blob: 2d04f9afafd7ea5cf59624777d12c9196b61e97c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Uwe Zeisbergerf30c2262006-10-03 23:01:26 +02002 * linux/fs/jbd/journal.c
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5 *
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Generic filesystem journal-writing code; part of the ext2fs
13 * journaling system.
14 *
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
18 *
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
23 */
24
25#include <linux/module.h>
26#include <linux/time.h>
27#include <linux/fs.h>
28#include <linux/jbd.h>
29#include <linux/errno.h>
30#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/init.h>
32#include <linux/mm.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080033#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/pagemap.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080035#include <linux/kthread.h>
Randy Dunlapc9cf5522006-06-27 02:53:52 -070036#include <linux/poison.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080037#include <linux/proc_fs.h>
Jose R. Santosc2a91592007-10-18 23:39:22 -070038#include <linux/debugfs.h>
Namhyung Kimf81e3d42010-10-04 19:12:13 +090039#include <linux/ratelimit.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080040
Lukas Czerner99cb1a32011-05-23 18:33:02 +020041#define CREATE_TRACE_POINTS
42#include <trace/events/jbd.h>
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/uaccess.h>
45#include <asm/page.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47EXPORT_SYMBOL(journal_start);
48EXPORT_SYMBOL(journal_restart);
49EXPORT_SYMBOL(journal_extend);
50EXPORT_SYMBOL(journal_stop);
51EXPORT_SYMBOL(journal_lock_updates);
52EXPORT_SYMBOL(journal_unlock_updates);
53EXPORT_SYMBOL(journal_get_write_access);
54EXPORT_SYMBOL(journal_get_create_access);
55EXPORT_SYMBOL(journal_get_undo_access);
56EXPORT_SYMBOL(journal_dirty_data);
57EXPORT_SYMBOL(journal_dirty_metadata);
58EXPORT_SYMBOL(journal_release_buffer);
59EXPORT_SYMBOL(journal_forget);
60#if 0
61EXPORT_SYMBOL(journal_sync_buffer);
62#endif
63EXPORT_SYMBOL(journal_flush);
64EXPORT_SYMBOL(journal_revoke);
65
66EXPORT_SYMBOL(journal_init_dev);
67EXPORT_SYMBOL(journal_init_inode);
68EXPORT_SYMBOL(journal_update_format);
69EXPORT_SYMBOL(journal_check_used_features);
70EXPORT_SYMBOL(journal_check_available_features);
71EXPORT_SYMBOL(journal_set_features);
72EXPORT_SYMBOL(journal_create);
73EXPORT_SYMBOL(journal_load);
74EXPORT_SYMBOL(journal_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075EXPORT_SYMBOL(journal_abort);
76EXPORT_SYMBOL(journal_errno);
77EXPORT_SYMBOL(journal_ack_err);
78EXPORT_SYMBOL(journal_clear_err);
79EXPORT_SYMBOL(log_wait_commit);
Stefan Schmidtff5e4b52009-11-12 09:53:50 +010080EXPORT_SYMBOL(log_start_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081EXPORT_SYMBOL(journal_start_commit);
82EXPORT_SYMBOL(journal_force_commit_nested);
83EXPORT_SYMBOL(journal_wipe);
84EXPORT_SYMBOL(journal_blocks_per_page);
85EXPORT_SYMBOL(journal_invalidatepage);
86EXPORT_SYMBOL(journal_try_to_free_buffers);
87EXPORT_SYMBOL(journal_force_commit);
88
89static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
Adrian Bunk022a4a72005-09-06 15:16:41 -070090static void __journal_abort_soft (journal_t *journal, int errno);
Darrick J. Wongdff68252010-10-04 12:35:05 -070091static const char *journal_dev_name(journal_t *journal, char *buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Paul Gortmakere99a03c2013-08-01 15:01:08 -040093#ifdef CONFIG_JBD_DEBUG
94void __jbd_debug(int level, const char *file, const char *func,
95 unsigned int line, const char *fmt, ...)
96{
97 struct va_format vaf;
98 va_list args;
99
100 if (level > journal_enable_debug)
101 return;
102 va_start(args, fmt);
103 vaf.fmt = fmt;
104 vaf.va = &args;
105 printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
106 va_end(args);
107}
108EXPORT_SYMBOL(__jbd_debug);
109#endif
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/*
112 * Helper function used to manage commit timeouts
113 */
114
115static void commit_timeout(unsigned long __data)
116{
117 struct task_struct * p = (struct task_struct *) __data;
118
119 wake_up_process(p);
120}
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122/*
123 * kjournald: The main thread function used to manage a logging device
124 * journal.
125 *
126 * This kernel thread is responsible for two things:
127 *
128 * 1) COMMIT: Every so often we need to commit the current state of the
129 * filesystem to disk. The journal thread is responsible for writing
130 * all of the metadata buffers to disk.
131 *
132 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
133 * of the data in that part of the log has been rewritten elsewhere on
134 * the disk. Flushing these old buffers to reclaim space in the log is
135 * known as checkpointing, and this thread is responsible for that job.
136 */
137
Adrian Bunk022a4a72005-09-06 15:16:41 -0700138static int kjournald(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Andrew Mortone3df1892006-03-25 03:06:53 -0800140 journal_t *journal = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 transaction_t *transaction;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
Andrew Mortone3df1892006-03-25 03:06:53 -0800143 /*
144 * Set up an interval timer which can be used to trigger a commit wakeup
145 * after the commit interval expires
146 */
147 setup_timer(&journal->j_commit_timer, commit_timeout,
148 (unsigned long)current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Nigel Cunningham35c80422012-02-03 19:59:41 +1100150 set_freezable();
151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 /* Record that the journal thread is running */
153 journal->j_task = current;
154 wake_up(&journal->j_wait_done_commit);
155
156 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
157 journal->j_commit_interval / HZ);
158
159 /*
160 * And now, wait forever for commit wakeup events.
161 */
162 spin_lock(&journal->j_state_lock);
163
164loop:
165 if (journal->j_flags & JFS_UNMOUNT)
166 goto end_loop;
167
168 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
169 journal->j_commit_sequence, journal->j_commit_request);
170
171 if (journal->j_commit_sequence != journal->j_commit_request) {
172 jbd_debug(1, "OK, requests differ\n");
173 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800174 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 journal_commit_transaction(journal);
176 spin_lock(&journal->j_state_lock);
177 goto loop;
178 }
179
180 wake_up(&journal->j_wait_done_commit);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700181 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 /*
183 * The simpler the better. Flushing journal isn't a
184 * good idea, because that depends on threads that may
185 * be already stopped.
186 */
187 jbd_debug(1, "Now suspending kjournald\n");
188 spin_unlock(&journal->j_state_lock);
Tejun Heoa0acae02011-11-21 12:32:22 -0800189 try_to_freeze();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 spin_lock(&journal->j_state_lock);
191 } else {
192 /*
193 * We assume on resume that commits are already there,
194 * so we don't sleep
195 */
196 DEFINE_WAIT(wait);
197 int should_sleep = 1;
198
199 prepare_to_wait(&journal->j_wait_commit, &wait,
200 TASK_INTERRUPTIBLE);
201 if (journal->j_commit_sequence != journal->j_commit_request)
202 should_sleep = 0;
203 transaction = journal->j_running_transaction;
204 if (transaction && time_after_eq(jiffies,
205 transaction->t_expires))
206 should_sleep = 0;
Mark Fashehcbf0d272005-09-06 15:19:08 -0700207 if (journal->j_flags & JFS_UNMOUNT)
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700208 should_sleep = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 if (should_sleep) {
210 spin_unlock(&journal->j_state_lock);
211 schedule();
212 spin_lock(&journal->j_state_lock);
213 }
214 finish_wait(&journal->j_wait_commit, &wait);
215 }
216
217 jbd_debug(1, "kjournald wakes\n");
218
219 /*
220 * Were we woken up by a commit wakeup event?
221 */
222 transaction = journal->j_running_transaction;
223 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
224 journal->j_commit_request = transaction->t_tid;
225 jbd_debug(1, "woke because of timeout\n");
226 }
227 goto loop;
228
229end_loop:
230 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800231 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 journal->j_task = NULL;
233 wake_up(&journal->j_wait_done_commit);
234 jbd_debug(1, "Journal thread exiting.\n");
235 return 0;
236}
237
Pavel Emelianov97f06782007-05-08 00:30:42 -0700238static int journal_start_thread(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700240 struct task_struct *t;
241
242 t = kthread_run(kjournald, journal, "kjournald");
243 if (IS_ERR(t))
244 return PTR_ERR(t);
245
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700246 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700247 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249
250static void journal_kill_thread(journal_t *journal)
251{
252 spin_lock(&journal->j_state_lock);
253 journal->j_flags |= JFS_UNMOUNT;
254
255 while (journal->j_task) {
256 wake_up(&journal->j_wait_commit);
257 spin_unlock(&journal->j_state_lock);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700258 wait_event(journal->j_wait_done_commit,
259 journal->j_task == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 spin_lock(&journal->j_state_lock);
261 }
262 spin_unlock(&journal->j_state_lock);
263}
264
265/*
266 * journal_write_metadata_buffer: write a metadata buffer to the journal.
267 *
268 * Writes a metadata buffer to a given disk block. The actual IO is not
269 * performed but a new buffer_head is constructed which labels the data
270 * to be written with the correct destination disk block.
271 *
272 * Any magic-number escaping which needs to be done will cause a
273 * copy-out here. If the buffer happens to start with the
274 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
275 * magic number is only written to the log for descripter blocks. In
276 * this case, we copy the data and replace the first word with 0, and we
277 * return a result code which indicates that this buffer needs to be
278 * marked as an escaped buffer in the corresponding log descriptor
279 * block. The missing word can then be restored when the block is read
280 * during recovery.
281 *
282 * If the source buffer has already been modified by a new transaction
283 * since we took the last commit snapshot, we use the frozen copy of
284 * that data for IO. If we end up using the existing buffer_head's data
285 * for the write, then we *have* to lock the buffer to prevent anyone
286 * else from using and possibly modifying it while the IO is in
287 * progress.
288 *
289 * The function returns a pointer to the buffer_heads to be used for IO.
290 *
291 * We assume that the journal has already been locked in this function.
292 *
293 * Return value:
294 * <0: Error
295 * >=0: Finished OK
296 *
297 * On success:
298 * Bit 0 set == escape performed on the data
299 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
300 */
301
302int journal_write_metadata_buffer(transaction_t *transaction,
303 struct journal_head *jh_in,
304 struct journal_head **jh_out,
Jan Kara9c28cbc2009-08-03 19:21:00 +0200305 unsigned int blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306{
307 int need_copy_out = 0;
308 int done_copy_out = 0;
309 int do_escape = 0;
310 char *mapped_data;
311 struct buffer_head *new_bh;
312 struct journal_head *new_jh;
313 struct page *new_page;
314 unsigned int new_offset;
315 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghuaf1015c42009-07-15 21:42:05 +0200316 journal_t *journal = transaction->t_journal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
318 /*
319 * The buffer really shouldn't be locked: only the current committing
320 * transaction is allowed to write it, so nobody else is allowed
321 * to do any IO.
322 *
323 * akpm: except if we're journalling data, and write() output is
324 * also part of a shared mapping, and another thread has
325 * decided to launch a writepage() against this buffer.
326 */
327 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
328
329 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
dingdinghuaf1015c42009-07-15 21:42:05 +0200330 /* keep subsequent assertions sane */
dingdinghuaf1015c42009-07-15 21:42:05 +0200331 atomic_set(&new_bh->b_count, 1);
332 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
334 /*
335 * If a new transaction has already done a buffer copy-out, then
336 * we use that version of the data for the commit.
337 */
338 jbd_lock_bh_state(bh_in);
339repeat:
340 if (jh_in->b_frozen_data) {
341 done_copy_out = 1;
342 new_page = virt_to_page(jh_in->b_frozen_data);
343 new_offset = offset_in_page(jh_in->b_frozen_data);
344 } else {
345 new_page = jh2bh(jh_in)->b_page;
346 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
347 }
348
Cong Wang8fb53c42011-11-25 23:14:31 +0800349 mapped_data = kmap_atomic(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 /*
351 * Check for escaping
352 */
353 if (*((__be32 *)(mapped_data + new_offset)) ==
354 cpu_to_be32(JFS_MAGIC_NUMBER)) {
355 need_copy_out = 1;
356 do_escape = 1;
357 }
Cong Wang8fb53c42011-11-25 23:14:31 +0800358 kunmap_atomic(mapped_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 /*
361 * Do we need to do a data copy?
362 */
363 if (need_copy_out && !done_copy_out) {
364 char *tmp;
365
366 jbd_unlock_bh_state(bh_in);
Mingming Caoc089d492007-10-16 18:38:25 -0400367 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 jbd_lock_bh_state(bh_in);
369 if (jh_in->b_frozen_data) {
Mingming Caoc089d492007-10-16 18:38:25 -0400370 jbd_free(tmp, bh_in->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 goto repeat;
372 }
373
374 jh_in->b_frozen_data = tmp;
Cong Wang8fb53c42011-11-25 23:14:31 +0800375 mapped_data = kmap_atomic(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
Cong Wang8fb53c42011-11-25 23:14:31 +0800377 kunmap_atomic(mapped_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379 new_page = virt_to_page(tmp);
380 new_offset = offset_in_page(tmp);
381 done_copy_out = 1;
382 }
383
384 /*
385 * Did we need to do an escaping? Now we've done all the
386 * copying, we can finally do so.
387 */
388 if (do_escape) {
Cong Wang8fb53c42011-11-25 23:14:31 +0800389 mapped_data = kmap_atomic(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 *((unsigned int *)(mapped_data + new_offset)) = 0;
Cong Wang8fb53c42011-11-25 23:14:31 +0800391 kunmap_atomic(mapped_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 }
393
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 set_bh_page(new_bh, new_page, new_offset);
395 new_jh->b_transaction = NULL;
396 new_bh->b_size = jh2bh(jh_in)->b_size;
397 new_bh->b_bdev = transaction->t_journal->j_dev;
398 new_bh->b_blocknr = blocknr;
399 set_buffer_mapped(new_bh);
400 set_buffer_dirty(new_bh);
401
402 *jh_out = new_jh;
403
404 /*
405 * The to-be-written buffer needs to get moved to the io queue,
406 * and the original buffer whose contents we are shadowing or
407 * copying is moved to the transaction's shadow queue.
408 */
409 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghuaf1015c42009-07-15 21:42:05 +0200410 spin_lock(&journal->j_list_lock);
411 __journal_file_buffer(jh_in, transaction, BJ_Shadow);
412 spin_unlock(&journal->j_list_lock);
413 jbd_unlock_bh_state(bh_in);
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 JBUFFER_TRACE(new_jh, "file as BJ_IO");
416 journal_file_buffer(new_jh, transaction, BJ_IO);
417
418 return do_escape | (done_copy_out << 1);
419}
420
421/*
422 * Allocation code for the journal file. Manage the space left in the
423 * journal, so that we can begin checkpointing when appropriate.
424 */
425
426/*
427 * __log_space_left: Return the number of free blocks left in the journal.
428 *
429 * Called with the journal already locked.
430 *
431 * Called under j_state_lock
432 */
433
434int __log_space_left(journal_t *journal)
435{
436 int left = journal->j_free;
437
438 assert_spin_locked(&journal->j_state_lock);
439
440 /*
441 * Be pessimistic here about the number of those free blocks which
442 * might be required for log descriptor control blocks.
443 */
444
445#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
446
447 left -= MIN_LOG_RESERVED_BLOCKS;
448
449 if (left <= 0)
450 return 0;
451 left -= (left >> 3);
452 return left;
453}
454
455/*
Jan Kara8fe4cd02009-02-11 13:04:25 -0800456 * Called under j_state_lock. Returns true if a transaction commit was started.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 */
458int __log_start_commit(journal_t *journal, tid_t target)
459{
460 /*
Ted Ts'od9b01932011-04-30 13:17:11 -0400461 * The only transaction we can possibly wait upon is the
462 * currently running transaction (if it exists). Otherwise,
463 * the target tid must be an old one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 */
Eric Sandeen7e2fb2d2012-12-18 11:03:57 -0600465 if (journal->j_commit_request != target &&
466 journal->j_running_transaction &&
Ted Ts'od9b01932011-04-30 13:17:11 -0400467 journal->j_running_transaction->t_tid == target) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200469 * We want a new commit: OK, mark the request and wakeup the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 * commit thread. We do _not_ do the commit ourselves.
471 */
472
473 journal->j_commit_request = target;
474 jbd_debug(1, "JBD: requesting commit %d/%d\n",
475 journal->j_commit_request,
476 journal->j_commit_sequence);
477 wake_up(&journal->j_wait_commit);
478 return 1;
Ted Ts'od9b01932011-04-30 13:17:11 -0400479 } else if (!tid_geq(journal->j_commit_request, target))
480 /* This should never happen, but if it does, preserve
481 the evidence before kjournald goes into a loop and
482 increments j_commit_sequence beyond all recognition. */
483 WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
484 journal->j_commit_request, journal->j_commit_sequence,
485 target, journal->j_running_transaction ?
486 journal->j_running_transaction->t_tid : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return 0;
488}
489
490int log_start_commit(journal_t *journal, tid_t tid)
491{
492 int ret;
493
494 spin_lock(&journal->j_state_lock);
495 ret = __log_start_commit(journal, tid);
496 spin_unlock(&journal->j_state_lock);
497 return ret;
498}
499
500/*
501 * Force and wait upon a commit if the calling process is not within
502 * transaction. This is used for forcing out undo-protected data which contains
503 * bitmaps, when the fs is running out of space.
504 *
505 * We can only force the running transaction if we don't have an active handle;
506 * otherwise, we will deadlock.
507 *
508 * Returns true if a transaction was started.
509 */
510int journal_force_commit_nested(journal_t *journal)
511{
512 transaction_t *transaction = NULL;
513 tid_t tid;
514
515 spin_lock(&journal->j_state_lock);
516 if (journal->j_running_transaction && !current->journal_info) {
517 transaction = journal->j_running_transaction;
518 __log_start_commit(journal, transaction->t_tid);
519 } else if (journal->j_committing_transaction)
520 transaction = journal->j_committing_transaction;
521
522 if (!transaction) {
523 spin_unlock(&journal->j_state_lock);
524 return 0; /* Nothing to retry */
525 }
526
527 tid = transaction->t_tid;
528 spin_unlock(&journal->j_state_lock);
529 log_wait_commit(journal, tid);
530 return 1;
531}
532
533/*
534 * Start a commit of the current running transaction (if any). Returns true
Jan Kara8fe4cd02009-02-11 13:04:25 -0800535 * if a transaction is going to be committed (or is currently already
536 * committing), and fills its tid in at *ptid
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 */
538int journal_start_commit(journal_t *journal, tid_t *ptid)
539{
540 int ret = 0;
541
542 spin_lock(&journal->j_state_lock);
543 if (journal->j_running_transaction) {
544 tid_t tid = journal->j_running_transaction->t_tid;
545
Jan Kara8fe4cd02009-02-11 13:04:25 -0800546 __log_start_commit(journal, tid);
547 /* There's a running transaction and we've just made sure
548 * it's commit has been scheduled. */
549 if (ptid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 *ptid = tid;
Jan Kara8fe4cd02009-02-11 13:04:25 -0800551 ret = 1;
552 } else if (journal->j_committing_transaction) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 /*
Artem Bityutskiy12810ad2012-07-25 18:12:07 +0300554 * If commit has been started, then we have to wait for
555 * completion of that transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 */
Jan Kara8fe4cd02009-02-11 13:04:25 -0800557 if (ptid)
558 *ptid = journal->j_committing_transaction->t_tid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 ret = 1;
560 }
561 spin_unlock(&journal->j_state_lock);
562 return ret;
563}
564
565/*
566 * Wait for a specified commit to complete.
567 * The caller may not hold the journal lock.
568 */
569int log_wait_commit(journal_t *journal, tid_t tid)
570{
571 int err = 0;
572
573#ifdef CONFIG_JBD_DEBUG
574 spin_lock(&journal->j_state_lock);
575 if (!tid_geq(journal->j_commit_request, tid)) {
576 printk(KERN_EMERG
577 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700578 __func__, journal->j_commit_request, tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580 spin_unlock(&journal->j_state_lock);
581#endif
582 spin_lock(&journal->j_state_lock);
Jan Karae678a4f2013-03-27 17:30:59 +0100583 /*
584 * Not running or committing trans? Must be already committed. This
585 * saves us from waiting for a *long* time when tid overflows.
586 */
587 if (!((journal->j_running_transaction &&
588 journal->j_running_transaction->t_tid == tid) ||
589 (journal->j_committing_transaction &&
590 journal->j_committing_transaction->t_tid == tid)))
591 goto out_unlock;
592
Jan Kara2db938b2011-02-21 17:25:37 +0100593 if (!tid_geq(journal->j_commit_waited, tid))
594 journal->j_commit_waited = tid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 while (tid_gt(tid, journal->j_commit_sequence)) {
596 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
597 tid, journal->j_commit_sequence);
598 wake_up(&journal->j_wait_commit);
599 spin_unlock(&journal->j_state_lock);
600 wait_event(journal->j_wait_done_commit,
601 !tid_gt(tid, journal->j_commit_sequence));
602 spin_lock(&journal->j_state_lock);
603 }
Jan Karae678a4f2013-03-27 17:30:59 +0100604out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 spin_unlock(&journal->j_state_lock);
606
607 if (unlikely(is_journal_aborted(journal))) {
608 printk(KERN_EMERG "journal commit I/O error\n");
609 err = -EIO;
610 }
611 return err;
612}
613
614/*
Jan Kara03f4d802010-04-15 22:16:24 +0200615 * Return 1 if a given transaction has not yet sent barrier request
616 * connected with a transaction commit. If 0 is returned, transaction
617 * may or may not have sent the barrier. Used to avoid sending barrier
618 * twice in common cases.
619 */
620int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
621{
622 int ret = 0;
623 transaction_t *commit_trans;
624
625 if (!(journal->j_flags & JFS_BARRIER))
626 return 0;
627 spin_lock(&journal->j_state_lock);
628 /* Transaction already committed? */
629 if (tid_geq(journal->j_commit_sequence, tid))
630 goto out;
631 /*
632 * Transaction is being committed and we already proceeded to
633 * writing commit record?
634 */
635 commit_trans = journal->j_committing_transaction;
636 if (commit_trans && commit_trans->t_tid == tid &&
637 commit_trans->t_state >= T_COMMIT_RECORD)
638 goto out;
639 ret = 1;
640out:
641 spin_unlock(&journal->j_state_lock);
642 return ret;
643}
Jan Kara52779702010-04-15 22:24:26 +0200644EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
Jan Kara03f4d802010-04-15 22:16:24 +0200645
646/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 * Log buffer allocation routines:
648 */
649
Jan Kara9c28cbc2009-08-03 19:21:00 +0200650int journal_next_log_block(journal_t *journal, unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200652 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 spin_lock(&journal->j_state_lock);
655 J_ASSERT(journal->j_free > 1);
656
657 blocknr = journal->j_head;
658 journal->j_head++;
659 journal->j_free--;
660 if (journal->j_head == journal->j_last)
661 journal->j_head = journal->j_first;
662 spin_unlock(&journal->j_state_lock);
663 return journal_bmap(journal, blocknr, retp);
664}
665
666/*
667 * Conversion of logical to physical block numbers for the journal
668 *
669 * On external journals the journal blocks are identity-mapped, so
670 * this is a no-op. If needed, we can use j_blk_offset - everything is
671 * ready.
672 */
Jan Kara9c28cbc2009-08-03 19:21:00 +0200673int journal_bmap(journal_t *journal, unsigned int blocknr,
674 unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
676 int err = 0;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200677 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679 if (journal->j_inode) {
680 ret = bmap(journal->j_inode, blocknr);
681 if (ret)
682 *retp = ret;
683 else {
684 char b[BDEVNAME_SIZE];
685
686 printk(KERN_ALERT "%s: journal block not found "
Jan Kara9c28cbc2009-08-03 19:21:00 +0200687 "at offset %u on %s\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700688 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 blocknr,
690 bdevname(journal->j_dev, b));
691 err = -EIO;
692 __journal_abort_soft(journal, err);
693 }
694 } else {
695 *retp = blocknr; /* +journal->j_blk_offset */
696 }
697 return err;
698}
699
700/*
701 * We play buffer_head aliasing tricks to write data/metadata blocks to
702 * the journal without copying their contents, but for journal
703 * descriptor blocks we do need to generate bona fide buffers.
704 *
705 * After the caller of journal_get_descriptor_buffer() has finished modifying
706 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
707 * But we don't bother doing that, so there will be coherency problems with
708 * mmaps of blockdevs which hold live JBD-controlled filesystems.
709 */
710struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
711{
712 struct buffer_head *bh;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200713 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 int err;
715
716 err = journal_next_log_block(journal, &blocknr);
717
718 if (err)
719 return NULL;
720
721 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700722 if (!bh)
723 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 lock_buffer(bh);
725 memset(bh->b_data, 0, journal->j_blocksize);
726 set_buffer_uptodate(bh);
727 unlock_buffer(bh);
728 BUFFER_TRACE(bh, "return this buffer");
729 return journal_add_journal_head(bh);
730}
731
732/*
733 * Management for journal control blocks: functions to create and
734 * destroy journal_t structures, and to initialise and read existing
735 * journal blocks from disk. */
736
737/* First: create and setup a journal_t object in memory. We initialise
738 * very few fields yet: that has to wait until we have created the
739 * journal structures from from scratch, or loaded them from disk. */
740
741static journal_t * journal_init_common (void)
742{
743 journal_t *journal;
744 int err;
745
Mingming Cao8c3478a2007-10-18 23:39:20 -0700746 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 if (!journal)
748 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750 init_waitqueue_head(&journal->j_wait_transaction_locked);
751 init_waitqueue_head(&journal->j_wait_logspace);
752 init_waitqueue_head(&journal->j_wait_done_commit);
753 init_waitqueue_head(&journal->j_wait_checkpoint);
754 init_waitqueue_head(&journal->j_wait_commit);
755 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800756 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 spin_lock_init(&journal->j_revoke_lock);
758 spin_lock_init(&journal->j_list_lock);
759 spin_lock_init(&journal->j_state_lock);
760
761 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
762
763 /* The journal is marked for error until we succeed with recovery! */
764 journal->j_flags = JFS_ABORT;
765
766 /* Set up a default-sized revoke table for the new mount. */
767 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
768 if (err) {
769 kfree(journal);
770 goto fail;
771 }
772 return journal;
773fail:
774 return NULL;
775}
776
777/* journal_init_dev and journal_init_inode:
778 *
779 * Create a journal structure assigned some fixed set of disk blocks to
780 * the journal. We don't actually touch those disk blocks yet, but we
781 * need to set up all of the mapping information to tell the journaling
782 * system where the journal blocks are.
783 *
784 */
785
786/**
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700787 * journal_t * journal_init_dev() - creates and initialises a journal structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 * @bdev: Block device on which to create the journal
789 * @fs_dev: Device which hold journalled filesystem for this journal.
790 * @start: Block nr Start of journal.
Eric Sandeen37ed3222006-09-27 01:49:31 -0700791 * @len: Length of the journal in blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 * @blocksize: blocksize of journalling device
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700793 *
794 * Returns: a newly created journal_t *
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700795 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 * journal_init_dev creates a journal which maps a fixed contiguous
797 * range of blocks on an arbitrary block device.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700798 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
800journal_t * journal_init_dev(struct block_device *bdev,
801 struct block_device *fs_dev,
802 int start, int len, int blocksize)
803{
804 journal_t *journal = journal_init_common();
805 struct buffer_head *bh;
806 int n;
807
808 if (!journal)
809 return NULL;
810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 /* journal descriptor can store up to n blocks -bzzz */
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700812 journal->j_blocksize = blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 n = journal->j_blocksize / sizeof(journal_block_tag_t);
814 journal->j_wbufsize = n;
815 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
816 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300817 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700818 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700819 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700821 journal->j_dev = bdev;
822 journal->j_fs_dev = fs_dev;
823 journal->j_blk_offset = start;
824 journal->j_maxlen = len;
825
826 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700827 if (!bh) {
828 printk(KERN_ERR
829 "%s: Cannot get buffer for journal superblock\n",
830 __func__);
831 goto out_err;
832 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700833 journal->j_sb_buffer = bh;
834 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Karaecca9af2009-04-02 16:57:13 -0700835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700837out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800838 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700839 kfree(journal);
840 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841}
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700842
843/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
845 * @inode: An inode to create the journal in
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700846 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 * journal_init_inode creates a journal which maps an on-disk inode as
848 * the journal. The inode must exist already, must support bmap() and
849 * must have all data blocks preallocated.
850 */
851journal_t * journal_init_inode (struct inode *inode)
852{
853 struct buffer_head *bh;
854 journal_t *journal = journal_init_common();
855 int err;
856 int n;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200857 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
859 if (!journal)
860 return NULL;
861
862 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
863 journal->j_inode = inode;
864 jbd_debug(1,
865 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700866 journal, inode->i_sb->s_id, inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 (long long) inode->i_size,
868 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
869
870 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
871 journal->j_blocksize = inode->i_sb->s_blocksize;
872
873 /* journal descriptor can store up to n blocks -bzzz */
874 n = journal->j_blocksize / sizeof(journal_block_tag_t);
875 journal->j_wbufsize = n;
876 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
877 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300878 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700879 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700880 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 }
882
883 err = journal_bmap(journal, 0, &blocknr);
884 /* If that failed, give up */
885 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -0800886 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700887 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700888 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890
891 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700892 if (!bh) {
893 printk(KERN_ERR
894 "%s: Cannot get buffer for journal superblock\n",
895 __func__);
896 goto out_err;
897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 journal->j_sb_buffer = bh;
899 journal->j_superblock = (journal_superblock_t *)bh->b_data;
900
901 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700902out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800903 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700904 kfree(journal);
905 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906}
907
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700908/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 * If the journal init or create aborts, we need to mark the journal
910 * superblock as being NULL to prevent the journal destroy from writing
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700911 * back a bogus superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 */
913static void journal_fail_superblock (journal_t *journal)
914{
915 struct buffer_head *bh = journal->j_sb_buffer;
916 brelse(bh);
917 journal->j_sb_buffer = NULL;
918}
919
920/*
921 * Given a journal_t structure, initialise the various fields for
922 * startup of a new journaling session. We use this both when creating
923 * a journal, and after recovering an old journal to reset it for
924 * subsequent use.
925 */
926
927static int journal_reset(journal_t *journal)
928{
929 journal_superblock_t *sb = journal->j_superblock;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200930 unsigned int first, last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 first = be32_to_cpu(sb->s_first);
933 last = be32_to_cpu(sb->s_maxlen);
Jan Kara7447a662009-07-15 20:36:08 +0200934 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200935 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
Jan Kara7447a662009-07-15 20:36:08 +0200936 first, last);
937 journal_fail_superblock(journal);
938 return -EINVAL;
939 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
941 journal->j_first = first;
942 journal->j_last = last;
943
944 journal->j_head = first;
945 journal->j_tail = first;
946 journal->j_free = last - first;
947
948 journal->j_tail_sequence = journal->j_transaction_sequence;
949 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
950 journal->j_commit_request = journal->j_commit_sequence;
951
952 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
953
Jan Kara9754e392012-04-07 12:33:03 +0200954 /*
955 * As a special case, if the on-disk copy is already marked as needing
956 * no recovery (s_start == 0), then we can safely defer the superblock
957 * update until the next commit by setting JFS_FLUSHED. This avoids
958 * attempting a write to a potential-readonly device.
959 */
960 if (sb->s_start == 0) {
961 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
962 "(start %u, seq %d, errno %d)\n",
963 journal->j_tail, journal->j_tail_sequence,
964 journal->j_errno);
965 journal->j_flags |= JFS_FLUSHED;
966 } else {
Jan Kara1ce84862012-04-07 12:50:13 +0200967 /* Lock here to make assertions happy... */
968 mutex_lock(&journal->j_checkpoint_mutex);
Jan Karafd2cbd42012-04-07 11:05:19 +0200969 /*
970 * Update log tail information. We use WRITE_FUA since new
971 * transaction will start reusing journal space and so we
972 * must make sure information about current log tail is on
973 * disk before that.
974 */
975 journal_update_sb_log_tail(journal,
976 journal->j_tail_sequence,
977 journal->j_tail,
978 WRITE_FUA);
Jan Kara1ce84862012-04-07 12:50:13 +0200979 mutex_unlock(&journal->j_checkpoint_mutex);
Jan Kara9754e392012-04-07 12:33:03 +0200980 }
Pavel Emelianov97f06782007-05-08 00:30:42 -0700981 return journal_start_thread(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982}
983
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700984/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 * int journal_create() - Initialise the new journal file
986 * @journal: Journal to create. This structure must have been initialised
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700987 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 * Given a journal_t structure which tells us which disk blocks we can
989 * use, create a new journal superblock and initialise all of the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700990 * journal fields from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 **/
992int journal_create(journal_t *journal)
993{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200994 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 struct buffer_head *bh;
996 journal_superblock_t *sb;
997 int i, err;
998
999 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
1000 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
1001 journal->j_maxlen);
1002 journal_fail_superblock(journal);
1003 return -EINVAL;
1004 }
1005
1006 if (journal->j_inode == NULL) {
1007 /*
1008 * We don't know what block to start at!
1009 */
1010 printk(KERN_EMERG
1011 "%s: creation of journal on external device!\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001012 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 BUG();
1014 }
1015
1016 /* Zero out the entire journal on disk. We cannot afford to
1017 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
1018 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1019 for (i = 0; i < journal->j_maxlen; i++) {
1020 err = journal_bmap(journal, i, &blocknr);
1021 if (err)
1022 return err;
1023 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Namhyung Kim2a0e3382010-10-13 17:17:18 +09001024 if (unlikely(!bh))
1025 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 lock_buffer(bh);
1027 memset (bh->b_data, 0, journal->j_blocksize);
1028 BUFFER_TRACE(bh, "marking dirty");
1029 mark_buffer_dirty(bh);
1030 BUFFER_TRACE(bh, "marking uptodate");
1031 set_buffer_uptodate(bh);
1032 unlock_buffer(bh);
1033 __brelse(bh);
1034 }
1035
1036 sync_blockdev(journal->j_dev);
1037 jbd_debug(1, "JBD: journal cleared.\n");
1038
1039 /* OK, fill in the initial static fields in the new superblock */
1040 sb = journal->j_superblock;
1041
1042 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
1043 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1044
1045 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1046 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
1047 sb->s_first = cpu_to_be32(1);
1048
1049 journal->j_transaction_sequence = 1;
1050
1051 journal->j_flags &= ~JFS_ABORT;
1052 journal->j_format_version = 2;
1053
1054 return journal_reset(journal);
1055}
1056
Jan Karafd2cbd42012-04-07 11:05:19 +02001057static void journal_write_superblock(journal_t *journal, int write_op)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 struct buffer_head *bh = journal->j_sb_buffer;
Jan Karafd2cbd42012-04-07 11:05:19 +02001060 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Jan Karafd2cbd42012-04-07 11:05:19 +02001062 trace_journal_write_superblock(journal, write_op);
1063 if (!(journal->j_flags & JFS_BARRIER))
1064 write_op &= ~(REQ_FUA | REQ_FLUSH);
1065 lock_buffer(bh);
Darrick J. Wongdff68252010-10-04 12:35:05 -07001066 if (buffer_write_io_error(bh)) {
1067 char b[BDEVNAME_SIZE];
1068 /*
1069 * Oh, dear. A previous attempt to write the journal
1070 * superblock failed. This could happen because the
1071 * USB device was yanked out. Or it could happen to
1072 * be a transient write error and maybe the block will
1073 * be remapped. Nothing we can do but to retry the
1074 * write and hope for the best.
1075 */
1076 printk(KERN_ERR "JBD: previous I/O error detected "
1077 "for journal superblock update for %s.\n",
1078 journal_dev_name(journal, b));
1079 clear_buffer_write_io_error(bh);
1080 set_buffer_uptodate(bh);
1081 }
1082
Jan Karafd2cbd42012-04-07 11:05:19 +02001083 get_bh(bh);
1084 bh->b_end_io = end_buffer_write_sync;
1085 ret = submit_bh(write_op, bh);
1086 wait_on_buffer(bh);
Jan Kara9754e392012-04-07 12:33:03 +02001087 if (buffer_write_io_error(bh)) {
Jan Kara9754e392012-04-07 12:33:03 +02001088 clear_buffer_write_io_error(bh);
1089 set_buffer_uptodate(bh);
Jan Karafd2cbd42012-04-07 11:05:19 +02001090 ret = -EIO;
1091 }
1092 if (ret) {
1093 char b[BDEVNAME_SIZE];
1094 printk(KERN_ERR "JBD: Error %d detected "
1095 "when updating journal superblock for %s.\n",
1096 ret, journal_dev_name(journal, b));
Jan Kara9754e392012-04-07 12:33:03 +02001097 }
1098}
1099
1100/**
1101 * journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1102 * @journal: The journal to update.
Jan Karafd2cbd42012-04-07 11:05:19 +02001103 * @tail_tid: TID of the new transaction at the tail of the log
1104 * @tail_block: The first block of the transaction at the tail of the log
1105 * @write_op: With which operation should we write the journal sb
Jan Kara9754e392012-04-07 12:33:03 +02001106 *
1107 * Update a journal's superblock information about log tail and write it to
1108 * disk, waiting for the IO to complete.
1109 */
Jan Karafd2cbd42012-04-07 11:05:19 +02001110void journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1111 unsigned int tail_block, int write_op)
Jan Kara9754e392012-04-07 12:33:03 +02001112{
1113 journal_superblock_t *sb = journal->j_superblock;
1114
Jan Kara1ce84862012-04-07 12:50:13 +02001115 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Jan Karafd2cbd42012-04-07 11:05:19 +02001116 jbd_debug(1,"JBD: updating superblock (start %u, seq %u)\n",
1117 tail_block, tail_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Jan Karafd2cbd42012-04-07 11:05:19 +02001119 sb->s_sequence = cpu_to_be32(tail_tid);
1120 sb->s_start = cpu_to_be32(tail_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Jan Karafd2cbd42012-04-07 11:05:19 +02001122 journal_write_superblock(journal, write_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Jan Kara9754e392012-04-07 12:33:03 +02001124 /* Log is no longer empty */
1125 spin_lock(&journal->j_state_lock);
1126 WARN_ON(!sb->s_sequence);
1127 journal->j_flags &= ~JFS_FLUSHED;
1128 spin_unlock(&journal->j_state_lock);
1129}
1130
1131/**
1132 * mark_journal_empty() - Mark on disk journal as empty.
1133 * @journal: The journal to update.
1134 *
1135 * Update a journal's dynamic superblock fields to show that journal is empty.
1136 * Write updated superblock to disk waiting for IO to complete.
1137 */
1138static void mark_journal_empty(journal_t *journal)
1139{
1140 journal_superblock_t *sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Jan Kara1ce84862012-04-07 12:50:13 +02001142 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 spin_lock(&journal->j_state_lock);
Jan Kara2e84f262012-08-15 13:50:27 +02001144 /* Is it already empty? */
1145 if (sb->s_start == 0) {
1146 spin_unlock(&journal->j_state_lock);
1147 return;
1148 }
Jan Kara9754e392012-04-07 12:33:03 +02001149 jbd_debug(1, "JBD: Marking journal as empty (seq %d)\n",
1150 journal->j_tail_sequence);
1151
1152 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1153 sb->s_start = cpu_to_be32(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 spin_unlock(&journal->j_state_lock);
Jan Kara9754e392012-04-07 12:33:03 +02001155
Jan Karafd2cbd42012-04-07 11:05:19 +02001156 journal_write_superblock(journal, WRITE_FUA);
Jan Kara9754e392012-04-07 12:33:03 +02001157
1158 spin_lock(&journal->j_state_lock);
1159 /* Log is empty */
1160 journal->j_flags |= JFS_FLUSHED;
1161 spin_unlock(&journal->j_state_lock);
1162}
1163
1164/**
1165 * journal_update_sb_errno() - Update error in the journal.
1166 * @journal: The journal to update.
1167 *
1168 * Update a journal's errno. Write updated superblock to disk waiting for IO
1169 * to complete.
1170 */
1171static void journal_update_sb_errno(journal_t *journal)
1172{
1173 journal_superblock_t *sb = journal->j_superblock;
1174
1175 spin_lock(&journal->j_state_lock);
1176 jbd_debug(1, "JBD: updating superblock error (errno %d)\n",
1177 journal->j_errno);
1178 sb->s_errno = cpu_to_be32(journal->j_errno);
1179 spin_unlock(&journal->j_state_lock);
1180
Jan Karafd2cbd42012-04-07 11:05:19 +02001181 journal_write_superblock(journal, WRITE_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182}
1183
1184/*
1185 * Read the superblock for a given journal, performing initial
1186 * validation of the format.
1187 */
1188
1189static int journal_get_superblock(journal_t *journal)
1190{
1191 struct buffer_head *bh;
1192 journal_superblock_t *sb;
1193 int err = -EIO;
1194
1195 bh = journal->j_sb_buffer;
1196
1197 J_ASSERT(bh != NULL);
1198 if (!buffer_uptodate(bh)) {
1199 ll_rw_block(READ, 1, &bh);
1200 wait_on_buffer(bh);
1201 if (!buffer_uptodate(bh)) {
1202 printk (KERN_ERR
1203 "JBD: IO error reading journal superblock\n");
1204 goto out;
1205 }
1206 }
1207
1208 sb = journal->j_superblock;
1209
1210 err = -EINVAL;
1211
1212 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1213 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1214 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1215 goto out;
1216 }
1217
1218 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1219 case JFS_SUPERBLOCK_V1:
1220 journal->j_format_version = 1;
1221 break;
1222 case JFS_SUPERBLOCK_V2:
1223 journal->j_format_version = 2;
1224 break;
1225 default:
1226 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1227 goto out;
1228 }
1229
1230 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1231 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1232 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1233 printk (KERN_WARNING "JBD: journal file too short\n");
1234 goto out;
1235 }
1236
Eryu Guan87622022011-11-01 19:04:59 -04001237 if (be32_to_cpu(sb->s_first) == 0 ||
1238 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1239 printk(KERN_WARNING
1240 "JBD: Invalid start block of journal: %u\n",
1241 be32_to_cpu(sb->s_first));
1242 goto out;
1243 }
1244
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 return 0;
1246
1247out:
1248 journal_fail_superblock(journal);
1249 return err;
1250}
1251
1252/*
1253 * Load the on-disk journal superblock and read the key fields into the
1254 * journal_t.
1255 */
1256
1257static int load_superblock(journal_t *journal)
1258{
1259 int err;
1260 journal_superblock_t *sb;
1261
1262 err = journal_get_superblock(journal);
1263 if (err)
1264 return err;
1265
1266 sb = journal->j_superblock;
1267
1268 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1269 journal->j_tail = be32_to_cpu(sb->s_start);
1270 journal->j_first = be32_to_cpu(sb->s_first);
1271 journal->j_last = be32_to_cpu(sb->s_maxlen);
1272 journal->j_errno = be32_to_cpu(sb->s_errno);
1273
1274 return 0;
1275}
1276
1277
1278/**
1279 * int journal_load() - Read journal from disk.
1280 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001281 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 * Given a journal_t structure which tells us which disk blocks contain
1283 * a journal, read the journal from disk to initialise the in-memory
1284 * structures.
1285 */
1286int journal_load(journal_t *journal)
1287{
1288 int err;
Badari Pulavartyea817392006-08-27 01:23:52 -07001289 journal_superblock_t *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 err = load_superblock(journal);
1292 if (err)
1293 return err;
1294
Badari Pulavartyea817392006-08-27 01:23:52 -07001295 sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 /* If this is a V2 superblock, then we have to check the
1297 * features flags on it. */
1298
1299 if (journal->j_format_version >= 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 if ((sb->s_feature_ro_compat &
1301 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1302 (sb->s_feature_incompat &
1303 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1304 printk (KERN_WARNING
1305 "JBD: Unrecognised features on journal\n");
1306 return -EINVAL;
1307 }
1308 }
1309
1310 /* Let the recovery code check whether it needs to recover any
1311 * data from the journal. */
1312 if (journal_recover(journal))
1313 goto recovery_error;
1314
1315 /* OK, we've finished with the dynamic journal bits:
1316 * reinitialise the dynamic contents of the superblock in memory
1317 * and reset them on disk. */
1318 if (journal_reset(journal))
1319 goto recovery_error;
1320
1321 journal->j_flags &= ~JFS_ABORT;
1322 journal->j_flags |= JFS_LOADED;
1323 return 0;
1324
1325recovery_error:
1326 printk (KERN_WARNING "JBD: recovery failed\n");
1327 return -EIO;
1328}
1329
1330/**
1331 * void journal_destroy() - Release a journal_t structure.
1332 * @journal: Journal to act on.
1333 *
1334 * Release a journal_t structure once it is no longer in use by the
1335 * journaled object.
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001336 * Return <0 if we couldn't clean up the journal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 */
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001338int journal_destroy(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339{
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001340 int err = 0;
1341
Jan Kara03f4d802010-04-15 22:16:24 +02001342
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 /* Wait for the commit thread to wake up and die. */
1344 journal_kill_thread(journal);
1345
1346 /* Force a final log commit */
1347 if (journal->j_running_transaction)
1348 journal_commit_transaction(journal);
1349
1350 /* Force any old transactions to disk */
1351
Jan Kara1ce84862012-04-07 12:50:13 +02001352 /* We cannot race with anybody but must keep assertions happy */
1353 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 /* Totally anal locking here... */
1355 spin_lock(&journal->j_list_lock);
1356 while (journal->j_checkpoint_transactions != NULL) {
1357 spin_unlock(&journal->j_list_lock);
1358 log_do_checkpoint(journal);
1359 spin_lock(&journal->j_list_lock);
1360 }
1361
1362 J_ASSERT(journal->j_running_transaction == NULL);
1363 J_ASSERT(journal->j_committing_transaction == NULL);
1364 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1365 spin_unlock(&journal->j_list_lock);
1366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 if (journal->j_sb_buffer) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001368 if (!is_journal_aborted(journal)) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001369 journal->j_tail_sequence =
1370 ++journal->j_transaction_sequence;
Jan Kara9754e392012-04-07 12:33:03 +02001371 mark_journal_empty(journal);
1372 } else
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001373 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 brelse(journal->j_sb_buffer);
1375 }
Jan Kara1ce84862012-04-07 12:50:13 +02001376 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
1378 if (journal->j_inode)
1379 iput(journal->j_inode);
1380 if (journal->j_revoke)
1381 journal_destroy_revoke(journal);
1382 kfree(journal->j_wbuf);
1383 kfree(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001384
1385 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386}
1387
1388
1389/**
1390 *int journal_check_used_features () - Check if features specified are used.
1391 * @journal: Journal to check.
1392 * @compat: bitmask of compatible features
1393 * @ro: bitmask of features that force read-only mount
1394 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001395 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 * Check whether the journal uses all of a given set of
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001397 * features. Return true (non-zero) if it does.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 **/
1399
1400int journal_check_used_features (journal_t *journal, unsigned long compat,
1401 unsigned long ro, unsigned long incompat)
1402{
1403 journal_superblock_t *sb;
1404
1405 if (!compat && !ro && !incompat)
1406 return 1;
1407 if (journal->j_format_version == 1)
1408 return 0;
1409
1410 sb = journal->j_superblock;
1411
1412 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1413 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1414 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1415 return 1;
1416
1417 return 0;
1418}
1419
1420/**
1421 * int journal_check_available_features() - Check feature set in journalling layer
1422 * @journal: Journal to check.
1423 * @compat: bitmask of compatible features
1424 * @ro: bitmask of features that force read-only mount
1425 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001426 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 * Check whether the journaling code supports the use of
1428 * all of a given set of features on this journal. Return true
1429 * (non-zero) if it can. */
1430
1431int journal_check_available_features (journal_t *journal, unsigned long compat,
1432 unsigned long ro, unsigned long incompat)
1433{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 if (!compat && !ro && !incompat)
1435 return 1;
1436
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 /* We can support any known requested features iff the
1438 * superblock is in version 2. Otherwise we fail to support any
1439 * extended sb features. */
1440
1441 if (journal->j_format_version != 2)
1442 return 0;
1443
1444 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1445 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1446 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1447 return 1;
1448
1449 return 0;
1450}
1451
1452/**
1453 * int journal_set_features () - Mark a given journal feature in the superblock
1454 * @journal: Journal to act on.
1455 * @compat: bitmask of compatible features
1456 * @ro: bitmask of features that force read-only mount
1457 * @incompat: bitmask of incompatible features
1458 *
1459 * Mark a given journal feature as present on the
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001460 * superblock. Returns true if the requested features could be set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 *
1462 */
1463
1464int journal_set_features (journal_t *journal, unsigned long compat,
1465 unsigned long ro, unsigned long incompat)
1466{
1467 journal_superblock_t *sb;
1468
1469 if (journal_check_used_features(journal, compat, ro, incompat))
1470 return 1;
1471
1472 if (!journal_check_available_features(journal, compat, ro, incompat))
1473 return 0;
1474
1475 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1476 compat, ro, incompat);
1477
1478 sb = journal->j_superblock;
1479
1480 sb->s_feature_compat |= cpu_to_be32(compat);
1481 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1482 sb->s_feature_incompat |= cpu_to_be32(incompat);
1483
1484 return 1;
1485}
1486
1487
1488/**
1489 * int journal_update_format () - Update on-disk journal structure.
1490 * @journal: Journal to act on.
1491 *
1492 * Given an initialised but unloaded journal struct, poke about in the
1493 * on-disk structure to update it to the most recent supported version.
1494 */
1495int journal_update_format (journal_t *journal)
1496{
1497 journal_superblock_t *sb;
1498 int err;
1499
1500 err = journal_get_superblock(journal);
1501 if (err)
1502 return err;
1503
1504 sb = journal->j_superblock;
1505
1506 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1507 case JFS_SUPERBLOCK_V2:
1508 return 0;
1509 case JFS_SUPERBLOCK_V1:
1510 return journal_convert_superblock_v1(journal, sb);
1511 default:
1512 break;
1513 }
1514 return -EINVAL;
1515}
1516
1517static int journal_convert_superblock_v1(journal_t *journal,
1518 journal_superblock_t *sb)
1519{
1520 int offset, blocksize;
1521 struct buffer_head *bh;
1522
1523 printk(KERN_WARNING
1524 "JBD: Converting superblock from version 1 to 2.\n");
1525
1526 /* Pre-initialise new fields to zero */
1527 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1528 blocksize = be32_to_cpu(sb->s_blocksize);
1529 memset(&sb->s_feature_compat, 0, blocksize-offset);
1530
1531 sb->s_nr_users = cpu_to_be32(1);
1532 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1533 journal->j_format_version = 2;
1534
1535 bh = journal->j_sb_buffer;
1536 BUFFER_TRACE(bh, "marking dirty");
1537 mark_buffer_dirty(bh);
1538 sync_dirty_buffer(bh);
1539 return 0;
1540}
1541
1542
1543/**
1544 * int journal_flush () - Flush journal
1545 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001546 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 * Flush all data for a given journal to disk and empty the journal.
1548 * Filesystems can use this when remounting readonly to ensure that
1549 * recovery does not need to happen on remount.
1550 */
1551
1552int journal_flush(journal_t *journal)
1553{
1554 int err = 0;
1555 transaction_t *transaction = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
1557 spin_lock(&journal->j_state_lock);
1558
1559 /* Force everything buffered to the log... */
1560 if (journal->j_running_transaction) {
1561 transaction = journal->j_running_transaction;
1562 __log_start_commit(journal, transaction->t_tid);
1563 } else if (journal->j_committing_transaction)
1564 transaction = journal->j_committing_transaction;
1565
1566 /* Wait for the log commit to complete... */
1567 if (transaction) {
1568 tid_t tid = transaction->t_tid;
1569
1570 spin_unlock(&journal->j_state_lock);
1571 log_wait_commit(journal, tid);
1572 } else {
1573 spin_unlock(&journal->j_state_lock);
1574 }
1575
1576 /* ...and flush everything in the log out to disk. */
1577 spin_lock(&journal->j_list_lock);
1578 while (!err && journal->j_checkpoint_transactions != NULL) {
1579 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001580 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 err = log_do_checkpoint(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001582 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 spin_lock(&journal->j_list_lock);
1584 }
1585 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001586
1587 if (is_journal_aborted(journal))
1588 return -EIO;
1589
Jan Kara1ce84862012-04-07 12:50:13 +02001590 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 cleanup_journal_tail(journal);
1592
1593 /* Finally, mark the journal as really needing no recovery.
1594 * This sets s_start==0 in the underlying superblock, which is
1595 * the magic code for a fully-recovered superblock. Any future
1596 * commits of data to the journal will restore the current
1597 * s_start value. */
Jan Kara9754e392012-04-07 12:33:03 +02001598 mark_journal_empty(journal);
Jan Kara1ce84862012-04-07 12:50:13 +02001599 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 spin_lock(&journal->j_state_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 J_ASSERT(!journal->j_running_transaction);
1602 J_ASSERT(!journal->j_committing_transaction);
1603 J_ASSERT(!journal->j_checkpoint_transactions);
1604 J_ASSERT(journal->j_head == journal->j_tail);
1605 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1606 spin_unlock(&journal->j_state_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001607 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608}
1609
1610/**
1611 * int journal_wipe() - Wipe journal contents
1612 * @journal: Journal to act on.
1613 * @write: flag (see below)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001614 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 * Wipe out all of the contents of a journal, safely. This will produce
1616 * a warning if the journal contains any valid recovery information.
1617 * Must be called between journal_init_*() and journal_load().
1618 *
1619 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1620 * we merely suppress recovery.
1621 */
1622
1623int journal_wipe(journal_t *journal, int write)
1624{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 int err = 0;
1626
1627 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1628
1629 err = load_superblock(journal);
1630 if (err)
1631 return err;
1632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 if (!journal->j_tail)
1634 goto no_recovery;
1635
1636 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1637 write ? "Clearing" : "Ignoring");
1638
1639 err = journal_skip_recovery(journal);
Jan Kara1ce84862012-04-07 12:50:13 +02001640 if (write) {
1641 /* Lock to make assertions happy... */
1642 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara9754e392012-04-07 12:33:03 +02001643 mark_journal_empty(journal);
Jan Kara1ce84862012-04-07 12:50:13 +02001644 mutex_unlock(&journal->j_checkpoint_mutex);
1645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
1647 no_recovery:
1648 return err;
1649}
1650
1651/*
1652 * journal_dev_name: format a character string to describe on what
1653 * device this journal is present.
1654 */
1655
Adrian Bunk022a4a72005-09-06 15:16:41 -07001656static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657{
1658 struct block_device *bdev;
1659
1660 if (journal->j_inode)
1661 bdev = journal->j_inode->i_sb->s_bdev;
1662 else
1663 bdev = journal->j_dev;
1664
1665 return bdevname(bdev, buffer);
1666}
1667
1668/*
1669 * Journal abort has very specific semantics, which we describe
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001670 * for journal abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 *
1672 * Two internal function, which provide abort to te jbd layer
1673 * itself are here.
1674 */
1675
1676/*
1677 * Quick version for internal journal use (doesn't lock the journal).
1678 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1679 * and don't attempt to make any other journal updates.
1680 */
Adrian Bunk53308382008-02-06 01:40:12 -08001681static void __journal_abort_hard(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682{
1683 transaction_t *transaction;
1684 char b[BDEVNAME_SIZE];
1685
1686 if (journal->j_flags & JFS_ABORT)
1687 return;
1688
1689 printk(KERN_ERR "Aborting journal on device %s.\n",
1690 journal_dev_name(journal, b));
1691
1692 spin_lock(&journal->j_state_lock);
1693 journal->j_flags |= JFS_ABORT;
1694 transaction = journal->j_running_transaction;
1695 if (transaction)
1696 __log_start_commit(journal, transaction->t_tid);
1697 spin_unlock(&journal->j_state_lock);
1698}
1699
1700/* Soft abort: record the abort error status in the journal superblock,
1701 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001702static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703{
1704 if (journal->j_flags & JFS_ABORT)
1705 return;
1706
1707 if (!journal->j_errno)
1708 journal->j_errno = errno;
1709
1710 __journal_abort_hard(journal);
1711
1712 if (errno)
Jan Kara9754e392012-04-07 12:33:03 +02001713 journal_update_sb_errno(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714}
1715
1716/**
1717 * void journal_abort () - Shutdown the journal immediately.
1718 * @journal: the journal to shutdown.
1719 * @errno: an error number to record in the journal indicating
1720 * the reason for the shutdown.
1721 *
1722 * Perform a complete, immediate shutdown of the ENTIRE
1723 * journal (not of a single transaction). This operation cannot be
1724 * undone without closing and reopening the journal.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001725 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 * The journal_abort function is intended to support higher level error
1727 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1728 * mode.
1729 *
1730 * Journal abort has very specific semantics. Any existing dirty,
1731 * unjournaled buffers in the main filesystem will still be written to
1732 * disk by bdflush, but the journaling mechanism will be suspended
1733 * immediately and no further transaction commits will be honoured.
1734 *
1735 * Any dirty, journaled buffers will be written back to disk without
1736 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1737 * filesystem, but we _do_ attempt to leave as much data as possible
1738 * behind for fsck to use for cleanup.
1739 *
1740 * Any attempt to get a new transaction handle on a journal which is in
1741 * ABORT state will just result in an -EROFS error return. A
1742 * journal_stop on an existing handle will return -EIO if we have
1743 * entered abort state during the update.
1744 *
1745 * Recursive transactions are not disturbed by journal abort until the
1746 * final journal_stop, which will receive the -EIO error.
1747 *
1748 * Finally, the journal_abort call allows the caller to supply an errno
1749 * which will be recorded (if possible) in the journal superblock. This
1750 * allows a client to record failure conditions in the middle of a
1751 * transaction without having to complete the transaction to record the
1752 * failure to disk. ext3_error, for example, now uses this
1753 * functionality.
1754 *
1755 * Errors which originate from within the journaling layer will NOT
1756 * supply an errno; a null errno implies that absolutely no further
1757 * writes are done to the journal (unless there are any already in
1758 * progress).
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001759 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 */
1761
1762void journal_abort(journal_t *journal, int errno)
1763{
1764 __journal_abort_soft(journal, errno);
1765}
1766
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001767/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 * int journal_errno () - returns the journal's error state.
1769 * @journal: journal to examine.
1770 *
1771 * This is the errno numbet set with journal_abort(), the last
1772 * time the journal was mounted - if the journal was stopped
1773 * without calling abort this will be 0.
1774 *
1775 * If the journal has been aborted on this mount time -EROFS will
1776 * be returned.
1777 */
1778int journal_errno(journal_t *journal)
1779{
1780 int err;
1781
1782 spin_lock(&journal->j_state_lock);
1783 if (journal->j_flags & JFS_ABORT)
1784 err = -EROFS;
1785 else
1786 err = journal->j_errno;
1787 spin_unlock(&journal->j_state_lock);
1788 return err;
1789}
1790
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001791/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 * int journal_clear_err () - clears the journal's error state
1793 * @journal: journal to act on.
1794 *
1795 * An error must be cleared or Acked to take a FS out of readonly
1796 * mode.
1797 */
1798int journal_clear_err(journal_t *journal)
1799{
1800 int err = 0;
1801
1802 spin_lock(&journal->j_state_lock);
1803 if (journal->j_flags & JFS_ABORT)
1804 err = -EROFS;
1805 else
1806 journal->j_errno = 0;
1807 spin_unlock(&journal->j_state_lock);
1808 return err;
1809}
1810
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001811/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 * void journal_ack_err() - Ack journal err.
1813 * @journal: journal to act on.
1814 *
1815 * An error must be cleared or Acked to take a FS out of readonly
1816 * mode.
1817 */
1818void journal_ack_err(journal_t *journal)
1819{
1820 spin_lock(&journal->j_state_lock);
1821 if (journal->j_errno)
1822 journal->j_flags |= JFS_ACK_ERR;
1823 spin_unlock(&journal->j_state_lock);
1824}
1825
1826int journal_blocks_per_page(struct inode *inode)
1827{
1828 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1829}
1830
1831/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 * Journal_head storage management
1833 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001834static struct kmem_cache *journal_head_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835#ifdef CONFIG_JBD_DEBUG
1836static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1837#endif
1838
1839static int journal_init_journal_head_cache(void)
1840{
1841 int retval;
1842
Al Viro1076d172008-03-29 03:07:18 +00001843 J_ASSERT(journal_head_cache == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 journal_head_cache = kmem_cache_create("journal_head",
1845 sizeof(struct journal_head),
1846 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001847 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001848 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00001850 if (!journal_head_cache) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 retval = -ENOMEM;
1852 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1853 }
1854 return retval;
1855}
1856
1857static void journal_destroy_journal_head_cache(void)
1858{
Duane Griffin3850f7a2008-07-25 01:46:19 -07001859 if (journal_head_cache) {
1860 kmem_cache_destroy(journal_head_cache);
1861 journal_head_cache = NULL;
1862 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863}
1864
1865/*
1866 * journal_head splicing and dicing
1867 */
1868static struct journal_head *journal_alloc_journal_head(void)
1869{
1870 struct journal_head *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871
1872#ifdef CONFIG_JBD_DEBUG
1873 atomic_inc(&nr_journal_heads);
1874#endif
Zheng Liu8bb9da92013-04-29 17:08:51 +08001875 ret = kmem_cache_zalloc(journal_head_cache, GFP_NOFS);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001876 if (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 jbd_debug(1, "out of memory for journal_head\n");
Namhyung Kimf81e3d42010-10-04 19:12:13 +09001878 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1879 __func__);
1880
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001881 while (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 yield();
Zheng Liu8bb9da92013-04-29 17:08:51 +08001883 ret = kmem_cache_zalloc(journal_head_cache, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 }
1885 }
1886 return ret;
1887}
1888
1889static void journal_free_journal_head(struct journal_head *jh)
1890{
1891#ifdef CONFIG_JBD_DEBUG
1892 atomic_dec(&nr_journal_heads);
Randy Dunlapc9cf5522006-06-27 02:53:52 -07001893 memset(jh, JBD_POISON_FREE, sizeof(*jh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894#endif
1895 kmem_cache_free(journal_head_cache, jh);
1896}
1897
1898/*
1899 * A journal_head is attached to a buffer_head whenever JBD has an
1900 * interest in the buffer.
1901 *
1902 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1903 * is set. This bit is tested in core kernel code where we need to take
1904 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1905 * there.
1906 *
1907 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1908 *
1909 * When a buffer has its BH_JBD bit set it is immune from being released by
1910 * core kernel code, mainly via ->b_count.
1911 *
Jan Karabb189242011-06-24 23:11:59 +02001912 * A journal_head is detached from its buffer_head when the journal_head's
1913 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
1914 * transaction (b_cp_transaction) hold their references to b_jcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 *
1916 * Various places in the kernel want to attach a journal_head to a buffer_head
1917 * _before_ attaching the journal_head to a transaction. To protect the
1918 * journal_head in this situation, journal_add_journal_head elevates the
1919 * journal_head's b_jcount refcount by one. The caller must call
1920 * journal_put_journal_head() to undo this.
1921 *
1922 * So the typical usage would be:
1923 *
1924 * (Attach a journal_head if needed. Increments b_jcount)
1925 * struct journal_head *jh = journal_add_journal_head(bh);
1926 * ...
Jan Karabb189242011-06-24 23:11:59 +02001927 * (Get another reference for transaction)
1928 * journal_grab_journal_head(bh);
1929 * jh->b_transaction = xxx;
1930 * (Put original reference)
1931 * journal_put_journal_head(jh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 */
1933
1934/*
1935 * Give a buffer_head a journal_head.
1936 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 * May sleep.
1938 */
1939struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1940{
1941 struct journal_head *jh;
1942 struct journal_head *new_jh = NULL;
1943
1944repeat:
Zheng Liu8bb9da92013-04-29 17:08:51 +08001945 if (!buffer_jbd(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 new_jh = journal_alloc_journal_head();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
1948 jbd_lock_bh_journal_head(bh);
1949 if (buffer_jbd(bh)) {
1950 jh = bh2jh(bh);
1951 } else {
1952 J_ASSERT_BH(bh,
1953 (atomic_read(&bh->b_count) > 0) ||
1954 (bh->b_page && bh->b_page->mapping));
1955
1956 if (!new_jh) {
1957 jbd_unlock_bh_journal_head(bh);
1958 goto repeat;
1959 }
1960
1961 jh = new_jh;
1962 new_jh = NULL; /* We consumed it */
1963 set_buffer_jbd(bh);
1964 bh->b_private = jh;
1965 jh->b_bh = bh;
1966 get_bh(bh);
1967 BUFFER_TRACE(bh, "added journal_head");
1968 }
1969 jh->b_jcount++;
1970 jbd_unlock_bh_journal_head(bh);
1971 if (new_jh)
1972 journal_free_journal_head(new_jh);
1973 return bh->b_private;
1974}
1975
1976/*
1977 * Grab a ref against this buffer_head's journal_head. If it ended up not
1978 * having a journal_head, return NULL
1979 */
1980struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1981{
1982 struct journal_head *jh = NULL;
1983
1984 jbd_lock_bh_journal_head(bh);
1985 if (buffer_jbd(bh)) {
1986 jh = bh2jh(bh);
1987 jh->b_jcount++;
1988 }
1989 jbd_unlock_bh_journal_head(bh);
1990 return jh;
1991}
1992
1993static void __journal_remove_journal_head(struct buffer_head *bh)
1994{
1995 struct journal_head *jh = bh2jh(bh);
1996
1997 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karabb189242011-06-24 23:11:59 +02001998 J_ASSERT_JH(jh, jh->b_transaction == NULL);
1999 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2000 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2001 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2002 J_ASSERT_BH(bh, buffer_jbd(bh));
2003 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2004 BUFFER_TRACE(bh, "remove journal_head");
2005 if (jh->b_frozen_data) {
2006 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2007 jbd_free(jh->b_frozen_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 }
Jan Karabb189242011-06-24 23:11:59 +02002009 if (jh->b_committed_data) {
2010 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2011 jbd_free(jh->b_committed_data, bh->b_size);
2012 }
2013 bh->b_private = NULL;
2014 jh->b_bh = NULL; /* debug, really */
2015 clear_buffer_jbd(bh);
2016 journal_free_journal_head(jh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017}
2018
2019/*
Jan Karabb189242011-06-24 23:11:59 +02002020 * Drop a reference on the passed journal_head. If it fell to zero then
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 * release the journal_head from the buffer_head.
2022 */
2023void journal_put_journal_head(struct journal_head *jh)
2024{
2025 struct buffer_head *bh = jh2bh(jh);
2026
2027 jbd_lock_bh_journal_head(bh);
2028 J_ASSERT_JH(jh, jh->b_jcount > 0);
2029 --jh->b_jcount;
Jan Karabb189242011-06-24 23:11:59 +02002030 if (!jh->b_jcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 __journal_remove_journal_head(bh);
Jan Karabb189242011-06-24 23:11:59 +02002032 jbd_unlock_bh_journal_head(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 __brelse(bh);
Jan Karabb189242011-06-24 23:11:59 +02002034 } else
2035 jbd_unlock_bh_journal_head(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036}
2037
2038/*
Jose R. Santosc2a91592007-10-18 23:39:22 -07002039 * debugfs tunables
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 */
Jose R. Santosc2a91592007-10-18 23:39:22 -07002041#ifdef CONFIG_JBD_DEBUG
2042
2043u8 journal_enable_debug __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044EXPORT_SYMBOL(journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Jose R. Santosc2a91592007-10-18 23:39:22 -07002046static struct dentry *jbd_debugfs_dir;
2047static struct dentry *jbd_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Jose R. Santosc2a91592007-10-18 23:39:22 -07002049static void __init jbd_create_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050{
Jose R. Santosc2a91592007-10-18 23:39:22 -07002051 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
2052 if (jbd_debugfs_dir)
Yin Kangkai765f8362009-12-15 14:48:25 -08002053 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
Jose R. Santosc2a91592007-10-18 23:39:22 -07002054 jbd_debugfs_dir,
2055 &journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056}
2057
Jose R. Santosc2a91592007-10-18 23:39:22 -07002058static void __exit jbd_remove_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059{
Jose R. Santosc2a91592007-10-18 23:39:22 -07002060 debugfs_remove(jbd_debug);
2061 debugfs_remove(jbd_debugfs_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062}
2063
2064#else
2065
Jose R. Santosc2a91592007-10-18 23:39:22 -07002066static inline void jbd_create_debugfs_entry(void)
2067{
2068}
2069
2070static inline void jbd_remove_debugfs_entry(void)
2071{
2072}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
2074#endif
2075
Christoph Lametere18b8902006-12-06 20:33:20 -08002076struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
2078static int __init journal_init_handle_cache(void)
2079{
2080 jbd_handle_cache = kmem_cache_create("journal_handle",
2081 sizeof(handle_t),
2082 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07002083 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09002084 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 if (jbd_handle_cache == NULL) {
2086 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2087 return -ENOMEM;
2088 }
2089 return 0;
2090}
2091
2092static void journal_destroy_handle_cache(void)
2093{
2094 if (jbd_handle_cache)
2095 kmem_cache_destroy(jbd_handle_cache);
2096}
2097
2098/*
2099 * Module startup and shutdown
2100 */
2101
2102static int __init journal_init_caches(void)
2103{
2104 int ret;
2105
2106 ret = journal_init_revoke_caches();
2107 if (ret == 0)
2108 ret = journal_init_journal_head_cache();
2109 if (ret == 0)
2110 ret = journal_init_handle_cache();
2111 return ret;
2112}
2113
2114static void journal_destroy_caches(void)
2115{
2116 journal_destroy_revoke_caches();
2117 journal_destroy_journal_head_cache();
2118 journal_destroy_handle_cache();
2119}
2120
2121static int __init journal_init(void)
2122{
2123 int ret;
2124
Alexey Dobriyan2aed3482006-09-27 01:49:28 -07002125 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
Adrian Bunk022a4a72005-09-06 15:16:41 -07002126
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 ret = journal_init_caches();
2128 if (ret != 0)
2129 journal_destroy_caches();
Jose R. Santosc2a91592007-10-18 23:39:22 -07002130 jbd_create_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 return ret;
2132}
2133
2134static void __exit journal_exit(void)
2135{
2136#ifdef CONFIG_JBD_DEBUG
2137 int n = atomic_read(&nr_journal_heads);
2138 if (n)
2139 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2140#endif
Jose R. Santosc2a91592007-10-18 23:39:22 -07002141 jbd_remove_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 journal_destroy_caches();
2143}
2144
2145MODULE_LICENSE("GPL");
2146module_init(journal_init);
2147module_exit(journal_exit);
2148