blob: c46a79adb6adf0fe74dec7173f3c1fc7b33e5002 [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)) {
Jan Kara301d4c92013-11-25 23:30:26 +0100576 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 "%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
Jan Kara301d4c92013-11-25 23:30:26 +0100607 if (unlikely(is_journal_aborted(journal)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 return err;
610}
611
612/*
Jan Kara03f4d802010-04-15 22:16:24 +0200613 * Return 1 if a given transaction has not yet sent barrier request
614 * connected with a transaction commit. If 0 is returned, transaction
615 * may or may not have sent the barrier. Used to avoid sending barrier
616 * twice in common cases.
617 */
618int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
619{
620 int ret = 0;
621 transaction_t *commit_trans;
622
623 if (!(journal->j_flags & JFS_BARRIER))
624 return 0;
625 spin_lock(&journal->j_state_lock);
626 /* Transaction already committed? */
627 if (tid_geq(journal->j_commit_sequence, tid))
628 goto out;
629 /*
630 * Transaction is being committed and we already proceeded to
631 * writing commit record?
632 */
633 commit_trans = journal->j_committing_transaction;
634 if (commit_trans && commit_trans->t_tid == tid &&
635 commit_trans->t_state >= T_COMMIT_RECORD)
636 goto out;
637 ret = 1;
638out:
639 spin_unlock(&journal->j_state_lock);
640 return ret;
641}
Jan Kara52779702010-04-15 22:24:26 +0200642EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
Jan Kara03f4d802010-04-15 22:16:24 +0200643
644/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 * Log buffer allocation routines:
646 */
647
Jan Kara9c28cbc2009-08-03 19:21:00 +0200648int journal_next_log_block(journal_t *journal, unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200650 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
652 spin_lock(&journal->j_state_lock);
653 J_ASSERT(journal->j_free > 1);
654
655 blocknr = journal->j_head;
656 journal->j_head++;
657 journal->j_free--;
658 if (journal->j_head == journal->j_last)
659 journal->j_head = journal->j_first;
660 spin_unlock(&journal->j_state_lock);
661 return journal_bmap(journal, blocknr, retp);
662}
663
664/*
665 * Conversion of logical to physical block numbers for the journal
666 *
667 * On external journals the journal blocks are identity-mapped, so
668 * this is a no-op. If needed, we can use j_blk_offset - everything is
669 * ready.
670 */
Jan Kara9c28cbc2009-08-03 19:21:00 +0200671int journal_bmap(journal_t *journal, unsigned int blocknr,
672 unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
674 int err = 0;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200675 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 if (journal->j_inode) {
678 ret = bmap(journal->j_inode, blocknr);
679 if (ret)
680 *retp = ret;
681 else {
682 char b[BDEVNAME_SIZE];
683
684 printk(KERN_ALERT "%s: journal block not found "
Jan Kara9c28cbc2009-08-03 19:21:00 +0200685 "at offset %u on %s\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700686 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 blocknr,
688 bdevname(journal->j_dev, b));
689 err = -EIO;
690 __journal_abort_soft(journal, err);
691 }
692 } else {
693 *retp = blocknr; /* +journal->j_blk_offset */
694 }
695 return err;
696}
697
698/*
699 * We play buffer_head aliasing tricks to write data/metadata blocks to
700 * the journal without copying their contents, but for journal
701 * descriptor blocks we do need to generate bona fide buffers.
702 *
703 * After the caller of journal_get_descriptor_buffer() has finished modifying
704 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
705 * But we don't bother doing that, so there will be coherency problems with
706 * mmaps of blockdevs which hold live JBD-controlled filesystems.
707 */
708struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
709{
710 struct buffer_head *bh;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200711 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 int err;
713
714 err = journal_next_log_block(journal, &blocknr);
715
716 if (err)
717 return NULL;
718
719 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700720 if (!bh)
721 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 lock_buffer(bh);
723 memset(bh->b_data, 0, journal->j_blocksize);
724 set_buffer_uptodate(bh);
725 unlock_buffer(bh);
726 BUFFER_TRACE(bh, "return this buffer");
727 return journal_add_journal_head(bh);
728}
729
730/*
731 * Management for journal control blocks: functions to create and
732 * destroy journal_t structures, and to initialise and read existing
733 * journal blocks from disk. */
734
735/* First: create and setup a journal_t object in memory. We initialise
736 * very few fields yet: that has to wait until we have created the
737 * journal structures from from scratch, or loaded them from disk. */
738
739static journal_t * journal_init_common (void)
740{
741 journal_t *journal;
742 int err;
743
Mingming Cao8c3478a2007-10-18 23:39:20 -0700744 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 if (!journal)
746 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748 init_waitqueue_head(&journal->j_wait_transaction_locked);
749 init_waitqueue_head(&journal->j_wait_logspace);
750 init_waitqueue_head(&journal->j_wait_done_commit);
751 init_waitqueue_head(&journal->j_wait_checkpoint);
752 init_waitqueue_head(&journal->j_wait_commit);
753 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800754 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 spin_lock_init(&journal->j_revoke_lock);
756 spin_lock_init(&journal->j_list_lock);
757 spin_lock_init(&journal->j_state_lock);
758
759 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
760
761 /* The journal is marked for error until we succeed with recovery! */
762 journal->j_flags = JFS_ABORT;
763
764 /* Set up a default-sized revoke table for the new mount. */
765 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
766 if (err) {
767 kfree(journal);
768 goto fail;
769 }
770 return journal;
771fail:
772 return NULL;
773}
774
775/* journal_init_dev and journal_init_inode:
776 *
777 * Create a journal structure assigned some fixed set of disk blocks to
778 * the journal. We don't actually touch those disk blocks yet, but we
779 * need to set up all of the mapping information to tell the journaling
780 * system where the journal blocks are.
781 *
782 */
783
784/**
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700785 * journal_t * journal_init_dev() - creates and initialises a journal structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 * @bdev: Block device on which to create the journal
787 * @fs_dev: Device which hold journalled filesystem for this journal.
788 * @start: Block nr Start of journal.
Eric Sandeen37ed3222006-09-27 01:49:31 -0700789 * @len: Length of the journal in blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 * @blocksize: blocksize of journalling device
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700791 *
792 * Returns: a newly created journal_t *
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700793 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 * journal_init_dev creates a journal which maps a fixed contiguous
795 * range of blocks on an arbitrary block device.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700796 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 */
798journal_t * journal_init_dev(struct block_device *bdev,
799 struct block_device *fs_dev,
800 int start, int len, int blocksize)
801{
802 journal_t *journal = journal_init_common();
803 struct buffer_head *bh;
804 int n;
805
806 if (!journal)
807 return NULL;
808
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 /* journal descriptor can store up to n blocks -bzzz */
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700810 journal->j_blocksize = blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 n = journal->j_blocksize / sizeof(journal_block_tag_t);
812 journal->j_wbufsize = n;
813 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
814 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300815 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700816 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700817 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700819 journal->j_dev = bdev;
820 journal->j_fs_dev = fs_dev;
821 journal->j_blk_offset = start;
822 journal->j_maxlen = len;
823
824 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700825 if (!bh) {
826 printk(KERN_ERR
827 "%s: Cannot get buffer for journal superblock\n",
828 __func__);
829 goto out_err;
830 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700831 journal->j_sb_buffer = bh;
832 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Karaecca9af2009-04-02 16:57:13 -0700833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700835out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800836 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700837 kfree(journal);
838 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839}
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700840
841/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
843 * @inode: An inode to create the journal in
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700844 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 * journal_init_inode creates a journal which maps an on-disk inode as
846 * the journal. The inode must exist already, must support bmap() and
847 * must have all data blocks preallocated.
848 */
849journal_t * journal_init_inode (struct inode *inode)
850{
851 struct buffer_head *bh;
852 journal_t *journal = journal_init_common();
853 int err;
854 int n;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200855 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 if (!journal)
858 return NULL;
859
860 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
861 journal->j_inode = inode;
862 jbd_debug(1,
863 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700864 journal, inode->i_sb->s_id, inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 (long long) inode->i_size,
866 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
867
868 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
869 journal->j_blocksize = inode->i_sb->s_blocksize;
870
871 /* journal descriptor can store up to n blocks -bzzz */
872 n = journal->j_blocksize / sizeof(journal_block_tag_t);
873 journal->j_wbufsize = n;
874 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
875 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300876 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700877 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700878 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
880
881 err = journal_bmap(journal, 0, &blocknr);
882 /* If that failed, give up */
883 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -0800884 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700885 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700886 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 }
888
Gioh Kima49058f2014-09-04 22:36:35 -0400889 bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700890 if (!bh) {
891 printk(KERN_ERR
892 "%s: Cannot get buffer for journal superblock\n",
893 __func__);
894 goto out_err;
895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 journal->j_sb_buffer = bh;
897 journal->j_superblock = (journal_superblock_t *)bh->b_data;
898
899 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700900out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800901 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700902 kfree(journal);
903 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904}
905
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700906/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 * If the journal init or create aborts, we need to mark the journal
908 * superblock as being NULL to prevent the journal destroy from writing
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700909 * back a bogus superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 */
911static void journal_fail_superblock (journal_t *journal)
912{
913 struct buffer_head *bh = journal->j_sb_buffer;
914 brelse(bh);
915 journal->j_sb_buffer = NULL;
916}
917
918/*
919 * Given a journal_t structure, initialise the various fields for
920 * startup of a new journaling session. We use this both when creating
921 * a journal, and after recovering an old journal to reset it for
922 * subsequent use.
923 */
924
925static int journal_reset(journal_t *journal)
926{
927 journal_superblock_t *sb = journal->j_superblock;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200928 unsigned int first, last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930 first = be32_to_cpu(sb->s_first);
931 last = be32_to_cpu(sb->s_maxlen);
Jan Kara7447a662009-07-15 20:36:08 +0200932 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200933 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
Jan Kara7447a662009-07-15 20:36:08 +0200934 first, last);
935 journal_fail_superblock(journal);
936 return -EINVAL;
937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
939 journal->j_first = first;
940 journal->j_last = last;
941
942 journal->j_head = first;
943 journal->j_tail = first;
944 journal->j_free = last - first;
945
946 journal->j_tail_sequence = journal->j_transaction_sequence;
947 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
948 journal->j_commit_request = journal->j_commit_sequence;
949
950 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
951
Jan Kara9754e392012-04-07 12:33:03 +0200952 /*
953 * As a special case, if the on-disk copy is already marked as needing
954 * no recovery (s_start == 0), then we can safely defer the superblock
955 * update until the next commit by setting JFS_FLUSHED. This avoids
956 * attempting a write to a potential-readonly device.
957 */
958 if (sb->s_start == 0) {
959 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
960 "(start %u, seq %d, errno %d)\n",
961 journal->j_tail, journal->j_tail_sequence,
962 journal->j_errno);
963 journal->j_flags |= JFS_FLUSHED;
964 } else {
Jan Kara1ce84862012-04-07 12:50:13 +0200965 /* Lock here to make assertions happy... */
966 mutex_lock(&journal->j_checkpoint_mutex);
Jan Karafd2cbd42012-04-07 11:05:19 +0200967 /*
968 * Update log tail information. We use WRITE_FUA since new
969 * transaction will start reusing journal space and so we
970 * must make sure information about current log tail is on
971 * disk before that.
972 */
973 journal_update_sb_log_tail(journal,
974 journal->j_tail_sequence,
975 journal->j_tail,
976 WRITE_FUA);
Jan Kara1ce84862012-04-07 12:50:13 +0200977 mutex_unlock(&journal->j_checkpoint_mutex);
Jan Kara9754e392012-04-07 12:33:03 +0200978 }
Pavel Emelianov97f06782007-05-08 00:30:42 -0700979 return journal_start_thread(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980}
981
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700982/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 * int journal_create() - Initialise the new journal file
984 * @journal: Journal to create. This structure must have been initialised
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700985 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 * Given a journal_t structure which tells us which disk blocks we can
987 * use, create a new journal superblock and initialise all of the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700988 * journal fields from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 **/
990int journal_create(journal_t *journal)
991{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200992 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 struct buffer_head *bh;
994 journal_superblock_t *sb;
995 int i, err;
996
997 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
998 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
999 journal->j_maxlen);
1000 journal_fail_superblock(journal);
1001 return -EINVAL;
1002 }
1003
1004 if (journal->j_inode == NULL) {
1005 /*
1006 * We don't know what block to start at!
1007 */
1008 printk(KERN_EMERG
1009 "%s: creation of journal on external device!\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001010 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 BUG();
1012 }
1013
1014 /* Zero out the entire journal on disk. We cannot afford to
1015 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
1016 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
1017 for (i = 0; i < journal->j_maxlen; i++) {
1018 err = journal_bmap(journal, i, &blocknr);
1019 if (err)
1020 return err;
1021 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Namhyung Kim2a0e3382010-10-13 17:17:18 +09001022 if (unlikely(!bh))
1023 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 lock_buffer(bh);
1025 memset (bh->b_data, 0, journal->j_blocksize);
1026 BUFFER_TRACE(bh, "marking dirty");
1027 mark_buffer_dirty(bh);
1028 BUFFER_TRACE(bh, "marking uptodate");
1029 set_buffer_uptodate(bh);
1030 unlock_buffer(bh);
1031 __brelse(bh);
1032 }
1033
1034 sync_blockdev(journal->j_dev);
1035 jbd_debug(1, "JBD: journal cleared.\n");
1036
1037 /* OK, fill in the initial static fields in the new superblock */
1038 sb = journal->j_superblock;
1039
1040 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
1041 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1042
1043 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1044 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
1045 sb->s_first = cpu_to_be32(1);
1046
1047 journal->j_transaction_sequence = 1;
1048
1049 journal->j_flags &= ~JFS_ABORT;
1050 journal->j_format_version = 2;
1051
1052 return journal_reset(journal);
1053}
1054
Jan Karafd2cbd42012-04-07 11:05:19 +02001055static void journal_write_superblock(journal_t *journal, int write_op)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 struct buffer_head *bh = journal->j_sb_buffer;
Jan Karafd2cbd42012-04-07 11:05:19 +02001058 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Jan Karafd2cbd42012-04-07 11:05:19 +02001060 trace_journal_write_superblock(journal, write_op);
1061 if (!(journal->j_flags & JFS_BARRIER))
1062 write_op &= ~(REQ_FUA | REQ_FLUSH);
1063 lock_buffer(bh);
Darrick J. Wongdff68252010-10-04 12:35:05 -07001064 if (buffer_write_io_error(bh)) {
1065 char b[BDEVNAME_SIZE];
1066 /*
1067 * Oh, dear. A previous attempt to write the journal
1068 * superblock failed. This could happen because the
1069 * USB device was yanked out. Or it could happen to
1070 * be a transient write error and maybe the block will
1071 * be remapped. Nothing we can do but to retry the
1072 * write and hope for the best.
1073 */
1074 printk(KERN_ERR "JBD: previous I/O error detected "
1075 "for journal superblock update for %s.\n",
1076 journal_dev_name(journal, b));
1077 clear_buffer_write_io_error(bh);
1078 set_buffer_uptodate(bh);
1079 }
1080
Jan Karafd2cbd42012-04-07 11:05:19 +02001081 get_bh(bh);
1082 bh->b_end_io = end_buffer_write_sync;
1083 ret = submit_bh(write_op, bh);
1084 wait_on_buffer(bh);
Jan Kara9754e392012-04-07 12:33:03 +02001085 if (buffer_write_io_error(bh)) {
Jan Kara9754e392012-04-07 12:33:03 +02001086 clear_buffer_write_io_error(bh);
1087 set_buffer_uptodate(bh);
Jan Karafd2cbd42012-04-07 11:05:19 +02001088 ret = -EIO;
1089 }
1090 if (ret) {
1091 char b[BDEVNAME_SIZE];
1092 printk(KERN_ERR "JBD: Error %d detected "
1093 "when updating journal superblock for %s.\n",
1094 ret, journal_dev_name(journal, b));
Jan Kara9754e392012-04-07 12:33:03 +02001095 }
1096}
1097
1098/**
1099 * journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1100 * @journal: The journal to update.
Jan Karafd2cbd42012-04-07 11:05:19 +02001101 * @tail_tid: TID of the new transaction at the tail of the log
1102 * @tail_block: The first block of the transaction at the tail of the log
1103 * @write_op: With which operation should we write the journal sb
Jan Kara9754e392012-04-07 12:33:03 +02001104 *
1105 * Update a journal's superblock information about log tail and write it to
1106 * disk, waiting for the IO to complete.
1107 */
Jan Karafd2cbd42012-04-07 11:05:19 +02001108void journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1109 unsigned int tail_block, int write_op)
Jan Kara9754e392012-04-07 12:33:03 +02001110{
1111 journal_superblock_t *sb = journal->j_superblock;
1112
Jan Kara1ce84862012-04-07 12:50:13 +02001113 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Jan Karafd2cbd42012-04-07 11:05:19 +02001114 jbd_debug(1,"JBD: updating superblock (start %u, seq %u)\n",
1115 tail_block, tail_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Jan Karafd2cbd42012-04-07 11:05:19 +02001117 sb->s_sequence = cpu_to_be32(tail_tid);
1118 sb->s_start = cpu_to_be32(tail_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Jan Karafd2cbd42012-04-07 11:05:19 +02001120 journal_write_superblock(journal, write_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Jan Kara9754e392012-04-07 12:33:03 +02001122 /* Log is no longer empty */
1123 spin_lock(&journal->j_state_lock);
1124 WARN_ON(!sb->s_sequence);
1125 journal->j_flags &= ~JFS_FLUSHED;
1126 spin_unlock(&journal->j_state_lock);
1127}
1128
1129/**
1130 * mark_journal_empty() - Mark on disk journal as empty.
1131 * @journal: The journal to update.
1132 *
1133 * Update a journal's dynamic superblock fields to show that journal is empty.
1134 * Write updated superblock to disk waiting for IO to complete.
1135 */
1136static void mark_journal_empty(journal_t *journal)
1137{
1138 journal_superblock_t *sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Jan Kara1ce84862012-04-07 12:50:13 +02001140 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 spin_lock(&journal->j_state_lock);
Jan Kara2e84f262012-08-15 13:50:27 +02001142 /* Is it already empty? */
1143 if (sb->s_start == 0) {
1144 spin_unlock(&journal->j_state_lock);
1145 return;
1146 }
Jan Kara9754e392012-04-07 12:33:03 +02001147 jbd_debug(1, "JBD: Marking journal as empty (seq %d)\n",
1148 journal->j_tail_sequence);
1149
1150 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1151 sb->s_start = cpu_to_be32(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 spin_unlock(&journal->j_state_lock);
Jan Kara9754e392012-04-07 12:33:03 +02001153
Jan Karafd2cbd42012-04-07 11:05:19 +02001154 journal_write_superblock(journal, WRITE_FUA);
Jan Kara9754e392012-04-07 12:33:03 +02001155
1156 spin_lock(&journal->j_state_lock);
1157 /* Log is empty */
1158 journal->j_flags |= JFS_FLUSHED;
1159 spin_unlock(&journal->j_state_lock);
1160}
1161
1162/**
1163 * journal_update_sb_errno() - Update error in the journal.
1164 * @journal: The journal to update.
1165 *
1166 * Update a journal's errno. Write updated superblock to disk waiting for IO
1167 * to complete.
1168 */
1169static void journal_update_sb_errno(journal_t *journal)
1170{
1171 journal_superblock_t *sb = journal->j_superblock;
1172
1173 spin_lock(&journal->j_state_lock);
1174 jbd_debug(1, "JBD: updating superblock error (errno %d)\n",
1175 journal->j_errno);
1176 sb->s_errno = cpu_to_be32(journal->j_errno);
1177 spin_unlock(&journal->j_state_lock);
1178
Jan Karafd2cbd42012-04-07 11:05:19 +02001179 journal_write_superblock(journal, WRITE_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
1181
1182/*
1183 * Read the superblock for a given journal, performing initial
1184 * validation of the format.
1185 */
1186
1187static int journal_get_superblock(journal_t *journal)
1188{
1189 struct buffer_head *bh;
1190 journal_superblock_t *sb;
1191 int err = -EIO;
1192
1193 bh = journal->j_sb_buffer;
1194
1195 J_ASSERT(bh != NULL);
1196 if (!buffer_uptodate(bh)) {
1197 ll_rw_block(READ, 1, &bh);
1198 wait_on_buffer(bh);
1199 if (!buffer_uptodate(bh)) {
1200 printk (KERN_ERR
1201 "JBD: IO error reading journal superblock\n");
1202 goto out;
1203 }
1204 }
1205
1206 sb = journal->j_superblock;
1207
1208 err = -EINVAL;
1209
1210 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1211 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1212 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1213 goto out;
1214 }
1215
1216 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1217 case JFS_SUPERBLOCK_V1:
1218 journal->j_format_version = 1;
1219 break;
1220 case JFS_SUPERBLOCK_V2:
1221 journal->j_format_version = 2;
1222 break;
1223 default:
1224 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1225 goto out;
1226 }
1227
1228 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1229 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1230 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1231 printk (KERN_WARNING "JBD: journal file too short\n");
1232 goto out;
1233 }
1234
Eryu Guan87622022011-11-01 19:04:59 -04001235 if (be32_to_cpu(sb->s_first) == 0 ||
1236 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1237 printk(KERN_WARNING
1238 "JBD: Invalid start block of journal: %u\n",
1239 be32_to_cpu(sb->s_first));
1240 goto out;
1241 }
1242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 return 0;
1244
1245out:
1246 journal_fail_superblock(journal);
1247 return err;
1248}
1249
1250/*
1251 * Load the on-disk journal superblock and read the key fields into the
1252 * journal_t.
1253 */
1254
1255static int load_superblock(journal_t *journal)
1256{
1257 int err;
1258 journal_superblock_t *sb;
1259
1260 err = journal_get_superblock(journal);
1261 if (err)
1262 return err;
1263
1264 sb = journal->j_superblock;
1265
1266 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1267 journal->j_tail = be32_to_cpu(sb->s_start);
1268 journal->j_first = be32_to_cpu(sb->s_first);
1269 journal->j_last = be32_to_cpu(sb->s_maxlen);
1270 journal->j_errno = be32_to_cpu(sb->s_errno);
1271
1272 return 0;
1273}
1274
1275
1276/**
1277 * int journal_load() - Read journal from disk.
1278 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001279 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 * Given a journal_t structure which tells us which disk blocks contain
1281 * a journal, read the journal from disk to initialise the in-memory
1282 * structures.
1283 */
1284int journal_load(journal_t *journal)
1285{
1286 int err;
Badari Pulavartyea817392006-08-27 01:23:52 -07001287 journal_superblock_t *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
1289 err = load_superblock(journal);
1290 if (err)
1291 return err;
1292
Badari Pulavartyea817392006-08-27 01:23:52 -07001293 sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 /* If this is a V2 superblock, then we have to check the
1295 * features flags on it. */
1296
1297 if (journal->j_format_version >= 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 if ((sb->s_feature_ro_compat &
1299 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1300 (sb->s_feature_incompat &
1301 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1302 printk (KERN_WARNING
1303 "JBD: Unrecognised features on journal\n");
1304 return -EINVAL;
1305 }
1306 }
1307
1308 /* Let the recovery code check whether it needs to recover any
1309 * data from the journal. */
1310 if (journal_recover(journal))
1311 goto recovery_error;
1312
1313 /* OK, we've finished with the dynamic journal bits:
1314 * reinitialise the dynamic contents of the superblock in memory
1315 * and reset them on disk. */
1316 if (journal_reset(journal))
1317 goto recovery_error;
1318
1319 journal->j_flags &= ~JFS_ABORT;
1320 journal->j_flags |= JFS_LOADED;
1321 return 0;
1322
1323recovery_error:
1324 printk (KERN_WARNING "JBD: recovery failed\n");
1325 return -EIO;
1326}
1327
1328/**
1329 * void journal_destroy() - Release a journal_t structure.
1330 * @journal: Journal to act on.
1331 *
1332 * Release a journal_t structure once it is no longer in use by the
1333 * journaled object.
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001334 * Return <0 if we couldn't clean up the journal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 */
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001336int journal_destroy(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001338 int err = 0;
1339
Jan Kara03f4d802010-04-15 22:16:24 +02001340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 /* Wait for the commit thread to wake up and die. */
1342 journal_kill_thread(journal);
1343
1344 /* Force a final log commit */
1345 if (journal->j_running_transaction)
1346 journal_commit_transaction(journal);
1347
1348 /* Force any old transactions to disk */
1349
Jan Kara1ce84862012-04-07 12:50:13 +02001350 /* We cannot race with anybody but must keep assertions happy */
1351 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 /* Totally anal locking here... */
1353 spin_lock(&journal->j_list_lock);
1354 while (journal->j_checkpoint_transactions != NULL) {
1355 spin_unlock(&journal->j_list_lock);
1356 log_do_checkpoint(journal);
1357 spin_lock(&journal->j_list_lock);
1358 }
1359
1360 J_ASSERT(journal->j_running_transaction == NULL);
1361 J_ASSERT(journal->j_committing_transaction == NULL);
1362 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1363 spin_unlock(&journal->j_list_lock);
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 if (journal->j_sb_buffer) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001366 if (!is_journal_aborted(journal)) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001367 journal->j_tail_sequence =
1368 ++journal->j_transaction_sequence;
Jan Kara9754e392012-04-07 12:33:03 +02001369 mark_journal_empty(journal);
1370 } else
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001371 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 brelse(journal->j_sb_buffer);
1373 }
Jan Kara1ce84862012-04-07 12:50:13 +02001374 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
Markus Elfring11cc9f52014-11-16 23:23:19 +01001376 iput(journal->j_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 if (journal->j_revoke)
1378 journal_destroy_revoke(journal);
1379 kfree(journal->j_wbuf);
1380 kfree(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001381
1382 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383}
1384
1385
1386/**
1387 *int journal_check_used_features () - Check if features specified are used.
1388 * @journal: Journal to check.
1389 * @compat: bitmask of compatible features
1390 * @ro: bitmask of features that force read-only mount
1391 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001392 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 * Check whether the journal uses all of a given set of
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001394 * features. Return true (non-zero) if it does.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 **/
1396
1397int journal_check_used_features (journal_t *journal, unsigned long compat,
1398 unsigned long ro, unsigned long incompat)
1399{
1400 journal_superblock_t *sb;
1401
1402 if (!compat && !ro && !incompat)
1403 return 1;
1404 if (journal->j_format_version == 1)
1405 return 0;
1406
1407 sb = journal->j_superblock;
1408
1409 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1410 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1411 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1412 return 1;
1413
1414 return 0;
1415}
1416
1417/**
1418 * int journal_check_available_features() - Check feature set in journalling layer
1419 * @journal: Journal to check.
1420 * @compat: bitmask of compatible features
1421 * @ro: bitmask of features that force read-only mount
1422 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001423 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 * Check whether the journaling code supports the use of
1425 * all of a given set of features on this journal. Return true
1426 * (non-zero) if it can. */
1427
1428int journal_check_available_features (journal_t *journal, unsigned long compat,
1429 unsigned long ro, unsigned long incompat)
1430{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 if (!compat && !ro && !incompat)
1432 return 1;
1433
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 /* We can support any known requested features iff the
1435 * superblock is in version 2. Otherwise we fail to support any
1436 * extended sb features. */
1437
1438 if (journal->j_format_version != 2)
1439 return 0;
1440
1441 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1442 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1443 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1444 return 1;
1445
1446 return 0;
1447}
1448
1449/**
1450 * int journal_set_features () - Mark a given journal feature in the superblock
1451 * @journal: Journal to act on.
1452 * @compat: bitmask of compatible features
1453 * @ro: bitmask of features that force read-only mount
1454 * @incompat: bitmask of incompatible features
1455 *
1456 * Mark a given journal feature as present on the
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001457 * superblock. Returns true if the requested features could be set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 *
1459 */
1460
1461int journal_set_features (journal_t *journal, unsigned long compat,
1462 unsigned long ro, unsigned long incompat)
1463{
1464 journal_superblock_t *sb;
1465
1466 if (journal_check_used_features(journal, compat, ro, incompat))
1467 return 1;
1468
1469 if (!journal_check_available_features(journal, compat, ro, incompat))
1470 return 0;
1471
1472 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1473 compat, ro, incompat);
1474
1475 sb = journal->j_superblock;
1476
1477 sb->s_feature_compat |= cpu_to_be32(compat);
1478 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1479 sb->s_feature_incompat |= cpu_to_be32(incompat);
1480
1481 return 1;
1482}
1483
1484
1485/**
1486 * int journal_update_format () - Update on-disk journal structure.
1487 * @journal: Journal to act on.
1488 *
1489 * Given an initialised but unloaded journal struct, poke about in the
1490 * on-disk structure to update it to the most recent supported version.
1491 */
1492int journal_update_format (journal_t *journal)
1493{
1494 journal_superblock_t *sb;
1495 int err;
1496
1497 err = journal_get_superblock(journal);
1498 if (err)
1499 return err;
1500
1501 sb = journal->j_superblock;
1502
1503 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1504 case JFS_SUPERBLOCK_V2:
1505 return 0;
1506 case JFS_SUPERBLOCK_V1:
1507 return journal_convert_superblock_v1(journal, sb);
1508 default:
1509 break;
1510 }
1511 return -EINVAL;
1512}
1513
1514static int journal_convert_superblock_v1(journal_t *journal,
1515 journal_superblock_t *sb)
1516{
1517 int offset, blocksize;
1518 struct buffer_head *bh;
1519
1520 printk(KERN_WARNING
1521 "JBD: Converting superblock from version 1 to 2.\n");
1522
1523 /* Pre-initialise new fields to zero */
1524 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1525 blocksize = be32_to_cpu(sb->s_blocksize);
1526 memset(&sb->s_feature_compat, 0, blocksize-offset);
1527
1528 sb->s_nr_users = cpu_to_be32(1);
1529 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1530 journal->j_format_version = 2;
1531
1532 bh = journal->j_sb_buffer;
1533 BUFFER_TRACE(bh, "marking dirty");
1534 mark_buffer_dirty(bh);
1535 sync_dirty_buffer(bh);
1536 return 0;
1537}
1538
1539
1540/**
1541 * int journal_flush () - Flush journal
1542 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001543 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 * Flush all data for a given journal to disk and empty the journal.
1545 * Filesystems can use this when remounting readonly to ensure that
1546 * recovery does not need to happen on remount.
1547 */
1548
1549int journal_flush(journal_t *journal)
1550{
1551 int err = 0;
1552 transaction_t *transaction = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554 spin_lock(&journal->j_state_lock);
1555
1556 /* Force everything buffered to the log... */
1557 if (journal->j_running_transaction) {
1558 transaction = journal->j_running_transaction;
1559 __log_start_commit(journal, transaction->t_tid);
1560 } else if (journal->j_committing_transaction)
1561 transaction = journal->j_committing_transaction;
1562
1563 /* Wait for the log commit to complete... */
1564 if (transaction) {
1565 tid_t tid = transaction->t_tid;
1566
1567 spin_unlock(&journal->j_state_lock);
1568 log_wait_commit(journal, tid);
1569 } else {
1570 spin_unlock(&journal->j_state_lock);
1571 }
1572
1573 /* ...and flush everything in the log out to disk. */
1574 spin_lock(&journal->j_list_lock);
1575 while (!err && journal->j_checkpoint_transactions != NULL) {
1576 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001577 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 err = log_do_checkpoint(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001579 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 spin_lock(&journal->j_list_lock);
1581 }
1582 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001583
1584 if (is_journal_aborted(journal))
1585 return -EIO;
1586
Jan Kara1ce84862012-04-07 12:50:13 +02001587 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 cleanup_journal_tail(journal);
1589
1590 /* Finally, mark the journal as really needing no recovery.
1591 * This sets s_start==0 in the underlying superblock, which is
1592 * the magic code for a fully-recovered superblock. Any future
1593 * commits of data to the journal will restore the current
1594 * s_start value. */
Jan Kara9754e392012-04-07 12:33:03 +02001595 mark_journal_empty(journal);
Jan Kara1ce84862012-04-07 12:50:13 +02001596 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 spin_lock(&journal->j_state_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 J_ASSERT(!journal->j_running_transaction);
1599 J_ASSERT(!journal->j_committing_transaction);
1600 J_ASSERT(!journal->j_checkpoint_transactions);
1601 J_ASSERT(journal->j_head == journal->j_tail);
1602 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1603 spin_unlock(&journal->j_state_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001604 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605}
1606
1607/**
1608 * int journal_wipe() - Wipe journal contents
1609 * @journal: Journal to act on.
1610 * @write: flag (see below)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001611 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 * Wipe out all of the contents of a journal, safely. This will produce
1613 * a warning if the journal contains any valid recovery information.
1614 * Must be called between journal_init_*() and journal_load().
1615 *
1616 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1617 * we merely suppress recovery.
1618 */
1619
1620int journal_wipe(journal_t *journal, int write)
1621{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 int err = 0;
1623
1624 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1625
1626 err = load_superblock(journal);
1627 if (err)
1628 return err;
1629
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 if (!journal->j_tail)
1631 goto no_recovery;
1632
1633 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1634 write ? "Clearing" : "Ignoring");
1635
1636 err = journal_skip_recovery(journal);
Jan Kara1ce84862012-04-07 12:50:13 +02001637 if (write) {
1638 /* Lock to make assertions happy... */
1639 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara9754e392012-04-07 12:33:03 +02001640 mark_journal_empty(journal);
Jan Kara1ce84862012-04-07 12:50:13 +02001641 mutex_unlock(&journal->j_checkpoint_mutex);
1642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
1644 no_recovery:
1645 return err;
1646}
1647
1648/*
1649 * journal_dev_name: format a character string to describe on what
1650 * device this journal is present.
1651 */
1652
Adrian Bunk022a4a72005-09-06 15:16:41 -07001653static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
1655 struct block_device *bdev;
1656
1657 if (journal->j_inode)
1658 bdev = journal->j_inode->i_sb->s_bdev;
1659 else
1660 bdev = journal->j_dev;
1661
1662 return bdevname(bdev, buffer);
1663}
1664
1665/*
1666 * Journal abort has very specific semantics, which we describe
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001667 * for journal abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 *
1669 * Two internal function, which provide abort to te jbd layer
1670 * itself are here.
1671 */
1672
1673/*
1674 * Quick version for internal journal use (doesn't lock the journal).
1675 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1676 * and don't attempt to make any other journal updates.
1677 */
Adrian Bunk53308382008-02-06 01:40:12 -08001678static void __journal_abort_hard(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
1680 transaction_t *transaction;
1681 char b[BDEVNAME_SIZE];
1682
1683 if (journal->j_flags & JFS_ABORT)
1684 return;
1685
1686 printk(KERN_ERR "Aborting journal on device %s.\n",
1687 journal_dev_name(journal, b));
1688
1689 spin_lock(&journal->j_state_lock);
1690 journal->j_flags |= JFS_ABORT;
1691 transaction = journal->j_running_transaction;
1692 if (transaction)
1693 __log_start_commit(journal, transaction->t_tid);
1694 spin_unlock(&journal->j_state_lock);
1695}
1696
1697/* Soft abort: record the abort error status in the journal superblock,
1698 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001699static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700{
1701 if (journal->j_flags & JFS_ABORT)
1702 return;
1703
1704 if (!journal->j_errno)
1705 journal->j_errno = errno;
1706
1707 __journal_abort_hard(journal);
1708
1709 if (errno)
Jan Kara9754e392012-04-07 12:33:03 +02001710 journal_update_sb_errno(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711}
1712
1713/**
1714 * void journal_abort () - Shutdown the journal immediately.
1715 * @journal: the journal to shutdown.
1716 * @errno: an error number to record in the journal indicating
1717 * the reason for the shutdown.
1718 *
1719 * Perform a complete, immediate shutdown of the ENTIRE
1720 * journal (not of a single transaction). This operation cannot be
1721 * undone without closing and reopening the journal.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001722 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 * The journal_abort function is intended to support higher level error
1724 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1725 * mode.
1726 *
1727 * Journal abort has very specific semantics. Any existing dirty,
1728 * unjournaled buffers in the main filesystem will still be written to
1729 * disk by bdflush, but the journaling mechanism will be suspended
1730 * immediately and no further transaction commits will be honoured.
1731 *
1732 * Any dirty, journaled buffers will be written back to disk without
1733 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1734 * filesystem, but we _do_ attempt to leave as much data as possible
1735 * behind for fsck to use for cleanup.
1736 *
1737 * Any attempt to get a new transaction handle on a journal which is in
1738 * ABORT state will just result in an -EROFS error return. A
1739 * journal_stop on an existing handle will return -EIO if we have
1740 * entered abort state during the update.
1741 *
1742 * Recursive transactions are not disturbed by journal abort until the
1743 * final journal_stop, which will receive the -EIO error.
1744 *
1745 * Finally, the journal_abort call allows the caller to supply an errno
1746 * which will be recorded (if possible) in the journal superblock. This
1747 * allows a client to record failure conditions in the middle of a
1748 * transaction without having to complete the transaction to record the
1749 * failure to disk. ext3_error, for example, now uses this
1750 * functionality.
1751 *
1752 * Errors which originate from within the journaling layer will NOT
1753 * supply an errno; a null errno implies that absolutely no further
1754 * writes are done to the journal (unless there are any already in
1755 * progress).
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001756 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 */
1758
1759void journal_abort(journal_t *journal, int errno)
1760{
1761 __journal_abort_soft(journal, errno);
1762}
1763
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001764/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 * int journal_errno () - returns the journal's error state.
1766 * @journal: journal to examine.
1767 *
1768 * This is the errno numbet set with journal_abort(), the last
1769 * time the journal was mounted - if the journal was stopped
1770 * without calling abort this will be 0.
1771 *
1772 * If the journal has been aborted on this mount time -EROFS will
1773 * be returned.
1774 */
1775int journal_errno(journal_t *journal)
1776{
1777 int err;
1778
1779 spin_lock(&journal->j_state_lock);
1780 if (journal->j_flags & JFS_ABORT)
1781 err = -EROFS;
1782 else
1783 err = journal->j_errno;
1784 spin_unlock(&journal->j_state_lock);
1785 return err;
1786}
1787
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001788/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 * int journal_clear_err () - clears the journal's error state
1790 * @journal: journal to act on.
1791 *
1792 * An error must be cleared or Acked to take a FS out of readonly
1793 * mode.
1794 */
1795int journal_clear_err(journal_t *journal)
1796{
1797 int err = 0;
1798
1799 spin_lock(&journal->j_state_lock);
1800 if (journal->j_flags & JFS_ABORT)
1801 err = -EROFS;
1802 else
1803 journal->j_errno = 0;
1804 spin_unlock(&journal->j_state_lock);
1805 return err;
1806}
1807
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001808/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 * void journal_ack_err() - Ack journal err.
1810 * @journal: journal to act on.
1811 *
1812 * An error must be cleared or Acked to take a FS out of readonly
1813 * mode.
1814 */
1815void journal_ack_err(journal_t *journal)
1816{
1817 spin_lock(&journal->j_state_lock);
1818 if (journal->j_errno)
1819 journal->j_flags |= JFS_ACK_ERR;
1820 spin_unlock(&journal->j_state_lock);
1821}
1822
1823int journal_blocks_per_page(struct inode *inode)
1824{
1825 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1826}
1827
1828/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 * Journal_head storage management
1830 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001831static struct kmem_cache *journal_head_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832#ifdef CONFIG_JBD_DEBUG
1833static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1834#endif
1835
1836static int journal_init_journal_head_cache(void)
1837{
1838 int retval;
1839
Al Viro1076d172008-03-29 03:07:18 +00001840 J_ASSERT(journal_head_cache == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 journal_head_cache = kmem_cache_create("journal_head",
1842 sizeof(struct journal_head),
1843 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001844 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001845 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00001847 if (!journal_head_cache) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 retval = -ENOMEM;
1849 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1850 }
1851 return retval;
1852}
1853
1854static void journal_destroy_journal_head_cache(void)
1855{
Duane Griffin3850f7a2008-07-25 01:46:19 -07001856 if (journal_head_cache) {
1857 kmem_cache_destroy(journal_head_cache);
1858 journal_head_cache = NULL;
1859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860}
1861
1862/*
1863 * journal_head splicing and dicing
1864 */
1865static struct journal_head *journal_alloc_journal_head(void)
1866{
1867 struct journal_head *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
1869#ifdef CONFIG_JBD_DEBUG
1870 atomic_inc(&nr_journal_heads);
1871#endif
Zheng Liu8bb9da92013-04-29 17:08:51 +08001872 ret = kmem_cache_zalloc(journal_head_cache, GFP_NOFS);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001873 if (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 jbd_debug(1, "out of memory for journal_head\n");
Namhyung Kimf81e3d42010-10-04 19:12:13 +09001875 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1876 __func__);
1877
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001878 while (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 yield();
Zheng Liu8bb9da92013-04-29 17:08:51 +08001880 ret = kmem_cache_zalloc(journal_head_cache, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 }
1882 }
1883 return ret;
1884}
1885
1886static void journal_free_journal_head(struct journal_head *jh)
1887{
1888#ifdef CONFIG_JBD_DEBUG
1889 atomic_dec(&nr_journal_heads);
Randy Dunlapc9cf5522006-06-27 02:53:52 -07001890 memset(jh, JBD_POISON_FREE, sizeof(*jh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891#endif
1892 kmem_cache_free(journal_head_cache, jh);
1893}
1894
1895/*
1896 * A journal_head is attached to a buffer_head whenever JBD has an
1897 * interest in the buffer.
1898 *
1899 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1900 * is set. This bit is tested in core kernel code where we need to take
1901 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1902 * there.
1903 *
1904 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1905 *
1906 * When a buffer has its BH_JBD bit set it is immune from being released by
1907 * core kernel code, mainly via ->b_count.
1908 *
Jan Karabb189242011-06-24 23:11:59 +02001909 * A journal_head is detached from its buffer_head when the journal_head's
1910 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
1911 * transaction (b_cp_transaction) hold their references to b_jcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 *
1913 * Various places in the kernel want to attach a journal_head to a buffer_head
1914 * _before_ attaching the journal_head to a transaction. To protect the
1915 * journal_head in this situation, journal_add_journal_head elevates the
1916 * journal_head's b_jcount refcount by one. The caller must call
1917 * journal_put_journal_head() to undo this.
1918 *
1919 * So the typical usage would be:
1920 *
1921 * (Attach a journal_head if needed. Increments b_jcount)
1922 * struct journal_head *jh = journal_add_journal_head(bh);
1923 * ...
Jan Karabb189242011-06-24 23:11:59 +02001924 * (Get another reference for transaction)
1925 * journal_grab_journal_head(bh);
1926 * jh->b_transaction = xxx;
1927 * (Put original reference)
1928 * journal_put_journal_head(jh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 */
1930
1931/*
1932 * Give a buffer_head a journal_head.
1933 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 * May sleep.
1935 */
1936struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1937{
1938 struct journal_head *jh;
1939 struct journal_head *new_jh = NULL;
1940
1941repeat:
Zheng Liu8bb9da92013-04-29 17:08:51 +08001942 if (!buffer_jbd(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 new_jh = journal_alloc_journal_head();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
1945 jbd_lock_bh_journal_head(bh);
1946 if (buffer_jbd(bh)) {
1947 jh = bh2jh(bh);
1948 } else {
1949 J_ASSERT_BH(bh,
1950 (atomic_read(&bh->b_count) > 0) ||
1951 (bh->b_page && bh->b_page->mapping));
1952
1953 if (!new_jh) {
1954 jbd_unlock_bh_journal_head(bh);
1955 goto repeat;
1956 }
1957
1958 jh = new_jh;
1959 new_jh = NULL; /* We consumed it */
1960 set_buffer_jbd(bh);
1961 bh->b_private = jh;
1962 jh->b_bh = bh;
1963 get_bh(bh);
1964 BUFFER_TRACE(bh, "added journal_head");
1965 }
1966 jh->b_jcount++;
1967 jbd_unlock_bh_journal_head(bh);
1968 if (new_jh)
1969 journal_free_journal_head(new_jh);
1970 return bh->b_private;
1971}
1972
1973/*
1974 * Grab a ref against this buffer_head's journal_head. If it ended up not
1975 * having a journal_head, return NULL
1976 */
1977struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1978{
1979 struct journal_head *jh = NULL;
1980
1981 jbd_lock_bh_journal_head(bh);
1982 if (buffer_jbd(bh)) {
1983 jh = bh2jh(bh);
1984 jh->b_jcount++;
1985 }
1986 jbd_unlock_bh_journal_head(bh);
1987 return jh;
1988}
1989
1990static void __journal_remove_journal_head(struct buffer_head *bh)
1991{
1992 struct journal_head *jh = bh2jh(bh);
1993
1994 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karabb189242011-06-24 23:11:59 +02001995 J_ASSERT_JH(jh, jh->b_transaction == NULL);
1996 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1997 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
1998 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1999 J_ASSERT_BH(bh, buffer_jbd(bh));
2000 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2001 BUFFER_TRACE(bh, "remove journal_head");
2002 if (jh->b_frozen_data) {
2003 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2004 jbd_free(jh->b_frozen_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 }
Jan Karabb189242011-06-24 23:11:59 +02002006 if (jh->b_committed_data) {
2007 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2008 jbd_free(jh->b_committed_data, bh->b_size);
2009 }
2010 bh->b_private = NULL;
2011 jh->b_bh = NULL; /* debug, really */
2012 clear_buffer_jbd(bh);
2013 journal_free_journal_head(jh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014}
2015
2016/*
Jan Karabb189242011-06-24 23:11:59 +02002017 * Drop a reference on the passed journal_head. If it fell to zero then
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 * release the journal_head from the buffer_head.
2019 */
2020void journal_put_journal_head(struct journal_head *jh)
2021{
2022 struct buffer_head *bh = jh2bh(jh);
2023
2024 jbd_lock_bh_journal_head(bh);
2025 J_ASSERT_JH(jh, jh->b_jcount > 0);
2026 --jh->b_jcount;
Jan Karabb189242011-06-24 23:11:59 +02002027 if (!jh->b_jcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 __journal_remove_journal_head(bh);
Jan Karabb189242011-06-24 23:11:59 +02002029 jbd_unlock_bh_journal_head(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 __brelse(bh);
Jan Karabb189242011-06-24 23:11:59 +02002031 } else
2032 jbd_unlock_bh_journal_head(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033}
2034
2035/*
Jose R. Santosc2a91592007-10-18 23:39:22 -07002036 * debugfs tunables
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 */
Jose R. Santosc2a91592007-10-18 23:39:22 -07002038#ifdef CONFIG_JBD_DEBUG
2039
2040u8 journal_enable_debug __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041EXPORT_SYMBOL(journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
Jose R. Santosc2a91592007-10-18 23:39:22 -07002043static struct dentry *jbd_debugfs_dir;
2044static struct dentry *jbd_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Jose R. Santosc2a91592007-10-18 23:39:22 -07002046static void __init jbd_create_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047{
Jose R. Santosc2a91592007-10-18 23:39:22 -07002048 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
2049 if (jbd_debugfs_dir)
Yin Kangkai765f8362009-12-15 14:48:25 -08002050 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
Jose R. Santosc2a91592007-10-18 23:39:22 -07002051 jbd_debugfs_dir,
2052 &journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053}
2054
Jose R. Santosc2a91592007-10-18 23:39:22 -07002055static void __exit jbd_remove_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056{
Jose R. Santosc2a91592007-10-18 23:39:22 -07002057 debugfs_remove(jbd_debug);
2058 debugfs_remove(jbd_debugfs_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059}
2060
2061#else
2062
Jose R. Santosc2a91592007-10-18 23:39:22 -07002063static inline void jbd_create_debugfs_entry(void)
2064{
2065}
2066
2067static inline void jbd_remove_debugfs_entry(void)
2068{
2069}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071#endif
2072
Christoph Lametere18b8902006-12-06 20:33:20 -08002073struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
2075static int __init journal_init_handle_cache(void)
2076{
2077 jbd_handle_cache = kmem_cache_create("journal_handle",
2078 sizeof(handle_t),
2079 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07002080 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09002081 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 if (jbd_handle_cache == NULL) {
2083 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2084 return -ENOMEM;
2085 }
2086 return 0;
2087}
2088
2089static void journal_destroy_handle_cache(void)
2090{
2091 if (jbd_handle_cache)
2092 kmem_cache_destroy(jbd_handle_cache);
2093}
2094
2095/*
2096 * Module startup and shutdown
2097 */
2098
2099static int __init journal_init_caches(void)
2100{
2101 int ret;
2102
2103 ret = journal_init_revoke_caches();
2104 if (ret == 0)
2105 ret = journal_init_journal_head_cache();
2106 if (ret == 0)
2107 ret = journal_init_handle_cache();
2108 return ret;
2109}
2110
2111static void journal_destroy_caches(void)
2112{
2113 journal_destroy_revoke_caches();
2114 journal_destroy_journal_head_cache();
2115 journal_destroy_handle_cache();
2116}
2117
2118static int __init journal_init(void)
2119{
2120 int ret;
2121
Alexey Dobriyan2aed3482006-09-27 01:49:28 -07002122 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
Adrian Bunk022a4a72005-09-06 15:16:41 -07002123
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 ret = journal_init_caches();
2125 if (ret != 0)
2126 journal_destroy_caches();
Jose R. Santosc2a91592007-10-18 23:39:22 -07002127 jbd_create_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 return ret;
2129}
2130
2131static void __exit journal_exit(void)
2132{
2133#ifdef CONFIG_JBD_DEBUG
2134 int n = atomic_read(&nr_journal_heads);
2135 if (n)
Jan Kara301d4c92013-11-25 23:30:26 +01002136 printk(KERN_ERR "JBD: leaked %d journal_heads!\n", n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137#endif
Jose R. Santosc2a91592007-10-18 23:39:22 -07002138 jbd_remove_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 journal_destroy_caches();
2140}
2141
2142MODULE_LICENSE("GPL");
2143module_init(journal_init);
2144module_exit(journal_exit);
2145