blob: 0971e9217808829a67f5fd340a1972a63a982930 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Uwe Zeisbergerf30c2262006-10-03 23:01:26 +02002 * linux/fs/jbd/journal.c
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5 *
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Generic filesystem journal-writing code; part of the ext2fs
13 * journaling system.
14 *
15 * This file manages journals: areas of disk reserved for logging
16 * transactional updates. This includes the kernel journaling thread
17 * which is responsible for scheduling updates to the log.
18 *
19 * We do not actually manage the physical storage of the journal in this
20 * file: that is left to a per-journal policy function, which allows us
21 * to store the journal within a filesystem-specified area for ext2
22 * journaling (ext2 can use a reserved inode for storing the log).
23 */
24
25#include <linux/module.h>
26#include <linux/time.h>
27#include <linux/fs.h>
28#include <linux/jbd.h>
29#include <linux/errno.h>
30#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/init.h>
32#include <linux/mm.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080033#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/pagemap.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080035#include <linux/kthread.h>
Randy Dunlapc9cf5522006-06-27 02:53:52 -070036#include <linux/poison.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080037#include <linux/proc_fs.h>
Jose R. Santosc2a91592007-10-18 23:39:22 -070038#include <linux/debugfs.h>
Namhyung Kimf81e3d42010-10-04 19:12:13 +090039#include <linux/ratelimit.h>
Andrew Morton8d8c8512006-03-25 03:06:53 -080040
Lukas Czerner99cb1a32011-05-23 18:33:02 +020041#define CREATE_TRACE_POINTS
42#include <trace/events/jbd.h>
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/uaccess.h>
45#include <asm/page.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47EXPORT_SYMBOL(journal_start);
48EXPORT_SYMBOL(journal_restart);
49EXPORT_SYMBOL(journal_extend);
50EXPORT_SYMBOL(journal_stop);
51EXPORT_SYMBOL(journal_lock_updates);
52EXPORT_SYMBOL(journal_unlock_updates);
53EXPORT_SYMBOL(journal_get_write_access);
54EXPORT_SYMBOL(journal_get_create_access);
55EXPORT_SYMBOL(journal_get_undo_access);
56EXPORT_SYMBOL(journal_dirty_data);
57EXPORT_SYMBOL(journal_dirty_metadata);
58EXPORT_SYMBOL(journal_release_buffer);
59EXPORT_SYMBOL(journal_forget);
60#if 0
61EXPORT_SYMBOL(journal_sync_buffer);
62#endif
63EXPORT_SYMBOL(journal_flush);
64EXPORT_SYMBOL(journal_revoke);
65
66EXPORT_SYMBOL(journal_init_dev);
67EXPORT_SYMBOL(journal_init_inode);
68EXPORT_SYMBOL(journal_update_format);
69EXPORT_SYMBOL(journal_check_used_features);
70EXPORT_SYMBOL(journal_check_available_features);
71EXPORT_SYMBOL(journal_set_features);
72EXPORT_SYMBOL(journal_create);
73EXPORT_SYMBOL(journal_load);
74EXPORT_SYMBOL(journal_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075EXPORT_SYMBOL(journal_abort);
76EXPORT_SYMBOL(journal_errno);
77EXPORT_SYMBOL(journal_ack_err);
78EXPORT_SYMBOL(journal_clear_err);
79EXPORT_SYMBOL(log_wait_commit);
Stefan Schmidtff5e4b52009-11-12 09:53:50 +010080EXPORT_SYMBOL(log_start_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081EXPORT_SYMBOL(journal_start_commit);
82EXPORT_SYMBOL(journal_force_commit_nested);
83EXPORT_SYMBOL(journal_wipe);
84EXPORT_SYMBOL(journal_blocks_per_page);
85EXPORT_SYMBOL(journal_invalidatepage);
86EXPORT_SYMBOL(journal_try_to_free_buffers);
87EXPORT_SYMBOL(journal_force_commit);
88
89static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
Adrian Bunk022a4a72005-09-06 15:16:41 -070090static void __journal_abort_soft (journal_t *journal, int errno);
Darrick J. Wongdff68252010-10-04 12:35:05 -070091static const char *journal_dev_name(journal_t *journal, char *buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93/*
94 * Helper function used to manage commit timeouts
95 */
96
97static void commit_timeout(unsigned long __data)
98{
99 struct task_struct * p = (struct task_struct *) __data;
100
101 wake_up_process(p);
102}
103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104/*
105 * kjournald: The main thread function used to manage a logging device
106 * journal.
107 *
108 * This kernel thread is responsible for two things:
109 *
110 * 1) COMMIT: Every so often we need to commit the current state of the
111 * filesystem to disk. The journal thread is responsible for writing
112 * all of the metadata buffers to disk.
113 *
114 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
115 * of the data in that part of the log has been rewritten elsewhere on
116 * the disk. Flushing these old buffers to reclaim space in the log is
117 * known as checkpointing, and this thread is responsible for that job.
118 */
119
Adrian Bunk022a4a72005-09-06 15:16:41 -0700120static int kjournald(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Andrew Mortone3df1892006-03-25 03:06:53 -0800122 journal_t *journal = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 transaction_t *transaction;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Andrew Mortone3df1892006-03-25 03:06:53 -0800125 /*
126 * Set up an interval timer which can be used to trigger a commit wakeup
127 * after the commit interval expires
128 */
129 setup_timer(&journal->j_commit_timer, commit_timeout,
130 (unsigned long)current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Nigel Cunningham35c80422012-02-03 19:59:41 +1100132 set_freezable();
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 /* Record that the journal thread is running */
135 journal->j_task = current;
136 wake_up(&journal->j_wait_done_commit);
137
138 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
139 journal->j_commit_interval / HZ);
140
141 /*
142 * And now, wait forever for commit wakeup events.
143 */
144 spin_lock(&journal->j_state_lock);
145
146loop:
147 if (journal->j_flags & JFS_UNMOUNT)
148 goto end_loop;
149
150 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
151 journal->j_commit_sequence, journal->j_commit_request);
152
153 if (journal->j_commit_sequence != journal->j_commit_request) {
154 jbd_debug(1, "OK, requests differ\n");
155 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800156 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 journal_commit_transaction(journal);
158 spin_lock(&journal->j_state_lock);
159 goto loop;
160 }
161
162 wake_up(&journal->j_wait_done_commit);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700163 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 /*
165 * The simpler the better. Flushing journal isn't a
166 * good idea, because that depends on threads that may
167 * be already stopped.
168 */
169 jbd_debug(1, "Now suspending kjournald\n");
170 spin_unlock(&journal->j_state_lock);
Tejun Heoa0acae02011-11-21 12:32:22 -0800171 try_to_freeze();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 spin_lock(&journal->j_state_lock);
173 } else {
174 /*
175 * We assume on resume that commits are already there,
176 * so we don't sleep
177 */
178 DEFINE_WAIT(wait);
179 int should_sleep = 1;
180
181 prepare_to_wait(&journal->j_wait_commit, &wait,
182 TASK_INTERRUPTIBLE);
183 if (journal->j_commit_sequence != journal->j_commit_request)
184 should_sleep = 0;
185 transaction = journal->j_running_transaction;
186 if (transaction && time_after_eq(jiffies,
187 transaction->t_expires))
188 should_sleep = 0;
Mark Fashehcbf0d272005-09-06 15:19:08 -0700189 if (journal->j_flags & JFS_UNMOUNT)
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700190 should_sleep = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 if (should_sleep) {
192 spin_unlock(&journal->j_state_lock);
193 schedule();
194 spin_lock(&journal->j_state_lock);
195 }
196 finish_wait(&journal->j_wait_commit, &wait);
197 }
198
199 jbd_debug(1, "kjournald wakes\n");
200
201 /*
202 * Were we woken up by a commit wakeup event?
203 */
204 transaction = journal->j_running_transaction;
205 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
206 journal->j_commit_request = transaction->t_tid;
207 jbd_debug(1, "woke because of timeout\n");
208 }
209 goto loop;
210
211end_loop:
212 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800213 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 journal->j_task = NULL;
215 wake_up(&journal->j_wait_done_commit);
216 jbd_debug(1, "Journal thread exiting.\n");
217 return 0;
218}
219
Pavel Emelianov97f06782007-05-08 00:30:42 -0700220static int journal_start_thread(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700222 struct task_struct *t;
223
224 t = kthread_run(kjournald, journal, "kjournald");
225 if (IS_ERR(t))
226 return PTR_ERR(t);
227
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700228 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700229 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230}
231
232static void journal_kill_thread(journal_t *journal)
233{
234 spin_lock(&journal->j_state_lock);
235 journal->j_flags |= JFS_UNMOUNT;
236
237 while (journal->j_task) {
238 wake_up(&journal->j_wait_commit);
239 spin_unlock(&journal->j_state_lock);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700240 wait_event(journal->j_wait_done_commit,
241 journal->j_task == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 spin_lock(&journal->j_state_lock);
243 }
244 spin_unlock(&journal->j_state_lock);
245}
246
247/*
248 * journal_write_metadata_buffer: write a metadata buffer to the journal.
249 *
250 * Writes a metadata buffer to a given disk block. The actual IO is not
251 * performed but a new buffer_head is constructed which labels the data
252 * to be written with the correct destination disk block.
253 *
254 * Any magic-number escaping which needs to be done will cause a
255 * copy-out here. If the buffer happens to start with the
256 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
257 * magic number is only written to the log for descripter blocks. In
258 * this case, we copy the data and replace the first word with 0, and we
259 * return a result code which indicates that this buffer needs to be
260 * marked as an escaped buffer in the corresponding log descriptor
261 * block. The missing word can then be restored when the block is read
262 * during recovery.
263 *
264 * If the source buffer has already been modified by a new transaction
265 * since we took the last commit snapshot, we use the frozen copy of
266 * that data for IO. If we end up using the existing buffer_head's data
267 * for the write, then we *have* to lock the buffer to prevent anyone
268 * else from using and possibly modifying it while the IO is in
269 * progress.
270 *
271 * The function returns a pointer to the buffer_heads to be used for IO.
272 *
273 * We assume that the journal has already been locked in this function.
274 *
275 * Return value:
276 * <0: Error
277 * >=0: Finished OK
278 *
279 * On success:
280 * Bit 0 set == escape performed on the data
281 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
282 */
283
284int journal_write_metadata_buffer(transaction_t *transaction,
285 struct journal_head *jh_in,
286 struct journal_head **jh_out,
Jan Kara9c28cbc2009-08-03 19:21:00 +0200287 unsigned int blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
289 int need_copy_out = 0;
290 int done_copy_out = 0;
291 int do_escape = 0;
292 char *mapped_data;
293 struct buffer_head *new_bh;
294 struct journal_head *new_jh;
295 struct page *new_page;
296 unsigned int new_offset;
297 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghuaf1015c42009-07-15 21:42:05 +0200298 journal_t *journal = transaction->t_journal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300 /*
301 * The buffer really shouldn't be locked: only the current committing
302 * transaction is allowed to write it, so nobody else is allowed
303 * to do any IO.
304 *
305 * akpm: except if we're journalling data, and write() output is
306 * also part of a shared mapping, and another thread has
307 * decided to launch a writepage() against this buffer.
308 */
309 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
310
311 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
dingdinghuaf1015c42009-07-15 21:42:05 +0200312 /* keep subsequent assertions sane */
313 new_bh->b_state = 0;
314 init_buffer(new_bh, NULL, NULL);
315 atomic_set(&new_bh->b_count, 1);
316 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
318 /*
319 * If a new transaction has already done a buffer copy-out, then
320 * we use that version of the data for the commit.
321 */
322 jbd_lock_bh_state(bh_in);
323repeat:
324 if (jh_in->b_frozen_data) {
325 done_copy_out = 1;
326 new_page = virt_to_page(jh_in->b_frozen_data);
327 new_offset = offset_in_page(jh_in->b_frozen_data);
328 } else {
329 new_page = jh2bh(jh_in)->b_page;
330 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
331 }
332
Cong Wang8fb53c42011-11-25 23:14:31 +0800333 mapped_data = kmap_atomic(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 /*
335 * Check for escaping
336 */
337 if (*((__be32 *)(mapped_data + new_offset)) ==
338 cpu_to_be32(JFS_MAGIC_NUMBER)) {
339 need_copy_out = 1;
340 do_escape = 1;
341 }
Cong Wang8fb53c42011-11-25 23:14:31 +0800342 kunmap_atomic(mapped_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
344 /*
345 * Do we need to do a data copy?
346 */
347 if (need_copy_out && !done_copy_out) {
348 char *tmp;
349
350 jbd_unlock_bh_state(bh_in);
Mingming Caoc089d492007-10-16 18:38:25 -0400351 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 jbd_lock_bh_state(bh_in);
353 if (jh_in->b_frozen_data) {
Mingming Caoc089d492007-10-16 18:38:25 -0400354 jbd_free(tmp, bh_in->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 goto repeat;
356 }
357
358 jh_in->b_frozen_data = tmp;
Cong Wang8fb53c42011-11-25 23:14:31 +0800359 mapped_data = kmap_atomic(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
Cong Wang8fb53c42011-11-25 23:14:31 +0800361 kunmap_atomic(mapped_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 new_page = virt_to_page(tmp);
364 new_offset = offset_in_page(tmp);
365 done_copy_out = 1;
366 }
367
368 /*
369 * Did we need to do an escaping? Now we've done all the
370 * copying, we can finally do so.
371 */
372 if (do_escape) {
Cong Wang8fb53c42011-11-25 23:14:31 +0800373 mapped_data = kmap_atomic(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 *((unsigned int *)(mapped_data + new_offset)) = 0;
Cong Wang8fb53c42011-11-25 23:14:31 +0800375 kunmap_atomic(mapped_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 }
377
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 set_bh_page(new_bh, new_page, new_offset);
379 new_jh->b_transaction = NULL;
380 new_bh->b_size = jh2bh(jh_in)->b_size;
381 new_bh->b_bdev = transaction->t_journal->j_dev;
382 new_bh->b_blocknr = blocknr;
383 set_buffer_mapped(new_bh);
384 set_buffer_dirty(new_bh);
385
386 *jh_out = new_jh;
387
388 /*
389 * The to-be-written buffer needs to get moved to the io queue,
390 * and the original buffer whose contents we are shadowing or
391 * copying is moved to the transaction's shadow queue.
392 */
393 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghuaf1015c42009-07-15 21:42:05 +0200394 spin_lock(&journal->j_list_lock);
395 __journal_file_buffer(jh_in, transaction, BJ_Shadow);
396 spin_unlock(&journal->j_list_lock);
397 jbd_unlock_bh_state(bh_in);
398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 JBUFFER_TRACE(new_jh, "file as BJ_IO");
400 journal_file_buffer(new_jh, transaction, BJ_IO);
401
402 return do_escape | (done_copy_out << 1);
403}
404
405/*
406 * Allocation code for the journal file. Manage the space left in the
407 * journal, so that we can begin checkpointing when appropriate.
408 */
409
410/*
411 * __log_space_left: Return the number of free blocks left in the journal.
412 *
413 * Called with the journal already locked.
414 *
415 * Called under j_state_lock
416 */
417
418int __log_space_left(journal_t *journal)
419{
420 int left = journal->j_free;
421
422 assert_spin_locked(&journal->j_state_lock);
423
424 /*
425 * Be pessimistic here about the number of those free blocks which
426 * might be required for log descriptor control blocks.
427 */
428
429#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
430
431 left -= MIN_LOG_RESERVED_BLOCKS;
432
433 if (left <= 0)
434 return 0;
435 left -= (left >> 3);
436 return left;
437}
438
439/*
Jan Kara8fe4cd02009-02-11 13:04:25 -0800440 * Called under j_state_lock. Returns true if a transaction commit was started.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 */
442int __log_start_commit(journal_t *journal, tid_t target)
443{
444 /*
Ted Ts'od9b01932011-04-30 13:17:11 -0400445 * The only transaction we can possibly wait upon is the
446 * currently running transaction (if it exists). Otherwise,
447 * the target tid must be an old one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 */
Ted Ts'od9b01932011-04-30 13:17:11 -0400449 if (journal->j_running_transaction &&
450 journal->j_running_transaction->t_tid == target) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200452 * We want a new commit: OK, mark the request and wakeup the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 * commit thread. We do _not_ do the commit ourselves.
454 */
455
456 journal->j_commit_request = target;
457 jbd_debug(1, "JBD: requesting commit %d/%d\n",
458 journal->j_commit_request,
459 journal->j_commit_sequence);
460 wake_up(&journal->j_wait_commit);
461 return 1;
Ted Ts'od9b01932011-04-30 13:17:11 -0400462 } else if (!tid_geq(journal->j_commit_request, target))
463 /* This should never happen, but if it does, preserve
464 the evidence before kjournald goes into a loop and
465 increments j_commit_sequence beyond all recognition. */
466 WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
467 journal->j_commit_request, journal->j_commit_sequence,
468 target, journal->j_running_transaction ?
469 journal->j_running_transaction->t_tid : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 return 0;
471}
472
473int log_start_commit(journal_t *journal, tid_t tid)
474{
475 int ret;
476
477 spin_lock(&journal->j_state_lock);
478 ret = __log_start_commit(journal, tid);
479 spin_unlock(&journal->j_state_lock);
480 return ret;
481}
482
483/*
484 * Force and wait upon a commit if the calling process is not within
485 * transaction. This is used for forcing out undo-protected data which contains
486 * bitmaps, when the fs is running out of space.
487 *
488 * We can only force the running transaction if we don't have an active handle;
489 * otherwise, we will deadlock.
490 *
491 * Returns true if a transaction was started.
492 */
493int journal_force_commit_nested(journal_t *journal)
494{
495 transaction_t *transaction = NULL;
496 tid_t tid;
497
498 spin_lock(&journal->j_state_lock);
499 if (journal->j_running_transaction && !current->journal_info) {
500 transaction = journal->j_running_transaction;
501 __log_start_commit(journal, transaction->t_tid);
502 } else if (journal->j_committing_transaction)
503 transaction = journal->j_committing_transaction;
504
505 if (!transaction) {
506 spin_unlock(&journal->j_state_lock);
507 return 0; /* Nothing to retry */
508 }
509
510 tid = transaction->t_tid;
511 spin_unlock(&journal->j_state_lock);
512 log_wait_commit(journal, tid);
513 return 1;
514}
515
516/*
517 * Start a commit of the current running transaction (if any). Returns true
Jan Kara8fe4cd02009-02-11 13:04:25 -0800518 * if a transaction is going to be committed (or is currently already
519 * committing), and fills its tid in at *ptid
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 */
521int journal_start_commit(journal_t *journal, tid_t *ptid)
522{
523 int ret = 0;
524
525 spin_lock(&journal->j_state_lock);
526 if (journal->j_running_transaction) {
527 tid_t tid = journal->j_running_transaction->t_tid;
528
Jan Kara8fe4cd02009-02-11 13:04:25 -0800529 __log_start_commit(journal, tid);
530 /* There's a running transaction and we've just made sure
531 * it's commit has been scheduled. */
532 if (ptid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 *ptid = tid;
Jan Kara8fe4cd02009-02-11 13:04:25 -0800534 ret = 1;
535 } else if (journal->j_committing_transaction) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 /*
537 * If ext3_write_super() recently started a commit, then we
538 * have to wait for completion of that transaction
539 */
Jan Kara8fe4cd02009-02-11 13:04:25 -0800540 if (ptid)
541 *ptid = journal->j_committing_transaction->t_tid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 ret = 1;
543 }
544 spin_unlock(&journal->j_state_lock);
545 return ret;
546}
547
548/*
549 * Wait for a specified commit to complete.
550 * The caller may not hold the journal lock.
551 */
552int log_wait_commit(journal_t *journal, tid_t tid)
553{
554 int err = 0;
555
556#ifdef CONFIG_JBD_DEBUG
557 spin_lock(&journal->j_state_lock);
558 if (!tid_geq(journal->j_commit_request, tid)) {
559 printk(KERN_EMERG
560 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700561 __func__, journal->j_commit_request, tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
563 spin_unlock(&journal->j_state_lock);
564#endif
565 spin_lock(&journal->j_state_lock);
566 while (tid_gt(tid, journal->j_commit_sequence)) {
567 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
568 tid, journal->j_commit_sequence);
569 wake_up(&journal->j_wait_commit);
570 spin_unlock(&journal->j_state_lock);
571 wait_event(journal->j_wait_done_commit,
572 !tid_gt(tid, journal->j_commit_sequence));
573 spin_lock(&journal->j_state_lock);
574 }
575 spin_unlock(&journal->j_state_lock);
576
577 if (unlikely(is_journal_aborted(journal))) {
578 printk(KERN_EMERG "journal commit I/O error\n");
579 err = -EIO;
580 }
581 return err;
582}
583
584/*
Jan Kara03f4d802010-04-15 22:16:24 +0200585 * Return 1 if a given transaction has not yet sent barrier request
586 * connected with a transaction commit. If 0 is returned, transaction
587 * may or may not have sent the barrier. Used to avoid sending barrier
588 * twice in common cases.
589 */
590int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
591{
592 int ret = 0;
593 transaction_t *commit_trans;
594
595 if (!(journal->j_flags & JFS_BARRIER))
596 return 0;
597 spin_lock(&journal->j_state_lock);
598 /* Transaction already committed? */
599 if (tid_geq(journal->j_commit_sequence, tid))
600 goto out;
601 /*
602 * Transaction is being committed and we already proceeded to
603 * writing commit record?
604 */
605 commit_trans = journal->j_committing_transaction;
606 if (commit_trans && commit_trans->t_tid == tid &&
607 commit_trans->t_state >= T_COMMIT_RECORD)
608 goto out;
609 ret = 1;
610out:
611 spin_unlock(&journal->j_state_lock);
612 return ret;
613}
Jan Kara52779702010-04-15 22:24:26 +0200614EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
Jan Kara03f4d802010-04-15 22:16:24 +0200615
616/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 * Log buffer allocation routines:
618 */
619
Jan Kara9c28cbc2009-08-03 19:21:00 +0200620int journal_next_log_block(journal_t *journal, unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200622 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624 spin_lock(&journal->j_state_lock);
625 J_ASSERT(journal->j_free > 1);
626
627 blocknr = journal->j_head;
628 journal->j_head++;
629 journal->j_free--;
630 if (journal->j_head == journal->j_last)
631 journal->j_head = journal->j_first;
632 spin_unlock(&journal->j_state_lock);
633 return journal_bmap(journal, blocknr, retp);
634}
635
636/*
637 * Conversion of logical to physical block numbers for the journal
638 *
639 * On external journals the journal blocks are identity-mapped, so
640 * this is a no-op. If needed, we can use j_blk_offset - everything is
641 * ready.
642 */
Jan Kara9c28cbc2009-08-03 19:21:00 +0200643int journal_bmap(journal_t *journal, unsigned int blocknr,
644 unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
646 int err = 0;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200647 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649 if (journal->j_inode) {
650 ret = bmap(journal->j_inode, blocknr);
651 if (ret)
652 *retp = ret;
653 else {
654 char b[BDEVNAME_SIZE];
655
656 printk(KERN_ALERT "%s: journal block not found "
Jan Kara9c28cbc2009-08-03 19:21:00 +0200657 "at offset %u on %s\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700658 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 blocknr,
660 bdevname(journal->j_dev, b));
661 err = -EIO;
662 __journal_abort_soft(journal, err);
663 }
664 } else {
665 *retp = blocknr; /* +journal->j_blk_offset */
666 }
667 return err;
668}
669
670/*
671 * We play buffer_head aliasing tricks to write data/metadata blocks to
672 * the journal without copying their contents, but for journal
673 * descriptor blocks we do need to generate bona fide buffers.
674 *
675 * After the caller of journal_get_descriptor_buffer() has finished modifying
676 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
677 * But we don't bother doing that, so there will be coherency problems with
678 * mmaps of blockdevs which hold live JBD-controlled filesystems.
679 */
680struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
681{
682 struct buffer_head *bh;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200683 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 int err;
685
686 err = journal_next_log_block(journal, &blocknr);
687
688 if (err)
689 return NULL;
690
691 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700692 if (!bh)
693 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 lock_buffer(bh);
695 memset(bh->b_data, 0, journal->j_blocksize);
696 set_buffer_uptodate(bh);
697 unlock_buffer(bh);
698 BUFFER_TRACE(bh, "return this buffer");
699 return journal_add_journal_head(bh);
700}
701
702/*
703 * Management for journal control blocks: functions to create and
704 * destroy journal_t structures, and to initialise and read existing
705 * journal blocks from disk. */
706
707/* First: create and setup a journal_t object in memory. We initialise
708 * very few fields yet: that has to wait until we have created the
709 * journal structures from from scratch, or loaded them from disk. */
710
711static journal_t * journal_init_common (void)
712{
713 journal_t *journal;
714 int err;
715
Mingming Cao8c3478a2007-10-18 23:39:20 -0700716 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (!journal)
718 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
720 init_waitqueue_head(&journal->j_wait_transaction_locked);
721 init_waitqueue_head(&journal->j_wait_logspace);
722 init_waitqueue_head(&journal->j_wait_done_commit);
723 init_waitqueue_head(&journal->j_wait_checkpoint);
724 init_waitqueue_head(&journal->j_wait_commit);
725 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800726 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 spin_lock_init(&journal->j_revoke_lock);
728 spin_lock_init(&journal->j_list_lock);
729 spin_lock_init(&journal->j_state_lock);
730
731 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
732
733 /* The journal is marked for error until we succeed with recovery! */
734 journal->j_flags = JFS_ABORT;
735
736 /* Set up a default-sized revoke table for the new mount. */
737 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
738 if (err) {
739 kfree(journal);
740 goto fail;
741 }
742 return journal;
743fail:
744 return NULL;
745}
746
747/* journal_init_dev and journal_init_inode:
748 *
749 * Create a journal structure assigned some fixed set of disk blocks to
750 * the journal. We don't actually touch those disk blocks yet, but we
751 * need to set up all of the mapping information to tell the journaling
752 * system where the journal blocks are.
753 *
754 */
755
756/**
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700757 * journal_t * journal_init_dev() - creates and initialises a journal structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 * @bdev: Block device on which to create the journal
759 * @fs_dev: Device which hold journalled filesystem for this journal.
760 * @start: Block nr Start of journal.
Eric Sandeen37ed3222006-09-27 01:49:31 -0700761 * @len: Length of the journal in blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 * @blocksize: blocksize of journalling device
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700763 *
764 * Returns: a newly created journal_t *
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700765 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 * journal_init_dev creates a journal which maps a fixed contiguous
767 * range of blocks on an arbitrary block device.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700768 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 */
770journal_t * journal_init_dev(struct block_device *bdev,
771 struct block_device *fs_dev,
772 int start, int len, int blocksize)
773{
774 journal_t *journal = journal_init_common();
775 struct buffer_head *bh;
776 int n;
777
778 if (!journal)
779 return NULL;
780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 /* journal descriptor can store up to n blocks -bzzz */
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700782 journal->j_blocksize = blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 n = journal->j_blocksize / sizeof(journal_block_tag_t);
784 journal->j_wbufsize = n;
785 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
786 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300787 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700788 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700789 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700791 journal->j_dev = bdev;
792 journal->j_fs_dev = fs_dev;
793 journal->j_blk_offset = start;
794 journal->j_maxlen = len;
795
796 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700797 if (!bh) {
798 printk(KERN_ERR
799 "%s: Cannot get buffer for journal superblock\n",
800 __func__);
801 goto out_err;
802 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700803 journal->j_sb_buffer = bh;
804 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Karaecca9af2009-04-02 16:57:13 -0700805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700807out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800808 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700809 kfree(journal);
810 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700812
813/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
815 * @inode: An inode to create the journal in
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700816 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 * journal_init_inode creates a journal which maps an on-disk inode as
818 * the journal. The inode must exist already, must support bmap() and
819 * must have all data blocks preallocated.
820 */
821journal_t * journal_init_inode (struct inode *inode)
822{
823 struct buffer_head *bh;
824 journal_t *journal = journal_init_common();
825 int err;
826 int n;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200827 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
829 if (!journal)
830 return NULL;
831
832 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
833 journal->j_inode = inode;
834 jbd_debug(1,
835 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700836 journal, inode->i_sb->s_id, inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 (long long) inode->i_size,
838 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
839
840 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
841 journal->j_blocksize = inode->i_sb->s_blocksize;
842
843 /* journal descriptor can store up to n blocks -bzzz */
844 n = journal->j_blocksize / sizeof(journal_block_tag_t);
845 journal->j_wbufsize = n;
846 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
847 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300848 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700849 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700850 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 }
852
853 err = journal_bmap(journal, 0, &blocknr);
854 /* If that failed, give up */
855 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -0800856 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700857 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700858 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 }
860
861 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700862 if (!bh) {
863 printk(KERN_ERR
864 "%s: Cannot get buffer for journal superblock\n",
865 __func__);
866 goto out_err;
867 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 journal->j_sb_buffer = bh;
869 journal->j_superblock = (journal_superblock_t *)bh->b_data;
870
871 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700872out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800873 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700874 kfree(journal);
875 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876}
877
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700878/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 * If the journal init or create aborts, we need to mark the journal
880 * superblock as being NULL to prevent the journal destroy from writing
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700881 * back a bogus superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 */
883static void journal_fail_superblock (journal_t *journal)
884{
885 struct buffer_head *bh = journal->j_sb_buffer;
886 brelse(bh);
887 journal->j_sb_buffer = NULL;
888}
889
890/*
891 * Given a journal_t structure, initialise the various fields for
892 * startup of a new journaling session. We use this both when creating
893 * a journal, and after recovering an old journal to reset it for
894 * subsequent use.
895 */
896
897static int journal_reset(journal_t *journal)
898{
899 journal_superblock_t *sb = journal->j_superblock;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200900 unsigned int first, last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 first = be32_to_cpu(sb->s_first);
903 last = be32_to_cpu(sb->s_maxlen);
Jan Kara7447a662009-07-15 20:36:08 +0200904 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200905 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
Jan Kara7447a662009-07-15 20:36:08 +0200906 first, last);
907 journal_fail_superblock(journal);
908 return -EINVAL;
909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
911 journal->j_first = first;
912 journal->j_last = last;
913
914 journal->j_head = first;
915 journal->j_tail = first;
916 journal->j_free = last - first;
917
918 journal->j_tail_sequence = journal->j_transaction_sequence;
919 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
920 journal->j_commit_request = journal->j_commit_sequence;
921
922 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
923
924 /* Add the dynamic fields and write it to disk. */
925 journal_update_superblock(journal, 1);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700926 return journal_start_thread(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927}
928
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700929/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 * int journal_create() - Initialise the new journal file
931 * @journal: Journal to create. This structure must have been initialised
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700932 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 * Given a journal_t structure which tells us which disk blocks we can
934 * use, create a new journal superblock and initialise all of the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700935 * journal fields from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 **/
937int journal_create(journal_t *journal)
938{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200939 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 struct buffer_head *bh;
941 journal_superblock_t *sb;
942 int i, err;
943
944 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
945 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
946 journal->j_maxlen);
947 journal_fail_superblock(journal);
948 return -EINVAL;
949 }
950
951 if (journal->j_inode == NULL) {
952 /*
953 * We don't know what block to start at!
954 */
955 printk(KERN_EMERG
956 "%s: creation of journal on external device!\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700957 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 BUG();
959 }
960
961 /* Zero out the entire journal on disk. We cannot afford to
962 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
963 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
964 for (i = 0; i < journal->j_maxlen; i++) {
965 err = journal_bmap(journal, i, &blocknr);
966 if (err)
967 return err;
968 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Namhyung Kim2a0e3382010-10-13 17:17:18 +0900969 if (unlikely(!bh))
970 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 lock_buffer(bh);
972 memset (bh->b_data, 0, journal->j_blocksize);
973 BUFFER_TRACE(bh, "marking dirty");
974 mark_buffer_dirty(bh);
975 BUFFER_TRACE(bh, "marking uptodate");
976 set_buffer_uptodate(bh);
977 unlock_buffer(bh);
978 __brelse(bh);
979 }
980
981 sync_blockdev(journal->j_dev);
982 jbd_debug(1, "JBD: journal cleared.\n");
983
984 /* OK, fill in the initial static fields in the new superblock */
985 sb = journal->j_superblock;
986
987 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
988 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
989
990 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
991 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
992 sb->s_first = cpu_to_be32(1);
993
994 journal->j_transaction_sequence = 1;
995
996 journal->j_flags &= ~JFS_ABORT;
997 journal->j_format_version = 2;
998
999 return journal_reset(journal);
1000}
1001
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001002/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 * void journal_update_superblock() - Update journal sb on disk.
1004 * @journal: The journal to update.
1005 * @wait: Set to '0' if you don't want to wait for IO completion.
1006 *
1007 * Update a journal's dynamic superblock fields and write it to disk,
1008 * optionally waiting for the IO to complete.
1009 */
1010void journal_update_superblock(journal_t *journal, int wait)
1011{
1012 journal_superblock_t *sb = journal->j_superblock;
1013 struct buffer_head *bh = journal->j_sb_buffer;
1014
1015 /*
1016 * As a special case, if the on-disk copy is already marked as needing
1017 * no recovery (s_start == 0) and there are no outstanding transactions
1018 * in the filesystem, then we can safely defer the superblock update
1019 * until the next commit by setting JFS_FLUSHED. This avoids
1020 * attempting a write to a potential-readonly device.
1021 */
1022 if (sb->s_start == 0 && journal->j_tail_sequence ==
1023 journal->j_transaction_sequence) {
1024 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
Jan Kara9c28cbc2009-08-03 19:21:00 +02001025 "(start %u, seq %d, errno %d)\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001026 journal->j_tail, journal->j_tail_sequence,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 journal->j_errno);
1028 goto out;
1029 }
1030
Darrick J. Wongdff68252010-10-04 12:35:05 -07001031 if (buffer_write_io_error(bh)) {
1032 char b[BDEVNAME_SIZE];
1033 /*
1034 * Oh, dear. A previous attempt to write the journal
1035 * superblock failed. This could happen because the
1036 * USB device was yanked out. Or it could happen to
1037 * be a transient write error and maybe the block will
1038 * be remapped. Nothing we can do but to retry the
1039 * write and hope for the best.
1040 */
1041 printk(KERN_ERR "JBD: previous I/O error detected "
1042 "for journal superblock update for %s.\n",
1043 journal_dev_name(journal, b));
1044 clear_buffer_write_io_error(bh);
1045 set_buffer_uptodate(bh);
1046 }
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 spin_lock(&journal->j_state_lock);
Jan Kara9c28cbc2009-08-03 19:21:00 +02001049 jbd_debug(1,"JBD: updating superblock (start %u, seq %d, errno %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1051
1052 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1053 sb->s_start = cpu_to_be32(journal->j_tail);
1054 sb->s_errno = cpu_to_be32(journal->j_errno);
1055 spin_unlock(&journal->j_state_lock);
1056
1057 BUFFER_TRACE(bh, "marking dirty");
1058 mark_buffer_dirty(bh);
Darrick J. Wongdff68252010-10-04 12:35:05 -07001059 if (wait) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 sync_dirty_buffer(bh);
Darrick J. Wongdff68252010-10-04 12:35:05 -07001061 if (buffer_write_io_error(bh)) {
1062 char b[BDEVNAME_SIZE];
1063 printk(KERN_ERR "JBD: I/O error detected "
1064 "when updating journal superblock for %s.\n",
1065 journal_dev_name(journal, b));
1066 clear_buffer_write_io_error(bh);
1067 set_buffer_uptodate(bh);
1068 }
1069 } else
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02001070 write_dirty_buffer(bh, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
Lukas Czerner99cb1a32011-05-23 18:33:02 +02001072 trace_jbd_update_superblock_end(journal, wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073out:
1074 /* If we have just flushed the log (by marking s_start==0), then
1075 * any future commit will have to be careful to update the
1076 * superblock again to re-record the true start of the log. */
1077
1078 spin_lock(&journal->j_state_lock);
1079 if (sb->s_start)
1080 journal->j_flags &= ~JFS_FLUSHED;
1081 else
1082 journal->j_flags |= JFS_FLUSHED;
1083 spin_unlock(&journal->j_state_lock);
1084}
1085
1086/*
1087 * Read the superblock for a given journal, performing initial
1088 * validation of the format.
1089 */
1090
1091static int journal_get_superblock(journal_t *journal)
1092{
1093 struct buffer_head *bh;
1094 journal_superblock_t *sb;
1095 int err = -EIO;
1096
1097 bh = journal->j_sb_buffer;
1098
1099 J_ASSERT(bh != NULL);
1100 if (!buffer_uptodate(bh)) {
1101 ll_rw_block(READ, 1, &bh);
1102 wait_on_buffer(bh);
1103 if (!buffer_uptodate(bh)) {
1104 printk (KERN_ERR
1105 "JBD: IO error reading journal superblock\n");
1106 goto out;
1107 }
1108 }
1109
1110 sb = journal->j_superblock;
1111
1112 err = -EINVAL;
1113
1114 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1115 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1116 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1117 goto out;
1118 }
1119
1120 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1121 case JFS_SUPERBLOCK_V1:
1122 journal->j_format_version = 1;
1123 break;
1124 case JFS_SUPERBLOCK_V2:
1125 journal->j_format_version = 2;
1126 break;
1127 default:
1128 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1129 goto out;
1130 }
1131
1132 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1133 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1134 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1135 printk (KERN_WARNING "JBD: journal file too short\n");
1136 goto out;
1137 }
1138
Eryu Guan87622022011-11-01 19:04:59 -04001139 if (be32_to_cpu(sb->s_first) == 0 ||
1140 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1141 printk(KERN_WARNING
1142 "JBD: Invalid start block of journal: %u\n",
1143 be32_to_cpu(sb->s_first));
1144 goto out;
1145 }
1146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 return 0;
1148
1149out:
1150 journal_fail_superblock(journal);
1151 return err;
1152}
1153
1154/*
1155 * Load the on-disk journal superblock and read the key fields into the
1156 * journal_t.
1157 */
1158
1159static int load_superblock(journal_t *journal)
1160{
1161 int err;
1162 journal_superblock_t *sb;
1163
1164 err = journal_get_superblock(journal);
1165 if (err)
1166 return err;
1167
1168 sb = journal->j_superblock;
1169
1170 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1171 journal->j_tail = be32_to_cpu(sb->s_start);
1172 journal->j_first = be32_to_cpu(sb->s_first);
1173 journal->j_last = be32_to_cpu(sb->s_maxlen);
1174 journal->j_errno = be32_to_cpu(sb->s_errno);
1175
1176 return 0;
1177}
1178
1179
1180/**
1181 * int journal_load() - Read journal from disk.
1182 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001183 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 * Given a journal_t structure which tells us which disk blocks contain
1185 * a journal, read the journal from disk to initialise the in-memory
1186 * structures.
1187 */
1188int journal_load(journal_t *journal)
1189{
1190 int err;
Badari Pulavartyea817392006-08-27 01:23:52 -07001191 journal_superblock_t *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
1193 err = load_superblock(journal);
1194 if (err)
1195 return err;
1196
Badari Pulavartyea817392006-08-27 01:23:52 -07001197 sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 /* If this is a V2 superblock, then we have to check the
1199 * features flags on it. */
1200
1201 if (journal->j_format_version >= 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 if ((sb->s_feature_ro_compat &
1203 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1204 (sb->s_feature_incompat &
1205 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1206 printk (KERN_WARNING
1207 "JBD: Unrecognised features on journal\n");
1208 return -EINVAL;
1209 }
1210 }
1211
1212 /* Let the recovery code check whether it needs to recover any
1213 * data from the journal. */
1214 if (journal_recover(journal))
1215 goto recovery_error;
1216
1217 /* OK, we've finished with the dynamic journal bits:
1218 * reinitialise the dynamic contents of the superblock in memory
1219 * and reset them on disk. */
1220 if (journal_reset(journal))
1221 goto recovery_error;
1222
1223 journal->j_flags &= ~JFS_ABORT;
1224 journal->j_flags |= JFS_LOADED;
1225 return 0;
1226
1227recovery_error:
1228 printk (KERN_WARNING "JBD: recovery failed\n");
1229 return -EIO;
1230}
1231
1232/**
1233 * void journal_destroy() - Release a journal_t structure.
1234 * @journal: Journal to act on.
1235 *
1236 * Release a journal_t structure once it is no longer in use by the
1237 * journaled object.
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001238 * Return <0 if we couldn't clean up the journal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 */
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001240int journal_destroy(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241{
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001242 int err = 0;
1243
Jan Kara03f4d802010-04-15 22:16:24 +02001244
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 /* Wait for the commit thread to wake up and die. */
1246 journal_kill_thread(journal);
1247
1248 /* Force a final log commit */
1249 if (journal->j_running_transaction)
1250 journal_commit_transaction(journal);
1251
1252 /* Force any old transactions to disk */
1253
1254 /* Totally anal locking here... */
1255 spin_lock(&journal->j_list_lock);
1256 while (journal->j_checkpoint_transactions != NULL) {
1257 spin_unlock(&journal->j_list_lock);
1258 log_do_checkpoint(journal);
1259 spin_lock(&journal->j_list_lock);
1260 }
1261
1262 J_ASSERT(journal->j_running_transaction == NULL);
1263 J_ASSERT(journal->j_committing_transaction == NULL);
1264 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1265 spin_unlock(&journal->j_list_lock);
1266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (journal->j_sb_buffer) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001268 if (!is_journal_aborted(journal)) {
1269 /* We can now mark the journal as empty. */
1270 journal->j_tail = 0;
1271 journal->j_tail_sequence =
1272 ++journal->j_transaction_sequence;
1273 journal_update_superblock(journal, 1);
1274 } else {
1275 err = -EIO;
1276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 brelse(journal->j_sb_buffer);
1278 }
1279
1280 if (journal->j_inode)
1281 iput(journal->j_inode);
1282 if (journal->j_revoke)
1283 journal_destroy_revoke(journal);
1284 kfree(journal->j_wbuf);
1285 kfree(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001286
1287 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288}
1289
1290
1291/**
1292 *int journal_check_used_features () - Check if features specified are used.
1293 * @journal: Journal to check.
1294 * @compat: bitmask of compatible features
1295 * @ro: bitmask of features that force read-only mount
1296 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001297 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 * Check whether the journal uses all of a given set of
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001299 * features. Return true (non-zero) if it does.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 **/
1301
1302int journal_check_used_features (journal_t *journal, unsigned long compat,
1303 unsigned long ro, unsigned long incompat)
1304{
1305 journal_superblock_t *sb;
1306
1307 if (!compat && !ro && !incompat)
1308 return 1;
1309 if (journal->j_format_version == 1)
1310 return 0;
1311
1312 sb = journal->j_superblock;
1313
1314 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1315 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1316 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1317 return 1;
1318
1319 return 0;
1320}
1321
1322/**
1323 * int journal_check_available_features() - Check feature set in journalling layer
1324 * @journal: Journal to check.
1325 * @compat: bitmask of compatible features
1326 * @ro: bitmask of features that force read-only mount
1327 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001328 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 * Check whether the journaling code supports the use of
1330 * all of a given set of features on this journal. Return true
1331 * (non-zero) if it can. */
1332
1333int journal_check_available_features (journal_t *journal, unsigned long compat,
1334 unsigned long ro, unsigned long incompat)
1335{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 if (!compat && !ro && !incompat)
1337 return 1;
1338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 /* We can support any known requested features iff the
1340 * superblock is in version 2. Otherwise we fail to support any
1341 * extended sb features. */
1342
1343 if (journal->j_format_version != 2)
1344 return 0;
1345
1346 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1347 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1348 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1349 return 1;
1350
1351 return 0;
1352}
1353
1354/**
1355 * int journal_set_features () - Mark a given journal feature in the superblock
1356 * @journal: Journal to act on.
1357 * @compat: bitmask of compatible features
1358 * @ro: bitmask of features that force read-only mount
1359 * @incompat: bitmask of incompatible features
1360 *
1361 * Mark a given journal feature as present on the
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001362 * superblock. Returns true if the requested features could be set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 *
1364 */
1365
1366int journal_set_features (journal_t *journal, unsigned long compat,
1367 unsigned long ro, unsigned long incompat)
1368{
1369 journal_superblock_t *sb;
1370
1371 if (journal_check_used_features(journal, compat, ro, incompat))
1372 return 1;
1373
1374 if (!journal_check_available_features(journal, compat, ro, incompat))
1375 return 0;
1376
1377 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1378 compat, ro, incompat);
1379
1380 sb = journal->j_superblock;
1381
1382 sb->s_feature_compat |= cpu_to_be32(compat);
1383 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1384 sb->s_feature_incompat |= cpu_to_be32(incompat);
1385
1386 return 1;
1387}
1388
1389
1390/**
1391 * int journal_update_format () - Update on-disk journal structure.
1392 * @journal: Journal to act on.
1393 *
1394 * Given an initialised but unloaded journal struct, poke about in the
1395 * on-disk structure to update it to the most recent supported version.
1396 */
1397int journal_update_format (journal_t *journal)
1398{
1399 journal_superblock_t *sb;
1400 int err;
1401
1402 err = journal_get_superblock(journal);
1403 if (err)
1404 return err;
1405
1406 sb = journal->j_superblock;
1407
1408 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1409 case JFS_SUPERBLOCK_V2:
1410 return 0;
1411 case JFS_SUPERBLOCK_V1:
1412 return journal_convert_superblock_v1(journal, sb);
1413 default:
1414 break;
1415 }
1416 return -EINVAL;
1417}
1418
1419static int journal_convert_superblock_v1(journal_t *journal,
1420 journal_superblock_t *sb)
1421{
1422 int offset, blocksize;
1423 struct buffer_head *bh;
1424
1425 printk(KERN_WARNING
1426 "JBD: Converting superblock from version 1 to 2.\n");
1427
1428 /* Pre-initialise new fields to zero */
1429 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1430 blocksize = be32_to_cpu(sb->s_blocksize);
1431 memset(&sb->s_feature_compat, 0, blocksize-offset);
1432
1433 sb->s_nr_users = cpu_to_be32(1);
1434 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1435 journal->j_format_version = 2;
1436
1437 bh = journal->j_sb_buffer;
1438 BUFFER_TRACE(bh, "marking dirty");
1439 mark_buffer_dirty(bh);
1440 sync_dirty_buffer(bh);
1441 return 0;
1442}
1443
1444
1445/**
1446 * int journal_flush () - Flush journal
1447 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001448 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 * Flush all data for a given journal to disk and empty the journal.
1450 * Filesystems can use this when remounting readonly to ensure that
1451 * recovery does not need to happen on remount.
1452 */
1453
1454int journal_flush(journal_t *journal)
1455{
1456 int err = 0;
1457 transaction_t *transaction = NULL;
Jan Kara9c28cbc2009-08-03 19:21:00 +02001458 unsigned int old_tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
1460 spin_lock(&journal->j_state_lock);
1461
1462 /* Force everything buffered to the log... */
1463 if (journal->j_running_transaction) {
1464 transaction = journal->j_running_transaction;
1465 __log_start_commit(journal, transaction->t_tid);
1466 } else if (journal->j_committing_transaction)
1467 transaction = journal->j_committing_transaction;
1468
1469 /* Wait for the log commit to complete... */
1470 if (transaction) {
1471 tid_t tid = transaction->t_tid;
1472
1473 spin_unlock(&journal->j_state_lock);
1474 log_wait_commit(journal, tid);
1475 } else {
1476 spin_unlock(&journal->j_state_lock);
1477 }
1478
1479 /* ...and flush everything in the log out to disk. */
1480 spin_lock(&journal->j_list_lock);
1481 while (!err && journal->j_checkpoint_transactions != NULL) {
1482 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001483 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 err = log_do_checkpoint(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001485 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 spin_lock(&journal->j_list_lock);
1487 }
1488 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001489
1490 if (is_journal_aborted(journal))
1491 return -EIO;
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 cleanup_journal_tail(journal);
1494
1495 /* Finally, mark the journal as really needing no recovery.
1496 * This sets s_start==0 in the underlying superblock, which is
1497 * the magic code for a fully-recovered superblock. Any future
1498 * commits of data to the journal will restore the current
1499 * s_start value. */
1500 spin_lock(&journal->j_state_lock);
1501 old_tail = journal->j_tail;
1502 journal->j_tail = 0;
1503 spin_unlock(&journal->j_state_lock);
1504 journal_update_superblock(journal, 1);
1505 spin_lock(&journal->j_state_lock);
1506 journal->j_tail = old_tail;
1507
1508 J_ASSERT(!journal->j_running_transaction);
1509 J_ASSERT(!journal->j_committing_transaction);
1510 J_ASSERT(!journal->j_checkpoint_transactions);
1511 J_ASSERT(journal->j_head == journal->j_tail);
1512 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1513 spin_unlock(&journal->j_state_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001514 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515}
1516
1517/**
1518 * int journal_wipe() - Wipe journal contents
1519 * @journal: Journal to act on.
1520 * @write: flag (see below)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001521 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 * Wipe out all of the contents of a journal, safely. This will produce
1523 * a warning if the journal contains any valid recovery information.
1524 * Must be called between journal_init_*() and journal_load().
1525 *
1526 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1527 * we merely suppress recovery.
1528 */
1529
1530int journal_wipe(journal_t *journal, int write)
1531{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 int err = 0;
1533
1534 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1535
1536 err = load_superblock(journal);
1537 if (err)
1538 return err;
1539
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 if (!journal->j_tail)
1541 goto no_recovery;
1542
1543 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1544 write ? "Clearing" : "Ignoring");
1545
1546 err = journal_skip_recovery(journal);
1547 if (write)
1548 journal_update_superblock(journal, 1);
1549
1550 no_recovery:
1551 return err;
1552}
1553
1554/*
1555 * journal_dev_name: format a character string to describe on what
1556 * device this journal is present.
1557 */
1558
Adrian Bunk022a4a72005-09-06 15:16:41 -07001559static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560{
1561 struct block_device *bdev;
1562
1563 if (journal->j_inode)
1564 bdev = journal->j_inode->i_sb->s_bdev;
1565 else
1566 bdev = journal->j_dev;
1567
1568 return bdevname(bdev, buffer);
1569}
1570
1571/*
1572 * Journal abort has very specific semantics, which we describe
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001573 * for journal abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 *
1575 * Two internal function, which provide abort to te jbd layer
1576 * itself are here.
1577 */
1578
1579/*
1580 * Quick version for internal journal use (doesn't lock the journal).
1581 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1582 * and don't attempt to make any other journal updates.
1583 */
Adrian Bunk53308382008-02-06 01:40:12 -08001584static void __journal_abort_hard(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585{
1586 transaction_t *transaction;
1587 char b[BDEVNAME_SIZE];
1588
1589 if (journal->j_flags & JFS_ABORT)
1590 return;
1591
1592 printk(KERN_ERR "Aborting journal on device %s.\n",
1593 journal_dev_name(journal, b));
1594
1595 spin_lock(&journal->j_state_lock);
1596 journal->j_flags |= JFS_ABORT;
1597 transaction = journal->j_running_transaction;
1598 if (transaction)
1599 __log_start_commit(journal, transaction->t_tid);
1600 spin_unlock(&journal->j_state_lock);
1601}
1602
1603/* Soft abort: record the abort error status in the journal superblock,
1604 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001605static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606{
1607 if (journal->j_flags & JFS_ABORT)
1608 return;
1609
1610 if (!journal->j_errno)
1611 journal->j_errno = errno;
1612
1613 __journal_abort_hard(journal);
1614
1615 if (errno)
1616 journal_update_superblock(journal, 1);
1617}
1618
1619/**
1620 * void journal_abort () - Shutdown the journal immediately.
1621 * @journal: the journal to shutdown.
1622 * @errno: an error number to record in the journal indicating
1623 * the reason for the shutdown.
1624 *
1625 * Perform a complete, immediate shutdown of the ENTIRE
1626 * journal (not of a single transaction). This operation cannot be
1627 * undone without closing and reopening the journal.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001628 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 * The journal_abort function is intended to support higher level error
1630 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1631 * mode.
1632 *
1633 * Journal abort has very specific semantics. Any existing dirty,
1634 * unjournaled buffers in the main filesystem will still be written to
1635 * disk by bdflush, but the journaling mechanism will be suspended
1636 * immediately and no further transaction commits will be honoured.
1637 *
1638 * Any dirty, journaled buffers will be written back to disk without
1639 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1640 * filesystem, but we _do_ attempt to leave as much data as possible
1641 * behind for fsck to use for cleanup.
1642 *
1643 * Any attempt to get a new transaction handle on a journal which is in
1644 * ABORT state will just result in an -EROFS error return. A
1645 * journal_stop on an existing handle will return -EIO if we have
1646 * entered abort state during the update.
1647 *
1648 * Recursive transactions are not disturbed by journal abort until the
1649 * final journal_stop, which will receive the -EIO error.
1650 *
1651 * Finally, the journal_abort call allows the caller to supply an errno
1652 * which will be recorded (if possible) in the journal superblock. This
1653 * allows a client to record failure conditions in the middle of a
1654 * transaction without having to complete the transaction to record the
1655 * failure to disk. ext3_error, for example, now uses this
1656 * functionality.
1657 *
1658 * Errors which originate from within the journaling layer will NOT
1659 * supply an errno; a null errno implies that absolutely no further
1660 * writes are done to the journal (unless there are any already in
1661 * progress).
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001662 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 */
1664
1665void journal_abort(journal_t *journal, int errno)
1666{
1667 __journal_abort_soft(journal, errno);
1668}
1669
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001670/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 * int journal_errno () - returns the journal's error state.
1672 * @journal: journal to examine.
1673 *
1674 * This is the errno numbet set with journal_abort(), the last
1675 * time the journal was mounted - if the journal was stopped
1676 * without calling abort this will be 0.
1677 *
1678 * If the journal has been aborted on this mount time -EROFS will
1679 * be returned.
1680 */
1681int journal_errno(journal_t *journal)
1682{
1683 int err;
1684
1685 spin_lock(&journal->j_state_lock);
1686 if (journal->j_flags & JFS_ABORT)
1687 err = -EROFS;
1688 else
1689 err = journal->j_errno;
1690 spin_unlock(&journal->j_state_lock);
1691 return err;
1692}
1693
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001694/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 * int journal_clear_err () - clears the journal's error state
1696 * @journal: journal to act on.
1697 *
1698 * An error must be cleared or Acked to take a FS out of readonly
1699 * mode.
1700 */
1701int journal_clear_err(journal_t *journal)
1702{
1703 int err = 0;
1704
1705 spin_lock(&journal->j_state_lock);
1706 if (journal->j_flags & JFS_ABORT)
1707 err = -EROFS;
1708 else
1709 journal->j_errno = 0;
1710 spin_unlock(&journal->j_state_lock);
1711 return err;
1712}
1713
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001714/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 * void journal_ack_err() - Ack journal err.
1716 * @journal: journal to act on.
1717 *
1718 * An error must be cleared or Acked to take a FS out of readonly
1719 * mode.
1720 */
1721void journal_ack_err(journal_t *journal)
1722{
1723 spin_lock(&journal->j_state_lock);
1724 if (journal->j_errno)
1725 journal->j_flags |= JFS_ACK_ERR;
1726 spin_unlock(&journal->j_state_lock);
1727}
1728
1729int journal_blocks_per_page(struct inode *inode)
1730{
1731 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1732}
1733
1734/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 * Journal_head storage management
1736 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001737static struct kmem_cache *journal_head_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738#ifdef CONFIG_JBD_DEBUG
1739static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1740#endif
1741
1742static int journal_init_journal_head_cache(void)
1743{
1744 int retval;
1745
Al Viro1076d172008-03-29 03:07:18 +00001746 J_ASSERT(journal_head_cache == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 journal_head_cache = kmem_cache_create("journal_head",
1748 sizeof(struct journal_head),
1749 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001750 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001751 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00001753 if (!journal_head_cache) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 retval = -ENOMEM;
1755 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1756 }
1757 return retval;
1758}
1759
1760static void journal_destroy_journal_head_cache(void)
1761{
Duane Griffin3850f7a2008-07-25 01:46:19 -07001762 if (journal_head_cache) {
1763 kmem_cache_destroy(journal_head_cache);
1764 journal_head_cache = NULL;
1765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
1767
1768/*
1769 * journal_head splicing and dicing
1770 */
1771static struct journal_head *journal_alloc_journal_head(void)
1772{
1773 struct journal_head *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
1775#ifdef CONFIG_JBD_DEBUG
1776 atomic_inc(&nr_journal_heads);
1777#endif
1778 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001779 if (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 jbd_debug(1, "out of memory for journal_head\n");
Namhyung Kimf81e3d42010-10-04 19:12:13 +09001781 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1782 __func__);
1783
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001784 while (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 yield();
1786 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1787 }
1788 }
1789 return ret;
1790}
1791
1792static void journal_free_journal_head(struct journal_head *jh)
1793{
1794#ifdef CONFIG_JBD_DEBUG
1795 atomic_dec(&nr_journal_heads);
Randy Dunlapc9cf5522006-06-27 02:53:52 -07001796 memset(jh, JBD_POISON_FREE, sizeof(*jh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797#endif
1798 kmem_cache_free(journal_head_cache, jh);
1799}
1800
1801/*
1802 * A journal_head is attached to a buffer_head whenever JBD has an
1803 * interest in the buffer.
1804 *
1805 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1806 * is set. This bit is tested in core kernel code where we need to take
1807 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1808 * there.
1809 *
1810 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1811 *
1812 * When a buffer has its BH_JBD bit set it is immune from being released by
1813 * core kernel code, mainly via ->b_count.
1814 *
Jan Karabb189242011-06-24 23:11:59 +02001815 * A journal_head is detached from its buffer_head when the journal_head's
1816 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
1817 * transaction (b_cp_transaction) hold their references to b_jcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 *
1819 * Various places in the kernel want to attach a journal_head to a buffer_head
1820 * _before_ attaching the journal_head to a transaction. To protect the
1821 * journal_head in this situation, journal_add_journal_head elevates the
1822 * journal_head's b_jcount refcount by one. The caller must call
1823 * journal_put_journal_head() to undo this.
1824 *
1825 * So the typical usage would be:
1826 *
1827 * (Attach a journal_head if needed. Increments b_jcount)
1828 * struct journal_head *jh = journal_add_journal_head(bh);
1829 * ...
Jan Karabb189242011-06-24 23:11:59 +02001830 * (Get another reference for transaction)
1831 * journal_grab_journal_head(bh);
1832 * jh->b_transaction = xxx;
1833 * (Put original reference)
1834 * journal_put_journal_head(jh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 */
1836
1837/*
1838 * Give a buffer_head a journal_head.
1839 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 * May sleep.
1841 */
1842struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1843{
1844 struct journal_head *jh;
1845 struct journal_head *new_jh = NULL;
1846
1847repeat:
1848 if (!buffer_jbd(bh)) {
1849 new_jh = journal_alloc_journal_head();
1850 memset(new_jh, 0, sizeof(*new_jh));
1851 }
1852
1853 jbd_lock_bh_journal_head(bh);
1854 if (buffer_jbd(bh)) {
1855 jh = bh2jh(bh);
1856 } else {
1857 J_ASSERT_BH(bh,
1858 (atomic_read(&bh->b_count) > 0) ||
1859 (bh->b_page && bh->b_page->mapping));
1860
1861 if (!new_jh) {
1862 jbd_unlock_bh_journal_head(bh);
1863 goto repeat;
1864 }
1865
1866 jh = new_jh;
1867 new_jh = NULL; /* We consumed it */
1868 set_buffer_jbd(bh);
1869 bh->b_private = jh;
1870 jh->b_bh = bh;
1871 get_bh(bh);
1872 BUFFER_TRACE(bh, "added journal_head");
1873 }
1874 jh->b_jcount++;
1875 jbd_unlock_bh_journal_head(bh);
1876 if (new_jh)
1877 journal_free_journal_head(new_jh);
1878 return bh->b_private;
1879}
1880
1881/*
1882 * Grab a ref against this buffer_head's journal_head. If it ended up not
1883 * having a journal_head, return NULL
1884 */
1885struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1886{
1887 struct journal_head *jh = NULL;
1888
1889 jbd_lock_bh_journal_head(bh);
1890 if (buffer_jbd(bh)) {
1891 jh = bh2jh(bh);
1892 jh->b_jcount++;
1893 }
1894 jbd_unlock_bh_journal_head(bh);
1895 return jh;
1896}
1897
1898static void __journal_remove_journal_head(struct buffer_head *bh)
1899{
1900 struct journal_head *jh = bh2jh(bh);
1901
1902 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karabb189242011-06-24 23:11:59 +02001903 J_ASSERT_JH(jh, jh->b_transaction == NULL);
1904 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1905 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
1906 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1907 J_ASSERT_BH(bh, buffer_jbd(bh));
1908 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1909 BUFFER_TRACE(bh, "remove journal_head");
1910 if (jh->b_frozen_data) {
1911 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
1912 jbd_free(jh->b_frozen_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 }
Jan Karabb189242011-06-24 23:11:59 +02001914 if (jh->b_committed_data) {
1915 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
1916 jbd_free(jh->b_committed_data, bh->b_size);
1917 }
1918 bh->b_private = NULL;
1919 jh->b_bh = NULL; /* debug, really */
1920 clear_buffer_jbd(bh);
1921 journal_free_journal_head(jh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922}
1923
1924/*
Jan Karabb189242011-06-24 23:11:59 +02001925 * Drop a reference on the passed journal_head. If it fell to zero then
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 * 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;
Jan Karabb189242011-06-24 23:11:59 +02001935 if (!jh->b_jcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 __journal_remove_journal_head(bh);
Jan Karabb189242011-06-24 23:11:59 +02001937 jbd_unlock_bh_journal_head(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 __brelse(bh);
Jan Karabb189242011-06-24 23:11:59 +02001939 } else
1940 jbd_unlock_bh_journal_head(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941}
1942
1943/*
Jose R. Santosc2a91592007-10-18 23:39:22 -07001944 * debugfs tunables
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 */
Jose R. Santosc2a91592007-10-18 23:39:22 -07001946#ifdef CONFIG_JBD_DEBUG
1947
1948u8 journal_enable_debug __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949EXPORT_SYMBOL(journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
Jose R. Santosc2a91592007-10-18 23:39:22 -07001951static struct dentry *jbd_debugfs_dir;
1952static struct dentry *jbd_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
Jose R. Santosc2a91592007-10-18 23:39:22 -07001954static void __init jbd_create_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001956 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
1957 if (jbd_debugfs_dir)
Yin Kangkai765f8362009-12-15 14:48:25 -08001958 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
Jose R. Santosc2a91592007-10-18 23:39:22 -07001959 jbd_debugfs_dir,
1960 &journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961}
1962
Jose R. Santosc2a91592007-10-18 23:39:22 -07001963static void __exit jbd_remove_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001965 debugfs_remove(jbd_debug);
1966 debugfs_remove(jbd_debugfs_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967}
1968
1969#else
1970
Jose R. Santosc2a91592007-10-18 23:39:22 -07001971static inline void jbd_create_debugfs_entry(void)
1972{
1973}
1974
1975static inline void jbd_remove_debugfs_entry(void)
1976{
1977}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
1979#endif
1980
Christoph Lametere18b8902006-12-06 20:33:20 -08001981struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
1983static int __init journal_init_handle_cache(void)
1984{
1985 jbd_handle_cache = kmem_cache_create("journal_handle",
1986 sizeof(handle_t),
1987 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001988 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001989 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 if (jbd_handle_cache == NULL) {
1991 printk(KERN_EMERG "JBD: failed to create handle cache\n");
1992 return -ENOMEM;
1993 }
1994 return 0;
1995}
1996
1997static void journal_destroy_handle_cache(void)
1998{
1999 if (jbd_handle_cache)
2000 kmem_cache_destroy(jbd_handle_cache);
2001}
2002
2003/*
2004 * Module startup and shutdown
2005 */
2006
2007static int __init journal_init_caches(void)
2008{
2009 int ret;
2010
2011 ret = journal_init_revoke_caches();
2012 if (ret == 0)
2013 ret = journal_init_journal_head_cache();
2014 if (ret == 0)
2015 ret = journal_init_handle_cache();
2016 return ret;
2017}
2018
2019static void journal_destroy_caches(void)
2020{
2021 journal_destroy_revoke_caches();
2022 journal_destroy_journal_head_cache();
2023 journal_destroy_handle_cache();
2024}
2025
2026static int __init journal_init(void)
2027{
2028 int ret;
2029
Alexey Dobriyan2aed3482006-09-27 01:49:28 -07002030 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
Adrian Bunk022a4a72005-09-06 15:16:41 -07002031
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 ret = journal_init_caches();
2033 if (ret != 0)
2034 journal_destroy_caches();
Jose R. Santosc2a91592007-10-18 23:39:22 -07002035 jbd_create_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 return ret;
2037}
2038
2039static void __exit journal_exit(void)
2040{
2041#ifdef CONFIG_JBD_DEBUG
2042 int n = atomic_read(&nr_journal_heads);
2043 if (n)
2044 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2045#endif
Jose R. Santosc2a91592007-10-18 23:39:22 -07002046 jbd_remove_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 journal_destroy_caches();
2048}
2049
2050MODULE_LICENSE("GPL");
2051module_init(journal_init);
2052module_exit(journal_exit);
2053