blob: f6974688e89fea72297bba1da0061c1949b242ee [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>
Theodore Ts'od2eecb02009-12-07 10:36:20 -050042#include <linux/log2.h>
43#include <linux/vmalloc.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040044#include <linux/backing-dev.h>
Brian King39e3ac22010-10-27 21:25:12 -040045#include <linux/bitops.h>
Theodore Ts'o670be5a2010-12-17 10:44:16 -050046#include <linux/ratelimit.h>
Theodore Ts'o879c5e62009-06-17 11:47:48 -040047
48#define CREATE_TRACE_POINTS
49#include <trace/events/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070050
51#include <asm/uaccess.h>
52#include <asm/page.h>
53
Mingming Caof7f4bcc2006-10-11 01:20:59 -070054EXPORT_SYMBOL(jbd2_journal_extend);
55EXPORT_SYMBOL(jbd2_journal_stop);
56EXPORT_SYMBOL(jbd2_journal_lock_updates);
57EXPORT_SYMBOL(jbd2_journal_unlock_updates);
58EXPORT_SYMBOL(jbd2_journal_get_write_access);
59EXPORT_SYMBOL(jbd2_journal_get_create_access);
60EXPORT_SYMBOL(jbd2_journal_get_undo_access);
Joel Beckere06c8222008-09-11 15:35:47 -070061EXPORT_SYMBOL(jbd2_journal_set_triggers);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070062EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
63EXPORT_SYMBOL(jbd2_journal_release_buffer);
64EXPORT_SYMBOL(jbd2_journal_forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -070065#if 0
66EXPORT_SYMBOL(journal_sync_buffer);
67#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -070068EXPORT_SYMBOL(jbd2_journal_flush);
69EXPORT_SYMBOL(jbd2_journal_revoke);
Dave Kleikamp470decc2006-10-11 01:20:57 -070070
Mingming Caof7f4bcc2006-10-11 01:20:59 -070071EXPORT_SYMBOL(jbd2_journal_init_dev);
72EXPORT_SYMBOL(jbd2_journal_init_inode);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070073EXPORT_SYMBOL(jbd2_journal_check_used_features);
74EXPORT_SYMBOL(jbd2_journal_check_available_features);
75EXPORT_SYMBOL(jbd2_journal_set_features);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070076EXPORT_SYMBOL(jbd2_journal_load);
77EXPORT_SYMBOL(jbd2_journal_destroy);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070078EXPORT_SYMBOL(jbd2_journal_abort);
79EXPORT_SYMBOL(jbd2_journal_errno);
80EXPORT_SYMBOL(jbd2_journal_ack_err);
81EXPORT_SYMBOL(jbd2_journal_clear_err);
82EXPORT_SYMBOL(jbd2_log_wait_commit);
Theodore Ts'o3b799d12009-12-09 20:42:53 -050083EXPORT_SYMBOL(jbd2_log_start_commit);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070084EXPORT_SYMBOL(jbd2_journal_start_commit);
85EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
86EXPORT_SYMBOL(jbd2_journal_wipe);
87EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
88EXPORT_SYMBOL(jbd2_journal_invalidatepage);
89EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
90EXPORT_SYMBOL(jbd2_journal_force_commit);
Jan Karac851ed52008-07-11 19:27:31 -040091EXPORT_SYMBOL(jbd2_journal_file_inode);
92EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
93EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
94EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -050095EXPORT_SYMBOL(jbd2_inode_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -070096
Dave Kleikamp470decc2006-10-11 01:20:57 -070097static void __journal_abort_soft (journal_t *journal, int errno);
Theodore Ts'od2eecb02009-12-07 10:36:20 -050098static int jbd2_journal_create_slab(size_t slab_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -070099
100/*
101 * Helper function used to manage commit timeouts
102 */
103
104static void commit_timeout(unsigned long __data)
105{
106 struct task_struct * p = (struct task_struct *) __data;
107
108 wake_up_process(p);
109}
110
111/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700112 * kjournald2: The main thread function used to manage a logging device
Dave Kleikamp470decc2006-10-11 01:20:57 -0700113 * journal.
114 *
115 * This kernel thread is responsible for two things:
116 *
117 * 1) COMMIT: Every so often we need to commit the current state of the
118 * filesystem to disk. The journal thread is responsible for writing
119 * all of the metadata buffers to disk.
120 *
121 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
122 * of the data in that part of the log has been rewritten elsewhere on
123 * the disk. Flushing these old buffers to reclaim space in the log is
124 * known as checkpointing, and this thread is responsible for that job.
125 */
126
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700127static int kjournald2(void *arg)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700128{
129 journal_t *journal = arg;
130 transaction_t *transaction;
131
132 /*
133 * Set up an interval timer which can be used to trigger a commit wakeup
134 * after the commit interval expires
135 */
136 setup_timer(&journal->j_commit_timer, commit_timeout,
137 (unsigned long)current);
138
Nigel Cunningham35c80422012-02-03 19:59:41 +1100139 set_freezable();
140
Dave Kleikamp470decc2006-10-11 01:20:57 -0700141 /* Record that the journal thread is running */
142 journal->j_task = current;
143 wake_up(&journal->j_wait_done_commit);
144
Dave Kleikamp470decc2006-10-11 01:20:57 -0700145 /*
146 * And now, wait forever for commit wakeup events.
147 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400148 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700149
150loop:
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700151 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700152 goto end_loop;
153
154 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
155 journal->j_commit_sequence, journal->j_commit_request);
156
157 if (journal->j_commit_sequence != journal->j_commit_request) {
158 jbd_debug(1, "OK, requests differ\n");
Theodore Ts'oa931da62010-08-03 21:35:12 -0400159 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700160 del_timer_sync(&journal->j_commit_timer);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700161 jbd2_journal_commit_transaction(journal);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400162 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700163 goto loop;
164 }
165
166 wake_up(&journal->j_wait_done_commit);
167 if (freezing(current)) {
168 /*
169 * The simpler the better. Flushing journal isn't a
170 * good idea, because that depends on threads that may
171 * be already stopped.
172 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700173 jbd_debug(1, "Now suspending kjournald2\n");
Theodore Ts'oa931da62010-08-03 21:35:12 -0400174 write_unlock(&journal->j_state_lock);
Tejun Heoa0acae02011-11-21 12:32:22 -0800175 try_to_freeze();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400176 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700177 } else {
178 /*
179 * We assume on resume that commits are already there,
180 * so we don't sleep
181 */
182 DEFINE_WAIT(wait);
183 int should_sleep = 1;
184
185 prepare_to_wait(&journal->j_wait_commit, &wait,
186 TASK_INTERRUPTIBLE);
187 if (journal->j_commit_sequence != journal->j_commit_request)
188 should_sleep = 0;
189 transaction = journal->j_running_transaction;
190 if (transaction && time_after_eq(jiffies,
191 transaction->t_expires))
192 should_sleep = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700193 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700194 should_sleep = 0;
195 if (should_sleep) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400196 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700197 schedule();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400198 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700199 }
200 finish_wait(&journal->j_wait_commit, &wait);
201 }
202
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700203 jbd_debug(1, "kjournald2 wakes\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700204
205 /*
206 * Were we woken up by a commit wakeup event?
207 */
208 transaction = journal->j_running_transaction;
209 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
210 journal->j_commit_request = transaction->t_tid;
211 jbd_debug(1, "woke because of timeout\n");
212 }
213 goto loop;
214
215end_loop:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400216 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700217 del_timer_sync(&journal->j_commit_timer);
218 journal->j_task = NULL;
219 wake_up(&journal->j_wait_done_commit);
220 jbd_debug(1, "Journal thread exiting.\n");
221 return 0;
222}
223
Pavel Emelianov97f06782007-05-08 00:30:42 -0700224static int jbd2_journal_start_thread(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700225{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700226 struct task_struct *t;
227
Theodore Ts'o90576c02009-09-29 15:51:30 -0400228 t = kthread_run(kjournald2, journal, "jbd2/%s",
229 journal->j_devname);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700230 if (IS_ERR(t))
231 return PTR_ERR(t);
232
Al Viro1076d172008-03-29 03:07:18 +0000233 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700234 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700235}
236
237static void journal_kill_thread(journal_t *journal)
238{
Theodore Ts'oa931da62010-08-03 21:35:12 -0400239 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700240 journal->j_flags |= JBD2_UNMOUNT;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700241
242 while (journal->j_task) {
243 wake_up(&journal->j_wait_commit);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400244 write_unlock(&journal->j_state_lock);
Al Viro1076d172008-03-29 03:07:18 +0000245 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400246 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700247 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400248 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700249}
250
251/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700252 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700253 *
254 * Writes a metadata buffer to a given disk block. The actual IO is not
255 * performed but a new buffer_head is constructed which labels the data
256 * to be written with the correct destination disk block.
257 *
258 * Any magic-number escaping which needs to be done will cause a
259 * copy-out here. If the buffer happens to start with the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700260 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700261 * magic number is only written to the log for descripter blocks. In
262 * this case, we copy the data and replace the first word with 0, and we
263 * return a result code which indicates that this buffer needs to be
264 * marked as an escaped buffer in the corresponding log descriptor
265 * block. The missing word can then be restored when the block is read
266 * during recovery.
267 *
268 * If the source buffer has already been modified by a new transaction
269 * since we took the last commit snapshot, we use the frozen copy of
270 * that data for IO. If we end up using the existing buffer_head's data
271 * for the write, then we *have* to lock the buffer to prevent anyone
272 * else from using and possibly modifying it while the IO is in
273 * progress.
274 *
275 * The function returns a pointer to the buffer_heads to be used for IO.
276 *
277 * We assume that the journal has already been locked in this function.
278 *
279 * Return value:
280 * <0: Error
281 * >=0: Finished OK
282 *
283 * On success:
284 * Bit 0 set == escape performed on the data
285 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
286 */
287
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700288int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700289 struct journal_head *jh_in,
290 struct journal_head **jh_out,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700291 unsigned long long blocknr)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700292{
293 int need_copy_out = 0;
294 int done_copy_out = 0;
295 int do_escape = 0;
296 char *mapped_data;
297 struct buffer_head *new_bh;
298 struct journal_head *new_jh;
299 struct page *new_page;
300 unsigned int new_offset;
301 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghua96577c42009-07-13 17:55:35 -0400302 journal_t *journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700303
304 /*
305 * The buffer really shouldn't be locked: only the current committing
306 * transaction is allowed to write it, so nobody else is allowed
307 * to do any IO.
308 *
309 * akpm: except if we're journalling data, and write() output is
310 * also part of a shared mapping, and another thread has
311 * decided to launch a writepage() against this buffer.
312 */
313 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
314
Theodore Ts'o47def822010-07-27 11:56:05 -0400315retry_alloc:
316 new_bh = alloc_buffer_head(GFP_NOFS);
317 if (!new_bh) {
318 /*
319 * Failure is not an option, but __GFP_NOFAIL is going
320 * away; so we retry ourselves here.
321 */
322 congestion_wait(BLK_RW_ASYNC, HZ/50);
323 goto retry_alloc;
324 }
325
dingdinghua96577c42009-07-13 17:55:35 -0400326 /* keep subsequent assertions sane */
327 new_bh->b_state = 0;
328 init_buffer(new_bh, NULL, NULL);
329 atomic_set(&new_bh->b_count, 1);
330 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700331
332 /*
333 * If a new transaction has already done a buffer copy-out, then
334 * we use that version of the data for the commit.
335 */
336 jbd_lock_bh_state(bh_in);
337repeat:
338 if (jh_in->b_frozen_data) {
339 done_copy_out = 1;
340 new_page = virt_to_page(jh_in->b_frozen_data);
341 new_offset = offset_in_page(jh_in->b_frozen_data);
342 } else {
343 new_page = jh2bh(jh_in)->b_page;
344 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
345 }
346
Cong Wang303a8f22011-11-25 23:14:31 +0800347 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700348 /*
Jan Kara13ceef02010-07-14 07:56:33 +0200349 * Fire data frozen trigger if data already wasn't frozen. Do this
350 * before checking for escaping, as the trigger may modify the magic
351 * offset. If a copy-out happens afterwards, it will have the correct
352 * data in the buffer.
Joel Beckere06c8222008-09-11 15:35:47 -0700353 */
Jan Kara13ceef02010-07-14 07:56:33 +0200354 if (!done_copy_out)
355 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
356 jh_in->b_triggers);
Joel Beckere06c8222008-09-11 15:35:47 -0700357
358 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700359 * Check for escaping
360 */
361 if (*((__be32 *)(mapped_data + new_offset)) ==
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700362 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700363 need_copy_out = 1;
364 do_escape = 1;
365 }
Cong Wang303a8f22011-11-25 23:14:31 +0800366 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700367
368 /*
369 * Do we need to do a data copy?
370 */
371 if (need_copy_out && !done_copy_out) {
372 char *tmp;
373
374 jbd_unlock_bh_state(bh_in);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400375 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500376 if (!tmp) {
377 jbd2_journal_put_journal_head(new_jh);
378 return -ENOMEM;
379 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700380 jbd_lock_bh_state(bh_in);
381 if (jh_in->b_frozen_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400382 jbd2_free(tmp, bh_in->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700383 goto repeat;
384 }
385
386 jh_in->b_frozen_data = tmp;
Cong Wang303a8f22011-11-25 23:14:31 +0800387 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700388 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
Cong Wang303a8f22011-11-25 23:14:31 +0800389 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700390
391 new_page = virt_to_page(tmp);
392 new_offset = offset_in_page(tmp);
393 done_copy_out = 1;
Joel Beckere06c8222008-09-11 15:35:47 -0700394
395 /*
396 * This isn't strictly necessary, as we're using frozen
397 * data for the escaping, but it keeps consistency with
398 * b_frozen_data usage.
399 */
400 jh_in->b_frozen_triggers = jh_in->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700401 }
402
403 /*
404 * Did we need to do an escaping? Now we've done all the
405 * copying, we can finally do so.
406 */
407 if (do_escape) {
Cong Wang303a8f22011-11-25 23:14:31 +0800408 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700409 *((unsigned int *)(mapped_data + new_offset)) = 0;
Cong Wang303a8f22011-11-25 23:14:31 +0800410 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700411 }
412
Dave Kleikamp470decc2006-10-11 01:20:57 -0700413 set_bh_page(new_bh, new_page, new_offset);
414 new_jh->b_transaction = NULL;
415 new_bh->b_size = jh2bh(jh_in)->b_size;
416 new_bh->b_bdev = transaction->t_journal->j_dev;
417 new_bh->b_blocknr = blocknr;
418 set_buffer_mapped(new_bh);
419 set_buffer_dirty(new_bh);
420
421 *jh_out = new_jh;
422
423 /*
424 * The to-be-written buffer needs to get moved to the io queue,
425 * and the original buffer whose contents we are shadowing or
426 * copying is moved to the transaction's shadow queue.
427 */
428 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghua96577c42009-07-13 17:55:35 -0400429 spin_lock(&journal->j_list_lock);
430 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
431 spin_unlock(&journal->j_list_lock);
432 jbd_unlock_bh_state(bh_in);
433
Dave Kleikamp470decc2006-10-11 01:20:57 -0700434 JBUFFER_TRACE(new_jh, "file as BJ_IO");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700435 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700436
437 return do_escape | (done_copy_out << 1);
438}
439
440/*
441 * Allocation code for the journal file. Manage the space left in the
442 * journal, so that we can begin checkpointing when appropriate.
443 */
444
445/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700446 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700447 *
448 * Called with the journal already locked.
449 *
450 * Called under j_state_lock
451 */
452
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700453int __jbd2_log_space_left(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700454{
455 int left = journal->j_free;
456
Theodore Ts'oa931da62010-08-03 21:35:12 -0400457 /* assert_spin_locked(&journal->j_state_lock); */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700458
459 /*
460 * Be pessimistic here about the number of those free blocks which
461 * might be required for log descriptor control blocks.
462 */
463
464#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
465
466 left -= MIN_LOG_RESERVED_BLOCKS;
467
468 if (left <= 0)
469 return 0;
470 left -= (left >> 3);
471 return left;
472}
473
474/*
Theodore Ts'oe4471832011-02-12 08:18:24 -0500475 * Called with j_state_lock locked for writing.
476 * Returns true if a transaction commit was started.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700477 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700478int __jbd2_log_start_commit(journal_t *journal, tid_t target)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700479{
480 /*
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400481 * The only transaction we can possibly wait upon is the
482 * currently running transaction (if it exists). Otherwise,
483 * the target tid must be an old one.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700484 */
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400485 if (journal->j_running_transaction &&
486 journal->j_running_transaction->t_tid == target) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700487 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200488 * We want a new commit: OK, mark the request and wakeup the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700489 * commit thread. We do _not_ do the commit ourselves.
490 */
491
492 journal->j_commit_request = target;
Eryu Guanf2a44522011-11-01 19:09:18 -0400493 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700494 journal->j_commit_request,
495 journal->j_commit_sequence);
496 wake_up(&journal->j_wait_commit);
497 return 1;
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400498 } else if (!tid_geq(journal->j_commit_request, target))
499 /* This should never happen, but if it does, preserve
500 the evidence before kjournald goes into a loop and
501 increments j_commit_sequence beyond all recognition. */
Eryu Guanf2a44522011-11-01 19:09:18 -0400502 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
Theodore Ts'o1be2add2011-05-08 19:37:54 -0400503 journal->j_commit_request,
504 journal->j_commit_sequence,
505 target, journal->j_running_transaction ?
506 journal->j_running_transaction->t_tid : 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700507 return 0;
508}
509
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700510int jbd2_log_start_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700511{
512 int ret;
513
Theodore Ts'oa931da62010-08-03 21:35:12 -0400514 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700515 ret = __jbd2_log_start_commit(journal, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400516 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700517 return ret;
518}
519
520/*
521 * Force and wait upon a commit if the calling process is not within
522 * transaction. This is used for forcing out undo-protected data which contains
523 * bitmaps, when the fs is running out of space.
524 *
525 * We can only force the running transaction if we don't have an active handle;
526 * otherwise, we will deadlock.
527 *
528 * Returns true if a transaction was started.
529 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700530int jbd2_journal_force_commit_nested(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700531{
532 transaction_t *transaction = NULL;
533 tid_t tid;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500534 int need_to_start = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700535
Theodore Ts'oa931da62010-08-03 21:35:12 -0400536 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700537 if (journal->j_running_transaction && !current->journal_info) {
538 transaction = journal->j_running_transaction;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500539 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
540 need_to_start = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700541 } else if (journal->j_committing_transaction)
542 transaction = journal->j_committing_transaction;
543
544 if (!transaction) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400545 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700546 return 0; /* Nothing to retry */
547 }
548
549 tid = transaction->t_tid;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400550 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500551 if (need_to_start)
552 jbd2_log_start_commit(journal, tid);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700553 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700554 return 1;
555}
556
557/*
558 * Start a commit of the current running transaction (if any). Returns true
Jan Karac88ccea2009-02-10 11:27:46 -0500559 * if a transaction is going to be committed (or is currently already
560 * committing), and fills its tid in at *ptid
Dave Kleikamp470decc2006-10-11 01:20:57 -0700561 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700562int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700563{
564 int ret = 0;
565
Theodore Ts'oa931da62010-08-03 21:35:12 -0400566 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700567 if (journal->j_running_transaction) {
568 tid_t tid = journal->j_running_transaction->t_tid;
569
Jan Karac88ccea2009-02-10 11:27:46 -0500570 __jbd2_log_start_commit(journal, tid);
571 /* There's a running transaction and we've just made sure
572 * it's commit has been scheduled. */
573 if (ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700574 *ptid = tid;
Jan Karac88ccea2009-02-10 11:27:46 -0500575 ret = 1;
576 } else if (journal->j_committing_transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700577 /*
578 * If ext3_write_super() recently started a commit, then we
579 * have to wait for completion of that transaction
580 */
Jan Karac88ccea2009-02-10 11:27:46 -0500581 if (ptid)
582 *ptid = journal->j_committing_transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700583 ret = 1;
584 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400585 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700586 return ret;
587}
588
589/*
Jan Karabbd2be32011-05-24 11:59:18 -0400590 * Return 1 if a given transaction has not yet sent barrier request
591 * connected with a transaction commit. If 0 is returned, transaction
592 * may or may not have sent the barrier. Used to avoid sending barrier
593 * twice in common cases.
594 */
595int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
596{
597 int ret = 0;
598 transaction_t *commit_trans;
599
600 if (!(journal->j_flags & JBD2_BARRIER))
601 return 0;
602 read_lock(&journal->j_state_lock);
603 /* Transaction already committed? */
604 if (tid_geq(journal->j_commit_sequence, tid))
605 goto out;
606 commit_trans = journal->j_committing_transaction;
607 if (!commit_trans || commit_trans->t_tid != tid) {
608 ret = 1;
609 goto out;
610 }
611 /*
612 * Transaction is being committed and we already proceeded to
613 * submitting a flush to fs partition?
614 */
615 if (journal->j_fs_dev != journal->j_dev) {
616 if (!commit_trans->t_need_data_flush ||
617 commit_trans->t_state >= T_COMMIT_DFLUSH)
618 goto out;
619 } else {
620 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
621 goto out;
622 }
623 ret = 1;
624out:
625 read_unlock(&journal->j_state_lock);
626 return ret;
627}
628EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
629
630/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700631 * Wait for a specified commit to complete.
632 * The caller may not hold the journal lock.
633 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700634int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700635{
636 int err = 0;
637
Theodore Ts'oa931da62010-08-03 21:35:12 -0400638 read_lock(&journal->j_state_lock);
Jose R. Santose23291b2007-07-18 08:57:06 -0400639#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -0700640 if (!tid_geq(journal->j_commit_request, tid)) {
641 printk(KERN_EMERG
642 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400643 __func__, journal->j_commit_request, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700644 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700645#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -0700646 while (tid_gt(tid, journal->j_commit_sequence)) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400647 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700648 tid, journal->j_commit_sequence);
649 wake_up(&journal->j_wait_commit);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400650 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700651 wait_event(journal->j_wait_done_commit,
652 !tid_gt(tid, journal->j_commit_sequence));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400653 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700654 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400655 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700656
657 if (unlikely(is_journal_aborted(journal))) {
658 printk(KERN_EMERG "journal commit I/O error\n");
659 err = -EIO;
660 }
661 return err;
662}
663
664/*
Theodore Ts'oebdc12a2013-04-03 22:02:52 -0400665 * When this function returns the transaction corresponding to tid
666 * will be completed. If the transaction has currently running, start
667 * committing that transaction before waiting for it to complete. If
668 * the transaction id is stale, it is by definition already completed,
669 * so just return SUCCESS.
670 */
671int jbd2_complete_transaction(journal_t *journal, tid_t tid)
672{
673 int need_to_wait = 1;
674
675 read_lock(&journal->j_state_lock);
676 if (journal->j_running_transaction &&
677 journal->j_running_transaction->t_tid == tid) {
678 if (journal->j_commit_request != tid) {
679 /* transaction not yet started, so request it */
680 read_unlock(&journal->j_state_lock);
681 jbd2_log_start_commit(journal, tid);
682 goto wait_commit;
683 }
684 } else if (!(journal->j_committing_transaction &&
685 journal->j_committing_transaction->t_tid == tid))
686 need_to_wait = 0;
687 read_unlock(&journal->j_state_lock);
688 if (!need_to_wait)
689 return 0;
690wait_commit:
691 return jbd2_log_wait_commit(journal, tid);
692}
693EXPORT_SYMBOL(jbd2_complete_transaction);
694
695/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700696 * Log buffer allocation routines:
697 */
698
Mingming Cao18eba7a2006-10-11 01:21:13 -0700699int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700700{
701 unsigned long blocknr;
702
Theodore Ts'oa931da62010-08-03 21:35:12 -0400703 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700704 J_ASSERT(journal->j_free > 1);
705
706 blocknr = journal->j_head;
707 journal->j_head++;
708 journal->j_free--;
709 if (journal->j_head == journal->j_last)
710 journal->j_head = journal->j_first;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400711 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700712 return jbd2_journal_bmap(journal, blocknr, retp);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700713}
714
715/*
716 * Conversion of logical to physical block numbers for the journal
717 *
718 * On external journals the journal blocks are identity-mapped, so
719 * this is a no-op. If needed, we can use j_blk_offset - everything is
720 * ready.
721 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700722int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700723 unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700724{
725 int err = 0;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700726 unsigned long long ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700727
728 if (journal->j_inode) {
729 ret = bmap(journal->j_inode, blocknr);
730 if (ret)
731 *retp = ret;
732 else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700733 printk(KERN_ALERT "%s: journal block not found "
734 "at offset %lu on %s\n",
Theodore Ts'o05496762008-09-16 14:36:17 -0400735 __func__, blocknr, journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700736 err = -EIO;
737 __journal_abort_soft(journal, err);
738 }
739 } else {
740 *retp = blocknr; /* +journal->j_blk_offset */
741 }
742 return err;
743}
744
745/*
746 * We play buffer_head aliasing tricks to write data/metadata blocks to
747 * the journal without copying their contents, but for journal
748 * descriptor blocks we do need to generate bona fide buffers.
749 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700750 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
Dave Kleikamp470decc2006-10-11 01:20:57 -0700751 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
752 * But we don't bother doing that, so there will be coherency problems with
753 * mmaps of blockdevs which hold live JBD-controlled filesystems.
754 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700755struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700756{
757 struct buffer_head *bh;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700758 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700759 int err;
760
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700761 err = jbd2_journal_next_log_block(journal, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700762
763 if (err)
764 return NULL;
765
766 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500767 if (!bh)
768 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700769 lock_buffer(bh);
770 memset(bh->b_data, 0, journal->j_blocksize);
771 set_buffer_uptodate(bh);
772 unlock_buffer(bh);
773 BUFFER_TRACE(bh, "return this buffer");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700774 return jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700775}
776
Jan Kara79feb522012-03-13 22:22:54 -0400777/*
778 * Return tid of the oldest transaction in the journal and block in the journal
779 * where the transaction starts.
780 *
781 * If the journal is now empty, return which will be the next transaction ID
782 * we will write and where will that transaction start.
783 *
784 * The return value is 0 if journal tail cannot be pushed any further, 1 if
785 * it can.
786 */
787int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
788 unsigned long *block)
789{
790 transaction_t *transaction;
791 int ret;
792
793 read_lock(&journal->j_state_lock);
794 spin_lock(&journal->j_list_lock);
795 transaction = journal->j_checkpoint_transactions;
796 if (transaction) {
797 *tid = transaction->t_tid;
798 *block = transaction->t_log_start;
799 } else if ((transaction = journal->j_committing_transaction) != NULL) {
800 *tid = transaction->t_tid;
801 *block = transaction->t_log_start;
802 } else if ((transaction = journal->j_running_transaction) != NULL) {
803 *tid = transaction->t_tid;
804 *block = journal->j_head;
805 } else {
806 *tid = journal->j_transaction_sequence;
807 *block = journal->j_head;
808 }
809 ret = tid_gt(*tid, journal->j_tail_sequence);
810 spin_unlock(&journal->j_list_lock);
811 read_unlock(&journal->j_state_lock);
812
813 return ret;
814}
815
816/*
817 * Update information in journal structure and in on disk journal superblock
818 * about log tail. This function does not check whether information passed in
819 * really pushes log tail further. It's responsibility of the caller to make
820 * sure provided log tail information is valid (e.g. by holding
821 * j_checkpoint_mutex all the time between computing log tail and calling this
822 * function as is the case with jbd2_cleanup_journal_tail()).
823 *
824 * Requires j_checkpoint_mutex
825 */
826void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
827{
828 unsigned long freed;
829
830 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
831
832 /*
833 * We cannot afford for write to remain in drive's caches since as
834 * soon as we update j_tail, next transaction can start reusing journal
835 * space and if we lose sb update during power failure we'd replay
836 * old transaction with possibly newly overwritten data.
837 */
838 jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
839 write_lock(&journal->j_state_lock);
840 freed = block - journal->j_tail;
841 if (block < journal->j_tail)
842 freed += journal->j_last - journal->j_first;
843
844 trace_jbd2_update_log_tail(journal, tid, block, freed);
845 jbd_debug(1,
846 "Cleaning journal tail from %d to %d (offset %lu), "
847 "freeing %lu\n",
848 journal->j_tail_sequence, tid, block, freed);
849
850 journal->j_free += freed;
851 journal->j_tail_sequence = tid;
852 journal->j_tail = block;
853 write_unlock(&journal->j_state_lock);
854}
855
Jan Kara33395782012-03-13 22:45:38 -0400856/*
857 * This is a variaon of __jbd2_update_log_tail which checks for validity of
858 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
859 * with other threads updating log tail.
860 */
861void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
862{
863 mutex_lock(&journal->j_checkpoint_mutex);
864 if (tid_gt(tid, journal->j_tail_sequence))
865 __jbd2_update_log_tail(journal, tid, block);
866 mutex_unlock(&journal->j_checkpoint_mutex);
867}
868
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500869struct jbd2_stats_proc_session {
870 journal_t *journal;
871 struct transaction_stats_s *stats;
872 int start;
873 int max;
874};
875
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500876static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
877{
878 return *pos ? NULL : SEQ_START_TOKEN;
879}
880
881static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
882{
883 return NULL;
884}
885
886static int jbd2_seq_info_show(struct seq_file *seq, void *v)
887{
888 struct jbd2_stats_proc_session *s = seq->private;
889
890 if (v != SEQ_START_TOKEN)
891 return 0;
Theodore Ts'obf699322009-09-30 00:32:06 -0400892 seq_printf(seq, "%lu transaction, each up to %u blocks\n",
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500893 s->stats->ts_tid,
894 s->journal->j_max_transaction_buffers);
895 if (s->stats->ts_tid == 0)
896 return 0;
897 seq_printf(seq, "average: \n %ums waiting for transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400898 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500899 seq_printf(seq, " %ums running transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400900 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500901 seq_printf(seq, " %ums transaction was being locked\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400902 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500903 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400904 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500905 seq_printf(seq, " %ums logging transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400906 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -0500907 seq_printf(seq, " %lluus average transaction commit time\n",
908 div_u64(s->journal->j_average_commit_time, 1000));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500909 seq_printf(seq, " %lu handles per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400910 s->stats->run.rs_handle_count / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500911 seq_printf(seq, " %lu blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400912 s->stats->run.rs_blocks / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500913 seq_printf(seq, " %lu logged blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400914 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500915 return 0;
916}
917
918static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
919{
920}
921
James Morris88e9d342009-09-22 16:43:43 -0700922static const struct seq_operations jbd2_seq_info_ops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500923 .start = jbd2_seq_info_start,
924 .next = jbd2_seq_info_next,
925 .stop = jbd2_seq_info_stop,
926 .show = jbd2_seq_info_show,
927};
928
929static int jbd2_seq_info_open(struct inode *inode, struct file *file)
930{
931 journal_t *journal = PDE(inode)->data;
932 struct jbd2_stats_proc_session *s;
933 int rc, size;
934
935 s = kmalloc(sizeof(*s), GFP_KERNEL);
936 if (s == NULL)
937 return -ENOMEM;
938 size = sizeof(struct transaction_stats_s);
939 s->stats = kmalloc(size, GFP_KERNEL);
940 if (s->stats == NULL) {
941 kfree(s);
942 return -ENOMEM;
943 }
944 spin_lock(&journal->j_history_lock);
945 memcpy(s->stats, &journal->j_stats, size);
946 s->journal = journal;
947 spin_unlock(&journal->j_history_lock);
948
949 rc = seq_open(file, &jbd2_seq_info_ops);
950 if (rc == 0) {
951 struct seq_file *m = file->private_data;
952 m->private = s;
953 } else {
954 kfree(s->stats);
955 kfree(s);
956 }
957 return rc;
958
959}
960
961static int jbd2_seq_info_release(struct inode *inode, struct file *file)
962{
963 struct seq_file *seq = file->private_data;
964 struct jbd2_stats_proc_session *s = seq->private;
965 kfree(s->stats);
966 kfree(s);
967 return seq_release(inode, file);
968}
969
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700970static const struct file_operations jbd2_seq_info_fops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500971 .owner = THIS_MODULE,
972 .open = jbd2_seq_info_open,
973 .read = seq_read,
974 .llseek = seq_lseek,
975 .release = jbd2_seq_info_release,
976};
977
978static struct proc_dir_entry *proc_jbd2_stats;
979
980static void jbd2_stats_proc_init(journal_t *journal)
981{
Theodore Ts'o05496762008-09-16 14:36:17 -0400982 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500983 if (journal->j_proc_entry) {
Denis V. Lunev79da3662008-04-29 01:02:11 -0700984 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
985 &jbd2_seq_info_fops, journal);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500986 }
987}
988
989static void jbd2_stats_proc_exit(journal_t *journal)
990{
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500991 remove_proc_entry("info", journal->j_proc_entry);
Theodore Ts'o05496762008-09-16 14:36:17 -0400992 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500993}
994
Dave Kleikamp470decc2006-10-11 01:20:57 -0700995/*
996 * Management for journal control blocks: functions to create and
997 * destroy journal_t structures, and to initialise and read existing
998 * journal blocks from disk. */
999
1000/* First: create and setup a journal_t object in memory. We initialise
1001 * very few fields yet: that has to wait until we have created the
1002 * journal structures from from scratch, or loaded them from disk. */
1003
1004static journal_t * journal_init_common (void)
1005{
1006 journal_t *journal;
1007 int err;
1008
Andrew Morton3ebfdf82009-12-23 08:05:15 -05001009 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001010 if (!journal)
Theodore Ts'ob7271b02010-12-18 13:39:38 -05001011 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001012
1013 init_waitqueue_head(&journal->j_wait_transaction_locked);
1014 init_waitqueue_head(&journal->j_wait_logspace);
1015 init_waitqueue_head(&journal->j_wait_done_commit);
1016 init_waitqueue_head(&journal->j_wait_checkpoint);
1017 init_waitqueue_head(&journal->j_wait_commit);
1018 init_waitqueue_head(&journal->j_wait_updates);
1019 mutex_init(&journal->j_barrier);
1020 mutex_init(&journal->j_checkpoint_mutex);
1021 spin_lock_init(&journal->j_revoke_lock);
1022 spin_lock_init(&journal->j_list_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001023 rwlock_init(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001024
Mingming Caocd02ff02007-10-16 18:38:25 -04001025 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
Theodore Ts'o30773842009-01-03 20:27:38 -05001026 journal->j_min_batch_time = 0;
1027 journal->j_max_batch_time = 15000; /* 15ms */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001028
1029 /* The journal is marked for error until we succeed with recovery! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001030 journal->j_flags = JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001031
1032 /* Set up a default-sized revoke table for the new mount. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001033 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001034 if (err) {
1035 kfree(journal);
Theodore Ts'ob7271b02010-12-18 13:39:38 -05001036 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001037 }
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001038
Theodore Ts'obf699322009-09-30 00:32:06 -04001039 spin_lock_init(&journal->j_history_lock);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001040
Dave Kleikamp470decc2006-10-11 01:20:57 -07001041 return journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001042}
1043
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001044/* jbd2_journal_init_dev and jbd2_journal_init_inode:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001045 *
1046 * Create a journal structure assigned some fixed set of disk blocks to
1047 * the journal. We don't actually touch those disk blocks yet, but we
1048 * need to set up all of the mapping information to tell the journaling
1049 * system where the journal blocks are.
1050 *
1051 */
1052
1053/**
Randy Dunlap5648ba52008-04-17 10:38:59 -04001054 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
Dave Kleikamp470decc2006-10-11 01:20:57 -07001055 * @bdev: Block device on which to create the journal
1056 * @fs_dev: Device which hold journalled filesystem for this journal.
1057 * @start: Block nr Start of journal.
1058 * @len: Length of the journal in blocks.
1059 * @blocksize: blocksize of journalling device
Randy Dunlap5648ba52008-04-17 10:38:59 -04001060 *
1061 * Returns: a newly created journal_t *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001062 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001063 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
Dave Kleikamp470decc2006-10-11 01:20:57 -07001064 * range of blocks on an arbitrary block device.
1065 *
1066 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001067journal_t * jbd2_journal_init_dev(struct block_device *bdev,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001068 struct block_device *fs_dev,
Mingming Cao18eba7a2006-10-11 01:21:13 -07001069 unsigned long long start, int len, int blocksize)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001070{
1071 journal_t *journal = journal_init_common();
1072 struct buffer_head *bh;
Theodore Ts'o05496762008-09-16 14:36:17 -04001073 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001074 int n;
1075
1076 if (!journal)
1077 return NULL;
1078
1079 /* journal descriptor can store up to n blocks -bzzz */
1080 journal->j_blocksize = blocksize;
yangsheng0587aa32010-11-17 21:46:26 -05001081 journal->j_dev = bdev;
1082 journal->j_fs_dev = fs_dev;
1083 journal->j_blk_offset = start;
1084 journal->j_maxlen = len;
1085 bdevname(journal->j_dev, journal->j_devname);
1086 p = journal->j_devname;
1087 while ((p = strchr(p, '/')))
1088 *p = '!';
Jan Kara4b905672009-01-06 14:53:35 -05001089 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001090 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1091 journal->j_wbufsize = n;
1092 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1093 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001094 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001095 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001096 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001097 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001098
1099 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001100 if (!bh) {
1101 printk(KERN_ERR
1102 "%s: Cannot get buffer for journal superblock\n",
1103 __func__);
1104 goto out_err;
1105 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001106 journal->j_sb_buffer = bh;
1107 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Kara4b905672009-01-06 14:53:35 -05001108
Dave Kleikamp470decc2006-10-11 01:20:57 -07001109 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001110out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001111 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001112 jbd2_stats_proc_exit(journal);
1113 kfree(journal);
1114 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001115}
1116
1117/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001118 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001119 * @inode: An inode to create the journal in
1120 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001121 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
Dave Kleikamp470decc2006-10-11 01:20:57 -07001122 * the journal. The inode must exist already, must support bmap() and
1123 * must have all data blocks preallocated.
1124 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001125journal_t * jbd2_journal_init_inode (struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001126{
1127 struct buffer_head *bh;
1128 journal_t *journal = journal_init_common();
Theodore Ts'o05496762008-09-16 14:36:17 -04001129 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001130 int err;
1131 int n;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001132 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001133
1134 if (!journal)
1135 return NULL;
1136
1137 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1138 journal->j_inode = inode;
Theodore Ts'o05496762008-09-16 14:36:17 -04001139 bdevname(journal->j_dev, journal->j_devname);
1140 p = journal->j_devname;
1141 while ((p = strchr(p, '/')))
1142 *p = '!';
1143 p = journal->j_devname + strlen(journal->j_devname);
Theodore Ts'o90576c02009-09-29 15:51:30 -04001144 sprintf(p, "-%lu", journal->j_inode->i_ino);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001145 jbd_debug(1,
1146 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1147 journal, inode->i_sb->s_id, inode->i_ino,
1148 (long long) inode->i_size,
1149 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1150
1151 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1152 journal->j_blocksize = inode->i_sb->s_blocksize;
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001153 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001154
1155 /* journal descriptor can store up to n blocks -bzzz */
1156 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1157 journal->j_wbufsize = n;
1158 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1159 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001160 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001161 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001162 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001163 }
1164
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001165 err = jbd2_journal_bmap(journal, 0, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001166 /* If that failed, give up */
1167 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -08001168 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001169 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001170 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001171 }
1172
1173 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001174 if (!bh) {
1175 printk(KERN_ERR
1176 "%s: Cannot get buffer for journal superblock\n",
1177 __func__);
1178 goto out_err;
1179 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001180 journal->j_sb_buffer = bh;
1181 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1182
1183 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001184out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001185 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001186 jbd2_stats_proc_exit(journal);
1187 kfree(journal);
1188 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001189}
1190
1191/*
1192 * If the journal init or create aborts, we need to mark the journal
1193 * superblock as being NULL to prevent the journal destroy from writing
1194 * back a bogus superblock.
1195 */
1196static void journal_fail_superblock (journal_t *journal)
1197{
1198 struct buffer_head *bh = journal->j_sb_buffer;
1199 brelse(bh);
1200 journal->j_sb_buffer = NULL;
1201}
1202
1203/*
1204 * Given a journal_t structure, initialise the various fields for
1205 * startup of a new journaling session. We use this both when creating
1206 * a journal, and after recovering an old journal to reset it for
1207 * subsequent use.
1208 */
1209
1210static int journal_reset(journal_t *journal)
1211{
1212 journal_superblock_t *sb = journal->j_superblock;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001213 unsigned long long first, last;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001214
1215 first = be32_to_cpu(sb->s_first);
1216 last = be32_to_cpu(sb->s_maxlen);
Jan Karaf6f50e22009-07-17 10:40:01 -04001217 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001218 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
Jan Karaf6f50e22009-07-17 10:40:01 -04001219 first, last);
1220 journal_fail_superblock(journal);
1221 return -EINVAL;
1222 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001223
1224 journal->j_first = first;
1225 journal->j_last = last;
1226
1227 journal->j_head = first;
1228 journal->j_tail = first;
1229 journal->j_free = last - first;
1230
1231 journal->j_tail_sequence = journal->j_transaction_sequence;
1232 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1233 journal->j_commit_request = journal->j_commit_sequence;
1234
1235 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1236
Dave Kleikamp470decc2006-10-11 01:20:57 -07001237 /*
1238 * As a special case, if the on-disk copy is already marked as needing
Jan Kara24bcc892012-03-13 15:41:04 -04001239 * no recovery (s_start == 0), then we can safely defer the superblock
1240 * update until the next commit by setting JBD2_FLUSHED. This avoids
Dave Kleikamp470decc2006-10-11 01:20:57 -07001241 * attempting a write to a potential-readonly device.
1242 */
Jan Kara24bcc892012-03-13 15:41:04 -04001243 if (sb->s_start == 0) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001244 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
Dave Kleikamp470decc2006-10-11 01:20:57 -07001245 "(start %ld, seq %d, errno %d)\n",
1246 journal->j_tail, journal->j_tail_sequence,
1247 journal->j_errno);
Jan Kara24bcc892012-03-13 15:41:04 -04001248 journal->j_flags |= JBD2_FLUSHED;
1249 } else {
Jan Karaa78bb112012-03-13 15:43:04 -04001250 /* Lock here to make assertions happy... */
1251 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara79feb522012-03-13 22:22:54 -04001252 /*
1253 * Update log tail information. We use WRITE_FUA since new
1254 * transaction will start reusing journal space and so we
1255 * must make sure information about current log tail is on
1256 * disk before that.
1257 */
1258 jbd2_journal_update_sb_log_tail(journal,
1259 journal->j_tail_sequence,
1260 journal->j_tail,
1261 WRITE_FUA);
Jan Karaa78bb112012-03-13 15:43:04 -04001262 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001263 }
Jan Kara24bcc892012-03-13 15:41:04 -04001264 return jbd2_journal_start_thread(journal);
1265}
Dave Kleikamp470decc2006-10-11 01:20:57 -07001266
Jan Kara79feb522012-03-13 22:22:54 -04001267static void jbd2_write_superblock(journal_t *journal, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001268{
1269 struct buffer_head *bh = journal->j_sb_buffer;
Jan Kara79feb522012-03-13 22:22:54 -04001270 int ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001271
Jan Kara79feb522012-03-13 22:22:54 -04001272 trace_jbd2_write_superblock(journal, write_op);
1273 if (!(journal->j_flags & JBD2_BARRIER))
1274 write_op &= ~(REQ_FUA | REQ_FLUSH);
1275 lock_buffer(bh);
Theodore Ts'o914258b2008-10-06 21:35:40 -04001276 if (buffer_write_io_error(bh)) {
1277 /*
1278 * Oh, dear. A previous attempt to write the journal
1279 * superblock failed. This could happen because the
1280 * USB device was yanked out. Or it could happen to
1281 * be a transient write error and maybe the block will
1282 * be remapped. Nothing we can do but to retry the
1283 * write and hope for the best.
1284 */
1285 printk(KERN_ERR "JBD2: previous I/O error detected "
1286 "for journal superblock update for %s.\n",
1287 journal->j_devname);
1288 clear_buffer_write_io_error(bh);
1289 set_buffer_uptodate(bh);
1290 }
Jan Kara79feb522012-03-13 22:22:54 -04001291 get_bh(bh);
1292 bh->b_end_io = end_buffer_write_sync;
1293 ret = submit_bh(write_op, bh);
1294 wait_on_buffer(bh);
Jan Kara24bcc892012-03-13 15:41:04 -04001295 if (buffer_write_io_error(bh)) {
Jan Kara24bcc892012-03-13 15:41:04 -04001296 clear_buffer_write_io_error(bh);
1297 set_buffer_uptodate(bh);
Jan Kara79feb522012-03-13 22:22:54 -04001298 ret = -EIO;
1299 }
1300 if (ret) {
1301 printk(KERN_ERR "JBD2: Error %d detected when updating "
1302 "journal superblock for %s.\n", ret,
1303 journal->j_devname);
Jan Kara24bcc892012-03-13 15:41:04 -04001304 }
1305}
Theodore Ts'o914258b2008-10-06 21:35:40 -04001306
Jan Kara24bcc892012-03-13 15:41:04 -04001307/**
1308 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1309 * @journal: The journal to update.
Jan Kara79feb522012-03-13 22:22:54 -04001310 * @tail_tid: TID of the new transaction at the tail of the log
1311 * @tail_block: The first block of the transaction at the tail of the log
1312 * @write_op: With which operation should we write the journal sb
Jan Kara24bcc892012-03-13 15:41:04 -04001313 *
1314 * Update a journal's superblock information about log tail and write it to
1315 * disk, waiting for the IO to complete.
1316 */
Jan Kara79feb522012-03-13 22:22:54 -04001317void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1318 unsigned long tail_block, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001319{
1320 journal_superblock_t *sb = journal->j_superblock;
1321
Jan Karaa78bb112012-03-13 15:43:04 -04001322 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Jan Kara79feb522012-03-13 22:22:54 -04001323 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1324 tail_block, tail_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001325
Jan Kara79feb522012-03-13 22:22:54 -04001326 sb->s_sequence = cpu_to_be32(tail_tid);
1327 sb->s_start = cpu_to_be32(tail_block);
Jan Kara24bcc892012-03-13 15:41:04 -04001328
Jan Kara79feb522012-03-13 22:22:54 -04001329 jbd2_write_superblock(journal, write_op);
Jan Kara24bcc892012-03-13 15:41:04 -04001330
1331 /* Log is no longer empty */
1332 write_lock(&journal->j_state_lock);
1333 WARN_ON(!sb->s_sequence);
1334 journal->j_flags &= ~JBD2_FLUSHED;
1335 write_unlock(&journal->j_state_lock);
1336}
1337
1338/**
1339 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1340 * @journal: The journal to update.
1341 *
1342 * Update a journal's dynamic superblock fields to show that journal is empty.
1343 * Write updated superblock to disk waiting for IO to complete.
1344 */
1345static void jbd2_mark_journal_empty(journal_t *journal)
1346{
1347 journal_superblock_t *sb = journal->j_superblock;
1348
Jan Karaa78bb112012-03-13 15:43:04 -04001349 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001350 read_lock(&journal->j_state_lock);
Eric Sandeen0699c6d2012-08-18 22:29:40 -04001351 /* Is it already empty? */
1352 if (sb->s_start == 0) {
1353 read_unlock(&journal->j_state_lock);
1354 return;
1355 }
Jan Kara24bcc892012-03-13 15:41:04 -04001356 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1357 journal->j_tail_sequence);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001358
1359 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
Jan Kara24bcc892012-03-13 15:41:04 -04001360 sb->s_start = cpu_to_be32(0);
1361 read_unlock(&journal->j_state_lock);
1362
Jan Kara79feb522012-03-13 22:22:54 -04001363 jbd2_write_superblock(journal, WRITE_FUA);
Jan Kara24bcc892012-03-13 15:41:04 -04001364
1365 /* Log is no longer empty */
1366 write_lock(&journal->j_state_lock);
1367 journal->j_flags |= JBD2_FLUSHED;
1368 write_unlock(&journal->j_state_lock);
1369}
1370
1371
1372/**
1373 * jbd2_journal_update_sb_errno() - Update error in the journal.
1374 * @journal: The journal to update.
1375 *
1376 * Update a journal's errno. Write updated superblock to disk waiting for IO
1377 * to complete.
1378 */
Theodore Ts'o8b9f3862012-08-05 19:04:57 -04001379void jbd2_journal_update_sb_errno(journal_t *journal)
Jan Kara24bcc892012-03-13 15:41:04 -04001380{
1381 journal_superblock_t *sb = journal->j_superblock;
1382
1383 read_lock(&journal->j_state_lock);
1384 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1385 journal->j_errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001386 sb->s_errno = cpu_to_be32(journal->j_errno);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001387 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001388
Jan Kara79feb522012-03-13 22:22:54 -04001389 jbd2_write_superblock(journal, WRITE_SYNC);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001390}
Theodore Ts'o8b9f3862012-08-05 19:04:57 -04001391EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001392
1393/*
1394 * Read the superblock for a given journal, performing initial
1395 * validation of the format.
1396 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001397static int journal_get_superblock(journal_t *journal)
1398{
1399 struct buffer_head *bh;
1400 journal_superblock_t *sb;
1401 int err = -EIO;
1402
1403 bh = journal->j_sb_buffer;
1404
1405 J_ASSERT(bh != NULL);
1406 if (!buffer_uptodate(bh)) {
1407 ll_rw_block(READ, 1, &bh);
1408 wait_on_buffer(bh);
1409 if (!buffer_uptodate(bh)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001410 printk(KERN_ERR
1411 "JBD2: IO error reading journal superblock\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001412 goto out;
1413 }
1414 }
1415
1416 sb = journal->j_superblock;
1417
1418 err = -EINVAL;
1419
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001420 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001421 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001422 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001423 goto out;
1424 }
1425
1426 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001427 case JBD2_SUPERBLOCK_V1:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001428 journal->j_format_version = 1;
1429 break;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001430 case JBD2_SUPERBLOCK_V2:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001431 journal->j_format_version = 2;
1432 break;
1433 default:
Eryu Guanf2a44522011-11-01 19:09:18 -04001434 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001435 goto out;
1436 }
1437
1438 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1439 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1440 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001441 printk(KERN_WARNING "JBD2: journal file too short\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001442 goto out;
1443 }
1444
Eryu Guan87622022011-11-01 19:04:59 -04001445 if (be32_to_cpu(sb->s_first) == 0 ||
1446 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1447 printk(KERN_WARNING
1448 "JBD2: Invalid start block of journal: %u\n",
1449 be32_to_cpu(sb->s_first));
1450 goto out;
1451 }
1452
Dave Kleikamp470decc2006-10-11 01:20:57 -07001453 return 0;
1454
1455out:
1456 journal_fail_superblock(journal);
1457 return err;
1458}
1459
1460/*
1461 * Load the on-disk journal superblock and read the key fields into the
1462 * journal_t.
1463 */
1464
1465static int load_superblock(journal_t *journal)
1466{
1467 int err;
1468 journal_superblock_t *sb;
1469
1470 err = journal_get_superblock(journal);
1471 if (err)
1472 return err;
1473
1474 sb = journal->j_superblock;
1475
1476 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1477 journal->j_tail = be32_to_cpu(sb->s_start);
1478 journal->j_first = be32_to_cpu(sb->s_first);
1479 journal->j_last = be32_to_cpu(sb->s_maxlen);
1480 journal->j_errno = be32_to_cpu(sb->s_errno);
1481
1482 return 0;
1483}
1484
1485
1486/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001487 * int jbd2_journal_load() - Read journal from disk.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001488 * @journal: Journal to act on.
1489 *
1490 * Given a journal_t structure which tells us which disk blocks contain
1491 * a journal, read the journal from disk to initialise the in-memory
1492 * structures.
1493 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001494int jbd2_journal_load(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001495{
1496 int err;
1497 journal_superblock_t *sb;
1498
1499 err = load_superblock(journal);
1500 if (err)
1501 return err;
1502
1503 sb = journal->j_superblock;
1504 /* If this is a V2 superblock, then we have to check the
1505 * features flags on it. */
1506
1507 if (journal->j_format_version >= 2) {
1508 if ((sb->s_feature_ro_compat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001509 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001510 (sb->s_feature_incompat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001511 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001512 printk(KERN_WARNING
1513 "JBD2: Unrecognised features on journal\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001514 return -EINVAL;
1515 }
1516 }
1517
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001518 /*
1519 * Create a slab for this blocksize
1520 */
1521 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1522 if (err)
1523 return err;
1524
Dave Kleikamp470decc2006-10-11 01:20:57 -07001525 /* Let the recovery code check whether it needs to recover any
1526 * data from the journal. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001527 if (jbd2_journal_recover(journal))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001528 goto recovery_error;
1529
Theodore Ts'oe6a47422009-11-15 15:31:37 -05001530 if (journal->j_failed_commit) {
1531 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1532 "is corrupt.\n", journal->j_failed_commit,
1533 journal->j_devname);
1534 return -EIO;
1535 }
1536
Dave Kleikamp470decc2006-10-11 01:20:57 -07001537 /* OK, we've finished with the dynamic journal bits:
1538 * reinitialise the dynamic contents of the superblock in memory
1539 * and reset them on disk. */
1540 if (journal_reset(journal))
1541 goto recovery_error;
1542
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001543 journal->j_flags &= ~JBD2_ABORT;
1544 journal->j_flags |= JBD2_LOADED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001545 return 0;
1546
1547recovery_error:
Eryu Guanf2a44522011-11-01 19:09:18 -04001548 printk(KERN_WARNING "JBD2: recovery failed\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001549 return -EIO;
1550}
1551
1552/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001553 * void jbd2_journal_destroy() - Release a journal_t structure.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001554 * @journal: Journal to act on.
1555 *
1556 * Release a journal_t structure once it is no longer in use by the
1557 * journaled object.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001558 * Return <0 if we couldn't clean up the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001559 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001560int jbd2_journal_destroy(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001561{
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001562 int err = 0;
1563
Dave Kleikamp470decc2006-10-11 01:20:57 -07001564 /* Wait for the commit thread to wake up and die. */
1565 journal_kill_thread(journal);
1566
1567 /* Force a final log commit */
1568 if (journal->j_running_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001569 jbd2_journal_commit_transaction(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001570
1571 /* Force any old transactions to disk */
1572
1573 /* Totally anal locking here... */
1574 spin_lock(&journal->j_list_lock);
1575 while (journal->j_checkpoint_transactions != NULL) {
1576 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001577 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001578 jbd2_log_do_checkpoint(journal);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001579 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001580 spin_lock(&journal->j_list_lock);
1581 }
1582
1583 J_ASSERT(journal->j_running_transaction == NULL);
1584 J_ASSERT(journal->j_committing_transaction == NULL);
1585 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1586 spin_unlock(&journal->j_list_lock);
1587
Dave Kleikamp470decc2006-10-11 01:20:57 -07001588 if (journal->j_sb_buffer) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001589 if (!is_journal_aborted(journal)) {
Jan Karaa78bb112012-03-13 15:43:04 -04001590 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara24bcc892012-03-13 15:41:04 -04001591 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04001592 mutex_unlock(&journal->j_checkpoint_mutex);
1593 } else
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001594 err = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001595 brelse(journal->j_sb_buffer);
1596 }
1597
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001598 if (journal->j_proc_entry)
1599 jbd2_stats_proc_exit(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001600 if (journal->j_inode)
1601 iput(journal->j_inode);
1602 if (journal->j_revoke)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001603 jbd2_journal_destroy_revoke(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001604 kfree(journal->j_wbuf);
1605 kfree(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001606
1607 return err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001608}
1609
1610
1611/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001612 *int jbd2_journal_check_used_features () - Check if features specified are used.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001613 * @journal: Journal to check.
1614 * @compat: bitmask of compatible features
1615 * @ro: bitmask of features that force read-only mount
1616 * @incompat: bitmask of incompatible features
1617 *
1618 * Check whether the journal uses all of a given set of
1619 * features. Return true (non-zero) if it does.
1620 **/
1621
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001622int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001623 unsigned long ro, unsigned long incompat)
1624{
1625 journal_superblock_t *sb;
1626
1627 if (!compat && !ro && !incompat)
1628 return 1;
Patrick J. LoPresti1113e1b2010-07-22 15:04:16 -07001629 /* Load journal superblock if it is not loaded yet. */
1630 if (journal->j_format_version == 0 &&
1631 journal_get_superblock(journal) != 0)
1632 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001633 if (journal->j_format_version == 1)
1634 return 0;
1635
1636 sb = journal->j_superblock;
1637
1638 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1639 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1640 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1641 return 1;
1642
1643 return 0;
1644}
1645
1646/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001647 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001648 * @journal: Journal to check.
1649 * @compat: bitmask of compatible features
1650 * @ro: bitmask of features that force read-only mount
1651 * @incompat: bitmask of incompatible features
1652 *
1653 * Check whether the journaling code supports the use of
1654 * all of a given set of features on this journal. Return true
1655 * (non-zero) if it can. */
1656
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001657int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001658 unsigned long ro, unsigned long incompat)
1659{
Dave Kleikamp470decc2006-10-11 01:20:57 -07001660 if (!compat && !ro && !incompat)
1661 return 1;
1662
Dave Kleikamp470decc2006-10-11 01:20:57 -07001663 /* We can support any known requested features iff the
1664 * superblock is in version 2. Otherwise we fail to support any
1665 * extended sb features. */
1666
1667 if (journal->j_format_version != 2)
1668 return 0;
1669
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001670 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1671 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1672 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001673 return 1;
1674
1675 return 0;
1676}
1677
1678/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001679 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
Dave Kleikamp470decc2006-10-11 01:20:57 -07001680 * @journal: Journal to act on.
1681 * @compat: bitmask of compatible features
1682 * @ro: bitmask of features that force read-only mount
1683 * @incompat: bitmask of incompatible features
1684 *
1685 * Mark a given journal feature as present on the
1686 * superblock. Returns true if the requested features could be set.
1687 *
1688 */
1689
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001690int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001691 unsigned long ro, unsigned long incompat)
1692{
1693 journal_superblock_t *sb;
1694
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001695 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001696 return 1;
1697
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001698 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001699 return 0;
1700
1701 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1702 compat, ro, incompat);
1703
1704 sb = journal->j_superblock;
1705
1706 sb->s_feature_compat |= cpu_to_be32(compat);
1707 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1708 sb->s_feature_incompat |= cpu_to_be32(incompat);
1709
1710 return 1;
1711}
1712
Girish Shilamkar818d2762008-01-28 23:58:27 -05001713/*
1714 * jbd2_journal_clear_features () - Clear a given journal feature in the
1715 * superblock
1716 * @journal: Journal to act on.
1717 * @compat: bitmask of compatible features
1718 * @ro: bitmask of features that force read-only mount
1719 * @incompat: bitmask of incompatible features
1720 *
1721 * Clear a given journal feature as present on the
1722 * superblock.
1723 */
1724void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1725 unsigned long ro, unsigned long incompat)
1726{
1727 journal_superblock_t *sb;
1728
1729 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1730 compat, ro, incompat);
1731
1732 sb = journal->j_superblock;
1733
1734 sb->s_feature_compat &= ~cpu_to_be32(compat);
1735 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1736 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1737}
1738EXPORT_SYMBOL(jbd2_journal_clear_features);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001739
1740/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001741 * int jbd2_journal_flush () - Flush journal
Dave Kleikamp470decc2006-10-11 01:20:57 -07001742 * @journal: Journal to act on.
1743 *
1744 * Flush all data for a given journal to disk and empty the journal.
1745 * Filesystems can use this when remounting readonly to ensure that
1746 * recovery does not need to happen on remount.
1747 */
1748
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001749int jbd2_journal_flush(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001750{
1751 int err = 0;
1752 transaction_t *transaction = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001753
Theodore Ts'oa931da62010-08-03 21:35:12 -04001754 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001755
1756 /* Force everything buffered to the log... */
1757 if (journal->j_running_transaction) {
1758 transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001759 __jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001760 } else if (journal->j_committing_transaction)
1761 transaction = journal->j_committing_transaction;
1762
1763 /* Wait for the log commit to complete... */
1764 if (transaction) {
1765 tid_t tid = transaction->t_tid;
1766
Theodore Ts'oa931da62010-08-03 21:35:12 -04001767 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001768 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001769 } else {
Theodore Ts'oa931da62010-08-03 21:35:12 -04001770 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001771 }
1772
1773 /* ...and flush everything in the log out to disk. */
1774 spin_lock(&journal->j_list_lock);
1775 while (!err && journal->j_checkpoint_transactions != NULL) {
1776 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001777 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001778 err = jbd2_log_do_checkpoint(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001779 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001780 spin_lock(&journal->j_list_lock);
1781 }
1782 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001783
1784 if (is_journal_aborted(journal))
1785 return -EIO;
1786
Jan Karaa78bb112012-03-13 15:43:04 -04001787 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001788 jbd2_cleanup_journal_tail(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001789
1790 /* Finally, mark the journal as really needing no recovery.
1791 * This sets s_start==0 in the underlying superblock, which is
1792 * the magic code for a fully-recovered superblock. Any future
1793 * commits of data to the journal will restore the current
1794 * s_start value. */
Jan Kara24bcc892012-03-13 15:41:04 -04001795 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04001796 mutex_unlock(&journal->j_checkpoint_mutex);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001797 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001798 J_ASSERT(!journal->j_running_transaction);
1799 J_ASSERT(!journal->j_committing_transaction);
1800 J_ASSERT(!journal->j_checkpoint_transactions);
1801 J_ASSERT(journal->j_head == journal->j_tail);
1802 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001803 write_unlock(&journal->j_state_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001804 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001805}
1806
1807/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001808 * int jbd2_journal_wipe() - Wipe journal contents
Dave Kleikamp470decc2006-10-11 01:20:57 -07001809 * @journal: Journal to act on.
1810 * @write: flag (see below)
1811 *
1812 * Wipe out all of the contents of a journal, safely. This will produce
1813 * a warning if the journal contains any valid recovery information.
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001814 * Must be called between journal_init_*() and jbd2_journal_load().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001815 *
1816 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1817 * we merely suppress recovery.
1818 */
1819
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001820int jbd2_journal_wipe(journal_t *journal, int write)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001821{
Dave Kleikamp470decc2006-10-11 01:20:57 -07001822 int err = 0;
1823
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001824 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001825
1826 err = load_superblock(journal);
1827 if (err)
1828 return err;
1829
Dave Kleikamp470decc2006-10-11 01:20:57 -07001830 if (!journal->j_tail)
1831 goto no_recovery;
1832
Eryu Guanf2a44522011-11-01 19:09:18 -04001833 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -07001834 write ? "Clearing" : "Ignoring");
1835
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001836 err = jbd2_journal_skip_recovery(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04001837 if (write) {
1838 /* Lock to make assertions happy... */
1839 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara24bcc892012-03-13 15:41:04 -04001840 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04001841 mutex_unlock(&journal->j_checkpoint_mutex);
1842 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001843
1844 no_recovery:
1845 return err;
1846}
1847
1848/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07001849 * Journal abort has very specific semantics, which we describe
1850 * for journal abort.
1851 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001852 * Two internal functions, which provide abort to the jbd layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001853 * itself are here.
1854 */
1855
1856/*
1857 * Quick version for internal journal use (doesn't lock the journal).
1858 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1859 * and don't attempt to make any other journal updates.
1860 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001861void __jbd2_journal_abort_hard(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001862{
1863 transaction_t *transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001864
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001865 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001866 return;
1867
1868 printk(KERN_ERR "Aborting journal on device %s.\n",
Theodore Ts'o05496762008-09-16 14:36:17 -04001869 journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001870
Theodore Ts'oa931da62010-08-03 21:35:12 -04001871 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001872 journal->j_flags |= JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001873 transaction = journal->j_running_transaction;
1874 if (transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001875 __jbd2_log_start_commit(journal, transaction->t_tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001876 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001877}
1878
1879/* Soft abort: record the abort error status in the journal superblock,
1880 * but don't do any other IO. */
1881static void __journal_abort_soft (journal_t *journal, int errno)
1882{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001883 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001884 return;
1885
1886 if (!journal->j_errno)
1887 journal->j_errno = errno;
1888
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001889 __jbd2_journal_abort_hard(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001890
1891 if (errno)
Jan Kara24bcc892012-03-13 15:41:04 -04001892 jbd2_journal_update_sb_errno(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001893}
1894
1895/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001896 * void jbd2_journal_abort () - Shutdown the journal immediately.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001897 * @journal: the journal to shutdown.
1898 * @errno: an error number to record in the journal indicating
1899 * the reason for the shutdown.
1900 *
1901 * Perform a complete, immediate shutdown of the ENTIRE
1902 * journal (not of a single transaction). This operation cannot be
1903 * undone without closing and reopening the journal.
1904 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001905 * The jbd2_journal_abort function is intended to support higher level error
Dave Kleikamp470decc2006-10-11 01:20:57 -07001906 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1907 * mode.
1908 *
1909 * Journal abort has very specific semantics. Any existing dirty,
1910 * unjournaled buffers in the main filesystem will still be written to
1911 * disk by bdflush, but the journaling mechanism will be suspended
1912 * immediately and no further transaction commits will be honoured.
1913 *
1914 * Any dirty, journaled buffers will be written back to disk without
1915 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1916 * filesystem, but we _do_ attempt to leave as much data as possible
1917 * behind for fsck to use for cleanup.
1918 *
1919 * Any attempt to get a new transaction handle on a journal which is in
1920 * ABORT state will just result in an -EROFS error return. A
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001921 * jbd2_journal_stop on an existing handle will return -EIO if we have
Dave Kleikamp470decc2006-10-11 01:20:57 -07001922 * entered abort state during the update.
1923 *
1924 * Recursive transactions are not disturbed by journal abort until the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001925 * final jbd2_journal_stop, which will receive the -EIO error.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001926 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001927 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
Dave Kleikamp470decc2006-10-11 01:20:57 -07001928 * which will be recorded (if possible) in the journal superblock. This
1929 * allows a client to record failure conditions in the middle of a
1930 * transaction without having to complete the transaction to record the
1931 * failure to disk. ext3_error, for example, now uses this
1932 * functionality.
1933 *
1934 * Errors which originate from within the journaling layer will NOT
1935 * supply an errno; a null errno implies that absolutely no further
1936 * writes are done to the journal (unless there are any already in
1937 * progress).
1938 *
1939 */
1940
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001941void jbd2_journal_abort(journal_t *journal, int errno)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001942{
1943 __journal_abort_soft(journal, errno);
1944}
1945
1946/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001947 * int jbd2_journal_errno () - returns the journal's error state.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001948 * @journal: journal to examine.
1949 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001950 * This is the errno number set with jbd2_journal_abort(), the last
Dave Kleikamp470decc2006-10-11 01:20:57 -07001951 * time the journal was mounted - if the journal was stopped
1952 * without calling abort this will be 0.
1953 *
1954 * If the journal has been aborted on this mount time -EROFS will
1955 * be returned.
1956 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001957int jbd2_journal_errno(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001958{
1959 int err;
1960
Theodore Ts'oa931da62010-08-03 21:35:12 -04001961 read_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001962 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001963 err = -EROFS;
1964 else
1965 err = journal->j_errno;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001966 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001967 return err;
1968}
1969
1970/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001971 * int jbd2_journal_clear_err () - clears the journal's error state
Dave Kleikamp470decc2006-10-11 01:20:57 -07001972 * @journal: journal to act on.
1973 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001974 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07001975 * mode.
1976 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001977int jbd2_journal_clear_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001978{
1979 int err = 0;
1980
Theodore Ts'oa931da62010-08-03 21:35:12 -04001981 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001982 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001983 err = -EROFS;
1984 else
1985 journal->j_errno = 0;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001986 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001987 return err;
1988}
1989
1990/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001991 * void jbd2_journal_ack_err() - Ack journal err.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001992 * @journal: journal to act on.
1993 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001994 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07001995 * mode.
1996 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001997void jbd2_journal_ack_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001998{
Theodore Ts'oa931da62010-08-03 21:35:12 -04001999 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002000 if (journal->j_errno)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002001 journal->j_flags |= JBD2_ACK_ERR;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002002 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002003}
2004
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002005int jbd2_journal_blocks_per_page(struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002006{
2007 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2008}
2009
2010/*
Zach Brownb517bea2006-10-11 01:21:08 -07002011 * helper functions to deal with 32 or 64bit block numbers.
2012 */
2013size_t journal_tag_bytes(journal_t *journal)
2014{
2015 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
Mingming Caocd02ff02007-10-16 18:38:25 -04002016 return JBD2_TAG_SIZE64;
Zach Brownb517bea2006-10-11 01:21:08 -07002017 else
Mingming Caocd02ff02007-10-16 18:38:25 -04002018 return JBD2_TAG_SIZE32;
Zach Brownb517bea2006-10-11 01:21:08 -07002019}
2020
2021/*
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002022 * JBD memory management
2023 *
2024 * These functions are used to allocate block-sized chunks of memory
2025 * used for making copies of buffer_head data. Very often it will be
2026 * page-sized chunks of data, but sometimes it will be in
2027 * sub-page-size chunks. (For example, 16k pages on Power systems
2028 * with a 4k block file system.) For blocks smaller than a page, we
2029 * use a SLAB allocator. There are slab caches for each block size,
2030 * which are allocated at mount time, if necessary, and we only free
2031 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2032 * this reason we don't need to a mutex to protect access to
2033 * jbd2_slab[] allocating or releasing memory; only in
2034 * jbd2_journal_create_slab().
2035 */
2036#define JBD2_MAX_SLABS 8
2037static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002038
2039static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2040 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2041 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2042};
2043
2044
2045static void jbd2_journal_destroy_slabs(void)
2046{
2047 int i;
2048
2049 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2050 if (jbd2_slab[i])
2051 kmem_cache_destroy(jbd2_slab[i]);
2052 jbd2_slab[i] = NULL;
2053 }
2054}
2055
2056static int jbd2_journal_create_slab(size_t size)
2057{
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002058 static DEFINE_MUTEX(jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002059 int i = order_base_2(size) - 10;
2060 size_t slab_size;
2061
2062 if (size == PAGE_SIZE)
2063 return 0;
2064
2065 if (i >= JBD2_MAX_SLABS)
2066 return -EINVAL;
2067
2068 if (unlikely(i < 0))
2069 i = 0;
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002070 mutex_lock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002071 if (jbd2_slab[i]) {
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002072 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002073 return 0; /* Already created */
2074 }
2075
2076 slab_size = 1 << (i+10);
2077 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2078 slab_size, 0, NULL);
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002079 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002080 if (!jbd2_slab[i]) {
2081 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2082 return -ENOMEM;
2083 }
2084 return 0;
2085}
2086
2087static struct kmem_cache *get_slab(size_t size)
2088{
2089 int i = order_base_2(size) - 10;
2090
2091 BUG_ON(i >= JBD2_MAX_SLABS);
2092 if (unlikely(i < 0))
2093 i = 0;
Bill Pemberton8ac97b72010-04-30 09:34:31 -04002094 BUG_ON(jbd2_slab[i] == NULL);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002095 return jbd2_slab[i];
2096}
2097
2098void *jbd2_alloc(size_t size, gfp_t flags)
2099{
2100 void *ptr;
2101
2102 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2103
2104 flags |= __GFP_REPEAT;
2105 if (size == PAGE_SIZE)
2106 ptr = (void *)__get_free_pages(flags, 0);
2107 else if (size > PAGE_SIZE) {
2108 int order = get_order(size);
2109
2110 if (order < 3)
2111 ptr = (void *)__get_free_pages(flags, order);
2112 else
2113 ptr = vmalloc(size);
2114 } else
2115 ptr = kmem_cache_alloc(get_slab(size), flags);
2116
2117 /* Check alignment; SLUB has gotten this wrong in the past,
2118 * and this can lead to user data corruption! */
2119 BUG_ON(((unsigned long) ptr) & (size-1));
2120
2121 return ptr;
2122}
2123
2124void jbd2_free(void *ptr, size_t size)
2125{
2126 if (size == PAGE_SIZE) {
2127 free_pages((unsigned long)ptr, 0);
2128 return;
2129 }
2130 if (size > PAGE_SIZE) {
2131 int order = get_order(size);
2132
2133 if (order < 3)
2134 free_pages((unsigned long)ptr, order);
2135 else
2136 vfree(ptr);
2137 return;
2138 }
2139 kmem_cache_free(get_slab(size), ptr);
2140};
2141
2142/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07002143 * Journal_head storage management
2144 */
Christoph Lametere18b8902006-12-06 20:33:20 -08002145static struct kmem_cache *jbd2_journal_head_cache;
Jose R. Santose23291b2007-07-18 08:57:06 -04002146#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002147static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2148#endif
2149
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002150static int jbd2_journal_init_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002151{
2152 int retval;
2153
Al Viro1076d172008-03-29 03:07:18 +00002154 J_ASSERT(jbd2_journal_head_cache == NULL);
Johann Lombardia920e942006-10-11 01:21:00 -07002155 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002156 sizeof(struct journal_head),
2157 0, /* offset */
Mingming Cao77160952008-01-28 23:58:27 -05002158 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09002159 NULL); /* ctor */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002160 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00002161 if (!jbd2_journal_head_cache) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002162 retval = -ENOMEM;
Eryu Guanf2a44522011-11-01 19:09:18 -04002163 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002164 }
2165 return retval;
2166}
2167
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002168static void jbd2_journal_destroy_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002169{
Duane Griffin8a9362e2008-04-17 10:38:59 -04002170 if (jbd2_journal_head_cache) {
2171 kmem_cache_destroy(jbd2_journal_head_cache);
2172 jbd2_journal_head_cache = NULL;
2173 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002174}
2175
2176/*
2177 * journal_head splicing and dicing
2178 */
2179static struct journal_head *journal_alloc_journal_head(void)
2180{
2181 struct journal_head *ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002182
Jose R. Santose23291b2007-07-18 08:57:06 -04002183#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002184 atomic_inc(&nr_journal_heads);
2185#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002186 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
Al Viro1076d172008-03-29 03:07:18 +00002187 if (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002188 jbd_debug(1, "out of memory for journal_head\n");
Theodore Ts'o670be5a2010-12-17 10:44:16 -05002189 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
Al Viro1076d172008-03-29 03:07:18 +00002190 while (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002191 yield();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002192 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002193 }
2194 }
2195 return ret;
2196}
2197
2198static void journal_free_journal_head(struct journal_head *jh)
2199{
Jose R. Santose23291b2007-07-18 08:57:06 -04002200#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002201 atomic_dec(&nr_journal_heads);
Mingming Caocd02ff02007-10-16 18:38:25 -04002202 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -07002203#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002204 kmem_cache_free(jbd2_journal_head_cache, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002205}
2206
2207/*
2208 * A journal_head is attached to a buffer_head whenever JBD has an
2209 * interest in the buffer.
2210 *
2211 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2212 * is set. This bit is tested in core kernel code where we need to take
2213 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2214 * there.
2215 *
2216 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2217 *
2218 * When a buffer has its BH_JBD bit set it is immune from being released by
2219 * core kernel code, mainly via ->b_count.
2220 *
Jan Karade1b7942011-06-13 15:38:22 -04002221 * A journal_head is detached from its buffer_head when the journal_head's
2222 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2223 * transaction (b_cp_transaction) hold their references to b_jcount.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002224 *
2225 * Various places in the kernel want to attach a journal_head to a buffer_head
2226 * _before_ attaching the journal_head to a transaction. To protect the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002227 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
Dave Kleikamp470decc2006-10-11 01:20:57 -07002228 * journal_head's b_jcount refcount by one. The caller must call
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002229 * jbd2_journal_put_journal_head() to undo this.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002230 *
2231 * So the typical usage would be:
2232 *
2233 * (Attach a journal_head if needed. Increments b_jcount)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002234 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002235 * ...
Jan Karade1b7942011-06-13 15:38:22 -04002236 * (Get another reference for transaction)
2237 * jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002238 * jh->b_transaction = xxx;
Jan Karade1b7942011-06-13 15:38:22 -04002239 * (Put original reference)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002240 * jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002241 */
2242
2243/*
2244 * Give a buffer_head a journal_head.
2245 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07002246 * May sleep.
2247 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002248struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002249{
2250 struct journal_head *jh;
2251 struct journal_head *new_jh = NULL;
2252
2253repeat:
2254 if (!buffer_jbd(bh)) {
2255 new_jh = journal_alloc_journal_head();
2256 memset(new_jh, 0, sizeof(*new_jh));
2257 }
2258
2259 jbd_lock_bh_journal_head(bh);
2260 if (buffer_jbd(bh)) {
2261 jh = bh2jh(bh);
2262 } else {
2263 J_ASSERT_BH(bh,
2264 (atomic_read(&bh->b_count) > 0) ||
2265 (bh->b_page && bh->b_page->mapping));
2266
2267 if (!new_jh) {
2268 jbd_unlock_bh_journal_head(bh);
2269 goto repeat;
2270 }
2271
2272 jh = new_jh;
2273 new_jh = NULL; /* We consumed it */
2274 set_buffer_jbd(bh);
2275 bh->b_private = jh;
2276 jh->b_bh = bh;
2277 get_bh(bh);
2278 BUFFER_TRACE(bh, "added journal_head");
2279 }
2280 jh->b_jcount++;
2281 jbd_unlock_bh_journal_head(bh);
2282 if (new_jh)
2283 journal_free_journal_head(new_jh);
2284 return bh->b_private;
2285}
2286
2287/*
2288 * Grab a ref against this buffer_head's journal_head. If it ended up not
2289 * having a journal_head, return NULL
2290 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002291struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002292{
2293 struct journal_head *jh = NULL;
2294
2295 jbd_lock_bh_journal_head(bh);
2296 if (buffer_jbd(bh)) {
2297 jh = bh2jh(bh);
2298 jh->b_jcount++;
2299 }
2300 jbd_unlock_bh_journal_head(bh);
2301 return jh;
2302}
2303
2304static void __journal_remove_journal_head(struct buffer_head *bh)
2305{
2306 struct journal_head *jh = bh2jh(bh);
2307
2308 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karade1b7942011-06-13 15:38:22 -04002309 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2310 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2311 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2312 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2313 J_ASSERT_BH(bh, buffer_jbd(bh));
2314 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2315 BUFFER_TRACE(bh, "remove journal_head");
2316 if (jh->b_frozen_data) {
2317 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2318 jbd2_free(jh->b_frozen_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002319 }
Jan Karade1b7942011-06-13 15:38:22 -04002320 if (jh->b_committed_data) {
2321 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2322 jbd2_free(jh->b_committed_data, bh->b_size);
2323 }
2324 bh->b_private = NULL;
2325 jh->b_bh = NULL; /* debug, really */
2326 clear_buffer_jbd(bh);
2327 journal_free_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002328}
2329
2330/*
Jan Karade1b7942011-06-13 15:38:22 -04002331 * Drop a reference on the passed journal_head. If it fell to zero then
Dave Kleikamp470decc2006-10-11 01:20:57 -07002332 * release the journal_head from the buffer_head.
2333 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002334void jbd2_journal_put_journal_head(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002335{
2336 struct buffer_head *bh = jh2bh(jh);
2337
2338 jbd_lock_bh_journal_head(bh);
2339 J_ASSERT_JH(jh, jh->b_jcount > 0);
2340 --jh->b_jcount;
Jan Karade1b7942011-06-13 15:38:22 -04002341 if (!jh->b_jcount) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002342 __journal_remove_journal_head(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002343 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002344 __brelse(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002345 } else
2346 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002347}
2348
2349/*
Jan Karac851ed52008-07-11 19:27:31 -04002350 * Initialize jbd inode head
2351 */
2352void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2353{
2354 jinode->i_transaction = NULL;
2355 jinode->i_next_transaction = NULL;
2356 jinode->i_vfs_inode = inode;
2357 jinode->i_flags = 0;
2358 INIT_LIST_HEAD(&jinode->i_list);
2359}
2360
2361/*
2362 * Function to be called before we start removing inode from memory (i.e.,
2363 * clear_inode() is a fine place to be called from). It removes inode from
2364 * transaction's lists.
2365 */
2366void jbd2_journal_release_jbd_inode(journal_t *journal,
2367 struct jbd2_inode *jinode)
2368{
Jan Karac851ed52008-07-11 19:27:31 -04002369 if (!journal)
2370 return;
2371restart:
2372 spin_lock(&journal->j_list_lock);
2373 /* Is commit writing out inode - we have to wait */
Brian King39e3ac22010-10-27 21:25:12 -04002374 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
Jan Karac851ed52008-07-11 19:27:31 -04002375 wait_queue_head_t *wq;
2376 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2377 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2378 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2379 spin_unlock(&journal->j_list_lock);
2380 schedule();
2381 finish_wait(wq, &wait.wait);
2382 goto restart;
2383 }
2384
Jan Karac851ed52008-07-11 19:27:31 -04002385 if (jinode->i_transaction) {
2386 list_del(&jinode->i_list);
2387 jinode->i_transaction = NULL;
2388 }
2389 spin_unlock(&journal->j_list_lock);
2390}
2391
2392/*
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002393 * debugfs tunables
Dave Kleikamp470decc2006-10-11 01:20:57 -07002394 */
Jose R. Santos6f38c742007-10-16 18:38:25 -04002395#ifdef CONFIG_JBD2_DEBUG
2396u8 jbd2_journal_enable_debug __read_mostly;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002397EXPORT_SYMBOL(jbd2_journal_enable_debug);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002398
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002399#define JBD2_DEBUG_NAME "jbd2-debug"
Dave Kleikamp470decc2006-10-11 01:20:57 -07002400
Jose R. Santos6f38c742007-10-16 18:38:25 -04002401static struct dentry *jbd2_debugfs_dir;
2402static struct dentry *jbd2_debug;
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002403
2404static void __init jbd2_create_debugfs_entry(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002405{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002406 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2407 if (jbd2_debugfs_dir)
Yin Kangkai765f8362009-12-15 14:48:25 -08002408 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2409 S_IRUGO | S_IWUSR,
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002410 jbd2_debugfs_dir,
2411 &jbd2_journal_enable_debug);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002412}
2413
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002414static void __exit jbd2_remove_debugfs_entry(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002415{
Jose R. Santos6f38c742007-10-16 18:38:25 -04002416 debugfs_remove(jbd2_debug);
2417 debugfs_remove(jbd2_debugfs_dir);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002418}
2419
2420#else
2421
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002422static void __init jbd2_create_debugfs_entry(void)
2423{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002424}
2425
2426static void __exit jbd2_remove_debugfs_entry(void)
2427{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002428}
Dave Kleikamp470decc2006-10-11 01:20:57 -07002429
2430#endif
2431
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002432#ifdef CONFIG_PROC_FS
2433
2434#define JBD2_STATS_PROC_NAME "fs/jbd2"
2435
2436static void __init jbd2_create_jbd_stats_proc_entry(void)
2437{
2438 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2439}
2440
2441static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2442{
2443 if (proc_jbd2_stats)
2444 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2445}
2446
2447#else
2448
2449#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2450#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2451
2452#endif
2453
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002454struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002455
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002456static int __init jbd2_journal_init_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002457{
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002458 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002459 if (jbd2_handle_cache == NULL) {
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002460 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2461 return -ENOMEM;
2462 }
2463 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2464 if (jbd2_inode_cache == NULL) {
2465 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2466 kmem_cache_destroy(jbd2_handle_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002467 return -ENOMEM;
2468 }
2469 return 0;
2470}
2471
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002472static void jbd2_journal_destroy_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002473{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002474 if (jbd2_handle_cache)
2475 kmem_cache_destroy(jbd2_handle_cache);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002476 if (jbd2_inode_cache)
2477 kmem_cache_destroy(jbd2_inode_cache);
2478
Dave Kleikamp470decc2006-10-11 01:20:57 -07002479}
2480
2481/*
2482 * Module startup and shutdown
2483 */
2484
2485static int __init journal_init_caches(void)
2486{
2487 int ret;
2488
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002489 ret = jbd2_journal_init_revoke_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002490 if (ret == 0)
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002491 ret = jbd2_journal_init_journal_head_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002492 if (ret == 0)
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002493 ret = jbd2_journal_init_handle_cache();
Yongqiang Yang0c2022e2012-02-20 17:53:02 -05002494 if (ret == 0)
2495 ret = jbd2_journal_init_transaction_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002496 return ret;
2497}
2498
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002499static void jbd2_journal_destroy_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002500{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002501 jbd2_journal_destroy_revoke_caches();
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002502 jbd2_journal_destroy_journal_head_cache();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002503 jbd2_journal_destroy_handle_cache();
Yongqiang Yang0c2022e2012-02-20 17:53:02 -05002504 jbd2_journal_destroy_transaction_cache();
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002505 jbd2_journal_destroy_slabs();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002506}
2507
2508static int __init journal_init(void)
2509{
2510 int ret;
2511
2512 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2513
2514 ret = journal_init_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002515 if (ret == 0) {
2516 jbd2_create_debugfs_entry();
2517 jbd2_create_jbd_stats_proc_entry();
2518 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002519 jbd2_journal_destroy_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002520 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002521 return ret;
2522}
2523
2524static void __exit journal_exit(void)
2525{
Jose R. Santose23291b2007-07-18 08:57:06 -04002526#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002527 int n = atomic_read(&nr_journal_heads);
2528 if (n)
Eryu Guanf2a44522011-11-01 19:09:18 -04002529 printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002530#endif
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002531 jbd2_remove_debugfs_entry();
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002532 jbd2_remove_jbd_stats_proc_entry();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002533 jbd2_journal_destroy_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002534}
2535
Dave Kleikamp470decc2006-10-11 01:20:57 -07002536MODULE_LICENSE("GPL");
2537module_init(journal_init);
2538module_exit(journal_exit);
2539