]> git.proxmox.com Git - mirror_ubuntu-kernels.git/blame - fs/jbd2/journal.c
Merge branches 'for-5.1/upstream-fixes', 'for-5.2/core', 'for-5.2/ish', 'for-5.2...
[mirror_ubuntu-kernels.git] / fs / jbd2 / journal.c
CommitLineData
f5166768 1// SPDX-License-Identifier: GPL-2.0+
470decc6 2/*
f7f4bccb 3 * linux/fs/jbd2/journal.c
470decc6
DK
4 *
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 *
7 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 *
470decc6
DK
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>
f7f4bccb 25#include <linux/jbd2.h>
470decc6
DK
26#include <linux/errno.h>
27#include <linux/slab.h>
470decc6
DK
28#include <linux/init.h>
29#include <linux/mm.h>
7dfb7103 30#include <linux/freezer.h>
470decc6
DK
31#include <linux/pagemap.h>
32#include <linux/kthread.h>
33#include <linux/poison.h>
34#include <linux/proc_fs.h>
8e85fb3f 35#include <linux/seq_file.h>
c225aa57 36#include <linux/math64.h>
879c5e6b 37#include <linux/hash.h>
d2eecb03
TT
38#include <linux/log2.h>
39#include <linux/vmalloc.h>
47def826 40#include <linux/backing-dev.h>
39e3ac25 41#include <linux/bitops.h>
670be5a7 42#include <linux/ratelimit.h>
eb52da3f 43#include <linux/sched/mm.h>
879c5e6b
TT
44
45#define CREATE_TRACE_POINTS
46#include <trace/events/jbd2.h>
470decc6 47
7c0f6ba6 48#include <linux/uaccess.h>
470decc6
DK
49#include <asm/page.h>
50
b6e96d00
TT
51#ifdef CONFIG_JBD2_DEBUG
52ushort jbd2_journal_enable_debug __read_mostly;
53EXPORT_SYMBOL(jbd2_journal_enable_debug);
54
55module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
56MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
57#endif
58
f7f4bccb
MC
59EXPORT_SYMBOL(jbd2_journal_extend);
60EXPORT_SYMBOL(jbd2_journal_stop);
61EXPORT_SYMBOL(jbd2_journal_lock_updates);
62EXPORT_SYMBOL(jbd2_journal_unlock_updates);
63EXPORT_SYMBOL(jbd2_journal_get_write_access);
64EXPORT_SYMBOL(jbd2_journal_get_create_access);
65EXPORT_SYMBOL(jbd2_journal_get_undo_access);
e06c8227 66EXPORT_SYMBOL(jbd2_journal_set_triggers);
f7f4bccb 67EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
f7f4bccb 68EXPORT_SYMBOL(jbd2_journal_forget);
470decc6
DK
69#if 0
70EXPORT_SYMBOL(journal_sync_buffer);
71#endif
f7f4bccb
MC
72EXPORT_SYMBOL(jbd2_journal_flush);
73EXPORT_SYMBOL(jbd2_journal_revoke);
74
75EXPORT_SYMBOL(jbd2_journal_init_dev);
76EXPORT_SYMBOL(jbd2_journal_init_inode);
f7f4bccb
MC
77EXPORT_SYMBOL(jbd2_journal_check_used_features);
78EXPORT_SYMBOL(jbd2_journal_check_available_features);
79EXPORT_SYMBOL(jbd2_journal_set_features);
f7f4bccb
MC
80EXPORT_SYMBOL(jbd2_journal_load);
81EXPORT_SYMBOL(jbd2_journal_destroy);
f7f4bccb
MC
82EXPORT_SYMBOL(jbd2_journal_abort);
83EXPORT_SYMBOL(jbd2_journal_errno);
84EXPORT_SYMBOL(jbd2_journal_ack_err);
85EXPORT_SYMBOL(jbd2_journal_clear_err);
86EXPORT_SYMBOL(jbd2_log_wait_commit);
3b799d15 87EXPORT_SYMBOL(jbd2_log_start_commit);
f7f4bccb
MC
88EXPORT_SYMBOL(jbd2_journal_start_commit);
89EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
90EXPORT_SYMBOL(jbd2_journal_wipe);
91EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
92EXPORT_SYMBOL(jbd2_journal_invalidatepage);
93EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
94EXPORT_SYMBOL(jbd2_journal_force_commit);
41617e1a
JK
95EXPORT_SYMBOL(jbd2_journal_inode_add_write);
96EXPORT_SYMBOL(jbd2_journal_inode_add_wait);
c851ed54
JK
97EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
98EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
99EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
8aefcd55 100EXPORT_SYMBOL(jbd2_inode_cache);
470decc6 101
470decc6 102static void __journal_abort_soft (journal_t *journal, int errno);
d2eecb03 103static int jbd2_journal_create_slab(size_t slab_size);
470decc6 104
169f1a2a
PG
105#ifdef CONFIG_JBD2_DEBUG
106void __jbd2_debug(int level, const char *file, const char *func,
107 unsigned int line, const char *fmt, ...)
108{
109 struct va_format vaf;
110 va_list args;
111
112 if (level > jbd2_journal_enable_debug)
113 return;
114 va_start(args, fmt);
115 vaf.fmt = fmt;
116 vaf.va = &args;
9196f571 117 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
169f1a2a
PG
118 va_end(args);
119}
120EXPORT_SYMBOL(__jbd2_debug);
121#endif
122
25ed6e8a 123/* Checksumming functions */
7747e6d0 124static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
25ed6e8a 125{
8595798c 126 if (!jbd2_journal_has_csum_v2or3_feature(j))
25ed6e8a
DW
127 return 1;
128
129 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
130}
131
18a6ea1e 132static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
4fd5ea43 133{
18a6ea1e
DW
134 __u32 csum;
135 __be32 old_csum;
4fd5ea43
DW
136
137 old_csum = sb->s_checksum;
138 sb->s_checksum = 0;
139 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
140 sb->s_checksum = old_csum;
141
142 return cpu_to_be32(csum);
143}
144
470decc6
DK
145/*
146 * Helper function used to manage commit timeouts
147 */
148
e3c95788 149static void commit_timeout(struct timer_list *t)
470decc6 150{
e3c95788 151 journal_t *journal = from_timer(journal, t, j_commit_timer);
470decc6 152
e3c95788 153 wake_up_process(journal->j_task);
470decc6
DK
154}
155
156/*
f7f4bccb 157 * kjournald2: The main thread function used to manage a logging device
470decc6
DK
158 * journal.
159 *
160 * This kernel thread is responsible for two things:
161 *
162 * 1) COMMIT: Every so often we need to commit the current state of the
163 * filesystem to disk. The journal thread is responsible for writing
164 * all of the metadata buffers to disk.
165 *
166 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
167 * of the data in that part of the log has been rewritten elsewhere on
168 * the disk. Flushing these old buffers to reclaim space in the log is
169 * known as checkpointing, and this thread is responsible for that job.
170 */
171
f7f4bccb 172static int kjournald2(void *arg)
470decc6
DK
173{
174 journal_t *journal = arg;
175 transaction_t *transaction;
176
177 /*
178 * Set up an interval timer which can be used to trigger a commit wakeup
179 * after the commit interval expires
180 */
e3c95788 181 timer_setup(&journal->j_commit_timer, commit_timeout, 0);
470decc6 182
35c80422
NC
183 set_freezable();
184
470decc6
DK
185 /* Record that the journal thread is running */
186 journal->j_task = current;
187 wake_up(&journal->j_wait_done_commit);
188
eb52da3f
MH
189 /*
190 * Make sure that no allocations from this kernel thread will ever
191 * recurse to the fs layer because we are responsible for the
192 * transaction commit and any fs involvement might get stuck waiting for
193 * the trasn. commit.
194 */
195 memalloc_nofs_save();
196
470decc6
DK
197 /*
198 * And now, wait forever for commit wakeup events.
199 */
a931da6a 200 write_lock(&journal->j_state_lock);
470decc6
DK
201
202loop:
f7f4bccb 203 if (journal->j_flags & JBD2_UNMOUNT)
470decc6
DK
204 goto end_loop;
205
206 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
207 journal->j_commit_sequence, journal->j_commit_request);
208
209 if (journal->j_commit_sequence != journal->j_commit_request) {
210 jbd_debug(1, "OK, requests differ\n");
a931da6a 211 write_unlock(&journal->j_state_lock);
470decc6 212 del_timer_sync(&journal->j_commit_timer);
f7f4bccb 213 jbd2_journal_commit_transaction(journal);
a931da6a 214 write_lock(&journal->j_state_lock);
470decc6
DK
215 goto loop;
216 }
217
218 wake_up(&journal->j_wait_done_commit);
219 if (freezing(current)) {
220 /*
221 * The simpler the better. Flushing journal isn't a
222 * good idea, because that depends on threads that may
223 * be already stopped.
224 */
f7f4bccb 225 jbd_debug(1, "Now suspending kjournald2\n");
a931da6a 226 write_unlock(&journal->j_state_lock);
a0acae0e 227 try_to_freeze();
a931da6a 228 write_lock(&journal->j_state_lock);
470decc6
DK
229 } else {
230 /*
231 * We assume on resume that commits are already there,
232 * so we don't sleep
233 */
234 DEFINE_WAIT(wait);
235 int should_sleep = 1;
236
237 prepare_to_wait(&journal->j_wait_commit, &wait,
238 TASK_INTERRUPTIBLE);
239 if (journal->j_commit_sequence != journal->j_commit_request)
240 should_sleep = 0;
241 transaction = journal->j_running_transaction;
242 if (transaction && time_after_eq(jiffies,
243 transaction->t_expires))
244 should_sleep = 0;
f7f4bccb 245 if (journal->j_flags & JBD2_UNMOUNT)
470decc6
DK
246 should_sleep = 0;
247 if (should_sleep) {
a931da6a 248 write_unlock(&journal->j_state_lock);
470decc6 249 schedule();
a931da6a 250 write_lock(&journal->j_state_lock);
470decc6
DK
251 }
252 finish_wait(&journal->j_wait_commit, &wait);
253 }
254
f7f4bccb 255 jbd_debug(1, "kjournald2 wakes\n");
470decc6
DK
256
257 /*
258 * Were we woken up by a commit wakeup event?
259 */
260 transaction = journal->j_running_transaction;
261 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
262 journal->j_commit_request = transaction->t_tid;
263 jbd_debug(1, "woke because of timeout\n");
264 }
265 goto loop;
266
267end_loop:
470decc6
DK
268 del_timer_sync(&journal->j_commit_timer);
269 journal->j_task = NULL;
270 wake_up(&journal->j_wait_done_commit);
271 jbd_debug(1, "Journal thread exiting.\n");
dbfcef6b 272 write_unlock(&journal->j_state_lock);
470decc6
DK
273 return 0;
274}
275
97f06784 276static int jbd2_journal_start_thread(journal_t *journal)
470decc6 277{
97f06784
PE
278 struct task_struct *t;
279
90576c0b
TT
280 t = kthread_run(kjournald2, journal, "jbd2/%s",
281 journal->j_devname);
97f06784
PE
282 if (IS_ERR(t))
283 return PTR_ERR(t);
284
1076d17a 285 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
97f06784 286 return 0;
470decc6
DK
287}
288
289static void journal_kill_thread(journal_t *journal)
290{
a931da6a 291 write_lock(&journal->j_state_lock);
f7f4bccb 292 journal->j_flags |= JBD2_UNMOUNT;
470decc6
DK
293
294 while (journal->j_task) {
a931da6a 295 write_unlock(&journal->j_state_lock);
3469a32a 296 wake_up(&journal->j_wait_commit);
1076d17a 297 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
a931da6a 298 write_lock(&journal->j_state_lock);
470decc6 299 }
a931da6a 300 write_unlock(&journal->j_state_lock);
470decc6
DK
301}
302
303/*
f7f4bccb 304 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
470decc6
DK
305 *
306 * Writes a metadata buffer to a given disk block. The actual IO is not
307 * performed but a new buffer_head is constructed which labels the data
308 * to be written with the correct destination disk block.
309 *
310 * Any magic-number escaping which needs to be done will cause a
311 * copy-out here. If the buffer happens to start with the
f7f4bccb 312 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
470decc6
DK
313 * magic number is only written to the log for descripter blocks. In
314 * this case, we copy the data and replace the first word with 0, and we
315 * return a result code which indicates that this buffer needs to be
316 * marked as an escaped buffer in the corresponding log descriptor
317 * block. The missing word can then be restored when the block is read
318 * during recovery.
319 *
320 * If the source buffer has already been modified by a new transaction
321 * since we took the last commit snapshot, we use the frozen copy of
f5113eff
JK
322 * that data for IO. If we end up using the existing buffer_head's data
323 * for the write, then we have to make sure nobody modifies it while the
324 * IO is in progress. do_get_write_access() handles this.
470decc6 325 *
f5113eff
JK
326 * The function returns a pointer to the buffer_head to be used for IO.
327 *
470decc6
DK
328 *
329 * Return value:
330 * <0: Error
331 * >=0: Finished OK
332 *
333 * On success:
334 * Bit 0 set == escape performed on the data
335 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
336 */
337
f7f4bccb 338int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
470decc6 339 struct journal_head *jh_in,
f5113eff
JK
340 struct buffer_head **bh_out,
341 sector_t blocknr)
470decc6
DK
342{
343 int need_copy_out = 0;
344 int done_copy_out = 0;
345 int do_escape = 0;
346 char *mapped_data;
347 struct buffer_head *new_bh;
470decc6
DK
348 struct page *new_page;
349 unsigned int new_offset;
350 struct buffer_head *bh_in = jh2bh(jh_in);
96577c43 351 journal_t *journal = transaction->t_journal;
470decc6
DK
352
353 /*
354 * The buffer really shouldn't be locked: only the current committing
355 * transaction is allowed to write it, so nobody else is allowed
356 * to do any IO.
357 *
358 * akpm: except if we're journalling data, and write() output is
359 * also part of a shared mapping, and another thread has
360 * decided to launch a writepage() against this buffer.
361 */
362 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
363
6ccaf3e2 364 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
47def826 365
96577c43 366 /* keep subsequent assertions sane */
96577c43 367 atomic_set(&new_bh->b_count, 1);
470decc6 368
f5113eff
JK
369 jbd_lock_bh_state(bh_in);
370repeat:
470decc6
DK
371 /*
372 * If a new transaction has already done a buffer copy-out, then
373 * we use that version of the data for the commit.
374 */
470decc6
DK
375 if (jh_in->b_frozen_data) {
376 done_copy_out = 1;
377 new_page = virt_to_page(jh_in->b_frozen_data);
378 new_offset = offset_in_page(jh_in->b_frozen_data);
379 } else {
380 new_page = jh2bh(jh_in)->b_page;
381 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
382 }
383
303a8f2a 384 mapped_data = kmap_atomic(new_page);
e06c8227 385 /*
13ceef09
JK
386 * Fire data frozen trigger if data already wasn't frozen. Do this
387 * before checking for escaping, as the trigger may modify the magic
388 * offset. If a copy-out happens afterwards, it will have the correct
389 * data in the buffer.
e06c8227 390 */
13ceef09
JK
391 if (!done_copy_out)
392 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
393 jh_in->b_triggers);
e06c8227 394
470decc6
DK
395 /*
396 * Check for escaping
397 */
398 if (*((__be32 *)(mapped_data + new_offset)) ==
f7f4bccb 399 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
470decc6
DK
400 need_copy_out = 1;
401 do_escape = 1;
402 }
303a8f2a 403 kunmap_atomic(mapped_data);
470decc6
DK
404
405 /*
406 * Do we need to do a data copy?
407 */
408 if (need_copy_out && !done_copy_out) {
409 char *tmp;
410
411 jbd_unlock_bh_state(bh_in);
af1e76d6 412 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
e6ec116b 413 if (!tmp) {
f5113eff 414 brelse(new_bh);
e6ec116b
TT
415 return -ENOMEM;
416 }
470decc6
DK
417 jbd_lock_bh_state(bh_in);
418 if (jh_in->b_frozen_data) {
af1e76d6 419 jbd2_free(tmp, bh_in->b_size);
470decc6
DK
420 goto repeat;
421 }
422
423 jh_in->b_frozen_data = tmp;
303a8f2a 424 mapped_data = kmap_atomic(new_page);
f5113eff 425 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
303a8f2a 426 kunmap_atomic(mapped_data);
470decc6
DK
427
428 new_page = virt_to_page(tmp);
429 new_offset = offset_in_page(tmp);
430 done_copy_out = 1;
e06c8227
JB
431
432 /*
433 * This isn't strictly necessary, as we're using frozen
434 * data for the escaping, but it keeps consistency with
435 * b_frozen_data usage.
436 */
437 jh_in->b_frozen_triggers = jh_in->b_triggers;
470decc6
DK
438 }
439
440 /*
441 * Did we need to do an escaping? Now we've done all the
442 * copying, we can finally do so.
443 */
444 if (do_escape) {
303a8f2a 445 mapped_data = kmap_atomic(new_page);
470decc6 446 *((unsigned int *)(mapped_data + new_offset)) = 0;
303a8f2a 447 kunmap_atomic(mapped_data);
470decc6
DK
448 }
449
470decc6 450 set_bh_page(new_bh, new_page, new_offset);
f5113eff
JK
451 new_bh->b_size = bh_in->b_size;
452 new_bh->b_bdev = journal->j_dev;
470decc6 453 new_bh->b_blocknr = blocknr;
b34090e5 454 new_bh->b_private = bh_in;
470decc6
DK
455 set_buffer_mapped(new_bh);
456 set_buffer_dirty(new_bh);
457
f5113eff 458 *bh_out = new_bh;
470decc6
DK
459
460 /*
461 * The to-be-written buffer needs to get moved to the io queue,
462 * and the original buffer whose contents we are shadowing or
463 * copying is moved to the transaction's shadow queue.
464 */
465 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
96577c43 466 spin_lock(&journal->j_list_lock);
467 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
468 spin_unlock(&journal->j_list_lock);
b34090e5 469 set_buffer_shadow(bh_in);
96577c43 470 jbd_unlock_bh_state(bh_in);
471
470decc6
DK
472 return do_escape | (done_copy_out << 1);
473}
474
475/*
476 * Allocation code for the journal file. Manage the space left in the
477 * journal, so that we can begin checkpointing when appropriate.
478 */
479
470decc6 480/*
e4471831
TT
481 * Called with j_state_lock locked for writing.
482 * Returns true if a transaction commit was started.
470decc6 483 */
f7f4bccb 484int __jbd2_log_start_commit(journal_t *journal, tid_t target)
470decc6 485{
e7b04ac0
ES
486 /* Return if the txn has already requested to be committed */
487 if (journal->j_commit_request == target)
488 return 0;
489
470decc6 490 /*
deeeaf13
TT
491 * The only transaction we can possibly wait upon is the
492 * currently running transaction (if it exists). Otherwise,
493 * the target tid must be an old one.
470decc6 494 */
deeeaf13
TT
495 if (journal->j_running_transaction &&
496 journal->j_running_transaction->t_tid == target) {
470decc6 497 /*
bcf3d0bc 498 * We want a new commit: OK, mark the request and wakeup the
470decc6
DK
499 * commit thread. We do _not_ do the commit ourselves.
500 */
501
502 journal->j_commit_request = target;
f2a44523 503 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
470decc6
DK
504 journal->j_commit_request,
505 journal->j_commit_sequence);
9fff24aa 506 journal->j_running_transaction->t_requested = jiffies;
470decc6
DK
507 wake_up(&journal->j_wait_commit);
508 return 1;
deeeaf13
TT
509 } else if (!tid_geq(journal->j_commit_request, target))
510 /* This should never happen, but if it does, preserve
511 the evidence before kjournald goes into a loop and
512 increments j_commit_sequence beyond all recognition. */
f2a44523 513 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
1be2add6
TT
514 journal->j_commit_request,
515 journal->j_commit_sequence,
516 target, journal->j_running_transaction ?
517 journal->j_running_transaction->t_tid : 0);
470decc6
DK
518 return 0;
519}
520
f7f4bccb 521int jbd2_log_start_commit(journal_t *journal, tid_t tid)
470decc6
DK
522{
523 int ret;
524
a931da6a 525 write_lock(&journal->j_state_lock);
f7f4bccb 526 ret = __jbd2_log_start_commit(journal, tid);
a931da6a 527 write_unlock(&journal->j_state_lock);
470decc6
DK
528 return ret;
529}
530
531/*
9ff86446
DM
532 * Force and wait any uncommitted transactions. We can only force the running
533 * transaction if we don't have an active handle, otherwise, we will deadlock.
534 * Returns: <0 in case of error,
535 * 0 if nothing to commit,
536 * 1 if transaction was successfully committed.
470decc6 537 */
9ff86446 538static int __jbd2_journal_force_commit(journal_t *journal)
470decc6
DK
539{
540 transaction_t *transaction = NULL;
541 tid_t tid;
9ff86446 542 int need_to_start = 0, ret = 0;
470decc6 543
a931da6a 544 read_lock(&journal->j_state_lock);
470decc6
DK
545 if (journal->j_running_transaction && !current->journal_info) {
546 transaction = journal->j_running_transaction;
e4471831
TT
547 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
548 need_to_start = 1;
470decc6
DK
549 } else if (journal->j_committing_transaction)
550 transaction = journal->j_committing_transaction;
551
552 if (!transaction) {
9ff86446 553 /* Nothing to commit */
a931da6a 554 read_unlock(&journal->j_state_lock);
9ff86446 555 return 0;
470decc6 556 }
470decc6 557 tid = transaction->t_tid;
a931da6a 558 read_unlock(&journal->j_state_lock);
e4471831
TT
559 if (need_to_start)
560 jbd2_log_start_commit(journal, tid);
9ff86446
DM
561 ret = jbd2_log_wait_commit(journal, tid);
562 if (!ret)
563 ret = 1;
564
565 return ret;
566}
567
568/**
569 * Force and wait upon a commit if the calling process is not within
570 * transaction. This is used for forcing out undo-protected data which contains
571 * bitmaps, when the fs is running out of space.
572 *
573 * @journal: journal to force
574 * Returns true if progress was made.
575 */
576int jbd2_journal_force_commit_nested(journal_t *journal)
577{
578 int ret;
579
580 ret = __jbd2_journal_force_commit(journal);
581 return ret > 0;
582}
583
584/**
585 * int journal_force_commit() - force any uncommitted transactions
586 * @journal: journal to force
587 *
588 * Caller want unconditional commit. We can only force the running transaction
589 * if we don't have an active handle, otherwise, we will deadlock.
590 */
591int jbd2_journal_force_commit(journal_t *journal)
592{
593 int ret;
594
595 J_ASSERT(!current->journal_info);
596 ret = __jbd2_journal_force_commit(journal);
597 if (ret > 0)
598 ret = 0;
599 return ret;
470decc6
DK
600}
601
602/*
603 * Start a commit of the current running transaction (if any). Returns true
c88ccea3
JK
604 * if a transaction is going to be committed (or is currently already
605 * committing), and fills its tid in at *ptid
470decc6 606 */
f7f4bccb 607int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
470decc6
DK
608{
609 int ret = 0;
610
a931da6a 611 write_lock(&journal->j_state_lock);
470decc6
DK
612 if (journal->j_running_transaction) {
613 tid_t tid = journal->j_running_transaction->t_tid;
614
c88ccea3
JK
615 __jbd2_log_start_commit(journal, tid);
616 /* There's a running transaction and we've just made sure
617 * it's commit has been scheduled. */
618 if (ptid)
470decc6 619 *ptid = tid;
c88ccea3
JK
620 ret = 1;
621 } else if (journal->j_committing_transaction) {
470decc6 622 /*
12810ad7
AB
623 * If commit has been started, then we have to wait for
624 * completion of that transaction.
470decc6 625 */
c88ccea3
JK
626 if (ptid)
627 *ptid = journal->j_committing_transaction->t_tid;
470decc6
DK
628 ret = 1;
629 }
a931da6a 630 write_unlock(&journal->j_state_lock);
470decc6
DK
631 return ret;
632}
633
bbd2be36
JK
634/*
635 * Return 1 if a given transaction has not yet sent barrier request
636 * connected with a transaction commit. If 0 is returned, transaction
637 * may or may not have sent the barrier. Used to avoid sending barrier
638 * twice in common cases.
639 */
640int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
641{
642 int ret = 0;
643 transaction_t *commit_trans;
644
645 if (!(journal->j_flags & JBD2_BARRIER))
646 return 0;
647 read_lock(&journal->j_state_lock);
648 /* Transaction already committed? */
649 if (tid_geq(journal->j_commit_sequence, tid))
650 goto out;
651 commit_trans = journal->j_committing_transaction;
652 if (!commit_trans || commit_trans->t_tid != tid) {
653 ret = 1;
654 goto out;
655 }
656 /*
657 * Transaction is being committed and we already proceeded to
658 * submitting a flush to fs partition?
659 */
660 if (journal->j_fs_dev != journal->j_dev) {
661 if (!commit_trans->t_need_data_flush ||
662 commit_trans->t_state >= T_COMMIT_DFLUSH)
663 goto out;
664 } else {
665 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
666 goto out;
667 }
668 ret = 1;
669out:
670 read_unlock(&journal->j_state_lock);
671 return ret;
672}
673EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
674
470decc6
DK
675/*
676 * Wait for a specified commit to complete.
677 * The caller may not hold the journal lock.
678 */
f7f4bccb 679int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
470decc6
DK
680{
681 int err = 0;
682
a931da6a 683 read_lock(&journal->j_state_lock);
c52c47e4
JK
684#ifdef CONFIG_PROVE_LOCKING
685 /*
686 * Some callers make sure transaction is already committing and in that
687 * case we cannot block on open handles anymore. So don't warn in that
688 * case.
689 */
690 if (tid_gt(tid, journal->j_commit_sequence) &&
691 (!journal->j_committing_transaction ||
692 journal->j_committing_transaction->t_tid != tid)) {
693 read_unlock(&journal->j_state_lock);
694 jbd2_might_wait_for_commit(journal);
695 read_lock(&journal->j_state_lock);
696 }
697#endif
e23291b9 698#ifdef CONFIG_JBD2_DEBUG
470decc6 699 if (!tid_geq(journal->j_commit_request, tid)) {
75685071 700 printk(KERN_ERR
470decc6 701 "%s: error: j_commit_request=%d, tid=%d\n",
329d291f 702 __func__, journal->j_commit_request, tid);
470decc6 703 }
470decc6 704#endif
470decc6 705 while (tid_gt(tid, journal->j_commit_sequence)) {
f2a44523 706 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
470decc6 707 tid, journal->j_commit_sequence);
a931da6a 708 read_unlock(&journal->j_state_lock);
3469a32a 709 wake_up(&journal->j_wait_commit);
470decc6
DK
710 wait_event(journal->j_wait_done_commit,
711 !tid_gt(tid, journal->j_commit_sequence));
a931da6a 712 read_lock(&journal->j_state_lock);
470decc6 713 }
a931da6a 714 read_unlock(&journal->j_state_lock);
470decc6 715
75685071 716 if (unlikely(is_journal_aborted(journal)))
470decc6 717 err = -EIO;
470decc6
DK
718 return err;
719}
720
b8a6176c
JK
721/* Return 1 when transaction with given tid has already committed. */
722int jbd2_transaction_committed(journal_t *journal, tid_t tid)
723{
724 int ret = 1;
725
726 read_lock(&journal->j_state_lock);
727 if (journal->j_running_transaction &&
728 journal->j_running_transaction->t_tid == tid)
729 ret = 0;
730 if (journal->j_committing_transaction &&
731 journal->j_committing_transaction->t_tid == tid)
732 ret = 0;
733 read_unlock(&journal->j_state_lock);
734 return ret;
735}
736EXPORT_SYMBOL(jbd2_transaction_committed);
737
d76a3a77
TT
738/*
739 * When this function returns the transaction corresponding to tid
740 * will be completed. If the transaction has currently running, start
741 * committing that transaction before waiting for it to complete. If
742 * the transaction id is stale, it is by definition already completed,
743 * so just return SUCCESS.
744 */
745int jbd2_complete_transaction(journal_t *journal, tid_t tid)
746{
747 int need_to_wait = 1;
748
749 read_lock(&journal->j_state_lock);
750 if (journal->j_running_transaction &&
751 journal->j_running_transaction->t_tid == tid) {
752 if (journal->j_commit_request != tid) {
753 /* transaction not yet started, so request it */
754 read_unlock(&journal->j_state_lock);
755 jbd2_log_start_commit(journal, tid);
756 goto wait_commit;
757 }
758 } else if (!(journal->j_committing_transaction &&
759 journal->j_committing_transaction->t_tid == tid))
760 need_to_wait = 0;
761 read_unlock(&journal->j_state_lock);
762 if (!need_to_wait)
763 return 0;
764wait_commit:
765 return jbd2_log_wait_commit(journal, tid);
766}
767EXPORT_SYMBOL(jbd2_complete_transaction);
768
470decc6
DK
769/*
770 * Log buffer allocation routines:
771 */
772
18eba7aa 773int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
470decc6
DK
774{
775 unsigned long blocknr;
776
a931da6a 777 write_lock(&journal->j_state_lock);
470decc6
DK
778 J_ASSERT(journal->j_free > 1);
779
780 blocknr = journal->j_head;
781 journal->j_head++;
782 journal->j_free--;
783 if (journal->j_head == journal->j_last)
784 journal->j_head = journal->j_first;
a931da6a 785 write_unlock(&journal->j_state_lock);
f7f4bccb 786 return jbd2_journal_bmap(journal, blocknr, retp);
470decc6
DK
787}
788
789/*
790 * Conversion of logical to physical block numbers for the journal
791 *
792 * On external journals the journal blocks are identity-mapped, so
793 * this is a no-op. If needed, we can use j_blk_offset - everything is
794 * ready.
795 */
f7f4bccb 796int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
18eba7aa 797 unsigned long long *retp)
470decc6
DK
798{
799 int err = 0;
18eba7aa 800 unsigned long long ret;
470decc6
DK
801
802 if (journal->j_inode) {
803 ret = bmap(journal->j_inode, blocknr);
804 if (ret)
805 *retp = ret;
806 else {
470decc6
DK
807 printk(KERN_ALERT "%s: journal block not found "
808 "at offset %lu on %s\n",
05496769 809 __func__, blocknr, journal->j_devname);
470decc6
DK
810 err = -EIO;
811 __journal_abort_soft(journal, err);
812 }
813 } else {
814 *retp = blocknr; /* +journal->j_blk_offset */
815 }
816 return err;
817}
818
819/*
820 * We play buffer_head aliasing tricks to write data/metadata blocks to
821 * the journal without copying their contents, but for journal
822 * descriptor blocks we do need to generate bona fide buffers.
823 *
f7f4bccb 824 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
470decc6
DK
825 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
826 * But we don't bother doing that, so there will be coherency problems with
827 * mmaps of blockdevs which hold live JBD-controlled filesystems.
828 */
32ab6715
JK
829struct buffer_head *
830jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
470decc6 831{
32ab6715 832 journal_t *journal = transaction->t_journal;
470decc6 833 struct buffer_head *bh;
18eba7aa 834 unsigned long long blocknr;
32ab6715 835 journal_header_t *header;
470decc6
DK
836 int err;
837
f7f4bccb 838 err = jbd2_journal_next_log_block(journal, &blocknr);
470decc6
DK
839
840 if (err)
841 return NULL;
842
843 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
4b905671
JK
844 if (!bh)
845 return NULL;
470decc6
DK
846 lock_buffer(bh);
847 memset(bh->b_data, 0, journal->j_blocksize);
32ab6715
JK
848 header = (journal_header_t *)bh->b_data;
849 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
850 header->h_blocktype = cpu_to_be32(type);
851 header->h_sequence = cpu_to_be32(transaction->t_tid);
470decc6
DK
852 set_buffer_uptodate(bh);
853 unlock_buffer(bh);
854 BUFFER_TRACE(bh, "return this buffer");
e5a120ae 855 return bh;
470decc6
DK
856}
857
1101cd4d
JK
858void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
859{
860 struct jbd2_journal_block_tail *tail;
861 __u32 csum;
862
863 if (!jbd2_journal_has_csum_v2or3(j))
864 return;
865
866 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
867 sizeof(struct jbd2_journal_block_tail));
868 tail->t_checksum = 0;
869 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
870 tail->t_checksum = cpu_to_be32(csum);
871}
872
79feb521
JK
873/*
874 * Return tid of the oldest transaction in the journal and block in the journal
875 * where the transaction starts.
876 *
877 * If the journal is now empty, return which will be the next transaction ID
878 * we will write and where will that transaction start.
879 *
880 * The return value is 0 if journal tail cannot be pushed any further, 1 if
881 * it can.
882 */
883int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
884 unsigned long *block)
885{
886 transaction_t *transaction;
887 int ret;
888
889 read_lock(&journal->j_state_lock);
890 spin_lock(&journal->j_list_lock);
891 transaction = journal->j_checkpoint_transactions;
892 if (transaction) {
893 *tid = transaction->t_tid;
894 *block = transaction->t_log_start;
895 } else if ((transaction = journal->j_committing_transaction) != NULL) {
896 *tid = transaction->t_tid;
897 *block = transaction->t_log_start;
898 } else if ((transaction = journal->j_running_transaction) != NULL) {
899 *tid = transaction->t_tid;
900 *block = journal->j_head;
901 } else {
902 *tid = journal->j_transaction_sequence;
903 *block = journal->j_head;
904 }
905 ret = tid_gt(*tid, journal->j_tail_sequence);
906 spin_unlock(&journal->j_list_lock);
907 read_unlock(&journal->j_state_lock);
908
909 return ret;
910}
911
912/*
913 * Update information in journal structure and in on disk journal superblock
914 * about log tail. This function does not check whether information passed in
915 * really pushes log tail further. It's responsibility of the caller to make
916 * sure provided log tail information is valid (e.g. by holding
917 * j_checkpoint_mutex all the time between computing log tail and calling this
918 * function as is the case with jbd2_cleanup_journal_tail()).
919 *
920 * Requires j_checkpoint_mutex
921 */
6f6a6fda 922int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
79feb521
JK
923{
924 unsigned long freed;
6f6a6fda 925 int ret;
79feb521
JK
926
927 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
928
929 /*
930 * We cannot afford for write to remain in drive's caches since as
931 * soon as we update j_tail, next transaction can start reusing journal
932 * space and if we lose sb update during power failure we'd replay
933 * old transaction with possibly newly overwritten data.
934 */
17f423b5
JK
935 ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
936 REQ_SYNC | REQ_FUA);
6f6a6fda
JQ
937 if (ret)
938 goto out;
939
79feb521
JK
940 write_lock(&journal->j_state_lock);
941 freed = block - journal->j_tail;
942 if (block < journal->j_tail)
943 freed += journal->j_last - journal->j_first;
944
945 trace_jbd2_update_log_tail(journal, tid, block, freed);
946 jbd_debug(1,
947 "Cleaning journal tail from %d to %d (offset %lu), "
948 "freeing %lu\n",
949 journal->j_tail_sequence, tid, block, freed);
950
951 journal->j_free += freed;
952 journal->j_tail_sequence = tid;
953 journal->j_tail = block;
954 write_unlock(&journal->j_state_lock);
6f6a6fda
JQ
955
956out:
957 return ret;
79feb521
JK
958}
959
3339578f 960/*
85e0c4e8 961 * This is a variation of __jbd2_update_log_tail which checks for validity of
3339578f
JK
962 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
963 * with other threads updating log tail.
964 */
965void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
966{
6fa7aa50 967 mutex_lock_io(&journal->j_checkpoint_mutex);
3339578f
JK
968 if (tid_gt(tid, journal->j_tail_sequence))
969 __jbd2_update_log_tail(journal, tid, block);
970 mutex_unlock(&journal->j_checkpoint_mutex);
971}
972
8e85fb3f
JL
973struct jbd2_stats_proc_session {
974 journal_t *journal;
975 struct transaction_stats_s *stats;
976 int start;
977 int max;
978};
979
8e85fb3f
JL
980static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
981{
982 return *pos ? NULL : SEQ_START_TOKEN;
983}
984
985static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
986{
987 return NULL;
988}
989
990static int jbd2_seq_info_show(struct seq_file *seq, void *v)
991{
992 struct jbd2_stats_proc_session *s = seq->private;
993
994 if (v != SEQ_START_TOKEN)
995 return 0;
9fff24aa
TT
996 seq_printf(seq, "%lu transactions (%lu requested), "
997 "each up to %u blocks\n",
998 s->stats->ts_tid, s->stats->ts_requested,
999 s->journal->j_max_transaction_buffers);
8e85fb3f
JL
1000 if (s->stats->ts_tid == 0)
1001 return 0;
1002 seq_printf(seq, "average: \n %ums waiting for transaction\n",
bf699327 1003 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
9fff24aa
TT
1004 seq_printf(seq, " %ums request delay\n",
1005 (s->stats->ts_requested == 0) ? 0 :
1006 jiffies_to_msecs(s->stats->run.rs_request_delay /
1007 s->stats->ts_requested));
8e85fb3f 1008 seq_printf(seq, " %ums running transaction\n",
bf699327 1009 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
8e85fb3f 1010 seq_printf(seq, " %ums transaction was being locked\n",
bf699327 1011 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
8e85fb3f 1012 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
bf699327 1013 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
8e85fb3f 1014 seq_printf(seq, " %ums logging transaction\n",
bf699327 1015 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
c225aa57
SHT
1016 seq_printf(seq, " %lluus average transaction commit time\n",
1017 div_u64(s->journal->j_average_commit_time, 1000));
8e85fb3f 1018 seq_printf(seq, " %lu handles per transaction\n",
bf699327 1019 s->stats->run.rs_handle_count / s->stats->ts_tid);
8e85fb3f 1020 seq_printf(seq, " %lu blocks per transaction\n",
bf699327 1021 s->stats->run.rs_blocks / s->stats->ts_tid);
8e85fb3f 1022 seq_printf(seq, " %lu logged blocks per transaction\n",
bf699327 1023 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
8e85fb3f
JL
1024 return 0;
1025}
1026
1027static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1028{
1029}
1030
88e9d34c 1031static const struct seq_operations jbd2_seq_info_ops = {
8e85fb3f
JL
1032 .start = jbd2_seq_info_start,
1033 .next = jbd2_seq_info_next,
1034 .stop = jbd2_seq_info_stop,
1035 .show = jbd2_seq_info_show,
1036};
1037
1038static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1039{
d9dda78b 1040 journal_t *journal = PDE_DATA(inode);
8e85fb3f
JL
1041 struct jbd2_stats_proc_session *s;
1042 int rc, size;
1043
1044 s = kmalloc(sizeof(*s), GFP_KERNEL);
1045 if (s == NULL)
1046 return -ENOMEM;
1047 size = sizeof(struct transaction_stats_s);
1048 s->stats = kmalloc(size, GFP_KERNEL);
1049 if (s->stats == NULL) {
1050 kfree(s);
1051 return -ENOMEM;
1052 }
1053 spin_lock(&journal->j_history_lock);
1054 memcpy(s->stats, &journal->j_stats, size);
1055 s->journal = journal;
1056 spin_unlock(&journal->j_history_lock);
1057
1058 rc = seq_open(file, &jbd2_seq_info_ops);
1059 if (rc == 0) {
1060 struct seq_file *m = file->private_data;
1061 m->private = s;
1062 } else {
1063 kfree(s->stats);
1064 kfree(s);
1065 }
1066 return rc;
1067
1068}
1069
1070static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1071{
1072 struct seq_file *seq = file->private_data;
1073 struct jbd2_stats_proc_session *s = seq->private;
1074 kfree(s->stats);
1075 kfree(s);
1076 return seq_release(inode, file);
1077}
1078
828c0950 1079static const struct file_operations jbd2_seq_info_fops = {
8e85fb3f
JL
1080 .owner = THIS_MODULE,
1081 .open = jbd2_seq_info_open,
1082 .read = seq_read,
1083 .llseek = seq_lseek,
1084 .release = jbd2_seq_info_release,
1085};
1086
1087static struct proc_dir_entry *proc_jbd2_stats;
1088
1089static void jbd2_stats_proc_init(journal_t *journal)
1090{
05496769 1091 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
8e85fb3f 1092 if (journal->j_proc_entry) {
79da3664
DL
1093 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1094 &jbd2_seq_info_fops, journal);
8e85fb3f
JL
1095 }
1096}
1097
1098static void jbd2_stats_proc_exit(journal_t *journal)
1099{
8e85fb3f 1100 remove_proc_entry("info", journal->j_proc_entry);
05496769 1101 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
8e85fb3f
JL
1102}
1103
470decc6
DK
1104/*
1105 * Management for journal control blocks: functions to create and
1106 * destroy journal_t structures, and to initialise and read existing
1107 * journal blocks from disk. */
1108
1109/* First: create and setup a journal_t object in memory. We initialise
1110 * very few fields yet: that has to wait until we have created the
1111 * journal structures from from scratch, or loaded them from disk. */
1112
f0c9fd54
GT
1113static journal_t *journal_init_common(struct block_device *bdev,
1114 struct block_device *fs_dev,
1115 unsigned long long start, int len, int blocksize)
470decc6 1116{
ab714aff 1117 static struct lock_class_key jbd2_trans_commit_key;
470decc6
DK
1118 journal_t *journal;
1119 int err;
f0c9fd54
GT
1120 struct buffer_head *bh;
1121 int n;
470decc6 1122
3ebfdf88 1123 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
470decc6 1124 if (!journal)
b7271b0a 1125 return NULL;
470decc6
DK
1126
1127 init_waitqueue_head(&journal->j_wait_transaction_locked);
470decc6 1128 init_waitqueue_head(&journal->j_wait_done_commit);
470decc6
DK
1129 init_waitqueue_head(&journal->j_wait_commit);
1130 init_waitqueue_head(&journal->j_wait_updates);
8f7d89f3 1131 init_waitqueue_head(&journal->j_wait_reserved);
470decc6
DK
1132 mutex_init(&journal->j_barrier);
1133 mutex_init(&journal->j_checkpoint_mutex);
1134 spin_lock_init(&journal->j_revoke_lock);
1135 spin_lock_init(&journal->j_list_lock);
a931da6a 1136 rwlock_init(&journal->j_state_lock);
470decc6 1137
cd02ff0b 1138 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
30773840
TT
1139 journal->j_min_batch_time = 0;
1140 journal->j_max_batch_time = 15000; /* 15ms */
8f7d89f3 1141 atomic_set(&journal->j_reserved_credits, 0);
470decc6
DK
1142
1143 /* The journal is marked for error until we succeed with recovery! */
f7f4bccb 1144 journal->j_flags = JBD2_ABORT;
470decc6
DK
1145
1146 /* Set up a default-sized revoke table for the new mount. */
f7f4bccb 1147 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
cd9cb405
EB
1148 if (err)
1149 goto err_cleanup;
8e85fb3f 1150
bf699327 1151 spin_lock_init(&journal->j_history_lock);
8e85fb3f 1152
ab714aff
JK
1153 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1154 &jbd2_trans_commit_key, 0);
1155
f0c9fd54
GT
1156 /* journal descriptor can store up to n blocks -bzzz */
1157 journal->j_blocksize = blocksize;
1158 journal->j_dev = bdev;
1159 journal->j_fs_dev = fs_dev;
1160 journal->j_blk_offset = start;
1161 journal->j_maxlen = len;
1162 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1163 journal->j_wbufsize = n;
1164 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1165 GFP_KERNEL);
cd9cb405
EB
1166 if (!journal->j_wbuf)
1167 goto err_cleanup;
f0c9fd54
GT
1168
1169 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1170 if (!bh) {
1171 pr_err("%s: Cannot get buffer for journal superblock\n",
1172 __func__);
cd9cb405 1173 goto err_cleanup;
f0c9fd54
GT
1174 }
1175 journal->j_sb_buffer = bh;
1176 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1177
470decc6 1178 return journal;
cd9cb405
EB
1179
1180err_cleanup:
1181 kfree(journal->j_wbuf);
1182 jbd2_journal_destroy_revoke(journal);
1183 kfree(journal);
1184 return NULL;
470decc6
DK
1185}
1186
f7f4bccb 1187/* jbd2_journal_init_dev and jbd2_journal_init_inode:
470decc6
DK
1188 *
1189 * Create a journal structure assigned some fixed set of disk blocks to
1190 * the journal. We don't actually touch those disk blocks yet, but we
1191 * need to set up all of the mapping information to tell the journaling
1192 * system where the journal blocks are.
1193 *
1194 */
1195
1196/**
5648ba5b 1197 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
470decc6
DK
1198 * @bdev: Block device on which to create the journal
1199 * @fs_dev: Device which hold journalled filesystem for this journal.
1200 * @start: Block nr Start of journal.
1201 * @len: Length of the journal in blocks.
1202 * @blocksize: blocksize of journalling device
5648ba5b
RD
1203 *
1204 * Returns: a newly created journal_t *
470decc6 1205 *
f7f4bccb 1206 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
470decc6
DK
1207 * range of blocks on an arbitrary block device.
1208 *
1209 */
f0c9fd54 1210journal_t *jbd2_journal_init_dev(struct block_device *bdev,
470decc6 1211 struct block_device *fs_dev,
18eba7aa 1212 unsigned long long start, int len, int blocksize)
470decc6 1213{
f0c9fd54 1214 journal_t *journal;
470decc6 1215
f0c9fd54 1216 journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
470decc6
DK
1217 if (!journal)
1218 return NULL;
1219
0587aa3d 1220 bdevname(journal->j_dev, journal->j_devname);
81ae394b 1221 strreplace(journal->j_devname, '/', '!');
4b905671 1222 jbd2_stats_proc_init(journal);
4b905671 1223
470decc6
DK
1224 return journal;
1225}
1226
1227/**
f7f4bccb 1228 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
470decc6
DK
1229 * @inode: An inode to create the journal in
1230 *
f7f4bccb 1231 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
470decc6
DK
1232 * the journal. The inode must exist already, must support bmap() and
1233 * must have all data blocks preallocated.
1234 */
f0c9fd54 1235journal_t *jbd2_journal_init_inode(struct inode *inode)
470decc6 1236{
f0c9fd54 1237 journal_t *journal;
05496769 1238 char *p;
18eba7aa 1239 unsigned long long blocknr;
470decc6 1240
f0c9fd54
GT
1241 blocknr = bmap(inode, 0);
1242 if (!blocknr) {
1243 pr_err("%s: Cannot locate journal superblock\n",
1244 __func__);
1245 return NULL;
1246 }
1247
1248 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1249 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1250 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1251
1252 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1253 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1254 inode->i_sb->s_blocksize);
470decc6
DK
1255 if (!journal)
1256 return NULL;
1257
470decc6 1258 journal->j_inode = inode;
05496769 1259 bdevname(journal->j_dev, journal->j_devname);
81ae394b 1260 p = strreplace(journal->j_devname, '/', '!');
90576c0b 1261 sprintf(p, "-%lu", journal->j_inode->i_ino);
8e85fb3f 1262 jbd2_stats_proc_init(journal);
470decc6 1263
470decc6
DK
1264 return journal;
1265}
1266
1267/*
1268 * If the journal init or create aborts, we need to mark the journal
1269 * superblock as being NULL to prevent the journal destroy from writing
1270 * back a bogus superblock.
1271 */
1272static void journal_fail_superblock (journal_t *journal)
1273{
1274 struct buffer_head *bh = journal->j_sb_buffer;
1275 brelse(bh);
1276 journal->j_sb_buffer = NULL;
1277}
1278
1279/*
1280 * Given a journal_t structure, initialise the various fields for
1281 * startup of a new journaling session. We use this both when creating
1282 * a journal, and after recovering an old journal to reset it for
1283 * subsequent use.
1284 */
1285
1286static int journal_reset(journal_t *journal)
1287{
1288 journal_superblock_t *sb = journal->j_superblock;
18eba7aa 1289 unsigned long long first, last;
470decc6
DK
1290
1291 first = be32_to_cpu(sb->s_first);
1292 last = be32_to_cpu(sb->s_maxlen);
f6f50e28 1293 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
f2a44523 1294 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
f6f50e28
JK
1295 first, last);
1296 journal_fail_superblock(journal);
1297 return -EINVAL;
1298 }
470decc6
DK
1299
1300 journal->j_first = first;
1301 journal->j_last = last;
1302
1303 journal->j_head = first;
1304 journal->j_tail = first;
1305 journal->j_free = last - first;
1306
1307 journal->j_tail_sequence = journal->j_transaction_sequence;
1308 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1309 journal->j_commit_request = journal->j_commit_sequence;
1310
1311 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1312
470decc6
DK
1313 /*
1314 * As a special case, if the on-disk copy is already marked as needing
24bcc89c
JK
1315 * no recovery (s_start == 0), then we can safely defer the superblock
1316 * update until the next commit by setting JBD2_FLUSHED. This avoids
470decc6
DK
1317 * attempting a write to a potential-readonly device.
1318 */
24bcc89c 1319 if (sb->s_start == 0) {
f2a44523 1320 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
470decc6
DK
1321 "(start %ld, seq %d, errno %d)\n",
1322 journal->j_tail, journal->j_tail_sequence,
1323 journal->j_errno);
24bcc89c
JK
1324 journal->j_flags |= JBD2_FLUSHED;
1325 } else {
a78bb11d 1326 /* Lock here to make assertions happy... */
6fa7aa50 1327 mutex_lock_io(&journal->j_checkpoint_mutex);
79feb521 1328 /*
70fd7614 1329 * Update log tail information. We use REQ_FUA since new
79feb521
JK
1330 * transaction will start reusing journal space and so we
1331 * must make sure information about current log tail is on
1332 * disk before that.
1333 */
1334 jbd2_journal_update_sb_log_tail(journal,
1335 journal->j_tail_sequence,
1336 journal->j_tail,
17f423b5 1337 REQ_SYNC | REQ_FUA);
a78bb11d 1338 mutex_unlock(&journal->j_checkpoint_mutex);
470decc6 1339 }
24bcc89c
JK
1340 return jbd2_journal_start_thread(journal);
1341}
470decc6 1342
538bcaa6
TT
1343/*
1344 * This function expects that the caller will have locked the journal
1345 * buffer head, and will return with it unlocked
1346 */
2a222ca9 1347static int jbd2_write_superblock(journal_t *journal, int write_flags)
24bcc89c
JK
1348{
1349 struct buffer_head *bh = journal->j_sb_buffer;
fe52d17c 1350 journal_superblock_t *sb = journal->j_superblock;
79feb521 1351 int ret;
470decc6 1352
2a222ca9 1353 trace_jbd2_write_superblock(journal, write_flags);
79feb521 1354 if (!(journal->j_flags & JBD2_BARRIER))
28a8f0d3 1355 write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
914258bf
TT
1356 if (buffer_write_io_error(bh)) {
1357 /*
1358 * Oh, dear. A previous attempt to write the journal
1359 * superblock failed. This could happen because the
1360 * USB device was yanked out. Or it could happen to
1361 * be a transient write error and maybe the block will
1362 * be remapped. Nothing we can do but to retry the
1363 * write and hope for the best.
1364 */
1365 printk(KERN_ERR "JBD2: previous I/O error detected "
1366 "for journal superblock update for %s.\n",
1367 journal->j_devname);
1368 clear_buffer_write_io_error(bh);
1369 set_buffer_uptodate(bh);
1370 }
a58ca992
TT
1371 if (jbd2_journal_has_csum_v2or3(journal))
1372 sb->s_checksum = jbd2_superblock_csum(journal, sb);
79feb521
JK
1373 get_bh(bh);
1374 bh->b_end_io = end_buffer_write_sync;
2a222ca9 1375 ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
79feb521 1376 wait_on_buffer(bh);
24bcc89c 1377 if (buffer_write_io_error(bh)) {
24bcc89c
JK
1378 clear_buffer_write_io_error(bh);
1379 set_buffer_uptodate(bh);
79feb521
JK
1380 ret = -EIO;
1381 }
1382 if (ret) {
1383 printk(KERN_ERR "JBD2: Error %d detected when updating "
1384 "journal superblock for %s.\n", ret,
1385 journal->j_devname);
6f6a6fda 1386 jbd2_journal_abort(journal, ret);
24bcc89c 1387 }
6f6a6fda
JQ
1388
1389 return ret;
24bcc89c
JK
1390}
1391
1392/**
1393 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1394 * @journal: The journal to update.
79feb521
JK
1395 * @tail_tid: TID of the new transaction at the tail of the log
1396 * @tail_block: The first block of the transaction at the tail of the log
1397 * @write_op: With which operation should we write the journal sb
24bcc89c
JK
1398 *
1399 * Update a journal's superblock information about log tail and write it to
1400 * disk, waiting for the IO to complete.
1401 */
6f6a6fda 1402int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
79feb521 1403 unsigned long tail_block, int write_op)
24bcc89c
JK
1404{
1405 journal_superblock_t *sb = journal->j_superblock;
6f6a6fda 1406 int ret;
24bcc89c 1407
85e0c4e8
TT
1408 if (is_journal_aborted(journal))
1409 return -EIO;
1410
a78bb11d 1411 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
79feb521
JK
1412 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1413 tail_block, tail_tid);
470decc6 1414
538bcaa6 1415 lock_buffer(journal->j_sb_buffer);
79feb521
JK
1416 sb->s_sequence = cpu_to_be32(tail_tid);
1417 sb->s_start = cpu_to_be32(tail_block);
470decc6 1418
6f6a6fda
JQ
1419 ret = jbd2_write_superblock(journal, write_op);
1420 if (ret)
1421 goto out;
470decc6 1422
24bcc89c
JK
1423 /* Log is no longer empty */
1424 write_lock(&journal->j_state_lock);
1425 WARN_ON(!sb->s_sequence);
1426 journal->j_flags &= ~JBD2_FLUSHED;
1427 write_unlock(&journal->j_state_lock);
6f6a6fda
JQ
1428
1429out:
1430 return ret;
24bcc89c 1431}
2201c590 1432
24bcc89c
JK
1433/**
1434 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1435 * @journal: The journal to update.
c0a2ad9b 1436 * @write_op: With which operation should we write the journal sb
24bcc89c
JK
1437 *
1438 * Update a journal's dynamic superblock fields to show that journal is empty.
1439 * Write updated superblock to disk waiting for IO to complete.
1440 */
c0a2ad9b 1441static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
24bcc89c
JK
1442{
1443 journal_superblock_t *sb = journal->j_superblock;
914258bf 1444
a78bb11d 1445 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
538bcaa6
TT
1446 lock_buffer(journal->j_sb_buffer);
1447 if (sb->s_start == 0) { /* Is it already empty? */
1448 unlock_buffer(journal->j_sb_buffer);
eeecef0a
ES
1449 return;
1450 }
538bcaa6 1451
24bcc89c
JK
1452 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1453 journal->j_tail_sequence);
470decc6
DK
1454
1455 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
24bcc89c 1456 sb->s_start = cpu_to_be32(0);
470decc6 1457
c0a2ad9b 1458 jbd2_write_superblock(journal, write_op);
470decc6 1459
24bcc89c 1460 /* Log is no longer empty */
a931da6a 1461 write_lock(&journal->j_state_lock);
24bcc89c 1462 journal->j_flags |= JBD2_FLUSHED;
a931da6a 1463 write_unlock(&journal->j_state_lock);
470decc6
DK
1464}
1465
24bcc89c
JK
1466
1467/**
1468 * jbd2_journal_update_sb_errno() - Update error in the journal.
1469 * @journal: The journal to update.
1470 *
1471 * Update a journal's errno. Write updated superblock to disk waiting for IO
1472 * to complete.
1473 */
d796c52e 1474void jbd2_journal_update_sb_errno(journal_t *journal)
24bcc89c
JK
1475{
1476 journal_superblock_t *sb = journal->j_superblock;
fb7c0244 1477 int errcode;
24bcc89c 1478
538bcaa6 1479 lock_buffer(journal->j_sb_buffer);
fb7c0244 1480 errcode = journal->j_errno;
fb7c0244
TT
1481 if (errcode == -ESHUTDOWN)
1482 errcode = 0;
1483 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1484 sb->s_errno = cpu_to_be32(errcode);
24bcc89c 1485
17f423b5 1486 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
24bcc89c 1487}
d796c52e 1488EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
24bcc89c 1489
470decc6
DK
1490/*
1491 * Read the superblock for a given journal, performing initial
1492 * validation of the format.
1493 */
470decc6
DK
1494static int journal_get_superblock(journal_t *journal)
1495{
1496 struct buffer_head *bh;
1497 journal_superblock_t *sb;
1498 int err = -EIO;
1499
1500 bh = journal->j_sb_buffer;
1501
1502 J_ASSERT(bh != NULL);
1503 if (!buffer_uptodate(bh)) {
dfec8a14 1504 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
470decc6
DK
1505 wait_on_buffer(bh);
1506 if (!buffer_uptodate(bh)) {
f2a44523
EG
1507 printk(KERN_ERR
1508 "JBD2: IO error reading journal superblock\n");
470decc6
DK
1509 goto out;
1510 }
1511 }
1512
25ed6e8a
DW
1513 if (buffer_verified(bh))
1514 return 0;
1515
470decc6
DK
1516 sb = journal->j_superblock;
1517
1518 err = -EINVAL;
1519
f7f4bccb 1520 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
470decc6 1521 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
f2a44523 1522 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
470decc6
DK
1523 goto out;
1524 }
1525
1526 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
f7f4bccb 1527 case JBD2_SUPERBLOCK_V1:
470decc6
DK
1528 journal->j_format_version = 1;
1529 break;
f7f4bccb 1530 case JBD2_SUPERBLOCK_V2:
470decc6
DK
1531 journal->j_format_version = 2;
1532 break;
1533 default:
f2a44523 1534 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
470decc6
DK
1535 goto out;
1536 }
1537
1538 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1539 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1540 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
f2a44523 1541 printk(KERN_WARNING "JBD2: journal file too short\n");
470decc6
DK
1542 goto out;
1543 }
1544
8762202d
EG
1545 if (be32_to_cpu(sb->s_first) == 0 ||
1546 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1547 printk(KERN_WARNING
1548 "JBD2: Invalid start block of journal: %u\n",
1549 be32_to_cpu(sb->s_first));
1550 goto out;
1551 }
1552
56316a0d
DW
1553 if (jbd2_has_feature_csum2(journal) &&
1554 jbd2_has_feature_csum3(journal)) {
db9ee220
DW
1555 /* Can't have checksum v2 and v3 at the same time! */
1556 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1557 "at the same time!\n");
1558 goto out;
1559 }
1560
8595798c 1561 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
56316a0d 1562 jbd2_has_feature_checksum(journal)) {
feb8c6d3
DW
1563 /* Can't have checksum v1 and v2 on at the same time! */
1564 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1565 "at the same time!\n");
1566 goto out;
1567 }
1568
25ed6e8a 1569 if (!jbd2_verify_csum_type(journal, sb)) {
a67c848a 1570 printk(KERN_ERR "JBD2: Unknown checksum type\n");
25ed6e8a
DW
1571 goto out;
1572 }
1573
01b5adce 1574 /* Load the checksum driver */
8595798c 1575 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
01b5adce
DW
1576 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1577 if (IS_ERR(journal->j_chksum_driver)) {
a67c848a 1578 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
01b5adce
DW
1579 err = PTR_ERR(journal->j_chksum_driver);
1580 journal->j_chksum_driver = NULL;
1581 goto out;
1582 }
1583 }
1584
a58ca992
TT
1585 if (jbd2_journal_has_csum_v2or3(journal)) {
1586 /* Check superblock checksum */
1587 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1588 printk(KERN_ERR "JBD2: journal checksum error\n");
1589 err = -EFSBADCRC;
1590 goto out;
1591 }
4fd5ea43 1592
a58ca992 1593 /* Precompute checksum seed for all metadata */
4fd5ea43
DW
1594 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1595 sizeof(sb->s_uuid));
a58ca992 1596 }
4fd5ea43 1597
25ed6e8a
DW
1598 set_buffer_verified(bh);
1599
470decc6
DK
1600 return 0;
1601
1602out:
1603 journal_fail_superblock(journal);
1604 return err;
1605}
1606
1607/*
1608 * Load the on-disk journal superblock and read the key fields into the
1609 * journal_t.
1610 */
1611
1612static int load_superblock(journal_t *journal)
1613{
1614 int err;
1615 journal_superblock_t *sb;
1616
1617 err = journal_get_superblock(journal);
1618 if (err)
1619 return err;
1620
1621 sb = journal->j_superblock;
1622
1623 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1624 journal->j_tail = be32_to_cpu(sb->s_start);
1625 journal->j_first = be32_to_cpu(sb->s_first);
1626 journal->j_last = be32_to_cpu(sb->s_maxlen);
1627 journal->j_errno = be32_to_cpu(sb->s_errno);
1628
1629 return 0;
1630}
1631
1632
1633/**
f7f4bccb 1634 * int jbd2_journal_load() - Read journal from disk.
470decc6
DK
1635 * @journal: Journal to act on.
1636 *
1637 * Given a journal_t structure which tells us which disk blocks contain
1638 * a journal, read the journal from disk to initialise the in-memory
1639 * structures.
1640 */
f7f4bccb 1641int jbd2_journal_load(journal_t *journal)
470decc6
DK
1642{
1643 int err;
1644 journal_superblock_t *sb;
1645
1646 err = load_superblock(journal);
1647 if (err)
1648 return err;
1649
1650 sb = journal->j_superblock;
1651 /* If this is a V2 superblock, then we have to check the
1652 * features flags on it. */
1653
1654 if (journal->j_format_version >= 2) {
1655 if ((sb->s_feature_ro_compat &
f7f4bccb 1656 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
470decc6 1657 (sb->s_feature_incompat &
f7f4bccb 1658 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
f2a44523
EG
1659 printk(KERN_WARNING
1660 "JBD2: Unrecognised features on journal\n");
470decc6
DK
1661 return -EINVAL;
1662 }
1663 }
1664
d2eecb03
TT
1665 /*
1666 * Create a slab for this blocksize
1667 */
1668 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1669 if (err)
1670 return err;
1671
470decc6
DK
1672 /* Let the recovery code check whether it needs to recover any
1673 * data from the journal. */
f7f4bccb 1674 if (jbd2_journal_recover(journal))
470decc6
DK
1675 goto recovery_error;
1676
e6a47428
TT
1677 if (journal->j_failed_commit) {
1678 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1679 "is corrupt.\n", journal->j_failed_commit,
1680 journal->j_devname);
6a797d27 1681 return -EFSCORRUPTED;
e6a47428
TT
1682 }
1683
470decc6
DK
1684 /* OK, we've finished with the dynamic journal bits:
1685 * reinitialise the dynamic contents of the superblock in memory
1686 * and reset them on disk. */
1687 if (journal_reset(journal))
1688 goto recovery_error;
1689
f7f4bccb
MC
1690 journal->j_flags &= ~JBD2_ABORT;
1691 journal->j_flags |= JBD2_LOADED;
470decc6
DK
1692 return 0;
1693
1694recovery_error:
f2a44523 1695 printk(KERN_WARNING "JBD2: recovery failed\n");
470decc6
DK
1696 return -EIO;
1697}
1698
1699/**
f7f4bccb 1700 * void jbd2_journal_destroy() - Release a journal_t structure.
470decc6
DK
1701 * @journal: Journal to act on.
1702 *
1703 * Release a journal_t structure once it is no longer in use by the
1704 * journaled object.
44519faf 1705 * Return <0 if we couldn't clean up the journal.
470decc6 1706 */
44519faf 1707int jbd2_journal_destroy(journal_t *journal)
470decc6 1708{
44519faf
HK
1709 int err = 0;
1710
470decc6
DK
1711 /* Wait for the commit thread to wake up and die. */
1712 journal_kill_thread(journal);
1713
1714 /* Force a final log commit */
1715 if (journal->j_running_transaction)
f7f4bccb 1716 jbd2_journal_commit_transaction(journal);
470decc6
DK
1717
1718 /* Force any old transactions to disk */
1719
1720 /* Totally anal locking here... */
1721 spin_lock(&journal->j_list_lock);
1722 while (journal->j_checkpoint_transactions != NULL) {
1723 spin_unlock(&journal->j_list_lock);
6fa7aa50 1724 mutex_lock_io(&journal->j_checkpoint_mutex);
841df7df 1725 err = jbd2_log_do_checkpoint(journal);
1a0d3786 1726 mutex_unlock(&journal->j_checkpoint_mutex);
841df7df
JK
1727 /*
1728 * If checkpointing failed, just free the buffers to avoid
1729 * looping forever
1730 */
1731 if (err) {
1732 jbd2_journal_destroy_checkpoint(journal);
1733 spin_lock(&journal->j_list_lock);
1734 break;
1735 }
470decc6
DK
1736 spin_lock(&journal->j_list_lock);
1737 }
1738
1739 J_ASSERT(journal->j_running_transaction == NULL);
1740 J_ASSERT(journal->j_committing_transaction == NULL);
1741 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1742 spin_unlock(&journal->j_list_lock);
1743
470decc6 1744 if (journal->j_sb_buffer) {
44519faf 1745 if (!is_journal_aborted(journal)) {
6fa7aa50 1746 mutex_lock_io(&journal->j_checkpoint_mutex);
c0a2ad9b
OH
1747
1748 write_lock(&journal->j_state_lock);
1749 journal->j_tail_sequence =
1750 ++journal->j_transaction_sequence;
1751 write_unlock(&journal->j_state_lock);
1752
70fd7614 1753 jbd2_mark_journal_empty(journal,
17f423b5 1754 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
a78bb11d
JK
1755 mutex_unlock(&journal->j_checkpoint_mutex);
1756 } else
44519faf 1757 err = -EIO;
470decc6
DK
1758 brelse(journal->j_sb_buffer);
1759 }
1760
8e85fb3f
JL
1761 if (journal->j_proc_entry)
1762 jbd2_stats_proc_exit(journal);
d9f39d1e 1763 iput(journal->j_inode);
470decc6 1764 if (journal->j_revoke)
f7f4bccb 1765 jbd2_journal_destroy_revoke(journal);
01b5adce
DW
1766 if (journal->j_chksum_driver)
1767 crypto_free_shash(journal->j_chksum_driver);
470decc6
DK
1768 kfree(journal->j_wbuf);
1769 kfree(journal);
44519faf
HK
1770
1771 return err;
470decc6
DK
1772}
1773
1774
1775/**
f7f4bccb 1776 *int jbd2_journal_check_used_features () - Check if features specified are used.
470decc6
DK
1777 * @journal: Journal to check.
1778 * @compat: bitmask of compatible features
1779 * @ro: bitmask of features that force read-only mount
1780 * @incompat: bitmask of incompatible features
1781 *
1782 * Check whether the journal uses all of a given set of
1783 * features. Return true (non-zero) if it does.
1784 **/
1785
f7f4bccb 1786int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
470decc6
DK
1787 unsigned long ro, unsigned long incompat)
1788{
1789 journal_superblock_t *sb;
1790
1791 if (!compat && !ro && !incompat)
1792 return 1;
1113e1b5
PL
1793 /* Load journal superblock if it is not loaded yet. */
1794 if (journal->j_format_version == 0 &&
1795 journal_get_superblock(journal) != 0)
1796 return 0;
470decc6
DK
1797 if (journal->j_format_version == 1)
1798 return 0;
1799
1800 sb = journal->j_superblock;
1801
1802 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1803 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1804 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1805 return 1;
1806
1807 return 0;
1808}
1809
1810/**
f7f4bccb 1811 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
470decc6
DK
1812 * @journal: Journal to check.
1813 * @compat: bitmask of compatible features
1814 * @ro: bitmask of features that force read-only mount
1815 * @incompat: bitmask of incompatible features
1816 *
1817 * Check whether the journaling code supports the use of
1818 * all of a given set of features on this journal. Return true
1819 * (non-zero) if it can. */
1820
f7f4bccb 1821int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
470decc6
DK
1822 unsigned long ro, unsigned long incompat)
1823{
470decc6
DK
1824 if (!compat && !ro && !incompat)
1825 return 1;
1826
470decc6
DK
1827 /* We can support any known requested features iff the
1828 * superblock is in version 2. Otherwise we fail to support any
1829 * extended sb features. */
1830
1831 if (journal->j_format_version != 2)
1832 return 0;
1833
f7f4bccb
MC
1834 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1835 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1836 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
470decc6
DK
1837 return 1;
1838
1839 return 0;
1840}
1841
1842/**
f7f4bccb 1843 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
470decc6
DK
1844 * @journal: Journal to act on.
1845 * @compat: bitmask of compatible features
1846 * @ro: bitmask of features that force read-only mount
1847 * @incompat: bitmask of incompatible features
1848 *
1849 * Mark a given journal feature as present on the
1850 * superblock. Returns true if the requested features could be set.
1851 *
1852 */
1853
f7f4bccb 1854int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
470decc6
DK
1855 unsigned long ro, unsigned long incompat)
1856{
25ed6e8a
DW
1857#define INCOMPAT_FEATURE_ON(f) \
1858 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1859#define COMPAT_FEATURE_ON(f) \
1860 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
470decc6
DK
1861 journal_superblock_t *sb;
1862
f7f4bccb 1863 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
470decc6
DK
1864 return 1;
1865
f7f4bccb 1866 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
470decc6
DK
1867 return 0;
1868
db9ee220
DW
1869 /* If enabling v2 checksums, turn on v3 instead */
1870 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1871 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1872 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1873 }
1874
1875 /* Asking for checksumming v3 and v1? Only give them v3. */
1876 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
25ed6e8a
DW
1877 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1878 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1879
470decc6
DK
1880 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1881 compat, ro, incompat);
1882
1883 sb = journal->j_superblock;
1884
538bcaa6
TT
1885 /* Load the checksum driver if necessary */
1886 if ((journal->j_chksum_driver == NULL) &&
1887 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1888 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1889 if (IS_ERR(journal->j_chksum_driver)) {
1890 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1891 journal->j_chksum_driver = NULL;
1892 return 0;
1893 }
1894 /* Precompute checksum seed for all metadata */
1895 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1896 sizeof(sb->s_uuid));
1897 }
1898
1899 lock_buffer(journal->j_sb_buffer);
1900
db9ee220
DW
1901 /* If enabling v3 checksums, update superblock */
1902 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
25ed6e8a
DW
1903 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1904 sb->s_feature_compat &=
1905 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1906 }
1907
1908 /* If enabling v1 checksums, downgrade superblock */
1909 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1910 sb->s_feature_incompat &=
db9ee220
DW
1911 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1912 JBD2_FEATURE_INCOMPAT_CSUM_V3);
25ed6e8a 1913
470decc6
DK
1914 sb->s_feature_compat |= cpu_to_be32(compat);
1915 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1916 sb->s_feature_incompat |= cpu_to_be32(incompat);
538bcaa6 1917 unlock_buffer(journal->j_sb_buffer);
470decc6
DK
1918
1919 return 1;
25ed6e8a
DW
1920#undef COMPAT_FEATURE_ON
1921#undef INCOMPAT_FEATURE_ON
470decc6
DK
1922}
1923
818d276c
GS
1924/*
1925 * jbd2_journal_clear_features () - Clear a given journal feature in the
1926 * superblock
1927 * @journal: Journal to act on.
1928 * @compat: bitmask of compatible features
1929 * @ro: bitmask of features that force read-only mount
1930 * @incompat: bitmask of incompatible features
1931 *
1932 * Clear a given journal feature as present on the
1933 * superblock.
1934 */
1935void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1936 unsigned long ro, unsigned long incompat)
1937{
1938 journal_superblock_t *sb;
1939
1940 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1941 compat, ro, incompat);
1942
1943 sb = journal->j_superblock;
1944
1945 sb->s_feature_compat &= ~cpu_to_be32(compat);
1946 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1947 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1948}
1949EXPORT_SYMBOL(jbd2_journal_clear_features);
470decc6 1950
470decc6 1951/**
f7f4bccb 1952 * int jbd2_journal_flush () - Flush journal
470decc6
DK
1953 * @journal: Journal to act on.
1954 *
1955 * Flush all data for a given journal to disk and empty the journal.
1956 * Filesystems can use this when remounting readonly to ensure that
1957 * recovery does not need to happen on remount.
1958 */
1959
f7f4bccb 1960int jbd2_journal_flush(journal_t *journal)
470decc6
DK
1961{
1962 int err = 0;
1963 transaction_t *transaction = NULL;
470decc6 1964
a931da6a 1965 write_lock(&journal->j_state_lock);
470decc6
DK
1966
1967 /* Force everything buffered to the log... */
1968 if (journal->j_running_transaction) {
1969 transaction = journal->j_running_transaction;
f7f4bccb 1970 __jbd2_log_start_commit(journal, transaction->t_tid);
470decc6
DK
1971 } else if (journal->j_committing_transaction)
1972 transaction = journal->j_committing_transaction;
1973
1974 /* Wait for the log commit to complete... */
1975 if (transaction) {
1976 tid_t tid = transaction->t_tid;
1977
a931da6a 1978 write_unlock(&journal->j_state_lock);
f7f4bccb 1979 jbd2_log_wait_commit(journal, tid);
470decc6 1980 } else {
a931da6a 1981 write_unlock(&journal->j_state_lock);
470decc6
DK
1982 }
1983
1984 /* ...and flush everything in the log out to disk. */
1985 spin_lock(&journal->j_list_lock);
1986 while (!err && journal->j_checkpoint_transactions != NULL) {
1987 spin_unlock(&journal->j_list_lock);
6fa7aa50 1988 mutex_lock_io(&journal->j_checkpoint_mutex);
f7f4bccb 1989 err = jbd2_log_do_checkpoint(journal);
44519faf 1990 mutex_unlock(&journal->j_checkpoint_mutex);
470decc6
DK
1991 spin_lock(&journal->j_list_lock);
1992 }
1993 spin_unlock(&journal->j_list_lock);
44519faf
HK
1994
1995 if (is_journal_aborted(journal))
1996 return -EIO;
1997
6fa7aa50 1998 mutex_lock_io(&journal->j_checkpoint_mutex);
6f6a6fda
JQ
1999 if (!err) {
2000 err = jbd2_cleanup_journal_tail(journal);
2001 if (err < 0) {
2002 mutex_unlock(&journal->j_checkpoint_mutex);
2003 goto out;
2004 }
2005 err = 0;
2006 }
470decc6
DK
2007
2008 /* Finally, mark the journal as really needing no recovery.
2009 * This sets s_start==0 in the underlying superblock, which is
2010 * the magic code for a fully-recovered superblock. Any future
2011 * commits of data to the journal will restore the current
2012 * s_start value. */
17f423b5 2013 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
a78bb11d 2014 mutex_unlock(&journal->j_checkpoint_mutex);
a931da6a 2015 write_lock(&journal->j_state_lock);
470decc6
DK
2016 J_ASSERT(!journal->j_running_transaction);
2017 J_ASSERT(!journal->j_committing_transaction);
2018 J_ASSERT(!journal->j_checkpoint_transactions);
2019 J_ASSERT(journal->j_head == journal->j_tail);
2020 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
a931da6a 2021 write_unlock(&journal->j_state_lock);
6f6a6fda
JQ
2022out:
2023 return err;
470decc6
DK
2024}
2025
2026/**
f7f4bccb 2027 * int jbd2_journal_wipe() - Wipe journal contents
470decc6
DK
2028 * @journal: Journal to act on.
2029 * @write: flag (see below)
2030 *
2031 * Wipe out all of the contents of a journal, safely. This will produce
2032 * a warning if the journal contains any valid recovery information.
f7f4bccb 2033 * Must be called between journal_init_*() and jbd2_journal_load().
470decc6
DK
2034 *
2035 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2036 * we merely suppress recovery.
2037 */
2038
f7f4bccb 2039int jbd2_journal_wipe(journal_t *journal, int write)
470decc6 2040{
470decc6
DK
2041 int err = 0;
2042
f7f4bccb 2043 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
470decc6
DK
2044
2045 err = load_superblock(journal);
2046 if (err)
2047 return err;
2048
470decc6
DK
2049 if (!journal->j_tail)
2050 goto no_recovery;
2051
f2a44523 2052 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
470decc6
DK
2053 write ? "Clearing" : "Ignoring");
2054
f7f4bccb 2055 err = jbd2_journal_skip_recovery(journal);
a78bb11d
JK
2056 if (write) {
2057 /* Lock to make assertions happy... */
53cf9784 2058 mutex_lock_io(&journal->j_checkpoint_mutex);
17f423b5 2059 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
a78bb11d
JK
2060 mutex_unlock(&journal->j_checkpoint_mutex);
2061 }
470decc6
DK
2062
2063 no_recovery:
2064 return err;
2065}
2066
470decc6
DK
2067/*
2068 * Journal abort has very specific semantics, which we describe
2069 * for journal abort.
2070 *
bfcd3555 2071 * Two internal functions, which provide abort to the jbd layer
470decc6
DK
2072 * itself are here.
2073 */
2074
2075/*
2076 * Quick version for internal journal use (doesn't lock the journal).
2077 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2078 * and don't attempt to make any other journal updates.
2079 */
f7f4bccb 2080void __jbd2_journal_abort_hard(journal_t *journal)
470decc6
DK
2081{
2082 transaction_t *transaction;
470decc6 2083
f7f4bccb 2084 if (journal->j_flags & JBD2_ABORT)
470decc6
DK
2085 return;
2086
2087 printk(KERN_ERR "Aborting journal on device %s.\n",
05496769 2088 journal->j_devname);
470decc6 2089
a931da6a 2090 write_lock(&journal->j_state_lock);
f7f4bccb 2091 journal->j_flags |= JBD2_ABORT;
470decc6
DK
2092 transaction = journal->j_running_transaction;
2093 if (transaction)
f7f4bccb 2094 __jbd2_log_start_commit(journal, transaction->t_tid);
a931da6a 2095 write_unlock(&journal->j_state_lock);
470decc6
DK
2096}
2097
2098/* Soft abort: record the abort error status in the journal superblock,
2099 * but don't do any other IO. */
2100static void __journal_abort_soft (journal_t *journal, int errno)
2101{
fb7c0244 2102 int old_errno;
470decc6 2103
fb7c0244
TT
2104 write_lock(&journal->j_state_lock);
2105 old_errno = journal->j_errno;
2106 if (!journal->j_errno || errno == -ESHUTDOWN)
470decc6
DK
2107 journal->j_errno = errno;
2108
fb7c0244
TT
2109 if (journal->j_flags & JBD2_ABORT) {
2110 write_unlock(&journal->j_state_lock);
2111 if (!old_errno && old_errno != -ESHUTDOWN &&
2112 errno == -ESHUTDOWN)
2113 jbd2_journal_update_sb_errno(journal);
2114 return;
2115 }
2116 write_unlock(&journal->j_state_lock);
2117
f7f4bccb 2118 __jbd2_journal_abort_hard(journal);
470decc6 2119
4327ba52 2120 if (errno) {
24bcc89c 2121 jbd2_journal_update_sb_errno(journal);
4327ba52
DJ
2122 write_lock(&journal->j_state_lock);
2123 journal->j_flags |= JBD2_REC_ERR;
2124 write_unlock(&journal->j_state_lock);
2125 }
470decc6
DK
2126}
2127
2128/**
f7f4bccb 2129 * void jbd2_journal_abort () - Shutdown the journal immediately.
470decc6
DK
2130 * @journal: the journal to shutdown.
2131 * @errno: an error number to record in the journal indicating
2132 * the reason for the shutdown.
2133 *
2134 * Perform a complete, immediate shutdown of the ENTIRE
2135 * journal (not of a single transaction). This operation cannot be
2136 * undone without closing and reopening the journal.
2137 *
f7f4bccb 2138 * The jbd2_journal_abort function is intended to support higher level error
470decc6
DK
2139 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2140 * mode.
2141 *
2142 * Journal abort has very specific semantics. Any existing dirty,
2143 * unjournaled buffers in the main filesystem will still be written to
2144 * disk by bdflush, but the journaling mechanism will be suspended
2145 * immediately and no further transaction commits will be honoured.
2146 *
2147 * Any dirty, journaled buffers will be written back to disk without
2148 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2149 * filesystem, but we _do_ attempt to leave as much data as possible
2150 * behind for fsck to use for cleanup.
2151 *
2152 * Any attempt to get a new transaction handle on a journal which is in
2153 * ABORT state will just result in an -EROFS error return. A
f7f4bccb 2154 * jbd2_journal_stop on an existing handle will return -EIO if we have
470decc6
DK
2155 * entered abort state during the update.
2156 *
2157 * Recursive transactions are not disturbed by journal abort until the
f7f4bccb 2158 * final jbd2_journal_stop, which will receive the -EIO error.
470decc6 2159 *
f7f4bccb 2160 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
470decc6
DK
2161 * which will be recorded (if possible) in the journal superblock. This
2162 * allows a client to record failure conditions in the middle of a
2163 * transaction without having to complete the transaction to record the
2164 * failure to disk. ext3_error, for example, now uses this
2165 * functionality.
2166 *
2167 * Errors which originate from within the journaling layer will NOT
2168 * supply an errno; a null errno implies that absolutely no further
2169 * writes are done to the journal (unless there are any already in
2170 * progress).
2171 *
2172 */
2173
f7f4bccb 2174void jbd2_journal_abort(journal_t *journal, int errno)
470decc6
DK
2175{
2176 __journal_abort_soft(journal, errno);
2177}
2178
2179/**
f7f4bccb 2180 * int jbd2_journal_errno () - returns the journal's error state.
470decc6
DK
2181 * @journal: journal to examine.
2182 *
bfcd3555 2183 * This is the errno number set with jbd2_journal_abort(), the last
470decc6
DK
2184 * time the journal was mounted - if the journal was stopped
2185 * without calling abort this will be 0.
2186 *
2187 * If the journal has been aborted on this mount time -EROFS will
2188 * be returned.
2189 */
f7f4bccb 2190int jbd2_journal_errno(journal_t *journal)
470decc6
DK
2191{
2192 int err;
2193
a931da6a 2194 read_lock(&journal->j_state_lock);
f7f4bccb 2195 if (journal->j_flags & JBD2_ABORT)
470decc6
DK
2196 err = -EROFS;
2197 else
2198 err = journal->j_errno;
a931da6a 2199 read_unlock(&journal->j_state_lock);
470decc6
DK
2200 return err;
2201}
2202
2203/**
f7f4bccb 2204 * int jbd2_journal_clear_err () - clears the journal's error state
470decc6
DK
2205 * @journal: journal to act on.
2206 *
bfcd3555 2207 * An error must be cleared or acked to take a FS out of readonly
470decc6
DK
2208 * mode.
2209 */
f7f4bccb 2210int jbd2_journal_clear_err(journal_t *journal)
470decc6
DK
2211{
2212 int err = 0;
2213
a931da6a 2214 write_lock(&journal->j_state_lock);
f7f4bccb 2215 if (journal->j_flags & JBD2_ABORT)
470decc6
DK
2216 err = -EROFS;
2217 else
2218 journal->j_errno = 0;
a931da6a 2219 write_unlock(&journal->j_state_lock);
470decc6
DK
2220 return err;
2221}
2222
2223/**
f7f4bccb 2224 * void jbd2_journal_ack_err() - Ack journal err.
470decc6
DK
2225 * @journal: journal to act on.
2226 *
bfcd3555 2227 * An error must be cleared or acked to take a FS out of readonly
470decc6
DK
2228 * mode.
2229 */
f7f4bccb 2230void jbd2_journal_ack_err(journal_t *journal)
470decc6 2231{
a931da6a 2232 write_lock(&journal->j_state_lock);
470decc6 2233 if (journal->j_errno)
f7f4bccb 2234 journal->j_flags |= JBD2_ACK_ERR;
a931da6a 2235 write_unlock(&journal->j_state_lock);
470decc6
DK
2236}
2237
f7f4bccb 2238int jbd2_journal_blocks_per_page(struct inode *inode)
470decc6 2239{
09cbfeaf 2240 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
470decc6
DK
2241}
2242
b517bea1
ZB
2243/*
2244 * helper functions to deal with 32 or 64bit block numbers.
2245 */
2246size_t journal_tag_bytes(journal_t *journal)
2247{
db9ee220
DW
2248 size_t sz;
2249
56316a0d 2250 if (jbd2_has_feature_csum3(journal))
db9ee220
DW
2251 return sizeof(journal_block_tag3_t);
2252
2253 sz = sizeof(journal_block_tag_t);
c3900875 2254
56316a0d 2255 if (jbd2_has_feature_csum2(journal))
db9ee220 2256 sz += sizeof(__u16);
c3900875 2257
56316a0d 2258 if (jbd2_has_feature_64bit(journal))
db9ee220 2259 return sz;
b517bea1 2260 else
db9ee220 2261 return sz - sizeof(__u32);
b517bea1
ZB
2262}
2263
d2eecb03
TT
2264/*
2265 * JBD memory management
2266 *
2267 * These functions are used to allocate block-sized chunks of memory
2268 * used for making copies of buffer_head data. Very often it will be
2269 * page-sized chunks of data, but sometimes it will be in
2270 * sub-page-size chunks. (For example, 16k pages on Power systems
2271 * with a 4k block file system.) For blocks smaller than a page, we
2272 * use a SLAB allocator. There are slab caches for each block size,
2273 * which are allocated at mount time, if necessary, and we only free
2274 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2275 * this reason we don't need to a mutex to protect access to
2276 * jbd2_slab[] allocating or releasing memory; only in
2277 * jbd2_journal_create_slab().
2278 */
2279#define JBD2_MAX_SLABS 8
2280static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
d2eecb03
TT
2281
2282static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2283 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2284 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2285};
2286
2287
2288static void jbd2_journal_destroy_slabs(void)
2289{
2290 int i;
2291
2292 for (i = 0; i < JBD2_MAX_SLABS; i++) {
8bdd5b60 2293 kmem_cache_destroy(jbd2_slab[i]);
d2eecb03
TT
2294 jbd2_slab[i] = NULL;
2295 }
2296}
2297
2298static int jbd2_journal_create_slab(size_t size)
2299{
51dfacde 2300 static DEFINE_MUTEX(jbd2_slab_create_mutex);
d2eecb03
TT
2301 int i = order_base_2(size) - 10;
2302 size_t slab_size;
2303
2304 if (size == PAGE_SIZE)
2305 return 0;
2306
2307 if (i >= JBD2_MAX_SLABS)
2308 return -EINVAL;
2309
2310 if (unlikely(i < 0))
2311 i = 0;
51dfacde 2312 mutex_lock(&jbd2_slab_create_mutex);
d2eecb03 2313 if (jbd2_slab[i]) {
51dfacde 2314 mutex_unlock(&jbd2_slab_create_mutex);
d2eecb03
TT
2315 return 0; /* Already created */
2316 }
2317
2318 slab_size = 1 << (i+10);
2319 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2320 slab_size, 0, NULL);
51dfacde 2321 mutex_unlock(&jbd2_slab_create_mutex);
d2eecb03
TT
2322 if (!jbd2_slab[i]) {
2323 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2324 return -ENOMEM;
2325 }
2326 return 0;
2327}
2328
2329static struct kmem_cache *get_slab(size_t size)
2330{
2331 int i = order_base_2(size) - 10;
2332
2333 BUG_ON(i >= JBD2_MAX_SLABS);
2334 if (unlikely(i < 0))
2335 i = 0;
8ac97b74 2336 BUG_ON(jbd2_slab[i] == NULL);
d2eecb03
TT
2337 return jbd2_slab[i];
2338}
2339
2340void *jbd2_alloc(size_t size, gfp_t flags)
2341{
2342 void *ptr;
2343
2344 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2345
f2db1971 2346 if (size < PAGE_SIZE)
d2eecb03 2347 ptr = kmem_cache_alloc(get_slab(size), flags);
f2db1971
MH
2348 else
2349 ptr = (void *)__get_free_pages(flags, get_order(size));
d2eecb03
TT
2350
2351 /* Check alignment; SLUB has gotten this wrong in the past,
2352 * and this can lead to user data corruption! */
2353 BUG_ON(((unsigned long) ptr) & (size-1));
2354
2355 return ptr;
2356}
2357
2358void jbd2_free(void *ptr, size_t size)
2359{
f2db1971
MH
2360 if (size < PAGE_SIZE)
2361 kmem_cache_free(get_slab(size), ptr);
2362 else
2363 free_pages((unsigned long)ptr, get_order(size));
d2eecb03
TT
2364};
2365
470decc6
DK
2366/*
2367 * Journal_head storage management
2368 */
e18b890b 2369static struct kmem_cache *jbd2_journal_head_cache;
e23291b9 2370#ifdef CONFIG_JBD2_DEBUG
470decc6
DK
2371static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2372#endif
2373
4185a2ac 2374static int jbd2_journal_init_journal_head_cache(void)
470decc6
DK
2375{
2376 int retval;
2377
1076d17a 2378 J_ASSERT(jbd2_journal_head_cache == NULL);
a920e941 2379 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
470decc6
DK
2380 sizeof(struct journal_head),
2381 0, /* offset */
5f0d5a3a 2382 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
20c2df83 2383 NULL); /* ctor */
470decc6 2384 retval = 0;
1076d17a 2385 if (!jbd2_journal_head_cache) {
470decc6 2386 retval = -ENOMEM;
f2a44523 2387 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
470decc6
DK
2388 }
2389 return retval;
2390}
2391
4185a2ac 2392static void jbd2_journal_destroy_journal_head_cache(void)
470decc6 2393{
8bdd5b60
WL
2394 kmem_cache_destroy(jbd2_journal_head_cache);
2395 jbd2_journal_head_cache = NULL;
470decc6
DK
2396}
2397
2398/*
2399 * journal_head splicing and dicing
2400 */
2401static struct journal_head *journal_alloc_journal_head(void)
2402{
2403 struct journal_head *ret;
470decc6 2404
e23291b9 2405#ifdef CONFIG_JBD2_DEBUG
470decc6
DK
2406 atomic_inc(&nr_journal_heads);
2407#endif
5d9cf9c6 2408 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
1076d17a 2409 if (!ret) {
470decc6 2410 jbd_debug(1, "out of memory for journal_head\n");
670be5a7 2411 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
7b506b10
MH
2412 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2413 GFP_NOFS | __GFP_NOFAIL);
470decc6
DK
2414 }
2415 return ret;
2416}
2417
2418static void journal_free_journal_head(struct journal_head *jh)
2419{
e23291b9 2420#ifdef CONFIG_JBD2_DEBUG
470decc6 2421 atomic_dec(&nr_journal_heads);
cd02ff0b 2422 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
470decc6 2423#endif
f7f4bccb 2424 kmem_cache_free(jbd2_journal_head_cache, jh);
470decc6
DK
2425}
2426
2427/*
2428 * A journal_head is attached to a buffer_head whenever JBD has an
2429 * interest in the buffer.
2430 *
2431 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2432 * is set. This bit is tested in core kernel code where we need to take
2433 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2434 * there.
2435 *
2436 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2437 *
2438 * When a buffer has its BH_JBD bit set it is immune from being released by
2439 * core kernel code, mainly via ->b_count.
2440 *
de1b7941
JK
2441 * A journal_head is detached from its buffer_head when the journal_head's
2442 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2443 * transaction (b_cp_transaction) hold their references to b_jcount.
470decc6
DK
2444 *
2445 * Various places in the kernel want to attach a journal_head to a buffer_head
2446 * _before_ attaching the journal_head to a transaction. To protect the
f7f4bccb 2447 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
470decc6 2448 * journal_head's b_jcount refcount by one. The caller must call
f7f4bccb 2449 * jbd2_journal_put_journal_head() to undo this.
470decc6
DK
2450 *
2451 * So the typical usage would be:
2452 *
2453 * (Attach a journal_head if needed. Increments b_jcount)
f7f4bccb 2454 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
470decc6 2455 * ...
de1b7941
JK
2456 * (Get another reference for transaction)
2457 * jbd2_journal_grab_journal_head(bh);
470decc6 2458 * jh->b_transaction = xxx;
de1b7941 2459 * (Put original reference)
f7f4bccb 2460 * jbd2_journal_put_journal_head(jh);
470decc6
DK
2461 */
2462
2463/*
2464 * Give a buffer_head a journal_head.
2465 *
470decc6
DK
2466 * May sleep.
2467 */
f7f4bccb 2468struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
470decc6
DK
2469{
2470 struct journal_head *jh;
2471 struct journal_head *new_jh = NULL;
2472
2473repeat:
5d9cf9c6 2474 if (!buffer_jbd(bh))
470decc6 2475 new_jh = journal_alloc_journal_head();
470decc6
DK
2476
2477 jbd_lock_bh_journal_head(bh);
2478 if (buffer_jbd(bh)) {
2479 jh = bh2jh(bh);
2480 } else {
2481 J_ASSERT_BH(bh,
2482 (atomic_read(&bh->b_count) > 0) ||
2483 (bh->b_page && bh->b_page->mapping));
2484
2485 if (!new_jh) {
2486 jbd_unlock_bh_journal_head(bh);
2487 goto repeat;
2488 }
2489
2490 jh = new_jh;
2491 new_jh = NULL; /* We consumed it */
2492 set_buffer_jbd(bh);
2493 bh->b_private = jh;
2494 jh->b_bh = bh;
2495 get_bh(bh);
2496 BUFFER_TRACE(bh, "added journal_head");
2497 }
2498 jh->b_jcount++;
2499 jbd_unlock_bh_journal_head(bh);
2500 if (new_jh)
2501 journal_free_journal_head(new_jh);
2502 return bh->b_private;
2503}
2504
2505/*
2506 * Grab a ref against this buffer_head's journal_head. If it ended up not
2507 * having a journal_head, return NULL
2508 */
f7f4bccb 2509struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
470decc6
DK
2510{
2511 struct journal_head *jh = NULL;
2512
2513 jbd_lock_bh_journal_head(bh);
2514 if (buffer_jbd(bh)) {
2515 jh = bh2jh(bh);
2516 jh->b_jcount++;
2517 }
2518 jbd_unlock_bh_journal_head(bh);
2519 return jh;
2520}
2521
2522static void __journal_remove_journal_head(struct buffer_head *bh)
2523{
2524 struct journal_head *jh = bh2jh(bh);
2525
2526 J_ASSERT_JH(jh, jh->b_jcount >= 0);
de1b7941
JK
2527 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2528 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2529 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2530 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2531 J_ASSERT_BH(bh, buffer_jbd(bh));
2532 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2533 BUFFER_TRACE(bh, "remove journal_head");
2534 if (jh->b_frozen_data) {
2535 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2536 jbd2_free(jh->b_frozen_data, bh->b_size);
470decc6 2537 }
de1b7941
JK
2538 if (jh->b_committed_data) {
2539 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2540 jbd2_free(jh->b_committed_data, bh->b_size);
2541 }
2542 bh->b_private = NULL;
2543 jh->b_bh = NULL; /* debug, really */
2544 clear_buffer_jbd(bh);
2545 journal_free_journal_head(jh);
470decc6
DK
2546}
2547
2548/*
de1b7941 2549 * Drop a reference on the passed journal_head. If it fell to zero then
470decc6
DK
2550 * release the journal_head from the buffer_head.
2551 */
f7f4bccb 2552void jbd2_journal_put_journal_head(struct journal_head *jh)
470decc6
DK
2553{
2554 struct buffer_head *bh = jh2bh(jh);
2555
2556 jbd_lock_bh_journal_head(bh);
2557 J_ASSERT_JH(jh, jh->b_jcount > 0);
2558 --jh->b_jcount;
de1b7941 2559 if (!jh->b_jcount) {
470decc6 2560 __journal_remove_journal_head(bh);
de1b7941 2561 jbd_unlock_bh_journal_head(bh);
470decc6 2562 __brelse(bh);
de1b7941
JK
2563 } else
2564 jbd_unlock_bh_journal_head(bh);
470decc6
DK
2565}
2566
c851ed54
JK
2567/*
2568 * Initialize jbd inode head
2569 */
2570void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2571{
2572 jinode->i_transaction = NULL;
2573 jinode->i_next_transaction = NULL;
2574 jinode->i_vfs_inode = inode;
2575 jinode->i_flags = 0;
2576 INIT_LIST_HEAD(&jinode->i_list);
2577}
2578
2579/*
2580 * Function to be called before we start removing inode from memory (i.e.,
2581 * clear_inode() is a fine place to be called from). It removes inode from
2582 * transaction's lists.
2583 */
2584void jbd2_journal_release_jbd_inode(journal_t *journal,
2585 struct jbd2_inode *jinode)
2586{
c851ed54
JK
2587 if (!journal)
2588 return;
2589restart:
2590 spin_lock(&journal->j_list_lock);
2591 /* Is commit writing out inode - we have to wait */
cb0d9d47 2592 if (jinode->i_flags & JI_COMMIT_RUNNING) {
c851ed54
JK
2593 wait_queue_head_t *wq;
2594 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2595 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
21417136 2596 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
c851ed54
JK
2597 spin_unlock(&journal->j_list_lock);
2598 schedule();
21417136 2599 finish_wait(wq, &wait.wq_entry);
c851ed54
JK
2600 goto restart;
2601 }
2602
c851ed54
JK
2603 if (jinode->i_transaction) {
2604 list_del(&jinode->i_list);
2605 jinode->i_transaction = NULL;
2606 }
2607 spin_unlock(&journal->j_list_lock);
2608}
2609
470decc6 2610
8e85fb3f
JL
2611#ifdef CONFIG_PROC_FS
2612
2613#define JBD2_STATS_PROC_NAME "fs/jbd2"
2614
2615static void __init jbd2_create_jbd_stats_proc_entry(void)
2616{
2617 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2618}
2619
2620static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2621{
2622 if (proc_jbd2_stats)
2623 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2624}
2625
2626#else
2627
2628#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2629#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2630
2631#endif
2632
8aefcd55 2633struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
470decc6 2634
4185a2ac 2635static int __init jbd2_journal_init_handle_cache(void)
470decc6 2636{
8aefcd55 2637 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
f7f4bccb 2638 if (jbd2_handle_cache == NULL) {
8aefcd55
TT
2639 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2640 return -ENOMEM;
2641 }
2642 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2643 if (jbd2_inode_cache == NULL) {
2644 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2645 kmem_cache_destroy(jbd2_handle_cache);
470decc6
DK
2646 return -ENOMEM;
2647 }
2648 return 0;
2649}
2650
f7f4bccb 2651static void jbd2_journal_destroy_handle_cache(void)
470decc6 2652{
8bdd5b60
WL
2653 kmem_cache_destroy(jbd2_handle_cache);
2654 jbd2_handle_cache = NULL;
2655 kmem_cache_destroy(jbd2_inode_cache);
2656 jbd2_inode_cache = NULL;
470decc6
DK
2657}
2658
2659/*
2660 * Module startup and shutdown
2661 */
2662
2663static int __init journal_init_caches(void)
2664{
2665 int ret;
2666
f7f4bccb 2667 ret = jbd2_journal_init_revoke_caches();
470decc6 2668 if (ret == 0)
4185a2ac 2669 ret = jbd2_journal_init_journal_head_cache();
470decc6 2670 if (ret == 0)
4185a2ac 2671 ret = jbd2_journal_init_handle_cache();
470decc6 2672 if (ret == 0)
0c2022ec 2673 ret = jbd2_journal_init_transaction_cache();
470decc6
DK
2674 return ret;
2675}
2676
f7f4bccb 2677static void jbd2_journal_destroy_caches(void)
470decc6 2678{
f7f4bccb 2679 jbd2_journal_destroy_revoke_caches();
4185a2ac 2680 jbd2_journal_destroy_journal_head_cache();
f7f4bccb 2681 jbd2_journal_destroy_handle_cache();
0c2022ec 2682 jbd2_journal_destroy_transaction_cache();
d2eecb03 2683 jbd2_journal_destroy_slabs();
470decc6
DK
2684}
2685
2686static int __init journal_init(void)
2687{
2688 int ret;
2689
2690 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2691
2692 ret = journal_init_caches();
620de4e1 2693 if (ret == 0) {
620de4e1
DG
2694 jbd2_create_jbd_stats_proc_entry();
2695 } else {
f7f4bccb 2696 jbd2_journal_destroy_caches();
620de4e1 2697 }
470decc6
DK
2698 return ret;
2699}
2700
2701static void __exit journal_exit(void)
2702{
e23291b9 2703#ifdef CONFIG_JBD2_DEBUG
470decc6
DK
2704 int n = atomic_read(&nr_journal_heads);
2705 if (n)
75685071 2706 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
470decc6 2707#endif
8e85fb3f 2708 jbd2_remove_jbd_stats_proc_entry();
f7f4bccb 2709 jbd2_journal_destroy_caches();
470decc6
DK
2710}
2711
2712MODULE_LICENSE("GPL");
2713module_init(journal_init);
2714module_exit(journal_exit);
2715