blob: 46fe7439fb919b6c602e8e751e33cc42e94c1b8e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Uwe Zeisbergerf30c2262006-10-03 23:01:26 +02002 * linux/fs/jbd/journal.c
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5 *
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Generic filesystem journal-writing code; part of the ext2fs
13 * journaling system.
14 *
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
18 *
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
23 */
24
25#include <linux/module.h>
26#include <linux/time.h>
27#include <linux/fs.h>
28#include <linux/jbd.h>
29#include <linux/errno.h>
30#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/init.h>
32#include <linux/mm.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080033#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/pagemap.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080035#include <linux/kthread.h>
Randy Dunlapc9cf5522006-06-27 02:53:52 -070036#include <linux/poison.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080037#include <linux/proc_fs.h>
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);
Badari Pulavartyea817392006-08-27 01:23:52 -070086static int journal_create_jbd_slab(size_t slab_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88/*
89 * Helper function used to manage commit timeouts
90 */
91
92static void commit_timeout(unsigned long __data)
93{
94 struct task_struct * p = (struct task_struct *) __data;
95
96 wake_up_process(p);
97}
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099/*
100 * kjournald: The main thread function used to manage a logging device
101 * journal.
102 *
103 * This kernel thread is responsible for two things:
104 *
105 * 1) COMMIT: Every so often we need to commit the current state of the
106 * filesystem to disk. The journal thread is responsible for writing
107 * all of the metadata buffers to disk.
108 *
109 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
110 * of the data in that part of the log has been rewritten elsewhere on
111 * the disk. Flushing these old buffers to reclaim space in the log is
112 * known as checkpointing, and this thread is responsible for that job.
113 */
114
Adrian Bunk022a4a72005-09-06 15:16:41 -0700115static int kjournald(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Andrew Mortone3df1892006-03-25 03:06:53 -0800117 journal_t *journal = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 transaction_t *transaction;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Andrew Mortone3df1892006-03-25 03:06:53 -0800120 /*
121 * Set up an interval timer which can be used to trigger a commit wakeup
122 * after the commit interval expires
123 */
124 setup_timer(&journal->j_commit_timer, commit_timeout,
125 (unsigned long)current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127 /* Record that the journal thread is running */
128 journal->j_task = current;
129 wake_up(&journal->j_wait_done_commit);
130
131 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
132 journal->j_commit_interval / HZ);
133
134 /*
135 * And now, wait forever for commit wakeup events.
136 */
137 spin_lock(&journal->j_state_lock);
138
139loop:
140 if (journal->j_flags & JFS_UNMOUNT)
141 goto end_loop;
142
143 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
144 journal->j_commit_sequence, journal->j_commit_request);
145
146 if (journal->j_commit_sequence != journal->j_commit_request) {
147 jbd_debug(1, "OK, requests differ\n");
148 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800149 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 journal_commit_transaction(journal);
151 spin_lock(&journal->j_state_lock);
152 goto loop;
153 }
154
155 wake_up(&journal->j_wait_done_commit);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700156 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 /*
158 * The simpler the better. Flushing journal isn't a
159 * good idea, because that depends on threads that may
160 * be already stopped.
161 */
162 jbd_debug(1, "Now suspending kjournald\n");
163 spin_unlock(&journal->j_state_lock);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700164 refrigerator();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 spin_lock(&journal->j_state_lock);
166 } else {
167 /*
168 * We assume on resume that commits are already there,
169 * so we don't sleep
170 */
171 DEFINE_WAIT(wait);
172 int should_sleep = 1;
173
174 prepare_to_wait(&journal->j_wait_commit, &wait,
175 TASK_INTERRUPTIBLE);
176 if (journal->j_commit_sequence != journal->j_commit_request)
177 should_sleep = 0;
178 transaction = journal->j_running_transaction;
179 if (transaction && time_after_eq(jiffies,
180 transaction->t_expires))
181 should_sleep = 0;
Mark Fashehcbf0d272005-09-06 15:19:08 -0700182 if (journal->j_flags & JFS_UNMOUNT)
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700183 should_sleep = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 if (should_sleep) {
185 spin_unlock(&journal->j_state_lock);
186 schedule();
187 spin_lock(&journal->j_state_lock);
188 }
189 finish_wait(&journal->j_wait_commit, &wait);
190 }
191
192 jbd_debug(1, "kjournald wakes\n");
193
194 /*
195 * Were we woken up by a commit wakeup event?
196 */
197 transaction = journal->j_running_transaction;
198 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
199 journal->j_commit_request = transaction->t_tid;
200 jbd_debug(1, "woke because of timeout\n");
201 }
202 goto loop;
203
204end_loop:
205 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800206 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 journal->j_task = NULL;
208 wake_up(&journal->j_wait_done_commit);
209 jbd_debug(1, "Journal thread exiting.\n");
210 return 0;
211}
212
Pavel Emelianov97f06782007-05-08 00:30:42 -0700213static int journal_start_thread(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700215 struct task_struct *t;
216
217 t = kthread_run(kjournald, journal, "kjournald");
218 if (IS_ERR(t))
219 return PTR_ERR(t);
220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 wait_event(journal->j_wait_done_commit, journal->j_task != 0);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700222 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225static void journal_kill_thread(journal_t *journal)
226{
227 spin_lock(&journal->j_state_lock);
228 journal->j_flags |= JFS_UNMOUNT;
229
230 while (journal->j_task) {
231 wake_up(&journal->j_wait_commit);
232 spin_unlock(&journal->j_state_lock);
233 wait_event(journal->j_wait_done_commit, journal->j_task == 0);
234 spin_lock(&journal->j_state_lock);
235 }
236 spin_unlock(&journal->j_state_lock);
237}
238
239/*
240 * journal_write_metadata_buffer: write a metadata buffer to the journal.
241 *
242 * Writes a metadata buffer to a given disk block. The actual IO is not
243 * performed but a new buffer_head is constructed which labels the data
244 * to be written with the correct destination disk block.
245 *
246 * Any magic-number escaping which needs to be done will cause a
247 * copy-out here. If the buffer happens to start with the
248 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
249 * magic number is only written to the log for descripter blocks. In
250 * this case, we copy the data and replace the first word with 0, and we
251 * return a result code which indicates that this buffer needs to be
252 * marked as an escaped buffer in the corresponding log descriptor
253 * block. The missing word can then be restored when the block is read
254 * during recovery.
255 *
256 * If the source buffer has already been modified by a new transaction
257 * since we took the last commit snapshot, we use the frozen copy of
258 * that data for IO. If we end up using the existing buffer_head's data
259 * for the write, then we *have* to lock the buffer to prevent anyone
260 * else from using and possibly modifying it while the IO is in
261 * progress.
262 *
263 * The function returns a pointer to the buffer_heads to be used for IO.
264 *
265 * We assume that the journal has already been locked in this function.
266 *
267 * Return value:
268 * <0: Error
269 * >=0: Finished OK
270 *
271 * On success:
272 * Bit 0 set == escape performed on the data
273 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
274 */
275
276int journal_write_metadata_buffer(transaction_t *transaction,
277 struct journal_head *jh_in,
278 struct journal_head **jh_out,
Eric Sandeen37ed3222006-09-27 01:49:31 -0700279 unsigned long blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 int need_copy_out = 0;
282 int done_copy_out = 0;
283 int do_escape = 0;
284 char *mapped_data;
285 struct buffer_head *new_bh;
286 struct journal_head *new_jh;
287 struct page *new_page;
288 unsigned int new_offset;
289 struct buffer_head *bh_in = jh2bh(jh_in);
290
291 /*
292 * The buffer really shouldn't be locked: only the current committing
293 * transaction is allowed to write it, so nobody else is allowed
294 * to do any IO.
295 *
296 * akpm: except if we're journalling data, and write() output is
297 * also part of a shared mapping, and another thread has
298 * decided to launch a writepage() against this buffer.
299 */
300 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
301
302 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
303
304 /*
305 * If a new transaction has already done a buffer copy-out, then
306 * we use that version of the data for the commit.
307 */
308 jbd_lock_bh_state(bh_in);
309repeat:
310 if (jh_in->b_frozen_data) {
311 done_copy_out = 1;
312 new_page = virt_to_page(jh_in->b_frozen_data);
313 new_offset = offset_in_page(jh_in->b_frozen_data);
314 } else {
315 new_page = jh2bh(jh_in)->b_page;
316 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
317 }
318
319 mapped_data = kmap_atomic(new_page, KM_USER0);
320 /*
321 * Check for escaping
322 */
323 if (*((__be32 *)(mapped_data + new_offset)) ==
324 cpu_to_be32(JFS_MAGIC_NUMBER)) {
325 need_copy_out = 1;
326 do_escape = 1;
327 }
328 kunmap_atomic(mapped_data, KM_USER0);
329
330 /*
331 * Do we need to do a data copy?
332 */
333 if (need_copy_out && !done_copy_out) {
334 char *tmp;
335
336 jbd_unlock_bh_state(bh_in);
Badari Pulavartyea817392006-08-27 01:23:52 -0700337 tmp = jbd_slab_alloc(bh_in->b_size, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 jbd_lock_bh_state(bh_in);
339 if (jh_in->b_frozen_data) {
Badari Pulavartyea817392006-08-27 01:23:52 -0700340 jbd_slab_free(tmp, bh_in->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 goto repeat;
342 }
343
344 jh_in->b_frozen_data = tmp;
345 mapped_data = kmap_atomic(new_page, KM_USER0);
346 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
347 kunmap_atomic(mapped_data, KM_USER0);
348
349 new_page = virt_to_page(tmp);
350 new_offset = offset_in_page(tmp);
351 done_copy_out = 1;
352 }
353
354 /*
355 * Did we need to do an escaping? Now we've done all the
356 * copying, we can finally do so.
357 */
358 if (do_escape) {
359 mapped_data = kmap_atomic(new_page, KM_USER0);
360 *((unsigned int *)(mapped_data + new_offset)) = 0;
361 kunmap_atomic(mapped_data, KM_USER0);
362 }
363
364 /* keep subsequent assertions sane */
365 new_bh->b_state = 0;
366 init_buffer(new_bh, NULL, NULL);
367 atomic_set(&new_bh->b_count, 1);
368 jbd_unlock_bh_state(bh_in);
369
370 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
371
372 set_bh_page(new_bh, new_page, new_offset);
373 new_jh->b_transaction = NULL;
374 new_bh->b_size = jh2bh(jh_in)->b_size;
375 new_bh->b_bdev = transaction->t_journal->j_dev;
376 new_bh->b_blocknr = blocknr;
377 set_buffer_mapped(new_bh);
378 set_buffer_dirty(new_bh);
379
380 *jh_out = new_jh;
381
382 /*
383 * The to-be-written buffer needs to get moved to the io queue,
384 * and the original buffer whose contents we are shadowing or
385 * copying is moved to the transaction's shadow queue.
386 */
387 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
388 journal_file_buffer(jh_in, transaction, BJ_Shadow);
389 JBUFFER_TRACE(new_jh, "file as BJ_IO");
390 journal_file_buffer(new_jh, transaction, BJ_IO);
391
392 return do_escape | (done_copy_out << 1);
393}
394
395/*
396 * Allocation code for the journal file. Manage the space left in the
397 * journal, so that we can begin checkpointing when appropriate.
398 */
399
400/*
401 * __log_space_left: Return the number of free blocks left in the journal.
402 *
403 * Called with the journal already locked.
404 *
405 * Called under j_state_lock
406 */
407
408int __log_space_left(journal_t *journal)
409{
410 int left = journal->j_free;
411
412 assert_spin_locked(&journal->j_state_lock);
413
414 /*
415 * Be pessimistic here about the number of those free blocks which
416 * might be required for log descriptor control blocks.
417 */
418
419#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
420
421 left -= MIN_LOG_RESERVED_BLOCKS;
422
423 if (left <= 0)
424 return 0;
425 left -= (left >> 3);
426 return left;
427}
428
429/*
430 * Called under j_state_lock. Returns true if a transaction was started.
431 */
432int __log_start_commit(journal_t *journal, tid_t target)
433{
434 /*
435 * Are we already doing a recent enough commit?
436 */
437 if (!tid_geq(journal->j_commit_request, target)) {
438 /*
439 * We want a new commit: OK, mark the request and wakup the
440 * commit thread. We do _not_ do the commit ourselves.
441 */
442
443 journal->j_commit_request = target;
444 jbd_debug(1, "JBD: requesting commit %d/%d\n",
445 journal->j_commit_request,
446 journal->j_commit_sequence);
447 wake_up(&journal->j_wait_commit);
448 return 1;
449 }
450 return 0;
451}
452
453int log_start_commit(journal_t *journal, tid_t tid)
454{
455 int ret;
456
457 spin_lock(&journal->j_state_lock);
458 ret = __log_start_commit(journal, tid);
459 spin_unlock(&journal->j_state_lock);
460 return ret;
461}
462
463/*
464 * Force and wait upon a commit if the calling process is not within
465 * transaction. This is used for forcing out undo-protected data which contains
466 * bitmaps, when the fs is running out of space.
467 *
468 * We can only force the running transaction if we don't have an active handle;
469 * otherwise, we will deadlock.
470 *
471 * Returns true if a transaction was started.
472 */
473int journal_force_commit_nested(journal_t *journal)
474{
475 transaction_t *transaction = NULL;
476 tid_t tid;
477
478 spin_lock(&journal->j_state_lock);
479 if (journal->j_running_transaction && !current->journal_info) {
480 transaction = journal->j_running_transaction;
481 __log_start_commit(journal, transaction->t_tid);
482 } else if (journal->j_committing_transaction)
483 transaction = journal->j_committing_transaction;
484
485 if (!transaction) {
486 spin_unlock(&journal->j_state_lock);
487 return 0; /* Nothing to retry */
488 }
489
490 tid = transaction->t_tid;
491 spin_unlock(&journal->j_state_lock);
492 log_wait_commit(journal, tid);
493 return 1;
494}
495
496/*
497 * Start a commit of the current running transaction (if any). Returns true
498 * if a transaction was started, and fills its tid in at *ptid
499 */
500int journal_start_commit(journal_t *journal, tid_t *ptid)
501{
502 int ret = 0;
503
504 spin_lock(&journal->j_state_lock);
505 if (journal->j_running_transaction) {
506 tid_t tid = journal->j_running_transaction->t_tid;
507
508 ret = __log_start_commit(journal, tid);
509 if (ret && ptid)
510 *ptid = tid;
511 } else if (journal->j_committing_transaction && ptid) {
512 /*
513 * If ext3_write_super() recently started a commit, then we
514 * have to wait for completion of that transaction
515 */
516 *ptid = journal->j_committing_transaction->t_tid;
517 ret = 1;
518 }
519 spin_unlock(&journal->j_state_lock);
520 return ret;
521}
522
523/*
524 * Wait for a specified commit to complete.
525 * The caller may not hold the journal lock.
526 */
527int log_wait_commit(journal_t *journal, tid_t tid)
528{
529 int err = 0;
530
531#ifdef CONFIG_JBD_DEBUG
532 spin_lock(&journal->j_state_lock);
533 if (!tid_geq(journal->j_commit_request, tid)) {
534 printk(KERN_EMERG
535 "%s: error: j_commit_request=%d, tid=%d\n",
536 __FUNCTION__, journal->j_commit_request, tid);
537 }
538 spin_unlock(&journal->j_state_lock);
539#endif
540 spin_lock(&journal->j_state_lock);
541 while (tid_gt(tid, journal->j_commit_sequence)) {
542 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
543 tid, journal->j_commit_sequence);
544 wake_up(&journal->j_wait_commit);
545 spin_unlock(&journal->j_state_lock);
546 wait_event(journal->j_wait_done_commit,
547 !tid_gt(tid, journal->j_commit_sequence));
548 spin_lock(&journal->j_state_lock);
549 }
550 spin_unlock(&journal->j_state_lock);
551
552 if (unlikely(is_journal_aborted(journal))) {
553 printk(KERN_EMERG "journal commit I/O error\n");
554 err = -EIO;
555 }
556 return err;
557}
558
559/*
560 * Log buffer allocation routines:
561 */
562
563int journal_next_log_block(journal_t *journal, unsigned long *retp)
564{
565 unsigned long blocknr;
566
567 spin_lock(&journal->j_state_lock);
568 J_ASSERT(journal->j_free > 1);
569
570 blocknr = journal->j_head;
571 journal->j_head++;
572 journal->j_free--;
573 if (journal->j_head == journal->j_last)
574 journal->j_head = journal->j_first;
575 spin_unlock(&journal->j_state_lock);
576 return journal_bmap(journal, blocknr, retp);
577}
578
579/*
580 * Conversion of logical to physical block numbers for the journal
581 *
582 * On external journals the journal blocks are identity-mapped, so
583 * this is a no-op. If needed, we can use j_blk_offset - everything is
584 * ready.
585 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700586int journal_bmap(journal_t *journal, unsigned long blocknr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 unsigned long *retp)
588{
589 int err = 0;
590 unsigned long ret;
591
592 if (journal->j_inode) {
593 ret = bmap(journal->j_inode, blocknr);
594 if (ret)
595 *retp = ret;
596 else {
597 char b[BDEVNAME_SIZE];
598
599 printk(KERN_ALERT "%s: journal block not found "
600 "at offset %lu on %s\n",
601 __FUNCTION__,
602 blocknr,
603 bdevname(journal->j_dev, b));
604 err = -EIO;
605 __journal_abort_soft(journal, err);
606 }
607 } else {
608 *retp = blocknr; /* +journal->j_blk_offset */
609 }
610 return err;
611}
612
613/*
614 * We play buffer_head aliasing tricks to write data/metadata blocks to
615 * the journal without copying their contents, but for journal
616 * descriptor blocks we do need to generate bona fide buffers.
617 *
618 * After the caller of journal_get_descriptor_buffer() has finished modifying
619 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
620 * But we don't bother doing that, so there will be coherency problems with
621 * mmaps of blockdevs which hold live JBD-controlled filesystems.
622 */
623struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
624{
625 struct buffer_head *bh;
626 unsigned long blocknr;
627 int err;
628
629 err = journal_next_log_block(journal, &blocknr);
630
631 if (err)
632 return NULL;
633
634 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
635 lock_buffer(bh);
636 memset(bh->b_data, 0, journal->j_blocksize);
637 set_buffer_uptodate(bh);
638 unlock_buffer(bh);
639 BUFFER_TRACE(bh, "return this buffer");
640 return journal_add_journal_head(bh);
641}
642
643/*
644 * Management for journal control blocks: functions to create and
645 * destroy journal_t structures, and to initialise and read existing
646 * journal blocks from disk. */
647
648/* First: create and setup a journal_t object in memory. We initialise
649 * very few fields yet: that has to wait until we have created the
650 * journal structures from from scratch, or loaded them from disk. */
651
652static journal_t * journal_init_common (void)
653{
654 journal_t *journal;
655 int err;
656
657 journal = jbd_kmalloc(sizeof(*journal), GFP_KERNEL);
658 if (!journal)
659 goto fail;
660 memset(journal, 0, sizeof(*journal));
661
662 init_waitqueue_head(&journal->j_wait_transaction_locked);
663 init_waitqueue_head(&journal->j_wait_logspace);
664 init_waitqueue_head(&journal->j_wait_done_commit);
665 init_waitqueue_head(&journal->j_wait_checkpoint);
666 init_waitqueue_head(&journal->j_wait_commit);
667 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800668 mutex_init(&journal->j_barrier);
669 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 spin_lock_init(&journal->j_revoke_lock);
671 spin_lock_init(&journal->j_list_lock);
672 spin_lock_init(&journal->j_state_lock);
673
674 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
675
676 /* The journal is marked for error until we succeed with recovery! */
677 journal->j_flags = JFS_ABORT;
678
679 /* Set up a default-sized revoke table for the new mount. */
680 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
681 if (err) {
682 kfree(journal);
683 goto fail;
684 }
685 return journal;
686fail:
687 return NULL;
688}
689
690/* journal_init_dev and journal_init_inode:
691 *
692 * Create a journal structure assigned some fixed set of disk blocks to
693 * the journal. We don't actually touch those disk blocks yet, but we
694 * need to set up all of the mapping information to tell the journaling
695 * system where the journal blocks are.
696 *
697 */
698
699/**
700 * journal_t * journal_init_dev() - creates an initialises a journal structure
701 * @bdev: Block device on which to create the journal
702 * @fs_dev: Device which hold journalled filesystem for this journal.
703 * @start: Block nr Start of journal.
Eric Sandeen37ed3222006-09-27 01:49:31 -0700704 * @len: Length of the journal in blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 * @blocksize: blocksize of journalling device
706 * @returns: a newly created journal_t *
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700707 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 * journal_init_dev creates a journal which maps a fixed contiguous
709 * range of blocks on an arbitrary block device.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700710 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 */
712journal_t * journal_init_dev(struct block_device *bdev,
713 struct block_device *fs_dev,
714 int start, int len, int blocksize)
715{
716 journal_t *journal = journal_init_common();
717 struct buffer_head *bh;
718 int n;
719
720 if (!journal)
721 return NULL;
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 /* journal descriptor can store up to n blocks -bzzz */
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700724 journal->j_blocksize = blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 n = journal->j_blocksize / sizeof(journal_block_tag_t);
726 journal->j_wbufsize = n;
727 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
728 if (!journal->j_wbuf) {
729 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
730 __FUNCTION__);
731 kfree(journal);
732 journal = NULL;
Eric Sesterhenn41716c72006-10-11 01:20:37 -0700733 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700735 journal->j_dev = bdev;
736 journal->j_fs_dev = fs_dev;
737 journal->j_blk_offset = start;
738 journal->j_maxlen = len;
739
740 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
741 J_ASSERT(bh != NULL);
742 journal->j_sb_buffer = bh;
743 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Eric Sesterhenn41716c72006-10-11 01:20:37 -0700744out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return journal;
746}
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700747
748/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
750 * @inode: An inode to create the journal in
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700751 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 * journal_init_inode creates a journal which maps an on-disk inode as
753 * the journal. The inode must exist already, must support bmap() and
754 * must have all data blocks preallocated.
755 */
756journal_t * journal_init_inode (struct inode *inode)
757{
758 struct buffer_head *bh;
759 journal_t *journal = journal_init_common();
760 int err;
761 int n;
762 unsigned long blocknr;
763
764 if (!journal)
765 return NULL;
766
767 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
768 journal->j_inode = inode;
769 jbd_debug(1,
770 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700771 journal, inode->i_sb->s_id, inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 (long long) inode->i_size,
773 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
774
775 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
776 journal->j_blocksize = inode->i_sb->s_blocksize;
777
778 /* journal descriptor can store up to n blocks -bzzz */
779 n = journal->j_blocksize / sizeof(journal_block_tag_t);
780 journal->j_wbufsize = n;
781 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
782 if (!journal->j_wbuf) {
783 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
784 __FUNCTION__);
785 kfree(journal);
786 return NULL;
787 }
788
789 err = journal_bmap(journal, 0, &blocknr);
790 /* If that failed, give up */
791 if (err) {
792 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
793 __FUNCTION__);
794 kfree(journal);
795 return NULL;
796 }
797
798 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
799 J_ASSERT(bh != NULL);
800 journal->j_sb_buffer = bh;
801 journal->j_superblock = (journal_superblock_t *)bh->b_data;
802
803 return journal;
804}
805
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700806/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 * If the journal init or create aborts, we need to mark the journal
808 * superblock as being NULL to prevent the journal destroy from writing
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700809 * back a bogus superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 */
811static void journal_fail_superblock (journal_t *journal)
812{
813 struct buffer_head *bh = journal->j_sb_buffer;
814 brelse(bh);
815 journal->j_sb_buffer = NULL;
816}
817
818/*
819 * Given a journal_t structure, initialise the various fields for
820 * startup of a new journaling session. We use this both when creating
821 * a journal, and after recovering an old journal to reset it for
822 * subsequent use.
823 */
824
825static int journal_reset(journal_t *journal)
826{
827 journal_superblock_t *sb = journal->j_superblock;
Eric Sandeen37ed3222006-09-27 01:49:31 -0700828 unsigned long first, last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
830 first = be32_to_cpu(sb->s_first);
831 last = be32_to_cpu(sb->s_maxlen);
832
833 journal->j_first = first;
834 journal->j_last = last;
835
836 journal->j_head = first;
837 journal->j_tail = first;
838 journal->j_free = last - first;
839
840 journal->j_tail_sequence = journal->j_transaction_sequence;
841 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
842 journal->j_commit_request = journal->j_commit_sequence;
843
844 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
845
846 /* Add the dynamic fields and write it to disk. */
847 journal_update_superblock(journal, 1);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700848 return journal_start_thread(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849}
850
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700851/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 * int journal_create() - Initialise the new journal file
853 * @journal: Journal to create. This structure must have been initialised
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700854 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 * Given a journal_t structure which tells us which disk blocks we can
856 * use, create a new journal superblock and initialise all of the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700857 * journal fields from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 **/
859int journal_create(journal_t *journal)
860{
861 unsigned long blocknr;
862 struct buffer_head *bh;
863 journal_superblock_t *sb;
864 int i, err;
865
866 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
867 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
868 journal->j_maxlen);
869 journal_fail_superblock(journal);
870 return -EINVAL;
871 }
872
873 if (journal->j_inode == NULL) {
874 /*
875 * We don't know what block to start at!
876 */
877 printk(KERN_EMERG
878 "%s: creation of journal on external device!\n",
879 __FUNCTION__);
880 BUG();
881 }
882
883 /* Zero out the entire journal on disk. We cannot afford to
884 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
885 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
886 for (i = 0; i < journal->j_maxlen; i++) {
887 err = journal_bmap(journal, i, &blocknr);
888 if (err)
889 return err;
890 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
891 lock_buffer(bh);
892 memset (bh->b_data, 0, journal->j_blocksize);
893 BUFFER_TRACE(bh, "marking dirty");
894 mark_buffer_dirty(bh);
895 BUFFER_TRACE(bh, "marking uptodate");
896 set_buffer_uptodate(bh);
897 unlock_buffer(bh);
898 __brelse(bh);
899 }
900
901 sync_blockdev(journal->j_dev);
902 jbd_debug(1, "JBD: journal cleared.\n");
903
904 /* OK, fill in the initial static fields in the new superblock */
905 sb = journal->j_superblock;
906
907 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
908 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
909
910 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
911 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
912 sb->s_first = cpu_to_be32(1);
913
914 journal->j_transaction_sequence = 1;
915
916 journal->j_flags &= ~JFS_ABORT;
917 journal->j_format_version = 2;
918
919 return journal_reset(journal);
920}
921
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700922/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 * void journal_update_superblock() - Update journal sb on disk.
924 * @journal: The journal to update.
925 * @wait: Set to '0' if you don't want to wait for IO completion.
926 *
927 * Update a journal's dynamic superblock fields and write it to disk,
928 * optionally waiting for the IO to complete.
929 */
930void journal_update_superblock(journal_t *journal, int wait)
931{
932 journal_superblock_t *sb = journal->j_superblock;
933 struct buffer_head *bh = journal->j_sb_buffer;
934
935 /*
936 * As a special case, if the on-disk copy is already marked as needing
937 * no recovery (s_start == 0) and there are no outstanding transactions
938 * in the filesystem, then we can safely defer the superblock update
939 * until the next commit by setting JFS_FLUSHED. This avoids
940 * attempting a write to a potential-readonly device.
941 */
942 if (sb->s_start == 0 && journal->j_tail_sequence ==
943 journal->j_transaction_sequence) {
944 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
945 "(start %ld, seq %d, errno %d)\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700946 journal->j_tail, journal->j_tail_sequence,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 journal->j_errno);
948 goto out;
949 }
950
951 spin_lock(&journal->j_state_lock);
952 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
953 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
954
955 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
956 sb->s_start = cpu_to_be32(journal->j_tail);
957 sb->s_errno = cpu_to_be32(journal->j_errno);
958 spin_unlock(&journal->j_state_lock);
959
960 BUFFER_TRACE(bh, "marking dirty");
961 mark_buffer_dirty(bh);
962 if (wait)
963 sync_dirty_buffer(bh);
964 else
Jan Kara26707692005-09-06 15:19:12 -0700965 ll_rw_block(SWRITE, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967out:
968 /* If we have just flushed the log (by marking s_start==0), then
969 * any future commit will have to be careful to update the
970 * superblock again to re-record the true start of the log. */
971
972 spin_lock(&journal->j_state_lock);
973 if (sb->s_start)
974 journal->j_flags &= ~JFS_FLUSHED;
975 else
976 journal->j_flags |= JFS_FLUSHED;
977 spin_unlock(&journal->j_state_lock);
978}
979
980/*
981 * Read the superblock for a given journal, performing initial
982 * validation of the format.
983 */
984
985static int journal_get_superblock(journal_t *journal)
986{
987 struct buffer_head *bh;
988 journal_superblock_t *sb;
989 int err = -EIO;
990
991 bh = journal->j_sb_buffer;
992
993 J_ASSERT(bh != NULL);
994 if (!buffer_uptodate(bh)) {
995 ll_rw_block(READ, 1, &bh);
996 wait_on_buffer(bh);
997 if (!buffer_uptodate(bh)) {
998 printk (KERN_ERR
999 "JBD: IO error reading journal superblock\n");
1000 goto out;
1001 }
1002 }
1003
1004 sb = journal->j_superblock;
1005
1006 err = -EINVAL;
1007
1008 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1009 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1010 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1011 goto out;
1012 }
1013
1014 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1015 case JFS_SUPERBLOCK_V1:
1016 journal->j_format_version = 1;
1017 break;
1018 case JFS_SUPERBLOCK_V2:
1019 journal->j_format_version = 2;
1020 break;
1021 default:
1022 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1023 goto out;
1024 }
1025
1026 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1027 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1028 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1029 printk (KERN_WARNING "JBD: journal file too short\n");
1030 goto out;
1031 }
1032
1033 return 0;
1034
1035out:
1036 journal_fail_superblock(journal);
1037 return err;
1038}
1039
1040/*
1041 * Load the on-disk journal superblock and read the key fields into the
1042 * journal_t.
1043 */
1044
1045static int load_superblock(journal_t *journal)
1046{
1047 int err;
1048 journal_superblock_t *sb;
1049
1050 err = journal_get_superblock(journal);
1051 if (err)
1052 return err;
1053
1054 sb = journal->j_superblock;
1055
1056 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1057 journal->j_tail = be32_to_cpu(sb->s_start);
1058 journal->j_first = be32_to_cpu(sb->s_first);
1059 journal->j_last = be32_to_cpu(sb->s_maxlen);
1060 journal->j_errno = be32_to_cpu(sb->s_errno);
1061
1062 return 0;
1063}
1064
1065
1066/**
1067 * int journal_load() - Read journal from disk.
1068 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001069 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 * Given a journal_t structure which tells us which disk blocks contain
1071 * a journal, read the journal from disk to initialise the in-memory
1072 * structures.
1073 */
1074int journal_load(journal_t *journal)
1075{
1076 int err;
Badari Pulavartyea817392006-08-27 01:23:52 -07001077 journal_superblock_t *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
1079 err = load_superblock(journal);
1080 if (err)
1081 return err;
1082
Badari Pulavartyea817392006-08-27 01:23:52 -07001083 sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 /* If this is a V2 superblock, then we have to check the
1085 * features flags on it. */
1086
1087 if (journal->j_format_version >= 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 if ((sb->s_feature_ro_compat &
1089 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1090 (sb->s_feature_incompat &
1091 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1092 printk (KERN_WARNING
1093 "JBD: Unrecognised features on journal\n");
1094 return -EINVAL;
1095 }
1096 }
1097
Badari Pulavartyea817392006-08-27 01:23:52 -07001098 /*
1099 * Create a slab for this blocksize
1100 */
Dave Kleikampa4e4de32006-09-27 01:49:36 -07001101 err = journal_create_jbd_slab(be32_to_cpu(sb->s_blocksize));
Badari Pulavartyea817392006-08-27 01:23:52 -07001102 if (err)
1103 return err;
1104
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 /* Let the recovery code check whether it needs to recover any
1106 * data from the journal. */
1107 if (journal_recover(journal))
1108 goto recovery_error;
1109
1110 /* OK, we've finished with the dynamic journal bits:
1111 * reinitialise the dynamic contents of the superblock in memory
1112 * and reset them on disk. */
1113 if (journal_reset(journal))
1114 goto recovery_error;
1115
1116 journal->j_flags &= ~JFS_ABORT;
1117 journal->j_flags |= JFS_LOADED;
1118 return 0;
1119
1120recovery_error:
1121 printk (KERN_WARNING "JBD: recovery failed\n");
1122 return -EIO;
1123}
1124
1125/**
1126 * void journal_destroy() - Release a journal_t structure.
1127 * @journal: Journal to act on.
1128 *
1129 * Release a journal_t structure once it is no longer in use by the
1130 * journaled object.
1131 */
1132void journal_destroy(journal_t *journal)
1133{
1134 /* Wait for the commit thread to wake up and die. */
1135 journal_kill_thread(journal);
1136
1137 /* Force a final log commit */
1138 if (journal->j_running_transaction)
1139 journal_commit_transaction(journal);
1140
1141 /* Force any old transactions to disk */
1142
1143 /* Totally anal locking here... */
1144 spin_lock(&journal->j_list_lock);
1145 while (journal->j_checkpoint_transactions != NULL) {
1146 spin_unlock(&journal->j_list_lock);
1147 log_do_checkpoint(journal);
1148 spin_lock(&journal->j_list_lock);
1149 }
1150
1151 J_ASSERT(journal->j_running_transaction == NULL);
1152 J_ASSERT(journal->j_committing_transaction == NULL);
1153 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1154 spin_unlock(&journal->j_list_lock);
1155
1156 /* We can now mark the journal as empty. */
1157 journal->j_tail = 0;
1158 journal->j_tail_sequence = ++journal->j_transaction_sequence;
1159 if (journal->j_sb_buffer) {
1160 journal_update_superblock(journal, 1);
1161 brelse(journal->j_sb_buffer);
1162 }
1163
1164 if (journal->j_inode)
1165 iput(journal->j_inode);
1166 if (journal->j_revoke)
1167 journal_destroy_revoke(journal);
1168 kfree(journal->j_wbuf);
1169 kfree(journal);
1170}
1171
1172
1173/**
1174 *int journal_check_used_features () - Check if features specified are used.
1175 * @journal: Journal to check.
1176 * @compat: bitmask of compatible features
1177 * @ro: bitmask of features that force read-only mount
1178 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001179 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 * Check whether the journal uses all of a given set of
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001181 * features. Return true (non-zero) if it does.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 **/
1183
1184int journal_check_used_features (journal_t *journal, unsigned long compat,
1185 unsigned long ro, unsigned long incompat)
1186{
1187 journal_superblock_t *sb;
1188
1189 if (!compat && !ro && !incompat)
1190 return 1;
1191 if (journal->j_format_version == 1)
1192 return 0;
1193
1194 sb = journal->j_superblock;
1195
1196 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1197 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1198 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1199 return 1;
1200
1201 return 0;
1202}
1203
1204/**
1205 * int journal_check_available_features() - Check feature set in journalling layer
1206 * @journal: Journal to check.
1207 * @compat: bitmask of compatible features
1208 * @ro: bitmask of features that force read-only mount
1209 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001210 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 * Check whether the journaling code supports the use of
1212 * all of a given set of features on this journal. Return true
1213 * (non-zero) if it can. */
1214
1215int journal_check_available_features (journal_t *journal, unsigned long compat,
1216 unsigned long ro, unsigned long incompat)
1217{
1218 journal_superblock_t *sb;
1219
1220 if (!compat && !ro && !incompat)
1221 return 1;
1222
1223 sb = journal->j_superblock;
1224
1225 /* We can support any known requested features iff the
1226 * superblock is in version 2. Otherwise we fail to support any
1227 * extended sb features. */
1228
1229 if (journal->j_format_version != 2)
1230 return 0;
1231
1232 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1233 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1234 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1235 return 1;
1236
1237 return 0;
1238}
1239
1240/**
1241 * int journal_set_features () - Mark a given journal feature in the superblock
1242 * @journal: Journal to act on.
1243 * @compat: bitmask of compatible features
1244 * @ro: bitmask of features that force read-only mount
1245 * @incompat: bitmask of incompatible features
1246 *
1247 * Mark a given journal feature as present on the
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001248 * superblock. Returns true if the requested features could be set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 *
1250 */
1251
1252int journal_set_features (journal_t *journal, unsigned long compat,
1253 unsigned long ro, unsigned long incompat)
1254{
1255 journal_superblock_t *sb;
1256
1257 if (journal_check_used_features(journal, compat, ro, incompat))
1258 return 1;
1259
1260 if (!journal_check_available_features(journal, compat, ro, incompat))
1261 return 0;
1262
1263 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1264 compat, ro, incompat);
1265
1266 sb = journal->j_superblock;
1267
1268 sb->s_feature_compat |= cpu_to_be32(compat);
1269 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1270 sb->s_feature_incompat |= cpu_to_be32(incompat);
1271
1272 return 1;
1273}
1274
1275
1276/**
1277 * int journal_update_format () - Update on-disk journal structure.
1278 * @journal: Journal to act on.
1279 *
1280 * Given an initialised but unloaded journal struct, poke about in the
1281 * on-disk structure to update it to the most recent supported version.
1282 */
1283int journal_update_format (journal_t *journal)
1284{
1285 journal_superblock_t *sb;
1286 int err;
1287
1288 err = journal_get_superblock(journal);
1289 if (err)
1290 return err;
1291
1292 sb = journal->j_superblock;
1293
1294 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1295 case JFS_SUPERBLOCK_V2:
1296 return 0;
1297 case JFS_SUPERBLOCK_V1:
1298 return journal_convert_superblock_v1(journal, sb);
1299 default:
1300 break;
1301 }
1302 return -EINVAL;
1303}
1304
1305static int journal_convert_superblock_v1(journal_t *journal,
1306 journal_superblock_t *sb)
1307{
1308 int offset, blocksize;
1309 struct buffer_head *bh;
1310
1311 printk(KERN_WARNING
1312 "JBD: Converting superblock from version 1 to 2.\n");
1313
1314 /* Pre-initialise new fields to zero */
1315 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1316 blocksize = be32_to_cpu(sb->s_blocksize);
1317 memset(&sb->s_feature_compat, 0, blocksize-offset);
1318
1319 sb->s_nr_users = cpu_to_be32(1);
1320 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1321 journal->j_format_version = 2;
1322
1323 bh = journal->j_sb_buffer;
1324 BUFFER_TRACE(bh, "marking dirty");
1325 mark_buffer_dirty(bh);
1326 sync_dirty_buffer(bh);
1327 return 0;
1328}
1329
1330
1331/**
1332 * int journal_flush () - Flush journal
1333 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001334 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 * Flush all data for a given journal to disk and empty the journal.
1336 * Filesystems can use this when remounting readonly to ensure that
1337 * recovery does not need to happen on remount.
1338 */
1339
1340int journal_flush(journal_t *journal)
1341{
1342 int err = 0;
1343 transaction_t *transaction = NULL;
1344 unsigned long old_tail;
1345
1346 spin_lock(&journal->j_state_lock);
1347
1348 /* Force everything buffered to the log... */
1349 if (journal->j_running_transaction) {
1350 transaction = journal->j_running_transaction;
1351 __log_start_commit(journal, transaction->t_tid);
1352 } else if (journal->j_committing_transaction)
1353 transaction = journal->j_committing_transaction;
1354
1355 /* Wait for the log commit to complete... */
1356 if (transaction) {
1357 tid_t tid = transaction->t_tid;
1358
1359 spin_unlock(&journal->j_state_lock);
1360 log_wait_commit(journal, tid);
1361 } else {
1362 spin_unlock(&journal->j_state_lock);
1363 }
1364
1365 /* ...and flush everything in the log out to disk. */
1366 spin_lock(&journal->j_list_lock);
1367 while (!err && journal->j_checkpoint_transactions != NULL) {
1368 spin_unlock(&journal->j_list_lock);
1369 err = log_do_checkpoint(journal);
1370 spin_lock(&journal->j_list_lock);
1371 }
1372 spin_unlock(&journal->j_list_lock);
1373 cleanup_journal_tail(journal);
1374
1375 /* Finally, mark the journal as really needing no recovery.
1376 * This sets s_start==0 in the underlying superblock, which is
1377 * the magic code for a fully-recovered superblock. Any future
1378 * commits of data to the journal will restore the current
1379 * s_start value. */
1380 spin_lock(&journal->j_state_lock);
1381 old_tail = journal->j_tail;
1382 journal->j_tail = 0;
1383 spin_unlock(&journal->j_state_lock);
1384 journal_update_superblock(journal, 1);
1385 spin_lock(&journal->j_state_lock);
1386 journal->j_tail = old_tail;
1387
1388 J_ASSERT(!journal->j_running_transaction);
1389 J_ASSERT(!journal->j_committing_transaction);
1390 J_ASSERT(!journal->j_checkpoint_transactions);
1391 J_ASSERT(journal->j_head == journal->j_tail);
1392 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1393 spin_unlock(&journal->j_state_lock);
1394 return err;
1395}
1396
1397/**
1398 * int journal_wipe() - Wipe journal contents
1399 * @journal: Journal to act on.
1400 * @write: flag (see below)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001401 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 * Wipe out all of the contents of a journal, safely. This will produce
1403 * a warning if the journal contains any valid recovery information.
1404 * Must be called between journal_init_*() and journal_load().
1405 *
1406 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1407 * we merely suppress recovery.
1408 */
1409
1410int journal_wipe(journal_t *journal, int write)
1411{
1412 journal_superblock_t *sb;
1413 int err = 0;
1414
1415 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1416
1417 err = load_superblock(journal);
1418 if (err)
1419 return err;
1420
1421 sb = journal->j_superblock;
1422
1423 if (!journal->j_tail)
1424 goto no_recovery;
1425
1426 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1427 write ? "Clearing" : "Ignoring");
1428
1429 err = journal_skip_recovery(journal);
1430 if (write)
1431 journal_update_superblock(journal, 1);
1432
1433 no_recovery:
1434 return err;
1435}
1436
1437/*
1438 * journal_dev_name: format a character string to describe on what
1439 * device this journal is present.
1440 */
1441
Adrian Bunk022a4a72005-09-06 15:16:41 -07001442static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443{
1444 struct block_device *bdev;
1445
1446 if (journal->j_inode)
1447 bdev = journal->j_inode->i_sb->s_bdev;
1448 else
1449 bdev = journal->j_dev;
1450
1451 return bdevname(bdev, buffer);
1452}
1453
1454/*
1455 * Journal abort has very specific semantics, which we describe
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001456 * for journal abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 *
1458 * Two internal function, which provide abort to te jbd layer
1459 * itself are here.
1460 */
1461
1462/*
1463 * Quick version for internal journal use (doesn't lock the journal).
1464 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1465 * and don't attempt to make any other journal updates.
1466 */
1467void __journal_abort_hard(journal_t *journal)
1468{
1469 transaction_t *transaction;
1470 char b[BDEVNAME_SIZE];
1471
1472 if (journal->j_flags & JFS_ABORT)
1473 return;
1474
1475 printk(KERN_ERR "Aborting journal on device %s.\n",
1476 journal_dev_name(journal, b));
1477
1478 spin_lock(&journal->j_state_lock);
1479 journal->j_flags |= JFS_ABORT;
1480 transaction = journal->j_running_transaction;
1481 if (transaction)
1482 __log_start_commit(journal, transaction->t_tid);
1483 spin_unlock(&journal->j_state_lock);
1484}
1485
1486/* Soft abort: record the abort error status in the journal superblock,
1487 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001488static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
1490 if (journal->j_flags & JFS_ABORT)
1491 return;
1492
1493 if (!journal->j_errno)
1494 journal->j_errno = errno;
1495
1496 __journal_abort_hard(journal);
1497
1498 if (errno)
1499 journal_update_superblock(journal, 1);
1500}
1501
1502/**
1503 * void journal_abort () - Shutdown the journal immediately.
1504 * @journal: the journal to shutdown.
1505 * @errno: an error number to record in the journal indicating
1506 * the reason for the shutdown.
1507 *
1508 * Perform a complete, immediate shutdown of the ENTIRE
1509 * journal (not of a single transaction). This operation cannot be
1510 * undone without closing and reopening the journal.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001511 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 * The journal_abort function is intended to support higher level error
1513 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1514 * mode.
1515 *
1516 * Journal abort has very specific semantics. Any existing dirty,
1517 * unjournaled buffers in the main filesystem will still be written to
1518 * disk by bdflush, but the journaling mechanism will be suspended
1519 * immediately and no further transaction commits will be honoured.
1520 *
1521 * Any dirty, journaled buffers will be written back to disk without
1522 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1523 * filesystem, but we _do_ attempt to leave as much data as possible
1524 * behind for fsck to use for cleanup.
1525 *
1526 * Any attempt to get a new transaction handle on a journal which is in
1527 * ABORT state will just result in an -EROFS error return. A
1528 * journal_stop on an existing handle will return -EIO if we have
1529 * entered abort state during the update.
1530 *
1531 * Recursive transactions are not disturbed by journal abort until the
1532 * final journal_stop, which will receive the -EIO error.
1533 *
1534 * Finally, the journal_abort call allows the caller to supply an errno
1535 * which will be recorded (if possible) in the journal superblock. This
1536 * allows a client to record failure conditions in the middle of a
1537 * transaction without having to complete the transaction to record the
1538 * failure to disk. ext3_error, for example, now uses this
1539 * functionality.
1540 *
1541 * Errors which originate from within the journaling layer will NOT
1542 * supply an errno; a null errno implies that absolutely no further
1543 * writes are done to the journal (unless there are any already in
1544 * progress).
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001545 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 */
1547
1548void journal_abort(journal_t *journal, int errno)
1549{
1550 __journal_abort_soft(journal, errno);
1551}
1552
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001553/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 * int journal_errno () - returns the journal's error state.
1555 * @journal: journal to examine.
1556 *
1557 * This is the errno numbet set with journal_abort(), the last
1558 * time the journal was mounted - if the journal was stopped
1559 * without calling abort this will be 0.
1560 *
1561 * If the journal has been aborted on this mount time -EROFS will
1562 * be returned.
1563 */
1564int journal_errno(journal_t *journal)
1565{
1566 int err;
1567
1568 spin_lock(&journal->j_state_lock);
1569 if (journal->j_flags & JFS_ABORT)
1570 err = -EROFS;
1571 else
1572 err = journal->j_errno;
1573 spin_unlock(&journal->j_state_lock);
1574 return err;
1575}
1576
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001577/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 * int journal_clear_err () - clears the journal's error state
1579 * @journal: journal to act on.
1580 *
1581 * An error must be cleared or Acked to take a FS out of readonly
1582 * mode.
1583 */
1584int journal_clear_err(journal_t *journal)
1585{
1586 int err = 0;
1587
1588 spin_lock(&journal->j_state_lock);
1589 if (journal->j_flags & JFS_ABORT)
1590 err = -EROFS;
1591 else
1592 journal->j_errno = 0;
1593 spin_unlock(&journal->j_state_lock);
1594 return err;
1595}
1596
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001597/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 * void journal_ack_err() - Ack journal err.
1599 * @journal: journal to act on.
1600 *
1601 * An error must be cleared or Acked to take a FS out of readonly
1602 * mode.
1603 */
1604void journal_ack_err(journal_t *journal)
1605{
1606 spin_lock(&journal->j_state_lock);
1607 if (journal->j_errno)
1608 journal->j_flags |= JFS_ACK_ERR;
1609 spin_unlock(&journal->j_state_lock);
1610}
1611
1612int journal_blocks_per_page(struct inode *inode)
1613{
1614 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1615}
1616
1617/*
1618 * Simple support for retrying memory allocations. Introduced to help to
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001619 * debug different VM deadlock avoidance strategies.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 */
Al Viro27496a82005-10-21 03:20:48 -04001621void * __jbd_kmalloc (const char *where, size_t size, gfp_t flags, int retry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622{
1623 return kmalloc(size, flags | (retry ? __GFP_NOFAIL : 0));
1624}
1625
1626/*
Badari Pulavartyea817392006-08-27 01:23:52 -07001627 * jbd slab management: create 1k, 2k, 4k, 8k slabs as needed
1628 * and allocate frozen and commit buffers from these slabs.
1629 *
1630 * Reason for doing this is to avoid, SLAB_DEBUG - since it could
1631 * cause bh to cross page boundary.
1632 */
1633
1634#define JBD_MAX_SLABS 5
1635#define JBD_SLAB_INDEX(size) (size >> 11)
1636
Christoph Lametere18b8902006-12-06 20:33:20 -08001637static struct kmem_cache *jbd_slab[JBD_MAX_SLABS];
Badari Pulavartyea817392006-08-27 01:23:52 -07001638static const char *jbd_slab_names[JBD_MAX_SLABS] = {
1639 "jbd_1k", "jbd_2k", "jbd_4k", NULL, "jbd_8k"
1640};
1641
1642static void journal_destroy_jbd_slabs(void)
1643{
1644 int i;
1645
1646 for (i = 0; i < JBD_MAX_SLABS; i++) {
1647 if (jbd_slab[i])
1648 kmem_cache_destroy(jbd_slab[i]);
1649 jbd_slab[i] = NULL;
1650 }
1651}
1652
1653static int journal_create_jbd_slab(size_t slab_size)
1654{
1655 int i = JBD_SLAB_INDEX(slab_size);
1656
1657 BUG_ON(i >= JBD_MAX_SLABS);
1658
1659 /*
1660 * Check if we already have a slab created for this size
1661 */
1662 if (jbd_slab[i])
1663 return 0;
1664
1665 /*
1666 * Create a slab and force alignment to be same as slabsize -
1667 * this will make sure that allocations won't cross the page
1668 * boundary.
1669 */
1670 jbd_slab[i] = kmem_cache_create(jbd_slab_names[i],
1671 slab_size, slab_size, 0, NULL, NULL);
1672 if (!jbd_slab[i]) {
1673 printk(KERN_EMERG "JBD: no memory for jbd_slab cache\n");
1674 return -ENOMEM;
1675 }
1676 return 0;
1677}
1678
1679void * jbd_slab_alloc(size_t size, gfp_t flags)
1680{
1681 int idx;
1682
1683 idx = JBD_SLAB_INDEX(size);
1684 BUG_ON(jbd_slab[idx] == NULL);
1685 return kmem_cache_alloc(jbd_slab[idx], flags | __GFP_NOFAIL);
1686}
1687
1688void jbd_slab_free(void *ptr, size_t size)
1689{
1690 int idx;
1691
1692 idx = JBD_SLAB_INDEX(size);
1693 BUG_ON(jbd_slab[idx] == NULL);
1694 kmem_cache_free(jbd_slab[idx], ptr);
1695}
1696
1697/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 * Journal_head storage management
1699 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001700static struct kmem_cache *journal_head_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701#ifdef CONFIG_JBD_DEBUG
1702static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1703#endif
1704
1705static int journal_init_journal_head_cache(void)
1706{
1707 int retval;
1708
1709 J_ASSERT(journal_head_cache == 0);
1710 journal_head_cache = kmem_cache_create("journal_head",
1711 sizeof(struct journal_head),
1712 0, /* offset */
1713 0, /* flags */
1714 NULL, /* ctor */
1715 NULL); /* dtor */
1716 retval = 0;
1717 if (journal_head_cache == 0) {
1718 retval = -ENOMEM;
1719 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1720 }
1721 return retval;
1722}
1723
1724static void journal_destroy_journal_head_cache(void)
1725{
1726 J_ASSERT(journal_head_cache != NULL);
1727 kmem_cache_destroy(journal_head_cache);
1728 journal_head_cache = NULL;
1729}
1730
1731/*
1732 * journal_head splicing and dicing
1733 */
1734static struct journal_head *journal_alloc_journal_head(void)
1735{
1736 struct journal_head *ret;
1737 static unsigned long last_warning;
1738
1739#ifdef CONFIG_JBD_DEBUG
1740 atomic_inc(&nr_journal_heads);
1741#endif
1742 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1743 if (ret == 0) {
1744 jbd_debug(1, "out of memory for journal_head\n");
1745 if (time_after(jiffies, last_warning + 5*HZ)) {
1746 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1747 __FUNCTION__);
1748 last_warning = jiffies;
1749 }
1750 while (ret == 0) {
1751 yield();
1752 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1753 }
1754 }
1755 return ret;
1756}
1757
1758static void journal_free_journal_head(struct journal_head *jh)
1759{
1760#ifdef CONFIG_JBD_DEBUG
1761 atomic_dec(&nr_journal_heads);
Randy Dunlapc9cf5522006-06-27 02:53:52 -07001762 memset(jh, JBD_POISON_FREE, sizeof(*jh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763#endif
1764 kmem_cache_free(journal_head_cache, jh);
1765}
1766
1767/*
1768 * A journal_head is attached to a buffer_head whenever JBD has an
1769 * interest in the buffer.
1770 *
1771 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1772 * is set. This bit is tested in core kernel code where we need to take
1773 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1774 * there.
1775 *
1776 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1777 *
1778 * When a buffer has its BH_JBD bit set it is immune from being released by
1779 * core kernel code, mainly via ->b_count.
1780 *
1781 * A journal_head may be detached from its buffer_head when the journal_head's
1782 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1783 * Various places in JBD call journal_remove_journal_head() to indicate that the
1784 * journal_head can be dropped if needed.
1785 *
1786 * Various places in the kernel want to attach a journal_head to a buffer_head
1787 * _before_ attaching the journal_head to a transaction. To protect the
1788 * journal_head in this situation, journal_add_journal_head elevates the
1789 * journal_head's b_jcount refcount by one. The caller must call
1790 * journal_put_journal_head() to undo this.
1791 *
1792 * So the typical usage would be:
1793 *
1794 * (Attach a journal_head if needed. Increments b_jcount)
1795 * struct journal_head *jh = journal_add_journal_head(bh);
1796 * ...
1797 * jh->b_transaction = xxx;
1798 * journal_put_journal_head(jh);
1799 *
1800 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1801 * because it has a non-zero b_transaction.
1802 */
1803
1804/*
1805 * Give a buffer_head a journal_head.
1806 *
1807 * Doesn't need the journal lock.
1808 * May sleep.
1809 */
1810struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1811{
1812 struct journal_head *jh;
1813 struct journal_head *new_jh = NULL;
1814
1815repeat:
1816 if (!buffer_jbd(bh)) {
1817 new_jh = journal_alloc_journal_head();
1818 memset(new_jh, 0, sizeof(*new_jh));
1819 }
1820
1821 jbd_lock_bh_journal_head(bh);
1822 if (buffer_jbd(bh)) {
1823 jh = bh2jh(bh);
1824 } else {
1825 J_ASSERT_BH(bh,
1826 (atomic_read(&bh->b_count) > 0) ||
1827 (bh->b_page && bh->b_page->mapping));
1828
1829 if (!new_jh) {
1830 jbd_unlock_bh_journal_head(bh);
1831 goto repeat;
1832 }
1833
1834 jh = new_jh;
1835 new_jh = NULL; /* We consumed it */
1836 set_buffer_jbd(bh);
1837 bh->b_private = jh;
1838 jh->b_bh = bh;
1839 get_bh(bh);
1840 BUFFER_TRACE(bh, "added journal_head");
1841 }
1842 jh->b_jcount++;
1843 jbd_unlock_bh_journal_head(bh);
1844 if (new_jh)
1845 journal_free_journal_head(new_jh);
1846 return bh->b_private;
1847}
1848
1849/*
1850 * Grab a ref against this buffer_head's journal_head. If it ended up not
1851 * having a journal_head, return NULL
1852 */
1853struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1854{
1855 struct journal_head *jh = NULL;
1856
1857 jbd_lock_bh_journal_head(bh);
1858 if (buffer_jbd(bh)) {
1859 jh = bh2jh(bh);
1860 jh->b_jcount++;
1861 }
1862 jbd_unlock_bh_journal_head(bh);
1863 return jh;
1864}
1865
1866static void __journal_remove_journal_head(struct buffer_head *bh)
1867{
1868 struct journal_head *jh = bh2jh(bh);
1869
1870 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1871
1872 get_bh(bh);
1873 if (jh->b_jcount == 0) {
1874 if (jh->b_transaction == NULL &&
1875 jh->b_next_transaction == NULL &&
1876 jh->b_cp_transaction == NULL) {
1877 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1878 J_ASSERT_BH(bh, buffer_jbd(bh));
1879 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1880 BUFFER_TRACE(bh, "remove journal_head");
1881 if (jh->b_frozen_data) {
1882 printk(KERN_WARNING "%s: freeing "
1883 "b_frozen_data\n",
1884 __FUNCTION__);
Badari Pulavartyea817392006-08-27 01:23:52 -07001885 jbd_slab_free(jh->b_frozen_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 }
1887 if (jh->b_committed_data) {
1888 printk(KERN_WARNING "%s: freeing "
1889 "b_committed_data\n",
1890 __FUNCTION__);
Badari Pulavartyea817392006-08-27 01:23:52 -07001891 jbd_slab_free(jh->b_committed_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 }
1893 bh->b_private = NULL;
1894 jh->b_bh = NULL; /* debug, really */
1895 clear_buffer_jbd(bh);
1896 __brelse(bh);
1897 journal_free_journal_head(jh);
1898 } else {
1899 BUFFER_TRACE(bh, "journal_head was locked");
1900 }
1901 }
1902}
1903
1904/*
1905 * journal_remove_journal_head(): if the buffer isn't attached to a transaction
1906 * and has a zero b_jcount then remove and release its journal_head. If we did
1907 * see that the buffer is not used by any transaction we also "logically"
1908 * decrement ->b_count.
1909 *
1910 * We in fact take an additional increment on ->b_count as a convenience,
1911 * because the caller usually wants to do additional things with the bh
1912 * after calling here.
1913 * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
1914 * time. Once the caller has run __brelse(), the buffer is eligible for
1915 * reaping by try_to_free_buffers().
1916 */
1917void journal_remove_journal_head(struct buffer_head *bh)
1918{
1919 jbd_lock_bh_journal_head(bh);
1920 __journal_remove_journal_head(bh);
1921 jbd_unlock_bh_journal_head(bh);
1922}
1923
1924/*
1925 * Drop a reference on the passed journal_head. If it fell to zero then try to
1926 * release the journal_head from the buffer_head.
1927 */
1928void journal_put_journal_head(struct journal_head *jh)
1929{
1930 struct buffer_head *bh = jh2bh(jh);
1931
1932 jbd_lock_bh_journal_head(bh);
1933 J_ASSERT_JH(jh, jh->b_jcount > 0);
1934 --jh->b_jcount;
1935 if (!jh->b_jcount && !jh->b_transaction) {
1936 __journal_remove_journal_head(bh);
1937 __brelse(bh);
1938 }
1939 jbd_unlock_bh_journal_head(bh);
1940}
1941
1942/*
1943 * /proc tunables
1944 */
1945#if defined(CONFIG_JBD_DEBUG)
1946int journal_enable_debug;
1947EXPORT_SYMBOL(journal_enable_debug);
1948#endif
1949
1950#if defined(CONFIG_JBD_DEBUG) && defined(CONFIG_PROC_FS)
1951
1952static struct proc_dir_entry *proc_jbd_debug;
1953
Adrian Bunk022a4a72005-09-06 15:16:41 -07001954static int read_jbd_debug(char *page, char **start, off_t off,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 int count, int *eof, void *data)
1956{
1957 int ret;
1958
1959 ret = sprintf(page + off, "%d\n", journal_enable_debug);
1960 *eof = 1;
1961 return ret;
1962}
1963
Adrian Bunk022a4a72005-09-06 15:16:41 -07001964static int write_jbd_debug(struct file *file, const char __user *buffer,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 unsigned long count, void *data)
1966{
1967 char buf[32];
1968
1969 if (count > ARRAY_SIZE(buf) - 1)
1970 count = ARRAY_SIZE(buf) - 1;
1971 if (copy_from_user(buf, buffer, count))
1972 return -EFAULT;
1973 buf[ARRAY_SIZE(buf) - 1] = '\0';
1974 journal_enable_debug = simple_strtoul(buf, NULL, 10);
1975 return count;
1976}
1977
1978#define JBD_PROC_NAME "sys/fs/jbd-debug"
1979
1980static void __init create_jbd_proc_entry(void)
1981{
1982 proc_jbd_debug = create_proc_entry(JBD_PROC_NAME, 0644, NULL);
1983 if (proc_jbd_debug) {
1984 /* Why is this so hard? */
1985 proc_jbd_debug->read_proc = read_jbd_debug;
1986 proc_jbd_debug->write_proc = write_jbd_debug;
1987 }
1988}
1989
1990static void __exit remove_jbd_proc_entry(void)
1991{
1992 if (proc_jbd_debug)
1993 remove_proc_entry(JBD_PROC_NAME, NULL);
1994}
1995
1996#else
1997
1998#define create_jbd_proc_entry() do {} while (0)
1999#define remove_jbd_proc_entry() do {} while (0)
2000
2001#endif
2002
Christoph Lametere18b8902006-12-06 20:33:20 -08002003struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
2005static int __init journal_init_handle_cache(void)
2006{
2007 jbd_handle_cache = kmem_cache_create("journal_handle",
2008 sizeof(handle_t),
2009 0, /* offset */
2010 0, /* flags */
2011 NULL, /* ctor */
2012 NULL); /* dtor */
2013 if (jbd_handle_cache == NULL) {
2014 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2015 return -ENOMEM;
2016 }
2017 return 0;
2018}
2019
2020static void journal_destroy_handle_cache(void)
2021{
2022 if (jbd_handle_cache)
2023 kmem_cache_destroy(jbd_handle_cache);
2024}
2025
2026/*
2027 * Module startup and shutdown
2028 */
2029
2030static int __init journal_init_caches(void)
2031{
2032 int ret;
2033
2034 ret = journal_init_revoke_caches();
2035 if (ret == 0)
2036 ret = journal_init_journal_head_cache();
2037 if (ret == 0)
2038 ret = journal_init_handle_cache();
2039 return ret;
2040}
2041
2042static void journal_destroy_caches(void)
2043{
2044 journal_destroy_revoke_caches();
2045 journal_destroy_journal_head_cache();
2046 journal_destroy_handle_cache();
Badari Pulavartyea817392006-08-27 01:23:52 -07002047 journal_destroy_jbd_slabs();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048}
2049
2050static int __init journal_init(void)
2051{
2052 int ret;
2053
Alexey Dobriyan2aed3482006-09-27 01:49:28 -07002054 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
Adrian Bunk022a4a72005-09-06 15:16:41 -07002055
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 ret = journal_init_caches();
2057 if (ret != 0)
2058 journal_destroy_caches();
2059 create_jbd_proc_entry();
2060 return ret;
2061}
2062
2063static void __exit journal_exit(void)
2064{
2065#ifdef CONFIG_JBD_DEBUG
2066 int n = atomic_read(&nr_journal_heads);
2067 if (n)
2068 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2069#endif
2070 remove_jbd_proc_entry();
2071 journal_destroy_caches();
2072}
2073
2074MODULE_LICENSE("GPL");
2075module_init(journal_init);
2076module_exit(journal_exit);
2077