blob: 93d1e47647bd48d0cd300358583a80b293ee1a77 [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>
Andrew Morton8d8c8512006-03-25 03:06:53 -080039
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <asm/uaccess.h>
41#include <asm/page.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43EXPORT_SYMBOL(journal_start);
44EXPORT_SYMBOL(journal_restart);
45EXPORT_SYMBOL(journal_extend);
46EXPORT_SYMBOL(journal_stop);
47EXPORT_SYMBOL(journal_lock_updates);
48EXPORT_SYMBOL(journal_unlock_updates);
49EXPORT_SYMBOL(journal_get_write_access);
50EXPORT_SYMBOL(journal_get_create_access);
51EXPORT_SYMBOL(journal_get_undo_access);
52EXPORT_SYMBOL(journal_dirty_data);
53EXPORT_SYMBOL(journal_dirty_metadata);
54EXPORT_SYMBOL(journal_release_buffer);
55EXPORT_SYMBOL(journal_forget);
56#if 0
57EXPORT_SYMBOL(journal_sync_buffer);
58#endif
59EXPORT_SYMBOL(journal_flush);
60EXPORT_SYMBOL(journal_revoke);
61
62EXPORT_SYMBOL(journal_init_dev);
63EXPORT_SYMBOL(journal_init_inode);
64EXPORT_SYMBOL(journal_update_format);
65EXPORT_SYMBOL(journal_check_used_features);
66EXPORT_SYMBOL(journal_check_available_features);
67EXPORT_SYMBOL(journal_set_features);
68EXPORT_SYMBOL(journal_create);
69EXPORT_SYMBOL(journal_load);
70EXPORT_SYMBOL(journal_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071EXPORT_SYMBOL(journal_abort);
72EXPORT_SYMBOL(journal_errno);
73EXPORT_SYMBOL(journal_ack_err);
74EXPORT_SYMBOL(journal_clear_err);
75EXPORT_SYMBOL(log_wait_commit);
Stefan Schmidtff5e4b52009-11-12 09:53:50 +010076EXPORT_SYMBOL(log_start_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077EXPORT_SYMBOL(journal_start_commit);
78EXPORT_SYMBOL(journal_force_commit_nested);
79EXPORT_SYMBOL(journal_wipe);
80EXPORT_SYMBOL(journal_blocks_per_page);
81EXPORT_SYMBOL(journal_invalidatepage);
82EXPORT_SYMBOL(journal_try_to_free_buffers);
83EXPORT_SYMBOL(journal_force_commit);
84
85static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
Adrian Bunk022a4a72005-09-06 15:16:41 -070086static void __journal_abort_soft (journal_t *journal, int errno);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88/*
89 * Helper function used to manage commit timeouts
90 */
91
92static void commit_timeout(unsigned long __data)
93{
94 struct task_struct * p = (struct task_struct *) __data;
95
96 wake_up_process(p);
97}
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099/*
100 * kjournald: The main thread function used to manage a logging device
101 * journal.
102 *
103 * This kernel thread is responsible for two things:
104 *
105 * 1) COMMIT: Every so often we need to commit the current state of the
106 * filesystem to disk. The journal thread is responsible for writing
107 * all of the metadata buffers to disk.
108 *
109 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
110 * of the data in that part of the log has been rewritten elsewhere on
111 * the disk. Flushing these old buffers to reclaim space in the log is
112 * known as checkpointing, and this thread is responsible for that job.
113 */
114
Adrian Bunk022a4a72005-09-06 15:16:41 -0700115static int kjournald(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Andrew Mortone3df1892006-03-25 03:06:53 -0800117 journal_t *journal = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 transaction_t *transaction;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Andrew Mortone3df1892006-03-25 03:06:53 -0800120 /*
121 * Set up an interval timer which can be used to trigger a commit wakeup
122 * after the commit interval expires
123 */
124 setup_timer(&journal->j_commit_timer, commit_timeout,
125 (unsigned long)current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127 /* Record that the journal thread is running */
128 journal->j_task = current;
129 wake_up(&journal->j_wait_done_commit);
130
131 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
132 journal->j_commit_interval / HZ);
133
134 /*
135 * And now, wait forever for commit wakeup events.
136 */
137 spin_lock(&journal->j_state_lock);
138
139loop:
140 if (journal->j_flags & JFS_UNMOUNT)
141 goto end_loop;
142
143 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
144 journal->j_commit_sequence, journal->j_commit_request);
145
146 if (journal->j_commit_sequence != journal->j_commit_request) {
147 jbd_debug(1, "OK, requests differ\n");
148 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800149 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 journal_commit_transaction(journal);
151 spin_lock(&journal->j_state_lock);
152 goto loop;
153 }
154
155 wake_up(&journal->j_wait_done_commit);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700156 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 /*
158 * The simpler the better. Flushing journal isn't a
159 * good idea, because that depends on threads that may
160 * be already stopped.
161 */
162 jbd_debug(1, "Now suspending kjournald\n");
163 spin_unlock(&journal->j_state_lock);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700164 refrigerator();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 spin_lock(&journal->j_state_lock);
166 } else {
167 /*
168 * We assume on resume that commits are already there,
169 * so we don't sleep
170 */
171 DEFINE_WAIT(wait);
172 int should_sleep = 1;
173
174 prepare_to_wait(&journal->j_wait_commit, &wait,
175 TASK_INTERRUPTIBLE);
176 if (journal->j_commit_sequence != journal->j_commit_request)
177 should_sleep = 0;
178 transaction = journal->j_running_transaction;
179 if (transaction && time_after_eq(jiffies,
180 transaction->t_expires))
181 should_sleep = 0;
Mark Fashehcbf0d272005-09-06 15:19:08 -0700182 if (journal->j_flags & JFS_UNMOUNT)
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700183 should_sleep = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 if (should_sleep) {
185 spin_unlock(&journal->j_state_lock);
186 schedule();
187 spin_lock(&journal->j_state_lock);
188 }
189 finish_wait(&journal->j_wait_commit, &wait);
190 }
191
192 jbd_debug(1, "kjournald wakes\n");
193
194 /*
195 * Were we woken up by a commit wakeup event?
196 */
197 transaction = journal->j_running_transaction;
198 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
199 journal->j_commit_request = transaction->t_tid;
200 jbd_debug(1, "woke because of timeout\n");
201 }
202 goto loop;
203
204end_loop:
205 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800206 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 journal->j_task = NULL;
208 wake_up(&journal->j_wait_done_commit);
209 jbd_debug(1, "Journal thread exiting.\n");
210 return 0;
211}
212
Pavel Emelianov97f06782007-05-08 00:30:42 -0700213static int journal_start_thread(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700215 struct task_struct *t;
216
217 t = kthread_run(kjournald, journal, "kjournald");
218 if (IS_ERR(t))
219 return PTR_ERR(t);
220
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700221 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700222 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225static void journal_kill_thread(journal_t *journal)
226{
227 spin_lock(&journal->j_state_lock);
228 journal->j_flags |= JFS_UNMOUNT;
229
230 while (journal->j_task) {
231 wake_up(&journal->j_wait_commit);
232 spin_unlock(&journal->j_state_lock);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700233 wait_event(journal->j_wait_done_commit,
234 journal->j_task == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 spin_lock(&journal->j_state_lock);
236 }
237 spin_unlock(&journal->j_state_lock);
238}
239
240/*
241 * journal_write_metadata_buffer: write a metadata buffer to the journal.
242 *
243 * Writes a metadata buffer to a given disk block. The actual IO is not
244 * performed but a new buffer_head is constructed which labels the data
245 * to be written with the correct destination disk block.
246 *
247 * Any magic-number escaping which needs to be done will cause a
248 * copy-out here. If the buffer happens to start with the
249 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
250 * magic number is only written to the log for descripter blocks. In
251 * this case, we copy the data and replace the first word with 0, and we
252 * return a result code which indicates that this buffer needs to be
253 * marked as an escaped buffer in the corresponding log descriptor
254 * block. The missing word can then be restored when the block is read
255 * during recovery.
256 *
257 * If the source buffer has already been modified by a new transaction
258 * since we took the last commit snapshot, we use the frozen copy of
259 * that data for IO. If we end up using the existing buffer_head's data
260 * for the write, then we *have* to lock the buffer to prevent anyone
261 * else from using and possibly modifying it while the IO is in
262 * progress.
263 *
264 * The function returns a pointer to the buffer_heads to be used for IO.
265 *
266 * We assume that the journal has already been locked in this function.
267 *
268 * Return value:
269 * <0: Error
270 * >=0: Finished OK
271 *
272 * On success:
273 * Bit 0 set == escape performed on the data
274 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
275 */
276
277int journal_write_metadata_buffer(transaction_t *transaction,
278 struct journal_head *jh_in,
279 struct journal_head **jh_out,
Jan Kara9c28cbc2009-08-03 19:21:00 +0200280 unsigned int blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
282 int need_copy_out = 0;
283 int done_copy_out = 0;
284 int do_escape = 0;
285 char *mapped_data;
286 struct buffer_head *new_bh;
287 struct journal_head *new_jh;
288 struct page *new_page;
289 unsigned int new_offset;
290 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghuaf1015c42009-07-15 21:42:05 +0200291 journal_t *journal = transaction->t_journal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
293 /*
294 * The buffer really shouldn't be locked: only the current committing
295 * transaction is allowed to write it, so nobody else is allowed
296 * to do any IO.
297 *
298 * akpm: except if we're journalling data, and write() output is
299 * also part of a shared mapping, and another thread has
300 * decided to launch a writepage() against this buffer.
301 */
302 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
303
304 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
dingdinghuaf1015c42009-07-15 21:42:05 +0200305 /* keep subsequent assertions sane */
306 new_bh->b_state = 0;
307 init_buffer(new_bh, NULL, NULL);
308 atomic_set(&new_bh->b_count, 1);
309 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
311 /*
312 * If a new transaction has already done a buffer copy-out, then
313 * we use that version of the data for the commit.
314 */
315 jbd_lock_bh_state(bh_in);
316repeat:
317 if (jh_in->b_frozen_data) {
318 done_copy_out = 1;
319 new_page = virt_to_page(jh_in->b_frozen_data);
320 new_offset = offset_in_page(jh_in->b_frozen_data);
321 } else {
322 new_page = jh2bh(jh_in)->b_page;
323 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
324 }
325
326 mapped_data = kmap_atomic(new_page, KM_USER0);
327 /*
328 * Check for escaping
329 */
330 if (*((__be32 *)(mapped_data + new_offset)) ==
331 cpu_to_be32(JFS_MAGIC_NUMBER)) {
332 need_copy_out = 1;
333 do_escape = 1;
334 }
335 kunmap_atomic(mapped_data, KM_USER0);
336
337 /*
338 * Do we need to do a data copy?
339 */
340 if (need_copy_out && !done_copy_out) {
341 char *tmp;
342
343 jbd_unlock_bh_state(bh_in);
Mingming Caoc089d492007-10-16 18:38:25 -0400344 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 jbd_lock_bh_state(bh_in);
346 if (jh_in->b_frozen_data) {
Mingming Caoc089d492007-10-16 18:38:25 -0400347 jbd_free(tmp, bh_in->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 goto repeat;
349 }
350
351 jh_in->b_frozen_data = tmp;
352 mapped_data = kmap_atomic(new_page, KM_USER0);
353 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
354 kunmap_atomic(mapped_data, KM_USER0);
355
356 new_page = virt_to_page(tmp);
357 new_offset = offset_in_page(tmp);
358 done_copy_out = 1;
359 }
360
361 /*
362 * Did we need to do an escaping? Now we've done all the
363 * copying, we can finally do so.
364 */
365 if (do_escape) {
366 mapped_data = kmap_atomic(new_page, KM_USER0);
367 *((unsigned int *)(mapped_data + new_offset)) = 0;
368 kunmap_atomic(mapped_data, KM_USER0);
369 }
370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 set_bh_page(new_bh, new_page, new_offset);
372 new_jh->b_transaction = NULL;
373 new_bh->b_size = jh2bh(jh_in)->b_size;
374 new_bh->b_bdev = transaction->t_journal->j_dev;
375 new_bh->b_blocknr = blocknr;
376 set_buffer_mapped(new_bh);
377 set_buffer_dirty(new_bh);
378
379 *jh_out = new_jh;
380
381 /*
382 * The to-be-written buffer needs to get moved to the io queue,
383 * and the original buffer whose contents we are shadowing or
384 * copying is moved to the transaction's shadow queue.
385 */
386 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghuaf1015c42009-07-15 21:42:05 +0200387 spin_lock(&journal->j_list_lock);
388 __journal_file_buffer(jh_in, transaction, BJ_Shadow);
389 spin_unlock(&journal->j_list_lock);
390 jbd_unlock_bh_state(bh_in);
391
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 JBUFFER_TRACE(new_jh, "file as BJ_IO");
393 journal_file_buffer(new_jh, transaction, BJ_IO);
394
395 return do_escape | (done_copy_out << 1);
396}
397
398/*
399 * Allocation code for the journal file. Manage the space left in the
400 * journal, so that we can begin checkpointing when appropriate.
401 */
402
403/*
404 * __log_space_left: Return the number of free blocks left in the journal.
405 *
406 * Called with the journal already locked.
407 *
408 * Called under j_state_lock
409 */
410
411int __log_space_left(journal_t *journal)
412{
413 int left = journal->j_free;
414
415 assert_spin_locked(&journal->j_state_lock);
416
417 /*
418 * Be pessimistic here about the number of those free blocks which
419 * might be required for log descriptor control blocks.
420 */
421
422#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
423
424 left -= MIN_LOG_RESERVED_BLOCKS;
425
426 if (left <= 0)
427 return 0;
428 left -= (left >> 3);
429 return left;
430}
431
432/*
Jan Kara8fe4cd02009-02-11 13:04:25 -0800433 * Called under j_state_lock. Returns true if a transaction commit was started.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 */
435int __log_start_commit(journal_t *journal, tid_t target)
436{
437 /*
438 * Are we already doing a recent enough commit?
439 */
440 if (!tid_geq(journal->j_commit_request, target)) {
441 /*
442 * We want a new commit: OK, mark the request and wakup the
443 * commit thread. We do _not_ do the commit ourselves.
444 */
445
446 journal->j_commit_request = target;
447 jbd_debug(1, "JBD: requesting commit %d/%d\n",
448 journal->j_commit_request,
449 journal->j_commit_sequence);
450 wake_up(&journal->j_wait_commit);
451 return 1;
452 }
453 return 0;
454}
455
456int log_start_commit(journal_t *journal, tid_t tid)
457{
458 int ret;
459
460 spin_lock(&journal->j_state_lock);
461 ret = __log_start_commit(journal, tid);
462 spin_unlock(&journal->j_state_lock);
463 return ret;
464}
465
466/*
467 * Force and wait upon a commit if the calling process is not within
468 * transaction. This is used for forcing out undo-protected data which contains
469 * bitmaps, when the fs is running out of space.
470 *
471 * We can only force the running transaction if we don't have an active handle;
472 * otherwise, we will deadlock.
473 *
474 * Returns true if a transaction was started.
475 */
476int journal_force_commit_nested(journal_t *journal)
477{
478 transaction_t *transaction = NULL;
479 tid_t tid;
480
481 spin_lock(&journal->j_state_lock);
482 if (journal->j_running_transaction && !current->journal_info) {
483 transaction = journal->j_running_transaction;
484 __log_start_commit(journal, transaction->t_tid);
485 } else if (journal->j_committing_transaction)
486 transaction = journal->j_committing_transaction;
487
488 if (!transaction) {
489 spin_unlock(&journal->j_state_lock);
490 return 0; /* Nothing to retry */
491 }
492
493 tid = transaction->t_tid;
494 spin_unlock(&journal->j_state_lock);
495 log_wait_commit(journal, tid);
496 return 1;
497}
498
499/*
500 * Start a commit of the current running transaction (if any). Returns true
Jan Kara8fe4cd02009-02-11 13:04:25 -0800501 * if a transaction is going to be committed (or is currently already
502 * committing), and fills its tid in at *ptid
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 */
504int journal_start_commit(journal_t *journal, tid_t *ptid)
505{
506 int ret = 0;
507
508 spin_lock(&journal->j_state_lock);
509 if (journal->j_running_transaction) {
510 tid_t tid = journal->j_running_transaction->t_tid;
511
Jan Kara8fe4cd02009-02-11 13:04:25 -0800512 __log_start_commit(journal, tid);
513 /* There's a running transaction and we've just made sure
514 * it's commit has been scheduled. */
515 if (ptid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 *ptid = tid;
Jan Kara8fe4cd02009-02-11 13:04:25 -0800517 ret = 1;
518 } else if (journal->j_committing_transaction) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 /*
520 * If ext3_write_super() recently started a commit, then we
521 * have to wait for completion of that transaction
522 */
Jan Kara8fe4cd02009-02-11 13:04:25 -0800523 if (ptid)
524 *ptid = journal->j_committing_transaction->t_tid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 ret = 1;
526 }
527 spin_unlock(&journal->j_state_lock);
528 return ret;
529}
530
531/*
532 * Wait for a specified commit to complete.
533 * The caller may not hold the journal lock.
534 */
535int log_wait_commit(journal_t *journal, tid_t tid)
536{
537 int err = 0;
538
539#ifdef CONFIG_JBD_DEBUG
540 spin_lock(&journal->j_state_lock);
541 if (!tid_geq(journal->j_commit_request, tid)) {
542 printk(KERN_EMERG
543 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700544 __func__, journal->j_commit_request, tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 }
546 spin_unlock(&journal->j_state_lock);
547#endif
548 spin_lock(&journal->j_state_lock);
549 while (tid_gt(tid, journal->j_commit_sequence)) {
550 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
551 tid, journal->j_commit_sequence);
552 wake_up(&journal->j_wait_commit);
553 spin_unlock(&journal->j_state_lock);
554 wait_event(journal->j_wait_done_commit,
555 !tid_gt(tid, journal->j_commit_sequence));
556 spin_lock(&journal->j_state_lock);
557 }
558 spin_unlock(&journal->j_state_lock);
559
560 if (unlikely(is_journal_aborted(journal))) {
561 printk(KERN_EMERG "journal commit I/O error\n");
562 err = -EIO;
563 }
564 return err;
565}
566
567/*
Jan Kara03f4d802010-04-15 22:16:24 +0200568 * Return 1 if a given transaction has not yet sent barrier request
569 * connected with a transaction commit. If 0 is returned, transaction
570 * may or may not have sent the barrier. Used to avoid sending barrier
571 * twice in common cases.
572 */
573int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
574{
575 int ret = 0;
576 transaction_t *commit_trans;
577
578 if (!(journal->j_flags & JFS_BARRIER))
579 return 0;
580 spin_lock(&journal->j_state_lock);
581 /* Transaction already committed? */
582 if (tid_geq(journal->j_commit_sequence, tid))
583 goto out;
584 /*
585 * Transaction is being committed and we already proceeded to
586 * writing commit record?
587 */
588 commit_trans = journal->j_committing_transaction;
589 if (commit_trans && commit_trans->t_tid == tid &&
590 commit_trans->t_state >= T_COMMIT_RECORD)
591 goto out;
592 ret = 1;
593out:
594 spin_unlock(&journal->j_state_lock);
595 return ret;
596}
Jan Kara52779702010-04-15 22:24:26 +0200597EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
Jan Kara03f4d802010-04-15 22:16:24 +0200598
599/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 * Log buffer allocation routines:
601 */
602
Jan Kara9c28cbc2009-08-03 19:21:00 +0200603int journal_next_log_block(journal_t *journal, unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200605 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607 spin_lock(&journal->j_state_lock);
608 J_ASSERT(journal->j_free > 1);
609
610 blocknr = journal->j_head;
611 journal->j_head++;
612 journal->j_free--;
613 if (journal->j_head == journal->j_last)
614 journal->j_head = journal->j_first;
615 spin_unlock(&journal->j_state_lock);
616 return journal_bmap(journal, blocknr, retp);
617}
618
619/*
620 * Conversion of logical to physical block numbers for the journal
621 *
622 * On external journals the journal blocks are identity-mapped, so
623 * this is a no-op. If needed, we can use j_blk_offset - everything is
624 * ready.
625 */
Jan Kara9c28cbc2009-08-03 19:21:00 +0200626int journal_bmap(journal_t *journal, unsigned int blocknr,
627 unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
629 int err = 0;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200630 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632 if (journal->j_inode) {
633 ret = bmap(journal->j_inode, blocknr);
634 if (ret)
635 *retp = ret;
636 else {
637 char b[BDEVNAME_SIZE];
638
639 printk(KERN_ALERT "%s: journal block not found "
Jan Kara9c28cbc2009-08-03 19:21:00 +0200640 "at offset %u on %s\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700641 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 blocknr,
643 bdevname(journal->j_dev, b));
644 err = -EIO;
645 __journal_abort_soft(journal, err);
646 }
647 } else {
648 *retp = blocknr; /* +journal->j_blk_offset */
649 }
650 return err;
651}
652
653/*
654 * We play buffer_head aliasing tricks to write data/metadata blocks to
655 * the journal without copying their contents, but for journal
656 * descriptor blocks we do need to generate bona fide buffers.
657 *
658 * After the caller of journal_get_descriptor_buffer() has finished modifying
659 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
660 * But we don't bother doing that, so there will be coherency problems with
661 * mmaps of blockdevs which hold live JBD-controlled filesystems.
662 */
663struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
664{
665 struct buffer_head *bh;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200666 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 int err;
668
669 err = journal_next_log_block(journal, &blocknr);
670
671 if (err)
672 return NULL;
673
674 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700675 if (!bh)
676 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 lock_buffer(bh);
678 memset(bh->b_data, 0, journal->j_blocksize);
679 set_buffer_uptodate(bh);
680 unlock_buffer(bh);
681 BUFFER_TRACE(bh, "return this buffer");
682 return journal_add_journal_head(bh);
683}
684
685/*
686 * Management for journal control blocks: functions to create and
687 * destroy journal_t structures, and to initialise and read existing
688 * journal blocks from disk. */
689
690/* First: create and setup a journal_t object in memory. We initialise
691 * very few fields yet: that has to wait until we have created the
692 * journal structures from from scratch, or loaded them from disk. */
693
694static journal_t * journal_init_common (void)
695{
696 journal_t *journal;
697 int err;
698
Mingming Cao8c3478a2007-10-18 23:39:20 -0700699 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (!journal)
701 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
703 init_waitqueue_head(&journal->j_wait_transaction_locked);
704 init_waitqueue_head(&journal->j_wait_logspace);
705 init_waitqueue_head(&journal->j_wait_done_commit);
706 init_waitqueue_head(&journal->j_wait_checkpoint);
707 init_waitqueue_head(&journal->j_wait_commit);
708 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800709 mutex_init(&journal->j_barrier);
710 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 spin_lock_init(&journal->j_revoke_lock);
712 spin_lock_init(&journal->j_list_lock);
713 spin_lock_init(&journal->j_state_lock);
714
715 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
716
717 /* The journal is marked for error until we succeed with recovery! */
718 journal->j_flags = JFS_ABORT;
719
720 /* Set up a default-sized revoke table for the new mount. */
721 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
722 if (err) {
723 kfree(journal);
724 goto fail;
725 }
726 return journal;
727fail:
728 return NULL;
729}
730
731/* journal_init_dev and journal_init_inode:
732 *
733 * Create a journal structure assigned some fixed set of disk blocks to
734 * the journal. We don't actually touch those disk blocks yet, but we
735 * need to set up all of the mapping information to tell the journaling
736 * system where the journal blocks are.
737 *
738 */
739
740/**
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700741 * journal_t * journal_init_dev() - creates and initialises a journal structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 * @bdev: Block device on which to create the journal
743 * @fs_dev: Device which hold journalled filesystem for this journal.
744 * @start: Block nr Start of journal.
Eric Sandeen37ed3222006-09-27 01:49:31 -0700745 * @len: Length of the journal in blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 * @blocksize: blocksize of journalling device
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700747 *
748 * Returns: a newly created journal_t *
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700749 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 * journal_init_dev creates a journal which maps a fixed contiguous
751 * range of blocks on an arbitrary block device.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700752 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 */
754journal_t * journal_init_dev(struct block_device *bdev,
755 struct block_device *fs_dev,
756 int start, int len, int blocksize)
757{
758 journal_t *journal = journal_init_common();
759 struct buffer_head *bh;
760 int n;
761
762 if (!journal)
763 return NULL;
764
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 /* journal descriptor can store up to n blocks -bzzz */
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700766 journal->j_blocksize = blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 n = journal->j_blocksize / sizeof(journal_block_tag_t);
768 journal->j_wbufsize = n;
769 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
770 if (!journal->j_wbuf) {
771 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700772 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700773 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700775 journal->j_dev = bdev;
776 journal->j_fs_dev = fs_dev;
777 journal->j_blk_offset = start;
778 journal->j_maxlen = len;
779
780 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700781 if (!bh) {
782 printk(KERN_ERR
783 "%s: Cannot get buffer for journal superblock\n",
784 __func__);
785 goto out_err;
786 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700787 journal->j_sb_buffer = bh;
788 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Karaecca9af2009-04-02 16:57:13 -0700789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700791out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800792 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700793 kfree(journal);
794 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795}
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700796
797/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
799 * @inode: An inode to create the journal in
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700800 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 * journal_init_inode creates a journal which maps an on-disk inode as
802 * the journal. The inode must exist already, must support bmap() and
803 * must have all data blocks preallocated.
804 */
805journal_t * journal_init_inode (struct inode *inode)
806{
807 struct buffer_head *bh;
808 journal_t *journal = journal_init_common();
809 int err;
810 int n;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200811 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
813 if (!journal)
814 return NULL;
815
816 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
817 journal->j_inode = inode;
818 jbd_debug(1,
819 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700820 journal, inode->i_sb->s_id, inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 (long long) inode->i_size,
822 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
823
824 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
825 journal->j_blocksize = inode->i_sb->s_blocksize;
826
827 /* journal descriptor can store up to n blocks -bzzz */
828 n = journal->j_blocksize / sizeof(journal_block_tag_t);
829 journal->j_wbufsize = n;
830 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
831 if (!journal->j_wbuf) {
832 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700833 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700834 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 }
836
837 err = journal_bmap(journal, 0, &blocknr);
838 /* If that failed, give up */
839 if (err) {
840 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700841 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700842 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 }
844
845 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700846 if (!bh) {
847 printk(KERN_ERR
848 "%s: Cannot get buffer for journal superblock\n",
849 __func__);
850 goto out_err;
851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 journal->j_sb_buffer = bh;
853 journal->j_superblock = (journal_superblock_t *)bh->b_data;
854
855 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700856out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800857 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700858 kfree(journal);
859 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
861
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700862/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 * If the journal init or create aborts, we need to mark the journal
864 * superblock as being NULL to prevent the journal destroy from writing
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700865 * back a bogus superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 */
867static void journal_fail_superblock (journal_t *journal)
868{
869 struct buffer_head *bh = journal->j_sb_buffer;
870 brelse(bh);
871 journal->j_sb_buffer = NULL;
872}
873
874/*
875 * Given a journal_t structure, initialise the various fields for
876 * startup of a new journaling session. We use this both when creating
877 * a journal, and after recovering an old journal to reset it for
878 * subsequent use.
879 */
880
881static int journal_reset(journal_t *journal)
882{
883 journal_superblock_t *sb = journal->j_superblock;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200884 unsigned int first, last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
886 first = be32_to_cpu(sb->s_first);
887 last = be32_to_cpu(sb->s_maxlen);
Jan Kara7447a662009-07-15 20:36:08 +0200888 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200889 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
Jan Kara7447a662009-07-15 20:36:08 +0200890 first, last);
891 journal_fail_superblock(journal);
892 return -EINVAL;
893 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 journal->j_first = first;
896 journal->j_last = last;
897
898 journal->j_head = first;
899 journal->j_tail = first;
900 journal->j_free = last - first;
901
902 journal->j_tail_sequence = journal->j_transaction_sequence;
903 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
904 journal->j_commit_request = journal->j_commit_sequence;
905
906 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
907
908 /* Add the dynamic fields and write it to disk. */
909 journal_update_superblock(journal, 1);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700910 return journal_start_thread(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911}
912
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700913/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 * int journal_create() - Initialise the new journal file
915 * @journal: Journal to create. This structure must have been initialised
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700916 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 * Given a journal_t structure which tells us which disk blocks we can
918 * use, create a new journal superblock and initialise all of the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700919 * journal fields from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 **/
921int journal_create(journal_t *journal)
922{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200923 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 struct buffer_head *bh;
925 journal_superblock_t *sb;
926 int i, err;
927
928 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
929 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
930 journal->j_maxlen);
931 journal_fail_superblock(journal);
932 return -EINVAL;
933 }
934
935 if (journal->j_inode == NULL) {
936 /*
937 * We don't know what block to start at!
938 */
939 printk(KERN_EMERG
940 "%s: creation of journal on external device!\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700941 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 BUG();
943 }
944
945 /* Zero out the entire journal on disk. We cannot afford to
946 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
947 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
948 for (i = 0; i < journal->j_maxlen; i++) {
949 err = journal_bmap(journal, i, &blocknr);
950 if (err)
951 return err;
952 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
953 lock_buffer(bh);
954 memset (bh->b_data, 0, journal->j_blocksize);
955 BUFFER_TRACE(bh, "marking dirty");
956 mark_buffer_dirty(bh);
957 BUFFER_TRACE(bh, "marking uptodate");
958 set_buffer_uptodate(bh);
959 unlock_buffer(bh);
960 __brelse(bh);
961 }
962
963 sync_blockdev(journal->j_dev);
964 jbd_debug(1, "JBD: journal cleared.\n");
965
966 /* OK, fill in the initial static fields in the new superblock */
967 sb = journal->j_superblock;
968
969 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
970 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
971
972 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
973 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
974 sb->s_first = cpu_to_be32(1);
975
976 journal->j_transaction_sequence = 1;
977
978 journal->j_flags &= ~JFS_ABORT;
979 journal->j_format_version = 2;
980
981 return journal_reset(journal);
982}
983
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700984/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 * void journal_update_superblock() - Update journal sb on disk.
986 * @journal: The journal to update.
987 * @wait: Set to '0' if you don't want to wait for IO completion.
988 *
989 * Update a journal's dynamic superblock fields and write it to disk,
990 * optionally waiting for the IO to complete.
991 */
992void journal_update_superblock(journal_t *journal, int wait)
993{
994 journal_superblock_t *sb = journal->j_superblock;
995 struct buffer_head *bh = journal->j_sb_buffer;
996
997 /*
998 * As a special case, if the on-disk copy is already marked as needing
999 * no recovery (s_start == 0) and there are no outstanding transactions
1000 * in the filesystem, then we can safely defer the superblock update
1001 * until the next commit by setting JFS_FLUSHED. This avoids
1002 * attempting a write to a potential-readonly device.
1003 */
1004 if (sb->s_start == 0 && journal->j_tail_sequence ==
1005 journal->j_transaction_sequence) {
1006 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
Jan Kara9c28cbc2009-08-03 19:21:00 +02001007 "(start %u, seq %d, errno %d)\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001008 journal->j_tail, journal->j_tail_sequence,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 journal->j_errno);
1010 goto out;
1011 }
1012
1013 spin_lock(&journal->j_state_lock);
Jan Kara9c28cbc2009-08-03 19:21:00 +02001014 jbd_debug(1,"JBD: updating superblock (start %u, seq %d, errno %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1016
1017 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1018 sb->s_start = cpu_to_be32(journal->j_tail);
1019 sb->s_errno = cpu_to_be32(journal->j_errno);
1020 spin_unlock(&journal->j_state_lock);
1021
1022 BUFFER_TRACE(bh, "marking dirty");
1023 mark_buffer_dirty(bh);
1024 if (wait)
1025 sync_dirty_buffer(bh);
1026 else
Jan Kara26707692005-09-06 15:19:12 -07001027 ll_rw_block(SWRITE, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029out:
1030 /* If we have just flushed the log (by marking s_start==0), then
1031 * any future commit will have to be careful to update the
1032 * superblock again to re-record the true start of the log. */
1033
1034 spin_lock(&journal->j_state_lock);
1035 if (sb->s_start)
1036 journal->j_flags &= ~JFS_FLUSHED;
1037 else
1038 journal->j_flags |= JFS_FLUSHED;
1039 spin_unlock(&journal->j_state_lock);
1040}
1041
1042/*
1043 * Read the superblock for a given journal, performing initial
1044 * validation of the format.
1045 */
1046
1047static int journal_get_superblock(journal_t *journal)
1048{
1049 struct buffer_head *bh;
1050 journal_superblock_t *sb;
1051 int err = -EIO;
1052
1053 bh = journal->j_sb_buffer;
1054
1055 J_ASSERT(bh != NULL);
1056 if (!buffer_uptodate(bh)) {
1057 ll_rw_block(READ, 1, &bh);
1058 wait_on_buffer(bh);
1059 if (!buffer_uptodate(bh)) {
1060 printk (KERN_ERR
1061 "JBD: IO error reading journal superblock\n");
1062 goto out;
1063 }
1064 }
1065
1066 sb = journal->j_superblock;
1067
1068 err = -EINVAL;
1069
1070 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1071 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1072 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1073 goto out;
1074 }
1075
1076 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1077 case JFS_SUPERBLOCK_V1:
1078 journal->j_format_version = 1;
1079 break;
1080 case JFS_SUPERBLOCK_V2:
1081 journal->j_format_version = 2;
1082 break;
1083 default:
1084 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1085 goto out;
1086 }
1087
1088 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1089 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1090 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1091 printk (KERN_WARNING "JBD: journal file too short\n");
1092 goto out;
1093 }
1094
1095 return 0;
1096
1097out:
1098 journal_fail_superblock(journal);
1099 return err;
1100}
1101
1102/*
1103 * Load the on-disk journal superblock and read the key fields into the
1104 * journal_t.
1105 */
1106
1107static int load_superblock(journal_t *journal)
1108{
1109 int err;
1110 journal_superblock_t *sb;
1111
1112 err = journal_get_superblock(journal);
1113 if (err)
1114 return err;
1115
1116 sb = journal->j_superblock;
1117
1118 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1119 journal->j_tail = be32_to_cpu(sb->s_start);
1120 journal->j_first = be32_to_cpu(sb->s_first);
1121 journal->j_last = be32_to_cpu(sb->s_maxlen);
1122 journal->j_errno = be32_to_cpu(sb->s_errno);
1123
1124 return 0;
1125}
1126
1127
1128/**
1129 * int journal_load() - Read journal from disk.
1130 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001131 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 * Given a journal_t structure which tells us which disk blocks contain
1133 * a journal, read the journal from disk to initialise the in-memory
1134 * structures.
1135 */
1136int journal_load(journal_t *journal)
1137{
1138 int err;
Badari Pulavartyea817392006-08-27 01:23:52 -07001139 journal_superblock_t *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
1141 err = load_superblock(journal);
1142 if (err)
1143 return err;
1144
Badari Pulavartyea817392006-08-27 01:23:52 -07001145 sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 /* If this is a V2 superblock, then we have to check the
1147 * features flags on it. */
1148
1149 if (journal->j_format_version >= 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if ((sb->s_feature_ro_compat &
1151 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1152 (sb->s_feature_incompat &
1153 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1154 printk (KERN_WARNING
1155 "JBD: Unrecognised features on journal\n");
1156 return -EINVAL;
1157 }
1158 }
1159
1160 /* Let the recovery code check whether it needs to recover any
1161 * data from the journal. */
1162 if (journal_recover(journal))
1163 goto recovery_error;
1164
1165 /* OK, we've finished with the dynamic journal bits:
1166 * reinitialise the dynamic contents of the superblock in memory
1167 * and reset them on disk. */
1168 if (journal_reset(journal))
1169 goto recovery_error;
1170
1171 journal->j_flags &= ~JFS_ABORT;
1172 journal->j_flags |= JFS_LOADED;
1173 return 0;
1174
1175recovery_error:
1176 printk (KERN_WARNING "JBD: recovery failed\n");
1177 return -EIO;
1178}
1179
1180/**
1181 * void journal_destroy() - Release a journal_t structure.
1182 * @journal: Journal to act on.
1183 *
1184 * Release a journal_t structure once it is no longer in use by the
1185 * journaled object.
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001186 * Return <0 if we couldn't clean up the journal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 */
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001188int journal_destroy(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001190 int err = 0;
1191
Jan Kara03f4d802010-04-15 22:16:24 +02001192
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 /* Wait for the commit thread to wake up and die. */
1194 journal_kill_thread(journal);
1195
1196 /* Force a final log commit */
1197 if (journal->j_running_transaction)
1198 journal_commit_transaction(journal);
1199
1200 /* Force any old transactions to disk */
1201
1202 /* Totally anal locking here... */
1203 spin_lock(&journal->j_list_lock);
1204 while (journal->j_checkpoint_transactions != NULL) {
1205 spin_unlock(&journal->j_list_lock);
1206 log_do_checkpoint(journal);
1207 spin_lock(&journal->j_list_lock);
1208 }
1209
1210 J_ASSERT(journal->j_running_transaction == NULL);
1211 J_ASSERT(journal->j_committing_transaction == NULL);
1212 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1213 spin_unlock(&journal->j_list_lock);
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (journal->j_sb_buffer) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001216 if (!is_journal_aborted(journal)) {
1217 /* We can now mark the journal as empty. */
1218 journal->j_tail = 0;
1219 journal->j_tail_sequence =
1220 ++journal->j_transaction_sequence;
1221 journal_update_superblock(journal, 1);
1222 } else {
1223 err = -EIO;
1224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 brelse(journal->j_sb_buffer);
1226 }
1227
1228 if (journal->j_inode)
1229 iput(journal->j_inode);
1230 if (journal->j_revoke)
1231 journal_destroy_revoke(journal);
1232 kfree(journal->j_wbuf);
1233 kfree(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001234
1235 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236}
1237
1238
1239/**
1240 *int journal_check_used_features () - Check if features specified are used.
1241 * @journal: Journal to check.
1242 * @compat: bitmask of compatible features
1243 * @ro: bitmask of features that force read-only mount
1244 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001245 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 * Check whether the journal uses all of a given set of
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001247 * features. Return true (non-zero) if it does.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 **/
1249
1250int journal_check_used_features (journal_t *journal, unsigned long compat,
1251 unsigned long ro, unsigned long incompat)
1252{
1253 journal_superblock_t *sb;
1254
1255 if (!compat && !ro && !incompat)
1256 return 1;
1257 if (journal->j_format_version == 1)
1258 return 0;
1259
1260 sb = journal->j_superblock;
1261
1262 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1263 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1264 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1265 return 1;
1266
1267 return 0;
1268}
1269
1270/**
1271 * int journal_check_available_features() - Check feature set in journalling layer
1272 * @journal: Journal to check.
1273 * @compat: bitmask of compatible features
1274 * @ro: bitmask of features that force read-only mount
1275 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001276 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 * Check whether the journaling code supports the use of
1278 * all of a given set of features on this journal. Return true
1279 * (non-zero) if it can. */
1280
1281int journal_check_available_features (journal_t *journal, unsigned long compat,
1282 unsigned long ro, unsigned long incompat)
1283{
1284 journal_superblock_t *sb;
1285
1286 if (!compat && !ro && !incompat)
1287 return 1;
1288
1289 sb = journal->j_superblock;
1290
1291 /* We can support any known requested features iff the
1292 * superblock is in version 2. Otherwise we fail to support any
1293 * extended sb features. */
1294
1295 if (journal->j_format_version != 2)
1296 return 0;
1297
1298 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1299 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1300 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1301 return 1;
1302
1303 return 0;
1304}
1305
1306/**
1307 * int journal_set_features () - Mark a given journal feature in the superblock
1308 * @journal: Journal to act on.
1309 * @compat: bitmask of compatible features
1310 * @ro: bitmask of features that force read-only mount
1311 * @incompat: bitmask of incompatible features
1312 *
1313 * Mark a given journal feature as present on the
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001314 * superblock. Returns true if the requested features could be set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 *
1316 */
1317
1318int journal_set_features (journal_t *journal, unsigned long compat,
1319 unsigned long ro, unsigned long incompat)
1320{
1321 journal_superblock_t *sb;
1322
1323 if (journal_check_used_features(journal, compat, ro, incompat))
1324 return 1;
1325
1326 if (!journal_check_available_features(journal, compat, ro, incompat))
1327 return 0;
1328
1329 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1330 compat, ro, incompat);
1331
1332 sb = journal->j_superblock;
1333
1334 sb->s_feature_compat |= cpu_to_be32(compat);
1335 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1336 sb->s_feature_incompat |= cpu_to_be32(incompat);
1337
1338 return 1;
1339}
1340
1341
1342/**
1343 * int journal_update_format () - Update on-disk journal structure.
1344 * @journal: Journal to act on.
1345 *
1346 * Given an initialised but unloaded journal struct, poke about in the
1347 * on-disk structure to update it to the most recent supported version.
1348 */
1349int journal_update_format (journal_t *journal)
1350{
1351 journal_superblock_t *sb;
1352 int err;
1353
1354 err = journal_get_superblock(journal);
1355 if (err)
1356 return err;
1357
1358 sb = journal->j_superblock;
1359
1360 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1361 case JFS_SUPERBLOCK_V2:
1362 return 0;
1363 case JFS_SUPERBLOCK_V1:
1364 return journal_convert_superblock_v1(journal, sb);
1365 default:
1366 break;
1367 }
1368 return -EINVAL;
1369}
1370
1371static int journal_convert_superblock_v1(journal_t *journal,
1372 journal_superblock_t *sb)
1373{
1374 int offset, blocksize;
1375 struct buffer_head *bh;
1376
1377 printk(KERN_WARNING
1378 "JBD: Converting superblock from version 1 to 2.\n");
1379
1380 /* Pre-initialise new fields to zero */
1381 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1382 blocksize = be32_to_cpu(sb->s_blocksize);
1383 memset(&sb->s_feature_compat, 0, blocksize-offset);
1384
1385 sb->s_nr_users = cpu_to_be32(1);
1386 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1387 journal->j_format_version = 2;
1388
1389 bh = journal->j_sb_buffer;
1390 BUFFER_TRACE(bh, "marking dirty");
1391 mark_buffer_dirty(bh);
1392 sync_dirty_buffer(bh);
1393 return 0;
1394}
1395
1396
1397/**
1398 * int journal_flush () - Flush journal
1399 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001400 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 * Flush all data for a given journal to disk and empty the journal.
1402 * Filesystems can use this when remounting readonly to ensure that
1403 * recovery does not need to happen on remount.
1404 */
1405
1406int journal_flush(journal_t *journal)
1407{
1408 int err = 0;
1409 transaction_t *transaction = NULL;
Jan Kara9c28cbc2009-08-03 19:21:00 +02001410 unsigned int old_tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
1412 spin_lock(&journal->j_state_lock);
1413
1414 /* Force everything buffered to the log... */
1415 if (journal->j_running_transaction) {
1416 transaction = journal->j_running_transaction;
1417 __log_start_commit(journal, transaction->t_tid);
1418 } else if (journal->j_committing_transaction)
1419 transaction = journal->j_committing_transaction;
1420
1421 /* Wait for the log commit to complete... */
1422 if (transaction) {
1423 tid_t tid = transaction->t_tid;
1424
1425 spin_unlock(&journal->j_state_lock);
1426 log_wait_commit(journal, tid);
1427 } else {
1428 spin_unlock(&journal->j_state_lock);
1429 }
1430
1431 /* ...and flush everything in the log out to disk. */
1432 spin_lock(&journal->j_list_lock);
1433 while (!err && journal->j_checkpoint_transactions != NULL) {
1434 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001435 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 err = log_do_checkpoint(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001437 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 spin_lock(&journal->j_list_lock);
1439 }
1440 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001441
1442 if (is_journal_aborted(journal))
1443 return -EIO;
1444
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 cleanup_journal_tail(journal);
1446
1447 /* Finally, mark the journal as really needing no recovery.
1448 * This sets s_start==0 in the underlying superblock, which is
1449 * the magic code for a fully-recovered superblock. Any future
1450 * commits of data to the journal will restore the current
1451 * s_start value. */
1452 spin_lock(&journal->j_state_lock);
1453 old_tail = journal->j_tail;
1454 journal->j_tail = 0;
1455 spin_unlock(&journal->j_state_lock);
1456 journal_update_superblock(journal, 1);
1457 spin_lock(&journal->j_state_lock);
1458 journal->j_tail = old_tail;
1459
1460 J_ASSERT(!journal->j_running_transaction);
1461 J_ASSERT(!journal->j_committing_transaction);
1462 J_ASSERT(!journal->j_checkpoint_transactions);
1463 J_ASSERT(journal->j_head == journal->j_tail);
1464 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1465 spin_unlock(&journal->j_state_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001466 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
1468
1469/**
1470 * int journal_wipe() - Wipe journal contents
1471 * @journal: Journal to act on.
1472 * @write: flag (see below)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001473 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 * Wipe out all of the contents of a journal, safely. This will produce
1475 * a warning if the journal contains any valid recovery information.
1476 * Must be called between journal_init_*() and journal_load().
1477 *
1478 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1479 * we merely suppress recovery.
1480 */
1481
1482int journal_wipe(journal_t *journal, int write)
1483{
1484 journal_superblock_t *sb;
1485 int err = 0;
1486
1487 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1488
1489 err = load_superblock(journal);
1490 if (err)
1491 return err;
1492
1493 sb = journal->j_superblock;
1494
1495 if (!journal->j_tail)
1496 goto no_recovery;
1497
1498 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1499 write ? "Clearing" : "Ignoring");
1500
1501 err = journal_skip_recovery(journal);
1502 if (write)
1503 journal_update_superblock(journal, 1);
1504
1505 no_recovery:
1506 return err;
1507}
1508
1509/*
1510 * journal_dev_name: format a character string to describe on what
1511 * device this journal is present.
1512 */
1513
Adrian Bunk022a4a72005-09-06 15:16:41 -07001514static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515{
1516 struct block_device *bdev;
1517
1518 if (journal->j_inode)
1519 bdev = journal->j_inode->i_sb->s_bdev;
1520 else
1521 bdev = journal->j_dev;
1522
1523 return bdevname(bdev, buffer);
1524}
1525
1526/*
1527 * Journal abort has very specific semantics, which we describe
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001528 * for journal abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 *
1530 * Two internal function, which provide abort to te jbd layer
1531 * itself are here.
1532 */
1533
1534/*
1535 * Quick version for internal journal use (doesn't lock the journal).
1536 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1537 * and don't attempt to make any other journal updates.
1538 */
Adrian Bunk53308382008-02-06 01:40:12 -08001539static void __journal_abort_hard(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
1541 transaction_t *transaction;
1542 char b[BDEVNAME_SIZE];
1543
1544 if (journal->j_flags & JFS_ABORT)
1545 return;
1546
1547 printk(KERN_ERR "Aborting journal on device %s.\n",
1548 journal_dev_name(journal, b));
1549
1550 spin_lock(&journal->j_state_lock);
1551 journal->j_flags |= JFS_ABORT;
1552 transaction = journal->j_running_transaction;
1553 if (transaction)
1554 __log_start_commit(journal, transaction->t_tid);
1555 spin_unlock(&journal->j_state_lock);
1556}
1557
1558/* Soft abort: record the abort error status in the journal superblock,
1559 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001560static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
1562 if (journal->j_flags & JFS_ABORT)
1563 return;
1564
1565 if (!journal->j_errno)
1566 journal->j_errno = errno;
1567
1568 __journal_abort_hard(journal);
1569
1570 if (errno)
1571 journal_update_superblock(journal, 1);
1572}
1573
1574/**
1575 * void journal_abort () - Shutdown the journal immediately.
1576 * @journal: the journal to shutdown.
1577 * @errno: an error number to record in the journal indicating
1578 * the reason for the shutdown.
1579 *
1580 * Perform a complete, immediate shutdown of the ENTIRE
1581 * journal (not of a single transaction). This operation cannot be
1582 * undone without closing and reopening the journal.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001583 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 * The journal_abort function is intended to support higher level error
1585 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1586 * mode.
1587 *
1588 * Journal abort has very specific semantics. Any existing dirty,
1589 * unjournaled buffers in the main filesystem will still be written to
1590 * disk by bdflush, but the journaling mechanism will be suspended
1591 * immediately and no further transaction commits will be honoured.
1592 *
1593 * Any dirty, journaled buffers will be written back to disk without
1594 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1595 * filesystem, but we _do_ attempt to leave as much data as possible
1596 * behind for fsck to use for cleanup.
1597 *
1598 * Any attempt to get a new transaction handle on a journal which is in
1599 * ABORT state will just result in an -EROFS error return. A
1600 * journal_stop on an existing handle will return -EIO if we have
1601 * entered abort state during the update.
1602 *
1603 * Recursive transactions are not disturbed by journal abort until the
1604 * final journal_stop, which will receive the -EIO error.
1605 *
1606 * Finally, the journal_abort call allows the caller to supply an errno
1607 * which will be recorded (if possible) in the journal superblock. This
1608 * allows a client to record failure conditions in the middle of a
1609 * transaction without having to complete the transaction to record the
1610 * failure to disk. ext3_error, for example, now uses this
1611 * functionality.
1612 *
1613 * Errors which originate from within the journaling layer will NOT
1614 * supply an errno; a null errno implies that absolutely no further
1615 * writes are done to the journal (unless there are any already in
1616 * progress).
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001617 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 */
1619
1620void journal_abort(journal_t *journal, int errno)
1621{
1622 __journal_abort_soft(journal, errno);
1623}
1624
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001625/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 * int journal_errno () - returns the journal's error state.
1627 * @journal: journal to examine.
1628 *
1629 * This is the errno numbet set with journal_abort(), the last
1630 * time the journal was mounted - if the journal was stopped
1631 * without calling abort this will be 0.
1632 *
1633 * If the journal has been aborted on this mount time -EROFS will
1634 * be returned.
1635 */
1636int journal_errno(journal_t *journal)
1637{
1638 int err;
1639
1640 spin_lock(&journal->j_state_lock);
1641 if (journal->j_flags & JFS_ABORT)
1642 err = -EROFS;
1643 else
1644 err = journal->j_errno;
1645 spin_unlock(&journal->j_state_lock);
1646 return err;
1647}
1648
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001649/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 * int journal_clear_err () - clears the journal's error state
1651 * @journal: journal to act on.
1652 *
1653 * An error must be cleared or Acked to take a FS out of readonly
1654 * mode.
1655 */
1656int journal_clear_err(journal_t *journal)
1657{
1658 int err = 0;
1659
1660 spin_lock(&journal->j_state_lock);
1661 if (journal->j_flags & JFS_ABORT)
1662 err = -EROFS;
1663 else
1664 journal->j_errno = 0;
1665 spin_unlock(&journal->j_state_lock);
1666 return err;
1667}
1668
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001669/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 * void journal_ack_err() - Ack journal err.
1671 * @journal: journal to act on.
1672 *
1673 * An error must be cleared or Acked to take a FS out of readonly
1674 * mode.
1675 */
1676void journal_ack_err(journal_t *journal)
1677{
1678 spin_lock(&journal->j_state_lock);
1679 if (journal->j_errno)
1680 journal->j_flags |= JFS_ACK_ERR;
1681 spin_unlock(&journal->j_state_lock);
1682}
1683
1684int journal_blocks_per_page(struct inode *inode)
1685{
1686 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1687}
1688
1689/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 * Journal_head storage management
1691 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001692static struct kmem_cache *journal_head_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693#ifdef CONFIG_JBD_DEBUG
1694static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1695#endif
1696
1697static int journal_init_journal_head_cache(void)
1698{
1699 int retval;
1700
Al Viro1076d172008-03-29 03:07:18 +00001701 J_ASSERT(journal_head_cache == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 journal_head_cache = kmem_cache_create("journal_head",
1703 sizeof(struct journal_head),
1704 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001705 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001706 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00001708 if (!journal_head_cache) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 retval = -ENOMEM;
1710 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1711 }
1712 return retval;
1713}
1714
1715static void journal_destroy_journal_head_cache(void)
1716{
Duane Griffin3850f7a2008-07-25 01:46:19 -07001717 if (journal_head_cache) {
1718 kmem_cache_destroy(journal_head_cache);
1719 journal_head_cache = NULL;
1720 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721}
1722
1723/*
1724 * journal_head splicing and dicing
1725 */
1726static struct journal_head *journal_alloc_journal_head(void)
1727{
1728 struct journal_head *ret;
1729 static unsigned long last_warning;
1730
1731#ifdef CONFIG_JBD_DEBUG
1732 atomic_inc(&nr_journal_heads);
1733#endif
1734 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001735 if (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 jbd_debug(1, "out of memory for journal_head\n");
1737 if (time_after(jiffies, last_warning + 5*HZ)) {
1738 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001739 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 last_warning = jiffies;
1741 }
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001742 while (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 yield();
1744 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1745 }
1746 }
1747 return ret;
1748}
1749
1750static void journal_free_journal_head(struct journal_head *jh)
1751{
1752#ifdef CONFIG_JBD_DEBUG
1753 atomic_dec(&nr_journal_heads);
Randy Dunlapc9cf5522006-06-27 02:53:52 -07001754 memset(jh, JBD_POISON_FREE, sizeof(*jh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755#endif
1756 kmem_cache_free(journal_head_cache, jh);
1757}
1758
1759/*
1760 * A journal_head is attached to a buffer_head whenever JBD has an
1761 * interest in the buffer.
1762 *
1763 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1764 * is set. This bit is tested in core kernel code where we need to take
1765 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1766 * there.
1767 *
1768 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1769 *
1770 * When a buffer has its BH_JBD bit set it is immune from being released by
1771 * core kernel code, mainly via ->b_count.
1772 *
1773 * A journal_head may be detached from its buffer_head when the journal_head's
1774 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1775 * Various places in JBD call journal_remove_journal_head() to indicate that the
1776 * journal_head can be dropped if needed.
1777 *
1778 * Various places in the kernel want to attach a journal_head to a buffer_head
1779 * _before_ attaching the journal_head to a transaction. To protect the
1780 * journal_head in this situation, journal_add_journal_head elevates the
1781 * journal_head's b_jcount refcount by one. The caller must call
1782 * journal_put_journal_head() to undo this.
1783 *
1784 * So the typical usage would be:
1785 *
1786 * (Attach a journal_head if needed. Increments b_jcount)
1787 * struct journal_head *jh = journal_add_journal_head(bh);
1788 * ...
1789 * jh->b_transaction = xxx;
1790 * journal_put_journal_head(jh);
1791 *
1792 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1793 * because it has a non-zero b_transaction.
1794 */
1795
1796/*
1797 * Give a buffer_head a journal_head.
1798 *
1799 * Doesn't need the journal lock.
1800 * May sleep.
1801 */
1802struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1803{
1804 struct journal_head *jh;
1805 struct journal_head *new_jh = NULL;
1806
1807repeat:
1808 if (!buffer_jbd(bh)) {
1809 new_jh = journal_alloc_journal_head();
1810 memset(new_jh, 0, sizeof(*new_jh));
1811 }
1812
1813 jbd_lock_bh_journal_head(bh);
1814 if (buffer_jbd(bh)) {
1815 jh = bh2jh(bh);
1816 } else {
1817 J_ASSERT_BH(bh,
1818 (atomic_read(&bh->b_count) > 0) ||
1819 (bh->b_page && bh->b_page->mapping));
1820
1821 if (!new_jh) {
1822 jbd_unlock_bh_journal_head(bh);
1823 goto repeat;
1824 }
1825
1826 jh = new_jh;
1827 new_jh = NULL; /* We consumed it */
1828 set_buffer_jbd(bh);
1829 bh->b_private = jh;
1830 jh->b_bh = bh;
1831 get_bh(bh);
1832 BUFFER_TRACE(bh, "added journal_head");
1833 }
1834 jh->b_jcount++;
1835 jbd_unlock_bh_journal_head(bh);
1836 if (new_jh)
1837 journal_free_journal_head(new_jh);
1838 return bh->b_private;
1839}
1840
1841/*
1842 * Grab a ref against this buffer_head's journal_head. If it ended up not
1843 * having a journal_head, return NULL
1844 */
1845struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1846{
1847 struct journal_head *jh = NULL;
1848
1849 jbd_lock_bh_journal_head(bh);
1850 if (buffer_jbd(bh)) {
1851 jh = bh2jh(bh);
1852 jh->b_jcount++;
1853 }
1854 jbd_unlock_bh_journal_head(bh);
1855 return jh;
1856}
1857
1858static void __journal_remove_journal_head(struct buffer_head *bh)
1859{
1860 struct journal_head *jh = bh2jh(bh);
1861
1862 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1863
1864 get_bh(bh);
1865 if (jh->b_jcount == 0) {
1866 if (jh->b_transaction == NULL &&
1867 jh->b_next_transaction == NULL &&
1868 jh->b_cp_transaction == NULL) {
1869 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1870 J_ASSERT_BH(bh, buffer_jbd(bh));
1871 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1872 BUFFER_TRACE(bh, "remove journal_head");
1873 if (jh->b_frozen_data) {
1874 printk(KERN_WARNING "%s: freeing "
1875 "b_frozen_data\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001876 __func__);
Mingming Caoc089d492007-10-16 18:38:25 -04001877 jbd_free(jh->b_frozen_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 }
1879 if (jh->b_committed_data) {
1880 printk(KERN_WARNING "%s: freeing "
1881 "b_committed_data\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001882 __func__);
Mingming Caoc089d492007-10-16 18:38:25 -04001883 jbd_free(jh->b_committed_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 }
1885 bh->b_private = NULL;
1886 jh->b_bh = NULL; /* debug, really */
1887 clear_buffer_jbd(bh);
1888 __brelse(bh);
1889 journal_free_journal_head(jh);
1890 } else {
1891 BUFFER_TRACE(bh, "journal_head was locked");
1892 }
1893 }
1894}
1895
1896/*
1897 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1898 * and has a zero b_jcount then remove and release its journal_head. If we did
1899 * see that the buffer is not used by any transaction we also "logically"
1900 * decrement ->b_count.
1901 *
1902 * We in fact take an additional increment on ->b_count as a convenience,
1903 * because the caller usually wants to do additional things with the bh
1904 * after calling here.
1905 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1906 * time. Once the caller has run __brelse(), the buffer is eligible for
1907 * reaping by try_to_free_buffers().
1908 */
1909void journal_remove_journal_head(struct buffer_head *bh)
1910{
1911 jbd_lock_bh_journal_head(bh);
1912 __journal_remove_journal_head(bh);
1913 jbd_unlock_bh_journal_head(bh);
1914}
1915
1916/*
1917 * Drop a reference on the passed journal_head. If it fell to zero then try to
1918 * release the journal_head from the buffer_head.
1919 */
1920void journal_put_journal_head(struct journal_head *jh)
1921{
1922 struct buffer_head *bh = jh2bh(jh);
1923
1924 jbd_lock_bh_journal_head(bh);
1925 J_ASSERT_JH(jh, jh->b_jcount > 0);
1926 --jh->b_jcount;
1927 if (!jh->b_jcount && !jh->b_transaction) {
1928 __journal_remove_journal_head(bh);
1929 __brelse(bh);
1930 }
1931 jbd_unlock_bh_journal_head(bh);
1932}
1933
1934/*
Jose R. Santosc2a91592007-10-18 23:39:22 -07001935 * debugfs tunables
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 */
Jose R. Santosc2a91592007-10-18 23:39:22 -07001937#ifdef CONFIG_JBD_DEBUG
1938
1939u8 journal_enable_debug __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940EXPORT_SYMBOL(journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
Jose R. Santosc2a91592007-10-18 23:39:22 -07001942static struct dentry *jbd_debugfs_dir;
1943static struct dentry *jbd_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
Jose R. Santosc2a91592007-10-18 23:39:22 -07001945static void __init jbd_create_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001947 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
1948 if (jbd_debugfs_dir)
Yin Kangkai765f8362009-12-15 14:48:25 -08001949 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
Jose R. Santosc2a91592007-10-18 23:39:22 -07001950 jbd_debugfs_dir,
1951 &journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952}
1953
Jose R. Santosc2a91592007-10-18 23:39:22 -07001954static void __exit jbd_remove_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001956 debugfs_remove(jbd_debug);
1957 debugfs_remove(jbd_debugfs_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958}
1959
1960#else
1961
Jose R. Santosc2a91592007-10-18 23:39:22 -07001962static inline void jbd_create_debugfs_entry(void)
1963{
1964}
1965
1966static inline void jbd_remove_debugfs_entry(void)
1967{
1968}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
1970#endif
1971
Christoph Lametere18b8902006-12-06 20:33:20 -08001972struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
1974static int __init journal_init_handle_cache(void)
1975{
1976 jbd_handle_cache = kmem_cache_create("journal_handle",
1977 sizeof(handle_t),
1978 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001979 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001980 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 if (jbd_handle_cache == NULL) {
1982 printk(KERN_EMERG "JBD: failed to create handle cache\n");
1983 return -ENOMEM;
1984 }
1985 return 0;
1986}
1987
1988static void journal_destroy_handle_cache(void)
1989{
1990 if (jbd_handle_cache)
1991 kmem_cache_destroy(jbd_handle_cache);
1992}
1993
1994/*
1995 * Module startup and shutdown
1996 */
1997
1998static int __init journal_init_caches(void)
1999{
2000 int ret;
2001
2002 ret = journal_init_revoke_caches();
2003 if (ret == 0)
2004 ret = journal_init_journal_head_cache();
2005 if (ret == 0)
2006 ret = journal_init_handle_cache();
2007 return ret;
2008}
2009
2010static void journal_destroy_caches(void)
2011{
2012 journal_destroy_revoke_caches();
2013 journal_destroy_journal_head_cache();
2014 journal_destroy_handle_cache();
2015}
2016
2017static int __init journal_init(void)
2018{
2019 int ret;
2020
Alexey Dobriyan2aed3482006-09-27 01:49:28 -07002021 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
Adrian Bunk022a4a72005-09-06 15:16:41 -07002022
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 ret = journal_init_caches();
2024 if (ret != 0)
2025 journal_destroy_caches();
Jose R. Santosc2a91592007-10-18 23:39:22 -07002026 jbd_create_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 return ret;
2028}
2029
2030static void __exit journal_exit(void)
2031{
2032#ifdef CONFIG_JBD_DEBUG
2033 int n = atomic_read(&nr_journal_heads);
2034 if (n)
2035 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2036#endif
Jose R. Santosc2a91592007-10-18 23:39:22 -07002037 jbd_remove_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 journal_destroy_caches();
2039}
2040
2041MODULE_LICENSE("GPL");
2042module_init(journal_init);
2043module_exit(journal_exit);
2044