blob: 7f96b5cb67816109e38a51b79a6ee80364e48c5d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/journal.c
3 *
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>
31#include <linux/smp_lock.h>
32#include <linux/init.h>
33#include <linux/mm.h>
34#include <linux/suspend.h>
35#include <linux/pagemap.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080036#include <linux/kthread.h>
37#include <linux/proc_fs.h>
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <asm/uaccess.h>
40#include <asm/page.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42EXPORT_SYMBOL(journal_start);
43EXPORT_SYMBOL(journal_restart);
44EXPORT_SYMBOL(journal_extend);
45EXPORT_SYMBOL(journal_stop);
46EXPORT_SYMBOL(journal_lock_updates);
47EXPORT_SYMBOL(journal_unlock_updates);
48EXPORT_SYMBOL(journal_get_write_access);
49EXPORT_SYMBOL(journal_get_create_access);
50EXPORT_SYMBOL(journal_get_undo_access);
51EXPORT_SYMBOL(journal_dirty_data);
52EXPORT_SYMBOL(journal_dirty_metadata);
53EXPORT_SYMBOL(journal_release_buffer);
54EXPORT_SYMBOL(journal_forget);
55#if 0
56EXPORT_SYMBOL(journal_sync_buffer);
57#endif
58EXPORT_SYMBOL(journal_flush);
59EXPORT_SYMBOL(journal_revoke);
60
61EXPORT_SYMBOL(journal_init_dev);
62EXPORT_SYMBOL(journal_init_inode);
63EXPORT_SYMBOL(journal_update_format);
64EXPORT_SYMBOL(journal_check_used_features);
65EXPORT_SYMBOL(journal_check_available_features);
66EXPORT_SYMBOL(journal_set_features);
67EXPORT_SYMBOL(journal_create);
68EXPORT_SYMBOL(journal_load);
69EXPORT_SYMBOL(journal_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070EXPORT_SYMBOL(journal_update_superblock);
71EXPORT_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)
182 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
212static void journal_start_thread(journal_t *journal)
213{
Andrew Morton8d8c8512006-03-25 03:06:53 -0800214 kthread_run(kjournald, journal, "kjournald");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 wait_event(journal->j_wait_done_commit, journal->j_task != 0);
216}
217
218static void journal_kill_thread(journal_t *journal)
219{
220 spin_lock(&journal->j_state_lock);
221 journal->j_flags |= JFS_UNMOUNT;
222
223 while (journal->j_task) {
224 wake_up(&journal->j_wait_commit);
225 spin_unlock(&journal->j_state_lock);
226 wait_event(journal->j_wait_done_commit, journal->j_task == 0);
227 spin_lock(&journal->j_state_lock);
228 }
229 spin_unlock(&journal->j_state_lock);
230}
231
232/*
233 * journal_write_metadata_buffer: write a metadata buffer to the journal.
234 *
235 * Writes a metadata buffer to a given disk block. The actual IO is not
236 * performed but a new buffer_head is constructed which labels the data
237 * to be written with the correct destination disk block.
238 *
239 * Any magic-number escaping which needs to be done will cause a
240 * copy-out here. If the buffer happens to start with the
241 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
242 * magic number is only written to the log for descripter blocks. In
243 * this case, we copy the data and replace the first word with 0, and we
244 * return a result code which indicates that this buffer needs to be
245 * marked as an escaped buffer in the corresponding log descriptor
246 * block. The missing word can then be restored when the block is read
247 * during recovery.
248 *
249 * If the source buffer has already been modified by a new transaction
250 * since we took the last commit snapshot, we use the frozen copy of
251 * that data for IO. If we end up using the existing buffer_head's data
252 * for the write, then we *have* to lock the buffer to prevent anyone
253 * else from using and possibly modifying it while the IO is in
254 * progress.
255 *
256 * The function returns a pointer to the buffer_heads to be used for IO.
257 *
258 * We assume that the journal has already been locked in this function.
259 *
260 * Return value:
261 * <0: Error
262 * >=0: Finished OK
263 *
264 * On success:
265 * Bit 0 set == escape performed on the data
266 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
267 */
268
269int journal_write_metadata_buffer(transaction_t *transaction,
270 struct journal_head *jh_in,
271 struct journal_head **jh_out,
272 int blocknr)
273{
274 int need_copy_out = 0;
275 int done_copy_out = 0;
276 int do_escape = 0;
277 char *mapped_data;
278 struct buffer_head *new_bh;
279 struct journal_head *new_jh;
280 struct page *new_page;
281 unsigned int new_offset;
282 struct buffer_head *bh_in = jh2bh(jh_in);
283
284 /*
285 * The buffer really shouldn't be locked: only the current committing
286 * transaction is allowed to write it, so nobody else is allowed
287 * to do any IO.
288 *
289 * akpm: except if we're journalling data, and write() output is
290 * also part of a shared mapping, and another thread has
291 * decided to launch a writepage() against this buffer.
292 */
293 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
294
295 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
296
297 /*
298 * If a new transaction has already done a buffer copy-out, then
299 * we use that version of the data for the commit.
300 */
301 jbd_lock_bh_state(bh_in);
302repeat:
303 if (jh_in->b_frozen_data) {
304 done_copy_out = 1;
305 new_page = virt_to_page(jh_in->b_frozen_data);
306 new_offset = offset_in_page(jh_in->b_frozen_data);
307 } else {
308 new_page = jh2bh(jh_in)->b_page;
309 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
310 }
311
312 mapped_data = kmap_atomic(new_page, KM_USER0);
313 /*
314 * Check for escaping
315 */
316 if (*((__be32 *)(mapped_data + new_offset)) ==
317 cpu_to_be32(JFS_MAGIC_NUMBER)) {
318 need_copy_out = 1;
319 do_escape = 1;
320 }
321 kunmap_atomic(mapped_data, KM_USER0);
322
323 /*
324 * Do we need to do a data copy?
325 */
326 if (need_copy_out && !done_copy_out) {
327 char *tmp;
328
329 jbd_unlock_bh_state(bh_in);
330 tmp = jbd_rep_kmalloc(bh_in->b_size, GFP_NOFS);
331 jbd_lock_bh_state(bh_in);
332 if (jh_in->b_frozen_data) {
333 kfree(tmp);
334 goto repeat;
335 }
336
337 jh_in->b_frozen_data = tmp;
338 mapped_data = kmap_atomic(new_page, KM_USER0);
339 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
340 kunmap_atomic(mapped_data, KM_USER0);
341
342 new_page = virt_to_page(tmp);
343 new_offset = offset_in_page(tmp);
344 done_copy_out = 1;
345 }
346
347 /*
348 * Did we need to do an escaping? Now we've done all the
349 * copying, we can finally do so.
350 */
351 if (do_escape) {
352 mapped_data = kmap_atomic(new_page, KM_USER0);
353 *((unsigned int *)(mapped_data + new_offset)) = 0;
354 kunmap_atomic(mapped_data, KM_USER0);
355 }
356
357 /* keep subsequent assertions sane */
358 new_bh->b_state = 0;
359 init_buffer(new_bh, NULL, NULL);
360 atomic_set(&new_bh->b_count, 1);
361 jbd_unlock_bh_state(bh_in);
362
363 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
364
365 set_bh_page(new_bh, new_page, new_offset);
366 new_jh->b_transaction = NULL;
367 new_bh->b_size = jh2bh(jh_in)->b_size;
368 new_bh->b_bdev = transaction->t_journal->j_dev;
369 new_bh->b_blocknr = blocknr;
370 set_buffer_mapped(new_bh);
371 set_buffer_dirty(new_bh);
372
373 *jh_out = new_jh;
374
375 /*
376 * The to-be-written buffer needs to get moved to the io queue,
377 * and the original buffer whose contents we are shadowing or
378 * copying is moved to the transaction's shadow queue.
379 */
380 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
381 journal_file_buffer(jh_in, transaction, BJ_Shadow);
382 JBUFFER_TRACE(new_jh, "file as BJ_IO");
383 journal_file_buffer(new_jh, transaction, BJ_IO);
384
385 return do_escape | (done_copy_out << 1);
386}
387
388/*
389 * Allocation code for the journal file. Manage the space left in the
390 * journal, so that we can begin checkpointing when appropriate.
391 */
392
393/*
394 * __log_space_left: Return the number of free blocks left in the journal.
395 *
396 * Called with the journal already locked.
397 *
398 * Called under j_state_lock
399 */
400
401int __log_space_left(journal_t *journal)
402{
403 int left = journal->j_free;
404
405 assert_spin_locked(&journal->j_state_lock);
406
407 /*
408 * Be pessimistic here about the number of those free blocks which
409 * might be required for log descriptor control blocks.
410 */
411
412#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
413
414 left -= MIN_LOG_RESERVED_BLOCKS;
415
416 if (left <= 0)
417 return 0;
418 left -= (left >> 3);
419 return left;
420}
421
422/*
423 * Called under j_state_lock. Returns true if a transaction was started.
424 */
425int __log_start_commit(journal_t *journal, tid_t target)
426{
427 /*
428 * Are we already doing a recent enough commit?
429 */
430 if (!tid_geq(journal->j_commit_request, target)) {
431 /*
432 * We want a new commit: OK, mark the request and wakup the
433 * commit thread. We do _not_ do the commit ourselves.
434 */
435
436 journal->j_commit_request = target;
437 jbd_debug(1, "JBD: requesting commit %d/%d\n",
438 journal->j_commit_request,
439 journal->j_commit_sequence);
440 wake_up(&journal->j_wait_commit);
441 return 1;
442 }
443 return 0;
444}
445
446int log_start_commit(journal_t *journal, tid_t tid)
447{
448 int ret;
449
450 spin_lock(&journal->j_state_lock);
451 ret = __log_start_commit(journal, tid);
452 spin_unlock(&journal->j_state_lock);
453 return ret;
454}
455
456/*
457 * Force and wait upon a commit if the calling process is not within
458 * transaction. This is used for forcing out undo-protected data which contains
459 * bitmaps, when the fs is running out of space.
460 *
461 * We can only force the running transaction if we don't have an active handle;
462 * otherwise, we will deadlock.
463 *
464 * Returns true if a transaction was started.
465 */
466int journal_force_commit_nested(journal_t *journal)
467{
468 transaction_t *transaction = NULL;
469 tid_t tid;
470
471 spin_lock(&journal->j_state_lock);
472 if (journal->j_running_transaction && !current->journal_info) {
473 transaction = journal->j_running_transaction;
474 __log_start_commit(journal, transaction->t_tid);
475 } else if (journal->j_committing_transaction)
476 transaction = journal->j_committing_transaction;
477
478 if (!transaction) {
479 spin_unlock(&journal->j_state_lock);
480 return 0; /* Nothing to retry */
481 }
482
483 tid = transaction->t_tid;
484 spin_unlock(&journal->j_state_lock);
485 log_wait_commit(journal, tid);
486 return 1;
487}
488
489/*
490 * Start a commit of the current running transaction (if any). Returns true
491 * if a transaction was started, and fills its tid in at *ptid
492 */
493int journal_start_commit(journal_t *journal, tid_t *ptid)
494{
495 int ret = 0;
496
497 spin_lock(&journal->j_state_lock);
498 if (journal->j_running_transaction) {
499 tid_t tid = journal->j_running_transaction->t_tid;
500
501 ret = __log_start_commit(journal, tid);
502 if (ret && ptid)
503 *ptid = tid;
504 } else if (journal->j_committing_transaction && ptid) {
505 /*
506 * If ext3_write_super() recently started a commit, then we
507 * have to wait for completion of that transaction
508 */
509 *ptid = journal->j_committing_transaction->t_tid;
510 ret = 1;
511 }
512 spin_unlock(&journal->j_state_lock);
513 return ret;
514}
515
516/*
517 * Wait for a specified commit to complete.
518 * The caller may not hold the journal lock.
519 */
520int log_wait_commit(journal_t *journal, tid_t tid)
521{
522 int err = 0;
523
524#ifdef CONFIG_JBD_DEBUG
525 spin_lock(&journal->j_state_lock);
526 if (!tid_geq(journal->j_commit_request, tid)) {
527 printk(KERN_EMERG
528 "%s: error: j_commit_request=%d, tid=%d\n",
529 __FUNCTION__, journal->j_commit_request, tid);
530 }
531 spin_unlock(&journal->j_state_lock);
532#endif
533 spin_lock(&journal->j_state_lock);
534 while (tid_gt(tid, journal->j_commit_sequence)) {
535 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
536 tid, journal->j_commit_sequence);
537 wake_up(&journal->j_wait_commit);
538 spin_unlock(&journal->j_state_lock);
539 wait_event(journal->j_wait_done_commit,
540 !tid_gt(tid, journal->j_commit_sequence));
541 spin_lock(&journal->j_state_lock);
542 }
543 spin_unlock(&journal->j_state_lock);
544
545 if (unlikely(is_journal_aborted(journal))) {
546 printk(KERN_EMERG "journal commit I/O error\n");
547 err = -EIO;
548 }
549 return err;
550}
551
552/*
553 * Log buffer allocation routines:
554 */
555
556int journal_next_log_block(journal_t *journal, unsigned long *retp)
557{
558 unsigned long blocknr;
559
560 spin_lock(&journal->j_state_lock);
561 J_ASSERT(journal->j_free > 1);
562
563 blocknr = journal->j_head;
564 journal->j_head++;
565 journal->j_free--;
566 if (journal->j_head == journal->j_last)
567 journal->j_head = journal->j_first;
568 spin_unlock(&journal->j_state_lock);
569 return journal_bmap(journal, blocknr, retp);
570}
571
572/*
573 * Conversion of logical to physical block numbers for the journal
574 *
575 * On external journals the journal blocks are identity-mapped, so
576 * this is a no-op. If needed, we can use j_blk_offset - everything is
577 * ready.
578 */
579int journal_bmap(journal_t *journal, unsigned long blocknr,
580 unsigned long *retp)
581{
582 int err = 0;
583 unsigned long ret;
584
585 if (journal->j_inode) {
586 ret = bmap(journal->j_inode, blocknr);
587 if (ret)
588 *retp = ret;
589 else {
590 char b[BDEVNAME_SIZE];
591
592 printk(KERN_ALERT "%s: journal block not found "
593 "at offset %lu on %s\n",
594 __FUNCTION__,
595 blocknr,
596 bdevname(journal->j_dev, b));
597 err = -EIO;
598 __journal_abort_soft(journal, err);
599 }
600 } else {
601 *retp = blocknr; /* +journal->j_blk_offset */
602 }
603 return err;
604}
605
606/*
607 * We play buffer_head aliasing tricks to write data/metadata blocks to
608 * the journal without copying their contents, but for journal
609 * descriptor blocks we do need to generate bona fide buffers.
610 *
611 * After the caller of journal_get_descriptor_buffer() has finished modifying
612 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
613 * But we don't bother doing that, so there will be coherency problems with
614 * mmaps of blockdevs which hold live JBD-controlled filesystems.
615 */
616struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
617{
618 struct buffer_head *bh;
619 unsigned long blocknr;
620 int err;
621
622 err = journal_next_log_block(journal, &blocknr);
623
624 if (err)
625 return NULL;
626
627 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
628 lock_buffer(bh);
629 memset(bh->b_data, 0, journal->j_blocksize);
630 set_buffer_uptodate(bh);
631 unlock_buffer(bh);
632 BUFFER_TRACE(bh, "return this buffer");
633 return journal_add_journal_head(bh);
634}
635
636/*
637 * Management for journal control blocks: functions to create and
638 * destroy journal_t structures, and to initialise and read existing
639 * journal blocks from disk. */
640
641/* First: create and setup a journal_t object in memory. We initialise
642 * very few fields yet: that has to wait until we have created the
643 * journal structures from from scratch, or loaded them from disk. */
644
645static journal_t * journal_init_common (void)
646{
647 journal_t *journal;
648 int err;
649
650 journal = jbd_kmalloc(sizeof(*journal), GFP_KERNEL);
651 if (!journal)
652 goto fail;
653 memset(journal, 0, sizeof(*journal));
654
655 init_waitqueue_head(&journal->j_wait_transaction_locked);
656 init_waitqueue_head(&journal->j_wait_logspace);
657 init_waitqueue_head(&journal->j_wait_done_commit);
658 init_waitqueue_head(&journal->j_wait_checkpoint);
659 init_waitqueue_head(&journal->j_wait_commit);
660 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800661 mutex_init(&journal->j_barrier);
662 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 spin_lock_init(&journal->j_revoke_lock);
664 spin_lock_init(&journal->j_list_lock);
665 spin_lock_init(&journal->j_state_lock);
666
667 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
668
669 /* The journal is marked for error until we succeed with recovery! */
670 journal->j_flags = JFS_ABORT;
671
672 /* Set up a default-sized revoke table for the new mount. */
673 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
674 if (err) {
675 kfree(journal);
676 goto fail;
677 }
678 return journal;
679fail:
680 return NULL;
681}
682
683/* journal_init_dev and journal_init_inode:
684 *
685 * Create a journal structure assigned some fixed set of disk blocks to
686 * the journal. We don't actually touch those disk blocks yet, but we
687 * need to set up all of the mapping information to tell the journaling
688 * system where the journal blocks are.
689 *
690 */
691
692/**
693 * journal_t * journal_init_dev() - creates an initialises a journal structure
694 * @bdev: Block device on which to create the journal
695 * @fs_dev: Device which hold journalled filesystem for this journal.
696 * @start: Block nr Start of journal.
697 * @len: Lenght of the journal in blocks.
698 * @blocksize: blocksize of journalling device
699 * @returns: a newly created journal_t *
700 *
701 * journal_init_dev creates a journal which maps a fixed contiguous
702 * range of blocks on an arbitrary block device.
703 *
704 */
705journal_t * journal_init_dev(struct block_device *bdev,
706 struct block_device *fs_dev,
707 int start, int len, int blocksize)
708{
709 journal_t *journal = journal_init_common();
710 struct buffer_head *bh;
711 int n;
712
713 if (!journal)
714 return NULL;
715
716 journal->j_dev = bdev;
717 journal->j_fs_dev = fs_dev;
718 journal->j_blk_offset = start;
719 journal->j_maxlen = len;
720 journal->j_blocksize = blocksize;
721
722 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
723 J_ASSERT(bh != NULL);
724 journal->j_sb_buffer = bh;
725 journal->j_superblock = (journal_superblock_t *)bh->b_data;
726
727 /* journal descriptor can store up to n blocks -bzzz */
728 n = journal->j_blocksize / sizeof(journal_block_tag_t);
729 journal->j_wbufsize = n;
730 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
731 if (!journal->j_wbuf) {
732 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
733 __FUNCTION__);
734 kfree(journal);
735 journal = NULL;
736 }
737
738 return journal;
739}
740
741/**
742 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
743 * @inode: An inode to create the journal in
744 *
745 * journal_init_inode creates a journal which maps an on-disk inode as
746 * the journal. The inode must exist already, must support bmap() and
747 * must have all data blocks preallocated.
748 */
749journal_t * journal_init_inode (struct inode *inode)
750{
751 struct buffer_head *bh;
752 journal_t *journal = journal_init_common();
753 int err;
754 int n;
755 unsigned long blocknr;
756
757 if (!journal)
758 return NULL;
759
760 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
761 journal->j_inode = inode;
762 jbd_debug(1,
763 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
764 journal, inode->i_sb->s_id, inode->i_ino,
765 (long long) inode->i_size,
766 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
767
768 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
769 journal->j_blocksize = inode->i_sb->s_blocksize;
770
771 /* journal descriptor can store up to n blocks -bzzz */
772 n = journal->j_blocksize / sizeof(journal_block_tag_t);
773 journal->j_wbufsize = n;
774 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
775 if (!journal->j_wbuf) {
776 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
777 __FUNCTION__);
778 kfree(journal);
779 return NULL;
780 }
781
782 err = journal_bmap(journal, 0, &blocknr);
783 /* If that failed, give up */
784 if (err) {
785 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
786 __FUNCTION__);
787 kfree(journal);
788 return NULL;
789 }
790
791 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
792 J_ASSERT(bh != NULL);
793 journal->j_sb_buffer = bh;
794 journal->j_superblock = (journal_superblock_t *)bh->b_data;
795
796 return journal;
797}
798
799/*
800 * If the journal init or create aborts, we need to mark the journal
801 * superblock as being NULL to prevent the journal destroy from writing
802 * back a bogus superblock.
803 */
804static void journal_fail_superblock (journal_t *journal)
805{
806 struct buffer_head *bh = journal->j_sb_buffer;
807 brelse(bh);
808 journal->j_sb_buffer = NULL;
809}
810
811/*
812 * Given a journal_t structure, initialise the various fields for
813 * startup of a new journaling session. We use this both when creating
814 * a journal, and after recovering an old journal to reset it for
815 * subsequent use.
816 */
817
818static int journal_reset(journal_t *journal)
819{
820 journal_superblock_t *sb = journal->j_superblock;
821 unsigned int first, last;
822
823 first = be32_to_cpu(sb->s_first);
824 last = be32_to_cpu(sb->s_maxlen);
825
826 journal->j_first = first;
827 journal->j_last = last;
828
829 journal->j_head = first;
830 journal->j_tail = first;
831 journal->j_free = last - first;
832
833 journal->j_tail_sequence = journal->j_transaction_sequence;
834 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
835 journal->j_commit_request = journal->j_commit_sequence;
836
837 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
838
839 /* Add the dynamic fields and write it to disk. */
840 journal_update_superblock(journal, 1);
841 journal_start_thread(journal);
842 return 0;
843}
844
845/**
846 * int journal_create() - Initialise the new journal file
847 * @journal: Journal to create. This structure must have been initialised
848 *
849 * Given a journal_t structure which tells us which disk blocks we can
850 * use, create a new journal superblock and initialise all of the
851 * journal fields from scratch.
852 **/
853int journal_create(journal_t *journal)
854{
855 unsigned long blocknr;
856 struct buffer_head *bh;
857 journal_superblock_t *sb;
858 int i, err;
859
860 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
861 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
862 journal->j_maxlen);
863 journal_fail_superblock(journal);
864 return -EINVAL;
865 }
866
867 if (journal->j_inode == NULL) {
868 /*
869 * We don't know what block to start at!
870 */
871 printk(KERN_EMERG
872 "%s: creation of journal on external device!\n",
873 __FUNCTION__);
874 BUG();
875 }
876
877 /* Zero out the entire journal on disk. We cannot afford to
878 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
879 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
880 for (i = 0; i < journal->j_maxlen; i++) {
881 err = journal_bmap(journal, i, &blocknr);
882 if (err)
883 return err;
884 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
885 lock_buffer(bh);
886 memset (bh->b_data, 0, journal->j_blocksize);
887 BUFFER_TRACE(bh, "marking dirty");
888 mark_buffer_dirty(bh);
889 BUFFER_TRACE(bh, "marking uptodate");
890 set_buffer_uptodate(bh);
891 unlock_buffer(bh);
892 __brelse(bh);
893 }
894
895 sync_blockdev(journal->j_dev);
896 jbd_debug(1, "JBD: journal cleared.\n");
897
898 /* OK, fill in the initial static fields in the new superblock */
899 sb = journal->j_superblock;
900
901 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
902 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
903
904 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
905 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
906 sb->s_first = cpu_to_be32(1);
907
908 journal->j_transaction_sequence = 1;
909
910 journal->j_flags &= ~JFS_ABORT;
911 journal->j_format_version = 2;
912
913 return journal_reset(journal);
914}
915
916/**
917 * void journal_update_superblock() - Update journal sb on disk.
918 * @journal: The journal to update.
919 * @wait: Set to '0' if you don't want to wait for IO completion.
920 *
921 * Update a journal's dynamic superblock fields and write it to disk,
922 * optionally waiting for the IO to complete.
923 */
924void journal_update_superblock(journal_t *journal, int wait)
925{
926 journal_superblock_t *sb = journal->j_superblock;
927 struct buffer_head *bh = journal->j_sb_buffer;
928
929 /*
930 * As a special case, if the on-disk copy is already marked as needing
931 * no recovery (s_start == 0) and there are no outstanding transactions
932 * in the filesystem, then we can safely defer the superblock update
933 * until the next commit by setting JFS_FLUSHED. This avoids
934 * attempting a write to a potential-readonly device.
935 */
936 if (sb->s_start == 0 && journal->j_tail_sequence ==
937 journal->j_transaction_sequence) {
938 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
939 "(start %ld, seq %d, errno %d)\n",
940 journal->j_tail, journal->j_tail_sequence,
941 journal->j_errno);
942 goto out;
943 }
944
945 spin_lock(&journal->j_state_lock);
946 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
947 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
948
949 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
950 sb->s_start = cpu_to_be32(journal->j_tail);
951 sb->s_errno = cpu_to_be32(journal->j_errno);
952 spin_unlock(&journal->j_state_lock);
953
954 BUFFER_TRACE(bh, "marking dirty");
955 mark_buffer_dirty(bh);
956 if (wait)
957 sync_dirty_buffer(bh);
958 else
Jan Kara26707692005-09-06 15:19:12 -0700959 ll_rw_block(SWRITE, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
961out:
962 /* If we have just flushed the log (by marking s_start==0), then
963 * any future commit will have to be careful to update the
964 * superblock again to re-record the true start of the log. */
965
966 spin_lock(&journal->j_state_lock);
967 if (sb->s_start)
968 journal->j_flags &= ~JFS_FLUSHED;
969 else
970 journal->j_flags |= JFS_FLUSHED;
971 spin_unlock(&journal->j_state_lock);
972}
973
974/*
975 * Read the superblock for a given journal, performing initial
976 * validation of the format.
977 */
978
979static int journal_get_superblock(journal_t *journal)
980{
981 struct buffer_head *bh;
982 journal_superblock_t *sb;
983 int err = -EIO;
984
985 bh = journal->j_sb_buffer;
986
987 J_ASSERT(bh != NULL);
988 if (!buffer_uptodate(bh)) {
989 ll_rw_block(READ, 1, &bh);
990 wait_on_buffer(bh);
991 if (!buffer_uptodate(bh)) {
992 printk (KERN_ERR
993 "JBD: IO error reading journal superblock\n");
994 goto out;
995 }
996 }
997
998 sb = journal->j_superblock;
999
1000 err = -EINVAL;
1001
1002 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1003 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1004 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1005 goto out;
1006 }
1007
1008 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1009 case JFS_SUPERBLOCK_V1:
1010 journal->j_format_version = 1;
1011 break;
1012 case JFS_SUPERBLOCK_V2:
1013 journal->j_format_version = 2;
1014 break;
1015 default:
1016 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1017 goto out;
1018 }
1019
1020 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1021 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1022 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1023 printk (KERN_WARNING "JBD: journal file too short\n");
1024 goto out;
1025 }
1026
1027 return 0;
1028
1029out:
1030 journal_fail_superblock(journal);
1031 return err;
1032}
1033
1034/*
1035 * Load the on-disk journal superblock and read the key fields into the
1036 * journal_t.
1037 */
1038
1039static int load_superblock(journal_t *journal)
1040{
1041 int err;
1042 journal_superblock_t *sb;
1043
1044 err = journal_get_superblock(journal);
1045 if (err)
1046 return err;
1047
1048 sb = journal->j_superblock;
1049
1050 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1051 journal->j_tail = be32_to_cpu(sb->s_start);
1052 journal->j_first = be32_to_cpu(sb->s_first);
1053 journal->j_last = be32_to_cpu(sb->s_maxlen);
1054 journal->j_errno = be32_to_cpu(sb->s_errno);
1055
1056 return 0;
1057}
1058
1059
1060/**
1061 * int journal_load() - Read journal from disk.
1062 * @journal: Journal to act on.
1063 *
1064 * Given a journal_t structure which tells us which disk blocks contain
1065 * a journal, read the journal from disk to initialise the in-memory
1066 * structures.
1067 */
1068int journal_load(journal_t *journal)
1069{
1070 int err;
1071
1072 err = load_superblock(journal);
1073 if (err)
1074 return err;
1075
1076 /* If this is a V2 superblock, then we have to check the
1077 * features flags on it. */
1078
1079 if (journal->j_format_version >= 2) {
1080 journal_superblock_t *sb = journal->j_superblock;
1081
1082 if ((sb->s_feature_ro_compat &
1083 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1084 (sb->s_feature_incompat &
1085 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1086 printk (KERN_WARNING
1087 "JBD: Unrecognised features on journal\n");
1088 return -EINVAL;
1089 }
1090 }
1091
1092 /* Let the recovery code check whether it needs to recover any
1093 * data from the journal. */
1094 if (journal_recover(journal))
1095 goto recovery_error;
1096
1097 /* OK, we've finished with the dynamic journal bits:
1098 * reinitialise the dynamic contents of the superblock in memory
1099 * and reset them on disk. */
1100 if (journal_reset(journal))
1101 goto recovery_error;
1102
1103 journal->j_flags &= ~JFS_ABORT;
1104 journal->j_flags |= JFS_LOADED;
1105 return 0;
1106
1107recovery_error:
1108 printk (KERN_WARNING "JBD: recovery failed\n");
1109 return -EIO;
1110}
1111
1112/**
1113 * void journal_destroy() - Release a journal_t structure.
1114 * @journal: Journal to act on.
1115 *
1116 * Release a journal_t structure once it is no longer in use by the
1117 * journaled object.
1118 */
1119void journal_destroy(journal_t *journal)
1120{
1121 /* Wait for the commit thread to wake up and die. */
1122 journal_kill_thread(journal);
1123
1124 /* Force a final log commit */
1125 if (journal->j_running_transaction)
1126 journal_commit_transaction(journal);
1127
1128 /* Force any old transactions to disk */
1129
1130 /* Totally anal locking here... */
1131 spin_lock(&journal->j_list_lock);
1132 while (journal->j_checkpoint_transactions != NULL) {
1133 spin_unlock(&journal->j_list_lock);
1134 log_do_checkpoint(journal);
1135 spin_lock(&journal->j_list_lock);
1136 }
1137
1138 J_ASSERT(journal->j_running_transaction == NULL);
1139 J_ASSERT(journal->j_committing_transaction == NULL);
1140 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1141 spin_unlock(&journal->j_list_lock);
1142
1143 /* We can now mark the journal as empty. */
1144 journal->j_tail = 0;
1145 journal->j_tail_sequence = ++journal->j_transaction_sequence;
1146 if (journal->j_sb_buffer) {
1147 journal_update_superblock(journal, 1);
1148 brelse(journal->j_sb_buffer);
1149 }
1150
1151 if (journal->j_inode)
1152 iput(journal->j_inode);
1153 if (journal->j_revoke)
1154 journal_destroy_revoke(journal);
1155 kfree(journal->j_wbuf);
1156 kfree(journal);
1157}
1158
1159
1160/**
1161 *int journal_check_used_features () - Check if features specified are used.
1162 * @journal: Journal to check.
1163 * @compat: bitmask of compatible features
1164 * @ro: bitmask of features that force read-only mount
1165 * @incompat: bitmask of incompatible features
1166 *
1167 * Check whether the journal uses all of a given set of
1168 * features. Return true (non-zero) if it does.
1169 **/
1170
1171int journal_check_used_features (journal_t *journal, unsigned long compat,
1172 unsigned long ro, unsigned long incompat)
1173{
1174 journal_superblock_t *sb;
1175
1176 if (!compat && !ro && !incompat)
1177 return 1;
1178 if (journal->j_format_version == 1)
1179 return 0;
1180
1181 sb = journal->j_superblock;
1182
1183 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1184 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1185 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1186 return 1;
1187
1188 return 0;
1189}
1190
1191/**
1192 * int journal_check_available_features() - Check feature set in journalling layer
1193 * @journal: Journal to check.
1194 * @compat: bitmask of compatible features
1195 * @ro: bitmask of features that force read-only mount
1196 * @incompat: bitmask of incompatible features
1197 *
1198 * Check whether the journaling code supports the use of
1199 * all of a given set of features on this journal. Return true
1200 * (non-zero) if it can. */
1201
1202int journal_check_available_features (journal_t *journal, unsigned long compat,
1203 unsigned long ro, unsigned long incompat)
1204{
1205 journal_superblock_t *sb;
1206
1207 if (!compat && !ro && !incompat)
1208 return 1;
1209
1210 sb = journal->j_superblock;
1211
1212 /* We can support any known requested features iff the
1213 * superblock is in version 2. Otherwise we fail to support any
1214 * extended sb features. */
1215
1216 if (journal->j_format_version != 2)
1217 return 0;
1218
1219 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1220 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1221 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1222 return 1;
1223
1224 return 0;
1225}
1226
1227/**
1228 * int journal_set_features () - Mark a given journal feature in the superblock
1229 * @journal: Journal to act on.
1230 * @compat: bitmask of compatible features
1231 * @ro: bitmask of features that force read-only mount
1232 * @incompat: bitmask of incompatible features
1233 *
1234 * Mark a given journal feature as present on the
1235 * superblock. Returns true if the requested features could be set.
1236 *
1237 */
1238
1239int journal_set_features (journal_t *journal, unsigned long compat,
1240 unsigned long ro, unsigned long incompat)
1241{
1242 journal_superblock_t *sb;
1243
1244 if (journal_check_used_features(journal, compat, ro, incompat))
1245 return 1;
1246
1247 if (!journal_check_available_features(journal, compat, ro, incompat))
1248 return 0;
1249
1250 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1251 compat, ro, incompat);
1252
1253 sb = journal->j_superblock;
1254
1255 sb->s_feature_compat |= cpu_to_be32(compat);
1256 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1257 sb->s_feature_incompat |= cpu_to_be32(incompat);
1258
1259 return 1;
1260}
1261
1262
1263/**
1264 * int journal_update_format () - Update on-disk journal structure.
1265 * @journal: Journal to act on.
1266 *
1267 * Given an initialised but unloaded journal struct, poke about in the
1268 * on-disk structure to update it to the most recent supported version.
1269 */
1270int journal_update_format (journal_t *journal)
1271{
1272 journal_superblock_t *sb;
1273 int err;
1274
1275 err = journal_get_superblock(journal);
1276 if (err)
1277 return err;
1278
1279 sb = journal->j_superblock;
1280
1281 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1282 case JFS_SUPERBLOCK_V2:
1283 return 0;
1284 case JFS_SUPERBLOCK_V1:
1285 return journal_convert_superblock_v1(journal, sb);
1286 default:
1287 break;
1288 }
1289 return -EINVAL;
1290}
1291
1292static int journal_convert_superblock_v1(journal_t *journal,
1293 journal_superblock_t *sb)
1294{
1295 int offset, blocksize;
1296 struct buffer_head *bh;
1297
1298 printk(KERN_WARNING
1299 "JBD: Converting superblock from version 1 to 2.\n");
1300
1301 /* Pre-initialise new fields to zero */
1302 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1303 blocksize = be32_to_cpu(sb->s_blocksize);
1304 memset(&sb->s_feature_compat, 0, blocksize-offset);
1305
1306 sb->s_nr_users = cpu_to_be32(1);
1307 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1308 journal->j_format_version = 2;
1309
1310 bh = journal->j_sb_buffer;
1311 BUFFER_TRACE(bh, "marking dirty");
1312 mark_buffer_dirty(bh);
1313 sync_dirty_buffer(bh);
1314 return 0;
1315}
1316
1317
1318/**
1319 * int journal_flush () - Flush journal
1320 * @journal: Journal to act on.
1321 *
1322 * Flush all data for a given journal to disk and empty the journal.
1323 * Filesystems can use this when remounting readonly to ensure that
1324 * recovery does not need to happen on remount.
1325 */
1326
1327int journal_flush(journal_t *journal)
1328{
1329 int err = 0;
1330 transaction_t *transaction = NULL;
1331 unsigned long old_tail;
1332
1333 spin_lock(&journal->j_state_lock);
1334
1335 /* Force everything buffered to the log... */
1336 if (journal->j_running_transaction) {
1337 transaction = journal->j_running_transaction;
1338 __log_start_commit(journal, transaction->t_tid);
1339 } else if (journal->j_committing_transaction)
1340 transaction = journal->j_committing_transaction;
1341
1342 /* Wait for the log commit to complete... */
1343 if (transaction) {
1344 tid_t tid = transaction->t_tid;
1345
1346 spin_unlock(&journal->j_state_lock);
1347 log_wait_commit(journal, tid);
1348 } else {
1349 spin_unlock(&journal->j_state_lock);
1350 }
1351
1352 /* ...and flush everything in the log out to disk. */
1353 spin_lock(&journal->j_list_lock);
1354 while (!err && journal->j_checkpoint_transactions != NULL) {
1355 spin_unlock(&journal->j_list_lock);
1356 err = log_do_checkpoint(journal);
1357 spin_lock(&journal->j_list_lock);
1358 }
1359 spin_unlock(&journal->j_list_lock);
1360 cleanup_journal_tail(journal);
1361
1362 /* Finally, mark the journal as really needing no recovery.
1363 * This sets s_start==0 in the underlying superblock, which is
1364 * the magic code for a fully-recovered superblock. Any future
1365 * commits of data to the journal will restore the current
1366 * s_start value. */
1367 spin_lock(&journal->j_state_lock);
1368 old_tail = journal->j_tail;
1369 journal->j_tail = 0;
1370 spin_unlock(&journal->j_state_lock);
1371 journal_update_superblock(journal, 1);
1372 spin_lock(&journal->j_state_lock);
1373 journal->j_tail = old_tail;
1374
1375 J_ASSERT(!journal->j_running_transaction);
1376 J_ASSERT(!journal->j_committing_transaction);
1377 J_ASSERT(!journal->j_checkpoint_transactions);
1378 J_ASSERT(journal->j_head == journal->j_tail);
1379 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1380 spin_unlock(&journal->j_state_lock);
1381 return err;
1382}
1383
1384/**
1385 * int journal_wipe() - Wipe journal contents
1386 * @journal: Journal to act on.
1387 * @write: flag (see below)
1388 *
1389 * Wipe out all of the contents of a journal, safely. This will produce
1390 * a warning if the journal contains any valid recovery information.
1391 * Must be called between journal_init_*() and journal_load().
1392 *
1393 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1394 * we merely suppress recovery.
1395 */
1396
1397int journal_wipe(journal_t *journal, int write)
1398{
1399 journal_superblock_t *sb;
1400 int err = 0;
1401
1402 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1403
1404 err = load_superblock(journal);
1405 if (err)
1406 return err;
1407
1408 sb = journal->j_superblock;
1409
1410 if (!journal->j_tail)
1411 goto no_recovery;
1412
1413 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1414 write ? "Clearing" : "Ignoring");
1415
1416 err = journal_skip_recovery(journal);
1417 if (write)
1418 journal_update_superblock(journal, 1);
1419
1420 no_recovery:
1421 return err;
1422}
1423
1424/*
1425 * journal_dev_name: format a character string to describe on what
1426 * device this journal is present.
1427 */
1428
Adrian Bunk022a4a72005-09-06 15:16:41 -07001429static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
1431 struct block_device *bdev;
1432
1433 if (journal->j_inode)
1434 bdev = journal->j_inode->i_sb->s_bdev;
1435 else
1436 bdev = journal->j_dev;
1437
1438 return bdevname(bdev, buffer);
1439}
1440
1441/*
1442 * Journal abort has very specific semantics, which we describe
1443 * for journal abort.
1444 *
1445 * Two internal function, which provide abort to te jbd layer
1446 * itself are here.
1447 */
1448
1449/*
1450 * Quick version for internal journal use (doesn't lock the journal).
1451 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1452 * and don't attempt to make any other journal updates.
1453 */
1454void __journal_abort_hard(journal_t *journal)
1455{
1456 transaction_t *transaction;
1457 char b[BDEVNAME_SIZE];
1458
1459 if (journal->j_flags & JFS_ABORT)
1460 return;
1461
1462 printk(KERN_ERR "Aborting journal on device %s.\n",
1463 journal_dev_name(journal, b));
1464
1465 spin_lock(&journal->j_state_lock);
1466 journal->j_flags |= JFS_ABORT;
1467 transaction = journal->j_running_transaction;
1468 if (transaction)
1469 __log_start_commit(journal, transaction->t_tid);
1470 spin_unlock(&journal->j_state_lock);
1471}
1472
1473/* Soft abort: record the abort error status in the journal superblock,
1474 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001475static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476{
1477 if (journal->j_flags & JFS_ABORT)
1478 return;
1479
1480 if (!journal->j_errno)
1481 journal->j_errno = errno;
1482
1483 __journal_abort_hard(journal);
1484
1485 if (errno)
1486 journal_update_superblock(journal, 1);
1487}
1488
1489/**
1490 * void journal_abort () - Shutdown the journal immediately.
1491 * @journal: the journal to shutdown.
1492 * @errno: an error number to record in the journal indicating
1493 * the reason for the shutdown.
1494 *
1495 * Perform a complete, immediate shutdown of the ENTIRE
1496 * journal (not of a single transaction). This operation cannot be
1497 * undone without closing and reopening the journal.
1498 *
1499 * The journal_abort function is intended to support higher level error
1500 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1501 * mode.
1502 *
1503 * Journal abort has very specific semantics. Any existing dirty,
1504 * unjournaled buffers in the main filesystem will still be written to
1505 * disk by bdflush, but the journaling mechanism will be suspended
1506 * immediately and no further transaction commits will be honoured.
1507 *
1508 * Any dirty, journaled buffers will be written back to disk without
1509 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1510 * filesystem, but we _do_ attempt to leave as much data as possible
1511 * behind for fsck to use for cleanup.
1512 *
1513 * Any attempt to get a new transaction handle on a journal which is in
1514 * ABORT state will just result in an -EROFS error return. A
1515 * journal_stop on an existing handle will return -EIO if we have
1516 * entered abort state during the update.
1517 *
1518 * Recursive transactions are not disturbed by journal abort until the
1519 * final journal_stop, which will receive the -EIO error.
1520 *
1521 * Finally, the journal_abort call allows the caller to supply an errno
1522 * which will be recorded (if possible) in the journal superblock. This
1523 * allows a client to record failure conditions in the middle of a
1524 * transaction without having to complete the transaction to record the
1525 * failure to disk. ext3_error, for example, now uses this
1526 * functionality.
1527 *
1528 * Errors which originate from within the journaling layer will NOT
1529 * supply an errno; a null errno implies that absolutely no further
1530 * writes are done to the journal (unless there are any already in
1531 * progress).
1532 *
1533 */
1534
1535void journal_abort(journal_t *journal, int errno)
1536{
1537 __journal_abort_soft(journal, errno);
1538}
1539
1540/**
1541 * int journal_errno () - returns the journal's error state.
1542 * @journal: journal to examine.
1543 *
1544 * This is the errno numbet set with journal_abort(), the last
1545 * time the journal was mounted - if the journal was stopped
1546 * without calling abort this will be 0.
1547 *
1548 * If the journal has been aborted on this mount time -EROFS will
1549 * be returned.
1550 */
1551int journal_errno(journal_t *journal)
1552{
1553 int err;
1554
1555 spin_lock(&journal->j_state_lock);
1556 if (journal->j_flags & JFS_ABORT)
1557 err = -EROFS;
1558 else
1559 err = journal->j_errno;
1560 spin_unlock(&journal->j_state_lock);
1561 return err;
1562}
1563
1564/**
1565 * int journal_clear_err () - clears the journal's error state
1566 * @journal: journal to act on.
1567 *
1568 * An error must be cleared or Acked to take a FS out of readonly
1569 * mode.
1570 */
1571int journal_clear_err(journal_t *journal)
1572{
1573 int err = 0;
1574
1575 spin_lock(&journal->j_state_lock);
1576 if (journal->j_flags & JFS_ABORT)
1577 err = -EROFS;
1578 else
1579 journal->j_errno = 0;
1580 spin_unlock(&journal->j_state_lock);
1581 return err;
1582}
1583
1584/**
1585 * void journal_ack_err() - Ack journal err.
1586 * @journal: journal to act on.
1587 *
1588 * An error must be cleared or Acked to take a FS out of readonly
1589 * mode.
1590 */
1591void journal_ack_err(journal_t *journal)
1592{
1593 spin_lock(&journal->j_state_lock);
1594 if (journal->j_errno)
1595 journal->j_flags |= JFS_ACK_ERR;
1596 spin_unlock(&journal->j_state_lock);
1597}
1598
1599int journal_blocks_per_page(struct inode *inode)
1600{
1601 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1602}
1603
1604/*
1605 * Simple support for retrying memory allocations. Introduced to help to
1606 * debug different VM deadlock avoidance strategies.
1607 */
Al Viro27496a82005-10-21 03:20:48 -04001608void * __jbd_kmalloc (const char *where, size_t size, gfp_t flags, int retry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609{
1610 return kmalloc(size, flags | (retry ? __GFP_NOFAIL : 0));
1611}
1612
1613/*
1614 * Journal_head storage management
1615 */
1616static kmem_cache_t *journal_head_cache;
1617#ifdef CONFIG_JBD_DEBUG
1618static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1619#endif
1620
1621static int journal_init_journal_head_cache(void)
1622{
1623 int retval;
1624
1625 J_ASSERT(journal_head_cache == 0);
1626 journal_head_cache = kmem_cache_create("journal_head",
1627 sizeof(struct journal_head),
1628 0, /* offset */
1629 0, /* flags */
1630 NULL, /* ctor */
1631 NULL); /* dtor */
1632 retval = 0;
1633 if (journal_head_cache == 0) {
1634 retval = -ENOMEM;
1635 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1636 }
1637 return retval;
1638}
1639
1640static void journal_destroy_journal_head_cache(void)
1641{
1642 J_ASSERT(journal_head_cache != NULL);
1643 kmem_cache_destroy(journal_head_cache);
1644 journal_head_cache = NULL;
1645}
1646
1647/*
1648 * journal_head splicing and dicing
1649 */
1650static struct journal_head *journal_alloc_journal_head(void)
1651{
1652 struct journal_head *ret;
1653 static unsigned long last_warning;
1654
1655#ifdef CONFIG_JBD_DEBUG
1656 atomic_inc(&nr_journal_heads);
1657#endif
1658 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1659 if (ret == 0) {
1660 jbd_debug(1, "out of memory for journal_head\n");
1661 if (time_after(jiffies, last_warning + 5*HZ)) {
1662 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1663 __FUNCTION__);
1664 last_warning = jiffies;
1665 }
1666 while (ret == 0) {
1667 yield();
1668 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1669 }
1670 }
1671 return ret;
1672}
1673
1674static void journal_free_journal_head(struct journal_head *jh)
1675{
1676#ifdef CONFIG_JBD_DEBUG
1677 atomic_dec(&nr_journal_heads);
1678 memset(jh, 0x5b, sizeof(*jh));
1679#endif
1680 kmem_cache_free(journal_head_cache, jh);
1681}
1682
1683/*
1684 * A journal_head is attached to a buffer_head whenever JBD has an
1685 * interest in the buffer.
1686 *
1687 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1688 * is set. This bit is tested in core kernel code where we need to take
1689 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1690 * there.
1691 *
1692 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1693 *
1694 * When a buffer has its BH_JBD bit set it is immune from being released by
1695 * core kernel code, mainly via ->b_count.
1696 *
1697 * A journal_head may be detached from its buffer_head when the journal_head's
1698 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1699 * Various places in JBD call journal_remove_journal_head() to indicate that the
1700 * journal_head can be dropped if needed.
1701 *
1702 * Various places in the kernel want to attach a journal_head to a buffer_head
1703 * _before_ attaching the journal_head to a transaction. To protect the
1704 * journal_head in this situation, journal_add_journal_head elevates the
1705 * journal_head's b_jcount refcount by one. The caller must call
1706 * journal_put_journal_head() to undo this.
1707 *
1708 * So the typical usage would be:
1709 *
1710 * (Attach a journal_head if needed. Increments b_jcount)
1711 * struct journal_head *jh = journal_add_journal_head(bh);
1712 * ...
1713 * jh->b_transaction = xxx;
1714 * journal_put_journal_head(jh);
1715 *
1716 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1717 * because it has a non-zero b_transaction.
1718 */
1719
1720/*
1721 * Give a buffer_head a journal_head.
1722 *
1723 * Doesn't need the journal lock.
1724 * May sleep.
1725 */
1726struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1727{
1728 struct journal_head *jh;
1729 struct journal_head *new_jh = NULL;
1730
1731repeat:
1732 if (!buffer_jbd(bh)) {
1733 new_jh = journal_alloc_journal_head();
1734 memset(new_jh, 0, sizeof(*new_jh));
1735 }
1736
1737 jbd_lock_bh_journal_head(bh);
1738 if (buffer_jbd(bh)) {
1739 jh = bh2jh(bh);
1740 } else {
1741 J_ASSERT_BH(bh,
1742 (atomic_read(&bh->b_count) > 0) ||
1743 (bh->b_page && bh->b_page->mapping));
1744
1745 if (!new_jh) {
1746 jbd_unlock_bh_journal_head(bh);
1747 goto repeat;
1748 }
1749
1750 jh = new_jh;
1751 new_jh = NULL; /* We consumed it */
1752 set_buffer_jbd(bh);
1753 bh->b_private = jh;
1754 jh->b_bh = bh;
1755 get_bh(bh);
1756 BUFFER_TRACE(bh, "added journal_head");
1757 }
1758 jh->b_jcount++;
1759 jbd_unlock_bh_journal_head(bh);
1760 if (new_jh)
1761 journal_free_journal_head(new_jh);
1762 return bh->b_private;
1763}
1764
1765/*
1766 * Grab a ref against this buffer_head's journal_head. If it ended up not
1767 * having a journal_head, return NULL
1768 */
1769struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1770{
1771 struct journal_head *jh = NULL;
1772
1773 jbd_lock_bh_journal_head(bh);
1774 if (buffer_jbd(bh)) {
1775 jh = bh2jh(bh);
1776 jh->b_jcount++;
1777 }
1778 jbd_unlock_bh_journal_head(bh);
1779 return jh;
1780}
1781
1782static void __journal_remove_journal_head(struct buffer_head *bh)
1783{
1784 struct journal_head *jh = bh2jh(bh);
1785
1786 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1787
1788 get_bh(bh);
1789 if (jh->b_jcount == 0) {
1790 if (jh->b_transaction == NULL &&
1791 jh->b_next_transaction == NULL &&
1792 jh->b_cp_transaction == NULL) {
1793 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1794 J_ASSERT_BH(bh, buffer_jbd(bh));
1795 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1796 BUFFER_TRACE(bh, "remove journal_head");
1797 if (jh->b_frozen_data) {
1798 printk(KERN_WARNING "%s: freeing "
1799 "b_frozen_data\n",
1800 __FUNCTION__);
1801 kfree(jh->b_frozen_data);
1802 }
1803 if (jh->b_committed_data) {
1804 printk(KERN_WARNING "%s: freeing "
1805 "b_committed_data\n",
1806 __FUNCTION__);
1807 kfree(jh->b_committed_data);
1808 }
1809 bh->b_private = NULL;
1810 jh->b_bh = NULL; /* debug, really */
1811 clear_buffer_jbd(bh);
1812 __brelse(bh);
1813 journal_free_journal_head(jh);
1814 } else {
1815 BUFFER_TRACE(bh, "journal_head was locked");
1816 }
1817 }
1818}
1819
1820/*
1821 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1822 * and has a zero b_jcount then remove and release its journal_head. If we did
1823 * see that the buffer is not used by any transaction we also "logically"
1824 * decrement ->b_count.
1825 *
1826 * We in fact take an additional increment on ->b_count as a convenience,
1827 * because the caller usually wants to do additional things with the bh
1828 * after calling here.
1829 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1830 * time. Once the caller has run __brelse(), the buffer is eligible for
1831 * reaping by try_to_free_buffers().
1832 */
1833void journal_remove_journal_head(struct buffer_head *bh)
1834{
1835 jbd_lock_bh_journal_head(bh);
1836 __journal_remove_journal_head(bh);
1837 jbd_unlock_bh_journal_head(bh);
1838}
1839
1840/*
1841 * Drop a reference on the passed journal_head. If it fell to zero then try to
1842 * release the journal_head from the buffer_head.
1843 */
1844void journal_put_journal_head(struct journal_head *jh)
1845{
1846 struct buffer_head *bh = jh2bh(jh);
1847
1848 jbd_lock_bh_journal_head(bh);
1849 J_ASSERT_JH(jh, jh->b_jcount > 0);
1850 --jh->b_jcount;
1851 if (!jh->b_jcount && !jh->b_transaction) {
1852 __journal_remove_journal_head(bh);
1853 __brelse(bh);
1854 }
1855 jbd_unlock_bh_journal_head(bh);
1856}
1857
1858/*
1859 * /proc tunables
1860 */
1861#if defined(CONFIG_JBD_DEBUG)
1862int journal_enable_debug;
1863EXPORT_SYMBOL(journal_enable_debug);
1864#endif
1865
1866#if defined(CONFIG_JBD_DEBUG) && defined(CONFIG_PROC_FS)
1867
1868static struct proc_dir_entry *proc_jbd_debug;
1869
Adrian Bunk022a4a72005-09-06 15:16:41 -07001870static int read_jbd_debug(char *page, char **start, off_t off,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 int count, int *eof, void *data)
1872{
1873 int ret;
1874
1875 ret = sprintf(page + off, "%d\n", journal_enable_debug);
1876 *eof = 1;
1877 return ret;
1878}
1879
Adrian Bunk022a4a72005-09-06 15:16:41 -07001880static int write_jbd_debug(struct file *file, const char __user *buffer,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 unsigned long count, void *data)
1882{
1883 char buf[32];
1884
1885 if (count > ARRAY_SIZE(buf) - 1)
1886 count = ARRAY_SIZE(buf) - 1;
1887 if (copy_from_user(buf, buffer, count))
1888 return -EFAULT;
1889 buf[ARRAY_SIZE(buf) - 1] = '\0';
1890 journal_enable_debug = simple_strtoul(buf, NULL, 10);
1891 return count;
1892}
1893
1894#define JBD_PROC_NAME "sys/fs/jbd-debug"
1895
1896static void __init create_jbd_proc_entry(void)
1897{
1898 proc_jbd_debug = create_proc_entry(JBD_PROC_NAME, 0644, NULL);
1899 if (proc_jbd_debug) {
1900 /* Why is this so hard? */
1901 proc_jbd_debug->read_proc = read_jbd_debug;
1902 proc_jbd_debug->write_proc = write_jbd_debug;
1903 }
1904}
1905
1906static void __exit remove_jbd_proc_entry(void)
1907{
1908 if (proc_jbd_debug)
1909 remove_proc_entry(JBD_PROC_NAME, NULL);
1910}
1911
1912#else
1913
1914#define create_jbd_proc_entry() do {} while (0)
1915#define remove_jbd_proc_entry() do {} while (0)
1916
1917#endif
1918
1919kmem_cache_t *jbd_handle_cache;
1920
1921static int __init journal_init_handle_cache(void)
1922{
1923 jbd_handle_cache = kmem_cache_create("journal_handle",
1924 sizeof(handle_t),
1925 0, /* offset */
1926 0, /* flags */
1927 NULL, /* ctor */
1928 NULL); /* dtor */
1929 if (jbd_handle_cache == NULL) {
1930 printk(KERN_EMERG "JBD: failed to create handle cache\n");
1931 return -ENOMEM;
1932 }
1933 return 0;
1934}
1935
1936static void journal_destroy_handle_cache(void)
1937{
1938 if (jbd_handle_cache)
1939 kmem_cache_destroy(jbd_handle_cache);
1940}
1941
1942/*
1943 * Module startup and shutdown
1944 */
1945
1946static int __init journal_init_caches(void)
1947{
1948 int ret;
1949
1950 ret = journal_init_revoke_caches();
1951 if (ret == 0)
1952 ret = journal_init_journal_head_cache();
1953 if (ret == 0)
1954 ret = journal_init_handle_cache();
1955 return ret;
1956}
1957
1958static void journal_destroy_caches(void)
1959{
1960 journal_destroy_revoke_caches();
1961 journal_destroy_journal_head_cache();
1962 journal_destroy_handle_cache();
1963}
1964
1965static int __init journal_init(void)
1966{
1967 int ret;
1968
Adrian Bunk022a4a72005-09-06 15:16:41 -07001969/* Static check for data structure consistency. There's no code
1970 * invoked --- we'll just get a linker failure if things aren't right.
1971 */
1972 extern void journal_bad_superblock_size(void);
1973 if (sizeof(struct journal_superblock_s) != 1024)
1974 journal_bad_superblock_size();
1975
1976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 ret = journal_init_caches();
1978 if (ret != 0)
1979 journal_destroy_caches();
1980 create_jbd_proc_entry();
1981 return ret;
1982}
1983
1984static void __exit journal_exit(void)
1985{
1986#ifdef CONFIG_JBD_DEBUG
1987 int n = atomic_read(&nr_journal_heads);
1988 if (n)
1989 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
1990#endif
1991 remove_jbd_proc_entry();
1992 journal_destroy_caches();
1993}
1994
1995MODULE_LICENSE("GPL");
1996module_init(journal_init);
1997module_exit(journal_exit);
1998