blob: b7ca3a92a4dba0b23f555a75cac79fea1b6ae9e4 [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002 * linux/fs/jbd2/journal.c
Dave Kleikamp470decc2006-10-11 01:20:57 -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>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070028#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070029#include <linux/errno.h>
30#include <linux/slab.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070031#include <linux/init.h>
32#include <linux/mm.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080033#include <linux/freezer.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070034#include <linux/pagemap.h>
35#include <linux/kthread.h>
36#include <linux/poison.h>
37#include <linux/proc_fs.h>
Jose R. Santos0f49d5d2007-07-18 08:50:18 -040038#include <linux/debugfs.h>
Johann Lombardi8e85fb32008-01-28 23:58:27 -050039#include <linux/seq_file.h>
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -050040#include <linux/math64.h>
Theodore Ts'o879c5e62009-06-17 11:47:48 -040041#include <linux/hash.h>
42
43#define CREATE_TRACE_POINTS
44#include <trace/events/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070045
46#include <asm/uaccess.h>
47#include <asm/page.h>
48
Mingming Caof7f4bcc2006-10-11 01:20:59 -070049EXPORT_SYMBOL(jbd2_journal_start);
50EXPORT_SYMBOL(jbd2_journal_restart);
51EXPORT_SYMBOL(jbd2_journal_extend);
52EXPORT_SYMBOL(jbd2_journal_stop);
53EXPORT_SYMBOL(jbd2_journal_lock_updates);
54EXPORT_SYMBOL(jbd2_journal_unlock_updates);
55EXPORT_SYMBOL(jbd2_journal_get_write_access);
56EXPORT_SYMBOL(jbd2_journal_get_create_access);
57EXPORT_SYMBOL(jbd2_journal_get_undo_access);
Joel Beckere06c8222008-09-11 15:35:47 -070058EXPORT_SYMBOL(jbd2_journal_set_triggers);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070059EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
60EXPORT_SYMBOL(jbd2_journal_release_buffer);
61EXPORT_SYMBOL(jbd2_journal_forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -070062#if 0
63EXPORT_SYMBOL(journal_sync_buffer);
64#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -070065EXPORT_SYMBOL(jbd2_journal_flush);
66EXPORT_SYMBOL(jbd2_journal_revoke);
Dave Kleikamp470decc2006-10-11 01:20:57 -070067
Mingming Caof7f4bcc2006-10-11 01:20:59 -070068EXPORT_SYMBOL(jbd2_journal_init_dev);
69EXPORT_SYMBOL(jbd2_journal_init_inode);
70EXPORT_SYMBOL(jbd2_journal_update_format);
71EXPORT_SYMBOL(jbd2_journal_check_used_features);
72EXPORT_SYMBOL(jbd2_journal_check_available_features);
73EXPORT_SYMBOL(jbd2_journal_set_features);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070074EXPORT_SYMBOL(jbd2_journal_load);
75EXPORT_SYMBOL(jbd2_journal_destroy);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070076EXPORT_SYMBOL(jbd2_journal_abort);
77EXPORT_SYMBOL(jbd2_journal_errno);
78EXPORT_SYMBOL(jbd2_journal_ack_err);
79EXPORT_SYMBOL(jbd2_journal_clear_err);
80EXPORT_SYMBOL(jbd2_log_wait_commit);
Theodore Ts'o3b799d12009-12-09 20:42:53 -050081EXPORT_SYMBOL(jbd2_log_start_commit);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070082EXPORT_SYMBOL(jbd2_journal_start_commit);
83EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
84EXPORT_SYMBOL(jbd2_journal_wipe);
85EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
86EXPORT_SYMBOL(jbd2_journal_invalidatepage);
87EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
88EXPORT_SYMBOL(jbd2_journal_force_commit);
Jan Karac851ed52008-07-11 19:27:31 -040089EXPORT_SYMBOL(jbd2_journal_file_inode);
90EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
91EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
92EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
Dave Kleikamp470decc2006-10-11 01:20:57 -070093
94static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
95static void __journal_abort_soft (journal_t *journal, int errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -070096
97/*
98 * Helper function used to manage commit timeouts
99 */
100
101static void commit_timeout(unsigned long __data)
102{
103 struct task_struct * p = (struct task_struct *) __data;
104
105 wake_up_process(p);
106}
107
108/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700109 * kjournald2: The main thread function used to manage a logging device
Dave Kleikamp470decc2006-10-11 01:20:57 -0700110 * journal.
111 *
112 * This kernel thread is responsible for two things:
113 *
114 * 1) COMMIT: Every so often we need to commit the current state of the
115 * filesystem to disk. The journal thread is responsible for writing
116 * all of the metadata buffers to disk.
117 *
118 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
119 * of the data in that part of the log has been rewritten elsewhere on
120 * the disk. Flushing these old buffers to reclaim space in the log is
121 * known as checkpointing, and this thread is responsible for that job.
122 */
123
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700124static int kjournald2(void *arg)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700125{
126 journal_t *journal = arg;
127 transaction_t *transaction;
128
129 /*
130 * Set up an interval timer which can be used to trigger a commit wakeup
131 * after the commit interval expires
132 */
133 setup_timer(&journal->j_commit_timer, commit_timeout,
134 (unsigned long)current);
135
136 /* Record that the journal thread is running */
137 journal->j_task = current;
138 wake_up(&journal->j_wait_done_commit);
139
Dave Kleikamp470decc2006-10-11 01:20:57 -0700140 /*
141 * And now, wait forever for commit wakeup events.
142 */
143 spin_lock(&journal->j_state_lock);
144
145loop:
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700146 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700147 goto end_loop;
148
149 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
150 journal->j_commit_sequence, journal->j_commit_request);
151
152 if (journal->j_commit_sequence != journal->j_commit_request) {
153 jbd_debug(1, "OK, requests differ\n");
154 spin_unlock(&journal->j_state_lock);
155 del_timer_sync(&journal->j_commit_timer);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700156 jbd2_journal_commit_transaction(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700157 spin_lock(&journal->j_state_lock);
158 goto loop;
159 }
160
161 wake_up(&journal->j_wait_done_commit);
162 if (freezing(current)) {
163 /*
164 * The simpler the better. Flushing journal isn't a
165 * good idea, because that depends on threads that may
166 * be already stopped.
167 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700168 jbd_debug(1, "Now suspending kjournald2\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700169 spin_unlock(&journal->j_state_lock);
170 refrigerator();
171 spin_lock(&journal->j_state_lock);
172 } else {
173 /*
174 * We assume on resume that commits are already there,
175 * so we don't sleep
176 */
177 DEFINE_WAIT(wait);
178 int should_sleep = 1;
179
180 prepare_to_wait(&journal->j_wait_commit, &wait,
181 TASK_INTERRUPTIBLE);
182 if (journal->j_commit_sequence != journal->j_commit_request)
183 should_sleep = 0;
184 transaction = journal->j_running_transaction;
185 if (transaction && time_after_eq(jiffies,
186 transaction->t_expires))
187 should_sleep = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700188 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700189 should_sleep = 0;
190 if (should_sleep) {
191 spin_unlock(&journal->j_state_lock);
192 schedule();
193 spin_lock(&journal->j_state_lock);
194 }
195 finish_wait(&journal->j_wait_commit, &wait);
196 }
197
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700198 jbd_debug(1, "kjournald2 wakes\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700199
200 /*
201 * Were we woken up by a commit wakeup event?
202 */
203 transaction = journal->j_running_transaction;
204 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
205 journal->j_commit_request = transaction->t_tid;
206 jbd_debug(1, "woke because of timeout\n");
207 }
208 goto loop;
209
210end_loop:
211 spin_unlock(&journal->j_state_lock);
212 del_timer_sync(&journal->j_commit_timer);
213 journal->j_task = NULL;
214 wake_up(&journal->j_wait_done_commit);
215 jbd_debug(1, "Journal thread exiting.\n");
216 return 0;
217}
218
Pavel Emelianov97f06782007-05-08 00:30:42 -0700219static int jbd2_journal_start_thread(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700220{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700221 struct task_struct *t;
222
Theodore Ts'o90576c02009-09-29 15:51:30 -0400223 t = kthread_run(kjournald2, journal, "jbd2/%s",
224 journal->j_devname);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700225 if (IS_ERR(t))
226 return PTR_ERR(t);
227
Al Viro1076d172008-03-29 03:07:18 +0000228 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700229 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700230}
231
232static void journal_kill_thread(journal_t *journal)
233{
234 spin_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700235 journal->j_flags |= JBD2_UNMOUNT;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700236
237 while (journal->j_task) {
238 wake_up(&journal->j_wait_commit);
239 spin_unlock(&journal->j_state_lock);
Al Viro1076d172008-03-29 03:07:18 +0000240 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700241 spin_lock(&journal->j_state_lock);
242 }
243 spin_unlock(&journal->j_state_lock);
244}
245
246/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700247 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700248 *
249 * Writes a metadata buffer to a given disk block. The actual IO is not
250 * performed but a new buffer_head is constructed which labels the data
251 * to be written with the correct destination disk block.
252 *
253 * Any magic-number escaping which needs to be done will cause a
254 * copy-out here. If the buffer happens to start with the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700255 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700256 * magic number is only written to the log for descripter blocks. In
257 * this case, we copy the data and replace the first word with 0, and we
258 * return a result code which indicates that this buffer needs to be
259 * marked as an escaped buffer in the corresponding log descriptor
260 * block. The missing word can then be restored when the block is read
261 * during recovery.
262 *
263 * If the source buffer has already been modified by a new transaction
264 * since we took the last commit snapshot, we use the frozen copy of
265 * that data for IO. If we end up using the existing buffer_head's data
266 * for the write, then we *have* to lock the buffer to prevent anyone
267 * else from using and possibly modifying it while the IO is in
268 * progress.
269 *
270 * The function returns a pointer to the buffer_heads to be used for IO.
271 *
272 * We assume that the journal has already been locked in this function.
273 *
274 * Return value:
275 * <0: Error
276 * >=0: Finished OK
277 *
278 * On success:
279 * Bit 0 set == escape performed on the data
280 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
281 */
282
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700283int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700284 struct journal_head *jh_in,
285 struct journal_head **jh_out,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700286 unsigned long long blocknr)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700287{
288 int need_copy_out = 0;
289 int done_copy_out = 0;
290 int do_escape = 0;
291 char *mapped_data;
292 struct buffer_head *new_bh;
293 struct journal_head *new_jh;
294 struct page *new_page;
295 unsigned int new_offset;
296 struct buffer_head *bh_in = jh2bh(jh_in);
Joel Beckere06c8222008-09-11 15:35:47 -0700297 struct jbd2_buffer_trigger_type *triggers;
dingdinghua96577c42009-07-13 17:55:35 -0400298 journal_t *journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -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);
dingdinghua96577c42009-07-13 17:55:35 -0400312 /* 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 = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
Dave Kleikamp470decc2006-10-11 01:20:57 -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);
Joel Beckere06c8222008-09-11 15:35:47 -0700328 triggers = jh_in->b_frozen_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700329 } else {
330 new_page = jh2bh(jh_in)->b_page;
331 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
Joel Beckere06c8222008-09-11 15:35:47 -0700332 triggers = jh_in->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700333 }
334
335 mapped_data = kmap_atomic(new_page, KM_USER0);
336 /*
Joel Beckere06c8222008-09-11 15:35:47 -0700337 * Fire any commit trigger. Do this before checking for escaping,
338 * as the trigger may modify the magic offset. If a copy-out
339 * happens afterwards, it will have the correct data in the buffer.
340 */
341 jbd2_buffer_commit_trigger(jh_in, mapped_data + new_offset,
342 triggers);
343
344 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700345 * Check for escaping
346 */
347 if (*((__be32 *)(mapped_data + new_offset)) ==
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700348 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700349 need_copy_out = 1;
350 do_escape = 1;
351 }
352 kunmap_atomic(mapped_data, KM_USER0);
353
354 /*
355 * Do we need to do a data copy?
356 */
357 if (need_copy_out && !done_copy_out) {
358 char *tmp;
359
360 jbd_unlock_bh_state(bh_in);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400361 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500362 if (!tmp) {
363 jbd2_journal_put_journal_head(new_jh);
364 return -ENOMEM;
365 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700366 jbd_lock_bh_state(bh_in);
367 if (jh_in->b_frozen_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400368 jbd2_free(tmp, bh_in->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700369 goto repeat;
370 }
371
372 jh_in->b_frozen_data = tmp;
373 mapped_data = kmap_atomic(new_page, KM_USER0);
374 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
375 kunmap_atomic(mapped_data, KM_USER0);
376
377 new_page = virt_to_page(tmp);
378 new_offset = offset_in_page(tmp);
379 done_copy_out = 1;
Joel Beckere06c8222008-09-11 15:35:47 -0700380
381 /*
382 * This isn't strictly necessary, as we're using frozen
383 * data for the escaping, but it keeps consistency with
384 * b_frozen_data usage.
385 */
386 jh_in->b_frozen_triggers = jh_in->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700387 }
388
389 /*
390 * Did we need to do an escaping? Now we've done all the
391 * copying, we can finally do so.
392 */
393 if (do_escape) {
394 mapped_data = kmap_atomic(new_page, KM_USER0);
395 *((unsigned int *)(mapped_data + new_offset)) = 0;
396 kunmap_atomic(mapped_data, KM_USER0);
397 }
398
Dave Kleikamp470decc2006-10-11 01:20:57 -0700399 set_bh_page(new_bh, new_page, new_offset);
400 new_jh->b_transaction = NULL;
401 new_bh->b_size = jh2bh(jh_in)->b_size;
402 new_bh->b_bdev = transaction->t_journal->j_dev;
403 new_bh->b_blocknr = blocknr;
404 set_buffer_mapped(new_bh);
405 set_buffer_dirty(new_bh);
406
407 *jh_out = new_jh;
408
409 /*
410 * The to-be-written buffer needs to get moved to the io queue,
411 * and the original buffer whose contents we are shadowing or
412 * copying is moved to the transaction's shadow queue.
413 */
414 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghua96577c42009-07-13 17:55:35 -0400415 spin_lock(&journal->j_list_lock);
416 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
417 spin_unlock(&journal->j_list_lock);
418 jbd_unlock_bh_state(bh_in);
419
Dave Kleikamp470decc2006-10-11 01:20:57 -0700420 JBUFFER_TRACE(new_jh, "file as BJ_IO");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700421 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700422
423 return do_escape | (done_copy_out << 1);
424}
425
426/*
427 * Allocation code for the journal file. Manage the space left in the
428 * journal, so that we can begin checkpointing when appropriate.
429 */
430
431/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700432 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700433 *
434 * Called with the journal already locked.
435 *
436 * Called under j_state_lock
437 */
438
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700439int __jbd2_log_space_left(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700440{
441 int left = journal->j_free;
442
443 assert_spin_locked(&journal->j_state_lock);
444
445 /*
446 * Be pessimistic here about the number of those free blocks which
447 * might be required for log descriptor control blocks.
448 */
449
450#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
451
452 left -= MIN_LOG_RESERVED_BLOCKS;
453
454 if (left <= 0)
455 return 0;
456 left -= (left >> 3);
457 return left;
458}
459
460/*
Jan Karac88ccea2009-02-10 11:27:46 -0500461 * Called under j_state_lock. Returns true if a transaction commit was started.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700462 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700463int __jbd2_log_start_commit(journal_t *journal, tid_t target)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700464{
465 /*
466 * Are we already doing a recent enough commit?
467 */
468 if (!tid_geq(journal->j_commit_request, target)) {
469 /*
470 * We want a new commit: OK, mark the request and wakup the
471 * commit thread. We do _not_ do the commit ourselves.
472 */
473
474 journal->j_commit_request = target;
475 jbd_debug(1, "JBD: requesting commit %d/%d\n",
476 journal->j_commit_request,
477 journal->j_commit_sequence);
478 wake_up(&journal->j_wait_commit);
479 return 1;
480 }
481 return 0;
482}
483
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700484int jbd2_log_start_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700485{
486 int ret;
487
488 spin_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700489 ret = __jbd2_log_start_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700490 spin_unlock(&journal->j_state_lock);
491 return ret;
492}
493
494/*
495 * Force and wait upon a commit if the calling process is not within
496 * transaction. This is used for forcing out undo-protected data which contains
497 * bitmaps, when the fs is running out of space.
498 *
499 * We can only force the running transaction if we don't have an active handle;
500 * otherwise, we will deadlock.
501 *
502 * Returns true if a transaction was started.
503 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700504int jbd2_journal_force_commit_nested(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700505{
506 transaction_t *transaction = NULL;
507 tid_t tid;
508
509 spin_lock(&journal->j_state_lock);
510 if (journal->j_running_transaction && !current->journal_info) {
511 transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700512 __jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700513 } else if (journal->j_committing_transaction)
514 transaction = journal->j_committing_transaction;
515
516 if (!transaction) {
517 spin_unlock(&journal->j_state_lock);
518 return 0; /* Nothing to retry */
519 }
520
521 tid = transaction->t_tid;
522 spin_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700523 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700524 return 1;
525}
526
527/*
528 * Start a commit of the current running transaction (if any). Returns true
Jan Karac88ccea2009-02-10 11:27:46 -0500529 * if a transaction is going to be committed (or is currently already
530 * committing), and fills its tid in at *ptid
Dave Kleikamp470decc2006-10-11 01:20:57 -0700531 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700532int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700533{
534 int ret = 0;
535
536 spin_lock(&journal->j_state_lock);
537 if (journal->j_running_transaction) {
538 tid_t tid = journal->j_running_transaction->t_tid;
539
Jan Karac88ccea2009-02-10 11:27:46 -0500540 __jbd2_log_start_commit(journal, tid);
541 /* There's a running transaction and we've just made sure
542 * it's commit has been scheduled. */
543 if (ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700544 *ptid = tid;
Jan Karac88ccea2009-02-10 11:27:46 -0500545 ret = 1;
546 } else if (journal->j_committing_transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700547 /*
548 * If ext3_write_super() recently started a commit, then we
549 * have to wait for completion of that transaction
550 */
Jan Karac88ccea2009-02-10 11:27:46 -0500551 if (ptid)
552 *ptid = journal->j_committing_transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700553 ret = 1;
554 }
555 spin_unlock(&journal->j_state_lock);
556 return ret;
557}
558
559/*
560 * Wait for a specified commit to complete.
561 * The caller may not hold the journal lock.
562 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700563int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700564{
565 int err = 0;
566
Jose R. Santose23291b2007-07-18 08:57:06 -0400567#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -0700568 spin_lock(&journal->j_state_lock);
569 if (!tid_geq(journal->j_commit_request, tid)) {
570 printk(KERN_EMERG
571 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400572 __func__, journal->j_commit_request, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700573 }
574 spin_unlock(&journal->j_state_lock);
575#endif
576 spin_lock(&journal->j_state_lock);
577 while (tid_gt(tid, journal->j_commit_sequence)) {
578 jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
579 tid, journal->j_commit_sequence);
580 wake_up(&journal->j_wait_commit);
581 spin_unlock(&journal->j_state_lock);
582 wait_event(journal->j_wait_done_commit,
583 !tid_gt(tid, journal->j_commit_sequence));
584 spin_lock(&journal->j_state_lock);
585 }
586 spin_unlock(&journal->j_state_lock);
587
588 if (unlikely(is_journal_aborted(journal))) {
589 printk(KERN_EMERG "journal commit I/O error\n");
590 err = -EIO;
591 }
592 return err;
593}
594
595/*
596 * Log buffer allocation routines:
597 */
598
Mingming Cao18eba7a2006-10-11 01:21:13 -0700599int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700600{
601 unsigned long blocknr;
602
603 spin_lock(&journal->j_state_lock);
604 J_ASSERT(journal->j_free > 1);
605
606 blocknr = journal->j_head;
607 journal->j_head++;
608 journal->j_free--;
609 if (journal->j_head == journal->j_last)
610 journal->j_head = journal->j_first;
611 spin_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700612 return jbd2_journal_bmap(journal, blocknr, retp);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700613}
614
615/*
616 * Conversion of logical to physical block numbers for the journal
617 *
618 * On external journals the journal blocks are identity-mapped, so
619 * this is a no-op. If needed, we can use j_blk_offset - everything is
620 * ready.
621 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700622int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700623 unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700624{
625 int err = 0;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700626 unsigned long long ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700627
628 if (journal->j_inode) {
629 ret = bmap(journal->j_inode, blocknr);
630 if (ret)
631 *retp = ret;
632 else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700633 printk(KERN_ALERT "%s: journal block not found "
634 "at offset %lu on %s\n",
Theodore Ts'o05496762008-09-16 14:36:17 -0400635 __func__, blocknr, journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700636 err = -EIO;
637 __journal_abort_soft(journal, err);
638 }
639 } else {
640 *retp = blocknr; /* +journal->j_blk_offset */
641 }
642 return err;
643}
644
645/*
646 * We play buffer_head aliasing tricks to write data/metadata blocks to
647 * the journal without copying their contents, but for journal
648 * descriptor blocks we do need to generate bona fide buffers.
649 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700650 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
Dave Kleikamp470decc2006-10-11 01:20:57 -0700651 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
652 * But we don't bother doing that, so there will be coherency problems with
653 * mmaps of blockdevs which hold live JBD-controlled filesystems.
654 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700655struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700656{
657 struct buffer_head *bh;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700658 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700659 int err;
660
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700661 err = jbd2_journal_next_log_block(journal, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700662
663 if (err)
664 return NULL;
665
666 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500667 if (!bh)
668 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700669 lock_buffer(bh);
670 memset(bh->b_data, 0, journal->j_blocksize);
671 set_buffer_uptodate(bh);
672 unlock_buffer(bh);
673 BUFFER_TRACE(bh, "return this buffer");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700674 return jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700675}
676
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500677struct jbd2_stats_proc_session {
678 journal_t *journal;
679 struct transaction_stats_s *stats;
680 int start;
681 int max;
682};
683
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500684static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
685{
686 return *pos ? NULL : SEQ_START_TOKEN;
687}
688
689static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
690{
691 return NULL;
692}
693
694static int jbd2_seq_info_show(struct seq_file *seq, void *v)
695{
696 struct jbd2_stats_proc_session *s = seq->private;
697
698 if (v != SEQ_START_TOKEN)
699 return 0;
Theodore Ts'obf699322009-09-30 00:32:06 -0400700 seq_printf(seq, "%lu transaction, each up to %u blocks\n",
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500701 s->stats->ts_tid,
702 s->journal->j_max_transaction_buffers);
703 if (s->stats->ts_tid == 0)
704 return 0;
705 seq_printf(seq, "average: \n %ums waiting for transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400706 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500707 seq_printf(seq, " %ums running transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400708 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500709 seq_printf(seq, " %ums transaction was being locked\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400710 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500711 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400712 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500713 seq_printf(seq, " %ums logging transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400714 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -0500715 seq_printf(seq, " %lluus average transaction commit time\n",
716 div_u64(s->journal->j_average_commit_time, 1000));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500717 seq_printf(seq, " %lu handles per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400718 s->stats->run.rs_handle_count / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500719 seq_printf(seq, " %lu blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400720 s->stats->run.rs_blocks / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500721 seq_printf(seq, " %lu logged blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400722 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500723 return 0;
724}
725
726static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
727{
728}
729
James Morris88e9d342009-09-22 16:43:43 -0700730static const struct seq_operations jbd2_seq_info_ops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500731 .start = jbd2_seq_info_start,
732 .next = jbd2_seq_info_next,
733 .stop = jbd2_seq_info_stop,
734 .show = jbd2_seq_info_show,
735};
736
737static int jbd2_seq_info_open(struct inode *inode, struct file *file)
738{
739 journal_t *journal = PDE(inode)->data;
740 struct jbd2_stats_proc_session *s;
741 int rc, size;
742
743 s = kmalloc(sizeof(*s), GFP_KERNEL);
744 if (s == NULL)
745 return -ENOMEM;
746 size = sizeof(struct transaction_stats_s);
747 s->stats = kmalloc(size, GFP_KERNEL);
748 if (s->stats == NULL) {
749 kfree(s);
750 return -ENOMEM;
751 }
752 spin_lock(&journal->j_history_lock);
753 memcpy(s->stats, &journal->j_stats, size);
754 s->journal = journal;
755 spin_unlock(&journal->j_history_lock);
756
757 rc = seq_open(file, &jbd2_seq_info_ops);
758 if (rc == 0) {
759 struct seq_file *m = file->private_data;
760 m->private = s;
761 } else {
762 kfree(s->stats);
763 kfree(s);
764 }
765 return rc;
766
767}
768
769static int jbd2_seq_info_release(struct inode *inode, struct file *file)
770{
771 struct seq_file *seq = file->private_data;
772 struct jbd2_stats_proc_session *s = seq->private;
773 kfree(s->stats);
774 kfree(s);
775 return seq_release(inode, file);
776}
777
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700778static const struct file_operations jbd2_seq_info_fops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500779 .owner = THIS_MODULE,
780 .open = jbd2_seq_info_open,
781 .read = seq_read,
782 .llseek = seq_lseek,
783 .release = jbd2_seq_info_release,
784};
785
786static struct proc_dir_entry *proc_jbd2_stats;
787
788static void jbd2_stats_proc_init(journal_t *journal)
789{
Theodore Ts'o05496762008-09-16 14:36:17 -0400790 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500791 if (journal->j_proc_entry) {
Denis V. Lunev79da3662008-04-29 01:02:11 -0700792 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
793 &jbd2_seq_info_fops, journal);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500794 }
795}
796
797static void jbd2_stats_proc_exit(journal_t *journal)
798{
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500799 remove_proc_entry("info", journal->j_proc_entry);
Theodore Ts'o05496762008-09-16 14:36:17 -0400800 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500801}
802
Dave Kleikamp470decc2006-10-11 01:20:57 -0700803/*
804 * Management for journal control blocks: functions to create and
805 * destroy journal_t structures, and to initialise and read existing
806 * journal blocks from disk. */
807
808/* First: create and setup a journal_t object in memory. We initialise
809 * very few fields yet: that has to wait until we have created the
810 * journal structures from from scratch, or loaded them from disk. */
811
812static journal_t * journal_init_common (void)
813{
814 journal_t *journal;
815 int err;
816
Mingming Caod802ffa2007-10-16 18:38:25 -0400817 journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700818 if (!journal)
819 goto fail;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700820
821 init_waitqueue_head(&journal->j_wait_transaction_locked);
822 init_waitqueue_head(&journal->j_wait_logspace);
823 init_waitqueue_head(&journal->j_wait_done_commit);
824 init_waitqueue_head(&journal->j_wait_checkpoint);
825 init_waitqueue_head(&journal->j_wait_commit);
826 init_waitqueue_head(&journal->j_wait_updates);
827 mutex_init(&journal->j_barrier);
828 mutex_init(&journal->j_checkpoint_mutex);
829 spin_lock_init(&journal->j_revoke_lock);
830 spin_lock_init(&journal->j_list_lock);
831 spin_lock_init(&journal->j_state_lock);
832
Mingming Caocd02ff02007-10-16 18:38:25 -0400833 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
Theodore Ts'o30773842009-01-03 20:27:38 -0500834 journal->j_min_batch_time = 0;
835 journal->j_max_batch_time = 15000; /* 15ms */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700836
837 /* The journal is marked for error until we succeed with recovery! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700838 journal->j_flags = JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700839
840 /* Set up a default-sized revoke table for the new mount. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700841 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700842 if (err) {
843 kfree(journal);
844 goto fail;
845 }
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500846
Theodore Ts'obf699322009-09-30 00:32:06 -0400847 spin_lock_init(&journal->j_history_lock);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500848
Dave Kleikamp470decc2006-10-11 01:20:57 -0700849 return journal;
850fail:
851 return NULL;
852}
853
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700854/* jbd2_journal_init_dev and jbd2_journal_init_inode:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700855 *
856 * Create a journal structure assigned some fixed set of disk blocks to
857 * the journal. We don't actually touch those disk blocks yet, but we
858 * need to set up all of the mapping information to tell the journaling
859 * system where the journal blocks are.
860 *
861 */
862
863/**
Randy Dunlap5648ba52008-04-17 10:38:59 -0400864 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
Dave Kleikamp470decc2006-10-11 01:20:57 -0700865 * @bdev: Block device on which to create the journal
866 * @fs_dev: Device which hold journalled filesystem for this journal.
867 * @start: Block nr Start of journal.
868 * @len: Length of the journal in blocks.
869 * @blocksize: blocksize of journalling device
Randy Dunlap5648ba52008-04-17 10:38:59 -0400870 *
871 * Returns: a newly created journal_t *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700872 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700873 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
Dave Kleikamp470decc2006-10-11 01:20:57 -0700874 * range of blocks on an arbitrary block device.
875 *
876 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700877journal_t * jbd2_journal_init_dev(struct block_device *bdev,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700878 struct block_device *fs_dev,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700879 unsigned long long start, int len, int blocksize)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700880{
881 journal_t *journal = journal_init_common();
882 struct buffer_head *bh;
Theodore Ts'o05496762008-09-16 14:36:17 -0400883 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700884 int n;
885
886 if (!journal)
887 return NULL;
888
889 /* journal descriptor can store up to n blocks -bzzz */
890 journal->j_blocksize = blocksize;
Jan Kara4b905672009-01-06 14:53:35 -0500891 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700892 n = journal->j_blocksize / sizeof(journal_block_tag_t);
893 journal->j_wbufsize = n;
894 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
895 if (!journal->j_wbuf) {
896 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400897 __func__);
Jan Kara4b905672009-01-06 14:53:35 -0500898 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700899 }
900 journal->j_dev = bdev;
901 journal->j_fs_dev = fs_dev;
902 journal->j_blk_offset = start;
903 journal->j_maxlen = len;
Theodore Ts'o05496762008-09-16 14:36:17 -0400904 bdevname(journal->j_dev, journal->j_devname);
905 p = journal->j_devname;
906 while ((p = strchr(p, '/')))
907 *p = '!';
Dave Kleikamp470decc2006-10-11 01:20:57 -0700908
909 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500910 if (!bh) {
911 printk(KERN_ERR
912 "%s: Cannot get buffer for journal superblock\n",
913 __func__);
914 goto out_err;
915 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700916 journal->j_sb_buffer = bh;
917 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Kara4b905672009-01-06 14:53:35 -0500918
Dave Kleikamp470decc2006-10-11 01:20:57 -0700919 return journal;
Jan Kara4b905672009-01-06 14:53:35 -0500920out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800921 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -0500922 jbd2_stats_proc_exit(journal);
923 kfree(journal);
924 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700925}
926
927/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700928 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700929 * @inode: An inode to create the journal in
930 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700931 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
Dave Kleikamp470decc2006-10-11 01:20:57 -0700932 * the journal. The inode must exist already, must support bmap() and
933 * must have all data blocks preallocated.
934 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700935journal_t * jbd2_journal_init_inode (struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700936{
937 struct buffer_head *bh;
938 journal_t *journal = journal_init_common();
Theodore Ts'o05496762008-09-16 14:36:17 -0400939 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700940 int err;
941 int n;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700942 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700943
944 if (!journal)
945 return NULL;
946
947 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
948 journal->j_inode = inode;
Theodore Ts'o05496762008-09-16 14:36:17 -0400949 bdevname(journal->j_dev, journal->j_devname);
950 p = journal->j_devname;
951 while ((p = strchr(p, '/')))
952 *p = '!';
953 p = journal->j_devname + strlen(journal->j_devname);
Theodore Ts'o90576c02009-09-29 15:51:30 -0400954 sprintf(p, "-%lu", journal->j_inode->i_ino);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700955 jbd_debug(1,
956 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
957 journal, inode->i_sb->s_id, inode->i_ino,
958 (long long) inode->i_size,
959 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
960
961 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
962 journal->j_blocksize = inode->i_sb->s_blocksize;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500963 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700964
965 /* journal descriptor can store up to n blocks -bzzz */
966 n = journal->j_blocksize / sizeof(journal_block_tag_t);
967 journal->j_wbufsize = n;
968 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
969 if (!journal->j_wbuf) {
970 printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400971 __func__);
Jan Kara4b905672009-01-06 14:53:35 -0500972 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700973 }
974
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700975 err = jbd2_journal_bmap(journal, 0, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700976 /* If that failed, give up */
977 if (err) {
978 printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400979 __func__);
Jan Kara4b905672009-01-06 14:53:35 -0500980 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700981 }
982
983 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500984 if (!bh) {
985 printk(KERN_ERR
986 "%s: Cannot get buffer for journal superblock\n",
987 __func__);
988 goto out_err;
989 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700990 journal->j_sb_buffer = bh;
991 journal->j_superblock = (journal_superblock_t *)bh->b_data;
992
993 return journal;
Jan Kara4b905672009-01-06 14:53:35 -0500994out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800995 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -0500996 jbd2_stats_proc_exit(journal);
997 kfree(journal);
998 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700999}
1000
1001/*
1002 * If the journal init or create aborts, we need to mark the journal
1003 * superblock as being NULL to prevent the journal destroy from writing
1004 * back a bogus superblock.
1005 */
1006static void journal_fail_superblock (journal_t *journal)
1007{
1008 struct buffer_head *bh = journal->j_sb_buffer;
1009 brelse(bh);
1010 journal->j_sb_buffer = NULL;
1011}
1012
1013/*
1014 * Given a journal_t structure, initialise the various fields for
1015 * startup of a new journaling session. We use this both when creating
1016 * a journal, and after recovering an old journal to reset it for
1017 * subsequent use.
1018 */
1019
1020static int journal_reset(journal_t *journal)
1021{
1022 journal_superblock_t *sb = journal->j_superblock;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001023 unsigned long long first, last;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001024
1025 first = be32_to_cpu(sb->s_first);
1026 last = be32_to_cpu(sb->s_maxlen);
Jan Karaf6f50e22009-07-17 10:40:01 -04001027 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1028 printk(KERN_ERR "JBD: Journal too short (blocks %llu-%llu).\n",
1029 first, last);
1030 journal_fail_superblock(journal);
1031 return -EINVAL;
1032 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001033
1034 journal->j_first = first;
1035 journal->j_last = last;
1036
1037 journal->j_head = first;
1038 journal->j_tail = first;
1039 journal->j_free = last - first;
1040
1041 journal->j_tail_sequence = journal->j_transaction_sequence;
1042 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1043 journal->j_commit_request = journal->j_commit_sequence;
1044
1045 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1046
1047 /* Add the dynamic fields and write it to disk. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001048 jbd2_journal_update_superblock(journal, 1);
Pavel Emelianov97f06782007-05-08 00:30:42 -07001049 return jbd2_journal_start_thread(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001050}
1051
1052/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001053 * void jbd2_journal_update_superblock() - Update journal sb on disk.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001054 * @journal: The journal to update.
1055 * @wait: Set to '0' if you don't want to wait for IO completion.
1056 *
1057 * Update a journal's dynamic superblock fields and write it to disk,
1058 * optionally waiting for the IO to complete.
1059 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001060void jbd2_journal_update_superblock(journal_t *journal, int wait)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001061{
1062 journal_superblock_t *sb = journal->j_superblock;
1063 struct buffer_head *bh = journal->j_sb_buffer;
1064
1065 /*
1066 * As a special case, if the on-disk copy is already marked as needing
1067 * no recovery (s_start == 0) and there are no outstanding transactions
1068 * in the filesystem, then we can safely defer the superblock update
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001069 * until the next commit by setting JBD2_FLUSHED. This avoids
Dave Kleikamp470decc2006-10-11 01:20:57 -07001070 * attempting a write to a potential-readonly device.
1071 */
1072 if (sb->s_start == 0 && journal->j_tail_sequence ==
1073 journal->j_transaction_sequence) {
1074 jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1075 "(start %ld, seq %d, errno %d)\n",
1076 journal->j_tail, journal->j_tail_sequence,
1077 journal->j_errno);
1078 goto out;
1079 }
1080
Theodore Ts'o914258b2008-10-06 21:35:40 -04001081 if (buffer_write_io_error(bh)) {
1082 /*
1083 * Oh, dear. A previous attempt to write the journal
1084 * superblock failed. This could happen because the
1085 * USB device was yanked out. Or it could happen to
1086 * be a transient write error and maybe the block will
1087 * be remapped. Nothing we can do but to retry the
1088 * write and hope for the best.
1089 */
1090 printk(KERN_ERR "JBD2: previous I/O error detected "
1091 "for journal superblock update for %s.\n",
1092 journal->j_devname);
1093 clear_buffer_write_io_error(bh);
1094 set_buffer_uptodate(bh);
1095 }
1096
Dave Kleikamp470decc2006-10-11 01:20:57 -07001097 spin_lock(&journal->j_state_lock);
1098 jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1099 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1100
1101 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1102 sb->s_start = cpu_to_be32(journal->j_tail);
1103 sb->s_errno = cpu_to_be32(journal->j_errno);
1104 spin_unlock(&journal->j_state_lock);
1105
1106 BUFFER_TRACE(bh, "marking dirty");
1107 mark_buffer_dirty(bh);
Theodore Ts'o914258b2008-10-06 21:35:40 -04001108 if (wait) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001109 sync_dirty_buffer(bh);
Theodore Ts'o914258b2008-10-06 21:35:40 -04001110 if (buffer_write_io_error(bh)) {
1111 printk(KERN_ERR "JBD2: I/O error detected "
1112 "when updating journal superblock for %s.\n",
1113 journal->j_devname);
1114 clear_buffer_write_io_error(bh);
1115 set_buffer_uptodate(bh);
1116 }
1117 } else
Dave Kleikamp470decc2006-10-11 01:20:57 -07001118 ll_rw_block(SWRITE, 1, &bh);
1119
1120out:
1121 /* If we have just flushed the log (by marking s_start==0), then
1122 * any future commit will have to be careful to update the
1123 * superblock again to re-record the true start of the log. */
1124
1125 spin_lock(&journal->j_state_lock);
1126 if (sb->s_start)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001127 journal->j_flags &= ~JBD2_FLUSHED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001128 else
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001129 journal->j_flags |= JBD2_FLUSHED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001130 spin_unlock(&journal->j_state_lock);
1131}
1132
1133/*
1134 * Read the superblock for a given journal, performing initial
1135 * validation of the format.
1136 */
1137
1138static int journal_get_superblock(journal_t *journal)
1139{
1140 struct buffer_head *bh;
1141 journal_superblock_t *sb;
1142 int err = -EIO;
1143
1144 bh = journal->j_sb_buffer;
1145
1146 J_ASSERT(bh != NULL);
1147 if (!buffer_uptodate(bh)) {
1148 ll_rw_block(READ, 1, &bh);
1149 wait_on_buffer(bh);
1150 if (!buffer_uptodate(bh)) {
1151 printk (KERN_ERR
1152 "JBD: IO error reading journal superblock\n");
1153 goto out;
1154 }
1155 }
1156
1157 sb = journal->j_superblock;
1158
1159 err = -EINVAL;
1160
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001161 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001162 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1163 printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1164 goto out;
1165 }
1166
1167 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001168 case JBD2_SUPERBLOCK_V1:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001169 journal->j_format_version = 1;
1170 break;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001171 case JBD2_SUPERBLOCK_V2:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001172 journal->j_format_version = 2;
1173 break;
1174 default:
1175 printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1176 goto out;
1177 }
1178
1179 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1180 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1181 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1182 printk (KERN_WARNING "JBD: journal file too short\n");
1183 goto out;
1184 }
1185
1186 return 0;
1187
1188out:
1189 journal_fail_superblock(journal);
1190 return err;
1191}
1192
1193/*
1194 * Load the on-disk journal superblock and read the key fields into the
1195 * journal_t.
1196 */
1197
1198static int load_superblock(journal_t *journal)
1199{
1200 int err;
1201 journal_superblock_t *sb;
1202
1203 err = journal_get_superblock(journal);
1204 if (err)
1205 return err;
1206
1207 sb = journal->j_superblock;
1208
1209 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1210 journal->j_tail = be32_to_cpu(sb->s_start);
1211 journal->j_first = be32_to_cpu(sb->s_first);
1212 journal->j_last = be32_to_cpu(sb->s_maxlen);
1213 journal->j_errno = be32_to_cpu(sb->s_errno);
1214
1215 return 0;
1216}
1217
1218
1219/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001220 * int jbd2_journal_load() - Read journal from disk.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001221 * @journal: Journal to act on.
1222 *
1223 * Given a journal_t structure which tells us which disk blocks contain
1224 * a journal, read the journal from disk to initialise the in-memory
1225 * structures.
1226 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001227int jbd2_journal_load(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001228{
1229 int err;
1230 journal_superblock_t *sb;
1231
1232 err = load_superblock(journal);
1233 if (err)
1234 return err;
1235
1236 sb = journal->j_superblock;
1237 /* If this is a V2 superblock, then we have to check the
1238 * features flags on it. */
1239
1240 if (journal->j_format_version >= 2) {
1241 if ((sb->s_feature_ro_compat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001242 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001243 (sb->s_feature_incompat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001244 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001245 printk (KERN_WARNING
1246 "JBD: Unrecognised features on journal\n");
1247 return -EINVAL;
1248 }
1249 }
1250
Dave Kleikamp470decc2006-10-11 01:20:57 -07001251 /* Let the recovery code check whether it needs to recover any
1252 * data from the journal. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001253 if (jbd2_journal_recover(journal))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001254 goto recovery_error;
1255
Theodore Ts'oe6a47422009-11-15 15:31:37 -05001256 if (journal->j_failed_commit) {
1257 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1258 "is corrupt.\n", journal->j_failed_commit,
1259 journal->j_devname);
1260 return -EIO;
1261 }
1262
Dave Kleikamp470decc2006-10-11 01:20:57 -07001263 /* OK, we've finished with the dynamic journal bits:
1264 * reinitialise the dynamic contents of the superblock in memory
1265 * and reset them on disk. */
1266 if (journal_reset(journal))
1267 goto recovery_error;
1268
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001269 journal->j_flags &= ~JBD2_ABORT;
1270 journal->j_flags |= JBD2_LOADED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001271 return 0;
1272
1273recovery_error:
1274 printk (KERN_WARNING "JBD: recovery failed\n");
1275 return -EIO;
1276}
1277
1278/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001279 * void jbd2_journal_destroy() - Release a journal_t structure.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001280 * @journal: Journal to act on.
1281 *
1282 * Release a journal_t structure once it is no longer in use by the
1283 * journaled object.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001284 * Return <0 if we couldn't clean up the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001285 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001286int jbd2_journal_destroy(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001287{
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001288 int err = 0;
1289
Dave Kleikamp470decc2006-10-11 01:20:57 -07001290 /* Wait for the commit thread to wake up and die. */
1291 journal_kill_thread(journal);
1292
1293 /* Force a final log commit */
1294 if (journal->j_running_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001295 jbd2_journal_commit_transaction(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001296
1297 /* Force any old transactions to disk */
1298
1299 /* Totally anal locking here... */
1300 spin_lock(&journal->j_list_lock);
1301 while (journal->j_checkpoint_transactions != NULL) {
1302 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001303 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001304 jbd2_log_do_checkpoint(journal);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001305 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001306 spin_lock(&journal->j_list_lock);
1307 }
1308
1309 J_ASSERT(journal->j_running_transaction == NULL);
1310 J_ASSERT(journal->j_committing_transaction == NULL);
1311 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1312 spin_unlock(&journal->j_list_lock);
1313
Dave Kleikamp470decc2006-10-11 01:20:57 -07001314 if (journal->j_sb_buffer) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001315 if (!is_journal_aborted(journal)) {
1316 /* We can now mark the journal as empty. */
1317 journal->j_tail = 0;
1318 journal->j_tail_sequence =
1319 ++journal->j_transaction_sequence;
1320 jbd2_journal_update_superblock(journal, 1);
1321 } else {
1322 err = -EIO;
1323 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001324 brelse(journal->j_sb_buffer);
1325 }
1326
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001327 if (journal->j_proc_entry)
1328 jbd2_stats_proc_exit(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001329 if (journal->j_inode)
1330 iput(journal->j_inode);
1331 if (journal->j_revoke)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001332 jbd2_journal_destroy_revoke(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001333 kfree(journal->j_wbuf);
1334 kfree(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001335
1336 return err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001337}
1338
1339
1340/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001341 *int jbd2_journal_check_used_features () - Check if features specified are used.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001342 * @journal: Journal to check.
1343 * @compat: bitmask of compatible features
1344 * @ro: bitmask of features that force read-only mount
1345 * @incompat: bitmask of incompatible features
1346 *
1347 * Check whether the journal uses all of a given set of
1348 * features. Return true (non-zero) if it does.
1349 **/
1350
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001351int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001352 unsigned long ro, unsigned long incompat)
1353{
1354 journal_superblock_t *sb;
1355
1356 if (!compat && !ro && !incompat)
1357 return 1;
1358 if (journal->j_format_version == 1)
1359 return 0;
1360
1361 sb = journal->j_superblock;
1362
1363 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1364 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1365 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1366 return 1;
1367
1368 return 0;
1369}
1370
1371/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001372 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001373 * @journal: Journal to check.
1374 * @compat: bitmask of compatible features
1375 * @ro: bitmask of features that force read-only mount
1376 * @incompat: bitmask of incompatible features
1377 *
1378 * Check whether the journaling code supports the use of
1379 * all of a given set of features on this journal. Return true
1380 * (non-zero) if it can. */
1381
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001382int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001383 unsigned long ro, unsigned long incompat)
1384{
1385 journal_superblock_t *sb;
1386
1387 if (!compat && !ro && !incompat)
1388 return 1;
1389
1390 sb = journal->j_superblock;
1391
1392 /* We can support any known requested features iff the
1393 * superblock is in version 2. Otherwise we fail to support any
1394 * extended sb features. */
1395
1396 if (journal->j_format_version != 2)
1397 return 0;
1398
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001399 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1400 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1401 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001402 return 1;
1403
1404 return 0;
1405}
1406
1407/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001408 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
Dave Kleikamp470decc2006-10-11 01:20:57 -07001409 * @journal: Journal to act on.
1410 * @compat: bitmask of compatible features
1411 * @ro: bitmask of features that force read-only mount
1412 * @incompat: bitmask of incompatible features
1413 *
1414 * Mark a given journal feature as present on the
1415 * superblock. Returns true if the requested features could be set.
1416 *
1417 */
1418
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001419int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001420 unsigned long ro, unsigned long incompat)
1421{
1422 journal_superblock_t *sb;
1423
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001424 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001425 return 1;
1426
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001427 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001428 return 0;
1429
1430 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1431 compat, ro, incompat);
1432
1433 sb = journal->j_superblock;
1434
1435 sb->s_feature_compat |= cpu_to_be32(compat);
1436 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1437 sb->s_feature_incompat |= cpu_to_be32(incompat);
1438
1439 return 1;
1440}
1441
Girish Shilamkar818d2762008-01-28 23:58:27 -05001442/*
1443 * jbd2_journal_clear_features () - Clear a given journal feature in the
1444 * superblock
1445 * @journal: Journal to act on.
1446 * @compat: bitmask of compatible features
1447 * @ro: bitmask of features that force read-only mount
1448 * @incompat: bitmask of incompatible features
1449 *
1450 * Clear a given journal feature as present on the
1451 * superblock.
1452 */
1453void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1454 unsigned long ro, unsigned long incompat)
1455{
1456 journal_superblock_t *sb;
1457
1458 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1459 compat, ro, incompat);
1460
1461 sb = journal->j_superblock;
1462
1463 sb->s_feature_compat &= ~cpu_to_be32(compat);
1464 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1465 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1466}
1467EXPORT_SYMBOL(jbd2_journal_clear_features);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001468
1469/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001470 * int jbd2_journal_update_format () - Update on-disk journal structure.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001471 * @journal: Journal to act on.
1472 *
1473 * Given an initialised but unloaded journal struct, poke about in the
1474 * on-disk structure to update it to the most recent supported version.
1475 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001476int jbd2_journal_update_format (journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001477{
1478 journal_superblock_t *sb;
1479 int err;
1480
1481 err = journal_get_superblock(journal);
1482 if (err)
1483 return err;
1484
1485 sb = journal->j_superblock;
1486
1487 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001488 case JBD2_SUPERBLOCK_V2:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001489 return 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001490 case JBD2_SUPERBLOCK_V1:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001491 return journal_convert_superblock_v1(journal, sb);
1492 default:
1493 break;
1494 }
1495 return -EINVAL;
1496}
1497
1498static int journal_convert_superblock_v1(journal_t *journal,
1499 journal_superblock_t *sb)
1500{
1501 int offset, blocksize;
1502 struct buffer_head *bh;
1503
1504 printk(KERN_WARNING
1505 "JBD: Converting superblock from version 1 to 2.\n");
1506
1507 /* Pre-initialise new fields to zero */
1508 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1509 blocksize = be32_to_cpu(sb->s_blocksize);
1510 memset(&sb->s_feature_compat, 0, blocksize-offset);
1511
1512 sb->s_nr_users = cpu_to_be32(1);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001513 sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001514 journal->j_format_version = 2;
1515
1516 bh = journal->j_sb_buffer;
1517 BUFFER_TRACE(bh, "marking dirty");
1518 mark_buffer_dirty(bh);
1519 sync_dirty_buffer(bh);
1520 return 0;
1521}
1522
1523
1524/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001525 * int jbd2_journal_flush () - Flush journal
Dave Kleikamp470decc2006-10-11 01:20:57 -07001526 * @journal: Journal to act on.
1527 *
1528 * Flush all data for a given journal to disk and empty the journal.
1529 * Filesystems can use this when remounting readonly to ensure that
1530 * recovery does not need to happen on remount.
1531 */
1532
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001533int jbd2_journal_flush(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001534{
1535 int err = 0;
1536 transaction_t *transaction = NULL;
1537 unsigned long old_tail;
1538
1539 spin_lock(&journal->j_state_lock);
1540
1541 /* Force everything buffered to the log... */
1542 if (journal->j_running_transaction) {
1543 transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001544 __jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001545 } else if (journal->j_committing_transaction)
1546 transaction = journal->j_committing_transaction;
1547
1548 /* Wait for the log commit to complete... */
1549 if (transaction) {
1550 tid_t tid = transaction->t_tid;
1551
1552 spin_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001553 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001554 } else {
1555 spin_unlock(&journal->j_state_lock);
1556 }
1557
1558 /* ...and flush everything in the log out to disk. */
1559 spin_lock(&journal->j_list_lock);
1560 while (!err && journal->j_checkpoint_transactions != NULL) {
1561 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001562 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001563 err = jbd2_log_do_checkpoint(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001564 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001565 spin_lock(&journal->j_list_lock);
1566 }
1567 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001568
1569 if (is_journal_aborted(journal))
1570 return -EIO;
1571
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001572 jbd2_cleanup_journal_tail(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001573
1574 /* Finally, mark the journal as really needing no recovery.
1575 * This sets s_start==0 in the underlying superblock, which is
1576 * the magic code for a fully-recovered superblock. Any future
1577 * commits of data to the journal will restore the current
1578 * s_start value. */
1579 spin_lock(&journal->j_state_lock);
1580 old_tail = journal->j_tail;
1581 journal->j_tail = 0;
1582 spin_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001583 jbd2_journal_update_superblock(journal, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001584 spin_lock(&journal->j_state_lock);
1585 journal->j_tail = old_tail;
1586
1587 J_ASSERT(!journal->j_running_transaction);
1588 J_ASSERT(!journal->j_committing_transaction);
1589 J_ASSERT(!journal->j_checkpoint_transactions);
1590 J_ASSERT(journal->j_head == journal->j_tail);
1591 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1592 spin_unlock(&journal->j_state_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001593 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001594}
1595
1596/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001597 * int jbd2_journal_wipe() - Wipe journal contents
Dave Kleikamp470decc2006-10-11 01:20:57 -07001598 * @journal: Journal to act on.
1599 * @write: flag (see below)
1600 *
1601 * Wipe out all of the contents of a journal, safely. This will produce
1602 * a warning if the journal contains any valid recovery information.
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001603 * Must be called between journal_init_*() and jbd2_journal_load().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001604 *
1605 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1606 * we merely suppress recovery.
1607 */
1608
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001609int jbd2_journal_wipe(journal_t *journal, int write)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001610{
1611 journal_superblock_t *sb;
1612 int err = 0;
1613
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001614 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001615
1616 err = load_superblock(journal);
1617 if (err)
1618 return err;
1619
1620 sb = journal->j_superblock;
1621
1622 if (!journal->j_tail)
1623 goto no_recovery;
1624
1625 printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1626 write ? "Clearing" : "Ignoring");
1627
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001628 err = jbd2_journal_skip_recovery(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001629 if (write)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001630 jbd2_journal_update_superblock(journal, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001631
1632 no_recovery:
1633 return err;
1634}
1635
1636/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07001637 * Journal abort has very specific semantics, which we describe
1638 * for journal abort.
1639 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001640 * Two internal functions, which provide abort to the jbd layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001641 * itself are here.
1642 */
1643
1644/*
1645 * Quick version for internal journal use (doesn't lock the journal).
1646 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1647 * and don't attempt to make any other journal updates.
1648 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001649void __jbd2_journal_abort_hard(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001650{
1651 transaction_t *transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001652
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001653 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001654 return;
1655
1656 printk(KERN_ERR "Aborting journal on device %s.\n",
Theodore Ts'o05496762008-09-16 14:36:17 -04001657 journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001658
1659 spin_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001660 journal->j_flags |= JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001661 transaction = journal->j_running_transaction;
1662 if (transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001663 __jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001664 spin_unlock(&journal->j_state_lock);
1665}
1666
1667/* Soft abort: record the abort error status in the journal superblock,
1668 * but don't do any other IO. */
1669static void __journal_abort_soft (journal_t *journal, int errno)
1670{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001671 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001672 return;
1673
1674 if (!journal->j_errno)
1675 journal->j_errno = errno;
1676
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001677 __jbd2_journal_abort_hard(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001678
1679 if (errno)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001680 jbd2_journal_update_superblock(journal, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001681}
1682
1683/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001684 * void jbd2_journal_abort () - Shutdown the journal immediately.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001685 * @journal: the journal to shutdown.
1686 * @errno: an error number to record in the journal indicating
1687 * the reason for the shutdown.
1688 *
1689 * Perform a complete, immediate shutdown of the ENTIRE
1690 * journal (not of a single transaction). This operation cannot be
1691 * undone without closing and reopening the journal.
1692 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001693 * The jbd2_journal_abort function is intended to support higher level error
Dave Kleikamp470decc2006-10-11 01:20:57 -07001694 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1695 * mode.
1696 *
1697 * Journal abort has very specific semantics. Any existing dirty,
1698 * unjournaled buffers in the main filesystem will still be written to
1699 * disk by bdflush, but the journaling mechanism will be suspended
1700 * immediately and no further transaction commits will be honoured.
1701 *
1702 * Any dirty, journaled buffers will be written back to disk without
1703 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1704 * filesystem, but we _do_ attempt to leave as much data as possible
1705 * behind for fsck to use for cleanup.
1706 *
1707 * Any attempt to get a new transaction handle on a journal which is in
1708 * ABORT state will just result in an -EROFS error return. A
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001709 * jbd2_journal_stop on an existing handle will return -EIO if we have
Dave Kleikamp470decc2006-10-11 01:20:57 -07001710 * entered abort state during the update.
1711 *
1712 * Recursive transactions are not disturbed by journal abort until the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001713 * final jbd2_journal_stop, which will receive the -EIO error.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001714 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001715 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
Dave Kleikamp470decc2006-10-11 01:20:57 -07001716 * which will be recorded (if possible) in the journal superblock. This
1717 * allows a client to record failure conditions in the middle of a
1718 * transaction without having to complete the transaction to record the
1719 * failure to disk. ext3_error, for example, now uses this
1720 * functionality.
1721 *
1722 * Errors which originate from within the journaling layer will NOT
1723 * supply an errno; a null errno implies that absolutely no further
1724 * writes are done to the journal (unless there are any already in
1725 * progress).
1726 *
1727 */
1728
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001729void jbd2_journal_abort(journal_t *journal, int errno)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001730{
1731 __journal_abort_soft(journal, errno);
1732}
1733
1734/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001735 * int jbd2_journal_errno () - returns the journal's error state.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001736 * @journal: journal to examine.
1737 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001738 * This is the errno number set with jbd2_journal_abort(), the last
Dave Kleikamp470decc2006-10-11 01:20:57 -07001739 * time the journal was mounted - if the journal was stopped
1740 * without calling abort this will be 0.
1741 *
1742 * If the journal has been aborted on this mount time -EROFS will
1743 * be returned.
1744 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001745int jbd2_journal_errno(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001746{
1747 int err;
1748
1749 spin_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001750 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001751 err = -EROFS;
1752 else
1753 err = journal->j_errno;
1754 spin_unlock(&journal->j_state_lock);
1755 return err;
1756}
1757
1758/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001759 * int jbd2_journal_clear_err () - clears the journal's error state
Dave Kleikamp470decc2006-10-11 01:20:57 -07001760 * @journal: journal to act on.
1761 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001762 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07001763 * mode.
1764 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001765int jbd2_journal_clear_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001766{
1767 int err = 0;
1768
1769 spin_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001770 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001771 err = -EROFS;
1772 else
1773 journal->j_errno = 0;
1774 spin_unlock(&journal->j_state_lock);
1775 return err;
1776}
1777
1778/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001779 * void jbd2_journal_ack_err() - Ack journal err.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001780 * @journal: journal to act on.
1781 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001782 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07001783 * mode.
1784 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001785void jbd2_journal_ack_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001786{
1787 spin_lock(&journal->j_state_lock);
1788 if (journal->j_errno)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001789 journal->j_flags |= JBD2_ACK_ERR;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001790 spin_unlock(&journal->j_state_lock);
1791}
1792
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001793int jbd2_journal_blocks_per_page(struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001794{
1795 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1796}
1797
1798/*
Zach Brownb517bea2006-10-11 01:21:08 -07001799 * helper functions to deal with 32 or 64bit block numbers.
1800 */
1801size_t journal_tag_bytes(journal_t *journal)
1802{
1803 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
Mingming Caocd02ff02007-10-16 18:38:25 -04001804 return JBD2_TAG_SIZE64;
Zach Brownb517bea2006-10-11 01:21:08 -07001805 else
Mingming Caocd02ff02007-10-16 18:38:25 -04001806 return JBD2_TAG_SIZE32;
Zach Brownb517bea2006-10-11 01:21:08 -07001807}
1808
1809/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07001810 * Journal_head storage management
1811 */
Christoph Lametere18b8902006-12-06 20:33:20 -08001812static struct kmem_cache *jbd2_journal_head_cache;
Jose R. Santose23291b2007-07-18 08:57:06 -04001813#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07001814static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1815#endif
1816
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001817static int journal_init_jbd2_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001818{
1819 int retval;
1820
Al Viro1076d172008-03-29 03:07:18 +00001821 J_ASSERT(jbd2_journal_head_cache == NULL);
Johann Lombardia920e942006-10-11 01:21:00 -07001822 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
Dave Kleikamp470decc2006-10-11 01:20:57 -07001823 sizeof(struct journal_head),
1824 0, /* offset */
Mingming Cao77160952008-01-28 23:58:27 -05001825 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09001826 NULL); /* ctor */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001827 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00001828 if (!jbd2_journal_head_cache) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001829 retval = -ENOMEM;
1830 printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1831 }
1832 return retval;
1833}
1834
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001835static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001836{
Duane Griffin8a9362e2008-04-17 10:38:59 -04001837 if (jbd2_journal_head_cache) {
1838 kmem_cache_destroy(jbd2_journal_head_cache);
1839 jbd2_journal_head_cache = NULL;
1840 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001841}
1842
1843/*
1844 * journal_head splicing and dicing
1845 */
1846static struct journal_head *journal_alloc_journal_head(void)
1847{
1848 struct journal_head *ret;
1849 static unsigned long last_warning;
1850
Jose R. Santose23291b2007-07-18 08:57:06 -04001851#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07001852 atomic_inc(&nr_journal_heads);
1853#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001854 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
Al Viro1076d172008-03-29 03:07:18 +00001855 if (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001856 jbd_debug(1, "out of memory for journal_head\n");
1857 if (time_after(jiffies, last_warning + 5*HZ)) {
1858 printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001859 __func__);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001860 last_warning = jiffies;
1861 }
Al Viro1076d172008-03-29 03:07:18 +00001862 while (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001863 yield();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001864 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001865 }
1866 }
1867 return ret;
1868}
1869
1870static void journal_free_journal_head(struct journal_head *jh)
1871{
Jose R. Santose23291b2007-07-18 08:57:06 -04001872#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07001873 atomic_dec(&nr_journal_heads);
Mingming Caocd02ff02007-10-16 18:38:25 -04001874 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001875#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001876 kmem_cache_free(jbd2_journal_head_cache, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001877}
1878
1879/*
1880 * A journal_head is attached to a buffer_head whenever JBD has an
1881 * interest in the buffer.
1882 *
1883 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1884 * is set. This bit is tested in core kernel code where we need to take
1885 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
1886 * there.
1887 *
1888 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1889 *
1890 * When a buffer has its BH_JBD bit set it is immune from being released by
1891 * core kernel code, mainly via ->b_count.
1892 *
1893 * A journal_head may be detached from its buffer_head when the journal_head's
1894 * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001895 * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
Dave Kleikamp470decc2006-10-11 01:20:57 -07001896 * journal_head can be dropped if needed.
1897 *
1898 * Various places in the kernel want to attach a journal_head to a buffer_head
1899 * _before_ attaching the journal_head to a transaction. To protect the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001900 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
Dave Kleikamp470decc2006-10-11 01:20:57 -07001901 * journal_head's b_jcount refcount by one. The caller must call
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001902 * jbd2_journal_put_journal_head() to undo this.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001903 *
1904 * So the typical usage would be:
1905 *
1906 * (Attach a journal_head if needed. Increments b_jcount)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001907 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001908 * ...
1909 * jh->b_transaction = xxx;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001910 * jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001911 *
1912 * Now, the journal_head's b_jcount is zero, but it is safe from being released
1913 * because it has a non-zero b_transaction.
1914 */
1915
1916/*
1917 * Give a buffer_head a journal_head.
1918 *
1919 * Doesn't need the journal lock.
1920 * May sleep.
1921 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001922struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001923{
1924 struct journal_head *jh;
1925 struct journal_head *new_jh = NULL;
1926
1927repeat:
1928 if (!buffer_jbd(bh)) {
1929 new_jh = journal_alloc_journal_head();
1930 memset(new_jh, 0, sizeof(*new_jh));
1931 }
1932
1933 jbd_lock_bh_journal_head(bh);
1934 if (buffer_jbd(bh)) {
1935 jh = bh2jh(bh);
1936 } else {
1937 J_ASSERT_BH(bh,
1938 (atomic_read(&bh->b_count) > 0) ||
1939 (bh->b_page && bh->b_page->mapping));
1940
1941 if (!new_jh) {
1942 jbd_unlock_bh_journal_head(bh);
1943 goto repeat;
1944 }
1945
1946 jh = new_jh;
1947 new_jh = NULL; /* We consumed it */
1948 set_buffer_jbd(bh);
1949 bh->b_private = jh;
1950 jh->b_bh = bh;
1951 get_bh(bh);
1952 BUFFER_TRACE(bh, "added journal_head");
1953 }
1954 jh->b_jcount++;
1955 jbd_unlock_bh_journal_head(bh);
1956 if (new_jh)
1957 journal_free_journal_head(new_jh);
1958 return bh->b_private;
1959}
1960
1961/*
1962 * Grab a ref against this buffer_head's journal_head. If it ended up not
1963 * having a journal_head, return NULL
1964 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001965struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001966{
1967 struct journal_head *jh = NULL;
1968
1969 jbd_lock_bh_journal_head(bh);
1970 if (buffer_jbd(bh)) {
1971 jh = bh2jh(bh);
1972 jh->b_jcount++;
1973 }
1974 jbd_unlock_bh_journal_head(bh);
1975 return jh;
1976}
1977
1978static void __journal_remove_journal_head(struct buffer_head *bh)
1979{
1980 struct journal_head *jh = bh2jh(bh);
1981
1982 J_ASSERT_JH(jh, jh->b_jcount >= 0);
1983
1984 get_bh(bh);
1985 if (jh->b_jcount == 0) {
1986 if (jh->b_transaction == NULL &&
1987 jh->b_next_transaction == NULL &&
1988 jh->b_cp_transaction == NULL) {
1989 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1990 J_ASSERT_BH(bh, buffer_jbd(bh));
1991 J_ASSERT_BH(bh, jh2bh(jh) == bh);
1992 BUFFER_TRACE(bh, "remove journal_head");
1993 if (jh->b_frozen_data) {
1994 printk(KERN_WARNING "%s: freeing "
1995 "b_frozen_data\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001996 __func__);
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001997 jbd2_free(jh->b_frozen_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001998 }
1999 if (jh->b_committed_data) {
2000 printk(KERN_WARNING "%s: freeing "
2001 "b_committed_data\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04002002 __func__);
Mingming Caoaf1e76d2007-10-16 18:38:25 -04002003 jbd2_free(jh->b_committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002004 }
2005 bh->b_private = NULL;
2006 jh->b_bh = NULL; /* debug, really */
2007 clear_buffer_jbd(bh);
2008 __brelse(bh);
2009 journal_free_journal_head(jh);
2010 } else {
2011 BUFFER_TRACE(bh, "journal_head was locked");
2012 }
2013 }
2014}
2015
2016/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002017 * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
Dave Kleikamp470decc2006-10-11 01:20:57 -07002018 * and has a zero b_jcount then remove and release its journal_head. If we did
2019 * see that the buffer is not used by any transaction we also "logically"
2020 * decrement ->b_count.
2021 *
2022 * We in fact take an additional increment on ->b_count as a convenience,
2023 * because the caller usually wants to do additional things with the bh
2024 * after calling here.
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002025 * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
Dave Kleikamp470decc2006-10-11 01:20:57 -07002026 * time. Once the caller has run __brelse(), the buffer is eligible for
2027 * reaping by try_to_free_buffers().
2028 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002029void jbd2_journal_remove_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002030{
2031 jbd_lock_bh_journal_head(bh);
2032 __journal_remove_journal_head(bh);
2033 jbd_unlock_bh_journal_head(bh);
2034}
2035
2036/*
2037 * Drop a reference on the passed journal_head. If it fell to zero then try to
2038 * release the journal_head from the buffer_head.
2039 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002040void jbd2_journal_put_journal_head(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002041{
2042 struct buffer_head *bh = jh2bh(jh);
2043
2044 jbd_lock_bh_journal_head(bh);
2045 J_ASSERT_JH(jh, jh->b_jcount > 0);
2046 --jh->b_jcount;
2047 if (!jh->b_jcount && !jh->b_transaction) {
2048 __journal_remove_journal_head(bh);
2049 __brelse(bh);
2050 }
2051 jbd_unlock_bh_journal_head(bh);
2052}
2053
2054/*
Jan Karac851ed52008-07-11 19:27:31 -04002055 * Initialize jbd inode head
2056 */
2057void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2058{
2059 jinode->i_transaction = NULL;
2060 jinode->i_next_transaction = NULL;
2061 jinode->i_vfs_inode = inode;
2062 jinode->i_flags = 0;
2063 INIT_LIST_HEAD(&jinode->i_list);
2064}
2065
2066/*
2067 * Function to be called before we start removing inode from memory (i.e.,
2068 * clear_inode() is a fine place to be called from). It removes inode from
2069 * transaction's lists.
2070 */
2071void jbd2_journal_release_jbd_inode(journal_t *journal,
2072 struct jbd2_inode *jinode)
2073{
2074 int writeout = 0;
2075
2076 if (!journal)
2077 return;
2078restart:
2079 spin_lock(&journal->j_list_lock);
2080 /* Is commit writing out inode - we have to wait */
2081 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2082 wait_queue_head_t *wq;
2083 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2084 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2085 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2086 spin_unlock(&journal->j_list_lock);
2087 schedule();
2088 finish_wait(wq, &wait.wait);
2089 goto restart;
2090 }
2091
2092 /* Do we need to wait for data writeback? */
2093 if (journal->j_committing_transaction == jinode->i_transaction)
2094 writeout = 1;
2095 if (jinode->i_transaction) {
2096 list_del(&jinode->i_list);
2097 jinode->i_transaction = NULL;
2098 }
2099 spin_unlock(&journal->j_list_lock);
2100}
2101
2102/*
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002103 * debugfs tunables
Dave Kleikamp470decc2006-10-11 01:20:57 -07002104 */
Jose R. Santos6f38c742007-10-16 18:38:25 -04002105#ifdef CONFIG_JBD2_DEBUG
2106u8 jbd2_journal_enable_debug __read_mostly;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002107EXPORT_SYMBOL(jbd2_journal_enable_debug);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002108
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002109#define JBD2_DEBUG_NAME "jbd2-debug"
Dave Kleikamp470decc2006-10-11 01:20:57 -07002110
Jose R. Santos6f38c742007-10-16 18:38:25 -04002111static struct dentry *jbd2_debugfs_dir;
2112static struct dentry *jbd2_debug;
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002113
2114static void __init jbd2_create_debugfs_entry(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002115{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002116 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2117 if (jbd2_debugfs_dir)
2118 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2119 jbd2_debugfs_dir,
2120 &jbd2_journal_enable_debug);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002121}
2122
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002123static void __exit jbd2_remove_debugfs_entry(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002124{
Jose R. Santos6f38c742007-10-16 18:38:25 -04002125 debugfs_remove(jbd2_debug);
2126 debugfs_remove(jbd2_debugfs_dir);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002127}
2128
2129#else
2130
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002131static void __init jbd2_create_debugfs_entry(void)
2132{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002133}
2134
2135static void __exit jbd2_remove_debugfs_entry(void)
2136{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002137}
Dave Kleikamp470decc2006-10-11 01:20:57 -07002138
2139#endif
2140
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002141#ifdef CONFIG_PROC_FS
2142
2143#define JBD2_STATS_PROC_NAME "fs/jbd2"
2144
2145static void __init jbd2_create_jbd_stats_proc_entry(void)
2146{
2147 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2148}
2149
2150static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2151{
2152 if (proc_jbd2_stats)
2153 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2154}
2155
2156#else
2157
2158#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2159#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2160
2161#endif
2162
Christoph Lametere18b8902006-12-06 20:33:20 -08002163struct kmem_cache *jbd2_handle_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002164
2165static int __init journal_init_handle_cache(void)
2166{
Johann Lombardia920e942006-10-11 01:21:00 -07002167 jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002168 sizeof(handle_t),
2169 0, /* offset */
Mingming Cao77160952008-01-28 23:58:27 -05002170 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09002171 NULL); /* ctor */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002172 if (jbd2_handle_cache == NULL) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002173 printk(KERN_EMERG "JBD: failed to create handle cache\n");
2174 return -ENOMEM;
2175 }
2176 return 0;
2177}
2178
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002179static void jbd2_journal_destroy_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002180{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002181 if (jbd2_handle_cache)
2182 kmem_cache_destroy(jbd2_handle_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002183}
2184
2185/*
2186 * Module startup and shutdown
2187 */
2188
2189static int __init journal_init_caches(void)
2190{
2191 int ret;
2192
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002193 ret = jbd2_journal_init_revoke_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002194 if (ret == 0)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002195 ret = journal_init_jbd2_journal_head_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002196 if (ret == 0)
2197 ret = journal_init_handle_cache();
2198 return ret;
2199}
2200
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002201static void jbd2_journal_destroy_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002202{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002203 jbd2_journal_destroy_revoke_caches();
2204 jbd2_journal_destroy_jbd2_journal_head_cache();
2205 jbd2_journal_destroy_handle_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002206}
2207
2208static int __init journal_init(void)
2209{
2210 int ret;
2211
2212 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2213
2214 ret = journal_init_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002215 if (ret == 0) {
2216 jbd2_create_debugfs_entry();
2217 jbd2_create_jbd_stats_proc_entry();
2218 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002219 jbd2_journal_destroy_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002220 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002221 return ret;
2222}
2223
2224static void __exit journal_exit(void)
2225{
Jose R. Santose23291b2007-07-18 08:57:06 -04002226#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002227 int n = atomic_read(&nr_journal_heads);
2228 if (n)
2229 printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2230#endif
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002231 jbd2_remove_debugfs_entry();
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002232 jbd2_remove_jbd_stats_proc_entry();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002233 jbd2_journal_destroy_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002234}
2235
Theodore Ts'o879c5e62009-06-17 11:47:48 -04002236/*
2237 * jbd2_dev_to_name is a utility function used by the jbd2 and ext4
2238 * tracing infrastructure to map a dev_t to a device name.
2239 *
2240 * The caller should use rcu_read_lock() in order to make sure the
2241 * device name stays valid until its done with it. We use
2242 * rcu_read_lock() as well to make sure we're safe in case the caller
2243 * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2244 * nested.
2245 */
2246struct devname_cache {
2247 struct rcu_head rcu;
2248 dev_t device;
2249 char devname[BDEVNAME_SIZE];
2250};
2251#define CACHE_SIZE_BITS 6
2252static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
2253static DEFINE_SPINLOCK(devname_cache_lock);
2254
2255static void free_devcache(struct rcu_head *rcu)
2256{
2257 kfree(rcu);
2258}
2259
2260const char *jbd2_dev_to_name(dev_t device)
2261{
2262 int i = hash_32(device, CACHE_SIZE_BITS);
2263 char *ret;
2264 struct block_device *bd;
Theodore Ts'ob5744802009-06-20 23:34:44 -04002265 static struct devname_cache *new_dev;
Theodore Ts'o879c5e62009-06-17 11:47:48 -04002266
2267 rcu_read_lock();
2268 if (devcache[i] && devcache[i]->device == device) {
2269 ret = devcache[i]->devname;
2270 rcu_read_unlock();
2271 return ret;
2272 }
2273 rcu_read_unlock();
2274
Theodore Ts'ob5744802009-06-20 23:34:44 -04002275 new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2276 if (!new_dev)
2277 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
Theodore Ts'o879c5e62009-06-17 11:47:48 -04002278 spin_lock(&devname_cache_lock);
2279 if (devcache[i]) {
2280 if (devcache[i]->device == device) {
Theodore Ts'ob5744802009-06-20 23:34:44 -04002281 kfree(new_dev);
Theodore Ts'o879c5e62009-06-17 11:47:48 -04002282 ret = devcache[i]->devname;
2283 spin_unlock(&devname_cache_lock);
2284 return ret;
2285 }
2286 call_rcu(&devcache[i]->rcu, free_devcache);
2287 }
Theodore Ts'ob5744802009-06-20 23:34:44 -04002288 devcache[i] = new_dev;
Theodore Ts'o879c5e62009-06-17 11:47:48 -04002289 devcache[i]->device = device;
2290 bd = bdget(device);
2291 if (bd) {
2292 bdevname(bd, devcache[i]->devname);
2293 bdput(bd);
2294 } else
2295 __bdevname(device, devcache[i]->devname);
2296 ret = devcache[i]->devname;
2297 spin_unlock(&devname_cache_lock);
2298 return ret;
2299}
2300EXPORT_SYMBOL(jbd2_dev_to_name);
2301
Dave Kleikamp470decc2006-10-11 01:20:57 -07002302MODULE_LICENSE("GPL");
2303module_init(journal_init);
2304module_exit(journal_exit);
2305