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 static ushort jbd2_journal_enable_debug __read_mostly;
53
54 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
55 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
56 #endif
57
58 EXPORT_SYMBOL(jbd2_journal_extend);
59 EXPORT_SYMBOL(jbd2_journal_stop);
60 EXPORT_SYMBOL(jbd2_journal_lock_updates);
61 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
62 EXPORT_SYMBOL(jbd2_journal_get_write_access);
63 EXPORT_SYMBOL(jbd2_journal_get_create_access);
64 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
65 EXPORT_SYMBOL(jbd2_journal_set_triggers);
66 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
67 EXPORT_SYMBOL(jbd2_journal_forget);
68 EXPORT_SYMBOL(jbd2_journal_flush);
69 EXPORT_SYMBOL(jbd2_journal_revoke);
70
71 EXPORT_SYMBOL(jbd2_journal_init_dev);
72 EXPORT_SYMBOL(jbd2_journal_init_inode);
73 EXPORT_SYMBOL(jbd2_journal_check_used_features);
74 EXPORT_SYMBOL(jbd2_journal_check_available_features);
75 EXPORT_SYMBOL(jbd2_journal_set_features);
76 EXPORT_SYMBOL(jbd2_journal_load);
77 EXPORT_SYMBOL(jbd2_journal_destroy);
78 EXPORT_SYMBOL(jbd2_journal_abort);
79 EXPORT_SYMBOL(jbd2_journal_errno);
80 EXPORT_SYMBOL(jbd2_journal_ack_err);
81 EXPORT_SYMBOL(jbd2_journal_clear_err);
82 EXPORT_SYMBOL(jbd2_log_wait_commit);
83 EXPORT_SYMBOL(jbd2_journal_start_commit);
84 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
85 EXPORT_SYMBOL(jbd2_journal_wipe);
86 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
87 EXPORT_SYMBOL(jbd2_journal_invalidate_folio);
88 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
89 EXPORT_SYMBOL(jbd2_journal_force_commit);
90 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
91 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
92 EXPORT_SYMBOL(jbd2_journal_submit_inode_data_buffers);
93 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers);
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
__jbd2_debug(int level,const char * file,const char * func,unsigned int line,const char * fmt,...)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 #endif
117
118 /* Checksumming functions */
jbd2_verify_csum_type(journal_t * j,journal_superblock_t * sb)119 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
120 {
121 if (!jbd2_journal_has_csum_v2or3_feature(j))
122 return 1;
123
124 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
125 }
126
jbd2_superblock_csum(journal_t * j,journal_superblock_t * sb)127 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
128 {
129 __u32 csum;
130 __be32 old_csum;
131
132 old_csum = sb->s_checksum;
133 sb->s_checksum = 0;
134 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
135 sb->s_checksum = old_csum;
136
137 return cpu_to_be32(csum);
138 }
139
140 /*
141 * Helper function used to manage commit timeouts
142 */
143
commit_timeout(struct timer_list * t)144 static void commit_timeout(struct timer_list *t)
145 {
146 journal_t *journal = from_timer(journal, t, j_commit_timer);
147
148 wake_up_process(journal->j_task);
149 }
150
151 /*
152 * kjournald2: The main thread function used to manage a logging device
153 * journal.
154 *
155 * This kernel thread is responsible for two things:
156 *
157 * 1) COMMIT: Every so often we need to commit the current state of the
158 * filesystem to disk. The journal thread is responsible for writing
159 * all of the metadata buffers to disk. If a fast commit is ongoing
160 * journal thread waits until it's done and then continues from
161 * there on.
162 *
163 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
164 * of the data in that part of the log has been rewritten elsewhere on
165 * the disk. Flushing these old buffers to reclaim space in the log is
166 * known as checkpointing, and this thread is responsible for that job.
167 */
168
kjournald2(void * arg)169 static int kjournald2(void *arg)
170 {
171 journal_t *journal = arg;
172 transaction_t *transaction;
173
174 /*
175 * Set up an interval timer which can be used to trigger a commit wakeup
176 * after the commit interval expires
177 */
178 timer_setup(&journal->j_commit_timer, commit_timeout, 0);
179
180 set_freezable();
181
182 /* Record that the journal thread is running */
183 journal->j_task = current;
184 wake_up(&journal->j_wait_done_commit);
185
186 /*
187 * Make sure that no allocations from this kernel thread will ever
188 * recurse to the fs layer because we are responsible for the
189 * transaction commit and any fs involvement might get stuck waiting for
190 * the trasn. commit.
191 */
192 memalloc_nofs_save();
193
194 /*
195 * And now, wait forever for commit wakeup events.
196 */
197 write_lock(&journal->j_state_lock);
198
199 loop:
200 if (journal->j_flags & JBD2_UNMOUNT)
201 goto end_loop;
202
203 jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n",
204 journal->j_commit_sequence, journal->j_commit_request);
205
206 if (journal->j_commit_sequence != journal->j_commit_request) {
207 jbd2_debug(1, "OK, requests differ\n");
208 write_unlock(&journal->j_state_lock);
209 del_timer_sync(&journal->j_commit_timer);
210 jbd2_journal_commit_transaction(journal);
211 write_lock(&journal->j_state_lock);
212 goto loop;
213 }
214
215 wake_up(&journal->j_wait_done_commit);
216 if (freezing(current)) {
217 /*
218 * The simpler the better. Flushing journal isn't a
219 * good idea, because that depends on threads that may
220 * be already stopped.
221 */
222 jbd2_debug(1, "Now suspending kjournald2\n");
223 write_unlock(&journal->j_state_lock);
224 try_to_freeze();
225 write_lock(&journal->j_state_lock);
226 } else {
227 /*
228 * We assume on resume that commits are already there,
229 * so we don't sleep
230 */
231 DEFINE_WAIT(wait);
232 int should_sleep = 1;
233
234 prepare_to_wait(&journal->j_wait_commit, &wait,
235 TASK_INTERRUPTIBLE);
236 if (journal->j_commit_sequence != journal->j_commit_request)
237 should_sleep = 0;
238 transaction = journal->j_running_transaction;
239 if (transaction && time_after_eq(jiffies,
240 transaction->t_expires))
241 should_sleep = 0;
242 if (journal->j_flags & JBD2_UNMOUNT)
243 should_sleep = 0;
244 if (should_sleep) {
245 write_unlock(&journal->j_state_lock);
246 schedule();
247 write_lock(&journal->j_state_lock);
248 }
249 finish_wait(&journal->j_wait_commit, &wait);
250 }
251
252 jbd2_debug(1, "kjournald2 wakes\n");
253
254 /*
255 * Were we woken up by a commit wakeup event?
256 */
257 transaction = journal->j_running_transaction;
258 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
259 journal->j_commit_request = transaction->t_tid;
260 jbd2_debug(1, "woke because of timeout\n");
261 }
262 goto loop;
263
264 end_loop:
265 del_timer_sync(&journal->j_commit_timer);
266 journal->j_task = NULL;
267 wake_up(&journal->j_wait_done_commit);
268 jbd2_debug(1, "Journal thread exiting.\n");
269 write_unlock(&journal->j_state_lock);
270 return 0;
271 }
272
jbd2_journal_start_thread(journal_t * journal)273 static int jbd2_journal_start_thread(journal_t *journal)
274 {
275 struct task_struct *t;
276
277 t = kthread_run(kjournald2, journal, "jbd2/%s",
278 journal->j_devname);
279 if (IS_ERR(t))
280 return PTR_ERR(t);
281
282 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
283 return 0;
284 }
285
journal_kill_thread(journal_t * journal)286 static void journal_kill_thread(journal_t *journal)
287 {
288 write_lock(&journal->j_state_lock);
289 journal->j_flags |= JBD2_UNMOUNT;
290
291 while (journal->j_task) {
292 write_unlock(&journal->j_state_lock);
293 wake_up(&journal->j_wait_commit);
294 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
295 write_lock(&journal->j_state_lock);
296 }
297 write_unlock(&journal->j_state_lock);
298 }
299
300 /*
301 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
302 *
303 * Writes a metadata buffer to a given disk block. The actual IO is not
304 * performed but a new buffer_head is constructed which labels the data
305 * to be written with the correct destination disk block.
306 *
307 * Any magic-number escaping which needs to be done will cause a
308 * copy-out here. If the buffer happens to start with the
309 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
310 * magic number is only written to the log for descripter blocks. In
311 * this case, we copy the data and replace the first word with 0, and we
312 * return a result code which indicates that this buffer needs to be
313 * marked as an escaped buffer in the corresponding log descriptor
314 * block. The missing word can then be restored when the block is read
315 * during recovery.
316 *
317 * If the source buffer has already been modified by a new transaction
318 * since we took the last commit snapshot, we use the frozen copy of
319 * that data for IO. If we end up using the existing buffer_head's data
320 * for the write, then we have to make sure nobody modifies it while the
321 * IO is in progress. do_get_write_access() handles this.
322 *
323 * The function returns a pointer to the buffer_head to be used for IO.
324 *
325 *
326 * Return value:
327 * <0: Error
328 * >=0: Finished OK
329 *
330 * On success:
331 * Bit 0 set == escape performed on the data
332 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
333 */
334
jbd2_journal_write_metadata_buffer(transaction_t * transaction,struct journal_head * jh_in,struct buffer_head ** bh_out,sector_t blocknr)335 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
336 struct journal_head *jh_in,
337 struct buffer_head **bh_out,
338 sector_t blocknr)
339 {
340 int need_copy_out = 0;
341 int done_copy_out = 0;
342 int do_escape = 0;
343 char *mapped_data;
344 struct buffer_head *new_bh;
345 struct page *new_page;
346 unsigned int new_offset;
347 struct buffer_head *bh_in = jh2bh(jh_in);
348 journal_t *journal = transaction->t_journal;
349
350 /*
351 * The buffer really shouldn't be locked: only the current committing
352 * transaction is allowed to write it, so nobody else is allowed
353 * to do any IO.
354 *
355 * akpm: except if we're journalling data, and write() output is
356 * also part of a shared mapping, and another thread has
357 * decided to launch a writepage() against this buffer.
358 */
359 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
360
361 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
362
363 /* keep subsequent assertions sane */
364 atomic_set(&new_bh->b_count, 1);
365
366 spin_lock(&jh_in->b_state_lock);
367 repeat:
368 /*
369 * If a new transaction has already done a buffer copy-out, then
370 * we use that version of the data for the commit.
371 */
372 if (jh_in->b_frozen_data) {
373 done_copy_out = 1;
374 new_page = virt_to_page(jh_in->b_frozen_data);
375 new_offset = offset_in_page(jh_in->b_frozen_data);
376 } else {
377 new_page = jh2bh(jh_in)->b_page;
378 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
379 }
380
381 mapped_data = kmap_atomic(new_page);
382 /*
383 * Fire data frozen trigger if data already wasn't frozen. Do this
384 * before checking for escaping, as the trigger may modify the magic
385 * offset. If a copy-out happens afterwards, it will have the correct
386 * data in the buffer.
387 */
388 if (!done_copy_out)
389 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
390 jh_in->b_triggers);
391
392 /*
393 * Check for escaping
394 */
395 if (*((__be32 *)(mapped_data + new_offset)) ==
396 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
397 need_copy_out = 1;
398 do_escape = 1;
399 }
400 kunmap_atomic(mapped_data);
401
402 /*
403 * Do we need to do a data copy?
404 */
405 if (need_copy_out && !done_copy_out) {
406 char *tmp;
407
408 spin_unlock(&jh_in->b_state_lock);
409 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
410 if (!tmp) {
411 brelse(new_bh);
412 return -ENOMEM;
413 }
414 spin_lock(&jh_in->b_state_lock);
415 if (jh_in->b_frozen_data) {
416 jbd2_free(tmp, bh_in->b_size);
417 goto repeat;
418 }
419
420 jh_in->b_frozen_data = tmp;
421 mapped_data = kmap_atomic(new_page);
422 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
423 kunmap_atomic(mapped_data);
424
425 new_page = virt_to_page(tmp);
426 new_offset = offset_in_page(tmp);
427 done_copy_out = 1;
428
429 /*
430 * This isn't strictly necessary, as we're using frozen
431 * data for the escaping, but it keeps consistency with
432 * b_frozen_data usage.
433 */
434 jh_in->b_frozen_triggers = jh_in->b_triggers;
435 }
436
437 /*
438 * Did we need to do an escaping? Now we've done all the
439 * copying, we can finally do so.
440 */
441 if (do_escape) {
442 mapped_data = kmap_atomic(new_page);
443 *((unsigned int *)(mapped_data + new_offset)) = 0;
444 kunmap_atomic(mapped_data);
445 }
446
447 set_bh_page(new_bh, new_page, new_offset);
448 new_bh->b_size = bh_in->b_size;
449 new_bh->b_bdev = journal->j_dev;
450 new_bh->b_blocknr = blocknr;
451 new_bh->b_private = bh_in;
452 set_buffer_mapped(new_bh);
453 set_buffer_dirty(new_bh);
454
455 *bh_out = new_bh;
456
457 /*
458 * The to-be-written buffer needs to get moved to the io queue,
459 * and the original buffer whose contents we are shadowing or
460 * copying is moved to the transaction's shadow queue.
461 */
462 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
463 spin_lock(&journal->j_list_lock);
464 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
465 spin_unlock(&journal->j_list_lock);
466 set_buffer_shadow(bh_in);
467 spin_unlock(&jh_in->b_state_lock);
468
469 return do_escape | (done_copy_out << 1);
470 }
471
472 /*
473 * Allocation code for the journal file. Manage the space left in the
474 * journal, so that we can begin checkpointing when appropriate.
475 */
476
477 /*
478 * Called with j_state_lock locked for writing.
479 * Returns true if a transaction commit was started.
480 */
__jbd2_log_start_commit(journal_t * journal,tid_t target)481 static int __jbd2_log_start_commit(journal_t *journal, tid_t target)
482 {
483 /* Return if the txn has already requested to be committed */
484 if (journal->j_commit_request == target)
485 return 0;
486
487 /*
488 * The only transaction we can possibly wait upon is the
489 * currently running transaction (if it exists). Otherwise,
490 * the target tid must be an old one.
491 */
492 if (journal->j_running_transaction &&
493 journal->j_running_transaction->t_tid == target) {
494 /*
495 * We want a new commit: OK, mark the request and wakeup the
496 * commit thread. We do _not_ do the commit ourselves.
497 */
498
499 journal->j_commit_request = target;
500 jbd2_debug(1, "JBD2: requesting commit %u/%u\n",
501 journal->j_commit_request,
502 journal->j_commit_sequence);
503 journal->j_running_transaction->t_requested = jiffies;
504 wake_up(&journal->j_wait_commit);
505 return 1;
506 } else if (!tid_geq(journal->j_commit_request, target))
507 /* This should never happen, but if it does, preserve
508 the evidence before kjournald goes into a loop and
509 increments j_commit_sequence beyond all recognition. */
510 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
511 journal->j_commit_request,
512 journal->j_commit_sequence,
513 target, journal->j_running_transaction ?
514 journal->j_running_transaction->t_tid : 0);
515 return 0;
516 }
517
jbd2_log_start_commit(journal_t * journal,tid_t tid)518 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
519 {
520 int ret;
521
522 write_lock(&journal->j_state_lock);
523 ret = __jbd2_log_start_commit(journal, tid);
524 write_unlock(&journal->j_state_lock);
525 return ret;
526 }
527
528 /*
529 * Force and wait any uncommitted transactions. We can only force the running
530 * transaction if we don't have an active handle, otherwise, we will deadlock.
531 * Returns: <0 in case of error,
532 * 0 if nothing to commit,
533 * 1 if transaction was successfully committed.
534 */
__jbd2_journal_force_commit(journal_t * journal)535 static int __jbd2_journal_force_commit(journal_t *journal)
536 {
537 transaction_t *transaction = NULL;
538 tid_t tid;
539 int need_to_start = 0, ret = 0;
540
541 read_lock(&journal->j_state_lock);
542 if (journal->j_running_transaction && !current->journal_info) {
543 transaction = journal->j_running_transaction;
544 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
545 need_to_start = 1;
546 } else if (journal->j_committing_transaction)
547 transaction = journal->j_committing_transaction;
548
549 if (!transaction) {
550 /* Nothing to commit */
551 read_unlock(&journal->j_state_lock);
552 return 0;
553 }
554 tid = transaction->t_tid;
555 read_unlock(&journal->j_state_lock);
556 if (need_to_start)
557 jbd2_log_start_commit(journal, tid);
558 ret = jbd2_log_wait_commit(journal, tid);
559 if (!ret)
560 ret = 1;
561
562 return ret;
563 }
564
565 /**
566 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
567 * calling process is not within transaction.
568 *
569 * @journal: journal to force
570 * Returns true if progress was made.
571 *
572 * This is used for forcing out undo-protected data which contains
573 * bitmaps, when the fs is running out of space.
574 */
jbd2_journal_force_commit_nested(journal_t * journal)575 int jbd2_journal_force_commit_nested(journal_t *journal)
576 {
577 int ret;
578
579 ret = __jbd2_journal_force_commit(journal);
580 return ret > 0;
581 }
582
583 /**
584 * jbd2_journal_force_commit() - force any uncommitted transactions
585 * @journal: journal to force
586 *
587 * Caller want unconditional commit. We can only force the running transaction
588 * if we don't have an active handle, otherwise, we will deadlock.
589 */
jbd2_journal_force_commit(journal_t * journal)590 int jbd2_journal_force_commit(journal_t *journal)
591 {
592 int ret;
593
594 J_ASSERT(!current->journal_info);
595 ret = __jbd2_journal_force_commit(journal);
596 if (ret > 0)
597 ret = 0;
598 return ret;
599 }
600
601 /*
602 * Start a commit of the current running transaction (if any). Returns true
603 * if a transaction is going to be committed (or is currently already
604 * committing), and fills its tid in at *ptid
605 */
jbd2_journal_start_commit(journal_t * journal,tid_t * ptid)606 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
607 {
608 int ret = 0;
609
610 write_lock(&journal->j_state_lock);
611 if (journal->j_running_transaction) {
612 tid_t tid = journal->j_running_transaction->t_tid;
613
614 __jbd2_log_start_commit(journal, tid);
615 /* There's a running transaction and we've just made sure
616 * it's commit has been scheduled. */
617 if (ptid)
618 *ptid = tid;
619 ret = 1;
620 } else if (journal->j_committing_transaction) {
621 /*
622 * If commit has been started, then we have to wait for
623 * completion of that transaction.
624 */
625 if (ptid)
626 *ptid = journal->j_committing_transaction->t_tid;
627 ret = 1;
628 }
629 write_unlock(&journal->j_state_lock);
630 return ret;
631 }
632
633 /*
634 * Return 1 if a given transaction has not yet sent barrier request
635 * connected with a transaction commit. If 0 is returned, transaction
636 * may or may not have sent the barrier. Used to avoid sending barrier
637 * twice in common cases.
638 */
jbd2_trans_will_send_data_barrier(journal_t * journal,tid_t tid)639 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
640 {
641 int ret = 0;
642 transaction_t *commit_trans;
643
644 if (!(journal->j_flags & JBD2_BARRIER))
645 return 0;
646 read_lock(&journal->j_state_lock);
647 /* Transaction already committed? */
648 if (tid_geq(journal->j_commit_sequence, tid))
649 goto out;
650 commit_trans = journal->j_committing_transaction;
651 if (!commit_trans || commit_trans->t_tid != tid) {
652 ret = 1;
653 goto out;
654 }
655 /*
656 * Transaction is being committed and we already proceeded to
657 * submitting a flush to fs partition?
658 */
659 if (journal->j_fs_dev != journal->j_dev) {
660 if (!commit_trans->t_need_data_flush ||
661 commit_trans->t_state >= T_COMMIT_DFLUSH)
662 goto out;
663 } else {
664 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
665 goto out;
666 }
667 ret = 1;
668 out:
669 read_unlock(&journal->j_state_lock);
670 return ret;
671 }
672 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
673
674 /*
675 * Wait for a specified commit to complete.
676 * The caller may not hold the journal lock.
677 */
jbd2_log_wait_commit(journal_t * journal,tid_t tid)678 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
679 {
680 int err = 0;
681
682 read_lock(&journal->j_state_lock);
683 #ifdef CONFIG_PROVE_LOCKING
684 /*
685 * Some callers make sure transaction is already committing and in that
686 * case we cannot block on open handles anymore. So don't warn in that
687 * case.
688 */
689 if (tid_gt(tid, journal->j_commit_sequence) &&
690 (!journal->j_committing_transaction ||
691 journal->j_committing_transaction->t_tid != tid)) {
692 read_unlock(&journal->j_state_lock);
693 jbd2_might_wait_for_commit(journal);
694 read_lock(&journal->j_state_lock);
695 }
696 #endif
697 #ifdef CONFIG_JBD2_DEBUG
698 if (!tid_geq(journal->j_commit_request, tid)) {
699 printk(KERN_ERR
700 "%s: error: j_commit_request=%u, tid=%u\n",
701 __func__, journal->j_commit_request, tid);
702 }
703 #endif
704 while (tid_gt(tid, journal->j_commit_sequence)) {
705 jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
706 tid, journal->j_commit_sequence);
707 read_unlock(&journal->j_state_lock);
708 wake_up(&journal->j_wait_commit);
709 wait_event(journal->j_wait_done_commit,
710 !tid_gt(tid, journal->j_commit_sequence));
711 read_lock(&journal->j_state_lock);
712 }
713 read_unlock(&journal->j_state_lock);
714
715 if (unlikely(is_journal_aborted(journal)))
716 err = -EIO;
717 return err;
718 }
719
720 /*
721 * Start a fast commit. If there's an ongoing fast or full commit wait for
722 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
723 * if a fast commit is not needed, either because there's an already a commit
724 * going on or this tid has already been committed. Returns -EINVAL if no jbd2
725 * commit has yet been performed.
726 */
jbd2_fc_begin_commit(journal_t * journal,tid_t tid)727 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
728 {
729 if (unlikely(is_journal_aborted(journal)))
730 return -EIO;
731 /*
732 * Fast commits only allowed if at least one full commit has
733 * been processed.
734 */
735 if (!journal->j_stats.ts_tid)
736 return -EINVAL;
737
738 write_lock(&journal->j_state_lock);
739 if (tid <= journal->j_commit_sequence) {
740 write_unlock(&journal->j_state_lock);
741 return -EALREADY;
742 }
743
744 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
745 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
746 DEFINE_WAIT(wait);
747
748 prepare_to_wait(&journal->j_fc_wait, &wait,
749 TASK_UNINTERRUPTIBLE);
750 write_unlock(&journal->j_state_lock);
751 schedule();
752 finish_wait(&journal->j_fc_wait, &wait);
753 return -EALREADY;
754 }
755 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
756 write_unlock(&journal->j_state_lock);
757 jbd2_journal_lock_updates(journal);
758
759 return 0;
760 }
761 EXPORT_SYMBOL(jbd2_fc_begin_commit);
762
763 /*
764 * Stop a fast commit. If fallback is set, this function starts commit of
765 * TID tid before any other fast commit can start.
766 */
__jbd2_fc_end_commit(journal_t * journal,tid_t tid,bool fallback)767 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
768 {
769 jbd2_journal_unlock_updates(journal);
770 if (journal->j_fc_cleanup_callback)
771 journal->j_fc_cleanup_callback(journal, 0, tid);
772 write_lock(&journal->j_state_lock);
773 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
774 if (fallback)
775 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
776 write_unlock(&journal->j_state_lock);
777 wake_up(&journal->j_fc_wait);
778 if (fallback)
779 return jbd2_complete_transaction(journal, tid);
780 return 0;
781 }
782
jbd2_fc_end_commit(journal_t * journal)783 int jbd2_fc_end_commit(journal_t *journal)
784 {
785 return __jbd2_fc_end_commit(journal, 0, false);
786 }
787 EXPORT_SYMBOL(jbd2_fc_end_commit);
788
jbd2_fc_end_commit_fallback(journal_t * journal)789 int jbd2_fc_end_commit_fallback(journal_t *journal)
790 {
791 tid_t tid;
792
793 read_lock(&journal->j_state_lock);
794 tid = journal->j_running_transaction ?
795 journal->j_running_transaction->t_tid : 0;
796 read_unlock(&journal->j_state_lock);
797 return __jbd2_fc_end_commit(journal, tid, true);
798 }
799 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
800
801 /* Return 1 when transaction with given tid has already committed. */
jbd2_transaction_committed(journal_t * journal,tid_t tid)802 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
803 {
804 int ret = 1;
805
806 read_lock(&journal->j_state_lock);
807 if (journal->j_running_transaction &&
808 journal->j_running_transaction->t_tid == tid)
809 ret = 0;
810 if (journal->j_committing_transaction &&
811 journal->j_committing_transaction->t_tid == tid)
812 ret = 0;
813 read_unlock(&journal->j_state_lock);
814 return ret;
815 }
816 EXPORT_SYMBOL(jbd2_transaction_committed);
817
818 /*
819 * When this function returns the transaction corresponding to tid
820 * will be completed. If the transaction has currently running, start
821 * committing that transaction before waiting for it to complete. If
822 * the transaction id is stale, it is by definition already completed,
823 * so just return SUCCESS.
824 */
jbd2_complete_transaction(journal_t * journal,tid_t tid)825 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
826 {
827 int need_to_wait = 1;
828
829 read_lock(&journal->j_state_lock);
830 if (journal->j_running_transaction &&
831 journal->j_running_transaction->t_tid == tid) {
832 if (journal->j_commit_request != tid) {
833 /* transaction not yet started, so request it */
834 read_unlock(&journal->j_state_lock);
835 jbd2_log_start_commit(journal, tid);
836 goto wait_commit;
837 }
838 } else if (!(journal->j_committing_transaction &&
839 journal->j_committing_transaction->t_tid == tid))
840 need_to_wait = 0;
841 read_unlock(&journal->j_state_lock);
842 if (!need_to_wait)
843 return 0;
844 wait_commit:
845 return jbd2_log_wait_commit(journal, tid);
846 }
847 EXPORT_SYMBOL(jbd2_complete_transaction);
848
849 /*
850 * Log buffer allocation routines:
851 */
852
jbd2_journal_next_log_block(journal_t * journal,unsigned long long * retp)853 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
854 {
855 unsigned long blocknr;
856
857 write_lock(&journal->j_state_lock);
858 J_ASSERT(journal->j_free > 1);
859
860 blocknr = journal->j_head;
861 journal->j_head++;
862 journal->j_free--;
863 if (journal->j_head == journal->j_last)
864 journal->j_head = journal->j_first;
865 write_unlock(&journal->j_state_lock);
866 return jbd2_journal_bmap(journal, blocknr, retp);
867 }
868
869 /* Map one fast commit buffer for use by the file system */
jbd2_fc_get_buf(journal_t * journal,struct buffer_head ** bh_out)870 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
871 {
872 unsigned long long pblock;
873 unsigned long blocknr;
874 int ret = 0;
875 struct buffer_head *bh;
876 int fc_off;
877
878 *bh_out = NULL;
879
880 if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
881 fc_off = journal->j_fc_off;
882 blocknr = journal->j_fc_first + fc_off;
883 journal->j_fc_off++;
884 } else {
885 ret = -EINVAL;
886 }
887
888 if (ret)
889 return ret;
890
891 ret = jbd2_journal_bmap(journal, blocknr, &pblock);
892 if (ret)
893 return ret;
894
895 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
896 if (!bh)
897 return -ENOMEM;
898
899
900 journal->j_fc_wbuf[fc_off] = bh;
901
902 *bh_out = bh;
903
904 return 0;
905 }
906 EXPORT_SYMBOL(jbd2_fc_get_buf);
907
908 /*
909 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
910 * for completion.
911 */
jbd2_fc_wait_bufs(journal_t * journal,int num_blks)912 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
913 {
914 struct buffer_head *bh;
915 int i, j_fc_off;
916
917 j_fc_off = journal->j_fc_off;
918
919 /*
920 * Wait in reverse order to minimize chances of us being woken up before
921 * all IOs have completed
922 */
923 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
924 bh = journal->j_fc_wbuf[i];
925 wait_on_buffer(bh);
926 /*
927 * Update j_fc_off so jbd2_fc_release_bufs can release remain
928 * buffer head.
929 */
930 if (unlikely(!buffer_uptodate(bh))) {
931 journal->j_fc_off = i + 1;
932 return -EIO;
933 }
934 put_bh(bh);
935 journal->j_fc_wbuf[i] = NULL;
936 }
937
938 return 0;
939 }
940 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
941
jbd2_fc_release_bufs(journal_t * journal)942 int jbd2_fc_release_bufs(journal_t *journal)
943 {
944 struct buffer_head *bh;
945 int i, j_fc_off;
946
947 j_fc_off = journal->j_fc_off;
948
949 for (i = j_fc_off - 1; i >= 0; i--) {
950 bh = journal->j_fc_wbuf[i];
951 if (!bh)
952 break;
953 put_bh(bh);
954 journal->j_fc_wbuf[i] = NULL;
955 }
956
957 return 0;
958 }
959 EXPORT_SYMBOL(jbd2_fc_release_bufs);
960
961 /*
962 * Conversion of logical to physical block numbers for the journal
963 *
964 * On external journals the journal blocks are identity-mapped, so
965 * this is a no-op. If needed, we can use j_blk_offset - everything is
966 * ready.
967 */
jbd2_journal_bmap(journal_t * journal,unsigned long blocknr,unsigned long long * retp)968 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
969 unsigned long long *retp)
970 {
971 int err = 0;
972 unsigned long long ret;
973 sector_t block = 0;
974
975 if (journal->j_inode) {
976 block = blocknr;
977 ret = bmap(journal->j_inode, &block);
978
979 if (ret || !block) {
980 printk(KERN_ALERT "%s: journal block not found "
981 "at offset %lu on %s\n",
982 __func__, blocknr, journal->j_devname);
983 err = -EIO;
984 jbd2_journal_abort(journal, err);
985 } else {
986 *retp = block;
987 }
988
989 } else {
990 *retp = blocknr; /* +journal->j_blk_offset */
991 }
992 return err;
993 }
994
995 /*
996 * We play buffer_head aliasing tricks to write data/metadata blocks to
997 * the journal without copying their contents, but for journal
998 * descriptor blocks we do need to generate bona fide buffers.
999 *
1000 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
1001 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
1002 * But we don't bother doing that, so there will be coherency problems with
1003 * mmaps of blockdevs which hold live JBD-controlled filesystems.
1004 */
1005 struct buffer_head *
jbd2_journal_get_descriptor_buffer(transaction_t * transaction,int type)1006 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1007 {
1008 journal_t *journal = transaction->t_journal;
1009 struct buffer_head *bh;
1010 unsigned long long blocknr;
1011 journal_header_t *header;
1012 int err;
1013
1014 err = jbd2_journal_next_log_block(journal, &blocknr);
1015
1016 if (err)
1017 return NULL;
1018
1019 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1020 if (!bh)
1021 return NULL;
1022 atomic_dec(&transaction->t_outstanding_credits);
1023 lock_buffer(bh);
1024 memset(bh->b_data, 0, journal->j_blocksize);
1025 header = (journal_header_t *)bh->b_data;
1026 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1027 header->h_blocktype = cpu_to_be32(type);
1028 header->h_sequence = cpu_to_be32(transaction->t_tid);
1029 set_buffer_uptodate(bh);
1030 unlock_buffer(bh);
1031 BUFFER_TRACE(bh, "return this buffer");
1032 return bh;
1033 }
1034
jbd2_descriptor_block_csum_set(journal_t * j,struct buffer_head * bh)1035 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1036 {
1037 struct jbd2_journal_block_tail *tail;
1038 __u32 csum;
1039
1040 if (!jbd2_journal_has_csum_v2or3(j))
1041 return;
1042
1043 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1044 sizeof(struct jbd2_journal_block_tail));
1045 tail->t_checksum = 0;
1046 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1047 tail->t_checksum = cpu_to_be32(csum);
1048 }
1049
1050 /*
1051 * Return tid of the oldest transaction in the journal and block in the journal
1052 * where the transaction starts.
1053 *
1054 * If the journal is now empty, return which will be the next transaction ID
1055 * we will write and where will that transaction start.
1056 *
1057 * The return value is 0 if journal tail cannot be pushed any further, 1 if
1058 * it can.
1059 */
jbd2_journal_get_log_tail(journal_t * journal,tid_t * tid,unsigned long * block)1060 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1061 unsigned long *block)
1062 {
1063 transaction_t *transaction;
1064 int ret;
1065
1066 read_lock(&journal->j_state_lock);
1067 spin_lock(&journal->j_list_lock);
1068 transaction = journal->j_checkpoint_transactions;
1069 if (transaction) {
1070 *tid = transaction->t_tid;
1071 *block = transaction->t_log_start;
1072 } else if ((transaction = journal->j_committing_transaction) != NULL) {
1073 *tid = transaction->t_tid;
1074 *block = transaction->t_log_start;
1075 } else if ((transaction = journal->j_running_transaction) != NULL) {
1076 *tid = transaction->t_tid;
1077 *block = journal->j_head;
1078 } else {
1079 *tid = journal->j_transaction_sequence;
1080 *block = journal->j_head;
1081 }
1082 ret = tid_gt(*tid, journal->j_tail_sequence);
1083 spin_unlock(&journal->j_list_lock);
1084 read_unlock(&journal->j_state_lock);
1085
1086 return ret;
1087 }
1088
1089 /*
1090 * Update information in journal structure and in on disk journal superblock
1091 * about log tail. This function does not check whether information passed in
1092 * really pushes log tail further. It's responsibility of the caller to make
1093 * sure provided log tail information is valid (e.g. by holding
1094 * j_checkpoint_mutex all the time between computing log tail and calling this
1095 * function as is the case with jbd2_cleanup_journal_tail()).
1096 *
1097 * Requires j_checkpoint_mutex
1098 */
__jbd2_update_log_tail(journal_t * journal,tid_t tid,unsigned long block)1099 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1100 {
1101 unsigned long freed;
1102 int ret;
1103
1104 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1105
1106 /*
1107 * We cannot afford for write to remain in drive's caches since as
1108 * soon as we update j_tail, next transaction can start reusing journal
1109 * space and if we lose sb update during power failure we'd replay
1110 * old transaction with possibly newly overwritten data.
1111 */
1112 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, REQ_FUA);
1113 if (ret)
1114 goto out;
1115
1116 write_lock(&journal->j_state_lock);
1117 freed = block - journal->j_tail;
1118 if (block < journal->j_tail)
1119 freed += journal->j_last - journal->j_first;
1120
1121 trace_jbd2_update_log_tail(journal, tid, block, freed);
1122 jbd2_debug(1,
1123 "Cleaning journal tail from %u to %u (offset %lu), "
1124 "freeing %lu\n",
1125 journal->j_tail_sequence, tid, block, freed);
1126
1127 journal->j_free += freed;
1128 journal->j_tail_sequence = tid;
1129 journal->j_tail = block;
1130 write_unlock(&journal->j_state_lock);
1131
1132 out:
1133 return ret;
1134 }
1135
1136 /*
1137 * This is a variation of __jbd2_update_log_tail which checks for validity of
1138 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1139 * with other threads updating log tail.
1140 */
jbd2_update_log_tail(journal_t * journal,tid_t tid,unsigned long block)1141 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1142 {
1143 mutex_lock_io(&journal->j_checkpoint_mutex);
1144 if (tid_gt(tid, journal->j_tail_sequence))
1145 __jbd2_update_log_tail(journal, tid, block);
1146 mutex_unlock(&journal->j_checkpoint_mutex);
1147 }
1148
1149 struct jbd2_stats_proc_session {
1150 journal_t *journal;
1151 struct transaction_stats_s *stats;
1152 int start;
1153 int max;
1154 };
1155
jbd2_seq_info_start(struct seq_file * seq,loff_t * pos)1156 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1157 {
1158 return *pos ? NULL : SEQ_START_TOKEN;
1159 }
1160
jbd2_seq_info_next(struct seq_file * seq,void * v,loff_t * pos)1161 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1162 {
1163 (*pos)++;
1164 return NULL;
1165 }
1166
jbd2_seq_info_show(struct seq_file * seq,void * v)1167 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1168 {
1169 struct jbd2_stats_proc_session *s = seq->private;
1170
1171 if (v != SEQ_START_TOKEN)
1172 return 0;
1173 seq_printf(seq, "%lu transactions (%lu requested), "
1174 "each up to %u blocks\n",
1175 s->stats->ts_tid, s->stats->ts_requested,
1176 s->journal->j_max_transaction_buffers);
1177 if (s->stats->ts_tid == 0)
1178 return 0;
1179 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1180 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1181 seq_printf(seq, " %ums request delay\n",
1182 (s->stats->ts_requested == 0) ? 0 :
1183 jiffies_to_msecs(s->stats->run.rs_request_delay /
1184 s->stats->ts_requested));
1185 seq_printf(seq, " %ums running transaction\n",
1186 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1187 seq_printf(seq, " %ums transaction was being locked\n",
1188 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1189 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1190 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1191 seq_printf(seq, " %ums logging transaction\n",
1192 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1193 seq_printf(seq, " %lluus average transaction commit time\n",
1194 div_u64(s->journal->j_average_commit_time, 1000));
1195 seq_printf(seq, " %lu handles per transaction\n",
1196 s->stats->run.rs_handle_count / s->stats->ts_tid);
1197 seq_printf(seq, " %lu blocks per transaction\n",
1198 s->stats->run.rs_blocks / s->stats->ts_tid);
1199 seq_printf(seq, " %lu logged blocks per transaction\n",
1200 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1201 return 0;
1202 }
1203
jbd2_seq_info_stop(struct seq_file * seq,void * v)1204 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1205 {
1206 }
1207
1208 static const struct seq_operations jbd2_seq_info_ops = {
1209 .start = jbd2_seq_info_start,
1210 .next = jbd2_seq_info_next,
1211 .stop = jbd2_seq_info_stop,
1212 .show = jbd2_seq_info_show,
1213 };
1214
jbd2_seq_info_open(struct inode * inode,struct file * file)1215 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1216 {
1217 journal_t *journal = pde_data(inode);
1218 struct jbd2_stats_proc_session *s;
1219 int rc, size;
1220
1221 s = kmalloc(sizeof(*s), GFP_KERNEL);
1222 if (s == NULL)
1223 return -ENOMEM;
1224 size = sizeof(struct transaction_stats_s);
1225 s->stats = kmalloc(size, GFP_KERNEL);
1226 if (s->stats == NULL) {
1227 kfree(s);
1228 return -ENOMEM;
1229 }
1230 spin_lock(&journal->j_history_lock);
1231 memcpy(s->stats, &journal->j_stats, size);
1232 s->journal = journal;
1233 spin_unlock(&journal->j_history_lock);
1234
1235 rc = seq_open(file, &jbd2_seq_info_ops);
1236 if (rc == 0) {
1237 struct seq_file *m = file->private_data;
1238 m->private = s;
1239 } else {
1240 kfree(s->stats);
1241 kfree(s);
1242 }
1243 return rc;
1244
1245 }
1246
jbd2_seq_info_release(struct inode * inode,struct file * file)1247 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1248 {
1249 struct seq_file *seq = file->private_data;
1250 struct jbd2_stats_proc_session *s = seq->private;
1251 kfree(s->stats);
1252 kfree(s);
1253 return seq_release(inode, file);
1254 }
1255
1256 static const struct proc_ops jbd2_info_proc_ops = {
1257 .proc_open = jbd2_seq_info_open,
1258 .proc_read = seq_read,
1259 .proc_lseek = seq_lseek,
1260 .proc_release = jbd2_seq_info_release,
1261 };
1262
1263 static struct proc_dir_entry *proc_jbd2_stats;
1264
jbd2_stats_proc_init(journal_t * journal)1265 static void jbd2_stats_proc_init(journal_t *journal)
1266 {
1267 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1268 if (journal->j_proc_entry) {
1269 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1270 &jbd2_info_proc_ops, journal);
1271 }
1272 }
1273
jbd2_stats_proc_exit(journal_t * journal)1274 static void jbd2_stats_proc_exit(journal_t *journal)
1275 {
1276 remove_proc_entry("info", journal->j_proc_entry);
1277 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1278 }
1279
1280 /* Minimum size of descriptor tag */
jbd2_min_tag_size(void)1281 static int jbd2_min_tag_size(void)
1282 {
1283 /*
1284 * Tag with 32-bit block numbers does not use last four bytes of the
1285 * structure
1286 */
1287 return sizeof(journal_block_tag_t) - 4;
1288 }
1289
1290 /**
1291 * jbd2_journal_shrink_scan()
1292 * @shrink: shrinker to work on
1293 * @sc: reclaim request to process
1294 *
1295 * Scan the checkpointed buffer on the checkpoint list and release the
1296 * journal_head.
1297 */
jbd2_journal_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)1298 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1299 struct shrink_control *sc)
1300 {
1301 journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1302 unsigned long nr_to_scan = sc->nr_to_scan;
1303 unsigned long nr_shrunk;
1304 unsigned long count;
1305
1306 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1307 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1308
1309 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1310
1311 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1312 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1313
1314 return nr_shrunk;
1315 }
1316
1317 /**
1318 * jbd2_journal_shrink_count()
1319 * @shrink: shrinker to work on
1320 * @sc: reclaim request to process
1321 *
1322 * Count the number of checkpoint buffers on the checkpoint list.
1323 */
jbd2_journal_shrink_count(struct shrinker * shrink,struct shrink_control * sc)1324 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1325 struct shrink_control *sc)
1326 {
1327 journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1328 unsigned long count;
1329
1330 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1331 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1332
1333 return count;
1334 }
1335
1336 /*
1337 * Management for journal control blocks: functions to create and
1338 * destroy journal_t structures, and to initialise and read existing
1339 * journal blocks from disk. */
1340
1341 /* First: create and setup a journal_t object in memory. We initialise
1342 * very few fields yet: that has to wait until we have created the
1343 * journal structures from from scratch, or loaded them from disk. */
1344
journal_init_common(struct block_device * bdev,struct block_device * fs_dev,unsigned long long start,int len,int blocksize)1345 static journal_t *journal_init_common(struct block_device *bdev,
1346 struct block_device *fs_dev,
1347 unsigned long long start, int len, int blocksize)
1348 {
1349 static struct lock_class_key jbd2_trans_commit_key;
1350 journal_t *journal;
1351 int err;
1352 struct buffer_head *bh;
1353 int n;
1354
1355 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1356 if (!journal)
1357 return NULL;
1358
1359 init_waitqueue_head(&journal->j_wait_transaction_locked);
1360 init_waitqueue_head(&journal->j_wait_done_commit);
1361 init_waitqueue_head(&journal->j_wait_commit);
1362 init_waitqueue_head(&journal->j_wait_updates);
1363 init_waitqueue_head(&journal->j_wait_reserved);
1364 init_waitqueue_head(&journal->j_fc_wait);
1365 mutex_init(&journal->j_abort_mutex);
1366 mutex_init(&journal->j_barrier);
1367 mutex_init(&journal->j_checkpoint_mutex);
1368 spin_lock_init(&journal->j_revoke_lock);
1369 spin_lock_init(&journal->j_list_lock);
1370 rwlock_init(&journal->j_state_lock);
1371
1372 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1373 journal->j_min_batch_time = 0;
1374 journal->j_max_batch_time = 15000; /* 15ms */
1375 atomic_set(&journal->j_reserved_credits, 0);
1376
1377 /* The journal is marked for error until we succeed with recovery! */
1378 journal->j_flags = JBD2_ABORT;
1379
1380 /* Set up a default-sized revoke table for the new mount. */
1381 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1382 if (err)
1383 goto err_cleanup;
1384
1385 spin_lock_init(&journal->j_history_lock);
1386
1387 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1388 &jbd2_trans_commit_key, 0);
1389
1390 /* journal descriptor can store up to n blocks -bzzz */
1391 journal->j_blocksize = blocksize;
1392 journal->j_dev = bdev;
1393 journal->j_fs_dev = fs_dev;
1394 journal->j_blk_offset = start;
1395 journal->j_total_len = len;
1396 /* We need enough buffers to write out full descriptor block. */
1397 n = journal->j_blocksize / jbd2_min_tag_size();
1398 journal->j_wbufsize = n;
1399 journal->j_fc_wbuf = NULL;
1400 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1401 GFP_KERNEL);
1402 if (!journal->j_wbuf)
1403 goto err_cleanup;
1404
1405 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1406 if (!bh) {
1407 pr_err("%s: Cannot get buffer for journal superblock\n",
1408 __func__);
1409 goto err_cleanup;
1410 }
1411 journal->j_sb_buffer = bh;
1412 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1413
1414 journal->j_shrink_transaction = NULL;
1415 journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
1416 journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
1417 journal->j_shrinker.seeks = DEFAULT_SEEKS;
1418 journal->j_shrinker.batch = journal->j_max_transaction_buffers;
1419
1420 if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL))
1421 goto err_cleanup;
1422
1423 if (register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)",
1424 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev))) {
1425 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1426 goto err_cleanup;
1427 }
1428 return journal;
1429
1430 err_cleanup:
1431 brelse(journal->j_sb_buffer);
1432 kfree(journal->j_wbuf);
1433 jbd2_journal_destroy_revoke(journal);
1434 kfree(journal);
1435 return NULL;
1436 }
1437
1438 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1439 *
1440 * Create a journal structure assigned some fixed set of disk blocks to
1441 * the journal. We don't actually touch those disk blocks yet, but we
1442 * need to set up all of the mapping information to tell the journaling
1443 * system where the journal blocks are.
1444 *
1445 */
1446
1447 /**
1448 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1449 * @bdev: Block device on which to create the journal
1450 * @fs_dev: Device which hold journalled filesystem for this journal.
1451 * @start: Block nr Start of journal.
1452 * @len: Length of the journal in blocks.
1453 * @blocksize: blocksize of journalling device
1454 *
1455 * Returns: a newly created journal_t *
1456 *
1457 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1458 * range of blocks on an arbitrary block device.
1459 *
1460 */
jbd2_journal_init_dev(struct block_device * bdev,struct block_device * fs_dev,unsigned long long start,int len,int blocksize)1461 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1462 struct block_device *fs_dev,
1463 unsigned long long start, int len, int blocksize)
1464 {
1465 journal_t *journal;
1466
1467 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1468 if (!journal)
1469 return NULL;
1470
1471 snprintf(journal->j_devname, sizeof(journal->j_devname),
1472 "%pg", journal->j_dev);
1473 strreplace(journal->j_devname, '/', '!');
1474 jbd2_stats_proc_init(journal);
1475
1476 return journal;
1477 }
1478
1479 /**
1480 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1481 * @inode: An inode to create the journal in
1482 *
1483 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1484 * the journal. The inode must exist already, must support bmap() and
1485 * must have all data blocks preallocated.
1486 */
jbd2_journal_init_inode(struct inode * inode)1487 journal_t *jbd2_journal_init_inode(struct inode *inode)
1488 {
1489 journal_t *journal;
1490 sector_t blocknr;
1491 char *p;
1492 int err = 0;
1493
1494 blocknr = 0;
1495 err = bmap(inode, &blocknr);
1496
1497 if (err || !blocknr) {
1498 pr_err("%s: Cannot locate journal superblock\n",
1499 __func__);
1500 return NULL;
1501 }
1502
1503 jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1504 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1505 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1506
1507 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1508 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1509 inode->i_sb->s_blocksize);
1510 if (!journal)
1511 return NULL;
1512
1513 journal->j_inode = inode;
1514 snprintf(journal->j_devname, sizeof(journal->j_devname),
1515 "%pg", journal->j_dev);
1516 p = strreplace(journal->j_devname, '/', '!');
1517 sprintf(p, "-%lu", journal->j_inode->i_ino);
1518 jbd2_stats_proc_init(journal);
1519
1520 return journal;
1521 }
1522
1523 /*
1524 * If the journal init or create aborts, we need to mark the journal
1525 * superblock as being NULL to prevent the journal destroy from writing
1526 * back a bogus superblock.
1527 */
journal_fail_superblock(journal_t * journal)1528 static void journal_fail_superblock(journal_t *journal)
1529 {
1530 struct buffer_head *bh = journal->j_sb_buffer;
1531 brelse(bh);
1532 journal->j_sb_buffer = NULL;
1533 }
1534
1535 /*
1536 * Given a journal_t structure, initialise the various fields for
1537 * startup of a new journaling session. We use this both when creating
1538 * a journal, and after recovering an old journal to reset it for
1539 * subsequent use.
1540 */
1541
journal_reset(journal_t * journal)1542 static int journal_reset(journal_t *journal)
1543 {
1544 journal_superblock_t *sb = journal->j_superblock;
1545 unsigned long long first, last;
1546
1547 first = be32_to_cpu(sb->s_first);
1548 last = be32_to_cpu(sb->s_maxlen);
1549 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1550 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1551 first, last);
1552 journal_fail_superblock(journal);
1553 return -EINVAL;
1554 }
1555
1556 journal->j_first = first;
1557 journal->j_last = last;
1558
1559 journal->j_head = journal->j_first;
1560 journal->j_tail = journal->j_first;
1561 journal->j_free = journal->j_last - journal->j_first;
1562
1563 journal->j_tail_sequence = journal->j_transaction_sequence;
1564 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1565 journal->j_commit_request = journal->j_commit_sequence;
1566
1567 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1568
1569 /*
1570 * Now that journal recovery is done, turn fast commits off here. This
1571 * way, if fast commit was enabled before the crash but if now FS has
1572 * disabled it, we don't enable fast commits.
1573 */
1574 jbd2_clear_feature_fast_commit(journal);
1575
1576 /*
1577 * As a special case, if the on-disk copy is already marked as needing
1578 * no recovery (s_start == 0), then we can safely defer the superblock
1579 * update until the next commit by setting JBD2_FLUSHED. This avoids
1580 * attempting a write to a potential-readonly device.
1581 */
1582 if (sb->s_start == 0) {
1583 jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1584 "(start %ld, seq %u, errno %d)\n",
1585 journal->j_tail, journal->j_tail_sequence,
1586 journal->j_errno);
1587 journal->j_flags |= JBD2_FLUSHED;
1588 } else {
1589 /* Lock here to make assertions happy... */
1590 mutex_lock_io(&journal->j_checkpoint_mutex);
1591 /*
1592 * Update log tail information. We use REQ_FUA since new
1593 * transaction will start reusing journal space and so we
1594 * must make sure information about current log tail is on
1595 * disk before that.
1596 */
1597 jbd2_journal_update_sb_log_tail(journal,
1598 journal->j_tail_sequence,
1599 journal->j_tail, REQ_FUA);
1600 mutex_unlock(&journal->j_checkpoint_mutex);
1601 }
1602 return jbd2_journal_start_thread(journal);
1603 }
1604
1605 /*
1606 * This function expects that the caller will have locked the journal
1607 * buffer head, and will return with it unlocked
1608 */
jbd2_write_superblock(journal_t * journal,blk_opf_t write_flags)1609 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags)
1610 {
1611 struct buffer_head *bh = journal->j_sb_buffer;
1612 journal_superblock_t *sb = journal->j_superblock;
1613 int ret = 0;
1614
1615 /* Buffer got discarded which means block device got invalidated */
1616 if (!buffer_mapped(bh)) {
1617 unlock_buffer(bh);
1618 return -EIO;
1619 }
1620
1621 /*
1622 * Always set high priority flags to exempt from block layer's
1623 * QOS policies, e.g. writeback throttle.
1624 */
1625 write_flags |= JBD2_JOURNAL_REQ_FLAGS;
1626 if (!(journal->j_flags & JBD2_BARRIER))
1627 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1628
1629 trace_jbd2_write_superblock(journal, write_flags);
1630
1631 if (buffer_write_io_error(bh)) {
1632 /*
1633 * Oh, dear. A previous attempt to write the journal
1634 * superblock failed. This could happen because the
1635 * USB device was yanked out. Or it could happen to
1636 * be a transient write error and maybe the block will
1637 * be remapped. Nothing we can do but to retry the
1638 * write and hope for the best.
1639 */
1640 printk(KERN_ERR "JBD2: previous I/O error detected "
1641 "for journal superblock update for %s.\n",
1642 journal->j_devname);
1643 clear_buffer_write_io_error(bh);
1644 set_buffer_uptodate(bh);
1645 }
1646 if (jbd2_journal_has_csum_v2or3(journal))
1647 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1648 get_bh(bh);
1649 bh->b_end_io = end_buffer_write_sync;
1650 submit_bh(REQ_OP_WRITE | write_flags, bh);
1651 wait_on_buffer(bh);
1652 if (buffer_write_io_error(bh)) {
1653 clear_buffer_write_io_error(bh);
1654 set_buffer_uptodate(bh);
1655 ret = -EIO;
1656 }
1657 if (ret) {
1658 printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n",
1659 journal->j_devname);
1660 if (!is_journal_aborted(journal))
1661 jbd2_journal_abort(journal, ret);
1662 }
1663
1664 return ret;
1665 }
1666
1667 /**
1668 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1669 * @journal: The journal to update.
1670 * @tail_tid: TID of the new transaction at the tail of the log
1671 * @tail_block: The first block of the transaction at the tail of the log
1672 * @write_flags: Flags for the journal sb write operation
1673 *
1674 * Update a journal's superblock information about log tail and write it to
1675 * disk, waiting for the IO to complete.
1676 */
jbd2_journal_update_sb_log_tail(journal_t * journal,tid_t tail_tid,unsigned long tail_block,blk_opf_t write_flags)1677 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1678 unsigned long tail_block,
1679 blk_opf_t write_flags)
1680 {
1681 journal_superblock_t *sb = journal->j_superblock;
1682 int ret;
1683
1684 if (is_journal_aborted(journal))
1685 return -EIO;
1686 if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) {
1687 jbd2_journal_abort(journal, -EIO);
1688 return -EIO;
1689 }
1690
1691 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1692 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1693 tail_block, tail_tid);
1694
1695 lock_buffer(journal->j_sb_buffer);
1696 sb->s_sequence = cpu_to_be32(tail_tid);
1697 sb->s_start = cpu_to_be32(tail_block);
1698
1699 ret = jbd2_write_superblock(journal, write_flags);
1700 if (ret)
1701 goto out;
1702
1703 /* Log is no longer empty */
1704 write_lock(&journal->j_state_lock);
1705 WARN_ON(!sb->s_sequence);
1706 journal->j_flags &= ~JBD2_FLUSHED;
1707 write_unlock(&journal->j_state_lock);
1708
1709 out:
1710 return ret;
1711 }
1712
1713 /**
1714 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1715 * @journal: The journal to update.
1716 * @write_flags: Flags for the journal sb write operation
1717 *
1718 * Update a journal's dynamic superblock fields to show that journal is empty.
1719 * Write updated superblock to disk waiting for IO to complete.
1720 */
jbd2_mark_journal_empty(journal_t * journal,blk_opf_t write_flags)1721 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags)
1722 {
1723 journal_superblock_t *sb = journal->j_superblock;
1724 bool had_fast_commit = false;
1725
1726 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1727 lock_buffer(journal->j_sb_buffer);
1728 if (sb->s_start == 0) { /* Is it already empty? */
1729 unlock_buffer(journal->j_sb_buffer);
1730 return;
1731 }
1732
1733 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1734 journal->j_tail_sequence);
1735
1736 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1737 sb->s_start = cpu_to_be32(0);
1738 if (jbd2_has_feature_fast_commit(journal)) {
1739 /*
1740 * When journal is clean, no need to commit fast commit flag and
1741 * make file system incompatible with older kernels.
1742 */
1743 jbd2_clear_feature_fast_commit(journal);
1744 had_fast_commit = true;
1745 }
1746
1747 jbd2_write_superblock(journal, write_flags);
1748
1749 if (had_fast_commit)
1750 jbd2_set_feature_fast_commit(journal);
1751
1752 /* Log is no longer empty */
1753 write_lock(&journal->j_state_lock);
1754 journal->j_flags |= JBD2_FLUSHED;
1755 write_unlock(&journal->j_state_lock);
1756 }
1757
1758 /**
1759 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1760 * @journal: The journal to erase.
1761 * @flags: A discard/zeroout request is sent for each physically contigous
1762 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1763 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1764 * to perform.
1765 *
1766 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1767 * will be explicitly written if no hardware offload is available, see
1768 * blkdev_issue_zeroout for more details.
1769 */
__jbd2_journal_erase(journal_t * journal,unsigned int flags)1770 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1771 {
1772 int err = 0;
1773 unsigned long block, log_offset; /* logical */
1774 unsigned long long phys_block, block_start, block_stop; /* physical */
1775 loff_t byte_start, byte_stop, byte_count;
1776
1777 /* flags must be set to either discard or zeroout */
1778 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1779 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1780 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1781 return -EINVAL;
1782
1783 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1784 !bdev_max_discard_sectors(journal->j_dev))
1785 return -EOPNOTSUPP;
1786
1787 /*
1788 * lookup block mapping and issue discard/zeroout for each
1789 * contiguous region
1790 */
1791 log_offset = be32_to_cpu(journal->j_superblock->s_first);
1792 block_start = ~0ULL;
1793 for (block = log_offset; block < journal->j_total_len; block++) {
1794 err = jbd2_journal_bmap(journal, block, &phys_block);
1795 if (err) {
1796 pr_err("JBD2: bad block at offset %lu", block);
1797 return err;
1798 }
1799
1800 if (block_start == ~0ULL) {
1801 block_start = phys_block;
1802 block_stop = block_start - 1;
1803 }
1804
1805 /*
1806 * last block not contiguous with current block,
1807 * process last contiguous region and return to this block on
1808 * next loop
1809 */
1810 if (phys_block != block_stop + 1) {
1811 block--;
1812 } else {
1813 block_stop++;
1814 /*
1815 * if this isn't the last block of journal,
1816 * no need to process now because next block may also
1817 * be part of this contiguous region
1818 */
1819 if (block != journal->j_total_len - 1)
1820 continue;
1821 }
1822
1823 /*
1824 * end of contiguous region or this is last block of journal,
1825 * take care of the region
1826 */
1827 byte_start = block_start * journal->j_blocksize;
1828 byte_stop = block_stop * journal->j_blocksize;
1829 byte_count = (block_stop - block_start + 1) *
1830 journal->j_blocksize;
1831
1832 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
1833 byte_start, byte_stop);
1834
1835 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
1836 err = blkdev_issue_discard(journal->j_dev,
1837 byte_start >> SECTOR_SHIFT,
1838 byte_count >> SECTOR_SHIFT,
1839 GFP_NOFS);
1840 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
1841 err = blkdev_issue_zeroout(journal->j_dev,
1842 byte_start >> SECTOR_SHIFT,
1843 byte_count >> SECTOR_SHIFT,
1844 GFP_NOFS, 0);
1845 }
1846
1847 if (unlikely(err != 0)) {
1848 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
1849 err, block_start, block_stop);
1850 return err;
1851 }
1852
1853 /* reset start and stop after processing a region */
1854 block_start = ~0ULL;
1855 }
1856
1857 return blkdev_issue_flush(journal->j_dev);
1858 }
1859
1860 /**
1861 * jbd2_journal_update_sb_errno() - Update error in the journal.
1862 * @journal: The journal to update.
1863 *
1864 * Update a journal's errno. Write updated superblock to disk waiting for IO
1865 * to complete.
1866 */
jbd2_journal_update_sb_errno(journal_t * journal)1867 void jbd2_journal_update_sb_errno(journal_t *journal)
1868 {
1869 journal_superblock_t *sb = journal->j_superblock;
1870 int errcode;
1871
1872 lock_buffer(journal->j_sb_buffer);
1873 errcode = journal->j_errno;
1874 if (errcode == -ESHUTDOWN)
1875 errcode = 0;
1876 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1877 sb->s_errno = cpu_to_be32(errcode);
1878
1879 jbd2_write_superblock(journal, REQ_FUA);
1880 }
1881 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1882
journal_revoke_records_per_block(journal_t * journal)1883 static int journal_revoke_records_per_block(journal_t *journal)
1884 {
1885 int record_size;
1886 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1887
1888 if (jbd2_has_feature_64bit(journal))
1889 record_size = 8;
1890 else
1891 record_size = 4;
1892
1893 if (jbd2_journal_has_csum_v2or3(journal))
1894 space -= sizeof(struct jbd2_journal_block_tail);
1895 return space / record_size;
1896 }
1897
1898 /*
1899 * Read the superblock for a given journal, performing initial
1900 * validation of the format.
1901 */
journal_get_superblock(journal_t * journal)1902 static int journal_get_superblock(journal_t *journal)
1903 {
1904 struct buffer_head *bh;
1905 journal_superblock_t *sb;
1906 int err;
1907
1908 bh = journal->j_sb_buffer;
1909
1910 J_ASSERT(bh != NULL);
1911 err = bh_read(bh, 0);
1912 if (err < 0) {
1913 printk(KERN_ERR
1914 "JBD2: IO error reading journal superblock\n");
1915 goto out;
1916 }
1917
1918 if (buffer_verified(bh))
1919 return 0;
1920
1921 sb = journal->j_superblock;
1922
1923 err = -EINVAL;
1924
1925 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1926 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1927 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1928 goto out;
1929 }
1930
1931 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1932 case JBD2_SUPERBLOCK_V1:
1933 journal->j_format_version = 1;
1934 break;
1935 case JBD2_SUPERBLOCK_V2:
1936 journal->j_format_version = 2;
1937 break;
1938 default:
1939 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1940 goto out;
1941 }
1942
1943 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1944 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1945 else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1946 printk(KERN_WARNING "JBD2: journal file too short\n");
1947 goto out;
1948 }
1949
1950 if (be32_to_cpu(sb->s_first) == 0 ||
1951 be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1952 printk(KERN_WARNING
1953 "JBD2: Invalid start block of journal: %u\n",
1954 be32_to_cpu(sb->s_first));
1955 goto out;
1956 }
1957
1958 if (jbd2_has_feature_csum2(journal) &&
1959 jbd2_has_feature_csum3(journal)) {
1960 /* Can't have checksum v2 and v3 at the same time! */
1961 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1962 "at the same time!\n");
1963 goto out;
1964 }
1965
1966 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1967 jbd2_has_feature_checksum(journal)) {
1968 /* Can't have checksum v1 and v2 on at the same time! */
1969 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1970 "at the same time!\n");
1971 goto out;
1972 }
1973
1974 if (!jbd2_verify_csum_type(journal, sb)) {
1975 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1976 goto out;
1977 }
1978
1979 /* Load the checksum driver */
1980 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1981 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1982 if (IS_ERR(journal->j_chksum_driver)) {
1983 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1984 err = PTR_ERR(journal->j_chksum_driver);
1985 journal->j_chksum_driver = NULL;
1986 goto out;
1987 }
1988 }
1989
1990 if (jbd2_journal_has_csum_v2or3(journal)) {
1991 /* Check superblock checksum */
1992 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1993 printk(KERN_ERR "JBD2: journal checksum error\n");
1994 err = -EFSBADCRC;
1995 goto out;
1996 }
1997
1998 /* Precompute checksum seed for all metadata */
1999 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2000 sizeof(sb->s_uuid));
2001 }
2002
2003 journal->j_revoke_records_per_block =
2004 journal_revoke_records_per_block(journal);
2005 set_buffer_verified(bh);
2006
2007 return 0;
2008
2009 out:
2010 journal_fail_superblock(journal);
2011 return err;
2012 }
2013
2014 /*
2015 * Load the on-disk journal superblock and read the key fields into the
2016 * journal_t.
2017 */
2018
load_superblock(journal_t * journal)2019 static int load_superblock(journal_t *journal)
2020 {
2021 int err;
2022 journal_superblock_t *sb;
2023 int num_fc_blocks;
2024
2025 err = journal_get_superblock(journal);
2026 if (err)
2027 return err;
2028
2029 sb = journal->j_superblock;
2030
2031 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
2032 journal->j_tail = be32_to_cpu(sb->s_start);
2033 journal->j_first = be32_to_cpu(sb->s_first);
2034 journal->j_errno = be32_to_cpu(sb->s_errno);
2035 journal->j_last = be32_to_cpu(sb->s_maxlen);
2036
2037 if (jbd2_has_feature_fast_commit(journal)) {
2038 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
2039 num_fc_blocks = jbd2_journal_get_num_fc_blks(sb);
2040 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
2041 journal->j_last = journal->j_fc_last - num_fc_blocks;
2042 journal->j_fc_first = journal->j_last + 1;
2043 journal->j_fc_off = 0;
2044 }
2045
2046 return 0;
2047 }
2048
2049
2050 /**
2051 * jbd2_journal_load() - Read journal from disk.
2052 * @journal: Journal to act on.
2053 *
2054 * Given a journal_t structure which tells us which disk blocks contain
2055 * a journal, read the journal from disk to initialise the in-memory
2056 * structures.
2057 */
jbd2_journal_load(journal_t * journal)2058 int jbd2_journal_load(journal_t *journal)
2059 {
2060 int err;
2061 journal_superblock_t *sb;
2062
2063 err = load_superblock(journal);
2064 if (err)
2065 return err;
2066
2067 sb = journal->j_superblock;
2068 /* If this is a V2 superblock, then we have to check the
2069 * features flags on it. */
2070
2071 if (journal->j_format_version >= 2) {
2072 if ((sb->s_feature_ro_compat &
2073 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
2074 (sb->s_feature_incompat &
2075 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
2076 printk(KERN_WARNING
2077 "JBD2: Unrecognised features on journal\n");
2078 return -EINVAL;
2079 }
2080 }
2081
2082 /*
2083 * Create a slab for this blocksize
2084 */
2085 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2086 if (err)
2087 return err;
2088
2089 /* Let the recovery code check whether it needs to recover any
2090 * data from the journal. */
2091 if (jbd2_journal_recover(journal))
2092 goto recovery_error;
2093
2094 if (journal->j_failed_commit) {
2095 printk(KERN_ERR "JBD2: journal transaction %u on %s "
2096 "is corrupt.\n", journal->j_failed_commit,
2097 journal->j_devname);
2098 return -EFSCORRUPTED;
2099 }
2100 /*
2101 * clear JBD2_ABORT flag initialized in journal_init_common
2102 * here to update log tail information with the newest seq.
2103 */
2104 journal->j_flags &= ~JBD2_ABORT;
2105
2106 /* OK, we've finished with the dynamic journal bits:
2107 * reinitialise the dynamic contents of the superblock in memory
2108 * and reset them on disk. */
2109 if (journal_reset(journal))
2110 goto recovery_error;
2111
2112 journal->j_flags |= JBD2_LOADED;
2113 return 0;
2114
2115 recovery_error:
2116 printk(KERN_WARNING "JBD2: recovery failed\n");
2117 return -EIO;
2118 }
2119
2120 /**
2121 * jbd2_journal_destroy() - Release a journal_t structure.
2122 * @journal: Journal to act on.
2123 *
2124 * Release a journal_t structure once it is no longer in use by the
2125 * journaled object.
2126 * Return <0 if we couldn't clean up the journal.
2127 */
jbd2_journal_destroy(journal_t * journal)2128 int jbd2_journal_destroy(journal_t *journal)
2129 {
2130 int err = 0;
2131
2132 /* Wait for the commit thread to wake up and die. */
2133 journal_kill_thread(journal);
2134
2135 /* Force a final log commit */
2136 if (journal->j_running_transaction)
2137 jbd2_journal_commit_transaction(journal);
2138
2139 /* Force any old transactions to disk */
2140
2141 /* Totally anal locking here... */
2142 spin_lock(&journal->j_list_lock);
2143 while (journal->j_checkpoint_transactions != NULL) {
2144 spin_unlock(&journal->j_list_lock);
2145 mutex_lock_io(&journal->j_checkpoint_mutex);
2146 err = jbd2_log_do_checkpoint(journal);
2147 mutex_unlock(&journal->j_checkpoint_mutex);
2148 /*
2149 * If checkpointing failed, just free the buffers to avoid
2150 * looping forever
2151 */
2152 if (err) {
2153 jbd2_journal_destroy_checkpoint(journal);
2154 spin_lock(&journal->j_list_lock);
2155 break;
2156 }
2157 spin_lock(&journal->j_list_lock);
2158 }
2159
2160 J_ASSERT(journal->j_running_transaction == NULL);
2161 J_ASSERT(journal->j_committing_transaction == NULL);
2162 J_ASSERT(journal->j_checkpoint_transactions == NULL);
2163 spin_unlock(&journal->j_list_lock);
2164
2165 /*
2166 * OK, all checkpoint transactions have been checked, now check the
2167 * write out io error flag and abort the journal if some buffer failed
2168 * to write back to the original location, otherwise the filesystem
2169 * may become inconsistent.
2170 */
2171 if (!is_journal_aborted(journal) &&
2172 test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags))
2173 jbd2_journal_abort(journal, -EIO);
2174
2175 if (journal->j_sb_buffer) {
2176 if (!is_journal_aborted(journal)) {
2177 mutex_lock_io(&journal->j_checkpoint_mutex);
2178
2179 write_lock(&journal->j_state_lock);
2180 journal->j_tail_sequence =
2181 ++journal->j_transaction_sequence;
2182 write_unlock(&journal->j_state_lock);
2183
2184 jbd2_mark_journal_empty(journal, REQ_PREFLUSH | REQ_FUA);
2185 mutex_unlock(&journal->j_checkpoint_mutex);
2186 } else
2187 err = -EIO;
2188 brelse(journal->j_sb_buffer);
2189 }
2190
2191 if (journal->j_shrinker.flags & SHRINKER_REGISTERED) {
2192 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2193 unregister_shrinker(&journal->j_shrinker);
2194 }
2195 if (journal->j_proc_entry)
2196 jbd2_stats_proc_exit(journal);
2197 iput(journal->j_inode);
2198 if (journal->j_revoke)
2199 jbd2_journal_destroy_revoke(journal);
2200 if (journal->j_chksum_driver)
2201 crypto_free_shash(journal->j_chksum_driver);
2202 kfree(journal->j_fc_wbuf);
2203 kfree(journal->j_wbuf);
2204 kfree(journal);
2205
2206 return err;
2207 }
2208
2209
2210 /**
2211 * jbd2_journal_check_used_features() - Check if features specified are used.
2212 * @journal: Journal to check.
2213 * @compat: bitmask of compatible features
2214 * @ro: bitmask of features that force read-only mount
2215 * @incompat: bitmask of incompatible features
2216 *
2217 * Check whether the journal uses all of a given set of
2218 * features. Return true (non-zero) if it does.
2219 **/
2220
jbd2_journal_check_used_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2221 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2222 unsigned long ro, unsigned long incompat)
2223 {
2224 journal_superblock_t *sb;
2225
2226 if (!compat && !ro && !incompat)
2227 return 1;
2228 /* Load journal superblock if it is not loaded yet. */
2229 if (journal->j_format_version == 0 &&
2230 journal_get_superblock(journal) != 0)
2231 return 0;
2232 if (journal->j_format_version == 1)
2233 return 0;
2234
2235 sb = journal->j_superblock;
2236
2237 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2238 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2239 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2240 return 1;
2241
2242 return 0;
2243 }
2244
2245 /**
2246 * jbd2_journal_check_available_features() - Check feature set in journalling layer
2247 * @journal: Journal to check.
2248 * @compat: bitmask of compatible features
2249 * @ro: bitmask of features that force read-only mount
2250 * @incompat: bitmask of incompatible features
2251 *
2252 * Check whether the journaling code supports the use of
2253 * all of a given set of features on this journal. Return true
2254 * (non-zero) if it can. */
2255
jbd2_journal_check_available_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2256 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2257 unsigned long ro, unsigned long incompat)
2258 {
2259 if (!compat && !ro && !incompat)
2260 return 1;
2261
2262 /* We can support any known requested features iff the
2263 * superblock is in version 2. Otherwise we fail to support any
2264 * extended sb features. */
2265
2266 if (journal->j_format_version != 2)
2267 return 0;
2268
2269 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2270 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2271 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2272 return 1;
2273
2274 return 0;
2275 }
2276
2277 static int
jbd2_journal_initialize_fast_commit(journal_t * journal)2278 jbd2_journal_initialize_fast_commit(journal_t *journal)
2279 {
2280 journal_superblock_t *sb = journal->j_superblock;
2281 unsigned long long num_fc_blks;
2282
2283 num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2284 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2285 return -ENOSPC;
2286
2287 /* Are we called twice? */
2288 WARN_ON(journal->j_fc_wbuf != NULL);
2289 journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2290 sizeof(struct buffer_head *), GFP_KERNEL);
2291 if (!journal->j_fc_wbuf)
2292 return -ENOMEM;
2293
2294 journal->j_fc_wbufsize = num_fc_blks;
2295 journal->j_fc_last = journal->j_last;
2296 journal->j_last = journal->j_fc_last - num_fc_blks;
2297 journal->j_fc_first = journal->j_last + 1;
2298 journal->j_fc_off = 0;
2299 journal->j_free = journal->j_last - journal->j_first;
2300 journal->j_max_transaction_buffers =
2301 jbd2_journal_get_max_txn_bufs(journal);
2302
2303 return 0;
2304 }
2305
2306 /**
2307 * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2308 * @journal: Journal to act on.
2309 * @compat: bitmask of compatible features
2310 * @ro: bitmask of features that force read-only mount
2311 * @incompat: bitmask of incompatible features
2312 *
2313 * Mark a given journal feature as present on the
2314 * superblock. Returns true if the requested features could be set.
2315 *
2316 */
2317
jbd2_journal_set_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2318 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2319 unsigned long ro, unsigned long incompat)
2320 {
2321 #define INCOMPAT_FEATURE_ON(f) \
2322 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2323 #define COMPAT_FEATURE_ON(f) \
2324 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2325 journal_superblock_t *sb;
2326
2327 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2328 return 1;
2329
2330 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2331 return 0;
2332
2333 /* If enabling v2 checksums, turn on v3 instead */
2334 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2335 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2336 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2337 }
2338
2339 /* Asking for checksumming v3 and v1? Only give them v3. */
2340 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2341 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2342 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2343
2344 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2345 compat, ro, incompat);
2346
2347 sb = journal->j_superblock;
2348
2349 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2350 if (jbd2_journal_initialize_fast_commit(journal)) {
2351 pr_err("JBD2: Cannot enable fast commits.\n");
2352 return 0;
2353 }
2354 }
2355
2356 /* Load the checksum driver if necessary */
2357 if ((journal->j_chksum_driver == NULL) &&
2358 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2359 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2360 if (IS_ERR(journal->j_chksum_driver)) {
2361 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2362 journal->j_chksum_driver = NULL;
2363 return 0;
2364 }
2365 /* Precompute checksum seed for all metadata */
2366 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2367 sizeof(sb->s_uuid));
2368 }
2369
2370 lock_buffer(journal->j_sb_buffer);
2371
2372 /* If enabling v3 checksums, update superblock */
2373 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2374 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2375 sb->s_feature_compat &=
2376 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2377 }
2378
2379 /* If enabling v1 checksums, downgrade superblock */
2380 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2381 sb->s_feature_incompat &=
2382 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2383 JBD2_FEATURE_INCOMPAT_CSUM_V3);
2384
2385 sb->s_feature_compat |= cpu_to_be32(compat);
2386 sb->s_feature_ro_compat |= cpu_to_be32(ro);
2387 sb->s_feature_incompat |= cpu_to_be32(incompat);
2388 unlock_buffer(journal->j_sb_buffer);
2389 journal->j_revoke_records_per_block =
2390 journal_revoke_records_per_block(journal);
2391
2392 return 1;
2393 #undef COMPAT_FEATURE_ON
2394 #undef INCOMPAT_FEATURE_ON
2395 }
2396
2397 /*
2398 * jbd2_journal_clear_features() - Clear a given journal feature in the
2399 * superblock
2400 * @journal: Journal to act on.
2401 * @compat: bitmask of compatible features
2402 * @ro: bitmask of features that force read-only mount
2403 * @incompat: bitmask of incompatible features
2404 *
2405 * Clear a given journal feature as present on the
2406 * superblock.
2407 */
jbd2_journal_clear_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2408 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2409 unsigned long ro, unsigned long incompat)
2410 {
2411 journal_superblock_t *sb;
2412
2413 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2414 compat, ro, incompat);
2415
2416 sb = journal->j_superblock;
2417
2418 sb->s_feature_compat &= ~cpu_to_be32(compat);
2419 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2420 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
2421 journal->j_revoke_records_per_block =
2422 journal_revoke_records_per_block(journal);
2423 }
2424 EXPORT_SYMBOL(jbd2_journal_clear_features);
2425
2426 /**
2427 * jbd2_journal_flush() - Flush journal
2428 * @journal: Journal to act on.
2429 * @flags: optional operation on the journal blocks after the flush (see below)
2430 *
2431 * Flush all data for a given journal to disk and empty the journal.
2432 * Filesystems can use this when remounting readonly to ensure that
2433 * recovery does not need to happen on remount. Optionally, a discard or zeroout
2434 * can be issued on the journal blocks after flushing.
2435 *
2436 * flags:
2437 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2438 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2439 */
jbd2_journal_flush(journal_t * journal,unsigned int flags)2440 int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2441 {
2442 int err = 0;
2443 transaction_t *transaction = NULL;
2444
2445 write_lock(&journal->j_state_lock);
2446
2447 /* Force everything buffered to the log... */
2448 if (journal->j_running_transaction) {
2449 transaction = journal->j_running_transaction;
2450 __jbd2_log_start_commit(journal, transaction->t_tid);
2451 } else if (journal->j_committing_transaction)
2452 transaction = journal->j_committing_transaction;
2453
2454 /* Wait for the log commit to complete... */
2455 if (transaction) {
2456 tid_t tid = transaction->t_tid;
2457
2458 write_unlock(&journal->j_state_lock);
2459 jbd2_log_wait_commit(journal, tid);
2460 } else {
2461 write_unlock(&journal->j_state_lock);
2462 }
2463
2464 /* ...and flush everything in the log out to disk. */
2465 spin_lock(&journal->j_list_lock);
2466 while (!err && journal->j_checkpoint_transactions != NULL) {
2467 spin_unlock(&journal->j_list_lock);
2468 mutex_lock_io(&journal->j_checkpoint_mutex);
2469 err = jbd2_log_do_checkpoint(journal);
2470 mutex_unlock(&journal->j_checkpoint_mutex);
2471 spin_lock(&journal->j_list_lock);
2472 }
2473 spin_unlock(&journal->j_list_lock);
2474
2475 if (is_journal_aborted(journal))
2476 return -EIO;
2477
2478 mutex_lock_io(&journal->j_checkpoint_mutex);
2479 if (!err) {
2480 err = jbd2_cleanup_journal_tail(journal);
2481 if (err < 0) {
2482 mutex_unlock(&journal->j_checkpoint_mutex);
2483 goto out;
2484 }
2485 err = 0;
2486 }
2487
2488 /* Finally, mark the journal as really needing no recovery.
2489 * This sets s_start==0 in the underlying superblock, which is
2490 * the magic code for a fully-recovered superblock. Any future
2491 * commits of data to the journal will restore the current
2492 * s_start value. */
2493 jbd2_mark_journal_empty(journal, REQ_FUA);
2494
2495 if (flags)
2496 err = __jbd2_journal_erase(journal, flags);
2497
2498 mutex_unlock(&journal->j_checkpoint_mutex);
2499 write_lock(&journal->j_state_lock);
2500 J_ASSERT(!journal->j_running_transaction);
2501 J_ASSERT(!journal->j_committing_transaction);
2502 J_ASSERT(!journal->j_checkpoint_transactions);
2503 J_ASSERT(journal->j_head == journal->j_tail);
2504 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2505 write_unlock(&journal->j_state_lock);
2506 out:
2507 return err;
2508 }
2509
2510 /**
2511 * jbd2_journal_wipe() - Wipe journal contents
2512 * @journal: Journal to act on.
2513 * @write: flag (see below)
2514 *
2515 * Wipe out all of the contents of a journal, safely. This will produce
2516 * a warning if the journal contains any valid recovery information.
2517 * Must be called between journal_init_*() and jbd2_journal_load().
2518 *
2519 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2520 * we merely suppress recovery.
2521 */
2522
jbd2_journal_wipe(journal_t * journal,int write)2523 int jbd2_journal_wipe(journal_t *journal, int write)
2524 {
2525 int err = 0;
2526
2527 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2528
2529 err = load_superblock(journal);
2530 if (err)
2531 return err;
2532
2533 if (!journal->j_tail)
2534 goto no_recovery;
2535
2536 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2537 write ? "Clearing" : "Ignoring");
2538
2539 err = jbd2_journal_skip_recovery(journal);
2540 if (write) {
2541 /* Lock to make assertions happy... */
2542 mutex_lock_io(&journal->j_checkpoint_mutex);
2543 jbd2_mark_journal_empty(journal, REQ_FUA);
2544 mutex_unlock(&journal->j_checkpoint_mutex);
2545 }
2546
2547 no_recovery:
2548 return err;
2549 }
2550
2551 /**
2552 * jbd2_journal_abort () - Shutdown the journal immediately.
2553 * @journal: the journal to shutdown.
2554 * @errno: an error number to record in the journal indicating
2555 * the reason for the shutdown.
2556 *
2557 * Perform a complete, immediate shutdown of the ENTIRE
2558 * journal (not of a single transaction). This operation cannot be
2559 * undone without closing and reopening the journal.
2560 *
2561 * The jbd2_journal_abort function is intended to support higher level error
2562 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2563 * mode.
2564 *
2565 * Journal abort has very specific semantics. Any existing dirty,
2566 * unjournaled buffers in the main filesystem will still be written to
2567 * disk by bdflush, but the journaling mechanism will be suspended
2568 * immediately and no further transaction commits will be honoured.
2569 *
2570 * Any dirty, journaled buffers will be written back to disk without
2571 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2572 * filesystem, but we _do_ attempt to leave as much data as possible
2573 * behind for fsck to use for cleanup.
2574 *
2575 * Any attempt to get a new transaction handle on a journal which is in
2576 * ABORT state will just result in an -EROFS error return. A
2577 * jbd2_journal_stop on an existing handle will return -EIO if we have
2578 * entered abort state during the update.
2579 *
2580 * Recursive transactions are not disturbed by journal abort until the
2581 * final jbd2_journal_stop, which will receive the -EIO error.
2582 *
2583 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2584 * which will be recorded (if possible) in the journal superblock. This
2585 * allows a client to record failure conditions in the middle of a
2586 * transaction without having to complete the transaction to record the
2587 * failure to disk. ext3_error, for example, now uses this
2588 * functionality.
2589 *
2590 */
2591
jbd2_journal_abort(journal_t * journal,int errno)2592 void jbd2_journal_abort(journal_t *journal, int errno)
2593 {
2594 transaction_t *transaction;
2595
2596 /*
2597 * Lock the aborting procedure until everything is done, this avoid
2598 * races between filesystem's error handling flow (e.g. ext4_abort()),
2599 * ensure panic after the error info is written into journal's
2600 * superblock.
2601 */
2602 mutex_lock(&journal->j_abort_mutex);
2603 /*
2604 * ESHUTDOWN always takes precedence because a file system check
2605 * caused by any other journal abort error is not required after
2606 * a shutdown triggered.
2607 */
2608 write_lock(&journal->j_state_lock);
2609 if (journal->j_flags & JBD2_ABORT) {
2610 int old_errno = journal->j_errno;
2611
2612 write_unlock(&journal->j_state_lock);
2613 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2614 journal->j_errno = errno;
2615 jbd2_journal_update_sb_errno(journal);
2616 }
2617 mutex_unlock(&journal->j_abort_mutex);
2618 return;
2619 }
2620
2621 /*
2622 * Mark the abort as occurred and start current running transaction
2623 * to release all journaled buffer.
2624 */
2625 pr_err("Aborting journal on device %s.\n", journal->j_devname);
2626
2627 journal->j_flags |= JBD2_ABORT;
2628 journal->j_errno = errno;
2629 transaction = journal->j_running_transaction;
2630 if (transaction)
2631 __jbd2_log_start_commit(journal, transaction->t_tid);
2632 write_unlock(&journal->j_state_lock);
2633
2634 /*
2635 * Record errno to the journal super block, so that fsck and jbd2
2636 * layer could realise that a filesystem check is needed.
2637 */
2638 jbd2_journal_update_sb_errno(journal);
2639 mutex_unlock(&journal->j_abort_mutex);
2640 }
2641
2642 /**
2643 * jbd2_journal_errno() - returns the journal's error state.
2644 * @journal: journal to examine.
2645 *
2646 * This is the errno number set with jbd2_journal_abort(), the last
2647 * time the journal was mounted - if the journal was stopped
2648 * without calling abort this will be 0.
2649 *
2650 * If the journal has been aborted on this mount time -EROFS will
2651 * be returned.
2652 */
jbd2_journal_errno(journal_t * journal)2653 int jbd2_journal_errno(journal_t *journal)
2654 {
2655 int err;
2656
2657 read_lock(&journal->j_state_lock);
2658 if (journal->j_flags & JBD2_ABORT)
2659 err = -EROFS;
2660 else
2661 err = journal->j_errno;
2662 read_unlock(&journal->j_state_lock);
2663 return err;
2664 }
2665
2666 /**
2667 * jbd2_journal_clear_err() - clears the journal's error state
2668 * @journal: journal to act on.
2669 *
2670 * An error must be cleared or acked to take a FS out of readonly
2671 * mode.
2672 */
jbd2_journal_clear_err(journal_t * journal)2673 int jbd2_journal_clear_err(journal_t *journal)
2674 {
2675 int err = 0;
2676
2677 write_lock(&journal->j_state_lock);
2678 if (journal->j_flags & JBD2_ABORT)
2679 err = -EROFS;
2680 else
2681 journal->j_errno = 0;
2682 write_unlock(&journal->j_state_lock);
2683 return err;
2684 }
2685
2686 /**
2687 * jbd2_journal_ack_err() - Ack journal err.
2688 * @journal: journal to act on.
2689 *
2690 * An error must be cleared or acked to take a FS out of readonly
2691 * mode.
2692 */
jbd2_journal_ack_err(journal_t * journal)2693 void jbd2_journal_ack_err(journal_t *journal)
2694 {
2695 write_lock(&journal->j_state_lock);
2696 if (journal->j_errno)
2697 journal->j_flags |= JBD2_ACK_ERR;
2698 write_unlock(&journal->j_state_lock);
2699 }
2700
jbd2_journal_blocks_per_page(struct inode * inode)2701 int jbd2_journal_blocks_per_page(struct inode *inode)
2702 {
2703 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2704 }
2705
2706 /*
2707 * helper functions to deal with 32 or 64bit block numbers.
2708 */
journal_tag_bytes(journal_t * journal)2709 size_t journal_tag_bytes(journal_t *journal)
2710 {
2711 size_t sz;
2712
2713 if (jbd2_has_feature_csum3(journal))
2714 return sizeof(journal_block_tag3_t);
2715
2716 sz = sizeof(journal_block_tag_t);
2717
2718 if (jbd2_has_feature_csum2(journal))
2719 sz += sizeof(__u16);
2720
2721 if (jbd2_has_feature_64bit(journal))
2722 return sz;
2723 else
2724 return sz - sizeof(__u32);
2725 }
2726
2727 /*
2728 * JBD memory management
2729 *
2730 * These functions are used to allocate block-sized chunks of memory
2731 * used for making copies of buffer_head data. Very often it will be
2732 * page-sized chunks of data, but sometimes it will be in
2733 * sub-page-size chunks. (For example, 16k pages on Power systems
2734 * with a 4k block file system.) For blocks smaller than a page, we
2735 * use a SLAB allocator. There are slab caches for each block size,
2736 * which are allocated at mount time, if necessary, and we only free
2737 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2738 * this reason we don't need to a mutex to protect access to
2739 * jbd2_slab[] allocating or releasing memory; only in
2740 * jbd2_journal_create_slab().
2741 */
2742 #define JBD2_MAX_SLABS 8
2743 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2744
2745 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2746 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2747 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2748 };
2749
2750
jbd2_journal_destroy_slabs(void)2751 static void jbd2_journal_destroy_slabs(void)
2752 {
2753 int i;
2754
2755 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2756 kmem_cache_destroy(jbd2_slab[i]);
2757 jbd2_slab[i] = NULL;
2758 }
2759 }
2760
jbd2_journal_create_slab(size_t size)2761 static int jbd2_journal_create_slab(size_t size)
2762 {
2763 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2764 int i = order_base_2(size) - 10;
2765 size_t slab_size;
2766
2767 if (size == PAGE_SIZE)
2768 return 0;
2769
2770 if (i >= JBD2_MAX_SLABS)
2771 return -EINVAL;
2772
2773 if (unlikely(i < 0))
2774 i = 0;
2775 mutex_lock(&jbd2_slab_create_mutex);
2776 if (jbd2_slab[i]) {
2777 mutex_unlock(&jbd2_slab_create_mutex);
2778 return 0; /* Already created */
2779 }
2780
2781 slab_size = 1 << (i+10);
2782 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2783 slab_size, 0, NULL);
2784 mutex_unlock(&jbd2_slab_create_mutex);
2785 if (!jbd2_slab[i]) {
2786 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2787 return -ENOMEM;
2788 }
2789 return 0;
2790 }
2791
get_slab(size_t size)2792 static struct kmem_cache *get_slab(size_t size)
2793 {
2794 int i = order_base_2(size) - 10;
2795
2796 BUG_ON(i >= JBD2_MAX_SLABS);
2797 if (unlikely(i < 0))
2798 i = 0;
2799 BUG_ON(jbd2_slab[i] == NULL);
2800 return jbd2_slab[i];
2801 }
2802
jbd2_alloc(size_t size,gfp_t flags)2803 void *jbd2_alloc(size_t size, gfp_t flags)
2804 {
2805 void *ptr;
2806
2807 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2808
2809 if (size < PAGE_SIZE)
2810 ptr = kmem_cache_alloc(get_slab(size), flags);
2811 else
2812 ptr = (void *)__get_free_pages(flags, get_order(size));
2813
2814 /* Check alignment; SLUB has gotten this wrong in the past,
2815 * and this can lead to user data corruption! */
2816 BUG_ON(((unsigned long) ptr) & (size-1));
2817
2818 return ptr;
2819 }
2820
jbd2_free(void * ptr,size_t size)2821 void jbd2_free(void *ptr, size_t size)
2822 {
2823 if (size < PAGE_SIZE)
2824 kmem_cache_free(get_slab(size), ptr);
2825 else
2826 free_pages((unsigned long)ptr, get_order(size));
2827 };
2828
2829 /*
2830 * Journal_head storage management
2831 */
2832 static struct kmem_cache *jbd2_journal_head_cache;
2833 #ifdef CONFIG_JBD2_DEBUG
2834 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2835 #endif
2836
jbd2_journal_init_journal_head_cache(void)2837 static int __init jbd2_journal_init_journal_head_cache(void)
2838 {
2839 J_ASSERT(!jbd2_journal_head_cache);
2840 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2841 sizeof(struct journal_head),
2842 0, /* offset */
2843 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2844 NULL); /* ctor */
2845 if (!jbd2_journal_head_cache) {
2846 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2847 return -ENOMEM;
2848 }
2849 return 0;
2850 }
2851
jbd2_journal_destroy_journal_head_cache(void)2852 static void jbd2_journal_destroy_journal_head_cache(void)
2853 {
2854 kmem_cache_destroy(jbd2_journal_head_cache);
2855 jbd2_journal_head_cache = NULL;
2856 }
2857
2858 /*
2859 * journal_head splicing and dicing
2860 */
journal_alloc_journal_head(void)2861 static struct journal_head *journal_alloc_journal_head(void)
2862 {
2863 struct journal_head *ret;
2864
2865 #ifdef CONFIG_JBD2_DEBUG
2866 atomic_inc(&nr_journal_heads);
2867 #endif
2868 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2869 if (!ret) {
2870 jbd2_debug(1, "out of memory for journal_head\n");
2871 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2872 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2873 GFP_NOFS | __GFP_NOFAIL);
2874 }
2875 if (ret)
2876 spin_lock_init(&ret->b_state_lock);
2877 return ret;
2878 }
2879
journal_free_journal_head(struct journal_head * jh)2880 static void journal_free_journal_head(struct journal_head *jh)
2881 {
2882 #ifdef CONFIG_JBD2_DEBUG
2883 atomic_dec(&nr_journal_heads);
2884 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2885 #endif
2886 kmem_cache_free(jbd2_journal_head_cache, jh);
2887 }
2888
2889 /*
2890 * A journal_head is attached to a buffer_head whenever JBD has an
2891 * interest in the buffer.
2892 *
2893 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2894 * is set. This bit is tested in core kernel code where we need to take
2895 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2896 * there.
2897 *
2898 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2899 *
2900 * When a buffer has its BH_JBD bit set it is immune from being released by
2901 * core kernel code, mainly via ->b_count.
2902 *
2903 * A journal_head is detached from its buffer_head when the journal_head's
2904 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2905 * transaction (b_cp_transaction) hold their references to b_jcount.
2906 *
2907 * Various places in the kernel want to attach a journal_head to a buffer_head
2908 * _before_ attaching the journal_head to a transaction. To protect the
2909 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2910 * journal_head's b_jcount refcount by one. The caller must call
2911 * jbd2_journal_put_journal_head() to undo this.
2912 *
2913 * So the typical usage would be:
2914 *
2915 * (Attach a journal_head if needed. Increments b_jcount)
2916 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2917 * ...
2918 * (Get another reference for transaction)
2919 * jbd2_journal_grab_journal_head(bh);
2920 * jh->b_transaction = xxx;
2921 * (Put original reference)
2922 * jbd2_journal_put_journal_head(jh);
2923 */
2924
2925 /*
2926 * Give a buffer_head a journal_head.
2927 *
2928 * May sleep.
2929 */
jbd2_journal_add_journal_head(struct buffer_head * bh)2930 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2931 {
2932 struct journal_head *jh;
2933 struct journal_head *new_jh = NULL;
2934
2935 repeat:
2936 if (!buffer_jbd(bh))
2937 new_jh = journal_alloc_journal_head();
2938
2939 jbd_lock_bh_journal_head(bh);
2940 if (buffer_jbd(bh)) {
2941 jh = bh2jh(bh);
2942 } else {
2943 J_ASSERT_BH(bh,
2944 (atomic_read(&bh->b_count) > 0) ||
2945 (bh->b_page && bh->b_page->mapping));
2946
2947 if (!new_jh) {
2948 jbd_unlock_bh_journal_head(bh);
2949 goto repeat;
2950 }
2951
2952 jh = new_jh;
2953 new_jh = NULL; /* We consumed it */
2954 set_buffer_jbd(bh);
2955 bh->b_private = jh;
2956 jh->b_bh = bh;
2957 get_bh(bh);
2958 BUFFER_TRACE(bh, "added journal_head");
2959 }
2960 jh->b_jcount++;
2961 jbd_unlock_bh_journal_head(bh);
2962 if (new_jh)
2963 journal_free_journal_head(new_jh);
2964 return bh->b_private;
2965 }
2966
2967 /*
2968 * Grab a ref against this buffer_head's journal_head. If it ended up not
2969 * having a journal_head, return NULL
2970 */
jbd2_journal_grab_journal_head(struct buffer_head * bh)2971 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2972 {
2973 struct journal_head *jh = NULL;
2974
2975 jbd_lock_bh_journal_head(bh);
2976 if (buffer_jbd(bh)) {
2977 jh = bh2jh(bh);
2978 jh->b_jcount++;
2979 }
2980 jbd_unlock_bh_journal_head(bh);
2981 return jh;
2982 }
2983 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2984
__journal_remove_journal_head(struct buffer_head * bh)2985 static void __journal_remove_journal_head(struct buffer_head *bh)
2986 {
2987 struct journal_head *jh = bh2jh(bh);
2988
2989 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2990 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2991 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2992 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2993 J_ASSERT_BH(bh, buffer_jbd(bh));
2994 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2995 BUFFER_TRACE(bh, "remove journal_head");
2996
2997 /* Unlink before dropping the lock */
2998 bh->b_private = NULL;
2999 jh->b_bh = NULL; /* debug, really */
3000 clear_buffer_jbd(bh);
3001 }
3002
journal_release_journal_head(struct journal_head * jh,size_t b_size)3003 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
3004 {
3005 if (jh->b_frozen_data) {
3006 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
3007 jbd2_free(jh->b_frozen_data, b_size);
3008 }
3009 if (jh->b_committed_data) {
3010 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
3011 jbd2_free(jh->b_committed_data, b_size);
3012 }
3013 journal_free_journal_head(jh);
3014 }
3015
3016 /*
3017 * Drop a reference on the passed journal_head. If it fell to zero then
3018 * release the journal_head from the buffer_head.
3019 */
jbd2_journal_put_journal_head(struct journal_head * jh)3020 void jbd2_journal_put_journal_head(struct journal_head *jh)
3021 {
3022 struct buffer_head *bh = jh2bh(jh);
3023
3024 jbd_lock_bh_journal_head(bh);
3025 J_ASSERT_JH(jh, jh->b_jcount > 0);
3026 --jh->b_jcount;
3027 if (!jh->b_jcount) {
3028 __journal_remove_journal_head(bh);
3029 jbd_unlock_bh_journal_head(bh);
3030 journal_release_journal_head(jh, bh->b_size);
3031 __brelse(bh);
3032 } else {
3033 jbd_unlock_bh_journal_head(bh);
3034 }
3035 }
3036 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3037
3038 /*
3039 * Initialize jbd inode head
3040 */
jbd2_journal_init_jbd_inode(struct jbd2_inode * jinode,struct inode * inode)3041 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3042 {
3043 jinode->i_transaction = NULL;
3044 jinode->i_next_transaction = NULL;
3045 jinode->i_vfs_inode = inode;
3046 jinode->i_flags = 0;
3047 jinode->i_dirty_start = 0;
3048 jinode->i_dirty_end = 0;
3049 INIT_LIST_HEAD(&jinode->i_list);
3050 }
3051
3052 /*
3053 * Function to be called before we start removing inode from memory (i.e.,
3054 * clear_inode() is a fine place to be called from). It removes inode from
3055 * transaction's lists.
3056 */
jbd2_journal_release_jbd_inode(journal_t * journal,struct jbd2_inode * jinode)3057 void jbd2_journal_release_jbd_inode(journal_t *journal,
3058 struct jbd2_inode *jinode)
3059 {
3060 if (!journal)
3061 return;
3062 restart:
3063 spin_lock(&journal->j_list_lock);
3064 /* Is commit writing out inode - we have to wait */
3065 if (jinode->i_flags & JI_COMMIT_RUNNING) {
3066 wait_queue_head_t *wq;
3067 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3068 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3069 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3070 spin_unlock(&journal->j_list_lock);
3071 schedule();
3072 finish_wait(wq, &wait.wq_entry);
3073 goto restart;
3074 }
3075
3076 if (jinode->i_transaction) {
3077 list_del(&jinode->i_list);
3078 jinode->i_transaction = NULL;
3079 }
3080 spin_unlock(&journal->j_list_lock);
3081 }
3082
3083
3084 #ifdef CONFIG_PROC_FS
3085
3086 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3087
jbd2_create_jbd_stats_proc_entry(void)3088 static void __init jbd2_create_jbd_stats_proc_entry(void)
3089 {
3090 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3091 }
3092
jbd2_remove_jbd_stats_proc_entry(void)3093 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3094 {
3095 if (proc_jbd2_stats)
3096 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3097 }
3098
3099 #else
3100
3101 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3102 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3103
3104 #endif
3105
3106 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3107
jbd2_journal_init_inode_cache(void)3108 static int __init jbd2_journal_init_inode_cache(void)
3109 {
3110 J_ASSERT(!jbd2_inode_cache);
3111 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3112 if (!jbd2_inode_cache) {
3113 pr_emerg("JBD2: failed to create inode cache\n");
3114 return -ENOMEM;
3115 }
3116 return 0;
3117 }
3118
jbd2_journal_init_handle_cache(void)3119 static int __init jbd2_journal_init_handle_cache(void)
3120 {
3121 J_ASSERT(!jbd2_handle_cache);
3122 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3123 if (!jbd2_handle_cache) {
3124 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3125 return -ENOMEM;
3126 }
3127 return 0;
3128 }
3129
jbd2_journal_destroy_inode_cache(void)3130 static void jbd2_journal_destroy_inode_cache(void)
3131 {
3132 kmem_cache_destroy(jbd2_inode_cache);
3133 jbd2_inode_cache = NULL;
3134 }
3135
jbd2_journal_destroy_handle_cache(void)3136 static void jbd2_journal_destroy_handle_cache(void)
3137 {
3138 kmem_cache_destroy(jbd2_handle_cache);
3139 jbd2_handle_cache = NULL;
3140 }
3141
3142 /*
3143 * Module startup and shutdown
3144 */
3145
journal_init_caches(void)3146 static int __init journal_init_caches(void)
3147 {
3148 int ret;
3149
3150 ret = jbd2_journal_init_revoke_record_cache();
3151 if (ret == 0)
3152 ret = jbd2_journal_init_revoke_table_cache();
3153 if (ret == 0)
3154 ret = jbd2_journal_init_journal_head_cache();
3155 if (ret == 0)
3156 ret = jbd2_journal_init_handle_cache();
3157 if (ret == 0)
3158 ret = jbd2_journal_init_inode_cache();
3159 if (ret == 0)
3160 ret = jbd2_journal_init_transaction_cache();
3161 return ret;
3162 }
3163
jbd2_journal_destroy_caches(void)3164 static void jbd2_journal_destroy_caches(void)
3165 {
3166 jbd2_journal_destroy_revoke_record_cache();
3167 jbd2_journal_destroy_revoke_table_cache();
3168 jbd2_journal_destroy_journal_head_cache();
3169 jbd2_journal_destroy_handle_cache();
3170 jbd2_journal_destroy_inode_cache();
3171 jbd2_journal_destroy_transaction_cache();
3172 jbd2_journal_destroy_slabs();
3173 }
3174
journal_init(void)3175 static int __init journal_init(void)
3176 {
3177 int ret;
3178
3179 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3180
3181 ret = journal_init_caches();
3182 if (ret == 0) {
3183 jbd2_create_jbd_stats_proc_entry();
3184 } else {
3185 jbd2_journal_destroy_caches();
3186 }
3187 return ret;
3188 }
3189
journal_exit(void)3190 static void __exit journal_exit(void)
3191 {
3192 #ifdef CONFIG_JBD2_DEBUG
3193 int n = atomic_read(&nr_journal_heads);
3194 if (n)
3195 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3196 #endif
3197 jbd2_remove_jbd_stats_proc_entry();
3198 jbd2_journal_destroy_caches();
3199 }
3200
3201 MODULE_LICENSE("GPL");
3202 module_init(journal_init);
3203 module_exit(journal_exit);
3204
3205