aboutsummaryrefslogtreecommitdiffstats
path: root/fs/jbd2/journal.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/jbd2/journal.c')
-rw-r--r--fs/jbd2/journal.c2072
1 files changed, 2072 insertions, 0 deletions
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
new file mode 100644
index 000000000000..c518dd8fe60a
--- /dev/null
+++ b/fs/jbd2/journal.c
@@ -0,0 +1,2072 @@
1/*
2 * linux/fs/jbd/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>
36#include <linux/kthread.h>
37#include <linux/poison.h>
38#include <linux/proc_fs.h>
39
40#include <asm/uaccess.h>
41#include <asm/page.h>
42
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);
71EXPORT_SYMBOL(journal_update_superblock);
72EXPORT_SYMBOL(journal_abort);
73EXPORT_SYMBOL(journal_errno);
74EXPORT_SYMBOL(journal_ack_err);
75EXPORT_SYMBOL(journal_clear_err);
76EXPORT_SYMBOL(log_wait_commit);
77EXPORT_SYMBOL(journal_start_commit);
78EXPORT_SYMBOL(journal_force_commit_nested);
79EXPORT_SYMBOL(journal_wipe);
80EXPORT_SYMBOL(journal_blocks_per_page);
81EXPORT_SYMBOL(journal_invalidatepage);
82EXPORT_SYMBOL(journal_try_to_free_buffers);
83EXPORT_SYMBOL(journal_force_commit);
84
85static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
86static void __journal_abort_soft (journal_t *journal, int errno);
87static int journal_create_jbd_slab(size_t slab_size);
88
89/*
90 * Helper function used to manage commit timeouts
91 */
92
93static void commit_timeout(unsigned long __data)
94{
95 struct task_struct * p = (struct task_struct *) __data;
96
97 wake_up_process(p);
98}
99
100/*
101 * kjournald: The main thread function used to manage a logging device
102 * journal.
103 *
104 * This kernel thread is responsible for two things:
105 *
106 * 1) COMMIT: Every so often we need to commit the current state of the
107 * filesystem to disk. The journal thread is responsible for writing
108 * all of the metadata buffers to disk.
109 *
110 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
111 * of the data in that part of the log has been rewritten elsewhere on
112 * the disk. Flushing these old buffers to reclaim space in the log is
113 * known as checkpointing, and this thread is responsible for that job.
114 */
115
116static int kjournald(void *arg)
117{
118 journal_t *journal = arg;
119 transaction_t *transaction;
120
121 /*
122 * Set up an interval timer which can be used to trigger a commit wakeup
123 * after the commit interval expires
124 */
125 setup_timer(&journal->j_commit_timer, commit_timeout,
126 (unsigned long)current);
127
128 /* Record that the journal thread is running */
129 journal->j_task = current;
130 wake_up(&journal->j_wait_done_commit);
131
132 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
133 journal->j_commit_interval / HZ);
134
135 /*
136 * And now, wait forever for commit wakeup events.
137 */
138 spin_lock(&journal->j_state_lock);
139
140loop:
141 if (journal->j_flags & JFS_UNMOUNT)
142 goto end_loop;
143
144 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
145 journal->j_commit_sequence, journal->j_commit_request);
146
147 if (journal->j_commit_sequence != journal->j_commit_request) {
148 jbd_debug(1, "OK, requests differ\n");
149 spin_unlock(&journal->j_state_lock);
150 del_timer_sync(&journal->j_commit_timer);
151 journal_commit_transaction(journal);
152 spin_lock(&journal->j_state_lock);
153 goto loop;
154 }
155
156 wake_up(&journal->j_wait_done_commit);
157 if (freezing(current)) {
158 /*
159 * The simpler the better. Flushing journal isn't a
160 * good idea, because that depends on threads that may
161 * be already stopped.
162 */
163 jbd_debug(1, "Now suspending kjournald\n");
164 spin_unlock(&journal->j_state_lock);
165 refrigerator();
166 spin_lock(&journal->j_state_lock);
167 } else {
168 /*
169 * We assume on resume that commits are already there,
170 * so we don't sleep
171 */
172 DEFINE_WAIT(wait);
173 int should_sleep = 1;
174
175 prepare_to_wait(&journal->j_wait_commit, &wait,
176 TASK_INTERRUPTIBLE);
177 if (journal->j_commit_sequence != journal->j_commit_request)
178 should_sleep = 0;
179 transaction = journal->j_running_transaction;
180 if (transaction && time_after_eq(jiffies,
181 transaction->t_expires))
182 should_sleep = 0;
183 if (journal->j_flags & JFS_UNMOUNT)
184 should_sleep = 0;
185 if (should_sleep) {
186 spin_unlock(&journal->j_state_lock);
187 schedule();
188 spin_lock(&journal->j_state_lock);
189 }
190 finish_wait(&journal->j_wait_commit, &wait);
191 }
192
193 jbd_debug(1, "kjournald wakes\n");
194
195 /*
196 * Were we woken up by a commit wakeup event?
197 */
198 transaction = journal->j_running_transaction;
199 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
200 journal->j_commit_request = transaction->t_tid;
201 jbd_debug(1, "woke because of timeout\n");
202 }
203 goto loop;
204
205end_loop:
206 spin_unlock(&journal->j_state_lock);
207 del_timer_sync(&journal->j_commit_timer);
208 journal->j_task = NULL;
209 wake_up(&journal->j_wait_done_commit);
210 jbd_debug(1, "Journal thread exiting.\n");
211 return 0;
212}
213
214static void journal_start_thread(journal_t *journal)
215{
216 kthread_run(kjournald, journal, "kjournald");
217 wait_event(journal->j_wait_done_commit, journal->j_task != 0);
218}
219
220static void journal_kill_thread(journal_t *journal)
221{
222 spin_lock(&journal->j_state_lock);
223 journal->j_flags |= JFS_UNMOUNT;
224
225 while (journal->j_task) {
226 wake_up(&journal->j_wait_commit);
227 spin_unlock(&journal->j_state_lock);
228 wait_event(journal->j_wait_done_commit, journal->j_task == 0);
229 spin_lock(&journal->j_state_lock);
230 }
231 spin_unlock(&journal->j_state_lock);
232}
233
234/*
235 * journal_write_metadata_buffer: write a metadata buffer to the journal.
236 *
237 * Writes a metadata buffer to a given disk block. The actual IO is not
238 * performed but a new buffer_head is constructed which labels the data
239 * to be written with the correct destination disk block.
240 *
241 * Any magic-number escaping which needs to be done will cause a
242 * copy-out here. If the buffer happens to start with the
243 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
244 * magic number is only written to the log for descripter blocks. In
245 * this case, we copy the data and replace the first word with 0, and we
246 * return a result code which indicates that this buffer needs to be
247 * marked as an escaped buffer in the corresponding log descriptor
248 * block. The missing word can then be restored when the block is read
249 * during recovery.
250 *
251 * If the source buffer has already been modified by a new transaction
252 * since we took the last commit snapshot, we use the frozen copy of
253 * that data for IO. If we end up using the existing buffer_head's data
254 * for the write, then we *have* to lock the buffer to prevent anyone
255 * else from using and possibly modifying it while the IO is in
256 * progress.
257 *
258 * The function returns a pointer to the buffer_heads to be used for IO.
259 *
260 * We assume that the journal has already been locked in this function.
261 *
262 * Return value:
263 * <0: Error
264 * >=0: Finished OK
265 *
266 * On success:
267 * Bit 0 set == escape performed on the data
268 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
269 */
270
271int journal_write_metadata_buffer(transaction_t *transaction,
272 struct journal_head *jh_in,
273 struct journal_head **jh_out,
274 unsigned long blocknr)
275{
276 int need_copy_out = 0;
277 int done_copy_out = 0;
278 int do_escape = 0;
279 char *mapped_data;
280 struct buffer_head *new_bh;
281 struct journal_head *new_jh;
282 struct page *new_page;
283 unsigned int new_offset;
284 struct buffer_head *bh_in = jh2bh(jh_in);
285
286 /*
287 * The buffer really shouldn't be locked: only the current committing
288 * transaction is allowed to write it, so nobody else is allowed
289 * to do any IO.
290 *
291 * akpm: except if we're journalling data, and write() output is
292 * also part of a shared mapping, and another thread has
293 * decided to launch a writepage() against this buffer.
294 */
295 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
296
297 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
298
299 /*
300 * If a new transaction has already done a buffer copy-out, then
301 * we use that version of the data for the commit.
302 */
303 jbd_lock_bh_state(bh_in);
304repeat:
305 if (jh_in->b_frozen_data) {
306 done_copy_out = 1;
307 new_page = virt_to_page(jh_in->b_frozen_data);
308 new_offset = offset_in_page(jh_in->b_frozen_data);
309 } else {
310 new_page = jh2bh(jh_in)->b_page;
311 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
312 }
313
314 mapped_data = kmap_atomic(new_page, KM_USER0);
315 /*
316 * Check for escaping
317 */
318 if (*((__be32 *)(mapped_data + new_offset)) ==
319 cpu_to_be32(JFS_MAGIC_NUMBER)) {
320 need_copy_out = 1;
321 do_escape = 1;
322 }
323 kunmap_atomic(mapped_data, KM_USER0);
324
325 /*
326 * Do we need to do a data copy?
327 */
328 if (need_copy_out && !done_copy_out) {
329 char *tmp;
330
331 jbd_unlock_bh_state(bh_in);
332 tmp = jbd_slab_alloc(bh_in->b_size, GFP_NOFS);
333 jbd_lock_bh_state(bh_in);
334 if (jh_in->b_frozen_data) {
335 jbd_slab_free(tmp, bh_in->b_size);
336 goto repeat;
337 }
338
339 jh_in->b_frozen_data = tmp;
340 mapped_data = kmap_atomic(new_page, KM_USER0);
341 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
342 kunmap_atomic(mapped_data, KM_USER0);
343
344 new_page = virt_to_page(tmp);
345 new_offset = offset_in_page(tmp);
346 done_copy_out = 1;
347 }
348
349 /*
350 * Did we need to do an escaping? Now we've done all the
351 * copying, we can finally do so.
352 */
353 if (do_escape) {
354 mapped_data = kmap_atomic(new_page, KM_USER0);
355 *((unsigned int *)(mapped_data + new_offset)) = 0;
356 kunmap_atomic(mapped_data, KM_USER0);
357 }
358
359 /* keep subsequent assertions sane */
360 new_bh->b_state = 0;
361 init_buffer(new_bh, NULL, NULL);
362 atomic_set(&new_bh->b_count, 1);
363 jbd_unlock_bh_state(bh_in);
364
365 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
366
367 set_bh_page(new_bh, new_page, new_offset);
368 new_jh->b_transaction = NULL;
369 new_bh->b_size = jh2bh(jh_in)->b_size;
370 new_bh->b_bdev = transaction->t_journal->j_dev;
371 new_bh->b_blocknr = blocknr;
372 set_buffer_mapped(new_bh);
373 set_buffer_dirty(new_bh);
374
375 *jh_out = new_jh;
376
377 /*
378 * The to-be-written buffer needs to get moved to the io queue,
379 * and the original buffer whose contents we are shadowing or
380 * copying is moved to the transaction's shadow queue.
381 */
382 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
383 journal_file_buffer(jh_in, transaction, BJ_Shadow);
384 JBUFFER_TRACE(new_jh, "file as BJ_IO");
385 journal_file_buffer(new_jh, transaction, BJ_IO);
386
387 return do_escape | (done_copy_out << 1);
388}
389
390/*
391 * Allocation code for the journal file. Manage the space left in the
392 * journal, so that we can begin checkpointing when appropriate.
393 */
394
395/*
396 * __log_space_left: Return the number of free blocks left in the journal.
397 *
398 * Called with the journal already locked.
399 *
400 * Called under j_state_lock
401 */
402
403int __log_space_left(journal_t *journal)
404{
405 int left = journal->j_free;
406
407 assert_spin_locked(&journal->j_state_lock);
408
409 /*
410 * Be pessimistic here about the number of those free blocks which
411 * might be required for log descriptor control blocks.
412 */
413
414#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
415
416 left -= MIN_LOG_RESERVED_BLOCKS;
417
418 if (left <= 0)
419 return 0;
420 left -= (left >> 3);
421 return left;
422}
423
424/*
425 * Called under j_state_lock. Returns true if a transaction was started.
426 */
427int __log_start_commit(journal_t *journal, tid_t target)
428{
429 /*
430 * Are we already doing a recent enough commit?
431 */
432 if (!tid_geq(journal->j_commit_request, target)) {
433 /*
434 * We want a new commit: OK, mark the request and wakup the
435 * commit thread. We do _not_ do the commit ourselves.
436 */
437
438 journal->j_commit_request = target;
439 jbd_debug(1, "JBD: requesting commit %d/%d\n",
440 journal->j_commit_request,
441 journal->j_commit_sequence);
442 wake_up(&journal->j_wait_commit);
443 return 1;
444 }
445 return 0;
446}
447
448int log_start_commit(journal_t *journal, tid_t tid)
449{
450 int ret;
451
452 spin_lock(&journal->j_state_lock);
453 ret = __log_start_commit(journal, tid);
454 spin_unlock(&journal->j_state_lock);
455 return ret;
456}
457
458/*
459 * Force and wait upon a commit if the calling process is not within
460 * transaction. This is used for forcing out undo-protected data which contains
461 * bitmaps, when the fs is running out of space.
462 *
463 * We can only force the running transaction if we don't have an active handle;
464 * otherwise, we will deadlock.
465 *
466 * Returns true if a transaction was started.
467 */
468int journal_force_commit_nested(journal_t *journal)
469{
470 transaction_t *transaction = NULL;
471 tid_t tid;
472
473 spin_lock(&journal->j_state_lock);
474 if (journal->j_running_transaction && !current->journal_info) {
475 transaction = journal->j_running_transaction;
476 __log_start_commit(journal, transaction->t_tid);
477 } else if (journal->j_committing_transaction)
478 transaction = journal->j_committing_transaction;
479
480 if (!transaction) {
481 spin_unlock(&journal->j_state_lock);
482 return 0; /* Nothing to retry */
483 }
484
485 tid = transaction->t_tid;
486 spin_unlock(&journal->j_state_lock);
487 log_wait_commit(journal, tid);
488 return 1;
489}
490
491/*
492 * Start a commit of the current running transaction (if any). Returns true
493 * if a transaction was started, and fills its tid in at *ptid
494 */
495int journal_start_commit(journal_t *journal, tid_t *ptid)
496{
497 int ret = 0;
498
499 spin_lock(&journal->j_state_lock);
500 if (journal->j_running_transaction) {
501 tid_t tid = journal->j_running_transaction->t_tid;
502
503 ret = __log_start_commit(journal, tid);
504 if (ret && ptid)
505 *ptid = tid;
506 } else if (journal->j_committing_transaction && ptid) {
507 /*
508 * If ext3_write_super() recently started a commit, then we
509 * have to wait for completion of that transaction
510 */
511 *ptid = journal->j_committing_transaction->t_tid;
512 ret = 1;
513 }
514 spin_unlock(&journal->j_state_lock);
515 return ret;
516}
517
518/*
519 * Wait for a specified commit to complete.
520 * The caller may not hold the journal lock.
521 */
522int log_wait_commit(journal_t *journal, tid_t tid)
523{
524 int err = 0;
525
526#ifdef CONFIG_JBD_DEBUG
527 spin_lock(&journal->j_state_lock);
528 if (!tid_geq(journal->j_commit_request, tid)) {
529 printk(KERN_EMERG
530 "%s: error: j_commit_request=%d, tid=%d\n",
531 __FUNCTION__, journal->j_commit_request, tid);
532 }
533 spin_unlock(&journal->j_state_lock);
534#endif
535 spin_lock(&journal->j_state_lock);
536 while (tid_gt(tid, journal->j_commit_sequence)) {
537 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
538 tid, journal->j_commit_sequence);
539 wake_up(&journal->j_wait_commit);
540 spin_unlock(&journal->j_state_lock);
541 wait_event(journal->j_wait_done_commit,
542 !tid_gt(tid, journal->j_commit_sequence));
543 spin_lock(&journal->j_state_lock);
544 }
545 spin_unlock(&journal->j_state_lock);
546
547 if (unlikely(is_journal_aborted(journal))) {
548 printk(KERN_EMERG "journal commit I/O error\n");
549 err = -EIO;
550 }
551 return err;
552}
553
554/*
555 * Log buffer allocation routines:
556 */
557
558int journal_next_log_block(journal_t *journal, unsigned long *retp)
559{
560 unsigned long blocknr;
561
562 spin_lock(&journal->j_state_lock);
563 J_ASSERT(journal->j_free > 1);
564
565 blocknr = journal->j_head;
566 journal->j_head++;
567 journal->j_free--;
568 if (journal->j_head == journal->j_last)
569 journal->j_head = journal->j_first;
570 spin_unlock(&journal->j_state_lock);
571 return journal_bmap(journal, blocknr, retp);
572}
573
574/*
575 * Conversion of logical to physical block numbers for the journal
576 *
577 * On external journals the journal blocks are identity-mapped, so
578 * this is a no-op. If needed, we can use j_blk_offset - everything is
579 * ready.
580 */
581int journal_bmap(journal_t *journal, unsigned long blocknr,
582 unsigned long *retp)
583{
584 int err = 0;
585 unsigned long ret;
586
587 if (journal->j_inode) {
588 ret = bmap(journal->j_inode, blocknr);
589 if (ret)
590 *retp = ret;
591 else {
592 char b[BDEVNAME_SIZE];
593
594 printk(KERN_ALERT "%s: journal block not found "
595 "at offset %lu on %s\n",
596 __FUNCTION__,
597 blocknr,
598 bdevname(journal->j_dev, b));
599 err = -EIO;
600 __journal_abort_soft(journal, err);
601 }
602 } else {
603 *retp = blocknr; /* +journal->j_blk_offset */
604 }
605 return err;
606}
607
608/*
609 * We play buffer_head aliasing tricks to write data/metadata blocks to
610 * the journal without copying their contents, but for journal
611 * descriptor blocks we do need to generate bona fide buffers.
612 *
613 * After the caller of journal_get_descriptor_buffer() has finished modifying
614 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
615 * But we don't bother doing that, so there will be coherency problems with
616 * mmaps of blockdevs which hold live JBD-controlled filesystems.
617 */
618struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
619{
620 struct buffer_head *bh;
621 unsigned long blocknr;
622 int err;
623
624 err = journal_next_log_block(journal, &blocknr);
625
626 if (err)
627 return NULL;
628
629 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
630 lock_buffer(bh);
631 memset(bh->b_data, 0, journal->j_blocksize);
632 set_buffer_uptodate(bh);
633 unlock_buffer(bh);
634 BUFFER_TRACE(bh, "return this buffer");
635 return journal_add_journal_head(bh);
636}
637
638/*
639 * Management for journal control blocks: functions to create and
640 * destroy journal_t structures, and to initialise and read existing
641 * journal blocks from disk. */
642
643/* First: create and setup a journal_t object in memory. We initialise
644 * very few fields yet: that has to wait until we have created the
645 * journal structures from from scratch, or loaded them from disk. */
646
647static journal_t * journal_init_common (void)
648{
649 journal_t *journal;
650 int err;
651
652 journal = jbd_kmalloc(sizeof(*journal), GFP_KERNEL);
653 if (!journal)
654 goto fail;
655 memset(journal, 0, sizeof(*journal));
656
657 init_waitqueue_head(&journal->j_wait_transaction_locked);
658 init_waitqueue_head(&journal->j_wait_logspace);
659 init_waitqueue_head(&journal->j_wait_done_commit);
660 init_waitqueue_head(&journal->j_wait_checkpoint);
661 init_waitqueue_head(&journal->j_wait_commit);
662 init_waitqueue_head(&journal->j_wait_updates);
663 mutex_init(&journal->j_barrier);
664 mutex_init(&journal->j_checkpoint_mutex);
665 spin_lock_init(&journal->j_revoke_lock);
666 spin_lock_init(&journal->j_list_lock);
667 spin_lock_init(&journal->j_state_lock);
668
669 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
670
671 /* The journal is marked for error until we succeed with recovery! */
672 journal->j_flags = JFS_ABORT;
673
674 /* Set up a default-sized revoke table for the new mount. */
675 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
676 if (err) {
677 kfree(journal);
678 goto fail;
679 }
680 return journal;
681fail:
682 return NULL;
683}
684
685/* journal_init_dev and journal_init_inode:
686 *
687 * Create a journal structure assigned some fixed set of disk blocks to
688 * the journal. We don't actually touch those disk blocks yet, but we
689 * need to set up all of the mapping information to tell the journaling
690 * system where the journal blocks are.
691 *
692 */
693
694/**
695 * journal_t * journal_init_dev() - creates an initialises a journal structure
696 * @bdev: Block device on which to create the journal
697 * @fs_dev: Device which hold journalled filesystem for this journal.
698 * @start: Block nr Start of journal.
699 * @len: Length of the journal in blocks.
700 * @blocksize: blocksize of journalling device
701 * @returns: a newly created journal_t *
702 *
703 * journal_init_dev creates a journal which maps a fixed contiguous
704 * range of blocks on an arbitrary block device.
705 *
706 */
707journal_t * journal_init_dev(struct block_device *bdev,
708 struct block_device *fs_dev,
709 int start, int len, int blocksize)
710{
711 journal_t *journal = journal_init_common();
712 struct buffer_head *bh;
713 int n;
714
715 if (!journal)
716 return NULL;
717
718 /* journal descriptor can store up to n blocks -bzzz */
719 journal->j_blocksize = blocksize;
720 n = journal->j_blocksize / sizeof(journal_block_tag_t);
721 journal->j_wbufsize = n;
722 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
723 if (!journal->j_wbuf) {
724 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
725 __FUNCTION__);
726 kfree(journal);
727 journal = NULL;
728 }
729 journal->j_dev = bdev;
730 journal->j_fs_dev = fs_dev;
731 journal->j_blk_offset = start;
732 journal->j_maxlen = len;
733
734 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
735 J_ASSERT(bh != NULL);
736 journal->j_sb_buffer = bh;
737 journal->j_superblock = (journal_superblock_t *)bh->b_data;
738
739 return journal;
740}
741
742/**
743 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
744 * @inode: An inode to create the journal in
745 *
746 * journal_init_inode creates a journal which maps an on-disk inode as
747 * the journal. The inode must exist already, must support bmap() and
748 * must have all data blocks preallocated.
749 */
750journal_t * journal_init_inode (struct inode *inode)
751{
752 struct buffer_head *bh;
753 journal_t *journal = journal_init_common();
754 int err;
755 int n;
756 unsigned long blocknr;
757
758 if (!journal)
759 return NULL;
760
761 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
762 journal->j_inode = inode;
763 jbd_debug(1,
764 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
765 journal, inode->i_sb->s_id, inode->i_ino,
766 (long long) inode->i_size,
767 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
768
769 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
770 journal->j_blocksize = inode->i_sb->s_blocksize;
771
772 /* journal descriptor can store up to n blocks -bzzz */
773 n = journal->j_blocksize / sizeof(journal_block_tag_t);
774 journal->j_wbufsize = n;
775 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
776 if (!journal->j_wbuf) {
777 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
778 __FUNCTION__);
779 kfree(journal);
780 return NULL;
781 }
782
783 err = journal_bmap(journal, 0, &blocknr);
784 /* If that failed, give up */
785 if (err) {
786 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
787 __FUNCTION__);
788 kfree(journal);
789 return NULL;
790 }
791
792 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
793 J_ASSERT(bh != NULL);
794 journal->j_sb_buffer = bh;
795 journal->j_superblock = (journal_superblock_t *)bh->b_data;
796
797 return journal;
798}
799
800/*
801 * If the journal init or create aborts, we need to mark the journal
802 * superblock as being NULL to prevent the journal destroy from writing
803 * back a bogus superblock.
804 */
805static void journal_fail_superblock (journal_t *journal)
806{
807 struct buffer_head *bh = journal->j_sb_buffer;
808 brelse(bh);
809 journal->j_sb_buffer = NULL;
810}
811
812/*
813 * Given a journal_t structure, initialise the various fields for
814 * startup of a new journaling session. We use this both when creating
815 * a journal, and after recovering an old journal to reset it for
816 * subsequent use.
817 */
818
819static int journal_reset(journal_t *journal)
820{
821 journal_superblock_t *sb = journal->j_superblock;
822 unsigned long first, last;
823
824 first = be32_to_cpu(sb->s_first);
825 last = be32_to_cpu(sb->s_maxlen);
826
827 journal->j_first = first;
828 journal->j_last = last;
829
830 journal->j_head = first;
831 journal->j_tail = first;
832 journal->j_free = last - first;
833
834 journal->j_tail_sequence = journal->j_transaction_sequence;
835 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
836 journal->j_commit_request = journal->j_commit_sequence;
837
838 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
839
840 /* Add the dynamic fields and write it to disk. */
841 journal_update_superblock(journal, 1);
842 journal_start_thread(journal);
843 return 0;
844}
845
846/**
847 * int journal_create() - Initialise the new journal file
848 * @journal: Journal to create. This structure must have been initialised
849 *
850 * Given a journal_t structure which tells us which disk blocks we can
851 * use, create a new journal superblock and initialise all of the
852 * journal fields from scratch.
853 **/
854int journal_create(journal_t *journal)
855{
856 unsigned long blocknr;
857 struct buffer_head *bh;
858 journal_superblock_t *sb;
859 int i, err;
860
861 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
862 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
863 journal->j_maxlen);
864 journal_fail_superblock(journal);
865 return -EINVAL;
866 }
867
868 if (journal->j_inode == NULL) {
869 /*
870 * We don't know what block to start at!
871 */
872 printk(KERN_EMERG
873 "%s: creation of journal on external device!\n",
874 __FUNCTION__);
875 BUG();
876 }
877
878 /* Zero out the entire journal on disk. We cannot afford to
879 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
880 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
881 for (i = 0; i < journal->j_maxlen; i++) {
882 err = journal_bmap(journal, i, &blocknr);
883 if (err)
884 return err;
885 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
886 lock_buffer(bh);
887 memset (bh->b_data, 0, journal->j_blocksize);
888 BUFFER_TRACE(bh, "marking dirty");
889 mark_buffer_dirty(bh);
890 BUFFER_TRACE(bh, "marking uptodate");
891 set_buffer_uptodate(bh);
892 unlock_buffer(bh);
893 __brelse(bh);
894 }
895
896 sync_blockdev(journal->j_dev);
897 jbd_debug(1, "JBD: journal cleared.\n");
898
899 /* OK, fill in the initial static fields in the new superblock */
900 sb = journal->j_superblock;
901
902 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
903 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
904
905 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
906 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
907 sb->s_first = cpu_to_be32(1);
908
909 journal->j_transaction_sequence = 1;
910
911 journal->j_flags &= ~JFS_ABORT;
912 journal->j_format_version = 2;
913
914 return journal_reset(journal);
915}
916
917/**
918 * void journal_update_superblock() - Update journal sb on disk.
919 * @journal: The journal to update.
920 * @wait: Set to '0' if you don't want to wait for IO completion.
921 *
922 * Update a journal's dynamic superblock fields and write it to disk,
923 * optionally waiting for the IO to complete.
924 */
925void journal_update_superblock(journal_t *journal, int wait)
926{
927 journal_superblock_t *sb = journal->j_superblock;
928 struct buffer_head *bh = journal->j_sb_buffer;
929
930 /*
931 * As a special case, if the on-disk copy is already marked as needing
932 * no recovery (s_start == 0) and there are no outstanding transactions
933 * in the filesystem, then we can safely defer the superblock update
934 * until the next commit by setting JFS_FLUSHED. This avoids
935 * attempting a write to a potential-readonly device.
936 */
937 if (sb->s_start == 0 && journal->j_tail_sequence ==
938 journal->j_transaction_sequence) {
939 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
940 "(start %ld, seq %d, errno %d)\n",
941 journal->j_tail, journal->j_tail_sequence,
942 journal->j_errno);
943 goto out;
944 }
945
946 spin_lock(&journal->j_state_lock);
947 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
948 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
949
950 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
951 sb->s_start = cpu_to_be32(journal->j_tail);
952 sb->s_errno = cpu_to_be32(journal->j_errno);
953 spin_unlock(&journal->j_state_lock);
954
955 BUFFER_TRACE(bh, "marking dirty");
956 mark_buffer_dirty(bh);
957 if (wait)
958 sync_dirty_buffer(bh);
959 else
960 ll_rw_block(SWRITE, 1, &bh);
961
962out:
963 /* If we have just flushed the log (by marking s_start==0), then
964 * any future commit will have to be careful to update the
965 * superblock again to re-record the true start of the log. */
966
967 spin_lock(&journal->j_state_lock);
968 if (sb->s_start)
969 journal->j_flags &= ~JFS_FLUSHED;
970 else
971 journal->j_flags |= JFS_FLUSHED;
972 spin_unlock(&journal->j_state_lock);
973}
974
975/*
976 * Read the superblock for a given journal, performing initial
977 * validation of the format.
978 */
979
980static int journal_get_superblock(journal_t *journal)
981{
982 struct buffer_head *bh;
983 journal_superblock_t *sb;
984 int err = -EIO;
985
986 bh = journal->j_sb_buffer;
987
988 J_ASSERT(bh != NULL);
989 if (!buffer_uptodate(bh)) {
990 ll_rw_block(READ, 1, &bh);
991 wait_on_buffer(bh);
992 if (!buffer_uptodate(bh)) {
993 printk (KERN_ERR
994 "JBD: IO error reading journal superblock\n");
995 goto out;
996 }
997 }
998
999 sb = journal->j_superblock;
1000
1001 err = -EINVAL;
1002
1003 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1004 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1005 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1006 goto out;
1007 }
1008
1009 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1010 case JFS_SUPERBLOCK_V1:
1011 journal->j_format_version = 1;
1012 break;
1013 case JFS_SUPERBLOCK_V2:
1014 journal->j_format_version = 2;
1015 break;
1016 default:
1017 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1018 goto out;
1019 }
1020
1021 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1022 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1023 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1024 printk (KERN_WARNING "JBD: journal file too short\n");
1025 goto out;
1026 }
1027
1028 return 0;
1029
1030out:
1031 journal_fail_superblock(journal);
1032 return err;
1033}
1034
1035/*
1036 * Load the on-disk journal superblock and read the key fields into the
1037 * journal_t.
1038 */
1039
1040static int load_superblock(journal_t *journal)
1041{
1042 int err;
1043 journal_superblock_t *sb;
1044
1045 err = journal_get_superblock(journal);
1046 if (err)
1047 return err;
1048
1049 sb = journal->j_superblock;
1050
1051 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1052 journal->j_tail = be32_to_cpu(sb->s_start);
1053 journal->j_first = be32_to_cpu(sb->s_first);
1054 journal->j_last = be32_to_cpu(sb->s_maxlen);
1055 journal->j_errno = be32_to_cpu(sb->s_errno);
1056
1057 return 0;
1058}
1059
1060
1061/**
1062 * int journal_load() - Read journal from disk.
1063 * @journal: Journal to act on.
1064 *
1065 * Given a journal_t structure which tells us which disk blocks contain
1066 * a journal, read the journal from disk to initialise the in-memory
1067 * structures.
1068 */
1069int journal_load(journal_t *journal)
1070{
1071 int err;
1072 journal_superblock_t *sb;
1073
1074 err = load_superblock(journal);
1075 if (err)
1076 return err;
1077
1078 sb = journal->j_superblock;
1079 /* If this is a V2 superblock, then we have to check the
1080 * features flags on it. */
1081
1082 if (journal->j_format_version >= 2) {
1083 if ((sb->s_feature_ro_compat &
1084 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1085 (sb->s_feature_incompat &
1086 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1087 printk (KERN_WARNING
1088 "JBD: Unrecognised features on journal\n");
1089 return -EINVAL;
1090 }
1091 }
1092
1093 /*
1094 * Create a slab for this blocksize
1095 */
1096 err = journal_create_jbd_slab(be32_to_cpu(sb->s_blocksize));
1097 if (err)
1098 return err;
1099
1100 /* Let the recovery code check whether it needs to recover any
1101 * data from the journal. */
1102 if (journal_recover(journal))
1103 goto recovery_error;
1104
1105 /* OK, we've finished with the dynamic journal bits:
1106 * reinitialise the dynamic contents of the superblock in memory
1107 * and reset them on disk. */
1108 if (journal_reset(journal))
1109 goto recovery_error;
1110
1111 journal->j_flags &= ~JFS_ABORT;
1112 journal->j_flags |= JFS_LOADED;
1113 return 0;
1114
1115recovery_error:
1116 printk (KERN_WARNING "JBD: recovery failed\n");
1117 return -EIO;
1118}
1119
1120/**
1121 * void journal_destroy() - Release a journal_t structure.
1122 * @journal: Journal to act on.
1123 *
1124 * Release a journal_t structure once it is no longer in use by the
1125 * journaled object.
1126 */
1127void journal_destroy(journal_t *journal)
1128{
1129 /* Wait for the commit thread to wake up and die. */
1130 journal_kill_thread(journal);
1131
1132 /* Force a final log commit */
1133 if (journal->j_running_transaction)
1134 journal_commit_transaction(journal);
1135
1136 /* Force any old transactions to disk */
1137
1138 /* Totally anal locking here... */
1139 spin_lock(&journal->j_list_lock);
1140 while (journal->j_checkpoint_transactions != NULL) {
1141 spin_unlock(&journal->j_list_lock);
1142 log_do_checkpoint(journal);
1143 spin_lock(&journal->j_list_lock);
1144 }
1145
1146 J_ASSERT(journal->j_running_transaction == NULL);
1147 J_ASSERT(journal->j_committing_transaction == NULL);
1148 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1149 spin_unlock(&journal->j_list_lock);
1150
1151 /* We can now mark the journal as empty. */
1152 journal->j_tail = 0;
1153 journal->j_tail_sequence = ++journal->j_transaction_sequence;
1154 if (journal->j_sb_buffer) {
1155 journal_update_superblock(journal, 1);
1156 brelse(journal->j_sb_buffer);
1157 }
1158
1159 if (journal->j_inode)
1160 iput(journal->j_inode);
1161 if (journal->j_revoke)
1162 journal_destroy_revoke(journal);
1163 kfree(journal->j_wbuf);
1164 kfree(journal);
1165}
1166
1167
1168/**
1169 *int journal_check_used_features () - Check if features specified are used.
1170 * @journal: Journal to check.
1171 * @compat: bitmask of compatible features
1172 * @ro: bitmask of features that force read-only mount
1173 * @incompat: bitmask of incompatible features
1174 *
1175 * Check whether the journal uses all of a given set of
1176 * features. Return true (non-zero) if it does.
1177 **/
1178
1179int journal_check_used_features (journal_t *journal, unsigned long compat,
1180 unsigned long ro, unsigned long incompat)
1181{
1182 journal_superblock_t *sb;
1183
1184 if (!compat && !ro && !incompat)
1185 return 1;
1186 if (journal->j_format_version == 1)
1187 return 0;
1188
1189 sb = journal->j_superblock;
1190
1191 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1192 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1193 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1194 return 1;
1195
1196 return 0;
1197}
1198
1199/**
1200 * int journal_check_available_features() - Check feature set in journalling layer
1201 * @journal: Journal to check.
1202 * @compat: bitmask of compatible features
1203 * @ro: bitmask of features that force read-only mount
1204 * @incompat: bitmask of incompatible features
1205 *
1206 * Check whether the journaling code supports the use of
1207 * all of a given set of features on this journal. Return true
1208 * (non-zero) if it can. */
1209
1210int journal_check_available_features (journal_t *journal, unsigned long compat,
1211 unsigned long ro, unsigned long incompat)
1212{
1213 journal_superblock_t *sb;
1214
1215 if (!compat && !ro && !incompat)
1216 return 1;
1217
1218 sb = journal->j_superblock;
1219
1220 /* We can support any known requested features iff the
1221 * superblock is in version 2. Otherwise we fail to support any
1222 * extended sb features. */
1223
1224 if (journal->j_format_version != 2)
1225 return 0;
1226
1227 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1228 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1229 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1230 return 1;
1231
1232 return 0;
1233}
1234
1235/**
1236 * int journal_set_features () - Mark a given journal feature in the superblock
1237 * @journal: Journal to act on.
1238 * @compat: bitmask of compatible features
1239 * @ro: bitmask of features that force read-only mount
1240 * @incompat: bitmask of incompatible features
1241 *
1242 * Mark a given journal feature as present on the
1243 * superblock. Returns true if the requested features could be set.
1244 *
1245 */
1246
1247int journal_set_features (journal_t *journal, unsigned long compat,
1248 unsigned long ro, unsigned long incompat)
1249{
1250 journal_superblock_t *sb;
1251
1252 if (journal_check_used_features(journal, compat, ro, incompat))
1253 return 1;
1254
1255 if (!journal_check_available_features(journal, compat, ro, incompat))
1256 return 0;
1257
1258 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1259 compat, ro, incompat);
1260
1261 sb = journal->j_superblock;
1262
1263 sb->s_feature_compat |= cpu_to_be32(compat);
1264 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1265 sb->s_feature_incompat |= cpu_to_be32(incompat);
1266
1267 return 1;
1268}
1269
1270
1271/**
1272 * int journal_update_format () - Update on-disk journal structure.
1273 * @journal: Journal to act on.
1274 *
1275 * Given an initialised but unloaded journal struct, poke about in the
1276 * on-disk structure to update it to the most recent supported version.
1277 */
1278int journal_update_format (journal_t *journal)
1279{
1280 journal_superblock_t *sb;
1281 int err;
1282
1283 err = journal_get_superblock(journal);
1284 if (err)
1285 return err;
1286
1287 sb = journal->j_superblock;
1288
1289 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1290 case JFS_SUPERBLOCK_V2:
1291 return 0;
1292 case JFS_SUPERBLOCK_V1:
1293 return journal_convert_superblock_v1(journal, sb);
1294 default:
1295 break;
1296 }
1297 return -EINVAL;
1298}
1299
1300static int journal_convert_superblock_v1(journal_t *journal,
1301 journal_superblock_t *sb)
1302{
1303 int offset, blocksize;
1304 struct buffer_head *bh;
1305
1306 printk(KERN_WARNING
1307 "JBD: Converting superblock from version 1 to 2.\n");
1308
1309 /* Pre-initialise new fields to zero */
1310 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1311 blocksize = be32_to_cpu(sb->s_blocksize);
1312 memset(&sb->s_feature_compat, 0, blocksize-offset);
1313
1314 sb->s_nr_users = cpu_to_be32(1);
1315 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1316 journal->j_format_version = 2;
1317
1318 bh = journal->j_sb_buffer;
1319 BUFFER_TRACE(bh, "marking dirty");
1320 mark_buffer_dirty(bh);
1321 sync_dirty_buffer(bh);
1322 return 0;
1323}
1324
1325
1326/**
1327 * int journal_flush () - Flush journal
1328 * @journal: Journal to act on.
1329 *
1330 * Flush all data for a given journal to disk and empty the journal.
1331 * Filesystems can use this when remounting readonly to ensure that
1332 * recovery does not need to happen on remount.
1333 */
1334
1335int journal_flush(journal_t *journal)
1336{
1337 int err = 0;
1338 transaction_t *transaction = NULL;
1339 unsigned long old_tail;
1340
1341 spin_lock(&journal->j_state_lock);
1342
1343 /* Force everything buffered to the log... */
1344 if (journal->j_running_transaction) {
1345 transaction = journal->j_running_transaction;
1346 __log_start_commit(journal, transaction->t_tid);
1347 } else if (journal->j_committing_transaction)
1348 transaction = journal->j_committing_transaction;
1349
1350 /* Wait for the log commit to complete... */
1351 if (transaction) {
1352 tid_t tid = transaction->t_tid;
1353
1354 spin_unlock(&journal->j_state_lock);
1355 log_wait_commit(journal, tid);
1356 } else {
1357 spin_unlock(&journal->j_state_lock);
1358 }
1359
1360 /* ...and flush everything in the log out to disk. */
1361 spin_lock(&journal->j_list_lock);
1362 while (!err && journal->j_checkpoint_transactions != NULL) {
1363 spin_unlock(&journal->j_list_lock);
1364 err = log_do_checkpoint(journal);
1365 spin_lock(&journal->j_list_lock);
1366 }
1367 spin_unlock(&journal->j_list_lock);
1368 cleanup_journal_tail(journal);
1369
1370 /* Finally, mark the journal as really needing no recovery.
1371 * This sets s_start==0 in the underlying superblock, which is
1372 * the magic code for a fully-recovered superblock. Any future
1373 * commits of data to the journal will restore the current
1374 * s_start value. */
1375 spin_lock(&journal->j_state_lock);
1376 old_tail = journal->j_tail;
1377 journal->j_tail = 0;
1378 spin_unlock(&journal->j_state_lock);
1379 journal_update_superblock(journal, 1);
1380 spin_lock(&journal->j_state_lock);
1381 journal->j_tail = old_tail;
1382
1383 J_ASSERT(!journal->j_running_transaction);
1384 J_ASSERT(!journal->j_committing_transaction);
1385 J_ASSERT(!journal->j_checkpoint_transactions);
1386 J_ASSERT(journal->j_head == journal->j_tail);
1387 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1388 spin_unlock(&journal->j_state_lock);
1389 return err;
1390}
1391
1392/**
1393 * int journal_wipe() - Wipe journal contents
1394 * @journal: Journal to act on.
1395 * @write: flag (see below)
1396 *
1397 * Wipe out all of the contents of a journal, safely. This will produce
1398 * a warning if the journal contains any valid recovery information.
1399 * Must be called between journal_init_*() and journal_load().
1400 *
1401 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1402 * we merely suppress recovery.
1403 */
1404
1405int journal_wipe(journal_t *journal, int write)
1406{
1407 journal_superblock_t *sb;
1408 int err = 0;
1409
1410 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1411
1412 err = load_superblock(journal);
1413 if (err)
1414 return err;
1415
1416 sb = journal->j_superblock;
1417
1418 if (!journal->j_tail)
1419 goto no_recovery;
1420
1421 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1422 write ? "Clearing" : "Ignoring");
1423
1424 err = journal_skip_recovery(journal);
1425 if (write)
1426 journal_update_superblock(journal, 1);
1427
1428 no_recovery:
1429 return err;
1430}
1431
1432/*
1433 * journal_dev_name: format a character string to describe on what
1434 * device this journal is present.
1435 */
1436
1437static const char *journal_dev_name(journal_t *journal, char *buffer)
1438{
1439 struct block_device *bdev;
1440
1441 if (journal->j_inode)
1442 bdev = journal->j_inode->i_sb->s_bdev;
1443 else
1444 bdev = journal->j_dev;
1445
1446 return bdevname(bdev, buffer);
1447}
1448
1449/*
1450 * Journal abort has very specific semantics, which we describe
1451 * for journal abort.
1452 *
1453 * Two internal function, which provide abort to te jbd layer
1454 * itself are here.
1455 */
1456
1457/*
1458 * Quick version for internal journal use (doesn't lock the journal).
1459 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1460 * and don't attempt to make any other journal updates.
1461 */
1462void __journal_abort_hard(journal_t *journal)
1463{
1464 transaction_t *transaction;
1465 char b[BDEVNAME_SIZE];
1466
1467 if (journal->j_flags & JFS_ABORT)
1468 return;
1469
1470 printk(KERN_ERR "Aborting journal on device %s.\n",
1471 journal_dev_name(journal, b));
1472
1473 spin_lock(&journal->j_state_lock);
1474 journal->j_flags |= JFS_ABORT;
1475 transaction = journal->j_running_transaction;
1476 if (transaction)
1477 __log_start_commit(journal, transaction->t_tid);
1478 spin_unlock(&journal->j_state_lock);
1479}
1480
1481/* Soft abort: record the abort error status in the journal superblock,
1482 * but don't do any other IO. */
1483static void __journal_abort_soft (journal_t *journal, int errno)
1484{
1485 if (journal->j_flags & JFS_ABORT)
1486 return;
1487
1488 if (!journal->j_errno)
1489 journal->j_errno = errno;
1490
1491 __journal_abort_hard(journal);
1492
1493 if (errno)
1494 journal_update_superblock(journal, 1);
1495}
1496
1497/**
1498 * void journal_abort () - Shutdown the journal immediately.
1499 * @journal: the journal to shutdown.
1500 * @errno: an error number to record in the journal indicating
1501 * the reason for the shutdown.
1502 *
1503 * Perform a complete, immediate shutdown of the ENTIRE
1504 * journal (not of a single transaction). This operation cannot be
1505 * undone without closing and reopening the journal.
1506 *
1507 * The journal_abort function is intended to support higher level error
1508 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1509 * mode.
1510 *
1511 * Journal abort has very specific semantics. Any existing dirty,
1512 * unjournaled buffers in the main filesystem will still be written to
1513 * disk by bdflush, but the journaling mechanism will be suspended
1514 * immediately and no further transaction commits will be honoured.
1515 *
1516 * Any dirty, journaled buffers will be written back to disk without
1517 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1518 * filesystem, but we _do_ attempt to leave as much data as possible
1519 * behind for fsck to use for cleanup.
1520 *
1521 * Any attempt to get a new transaction handle on a journal which is in
1522 * ABORT state will just result in an -EROFS error return. A
1523 * journal_stop on an existing handle will return -EIO if we have
1524 * entered abort state during the update.
1525 *
1526 * Recursive transactions are not disturbed by journal abort until the
1527 * final journal_stop, which will receive the -EIO error.
1528 *
1529 * Finally, the journal_abort call allows the caller to supply an errno
1530 * which will be recorded (if possible) in the journal superblock. This
1531 * allows a client to record failure conditions in the middle of a
1532 * transaction without having to complete the transaction to record the
1533 * failure to disk. ext3_error, for example, now uses this
1534 * functionality.
1535 *
1536 * Errors which originate from within the journaling layer will NOT
1537 * supply an errno; a null errno implies that absolutely no further
1538 * writes are done to the journal (unless there are any already in
1539 * progress).
1540 *
1541 */
1542
1543void journal_abort(journal_t *journal, int errno)
1544{
1545 __journal_abort_soft(journal, errno);
1546}
1547
1548/**
1549 * int journal_errno () - returns the journal's error state.
1550 * @journal: journal to examine.
1551 *
1552 * This is the errno numbet set with journal_abort(), the last
1553 * time the journal was mounted - if the journal was stopped
1554 * without calling abort this will be 0.
1555 *
1556 * If the journal has been aborted on this mount time -EROFS will
1557 * be returned.
1558 */
1559int journal_errno(journal_t *journal)
1560{
1561 int err;
1562
1563 spin_lock(&journal->j_state_lock);
1564 if (journal->j_flags & JFS_ABORT)
1565 err = -EROFS;
1566 else
1567 err = journal->j_errno;
1568 spin_unlock(&journal->j_state_lock);
1569 return err;
1570}
1571
1572/**
1573 * int journal_clear_err () - clears the journal's error state
1574 * @journal: journal to act on.
1575 *
1576 * An error must be cleared or Acked to take a FS out of readonly
1577 * mode.
1578 */
1579int journal_clear_err(journal_t *journal)
1580{
1581 int err = 0;
1582
1583 spin_lock(&journal->j_state_lock);
1584 if (journal->j_flags & JFS_ABORT)
1585 err = -EROFS;
1586 else
1587 journal->j_errno = 0;
1588 spin_unlock(&journal->j_state_lock);
1589 return err;
1590}
1591
1592/**
1593 * void journal_ack_err() - Ack journal err.
1594 * @journal: journal to act on.
1595 *
1596 * An error must be cleared or Acked to take a FS out of readonly
1597 * mode.
1598 */
1599void journal_ack_err(journal_t *journal)
1600{
1601 spin_lock(&journal->j_state_lock);
1602 if (journal->j_errno)
1603 journal->j_flags |= JFS_ACK_ERR;
1604 spin_unlock(&journal->j_state_lock);
1605}
1606
1607int journal_blocks_per_page(struct inode *inode)
1608{
1609 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1610}
1611
1612/*
1613 * Simple support for retrying memory allocations. Introduced to help to
1614 * debug different VM deadlock avoidance strategies.
1615 */
1616void * __jbd_kmalloc (const char *where, size_t size, gfp_t flags, int retry)
1617{
1618 return kmalloc(size, flags | (retry ? __GFP_NOFAIL : 0));
1619}
1620
1621/*
1622 * jbd slab management: create 1k, 2k, 4k, 8k slabs as needed
1623 * and allocate frozen and commit buffers from these slabs.
1624 *
1625 * Reason for doing this is to avoid, SLAB_DEBUG - since it could
1626 * cause bh to cross page boundary.
1627 */
1628
1629#define JBD_MAX_SLABS 5
1630#define JBD_SLAB_INDEX(size) (size >> 11)
1631
1632static kmem_cache_t *jbd_slab[JBD_MAX_SLABS];
1633static const char *jbd_slab_names[JBD_MAX_SLABS] = {
1634 "jbd_1k", "jbd_2k", "jbd_4k", NULL, "jbd_8k"
1635};
1636
1637static void journal_destroy_jbd_slabs(void)
1638{
1639 int i;
1640
1641 for (i = 0; i < JBD_MAX_SLABS; i++) {
1642 if (jbd_slab[i])
1643 kmem_cache_destroy(jbd_slab[i]);
1644 jbd_slab[i] = NULL;
1645 }
1646}
1647
1648static int journal_create_jbd_slab(size_t slab_size)
1649{
1650 int i = JBD_SLAB_INDEX(slab_size);
1651
1652 BUG_ON(i >= JBD_MAX_SLABS);
1653
1654 /*
1655 * Check if we already have a slab created for this size
1656 */
1657 if (jbd_slab[i])
1658 return 0;
1659
1660 /*
1661 * Create a slab and force alignment to be same as slabsize -
1662 * this will make sure that allocations won't cross the page
1663 * boundary.
1664 */
1665 jbd_slab[i] = kmem_cache_create(jbd_slab_names[i],
1666 slab_size, slab_size, 0, NULL, NULL);
1667 if (!jbd_slab[i]) {
1668 printk(KERN_EMERG "JBD: no memory for jbd_slab cache\n");
1669 return -ENOMEM;
1670 }
1671 return 0;
1672}
1673
1674void * jbd_slab_alloc(size_t size, gfp_t flags)
1675{
1676 int idx;
1677
1678 idx = JBD_SLAB_INDEX(size);
1679 BUG_ON(jbd_slab[idx] == NULL);
1680 return kmem_cache_alloc(jbd_slab[idx], flags | __GFP_NOFAIL);
1681}
1682
1683void jbd_slab_free(void *ptr, size_t size)
1684{
1685 int idx;
1686
1687 idx = JBD_SLAB_INDEX(size);
1688 BUG_ON(jbd_slab[idx] == NULL);
1689 kmem_cache_free(jbd_slab[idx], ptr);
1690}
1691
1692/*
1693 * Journal_head storage management
1694 */
1695static kmem_cache_t *journal_head_cache;
1696#ifdef CONFIG_JBD_DEBUG
1697static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1698#endif
1699
1700static int journal_init_journal_head_cache(void)
1701{
1702 int retval;
1703
1704 J_ASSERT(journal_head_cache == 0);
1705 journal_head_cache = kmem_cache_create("journal_head",
1706 sizeof(struct journal_head),
1707 0, /* offset */
1708 0, /* flags */
1709 NULL, /* ctor */
1710 NULL); /* dtor */
1711 retval = 0;
1712 if (journal_head_cache == 0) {
1713 retval = -ENOMEM;
1714 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1715 }
1716 return retval;
1717}
1718
1719static void journal_destroy_journal_head_cache(void)
1720{
1721 J_ASSERT(journal_head_cache != NULL);
1722 kmem_cache_destroy(journal_head_cache);
1723 journal_head_cache = NULL;
1724}
1725
1726/*
1727 * journal_head splicing and dicing
1728 */
1729static struct journal_head *journal_alloc_journal_head(void)
1730{
1731 struct journal_head *ret;
1732 static unsigned long last_warning;
1733
1734#ifdef CONFIG_JBD_DEBUG
1735 atomic_inc(&nr_journal_heads);
1736#endif
1737 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1738 if (ret == 0) {
1739 jbd_debug(1, "out of memory for journal_head\n");
1740 if (time_after(jiffies, last_warning + 5*HZ)) {
1741 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1742 __FUNCTION__);
1743 last_warning = jiffies;
1744 }
1745 while (ret == 0) {
1746 yield();
1747 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1748 }
1749 }
1750 return ret;
1751}
1752
1753static void journal_free_journal_head(struct journal_head *jh)
1754{
1755#ifdef CONFIG_JBD_DEBUG
1756 atomic_dec(&nr_journal_heads);
1757 memset(jh, JBD_POISON_FREE, sizeof(*jh));
1758#endif
1759 kmem_cache_free(journal_head_cache, jh);
1760}
1761
1762/*
1763 * A journal_head is attached to a buffer_head whenever JBD has an
1764 * interest in the buffer.
1765 *
1766 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1767 * is set. This bit is tested in core kernel code where we need to take
1768 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1769 * there.
1770 *
1771 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1772 *
1773 * When a buffer has its BH_JBD bit set it is immune from being released by
1774 * core kernel code, mainly via ->b_count.
1775 *
1776 * A journal_head may be detached from its buffer_head when the journal_head's
1777 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1778 * Various places in JBD call journal_remove_journal_head() to indicate that the
1779 * journal_head can be dropped if needed.
1780 *
1781 * Various places in the kernel want to attach a journal_head to a buffer_head
1782 * _before_ attaching the journal_head to a transaction. To protect the
1783 * journal_head in this situation, journal_add_journal_head elevates the
1784 * journal_head's b_jcount refcount by one. The caller must call
1785 * journal_put_journal_head() to undo this.
1786 *
1787 * So the typical usage would be:
1788 *
1789 * (Attach a journal_head if needed. Increments b_jcount)
1790 * struct journal_head *jh = journal_add_journal_head(bh);
1791 * ...
1792 * jh->b_transaction = xxx;
1793 * journal_put_journal_head(jh);
1794 *
1795 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1796 * because it has a non-zero b_transaction.
1797 */
1798
1799/*
1800 * Give a buffer_head a journal_head.
1801 *
1802 * Doesn't need the journal lock.
1803 * May sleep.
1804 */
1805struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1806{
1807 struct journal_head *jh;
1808 struct journal_head *new_jh = NULL;
1809
1810repeat:
1811 if (!buffer_jbd(bh)) {
1812 new_jh = journal_alloc_journal_head();
1813 memset(new_jh, 0, sizeof(*new_jh));
1814 }
1815
1816 jbd_lock_bh_journal_head(bh);
1817 if (buffer_jbd(bh)) {
1818 jh = bh2jh(bh);
1819 } else {
1820 J_ASSERT_BH(bh,
1821 (atomic_read(&bh->b_count) > 0) ||
1822 (bh->b_page && bh->b_page->mapping));
1823
1824 if (!new_jh) {
1825 jbd_unlock_bh_journal_head(bh);
1826 goto repeat;
1827 }
1828
1829 jh = new_jh;
1830 new_jh = NULL; /* We consumed it */
1831 set_buffer_jbd(bh);
1832 bh->b_private = jh;
1833 jh->b_bh = bh;
1834 get_bh(bh);
1835 BUFFER_TRACE(bh, "added journal_head");
1836 }
1837 jh->b_jcount++;
1838 jbd_unlock_bh_journal_head(bh);
1839 if (new_jh)
1840 journal_free_journal_head(new_jh);
1841 return bh->b_private;
1842}
1843
1844/*
1845 * Grab a ref against this buffer_head's journal_head. If it ended up not
1846 * having a journal_head, return NULL
1847 */
1848struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1849{
1850 struct journal_head *jh = NULL;
1851
1852 jbd_lock_bh_journal_head(bh);
1853 if (buffer_jbd(bh)) {
1854 jh = bh2jh(bh);
1855 jh->b_jcount++;
1856 }
1857 jbd_unlock_bh_journal_head(bh);
1858 return jh;
1859}
1860
1861static void __journal_remove_journal_head(struct buffer_head *bh)
1862{
1863 struct journal_head *jh = bh2jh(bh);
1864
1865 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1866
1867 get_bh(bh);
1868 if (jh->b_jcount == 0) {
1869 if (jh->b_transaction == NULL &&
1870 jh->b_next_transaction == NULL &&
1871 jh->b_cp_transaction == NULL) {
1872 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1873 J_ASSERT_BH(bh, buffer_jbd(bh));
1874 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1875 BUFFER_TRACE(bh, "remove journal_head");
1876 if (jh->b_frozen_data) {
1877 printk(KERN_WARNING "%s: freeing "
1878 "b_frozen_data\n",
1879 __FUNCTION__);
1880 jbd_slab_free(jh->b_frozen_data, bh->b_size);
1881 }
1882 if (jh->b_committed_data) {
1883 printk(KERN_WARNING "%s: freeing "
1884 "b_committed_data\n",
1885 __FUNCTION__);
1886 jbd_slab_free(jh->b_committed_data, bh->b_size);
1887 }
1888 bh->b_private = NULL;
1889 jh->b_bh = NULL; /* debug, really */
1890 clear_buffer_jbd(bh);
1891 __brelse(bh);
1892 journal_free_journal_head(jh);
1893 } else {
1894 BUFFER_TRACE(bh, "journal_head was locked");
1895 }
1896 }
1897}
1898
1899/*
1900 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1901 * and has a zero b_jcount then remove and release its journal_head. If we did
1902 * see that the buffer is not used by any transaction we also "logically"
1903 * decrement ->b_count.
1904 *
1905 * We in fact take an additional increment on ->b_count as a convenience,
1906 * because the caller usually wants to do additional things with the bh
1907 * after calling here.
1908 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1909 * time. Once the caller has run __brelse(), the buffer is eligible for
1910 * reaping by try_to_free_buffers().
1911 */
1912void journal_remove_journal_head(struct buffer_head *bh)
1913{
1914 jbd_lock_bh_journal_head(bh);
1915 __journal_remove_journal_head(bh);
1916 jbd_unlock_bh_journal_head(bh);
1917}
1918
1919/*
1920 * Drop a reference on the passed journal_head. If it fell to zero then try to
1921 * release the journal_head from the buffer_head.
1922 */
1923void journal_put_journal_head(struct journal_head *jh)
1924{
1925 struct buffer_head *bh = jh2bh(jh);
1926
1927 jbd_lock_bh_journal_head(bh);
1928 J_ASSERT_JH(jh, jh->b_jcount > 0);
1929 --jh->b_jcount;
1930 if (!jh->b_jcount && !jh->b_transaction) {
1931 __journal_remove_journal_head(bh);
1932 __brelse(bh);
1933 }
1934 jbd_unlock_bh_journal_head(bh);
1935}
1936
1937/*
1938 * /proc tunables
1939 */
1940#if defined(CONFIG_JBD_DEBUG)
1941int journal_enable_debug;
1942EXPORT_SYMBOL(journal_enable_debug);
1943#endif
1944
1945#if defined(CONFIG_JBD_DEBUG) && defined(CONFIG_PROC_FS)
1946
1947static struct proc_dir_entry *proc_jbd_debug;
1948
1949static int read_jbd_debug(char *page, char **start, off_t off,
1950 int count, int *eof, void *data)
1951{
1952 int ret;
1953
1954 ret = sprintf(page + off, "%d\n", journal_enable_debug);
1955 *eof = 1;
1956 return ret;
1957}
1958
1959static int write_jbd_debug(struct file *file, const char __user *buffer,
1960 unsigned long count, void *data)
1961{
1962 char buf[32];
1963
1964 if (count > ARRAY_SIZE(buf) - 1)
1965 count = ARRAY_SIZE(buf) - 1;
1966 if (copy_from_user(buf, buffer, count))
1967 return -EFAULT;
1968 buf[ARRAY_SIZE(buf) - 1] = '\0';
1969 journal_enable_debug = simple_strtoul(buf, NULL, 10);
1970 return count;
1971}
1972
1973#define JBD_PROC_NAME "sys/fs/jbd-debug"
1974
1975static void __init create_jbd_proc_entry(void)
1976{
1977 proc_jbd_debug = create_proc_entry(JBD_PROC_NAME, 0644, NULL);
1978 if (proc_jbd_debug) {
1979 /* Why is this so hard? */
1980 proc_jbd_debug->read_proc = read_jbd_debug;
1981 proc_jbd_debug->write_proc = write_jbd_debug;
1982 }
1983}
1984
1985static void __exit remove_jbd_proc_entry(void)
1986{
1987 if (proc_jbd_debug)
1988 remove_proc_entry(JBD_PROC_NAME, NULL);
1989}
1990
1991#else
1992
1993#define create_jbd_proc_entry() do {} while (0)
1994#define remove_jbd_proc_entry() do {} while (0)
1995
1996#endif
1997
1998kmem_cache_t *jbd_handle_cache;
1999
2000static int __init journal_init_handle_cache(void)
2001{
2002 jbd_handle_cache = kmem_cache_create("journal_handle",
2003 sizeof(handle_t),
2004 0, /* offset */
2005 0, /* flags */
2006 NULL, /* ctor */
2007 NULL); /* dtor */
2008 if (jbd_handle_cache == NULL) {
2009 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2010 return -ENOMEM;
2011 }
2012 return 0;
2013}
2014
2015static void journal_destroy_handle_cache(void)
2016{
2017 if (jbd_handle_cache)
2018 kmem_cache_destroy(jbd_handle_cache);
2019}
2020
2021/*
2022 * Module startup and shutdown
2023 */
2024
2025static int __init journal_init_caches(void)
2026{
2027 int ret;
2028
2029 ret = journal_init_revoke_caches();
2030 if (ret == 0)
2031 ret = journal_init_journal_head_cache();
2032 if (ret == 0)
2033 ret = journal_init_handle_cache();
2034 return ret;
2035}
2036
2037static void journal_destroy_caches(void)
2038{
2039 journal_destroy_revoke_caches();
2040 journal_destroy_journal_head_cache();
2041 journal_destroy_handle_cache();
2042 journal_destroy_jbd_slabs();
2043}
2044
2045static int __init journal_init(void)
2046{
2047 int ret;
2048
2049 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2050
2051 ret = journal_init_caches();
2052 if (ret != 0)
2053 journal_destroy_caches();
2054 create_jbd_proc_entry();
2055 return ret;
2056}
2057
2058static void __exit journal_exit(void)
2059{
2060#ifdef CONFIG_JBD_DEBUG
2061 int n = atomic_read(&nr_journal_heads);
2062 if (n)
2063 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2064#endif
2065 remove_jbd_proc_entry();
2066 journal_destroy_caches();
2067}
2068
2069MODULE_LICENSE("GPL");
2070module_init(journal_init);
2071module_exit(journal_exit);
2072
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-02 08:02:07 -0500