Loading...
1/*
2 * linux/fs/jbd2/journal.c
3 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5 *
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Generic filesystem journal-writing code; part of the ext2fs
13 * journaling system.
14 *
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
18 *
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
23 */
24
25#include <linux/module.h>
26#include <linux/time.h>
27#include <linux/fs.h>
28#include <linux/jbd2.h>
29#include <linux/errno.h>
30#include <linux/slab.h>
31#include <linux/init.h>
32#include <linux/mm.h>
33#include <linux/freezer.h>
34#include <linux/pagemap.h>
35#include <linux/kthread.h>
36#include <linux/poison.h>
37#include <linux/proc_fs.h>
38#include <linux/seq_file.h>
39#include <linux/math64.h>
40#include <linux/hash.h>
41#include <linux/log2.h>
42#include <linux/vmalloc.h>
43#include <linux/backing-dev.h>
44#include <linux/bitops.h>
45#include <linux/ratelimit.h>
46
47#define CREATE_TRACE_POINTS
48#include <trace/events/jbd2.h>
49
50#include <linux/uaccess.h>
51#include <asm/page.h>
52
53#ifdef CONFIG_JBD2_DEBUG
54ushort jbd2_journal_enable_debug __read_mostly;
55EXPORT_SYMBOL(jbd2_journal_enable_debug);
56
57module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
58MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
59#endif
60
61EXPORT_SYMBOL(jbd2_journal_extend);
62EXPORT_SYMBOL(jbd2_journal_stop);
63EXPORT_SYMBOL(jbd2_journal_lock_updates);
64EXPORT_SYMBOL(jbd2_journal_unlock_updates);
65EXPORT_SYMBOL(jbd2_journal_get_write_access);
66EXPORT_SYMBOL(jbd2_journal_get_create_access);
67EXPORT_SYMBOL(jbd2_journal_get_undo_access);
68EXPORT_SYMBOL(jbd2_journal_set_triggers);
69EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
70EXPORT_SYMBOL(jbd2_journal_forget);
71#if 0
72EXPORT_SYMBOL(journal_sync_buffer);
73#endif
74EXPORT_SYMBOL(jbd2_journal_flush);
75EXPORT_SYMBOL(jbd2_journal_revoke);
76
77EXPORT_SYMBOL(jbd2_journal_init_dev);
78EXPORT_SYMBOL(jbd2_journal_init_inode);
79EXPORT_SYMBOL(jbd2_journal_check_used_features);
80EXPORT_SYMBOL(jbd2_journal_check_available_features);
81EXPORT_SYMBOL(jbd2_journal_set_features);
82EXPORT_SYMBOL(jbd2_journal_load);
83EXPORT_SYMBOL(jbd2_journal_destroy);
84EXPORT_SYMBOL(jbd2_journal_abort);
85EXPORT_SYMBOL(jbd2_journal_errno);
86EXPORT_SYMBOL(jbd2_journal_ack_err);
87EXPORT_SYMBOL(jbd2_journal_clear_err);
88EXPORT_SYMBOL(jbd2_log_wait_commit);
89EXPORT_SYMBOL(jbd2_log_start_commit);
90EXPORT_SYMBOL(jbd2_journal_start_commit);
91EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
92EXPORT_SYMBOL(jbd2_journal_wipe);
93EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
94EXPORT_SYMBOL(jbd2_journal_invalidatepage);
95EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
96EXPORT_SYMBOL(jbd2_journal_force_commit);
97EXPORT_SYMBOL(jbd2_journal_inode_add_write);
98EXPORT_SYMBOL(jbd2_journal_inode_add_wait);
99EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
100EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
101EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
102EXPORT_SYMBOL(jbd2_inode_cache);
103
104static void __journal_abort_soft (journal_t *journal, int errno);
105static int jbd2_journal_create_slab(size_t slab_size);
106
107#ifdef CONFIG_JBD2_DEBUG
108void __jbd2_debug(int level, const char *file, const char *func,
109 unsigned int line, const char *fmt, ...)
110{
111 struct va_format vaf;
112 va_list args;
113
114 if (level > jbd2_journal_enable_debug)
115 return;
116 va_start(args, fmt);
117 vaf.fmt = fmt;
118 vaf.va = &args;
119 printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
120 va_end(args);
121}
122EXPORT_SYMBOL(__jbd2_debug);
123#endif
124
125/* Checksumming functions */
126static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
127{
128 if (!jbd2_journal_has_csum_v2or3_feature(j))
129 return 1;
130
131 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
132}
133
134static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
135{
136 __u32 csum;
137 __be32 old_csum;
138
139 old_csum = sb->s_checksum;
140 sb->s_checksum = 0;
141 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
142 sb->s_checksum = old_csum;
143
144 return cpu_to_be32(csum);
145}
146
147static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
148{
149 if (!jbd2_journal_has_csum_v2or3(j))
150 return 1;
151
152 return sb->s_checksum == jbd2_superblock_csum(j, sb);
153}
154
155static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
156{
157 if (!jbd2_journal_has_csum_v2or3(j))
158 return;
159
160 sb->s_checksum = jbd2_superblock_csum(j, sb);
161}
162
163/*
164 * Helper function used to manage commit timeouts
165 */
166
167static void commit_timeout(unsigned long __data)
168{
169 struct task_struct * p = (struct task_struct *) __data;
170
171 wake_up_process(p);
172}
173
174/*
175 * kjournald2: The main thread function used to manage a logging device
176 * journal.
177 *
178 * This kernel thread is responsible for two things:
179 *
180 * 1) COMMIT: Every so often we need to commit the current state of the
181 * filesystem to disk. The journal thread is responsible for writing
182 * all of the metadata buffers to disk.
183 *
184 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
185 * of the data in that part of the log has been rewritten elsewhere on
186 * the disk. Flushing these old buffers to reclaim space in the log is
187 * known as checkpointing, and this thread is responsible for that job.
188 */
189
190static int kjournald2(void *arg)
191{
192 journal_t *journal = arg;
193 transaction_t *transaction;
194
195 /*
196 * Set up an interval timer which can be used to trigger a commit wakeup
197 * after the commit interval expires
198 */
199 setup_timer(&journal->j_commit_timer, commit_timeout,
200 (unsigned long)current);
201
202 set_freezable();
203
204 /* Record that the journal thread is running */
205 journal->j_task = current;
206 wake_up(&journal->j_wait_done_commit);
207
208 /*
209 * And now, wait forever for commit wakeup events.
210 */
211 write_lock(&journal->j_state_lock);
212
213loop:
214 if (journal->j_flags & JBD2_UNMOUNT)
215 goto end_loop;
216
217 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
218 journal->j_commit_sequence, journal->j_commit_request);
219
220 if (journal->j_commit_sequence != journal->j_commit_request) {
221 jbd_debug(1, "OK, requests differ\n");
222 write_unlock(&journal->j_state_lock);
223 del_timer_sync(&journal->j_commit_timer);
224 jbd2_journal_commit_transaction(journal);
225 write_lock(&journal->j_state_lock);
226 goto loop;
227 }
228
229 wake_up(&journal->j_wait_done_commit);
230 if (freezing(current)) {
231 /*
232 * The simpler the better. Flushing journal isn't a
233 * good idea, because that depends on threads that may
234 * be already stopped.
235 */
236 jbd_debug(1, "Now suspending kjournald2\n");
237 write_unlock(&journal->j_state_lock);
238 try_to_freeze();
239 write_lock(&journal->j_state_lock);
240 } else {
241 /*
242 * We assume on resume that commits are already there,
243 * so we don't sleep
244 */
245 DEFINE_WAIT(wait);
246 int should_sleep = 1;
247
248 prepare_to_wait(&journal->j_wait_commit, &wait,
249 TASK_INTERRUPTIBLE);
250 if (journal->j_commit_sequence != journal->j_commit_request)
251 should_sleep = 0;
252 transaction = journal->j_running_transaction;
253 if (transaction && time_after_eq(jiffies,
254 transaction->t_expires))
255 should_sleep = 0;
256 if (journal->j_flags & JBD2_UNMOUNT)
257 should_sleep = 0;
258 if (should_sleep) {
259 write_unlock(&journal->j_state_lock);
260 schedule();
261 write_lock(&journal->j_state_lock);
262 }
263 finish_wait(&journal->j_wait_commit, &wait);
264 }
265
266 jbd_debug(1, "kjournald2 wakes\n");
267
268 /*
269 * Were we woken up by a commit wakeup event?
270 */
271 transaction = journal->j_running_transaction;
272 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
273 journal->j_commit_request = transaction->t_tid;
274 jbd_debug(1, "woke because of timeout\n");
275 }
276 goto loop;
277
278end_loop:
279 write_unlock(&journal->j_state_lock);
280 del_timer_sync(&journal->j_commit_timer);
281 journal->j_task = NULL;
282 wake_up(&journal->j_wait_done_commit);
283 jbd_debug(1, "Journal thread exiting.\n");
284 return 0;
285}
286
287static int jbd2_journal_start_thread(journal_t *journal)
288{
289 struct task_struct *t;
290
291 t = kthread_run(kjournald2, journal, "jbd2/%s",
292 journal->j_devname);
293 if (IS_ERR(t))
294 return PTR_ERR(t);
295
296 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
297 return 0;
298}
299
300static void journal_kill_thread(journal_t *journal)
301{
302 write_lock(&journal->j_state_lock);
303 journal->j_flags |= JBD2_UNMOUNT;
304
305 while (journal->j_task) {
306 write_unlock(&journal->j_state_lock);
307 wake_up(&journal->j_wait_commit);
308 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
309 write_lock(&journal->j_state_lock);
310 }
311 write_unlock(&journal->j_state_lock);
312}
313
314/*
315 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
316 *
317 * Writes a metadata buffer to a given disk block. The actual IO is not
318 * performed but a new buffer_head is constructed which labels the data
319 * to be written with the correct destination disk block.
320 *
321 * Any magic-number escaping which needs to be done will cause a
322 * copy-out here. If the buffer happens to start with the
323 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
324 * magic number is only written to the log for descripter blocks. In
325 * this case, we copy the data and replace the first word with 0, and we
326 * return a result code which indicates that this buffer needs to be
327 * marked as an escaped buffer in the corresponding log descriptor
328 * block. The missing word can then be restored when the block is read
329 * during recovery.
330 *
331 * If the source buffer has already been modified by a new transaction
332 * since we took the last commit snapshot, we use the frozen copy of
333 * that data for IO. If we end up using the existing buffer_head's data
334 * for the write, then we have to make sure nobody modifies it while the
335 * IO is in progress. do_get_write_access() handles this.
336 *
337 * The function returns a pointer to the buffer_head to be used for IO.
338 *
339 *
340 * Return value:
341 * <0: Error
342 * >=0: Finished OK
343 *
344 * On success:
345 * Bit 0 set == escape performed on the data
346 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
347 */
348
349int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
350 struct journal_head *jh_in,
351 struct buffer_head **bh_out,
352 sector_t blocknr)
353{
354 int need_copy_out = 0;
355 int done_copy_out = 0;
356 int do_escape = 0;
357 char *mapped_data;
358 struct buffer_head *new_bh;
359 struct page *new_page;
360 unsigned int new_offset;
361 struct buffer_head *bh_in = jh2bh(jh_in);
362 journal_t *journal = transaction->t_journal;
363
364 /*
365 * The buffer really shouldn't be locked: only the current committing
366 * transaction is allowed to write it, so nobody else is allowed
367 * to do any IO.
368 *
369 * akpm: except if we're journalling data, and write() output is
370 * also part of a shared mapping, and another thread has
371 * decided to launch a writepage() against this buffer.
372 */
373 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
374
375 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
376
377 /* keep subsequent assertions sane */
378 atomic_set(&new_bh->b_count, 1);
379
380 jbd_lock_bh_state(bh_in);
381repeat:
382 /*
383 * If a new transaction has already done a buffer copy-out, then
384 * we use that version of the data for the commit.
385 */
386 if (jh_in->b_frozen_data) {
387 done_copy_out = 1;
388 new_page = virt_to_page(jh_in->b_frozen_data);
389 new_offset = offset_in_page(jh_in->b_frozen_data);
390 } else {
391 new_page = jh2bh(jh_in)->b_page;
392 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
393 }
394
395 mapped_data = kmap_atomic(new_page);
396 /*
397 * Fire data frozen trigger if data already wasn't frozen. Do this
398 * before checking for escaping, as the trigger may modify the magic
399 * offset. If a copy-out happens afterwards, it will have the correct
400 * data in the buffer.
401 */
402 if (!done_copy_out)
403 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
404 jh_in->b_triggers);
405
406 /*
407 * Check for escaping
408 */
409 if (*((__be32 *)(mapped_data + new_offset)) ==
410 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
411 need_copy_out = 1;
412 do_escape = 1;
413 }
414 kunmap_atomic(mapped_data);
415
416 /*
417 * Do we need to do a data copy?
418 */
419 if (need_copy_out && !done_copy_out) {
420 char *tmp;
421
422 jbd_unlock_bh_state(bh_in);
423 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
424 if (!tmp) {
425 brelse(new_bh);
426 return -ENOMEM;
427 }
428 jbd_lock_bh_state(bh_in);
429 if (jh_in->b_frozen_data) {
430 jbd2_free(tmp, bh_in->b_size);
431 goto repeat;
432 }
433
434 jh_in->b_frozen_data = tmp;
435 mapped_data = kmap_atomic(new_page);
436 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
437 kunmap_atomic(mapped_data);
438
439 new_page = virt_to_page(tmp);
440 new_offset = offset_in_page(tmp);
441 done_copy_out = 1;
442
443 /*
444 * This isn't strictly necessary, as we're using frozen
445 * data for the escaping, but it keeps consistency with
446 * b_frozen_data usage.
447 */
448 jh_in->b_frozen_triggers = jh_in->b_triggers;
449 }
450
451 /*
452 * Did we need to do an escaping? Now we've done all the
453 * copying, we can finally do so.
454 */
455 if (do_escape) {
456 mapped_data = kmap_atomic(new_page);
457 *((unsigned int *)(mapped_data + new_offset)) = 0;
458 kunmap_atomic(mapped_data);
459 }
460
461 set_bh_page(new_bh, new_page, new_offset);
462 new_bh->b_size = bh_in->b_size;
463 new_bh->b_bdev = journal->j_dev;
464 new_bh->b_blocknr = blocknr;
465 new_bh->b_private = bh_in;
466 set_buffer_mapped(new_bh);
467 set_buffer_dirty(new_bh);
468
469 *bh_out = new_bh;
470
471 /*
472 * The to-be-written buffer needs to get moved to the io queue,
473 * and the original buffer whose contents we are shadowing or
474 * copying is moved to the transaction's shadow queue.
475 */
476 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
477 spin_lock(&journal->j_list_lock);
478 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
479 spin_unlock(&journal->j_list_lock);
480 set_buffer_shadow(bh_in);
481 jbd_unlock_bh_state(bh_in);
482
483 return do_escape | (done_copy_out << 1);
484}
485
486/*
487 * Allocation code for the journal file. Manage the space left in the
488 * journal, so that we can begin checkpointing when appropriate.
489 */
490
491/*
492 * Called with j_state_lock locked for writing.
493 * Returns true if a transaction commit was started.
494 */
495int __jbd2_log_start_commit(journal_t *journal, tid_t target)
496{
497 /* Return if the txn has already requested to be committed */
498 if (journal->j_commit_request == target)
499 return 0;
500
501 /*
502 * The only transaction we can possibly wait upon is the
503 * currently running transaction (if it exists). Otherwise,
504 * the target tid must be an old one.
505 */
506 if (journal->j_running_transaction &&
507 journal->j_running_transaction->t_tid == target) {
508 /*
509 * We want a new commit: OK, mark the request and wakeup the
510 * commit thread. We do _not_ do the commit ourselves.
511 */
512
513 journal->j_commit_request = target;
514 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
515 journal->j_commit_request,
516 journal->j_commit_sequence);
517 journal->j_running_transaction->t_requested = jiffies;
518 wake_up(&journal->j_wait_commit);
519 return 1;
520 } else if (!tid_geq(journal->j_commit_request, target))
521 /* This should never happen, but if it does, preserve
522 the evidence before kjournald goes into a loop and
523 increments j_commit_sequence beyond all recognition. */
524 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
525 journal->j_commit_request,
526 journal->j_commit_sequence,
527 target, journal->j_running_transaction ?
528 journal->j_running_transaction->t_tid : 0);
529 return 0;
530}
531
532int jbd2_log_start_commit(journal_t *journal, tid_t tid)
533{
534 int ret;
535
536 write_lock(&journal->j_state_lock);
537 ret = __jbd2_log_start_commit(journal, tid);
538 write_unlock(&journal->j_state_lock);
539 return ret;
540}
541
542/*
543 * Force and wait any uncommitted transactions. We can only force the running
544 * transaction if we don't have an active handle, otherwise, we will deadlock.
545 * Returns: <0 in case of error,
546 * 0 if nothing to commit,
547 * 1 if transaction was successfully committed.
548 */
549static int __jbd2_journal_force_commit(journal_t *journal)
550{
551 transaction_t *transaction = NULL;
552 tid_t tid;
553 int need_to_start = 0, ret = 0;
554
555 read_lock(&journal->j_state_lock);
556 if (journal->j_running_transaction && !current->journal_info) {
557 transaction = journal->j_running_transaction;
558 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
559 need_to_start = 1;
560 } else if (journal->j_committing_transaction)
561 transaction = journal->j_committing_transaction;
562
563 if (!transaction) {
564 /* Nothing to commit */
565 read_unlock(&journal->j_state_lock);
566 return 0;
567 }
568 tid = transaction->t_tid;
569 read_unlock(&journal->j_state_lock);
570 if (need_to_start)
571 jbd2_log_start_commit(journal, tid);
572 ret = jbd2_log_wait_commit(journal, tid);
573 if (!ret)
574 ret = 1;
575
576 return ret;
577}
578
579/**
580 * Force and wait upon a commit if the calling process is not within
581 * transaction. This is used for forcing out undo-protected data which contains
582 * bitmaps, when the fs is running out of space.
583 *
584 * @journal: journal to force
585 * Returns true if progress was made.
586 */
587int jbd2_journal_force_commit_nested(journal_t *journal)
588{
589 int ret;
590
591 ret = __jbd2_journal_force_commit(journal);
592 return ret > 0;
593}
594
595/**
596 * int journal_force_commit() - force any uncommitted transactions
597 * @journal: journal to force
598 *
599 * Caller want unconditional commit. We can only force the running transaction
600 * if we don't have an active handle, otherwise, we will deadlock.
601 */
602int jbd2_journal_force_commit(journal_t *journal)
603{
604 int ret;
605
606 J_ASSERT(!current->journal_info);
607 ret = __jbd2_journal_force_commit(journal);
608 if (ret > 0)
609 ret = 0;
610 return ret;
611}
612
613/*
614 * Start a commit of the current running transaction (if any). Returns true
615 * if a transaction is going to be committed (or is currently already
616 * committing), and fills its tid in at *ptid
617 */
618int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
619{
620 int ret = 0;
621
622 write_lock(&journal->j_state_lock);
623 if (journal->j_running_transaction) {
624 tid_t tid = journal->j_running_transaction->t_tid;
625
626 __jbd2_log_start_commit(journal, tid);
627 /* There's a running transaction and we've just made sure
628 * it's commit has been scheduled. */
629 if (ptid)
630 *ptid = tid;
631 ret = 1;
632 } else if (journal->j_committing_transaction) {
633 /*
634 * If commit has been started, then we have to wait for
635 * completion of that transaction.
636 */
637 if (ptid)
638 *ptid = journal->j_committing_transaction->t_tid;
639 ret = 1;
640 }
641 write_unlock(&journal->j_state_lock);
642 return ret;
643}
644
645/*
646 * Return 1 if a given transaction has not yet sent barrier request
647 * connected with a transaction commit. If 0 is returned, transaction
648 * may or may not have sent the barrier. Used to avoid sending barrier
649 * twice in common cases.
650 */
651int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
652{
653 int ret = 0;
654 transaction_t *commit_trans;
655
656 if (!(journal->j_flags & JBD2_BARRIER))
657 return 0;
658 read_lock(&journal->j_state_lock);
659 /* Transaction already committed? */
660 if (tid_geq(journal->j_commit_sequence, tid))
661 goto out;
662 commit_trans = journal->j_committing_transaction;
663 if (!commit_trans || commit_trans->t_tid != tid) {
664 ret = 1;
665 goto out;
666 }
667 /*
668 * Transaction is being committed and we already proceeded to
669 * submitting a flush to fs partition?
670 */
671 if (journal->j_fs_dev != journal->j_dev) {
672 if (!commit_trans->t_need_data_flush ||
673 commit_trans->t_state >= T_COMMIT_DFLUSH)
674 goto out;
675 } else {
676 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
677 goto out;
678 }
679 ret = 1;
680out:
681 read_unlock(&journal->j_state_lock);
682 return ret;
683}
684EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
685
686/*
687 * Wait for a specified commit to complete.
688 * The caller may not hold the journal lock.
689 */
690int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
691{
692 int err = 0;
693
694 jbd2_might_wait_for_commit(journal);
695 read_lock(&journal->j_state_lock);
696#ifdef CONFIG_JBD2_DEBUG
697 if (!tid_geq(journal->j_commit_request, tid)) {
698 printk(KERN_ERR
699 "%s: error: j_commit_request=%d, tid=%d\n",
700 __func__, journal->j_commit_request, tid);
701 }
702#endif
703 while (tid_gt(tid, journal->j_commit_sequence)) {
704 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
705 tid, journal->j_commit_sequence);
706 read_unlock(&journal->j_state_lock);
707 wake_up(&journal->j_wait_commit);
708 wait_event(journal->j_wait_done_commit,
709 !tid_gt(tid, journal->j_commit_sequence));
710 read_lock(&journal->j_state_lock);
711 }
712 read_unlock(&journal->j_state_lock);
713
714 if (unlikely(is_journal_aborted(journal)))
715 err = -EIO;
716 return err;
717}
718
719/*
720 * When this function returns the transaction corresponding to tid
721 * will be completed. If the transaction has currently running, start
722 * committing that transaction before waiting for it to complete. If
723 * the transaction id is stale, it is by definition already completed,
724 * so just return SUCCESS.
725 */
726int jbd2_complete_transaction(journal_t *journal, tid_t tid)
727{
728 int need_to_wait = 1;
729
730 read_lock(&journal->j_state_lock);
731 if (journal->j_running_transaction &&
732 journal->j_running_transaction->t_tid == tid) {
733 if (journal->j_commit_request != tid) {
734 /* transaction not yet started, so request it */
735 read_unlock(&journal->j_state_lock);
736 jbd2_log_start_commit(journal, tid);
737 goto wait_commit;
738 }
739 } else if (!(journal->j_committing_transaction &&
740 journal->j_committing_transaction->t_tid == tid))
741 need_to_wait = 0;
742 read_unlock(&journal->j_state_lock);
743 if (!need_to_wait)
744 return 0;
745wait_commit:
746 return jbd2_log_wait_commit(journal, tid);
747}
748EXPORT_SYMBOL(jbd2_complete_transaction);
749
750/*
751 * Log buffer allocation routines:
752 */
753
754int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
755{
756 unsigned long blocknr;
757
758 write_lock(&journal->j_state_lock);
759 J_ASSERT(journal->j_free > 1);
760
761 blocknr = journal->j_head;
762 journal->j_head++;
763 journal->j_free--;
764 if (journal->j_head == journal->j_last)
765 journal->j_head = journal->j_first;
766 write_unlock(&journal->j_state_lock);
767 return jbd2_journal_bmap(journal, blocknr, retp);
768}
769
770/*
771 * Conversion of logical to physical block numbers for the journal
772 *
773 * On external journals the journal blocks are identity-mapped, so
774 * this is a no-op. If needed, we can use j_blk_offset - everything is
775 * ready.
776 */
777int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
778 unsigned long long *retp)
779{
780 int err = 0;
781 unsigned long long ret;
782
783 if (journal->j_inode) {
784 ret = bmap(journal->j_inode, blocknr);
785 if (ret)
786 *retp = ret;
787 else {
788 printk(KERN_ALERT "%s: journal block not found "
789 "at offset %lu on %s\n",
790 __func__, blocknr, journal->j_devname);
791 err = -EIO;
792 __journal_abort_soft(journal, err);
793 }
794 } else {
795 *retp = blocknr; /* +journal->j_blk_offset */
796 }
797 return err;
798}
799
800/*
801 * We play buffer_head aliasing tricks to write data/metadata blocks to
802 * the journal without copying their contents, but for journal
803 * descriptor blocks we do need to generate bona fide buffers.
804 *
805 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
806 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
807 * But we don't bother doing that, so there will be coherency problems with
808 * mmaps of blockdevs which hold live JBD-controlled filesystems.
809 */
810struct buffer_head *
811jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
812{
813 journal_t *journal = transaction->t_journal;
814 struct buffer_head *bh;
815 unsigned long long blocknr;
816 journal_header_t *header;
817 int err;
818
819 err = jbd2_journal_next_log_block(journal, &blocknr);
820
821 if (err)
822 return NULL;
823
824 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
825 if (!bh)
826 return NULL;
827 lock_buffer(bh);
828 memset(bh->b_data, 0, journal->j_blocksize);
829 header = (journal_header_t *)bh->b_data;
830 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
831 header->h_blocktype = cpu_to_be32(type);
832 header->h_sequence = cpu_to_be32(transaction->t_tid);
833 set_buffer_uptodate(bh);
834 unlock_buffer(bh);
835 BUFFER_TRACE(bh, "return this buffer");
836 return bh;
837}
838
839void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
840{
841 struct jbd2_journal_block_tail *tail;
842 __u32 csum;
843
844 if (!jbd2_journal_has_csum_v2or3(j))
845 return;
846
847 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
848 sizeof(struct jbd2_journal_block_tail));
849 tail->t_checksum = 0;
850 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
851 tail->t_checksum = cpu_to_be32(csum);
852}
853
854/*
855 * Return tid of the oldest transaction in the journal and block in the journal
856 * where the transaction starts.
857 *
858 * If the journal is now empty, return which will be the next transaction ID
859 * we will write and where will that transaction start.
860 *
861 * The return value is 0 if journal tail cannot be pushed any further, 1 if
862 * it can.
863 */
864int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
865 unsigned long *block)
866{
867 transaction_t *transaction;
868 int ret;
869
870 read_lock(&journal->j_state_lock);
871 spin_lock(&journal->j_list_lock);
872 transaction = journal->j_checkpoint_transactions;
873 if (transaction) {
874 *tid = transaction->t_tid;
875 *block = transaction->t_log_start;
876 } else if ((transaction = journal->j_committing_transaction) != NULL) {
877 *tid = transaction->t_tid;
878 *block = transaction->t_log_start;
879 } else if ((transaction = journal->j_running_transaction) != NULL) {
880 *tid = transaction->t_tid;
881 *block = journal->j_head;
882 } else {
883 *tid = journal->j_transaction_sequence;
884 *block = journal->j_head;
885 }
886 ret = tid_gt(*tid, journal->j_tail_sequence);
887 spin_unlock(&journal->j_list_lock);
888 read_unlock(&journal->j_state_lock);
889
890 return ret;
891}
892
893/*
894 * Update information in journal structure and in on disk journal superblock
895 * about log tail. This function does not check whether information passed in
896 * really pushes log tail further. It's responsibility of the caller to make
897 * sure provided log tail information is valid (e.g. by holding
898 * j_checkpoint_mutex all the time between computing log tail and calling this
899 * function as is the case with jbd2_cleanup_journal_tail()).
900 *
901 * Requires j_checkpoint_mutex
902 */
903int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
904{
905 unsigned long freed;
906 int ret;
907
908 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
909
910 /*
911 * We cannot afford for write to remain in drive's caches since as
912 * soon as we update j_tail, next transaction can start reusing journal
913 * space and if we lose sb update during power failure we'd replay
914 * old transaction with possibly newly overwritten data.
915 */
916 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, REQ_FUA);
917 if (ret)
918 goto out;
919
920 write_lock(&journal->j_state_lock);
921 freed = block - journal->j_tail;
922 if (block < journal->j_tail)
923 freed += journal->j_last - journal->j_first;
924
925 trace_jbd2_update_log_tail(journal, tid, block, freed);
926 jbd_debug(1,
927 "Cleaning journal tail from %d to %d (offset %lu), "
928 "freeing %lu\n",
929 journal->j_tail_sequence, tid, block, freed);
930
931 journal->j_free += freed;
932 journal->j_tail_sequence = tid;
933 journal->j_tail = block;
934 write_unlock(&journal->j_state_lock);
935
936out:
937 return ret;
938}
939
940/*
941 * This is a variaon of __jbd2_update_log_tail which checks for validity of
942 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
943 * with other threads updating log tail.
944 */
945void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
946{
947 mutex_lock(&journal->j_checkpoint_mutex);
948 if (tid_gt(tid, journal->j_tail_sequence))
949 __jbd2_update_log_tail(journal, tid, block);
950 mutex_unlock(&journal->j_checkpoint_mutex);
951}
952
953struct jbd2_stats_proc_session {
954 journal_t *journal;
955 struct transaction_stats_s *stats;
956 int start;
957 int max;
958};
959
960static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
961{
962 return *pos ? NULL : SEQ_START_TOKEN;
963}
964
965static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
966{
967 return NULL;
968}
969
970static int jbd2_seq_info_show(struct seq_file *seq, void *v)
971{
972 struct jbd2_stats_proc_session *s = seq->private;
973
974 if (v != SEQ_START_TOKEN)
975 return 0;
976 seq_printf(seq, "%lu transactions (%lu requested), "
977 "each up to %u blocks\n",
978 s->stats->ts_tid, s->stats->ts_requested,
979 s->journal->j_max_transaction_buffers);
980 if (s->stats->ts_tid == 0)
981 return 0;
982 seq_printf(seq, "average: \n %ums waiting for transaction\n",
983 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
984 seq_printf(seq, " %ums request delay\n",
985 (s->stats->ts_requested == 0) ? 0 :
986 jiffies_to_msecs(s->stats->run.rs_request_delay /
987 s->stats->ts_requested));
988 seq_printf(seq, " %ums running transaction\n",
989 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
990 seq_printf(seq, " %ums transaction was being locked\n",
991 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
992 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
993 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
994 seq_printf(seq, " %ums logging transaction\n",
995 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
996 seq_printf(seq, " %lluus average transaction commit time\n",
997 div_u64(s->journal->j_average_commit_time, 1000));
998 seq_printf(seq, " %lu handles per transaction\n",
999 s->stats->run.rs_handle_count / s->stats->ts_tid);
1000 seq_printf(seq, " %lu blocks per transaction\n",
1001 s->stats->run.rs_blocks / s->stats->ts_tid);
1002 seq_printf(seq, " %lu logged blocks per transaction\n",
1003 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1004 return 0;
1005}
1006
1007static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1008{
1009}
1010
1011static const struct seq_operations jbd2_seq_info_ops = {
1012 .start = jbd2_seq_info_start,
1013 .next = jbd2_seq_info_next,
1014 .stop = jbd2_seq_info_stop,
1015 .show = jbd2_seq_info_show,
1016};
1017
1018static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1019{
1020 journal_t *journal = PDE_DATA(inode);
1021 struct jbd2_stats_proc_session *s;
1022 int rc, size;
1023
1024 s = kmalloc(sizeof(*s), GFP_KERNEL);
1025 if (s == NULL)
1026 return -ENOMEM;
1027 size = sizeof(struct transaction_stats_s);
1028 s->stats = kmalloc(size, GFP_KERNEL);
1029 if (s->stats == NULL) {
1030 kfree(s);
1031 return -ENOMEM;
1032 }
1033 spin_lock(&journal->j_history_lock);
1034 memcpy(s->stats, &journal->j_stats, size);
1035 s->journal = journal;
1036 spin_unlock(&journal->j_history_lock);
1037
1038 rc = seq_open(file, &jbd2_seq_info_ops);
1039 if (rc == 0) {
1040 struct seq_file *m = file->private_data;
1041 m->private = s;
1042 } else {
1043 kfree(s->stats);
1044 kfree(s);
1045 }
1046 return rc;
1047
1048}
1049
1050static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1051{
1052 struct seq_file *seq = file->private_data;
1053 struct jbd2_stats_proc_session *s = seq->private;
1054 kfree(s->stats);
1055 kfree(s);
1056 return seq_release(inode, file);
1057}
1058
1059static const struct file_operations jbd2_seq_info_fops = {
1060 .owner = THIS_MODULE,
1061 .open = jbd2_seq_info_open,
1062 .read = seq_read,
1063 .llseek = seq_lseek,
1064 .release = jbd2_seq_info_release,
1065};
1066
1067static struct proc_dir_entry *proc_jbd2_stats;
1068
1069static void jbd2_stats_proc_init(journal_t *journal)
1070{
1071 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1072 if (journal->j_proc_entry) {
1073 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1074 &jbd2_seq_info_fops, journal);
1075 }
1076}
1077
1078static void jbd2_stats_proc_exit(journal_t *journal)
1079{
1080 remove_proc_entry("info", journal->j_proc_entry);
1081 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1082}
1083
1084/*
1085 * Management for journal control blocks: functions to create and
1086 * destroy journal_t structures, and to initialise and read existing
1087 * journal blocks from disk. */
1088
1089/* First: create and setup a journal_t object in memory. We initialise
1090 * very few fields yet: that has to wait until we have created the
1091 * journal structures from from scratch, or loaded them from disk. */
1092
1093static journal_t *journal_init_common(struct block_device *bdev,
1094 struct block_device *fs_dev,
1095 unsigned long long start, int len, int blocksize)
1096{
1097 static struct lock_class_key jbd2_trans_commit_key;
1098 journal_t *journal;
1099 int err;
1100 struct buffer_head *bh;
1101 int n;
1102
1103 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1104 if (!journal)
1105 return NULL;
1106
1107 init_waitqueue_head(&journal->j_wait_transaction_locked);
1108 init_waitqueue_head(&journal->j_wait_done_commit);
1109 init_waitqueue_head(&journal->j_wait_commit);
1110 init_waitqueue_head(&journal->j_wait_updates);
1111 init_waitqueue_head(&journal->j_wait_reserved);
1112 mutex_init(&journal->j_barrier);
1113 mutex_init(&journal->j_checkpoint_mutex);
1114 spin_lock_init(&journal->j_revoke_lock);
1115 spin_lock_init(&journal->j_list_lock);
1116 rwlock_init(&journal->j_state_lock);
1117
1118 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1119 journal->j_min_batch_time = 0;
1120 journal->j_max_batch_time = 15000; /* 15ms */
1121 atomic_set(&journal->j_reserved_credits, 0);
1122
1123 /* The journal is marked for error until we succeed with recovery! */
1124 journal->j_flags = JBD2_ABORT;
1125
1126 /* Set up a default-sized revoke table for the new mount. */
1127 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1128 if (err)
1129 goto err_cleanup;
1130
1131 spin_lock_init(&journal->j_history_lock);
1132
1133 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1134 &jbd2_trans_commit_key, 0);
1135
1136 /* journal descriptor can store up to n blocks -bzzz */
1137 journal->j_blocksize = blocksize;
1138 journal->j_dev = bdev;
1139 journal->j_fs_dev = fs_dev;
1140 journal->j_blk_offset = start;
1141 journal->j_maxlen = len;
1142 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1143 journal->j_wbufsize = n;
1144 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1145 GFP_KERNEL);
1146 if (!journal->j_wbuf)
1147 goto err_cleanup;
1148
1149 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1150 if (!bh) {
1151 pr_err("%s: Cannot get buffer for journal superblock\n",
1152 __func__);
1153 goto err_cleanup;
1154 }
1155 journal->j_sb_buffer = bh;
1156 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1157
1158 return journal;
1159
1160err_cleanup:
1161 kfree(journal->j_wbuf);
1162 jbd2_journal_destroy_revoke(journal);
1163 kfree(journal);
1164 return NULL;
1165}
1166
1167/* jbd2_journal_init_dev and jbd2_journal_init_inode:
1168 *
1169 * Create a journal structure assigned some fixed set of disk blocks to
1170 * the journal. We don't actually touch those disk blocks yet, but we
1171 * need to set up all of the mapping information to tell the journaling
1172 * system where the journal blocks are.
1173 *
1174 */
1175
1176/**
1177 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1178 * @bdev: Block device on which to create the journal
1179 * @fs_dev: Device which hold journalled filesystem for this journal.
1180 * @start: Block nr Start of journal.
1181 * @len: Length of the journal in blocks.
1182 * @blocksize: blocksize of journalling device
1183 *
1184 * Returns: a newly created journal_t *
1185 *
1186 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1187 * range of blocks on an arbitrary block device.
1188 *
1189 */
1190journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1191 struct block_device *fs_dev,
1192 unsigned long long start, int len, int blocksize)
1193{
1194 journal_t *journal;
1195
1196 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1197 if (!journal)
1198 return NULL;
1199
1200 bdevname(journal->j_dev, journal->j_devname);
1201 strreplace(journal->j_devname, '/', '!');
1202 jbd2_stats_proc_init(journal);
1203
1204 return journal;
1205}
1206
1207/**
1208 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1209 * @inode: An inode to create the journal in
1210 *
1211 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1212 * the journal. The inode must exist already, must support bmap() and
1213 * must have all data blocks preallocated.
1214 */
1215journal_t *jbd2_journal_init_inode(struct inode *inode)
1216{
1217 journal_t *journal;
1218 char *p;
1219 unsigned long long blocknr;
1220
1221 blocknr = bmap(inode, 0);
1222 if (!blocknr) {
1223 pr_err("%s: Cannot locate journal superblock\n",
1224 __func__);
1225 return NULL;
1226 }
1227
1228 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1229 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1230 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1231
1232 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1233 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1234 inode->i_sb->s_blocksize);
1235 if (!journal)
1236 return NULL;
1237
1238 journal->j_inode = inode;
1239 bdevname(journal->j_dev, journal->j_devname);
1240 p = strreplace(journal->j_devname, '/', '!');
1241 sprintf(p, "-%lu", journal->j_inode->i_ino);
1242 jbd2_stats_proc_init(journal);
1243
1244 return journal;
1245}
1246
1247/*
1248 * If the journal init or create aborts, we need to mark the journal
1249 * superblock as being NULL to prevent the journal destroy from writing
1250 * back a bogus superblock.
1251 */
1252static void journal_fail_superblock (journal_t *journal)
1253{
1254 struct buffer_head *bh = journal->j_sb_buffer;
1255 brelse(bh);
1256 journal->j_sb_buffer = NULL;
1257}
1258
1259/*
1260 * Given a journal_t structure, initialise the various fields for
1261 * startup of a new journaling session. We use this both when creating
1262 * a journal, and after recovering an old journal to reset it for
1263 * subsequent use.
1264 */
1265
1266static int journal_reset(journal_t *journal)
1267{
1268 journal_superblock_t *sb = journal->j_superblock;
1269 unsigned long long first, last;
1270
1271 first = be32_to_cpu(sb->s_first);
1272 last = be32_to_cpu(sb->s_maxlen);
1273 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1274 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1275 first, last);
1276 journal_fail_superblock(journal);
1277 return -EINVAL;
1278 }
1279
1280 journal->j_first = first;
1281 journal->j_last = last;
1282
1283 journal->j_head = first;
1284 journal->j_tail = first;
1285 journal->j_free = last - first;
1286
1287 journal->j_tail_sequence = journal->j_transaction_sequence;
1288 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1289 journal->j_commit_request = journal->j_commit_sequence;
1290
1291 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1292
1293 /*
1294 * As a special case, if the on-disk copy is already marked as needing
1295 * no recovery (s_start == 0), then we can safely defer the superblock
1296 * update until the next commit by setting JBD2_FLUSHED. This avoids
1297 * attempting a write to a potential-readonly device.
1298 */
1299 if (sb->s_start == 0) {
1300 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1301 "(start %ld, seq %d, errno %d)\n",
1302 journal->j_tail, journal->j_tail_sequence,
1303 journal->j_errno);
1304 journal->j_flags |= JBD2_FLUSHED;
1305 } else {
1306 /* Lock here to make assertions happy... */
1307 mutex_lock(&journal->j_checkpoint_mutex);
1308 /*
1309 * Update log tail information. We use REQ_FUA since new
1310 * transaction will start reusing journal space and so we
1311 * must make sure information about current log tail is on
1312 * disk before that.
1313 */
1314 jbd2_journal_update_sb_log_tail(journal,
1315 journal->j_tail_sequence,
1316 journal->j_tail,
1317 REQ_FUA);
1318 mutex_unlock(&journal->j_checkpoint_mutex);
1319 }
1320 return jbd2_journal_start_thread(journal);
1321}
1322
1323static int jbd2_write_superblock(journal_t *journal, int write_flags)
1324{
1325 struct buffer_head *bh = journal->j_sb_buffer;
1326 journal_superblock_t *sb = journal->j_superblock;
1327 int ret;
1328
1329 trace_jbd2_write_superblock(journal, write_flags);
1330 if (!(journal->j_flags & JBD2_BARRIER))
1331 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1332 lock_buffer(bh);
1333 if (buffer_write_io_error(bh)) {
1334 /*
1335 * Oh, dear. A previous attempt to write the journal
1336 * superblock failed. This could happen because the
1337 * USB device was yanked out. Or it could happen to
1338 * be a transient write error and maybe the block will
1339 * be remapped. Nothing we can do but to retry the
1340 * write and hope for the best.
1341 */
1342 printk(KERN_ERR "JBD2: previous I/O error detected "
1343 "for journal superblock update for %s.\n",
1344 journal->j_devname);
1345 clear_buffer_write_io_error(bh);
1346 set_buffer_uptodate(bh);
1347 }
1348 jbd2_superblock_csum_set(journal, sb);
1349 get_bh(bh);
1350 bh->b_end_io = end_buffer_write_sync;
1351 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1352 wait_on_buffer(bh);
1353 if (buffer_write_io_error(bh)) {
1354 clear_buffer_write_io_error(bh);
1355 set_buffer_uptodate(bh);
1356 ret = -EIO;
1357 }
1358 if (ret) {
1359 printk(KERN_ERR "JBD2: Error %d detected when updating "
1360 "journal superblock for %s.\n", ret,
1361 journal->j_devname);
1362 jbd2_journal_abort(journal, ret);
1363 }
1364
1365 return ret;
1366}
1367
1368/**
1369 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1370 * @journal: The journal to update.
1371 * @tail_tid: TID of the new transaction at the tail of the log
1372 * @tail_block: The first block of the transaction at the tail of the log
1373 * @write_op: With which operation should we write the journal sb
1374 *
1375 * Update a journal's superblock information about log tail and write it to
1376 * disk, waiting for the IO to complete.
1377 */
1378int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1379 unsigned long tail_block, int write_op)
1380{
1381 journal_superblock_t *sb = journal->j_superblock;
1382 int ret;
1383
1384 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1385 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1386 tail_block, tail_tid);
1387
1388 sb->s_sequence = cpu_to_be32(tail_tid);
1389 sb->s_start = cpu_to_be32(tail_block);
1390
1391 ret = jbd2_write_superblock(journal, write_op);
1392 if (ret)
1393 goto out;
1394
1395 /* Log is no longer empty */
1396 write_lock(&journal->j_state_lock);
1397 WARN_ON(!sb->s_sequence);
1398 journal->j_flags &= ~JBD2_FLUSHED;
1399 write_unlock(&journal->j_state_lock);
1400
1401out:
1402 return ret;
1403}
1404
1405/**
1406 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1407 * @journal: The journal to update.
1408 * @write_op: With which operation should we write the journal sb
1409 *
1410 * Update a journal's dynamic superblock fields to show that journal is empty.
1411 * Write updated superblock to disk waiting for IO to complete.
1412 */
1413static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1414{
1415 journal_superblock_t *sb = journal->j_superblock;
1416
1417 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1418 read_lock(&journal->j_state_lock);
1419 /* Is it already empty? */
1420 if (sb->s_start == 0) {
1421 read_unlock(&journal->j_state_lock);
1422 return;
1423 }
1424 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1425 journal->j_tail_sequence);
1426
1427 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1428 sb->s_start = cpu_to_be32(0);
1429 read_unlock(&journal->j_state_lock);
1430
1431 jbd2_write_superblock(journal, write_op);
1432
1433 /* Log is no longer empty */
1434 write_lock(&journal->j_state_lock);
1435 journal->j_flags |= JBD2_FLUSHED;
1436 write_unlock(&journal->j_state_lock);
1437}
1438
1439
1440/**
1441 * jbd2_journal_update_sb_errno() - Update error in the journal.
1442 * @journal: The journal to update.
1443 *
1444 * Update a journal's errno. Write updated superblock to disk waiting for IO
1445 * to complete.
1446 */
1447void jbd2_journal_update_sb_errno(journal_t *journal)
1448{
1449 journal_superblock_t *sb = journal->j_superblock;
1450
1451 read_lock(&journal->j_state_lock);
1452 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1453 journal->j_errno);
1454 sb->s_errno = cpu_to_be32(journal->j_errno);
1455 read_unlock(&journal->j_state_lock);
1456
1457 jbd2_write_superblock(journal, REQ_FUA);
1458}
1459EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1460
1461/*
1462 * Read the superblock for a given journal, performing initial
1463 * validation of the format.
1464 */
1465static int journal_get_superblock(journal_t *journal)
1466{
1467 struct buffer_head *bh;
1468 journal_superblock_t *sb;
1469 int err = -EIO;
1470
1471 bh = journal->j_sb_buffer;
1472
1473 J_ASSERT(bh != NULL);
1474 if (!buffer_uptodate(bh)) {
1475 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1476 wait_on_buffer(bh);
1477 if (!buffer_uptodate(bh)) {
1478 printk(KERN_ERR
1479 "JBD2: IO error reading journal superblock\n");
1480 goto out;
1481 }
1482 }
1483
1484 if (buffer_verified(bh))
1485 return 0;
1486
1487 sb = journal->j_superblock;
1488
1489 err = -EINVAL;
1490
1491 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1492 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1493 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1494 goto out;
1495 }
1496
1497 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1498 case JBD2_SUPERBLOCK_V1:
1499 journal->j_format_version = 1;
1500 break;
1501 case JBD2_SUPERBLOCK_V2:
1502 journal->j_format_version = 2;
1503 break;
1504 default:
1505 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1506 goto out;
1507 }
1508
1509 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1510 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1511 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1512 printk(KERN_WARNING "JBD2: journal file too short\n");
1513 goto out;
1514 }
1515
1516 if (be32_to_cpu(sb->s_first) == 0 ||
1517 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1518 printk(KERN_WARNING
1519 "JBD2: Invalid start block of journal: %u\n",
1520 be32_to_cpu(sb->s_first));
1521 goto out;
1522 }
1523
1524 if (jbd2_has_feature_csum2(journal) &&
1525 jbd2_has_feature_csum3(journal)) {
1526 /* Can't have checksum v2 and v3 at the same time! */
1527 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1528 "at the same time!\n");
1529 goto out;
1530 }
1531
1532 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1533 jbd2_has_feature_checksum(journal)) {
1534 /* Can't have checksum v1 and v2 on at the same time! */
1535 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1536 "at the same time!\n");
1537 goto out;
1538 }
1539
1540 if (!jbd2_verify_csum_type(journal, sb)) {
1541 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1542 goto out;
1543 }
1544
1545 /* Load the checksum driver */
1546 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1547 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1548 if (IS_ERR(journal->j_chksum_driver)) {
1549 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1550 err = PTR_ERR(journal->j_chksum_driver);
1551 journal->j_chksum_driver = NULL;
1552 goto out;
1553 }
1554 }
1555
1556 /* Check superblock checksum */
1557 if (!jbd2_superblock_csum_verify(journal, sb)) {
1558 printk(KERN_ERR "JBD2: journal checksum error\n");
1559 err = -EFSBADCRC;
1560 goto out;
1561 }
1562
1563 /* Precompute checksum seed for all metadata */
1564 if (jbd2_journal_has_csum_v2or3(journal))
1565 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1566 sizeof(sb->s_uuid));
1567
1568 set_buffer_verified(bh);
1569
1570 return 0;
1571
1572out:
1573 journal_fail_superblock(journal);
1574 return err;
1575}
1576
1577/*
1578 * Load the on-disk journal superblock and read the key fields into the
1579 * journal_t.
1580 */
1581
1582static int load_superblock(journal_t *journal)
1583{
1584 int err;
1585 journal_superblock_t *sb;
1586
1587 err = journal_get_superblock(journal);
1588 if (err)
1589 return err;
1590
1591 sb = journal->j_superblock;
1592
1593 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1594 journal->j_tail = be32_to_cpu(sb->s_start);
1595 journal->j_first = be32_to_cpu(sb->s_first);
1596 journal->j_last = be32_to_cpu(sb->s_maxlen);
1597 journal->j_errno = be32_to_cpu(sb->s_errno);
1598
1599 return 0;
1600}
1601
1602
1603/**
1604 * int jbd2_journal_load() - Read journal from disk.
1605 * @journal: Journal to act on.
1606 *
1607 * Given a journal_t structure which tells us which disk blocks contain
1608 * a journal, read the journal from disk to initialise the in-memory
1609 * structures.
1610 */
1611int jbd2_journal_load(journal_t *journal)
1612{
1613 int err;
1614 journal_superblock_t *sb;
1615
1616 err = load_superblock(journal);
1617 if (err)
1618 return err;
1619
1620 sb = journal->j_superblock;
1621 /* If this is a V2 superblock, then we have to check the
1622 * features flags on it. */
1623
1624 if (journal->j_format_version >= 2) {
1625 if ((sb->s_feature_ro_compat &
1626 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1627 (sb->s_feature_incompat &
1628 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1629 printk(KERN_WARNING
1630 "JBD2: Unrecognised features on journal\n");
1631 return -EINVAL;
1632 }
1633 }
1634
1635 /*
1636 * Create a slab for this blocksize
1637 */
1638 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1639 if (err)
1640 return err;
1641
1642 /* Let the recovery code check whether it needs to recover any
1643 * data from the journal. */
1644 if (jbd2_journal_recover(journal))
1645 goto recovery_error;
1646
1647 if (journal->j_failed_commit) {
1648 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1649 "is corrupt.\n", journal->j_failed_commit,
1650 journal->j_devname);
1651 return -EFSCORRUPTED;
1652 }
1653
1654 /* OK, we've finished with the dynamic journal bits:
1655 * reinitialise the dynamic contents of the superblock in memory
1656 * and reset them on disk. */
1657 if (journal_reset(journal))
1658 goto recovery_error;
1659
1660 journal->j_flags &= ~JBD2_ABORT;
1661 journal->j_flags |= JBD2_LOADED;
1662 return 0;
1663
1664recovery_error:
1665 printk(KERN_WARNING "JBD2: recovery failed\n");
1666 return -EIO;
1667}
1668
1669/**
1670 * void jbd2_journal_destroy() - Release a journal_t structure.
1671 * @journal: Journal to act on.
1672 *
1673 * Release a journal_t structure once it is no longer in use by the
1674 * journaled object.
1675 * Return <0 if we couldn't clean up the journal.
1676 */
1677int jbd2_journal_destroy(journal_t *journal)
1678{
1679 int err = 0;
1680
1681 /* Wait for the commit thread to wake up and die. */
1682 journal_kill_thread(journal);
1683
1684 /* Force a final log commit */
1685 if (journal->j_running_transaction)
1686 jbd2_journal_commit_transaction(journal);
1687
1688 /* Force any old transactions to disk */
1689
1690 /* Totally anal locking here... */
1691 spin_lock(&journal->j_list_lock);
1692 while (journal->j_checkpoint_transactions != NULL) {
1693 spin_unlock(&journal->j_list_lock);
1694 mutex_lock(&journal->j_checkpoint_mutex);
1695 err = jbd2_log_do_checkpoint(journal);
1696 mutex_unlock(&journal->j_checkpoint_mutex);
1697 /*
1698 * If checkpointing failed, just free the buffers to avoid
1699 * looping forever
1700 */
1701 if (err) {
1702 jbd2_journal_destroy_checkpoint(journal);
1703 spin_lock(&journal->j_list_lock);
1704 break;
1705 }
1706 spin_lock(&journal->j_list_lock);
1707 }
1708
1709 J_ASSERT(journal->j_running_transaction == NULL);
1710 J_ASSERT(journal->j_committing_transaction == NULL);
1711 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1712 spin_unlock(&journal->j_list_lock);
1713
1714 if (journal->j_sb_buffer) {
1715 if (!is_journal_aborted(journal)) {
1716 mutex_lock(&journal->j_checkpoint_mutex);
1717
1718 write_lock(&journal->j_state_lock);
1719 journal->j_tail_sequence =
1720 ++journal->j_transaction_sequence;
1721 write_unlock(&journal->j_state_lock);
1722
1723 jbd2_mark_journal_empty(journal,
1724 REQ_PREFLUSH | REQ_FUA);
1725 mutex_unlock(&journal->j_checkpoint_mutex);
1726 } else
1727 err = -EIO;
1728 brelse(journal->j_sb_buffer);
1729 }
1730
1731 if (journal->j_proc_entry)
1732 jbd2_stats_proc_exit(journal);
1733 iput(journal->j_inode);
1734 if (journal->j_revoke)
1735 jbd2_journal_destroy_revoke(journal);
1736 if (journal->j_chksum_driver)
1737 crypto_free_shash(journal->j_chksum_driver);
1738 kfree(journal->j_wbuf);
1739 kfree(journal);
1740
1741 return err;
1742}
1743
1744
1745/**
1746 *int jbd2_journal_check_used_features () - Check if features specified are used.
1747 * @journal: Journal to check.
1748 * @compat: bitmask of compatible features
1749 * @ro: bitmask of features that force read-only mount
1750 * @incompat: bitmask of incompatible features
1751 *
1752 * Check whether the journal uses all of a given set of
1753 * features. Return true (non-zero) if it does.
1754 **/
1755
1756int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1757 unsigned long ro, unsigned long incompat)
1758{
1759 journal_superblock_t *sb;
1760
1761 if (!compat && !ro && !incompat)
1762 return 1;
1763 /* Load journal superblock if it is not loaded yet. */
1764 if (journal->j_format_version == 0 &&
1765 journal_get_superblock(journal) != 0)
1766 return 0;
1767 if (journal->j_format_version == 1)
1768 return 0;
1769
1770 sb = journal->j_superblock;
1771
1772 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1773 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1774 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1775 return 1;
1776
1777 return 0;
1778}
1779
1780/**
1781 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1782 * @journal: Journal to check.
1783 * @compat: bitmask of compatible features
1784 * @ro: bitmask of features that force read-only mount
1785 * @incompat: bitmask of incompatible features
1786 *
1787 * Check whether the journaling code supports the use of
1788 * all of a given set of features on this journal. Return true
1789 * (non-zero) if it can. */
1790
1791int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1792 unsigned long ro, unsigned long incompat)
1793{
1794 if (!compat && !ro && !incompat)
1795 return 1;
1796
1797 /* We can support any known requested features iff the
1798 * superblock is in version 2. Otherwise we fail to support any
1799 * extended sb features. */
1800
1801 if (journal->j_format_version != 2)
1802 return 0;
1803
1804 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1805 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1806 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1807 return 1;
1808
1809 return 0;
1810}
1811
1812/**
1813 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1814 * @journal: Journal to act on.
1815 * @compat: bitmask of compatible features
1816 * @ro: bitmask of features that force read-only mount
1817 * @incompat: bitmask of incompatible features
1818 *
1819 * Mark a given journal feature as present on the
1820 * superblock. Returns true if the requested features could be set.
1821 *
1822 */
1823
1824int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1825 unsigned long ro, unsigned long incompat)
1826{
1827#define INCOMPAT_FEATURE_ON(f) \
1828 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1829#define COMPAT_FEATURE_ON(f) \
1830 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1831 journal_superblock_t *sb;
1832
1833 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1834 return 1;
1835
1836 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1837 return 0;
1838
1839 /* If enabling v2 checksums, turn on v3 instead */
1840 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1841 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1842 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1843 }
1844
1845 /* Asking for checksumming v3 and v1? Only give them v3. */
1846 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1847 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1848 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1849
1850 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1851 compat, ro, incompat);
1852
1853 sb = journal->j_superblock;
1854
1855 /* If enabling v3 checksums, update superblock */
1856 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1857 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1858 sb->s_feature_compat &=
1859 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1860
1861 /* Load the checksum driver */
1862 if (journal->j_chksum_driver == NULL) {
1863 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1864 0, 0);
1865 if (IS_ERR(journal->j_chksum_driver)) {
1866 printk(KERN_ERR "JBD2: Cannot load crc32c "
1867 "driver.\n");
1868 journal->j_chksum_driver = NULL;
1869 return 0;
1870 }
1871
1872 /* Precompute checksum seed for all metadata */
1873 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1874 sb->s_uuid,
1875 sizeof(sb->s_uuid));
1876 }
1877 }
1878
1879 /* If enabling v1 checksums, downgrade superblock */
1880 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1881 sb->s_feature_incompat &=
1882 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1883 JBD2_FEATURE_INCOMPAT_CSUM_V3);
1884
1885 sb->s_feature_compat |= cpu_to_be32(compat);
1886 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1887 sb->s_feature_incompat |= cpu_to_be32(incompat);
1888
1889 return 1;
1890#undef COMPAT_FEATURE_ON
1891#undef INCOMPAT_FEATURE_ON
1892}
1893
1894/*
1895 * jbd2_journal_clear_features () - Clear a given journal feature in the
1896 * superblock
1897 * @journal: Journal to act on.
1898 * @compat: bitmask of compatible features
1899 * @ro: bitmask of features that force read-only mount
1900 * @incompat: bitmask of incompatible features
1901 *
1902 * Clear a given journal feature as present on the
1903 * superblock.
1904 */
1905void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1906 unsigned long ro, unsigned long incompat)
1907{
1908 journal_superblock_t *sb;
1909
1910 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1911 compat, ro, incompat);
1912
1913 sb = journal->j_superblock;
1914
1915 sb->s_feature_compat &= ~cpu_to_be32(compat);
1916 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1917 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1918}
1919EXPORT_SYMBOL(jbd2_journal_clear_features);
1920
1921/**
1922 * int jbd2_journal_flush () - Flush journal
1923 * @journal: Journal to act on.
1924 *
1925 * Flush all data for a given journal to disk and empty the journal.
1926 * Filesystems can use this when remounting readonly to ensure that
1927 * recovery does not need to happen on remount.
1928 */
1929
1930int jbd2_journal_flush(journal_t *journal)
1931{
1932 int err = 0;
1933 transaction_t *transaction = NULL;
1934
1935 write_lock(&journal->j_state_lock);
1936
1937 /* Force everything buffered to the log... */
1938 if (journal->j_running_transaction) {
1939 transaction = journal->j_running_transaction;
1940 __jbd2_log_start_commit(journal, transaction->t_tid);
1941 } else if (journal->j_committing_transaction)
1942 transaction = journal->j_committing_transaction;
1943
1944 /* Wait for the log commit to complete... */
1945 if (transaction) {
1946 tid_t tid = transaction->t_tid;
1947
1948 write_unlock(&journal->j_state_lock);
1949 jbd2_log_wait_commit(journal, tid);
1950 } else {
1951 write_unlock(&journal->j_state_lock);
1952 }
1953
1954 /* ...and flush everything in the log out to disk. */
1955 spin_lock(&journal->j_list_lock);
1956 while (!err && journal->j_checkpoint_transactions != NULL) {
1957 spin_unlock(&journal->j_list_lock);
1958 mutex_lock(&journal->j_checkpoint_mutex);
1959 err = jbd2_log_do_checkpoint(journal);
1960 mutex_unlock(&journal->j_checkpoint_mutex);
1961 spin_lock(&journal->j_list_lock);
1962 }
1963 spin_unlock(&journal->j_list_lock);
1964
1965 if (is_journal_aborted(journal))
1966 return -EIO;
1967
1968 mutex_lock(&journal->j_checkpoint_mutex);
1969 if (!err) {
1970 err = jbd2_cleanup_journal_tail(journal);
1971 if (err < 0) {
1972 mutex_unlock(&journal->j_checkpoint_mutex);
1973 goto out;
1974 }
1975 err = 0;
1976 }
1977
1978 /* Finally, mark the journal as really needing no recovery.
1979 * This sets s_start==0 in the underlying superblock, which is
1980 * the magic code for a fully-recovered superblock. Any future
1981 * commits of data to the journal will restore the current
1982 * s_start value. */
1983 jbd2_mark_journal_empty(journal, REQ_FUA);
1984 mutex_unlock(&journal->j_checkpoint_mutex);
1985 write_lock(&journal->j_state_lock);
1986 J_ASSERT(!journal->j_running_transaction);
1987 J_ASSERT(!journal->j_committing_transaction);
1988 J_ASSERT(!journal->j_checkpoint_transactions);
1989 J_ASSERT(journal->j_head == journal->j_tail);
1990 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1991 write_unlock(&journal->j_state_lock);
1992out:
1993 return err;
1994}
1995
1996/**
1997 * int jbd2_journal_wipe() - Wipe journal contents
1998 * @journal: Journal to act on.
1999 * @write: flag (see below)
2000 *
2001 * Wipe out all of the contents of a journal, safely. This will produce
2002 * a warning if the journal contains any valid recovery information.
2003 * Must be called between journal_init_*() and jbd2_journal_load().
2004 *
2005 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2006 * we merely suppress recovery.
2007 */
2008
2009int jbd2_journal_wipe(journal_t *journal, int write)
2010{
2011 int err = 0;
2012
2013 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2014
2015 err = load_superblock(journal);
2016 if (err)
2017 return err;
2018
2019 if (!journal->j_tail)
2020 goto no_recovery;
2021
2022 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2023 write ? "Clearing" : "Ignoring");
2024
2025 err = jbd2_journal_skip_recovery(journal);
2026 if (write) {
2027 /* Lock to make assertions happy... */
2028 mutex_lock(&journal->j_checkpoint_mutex);
2029 jbd2_mark_journal_empty(journal, REQ_FUA);
2030 mutex_unlock(&journal->j_checkpoint_mutex);
2031 }
2032
2033 no_recovery:
2034 return err;
2035}
2036
2037/*
2038 * Journal abort has very specific semantics, which we describe
2039 * for journal abort.
2040 *
2041 * Two internal functions, which provide abort to the jbd layer
2042 * itself are here.
2043 */
2044
2045/*
2046 * Quick version for internal journal use (doesn't lock the journal).
2047 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2048 * and don't attempt to make any other journal updates.
2049 */
2050void __jbd2_journal_abort_hard(journal_t *journal)
2051{
2052 transaction_t *transaction;
2053
2054 if (journal->j_flags & JBD2_ABORT)
2055 return;
2056
2057 printk(KERN_ERR "Aborting journal on device %s.\n",
2058 journal->j_devname);
2059
2060 write_lock(&journal->j_state_lock);
2061 journal->j_flags |= JBD2_ABORT;
2062 transaction = journal->j_running_transaction;
2063 if (transaction)
2064 __jbd2_log_start_commit(journal, transaction->t_tid);
2065 write_unlock(&journal->j_state_lock);
2066}
2067
2068/* Soft abort: record the abort error status in the journal superblock,
2069 * but don't do any other IO. */
2070static void __journal_abort_soft (journal_t *journal, int errno)
2071{
2072 if (journal->j_flags & JBD2_ABORT)
2073 return;
2074
2075 if (!journal->j_errno)
2076 journal->j_errno = errno;
2077
2078 __jbd2_journal_abort_hard(journal);
2079
2080 if (errno) {
2081 jbd2_journal_update_sb_errno(journal);
2082 write_lock(&journal->j_state_lock);
2083 journal->j_flags |= JBD2_REC_ERR;
2084 write_unlock(&journal->j_state_lock);
2085 }
2086}
2087
2088/**
2089 * void jbd2_journal_abort () - Shutdown the journal immediately.
2090 * @journal: the journal to shutdown.
2091 * @errno: an error number to record in the journal indicating
2092 * the reason for the shutdown.
2093 *
2094 * Perform a complete, immediate shutdown of the ENTIRE
2095 * journal (not of a single transaction). This operation cannot be
2096 * undone without closing and reopening the journal.
2097 *
2098 * The jbd2_journal_abort function is intended to support higher level error
2099 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2100 * mode.
2101 *
2102 * Journal abort has very specific semantics. Any existing dirty,
2103 * unjournaled buffers in the main filesystem will still be written to
2104 * disk by bdflush, but the journaling mechanism will be suspended
2105 * immediately and no further transaction commits will be honoured.
2106 *
2107 * Any dirty, journaled buffers will be written back to disk without
2108 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2109 * filesystem, but we _do_ attempt to leave as much data as possible
2110 * behind for fsck to use for cleanup.
2111 *
2112 * Any attempt to get a new transaction handle on a journal which is in
2113 * ABORT state will just result in an -EROFS error return. A
2114 * jbd2_journal_stop on an existing handle will return -EIO if we have
2115 * entered abort state during the update.
2116 *
2117 * Recursive transactions are not disturbed by journal abort until the
2118 * final jbd2_journal_stop, which will receive the -EIO error.
2119 *
2120 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2121 * which will be recorded (if possible) in the journal superblock. This
2122 * allows a client to record failure conditions in the middle of a
2123 * transaction without having to complete the transaction to record the
2124 * failure to disk. ext3_error, for example, now uses this
2125 * functionality.
2126 *
2127 * Errors which originate from within the journaling layer will NOT
2128 * supply an errno; a null errno implies that absolutely no further
2129 * writes are done to the journal (unless there are any already in
2130 * progress).
2131 *
2132 */
2133
2134void jbd2_journal_abort(journal_t *journal, int errno)
2135{
2136 __journal_abort_soft(journal, errno);
2137}
2138
2139/**
2140 * int jbd2_journal_errno () - returns the journal's error state.
2141 * @journal: journal to examine.
2142 *
2143 * This is the errno number set with jbd2_journal_abort(), the last
2144 * time the journal was mounted - if the journal was stopped
2145 * without calling abort this will be 0.
2146 *
2147 * If the journal has been aborted on this mount time -EROFS will
2148 * be returned.
2149 */
2150int jbd2_journal_errno(journal_t *journal)
2151{
2152 int err;
2153
2154 read_lock(&journal->j_state_lock);
2155 if (journal->j_flags & JBD2_ABORT)
2156 err = -EROFS;
2157 else
2158 err = journal->j_errno;
2159 read_unlock(&journal->j_state_lock);
2160 return err;
2161}
2162
2163/**
2164 * int jbd2_journal_clear_err () - clears the journal's error state
2165 * @journal: journal to act on.
2166 *
2167 * An error must be cleared or acked to take a FS out of readonly
2168 * mode.
2169 */
2170int jbd2_journal_clear_err(journal_t *journal)
2171{
2172 int err = 0;
2173
2174 write_lock(&journal->j_state_lock);
2175 if (journal->j_flags & JBD2_ABORT)
2176 err = -EROFS;
2177 else
2178 journal->j_errno = 0;
2179 write_unlock(&journal->j_state_lock);
2180 return err;
2181}
2182
2183/**
2184 * void jbd2_journal_ack_err() - Ack journal err.
2185 * @journal: journal to act on.
2186 *
2187 * An error must be cleared or acked to take a FS out of readonly
2188 * mode.
2189 */
2190void jbd2_journal_ack_err(journal_t *journal)
2191{
2192 write_lock(&journal->j_state_lock);
2193 if (journal->j_errno)
2194 journal->j_flags |= JBD2_ACK_ERR;
2195 write_unlock(&journal->j_state_lock);
2196}
2197
2198int jbd2_journal_blocks_per_page(struct inode *inode)
2199{
2200 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2201}
2202
2203/*
2204 * helper functions to deal with 32 or 64bit block numbers.
2205 */
2206size_t journal_tag_bytes(journal_t *journal)
2207{
2208 size_t sz;
2209
2210 if (jbd2_has_feature_csum3(journal))
2211 return sizeof(journal_block_tag3_t);
2212
2213 sz = sizeof(journal_block_tag_t);
2214
2215 if (jbd2_has_feature_csum2(journal))
2216 sz += sizeof(__u16);
2217
2218 if (jbd2_has_feature_64bit(journal))
2219 return sz;
2220 else
2221 return sz - sizeof(__u32);
2222}
2223
2224/*
2225 * JBD memory management
2226 *
2227 * These functions are used to allocate block-sized chunks of memory
2228 * used for making copies of buffer_head data. Very often it will be
2229 * page-sized chunks of data, but sometimes it will be in
2230 * sub-page-size chunks. (For example, 16k pages on Power systems
2231 * with a 4k block file system.) For blocks smaller than a page, we
2232 * use a SLAB allocator. There are slab caches for each block size,
2233 * which are allocated at mount time, if necessary, and we only free
2234 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2235 * this reason we don't need to a mutex to protect access to
2236 * jbd2_slab[] allocating or releasing memory; only in
2237 * jbd2_journal_create_slab().
2238 */
2239#define JBD2_MAX_SLABS 8
2240static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2241
2242static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2243 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2244 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2245};
2246
2247
2248static void jbd2_journal_destroy_slabs(void)
2249{
2250 int i;
2251
2252 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2253 if (jbd2_slab[i])
2254 kmem_cache_destroy(jbd2_slab[i]);
2255 jbd2_slab[i] = NULL;
2256 }
2257}
2258
2259static int jbd2_journal_create_slab(size_t size)
2260{
2261 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2262 int i = order_base_2(size) - 10;
2263 size_t slab_size;
2264
2265 if (size == PAGE_SIZE)
2266 return 0;
2267
2268 if (i >= JBD2_MAX_SLABS)
2269 return -EINVAL;
2270
2271 if (unlikely(i < 0))
2272 i = 0;
2273 mutex_lock(&jbd2_slab_create_mutex);
2274 if (jbd2_slab[i]) {
2275 mutex_unlock(&jbd2_slab_create_mutex);
2276 return 0; /* Already created */
2277 }
2278
2279 slab_size = 1 << (i+10);
2280 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2281 slab_size, 0, NULL);
2282 mutex_unlock(&jbd2_slab_create_mutex);
2283 if (!jbd2_slab[i]) {
2284 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2285 return -ENOMEM;
2286 }
2287 return 0;
2288}
2289
2290static struct kmem_cache *get_slab(size_t size)
2291{
2292 int i = order_base_2(size) - 10;
2293
2294 BUG_ON(i >= JBD2_MAX_SLABS);
2295 if (unlikely(i < 0))
2296 i = 0;
2297 BUG_ON(jbd2_slab[i] == NULL);
2298 return jbd2_slab[i];
2299}
2300
2301void *jbd2_alloc(size_t size, gfp_t flags)
2302{
2303 void *ptr;
2304
2305 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2306
2307 if (size < PAGE_SIZE)
2308 ptr = kmem_cache_alloc(get_slab(size), flags);
2309 else
2310 ptr = (void *)__get_free_pages(flags, get_order(size));
2311
2312 /* Check alignment; SLUB has gotten this wrong in the past,
2313 * and this can lead to user data corruption! */
2314 BUG_ON(((unsigned long) ptr) & (size-1));
2315
2316 return ptr;
2317}
2318
2319void jbd2_free(void *ptr, size_t size)
2320{
2321 if (size < PAGE_SIZE)
2322 kmem_cache_free(get_slab(size), ptr);
2323 else
2324 free_pages((unsigned long)ptr, get_order(size));
2325};
2326
2327/*
2328 * Journal_head storage management
2329 */
2330static struct kmem_cache *jbd2_journal_head_cache;
2331#ifdef CONFIG_JBD2_DEBUG
2332static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2333#endif
2334
2335static int jbd2_journal_init_journal_head_cache(void)
2336{
2337 int retval;
2338
2339 J_ASSERT(jbd2_journal_head_cache == NULL);
2340 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2341 sizeof(struct journal_head),
2342 0, /* offset */
2343 SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
2344 NULL); /* ctor */
2345 retval = 0;
2346 if (!jbd2_journal_head_cache) {
2347 retval = -ENOMEM;
2348 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2349 }
2350 return retval;
2351}
2352
2353static void jbd2_journal_destroy_journal_head_cache(void)
2354{
2355 if (jbd2_journal_head_cache) {
2356 kmem_cache_destroy(jbd2_journal_head_cache);
2357 jbd2_journal_head_cache = NULL;
2358 }
2359}
2360
2361/*
2362 * journal_head splicing and dicing
2363 */
2364static struct journal_head *journal_alloc_journal_head(void)
2365{
2366 struct journal_head *ret;
2367
2368#ifdef CONFIG_JBD2_DEBUG
2369 atomic_inc(&nr_journal_heads);
2370#endif
2371 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2372 if (!ret) {
2373 jbd_debug(1, "out of memory for journal_head\n");
2374 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2375 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2376 GFP_NOFS | __GFP_NOFAIL);
2377 }
2378 return ret;
2379}
2380
2381static void journal_free_journal_head(struct journal_head *jh)
2382{
2383#ifdef CONFIG_JBD2_DEBUG
2384 atomic_dec(&nr_journal_heads);
2385 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2386#endif
2387 kmem_cache_free(jbd2_journal_head_cache, jh);
2388}
2389
2390/*
2391 * A journal_head is attached to a buffer_head whenever JBD has an
2392 * interest in the buffer.
2393 *
2394 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2395 * is set. This bit is tested in core kernel code where we need to take
2396 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2397 * there.
2398 *
2399 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2400 *
2401 * When a buffer has its BH_JBD bit set it is immune from being released by
2402 * core kernel code, mainly via ->b_count.
2403 *
2404 * A journal_head is detached from its buffer_head when the journal_head's
2405 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2406 * transaction (b_cp_transaction) hold their references to b_jcount.
2407 *
2408 * Various places in the kernel want to attach a journal_head to a buffer_head
2409 * _before_ attaching the journal_head to a transaction. To protect the
2410 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2411 * journal_head's b_jcount refcount by one. The caller must call
2412 * jbd2_journal_put_journal_head() to undo this.
2413 *
2414 * So the typical usage would be:
2415 *
2416 * (Attach a journal_head if needed. Increments b_jcount)
2417 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2418 * ...
2419 * (Get another reference for transaction)
2420 * jbd2_journal_grab_journal_head(bh);
2421 * jh->b_transaction = xxx;
2422 * (Put original reference)
2423 * jbd2_journal_put_journal_head(jh);
2424 */
2425
2426/*
2427 * Give a buffer_head a journal_head.
2428 *
2429 * May sleep.
2430 */
2431struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2432{
2433 struct journal_head *jh;
2434 struct journal_head *new_jh = NULL;
2435
2436repeat:
2437 if (!buffer_jbd(bh))
2438 new_jh = journal_alloc_journal_head();
2439
2440 jbd_lock_bh_journal_head(bh);
2441 if (buffer_jbd(bh)) {
2442 jh = bh2jh(bh);
2443 } else {
2444 J_ASSERT_BH(bh,
2445 (atomic_read(&bh->b_count) > 0) ||
2446 (bh->b_page && bh->b_page->mapping));
2447
2448 if (!new_jh) {
2449 jbd_unlock_bh_journal_head(bh);
2450 goto repeat;
2451 }
2452
2453 jh = new_jh;
2454 new_jh = NULL; /* We consumed it */
2455 set_buffer_jbd(bh);
2456 bh->b_private = jh;
2457 jh->b_bh = bh;
2458 get_bh(bh);
2459 BUFFER_TRACE(bh, "added journal_head");
2460 }
2461 jh->b_jcount++;
2462 jbd_unlock_bh_journal_head(bh);
2463 if (new_jh)
2464 journal_free_journal_head(new_jh);
2465 return bh->b_private;
2466}
2467
2468/*
2469 * Grab a ref against this buffer_head's journal_head. If it ended up not
2470 * having a journal_head, return NULL
2471 */
2472struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2473{
2474 struct journal_head *jh = NULL;
2475
2476 jbd_lock_bh_journal_head(bh);
2477 if (buffer_jbd(bh)) {
2478 jh = bh2jh(bh);
2479 jh->b_jcount++;
2480 }
2481 jbd_unlock_bh_journal_head(bh);
2482 return jh;
2483}
2484
2485static void __journal_remove_journal_head(struct buffer_head *bh)
2486{
2487 struct journal_head *jh = bh2jh(bh);
2488
2489 J_ASSERT_JH(jh, jh->b_jcount >= 0);
2490 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2491 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2492 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2493 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2494 J_ASSERT_BH(bh, buffer_jbd(bh));
2495 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2496 BUFFER_TRACE(bh, "remove journal_head");
2497 if (jh->b_frozen_data) {
2498 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2499 jbd2_free(jh->b_frozen_data, bh->b_size);
2500 }
2501 if (jh->b_committed_data) {
2502 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2503 jbd2_free(jh->b_committed_data, bh->b_size);
2504 }
2505 bh->b_private = NULL;
2506 jh->b_bh = NULL; /* debug, really */
2507 clear_buffer_jbd(bh);
2508 journal_free_journal_head(jh);
2509}
2510
2511/*
2512 * Drop a reference on the passed journal_head. If it fell to zero then
2513 * release the journal_head from the buffer_head.
2514 */
2515void jbd2_journal_put_journal_head(struct journal_head *jh)
2516{
2517 struct buffer_head *bh = jh2bh(jh);
2518
2519 jbd_lock_bh_journal_head(bh);
2520 J_ASSERT_JH(jh, jh->b_jcount > 0);
2521 --jh->b_jcount;
2522 if (!jh->b_jcount) {
2523 __journal_remove_journal_head(bh);
2524 jbd_unlock_bh_journal_head(bh);
2525 __brelse(bh);
2526 } else
2527 jbd_unlock_bh_journal_head(bh);
2528}
2529
2530/*
2531 * Initialize jbd inode head
2532 */
2533void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2534{
2535 jinode->i_transaction = NULL;
2536 jinode->i_next_transaction = NULL;
2537 jinode->i_vfs_inode = inode;
2538 jinode->i_flags = 0;
2539 INIT_LIST_HEAD(&jinode->i_list);
2540}
2541
2542/*
2543 * Function to be called before we start removing inode from memory (i.e.,
2544 * clear_inode() is a fine place to be called from). It removes inode from
2545 * transaction's lists.
2546 */
2547void jbd2_journal_release_jbd_inode(journal_t *journal,
2548 struct jbd2_inode *jinode)
2549{
2550 if (!journal)
2551 return;
2552restart:
2553 spin_lock(&journal->j_list_lock);
2554 /* Is commit writing out inode - we have to wait */
2555 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2556 wait_queue_head_t *wq;
2557 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2558 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2559 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2560 spin_unlock(&journal->j_list_lock);
2561 schedule();
2562 finish_wait(wq, &wait.wait);
2563 goto restart;
2564 }
2565
2566 if (jinode->i_transaction) {
2567 list_del(&jinode->i_list);
2568 jinode->i_transaction = NULL;
2569 }
2570 spin_unlock(&journal->j_list_lock);
2571}
2572
2573
2574#ifdef CONFIG_PROC_FS
2575
2576#define JBD2_STATS_PROC_NAME "fs/jbd2"
2577
2578static void __init jbd2_create_jbd_stats_proc_entry(void)
2579{
2580 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2581}
2582
2583static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2584{
2585 if (proc_jbd2_stats)
2586 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2587}
2588
2589#else
2590
2591#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2592#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2593
2594#endif
2595
2596struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2597
2598static int __init jbd2_journal_init_handle_cache(void)
2599{
2600 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2601 if (jbd2_handle_cache == NULL) {
2602 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2603 return -ENOMEM;
2604 }
2605 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2606 if (jbd2_inode_cache == NULL) {
2607 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2608 kmem_cache_destroy(jbd2_handle_cache);
2609 return -ENOMEM;
2610 }
2611 return 0;
2612}
2613
2614static void jbd2_journal_destroy_handle_cache(void)
2615{
2616 if (jbd2_handle_cache)
2617 kmem_cache_destroy(jbd2_handle_cache);
2618 if (jbd2_inode_cache)
2619 kmem_cache_destroy(jbd2_inode_cache);
2620
2621}
2622
2623/*
2624 * Module startup and shutdown
2625 */
2626
2627static int __init journal_init_caches(void)
2628{
2629 int ret;
2630
2631 ret = jbd2_journal_init_revoke_caches();
2632 if (ret == 0)
2633 ret = jbd2_journal_init_journal_head_cache();
2634 if (ret == 0)
2635 ret = jbd2_journal_init_handle_cache();
2636 if (ret == 0)
2637 ret = jbd2_journal_init_transaction_cache();
2638 return ret;
2639}
2640
2641static void jbd2_journal_destroy_caches(void)
2642{
2643 jbd2_journal_destroy_revoke_caches();
2644 jbd2_journal_destroy_journal_head_cache();
2645 jbd2_journal_destroy_handle_cache();
2646 jbd2_journal_destroy_transaction_cache();
2647 jbd2_journal_destroy_slabs();
2648}
2649
2650static int __init journal_init(void)
2651{
2652 int ret;
2653
2654 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2655
2656 ret = journal_init_caches();
2657 if (ret == 0) {
2658 jbd2_create_jbd_stats_proc_entry();
2659 } else {
2660 jbd2_journal_destroy_caches();
2661 }
2662 return ret;
2663}
2664
2665static void __exit journal_exit(void)
2666{
2667#ifdef CONFIG_JBD2_DEBUG
2668 int n = atomic_read(&nr_journal_heads);
2669 if (n)
2670 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2671#endif
2672 jbd2_remove_jbd_stats_proc_entry();
2673 jbd2_journal_destroy_caches();
2674}
2675
2676MODULE_LICENSE("GPL");
2677module_init(journal_init);
2678module_exit(journal_exit);
2679
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
52static ushort jbd2_journal_enable_debug __read_mostly;
53
54module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
55MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
56#endif
57
58EXPORT_SYMBOL(jbd2_journal_extend);
59EXPORT_SYMBOL(jbd2_journal_stop);
60EXPORT_SYMBOL(jbd2_journal_lock_updates);
61EXPORT_SYMBOL(jbd2_journal_unlock_updates);
62EXPORT_SYMBOL(jbd2_journal_get_write_access);
63EXPORT_SYMBOL(jbd2_journal_get_create_access);
64EXPORT_SYMBOL(jbd2_journal_get_undo_access);
65EXPORT_SYMBOL(jbd2_journal_set_triggers);
66EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
67EXPORT_SYMBOL(jbd2_journal_forget);
68EXPORT_SYMBOL(jbd2_journal_flush);
69EXPORT_SYMBOL(jbd2_journal_revoke);
70
71EXPORT_SYMBOL(jbd2_journal_init_dev);
72EXPORT_SYMBOL(jbd2_journal_init_inode);
73EXPORT_SYMBOL(jbd2_journal_check_used_features);
74EXPORT_SYMBOL(jbd2_journal_check_available_features);
75EXPORT_SYMBOL(jbd2_journal_set_features);
76EXPORT_SYMBOL(jbd2_journal_load);
77EXPORT_SYMBOL(jbd2_journal_destroy);
78EXPORT_SYMBOL(jbd2_journal_abort);
79EXPORT_SYMBOL(jbd2_journal_errno);
80EXPORT_SYMBOL(jbd2_journal_ack_err);
81EXPORT_SYMBOL(jbd2_journal_clear_err);
82EXPORT_SYMBOL(jbd2_log_wait_commit);
83EXPORT_SYMBOL(jbd2_journal_start_commit);
84EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
85EXPORT_SYMBOL(jbd2_journal_wipe);
86EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
87EXPORT_SYMBOL(jbd2_journal_invalidate_folio);
88EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
89EXPORT_SYMBOL(jbd2_journal_force_commit);
90EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
91EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
92EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers);
93EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
94EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
95EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
96EXPORT_SYMBOL(jbd2_inode_cache);
97
98static int jbd2_journal_create_slab(size_t slab_size);
99
100#ifdef CONFIG_JBD2_DEBUG
101void __jbd2_debug(int level, const char *file, const char *func,
102 unsigned int line, const char *fmt, ...)
103{
104 struct va_format vaf;
105 va_list args;
106
107 if (level > jbd2_journal_enable_debug)
108 return;
109 va_start(args, fmt);
110 vaf.fmt = fmt;
111 vaf.va = &args;
112 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
113 va_end(args);
114}
115#endif
116
117/* Checksumming functions */
118static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
119{
120 __u32 csum;
121 __be32 old_csum;
122
123 old_csum = sb->s_checksum;
124 sb->s_checksum = 0;
125 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
126 sb->s_checksum = old_csum;
127
128 return cpu_to_be32(csum);
129}
130
131/*
132 * Helper function used to manage commit timeouts
133 */
134
135static void commit_timeout(struct timer_list *t)
136{
137 journal_t *journal = from_timer(journal, t, j_commit_timer);
138
139 wake_up_process(journal->j_task);
140}
141
142/*
143 * kjournald2: The main thread function used to manage a logging device
144 * journal.
145 *
146 * This kernel thread is responsible for two things:
147 *
148 * 1) COMMIT: Every so often we need to commit the current state of the
149 * filesystem to disk. The journal thread is responsible for writing
150 * all of the metadata buffers to disk. If a fast commit is ongoing
151 * journal thread waits until it's done and then continues from
152 * there on.
153 *
154 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
155 * of the data in that part of the log has been rewritten elsewhere on
156 * the disk. Flushing these old buffers to reclaim space in the log is
157 * known as checkpointing, and this thread is responsible for that job.
158 */
159
160static int kjournald2(void *arg)
161{
162 journal_t *journal = arg;
163 transaction_t *transaction;
164
165 /*
166 * Set up an interval timer which can be used to trigger a commit wakeup
167 * after the commit interval expires
168 */
169 timer_setup(&journal->j_commit_timer, commit_timeout, 0);
170
171 set_freezable();
172
173 /* Record that the journal thread is running */
174 journal->j_task = current;
175 wake_up(&journal->j_wait_done_commit);
176
177 /*
178 * Make sure that no allocations from this kernel thread will ever
179 * recurse to the fs layer because we are responsible for the
180 * transaction commit and any fs involvement might get stuck waiting for
181 * the trasn. commit.
182 */
183 memalloc_nofs_save();
184
185 /*
186 * And now, wait forever for commit wakeup events.
187 */
188 write_lock(&journal->j_state_lock);
189
190loop:
191 if (journal->j_flags & JBD2_UNMOUNT)
192 goto end_loop;
193
194 jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n",
195 journal->j_commit_sequence, journal->j_commit_request);
196
197 if (journal->j_commit_sequence != journal->j_commit_request) {
198 jbd2_debug(1, "OK, requests differ\n");
199 write_unlock(&journal->j_state_lock);
200 del_timer_sync(&journal->j_commit_timer);
201 jbd2_journal_commit_transaction(journal);
202 write_lock(&journal->j_state_lock);
203 goto loop;
204 }
205
206 wake_up(&journal->j_wait_done_commit);
207 if (freezing(current)) {
208 /*
209 * The simpler the better. Flushing journal isn't a
210 * good idea, because that depends on threads that may
211 * be already stopped.
212 */
213 jbd2_debug(1, "Now suspending kjournald2\n");
214 write_unlock(&journal->j_state_lock);
215 try_to_freeze();
216 write_lock(&journal->j_state_lock);
217 } else {
218 /*
219 * We assume on resume that commits are already there,
220 * so we don't sleep
221 */
222 DEFINE_WAIT(wait);
223 int should_sleep = 1;
224
225 prepare_to_wait(&journal->j_wait_commit, &wait,
226 TASK_INTERRUPTIBLE);
227 if (journal->j_commit_sequence != journal->j_commit_request)
228 should_sleep = 0;
229 transaction = journal->j_running_transaction;
230 if (transaction && time_after_eq(jiffies,
231 transaction->t_expires))
232 should_sleep = 0;
233 if (journal->j_flags & JBD2_UNMOUNT)
234 should_sleep = 0;
235 if (should_sleep) {
236 write_unlock(&journal->j_state_lock);
237 schedule();
238 write_lock(&journal->j_state_lock);
239 }
240 finish_wait(&journal->j_wait_commit, &wait);
241 }
242
243 jbd2_debug(1, "kjournald2 wakes\n");
244
245 /*
246 * Were we woken up by a commit wakeup event?
247 */
248 transaction = journal->j_running_transaction;
249 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
250 journal->j_commit_request = transaction->t_tid;
251 jbd2_debug(1, "woke because of timeout\n");
252 }
253 goto loop;
254
255end_loop:
256 del_timer_sync(&journal->j_commit_timer);
257 journal->j_task = NULL;
258 wake_up(&journal->j_wait_done_commit);
259 jbd2_debug(1, "Journal thread exiting.\n");
260 write_unlock(&journal->j_state_lock);
261 return 0;
262}
263
264static int jbd2_journal_start_thread(journal_t *journal)
265{
266 struct task_struct *t;
267
268 t = kthread_run(kjournald2, journal, "jbd2/%s",
269 journal->j_devname);
270 if (IS_ERR(t))
271 return PTR_ERR(t);
272
273 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
274 return 0;
275}
276
277static void journal_kill_thread(journal_t *journal)
278{
279 write_lock(&journal->j_state_lock);
280 journal->j_flags |= JBD2_UNMOUNT;
281
282 while (journal->j_task) {
283 write_unlock(&journal->j_state_lock);
284 wake_up(&journal->j_wait_commit);
285 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
286 write_lock(&journal->j_state_lock);
287 }
288 write_unlock(&journal->j_state_lock);
289}
290
291/*
292 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
293 *
294 * Writes a metadata buffer to a given disk block. The actual IO is not
295 * performed but a new buffer_head is constructed which labels the data
296 * to be written with the correct destination disk block.
297 *
298 * Any magic-number escaping which needs to be done will cause a
299 * copy-out here. If the buffer happens to start with the
300 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
301 * magic number is only written to the log for descripter blocks. In
302 * this case, we copy the data and replace the first word with 0, and we
303 * return a result code which indicates that this buffer needs to be
304 * marked as an escaped buffer in the corresponding log descriptor
305 * block. The missing word can then be restored when the block is read
306 * during recovery.
307 *
308 * If the source buffer has already been modified by a new transaction
309 * since we took the last commit snapshot, we use the frozen copy of
310 * that data for IO. If we end up using the existing buffer_head's data
311 * for the write, then we have to make sure nobody modifies it while the
312 * IO is in progress. do_get_write_access() handles this.
313 *
314 * The function returns a pointer to the buffer_head to be used for IO.
315 *
316 *
317 * Return value:
318 * <0: Error
319 * >=0: Finished OK
320 *
321 * On success:
322 * Bit 0 set == escape performed on the data
323 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
324 */
325
326int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
327 struct journal_head *jh_in,
328 struct buffer_head **bh_out,
329 sector_t blocknr)
330{
331 int need_copy_out = 0;
332 int done_copy_out = 0;
333 int do_escape = 0;
334 char *mapped_data;
335 struct buffer_head *new_bh;
336 struct folio *new_folio;
337 unsigned int new_offset;
338 struct buffer_head *bh_in = jh2bh(jh_in);
339 journal_t *journal = transaction->t_journal;
340
341 /*
342 * The buffer really shouldn't be locked: only the current committing
343 * transaction is allowed to write it, so nobody else is allowed
344 * to do any IO.
345 *
346 * akpm: except if we're journalling data, and write() output is
347 * also part of a shared mapping, and another thread has
348 * decided to launch a writepage() against this buffer.
349 */
350 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
351
352 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
353
354 /* keep subsequent assertions sane */
355 atomic_set(&new_bh->b_count, 1);
356
357 spin_lock(&jh_in->b_state_lock);
358repeat:
359 /*
360 * If a new transaction has already done a buffer copy-out, then
361 * we use that version of the data for the commit.
362 */
363 if (jh_in->b_frozen_data) {
364 done_copy_out = 1;
365 new_folio = virt_to_folio(jh_in->b_frozen_data);
366 new_offset = offset_in_folio(new_folio, jh_in->b_frozen_data);
367 } else {
368 new_folio = jh2bh(jh_in)->b_folio;
369 new_offset = offset_in_folio(new_folio, jh2bh(jh_in)->b_data);
370 }
371
372 mapped_data = kmap_local_folio(new_folio, new_offset);
373 /*
374 * Fire data frozen trigger if data already wasn't frozen. Do this
375 * before checking for escaping, as the trigger may modify the magic
376 * offset. If a copy-out happens afterwards, it will have the correct
377 * data in the buffer.
378 */
379 if (!done_copy_out)
380 jbd2_buffer_frozen_trigger(jh_in, mapped_data,
381 jh_in->b_triggers);
382
383 /*
384 * Check for escaping
385 */
386 if (*((__be32 *)mapped_data) == cpu_to_be32(JBD2_MAGIC_NUMBER)) {
387 need_copy_out = 1;
388 do_escape = 1;
389 }
390 kunmap_local(mapped_data);
391
392 /*
393 * Do we need to do a data copy?
394 */
395 if (need_copy_out && !done_copy_out) {
396 char *tmp;
397
398 spin_unlock(&jh_in->b_state_lock);
399 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
400 if (!tmp) {
401 brelse(new_bh);
402 return -ENOMEM;
403 }
404 spin_lock(&jh_in->b_state_lock);
405 if (jh_in->b_frozen_data) {
406 jbd2_free(tmp, bh_in->b_size);
407 goto repeat;
408 }
409
410 jh_in->b_frozen_data = tmp;
411 memcpy_from_folio(tmp, new_folio, new_offset, bh_in->b_size);
412
413 new_folio = virt_to_folio(tmp);
414 new_offset = offset_in_folio(new_folio, tmp);
415 done_copy_out = 1;
416
417 /*
418 * This isn't strictly necessary, as we're using frozen
419 * data for the escaping, but it keeps consistency with
420 * b_frozen_data usage.
421 */
422 jh_in->b_frozen_triggers = jh_in->b_triggers;
423 }
424
425 /*
426 * Did we need to do an escaping? Now we've done all the
427 * copying, we can finally do so.
428 */
429 if (do_escape) {
430 mapped_data = kmap_local_folio(new_folio, new_offset);
431 *((unsigned int *)mapped_data) = 0;
432 kunmap_local(mapped_data);
433 }
434
435 folio_set_bh(new_bh, new_folio, new_offset);
436 new_bh->b_size = bh_in->b_size;
437 new_bh->b_bdev = journal->j_dev;
438 new_bh->b_blocknr = blocknr;
439 new_bh->b_private = bh_in;
440 set_buffer_mapped(new_bh);
441 set_buffer_dirty(new_bh);
442
443 *bh_out = new_bh;
444
445 /*
446 * The to-be-written buffer needs to get moved to the io queue,
447 * and the original buffer whose contents we are shadowing or
448 * copying is moved to the transaction's shadow queue.
449 */
450 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
451 spin_lock(&journal->j_list_lock);
452 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
453 spin_unlock(&journal->j_list_lock);
454 set_buffer_shadow(bh_in);
455 spin_unlock(&jh_in->b_state_lock);
456
457 return do_escape | (done_copy_out << 1);
458}
459
460/*
461 * Allocation code for the journal file. Manage the space left in the
462 * journal, so that we can begin checkpointing when appropriate.
463 */
464
465/*
466 * Called with j_state_lock locked for writing.
467 * Returns true if a transaction commit was started.
468 */
469static int __jbd2_log_start_commit(journal_t *journal, tid_t target)
470{
471 /* Return if the txn has already requested to be committed */
472 if (journal->j_commit_request == target)
473 return 0;
474
475 /*
476 * The only transaction we can possibly wait upon is the
477 * currently running transaction (if it exists). Otherwise,
478 * the target tid must be an old one.
479 */
480 if (journal->j_running_transaction &&
481 journal->j_running_transaction->t_tid == target) {
482 /*
483 * We want a new commit: OK, mark the request and wakeup the
484 * commit thread. We do _not_ do the commit ourselves.
485 */
486
487 journal->j_commit_request = target;
488 jbd2_debug(1, "JBD2: requesting commit %u/%u\n",
489 journal->j_commit_request,
490 journal->j_commit_sequence);
491 journal->j_running_transaction->t_requested = jiffies;
492 wake_up(&journal->j_wait_commit);
493 return 1;
494 } else if (!tid_geq(journal->j_commit_request, target))
495 /* This should never happen, but if it does, preserve
496 the evidence before kjournald goes into a loop and
497 increments j_commit_sequence beyond all recognition. */
498 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
499 journal->j_commit_request,
500 journal->j_commit_sequence,
501 target, journal->j_running_transaction ?
502 journal->j_running_transaction->t_tid : 0);
503 return 0;
504}
505
506int jbd2_log_start_commit(journal_t *journal, tid_t tid)
507{
508 int ret;
509
510 write_lock(&journal->j_state_lock);
511 ret = __jbd2_log_start_commit(journal, tid);
512 write_unlock(&journal->j_state_lock);
513 return ret;
514}
515
516/*
517 * Force and wait any uncommitted transactions. We can only force the running
518 * transaction if we don't have an active handle, otherwise, we will deadlock.
519 * Returns: <0 in case of error,
520 * 0 if nothing to commit,
521 * 1 if transaction was successfully committed.
522 */
523static int __jbd2_journal_force_commit(journal_t *journal)
524{
525 transaction_t *transaction = NULL;
526 tid_t tid;
527 int need_to_start = 0, ret = 0;
528
529 read_lock(&journal->j_state_lock);
530 if (journal->j_running_transaction && !current->journal_info) {
531 transaction = journal->j_running_transaction;
532 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
533 need_to_start = 1;
534 } else if (journal->j_committing_transaction)
535 transaction = journal->j_committing_transaction;
536
537 if (!transaction) {
538 /* Nothing to commit */
539 read_unlock(&journal->j_state_lock);
540 return 0;
541 }
542 tid = transaction->t_tid;
543 read_unlock(&journal->j_state_lock);
544 if (need_to_start)
545 jbd2_log_start_commit(journal, tid);
546 ret = jbd2_log_wait_commit(journal, tid);
547 if (!ret)
548 ret = 1;
549
550 return ret;
551}
552
553/**
554 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
555 * calling process is not within transaction.
556 *
557 * @journal: journal to force
558 * Returns true if progress was made.
559 *
560 * This is used for forcing out undo-protected data which contains
561 * bitmaps, when the fs is running out of space.
562 */
563int jbd2_journal_force_commit_nested(journal_t *journal)
564{
565 int ret;
566
567 ret = __jbd2_journal_force_commit(journal);
568 return ret > 0;
569}
570
571/**
572 * jbd2_journal_force_commit() - force any uncommitted transactions
573 * @journal: journal to force
574 *
575 * Caller want unconditional commit. We can only force the running transaction
576 * if we don't have an active handle, otherwise, we will deadlock.
577 */
578int jbd2_journal_force_commit(journal_t *journal)
579{
580 int ret;
581
582 J_ASSERT(!current->journal_info);
583 ret = __jbd2_journal_force_commit(journal);
584 if (ret > 0)
585 ret = 0;
586 return ret;
587}
588
589/*
590 * Start a commit of the current running transaction (if any). Returns true
591 * if a transaction is going to be committed (or is currently already
592 * committing), and fills its tid in at *ptid
593 */
594int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
595{
596 int ret = 0;
597
598 write_lock(&journal->j_state_lock);
599 if (journal->j_running_transaction) {
600 tid_t tid = journal->j_running_transaction->t_tid;
601
602 __jbd2_log_start_commit(journal, tid);
603 /* There's a running transaction and we've just made sure
604 * it's commit has been scheduled. */
605 if (ptid)
606 *ptid = tid;
607 ret = 1;
608 } else if (journal->j_committing_transaction) {
609 /*
610 * If commit has been started, then we have to wait for
611 * completion of that transaction.
612 */
613 if (ptid)
614 *ptid = journal->j_committing_transaction->t_tid;
615 ret = 1;
616 }
617 write_unlock(&journal->j_state_lock);
618 return ret;
619}
620
621/*
622 * Return 1 if a given transaction has not yet sent barrier request
623 * connected with a transaction commit. If 0 is returned, transaction
624 * may or may not have sent the barrier. Used to avoid sending barrier
625 * twice in common cases.
626 */
627int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
628{
629 int ret = 0;
630 transaction_t *commit_trans;
631
632 if (!(journal->j_flags & JBD2_BARRIER))
633 return 0;
634 read_lock(&journal->j_state_lock);
635 /* Transaction already committed? */
636 if (tid_geq(journal->j_commit_sequence, tid))
637 goto out;
638 commit_trans = journal->j_committing_transaction;
639 if (!commit_trans || commit_trans->t_tid != tid) {
640 ret = 1;
641 goto out;
642 }
643 /*
644 * Transaction is being committed and we already proceeded to
645 * submitting a flush to fs partition?
646 */
647 if (journal->j_fs_dev != journal->j_dev) {
648 if (!commit_trans->t_need_data_flush ||
649 commit_trans->t_state >= T_COMMIT_DFLUSH)
650 goto out;
651 } else {
652 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
653 goto out;
654 }
655 ret = 1;
656out:
657 read_unlock(&journal->j_state_lock);
658 return ret;
659}
660EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
661
662/*
663 * Wait for a specified commit to complete.
664 * The caller may not hold the journal lock.
665 */
666int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
667{
668 int err = 0;
669
670 read_lock(&journal->j_state_lock);
671#ifdef CONFIG_PROVE_LOCKING
672 /*
673 * Some callers make sure transaction is already committing and in that
674 * case we cannot block on open handles anymore. So don't warn in that
675 * case.
676 */
677 if (tid_gt(tid, journal->j_commit_sequence) &&
678 (!journal->j_committing_transaction ||
679 journal->j_committing_transaction->t_tid != tid)) {
680 read_unlock(&journal->j_state_lock);
681 jbd2_might_wait_for_commit(journal);
682 read_lock(&journal->j_state_lock);
683 }
684#endif
685#ifdef CONFIG_JBD2_DEBUG
686 if (!tid_geq(journal->j_commit_request, tid)) {
687 printk(KERN_ERR
688 "%s: error: j_commit_request=%u, tid=%u\n",
689 __func__, journal->j_commit_request, tid);
690 }
691#endif
692 while (tid_gt(tid, journal->j_commit_sequence)) {
693 jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
694 tid, journal->j_commit_sequence);
695 read_unlock(&journal->j_state_lock);
696 wake_up(&journal->j_wait_commit);
697 wait_event(journal->j_wait_done_commit,
698 !tid_gt(tid, journal->j_commit_sequence));
699 read_lock(&journal->j_state_lock);
700 }
701 read_unlock(&journal->j_state_lock);
702
703 if (unlikely(is_journal_aborted(journal)))
704 err = -EIO;
705 return err;
706}
707
708/*
709 * Start a fast commit. If there's an ongoing fast or full commit wait for
710 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
711 * if a fast commit is not needed, either because there's an already a commit
712 * going on or this tid has already been committed. Returns -EINVAL if no jbd2
713 * commit has yet been performed.
714 */
715int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
716{
717 if (unlikely(is_journal_aborted(journal)))
718 return -EIO;
719 /*
720 * Fast commits only allowed if at least one full commit has
721 * been processed.
722 */
723 if (!journal->j_stats.ts_tid)
724 return -EINVAL;
725
726 write_lock(&journal->j_state_lock);
727 if (tid <= journal->j_commit_sequence) {
728 write_unlock(&journal->j_state_lock);
729 return -EALREADY;
730 }
731
732 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
733 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
734 DEFINE_WAIT(wait);
735
736 prepare_to_wait(&journal->j_fc_wait, &wait,
737 TASK_UNINTERRUPTIBLE);
738 write_unlock(&journal->j_state_lock);
739 schedule();
740 finish_wait(&journal->j_fc_wait, &wait);
741 return -EALREADY;
742 }
743 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
744 write_unlock(&journal->j_state_lock);
745 jbd2_journal_lock_updates(journal);
746
747 return 0;
748}
749EXPORT_SYMBOL(jbd2_fc_begin_commit);
750
751/*
752 * Stop a fast commit. If fallback is set, this function starts commit of
753 * TID tid before any other fast commit can start.
754 */
755static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
756{
757 jbd2_journal_unlock_updates(journal);
758 if (journal->j_fc_cleanup_callback)
759 journal->j_fc_cleanup_callback(journal, 0, tid);
760 write_lock(&journal->j_state_lock);
761 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
762 if (fallback)
763 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
764 write_unlock(&journal->j_state_lock);
765 wake_up(&journal->j_fc_wait);
766 if (fallback)
767 return jbd2_complete_transaction(journal, tid);
768 return 0;
769}
770
771int jbd2_fc_end_commit(journal_t *journal)
772{
773 return __jbd2_fc_end_commit(journal, 0, false);
774}
775EXPORT_SYMBOL(jbd2_fc_end_commit);
776
777int jbd2_fc_end_commit_fallback(journal_t *journal)
778{
779 tid_t tid;
780
781 read_lock(&journal->j_state_lock);
782 tid = journal->j_running_transaction ?
783 journal->j_running_transaction->t_tid : 0;
784 read_unlock(&journal->j_state_lock);
785 return __jbd2_fc_end_commit(journal, tid, true);
786}
787EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
788
789/* Return 1 when transaction with given tid has already committed. */
790int jbd2_transaction_committed(journal_t *journal, tid_t tid)
791{
792 int ret = 1;
793
794 read_lock(&journal->j_state_lock);
795 if (journal->j_running_transaction &&
796 journal->j_running_transaction->t_tid == tid)
797 ret = 0;
798 if (journal->j_committing_transaction &&
799 journal->j_committing_transaction->t_tid == tid)
800 ret = 0;
801 read_unlock(&journal->j_state_lock);
802 return ret;
803}
804EXPORT_SYMBOL(jbd2_transaction_committed);
805
806/*
807 * When this function returns the transaction corresponding to tid
808 * will be completed. If the transaction has currently running, start
809 * committing that transaction before waiting for it to complete. If
810 * the transaction id is stale, it is by definition already completed,
811 * so just return SUCCESS.
812 */
813int jbd2_complete_transaction(journal_t *journal, tid_t tid)
814{
815 int need_to_wait = 1;
816
817 read_lock(&journal->j_state_lock);
818 if (journal->j_running_transaction &&
819 journal->j_running_transaction->t_tid == tid) {
820 if (journal->j_commit_request != tid) {
821 /* transaction not yet started, so request it */
822 read_unlock(&journal->j_state_lock);
823 jbd2_log_start_commit(journal, tid);
824 goto wait_commit;
825 }
826 } else if (!(journal->j_committing_transaction &&
827 journal->j_committing_transaction->t_tid == tid))
828 need_to_wait = 0;
829 read_unlock(&journal->j_state_lock);
830 if (!need_to_wait)
831 return 0;
832wait_commit:
833 return jbd2_log_wait_commit(journal, tid);
834}
835EXPORT_SYMBOL(jbd2_complete_transaction);
836
837/*
838 * Log buffer allocation routines:
839 */
840
841int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
842{
843 unsigned long blocknr;
844
845 write_lock(&journal->j_state_lock);
846 J_ASSERT(journal->j_free > 1);
847
848 blocknr = journal->j_head;
849 journal->j_head++;
850 journal->j_free--;
851 if (journal->j_head == journal->j_last)
852 journal->j_head = journal->j_first;
853 write_unlock(&journal->j_state_lock);
854 return jbd2_journal_bmap(journal, blocknr, retp);
855}
856
857/* Map one fast commit buffer for use by the file system */
858int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
859{
860 unsigned long long pblock;
861 unsigned long blocknr;
862 int ret = 0;
863 struct buffer_head *bh;
864 int fc_off;
865
866 *bh_out = NULL;
867
868 if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
869 fc_off = journal->j_fc_off;
870 blocknr = journal->j_fc_first + fc_off;
871 journal->j_fc_off++;
872 } else {
873 ret = -EINVAL;
874 }
875
876 if (ret)
877 return ret;
878
879 ret = jbd2_journal_bmap(journal, blocknr, &pblock);
880 if (ret)
881 return ret;
882
883 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
884 if (!bh)
885 return -ENOMEM;
886
887
888 journal->j_fc_wbuf[fc_off] = bh;
889
890 *bh_out = bh;
891
892 return 0;
893}
894EXPORT_SYMBOL(jbd2_fc_get_buf);
895
896/*
897 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
898 * for completion.
899 */
900int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
901{
902 struct buffer_head *bh;
903 int i, j_fc_off;
904
905 j_fc_off = journal->j_fc_off;
906
907 /*
908 * Wait in reverse order to minimize chances of us being woken up before
909 * all IOs have completed
910 */
911 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
912 bh = journal->j_fc_wbuf[i];
913 wait_on_buffer(bh);
914 /*
915 * Update j_fc_off so jbd2_fc_release_bufs can release remain
916 * buffer head.
917 */
918 if (unlikely(!buffer_uptodate(bh))) {
919 journal->j_fc_off = i + 1;
920 return -EIO;
921 }
922 put_bh(bh);
923 journal->j_fc_wbuf[i] = NULL;
924 }
925
926 return 0;
927}
928EXPORT_SYMBOL(jbd2_fc_wait_bufs);
929
930int jbd2_fc_release_bufs(journal_t *journal)
931{
932 struct buffer_head *bh;
933 int i, j_fc_off;
934
935 j_fc_off = journal->j_fc_off;
936
937 for (i = j_fc_off - 1; i >= 0; i--) {
938 bh = journal->j_fc_wbuf[i];
939 if (!bh)
940 break;
941 put_bh(bh);
942 journal->j_fc_wbuf[i] = NULL;
943 }
944
945 return 0;
946}
947EXPORT_SYMBOL(jbd2_fc_release_bufs);
948
949/*
950 * Conversion of logical to physical block numbers for the journal
951 *
952 * On external journals the journal blocks are identity-mapped, so
953 * this is a no-op. If needed, we can use j_blk_offset - everything is
954 * ready.
955 */
956int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
957 unsigned long long *retp)
958{
959 int err = 0;
960 unsigned long long ret;
961 sector_t block = blocknr;
962
963 if (journal->j_bmap) {
964 err = journal->j_bmap(journal, &block);
965 if (err == 0)
966 *retp = block;
967 } else if (journal->j_inode) {
968 ret = bmap(journal->j_inode, &block);
969
970 if (ret || !block) {
971 printk(KERN_ALERT "%s: journal block not found "
972 "at offset %lu on %s\n",
973 __func__, blocknr, journal->j_devname);
974 err = -EIO;
975 jbd2_journal_abort(journal, err);
976 } else {
977 *retp = block;
978 }
979
980 } else {
981 *retp = blocknr; /* +journal->j_blk_offset */
982 }
983 return err;
984}
985
986/*
987 * We play buffer_head aliasing tricks to write data/metadata blocks to
988 * the journal without copying their contents, but for journal
989 * descriptor blocks we do need to generate bona fide buffers.
990 *
991 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
992 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
993 * But we don't bother doing that, so there will be coherency problems with
994 * mmaps of blockdevs which hold live JBD-controlled filesystems.
995 */
996struct buffer_head *
997jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
998{
999 journal_t *journal = transaction->t_journal;
1000 struct buffer_head *bh;
1001 unsigned long long blocknr;
1002 journal_header_t *header;
1003 int err;
1004
1005 err = jbd2_journal_next_log_block(journal, &blocknr);
1006
1007 if (err)
1008 return NULL;
1009
1010 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1011 if (!bh)
1012 return NULL;
1013 atomic_dec(&transaction->t_outstanding_credits);
1014 lock_buffer(bh);
1015 memset(bh->b_data, 0, journal->j_blocksize);
1016 header = (journal_header_t *)bh->b_data;
1017 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1018 header->h_blocktype = cpu_to_be32(type);
1019 header->h_sequence = cpu_to_be32(transaction->t_tid);
1020 set_buffer_uptodate(bh);
1021 unlock_buffer(bh);
1022 BUFFER_TRACE(bh, "return this buffer");
1023 return bh;
1024}
1025
1026void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1027{
1028 struct jbd2_journal_block_tail *tail;
1029 __u32 csum;
1030
1031 if (!jbd2_journal_has_csum_v2or3(j))
1032 return;
1033
1034 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1035 sizeof(struct jbd2_journal_block_tail));
1036 tail->t_checksum = 0;
1037 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1038 tail->t_checksum = cpu_to_be32(csum);
1039}
1040
1041/*
1042 * Return tid of the oldest transaction in the journal and block in the journal
1043 * where the transaction starts.
1044 *
1045 * If the journal is now empty, return which will be the next transaction ID
1046 * we will write and where will that transaction start.
1047 *
1048 * The return value is 0 if journal tail cannot be pushed any further, 1 if
1049 * it can.
1050 */
1051int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1052 unsigned long *block)
1053{
1054 transaction_t *transaction;
1055 int ret;
1056
1057 read_lock(&journal->j_state_lock);
1058 spin_lock(&journal->j_list_lock);
1059 transaction = journal->j_checkpoint_transactions;
1060 if (transaction) {
1061 *tid = transaction->t_tid;
1062 *block = transaction->t_log_start;
1063 } else if ((transaction = journal->j_committing_transaction) != NULL) {
1064 *tid = transaction->t_tid;
1065 *block = transaction->t_log_start;
1066 } else if ((transaction = journal->j_running_transaction) != NULL) {
1067 *tid = transaction->t_tid;
1068 *block = journal->j_head;
1069 } else {
1070 *tid = journal->j_transaction_sequence;
1071 *block = journal->j_head;
1072 }
1073 ret = tid_gt(*tid, journal->j_tail_sequence);
1074 spin_unlock(&journal->j_list_lock);
1075 read_unlock(&journal->j_state_lock);
1076
1077 return ret;
1078}
1079
1080/*
1081 * Update information in journal structure and in on disk journal superblock
1082 * about log tail. This function does not check whether information passed in
1083 * really pushes log tail further. It's responsibility of the caller to make
1084 * sure provided log tail information is valid (e.g. by holding
1085 * j_checkpoint_mutex all the time between computing log tail and calling this
1086 * function as is the case with jbd2_cleanup_journal_tail()).
1087 *
1088 * Requires j_checkpoint_mutex
1089 */
1090int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1091{
1092 unsigned long freed;
1093 int ret;
1094
1095 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1096
1097 /*
1098 * We cannot afford for write to remain in drive's caches since as
1099 * soon as we update j_tail, next transaction can start reusing journal
1100 * space and if we lose sb update during power failure we'd replay
1101 * old transaction with possibly newly overwritten data.
1102 */
1103 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, REQ_FUA);
1104 if (ret)
1105 goto out;
1106
1107 write_lock(&journal->j_state_lock);
1108 freed = block - journal->j_tail;
1109 if (block < journal->j_tail)
1110 freed += journal->j_last - journal->j_first;
1111
1112 trace_jbd2_update_log_tail(journal, tid, block, freed);
1113 jbd2_debug(1,
1114 "Cleaning journal tail from %u to %u (offset %lu), "
1115 "freeing %lu\n",
1116 journal->j_tail_sequence, tid, block, freed);
1117
1118 journal->j_free += freed;
1119 journal->j_tail_sequence = tid;
1120 journal->j_tail = block;
1121 write_unlock(&journal->j_state_lock);
1122
1123out:
1124 return ret;
1125}
1126
1127/*
1128 * This is a variation of __jbd2_update_log_tail which checks for validity of
1129 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1130 * with other threads updating log tail.
1131 */
1132void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1133{
1134 mutex_lock_io(&journal->j_checkpoint_mutex);
1135 if (tid_gt(tid, journal->j_tail_sequence))
1136 __jbd2_update_log_tail(journal, tid, block);
1137 mutex_unlock(&journal->j_checkpoint_mutex);
1138}
1139
1140struct jbd2_stats_proc_session {
1141 journal_t *journal;
1142 struct transaction_stats_s *stats;
1143 int start;
1144 int max;
1145};
1146
1147static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1148{
1149 return *pos ? NULL : SEQ_START_TOKEN;
1150}
1151
1152static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1153{
1154 (*pos)++;
1155 return NULL;
1156}
1157
1158static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1159{
1160 struct jbd2_stats_proc_session *s = seq->private;
1161
1162 if (v != SEQ_START_TOKEN)
1163 return 0;
1164 seq_printf(seq, "%lu transactions (%lu requested), "
1165 "each up to %u blocks\n",
1166 s->stats->ts_tid, s->stats->ts_requested,
1167 s->journal->j_max_transaction_buffers);
1168 if (s->stats->ts_tid == 0)
1169 return 0;
1170 seq_printf(seq, "average: \n %ums waiting for transaction\n",
1171 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1172 seq_printf(seq, " %ums request delay\n",
1173 (s->stats->ts_requested == 0) ? 0 :
1174 jiffies_to_msecs(s->stats->run.rs_request_delay /
1175 s->stats->ts_requested));
1176 seq_printf(seq, " %ums running transaction\n",
1177 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1178 seq_printf(seq, " %ums transaction was being locked\n",
1179 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1180 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
1181 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1182 seq_printf(seq, " %ums logging transaction\n",
1183 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1184 seq_printf(seq, " %lluus average transaction commit time\n",
1185 div_u64(s->journal->j_average_commit_time, 1000));
1186 seq_printf(seq, " %lu handles per transaction\n",
1187 s->stats->run.rs_handle_count / s->stats->ts_tid);
1188 seq_printf(seq, " %lu blocks per transaction\n",
1189 s->stats->run.rs_blocks / s->stats->ts_tid);
1190 seq_printf(seq, " %lu logged blocks per transaction\n",
1191 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1192 return 0;
1193}
1194
1195static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1196{
1197}
1198
1199static const struct seq_operations jbd2_seq_info_ops = {
1200 .start = jbd2_seq_info_start,
1201 .next = jbd2_seq_info_next,
1202 .stop = jbd2_seq_info_stop,
1203 .show = jbd2_seq_info_show,
1204};
1205
1206static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1207{
1208 journal_t *journal = pde_data(inode);
1209 struct jbd2_stats_proc_session *s;
1210 int rc, size;
1211
1212 s = kmalloc(sizeof(*s), GFP_KERNEL);
1213 if (s == NULL)
1214 return -ENOMEM;
1215 size = sizeof(struct transaction_stats_s);
1216 s->stats = kmalloc(size, GFP_KERNEL);
1217 if (s->stats == NULL) {
1218 kfree(s);
1219 return -ENOMEM;
1220 }
1221 spin_lock(&journal->j_history_lock);
1222 memcpy(s->stats, &journal->j_stats, size);
1223 s->journal = journal;
1224 spin_unlock(&journal->j_history_lock);
1225
1226 rc = seq_open(file, &jbd2_seq_info_ops);
1227 if (rc == 0) {
1228 struct seq_file *m = file->private_data;
1229 m->private = s;
1230 } else {
1231 kfree(s->stats);
1232 kfree(s);
1233 }
1234 return rc;
1235
1236}
1237
1238static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1239{
1240 struct seq_file *seq = file->private_data;
1241 struct jbd2_stats_proc_session *s = seq->private;
1242 kfree(s->stats);
1243 kfree(s);
1244 return seq_release(inode, file);
1245}
1246
1247static const struct proc_ops jbd2_info_proc_ops = {
1248 .proc_open = jbd2_seq_info_open,
1249 .proc_read = seq_read,
1250 .proc_lseek = seq_lseek,
1251 .proc_release = jbd2_seq_info_release,
1252};
1253
1254static struct proc_dir_entry *proc_jbd2_stats;
1255
1256static void jbd2_stats_proc_init(journal_t *journal)
1257{
1258 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1259 if (journal->j_proc_entry) {
1260 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1261 &jbd2_info_proc_ops, journal);
1262 }
1263}
1264
1265static void jbd2_stats_proc_exit(journal_t *journal)
1266{
1267 remove_proc_entry("info", journal->j_proc_entry);
1268 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1269}
1270
1271/* Minimum size of descriptor tag */
1272static int jbd2_min_tag_size(void)
1273{
1274 /*
1275 * Tag with 32-bit block numbers does not use last four bytes of the
1276 * structure
1277 */
1278 return sizeof(journal_block_tag_t) - 4;
1279}
1280
1281/**
1282 * jbd2_journal_shrink_scan()
1283 * @shrink: shrinker to work on
1284 * @sc: reclaim request to process
1285 *
1286 * Scan the checkpointed buffer on the checkpoint list and release the
1287 * journal_head.
1288 */
1289static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1290 struct shrink_control *sc)
1291{
1292 journal_t *journal = shrink->private_data;
1293 unsigned long nr_to_scan = sc->nr_to_scan;
1294 unsigned long nr_shrunk;
1295 unsigned long count;
1296
1297 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1298 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1299
1300 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1301
1302 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1303 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1304
1305 return nr_shrunk;
1306}
1307
1308/**
1309 * jbd2_journal_shrink_count()
1310 * @shrink: shrinker to work on
1311 * @sc: reclaim request to process
1312 *
1313 * Count the number of checkpoint buffers on the checkpoint list.
1314 */
1315static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1316 struct shrink_control *sc)
1317{
1318 journal_t *journal = shrink->private_data;
1319 unsigned long count;
1320
1321 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1322 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1323
1324 return count;
1325}
1326
1327/*
1328 * If the journal init or create aborts, we need to mark the journal
1329 * superblock as being NULL to prevent the journal destroy from writing
1330 * back a bogus superblock.
1331 */
1332static void journal_fail_superblock(journal_t *journal)
1333{
1334 struct buffer_head *bh = journal->j_sb_buffer;
1335 brelse(bh);
1336 journal->j_sb_buffer = NULL;
1337}
1338
1339/*
1340 * Check the superblock for a given journal, performing initial
1341 * validation of the format.
1342 */
1343static int journal_check_superblock(journal_t *journal)
1344{
1345 journal_superblock_t *sb = journal->j_superblock;
1346 int num_fc_blks;
1347 int err = -EINVAL;
1348
1349 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1350 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1351 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1352 return err;
1353 }
1354
1355 if (be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V1 &&
1356 be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V2) {
1357 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1358 return err;
1359 }
1360
1361 if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1362 printk(KERN_WARNING "JBD2: journal file too short\n");
1363 return err;
1364 }
1365
1366 if (be32_to_cpu(sb->s_first) == 0 ||
1367 be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1368 printk(KERN_WARNING
1369 "JBD2: Invalid start block of journal: %u\n",
1370 be32_to_cpu(sb->s_first));
1371 return err;
1372 }
1373
1374 /*
1375 * If this is a V2 superblock, then we have to check the
1376 * features flags on it.
1377 */
1378 if (!jbd2_format_support_feature(journal))
1379 return 0;
1380
1381 if ((sb->s_feature_ro_compat &
1382 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1383 (sb->s_feature_incompat &
1384 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1385 printk(KERN_WARNING "JBD2: Unrecognised features on journal\n");
1386 return err;
1387 }
1388
1389 num_fc_blks = jbd2_has_feature_fast_commit(journal) ?
1390 jbd2_journal_get_num_fc_blks(sb) : 0;
1391 if (be32_to_cpu(sb->s_maxlen) < JBD2_MIN_JOURNAL_BLOCKS ||
1392 be32_to_cpu(sb->s_maxlen) - JBD2_MIN_JOURNAL_BLOCKS < num_fc_blks) {
1393 printk(KERN_ERR "JBD2: journal file too short %u,%d\n",
1394 be32_to_cpu(sb->s_maxlen), num_fc_blks);
1395 return err;
1396 }
1397
1398 if (jbd2_has_feature_csum2(journal) &&
1399 jbd2_has_feature_csum3(journal)) {
1400 /* Can't have checksum v2 and v3 at the same time! */
1401 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1402 "at the same time!\n");
1403 return err;
1404 }
1405
1406 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1407 jbd2_has_feature_checksum(journal)) {
1408 /* Can't have checksum v1 and v2 on at the same time! */
1409 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1410 "at the same time!\n");
1411 return err;
1412 }
1413
1414 /* Load the checksum driver */
1415 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1416 if (sb->s_checksum_type != JBD2_CRC32C_CHKSUM) {
1417 printk(KERN_ERR "JBD2: Unknown checksum type\n");
1418 return err;
1419 }
1420
1421 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1422 if (IS_ERR(journal->j_chksum_driver)) {
1423 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1424 err = PTR_ERR(journal->j_chksum_driver);
1425 journal->j_chksum_driver = NULL;
1426 return err;
1427 }
1428 /* Check superblock checksum */
1429 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1430 printk(KERN_ERR "JBD2: journal checksum error\n");
1431 err = -EFSBADCRC;
1432 return err;
1433 }
1434 }
1435
1436 return 0;
1437}
1438
1439static int journal_revoke_records_per_block(journal_t *journal)
1440{
1441 int record_size;
1442 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1443
1444 if (jbd2_has_feature_64bit(journal))
1445 record_size = 8;
1446 else
1447 record_size = 4;
1448
1449 if (jbd2_journal_has_csum_v2or3(journal))
1450 space -= sizeof(struct jbd2_journal_block_tail);
1451 return space / record_size;
1452}
1453
1454/*
1455 * Load the on-disk journal superblock and read the key fields into the
1456 * journal_t.
1457 */
1458static int journal_load_superblock(journal_t *journal)
1459{
1460 int err;
1461 struct buffer_head *bh;
1462 journal_superblock_t *sb;
1463
1464 bh = getblk_unmovable(journal->j_dev, journal->j_blk_offset,
1465 journal->j_blocksize);
1466 if (bh)
1467 err = bh_read(bh, 0);
1468 if (!bh || err < 0) {
1469 pr_err("%s: Cannot read journal superblock\n", __func__);
1470 brelse(bh);
1471 return -EIO;
1472 }
1473
1474 journal->j_sb_buffer = bh;
1475 sb = (journal_superblock_t *)bh->b_data;
1476 journal->j_superblock = sb;
1477 err = journal_check_superblock(journal);
1478 if (err) {
1479 journal_fail_superblock(journal);
1480 return err;
1481 }
1482
1483 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1484 journal->j_tail = be32_to_cpu(sb->s_start);
1485 journal->j_first = be32_to_cpu(sb->s_first);
1486 journal->j_errno = be32_to_cpu(sb->s_errno);
1487 journal->j_last = be32_to_cpu(sb->s_maxlen);
1488
1489 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1490 journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1491 /* Precompute checksum seed for all metadata */
1492 if (jbd2_journal_has_csum_v2or3(journal))
1493 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1494 sizeof(sb->s_uuid));
1495 journal->j_revoke_records_per_block =
1496 journal_revoke_records_per_block(journal);
1497
1498 if (jbd2_has_feature_fast_commit(journal)) {
1499 journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
1500 journal->j_last = journal->j_fc_last -
1501 jbd2_journal_get_num_fc_blks(sb);
1502 journal->j_fc_first = journal->j_last + 1;
1503 journal->j_fc_off = 0;
1504 }
1505
1506 return 0;
1507}
1508
1509
1510/*
1511 * Management for journal control blocks: functions to create and
1512 * destroy journal_t structures, and to initialise and read existing
1513 * journal blocks from disk. */
1514
1515/* First: create and setup a journal_t object in memory. We initialise
1516 * very few fields yet: that has to wait until we have created the
1517 * journal structures from from scratch, or loaded them from disk. */
1518
1519static journal_t *journal_init_common(struct block_device *bdev,
1520 struct block_device *fs_dev,
1521 unsigned long long start, int len, int blocksize)
1522{
1523 static struct lock_class_key jbd2_trans_commit_key;
1524 journal_t *journal;
1525 int err;
1526 int n;
1527
1528 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1529 if (!journal)
1530 return ERR_PTR(-ENOMEM);
1531
1532 journal->j_blocksize = blocksize;
1533 journal->j_dev = bdev;
1534 journal->j_fs_dev = fs_dev;
1535 journal->j_blk_offset = start;
1536 journal->j_total_len = len;
1537 jbd2_init_fs_dev_write_error(journal);
1538
1539 err = journal_load_superblock(journal);
1540 if (err)
1541 goto err_cleanup;
1542
1543 init_waitqueue_head(&journal->j_wait_transaction_locked);
1544 init_waitqueue_head(&journal->j_wait_done_commit);
1545 init_waitqueue_head(&journal->j_wait_commit);
1546 init_waitqueue_head(&journal->j_wait_updates);
1547 init_waitqueue_head(&journal->j_wait_reserved);
1548 init_waitqueue_head(&journal->j_fc_wait);
1549 mutex_init(&journal->j_abort_mutex);
1550 mutex_init(&journal->j_barrier);
1551 mutex_init(&journal->j_checkpoint_mutex);
1552 spin_lock_init(&journal->j_revoke_lock);
1553 spin_lock_init(&journal->j_list_lock);
1554 spin_lock_init(&journal->j_history_lock);
1555 rwlock_init(&journal->j_state_lock);
1556
1557 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1558 journal->j_min_batch_time = 0;
1559 journal->j_max_batch_time = 15000; /* 15ms */
1560 atomic_set(&journal->j_reserved_credits, 0);
1561 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1562 &jbd2_trans_commit_key, 0);
1563
1564 /* The journal is marked for error until we succeed with recovery! */
1565 journal->j_flags = JBD2_ABORT;
1566
1567 /* Set up a default-sized revoke table for the new mount. */
1568 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1569 if (err)
1570 goto err_cleanup;
1571
1572 /*
1573 * journal descriptor can store up to n blocks, we need enough
1574 * buffers to write out full descriptor block.
1575 */
1576 err = -ENOMEM;
1577 n = journal->j_blocksize / jbd2_min_tag_size();
1578 journal->j_wbufsize = n;
1579 journal->j_fc_wbuf = NULL;
1580 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1581 GFP_KERNEL);
1582 if (!journal->j_wbuf)
1583 goto err_cleanup;
1584
1585 err = percpu_counter_init(&journal->j_checkpoint_jh_count, 0,
1586 GFP_KERNEL);
1587 if (err)
1588 goto err_cleanup;
1589
1590 journal->j_shrink_transaction = NULL;
1591
1592 journal->j_shrinker = shrinker_alloc(0, "jbd2-journal:(%u:%u)",
1593 MAJOR(bdev->bd_dev),
1594 MINOR(bdev->bd_dev));
1595 if (!journal->j_shrinker) {
1596 err = -ENOMEM;
1597 goto err_cleanup;
1598 }
1599
1600 journal->j_shrinker->scan_objects = jbd2_journal_shrink_scan;
1601 journal->j_shrinker->count_objects = jbd2_journal_shrink_count;
1602 journal->j_shrinker->batch = journal->j_max_transaction_buffers;
1603 journal->j_shrinker->private_data = journal;
1604
1605 shrinker_register(journal->j_shrinker);
1606
1607 return journal;
1608
1609err_cleanup:
1610 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1611 if (journal->j_chksum_driver)
1612 crypto_free_shash(journal->j_chksum_driver);
1613 kfree(journal->j_wbuf);
1614 jbd2_journal_destroy_revoke(journal);
1615 journal_fail_superblock(journal);
1616 kfree(journal);
1617 return ERR_PTR(err);
1618}
1619
1620/* jbd2_journal_init_dev and jbd2_journal_init_inode:
1621 *
1622 * Create a journal structure assigned some fixed set of disk blocks to
1623 * the journal. We don't actually touch those disk blocks yet, but we
1624 * need to set up all of the mapping information to tell the journaling
1625 * system where the journal blocks are.
1626 *
1627 */
1628
1629/**
1630 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1631 * @bdev: Block device on which to create the journal
1632 * @fs_dev: Device which hold journalled filesystem for this journal.
1633 * @start: Block nr Start of journal.
1634 * @len: Length of the journal in blocks.
1635 * @blocksize: blocksize of journalling device
1636 *
1637 * Returns: a newly created journal_t *
1638 *
1639 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1640 * range of blocks on an arbitrary block device.
1641 *
1642 */
1643journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1644 struct block_device *fs_dev,
1645 unsigned long long start, int len, int blocksize)
1646{
1647 journal_t *journal;
1648
1649 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1650 if (IS_ERR(journal))
1651 return ERR_CAST(journal);
1652
1653 snprintf(journal->j_devname, sizeof(journal->j_devname),
1654 "%pg", journal->j_dev);
1655 strreplace(journal->j_devname, '/', '!');
1656 jbd2_stats_proc_init(journal);
1657
1658 return journal;
1659}
1660
1661/**
1662 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1663 * @inode: An inode to create the journal in
1664 *
1665 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1666 * the journal. The inode must exist already, must support bmap() and
1667 * must have all data blocks preallocated.
1668 */
1669journal_t *jbd2_journal_init_inode(struct inode *inode)
1670{
1671 journal_t *journal;
1672 sector_t blocknr;
1673 int err = 0;
1674
1675 blocknr = 0;
1676 err = bmap(inode, &blocknr);
1677 if (err || !blocknr) {
1678 pr_err("%s: Cannot locate journal superblock\n", __func__);
1679 return err ? ERR_PTR(err) : ERR_PTR(-EINVAL);
1680 }
1681
1682 jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1683 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1684 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1685
1686 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1687 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1688 inode->i_sb->s_blocksize);
1689 if (IS_ERR(journal))
1690 return ERR_CAST(journal);
1691
1692 journal->j_inode = inode;
1693 snprintf(journal->j_devname, sizeof(journal->j_devname),
1694 "%pg-%lu", journal->j_dev, journal->j_inode->i_ino);
1695 strreplace(journal->j_devname, '/', '!');
1696 jbd2_stats_proc_init(journal);
1697
1698 return journal;
1699}
1700
1701/*
1702 * Given a journal_t structure, initialise the various fields for
1703 * startup of a new journaling session. We use this both when creating
1704 * a journal, and after recovering an old journal to reset it for
1705 * subsequent use.
1706 */
1707
1708static int journal_reset(journal_t *journal)
1709{
1710 journal_superblock_t *sb = journal->j_superblock;
1711 unsigned long long first, last;
1712
1713 first = be32_to_cpu(sb->s_first);
1714 last = be32_to_cpu(sb->s_maxlen);
1715 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1716 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1717 first, last);
1718 journal_fail_superblock(journal);
1719 return -EINVAL;
1720 }
1721
1722 journal->j_first = first;
1723 journal->j_last = last;
1724
1725 if (journal->j_head != 0 && journal->j_flags & JBD2_CYCLE_RECORD) {
1726 /*
1727 * Disable the cycled recording mode if the journal head block
1728 * number is not correct.
1729 */
1730 if (journal->j_head < first || journal->j_head >= last) {
1731 printk(KERN_WARNING "JBD2: Incorrect Journal head block %lu, "
1732 "disable journal_cycle_record\n",
1733 journal->j_head);
1734 journal->j_head = journal->j_first;
1735 }
1736 } else {
1737 journal->j_head = journal->j_first;
1738 }
1739 journal->j_tail = journal->j_head;
1740 journal->j_free = journal->j_last - journal->j_first;
1741
1742 journal->j_tail_sequence = journal->j_transaction_sequence;
1743 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1744 journal->j_commit_request = journal->j_commit_sequence;
1745
1746 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1747
1748 /*
1749 * Now that journal recovery is done, turn fast commits off here. This
1750 * way, if fast commit was enabled before the crash but if now FS has
1751 * disabled it, we don't enable fast commits.
1752 */
1753 jbd2_clear_feature_fast_commit(journal);
1754
1755 /*
1756 * As a special case, if the on-disk copy is already marked as needing
1757 * no recovery (s_start == 0), then we can safely defer the superblock
1758 * update until the next commit by setting JBD2_FLUSHED. This avoids
1759 * attempting a write to a potential-readonly device.
1760 */
1761 if (sb->s_start == 0) {
1762 jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1763 "(start %ld, seq %u, errno %d)\n",
1764 journal->j_tail, journal->j_tail_sequence,
1765 journal->j_errno);
1766 journal->j_flags |= JBD2_FLUSHED;
1767 } else {
1768 /* Lock here to make assertions happy... */
1769 mutex_lock_io(&journal->j_checkpoint_mutex);
1770 /*
1771 * Update log tail information. We use REQ_FUA since new
1772 * transaction will start reusing journal space and so we
1773 * must make sure information about current log tail is on
1774 * disk before that.
1775 */
1776 jbd2_journal_update_sb_log_tail(journal,
1777 journal->j_tail_sequence,
1778 journal->j_tail, REQ_FUA);
1779 mutex_unlock(&journal->j_checkpoint_mutex);
1780 }
1781 return jbd2_journal_start_thread(journal);
1782}
1783
1784/*
1785 * This function expects that the caller will have locked the journal
1786 * buffer head, and will return with it unlocked
1787 */
1788static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags)
1789{
1790 struct buffer_head *bh = journal->j_sb_buffer;
1791 journal_superblock_t *sb = journal->j_superblock;
1792 int ret = 0;
1793
1794 /* Buffer got discarded which means block device got invalidated */
1795 if (!buffer_mapped(bh)) {
1796 unlock_buffer(bh);
1797 return -EIO;
1798 }
1799
1800 /*
1801 * Always set high priority flags to exempt from block layer's
1802 * QOS policies, e.g. writeback throttle.
1803 */
1804 write_flags |= JBD2_JOURNAL_REQ_FLAGS;
1805 if (!(journal->j_flags & JBD2_BARRIER))
1806 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1807
1808 trace_jbd2_write_superblock(journal, write_flags);
1809
1810 if (buffer_write_io_error(bh)) {
1811 /*
1812 * Oh, dear. A previous attempt to write the journal
1813 * superblock failed. This could happen because the
1814 * USB device was yanked out. Or it could happen to
1815 * be a transient write error and maybe the block will
1816 * be remapped. Nothing we can do but to retry the
1817 * write and hope for the best.
1818 */
1819 printk(KERN_ERR "JBD2: previous I/O error detected "
1820 "for journal superblock update for %s.\n",
1821 journal->j_devname);
1822 clear_buffer_write_io_error(bh);
1823 set_buffer_uptodate(bh);
1824 }
1825 if (jbd2_journal_has_csum_v2or3(journal))
1826 sb->s_checksum = jbd2_superblock_csum(journal, sb);
1827 get_bh(bh);
1828 bh->b_end_io = end_buffer_write_sync;
1829 submit_bh(REQ_OP_WRITE | write_flags, bh);
1830 wait_on_buffer(bh);
1831 if (buffer_write_io_error(bh)) {
1832 clear_buffer_write_io_error(bh);
1833 set_buffer_uptodate(bh);
1834 ret = -EIO;
1835 }
1836 if (ret) {
1837 printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n",
1838 journal->j_devname);
1839 if (!is_journal_aborted(journal))
1840 jbd2_journal_abort(journal, ret);
1841 }
1842
1843 return ret;
1844}
1845
1846/**
1847 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1848 * @journal: The journal to update.
1849 * @tail_tid: TID of the new transaction at the tail of the log
1850 * @tail_block: The first block of the transaction at the tail of the log
1851 * @write_flags: Flags for the journal sb write operation
1852 *
1853 * Update a journal's superblock information about log tail and write it to
1854 * disk, waiting for the IO to complete.
1855 */
1856int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1857 unsigned long tail_block,
1858 blk_opf_t write_flags)
1859{
1860 journal_superblock_t *sb = journal->j_superblock;
1861 int ret;
1862
1863 if (is_journal_aborted(journal))
1864 return -EIO;
1865 if (jbd2_check_fs_dev_write_error(journal)) {
1866 jbd2_journal_abort(journal, -EIO);
1867 return -EIO;
1868 }
1869
1870 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1871 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1872 tail_block, tail_tid);
1873
1874 lock_buffer(journal->j_sb_buffer);
1875 sb->s_sequence = cpu_to_be32(tail_tid);
1876 sb->s_start = cpu_to_be32(tail_block);
1877
1878 ret = jbd2_write_superblock(journal, write_flags);
1879 if (ret)
1880 goto out;
1881
1882 /* Log is no longer empty */
1883 write_lock(&journal->j_state_lock);
1884 WARN_ON(!sb->s_sequence);
1885 journal->j_flags &= ~JBD2_FLUSHED;
1886 write_unlock(&journal->j_state_lock);
1887
1888out:
1889 return ret;
1890}
1891
1892/**
1893 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1894 * @journal: The journal to update.
1895 * @write_flags: Flags for the journal sb write operation
1896 *
1897 * Update a journal's dynamic superblock fields to show that journal is empty.
1898 * Write updated superblock to disk waiting for IO to complete.
1899 */
1900static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags)
1901{
1902 journal_superblock_t *sb = journal->j_superblock;
1903 bool had_fast_commit = false;
1904
1905 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1906 lock_buffer(journal->j_sb_buffer);
1907 if (sb->s_start == 0) { /* Is it already empty? */
1908 unlock_buffer(journal->j_sb_buffer);
1909 return;
1910 }
1911
1912 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1913 journal->j_tail_sequence);
1914
1915 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1916 sb->s_start = cpu_to_be32(0);
1917 sb->s_head = cpu_to_be32(journal->j_head);
1918 if (jbd2_has_feature_fast_commit(journal)) {
1919 /*
1920 * When journal is clean, no need to commit fast commit flag and
1921 * make file system incompatible with older kernels.
1922 */
1923 jbd2_clear_feature_fast_commit(journal);
1924 had_fast_commit = true;
1925 }
1926
1927 jbd2_write_superblock(journal, write_flags);
1928
1929 if (had_fast_commit)
1930 jbd2_set_feature_fast_commit(journal);
1931
1932 /* Log is no longer empty */
1933 write_lock(&journal->j_state_lock);
1934 journal->j_flags |= JBD2_FLUSHED;
1935 write_unlock(&journal->j_state_lock);
1936}
1937
1938/**
1939 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1940 * @journal: The journal to erase.
1941 * @flags: A discard/zeroout request is sent for each physically contigous
1942 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1943 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1944 * to perform.
1945 *
1946 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1947 * will be explicitly written if no hardware offload is available, see
1948 * blkdev_issue_zeroout for more details.
1949 */
1950static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1951{
1952 int err = 0;
1953 unsigned long block, log_offset; /* logical */
1954 unsigned long long phys_block, block_start, block_stop; /* physical */
1955 loff_t byte_start, byte_stop, byte_count;
1956
1957 /* flags must be set to either discard or zeroout */
1958 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1959 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1960 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1961 return -EINVAL;
1962
1963 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1964 !bdev_max_discard_sectors(journal->j_dev))
1965 return -EOPNOTSUPP;
1966
1967 /*
1968 * lookup block mapping and issue discard/zeroout for each
1969 * contiguous region
1970 */
1971 log_offset = be32_to_cpu(journal->j_superblock->s_first);
1972 block_start = ~0ULL;
1973 for (block = log_offset; block < journal->j_total_len; block++) {
1974 err = jbd2_journal_bmap(journal, block, &phys_block);
1975 if (err) {
1976 pr_err("JBD2: bad block at offset %lu", block);
1977 return err;
1978 }
1979
1980 if (block_start == ~0ULL) {
1981 block_start = phys_block;
1982 block_stop = block_start - 1;
1983 }
1984
1985 /*
1986 * last block not contiguous with current block,
1987 * process last contiguous region and return to this block on
1988 * next loop
1989 */
1990 if (phys_block != block_stop + 1) {
1991 block--;
1992 } else {
1993 block_stop++;
1994 /*
1995 * if this isn't the last block of journal,
1996 * no need to process now because next block may also
1997 * be part of this contiguous region
1998 */
1999 if (block != journal->j_total_len - 1)
2000 continue;
2001 }
2002
2003 /*
2004 * end of contiguous region or this is last block of journal,
2005 * take care of the region
2006 */
2007 byte_start = block_start * journal->j_blocksize;
2008 byte_stop = block_stop * journal->j_blocksize;
2009 byte_count = (block_stop - block_start + 1) *
2010 journal->j_blocksize;
2011
2012 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
2013 byte_start, byte_stop);
2014
2015 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
2016 err = blkdev_issue_discard(journal->j_dev,
2017 byte_start >> SECTOR_SHIFT,
2018 byte_count >> SECTOR_SHIFT,
2019 GFP_NOFS);
2020 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
2021 err = blkdev_issue_zeroout(journal->j_dev,
2022 byte_start >> SECTOR_SHIFT,
2023 byte_count >> SECTOR_SHIFT,
2024 GFP_NOFS, 0);
2025 }
2026
2027 if (unlikely(err != 0)) {
2028 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
2029 err, block_start, block_stop);
2030 return err;
2031 }
2032
2033 /* reset start and stop after processing a region */
2034 block_start = ~0ULL;
2035 }
2036
2037 return blkdev_issue_flush(journal->j_dev);
2038}
2039
2040/**
2041 * jbd2_journal_update_sb_errno() - Update error in the journal.
2042 * @journal: The journal to update.
2043 *
2044 * Update a journal's errno. Write updated superblock to disk waiting for IO
2045 * to complete.
2046 */
2047void jbd2_journal_update_sb_errno(journal_t *journal)
2048{
2049 journal_superblock_t *sb = journal->j_superblock;
2050 int errcode;
2051
2052 lock_buffer(journal->j_sb_buffer);
2053 errcode = journal->j_errno;
2054 if (errcode == -ESHUTDOWN)
2055 errcode = 0;
2056 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
2057 sb->s_errno = cpu_to_be32(errcode);
2058
2059 jbd2_write_superblock(journal, REQ_FUA);
2060}
2061EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
2062
2063/**
2064 * jbd2_journal_load() - Read journal from disk.
2065 * @journal: Journal to act on.
2066 *
2067 * Given a journal_t structure which tells us which disk blocks contain
2068 * a journal, read the journal from disk to initialise the in-memory
2069 * structures.
2070 */
2071int jbd2_journal_load(journal_t *journal)
2072{
2073 int err;
2074 journal_superblock_t *sb = journal->j_superblock;
2075
2076 /*
2077 * Create a slab for this blocksize
2078 */
2079 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2080 if (err)
2081 return err;
2082
2083 /* Let the recovery code check whether it needs to recover any
2084 * data from the journal. */
2085 err = jbd2_journal_recover(journal);
2086 if (err) {
2087 pr_warn("JBD2: journal recovery failed\n");
2088 return err;
2089 }
2090
2091 if (journal->j_failed_commit) {
2092 printk(KERN_ERR "JBD2: journal transaction %u on %s "
2093 "is corrupt.\n", journal->j_failed_commit,
2094 journal->j_devname);
2095 return -EFSCORRUPTED;
2096 }
2097 /*
2098 * clear JBD2_ABORT flag initialized in journal_init_common
2099 * here to update log tail information with the newest seq.
2100 */
2101 journal->j_flags &= ~JBD2_ABORT;
2102
2103 /* OK, we've finished with the dynamic journal bits:
2104 * reinitialise the dynamic contents of the superblock in memory
2105 * and reset them on disk. */
2106 err = journal_reset(journal);
2107 if (err) {
2108 pr_warn("JBD2: journal reset failed\n");
2109 return err;
2110 }
2111
2112 journal->j_flags |= JBD2_LOADED;
2113 return 0;
2114}
2115
2116/**
2117 * jbd2_journal_destroy() - Release a journal_t structure.
2118 * @journal: Journal to act on.
2119 *
2120 * Release a journal_t structure once it is no longer in use by the
2121 * journaled object.
2122 * Return <0 if we couldn't clean up the journal.
2123 */
2124int jbd2_journal_destroy(journal_t *journal)
2125{
2126 int err = 0;
2127
2128 /* Wait for the commit thread to wake up and die. */
2129 journal_kill_thread(journal);
2130
2131 /* Force a final log commit */
2132 if (journal->j_running_transaction)
2133 jbd2_journal_commit_transaction(journal);
2134
2135 /* Force any old transactions to disk */
2136
2137 /* Totally anal locking here... */
2138 spin_lock(&journal->j_list_lock);
2139 while (journal->j_checkpoint_transactions != NULL) {
2140 spin_unlock(&journal->j_list_lock);
2141 mutex_lock_io(&journal->j_checkpoint_mutex);
2142 err = jbd2_log_do_checkpoint(journal);
2143 mutex_unlock(&journal->j_checkpoint_mutex);
2144 /*
2145 * If checkpointing failed, just free the buffers to avoid
2146 * looping forever
2147 */
2148 if (err) {
2149 jbd2_journal_destroy_checkpoint(journal);
2150 spin_lock(&journal->j_list_lock);
2151 break;
2152 }
2153 spin_lock(&journal->j_list_lock);
2154 }
2155
2156 J_ASSERT(journal->j_running_transaction == NULL);
2157 J_ASSERT(journal->j_committing_transaction == NULL);
2158 J_ASSERT(journal->j_checkpoint_transactions == NULL);
2159 spin_unlock(&journal->j_list_lock);
2160
2161 /*
2162 * OK, all checkpoint transactions have been checked, now check the
2163 * writeback errseq of fs dev and abort the journal if some buffer
2164 * failed to write back to the original location, otherwise the
2165 * filesystem may become inconsistent.
2166 */
2167 if (!is_journal_aborted(journal) &&
2168 jbd2_check_fs_dev_write_error(journal))
2169 jbd2_journal_abort(journal, -EIO);
2170
2171 if (journal->j_sb_buffer) {
2172 if (!is_journal_aborted(journal)) {
2173 mutex_lock_io(&journal->j_checkpoint_mutex);
2174
2175 write_lock(&journal->j_state_lock);
2176 journal->j_tail_sequence =
2177 ++journal->j_transaction_sequence;
2178 write_unlock(&journal->j_state_lock);
2179
2180 jbd2_mark_journal_empty(journal, REQ_PREFLUSH | REQ_FUA);
2181 mutex_unlock(&journal->j_checkpoint_mutex);
2182 } else
2183 err = -EIO;
2184 brelse(journal->j_sb_buffer);
2185 }
2186
2187 if (journal->j_shrinker) {
2188 percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2189 shrinker_free(journal->j_shrinker);
2190 }
2191 if (journal->j_proc_entry)
2192 jbd2_stats_proc_exit(journal);
2193 iput(journal->j_inode);
2194 if (journal->j_revoke)
2195 jbd2_journal_destroy_revoke(journal);
2196 if (journal->j_chksum_driver)
2197 crypto_free_shash(journal->j_chksum_driver);
2198 kfree(journal->j_fc_wbuf);
2199 kfree(journal->j_wbuf);
2200 kfree(journal);
2201
2202 return err;
2203}
2204
2205
2206/**
2207 * jbd2_journal_check_used_features() - Check if features specified are used.
2208 * @journal: Journal to check.
2209 * @compat: bitmask of compatible features
2210 * @ro: bitmask of features that force read-only mount
2211 * @incompat: bitmask of incompatible features
2212 *
2213 * Check whether the journal uses all of a given set of
2214 * features. Return true (non-zero) if it does.
2215 **/
2216
2217int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2218 unsigned long ro, unsigned long incompat)
2219{
2220 journal_superblock_t *sb;
2221
2222 if (!compat && !ro && !incompat)
2223 return 1;
2224 if (!jbd2_format_support_feature(journal))
2225 return 0;
2226
2227 sb = journal->j_superblock;
2228
2229 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2230 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2231 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2232 return 1;
2233
2234 return 0;
2235}
2236
2237/**
2238 * jbd2_journal_check_available_features() - Check feature set in journalling layer
2239 * @journal: Journal to check.
2240 * @compat: bitmask of compatible features
2241 * @ro: bitmask of features that force read-only mount
2242 * @incompat: bitmask of incompatible features
2243 *
2244 * Check whether the journaling code supports the use of
2245 * all of a given set of features on this journal. Return true
2246 * (non-zero) if it can. */
2247
2248int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2249 unsigned long ro, unsigned long incompat)
2250{
2251 if (!compat && !ro && !incompat)
2252 return 1;
2253
2254 if (!jbd2_format_support_feature(journal))
2255 return 0;
2256
2257 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2258 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2259 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2260 return 1;
2261
2262 return 0;
2263}
2264
2265static int
2266jbd2_journal_initialize_fast_commit(journal_t *journal)
2267{
2268 journal_superblock_t *sb = journal->j_superblock;
2269 unsigned long long num_fc_blks;
2270
2271 num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2272 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2273 return -ENOSPC;
2274
2275 /* Are we called twice? */
2276 WARN_ON(journal->j_fc_wbuf != NULL);
2277 journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2278 sizeof(struct buffer_head *), GFP_KERNEL);
2279 if (!journal->j_fc_wbuf)
2280 return -ENOMEM;
2281
2282 journal->j_fc_wbufsize = num_fc_blks;
2283 journal->j_fc_last = journal->j_last;
2284 journal->j_last = journal->j_fc_last - num_fc_blks;
2285 journal->j_fc_first = journal->j_last + 1;
2286 journal->j_fc_off = 0;
2287 journal->j_free = journal->j_last - journal->j_first;
2288 journal->j_max_transaction_buffers =
2289 jbd2_journal_get_max_txn_bufs(journal);
2290
2291 return 0;
2292}
2293
2294/**
2295 * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2296 * @journal: Journal to act on.
2297 * @compat: bitmask of compatible features
2298 * @ro: bitmask of features that force read-only mount
2299 * @incompat: bitmask of incompatible features
2300 *
2301 * Mark a given journal feature as present on the
2302 * superblock. Returns true if the requested features could be set.
2303 *
2304 */
2305
2306int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2307 unsigned long ro, unsigned long incompat)
2308{
2309#define INCOMPAT_FEATURE_ON(f) \
2310 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2311#define COMPAT_FEATURE_ON(f) \
2312 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2313 journal_superblock_t *sb;
2314
2315 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2316 return 1;
2317
2318 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2319 return 0;
2320
2321 /* If enabling v2 checksums, turn on v3 instead */
2322 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2323 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2324 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2325 }
2326
2327 /* Asking for checksumming v3 and v1? Only give them v3. */
2328 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2329 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2330 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2331
2332 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2333 compat, ro, incompat);
2334
2335 sb = journal->j_superblock;
2336
2337 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2338 if (jbd2_journal_initialize_fast_commit(journal)) {
2339 pr_err("JBD2: Cannot enable fast commits.\n");
2340 return 0;
2341 }
2342 }
2343
2344 /* Load the checksum driver if necessary */
2345 if ((journal->j_chksum_driver == NULL) &&
2346 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2347 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2348 if (IS_ERR(journal->j_chksum_driver)) {
2349 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2350 journal->j_chksum_driver = NULL;
2351 return 0;
2352 }
2353 /* Precompute checksum seed for all metadata */
2354 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2355 sizeof(sb->s_uuid));
2356 }
2357
2358 lock_buffer(journal->j_sb_buffer);
2359
2360 /* If enabling v3 checksums, update superblock */
2361 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2362 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2363 sb->s_feature_compat &=
2364 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2365 }
2366
2367 /* If enabling v1 checksums, downgrade superblock */
2368 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2369 sb->s_feature_incompat &=
2370 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2371 JBD2_FEATURE_INCOMPAT_CSUM_V3);
2372
2373 sb->s_feature_compat |= cpu_to_be32(compat);
2374 sb->s_feature_ro_compat |= cpu_to_be32(ro);
2375 sb->s_feature_incompat |= cpu_to_be32(incompat);
2376 unlock_buffer(journal->j_sb_buffer);
2377 journal->j_revoke_records_per_block =
2378 journal_revoke_records_per_block(journal);
2379
2380 return 1;
2381#undef COMPAT_FEATURE_ON
2382#undef INCOMPAT_FEATURE_ON
2383}
2384
2385/*
2386 * jbd2_journal_clear_features() - Clear a given journal feature in the
2387 * superblock
2388 * @journal: Journal to act on.
2389 * @compat: bitmask of compatible features
2390 * @ro: bitmask of features that force read-only mount
2391 * @incompat: bitmask of incompatible features
2392 *
2393 * Clear a given journal feature as present on the
2394 * superblock.
2395 */
2396void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2397 unsigned long ro, unsigned long incompat)
2398{
2399 journal_superblock_t *sb;
2400
2401 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2402 compat, ro, incompat);
2403
2404 sb = journal->j_superblock;
2405
2406 sb->s_feature_compat &= ~cpu_to_be32(compat);
2407 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2408 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
2409 journal->j_revoke_records_per_block =
2410 journal_revoke_records_per_block(journal);
2411}
2412EXPORT_SYMBOL(jbd2_journal_clear_features);
2413
2414/**
2415 * jbd2_journal_flush() - Flush journal
2416 * @journal: Journal to act on.
2417 * @flags: optional operation on the journal blocks after the flush (see below)
2418 *
2419 * Flush all data for a given journal to disk and empty the journal.
2420 * Filesystems can use this when remounting readonly to ensure that
2421 * recovery does not need to happen on remount. Optionally, a discard or zeroout
2422 * can be issued on the journal blocks after flushing.
2423 *
2424 * flags:
2425 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2426 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2427 */
2428int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2429{
2430 int err = 0;
2431 transaction_t *transaction = NULL;
2432
2433 write_lock(&journal->j_state_lock);
2434
2435 /* Force everything buffered to the log... */
2436 if (journal->j_running_transaction) {
2437 transaction = journal->j_running_transaction;
2438 __jbd2_log_start_commit(journal, transaction->t_tid);
2439 } else if (journal->j_committing_transaction)
2440 transaction = journal->j_committing_transaction;
2441
2442 /* Wait for the log commit to complete... */
2443 if (transaction) {
2444 tid_t tid = transaction->t_tid;
2445
2446 write_unlock(&journal->j_state_lock);
2447 jbd2_log_wait_commit(journal, tid);
2448 } else {
2449 write_unlock(&journal->j_state_lock);
2450 }
2451
2452 /* ...and flush everything in the log out to disk. */
2453 spin_lock(&journal->j_list_lock);
2454 while (!err && journal->j_checkpoint_transactions != NULL) {
2455 spin_unlock(&journal->j_list_lock);
2456 mutex_lock_io(&journal->j_checkpoint_mutex);
2457 err = jbd2_log_do_checkpoint(journal);
2458 mutex_unlock(&journal->j_checkpoint_mutex);
2459 spin_lock(&journal->j_list_lock);
2460 }
2461 spin_unlock(&journal->j_list_lock);
2462
2463 if (is_journal_aborted(journal))
2464 return -EIO;
2465
2466 mutex_lock_io(&journal->j_checkpoint_mutex);
2467 if (!err) {
2468 err = jbd2_cleanup_journal_tail(journal);
2469 if (err < 0) {
2470 mutex_unlock(&journal->j_checkpoint_mutex);
2471 goto out;
2472 }
2473 err = 0;
2474 }
2475
2476 /* Finally, mark the journal as really needing no recovery.
2477 * This sets s_start==0 in the underlying superblock, which is
2478 * the magic code for a fully-recovered superblock. Any future
2479 * commits of data to the journal will restore the current
2480 * s_start value. */
2481 jbd2_mark_journal_empty(journal, REQ_FUA);
2482
2483 if (flags)
2484 err = __jbd2_journal_erase(journal, flags);
2485
2486 mutex_unlock(&journal->j_checkpoint_mutex);
2487 write_lock(&journal->j_state_lock);
2488 J_ASSERT(!journal->j_running_transaction);
2489 J_ASSERT(!journal->j_committing_transaction);
2490 J_ASSERT(!journal->j_checkpoint_transactions);
2491 J_ASSERT(journal->j_head == journal->j_tail);
2492 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2493 write_unlock(&journal->j_state_lock);
2494out:
2495 return err;
2496}
2497
2498/**
2499 * jbd2_journal_wipe() - Wipe journal contents
2500 * @journal: Journal to act on.
2501 * @write: flag (see below)
2502 *
2503 * Wipe out all of the contents of a journal, safely. This will produce
2504 * a warning if the journal contains any valid recovery information.
2505 * Must be called between journal_init_*() and jbd2_journal_load().
2506 *
2507 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2508 * we merely suppress recovery.
2509 */
2510
2511int jbd2_journal_wipe(journal_t *journal, int write)
2512{
2513 int err;
2514
2515 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2516
2517 if (!journal->j_tail)
2518 return 0;
2519
2520 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2521 write ? "Clearing" : "Ignoring");
2522
2523 err = jbd2_journal_skip_recovery(journal);
2524 if (write) {
2525 /* Lock to make assertions happy... */
2526 mutex_lock_io(&journal->j_checkpoint_mutex);
2527 jbd2_mark_journal_empty(journal, REQ_FUA);
2528 mutex_unlock(&journal->j_checkpoint_mutex);
2529 }
2530
2531 return err;
2532}
2533
2534/**
2535 * jbd2_journal_abort () - Shutdown the journal immediately.
2536 * @journal: the journal to shutdown.
2537 * @errno: an error number to record in the journal indicating
2538 * the reason for the shutdown.
2539 *
2540 * Perform a complete, immediate shutdown of the ENTIRE
2541 * journal (not of a single transaction). This operation cannot be
2542 * undone without closing and reopening the journal.
2543 *
2544 * The jbd2_journal_abort function is intended to support higher level error
2545 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2546 * mode.
2547 *
2548 * Journal abort has very specific semantics. Any existing dirty,
2549 * unjournaled buffers in the main filesystem will still be written to
2550 * disk by bdflush, but the journaling mechanism will be suspended
2551 * immediately and no further transaction commits will be honoured.
2552 *
2553 * Any dirty, journaled buffers will be written back to disk without
2554 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2555 * filesystem, but we _do_ attempt to leave as much data as possible
2556 * behind for fsck to use for cleanup.
2557 *
2558 * Any attempt to get a new transaction handle on a journal which is in
2559 * ABORT state will just result in an -EROFS error return. A
2560 * jbd2_journal_stop on an existing handle will return -EIO if we have
2561 * entered abort state during the update.
2562 *
2563 * Recursive transactions are not disturbed by journal abort until the
2564 * final jbd2_journal_stop, which will receive the -EIO error.
2565 *
2566 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2567 * which will be recorded (if possible) in the journal superblock. This
2568 * allows a client to record failure conditions in the middle of a
2569 * transaction without having to complete the transaction to record the
2570 * failure to disk. ext3_error, for example, now uses this
2571 * functionality.
2572 *
2573 */
2574
2575void jbd2_journal_abort(journal_t *journal, int errno)
2576{
2577 transaction_t *transaction;
2578
2579 /*
2580 * Lock the aborting procedure until everything is done, this avoid
2581 * races between filesystem's error handling flow (e.g. ext4_abort()),
2582 * ensure panic after the error info is written into journal's
2583 * superblock.
2584 */
2585 mutex_lock(&journal->j_abort_mutex);
2586 /*
2587 * ESHUTDOWN always takes precedence because a file system check
2588 * caused by any other journal abort error is not required after
2589 * a shutdown triggered.
2590 */
2591 write_lock(&journal->j_state_lock);
2592 if (journal->j_flags & JBD2_ABORT) {
2593 int old_errno = journal->j_errno;
2594
2595 write_unlock(&journal->j_state_lock);
2596 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2597 journal->j_errno = errno;
2598 jbd2_journal_update_sb_errno(journal);
2599 }
2600 mutex_unlock(&journal->j_abort_mutex);
2601 return;
2602 }
2603
2604 /*
2605 * Mark the abort as occurred and start current running transaction
2606 * to release all journaled buffer.
2607 */
2608 pr_err("Aborting journal on device %s.\n", journal->j_devname);
2609
2610 journal->j_flags |= JBD2_ABORT;
2611 journal->j_errno = errno;
2612 transaction = journal->j_running_transaction;
2613 if (transaction)
2614 __jbd2_log_start_commit(journal, transaction->t_tid);
2615 write_unlock(&journal->j_state_lock);
2616
2617 /*
2618 * Record errno to the journal super block, so that fsck and jbd2
2619 * layer could realise that a filesystem check is needed.
2620 */
2621 jbd2_journal_update_sb_errno(journal);
2622 mutex_unlock(&journal->j_abort_mutex);
2623}
2624
2625/**
2626 * jbd2_journal_errno() - returns the journal's error state.
2627 * @journal: journal to examine.
2628 *
2629 * This is the errno number set with jbd2_journal_abort(), the last
2630 * time the journal was mounted - if the journal was stopped
2631 * without calling abort this will be 0.
2632 *
2633 * If the journal has been aborted on this mount time -EROFS will
2634 * be returned.
2635 */
2636int jbd2_journal_errno(journal_t *journal)
2637{
2638 int err;
2639
2640 read_lock(&journal->j_state_lock);
2641 if (journal->j_flags & JBD2_ABORT)
2642 err = -EROFS;
2643 else
2644 err = journal->j_errno;
2645 read_unlock(&journal->j_state_lock);
2646 return err;
2647}
2648
2649/**
2650 * jbd2_journal_clear_err() - clears the journal's error state
2651 * @journal: journal to act on.
2652 *
2653 * An error must be cleared or acked to take a FS out of readonly
2654 * mode.
2655 */
2656int jbd2_journal_clear_err(journal_t *journal)
2657{
2658 int err = 0;
2659
2660 write_lock(&journal->j_state_lock);
2661 if (journal->j_flags & JBD2_ABORT)
2662 err = -EROFS;
2663 else
2664 journal->j_errno = 0;
2665 write_unlock(&journal->j_state_lock);
2666 return err;
2667}
2668
2669/**
2670 * jbd2_journal_ack_err() - Ack journal err.
2671 * @journal: journal to act on.
2672 *
2673 * An error must be cleared or acked to take a FS out of readonly
2674 * mode.
2675 */
2676void jbd2_journal_ack_err(journal_t *journal)
2677{
2678 write_lock(&journal->j_state_lock);
2679 if (journal->j_errno)
2680 journal->j_flags |= JBD2_ACK_ERR;
2681 write_unlock(&journal->j_state_lock);
2682}
2683
2684int jbd2_journal_blocks_per_page(struct inode *inode)
2685{
2686 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2687}
2688
2689/*
2690 * helper functions to deal with 32 or 64bit block numbers.
2691 */
2692size_t journal_tag_bytes(journal_t *journal)
2693{
2694 size_t sz;
2695
2696 if (jbd2_has_feature_csum3(journal))
2697 return sizeof(journal_block_tag3_t);
2698
2699 sz = sizeof(journal_block_tag_t);
2700
2701 if (jbd2_has_feature_csum2(journal))
2702 sz += sizeof(__u16);
2703
2704 if (jbd2_has_feature_64bit(journal))
2705 return sz;
2706 else
2707 return sz - sizeof(__u32);
2708}
2709
2710/*
2711 * JBD memory management
2712 *
2713 * These functions are used to allocate block-sized chunks of memory
2714 * used for making copies of buffer_head data. Very often it will be
2715 * page-sized chunks of data, but sometimes it will be in
2716 * sub-page-size chunks. (For example, 16k pages on Power systems
2717 * with a 4k block file system.) For blocks smaller than a page, we
2718 * use a SLAB allocator. There are slab caches for each block size,
2719 * which are allocated at mount time, if necessary, and we only free
2720 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2721 * this reason we don't need to a mutex to protect access to
2722 * jbd2_slab[] allocating or releasing memory; only in
2723 * jbd2_journal_create_slab().
2724 */
2725#define JBD2_MAX_SLABS 8
2726static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2727
2728static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2729 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2730 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2731};
2732
2733
2734static void jbd2_journal_destroy_slabs(void)
2735{
2736 int i;
2737
2738 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2739 kmem_cache_destroy(jbd2_slab[i]);
2740 jbd2_slab[i] = NULL;
2741 }
2742}
2743
2744static int jbd2_journal_create_slab(size_t size)
2745{
2746 static DEFINE_MUTEX(jbd2_slab_create_mutex);
2747 int i = order_base_2(size) - 10;
2748 size_t slab_size;
2749
2750 if (size == PAGE_SIZE)
2751 return 0;
2752
2753 if (i >= JBD2_MAX_SLABS)
2754 return -EINVAL;
2755
2756 if (unlikely(i < 0))
2757 i = 0;
2758 mutex_lock(&jbd2_slab_create_mutex);
2759 if (jbd2_slab[i]) {
2760 mutex_unlock(&jbd2_slab_create_mutex);
2761 return 0; /* Already created */
2762 }
2763
2764 slab_size = 1 << (i+10);
2765 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2766 slab_size, 0, NULL);
2767 mutex_unlock(&jbd2_slab_create_mutex);
2768 if (!jbd2_slab[i]) {
2769 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2770 return -ENOMEM;
2771 }
2772 return 0;
2773}
2774
2775static struct kmem_cache *get_slab(size_t size)
2776{
2777 int i = order_base_2(size) - 10;
2778
2779 BUG_ON(i >= JBD2_MAX_SLABS);
2780 if (unlikely(i < 0))
2781 i = 0;
2782 BUG_ON(jbd2_slab[i] == NULL);
2783 return jbd2_slab[i];
2784}
2785
2786void *jbd2_alloc(size_t size, gfp_t flags)
2787{
2788 void *ptr;
2789
2790 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2791
2792 if (size < PAGE_SIZE)
2793 ptr = kmem_cache_alloc(get_slab(size), flags);
2794 else
2795 ptr = (void *)__get_free_pages(flags, get_order(size));
2796
2797 /* Check alignment; SLUB has gotten this wrong in the past,
2798 * and this can lead to user data corruption! */
2799 BUG_ON(((unsigned long) ptr) & (size-1));
2800
2801 return ptr;
2802}
2803
2804void jbd2_free(void *ptr, size_t size)
2805{
2806 if (size < PAGE_SIZE)
2807 kmem_cache_free(get_slab(size), ptr);
2808 else
2809 free_pages((unsigned long)ptr, get_order(size));
2810};
2811
2812/*
2813 * Journal_head storage management
2814 */
2815static struct kmem_cache *jbd2_journal_head_cache;
2816#ifdef CONFIG_JBD2_DEBUG
2817static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2818#endif
2819
2820static int __init jbd2_journal_init_journal_head_cache(void)
2821{
2822 J_ASSERT(!jbd2_journal_head_cache);
2823 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2824 sizeof(struct journal_head),
2825 0, /* offset */
2826 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2827 NULL); /* ctor */
2828 if (!jbd2_journal_head_cache) {
2829 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2830 return -ENOMEM;
2831 }
2832 return 0;
2833}
2834
2835static void jbd2_journal_destroy_journal_head_cache(void)
2836{
2837 kmem_cache_destroy(jbd2_journal_head_cache);
2838 jbd2_journal_head_cache = NULL;
2839}
2840
2841/*
2842 * journal_head splicing and dicing
2843 */
2844static struct journal_head *journal_alloc_journal_head(void)
2845{
2846 struct journal_head *ret;
2847
2848#ifdef CONFIG_JBD2_DEBUG
2849 atomic_inc(&nr_journal_heads);
2850#endif
2851 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2852 if (!ret) {
2853 jbd2_debug(1, "out of memory for journal_head\n");
2854 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2855 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2856 GFP_NOFS | __GFP_NOFAIL);
2857 }
2858 if (ret)
2859 spin_lock_init(&ret->b_state_lock);
2860 return ret;
2861}
2862
2863static void journal_free_journal_head(struct journal_head *jh)
2864{
2865#ifdef CONFIG_JBD2_DEBUG
2866 atomic_dec(&nr_journal_heads);
2867 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2868#endif
2869 kmem_cache_free(jbd2_journal_head_cache, jh);
2870}
2871
2872/*
2873 * A journal_head is attached to a buffer_head whenever JBD has an
2874 * interest in the buffer.
2875 *
2876 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2877 * is set. This bit is tested in core kernel code where we need to take
2878 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2879 * there.
2880 *
2881 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2882 *
2883 * When a buffer has its BH_JBD bit set it is immune from being released by
2884 * core kernel code, mainly via ->b_count.
2885 *
2886 * A journal_head is detached from its buffer_head when the journal_head's
2887 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2888 * transaction (b_cp_transaction) hold their references to b_jcount.
2889 *
2890 * Various places in the kernel want to attach a journal_head to a buffer_head
2891 * _before_ attaching the journal_head to a transaction. To protect the
2892 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2893 * journal_head's b_jcount refcount by one. The caller must call
2894 * jbd2_journal_put_journal_head() to undo this.
2895 *
2896 * So the typical usage would be:
2897 *
2898 * (Attach a journal_head if needed. Increments b_jcount)
2899 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2900 * ...
2901 * (Get another reference for transaction)
2902 * jbd2_journal_grab_journal_head(bh);
2903 * jh->b_transaction = xxx;
2904 * (Put original reference)
2905 * jbd2_journal_put_journal_head(jh);
2906 */
2907
2908/*
2909 * Give a buffer_head a journal_head.
2910 *
2911 * May sleep.
2912 */
2913struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2914{
2915 struct journal_head *jh;
2916 struct journal_head *new_jh = NULL;
2917
2918repeat:
2919 if (!buffer_jbd(bh))
2920 new_jh = journal_alloc_journal_head();
2921
2922 jbd_lock_bh_journal_head(bh);
2923 if (buffer_jbd(bh)) {
2924 jh = bh2jh(bh);
2925 } else {
2926 J_ASSERT_BH(bh,
2927 (atomic_read(&bh->b_count) > 0) ||
2928 (bh->b_folio && bh->b_folio->mapping));
2929
2930 if (!new_jh) {
2931 jbd_unlock_bh_journal_head(bh);
2932 goto repeat;
2933 }
2934
2935 jh = new_jh;
2936 new_jh = NULL; /* We consumed it */
2937 set_buffer_jbd(bh);
2938 bh->b_private = jh;
2939 jh->b_bh = bh;
2940 get_bh(bh);
2941 BUFFER_TRACE(bh, "added journal_head");
2942 }
2943 jh->b_jcount++;
2944 jbd_unlock_bh_journal_head(bh);
2945 if (new_jh)
2946 journal_free_journal_head(new_jh);
2947 return bh->b_private;
2948}
2949
2950/*
2951 * Grab a ref against this buffer_head's journal_head. If it ended up not
2952 * having a journal_head, return NULL
2953 */
2954struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2955{
2956 struct journal_head *jh = NULL;
2957
2958 jbd_lock_bh_journal_head(bh);
2959 if (buffer_jbd(bh)) {
2960 jh = bh2jh(bh);
2961 jh->b_jcount++;
2962 }
2963 jbd_unlock_bh_journal_head(bh);
2964 return jh;
2965}
2966EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2967
2968static void __journal_remove_journal_head(struct buffer_head *bh)
2969{
2970 struct journal_head *jh = bh2jh(bh);
2971
2972 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2973 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2974 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2975 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2976 J_ASSERT_BH(bh, buffer_jbd(bh));
2977 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2978 BUFFER_TRACE(bh, "remove journal_head");
2979
2980 /* Unlink before dropping the lock */
2981 bh->b_private = NULL;
2982 jh->b_bh = NULL; /* debug, really */
2983 clear_buffer_jbd(bh);
2984}
2985
2986static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2987{
2988 if (jh->b_frozen_data) {
2989 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2990 jbd2_free(jh->b_frozen_data, b_size);
2991 }
2992 if (jh->b_committed_data) {
2993 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2994 jbd2_free(jh->b_committed_data, b_size);
2995 }
2996 journal_free_journal_head(jh);
2997}
2998
2999/*
3000 * Drop a reference on the passed journal_head. If it fell to zero then
3001 * release the journal_head from the buffer_head.
3002 */
3003void jbd2_journal_put_journal_head(struct journal_head *jh)
3004{
3005 struct buffer_head *bh = jh2bh(jh);
3006
3007 jbd_lock_bh_journal_head(bh);
3008 J_ASSERT_JH(jh, jh->b_jcount > 0);
3009 --jh->b_jcount;
3010 if (!jh->b_jcount) {
3011 __journal_remove_journal_head(bh);
3012 jbd_unlock_bh_journal_head(bh);
3013 journal_release_journal_head(jh, bh->b_size);
3014 __brelse(bh);
3015 } else {
3016 jbd_unlock_bh_journal_head(bh);
3017 }
3018}
3019EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3020
3021/*
3022 * Initialize jbd inode head
3023 */
3024void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3025{
3026 jinode->i_transaction = NULL;
3027 jinode->i_next_transaction = NULL;
3028 jinode->i_vfs_inode = inode;
3029 jinode->i_flags = 0;
3030 jinode->i_dirty_start = 0;
3031 jinode->i_dirty_end = 0;
3032 INIT_LIST_HEAD(&jinode->i_list);
3033}
3034
3035/*
3036 * Function to be called before we start removing inode from memory (i.e.,
3037 * clear_inode() is a fine place to be called from). It removes inode from
3038 * transaction's lists.
3039 */
3040void jbd2_journal_release_jbd_inode(journal_t *journal,
3041 struct jbd2_inode *jinode)
3042{
3043 if (!journal)
3044 return;
3045restart:
3046 spin_lock(&journal->j_list_lock);
3047 /* Is commit writing out inode - we have to wait */
3048 if (jinode->i_flags & JI_COMMIT_RUNNING) {
3049 wait_queue_head_t *wq;
3050 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3051 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3052 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3053 spin_unlock(&journal->j_list_lock);
3054 schedule();
3055 finish_wait(wq, &wait.wq_entry);
3056 goto restart;
3057 }
3058
3059 if (jinode->i_transaction) {
3060 list_del(&jinode->i_list);
3061 jinode->i_transaction = NULL;
3062 }
3063 spin_unlock(&journal->j_list_lock);
3064}
3065
3066
3067#ifdef CONFIG_PROC_FS
3068
3069#define JBD2_STATS_PROC_NAME "fs/jbd2"
3070
3071static void __init jbd2_create_jbd_stats_proc_entry(void)
3072{
3073 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3074}
3075
3076static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3077{
3078 if (proc_jbd2_stats)
3079 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3080}
3081
3082#else
3083
3084#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3085#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3086
3087#endif
3088
3089struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3090
3091static int __init jbd2_journal_init_inode_cache(void)
3092{
3093 J_ASSERT(!jbd2_inode_cache);
3094 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3095 if (!jbd2_inode_cache) {
3096 pr_emerg("JBD2: failed to create inode cache\n");
3097 return -ENOMEM;
3098 }
3099 return 0;
3100}
3101
3102static int __init jbd2_journal_init_handle_cache(void)
3103{
3104 J_ASSERT(!jbd2_handle_cache);
3105 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3106 if (!jbd2_handle_cache) {
3107 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3108 return -ENOMEM;
3109 }
3110 return 0;
3111}
3112
3113static void jbd2_journal_destroy_inode_cache(void)
3114{
3115 kmem_cache_destroy(jbd2_inode_cache);
3116 jbd2_inode_cache = NULL;
3117}
3118
3119static void jbd2_journal_destroy_handle_cache(void)
3120{
3121 kmem_cache_destroy(jbd2_handle_cache);
3122 jbd2_handle_cache = NULL;
3123}
3124
3125/*
3126 * Module startup and shutdown
3127 */
3128
3129static int __init journal_init_caches(void)
3130{
3131 int ret;
3132
3133 ret = jbd2_journal_init_revoke_record_cache();
3134 if (ret == 0)
3135 ret = jbd2_journal_init_revoke_table_cache();
3136 if (ret == 0)
3137 ret = jbd2_journal_init_journal_head_cache();
3138 if (ret == 0)
3139 ret = jbd2_journal_init_handle_cache();
3140 if (ret == 0)
3141 ret = jbd2_journal_init_inode_cache();
3142 if (ret == 0)
3143 ret = jbd2_journal_init_transaction_cache();
3144 return ret;
3145}
3146
3147static void jbd2_journal_destroy_caches(void)
3148{
3149 jbd2_journal_destroy_revoke_record_cache();
3150 jbd2_journal_destroy_revoke_table_cache();
3151 jbd2_journal_destroy_journal_head_cache();
3152 jbd2_journal_destroy_handle_cache();
3153 jbd2_journal_destroy_inode_cache();
3154 jbd2_journal_destroy_transaction_cache();
3155 jbd2_journal_destroy_slabs();
3156}
3157
3158static int __init journal_init(void)
3159{
3160 int ret;
3161
3162 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3163
3164 ret = journal_init_caches();
3165 if (ret == 0) {
3166 jbd2_create_jbd_stats_proc_entry();
3167 } else {
3168 jbd2_journal_destroy_caches();
3169 }
3170 return ret;
3171}
3172
3173static void __exit journal_exit(void)
3174{
3175#ifdef CONFIG_JBD2_DEBUG
3176 int n = atomic_read(&nr_journal_heads);
3177 if (n)
3178 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3179#endif
3180 jbd2_remove_jbd_stats_proc_entry();
3181 jbd2_journal_destroy_caches();
3182}
3183
3184MODULE_LICENSE("GPL");
3185module_init(journal_init);
3186module_exit(journal_exit);
3187