blob: fea8dd661d2bbca7820013f10ef34baf87bb1190 [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
132 /* Record that the journal thread is running */
133 journal->j_task = current;
134 wake_up(&journal->j_wait_done_commit);
135
136 printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n",
137 journal->j_commit_interval / HZ);
138
139 /*
140 * And now, wait forever for commit wakeup events.
141 */
142 spin_lock(&journal->j_state_lock);
143
144loop:
145 if (journal->j_flags & JFS_UNMOUNT)
146 goto end_loop;
147
148 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
149 journal->j_commit_sequence, journal->j_commit_request);
150
151 if (journal->j_commit_sequence != journal->j_commit_request) {
152 jbd_debug(1, "OK, requests differ\n");
153 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800154 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 journal_commit_transaction(journal);
156 spin_lock(&journal->j_state_lock);
157 goto loop;
158 }
159
160 wake_up(&journal->j_wait_done_commit);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700161 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 /*
163 * The simpler the better. Flushing journal isn't a
164 * good idea, because that depends on threads that may
165 * be already stopped.
166 */
167 jbd_debug(1, "Now suspending kjournald\n");
168 spin_unlock(&journal->j_state_lock);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -0700169 refrigerator();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 spin_lock(&journal->j_state_lock);
171 } else {
172 /*
173 * We assume on resume that commits are already there,
174 * so we don't sleep
175 */
176 DEFINE_WAIT(wait);
177 int should_sleep = 1;
178
179 prepare_to_wait(&journal->j_wait_commit, &wait,
180 TASK_INTERRUPTIBLE);
181 if (journal->j_commit_sequence != journal->j_commit_request)
182 should_sleep = 0;
183 transaction = journal->j_running_transaction;
184 if (transaction && time_after_eq(jiffies,
185 transaction->t_expires))
186 should_sleep = 0;
Mark Fashehcbf0d272005-09-06 15:19:08 -0700187 if (journal->j_flags & JFS_UNMOUNT)
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700188 should_sleep = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 if (should_sleep) {
190 spin_unlock(&journal->j_state_lock);
191 schedule();
192 spin_lock(&journal->j_state_lock);
193 }
194 finish_wait(&journal->j_wait_commit, &wait);
195 }
196
197 jbd_debug(1, "kjournald wakes\n");
198
199 /*
200 * Were we woken up by a commit wakeup event?
201 */
202 transaction = journal->j_running_transaction;
203 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
204 journal->j_commit_request = transaction->t_tid;
205 jbd_debug(1, "woke because of timeout\n");
206 }
207 goto loop;
208
209end_loop:
210 spin_unlock(&journal->j_state_lock);
Andrew Mortone3df1892006-03-25 03:06:53 -0800211 del_timer_sync(&journal->j_commit_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 journal->j_task = NULL;
213 wake_up(&journal->j_wait_done_commit);
214 jbd_debug(1, "Journal thread exiting.\n");
215 return 0;
216}
217
Pavel Emelianov97f06782007-05-08 00:30:42 -0700218static int journal_start_thread(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700220 struct task_struct *t;
221
222 t = kthread_run(kjournald, journal, "kjournald");
223 if (IS_ERR(t))
224 return PTR_ERR(t);
225
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700226 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700227 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228}
229
230static void journal_kill_thread(journal_t *journal)
231{
232 spin_lock(&journal->j_state_lock);
233 journal->j_flags |= JFS_UNMOUNT;
234
235 while (journal->j_task) {
236 wake_up(&journal->j_wait_commit);
237 spin_unlock(&journal->j_state_lock);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700238 wait_event(journal->j_wait_done_commit,
239 journal->j_task == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 spin_lock(&journal->j_state_lock);
241 }
242 spin_unlock(&journal->j_state_lock);
243}
244
245/*
246 * journal_write_metadata_buffer: write a metadata buffer to the journal.
247 *
248 * Writes a metadata buffer to a given disk block. The actual IO is not
249 * performed but a new buffer_head is constructed which labels the data
250 * to be written with the correct destination disk block.
251 *
252 * Any magic-number escaping which needs to be done will cause a
253 * copy-out here. If the buffer happens to start with the
254 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
255 * magic number is only written to the log for descripter blocks. In
256 * this case, we copy the data and replace the first word with 0, and we
257 * return a result code which indicates that this buffer needs to be
258 * marked as an escaped buffer in the corresponding log descriptor
259 * block. The missing word can then be restored when the block is read
260 * during recovery.
261 *
262 * If the source buffer has already been modified by a new transaction
263 * since we took the last commit snapshot, we use the frozen copy of
264 * that data for IO. If we end up using the existing buffer_head's data
265 * for the write, then we *have* to lock the buffer to prevent anyone
266 * else from using and possibly modifying it while the IO is in
267 * progress.
268 *
269 * The function returns a pointer to the buffer_heads to be used for IO.
270 *
271 * We assume that the journal has already been locked in this function.
272 *
273 * Return value:
274 * <0: Error
275 * >=0: Finished OK
276 *
277 * On success:
278 * Bit 0 set == escape performed on the data
279 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
280 */
281
282int journal_write_metadata_buffer(transaction_t *transaction,
283 struct journal_head *jh_in,
284 struct journal_head **jh_out,
Jan Kara9c28cbc2009-08-03 19:21:00 +0200285 unsigned int blocknr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
287 int need_copy_out = 0;
288 int done_copy_out = 0;
289 int do_escape = 0;
290 char *mapped_data;
291 struct buffer_head *new_bh;
292 struct journal_head *new_jh;
293 struct page *new_page;
294 unsigned int new_offset;
295 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghuaf1015c42009-07-15 21:42:05 +0200296 journal_t *journal = transaction->t_journal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
298 /*
299 * The buffer really shouldn't be locked: only the current committing
300 * transaction is allowed to write it, so nobody else is allowed
301 * to do any IO.
302 *
303 * akpm: except if we're journalling data, and write() output is
304 * also part of a shared mapping, and another thread has
305 * decided to launch a writepage() against this buffer.
306 */
307 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
308
309 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
dingdinghuaf1015c42009-07-15 21:42:05 +0200310 /* keep subsequent assertions sane */
311 new_bh->b_state = 0;
312 init_buffer(new_bh, NULL, NULL);
313 atomic_set(&new_bh->b_count, 1);
314 new_jh = journal_add_journal_head(new_bh); /* This sleeps */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316 /*
317 * If a new transaction has already done a buffer copy-out, then
318 * we use that version of the data for the commit.
319 */
320 jbd_lock_bh_state(bh_in);
321repeat:
322 if (jh_in->b_frozen_data) {
323 done_copy_out = 1;
324 new_page = virt_to_page(jh_in->b_frozen_data);
325 new_offset = offset_in_page(jh_in->b_frozen_data);
326 } else {
327 new_page = jh2bh(jh_in)->b_page;
328 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
329 }
330
331 mapped_data = kmap_atomic(new_page, KM_USER0);
332 /*
333 * Check for escaping
334 */
335 if (*((__be32 *)(mapped_data + new_offset)) ==
336 cpu_to_be32(JFS_MAGIC_NUMBER)) {
337 need_copy_out = 1;
338 do_escape = 1;
339 }
340 kunmap_atomic(mapped_data, KM_USER0);
341
342 /*
343 * Do we need to do a data copy?
344 */
345 if (need_copy_out && !done_copy_out) {
346 char *tmp;
347
348 jbd_unlock_bh_state(bh_in);
Mingming Caoc089d492007-10-16 18:38:25 -0400349 tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 jbd_lock_bh_state(bh_in);
351 if (jh_in->b_frozen_data) {
Mingming Caoc089d492007-10-16 18:38:25 -0400352 jbd_free(tmp, bh_in->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 goto repeat;
354 }
355
356 jh_in->b_frozen_data = tmp;
357 mapped_data = kmap_atomic(new_page, KM_USER0);
358 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
359 kunmap_atomic(mapped_data, KM_USER0);
360
361 new_page = virt_to_page(tmp);
362 new_offset = offset_in_page(tmp);
363 done_copy_out = 1;
364 }
365
366 /*
367 * Did we need to do an escaping? Now we've done all the
368 * copying, we can finally do so.
369 */
370 if (do_escape) {
371 mapped_data = kmap_atomic(new_page, KM_USER0);
372 *((unsigned int *)(mapped_data + new_offset)) = 0;
373 kunmap_atomic(mapped_data, KM_USER0);
374 }
375
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 set_bh_page(new_bh, new_page, new_offset);
377 new_jh->b_transaction = NULL;
378 new_bh->b_size = jh2bh(jh_in)->b_size;
379 new_bh->b_bdev = transaction->t_journal->j_dev;
380 new_bh->b_blocknr = blocknr;
381 set_buffer_mapped(new_bh);
382 set_buffer_dirty(new_bh);
383
384 *jh_out = new_jh;
385
386 /*
387 * The to-be-written buffer needs to get moved to the io queue,
388 * and the original buffer whose contents we are shadowing or
389 * copying is moved to the transaction's shadow queue.
390 */
391 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghuaf1015c42009-07-15 21:42:05 +0200392 spin_lock(&journal->j_list_lock);
393 __journal_file_buffer(jh_in, transaction, BJ_Shadow);
394 spin_unlock(&journal->j_list_lock);
395 jbd_unlock_bh_state(bh_in);
396
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 JBUFFER_TRACE(new_jh, "file as BJ_IO");
398 journal_file_buffer(new_jh, transaction, BJ_IO);
399
400 return do_escape | (done_copy_out << 1);
401}
402
403/*
404 * Allocation code for the journal file. Manage the space left in the
405 * journal, so that we can begin checkpointing when appropriate.
406 */
407
408/*
409 * __log_space_left: Return the number of free blocks left in the journal.
410 *
411 * Called with the journal already locked.
412 *
413 * Called under j_state_lock
414 */
415
416int __log_space_left(journal_t *journal)
417{
418 int left = journal->j_free;
419
420 assert_spin_locked(&journal->j_state_lock);
421
422 /*
423 * Be pessimistic here about the number of those free blocks which
424 * might be required for log descriptor control blocks.
425 */
426
427#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
428
429 left -= MIN_LOG_RESERVED_BLOCKS;
430
431 if (left <= 0)
432 return 0;
433 left -= (left >> 3);
434 return left;
435}
436
437/*
Jan Kara8fe4cd02009-02-11 13:04:25 -0800438 * Called under j_state_lock. Returns true if a transaction commit was started.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 */
440int __log_start_commit(journal_t *journal, tid_t target)
441{
442 /*
Ted Ts'od9b01932011-04-30 13:17:11 -0400443 * The only transaction we can possibly wait upon is the
444 * currently running transaction (if it exists). Otherwise,
445 * the target tid must be an old one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 */
Ted Ts'od9b01932011-04-30 13:17:11 -0400447 if (journal->j_running_transaction &&
448 journal->j_running_transaction->t_tid == target) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200450 * We want a new commit: OK, mark the request and wakeup the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 * commit thread. We do _not_ do the commit ourselves.
452 */
453
454 journal->j_commit_request = target;
455 jbd_debug(1, "JBD: requesting commit %d/%d\n",
456 journal->j_commit_request,
457 journal->j_commit_sequence);
458 wake_up(&journal->j_wait_commit);
459 return 1;
Ted Ts'od9b01932011-04-30 13:17:11 -0400460 } else if (!tid_geq(journal->j_commit_request, target))
461 /* This should never happen, but if it does, preserve
462 the evidence before kjournald goes into a loop and
463 increments j_commit_sequence beyond all recognition. */
464 WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
465 journal->j_commit_request, journal->j_commit_sequence,
466 target, journal->j_running_transaction ?
467 journal->j_running_transaction->t_tid : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 return 0;
469}
470
471int log_start_commit(journal_t *journal, tid_t tid)
472{
473 int ret;
474
475 spin_lock(&journal->j_state_lock);
476 ret = __log_start_commit(journal, tid);
477 spin_unlock(&journal->j_state_lock);
478 return ret;
479}
480
481/*
482 * Force and wait upon a commit if the calling process is not within
483 * transaction. This is used for forcing out undo-protected data which contains
484 * bitmaps, when the fs is running out of space.
485 *
486 * We can only force the running transaction if we don't have an active handle;
487 * otherwise, we will deadlock.
488 *
489 * Returns true if a transaction was started.
490 */
491int journal_force_commit_nested(journal_t *journal)
492{
493 transaction_t *transaction = NULL;
494 tid_t tid;
495
496 spin_lock(&journal->j_state_lock);
497 if (journal->j_running_transaction && !current->journal_info) {
498 transaction = journal->j_running_transaction;
499 __log_start_commit(journal, transaction->t_tid);
500 } else if (journal->j_committing_transaction)
501 transaction = journal->j_committing_transaction;
502
503 if (!transaction) {
504 spin_unlock(&journal->j_state_lock);
505 return 0; /* Nothing to retry */
506 }
507
508 tid = transaction->t_tid;
509 spin_unlock(&journal->j_state_lock);
510 log_wait_commit(journal, tid);
511 return 1;
512}
513
514/*
515 * Start a commit of the current running transaction (if any). Returns true
Jan Kara8fe4cd02009-02-11 13:04:25 -0800516 * if a transaction is going to be committed (or is currently already
517 * committing), and fills its tid in at *ptid
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 */
519int journal_start_commit(journal_t *journal, tid_t *ptid)
520{
521 int ret = 0;
522
523 spin_lock(&journal->j_state_lock);
524 if (journal->j_running_transaction) {
525 tid_t tid = journal->j_running_transaction->t_tid;
526
Jan Kara8fe4cd02009-02-11 13:04:25 -0800527 __log_start_commit(journal, tid);
528 /* There's a running transaction and we've just made sure
529 * it's commit has been scheduled. */
530 if (ptid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 *ptid = tid;
Jan Kara8fe4cd02009-02-11 13:04:25 -0800532 ret = 1;
533 } else if (journal->j_committing_transaction) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 /*
535 * If ext3_write_super() recently started a commit, then we
536 * have to wait for completion of that transaction
537 */
Jan Kara8fe4cd02009-02-11 13:04:25 -0800538 if (ptid)
539 *ptid = journal->j_committing_transaction->t_tid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 ret = 1;
541 }
542 spin_unlock(&journal->j_state_lock);
543 return ret;
544}
545
546/*
547 * Wait for a specified commit to complete.
548 * The caller may not hold the journal lock.
549 */
550int log_wait_commit(journal_t *journal, tid_t tid)
551{
552 int err = 0;
553
554#ifdef CONFIG_JBD_DEBUG
555 spin_lock(&journal->j_state_lock);
556 if (!tid_geq(journal->j_commit_request, tid)) {
557 printk(KERN_EMERG
558 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700559 __func__, journal->j_commit_request, tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 }
561 spin_unlock(&journal->j_state_lock);
562#endif
563 spin_lock(&journal->j_state_lock);
564 while (tid_gt(tid, journal->j_commit_sequence)) {
565 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
566 tid, journal->j_commit_sequence);
567 wake_up(&journal->j_wait_commit);
568 spin_unlock(&journal->j_state_lock);
569 wait_event(journal->j_wait_done_commit,
570 !tid_gt(tid, journal->j_commit_sequence));
571 spin_lock(&journal->j_state_lock);
572 }
573 spin_unlock(&journal->j_state_lock);
574
575 if (unlikely(is_journal_aborted(journal))) {
576 printk(KERN_EMERG "journal commit I/O error\n");
577 err = -EIO;
578 }
579 return err;
580}
581
582/*
Jan Kara03f4d802010-04-15 22:16:24 +0200583 * Return 1 if a given transaction has not yet sent barrier request
584 * connected with a transaction commit. If 0 is returned, transaction
585 * may or may not have sent the barrier. Used to avoid sending barrier
586 * twice in common cases.
587 */
588int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
589{
590 int ret = 0;
591 transaction_t *commit_trans;
592
593 if (!(journal->j_flags & JFS_BARRIER))
594 return 0;
595 spin_lock(&journal->j_state_lock);
596 /* Transaction already committed? */
597 if (tid_geq(journal->j_commit_sequence, tid))
598 goto out;
599 /*
600 * Transaction is being committed and we already proceeded to
601 * writing commit record?
602 */
603 commit_trans = journal->j_committing_transaction;
604 if (commit_trans && commit_trans->t_tid == tid &&
605 commit_trans->t_state >= T_COMMIT_RECORD)
606 goto out;
607 ret = 1;
608out:
609 spin_unlock(&journal->j_state_lock);
610 return ret;
611}
Jan Kara52779702010-04-15 22:24:26 +0200612EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
Jan Kara03f4d802010-04-15 22:16:24 +0200613
614/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 * Log buffer allocation routines:
616 */
617
Jan Kara9c28cbc2009-08-03 19:21:00 +0200618int journal_next_log_block(journal_t *journal, unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200620 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622 spin_lock(&journal->j_state_lock);
623 J_ASSERT(journal->j_free > 1);
624
625 blocknr = journal->j_head;
626 journal->j_head++;
627 journal->j_free--;
628 if (journal->j_head == journal->j_last)
629 journal->j_head = journal->j_first;
630 spin_unlock(&journal->j_state_lock);
631 return journal_bmap(journal, blocknr, retp);
632}
633
634/*
635 * Conversion of logical to physical block numbers for the journal
636 *
637 * On external journals the journal blocks are identity-mapped, so
638 * this is a no-op. If needed, we can use j_blk_offset - everything is
639 * ready.
640 */
Jan Kara9c28cbc2009-08-03 19:21:00 +0200641int journal_bmap(journal_t *journal, unsigned int blocknr,
642 unsigned int *retp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
644 int err = 0;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200645 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647 if (journal->j_inode) {
648 ret = bmap(journal->j_inode, blocknr);
649 if (ret)
650 *retp = ret;
651 else {
652 char b[BDEVNAME_SIZE];
653
654 printk(KERN_ALERT "%s: journal block not found "
Jan Kara9c28cbc2009-08-03 19:21:00 +0200655 "at offset %u on %s\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700656 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 blocknr,
658 bdevname(journal->j_dev, b));
659 err = -EIO;
660 __journal_abort_soft(journal, err);
661 }
662 } else {
663 *retp = blocknr; /* +journal->j_blk_offset */
664 }
665 return err;
666}
667
668/*
669 * We play buffer_head aliasing tricks to write data/metadata blocks to
670 * the journal without copying their contents, but for journal
671 * descriptor blocks we do need to generate bona fide buffers.
672 *
673 * After the caller of journal_get_descriptor_buffer() has finished modifying
674 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
675 * But we don't bother doing that, so there will be coherency problems with
676 * mmaps of blockdevs which hold live JBD-controlled filesystems.
677 */
678struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
679{
680 struct buffer_head *bh;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200681 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 int err;
683
684 err = journal_next_log_block(journal, &blocknr);
685
686 if (err)
687 return NULL;
688
689 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700690 if (!bh)
691 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 lock_buffer(bh);
693 memset(bh->b_data, 0, journal->j_blocksize);
694 set_buffer_uptodate(bh);
695 unlock_buffer(bh);
696 BUFFER_TRACE(bh, "return this buffer");
697 return journal_add_journal_head(bh);
698}
699
700/*
701 * Management for journal control blocks: functions to create and
702 * destroy journal_t structures, and to initialise and read existing
703 * journal blocks from disk. */
704
705/* First: create and setup a journal_t object in memory. We initialise
706 * very few fields yet: that has to wait until we have created the
707 * journal structures from from scratch, or loaded them from disk. */
708
709static journal_t * journal_init_common (void)
710{
711 journal_t *journal;
712 int err;
713
Mingming Cao8c3478a2007-10-18 23:39:20 -0700714 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 if (!journal)
716 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
718 init_waitqueue_head(&journal->j_wait_transaction_locked);
719 init_waitqueue_head(&journal->j_wait_logspace);
720 init_waitqueue_head(&journal->j_wait_done_commit);
721 init_waitqueue_head(&journal->j_wait_checkpoint);
722 init_waitqueue_head(&journal->j_wait_commit);
723 init_waitqueue_head(&journal->j_wait_updates);
Arjan van de Ven2c68ee72006-03-23 03:00:35 -0800724 mutex_init(&journal->j_barrier);
725 mutex_init(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 spin_lock_init(&journal->j_revoke_lock);
727 spin_lock_init(&journal->j_list_lock);
728 spin_lock_init(&journal->j_state_lock);
729
730 journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
731
732 /* The journal is marked for error until we succeed with recovery! */
733 journal->j_flags = JFS_ABORT;
734
735 /* Set up a default-sized revoke table for the new mount. */
736 err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
737 if (err) {
738 kfree(journal);
739 goto fail;
740 }
741 return journal;
742fail:
743 return NULL;
744}
745
746/* journal_init_dev and journal_init_inode:
747 *
748 * Create a journal structure assigned some fixed set of disk blocks to
749 * the journal. We don't actually touch those disk blocks yet, but we
750 * need to set up all of the mapping information to tell the journaling
751 * system where the journal blocks are.
752 *
753 */
754
755/**
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700756 * journal_t * journal_init_dev() - creates and initialises a journal structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 * @bdev: Block device on which to create the journal
758 * @fs_dev: Device which hold journalled filesystem for this journal.
759 * @start: Block nr Start of journal.
Eric Sandeen37ed3222006-09-27 01:49:31 -0700760 * @len: Length of the journal in blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 * @blocksize: blocksize of journalling device
Randy Dunlap0cf01f62008-03-19 17:00:44 -0700762 *
763 * Returns: a newly created journal_t *
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700764 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 * journal_init_dev creates a journal which maps a fixed contiguous
766 * range of blocks on an arbitrary block device.
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700767 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 */
769journal_t * journal_init_dev(struct block_device *bdev,
770 struct block_device *fs_dev,
771 int start, int len, int blocksize)
772{
773 journal_t *journal = journal_init_common();
774 struct buffer_head *bh;
775 int n;
776
777 if (!journal)
778 return NULL;
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 /* journal descriptor can store up to n blocks -bzzz */
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700781 journal->j_blocksize = blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 n = journal->j_blocksize / sizeof(journal_block_tag_t);
783 journal->j_wbufsize = n;
784 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
785 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300786 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700787 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700788 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700790 journal->j_dev = bdev;
791 journal->j_fs_dev = fs_dev;
792 journal->j_blk_offset = start;
793 journal->j_maxlen = len;
794
795 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700796 if (!bh) {
797 printk(KERN_ERR
798 "%s: Cannot get buffer for journal superblock\n",
799 __func__);
800 goto out_err;
801 }
Zoltan Menyhartd1807792006-09-29 01:58:40 -0700802 journal->j_sb_buffer = bh;
803 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Karaecca9af2009-04-02 16:57:13 -0700804
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700806out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800807 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700808 kfree(journal);
809 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700811
812/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 * journal_t * journal_init_inode () - creates a journal which maps to a inode.
814 * @inode: An inode to create the journal in
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700815 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 * journal_init_inode creates a journal which maps an on-disk inode as
817 * the journal. The inode must exist already, must support bmap() and
818 * must have all data blocks preallocated.
819 */
820journal_t * journal_init_inode (struct inode *inode)
821{
822 struct buffer_head *bh;
823 journal_t *journal = journal_init_common();
824 int err;
825 int n;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200826 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
828 if (!journal)
829 return NULL;
830
831 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
832 journal->j_inode = inode;
833 jbd_debug(1,
834 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700835 journal, inode->i_sb->s_id, inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 (long long) inode->i_size,
837 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
838
839 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
840 journal->j_blocksize = inode->i_sb->s_blocksize;
841
842 /* journal descriptor can store up to n blocks -bzzz */
843 n = journal->j_blocksize / sizeof(journal_block_tag_t);
844 journal->j_wbufsize = n;
845 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
846 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300847 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700848 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700849 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
851
852 err = journal_bmap(journal, 0, &blocknr);
853 /* If that failed, give up */
854 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -0800855 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700856 __func__);
Jan Karaecca9af2009-04-02 16:57:13 -0700857 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 }
859
860 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Karaecca9af2009-04-02 16:57:13 -0700861 if (!bh) {
862 printk(KERN_ERR
863 "%s: Cannot get buffer for journal superblock\n",
864 __func__);
865 goto out_err;
866 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 journal->j_sb_buffer = bh;
868 journal->j_superblock = (journal_superblock_t *)bh->b_data;
869
870 return journal;
Jan Karaecca9af2009-04-02 16:57:13 -0700871out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800872 kfree(journal->j_wbuf);
Jan Karaecca9af2009-04-02 16:57:13 -0700873 kfree(journal);
874 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700877/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 * If the journal init or create aborts, we need to mark the journal
879 * superblock as being NULL to prevent the journal destroy from writing
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700880 * back a bogus superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 */
882static void journal_fail_superblock (journal_t *journal)
883{
884 struct buffer_head *bh = journal->j_sb_buffer;
885 brelse(bh);
886 journal->j_sb_buffer = NULL;
887}
888
889/*
890 * Given a journal_t structure, initialise the various fields for
891 * startup of a new journaling session. We use this both when creating
892 * a journal, and after recovering an old journal to reset it for
893 * subsequent use.
894 */
895
896static int journal_reset(journal_t *journal)
897{
898 journal_superblock_t *sb = journal->j_superblock;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200899 unsigned int first, last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901 first = be32_to_cpu(sb->s_first);
902 last = be32_to_cpu(sb->s_maxlen);
Jan Kara7447a662009-07-15 20:36:08 +0200903 if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200904 printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
Jan Kara7447a662009-07-15 20:36:08 +0200905 first, last);
906 journal_fail_superblock(journal);
907 return -EINVAL;
908 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
910 journal->j_first = first;
911 journal->j_last = last;
912
913 journal->j_head = first;
914 journal->j_tail = first;
915 journal->j_free = last - first;
916
917 journal->j_tail_sequence = journal->j_transaction_sequence;
918 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
919 journal->j_commit_request = journal->j_commit_sequence;
920
921 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
922
923 /* Add the dynamic fields and write it to disk. */
924 journal_update_superblock(journal, 1);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700925 return journal_start_thread(journal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926}
927
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700928/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 * int journal_create() - Initialise the new journal file
930 * @journal: Journal to create. This structure must have been initialised
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700931 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 * Given a journal_t structure which tells us which disk blocks we can
933 * use, create a new journal superblock and initialise all of the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700934 * journal fields from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 **/
936int journal_create(journal_t *journal)
937{
Jan Kara9c28cbc2009-08-03 19:21:00 +0200938 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 struct buffer_head *bh;
940 journal_superblock_t *sb;
941 int i, err;
942
943 if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
944 printk (KERN_ERR "Journal length (%d blocks) too short.\n",
945 journal->j_maxlen);
946 journal_fail_superblock(journal);
947 return -EINVAL;
948 }
949
950 if (journal->j_inode == NULL) {
951 /*
952 * We don't know what block to start at!
953 */
954 printk(KERN_EMERG
955 "%s: creation of journal on external device!\n",
Harvey Harrison08fc99b2008-04-28 02:16:16 -0700956 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 BUG();
958 }
959
960 /* Zero out the entire journal on disk. We cannot afford to
961 have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
962 jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
963 for (i = 0; i < journal->j_maxlen; i++) {
964 err = journal_bmap(journal, i, &blocknr);
965 if (err)
966 return err;
967 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Namhyung Kim2a0e3382010-10-13 17:17:18 +0900968 if (unlikely(!bh))
969 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 lock_buffer(bh);
971 memset (bh->b_data, 0, journal->j_blocksize);
972 BUFFER_TRACE(bh, "marking dirty");
973 mark_buffer_dirty(bh);
974 BUFFER_TRACE(bh, "marking uptodate");
975 set_buffer_uptodate(bh);
976 unlock_buffer(bh);
977 __brelse(bh);
978 }
979
980 sync_blockdev(journal->j_dev);
981 jbd_debug(1, "JBD: journal cleared.\n");
982
983 /* OK, fill in the initial static fields in the new superblock */
984 sb = journal->j_superblock;
985
986 sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
987 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
988
989 sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
990 sb->s_maxlen = cpu_to_be32(journal->j_maxlen);
991 sb->s_first = cpu_to_be32(1);
992
993 journal->j_transaction_sequence = 1;
994
995 journal->j_flags &= ~JFS_ABORT;
996 journal->j_format_version = 2;
997
998 return journal_reset(journal);
999}
1000
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001001/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 * void journal_update_superblock() - Update journal sb on disk.
1003 * @journal: The journal to update.
1004 * @wait: Set to '0' if you don't want to wait for IO completion.
1005 *
1006 * Update a journal's dynamic superblock fields and write it to disk,
1007 * optionally waiting for the IO to complete.
1008 */
1009void journal_update_superblock(journal_t *journal, int wait)
1010{
1011 journal_superblock_t *sb = journal->j_superblock;
1012 struct buffer_head *bh = journal->j_sb_buffer;
1013
1014 /*
1015 * As a special case, if the on-disk copy is already marked as needing
1016 * no recovery (s_start == 0) and there are no outstanding transactions
1017 * in the filesystem, then we can safely defer the superblock update
1018 * until the next commit by setting JFS_FLUSHED. This avoids
1019 * attempting a write to a potential-readonly device.
1020 */
1021 if (sb->s_start == 0 && journal->j_tail_sequence ==
1022 journal->j_transaction_sequence) {
1023 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
Jan Kara9c28cbc2009-08-03 19:21:00 +02001024 "(start %u, seq %d, errno %d)\n",
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001025 journal->j_tail, journal->j_tail_sequence,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 journal->j_errno);
1027 goto out;
1028 }
1029
Darrick J. Wongdff68252010-10-04 12:35:05 -07001030 if (buffer_write_io_error(bh)) {
1031 char b[BDEVNAME_SIZE];
1032 /*
1033 * Oh, dear. A previous attempt to write the journal
1034 * superblock failed. This could happen because the
1035 * USB device was yanked out. Or it could happen to
1036 * be a transient write error and maybe the block will
1037 * be remapped. Nothing we can do but to retry the
1038 * write and hope for the best.
1039 */
1040 printk(KERN_ERR "JBD: previous I/O error detected "
1041 "for journal superblock update for %s.\n",
1042 journal_dev_name(journal, b));
1043 clear_buffer_write_io_error(bh);
1044 set_buffer_uptodate(bh);
1045 }
1046
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 spin_lock(&journal->j_state_lock);
Jan Kara9c28cbc2009-08-03 19:21:00 +02001048 jbd_debug(1,"JBD: updating superblock (start %u, seq %d, errno %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1050
1051 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1052 sb->s_start = cpu_to_be32(journal->j_tail);
1053 sb->s_errno = cpu_to_be32(journal->j_errno);
1054 spin_unlock(&journal->j_state_lock);
1055
1056 BUFFER_TRACE(bh, "marking dirty");
1057 mark_buffer_dirty(bh);
Darrick J. Wongdff68252010-10-04 12:35:05 -07001058 if (wait) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 sync_dirty_buffer(bh);
Darrick J. Wongdff68252010-10-04 12:35:05 -07001060 if (buffer_write_io_error(bh)) {
1061 char b[BDEVNAME_SIZE];
1062 printk(KERN_ERR "JBD: I/O error detected "
1063 "when updating journal superblock for %s.\n",
1064 journal_dev_name(journal, b));
1065 clear_buffer_write_io_error(bh);
1066 set_buffer_uptodate(bh);
1067 }
1068 } else
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02001069 write_dirty_buffer(bh, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Lukas Czerner99cb1a32011-05-23 18:33:02 +02001071 trace_jbd_update_superblock_end(journal, wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072out:
1073 /* If we have just flushed the log (by marking s_start==0), then
1074 * any future commit will have to be careful to update the
1075 * superblock again to re-record the true start of the log. */
1076
1077 spin_lock(&journal->j_state_lock);
1078 if (sb->s_start)
1079 journal->j_flags &= ~JFS_FLUSHED;
1080 else
1081 journal->j_flags |= JFS_FLUSHED;
1082 spin_unlock(&journal->j_state_lock);
1083}
1084
1085/*
1086 * Read the superblock for a given journal, performing initial
1087 * validation of the format.
1088 */
1089
1090static int journal_get_superblock(journal_t *journal)
1091{
1092 struct buffer_head *bh;
1093 journal_superblock_t *sb;
1094 int err = -EIO;
1095
1096 bh = journal->j_sb_buffer;
1097
1098 J_ASSERT(bh != NULL);
1099 if (!buffer_uptodate(bh)) {
1100 ll_rw_block(READ, 1, &bh);
1101 wait_on_buffer(bh);
1102 if (!buffer_uptodate(bh)) {
1103 printk (KERN_ERR
1104 "JBD: IO error reading journal superblock\n");
1105 goto out;
1106 }
1107 }
1108
1109 sb = journal->j_superblock;
1110
1111 err = -EINVAL;
1112
1113 if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1114 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1115 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1116 goto out;
1117 }
1118
1119 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1120 case JFS_SUPERBLOCK_V1:
1121 journal->j_format_version = 1;
1122 break;
1123 case JFS_SUPERBLOCK_V2:
1124 journal->j_format_version = 2;
1125 break;
1126 default:
1127 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1128 goto out;
1129 }
1130
1131 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1132 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1133 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1134 printk (KERN_WARNING "JBD: journal file too short\n");
1135 goto out;
1136 }
1137
Eryu Guan87622022011-11-01 19:04:59 -04001138 if (be32_to_cpu(sb->s_first) == 0 ||
1139 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1140 printk(KERN_WARNING
1141 "JBD: Invalid start block of journal: %u\n",
1142 be32_to_cpu(sb->s_first));
1143 goto out;
1144 }
1145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 return 0;
1147
1148out:
1149 journal_fail_superblock(journal);
1150 return err;
1151}
1152
1153/*
1154 * Load the on-disk journal superblock and read the key fields into the
1155 * journal_t.
1156 */
1157
1158static int load_superblock(journal_t *journal)
1159{
1160 int err;
1161 journal_superblock_t *sb;
1162
1163 err = journal_get_superblock(journal);
1164 if (err)
1165 return err;
1166
1167 sb = journal->j_superblock;
1168
1169 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1170 journal->j_tail = be32_to_cpu(sb->s_start);
1171 journal->j_first = be32_to_cpu(sb->s_first);
1172 journal->j_last = be32_to_cpu(sb->s_maxlen);
1173 journal->j_errno = be32_to_cpu(sb->s_errno);
1174
1175 return 0;
1176}
1177
1178
1179/**
1180 * int journal_load() - Read journal from disk.
1181 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001182 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 * Given a journal_t structure which tells us which disk blocks contain
1184 * a journal, read the journal from disk to initialise the in-memory
1185 * structures.
1186 */
1187int journal_load(journal_t *journal)
1188{
1189 int err;
Badari Pulavartyea817392006-08-27 01:23:52 -07001190 journal_superblock_t *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
1192 err = load_superblock(journal);
1193 if (err)
1194 return err;
1195
Badari Pulavartyea817392006-08-27 01:23:52 -07001196 sb = journal->j_superblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 /* If this is a V2 superblock, then we have to check the
1198 * features flags on it. */
1199
1200 if (journal->j_format_version >= 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 if ((sb->s_feature_ro_compat &
1202 ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1203 (sb->s_feature_incompat &
1204 ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1205 printk (KERN_WARNING
1206 "JBD: Unrecognised features on journal\n");
1207 return -EINVAL;
1208 }
1209 }
1210
1211 /* Let the recovery code check whether it needs to recover any
1212 * data from the journal. */
1213 if (journal_recover(journal))
1214 goto recovery_error;
1215
1216 /* OK, we've finished with the dynamic journal bits:
1217 * reinitialise the dynamic contents of the superblock in memory
1218 * and reset them on disk. */
1219 if (journal_reset(journal))
1220 goto recovery_error;
1221
1222 journal->j_flags &= ~JFS_ABORT;
1223 journal->j_flags |= JFS_LOADED;
1224 return 0;
1225
1226recovery_error:
1227 printk (KERN_WARNING "JBD: recovery failed\n");
1228 return -EIO;
1229}
1230
1231/**
1232 * void journal_destroy() - Release a journal_t structure.
1233 * @journal: Journal to act on.
1234 *
1235 * Release a journal_t structure once it is no longer in use by the
1236 * journaled object.
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001237 * Return <0 if we couldn't clean up the journal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 */
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001239int journal_destroy(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240{
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001241 int err = 0;
1242
Jan Kara03f4d802010-04-15 22:16:24 +02001243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 /* Wait for the commit thread to wake up and die. */
1245 journal_kill_thread(journal);
1246
1247 /* Force a final log commit */
1248 if (journal->j_running_transaction)
1249 journal_commit_transaction(journal);
1250
1251 /* Force any old transactions to disk */
1252
1253 /* Totally anal locking here... */
1254 spin_lock(&journal->j_list_lock);
1255 while (journal->j_checkpoint_transactions != NULL) {
1256 spin_unlock(&journal->j_list_lock);
1257 log_do_checkpoint(journal);
1258 spin_lock(&journal->j_list_lock);
1259 }
1260
1261 J_ASSERT(journal->j_running_transaction == NULL);
1262 J_ASSERT(journal->j_committing_transaction == NULL);
1263 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1264 spin_unlock(&journal->j_list_lock);
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 if (journal->j_sb_buffer) {
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001267 if (!is_journal_aborted(journal)) {
1268 /* We can now mark the journal as empty. */
1269 journal->j_tail = 0;
1270 journal->j_tail_sequence =
1271 ++journal->j_transaction_sequence;
1272 journal_update_superblock(journal, 1);
1273 } else {
1274 err = -EIO;
1275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 brelse(journal->j_sb_buffer);
1277 }
1278
1279 if (journal->j_inode)
1280 iput(journal->j_inode);
1281 if (journal->j_revoke)
1282 journal_destroy_revoke(journal);
1283 kfree(journal->j_wbuf);
1284 kfree(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001285
1286 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
1288
1289
1290/**
1291 *int journal_check_used_features () - Check if features specified are used.
1292 * @journal: Journal to check.
1293 * @compat: bitmask of compatible features
1294 * @ro: bitmask of features that force read-only mount
1295 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001296 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 * Check whether the journal uses all of a given set of
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001298 * features. Return true (non-zero) if it does.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 **/
1300
1301int journal_check_used_features (journal_t *journal, unsigned long compat,
1302 unsigned long ro, unsigned long incompat)
1303{
1304 journal_superblock_t *sb;
1305
1306 if (!compat && !ro && !incompat)
1307 return 1;
1308 if (journal->j_format_version == 1)
1309 return 0;
1310
1311 sb = journal->j_superblock;
1312
1313 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1314 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1315 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1316 return 1;
1317
1318 return 0;
1319}
1320
1321/**
1322 * int journal_check_available_features() - Check feature set in journalling layer
1323 * @journal: Journal to check.
1324 * @compat: bitmask of compatible features
1325 * @ro: bitmask of features that force read-only mount
1326 * @incompat: bitmask of incompatible features
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001327 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 * Check whether the journaling code supports the use of
1329 * all of a given set of features on this journal. Return true
1330 * (non-zero) if it can. */
1331
1332int journal_check_available_features (journal_t *journal, unsigned long compat,
1333 unsigned long ro, unsigned long incompat)
1334{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 if (!compat && !ro && !incompat)
1336 return 1;
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 /* We can support any known requested features iff the
1339 * superblock is in version 2. Otherwise we fail to support any
1340 * extended sb features. */
1341
1342 if (journal->j_format_version != 2)
1343 return 0;
1344
1345 if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1346 (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1347 (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1348 return 1;
1349
1350 return 0;
1351}
1352
1353/**
1354 * int journal_set_features () - Mark a given journal feature in the superblock
1355 * @journal: Journal to act on.
1356 * @compat: bitmask of compatible features
1357 * @ro: bitmask of features that force read-only mount
1358 * @incompat: bitmask of incompatible features
1359 *
1360 * Mark a given journal feature as present on the
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001361 * superblock. Returns true if the requested features could be set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 *
1363 */
1364
1365int journal_set_features (journal_t *journal, unsigned long compat,
1366 unsigned long ro, unsigned long incompat)
1367{
1368 journal_superblock_t *sb;
1369
1370 if (journal_check_used_features(journal, compat, ro, incompat))
1371 return 1;
1372
1373 if (!journal_check_available_features(journal, compat, ro, incompat))
1374 return 0;
1375
1376 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1377 compat, ro, incompat);
1378
1379 sb = journal->j_superblock;
1380
1381 sb->s_feature_compat |= cpu_to_be32(compat);
1382 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1383 sb->s_feature_incompat |= cpu_to_be32(incompat);
1384
1385 return 1;
1386}
1387
1388
1389/**
1390 * int journal_update_format () - Update on-disk journal structure.
1391 * @journal: Journal to act on.
1392 *
1393 * Given an initialised but unloaded journal struct, poke about in the
1394 * on-disk structure to update it to the most recent supported version.
1395 */
1396int journal_update_format (journal_t *journal)
1397{
1398 journal_superblock_t *sb;
1399 int err;
1400
1401 err = journal_get_superblock(journal);
1402 if (err)
1403 return err;
1404
1405 sb = journal->j_superblock;
1406
1407 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1408 case JFS_SUPERBLOCK_V2:
1409 return 0;
1410 case JFS_SUPERBLOCK_V1:
1411 return journal_convert_superblock_v1(journal, sb);
1412 default:
1413 break;
1414 }
1415 return -EINVAL;
1416}
1417
1418static int journal_convert_superblock_v1(journal_t *journal,
1419 journal_superblock_t *sb)
1420{
1421 int offset, blocksize;
1422 struct buffer_head *bh;
1423
1424 printk(KERN_WARNING
1425 "JBD: Converting superblock from version 1 to 2.\n");
1426
1427 /* Pre-initialise new fields to zero */
1428 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1429 blocksize = be32_to_cpu(sb->s_blocksize);
1430 memset(&sb->s_feature_compat, 0, blocksize-offset);
1431
1432 sb->s_nr_users = cpu_to_be32(1);
1433 sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1434 journal->j_format_version = 2;
1435
1436 bh = journal->j_sb_buffer;
1437 BUFFER_TRACE(bh, "marking dirty");
1438 mark_buffer_dirty(bh);
1439 sync_dirty_buffer(bh);
1440 return 0;
1441}
1442
1443
1444/**
1445 * int journal_flush () - Flush journal
1446 * @journal: Journal to act on.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001447 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 * Flush all data for a given journal to disk and empty the journal.
1449 * Filesystems can use this when remounting readonly to ensure that
1450 * recovery does not need to happen on remount.
1451 */
1452
1453int journal_flush(journal_t *journal)
1454{
1455 int err = 0;
1456 transaction_t *transaction = NULL;
Jan Kara9c28cbc2009-08-03 19:21:00 +02001457 unsigned int old_tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459 spin_lock(&journal->j_state_lock);
1460
1461 /* Force everything buffered to the log... */
1462 if (journal->j_running_transaction) {
1463 transaction = journal->j_running_transaction;
1464 __log_start_commit(journal, transaction->t_tid);
1465 } else if (journal->j_committing_transaction)
1466 transaction = journal->j_committing_transaction;
1467
1468 /* Wait for the log commit to complete... */
1469 if (transaction) {
1470 tid_t tid = transaction->t_tid;
1471
1472 spin_unlock(&journal->j_state_lock);
1473 log_wait_commit(journal, tid);
1474 } else {
1475 spin_unlock(&journal->j_state_lock);
1476 }
1477
1478 /* ...and flush everything in the log out to disk. */
1479 spin_lock(&journal->j_list_lock);
1480 while (!err && journal->j_checkpoint_transactions != NULL) {
1481 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001482 mutex_lock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 err = log_do_checkpoint(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001484 mutex_unlock(&journal->j_checkpoint_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 spin_lock(&journal->j_list_lock);
1486 }
1487 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001488
1489 if (is_journal_aborted(journal))
1490 return -EIO;
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 cleanup_journal_tail(journal);
1493
1494 /* Finally, mark the journal as really needing no recovery.
1495 * This sets s_start==0 in the underlying superblock, which is
1496 * the magic code for a fully-recovered superblock. Any future
1497 * commits of data to the journal will restore the current
1498 * s_start value. */
1499 spin_lock(&journal->j_state_lock);
1500 old_tail = journal->j_tail;
1501 journal->j_tail = 0;
1502 spin_unlock(&journal->j_state_lock);
1503 journal_update_superblock(journal, 1);
1504 spin_lock(&journal->j_state_lock);
1505 journal->j_tail = old_tail;
1506
1507 J_ASSERT(!journal->j_running_transaction);
1508 J_ASSERT(!journal->j_committing_transaction);
1509 J_ASSERT(!journal->j_checkpoint_transactions);
1510 J_ASSERT(journal->j_head == journal->j_tail);
1511 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1512 spin_unlock(&journal->j_state_lock);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -07001513 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514}
1515
1516/**
1517 * int journal_wipe() - Wipe journal contents
1518 * @journal: Journal to act on.
1519 * @write: flag (see below)
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001520 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 * Wipe out all of the contents of a journal, safely. This will produce
1522 * a warning if the journal contains any valid recovery information.
1523 * Must be called between journal_init_*() and journal_load().
1524 *
1525 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1526 * we merely suppress recovery.
1527 */
1528
1529int journal_wipe(journal_t *journal, int write)
1530{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 int err = 0;
1532
1533 J_ASSERT (!(journal->j_flags & JFS_LOADED));
1534
1535 err = load_superblock(journal);
1536 if (err)
1537 return err;
1538
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if (!journal->j_tail)
1540 goto no_recovery;
1541
1542 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1543 write ? "Clearing" : "Ignoring");
1544
1545 err = journal_skip_recovery(journal);
1546 if (write)
1547 journal_update_superblock(journal, 1);
1548
1549 no_recovery:
1550 return err;
1551}
1552
1553/*
1554 * journal_dev_name: format a character string to describe on what
1555 * device this journal is present.
1556 */
1557
Adrian Bunk022a4a72005-09-06 15:16:41 -07001558static const char *journal_dev_name(journal_t *journal, char *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
1560 struct block_device *bdev;
1561
1562 if (journal->j_inode)
1563 bdev = journal->j_inode->i_sb->s_bdev;
1564 else
1565 bdev = journal->j_dev;
1566
1567 return bdevname(bdev, buffer);
1568}
1569
1570/*
1571 * Journal abort has very specific semantics, which we describe
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001572 * for journal abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 *
1574 * Two internal function, which provide abort to te jbd layer
1575 * itself are here.
1576 */
1577
1578/*
1579 * Quick version for internal journal use (doesn't lock the journal).
1580 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1581 * and don't attempt to make any other journal updates.
1582 */
Adrian Bunk53308382008-02-06 01:40:12 -08001583static void __journal_abort_hard(journal_t *journal)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584{
1585 transaction_t *transaction;
1586 char b[BDEVNAME_SIZE];
1587
1588 if (journal->j_flags & JFS_ABORT)
1589 return;
1590
1591 printk(KERN_ERR "Aborting journal on device %s.\n",
1592 journal_dev_name(journal, b));
1593
1594 spin_lock(&journal->j_state_lock);
1595 journal->j_flags |= JFS_ABORT;
1596 transaction = journal->j_running_transaction;
1597 if (transaction)
1598 __log_start_commit(journal, transaction->t_tid);
1599 spin_unlock(&journal->j_state_lock);
1600}
1601
1602/* Soft abort: record the abort error status in the journal superblock,
1603 * but don't do any other IO. */
Adrian Bunk022a4a72005-09-06 15:16:41 -07001604static void __journal_abort_soft (journal_t *journal, int errno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605{
1606 if (journal->j_flags & JFS_ABORT)
1607 return;
1608
1609 if (!journal->j_errno)
1610 journal->j_errno = errno;
1611
1612 __journal_abort_hard(journal);
1613
1614 if (errno)
1615 journal_update_superblock(journal, 1);
1616}
1617
1618/**
1619 * void journal_abort () - Shutdown the journal immediately.
1620 * @journal: the journal to shutdown.
1621 * @errno: an error number to record in the journal indicating
1622 * the reason for the shutdown.
1623 *
1624 * Perform a complete, immediate shutdown of the ENTIRE
1625 * journal (not of a single transaction). This operation cannot be
1626 * undone without closing and reopening the journal.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001627 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 * The journal_abort function is intended to support higher level error
1629 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1630 * mode.
1631 *
1632 * Journal abort has very specific semantics. Any existing dirty,
1633 * unjournaled buffers in the main filesystem will still be written to
1634 * disk by bdflush, but the journaling mechanism will be suspended
1635 * immediately and no further transaction commits will be honoured.
1636 *
1637 * Any dirty, journaled buffers will be written back to disk without
1638 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1639 * filesystem, but we _do_ attempt to leave as much data as possible
1640 * behind for fsck to use for cleanup.
1641 *
1642 * Any attempt to get a new transaction handle on a journal which is in
1643 * ABORT state will just result in an -EROFS error return. A
1644 * journal_stop on an existing handle will return -EIO if we have
1645 * entered abort state during the update.
1646 *
1647 * Recursive transactions are not disturbed by journal abort until the
1648 * final journal_stop, which will receive the -EIO error.
1649 *
1650 * Finally, the journal_abort call allows the caller to supply an errno
1651 * which will be recorded (if possible) in the journal superblock. This
1652 * allows a client to record failure conditions in the middle of a
1653 * transaction without having to complete the transaction to record the
1654 * failure to disk. ext3_error, for example, now uses this
1655 * functionality.
1656 *
1657 * Errors which originate from within the journaling layer will NOT
1658 * supply an errno; a null errno implies that absolutely no further
1659 * writes are done to the journal (unless there are any already in
1660 * progress).
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001661 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 */
1663
1664void journal_abort(journal_t *journal, int errno)
1665{
1666 __journal_abort_soft(journal, errno);
1667}
1668
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001669/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 * int journal_errno () - returns the journal's error state.
1671 * @journal: journal to examine.
1672 *
1673 * This is the errno numbet set with journal_abort(), the last
1674 * time the journal was mounted - if the journal was stopped
1675 * without calling abort this will be 0.
1676 *
1677 * If the journal has been aborted on this mount time -EROFS will
1678 * be returned.
1679 */
1680int journal_errno(journal_t *journal)
1681{
1682 int err;
1683
1684 spin_lock(&journal->j_state_lock);
1685 if (journal->j_flags & JFS_ABORT)
1686 err = -EROFS;
1687 else
1688 err = journal->j_errno;
1689 spin_unlock(&journal->j_state_lock);
1690 return err;
1691}
1692
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001693/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 * int journal_clear_err () - clears the journal's error state
1695 * @journal: journal to act on.
1696 *
1697 * An error must be cleared or Acked to take a FS out of readonly
1698 * mode.
1699 */
1700int journal_clear_err(journal_t *journal)
1701{
1702 int err = 0;
1703
1704 spin_lock(&journal->j_state_lock);
1705 if (journal->j_flags & JFS_ABORT)
1706 err = -EROFS;
1707 else
1708 journal->j_errno = 0;
1709 spin_unlock(&journal->j_state_lock);
1710 return err;
1711}
1712
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001713/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 * void journal_ack_err() - Ack journal err.
1715 * @journal: journal to act on.
1716 *
1717 * An error must be cleared or Acked to take a FS out of readonly
1718 * mode.
1719 */
1720void journal_ack_err(journal_t *journal)
1721{
1722 spin_lock(&journal->j_state_lock);
1723 if (journal->j_errno)
1724 journal->j_flags |= JFS_ACK_ERR;
1725 spin_unlock(&journal->j_state_lock);
1726}
1727
1728int journal_blocks_per_page(struct inode *inode)
1729{
1730 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1731}
1732
1733/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 * Journal_head storage management
1735 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001736static struct kmem_cache *journal_head_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737#ifdef CONFIG_JBD_DEBUG
1738static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1739#endif
1740
1741static int journal_init_journal_head_cache(void)
1742{
1743 int retval;
1744
Al Viro1076d172008-03-29 03:07:18 +00001745 J_ASSERT(journal_head_cache == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 journal_head_cache = kmem_cache_create("journal_head",
1747 sizeof(struct journal_head),
1748 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001749 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001750 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00001752 if (!journal_head_cache) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 retval = -ENOMEM;
1754 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1755 }
1756 return retval;
1757}
1758
1759static void journal_destroy_journal_head_cache(void)
1760{
Duane Griffin3850f7a2008-07-25 01:46:19 -07001761 if (journal_head_cache) {
1762 kmem_cache_destroy(journal_head_cache);
1763 journal_head_cache = NULL;
1764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765}
1766
1767/*
1768 * journal_head splicing and dicing
1769 */
1770static struct journal_head *journal_alloc_journal_head(void)
1771{
1772 struct journal_head *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774#ifdef CONFIG_JBD_DEBUG
1775 atomic_inc(&nr_journal_heads);
1776#endif
1777 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001778 if (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 jbd_debug(1, "out of memory for journal_head\n");
Namhyung Kimf81e3d42010-10-04 19:12:13 +09001780 printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1781 __func__);
1782
Stephen Hemmingerc80544d2007-10-18 03:07:05 -07001783 while (ret == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 yield();
1785 ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1786 }
1787 }
1788 return ret;
1789}
1790
1791static void journal_free_journal_head(struct journal_head *jh)
1792{
1793#ifdef CONFIG_JBD_DEBUG
1794 atomic_dec(&nr_journal_heads);
Randy Dunlapc9cf5522006-06-27 02:53:52 -07001795 memset(jh, JBD_POISON_FREE, sizeof(*jh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796#endif
1797 kmem_cache_free(journal_head_cache, jh);
1798}
1799
1800/*
1801 * A journal_head is attached to a buffer_head whenever JBD has an
1802 * interest in the buffer.
1803 *
1804 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1805 * is set. This bit is tested in core kernel code where we need to take
1806 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1807 * there.
1808 *
1809 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1810 *
1811 * When a buffer has its BH_JBD bit set it is immune from being released by
1812 * core kernel code, mainly via ->b_count.
1813 *
Jan Karabb189242011-06-24 23:11:59 +02001814 * A journal_head is detached from its buffer_head when the journal_head's
1815 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
1816 * transaction (b_cp_transaction) hold their references to b_jcount.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 *
1818 * Various places in the kernel want to attach a journal_head to a buffer_head
1819 * _before_ attaching the journal_head to a transaction. To protect the
1820 * journal_head in this situation, journal_add_journal_head elevates the
1821 * journal_head's b_jcount refcount by one. The caller must call
1822 * journal_put_journal_head() to undo this.
1823 *
1824 * So the typical usage would be:
1825 *
1826 * (Attach a journal_head if needed. Increments b_jcount)
1827 * struct journal_head *jh = journal_add_journal_head(bh);
1828 * ...
Jan Karabb189242011-06-24 23:11:59 +02001829 * (Get another reference for transaction)
1830 * journal_grab_journal_head(bh);
1831 * jh->b_transaction = xxx;
1832 * (Put original reference)
1833 * journal_put_journal_head(jh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 */
1835
1836/*
1837 * Give a buffer_head a journal_head.
1838 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 * May sleep.
1840 */
1841struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1842{
1843 struct journal_head *jh;
1844 struct journal_head *new_jh = NULL;
1845
1846repeat:
1847 if (!buffer_jbd(bh)) {
1848 new_jh = journal_alloc_journal_head();
1849 memset(new_jh, 0, sizeof(*new_jh));
1850 }
1851
1852 jbd_lock_bh_journal_head(bh);
1853 if (buffer_jbd(bh)) {
1854 jh = bh2jh(bh);
1855 } else {
1856 J_ASSERT_BH(bh,
1857 (atomic_read(&bh->b_count) > 0) ||
1858 (bh->b_page && bh->b_page->mapping));
1859
1860 if (!new_jh) {
1861 jbd_unlock_bh_journal_head(bh);
1862 goto repeat;
1863 }
1864
1865 jh = new_jh;
1866 new_jh = NULL; /* We consumed it */
1867 set_buffer_jbd(bh);
1868 bh->b_private = jh;
1869 jh->b_bh = bh;
1870 get_bh(bh);
1871 BUFFER_TRACE(bh, "added journal_head");
1872 }
1873 jh->b_jcount++;
1874 jbd_unlock_bh_journal_head(bh);
1875 if (new_jh)
1876 journal_free_journal_head(new_jh);
1877 return bh->b_private;
1878}
1879
1880/*
1881 * Grab a ref against this buffer_head's journal_head. If it ended up not
1882 * having a journal_head, return NULL
1883 */
1884struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1885{
1886 struct journal_head *jh = NULL;
1887
1888 jbd_lock_bh_journal_head(bh);
1889 if (buffer_jbd(bh)) {
1890 jh = bh2jh(bh);
1891 jh->b_jcount++;
1892 }
1893 jbd_unlock_bh_journal_head(bh);
1894 return jh;
1895}
1896
1897static void __journal_remove_journal_head(struct buffer_head *bh)
1898{
1899 struct journal_head *jh = bh2jh(bh);
1900
1901 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karabb189242011-06-24 23:11:59 +02001902 J_ASSERT_JH(jh, jh->b_transaction == NULL);
1903 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1904 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
1905 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1906 J_ASSERT_BH(bh, buffer_jbd(bh));
1907 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1908 BUFFER_TRACE(bh, "remove journal_head");
1909 if (jh->b_frozen_data) {
1910 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
1911 jbd_free(jh->b_frozen_data, bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 }
Jan Karabb189242011-06-24 23:11:59 +02001913 if (jh->b_committed_data) {
1914 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
1915 jbd_free(jh->b_committed_data, bh->b_size);
1916 }
1917 bh->b_private = NULL;
1918 jh->b_bh = NULL; /* debug, really */
1919 clear_buffer_jbd(bh);
1920 journal_free_journal_head(jh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921}
1922
1923/*
Jan Karabb189242011-06-24 23:11:59 +02001924 * Drop a reference on the passed journal_head. If it fell to zero then
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 * release the journal_head from the buffer_head.
1926 */
1927void journal_put_journal_head(struct journal_head *jh)
1928{
1929 struct buffer_head *bh = jh2bh(jh);
1930
1931 jbd_lock_bh_journal_head(bh);
1932 J_ASSERT_JH(jh, jh->b_jcount > 0);
1933 --jh->b_jcount;
Jan Karabb189242011-06-24 23:11:59 +02001934 if (!jh->b_jcount) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 __journal_remove_journal_head(bh);
Jan Karabb189242011-06-24 23:11:59 +02001936 jbd_unlock_bh_journal_head(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 __brelse(bh);
Jan Karabb189242011-06-24 23:11:59 +02001938 } else
1939 jbd_unlock_bh_journal_head(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940}
1941
1942/*
Jose R. Santosc2a91592007-10-18 23:39:22 -07001943 * debugfs tunables
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 */
Jose R. Santosc2a91592007-10-18 23:39:22 -07001945#ifdef CONFIG_JBD_DEBUG
1946
1947u8 journal_enable_debug __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948EXPORT_SYMBOL(journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
Jose R. Santosc2a91592007-10-18 23:39:22 -07001950static struct dentry *jbd_debugfs_dir;
1951static struct dentry *jbd_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
Jose R. Santosc2a91592007-10-18 23:39:22 -07001953static void __init jbd_create_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001955 jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
1956 if (jbd_debugfs_dir)
Yin Kangkai765f8362009-12-15 14:48:25 -08001957 jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
Jose R. Santosc2a91592007-10-18 23:39:22 -07001958 jbd_debugfs_dir,
1959 &journal_enable_debug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960}
1961
Jose R. Santosc2a91592007-10-18 23:39:22 -07001962static void __exit jbd_remove_debugfs_entry(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963{
Jose R. Santosc2a91592007-10-18 23:39:22 -07001964 debugfs_remove(jbd_debug);
1965 debugfs_remove(jbd_debugfs_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966}
1967
1968#else
1969
Jose R. Santosc2a91592007-10-18 23:39:22 -07001970static inline void jbd_create_debugfs_entry(void)
1971{
1972}
1973
1974static inline void jbd_remove_debugfs_entry(void)
1975{
1976}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
1978#endif
1979
Christoph Lametere18b8902006-12-06 20:33:20 -08001980struct kmem_cache *jbd_handle_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
1982static int __init journal_init_handle_cache(void)
1983{
1984 jbd_handle_cache = kmem_cache_create("journal_handle",
1985 sizeof(handle_t),
1986 0, /* offset */
Mel Gormane12ba742007-10-16 01:25:52 -07001987 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001988 NULL); /* ctor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 if (jbd_handle_cache == NULL) {
1990 printk(KERN_EMERG "JBD: failed to create handle cache\n");
1991 return -ENOMEM;
1992 }
1993 return 0;
1994}
1995
1996static void journal_destroy_handle_cache(void)
1997{
1998 if (jbd_handle_cache)
1999 kmem_cache_destroy(jbd_handle_cache);
2000}
2001
2002/*
2003 * Module startup and shutdown
2004 */
2005
2006static int __init journal_init_caches(void)
2007{
2008 int ret;
2009
2010 ret = journal_init_revoke_caches();
2011 if (ret == 0)
2012 ret = journal_init_journal_head_cache();
2013 if (ret == 0)
2014 ret = journal_init_handle_cache();
2015 return ret;
2016}
2017
2018static void journal_destroy_caches(void)
2019{
2020 journal_destroy_revoke_caches();
2021 journal_destroy_journal_head_cache();
2022 journal_destroy_handle_cache();
2023}
2024
2025static int __init journal_init(void)
2026{
2027 int ret;
2028
Alexey Dobriyan2aed3482006-09-27 01:49:28 -07002029 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
Adrian Bunk022a4a72005-09-06 15:16:41 -07002030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 ret = journal_init_caches();
2032 if (ret != 0)
2033 journal_destroy_caches();
Jose R. Santosc2a91592007-10-18 23:39:22 -07002034 jbd_create_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 return ret;
2036}
2037
2038static void __exit journal_exit(void)
2039{
2040#ifdef CONFIG_JBD_DEBUG
2041 int n = atomic_read(&nr_journal_heads);
2042 if (n)
2043 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2044#endif
Jose R. Santosc2a91592007-10-18 23:39:22 -07002045 jbd_remove_debugfs_entry();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 journal_destroy_caches();
2047}
2048
2049MODULE_LICENSE("GPL");
2050module_init(journal_init);
2051module_exit(journal_exit);
2052