blob: b3713afaaa9e698337b40c7e1fecac6afbebc1bf [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <asm/uaccess.h>
42#include <asm/page.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44EXPORT_SYMBOL(journal_start);
45EXPORT_SYMBOL(journal_restart);
46EXPORT_SYMBOL(journal_extend);
47EXPORT_SYMBOL(journal_stop);
48EXPORT_SYMBOL(journal_lock_updates);
49EXPORT_SYMBOL(journal_unlock_updates);
50EXPORT_SYMBOL(journal_get_write_access);
51EXPORT_SYMBOL(journal_get_create_access);
52EXPORT_SYMBOL(journal_get_undo_access);
53EXPORT_SYMBOL(journal_dirty_data);
54EXPORT_SYMBOL(journal_dirty_metadata);
55EXPORT_SYMBOL(journal_release_buffer);
56EXPORT_SYMBOL(journal_forget);
57#if 0
58EXPORT_SYMBOL(journal_sync_buffer);
59#endif
60EXPORT_SYMBOL(journal_flush);
61EXPORT_SYMBOL(journal_revoke);
62
63EXPORT_SYMBOL(journal_init_dev);
64EXPORT_SYMBOL(journal_init_inode);
65EXPORT_SYMBOL(journal_update_format);
66EXPORT_SYMBOL(journal_check_used_features);
67EXPORT_SYMBOL(journal_check_available_features);
68EXPORT_SYMBOL(journal_set_features);
69EXPORT_SYMBOL(journal_create);
70EXPORT_SYMBOL(journal_load);
71EXPORT_SYMBOL(journal_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072EXPORT_SYMBOL(journal_abort);
73EXPORT_SYMBOL(journal_errno);
74EXPORT_SYMBOL(journal_ack_err);
75EXPORT_SYMBOL(journal_clear_err);
76EXPORT_SYMBOL(log_wait_commit);
Stefan Schmidtff5e4b52009-11-12 09:53:50 +010077EXPORT_SYMBOL(log_start_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078EXPORT_SYMBOL(journal_start_commit);
79EXPORT_SYMBOL(journal_force_commit_nested);
80EXPORT_SYMBOL(journal_wipe);
81EXPORT_SYMBOL(journal_blocks_per_page);
82EXPORT_SYMBOL(journal_invalidatepage);
83EXPORT_SYMBOL(journal_try_to_free_buffers);
84EXPORT_SYMBOL(journal_force_commit);
85
86static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
Adrian Bunk022a4a72005-09-06 15:16:41 -070087static void __journal_abort_soft (journal_t *journal, int errno);
Darrick J. Wongdff68252010-10-04 12:35:05 -070088static const char *journal_dev_name(journal_t *journal, char *buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90/*
91 * Helper function used to manage commit timeouts
92 */
93
94static void commit_timeout(unsigned long __data)
95{
96 struct task_struct * p = (struct task_struct *) __data;
97
98 wake_up_process(p);
99}
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101/*
102 * kjournald: The main thread function used to manage a logging device
103 * journal.
104 *
105 * This kernel thread is responsible for two things:
106 *
107 * 1) COMMIT: Every so often we need to commit the current state of the
108 * filesystem to disk. The journal thread is responsible for writing
109 * all of the metadata buffers to disk.
110 *
111 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
112 * of the data in that part of the log has been rewritten elsewhere on
113 * the disk. Flushing these old buffers to reclaim space in the log is
114 * known as checkpointing, and this thread is responsible for that job.
115 */
116
Adrian Bunk022a4a72005-09-06 15:16:41 -0700117static int kjournald(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
Andrew Mortone3df1892006-03-25 03:06:53 -0800119 journal_t *journal = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 transaction_t *transaction;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Andrew Mortone3df1892006-03-25 03:06:53 -0800122 /*
123 * Set up an interval timer which can be used to trigger a commit wakeup
124 * after the commit interval expires
125 */
126 setup_timer(&journal->j_commit_timer, commit_timeout,
127 (unsigned long)current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129 /* Record that the journal thread is running */
130 journal->j_task = current;
131 wake_up(&journal->j_wait_done_commit);
132
133 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
134 journal->j_commit_interval / HZ);
135
136 /*
137 * And now, wait forever for commit wakeup events.
138 */
139 spin_lock(&journal->j_state_lock);
140
141loop:
142 if (journal->j_flags & JFS_UNMOUNT)
143 goto end_loop;
144
145 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
146 journal->j_commit_sequence, journal->j_commit_request);
147
148 if (journal->j_commit_sequence != journal->j_commit_request) {
149 jbd_debug(1, "OK, requests differ\n");
150 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800151 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 journal_commit_transaction(journal);
153 spin_lock(&journal->j_state_lock);
154 goto loop;
155 }
156
157 wake_up(&journal->j_wait_done_commit);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700158 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 /*
160 * The simpler the better. Flushing journal isn't a
161 * good idea, because that depends on threads that may
162 * be already stopped.
163 */
164 jbd_debug(1, "Now suspending kjournald\n");
165 spin_unlock(&journal->j_state_lock);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700166 refrigerator();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 spin_lock(&journal->j_state_lock);
168 } else {
169 /*
170 * We assume on resume that commits are already there,
171 * so we don't sleep
172 */
173 DEFINE_WAIT(wait);
174 int should_sleep = 1;
175
176 prepare_to_wait(&journal->j_wait_commit, &wait,
177 TASK_INTERRUPTIBLE);
178 if (journal->j_commit_sequence != journal->j_commit_request)
179 should_sleep = 0;
180 transaction = journal->j_running_transaction;
181 if (transaction && time_after_eq(jiffies,
182 transaction->t_expires))
183 should_sleep = 0;
Mark Fashehcbf0d272005-09-06 15:19:08 -0700184 if (journal->j_flags & JFS_UNMOUNT)
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700185 should_sleep = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 if (should_sleep) {
187 spin_unlock(&journal->j_state_lock);
188 schedule();
189 spin_lock(&journal->j_state_lock);
190 }
191 finish_wait(&journal->j_wait_commit, &wait);
192 }
193
194 jbd_debug(1, "kjournald wakes\n");
195
196 /*
197 * Were we woken up by a commit wakeup event?
198 */
199 transaction = journal->j_running_transaction;
200 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
201 journal->j_commit_request = transaction->t_tid;
202 jbd_debug(1, "woke because of timeout\n");
203 }
204 goto loop;
205
206end_loop:
207 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800208 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 journal->j_task = NULL;
210 wake_up(&journal->j_wait_done_commit);
211 jbd_debug(1, "Journal thread exiting.\n");
212 return 0;
213}
214
Pavel Emelianov97f06782007-05-08 00:30:42 -0700215static int journal_start_thread(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700217 struct task_struct *t;
218
219 t = kthread_run(kjournald, journal, "kjournald");
220 if (IS_ERR(t))
221 return PTR_ERR(t);
222
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700223 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700224 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
227static void journal_kill_thread(journal_t *journal)
228{
229 spin_lock(&journal->j_state_lock);
230 journal->j_flags |= JFS_UNMOUNT;
231
232 while (journal->j_task) {
233 wake_up(&journal->j_wait_commit);
234 spin_unlock(&journal->j_state_lock);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700235 wait_event(journal->j_wait_done_commit,
236 journal->j_task == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 spin_lock(&journal->j_state_lock);
238 }
239 spin_unlock(&journal->j_state_lock);
240}
241
242/*
243 * journal_write_metadata_buffer: write a metadata buffer to the journal.
244 *
245 * Writes a metadata buffer to a given disk block. The actual IO is not
246 * performed but a new buffer_head is constructed which labels the data
247 * to be written with the correct destination disk block.
248 *
249 * Any magic-number escaping which needs to be done will cause a
250 * copy-out here. If the buffer happens to start with the
251 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
252 * magic number is only written to the log for descripter blocks. In
253 * this case, we copy the data and replace the first word with 0, and we
254 * return a result code which indicates that this buffer needs to be
255 * marked as an escaped buffer in the corresponding log descriptor
256 * block. The missing word can then be restored when the block is read
257 * during recovery.
258 *
259 * If the source buffer has already been modified by a new transaction
260 * since we took the last commit snapshot, we use the frozen copy of
261 * that data for IO. If we end up using the existing buffer_head's data
262 * for the write, then we *have* to lock the buffer to prevent anyone
263 * else from using and possibly modifying it while the IO is in
264 * progress.
265 *
266 * The function returns a pointer to the buffer_heads to be used for IO.
267 *
268 * We assume that the journal has already been locked in this function.
269 *
270 * Return value:
271 * <0: Error
272 * >=0: Finished OK
273 *
274 * On success:
275 * Bit 0 set == escape performed on the data
276 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
277 */
278
279int journal_write_metadata_buffer(transaction_t *transaction,
280 struct journal_head *jh_in,
281 struct journal_head **jh_out,
Jan Kara9c28cbc2009-08-03 19:21:00 +0200282 unsigned int blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
284 int need_copy_out = 0;
285 int done_copy_out = 0;
286 int do_escape = 0;
287 char *mapped_data;
288 struct buffer_head *new_bh;
289 struct journal_head *new_jh;
290 struct page *new_page;
291 unsigned int new_offset;
292 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghuaf1015c42009-07-15 21:42:05 +0200293 journal_t *journal = transaction->t_journal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
295 /*
296 * The buffer really shouldn't be locked: only the current committing
297 * transaction is allowed to write it, so nobody else is allowed
298 * to do any IO.
299 *
300 * akpm: except if we're journalling data, and write() output is
301 * also part of a shared mapping, and another thread has
302 * decided to launch a writepage() against this buffer.
303 */
304 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
305
306 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
dingdinghuaf1015c42009-07-15 21:42:05 +0200307 /* keep subsequent assertions sane */
308 new_bh->b_state = 0;
309 init_buffer(new_bh, NULL, NULL);
310 atomic_set(&new_bh->b_count, 1);
311 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
313 /*
314 * If a new transaction has already done a buffer copy-out, then
315 * we use that version of the data for the commit.
316 */
317 jbd_lock_bh_state(bh_in);
318repeat:
319 if (jh_in->b_frozen_data) {
320 done_copy_out = 1;
321 new_page = virt_to_page(jh_in->b_frozen_data);
322 new_offset = offset_in_page(jh_in->b_frozen_data);
323 } else {
324 new_page = jh2bh(jh_in)->b_page;
325 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
326 }
327
328 mapped_data = kmap_atomic(new_page, KM_USER0);
329 /*
330 * Check for escaping
331 */
332 if (*((__be32 *)(mapped_data + new_offset)) ==
333 cpu_to_be32(JFS_MAGIC_NUMBER)) {
334 need_copy_out = 1;
335 do_escape = 1;
336 }
337 kunmap_atomic(mapped_data, KM_USER0);
338
339 /*
340 * Do we need to do a data copy?
341 */
342 if (need_copy_out && !done_copy_out) {
343 char *tmp;
344
345 jbd_unlock_bh_state(bh_in);
Mingming Caoc089d492007-10-16 18:38:25 -0400346 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 jbd_lock_bh_state(bh_in);
348 if (jh_in->b_frozen_data) {
Mingming Caoc089d492007-10-16 18:38:25 -0400349 jbd_free(tmp, bh_in->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 goto repeat;
351 }
352
353 jh_in->b_frozen_data = tmp;
354 mapped_data = kmap_atomic(new_page, KM_USER0);
355 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
356 kunmap_atomic(mapped_data, KM_USER0);
357
358 new_page = virt_to_page(tmp);
359 new_offset = offset_in_page(tmp);
360 done_copy_out = 1;
361 }
362
363 /*
364 * Did we need to do an escaping? Now we've done all the
365 * copying, we can finally do so.
366 */
367 if (do_escape) {
368 mapped_data = kmap_atomic(new_page, KM_USER0);
369 *((unsigned int *)(mapped_data + new_offset)) = 0;
370 kunmap_atomic(mapped_data, KM_USER0);
371 }
372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 set_bh_page(new_bh, new_page, new_offset);
374 new_jh->b_transaction = NULL;
375 new_bh->b_size = jh2bh(jh_in)->b_size;
376 new_bh->b_bdev = transaction->t_journal->j_dev;
377 new_bh->b_blocknr = blocknr;
378 set_buffer_mapped(new_bh);
379 set_buffer_dirty(new_bh);
380
381 *jh_out = new_jh;
382
383 /*
384 * The to-be-written buffer needs to get moved to the io queue,
385 * and the original buffer whose contents we are shadowing or
386 * copying is moved to the transaction's shadow queue.
387 */
388 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghuaf1015c42009-07-15 21:42:05 +0200389 spin_lock(&journal->j_list_lock);
390 __journal_file_buffer(jh_in, transaction, BJ_Shadow);
391 spin_unlock(&journal->j_list_lock);
392 jbd_unlock_bh_state(bh_in);
393
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 JBUFFER_TRACE(new_jh, "file as BJ_IO");
395 journal_file_buffer(new_jh, transaction, BJ_IO);
396
397 return do_escape | (done_copy_out << 1);
398}
399
400/*
401 * Allocation code for the journal file. Manage the space left in the
402 * journal, so that we can begin checkpointing when appropriate.
403 */
404
405/*
406 * __log_space_left: Return the number of free blocks left in the journal.
407 *
408 * Called with the journal already locked.
409 *
410 * Called under j_state_lock
411 */
412
413int __log_space_left(journal_t *journal)
414{
415 int left = journal->j_free;
416
417 assert_spin_locked(&journal->j_state_lock);
418
419 /*
420 * Be pessimistic here about the number of those free blocks which
421 * might be required for log descriptor control blocks.
422 */
423
424#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
425
426 left -= MIN_LOG_RESERVED_BLOCKS;
427
428 if (left <= 0)
429 return 0;
430 left -= (left >> 3);
431 return left;
432}
433
434/*
Jan Kara8fe4cd02009-02-11 13:04:25 -0800435 * Called under j_state_lock. Returns true if a transaction commit was started.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 */
437int __log_start_commit(journal_t *journal, tid_t target)
438{
439 /*
440 * Are we already doing a recent enough commit?
441 */
442 if (!tid_geq(journal->j_commit_request, target)) {
443 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200444 * We want a new commit: OK, mark the request and wakeup the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 * commit thread. We do _not_ do the commit ourselves.
446 */
447
448 journal->j_commit_request = target;
449 jbd_debug(1, "JBD: requesting commit %d/%d\n",
450 journal->j_commit_request,
451 journal->j_commit_sequence);
452 wake_up(&journal->j_wait_commit);
453 return 1;
454 }
455 return 0;
456}
457
458int log_start_commit(journal_t *journal, tid_t tid)
459{
460 int ret;
461
462 spin_lock(&journal->j_state_lock);
463 ret = __log_start_commit(journal, tid);
464 spin_unlock(&journal->j_state_lock);
465 return ret;
466}
467
468/*
469 * Force and wait upon a commit if the calling process is not within
470 * transaction. This is used for forcing out undo-protected data which contains
471 * bitmaps, when the fs is running out of space.
472 *
473 * We can only force the running transaction if we don't have an active handle;
474 * otherwise, we will deadlock.
475 *
476 * Returns true if a transaction was started.
477 */
478int journal_force_commit_nested(journal_t *journal)
479{
480 transaction_t *transaction = NULL;
481 tid_t tid;
482
483 spin_lock(&journal->j_state_lock);
484 if (journal->j_running_transaction && !current->journal_info) {
485 transaction = journal->j_running_transaction;
486 __log_start_commit(journal, transaction->t_tid);
487 } else if (journal->j_committing_transaction)
488 transaction = journal->j_committing_transaction;
489
490 if (!transaction) {
491 spin_unlock(&journal->j_state_lock);
492 return 0; /* Nothing to retry */
493 }
494
495 tid = transaction->t_tid;
496 spin_unlock(&journal->j_state_lock);
497 log_wait_commit(journal, tid);
498 return 1;
499}
500
501/*
502 * Start a commit of the current running transaction (if any). Returns true
Jan Kara8fe4cd02009-02-11 13:04:25 -0800503 * if a transaction is going to be committed (or is currently already
504 * committing), and fills its tid in at *ptid
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 */
506int journal_start_commit(journal_t *journal, tid_t *ptid)
507{
508 int ret = 0;
509
510 spin_lock(&journal->j_state_lock);
511 if (journal->j_running_transaction) {
512 tid_t tid = journal->j_running_transaction->t_tid;
513
Jan Kara8fe4cd02009-02-11 13:04:25 -0800514 __log_start_commit(journal, tid);
515 /* There's a running transaction and we've just made sure
516 * it's commit has been scheduled. */
517 if (ptid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 *ptid = tid;
Jan Kara8fe4cd02009-02-11 13:04:25 -0800519 ret = 1;
520 } else if (journal->j_committing_transaction) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 /*
522 * If ext3_write_super() recently started a commit, then we
523 * have to wait for completion of that transaction
524 */
Jan Kara8fe4cd02009-02-11 13:04:25 -0800525 if (ptid)
526 *ptid = journal->j_committing_transaction->t_tid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 ret = 1;
528 }
529 spin_unlock(&journal->j_state_lock);
530 return ret;
531}
532
533/*
534 * Wait for a specified commit to complete.
535 * The caller may not hold the journal lock.
536 */
537int log_wait_commit(journal_t *journal, tid_t tid)
538{
539 int err = 0;
540
541#ifdef CONFIG_JBD_DEBUG
542 spin_lock(&journal->j_state_lock);
543 if (!tid_geq(journal->j_commit_request, tid)) {
544 printk(KERN_EMERG
545 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700546 __func__, journal->j_commit_request, tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 }
548 spin_unlock(&journal->j_state_lock);
549#endif
550 spin_lock(&journal->j_state_lock);
551 while (tid_gt(tid, journal->j_commit_sequence)) {
552 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
553 tid, journal->j_commit_sequence);
554 wake_up(&journal->j_wait_commit);
555 spin_unlock(&journal->j_state_lock);
556 wait_event(journal->j_wait_done_commit,
557 !tid_gt(tid, journal->j_commit_sequence));
558 spin_lock(&journal->j_state_lock);
559 }
560 spin_unlock(&journal->j_state_lock);
561
562 if (unlikely(is_journal_aborted(journal))) {
563 printk(KERN_EMERG "journal commit I/O error\n");
564 err = -EIO;
565 }
566 return err;
567}
568
569/*
Jan Kara03f4d802010-04-15 22:16:24 +0200570 * Return 1 if a given transaction has not yet sent barrier request
571 * connected with a transaction commit. If 0 is returned, transaction
572 * may or may not have sent the barrier. Used to avoid sending barrier
573 * twice in common cases.
574 */
575int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
576{
577 int ret = 0;
578 transaction_t *commit_trans;
579
580 if (!(journal->j_flags & JFS_BARRIER))
581 return 0;
582 spin_lock(&journal->j_state_lock);
583 /* Transaction already committed? */
584 if (tid_geq(journal->j_commit_sequence, tid))
585 goto out;
586 /*
587 * Transaction is being committed and we already proceeded to
588 * writing commit record?
589 */
590 commit_trans = journal->j_committing_transaction;
591 if (commit_trans && commit_trans->t_tid == tid &&
592 commit_trans->t_state >= T_COMMIT_RECORD)
593 goto out;
594 ret = 1;
595out:
596 spin_unlock(&journal->j_state_lock);
597 return ret;
598}
Jan Kara52779702010-04-15 22:24:26 +0200599EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
Jan Kara03f4d802010-04-15 22:16:24 +0200600
601/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 * Log buffer allocation routines:
603 */
604
Jan Kara9c28cbc2009-08-03 19:21:00 +0200605int journal_next_log_block(journal_t *journal, unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200607 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
609 spin_lock(&journal->j_state_lock);
610 J_ASSERT(journal->j_free > 1);
611
612 blocknr = journal->j_head;
613 journal->j_head++;
614 journal->j_free--;
615 if (journal->j_head == journal->j_last)
616 journal->j_head = journal->j_first;
617 spin_unlock(&journal->j_state_lock);
618 return journal_bmap(journal, blocknr, retp);
619}
620
621/*
622 * Conversion of logical to physical block numbers for the journal
623 *
624 * On external journals the journal blocks are identity-mapped, so
625 * this is a no-op. If needed, we can use j_blk_offset - everything is
626 * ready.
627 */
Jan Kara9c28cbc2009-08-03 19:21:00 +0200628int journal_bmap(journal_t *journal, unsigned int blocknr,
629 unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
631 int err = 0;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200632 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
634 if (journal->j_inode) {
635 ret = bmap(journal->j_inode, blocknr);
636 if (ret)
637 *retp = ret;
638 else {
639 char b[BDEVNAME_SIZE];
640
641 printk(KERN_ALERT "%s: journal block not found "
Jan Kara9c28cbc2009-08-03 19:21:00 +0200642 "at offset %u on %s\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700643 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 blocknr,
645 bdevname(journal->j_dev, b));
646 err = -EIO;
647 __journal_abort_soft(journal, err);
648 }
649 } else {
650 *retp = blocknr; /* +journal->j_blk_offset */
651 }
652 return err;
653}
654
655/*
656 * We play buffer_head aliasing tricks to write data/metadata blocks to
657 * the journal without copying their contents, but for journal
658 * descriptor blocks we do need to generate bona fide buffers.
659 *
660 * After the caller of journal_get_descriptor_buffer() has finished modifying
661 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
662 * But we don't bother doing that, so there will be coherency problems with
663 * mmaps of blockdevs which hold live JBD-controlled filesystems.
664 */
665struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
666{
667 struct buffer_head *bh;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200668 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 int err;
670
671 err = journal_next_log_block(journal, &blocknr);
672
673 if (err)
674 return NULL;
675
676 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700677 if (!bh)
678 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 lock_buffer(bh);
680 memset(bh->b_data, 0, journal->j_blocksize);
681 set_buffer_uptodate(bh);
682 unlock_buffer(bh);
683 BUFFER_TRACE(bh, "return this buffer");
684 return journal_add_journal_head(bh);
685}
686
687/*
688 * Management for journal control blocks: functions to create and
689 * destroy journal_t structures, and to initialise and read existing
690 * journal blocks from disk. */
691
692/* First: create and setup a journal_t object in memory. We initialise
693 * very few fields yet: that has to wait until we have created the
694 * journal structures from from scratch, or loaded them from disk. */
695
696static journal_t * journal_init_common (void)
697{
698 journal_t *journal;
699 int err;
700
Mingming Cao8c3478a2007-10-18 23:39:20 -0700701 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (!journal)
703 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
705 init_waitqueue_head(&journal->j_wait_transaction_locked);
706 init_waitqueue_head(&journal->j_wait_logspace);
707 init_waitqueue_head(&journal->j_wait_done_commit);
708 init_waitqueue_head(&journal->j_wait_checkpoint);
709 init_waitqueue_head(&journal->j_wait_commit);
710 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800711 mutex_init(&journal->j_barrier);
712 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 spin_lock_init(&journal->j_revoke_lock);
714 spin_lock_init(&journal->j_list_lock);
715 spin_lock_init(&journal->j_state_lock);
716
717 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
718
719 /* The journal is marked for error until we succeed with recovery! */
720 journal->j_flags = JFS_ABORT;
721
722 /* Set up a default-sized revoke table for the new mount. */
723 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
724 if (err) {
725 kfree(journal);
726 goto fail;
727 }
728 return journal;
729fail:
730 return NULL;
731}
732
733/* journal_init_dev and journal_init_inode:
734 *
735 * Create a journal structure assigned some fixed set of disk blocks to
736 * the journal. We don't actually touch those disk blocks yet, but we
737 * need to set up all of the mapping information to tell the journaling
738 * system where the journal blocks are.
739 *
740 */
741
742/**
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700743 * journal_t * journal_init_dev() - creates and initialises a journal structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 * @bdev: Block device on which to create the journal
745 * @fs_dev: Device which hold journalled filesystem for this journal.
746 * @start: Block nr Start of journal.
Eric Sandeen37ed3222006-09-27 01:49:31 -0700747 * @len: Length of the journal in blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 * @blocksize: blocksize of journalling device
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700749 *
750 * Returns: a newly created journal_t *
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700751 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 * journal_init_dev creates a journal which maps a fixed contiguous
753 * range of blocks on an arbitrary block device.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700754 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 */
756journal_t * journal_init_dev(struct block_device *bdev,
757 struct block_device *fs_dev,
758 int start, int len, int blocksize)
759{
760 journal_t *journal = journal_init_common();
761 struct buffer_head *bh;
762 int n;
763
764 if (!journal)
765 return NULL;
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 /* journal descriptor can store up to n blocks -bzzz */
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700768 journal->j_blocksize = blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 n = journal->j_blocksize / sizeof(journal_block_tag_t);
770 journal->j_wbufsize = n;
771 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
772 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300773 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700774 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700775 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700777 journal->j_dev = bdev;
778 journal->j_fs_dev = fs_dev;
779 journal->j_blk_offset = start;
780 journal->j_maxlen = len;
781
782 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700783 if (!bh) {
784 printk(KERN_ERR
785 "%s: Cannot get buffer for journal superblock\n",
786 __func__);
787 goto out_err;
788 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700789 journal->j_sb_buffer = bh;
790 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Karaecca9af2009-04-02 16:57:13 -0700791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700793out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800794 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700795 kfree(journal);
796 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797}
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700798
799/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
801 * @inode: An inode to create the journal in
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700802 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 * journal_init_inode creates a journal which maps an on-disk inode as
804 * the journal. The inode must exist already, must support bmap() and
805 * must have all data blocks preallocated.
806 */
807journal_t * journal_init_inode (struct inode *inode)
808{
809 struct buffer_head *bh;
810 journal_t *journal = journal_init_common();
811 int err;
812 int n;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200813 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815 if (!journal)
816 return NULL;
817
818 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
819 journal->j_inode = inode;
820 jbd_debug(1,
821 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700822 journal, inode->i_sb->s_id, inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 (long long) inode->i_size,
824 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
825
826 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
827 journal->j_blocksize = inode->i_sb->s_blocksize;
828
829 /* journal descriptor can store up to n blocks -bzzz */
830 n = journal->j_blocksize / sizeof(journal_block_tag_t);
831 journal->j_wbufsize = n;
832 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
833 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300834 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700835 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700836 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 }
838
839 err = journal_bmap(journal, 0, &blocknr);
840 /* If that failed, give up */
841 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -0800842 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700843 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700844 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 }
846
847 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700848 if (!bh) {
849 printk(KERN_ERR
850 "%s: Cannot get buffer for journal superblock\n",
851 __func__);
852 goto out_err;
853 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 journal->j_sb_buffer = bh;
855 journal->j_superblock = (journal_superblock_t *)bh->b_data;
856
857 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700858out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800859 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700860 kfree(journal);
861 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862}
863
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700864/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 * If the journal init or create aborts, we need to mark the journal
866 * superblock as being NULL to prevent the journal destroy from writing
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700867 * back a bogus superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 */
869static void journal_fail_superblock (journal_t *journal)
870{
871 struct buffer_head *bh = journal->j_sb_buffer;
872 brelse(bh);
873 journal->j_sb_buffer = NULL;
874}
875
876/*
877 * Given a journal_t structure, initialise the various fields for
878 * startup of a new journaling session. We use this both when creating
879 * a journal, and after recovering an old journal to reset it for
880 * subsequent use.
881 */
882
883static int journal_reset(journal_t *journal)
884{
885 journal_superblock_t *sb = journal->j_superblock;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200886 unsigned int first, last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
888 first = be32_to_cpu(sb->s_first);
889 last = be32_to_cpu(sb->s_maxlen);
Jan Kara7447a662009-07-15 20:36:08 +0200890 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200891 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
Jan Kara7447a662009-07-15 20:36:08 +0200892 first, last);
893 journal_fail_superblock(journal);
894 return -EINVAL;
895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 journal->j_first = first;
898 journal->j_last = last;
899
900 journal->j_head = first;
901 journal->j_tail = first;
902 journal->j_free = last - first;
903
904 journal->j_tail_sequence = journal->j_transaction_sequence;
905 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
906 journal->j_commit_request = journal->j_commit_sequence;
907
908 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
909
910 /* Add the dynamic fields and write it to disk. */
911 journal_update_superblock(journal, 1);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700912 return journal_start_thread(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913}
914
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700915/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 * int journal_create() - Initialise the new journal file
917 * @journal: Journal to create. This structure must have been initialised
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700918 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 * Given a journal_t structure which tells us which disk blocks we can
920 * use, create a new journal superblock and initialise all of the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700921 * journal fields from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 **/
923int journal_create(journal_t *journal)
924{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200925 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 struct buffer_head *bh;
927 journal_superblock_t *sb;
928 int i, err;
929
930 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
931 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
932 journal->j_maxlen);
933 journal_fail_superblock(journal);
934 return -EINVAL;
935 }
936
937 if (journal->j_inode == NULL) {
938 /*
939 * We don't know what block to start at!
940 */
941 printk(KERN_EMERG
942 "%s: creation of journal on external device!\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700943 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 BUG();
945 }
946
947 /* Zero out the entire journal on disk. We cannot afford to
948 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
949 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
950 for (i = 0; i < journal->j_maxlen; i++) {
951 err = journal_bmap(journal, i, &blocknr);
952 if (err)
953 return err;
954 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Namhyung Kim2a0e3382010-10-13 17:17:18 +0900955 if (unlikely(!bh))
956 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 lock_buffer(bh);
958 memset (bh->b_data, 0, journal->j_blocksize);
959 BUFFER_TRACE(bh, "marking dirty");
960 mark_buffer_dirty(bh);
961 BUFFER_TRACE(bh, "marking uptodate");
962 set_buffer_uptodate(bh);
963 unlock_buffer(bh);
964 __brelse(bh);
965 }
966
967 sync_blockdev(journal->j_dev);
968 jbd_debug(1, "JBD: journal cleared.\n");
969
970 /* OK, fill in the initial static fields in the new superblock */
971 sb = journal->j_superblock;
972
973 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
974 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
975
976 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
977 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
978 sb->s_first = cpu_to_be32(1);
979
980 journal->j_transaction_sequence = 1;
981
982 journal->j_flags &= ~JFS_ABORT;
983 journal->j_format_version = 2;
984
985 return journal_reset(journal);
986}
987
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700988/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 * void journal_update_superblock() - Update journal sb on disk.
990 * @journal: The journal to update.
991 * @wait: Set to '0' if you don't want to wait for IO completion.
992 *
993 * Update a journal's dynamic superblock fields and write it to disk,
994 * optionally waiting for the IO to complete.
995 */
996void journal_update_superblock(journal_t *journal, int wait)
997{
998 journal_superblock_t *sb = journal->j_superblock;
999 struct buffer_head *bh = journal->j_sb_buffer;
1000
1001 /*
1002 * As a special case, if the on-disk copy is already marked as needing
1003 * no recovery (s_start == 0) and there are no outstanding transactions
1004 * in the filesystem, then we can safely defer the superblock update
1005 * until the next commit by setting JFS_FLUSHED. This avoids
1006 * attempting a write to a potential-readonly device.
1007 */
1008 if (sb->s_start == 0 && journal->j_tail_sequence ==
1009 journal->j_transaction_sequence) {
1010 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
Jan Kara9c28cbc2009-08-03 19:21:00 +02001011 "(start %u, seq %d, errno %d)\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001012 journal->j_tail, journal->j_tail_sequence,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 journal->j_errno);
1014 goto out;
1015 }
1016
Darrick J. Wongdff68252010-10-04 12:35:05 -07001017 if (buffer_write_io_error(bh)) {
1018 char b[BDEVNAME_SIZE];
1019 /*
1020 * Oh, dear. A previous attempt to write the journal
1021 * superblock failed. This could happen because the
1022 * USB device was yanked out. Or it could happen to
1023 * be a transient write error and maybe the block will
1024 * be remapped. Nothing we can do but to retry the
1025 * write and hope for the best.
1026 */
1027 printk(KERN_ERR "JBD: previous I/O error detected "
1028 "for journal superblock update for %s.\n",
1029 journal_dev_name(journal, b));
1030 clear_buffer_write_io_error(bh);
1031 set_buffer_uptodate(bh);
1032 }
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 spin_lock(&journal->j_state_lock);
Jan Kara9c28cbc2009-08-03 19:21:00 +02001035 jbd_debug(1,"JBD: updating superblock (start %u, seq %d, errno %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1037
1038 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1039 sb->s_start = cpu_to_be32(journal->j_tail);
1040 sb->s_errno = cpu_to_be32(journal->j_errno);
1041 spin_unlock(&journal->j_state_lock);
1042
1043 BUFFER_TRACE(bh, "marking dirty");
1044 mark_buffer_dirty(bh);
Darrick J. Wongdff68252010-10-04 12:35:05 -07001045 if (wait) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 sync_dirty_buffer(bh);
Darrick J. Wongdff68252010-10-04 12:35:05 -07001047 if (buffer_write_io_error(bh)) {
1048 char b[BDEVNAME_SIZE];
1049 printk(KERN_ERR "JBD: I/O error detected "
1050 "when updating journal superblock for %s.\n",
1051 journal_dev_name(journal, b));
1052 clear_buffer_write_io_error(bh);
1053 set_buffer_uptodate(bh);
1054 }
1055 } else
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02001056 write_dirty_buffer(bh, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
1058out:
1059 /* If we have just flushed the log (by marking s_start==0), then
1060 * any future commit will have to be careful to update the
1061 * superblock again to re-record the true start of the log. */
1062
1063 spin_lock(&journal->j_state_lock);
1064 if (sb->s_start)
1065 journal->j_flags &= ~JFS_FLUSHED;
1066 else
1067 journal->j_flags |= JFS_FLUSHED;
1068 spin_unlock(&journal->j_state_lock);
1069}
1070
1071/*
1072 * Read the superblock for a given journal, performing initial
1073 * validation of the format.
1074 */
1075
1076static int journal_get_superblock(journal_t *journal)
1077{
1078 struct buffer_head *bh;
1079 journal_superblock_t *sb;
1080 int err = -EIO;
1081
1082 bh = journal->j_sb_buffer;
1083
1084 J_ASSERT(bh != NULL);
1085 if (!buffer_uptodate(bh)) {
1086 ll_rw_block(READ, 1, &bh);
1087 wait_on_buffer(bh);
1088 if (!buffer_uptodate(bh)) {
1089 printk (KERN_ERR
1090 "JBD: IO error reading journal superblock\n");
1091 goto out;
1092 }
1093 }
1094
1095 sb = journal->j_superblock;
1096
1097 err = -EINVAL;
1098
1099 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1100 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1101 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1102 goto out;
1103 }
1104
1105 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1106 case JFS_SUPERBLOCK_V1:
1107 journal->j_format_version = 1;
1108 break;
1109 case JFS_SUPERBLOCK_V2:
1110 journal->j_format_version = 2;
1111 break;
1112 default:
1113 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1114 goto out;
1115 }
1116
1117 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1118 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1119 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1120 printk (KERN_WARNING "JBD: journal file too short\n");
1121 goto out;
1122 }
1123
1124 return 0;
1125
1126out:
1127 journal_fail_superblock(journal);
1128 return err;
1129}
1130
1131/*
1132 * Load the on-disk journal superblock and read the key fields into the
1133 * journal_t.
1134 */
1135
1136static int load_superblock(journal_t *journal)
1137{
1138 int err;
1139 journal_superblock_t *sb;
1140
1141 err = journal_get_superblock(journal);
1142 if (err)
1143 return err;
1144
1145 sb = journal->j_superblock;
1146
1147 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1148 journal->j_tail = be32_to_cpu(sb->s_start);
1149 journal->j_first = be32_to_cpu(sb->s_first);
1150 journal->j_last = be32_to_cpu(sb->s_maxlen);
1151 journal->j_errno = be32_to_cpu(sb->s_errno);
1152
1153 return 0;
1154}
1155
1156
1157/**
1158 * int journal_load() - Read journal from disk.
1159 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001160 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 * Given a journal_t structure which tells us which disk blocks contain
1162 * a journal, read the journal from disk to initialise the in-memory
1163 * structures.
1164 */
1165int journal_load(journal_t *journal)
1166{
1167 int err;
Badari Pulavartyea817392006-08-27 01:23:52 -07001168 journal_superblock_t *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 err = load_superblock(journal);
1171 if (err)
1172 return err;
1173
Badari Pulavartyea817392006-08-27 01:23:52 -07001174 sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 /* If this is a V2 superblock, then we have to check the
1176 * features flags on it. */
1177
1178 if (journal->j_format_version >= 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if ((sb->s_feature_ro_compat &
1180 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1181 (sb->s_feature_incompat &
1182 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1183 printk (KERN_WARNING
1184 "JBD: Unrecognised features on journal\n");
1185 return -EINVAL;
1186 }
1187 }
1188
1189 /* Let the recovery code check whether it needs to recover any
1190 * data from the journal. */
1191 if (journal_recover(journal))
1192 goto recovery_error;
1193
1194 /* OK, we've finished with the dynamic journal bits:
1195 * reinitialise the dynamic contents of the superblock in memory
1196 * and reset them on disk. */
1197 if (journal_reset(journal))
1198 goto recovery_error;
1199
1200 journal->j_flags &= ~JFS_ABORT;
1201 journal->j_flags |= JFS_LOADED;
1202 return 0;
1203
1204recovery_error:
1205 printk (KERN_WARNING "JBD: recovery failed\n");
1206 return -EIO;
1207}
1208
1209/**
1210 * void journal_destroy() - Release a journal_t structure.
1211 * @journal: Journal to act on.
1212 *
1213 * Release a journal_t structure once it is no longer in use by the
1214 * journaled object.
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001215 * Return <0 if we couldn't clean up the journal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 */
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001217int journal_destroy(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218{
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001219 int err = 0;
1220
Jan Kara03f4d802010-04-15 22:16:24 +02001221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 /* Wait for the commit thread to wake up and die. */
1223 journal_kill_thread(journal);
1224
1225 /* Force a final log commit */
1226 if (journal->j_running_transaction)
1227 journal_commit_transaction(journal);
1228
1229 /* Force any old transactions to disk */
1230
1231 /* Totally anal locking here... */
1232 spin_lock(&journal->j_list_lock);
1233 while (journal->j_checkpoint_transactions != NULL) {
1234 spin_unlock(&journal->j_list_lock);
1235 log_do_checkpoint(journal);
1236 spin_lock(&journal->j_list_lock);
1237 }
1238
1239 J_ASSERT(journal->j_running_transaction == NULL);
1240 J_ASSERT(journal->j_committing_transaction == NULL);
1241 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1242 spin_unlock(&journal->j_list_lock);
1243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 if (journal->j_sb_buffer) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001245 if (!is_journal_aborted(journal)) {
1246 /* We can now mark the journal as empty. */
1247 journal->j_tail = 0;
1248 journal->j_tail_sequence =
1249 ++journal->j_transaction_sequence;
1250 journal_update_superblock(journal, 1);
1251 } else {
1252 err = -EIO;
1253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 brelse(journal->j_sb_buffer);
1255 }
1256
1257 if (journal->j_inode)
1258 iput(journal->j_inode);
1259 if (journal->j_revoke)
1260 journal_destroy_revoke(journal);
1261 kfree(journal->j_wbuf);
1262 kfree(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001263
1264 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265}
1266
1267
1268/**
1269 *int journal_check_used_features () - Check if features specified are used.
1270 * @journal: Journal to check.
1271 * @compat: bitmask of compatible features
1272 * @ro: bitmask of features that force read-only mount
1273 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001274 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 * Check whether the journal uses all of a given set of
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001276 * features. Return true (non-zero) if it does.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 **/
1278
1279int journal_check_used_features (journal_t *journal, unsigned long compat,
1280 unsigned long ro, unsigned long incompat)
1281{
1282 journal_superblock_t *sb;
1283
1284 if (!compat && !ro && !incompat)
1285 return 1;
1286 if (journal->j_format_version == 1)
1287 return 0;
1288
1289 sb = journal->j_superblock;
1290
1291 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1292 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1293 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1294 return 1;
1295
1296 return 0;
1297}
1298
1299/**
1300 * int journal_check_available_features() - Check feature set in journalling layer
1301 * @journal: Journal to check.
1302 * @compat: bitmask of compatible features
1303 * @ro: bitmask of features that force read-only mount
1304 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001305 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 * Check whether the journaling code supports the use of
1307 * all of a given set of features on this journal. Return true
1308 * (non-zero) if it can. */
1309
1310int journal_check_available_features (journal_t *journal, unsigned long compat,
1311 unsigned long ro, unsigned long incompat)
1312{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 if (!compat && !ro && !incompat)
1314 return 1;
1315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 /* We can support any known requested features iff the
1317 * superblock is in version 2. Otherwise we fail to support any
1318 * extended sb features. */
1319
1320 if (journal->j_format_version != 2)
1321 return 0;
1322
1323 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1324 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1325 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1326 return 1;
1327
1328 return 0;
1329}
1330
1331/**
1332 * int journal_set_features () - Mark a given journal feature in the superblock
1333 * @journal: Journal to act on.
1334 * @compat: bitmask of compatible features
1335 * @ro: bitmask of features that force read-only mount
1336 * @incompat: bitmask of incompatible features
1337 *
1338 * Mark a given journal feature as present on the
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001339 * superblock. Returns true if the requested features could be set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 *
1341 */
1342
1343int journal_set_features (journal_t *journal, unsigned long compat,
1344 unsigned long ro, unsigned long incompat)
1345{
1346 journal_superblock_t *sb;
1347
1348 if (journal_check_used_features(journal, compat, ro, incompat))
1349 return 1;
1350
1351 if (!journal_check_available_features(journal, compat, ro, incompat))
1352 return 0;
1353
1354 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1355 compat, ro, incompat);
1356
1357 sb = journal->j_superblock;
1358
1359 sb->s_feature_compat |= cpu_to_be32(compat);
1360 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1361 sb->s_feature_incompat |= cpu_to_be32(incompat);
1362
1363 return 1;
1364}
1365
1366
1367/**
1368 * int journal_update_format () - Update on-disk journal structure.
1369 * @journal: Journal to act on.
1370 *
1371 * Given an initialised but unloaded journal struct, poke about in the
1372 * on-disk structure to update it to the most recent supported version.
1373 */
1374int journal_update_format (journal_t *journal)
1375{
1376 journal_superblock_t *sb;
1377 int err;
1378
1379 err = journal_get_superblock(journal);
1380 if (err)
1381 return err;
1382
1383 sb = journal->j_superblock;
1384
1385 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1386 case JFS_SUPERBLOCK_V2:
1387 return 0;
1388 case JFS_SUPERBLOCK_V1:
1389 return journal_convert_superblock_v1(journal, sb);
1390 default:
1391 break;
1392 }
1393 return -EINVAL;
1394}
1395
1396static int journal_convert_superblock_v1(journal_t *journal,
1397 journal_superblock_t *sb)
1398{
1399 int offset, blocksize;
1400 struct buffer_head *bh;
1401
1402 printk(KERN_WARNING
1403 "JBD: Converting superblock from version 1 to 2.\n");
1404
1405 /* Pre-initialise new fields to zero */
1406 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1407 blocksize = be32_to_cpu(sb->s_blocksize);
1408 memset(&sb->s_feature_compat, 0, blocksize-offset);
1409
1410 sb->s_nr_users = cpu_to_be32(1);
1411 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1412 journal->j_format_version = 2;
1413
1414 bh = journal->j_sb_buffer;
1415 BUFFER_TRACE(bh, "marking dirty");
1416 mark_buffer_dirty(bh);
1417 sync_dirty_buffer(bh);
1418 return 0;
1419}
1420
1421
1422/**
1423 * int journal_flush () - Flush journal
1424 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001425 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 * Flush all data for a given journal to disk and empty the journal.
1427 * Filesystems can use this when remounting readonly to ensure that
1428 * recovery does not need to happen on remount.
1429 */
1430
1431int journal_flush(journal_t *journal)
1432{
1433 int err = 0;
1434 transaction_t *transaction = NULL;
Jan Kara9c28cbc2009-08-03 19:21:00 +02001435 unsigned int old_tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
1437 spin_lock(&journal->j_state_lock);
1438
1439 /* Force everything buffered to the log... */
1440 if (journal->j_running_transaction) {
1441 transaction = journal->j_running_transaction;
1442 __log_start_commit(journal, transaction->t_tid);
1443 } else if (journal->j_committing_transaction)
1444 transaction = journal->j_committing_transaction;
1445
1446 /* Wait for the log commit to complete... */
1447 if (transaction) {
1448 tid_t tid = transaction->t_tid;
1449
1450 spin_unlock(&journal->j_state_lock);
1451 log_wait_commit(journal, tid);
1452 } else {
1453 spin_unlock(&journal->j_state_lock);
1454 }
1455
1456 /* ...and flush everything in the log out to disk. */
1457 spin_lock(&journal->j_list_lock);
1458 while (!err && journal->j_checkpoint_transactions != NULL) {
1459 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001460 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 err = log_do_checkpoint(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001462 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 spin_lock(&journal->j_list_lock);
1464 }
1465 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001466
1467 if (is_journal_aborted(journal))
1468 return -EIO;
1469
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 cleanup_journal_tail(journal);
1471
1472 /* Finally, mark the journal as really needing no recovery.
1473 * This sets s_start==0 in the underlying superblock, which is
1474 * the magic code for a fully-recovered superblock. Any future
1475 * commits of data to the journal will restore the current
1476 * s_start value. */
1477 spin_lock(&journal->j_state_lock);
1478 old_tail = journal->j_tail;
1479 journal->j_tail = 0;
1480 spin_unlock(&journal->j_state_lock);
1481 journal_update_superblock(journal, 1);
1482 spin_lock(&journal->j_state_lock);
1483 journal->j_tail = old_tail;
1484
1485 J_ASSERT(!journal->j_running_transaction);
1486 J_ASSERT(!journal->j_committing_transaction);
1487 J_ASSERT(!journal->j_checkpoint_transactions);
1488 J_ASSERT(journal->j_head == journal->j_tail);
1489 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1490 spin_unlock(&journal->j_state_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001491 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492}
1493
1494/**
1495 * int journal_wipe() - Wipe journal contents
1496 * @journal: Journal to act on.
1497 * @write: flag (see below)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001498 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 * Wipe out all of the contents of a journal, safely. This will produce
1500 * a warning if the journal contains any valid recovery information.
1501 * Must be called between journal_init_*() and journal_load().
1502 *
1503 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1504 * we merely suppress recovery.
1505 */
1506
1507int journal_wipe(journal_t *journal, int write)
1508{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 int err = 0;
1510
1511 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1512
1513 err = load_superblock(journal);
1514 if (err)
1515 return err;
1516
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 if (!journal->j_tail)
1518 goto no_recovery;
1519
1520 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1521 write ? "Clearing" : "Ignoring");
1522
1523 err = journal_skip_recovery(journal);
1524 if (write)
1525 journal_update_superblock(journal, 1);
1526
1527 no_recovery:
1528 return err;
1529}
1530
1531/*
1532 * journal_dev_name: format a character string to describe on what
1533 * device this journal is present.
1534 */
1535
Adrian Bunk022a4a72005-09-06 15:16:41 -07001536static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537{
1538 struct block_device *bdev;
1539
1540 if (journal->j_inode)
1541 bdev = journal->j_inode->i_sb->s_bdev;
1542 else
1543 bdev = journal->j_dev;
1544
1545 return bdevname(bdev, buffer);
1546}
1547
1548/*
1549 * Journal abort has very specific semantics, which we describe
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001550 * for journal abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 *
1552 * Two internal function, which provide abort to te jbd layer
1553 * itself are here.
1554 */
1555
1556/*
1557 * Quick version for internal journal use (doesn't lock the journal).
1558 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1559 * and don't attempt to make any other journal updates.
1560 */
Adrian Bunk53308382008-02-06 01:40:12 -08001561static void __journal_abort_hard(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562{
1563 transaction_t *transaction;
1564 char b[BDEVNAME_SIZE];
1565
1566 if (journal->j_flags & JFS_ABORT)
1567 return;
1568
1569 printk(KERN_ERR "Aborting journal on device %s.\n",
1570 journal_dev_name(journal, b));
1571
1572 spin_lock(&journal->j_state_lock);
1573 journal->j_flags |= JFS_ABORT;
1574 transaction = journal->j_running_transaction;
1575 if (transaction)
1576 __log_start_commit(journal, transaction->t_tid);
1577 spin_unlock(&journal->j_state_lock);
1578}
1579
1580/* Soft abort: record the abort error status in the journal superblock,
1581 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001582static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583{
1584 if (journal->j_flags & JFS_ABORT)
1585 return;
1586
1587 if (!journal->j_errno)
1588 journal->j_errno = errno;
1589
1590 __journal_abort_hard(journal);
1591
1592 if (errno)
1593 journal_update_superblock(journal, 1);
1594}
1595
1596/**
1597 * void journal_abort () - Shutdown the journal immediately.
1598 * @journal: the journal to shutdown.
1599 * @errno: an error number to record in the journal indicating
1600 * the reason for the shutdown.
1601 *
1602 * Perform a complete, immediate shutdown of the ENTIRE
1603 * journal (not of a single transaction). This operation cannot be
1604 * undone without closing and reopening the journal.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001605 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 * The journal_abort function is intended to support higher level error
1607 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1608 * mode.
1609 *
1610 * Journal abort has very specific semantics. Any existing dirty,
1611 * unjournaled buffers in the main filesystem will still be written to
1612 * disk by bdflush, but the journaling mechanism will be suspended
1613 * immediately and no further transaction commits will be honoured.
1614 *
1615 * Any dirty, journaled buffers will be written back to disk without
1616 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1617 * filesystem, but we _do_ attempt to leave as much data as possible
1618 * behind for fsck to use for cleanup.
1619 *
1620 * Any attempt to get a new transaction handle on a journal which is in
1621 * ABORT state will just result in an -EROFS error return. A
1622 * journal_stop on an existing handle will return -EIO if we have
1623 * entered abort state during the update.
1624 *
1625 * Recursive transactions are not disturbed by journal abort until the
1626 * final journal_stop, which will receive the -EIO error.
1627 *
1628 * Finally, the journal_abort call allows the caller to supply an errno
1629 * which will be recorded (if possible) in the journal superblock. This
1630 * allows a client to record failure conditions in the middle of a
1631 * transaction without having to complete the transaction to record the
1632 * failure to disk. ext3_error, for example, now uses this
1633 * functionality.
1634 *
1635 * Errors which originate from within the journaling layer will NOT
1636 * supply an errno; a null errno implies that absolutely no further
1637 * writes are done to the journal (unless there are any already in
1638 * progress).
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001639 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 */
1641
1642void journal_abort(journal_t *journal, int errno)
1643{
1644 __journal_abort_soft(journal, errno);
1645}
1646
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001647/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 * int journal_errno () - returns the journal's error state.
1649 * @journal: journal to examine.
1650 *
1651 * This is the errno numbet set with journal_abort(), the last
1652 * time the journal was mounted - if the journal was stopped
1653 * without calling abort this will be 0.
1654 *
1655 * If the journal has been aborted on this mount time -EROFS will
1656 * be returned.
1657 */
1658int journal_errno(journal_t *journal)
1659{
1660 int err;
1661
1662 spin_lock(&journal->j_state_lock);
1663 if (journal->j_flags & JFS_ABORT)
1664 err = -EROFS;
1665 else
1666 err = journal->j_errno;
1667 spin_unlock(&journal->j_state_lock);
1668 return err;
1669}
1670
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001671/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 * int journal_clear_err () - clears the journal's error state
1673 * @journal: journal to act on.
1674 *
1675 * An error must be cleared or Acked to take a FS out of readonly
1676 * mode.
1677 */
1678int journal_clear_err(journal_t *journal)
1679{
1680 int err = 0;
1681
1682 spin_lock(&journal->j_state_lock);
1683 if (journal->j_flags & JFS_ABORT)
1684 err = -EROFS;
1685 else
1686 journal->j_errno = 0;
1687 spin_unlock(&journal->j_state_lock);
1688 return err;
1689}
1690
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001691/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 * void journal_ack_err() - Ack journal err.
1693 * @journal: journal to act on.
1694 *
1695 * An error must be cleared or Acked to take a FS out of readonly
1696 * mode.
1697 */
1698void journal_ack_err(journal_t *journal)
1699{
1700 spin_lock(&journal->j_state_lock);
1701 if (journal->j_errno)
1702 journal->j_flags |= JFS_ACK_ERR;
1703 spin_unlock(&journal->j_state_lock);
1704}
1705
1706int journal_blocks_per_page(struct inode *inode)
1707{
1708 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1709}
1710
1711/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 * Journal_head storage management
1713 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001714static struct kmem_cache *journal_head_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715#ifdef CONFIG_JBD_DEBUG
1716static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1717#endif
1718
1719static int journal_init_journal_head_cache(void)
1720{
1721 int retval;
1722
Al Viro1076d172008-03-29 03:07:18 +00001723 J_ASSERT(journal_head_cache == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 journal_head_cache = kmem_cache_create("journal_head",
1725 sizeof(struct journal_head),
1726 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001727 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001728 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00001730 if (!journal_head_cache) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 retval = -ENOMEM;
1732 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1733 }
1734 return retval;
1735}
1736
1737static void journal_destroy_journal_head_cache(void)
1738{
Duane Griffin3850f7a2008-07-25 01:46:19 -07001739 if (journal_head_cache) {
1740 kmem_cache_destroy(journal_head_cache);
1741 journal_head_cache = NULL;
1742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743}
1744
1745/*
1746 * journal_head splicing and dicing
1747 */
1748static struct journal_head *journal_alloc_journal_head(void)
1749{
1750 struct journal_head *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
1752#ifdef CONFIG_JBD_DEBUG
1753 atomic_inc(&nr_journal_heads);
1754#endif
1755 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001756 if (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 jbd_debug(1, "out of memory for journal_head\n");
Namhyung Kimf81e3d42010-10-04 19:12:13 +09001758 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1759 __func__);
1760
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001761 while (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 yield();
1763 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1764 }
1765 }
1766 return ret;
1767}
1768
1769static void journal_free_journal_head(struct journal_head *jh)
1770{
1771#ifdef CONFIG_JBD_DEBUG
1772 atomic_dec(&nr_journal_heads);
Randy Dunlapc9cf5522006-06-27 02:53:52 -07001773 memset(jh, JBD_POISON_FREE, sizeof(*jh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774#endif
1775 kmem_cache_free(journal_head_cache, jh);
1776}
1777
1778/*
1779 * A journal_head is attached to a buffer_head whenever JBD has an
1780 * interest in the buffer.
1781 *
1782 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1783 * is set. This bit is tested in core kernel code where we need to take
1784 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1785 * there.
1786 *
1787 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1788 *
1789 * When a buffer has its BH_JBD bit set it is immune from being released by
1790 * core kernel code, mainly via ->b_count.
1791 *
1792 * A journal_head may be detached from its buffer_head when the journal_head's
1793 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1794 * Various places in JBD call journal_remove_journal_head() to indicate that the
1795 * journal_head can be dropped if needed.
1796 *
1797 * Various places in the kernel want to attach a journal_head to a buffer_head
1798 * _before_ attaching the journal_head to a transaction. To protect the
1799 * journal_head in this situation, journal_add_journal_head elevates the
1800 * journal_head's b_jcount refcount by one. The caller must call
1801 * journal_put_journal_head() to undo this.
1802 *
1803 * So the typical usage would be:
1804 *
1805 * (Attach a journal_head if needed. Increments b_jcount)
1806 * struct journal_head *jh = journal_add_journal_head(bh);
1807 * ...
1808 * jh->b_transaction = xxx;
1809 * journal_put_journal_head(jh);
1810 *
1811 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1812 * because it has a non-zero b_transaction.
1813 */
1814
1815/*
1816 * Give a buffer_head a journal_head.
1817 *
1818 * Doesn't need the journal lock.
1819 * May sleep.
1820 */
1821struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1822{
1823 struct journal_head *jh;
1824 struct journal_head *new_jh = NULL;
1825
1826repeat:
1827 if (!buffer_jbd(bh)) {
1828 new_jh = journal_alloc_journal_head();
1829 memset(new_jh, 0, sizeof(*new_jh));
1830 }
1831
1832 jbd_lock_bh_journal_head(bh);
1833 if (buffer_jbd(bh)) {
1834 jh = bh2jh(bh);
1835 } else {
1836 J_ASSERT_BH(bh,
1837 (atomic_read(&bh->b_count) > 0) ||
1838 (bh->b_page && bh->b_page->mapping));
1839
1840 if (!new_jh) {
1841 jbd_unlock_bh_journal_head(bh);
1842 goto repeat;
1843 }
1844
1845 jh = new_jh;
1846 new_jh = NULL; /* We consumed it */
1847 set_buffer_jbd(bh);
1848 bh->b_private = jh;
1849 jh->b_bh = bh;
1850 get_bh(bh);
1851 BUFFER_TRACE(bh, "added journal_head");
1852 }
1853 jh->b_jcount++;
1854 jbd_unlock_bh_journal_head(bh);
1855 if (new_jh)
1856 journal_free_journal_head(new_jh);
1857 return bh->b_private;
1858}
1859
1860/*
1861 * Grab a ref against this buffer_head's journal_head. If it ended up not
1862 * having a journal_head, return NULL
1863 */
1864struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1865{
1866 struct journal_head *jh = NULL;
1867
1868 jbd_lock_bh_journal_head(bh);
1869 if (buffer_jbd(bh)) {
1870 jh = bh2jh(bh);
1871 jh->b_jcount++;
1872 }
1873 jbd_unlock_bh_journal_head(bh);
1874 return jh;
1875}
1876
1877static void __journal_remove_journal_head(struct buffer_head *bh)
1878{
1879 struct journal_head *jh = bh2jh(bh);
1880
1881 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1882
1883 get_bh(bh);
1884 if (jh->b_jcount == 0) {
1885 if (jh->b_transaction == NULL &&
1886 jh->b_next_transaction == NULL &&
1887 jh->b_cp_transaction == NULL) {
1888 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1889 J_ASSERT_BH(bh, buffer_jbd(bh));
1890 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1891 BUFFER_TRACE(bh, "remove journal_head");
1892 if (jh->b_frozen_data) {
1893 printk(KERN_WARNING "%s: freeing "
1894 "b_frozen_data\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001895 __func__);
Mingming Caoc089d492007-10-16 18:38:25 -04001896 jbd_free(jh->b_frozen_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 }
1898 if (jh->b_committed_data) {
1899 printk(KERN_WARNING "%s: freeing "
1900 "b_committed_data\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001901 __func__);
Mingming Caoc089d492007-10-16 18:38:25 -04001902 jbd_free(jh->b_committed_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 }
1904 bh->b_private = NULL;
1905 jh->b_bh = NULL; /* debug, really */
1906 clear_buffer_jbd(bh);
1907 __brelse(bh);
1908 journal_free_journal_head(jh);
1909 } else {
1910 BUFFER_TRACE(bh, "journal_head was locked");
1911 }
1912 }
1913}
1914
1915/*
1916 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1917 * and has a zero b_jcount then remove and release its journal_head. If we did
1918 * see that the buffer is not used by any transaction we also "logically"
1919 * decrement ->b_count.
1920 *
1921 * We in fact take an additional increment on ->b_count as a convenience,
1922 * because the caller usually wants to do additional things with the bh
1923 * after calling here.
1924 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1925 * time. Once the caller has run __brelse(), the buffer is eligible for
1926 * reaping by try_to_free_buffers().
1927 */
1928void journal_remove_journal_head(struct buffer_head *bh)
1929{
1930 jbd_lock_bh_journal_head(bh);
1931 __journal_remove_journal_head(bh);
1932 jbd_unlock_bh_journal_head(bh);
1933}
1934
1935/*
1936 * Drop a reference on the passed journal_head. If it fell to zero then try to
1937 * release the journal_head from the buffer_head.
1938 */
1939void journal_put_journal_head(struct journal_head *jh)
1940{
1941 struct buffer_head *bh = jh2bh(jh);
1942
1943 jbd_lock_bh_journal_head(bh);
1944 J_ASSERT_JH(jh, jh->b_jcount > 0);
1945 --jh->b_jcount;
1946 if (!jh->b_jcount && !jh->b_transaction) {
1947 __journal_remove_journal_head(bh);
1948 __brelse(bh);
1949 }
1950 jbd_unlock_bh_journal_head(bh);
1951}
1952
1953/*
Jose R. Santosc2a91592007-10-18 23:39:22 -07001954 * debugfs tunables
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 */
Jose R. Santosc2a91592007-10-18 23:39:22 -07001956#ifdef CONFIG_JBD_DEBUG
1957
1958u8 journal_enable_debug __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959EXPORT_SYMBOL(journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
Jose R. Santosc2a91592007-10-18 23:39:22 -07001961static struct dentry *jbd_debugfs_dir;
1962static struct dentry *jbd_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
Jose R. Santosc2a91592007-10-18 23:39:22 -07001964static void __init jbd_create_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001966 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
1967 if (jbd_debugfs_dir)
Yin Kangkai765f8362009-12-15 14:48:25 -08001968 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
Jose R. Santosc2a91592007-10-18 23:39:22 -07001969 jbd_debugfs_dir,
1970 &journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971}
1972
Jose R. Santosc2a91592007-10-18 23:39:22 -07001973static void __exit jbd_remove_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001975 debugfs_remove(jbd_debug);
1976 debugfs_remove(jbd_debugfs_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977}
1978
1979#else
1980
Jose R. Santosc2a91592007-10-18 23:39:22 -07001981static inline void jbd_create_debugfs_entry(void)
1982{
1983}
1984
1985static inline void jbd_remove_debugfs_entry(void)
1986{
1987}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
1989#endif
1990
Christoph Lametere18b8902006-12-06 20:33:20 -08001991struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
1993static int __init journal_init_handle_cache(void)
1994{
1995 jbd_handle_cache = kmem_cache_create("journal_handle",
1996 sizeof(handle_t),
1997 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001998 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001999 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 if (jbd_handle_cache == NULL) {
2001 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2002 return -ENOMEM;
2003 }
2004 return 0;
2005}
2006
2007static void journal_destroy_handle_cache(void)
2008{
2009 if (jbd_handle_cache)
2010 kmem_cache_destroy(jbd_handle_cache);
2011}
2012
2013/*
2014 * Module startup and shutdown
2015 */
2016
2017static int __init journal_init_caches(void)
2018{
2019 int ret;
2020
2021 ret = journal_init_revoke_caches();
2022 if (ret == 0)
2023 ret = journal_init_journal_head_cache();
2024 if (ret == 0)
2025 ret = journal_init_handle_cache();
2026 return ret;
2027}
2028
2029static void journal_destroy_caches(void)
2030{
2031 journal_destroy_revoke_caches();
2032 journal_destroy_journal_head_cache();
2033 journal_destroy_handle_cache();
2034}
2035
2036static int __init journal_init(void)
2037{
2038 int ret;
2039
Alexey Dobriyan2aed3482006-09-27 01:49:28 -07002040 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
Adrian Bunk022a4a72005-09-06 15:16:41 -07002041
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 ret = journal_init_caches();
2043 if (ret != 0)
2044 journal_destroy_caches();
Jose R. Santosc2a91592007-10-18 23:39:22 -07002045 jbd_create_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 return ret;
2047}
2048
2049static void __exit journal_exit(void)
2050{
2051#ifdef CONFIG_JBD_DEBUG
2052 int n = atomic_read(&nr_journal_heads);
2053 if (n)
2054 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2055#endif
Jose R. Santosc2a91592007-10-18 23:39:22 -07002056 jbd_remove_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 journal_destroy_caches();
2058}
2059
2060MODULE_LICENSE("GPL");
2061module_init(journal_init);
2062module_exit(journal_exit);
2063