blob: e79c07812afa4d044ec46bc92d77ba8f759aacb4 [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);
76EXPORT_SYMBOL(journal_start_commit);
77EXPORT_SYMBOL(journal_force_commit_nested);
78EXPORT_SYMBOL(journal_wipe);
79EXPORT_SYMBOL(journal_blocks_per_page);
80EXPORT_SYMBOL(journal_invalidatepage);
81EXPORT_SYMBOL(journal_try_to_free_buffers);
82EXPORT_SYMBOL(journal_force_commit);
83
84static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
Adrian Bunk022a4a72005-09-06 15:16:41 -070085static void __journal_abort_soft (journal_t *journal, int errno);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
87/*
88 * Helper function used to manage commit timeouts
89 */
90
91static void commit_timeout(unsigned long __data)
92{
93 struct task_struct * p = (struct task_struct *) __data;
94
95 wake_up_process(p);
96}
97
Linus Torvalds1da177e2005-04-16 15:20:36 -070098/*
99 * kjournald: The main thread function used to manage a logging device
100 * journal.
101 *
102 * This kernel thread is responsible for two things:
103 *
104 * 1) COMMIT: Every so often we need to commit the current state of the
105 * filesystem to disk. The journal thread is responsible for writing
106 * all of the metadata buffers to disk.
107 *
108 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
109 * of the data in that part of the log has been rewritten elsewhere on
110 * the disk. Flushing these old buffers to reclaim space in the log is
111 * known as checkpointing, and this thread is responsible for that job.
112 */
113
Adrian Bunk022a4a72005-09-06 15:16:41 -0700114static int kjournald(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115{
Andrew Mortone3df1892006-03-25 03:06:53 -0800116 journal_t *journal = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 transaction_t *transaction;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Andrew Mortone3df1892006-03-25 03:06:53 -0800119 /*
120 * Set up an interval timer which can be used to trigger a commit wakeup
121 * after the commit interval expires
122 */
123 setup_timer(&journal->j_commit_timer, commit_timeout,
124 (unsigned long)current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
126 /* Record that the journal thread is running */
127 journal->j_task = current;
128 wake_up(&journal->j_wait_done_commit);
129
130 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
131 journal->j_commit_interval / HZ);
132
133 /*
134 * And now, wait forever for commit wakeup events.
135 */
136 spin_lock(&journal->j_state_lock);
137
138loop:
139 if (journal->j_flags & JFS_UNMOUNT)
140 goto end_loop;
141
142 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
143 journal->j_commit_sequence, journal->j_commit_request);
144
145 if (journal->j_commit_sequence != journal->j_commit_request) {
146 jbd_debug(1, "OK, requests differ\n");
147 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800148 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 journal_commit_transaction(journal);
150 spin_lock(&journal->j_state_lock);
151 goto loop;
152 }
153
154 wake_up(&journal->j_wait_done_commit);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700155 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 /*
157 * The simpler the better. Flushing journal isn't a
158 * good idea, because that depends on threads that may
159 * be already stopped.
160 */
161 jbd_debug(1, "Now suspending kjournald\n");
162 spin_unlock(&journal->j_state_lock);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700163 refrigerator();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 spin_lock(&journal->j_state_lock);
165 } else {
166 /*
167 * We assume on resume that commits are already there,
168 * so we don't sleep
169 */
170 DEFINE_WAIT(wait);
171 int should_sleep = 1;
172
173 prepare_to_wait(&journal->j_wait_commit, &wait,
174 TASK_INTERRUPTIBLE);
175 if (journal->j_commit_sequence != journal->j_commit_request)
176 should_sleep = 0;
177 transaction = journal->j_running_transaction;
178 if (transaction && time_after_eq(jiffies,
179 transaction->t_expires))
180 should_sleep = 0;
Mark Fashehcbf0d272005-09-06 15:19:08 -0700181 if (journal->j_flags & JFS_UNMOUNT)
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700182 should_sleep = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 if (should_sleep) {
184 spin_unlock(&journal->j_state_lock);
185 schedule();
186 spin_lock(&journal->j_state_lock);
187 }
188 finish_wait(&journal->j_wait_commit, &wait);
189 }
190
191 jbd_debug(1, "kjournald wakes\n");
192
193 /*
194 * Were we woken up by a commit wakeup event?
195 */
196 transaction = journal->j_running_transaction;
197 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
198 journal->j_commit_request = transaction->t_tid;
199 jbd_debug(1, "woke because of timeout\n");
200 }
201 goto loop;
202
203end_loop:
204 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800205 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 journal->j_task = NULL;
207 wake_up(&journal->j_wait_done_commit);
208 jbd_debug(1, "Journal thread exiting.\n");
209 return 0;
210}
211
Pavel Emelianov97f06782007-05-08 00:30:42 -0700212static int journal_start_thread(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700214 struct task_struct *t;
215
216 t = kthread_run(kjournald, journal, "kjournald");
217 if (IS_ERR(t))
218 return PTR_ERR(t);
219
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700220 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700221 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
223
224static void journal_kill_thread(journal_t *journal)
225{
226 spin_lock(&journal->j_state_lock);
227 journal->j_flags |= JFS_UNMOUNT;
228
229 while (journal->j_task) {
230 wake_up(&journal->j_wait_commit);
231 spin_unlock(&journal->j_state_lock);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700232 wait_event(journal->j_wait_done_commit,
233 journal->j_task == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 spin_lock(&journal->j_state_lock);
235 }
236 spin_unlock(&journal->j_state_lock);
237}
238
239/*
240 * journal_write_metadata_buffer: write a metadata buffer to the journal.
241 *
242 * Writes a metadata buffer to a given disk block. The actual IO is not
243 * performed but a new buffer_head is constructed which labels the data
244 * to be written with the correct destination disk block.
245 *
246 * Any magic-number escaping which needs to be done will cause a
247 * copy-out here. If the buffer happens to start with the
248 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
249 * magic number is only written to the log for descripter blocks. In
250 * this case, we copy the data and replace the first word with 0, and we
251 * return a result code which indicates that this buffer needs to be
252 * marked as an escaped buffer in the corresponding log descriptor
253 * block. The missing word can then be restored when the block is read
254 * during recovery.
255 *
256 * If the source buffer has already been modified by a new transaction
257 * since we took the last commit snapshot, we use the frozen copy of
258 * that data for IO. If we end up using the existing buffer_head's data
259 * for the write, then we *have* to lock the buffer to prevent anyone
260 * else from using and possibly modifying it while the IO is in
261 * progress.
262 *
263 * The function returns a pointer to the buffer_heads to be used for IO.
264 *
265 * We assume that the journal has already been locked in this function.
266 *
267 * Return value:
268 * <0: Error
269 * >=0: Finished OK
270 *
271 * On success:
272 * Bit 0 set == escape performed on the data
273 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
274 */
275
276int journal_write_metadata_buffer(transaction_t *transaction,
277 struct journal_head *jh_in,
278 struct journal_head **jh_out,
Eric Sandeen37ed3222006-09-27 01:49:31 -0700279 unsigned long blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 int need_copy_out = 0;
282 int done_copy_out = 0;
283 int do_escape = 0;
284 char *mapped_data;
285 struct buffer_head *new_bh;
286 struct journal_head *new_jh;
287 struct page *new_page;
288 unsigned int new_offset;
289 struct buffer_head *bh_in = jh2bh(jh_in);
290
291 /*
292 * The buffer really shouldn't be locked: only the current committing
293 * transaction is allowed to write it, so nobody else is allowed
294 * to do any IO.
295 *
296 * akpm: except if we're journalling data, and write() output is
297 * also part of a shared mapping, and another thread has
298 * decided to launch a writepage() against this buffer.
299 */
300 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
301
302 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
303
304 /*
305 * If a new transaction has already done a buffer copy-out, then
306 * we use that version of the data for the commit.
307 */
308 jbd_lock_bh_state(bh_in);
309repeat:
310 if (jh_in->b_frozen_data) {
311 done_copy_out = 1;
312 new_page = virt_to_page(jh_in->b_frozen_data);
313 new_offset = offset_in_page(jh_in->b_frozen_data);
314 } else {
315 new_page = jh2bh(jh_in)->b_page;
316 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
317 }
318
319 mapped_data = kmap_atomic(new_page, KM_USER0);
320 /*
321 * Check for escaping
322 */
323 if (*((__be32 *)(mapped_data + new_offset)) ==
324 cpu_to_be32(JFS_MAGIC_NUMBER)) {
325 need_copy_out = 1;
326 do_escape = 1;
327 }
328 kunmap_atomic(mapped_data, KM_USER0);
329
330 /*
331 * Do we need to do a data copy?
332 */
333 if (need_copy_out && !done_copy_out) {
334 char *tmp;
335
336 jbd_unlock_bh_state(bh_in);
Mingming Caoc089d492007-10-16 18:38:25 -0400337 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 jbd_lock_bh_state(bh_in);
339 if (jh_in->b_frozen_data) {
Mingming Caoc089d492007-10-16 18:38:25 -0400340 jbd_free(tmp, bh_in->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 goto repeat;
342 }
343
344 jh_in->b_frozen_data = tmp;
345 mapped_data = kmap_atomic(new_page, KM_USER0);
346 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
347 kunmap_atomic(mapped_data, KM_USER0);
348
349 new_page = virt_to_page(tmp);
350 new_offset = offset_in_page(tmp);
351 done_copy_out = 1;
352 }
353
354 /*
355 * Did we need to do an escaping? Now we've done all the
356 * copying, we can finally do so.
357 */
358 if (do_escape) {
359 mapped_data = kmap_atomic(new_page, KM_USER0);
360 *((unsigned int *)(mapped_data + new_offset)) = 0;
361 kunmap_atomic(mapped_data, KM_USER0);
362 }
363
364 /* keep subsequent assertions sane */
365 new_bh->b_state = 0;
366 init_buffer(new_bh, NULL, NULL);
367 atomic_set(&new_bh->b_count, 1);
368 jbd_unlock_bh_state(bh_in);
369
370 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
371
372 set_bh_page(new_bh, new_page, new_offset);
373 new_jh->b_transaction = NULL;
374 new_bh->b_size = jh2bh(jh_in)->b_size;
375 new_bh->b_bdev = transaction->t_journal->j_dev;
376 new_bh->b_blocknr = blocknr;
377 set_buffer_mapped(new_bh);
378 set_buffer_dirty(new_bh);
379
380 *jh_out = new_jh;
381
382 /*
383 * The to-be-written buffer needs to get moved to the io queue,
384 * and the original buffer whose contents we are shadowing or
385 * copying is moved to the transaction's shadow queue.
386 */
387 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
388 journal_file_buffer(jh_in, transaction, BJ_Shadow);
389 JBUFFER_TRACE(new_jh, "file as BJ_IO");
390 journal_file_buffer(new_jh, transaction, BJ_IO);
391
392 return do_escape | (done_copy_out << 1);
393}
394
395/*
396 * Allocation code for the journal file. Manage the space left in the
397 * journal, so that we can begin checkpointing when appropriate.
398 */
399
400/*
401 * __log_space_left: Return the number of free blocks left in the journal.
402 *
403 * Called with the journal already locked.
404 *
405 * Called under j_state_lock
406 */
407
408int __log_space_left(journal_t *journal)
409{
410 int left = journal->j_free;
411
412 assert_spin_locked(&journal->j_state_lock);
413
414 /*
415 * Be pessimistic here about the number of those free blocks which
416 * might be required for log descriptor control blocks.
417 */
418
419#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
420
421 left -= MIN_LOG_RESERVED_BLOCKS;
422
423 if (left <= 0)
424 return 0;
425 left -= (left >> 3);
426 return left;
427}
428
429/*
Jan Kara8fe4cd02009-02-11 13:04:25 -0800430 * Called under j_state_lock. Returns true if a transaction commit was started.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 */
432int __log_start_commit(journal_t *journal, tid_t target)
433{
434 /*
435 * Are we already doing a recent enough commit?
436 */
437 if (!tid_geq(journal->j_commit_request, target)) {
438 /*
439 * We want a new commit: OK, mark the request and wakup the
440 * commit thread. We do _not_ do the commit ourselves.
441 */
442
443 journal->j_commit_request = target;
444 jbd_debug(1, "JBD: requesting commit %d/%d\n",
445 journal->j_commit_request,
446 journal->j_commit_sequence);
447 wake_up(&journal->j_wait_commit);
448 return 1;
449 }
450 return 0;
451}
452
453int log_start_commit(journal_t *journal, tid_t tid)
454{
455 int ret;
456
457 spin_lock(&journal->j_state_lock);
458 ret = __log_start_commit(journal, tid);
459 spin_unlock(&journal->j_state_lock);
460 return ret;
461}
462
463/*
464 * Force and wait upon a commit if the calling process is not within
465 * transaction. This is used for forcing out undo-protected data which contains
466 * bitmaps, when the fs is running out of space.
467 *
468 * We can only force the running transaction if we don't have an active handle;
469 * otherwise, we will deadlock.
470 *
471 * Returns true if a transaction was started.
472 */
473int journal_force_commit_nested(journal_t *journal)
474{
475 transaction_t *transaction = NULL;
476 tid_t tid;
477
478 spin_lock(&journal->j_state_lock);
479 if (journal->j_running_transaction && !current->journal_info) {
480 transaction = journal->j_running_transaction;
481 __log_start_commit(journal, transaction->t_tid);
482 } else if (journal->j_committing_transaction)
483 transaction = journal->j_committing_transaction;
484
485 if (!transaction) {
486 spin_unlock(&journal->j_state_lock);
487 return 0; /* Nothing to retry */
488 }
489
490 tid = transaction->t_tid;
491 spin_unlock(&journal->j_state_lock);
492 log_wait_commit(journal, tid);
493 return 1;
494}
495
496/*
497 * Start a commit of the current running transaction (if any). Returns true
Jan Kara8fe4cd02009-02-11 13:04:25 -0800498 * if a transaction is going to be committed (or is currently already
499 * committing), and fills its tid in at *ptid
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 */
501int journal_start_commit(journal_t *journal, tid_t *ptid)
502{
503 int ret = 0;
504
505 spin_lock(&journal->j_state_lock);
506 if (journal->j_running_transaction) {
507 tid_t tid = journal->j_running_transaction->t_tid;
508
Jan Kara8fe4cd02009-02-11 13:04:25 -0800509 __log_start_commit(journal, tid);
510 /* There's a running transaction and we've just made sure
511 * it's commit has been scheduled. */
512 if (ptid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 *ptid = tid;
Jan Kara8fe4cd02009-02-11 13:04:25 -0800514 ret = 1;
515 } else if (journal->j_committing_transaction) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 /*
517 * If ext3_write_super() recently started a commit, then we
518 * have to wait for completion of that transaction
519 */
Jan Kara8fe4cd02009-02-11 13:04:25 -0800520 if (ptid)
521 *ptid = journal->j_committing_transaction->t_tid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 ret = 1;
523 }
524 spin_unlock(&journal->j_state_lock);
525 return ret;
526}
527
528/*
529 * Wait for a specified commit to complete.
530 * The caller may not hold the journal lock.
531 */
532int log_wait_commit(journal_t *journal, tid_t tid)
533{
534 int err = 0;
535
536#ifdef CONFIG_JBD_DEBUG
537 spin_lock(&journal->j_state_lock);
538 if (!tid_geq(journal->j_commit_request, tid)) {
539 printk(KERN_EMERG
540 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700541 __func__, journal->j_commit_request, tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
543 spin_unlock(&journal->j_state_lock);
544#endif
545 spin_lock(&journal->j_state_lock);
546 while (tid_gt(tid, journal->j_commit_sequence)) {
547 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
548 tid, journal->j_commit_sequence);
549 wake_up(&journal->j_wait_commit);
550 spin_unlock(&journal->j_state_lock);
551 wait_event(journal->j_wait_done_commit,
552 !tid_gt(tid, journal->j_commit_sequence));
553 spin_lock(&journal->j_state_lock);
554 }
555 spin_unlock(&journal->j_state_lock);
556
557 if (unlikely(is_journal_aborted(journal))) {
558 printk(KERN_EMERG "journal commit I/O error\n");
559 err = -EIO;
560 }
561 return err;
562}
563
564/*
565 * Log buffer allocation routines:
566 */
567
568int journal_next_log_block(journal_t *journal, unsigned long *retp)
569{
570 unsigned long blocknr;
571
572 spin_lock(&journal->j_state_lock);
573 J_ASSERT(journal->j_free > 1);
574
575 blocknr = journal->j_head;
576 journal->j_head++;
577 journal->j_free--;
578 if (journal->j_head == journal->j_last)
579 journal->j_head = journal->j_first;
580 spin_unlock(&journal->j_state_lock);
581 return journal_bmap(journal, blocknr, retp);
582}
583
584/*
585 * Conversion of logical to physical block numbers for the journal
586 *
587 * On external journals the journal blocks are identity-mapped, so
588 * this is a no-op. If needed, we can use j_blk_offset - everything is
589 * ready.
590 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700591int journal_bmap(journal_t *journal, unsigned long blocknr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 unsigned long *retp)
593{
594 int err = 0;
595 unsigned long ret;
596
597 if (journal->j_inode) {
598 ret = bmap(journal->j_inode, blocknr);
599 if (ret)
600 *retp = ret;
601 else {
602 char b[BDEVNAME_SIZE];
603
604 printk(KERN_ALERT "%s: journal block not found "
605 "at offset %lu on %s\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700606 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 blocknr,
608 bdevname(journal->j_dev, b));
609 err = -EIO;
610 __journal_abort_soft(journal, err);
611 }
612 } else {
613 *retp = blocknr; /* +journal->j_blk_offset */
614 }
615 return err;
616}
617
618/*
619 * We play buffer_head aliasing tricks to write data/metadata blocks to
620 * the journal without copying their contents, but for journal
621 * descriptor blocks we do need to generate bona fide buffers.
622 *
623 * After the caller of journal_get_descriptor_buffer() has finished modifying
624 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
625 * But we don't bother doing that, so there will be coherency problems with
626 * mmaps of blockdevs which hold live JBD-controlled filesystems.
627 */
628struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
629{
630 struct buffer_head *bh;
631 unsigned long blocknr;
632 int err;
633
634 err = journal_next_log_block(journal, &blocknr);
635
636 if (err)
637 return NULL;
638
639 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
640 lock_buffer(bh);
641 memset(bh->b_data, 0, journal->j_blocksize);
642 set_buffer_uptodate(bh);
643 unlock_buffer(bh);
644 BUFFER_TRACE(bh, "return this buffer");
645 return journal_add_journal_head(bh);
646}
647
648/*
649 * Management for journal control blocks: functions to create and
650 * destroy journal_t structures, and to initialise and read existing
651 * journal blocks from disk. */
652
653/* First: create and setup a journal_t object in memory. We initialise
654 * very few fields yet: that has to wait until we have created the
655 * journal structures from from scratch, or loaded them from disk. */
656
657static journal_t * journal_init_common (void)
658{
659 journal_t *journal;
660 int err;
661
Mingming Cao8c3478a2007-10-18 23:39:20 -0700662 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 if (!journal)
664 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
666 init_waitqueue_head(&journal->j_wait_transaction_locked);
667 init_waitqueue_head(&journal->j_wait_logspace);
668 init_waitqueue_head(&journal->j_wait_done_commit);
669 init_waitqueue_head(&journal->j_wait_checkpoint);
670 init_waitqueue_head(&journal->j_wait_commit);
671 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800672 mutex_init(&journal->j_barrier);
673 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 spin_lock_init(&journal->j_revoke_lock);
675 spin_lock_init(&journal->j_list_lock);
676 spin_lock_init(&journal->j_state_lock);
677
678 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
679
680 /* The journal is marked for error until we succeed with recovery! */
681 journal->j_flags = JFS_ABORT;
682
683 /* Set up a default-sized revoke table for the new mount. */
684 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
685 if (err) {
686 kfree(journal);
687 goto fail;
688 }
689 return journal;
690fail:
691 return NULL;
692}
693
694/* journal_init_dev and journal_init_inode:
695 *
696 * Create a journal structure assigned some fixed set of disk blocks to
697 * the journal. We don't actually touch those disk blocks yet, but we
698 * need to set up all of the mapping information to tell the journaling
699 * system where the journal blocks are.
700 *
701 */
702
703/**
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700704 * journal_t * journal_init_dev() - creates and initialises a journal structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 * @bdev: Block device on which to create the journal
706 * @fs_dev: Device which hold journalled filesystem for this journal.
707 * @start: Block nr Start of journal.
Eric Sandeen37ed3222006-09-27 01:49:31 -0700708 * @len: Length of the journal in blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 * @blocksize: blocksize of journalling device
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700710 *
711 * Returns: a newly created journal_t *
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700712 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 * journal_init_dev creates a journal which maps a fixed contiguous
714 * range of blocks on an arbitrary block device.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700715 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 */
717journal_t * journal_init_dev(struct block_device *bdev,
718 struct block_device *fs_dev,
719 int start, int len, int blocksize)
720{
721 journal_t *journal = journal_init_common();
722 struct buffer_head *bh;
723 int n;
724
725 if (!journal)
726 return NULL;
727
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 /* journal descriptor can store up to n blocks -bzzz */
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700729 journal->j_blocksize = blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 n = journal->j_blocksize / sizeof(journal_block_tag_t);
731 journal->j_wbufsize = n;
732 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
733 if (!journal->j_wbuf) {
734 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700735 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 kfree(journal);
737 journal = NULL;
Eric Sesterhenn41716c72006-10-11 01:20:37 -0700738 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700740 journal->j_dev = bdev;
741 journal->j_fs_dev = fs_dev;
742 journal->j_blk_offset = start;
743 journal->j_maxlen = len;
744
745 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
746 J_ASSERT(bh != NULL);
747 journal->j_sb_buffer = bh;
748 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Eric Sesterhenn41716c72006-10-11 01:20:37 -0700749out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 return journal;
751}
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700752
753/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
755 * @inode: An inode to create the journal in
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700756 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 * journal_init_inode creates a journal which maps an on-disk inode as
758 * the journal. The inode must exist already, must support bmap() and
759 * must have all data blocks preallocated.
760 */
761journal_t * journal_init_inode (struct inode *inode)
762{
763 struct buffer_head *bh;
764 journal_t *journal = journal_init_common();
765 int err;
766 int n;
767 unsigned long blocknr;
768
769 if (!journal)
770 return NULL;
771
772 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
773 journal->j_inode = inode;
774 jbd_debug(1,
775 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700776 journal, inode->i_sb->s_id, inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 (long long) inode->i_size,
778 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
779
780 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
781 journal->j_blocksize = inode->i_sb->s_blocksize;
782
783 /* journal descriptor can store up to n blocks -bzzz */
784 n = journal->j_blocksize / sizeof(journal_block_tag_t);
785 journal->j_wbufsize = n;
786 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
787 if (!journal->j_wbuf) {
788 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700789 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 kfree(journal);
791 return NULL;
792 }
793
794 err = journal_bmap(journal, 0, &blocknr);
795 /* If that failed, give up */
796 if (err) {
797 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700798 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 kfree(journal);
800 return NULL;
801 }
802
803 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
804 J_ASSERT(bh != NULL);
805 journal->j_sb_buffer = bh;
806 journal->j_superblock = (journal_superblock_t *)bh->b_data;
807
808 return journal;
809}
810
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700811/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 * If the journal init or create aborts, we need to mark the journal
813 * superblock as being NULL to prevent the journal destroy from writing
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700814 * back a bogus superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 */
816static void journal_fail_superblock (journal_t *journal)
817{
818 struct buffer_head *bh = journal->j_sb_buffer;
819 brelse(bh);
820 journal->j_sb_buffer = NULL;
821}
822
823/*
824 * Given a journal_t structure, initialise the various fields for
825 * startup of a new journaling session. We use this both when creating
826 * a journal, and after recovering an old journal to reset it for
827 * subsequent use.
828 */
829
830static int journal_reset(journal_t *journal)
831{
832 journal_superblock_t *sb = journal->j_superblock;
Eric Sandeen37ed3222006-09-27 01:49:31 -0700833 unsigned long first, last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
835 first = be32_to_cpu(sb->s_first);
836 last = be32_to_cpu(sb->s_maxlen);
837
838 journal->j_first = first;
839 journal->j_last = last;
840
841 journal->j_head = first;
842 journal->j_tail = first;
843 journal->j_free = last - first;
844
845 journal->j_tail_sequence = journal->j_transaction_sequence;
846 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
847 journal->j_commit_request = journal->j_commit_sequence;
848
849 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
850
851 /* Add the dynamic fields and write it to disk. */
852 journal_update_superblock(journal, 1);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700853 return journal_start_thread(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854}
855
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700856/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 * int journal_create() - Initialise the new journal file
858 * @journal: Journal to create. This structure must have been initialised
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700859 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 * Given a journal_t structure which tells us which disk blocks we can
861 * use, create a new journal superblock and initialise all of the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700862 * journal fields from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 **/
864int journal_create(journal_t *journal)
865{
866 unsigned long blocknr;
867 struct buffer_head *bh;
868 journal_superblock_t *sb;
869 int i, err;
870
871 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
872 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
873 journal->j_maxlen);
874 journal_fail_superblock(journal);
875 return -EINVAL;
876 }
877
878 if (journal->j_inode == NULL) {
879 /*
880 * We don't know what block to start at!
881 */
882 printk(KERN_EMERG
883 "%s: creation of journal on external device!\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700884 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 BUG();
886 }
887
888 /* Zero out the entire journal on disk. We cannot afford to
889 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
890 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
891 for (i = 0; i < journal->j_maxlen; i++) {
892 err = journal_bmap(journal, i, &blocknr);
893 if (err)
894 return err;
895 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
896 lock_buffer(bh);
897 memset (bh->b_data, 0, journal->j_blocksize);
898 BUFFER_TRACE(bh, "marking dirty");
899 mark_buffer_dirty(bh);
900 BUFFER_TRACE(bh, "marking uptodate");
901 set_buffer_uptodate(bh);
902 unlock_buffer(bh);
903 __brelse(bh);
904 }
905
906 sync_blockdev(journal->j_dev);
907 jbd_debug(1, "JBD: journal cleared.\n");
908
909 /* OK, fill in the initial static fields in the new superblock */
910 sb = journal->j_superblock;
911
912 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
913 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
914
915 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
916 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
917 sb->s_first = cpu_to_be32(1);
918
919 journal->j_transaction_sequence = 1;
920
921 journal->j_flags &= ~JFS_ABORT;
922 journal->j_format_version = 2;
923
924 return journal_reset(journal);
925}
926
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700927/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 * void journal_update_superblock() - Update journal sb on disk.
929 * @journal: The journal to update.
930 * @wait: Set to '0' if you don't want to wait for IO completion.
931 *
932 * Update a journal's dynamic superblock fields and write it to disk,
933 * optionally waiting for the IO to complete.
934 */
935void journal_update_superblock(journal_t *journal, int wait)
936{
937 journal_superblock_t *sb = journal->j_superblock;
938 struct buffer_head *bh = journal->j_sb_buffer;
939
940 /*
941 * As a special case, if the on-disk copy is already marked as needing
942 * no recovery (s_start == 0) and there are no outstanding transactions
943 * in the filesystem, then we can safely defer the superblock update
944 * until the next commit by setting JFS_FLUSHED. This avoids
945 * attempting a write to a potential-readonly device.
946 */
947 if (sb->s_start == 0 && journal->j_tail_sequence ==
948 journal->j_transaction_sequence) {
949 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
950 "(start %ld, seq %d, errno %d)\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700951 journal->j_tail, journal->j_tail_sequence,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 journal->j_errno);
953 goto out;
954 }
955
956 spin_lock(&journal->j_state_lock);
957 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
958 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
959
960 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
961 sb->s_start = cpu_to_be32(journal->j_tail);
962 sb->s_errno = cpu_to_be32(journal->j_errno);
963 spin_unlock(&journal->j_state_lock);
964
965 BUFFER_TRACE(bh, "marking dirty");
966 mark_buffer_dirty(bh);
967 if (wait)
968 sync_dirty_buffer(bh);
969 else
Jan Kara26707692005-09-06 15:19:12 -0700970 ll_rw_block(SWRITE, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
972out:
973 /* If we have just flushed the log (by marking s_start==0), then
974 * any future commit will have to be careful to update the
975 * superblock again to re-record the true start of the log. */
976
977 spin_lock(&journal->j_state_lock);
978 if (sb->s_start)
979 journal->j_flags &= ~JFS_FLUSHED;
980 else
981 journal->j_flags |= JFS_FLUSHED;
982 spin_unlock(&journal->j_state_lock);
983}
984
985/*
986 * Read the superblock for a given journal, performing initial
987 * validation of the format.
988 */
989
990static int journal_get_superblock(journal_t *journal)
991{
992 struct buffer_head *bh;
993 journal_superblock_t *sb;
994 int err = -EIO;
995
996 bh = journal->j_sb_buffer;
997
998 J_ASSERT(bh != NULL);
999 if (!buffer_uptodate(bh)) {
1000 ll_rw_block(READ, 1, &bh);
1001 wait_on_buffer(bh);
1002 if (!buffer_uptodate(bh)) {
1003 printk (KERN_ERR
1004 "JBD: IO error reading journal superblock\n");
1005 goto out;
1006 }
1007 }
1008
1009 sb = journal->j_superblock;
1010
1011 err = -EINVAL;
1012
1013 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1014 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1015 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1016 goto out;
1017 }
1018
1019 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1020 case JFS_SUPERBLOCK_V1:
1021 journal->j_format_version = 1;
1022 break;
1023 case JFS_SUPERBLOCK_V2:
1024 journal->j_format_version = 2;
1025 break;
1026 default:
1027 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1028 goto out;
1029 }
1030
1031 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1032 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1033 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1034 printk (KERN_WARNING "JBD: journal file too short\n");
1035 goto out;
1036 }
1037
1038 return 0;
1039
1040out:
1041 journal_fail_superblock(journal);
1042 return err;
1043}
1044
1045/*
1046 * Load the on-disk journal superblock and read the key fields into the
1047 * journal_t.
1048 */
1049
1050static int load_superblock(journal_t *journal)
1051{
1052 int err;
1053 journal_superblock_t *sb;
1054
1055 err = journal_get_superblock(journal);
1056 if (err)
1057 return err;
1058
1059 sb = journal->j_superblock;
1060
1061 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1062 journal->j_tail = be32_to_cpu(sb->s_start);
1063 journal->j_first = be32_to_cpu(sb->s_first);
1064 journal->j_last = be32_to_cpu(sb->s_maxlen);
1065 journal->j_errno = be32_to_cpu(sb->s_errno);
1066
1067 return 0;
1068}
1069
1070
1071/**
1072 * int journal_load() - Read journal from disk.
1073 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001074 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 * Given a journal_t structure which tells us which disk blocks contain
1076 * a journal, read the journal from disk to initialise the in-memory
1077 * structures.
1078 */
1079int journal_load(journal_t *journal)
1080{
1081 int err;
Badari Pulavartyea817392006-08-27 01:23:52 -07001082 journal_superblock_t *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
1084 err = load_superblock(journal);
1085 if (err)
1086 return err;
1087
Badari Pulavartyea817392006-08-27 01:23:52 -07001088 sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 /* If this is a V2 superblock, then we have to check the
1090 * features flags on it. */
1091
1092 if (journal->j_format_version >= 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 if ((sb->s_feature_ro_compat &
1094 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1095 (sb->s_feature_incompat &
1096 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1097 printk (KERN_WARNING
1098 "JBD: Unrecognised features on journal\n");
1099 return -EINVAL;
1100 }
1101 }
1102
1103 /* Let the recovery code check whether it needs to recover any
1104 * data from the journal. */
1105 if (journal_recover(journal))
1106 goto recovery_error;
1107
1108 /* OK, we've finished with the dynamic journal bits:
1109 * reinitialise the dynamic contents of the superblock in memory
1110 * and reset them on disk. */
1111 if (journal_reset(journal))
1112 goto recovery_error;
1113
1114 journal->j_flags &= ~JFS_ABORT;
1115 journal->j_flags |= JFS_LOADED;
1116 return 0;
1117
1118recovery_error:
1119 printk (KERN_WARNING "JBD: recovery failed\n");
1120 return -EIO;
1121}
1122
1123/**
1124 * void journal_destroy() - Release a journal_t structure.
1125 * @journal: Journal to act on.
1126 *
1127 * Release a journal_t structure once it is no longer in use by the
1128 * journaled object.
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001129 * Return <0 if we couldn't clean up the journal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 */
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001131int journal_destroy(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132{
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001133 int err = 0;
1134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 /* Wait for the commit thread to wake up and die. */
1136 journal_kill_thread(journal);
1137
1138 /* Force a final log commit */
1139 if (journal->j_running_transaction)
1140 journal_commit_transaction(journal);
1141
1142 /* Force any old transactions to disk */
1143
1144 /* Totally anal locking here... */
1145 spin_lock(&journal->j_list_lock);
1146 while (journal->j_checkpoint_transactions != NULL) {
1147 spin_unlock(&journal->j_list_lock);
1148 log_do_checkpoint(journal);
1149 spin_lock(&journal->j_list_lock);
1150 }
1151
1152 J_ASSERT(journal->j_running_transaction == NULL);
1153 J_ASSERT(journal->j_committing_transaction == NULL);
1154 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1155 spin_unlock(&journal->j_list_lock);
1156
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (journal->j_sb_buffer) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001158 if (!is_journal_aborted(journal)) {
1159 /* We can now mark the journal as empty. */
1160 journal->j_tail = 0;
1161 journal->j_tail_sequence =
1162 ++journal->j_transaction_sequence;
1163 journal_update_superblock(journal, 1);
1164 } else {
1165 err = -EIO;
1166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 brelse(journal->j_sb_buffer);
1168 }
1169
1170 if (journal->j_inode)
1171 iput(journal->j_inode);
1172 if (journal->j_revoke)
1173 journal_destroy_revoke(journal);
1174 kfree(journal->j_wbuf);
1175 kfree(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001176
1177 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178}
1179
1180
1181/**
1182 *int journal_check_used_features () - Check if features specified are used.
1183 * @journal: Journal to check.
1184 * @compat: bitmask of compatible features
1185 * @ro: bitmask of features that force read-only mount
1186 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001187 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 * Check whether the journal uses all of a given set of
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001189 * features. Return true (non-zero) if it does.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 **/
1191
1192int journal_check_used_features (journal_t *journal, unsigned long compat,
1193 unsigned long ro, unsigned long incompat)
1194{
1195 journal_superblock_t *sb;
1196
1197 if (!compat && !ro && !incompat)
1198 return 1;
1199 if (journal->j_format_version == 1)
1200 return 0;
1201
1202 sb = journal->j_superblock;
1203
1204 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1205 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1206 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1207 return 1;
1208
1209 return 0;
1210}
1211
1212/**
1213 * int journal_check_available_features() - Check feature set in journalling layer
1214 * @journal: Journal to check.
1215 * @compat: bitmask of compatible features
1216 * @ro: bitmask of features that force read-only mount
1217 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001218 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 * Check whether the journaling code supports the use of
1220 * all of a given set of features on this journal. Return true
1221 * (non-zero) if it can. */
1222
1223int journal_check_available_features (journal_t *journal, unsigned long compat,
1224 unsigned long ro, unsigned long incompat)
1225{
1226 journal_superblock_t *sb;
1227
1228 if (!compat && !ro && !incompat)
1229 return 1;
1230
1231 sb = journal->j_superblock;
1232
1233 /* We can support any known requested features iff the
1234 * superblock is in version 2. Otherwise we fail to support any
1235 * extended sb features. */
1236
1237 if (journal->j_format_version != 2)
1238 return 0;
1239
1240 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1241 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1242 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1243 return 1;
1244
1245 return 0;
1246}
1247
1248/**
1249 * int journal_set_features () - Mark a given journal feature in the superblock
1250 * @journal: Journal to act on.
1251 * @compat: bitmask of compatible features
1252 * @ro: bitmask of features that force read-only mount
1253 * @incompat: bitmask of incompatible features
1254 *
1255 * Mark a given journal feature as present on the
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001256 * superblock. Returns true if the requested features could be set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 *
1258 */
1259
1260int journal_set_features (journal_t *journal, unsigned long compat,
1261 unsigned long ro, unsigned long incompat)
1262{
1263 journal_superblock_t *sb;
1264
1265 if (journal_check_used_features(journal, compat, ro, incompat))
1266 return 1;
1267
1268 if (!journal_check_available_features(journal, compat, ro, incompat))
1269 return 0;
1270
1271 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1272 compat, ro, incompat);
1273
1274 sb = journal->j_superblock;
1275
1276 sb->s_feature_compat |= cpu_to_be32(compat);
1277 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1278 sb->s_feature_incompat |= cpu_to_be32(incompat);
1279
1280 return 1;
1281}
1282
1283
1284/**
1285 * int journal_update_format () - Update on-disk journal structure.
1286 * @journal: Journal to act on.
1287 *
1288 * Given an initialised but unloaded journal struct, poke about in the
1289 * on-disk structure to update it to the most recent supported version.
1290 */
1291int journal_update_format (journal_t *journal)
1292{
1293 journal_superblock_t *sb;
1294 int err;
1295
1296 err = journal_get_superblock(journal);
1297 if (err)
1298 return err;
1299
1300 sb = journal->j_superblock;
1301
1302 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1303 case JFS_SUPERBLOCK_V2:
1304 return 0;
1305 case JFS_SUPERBLOCK_V1:
1306 return journal_convert_superblock_v1(journal, sb);
1307 default:
1308 break;
1309 }
1310 return -EINVAL;
1311}
1312
1313static int journal_convert_superblock_v1(journal_t *journal,
1314 journal_superblock_t *sb)
1315{
1316 int offset, blocksize;
1317 struct buffer_head *bh;
1318
1319 printk(KERN_WARNING
1320 "JBD: Converting superblock from version 1 to 2.\n");
1321
1322 /* Pre-initialise new fields to zero */
1323 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1324 blocksize = be32_to_cpu(sb->s_blocksize);
1325 memset(&sb->s_feature_compat, 0, blocksize-offset);
1326
1327 sb->s_nr_users = cpu_to_be32(1);
1328 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1329 journal->j_format_version = 2;
1330
1331 bh = journal->j_sb_buffer;
1332 BUFFER_TRACE(bh, "marking dirty");
1333 mark_buffer_dirty(bh);
1334 sync_dirty_buffer(bh);
1335 return 0;
1336}
1337
1338
1339/**
1340 * int journal_flush () - Flush journal
1341 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001342 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 * Flush all data for a given journal to disk and empty the journal.
1344 * Filesystems can use this when remounting readonly to ensure that
1345 * recovery does not need to happen on remount.
1346 */
1347
1348int journal_flush(journal_t *journal)
1349{
1350 int err = 0;
1351 transaction_t *transaction = NULL;
1352 unsigned long old_tail;
1353
1354 spin_lock(&journal->j_state_lock);
1355
1356 /* Force everything buffered to the log... */
1357 if (journal->j_running_transaction) {
1358 transaction = journal->j_running_transaction;
1359 __log_start_commit(journal, transaction->t_tid);
1360 } else if (journal->j_committing_transaction)
1361 transaction = journal->j_committing_transaction;
1362
1363 /* Wait for the log commit to complete... */
1364 if (transaction) {
1365 tid_t tid = transaction->t_tid;
1366
1367 spin_unlock(&journal->j_state_lock);
1368 log_wait_commit(journal, tid);
1369 } else {
1370 spin_unlock(&journal->j_state_lock);
1371 }
1372
1373 /* ...and flush everything in the log out to disk. */
1374 spin_lock(&journal->j_list_lock);
1375 while (!err && journal->j_checkpoint_transactions != NULL) {
1376 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001377 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 err = log_do_checkpoint(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001379 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 spin_lock(&journal->j_list_lock);
1381 }
1382 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001383
1384 if (is_journal_aborted(journal))
1385 return -EIO;
1386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 cleanup_journal_tail(journal);
1388
1389 /* Finally, mark the journal as really needing no recovery.
1390 * This sets s_start==0 in the underlying superblock, which is
1391 * the magic code for a fully-recovered superblock. Any future
1392 * commits of data to the journal will restore the current
1393 * s_start value. */
1394 spin_lock(&journal->j_state_lock);
1395 old_tail = journal->j_tail;
1396 journal->j_tail = 0;
1397 spin_unlock(&journal->j_state_lock);
1398 journal_update_superblock(journal, 1);
1399 spin_lock(&journal->j_state_lock);
1400 journal->j_tail = old_tail;
1401
1402 J_ASSERT(!journal->j_running_transaction);
1403 J_ASSERT(!journal->j_committing_transaction);
1404 J_ASSERT(!journal->j_checkpoint_transactions);
1405 J_ASSERT(journal->j_head == journal->j_tail);
1406 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1407 spin_unlock(&journal->j_state_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001408 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409}
1410
1411/**
1412 * int journal_wipe() - Wipe journal contents
1413 * @journal: Journal to act on.
1414 * @write: flag (see below)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001415 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 * Wipe out all of the contents of a journal, safely. This will produce
1417 * a warning if the journal contains any valid recovery information.
1418 * Must be called between journal_init_*() and journal_load().
1419 *
1420 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1421 * we merely suppress recovery.
1422 */
1423
1424int journal_wipe(journal_t *journal, int write)
1425{
1426 journal_superblock_t *sb;
1427 int err = 0;
1428
1429 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1430
1431 err = load_superblock(journal);
1432 if (err)
1433 return err;
1434
1435 sb = journal->j_superblock;
1436
1437 if (!journal->j_tail)
1438 goto no_recovery;
1439
1440 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1441 write ? "Clearing" : "Ignoring");
1442
1443 err = journal_skip_recovery(journal);
1444 if (write)
1445 journal_update_superblock(journal, 1);
1446
1447 no_recovery:
1448 return err;
1449}
1450
1451/*
1452 * journal_dev_name: format a character string to describe on what
1453 * device this journal is present.
1454 */
1455
Adrian Bunk022a4a72005-09-06 15:16:41 -07001456static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457{
1458 struct block_device *bdev;
1459
1460 if (journal->j_inode)
1461 bdev = journal->j_inode->i_sb->s_bdev;
1462 else
1463 bdev = journal->j_dev;
1464
1465 return bdevname(bdev, buffer);
1466}
1467
1468/*
1469 * Journal abort has very specific semantics, which we describe
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001470 * for journal abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 *
1472 * Two internal function, which provide abort to te jbd layer
1473 * itself are here.
1474 */
1475
1476/*
1477 * Quick version for internal journal use (doesn't lock the journal).
1478 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1479 * and don't attempt to make any other journal updates.
1480 */
Adrian Bunk53308382008-02-06 01:40:12 -08001481static void __journal_abort_hard(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482{
1483 transaction_t *transaction;
1484 char b[BDEVNAME_SIZE];
1485
1486 if (journal->j_flags & JFS_ABORT)
1487 return;
1488
1489 printk(KERN_ERR "Aborting journal on device %s.\n",
1490 journal_dev_name(journal, b));
1491
1492 spin_lock(&journal->j_state_lock);
1493 journal->j_flags |= JFS_ABORT;
1494 transaction = journal->j_running_transaction;
1495 if (transaction)
1496 __log_start_commit(journal, transaction->t_tid);
1497 spin_unlock(&journal->j_state_lock);
1498}
1499
1500/* Soft abort: record the abort error status in the journal superblock,
1501 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001502static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
1504 if (journal->j_flags & JFS_ABORT)
1505 return;
1506
1507 if (!journal->j_errno)
1508 journal->j_errno = errno;
1509
1510 __journal_abort_hard(journal);
1511
1512 if (errno)
1513 journal_update_superblock(journal, 1);
1514}
1515
1516/**
1517 * void journal_abort () - Shutdown the journal immediately.
1518 * @journal: the journal to shutdown.
1519 * @errno: an error number to record in the journal indicating
1520 * the reason for the shutdown.
1521 *
1522 * Perform a complete, immediate shutdown of the ENTIRE
1523 * journal (not of a single transaction). This operation cannot be
1524 * undone without closing and reopening the journal.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001525 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 * The journal_abort function is intended to support higher level error
1527 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1528 * mode.
1529 *
1530 * Journal abort has very specific semantics. Any existing dirty,
1531 * unjournaled buffers in the main filesystem will still be written to
1532 * disk by bdflush, but the journaling mechanism will be suspended
1533 * immediately and no further transaction commits will be honoured.
1534 *
1535 * Any dirty, journaled buffers will be written back to disk without
1536 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1537 * filesystem, but we _do_ attempt to leave as much data as possible
1538 * behind for fsck to use for cleanup.
1539 *
1540 * Any attempt to get a new transaction handle on a journal which is in
1541 * ABORT state will just result in an -EROFS error return. A
1542 * journal_stop on an existing handle will return -EIO if we have
1543 * entered abort state during the update.
1544 *
1545 * Recursive transactions are not disturbed by journal abort until the
1546 * final journal_stop, which will receive the -EIO error.
1547 *
1548 * Finally, the journal_abort call allows the caller to supply an errno
1549 * which will be recorded (if possible) in the journal superblock. This
1550 * allows a client to record failure conditions in the middle of a
1551 * transaction without having to complete the transaction to record the
1552 * failure to disk. ext3_error, for example, now uses this
1553 * functionality.
1554 *
1555 * Errors which originate from within the journaling layer will NOT
1556 * supply an errno; a null errno implies that absolutely no further
1557 * writes are done to the journal (unless there are any already in
1558 * progress).
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001559 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 */
1561
1562void journal_abort(journal_t *journal, int errno)
1563{
1564 __journal_abort_soft(journal, errno);
1565}
1566
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001567/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 * int journal_errno () - returns the journal's error state.
1569 * @journal: journal to examine.
1570 *
1571 * This is the errno numbet set with journal_abort(), the last
1572 * time the journal was mounted - if the journal was stopped
1573 * without calling abort this will be 0.
1574 *
1575 * If the journal has been aborted on this mount time -EROFS will
1576 * be returned.
1577 */
1578int journal_errno(journal_t *journal)
1579{
1580 int err;
1581
1582 spin_lock(&journal->j_state_lock);
1583 if (journal->j_flags & JFS_ABORT)
1584 err = -EROFS;
1585 else
1586 err = journal->j_errno;
1587 spin_unlock(&journal->j_state_lock);
1588 return err;
1589}
1590
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001591/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 * int journal_clear_err () - clears the journal's error state
1593 * @journal: journal to act on.
1594 *
1595 * An error must be cleared or Acked to take a FS out of readonly
1596 * mode.
1597 */
1598int journal_clear_err(journal_t *journal)
1599{
1600 int err = 0;
1601
1602 spin_lock(&journal->j_state_lock);
1603 if (journal->j_flags & JFS_ABORT)
1604 err = -EROFS;
1605 else
1606 journal->j_errno = 0;
1607 spin_unlock(&journal->j_state_lock);
1608 return err;
1609}
1610
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001611/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 * void journal_ack_err() - Ack journal err.
1613 * @journal: journal to act on.
1614 *
1615 * An error must be cleared or Acked to take a FS out of readonly
1616 * mode.
1617 */
1618void journal_ack_err(journal_t *journal)
1619{
1620 spin_lock(&journal->j_state_lock);
1621 if (journal->j_errno)
1622 journal->j_flags |= JFS_ACK_ERR;
1623 spin_unlock(&journal->j_state_lock);
1624}
1625
1626int journal_blocks_per_page(struct inode *inode)
1627{
1628 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1629}
1630
1631/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 * Journal_head storage management
1633 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001634static struct kmem_cache *journal_head_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635#ifdef CONFIG_JBD_DEBUG
1636static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1637#endif
1638
1639static int journal_init_journal_head_cache(void)
1640{
1641 int retval;
1642
Al Viro1076d172008-03-29 03:07:18 +00001643 J_ASSERT(journal_head_cache == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 journal_head_cache = kmem_cache_create("journal_head",
1645 sizeof(struct journal_head),
1646 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001647 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001648 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00001650 if (!journal_head_cache) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 retval = -ENOMEM;
1652 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1653 }
1654 return retval;
1655}
1656
1657static void journal_destroy_journal_head_cache(void)
1658{
Duane Griffin3850f7a2008-07-25 01:46:19 -07001659 if (journal_head_cache) {
1660 kmem_cache_destroy(journal_head_cache);
1661 journal_head_cache = NULL;
1662 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663}
1664
1665/*
1666 * journal_head splicing and dicing
1667 */
1668static struct journal_head *journal_alloc_journal_head(void)
1669{
1670 struct journal_head *ret;
1671 static unsigned long last_warning;
1672
1673#ifdef CONFIG_JBD_DEBUG
1674 atomic_inc(&nr_journal_heads);
1675#endif
1676 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001677 if (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 jbd_debug(1, "out of memory for journal_head\n");
1679 if (time_after(jiffies, last_warning + 5*HZ)) {
1680 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001681 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 last_warning = jiffies;
1683 }
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001684 while (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 yield();
1686 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1687 }
1688 }
1689 return ret;
1690}
1691
1692static void journal_free_journal_head(struct journal_head *jh)
1693{
1694#ifdef CONFIG_JBD_DEBUG
1695 atomic_dec(&nr_journal_heads);
Randy Dunlapc9cf5522006-06-27 02:53:52 -07001696 memset(jh, JBD_POISON_FREE, sizeof(*jh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697#endif
1698 kmem_cache_free(journal_head_cache, jh);
1699}
1700
1701/*
1702 * A journal_head is attached to a buffer_head whenever JBD has an
1703 * interest in the buffer.
1704 *
1705 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1706 * is set. This bit is tested in core kernel code where we need to take
1707 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1708 * there.
1709 *
1710 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1711 *
1712 * When a buffer has its BH_JBD bit set it is immune from being released by
1713 * core kernel code, mainly via ->b_count.
1714 *
1715 * A journal_head may be detached from its buffer_head when the journal_head's
1716 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1717 * Various places in JBD call journal_remove_journal_head() to indicate that the
1718 * journal_head can be dropped if needed.
1719 *
1720 * Various places in the kernel want to attach a journal_head to a buffer_head
1721 * _before_ attaching the journal_head to a transaction. To protect the
1722 * journal_head in this situation, journal_add_journal_head elevates the
1723 * journal_head's b_jcount refcount by one. The caller must call
1724 * journal_put_journal_head() to undo this.
1725 *
1726 * So the typical usage would be:
1727 *
1728 * (Attach a journal_head if needed. Increments b_jcount)
1729 * struct journal_head *jh = journal_add_journal_head(bh);
1730 * ...
1731 * jh->b_transaction = xxx;
1732 * journal_put_journal_head(jh);
1733 *
1734 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1735 * because it has a non-zero b_transaction.
1736 */
1737
1738/*
1739 * Give a buffer_head a journal_head.
1740 *
1741 * Doesn't need the journal lock.
1742 * May sleep.
1743 */
1744struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1745{
1746 struct journal_head *jh;
1747 struct journal_head *new_jh = NULL;
1748
1749repeat:
1750 if (!buffer_jbd(bh)) {
1751 new_jh = journal_alloc_journal_head();
1752 memset(new_jh, 0, sizeof(*new_jh));
1753 }
1754
1755 jbd_lock_bh_journal_head(bh);
1756 if (buffer_jbd(bh)) {
1757 jh = bh2jh(bh);
1758 } else {
1759 J_ASSERT_BH(bh,
1760 (atomic_read(&bh->b_count) > 0) ||
1761 (bh->b_page && bh->b_page->mapping));
1762
1763 if (!new_jh) {
1764 jbd_unlock_bh_journal_head(bh);
1765 goto repeat;
1766 }
1767
1768 jh = new_jh;
1769 new_jh = NULL; /* We consumed it */
1770 set_buffer_jbd(bh);
1771 bh->b_private = jh;
1772 jh->b_bh = bh;
1773 get_bh(bh);
1774 BUFFER_TRACE(bh, "added journal_head");
1775 }
1776 jh->b_jcount++;
1777 jbd_unlock_bh_journal_head(bh);
1778 if (new_jh)
1779 journal_free_journal_head(new_jh);
1780 return bh->b_private;
1781}
1782
1783/*
1784 * Grab a ref against this buffer_head's journal_head. If it ended up not
1785 * having a journal_head, return NULL
1786 */
1787struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1788{
1789 struct journal_head *jh = NULL;
1790
1791 jbd_lock_bh_journal_head(bh);
1792 if (buffer_jbd(bh)) {
1793 jh = bh2jh(bh);
1794 jh->b_jcount++;
1795 }
1796 jbd_unlock_bh_journal_head(bh);
1797 return jh;
1798}
1799
1800static void __journal_remove_journal_head(struct buffer_head *bh)
1801{
1802 struct journal_head *jh = bh2jh(bh);
1803
1804 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1805
1806 get_bh(bh);
1807 if (jh->b_jcount == 0) {
1808 if (jh->b_transaction == NULL &&
1809 jh->b_next_transaction == NULL &&
1810 jh->b_cp_transaction == NULL) {
1811 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1812 J_ASSERT_BH(bh, buffer_jbd(bh));
1813 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1814 BUFFER_TRACE(bh, "remove journal_head");
1815 if (jh->b_frozen_data) {
1816 printk(KERN_WARNING "%s: freeing "
1817 "b_frozen_data\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001818 __func__);
Mingming Caoc089d492007-10-16 18:38:25 -04001819 jbd_free(jh->b_frozen_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 }
1821 if (jh->b_committed_data) {
1822 printk(KERN_WARNING "%s: freeing "
1823 "b_committed_data\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -07001824 __func__);
Mingming Caoc089d492007-10-16 18:38:25 -04001825 jbd_free(jh->b_committed_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 }
1827 bh->b_private = NULL;
1828 jh->b_bh = NULL; /* debug, really */
1829 clear_buffer_jbd(bh);
1830 __brelse(bh);
1831 journal_free_journal_head(jh);
1832 } else {
1833 BUFFER_TRACE(bh, "journal_head was locked");
1834 }
1835 }
1836}
1837
1838/*
1839 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1840 * and has a zero b_jcount then remove and release its journal_head. If we did
1841 * see that the buffer is not used by any transaction we also "logically"
1842 * decrement ->b_count.
1843 *
1844 * We in fact take an additional increment on ->b_count as a convenience,
1845 * because the caller usually wants to do additional things with the bh
1846 * after calling here.
1847 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1848 * time. Once the caller has run __brelse(), the buffer is eligible for
1849 * reaping by try_to_free_buffers().
1850 */
1851void journal_remove_journal_head(struct buffer_head *bh)
1852{
1853 jbd_lock_bh_journal_head(bh);
1854 __journal_remove_journal_head(bh);
1855 jbd_unlock_bh_journal_head(bh);
1856}
1857
1858/*
1859 * Drop a reference on the passed journal_head. If it fell to zero then try to
1860 * release the journal_head from the buffer_head.
1861 */
1862void journal_put_journal_head(struct journal_head *jh)
1863{
1864 struct buffer_head *bh = jh2bh(jh);
1865
1866 jbd_lock_bh_journal_head(bh);
1867 J_ASSERT_JH(jh, jh->b_jcount > 0);
1868 --jh->b_jcount;
1869 if (!jh->b_jcount && !jh->b_transaction) {
1870 __journal_remove_journal_head(bh);
1871 __brelse(bh);
1872 }
1873 jbd_unlock_bh_journal_head(bh);
1874}
1875
1876/*
Jose R. Santosc2a91592007-10-18 23:39:22 -07001877 * debugfs tunables
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 */
Jose R. Santosc2a91592007-10-18 23:39:22 -07001879#ifdef CONFIG_JBD_DEBUG
1880
1881u8 journal_enable_debug __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882EXPORT_SYMBOL(journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
Jose R. Santosc2a91592007-10-18 23:39:22 -07001884static struct dentry *jbd_debugfs_dir;
1885static struct dentry *jbd_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Jose R. Santosc2a91592007-10-18 23:39:22 -07001887static void __init jbd_create_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001889 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
1890 if (jbd_debugfs_dir)
1891 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO,
1892 jbd_debugfs_dir,
1893 &journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894}
1895
Jose R. Santosc2a91592007-10-18 23:39:22 -07001896static void __exit jbd_remove_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001898 debugfs_remove(jbd_debug);
1899 debugfs_remove(jbd_debugfs_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900}
1901
1902#else
1903
Jose R. Santosc2a91592007-10-18 23:39:22 -07001904static inline void jbd_create_debugfs_entry(void)
1905{
1906}
1907
1908static inline void jbd_remove_debugfs_entry(void)
1909{
1910}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912#endif
1913
Christoph Lametere18b8902006-12-06 20:33:20 -08001914struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
1916static int __init journal_init_handle_cache(void)
1917{
1918 jbd_handle_cache = kmem_cache_create("journal_handle",
1919 sizeof(handle_t),
1920 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001921 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001922 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 if (jbd_handle_cache == NULL) {
1924 printk(KERN_EMERG "JBD: failed to create handle cache\n");
1925 return -ENOMEM;
1926 }
1927 return 0;
1928}
1929
1930static void journal_destroy_handle_cache(void)
1931{
1932 if (jbd_handle_cache)
1933 kmem_cache_destroy(jbd_handle_cache);
1934}
1935
1936/*
1937 * Module startup and shutdown
1938 */
1939
1940static int __init journal_init_caches(void)
1941{
1942 int ret;
1943
1944 ret = journal_init_revoke_caches();
1945 if (ret == 0)
1946 ret = journal_init_journal_head_cache();
1947 if (ret == 0)
1948 ret = journal_init_handle_cache();
1949 return ret;
1950}
1951
1952static void journal_destroy_caches(void)
1953{
1954 journal_destroy_revoke_caches();
1955 journal_destroy_journal_head_cache();
1956 journal_destroy_handle_cache();
1957}
1958
1959static int __init journal_init(void)
1960{
1961 int ret;
1962
Alexey Dobriyan2aed3482006-09-27 01:49:28 -07001963 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
Adrian Bunk022a4a72005-09-06 15:16:41 -07001964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 ret = journal_init_caches();
1966 if (ret != 0)
1967 journal_destroy_caches();
Jose R. Santosc2a91592007-10-18 23:39:22 -07001968 jbd_create_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 return ret;
1970}
1971
1972static void __exit journal_exit(void)
1973{
1974#ifdef CONFIG_JBD_DEBUG
1975 int n = atomic_read(&nr_journal_heads);
1976 if (n)
1977 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
1978#endif
Jose R. Santosc2a91592007-10-18 23:39:22 -07001979 jbd_remove_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 journal_destroy_caches();
1981}
1982
1983MODULE_LICENSE("GPL");
1984module_init(journal_init);
1985module_exit(journal_exit);
1986