554         if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1555             sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1556                 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1557                 goto out;
1558         }
1559 
1560         switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1561         case JBD2_SUPERBLOCK_V1:
1562                 journal->j_format_version = 1;
1563                 break;
1564         case JBD2_SUPERBLOCK_V2:
1565                 journal->j_format_version = 2;
1566                 break;
1567         default:
1568                 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1569                 goto out;
1570         }
1571 
1572         if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1573                 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1574         else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1575                 printk(KERN_WARNING "JBD2: journal file too short\n");
1576                 goto out;
1577         }
1578 
1579         if (be32_to_cpu(sb->s_first) == 0 ||
1580             be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1581                 printk(KERN_WARNING
1582                         "JBD2: Invalid start block of journal: %u\n",
1583                         be32_to_cpu(sb->s_first));
1584                 goto out;
1585         }
1586 
1587         if (jbd2_has_feature_csum2(journal) &&
1588             jbd2_has_feature_csum3(journal)) {
1589                 /* Can't have checksum v2 and v3 at the same time! */
1590                 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1591                        "at the same time!\n");
1592                 goto out;
1593         }
1594 
1595         if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1596             jbd2_has_feature_checksum(journal)) {
1597                 /* Can't have checksum v1 and v2 on at the same time! */
1598                 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1599                        "at the same time!\n");
1600                 goto out;
1601         }
1602 
1603         if (!jbd2_verify_csum_type(journal, sb)) {
1604                 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1605                 goto out;
1606         }
1607 
1608         /* Load the checksum driver */
1609         if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1610                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1611                 if (IS_ERR(journal->j_chksum_driver)) {
1612                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1613                         err = PTR_ERR(journal->j_chksum_driver);
1614                         journal->j_chksum_driver = NULL;
1615                         goto out;
1616                 }
1617         }
1618 
1619         if (jbd2_journal_has_csum_v2or3(journal)) {
1620                 /* Check superblock checksum */
1621                 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1622                         printk(KERN_ERR "JBD2: journal checksum error\n");
1623                         err = -EFSBADCRC;
1624                         goto out;
1625                 }
1626 
1627                 /* Precompute checksum seed for all metadata */
1628                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1629                                                    sizeof(sb->s_uuid));
1630         }
1631 
1632         journal->j_revoke_records_per_block =
1633                                 journal_revoke_records_per_block(journal);
1634         set_buffer_verified(bh);
1635 
1636         return 0;
1637 
1638 out:
1639         journal_fail_superblock(journal);
1640         return err;
1641 }
1642 
1643 /*
1644  * Load the on-disk journal superblock and read the key fields into the
1645  * journal_t.
1646  */
1647 
1648 static int load_superblock(journal_t *journal)
1649 {
1650         int err;
1651         journal_superblock_t *sb;
1652 
1653         err = journal_get_superblock(journal);
1654         if (err)
1655                 return err;
1656 
1657         sb = journal->j_superblock;
1658 
1659         journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1660         journal->j_tail = be32_to_cpu(sb->s_start);
1661         journal->j_first = be32_to_cpu(sb->s_first);
1662         journal->j_last = be32_to_cpu(sb->s_maxlen);
1663         journal->j_errno = be32_to_cpu(sb->s_errno);
1664 
1665         return 0;
1666 }
1667 
1668 
1669 /**
1670  * int jbd2_journal_load() - Read journal from disk.
1671  * @journal: Journal to act on.
1672  *
1673  * Given a journal_t structure which tells us which disk blocks contain
1674  * a journal, read the journal from disk to initialise the in-memory
1675  * structures.
1676  */
1677 int jbd2_journal_load(journal_t *journal)
1678 {
1679         int err;
1680         journal_superblock_t *sb;
1681 
1682         err = load_superblock(journal);
1683         if (err)
1684                 return err;
1685 
1686         sb = journal->j_superblock;
1687         /* If this is a V2 superblock, then we have to check the
1688          * features flags on it. */
1689 
1690         if (journal->j_format_version >= 2) {
1691                 if ((sb->s_feature_ro_compat &
1692                      ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1693                     (sb->s_feature_incompat &
1694                      ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1695                         printk(KERN_WARNING
1696                                 "JBD2: Unrecognised features on journal\n");
1697                         return -EINVAL;
1698                 }
1699         }
1700 
1701         /*
1702          * Create a slab for this blocksize
1703          */
1704         err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1705         if (err)
1706                 return err;
1707 
1708         /* Let the recovery code check whether it needs to recover any
1709          * data from the journal. */
1710         if (jbd2_journal_recover(journal))
1711                 goto recovery_error;
1712 
1713         if (journal->j_failed_commit) {
1714                 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1715                        "is corrupt.\n", journal->j_failed_commit,
1716                        journal->j_devname);
1717                 return -EFSCORRUPTED;
1718         }
1719         /*
1720          * clear JBD2_ABORT flag initialized in journal_init_common
1721          * here to update log tail information with the newest seq.
1722          */
1723         journal->j_flags &= ~JBD2_ABORT;
1724 
1725         /* OK, we've finished with the dynamic journal bits:
1726          * reinitialise the dynamic contents of the superblock in memory
1727          * and reset them on disk. */
1728         if (journal_reset(journal))
1729                 goto recovery_error;
1730 
1731         journal->j_flags |= JBD2_LOADED;
1732         return 0;
1733 
1734 recovery_error:
1735         printk(KERN_WARNING "JBD2: recovery failed\n");
1736         return -EIO;
1737 }
1738 
1739 /**
1740  * void jbd2_journal_destroy() - Release a journal_t structure.
1741  * @journal: Journal to act on.
1742  *
1743  * Release a journal_t structure once it is no longer in use by the
1744  * journaled object.
1745  * Return <0 if we couldn't clean up the journal.
1746  */
1747 int jbd2_journal_destroy(journal_t *journal)
1748 {
1749         int err = 0;
1750 
1751         /* Wait for the commit thread to wake up and die. */
1752         journal_kill_thread(journal);
1753 
1754         /* Force a final log commit */
1755         if (journal->j_running_transaction)
1756                 jbd2_journal_commit_transaction(journal);
1757 
1758         /* Force any old transactions to disk */
1759 
1760         /* Totally anal locking here... */
1761         spin_lock(&journal->j_list_lock);
1762         while (journal->j_checkpoint_transactions != NULL) {
1763                 spin_unlock(&journal->j_list_lock);
1764                 mutex_lock_io(&journal->j_checkpoint_mutex);
1765                 err = jbd2_log_do_checkpoint(journal);
1766                 mutex_unlock(&journal->j_checkpoint_mutex);
1767                 /*
1768                  * If checkpointing failed, just free the buffers to avoid
1769                  * looping forever
1770                  */
1771                 if (err) {
1772                         jbd2_journal_destroy_checkpoint(journal);
1773                         spin_lock(&journal->j_list_lock);
1774                         break;
1775                 }
1776                 spin_lock(&journal->j_list_lock);
1777         }
1778 
1779         J_ASSERT(journal->j_running_transaction == NULL);
1780         J_ASSERT(journal->j_committing_transaction == NULL);
1781         J_ASSERT(journal->j_checkpoint_transactions == NULL);
1782         spin_unlock(&journal->j_list_lock);
1783 
1784         if (journal->j_sb_buffer) {
1785                 if (!is_journal_aborted(journal)) {
1786                         mutex_lock_io(&journal->j_checkpoint_mutex);
1787 
1788                         write_lock(&journal->j_state_lock);
1789                         journal->j_tail_sequence =
1790                                 ++journal->j_transaction_sequence;
1791                         write_unlock(&journal->j_state_lock);
1792 
1793                         jbd2_mark_journal_empty(journal,
1794                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1795                         mutex_unlock(&journal->j_checkpoint_mutex);
1796                 } else
1797                         err = -EIO;
1798                 brelse(journal->j_sb_buffer);
1799         }
1800 
1801         if (journal->j_proc_entry)
1802                 jbd2_stats_proc_exit(journal);
1803         iput(journal->j_inode);
1804         if (journal->j_revoke)
1805                 jbd2_journal_destroy_revoke(journal);
1806         if (journal->j_chksum_driver)
1807                 crypto_free_shash(journal->j_chksum_driver);
1808         kfree(journal->j_wbuf);
1809         kfree(journal);
1810 
1811         return err;
1812 }
1813 
1814 
1815 /**
1816  *int jbd2_journal_check_used_features () - Check if features specified are used.
1817  * @journal: Journal to check.
1818  * @compat: bitmask of compatible features
1819  * @ro: bitmask of features that force read-only mount
1820  * @incompat: bitmask of incompatible features
1821  *
1822  * Check whether the journal uses all of a given set of
1823  * features.  Return true (non-zero) if it does.
1824  **/
1825 
1826 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1827                                  unsigned long ro, unsigned long incompat)
1828 {
1829         journal_superblock_t *sb;
1830 
1831         if (!compat && !ro && !incompat)
1832                 return 1;
1833         /* Load journal superblock if it is not loaded yet. */
1834         if (journal->j_format_version == 0 &&
1835             journal_get_superblock(journal) != 0)
1836                 return 0;
1837         if (journal->j_format_version == 1)
1838                 return 0;
1839 
1840         sb = journal->j_superblock;
1841 
1842         if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1843             ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1844             ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1845                 return 1;
1846 
1847         return 0;
1848 }
1849 
1850 /**
1851  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1852  * @journal: Journal to check.
1853  * @compat: bitmask of compatible features
1854  * @ro: bitmask of features that force read-only mount
1855  * @incompat: bitmask of incompatible features
1856  *
1857  * Check whether the journaling code supports the use of
1858  * all of a given set of features on this journal.  Return true
1859  * (non-zero) if it can. */
1860 
1861 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1862                                       unsigned long ro, unsigned long incompat)
1863 {
1864         if (!compat && !ro && !incompat)
1865                 return 1;
1866 
1867         /* We can support any known requested features iff the
1868          * superblock is in version 2.  Otherwise we fail to support any
1869          * extended sb features. */
1870 
1871         if (journal->j_format_version != 2)
1872                 return 0;
1873 
1874         if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1875             (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1876             (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1877                 return 1;
1878 
1879         return 0;
1880 }
1881 
1882 /**
1883  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1884  * @journal: Journal to act on.
1885  * @compat: bitmask of compatible features
1886  * @ro: bitmask of features that force read-only mount
1887  * @incompat: bitmask of incompatible features
1888  *
1889  * Mark a given journal feature as present on the
1890  * superblock.  Returns true if the requested features could be set.
1891  *
1892  */
1893 
1894 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1895                           unsigned long ro, unsigned long incompat)
1896 {
1897 #define INCOMPAT_FEATURE_ON(f) \
1898                 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1899 #define COMPAT_FEATURE_ON(f) \
1900                 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1901         journal_superblock_t *sb;
1902 
1903         if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1904                 return 1;
1905 
1906         if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1907                 return 0;
1908 
1909         /* If enabling v2 checksums, turn on v3 instead */
1910         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1911                 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1912                 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1913         }
1914 
1915         /* Asking for checksumming v3 and v1?  Only give them v3. */
1916         if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1917             compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1918                 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1919 
1920         jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1921                   compat, ro, incompat);
1922 
1923         sb = journal->j_superblock;
1924 
1925         /* Load the checksum driver if necessary */
1926         if ((journal->j_chksum_driver == NULL) &&
1927             INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1928                 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1929                 if (IS_ERR(journal->j_chksum_driver)) {
1930                         printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1931                         journal->j_chksum_driver = NULL;
1932                         return 0;
1933                 }
1934                 /* Precompute checksum seed for all metadata */
1935                 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1936                                                    sizeof(sb->s_uuid));
1937         }
1938 
1939         lock_buffer(journal->j_sb_buffer);
1940 
1941         /* If enabling v3 checksums, update superblock */
1942         if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1943                 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1944                 sb->s_feature_compat &=
1945                         ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1946         }
1947 
1948         /* If enabling v1 checksums, downgrade superblock */
1949         if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1950                 sb->s_feature_incompat &=
1951                         ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1952                                      JBD2_FEATURE_INCOMPAT_CSUM_V3);
1953 
1954         sb->s_feature_compat    |= cpu_to_be32(compat);
1955         sb->s_feature_ro_compat |= cpu_to_be32(ro);
1956         sb->s_feature_incompat  |= cpu_to_be32(incompat);
1957         unlock_buffer(journal->j_sb_buffer);
1958         journal->j_revoke_records_per_block =
1959                                 journal_revoke_records_per_block(journal);
1960 
1961         return 1;
1962 #undef COMPAT_FEATURE_ON
1963 #undef INCOMPAT_FEATURE_ON
1964 }
1965 
1966 /*
1967  * jbd2_journal_clear_features () - Clear a given journal feature in the
1968  *                                  superblock
1969  * @journal: Journal to act on.
1970  * @compat: bitmask of compatible features
1971  * @ro: bitmask of features that force read-only mount
1972  * @incompat: bitmask of incompatible features
1973  *
1974  * Clear a given journal feature as present on the
1975  * superblock.
1976  */
1977 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1978                                 unsigned long ro, unsigned long incompat)
1979 {
1980         journal_superblock_t *sb;
1981 
1982         jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1983                   compat, ro, incompat);
1984 
1985         sb = journal->j_superblock;
1986 
1987         sb->s_feature_compat    &= ~cpu_to_be32(compat);
1988         sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1989         sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1990         journal->j_revoke_records_per_block =
1991                                 journal_revoke_records_per_block(journal);
1992 }
1993 EXPORT_SYMBOL(jbd2_journal_clear_features);
1994 
1995 /**
1996  * int jbd2_journal_flush () - Flush journal
1997  * @journal: Journal to act on.
1998  *
1999  * Flush all data for a given journal to disk and empty the journal.
2000  * Filesystems can use this when remounting readonly to ensure that
2001  * recovery does not need to happen on remount.
2002  */
2003 
2004 int jbd2_journal_flush(journal_t *journal)
2005 {
2006         int err = 0;
2007         transaction_t *transaction = NULL;
2008 
2009         write_lock(&journal->j_state_lock);
2010 
2011         /* Force everything buffered to the log... */
2012         if (journal->j_running_transaction) {
2013                 transaction = journal->j_running_transaction;
2014                 __jbd2_log_start_commit(journal, transaction->t_tid);
2015         } else if (journal->j_committing_transaction)
2016                 transaction = journal->j_committing_transaction;
2017 
2018         /* Wait for the log commit to complete... */
2019         if (transaction) {
2020                 tid_t tid = transaction->t_tid;
2021 
2022                 write_unlock(&journal->j_state_lock);
2023                 jbd2_log_wait_commit(journal, tid);
2024         } else {
2025                 write_unlock(&journal->j_state_lock);
2026         }
2027 
2028         /* ...and flush everything in the log out to disk. */
2029         spin_lock(&journal->j_list_lock);
2030         while (!err && journal->j_checkpoint_transactions != NULL) {
2031                 spin_unlock(&journal->j_list_lock);
2032                 mutex_lock_io(&journal->j_checkpoint_mutex);
2033                 err = jbd2_log_do_checkpoint(journal);
2034                 mutex_unlock(&journal->j_checkpoint_mutex);
2035                 spin_lock(&journal->j_list_lock);
2036         }
2037         spin_unlock(&journal->j_list_lock);
2038 
2039         if (is_journal_aborted(journal))
2040                 return -EIO;
2041 
2042         mutex_lock_io(&journal->j_checkpoint_mutex);
2043         if (!err) {
2044                 err = jbd2_cleanup_journal_tail(journal);
2045                 if (err < 0) {
2046                         mutex_unlock(&journal->j_checkpoint_mutex);
2047                         goto out;
2048                 }
2049                 err = 0;
2050         }
2051 
2052         /* Finally, mark the journal as really needing no recovery.
2053          * This sets s_start==0 in the underlying superblock, which is
2054          * the magic code for a fully-recovered superblock.  Any future
2055          * commits of data to the journal will restore the current
2056          * s_start value. */
2057         jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2058         mutex_unlock(&journal->j_checkpoint_mutex);
2059         write_lock(&journal->j_state_lock);
2060         J_ASSERT(!journal->j_running_transaction);
2061         J_ASSERT(!journal->j_committing_transaction);
2062         J_ASSERT(!journal->j_checkpoint_transactions);
2063         J_ASSERT(journal->j_head == journal->j_tail);
2064         J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2065         write_unlock(&journal->j_state_lock);
2066 out:
2067         return err;
2068 }
2069 
2070 /**
2071  * int jbd2_journal_wipe() - Wipe journal contents
2072  * @journal: Journal to act on.
2073  * @write: flag (see below)
2074  *
2075  * Wipe out all of the contents of a journal, safely.  This will produce
2076  * a warning if the journal contains any valid recovery information.
2077  * Must be called between journal_init_*() and jbd2_journal_load().
2078  *
2079  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2080  * we merely suppress recovery.
2081  */
2082 
2083 int jbd2_journal_wipe(journal_t *journal, int write)
2084 {
2085         int err = 0;
2086 
2087         J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2088 
2089         err = load_superblock(journal);
2090         if (err)
2091                 return err;
2092 
2093         if (!journal->j_tail)
2094                 goto no_recovery;
2095 
2096         printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2097                 write ? "Clearing" : "Ignoring");
2098 
2099         err = jbd2_journal_skip_recovery(journal);
2100         if (write) {
2101                 /* Lock to make assertions happy... */
2102                 mutex_lock_io(&journal->j_checkpoint_mutex);
2103                 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2104                 mutex_unlock(&journal->j_checkpoint_mutex);
2105         }
2106 
2107  no_recovery:
2108         return err;
2109 }
2110 
2111 /**
2112  * void jbd2_journal_abort () - Shutdown the journal immediately.
2113  * @journal: the journal to shutdown.
2114  * @errno:   an error number to record in the journal indicating
2115  *           the reason for the shutdown.
2116  *
2117  * Perform a complete, immediate shutdown of the ENTIRE
2118  * journal (not of a single transaction).  This operation cannot be
2119  * undone without closing and reopening the journal.
2120  *
2121  * The jbd2_journal_abort function is intended to support higher level error
2122  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2123  * mode.
2124  *
2125  * Journal abort has very specific semantics.  Any existing dirty,
2126  * unjournaled buffers in the main filesystem will still be written to
2127  * disk by bdflush, but the journaling mechanism will be suspended
2128  * immediately and no further transaction commits will be honoured.
2129  *
2130  * Any dirty, journaled buffers will be written back to disk without
2131  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2132  * filesystem, but we _do_ attempt to leave as much data as possible
2133  * behind for fsck to use for cleanup.
2134  *
2135  * Any attempt to get a new transaction handle on a journal which is in
2136  * ABORT state will just result in an -EROFS error return.  A
2137  * jbd2_journal_stop on an existing handle will return -EIO if we have
2138  * entered abort state during the update.
2139  *
2140  * Recursive transactions are not disturbed by journal abort until the
2141  * final jbd2_journal_stop, which will receive the -EIO error.
2142  *
2143  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2144  * which will be recorded (if possible) in the journal superblock.  This
2145  * allows a client to record failure conditions in the middle of a
2146  * transaction without having to complete the transaction to record the
2147  * failure to disk.  ext3_error, for example, now uses this
2148  * functionality.
2149  *
2150  */
2151 
2152 void jbd2_journal_abort(journal_t *journal, int errno)
2153 {
2154         transaction_t *transaction;
2155 
2156         /*
2157          * ESHUTDOWN always takes precedence because a file system check
2158          * caused by any other journal abort error is not required after
2159          * a shutdown triggered.
2160          */
2161         write_lock(&journal->j_state_lock);
2162         if (journal->j_flags & JBD2_ABORT) {
2163                 int old_errno = journal->j_errno;
2164 
2165                 write_unlock(&journal->j_state_lock);
2166                 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2167                         journal->j_errno = errno;
2168                         jbd2_journal_update_sb_errno(journal);
2169                 }
2170                 return;
2171         }
2172 
2173         /*
2174          * Mark the abort as occurred and start current running transaction
2175          * to release all journaled buffer.
2176          */
2177         pr_err("Aborting journal on device %s.\n", journal->j_devname);
2178 
2179         journal->j_flags |= JBD2_ABORT;
2180         journal->j_errno = errno;
2181         transaction = journal->j_running_transaction;
2182         if (transaction)
2183                 __jbd2_log_start_commit(journal, transaction->t_tid);
2184         write_unlock(&journal->j_state_lock);
2185 
2186         /*
2187          * Record errno to the journal super block, so that fsck and jbd2
2188          * layer could realise that a filesystem check is needed.
2189          */
2190         jbd2_journal_update_sb_errno(journal);
2191 
2192         write_lock(&journal->j_state_lock);
2193         journal->j_flags |= JBD2_REC_ERR;
2194         write_unlock(&journal->j_state_lock);
2195 }
2196 
2197 /**
2198  * int jbd2_journal_errno () - returns the journal's error state.
2199  * @journal: journal to examine.
2200  *
2201  * This is the errno number set with jbd2_journal_abort(), the last
2202  * time the journal was mounted - if the journal was stopped
2203  * without calling abort this will be 0.
2204  *
2205  * If the journal has been aborted on this mount time -EROFS will
2206  * be returned.
2207  */
2208 int jbd2_journal_errno(journal_t *journal)
2209 {
2210         int err;
2211 
2212         read_lock(&journal->j_state_lock);
2213         if (journal->j_flags & JBD2_ABORT)
2214                 err = -EROFS;
2215         else
2216                 err = journal->j_errno;
2217         read_unlock(&journal->j_state_lock);
2218         return err;
2219 }
2220 
2221 /**
2222  * int jbd2_journal_clear_err () - clears the journal's error state
2223  * @journal: journal to act on.
2224  *
2225  * An error must be cleared or acked to take a FS out of readonly
2226  * mode.
2227  */
2228 int jbd2_journal_clear_err(journal_t *journal)
2229 {
2230         int err = 0;
2231 
2232         write_lock(&journal->j_state_lock);
2233         if (journal->j_flags & JBD2_ABORT)
2234                 err = -EROFS;
2235         else
2236                 journal->j_errno = 0;
2237         write_unlock(&journal->j_state_lock);
2238         return err;
2239 }
2240 
2241 /**
2242  * void jbd2_journal_ack_err() - Ack journal err.
2243  * @journal: journal to act on.
2244  *
2245  * An error must be cleared or acked to take a FS out of readonly
2246  * mode.
2247  */
2248 void jbd2_journal_ack_err(journal_t *journal)
2249 {
2250         write_lock(&journal->j_state_lock);
2251         if (journal->j_errno)
2252                 journal->j_flags |= JBD2_ACK_ERR;
2253         write_unlock(&journal->j_state_lock);
2254 }
2255 
2256 int jbd2_journal_blocks_per_page(struct inode *inode)
2257 {
2258         return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2259 }
2260 
2261 /*
2262  * helper functions to deal with 32 or 64bit block numbers.
2263  */
2264 size_t journal_tag_bytes(journal_t *journal)
2265 {
2266         size_t sz;
2267 
2268         if (jbd2_has_feature_csum3(journal))
2269                 return sizeof(journal_block_tag3_t);
2270 
2271         sz = sizeof(journal_block_tag_t);
2272 
2273         if (jbd2_has_feature_csum2(journal))
2274                 sz += sizeof(__u16);
2275 
2276         if (jbd2_has_feature_64bit(journal))
2277                 return sz;
2278         else
2279                 return sz - sizeof(__u32);
2280 }
2281 
2282 /*
2283  * JBD memory management
2284  *
2285  * These functions are used to allocate block-sized chunks of memory
2286  * used for making copies of buffer_head data.  Very often it will be
2287  * page-sized chunks of data, but sometimes it will be in
2288  * sub-page-size chunks.  (For example, 16k pages on Power systems
2289  * with a 4k block file system.)  For blocks smaller than a page, we
2290  * use a SLAB allocator.  There are slab caches for each block size,
2291  * which are allocated at mount time, if necessary, and we only free
2292  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2293  * this reason we don't need to a mutex to protect access to
2294  * jbd2_slab[] allocating or releasing memory; only in
2295  * jbd2_journal_create_slab().
2296  */
2297 #define JBD2_MAX_SLABS 8
2298 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2299 
2300 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2301         "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2302         "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2303 };
2304 
2305 
2306 static void jbd2_journal_destroy_slabs(void)
2307 {
2308         int i;
2309 
2310         for (i = 0; i < JBD2_MAX_SLABS; i++) {
2311                 kmem_cache_destroy(jbd2_slab[i]);
2312                 jbd2_slab[i] = NULL;
2313         }
2314 }
2315 
2316 static int jbd2_journal_create_slab(size_t size)
2317 {
2318         static DEFINE_MUTEX(jbd2_slab_create_mutex);
2319         int i = order_base_2(size) - 10;
2320         size_t slab_size;
2321 
2322         if (size == PAGE_SIZE)
2323                 return 0;
2324 
2325         if (i >= JBD2_MAX_SLABS)
2326                 return -EINVAL;
2327 
2328         if (unlikely(i < 0))
2329                 i = 0;
2330         mutex_lock(&jbd2_slab_create_mutex);
2331         if (jbd2_slab[i]) {
2332                 mutex_unlock(&jbd2_slab_create_mutex);
2333                 return 0;       /* Already created */
2334         }
2335 
2336         slab_size = 1 << (i+10);
2337         jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2338                                          slab_size, 0, NULL);
2339         mutex_unlock(&jbd2_slab_create_mutex);
2340         if (!jbd2_slab[i]) {
2341                 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2342                 return -ENOMEM;
2343         }
2344         return 0;
2345 }
2346 
2347 static struct kmem_cache *get_slab(size_t size)
2348 {
2349         int i = order_base_2(size) - 10;
2350 
2351         BUG_ON(i >= JBD2_MAX_SLABS);
2352         if (unlikely(i < 0))
2353                 i = 0;
2354         BUG_ON(jbd2_slab[i] == NULL);
2355         return jbd2_slab[i];
2356 }
2357 
2358 void *jbd2_alloc(size_t size, gfp_t flags)
2359 {
2360         void *ptr;
2361 
2362         BUG_ON(size & (size-1)); /* Must be a power of 2 */
2363 
2364         if (size < PAGE_SIZE)
2365                 ptr = kmem_cache_alloc(get_slab(size), flags);
2366         else
2367                 ptr = (void *)__get_free_pages(flags, get_order(size));
2368 
2369         /* Check alignment; SLUB has gotten this wrong in the past,
2370          * and this can lead to user data corruption! */
2371         BUG_ON(((unsigned long) ptr) & (size-1));
2372 
2373         return ptr;
2374 }
2375 
2376 void jbd2_free(void *ptr, size_t size)
2377 {
2378         if (size < PAGE_SIZE)
2379                 kmem_cache_free(get_slab(size), ptr);
2380         else
2381                 free_pages((unsigned long)ptr, get_order(size));
2382 };
2383 
2384 /*
2385  * Journal_head storage management
2386  */
2387 static struct kmem_cache *jbd2_journal_head_cache;
2388 #ifdef CONFIG_JBD2_DEBUG
2389 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2390 #endif
2391 
2392 static int __init jbd2_journal_init_journal_head_cache(void)
2393 {
2394         J_ASSERT(!jbd2_journal_head_cache);
2395         jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2396                                 sizeof(struct journal_head),
2397                                 0,              /* offset */
2398                                 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2399                                 NULL);          /* ctor */
2400         if (!jbd2_journal_head_cache) {
2401                 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2402                 return -ENOMEM;
2403         }
2404         return 0;
2405 }
2406 
2407 static void jbd2_journal_destroy_journal_head_cache(void)
2408 {
2409         kmem_cache_destroy(jbd2_journal_head_cache);
2410         jbd2_journal_head_cache = NULL;
2411 }
2412 
2413 /*
2414  * journal_head splicing and dicing
2415  */
2416 static struct journal_head *journal_alloc_journal_head(void)
2417 {
2418         struct journal_head *ret;
2419 
2420 #ifdef CONFIG_JBD2_DEBUG
2421         atomic_inc(&nr_journal_heads);
2422 #endif
2423         ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2424         if (!ret) {
2425                 jbd_debug(1, "out of memory for journal_head\n");
2426                 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2427                 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2428                                 GFP_NOFS | __GFP_NOFAIL);
2429         }
2430         if (ret)
2431                 spin_lock_init(&ret->b_state_lock);
2432         return ret;
2433 }
2434 
2435 static void journal_free_journal_head(struct journal_head *jh)
2436 {
2437 #ifdef CONFIG_JBD2_DEBUG
2438         atomic_dec(&nr_journal_heads);
2439         memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2440 #endif
2441         kmem_cache_free(jbd2_journal_head_cache, jh);
2442 }
2443 
2444 /*
2445  * A journal_head is attached to a buffer_head whenever JBD has an
2446  * interest in the buffer.
2447  *
2448  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2449  * is set.  This bit is tested in core kernel code where we need to take
2450  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2451  * there.
2452  *
2453  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2454  *
2455  * When a buffer has its BH_JBD bit set it is immune from being released by
2456  * core kernel code, mainly via ->b_count.
2457  *
2458  * A journal_head is detached from its buffer_head when the journal_head's
2459  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2460  * transaction (b_cp_transaction) hold their references to b_jcount.
2461  *
2462  * Various places in the kernel want to attach a journal_head to a buffer_head
2463  * _before_ attaching the journal_head to a transaction.  To protect the
2464  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2465  * journal_head's b_jcount refcount by one.  The caller must call
2466  * jbd2_journal_put_journal_head() to undo this.
2467  *
2468  * So the typical usage would be:
2469  *
2470  *      (Attach a journal_head if needed.  Increments b_jcount)
2471  *      struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2472  *      ...
2473  *      (Get another reference for transaction)
2474  *      jbd2_journal_grab_journal_head(bh);
2475  *      jh->b_transaction = xxx;
2476  *      (Put original reference)
2477  *      jbd2_journal_put_journal_head(jh);
2478  */
2479 
2480 /*
2481  * Give a buffer_head a journal_head.
2482  *
2483  * May sleep.
2484  */
2485 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2486 {
2487         struct journal_head *jh;
2488         struct journal_head *new_jh = NULL;
2489 
2490 repeat:
2491         if (!buffer_jbd(bh))
2492                 new_jh = journal_alloc_journal_head();
2493 
2494         jbd_lock_bh_journal_head(bh);
2495         if (buffer_jbd(bh)) {
2496                 jh = bh2jh(bh);
2497         } else {
2498                 J_ASSERT_BH(bh,
2499                         (atomic_read(&bh->b_count) > 0) ||
2500                         (bh->b_page && bh->b_page->mapping));
2501 
2502                 if (!new_jh) {
2503                         jbd_unlock_bh_journal_head(bh);
2504                         goto repeat;
2505                 }
2506 
2507                 jh = new_jh;
2508                 new_jh = NULL;          /* We consumed it */
2509                 set_buffer_jbd(bh);
2510                 bh->b_private = jh;
2511                 jh->b_bh = bh;
2512                 get_bh(bh);
2513                 BUFFER_TRACE(bh, "added journal_head");
2514         }
2515         jh->b_jcount++;
2516         jbd_unlock_bh_journal_head(bh);
2517         if (new_jh)
2518                 journal_free_journal_head(new_jh);
2519         return bh->b_private;
2520 }
2521 
2522 /*
2523  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2524  * having a journal_head, return NULL
2525  */
2526 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2527 {
2528         struct journal_head *jh = NULL;
2529 
2530         jbd_lock_bh_journal_head(bh);
2531         if (buffer_jbd(bh)) {
2532                 jh = bh2jh(bh);
2533                 jh->b_jcount++;
2534         }
2535         jbd_unlock_bh_journal_head(bh);
2536         return jh;
2537 }
2538 
2539 static void __journal_remove_journal_head(struct buffer_head *bh)
2540 {
2541         struct journal_head *jh = bh2jh(bh);
2542 
2543         J_ASSERT_JH(jh, jh->b_transaction == NULL);
2544         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2545         J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2546         J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2547         J_ASSERT_BH(bh, buffer_jbd(bh));
2548         J_ASSERT_BH(bh, jh2bh(jh) == bh);
2549         BUFFER_TRACE(bh, "remove journal_head");
2550 
2551         /* Unlink before dropping the lock */
2552         bh->b_private = NULL;
2553         jh->b_bh = NULL;        /* debug, really */
2554         clear_buffer_jbd(bh);
2555 }
2556 
2557 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2558 {
2559         if (jh->b_frozen_data) {
2560                 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2561                 jbd2_free(jh->b_frozen_data, b_size);
2562         }
2563         if (jh->b_committed_data) {
2564                 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2565                 jbd2_free(jh->b_committed_data, b_size);
2566         }
2567         journal_free_journal_head(jh);
2568 }
2569 
2570 /*
2571  * Drop a reference on the passed journal_head.  If it fell to zero then
2572  * release the journal_head from the buffer_head.
2573  */
2574 void jbd2_journal_put_journal_head(struct journal_head *jh)
2575 {
2576         struct buffer_head *bh = jh2bh(jh);
2577 
2578         jbd_lock_bh_journal_head(bh);
2579         J_ASSERT_JH(jh, jh->b_jcount > 0);
2580         --jh->b_jcount;
2581         if (!jh->b_jcount) {
2582                 __journal_remove_journal_head(bh);
2583                 jbd_unlock_bh_journal_head(bh);
2584                 journal_release_journal_head(jh, bh->b_size);
2585                 __brelse(bh);
2586         } else {
2587                 jbd_unlock_bh_journal_head(bh);
2588         }
2589 }
2590 
2591 /*
2592  * Initialize jbd inode head
2593  */
2594 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2595 {
2596         jinode->i_transaction = NULL;
2597         jinode->i_next_transaction = NULL;
2598         jinode->i_vfs_inode = inode;
2599         jinode->i_flags = 0;
2600         jinode->i_dirty_start = 0;
2601         jinode->i_dirty_end = 0;
2602         INIT_LIST_HEAD(&jinode->i_list);
2603 }
2604 
2605 /*
2606  * Function to be called before we start removing inode from memory (i.e.,
2607  * clear_inode() is a fine place to be called from). It removes inode from
2608  * transaction's lists.
2609  */
2610 void jbd2_journal_release_jbd_inode(journal_t *journal,
2611                                     struct jbd2_inode *jinode)
2612 {
2613         if (!journal)
2614                 return;
2615 restart:
2616         spin_lock(&journal->j_list_lock);
2617         /* Is commit writing out inode - we have to wait */
2618         if (jinode->i_flags & JI_COMMIT_RUNNING) {
2619                 wait_queue_head_t *wq;
2620                 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2621                 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2622                 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2623                 spin_unlock(&journal->j_list_lock);
2624                 schedule();
2625                 finish_wait(wq, &wait.wq_entry);
2626                 goto restart;
2627         }
2628 
2629         if (jinode->i_transaction) {
2630                 list_del(&jinode->i_list);
2631                 jinode->i_transaction = NULL;
2632         }
2633         spin_unlock(&journal->j_list_lock);
2634 }
2635 
2636 
2637 #ifdef CONFIG_PROC_FS
2638 
2639 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2640 
2641 static void __init jbd2_create_jbd_stats_proc_entry(void)
2642 {
2643         proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2644 }
2645 
2646 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2647 {
2648         if (proc_jbd2_stats)
2649                 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2650 }
2651 
2652 #else
2653 
2654 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2655 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2656 
2657 #endif
2658 
2659 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2660 
2661 static int __init jbd2_journal_init_inode_cache(void)
2662 {
2663         J_ASSERT(!jbd2_inode_cache);
2664         jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2665         if (!jbd2_inode_cache) {
2666                 pr_emerg("JBD2: failed to create inode cache\n");
2667                 return -ENOMEM;
2668         }
2669         return 0;
2670 }
2671 
2672 static int __init jbd2_journal_init_handle_cache(void)
2673 {
2674         J_ASSERT(!jbd2_handle_cache);
2675         jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2676         if (!jbd2_handle_cache) {
2677                 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2678                 return -ENOMEM;
2679         }
2680         return 0;
2681 }
2682 
2683 static void jbd2_journal_destroy_inode_cache(void)
2684 {
2685         kmem_cache_destroy(jbd2_inode_cache);
2686         jbd2_inode_cache = NULL;
2687 }
2688 
2689 static void jbd2_journal_destroy_handle_cache(void)
2690 {
2691         kmem_cache_destroy(jbd2_handle_cache);
2692         jbd2_handle_cache = NULL;
2693 }
2694 
2695 /*
2696  * Module startup and shutdown
2697  */
2698 
2699 static int __init journal_init_caches(void)
2700 {
2701         int ret;
2702 
2703         ret = jbd2_journal_init_revoke_record_cache();
2704         if (ret == 0)
2705                 ret = jbd2_journal_init_revoke_table_cache();
2706         if (ret == 0)
2707                 ret = jbd2_journal_init_journal_head_cache();
2708         if (ret == 0)
2709                 ret = jbd2_journal_init_handle_cache();
2710         if (ret == 0)
2711                 ret = jbd2_journal_init_inode_cache();
2712         if (ret == 0)
2713                 ret = jbd2_journal_init_transaction_cache();
2714         return ret;
2715 }
2716 
2717 static void jbd2_journal_destroy_caches(void)
2718 {
2719         jbd2_journal_destroy_revoke_record_cache();
2720         jbd2_journal_destroy_revoke_table_cache();
2721         jbd2_journal_destroy_journal_head_cache();
2722         jbd2_journal_destroy_handle_cache();
2723         jbd2_journal_destroy_inode_cache();
2724         jbd2_journal_destroy_transaction_cache();
2725         jbd2_journal_destroy_slabs();
2726 }
2727 
2728 static int __init journal_init(void)
2729 {
2730         int ret;
2731 
2732         BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2733 
2734         ret = journal_init_caches();
2735         if (ret == 0) {
2736                 jbd2_create_jbd_stats_proc_entry();
2737         } else {
2738                 jbd2_journal_destroy_caches();
2739         }
2740         return ret;
2741 }
2742 
2743 static void __exit journal_exit(void)
2744 {
2745 #ifdef CONFIG_JBD2_DEBUG
2746         int n = atomic_read(&nr_journal_heads);
2747         if (n)
2748                 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2749 #endif
2750         jbd2_remove_jbd_stats_proc_entry();
2751         jbd2_journal_destroy_caches();
2752 }
2753 
2754 MODULE_LICENSE("GPL");
2755 module_init(journal_init);
2756 module_exit(journal_exit);
2757 
2758 

~ [ 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