blob: 81e622681c82273aa050eb85cdb057b1fea0e79e [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>
Johann Lombardi8e85fb32008-01-28 23:58:27 -050038#include <linux/seq_file.h>
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -050039#include <linux/math64.h>
Theodore Ts'o879c5e62009-06-17 11:47:48 -040040#include <linux/hash.h>
Theodore Ts'od2eecb02009-12-07 10:36:20 -050041#include <linux/log2.h>
42#include <linux/vmalloc.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040043#include <linux/backing-dev.h>
Brian King39e3ac22010-10-27 21:25:12 -040044#include <linux/bitops.h>
Theodore Ts'o670be5a2010-12-17 10:44:16 -050045#include <linux/ratelimit.h>
Theodore Ts'o879c5e62009-06-17 11:47:48 -040046
47#define CREATE_TRACE_POINTS
48#include <trace/events/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070049
50#include <asm/uaccess.h>
51#include <asm/page.h>
52
Theodore Ts'ob6e96d02013-02-09 16:29:20 -050053#ifdef CONFIG_JBD2_DEBUG
54ushort jbd2_journal_enable_debug __read_mostly;
55EXPORT_SYMBOL(jbd2_journal_enable_debug);
56
57module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
58MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
59#endif
60
Mingming Caof7f4bcc2006-10-11 01:20:59 -070061EXPORT_SYMBOL(jbd2_journal_extend);
62EXPORT_SYMBOL(jbd2_journal_stop);
63EXPORT_SYMBOL(jbd2_journal_lock_updates);
64EXPORT_SYMBOL(jbd2_journal_unlock_updates);
65EXPORT_SYMBOL(jbd2_journal_get_write_access);
66EXPORT_SYMBOL(jbd2_journal_get_create_access);
67EXPORT_SYMBOL(jbd2_journal_get_undo_access);
Joel Beckere06c8222008-09-11 15:35:47 -070068EXPORT_SYMBOL(jbd2_journal_set_triggers);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070069EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070070EXPORT_SYMBOL(jbd2_journal_forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -070071#if 0
72EXPORT_SYMBOL(journal_sync_buffer);
73#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -070074EXPORT_SYMBOL(jbd2_journal_flush);
75EXPORT_SYMBOL(jbd2_journal_revoke);
Dave Kleikamp470decc2006-10-11 01:20:57 -070076
Mingming Caof7f4bcc2006-10-11 01:20:59 -070077EXPORT_SYMBOL(jbd2_journal_init_dev);
78EXPORT_SYMBOL(jbd2_journal_init_inode);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070079EXPORT_SYMBOL(jbd2_journal_check_used_features);
80EXPORT_SYMBOL(jbd2_journal_check_available_features);
81EXPORT_SYMBOL(jbd2_journal_set_features);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070082EXPORT_SYMBOL(jbd2_journal_load);
83EXPORT_SYMBOL(jbd2_journal_destroy);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070084EXPORT_SYMBOL(jbd2_journal_abort);
85EXPORT_SYMBOL(jbd2_journal_errno);
86EXPORT_SYMBOL(jbd2_journal_ack_err);
87EXPORT_SYMBOL(jbd2_journal_clear_err);
88EXPORT_SYMBOL(jbd2_log_wait_commit);
Theodore Ts'o3b799d12009-12-09 20:42:53 -050089EXPORT_SYMBOL(jbd2_log_start_commit);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070090EXPORT_SYMBOL(jbd2_journal_start_commit);
91EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
92EXPORT_SYMBOL(jbd2_journal_wipe);
93EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
94EXPORT_SYMBOL(jbd2_journal_invalidatepage);
95EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
96EXPORT_SYMBOL(jbd2_journal_force_commit);
Jan Karac851ed52008-07-11 19:27:31 -040097EXPORT_SYMBOL(jbd2_journal_file_inode);
98EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
99EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
100EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500101EXPORT_SYMBOL(jbd2_inode_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700102
Dave Kleikamp470decc2006-10-11 01:20:57 -0700103static void __journal_abort_soft (journal_t *journal, int errno);
Theodore Ts'od2eecb02009-12-07 10:36:20 -0500104static int jbd2_journal_create_slab(size_t slab_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700105
Paul Gortmaker169f1a22013-06-12 23:04:04 -0400106#ifdef CONFIG_JBD2_DEBUG
107void __jbd2_debug(int level, const char *file, const char *func,
108 unsigned int line, const char *fmt, ...)
109{
110 struct va_format vaf;
111 va_list args;
112
113 if (level > jbd2_journal_enable_debug)
114 return;
115 va_start(args, fmt);
116 vaf.fmt = fmt;
117 vaf.va = &args;
118 printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
119 va_end(args);
120}
121EXPORT_SYMBOL(__jbd2_debug);
122#endif
123
Darrick J. Wong25ed6e82012-05-27 07:48:56 -0400124/* Checksumming functions */
Rashika Kheria7747e6d2014-02-17 20:49:04 -0500125static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
Darrick J. Wong25ed6e82012-05-27 07:48:56 -0400126{
Darrick J. Wong85957982015-10-15 10:30:36 -0400127 if (!jbd2_journal_has_csum_v2or3_feature(j))
Darrick J. Wong25ed6e82012-05-27 07:48:56 -0400128 return 1;
129
130 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
131}
132
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400133static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400134{
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400135 __u32 csum;
136 __be32 old_csum;
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400137
138 old_csum = sb->s_checksum;
139 sb->s_checksum = 0;
140 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
141 sb->s_checksum = old_csum;
142
143 return cpu_to_be32(csum);
144}
145
Rashika Kheria7747e6d2014-02-17 20:49:04 -0500146static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400147{
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400148 if (!jbd2_journal_has_csum_v2or3(j))
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400149 return 1;
150
151 return sb->s_checksum == jbd2_superblock_csum(j, sb);
152}
153
Rashika Kheria7747e6d2014-02-17 20:49:04 -0500154static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400155{
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400156 if (!jbd2_journal_has_csum_v2or3(j))
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400157 return;
158
159 sb->s_checksum = jbd2_superblock_csum(j, sb);
160}
161
Dave Kleikamp470decc2006-10-11 01:20:57 -0700162/*
163 * Helper function used to manage commit timeouts
164 */
165
166static void commit_timeout(unsigned long __data)
167{
168 struct task_struct * p = (struct task_struct *) __data;
169
170 wake_up_process(p);
171}
172
173/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700174 * kjournald2: The main thread function used to manage a logging device
Dave Kleikamp470decc2006-10-11 01:20:57 -0700175 * journal.
176 *
177 * This kernel thread is responsible for two things:
178 *
179 * 1) COMMIT: Every so often we need to commit the current state of the
180 * filesystem to disk. The journal thread is responsible for writing
181 * all of the metadata buffers to disk.
182 *
183 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
184 * of the data in that part of the log has been rewritten elsewhere on
185 * the disk. Flushing these old buffers to reclaim space in the log is
186 * known as checkpointing, and this thread is responsible for that job.
187 */
188
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700189static int kjournald2(void *arg)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700190{
191 journal_t *journal = arg;
192 transaction_t *transaction;
193
194 /*
195 * Set up an interval timer which can be used to trigger a commit wakeup
196 * after the commit interval expires
197 */
198 setup_timer(&journal->j_commit_timer, commit_timeout,
199 (unsigned long)current);
200
Nigel Cunningham35c80422012-02-03 19:59:41 +1100201 set_freezable();
202
Dave Kleikamp470decc2006-10-11 01:20:57 -0700203 /* Record that the journal thread is running */
204 journal->j_task = current;
205 wake_up(&journal->j_wait_done_commit);
206
Dave Kleikamp470decc2006-10-11 01:20:57 -0700207 /*
208 * And now, wait forever for commit wakeup events.
209 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400210 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700211
212loop:
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700213 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700214 goto end_loop;
215
216 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
217 journal->j_commit_sequence, journal->j_commit_request);
218
219 if (journal->j_commit_sequence != journal->j_commit_request) {
220 jbd_debug(1, "OK, requests differ\n");
Theodore Ts'oa931da62010-08-03 21:35:12 -0400221 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700222 del_timer_sync(&journal->j_commit_timer);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700223 jbd2_journal_commit_transaction(journal);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400224 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700225 goto loop;
226 }
227
228 wake_up(&journal->j_wait_done_commit);
229 if (freezing(current)) {
230 /*
231 * The simpler the better. Flushing journal isn't a
232 * good idea, because that depends on threads that may
233 * be already stopped.
234 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700235 jbd_debug(1, "Now suspending kjournald2\n");
Theodore Ts'oa931da62010-08-03 21:35:12 -0400236 write_unlock(&journal->j_state_lock);
Tejun Heoa0acae02011-11-21 12:32:22 -0800237 try_to_freeze();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400238 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700239 } else {
240 /*
241 * We assume on resume that commits are already there,
242 * so we don't sleep
243 */
244 DEFINE_WAIT(wait);
245 int should_sleep = 1;
246
247 prepare_to_wait(&journal->j_wait_commit, &wait,
248 TASK_INTERRUPTIBLE);
249 if (journal->j_commit_sequence != journal->j_commit_request)
250 should_sleep = 0;
251 transaction = journal->j_running_transaction;
252 if (transaction && time_after_eq(jiffies,
253 transaction->t_expires))
254 should_sleep = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700255 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700256 should_sleep = 0;
257 if (should_sleep) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400258 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700259 schedule();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400260 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700261 }
262 finish_wait(&journal->j_wait_commit, &wait);
263 }
264
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700265 jbd_debug(1, "kjournald2 wakes\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700266
267 /*
268 * Were we woken up by a commit wakeup event?
269 */
270 transaction = journal->j_running_transaction;
271 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
272 journal->j_commit_request = transaction->t_tid;
273 jbd_debug(1, "woke because of timeout\n");
274 }
275 goto loop;
276
277end_loop:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400278 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700279 del_timer_sync(&journal->j_commit_timer);
280 journal->j_task = NULL;
281 wake_up(&journal->j_wait_done_commit);
282 jbd_debug(1, "Journal thread exiting.\n");
283 return 0;
284}
285
Pavel Emelianov97f06782007-05-08 00:30:42 -0700286static int jbd2_journal_start_thread(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700287{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700288 struct task_struct *t;
289
Theodore Ts'o90576c02009-09-29 15:51:30 -0400290 t = kthread_run(kjournald2, journal, "jbd2/%s",
291 journal->j_devname);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700292 if (IS_ERR(t))
293 return PTR_ERR(t);
294
Al Viro1076d172008-03-29 03:07:18 +0000295 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700296 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700297}
298
299static void journal_kill_thread(journal_t *journal)
300{
Theodore Ts'oa931da62010-08-03 21:35:12 -0400301 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700302 journal->j_flags |= JBD2_UNMOUNT;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700303
304 while (journal->j_task) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400305 write_unlock(&journal->j_state_lock);
Theodore Ts'o3469a322014-03-08 19:11:36 -0500306 wake_up(&journal->j_wait_commit);
Al Viro1076d172008-03-29 03:07:18 +0000307 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400308 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700309 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400310 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700311}
312
313/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700314 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700315 *
316 * Writes a metadata buffer to a given disk block. The actual IO is not
317 * performed but a new buffer_head is constructed which labels the data
318 * to be written with the correct destination disk block.
319 *
320 * Any magic-number escaping which needs to be done will cause a
321 * copy-out here. If the buffer happens to start with the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700322 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700323 * magic number is only written to the log for descripter blocks. In
324 * this case, we copy the data and replace the first word with 0, and we
325 * return a result code which indicates that this buffer needs to be
326 * marked as an escaped buffer in the corresponding log descriptor
327 * block. The missing word can then be restored when the block is read
328 * during recovery.
329 *
330 * If the source buffer has already been modified by a new transaction
331 * since we took the last commit snapshot, we use the frozen copy of
Jan Karaf5113ef2013-06-04 12:01:45 -0400332 * that data for IO. If we end up using the existing buffer_head's data
333 * for the write, then we have to make sure nobody modifies it while the
334 * IO is in progress. do_get_write_access() handles this.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700335 *
Jan Karaf5113ef2013-06-04 12:01:45 -0400336 * The function returns a pointer to the buffer_head to be used for IO.
337 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700338 *
339 * Return value:
340 * <0: Error
341 * >=0: Finished OK
342 *
343 * On success:
344 * Bit 0 set == escape performed on the data
345 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
346 */
347
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700348int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700349 struct journal_head *jh_in,
Jan Karaf5113ef2013-06-04 12:01:45 -0400350 struct buffer_head **bh_out,
351 sector_t blocknr)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700352{
353 int need_copy_out = 0;
354 int done_copy_out = 0;
355 int do_escape = 0;
356 char *mapped_data;
357 struct buffer_head *new_bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700358 struct page *new_page;
359 unsigned int new_offset;
360 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghua96577c42009-07-13 17:55:35 -0400361 journal_t *journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700362
363 /*
364 * The buffer really shouldn't be locked: only the current committing
365 * transaction is allowed to write it, so nobody else is allowed
366 * to do any IO.
367 *
368 * akpm: except if we're journalling data, and write() output is
369 * also part of a shared mapping, and another thread has
370 * decided to launch a writepage() against this buffer.
371 */
372 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
373
Michal Hocko6ccaf3e2015-06-08 10:53:10 -0400374 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
Theodore Ts'o47def822010-07-27 11:56:05 -0400375
dingdinghua96577c42009-07-13 17:55:35 -0400376 /* keep subsequent assertions sane */
dingdinghua96577c42009-07-13 17:55:35 -0400377 atomic_set(&new_bh->b_count, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700378
Jan Karaf5113ef2013-06-04 12:01:45 -0400379 jbd_lock_bh_state(bh_in);
380repeat:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700381 /*
382 * If a new transaction has already done a buffer copy-out, then
383 * we use that version of the data for the commit.
384 */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700385 if (jh_in->b_frozen_data) {
386 done_copy_out = 1;
387 new_page = virt_to_page(jh_in->b_frozen_data);
388 new_offset = offset_in_page(jh_in->b_frozen_data);
389 } else {
390 new_page = jh2bh(jh_in)->b_page;
391 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
392 }
393
Cong Wang303a8f22011-11-25 23:14:31 +0800394 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700395 /*
Jan Kara13ceef02010-07-14 07:56:33 +0200396 * Fire data frozen trigger if data already wasn't frozen. Do this
397 * before checking for escaping, as the trigger may modify the magic
398 * offset. If a copy-out happens afterwards, it will have the correct
399 * data in the buffer.
Joel Beckere06c8222008-09-11 15:35:47 -0700400 */
Jan Kara13ceef02010-07-14 07:56:33 +0200401 if (!done_copy_out)
402 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
403 jh_in->b_triggers);
Joel Beckere06c8222008-09-11 15:35:47 -0700404
405 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700406 * Check for escaping
407 */
408 if (*((__be32 *)(mapped_data + new_offset)) ==
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700409 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700410 need_copy_out = 1;
411 do_escape = 1;
412 }
Cong Wang303a8f22011-11-25 23:14:31 +0800413 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700414
415 /*
416 * Do we need to do a data copy?
417 */
418 if (need_copy_out && !done_copy_out) {
419 char *tmp;
420
421 jbd_unlock_bh_state(bh_in);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400422 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500423 if (!tmp) {
Jan Karaf5113ef2013-06-04 12:01:45 -0400424 brelse(new_bh);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500425 return -ENOMEM;
426 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700427 jbd_lock_bh_state(bh_in);
428 if (jh_in->b_frozen_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400429 jbd2_free(tmp, bh_in->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700430 goto repeat;
431 }
432
433 jh_in->b_frozen_data = tmp;
Cong Wang303a8f22011-11-25 23:14:31 +0800434 mapped_data = kmap_atomic(new_page);
Jan Karaf5113ef2013-06-04 12:01:45 -0400435 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
Cong Wang303a8f22011-11-25 23:14:31 +0800436 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700437
438 new_page = virt_to_page(tmp);
439 new_offset = offset_in_page(tmp);
440 done_copy_out = 1;
Joel Beckere06c8222008-09-11 15:35:47 -0700441
442 /*
443 * This isn't strictly necessary, as we're using frozen
444 * data for the escaping, but it keeps consistency with
445 * b_frozen_data usage.
446 */
447 jh_in->b_frozen_triggers = jh_in->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700448 }
449
450 /*
451 * Did we need to do an escaping? Now we've done all the
452 * copying, we can finally do so.
453 */
454 if (do_escape) {
Cong Wang303a8f22011-11-25 23:14:31 +0800455 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700456 *((unsigned int *)(mapped_data + new_offset)) = 0;
Cong Wang303a8f22011-11-25 23:14:31 +0800457 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700458 }
459
Dave Kleikamp470decc2006-10-11 01:20:57 -0700460 set_bh_page(new_bh, new_page, new_offset);
Jan Karaf5113ef2013-06-04 12:01:45 -0400461 new_bh->b_size = bh_in->b_size;
462 new_bh->b_bdev = journal->j_dev;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700463 new_bh->b_blocknr = blocknr;
Jan Karab34090e2013-06-04 12:08:56 -0400464 new_bh->b_private = bh_in;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700465 set_buffer_mapped(new_bh);
466 set_buffer_dirty(new_bh);
467
Jan Karaf5113ef2013-06-04 12:01:45 -0400468 *bh_out = new_bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700469
470 /*
471 * The to-be-written buffer needs to get moved to the io queue,
472 * and the original buffer whose contents we are shadowing or
473 * copying is moved to the transaction's shadow queue.
474 */
475 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghua96577c42009-07-13 17:55:35 -0400476 spin_lock(&journal->j_list_lock);
477 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
478 spin_unlock(&journal->j_list_lock);
Jan Karab34090e2013-06-04 12:08:56 -0400479 set_buffer_shadow(bh_in);
dingdinghua96577c42009-07-13 17:55:35 -0400480 jbd_unlock_bh_state(bh_in);
481
Dave Kleikamp470decc2006-10-11 01:20:57 -0700482 return do_escape | (done_copy_out << 1);
483}
484
485/*
486 * Allocation code for the journal file. Manage the space left in the
487 * journal, so that we can begin checkpointing when appropriate.
488 */
489
490/*
Theodore Ts'oe4471832011-02-12 08:18:24 -0500491 * Called with j_state_lock locked for writing.
492 * Returns true if a transaction commit was started.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700493 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700494int __jbd2_log_start_commit(journal_t *journal, tid_t target)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700495{
Eric Sandeene7b04ac2013-01-30 00:39:28 -0500496 /* Return if the txn has already requested to be committed */
497 if (journal->j_commit_request == target)
498 return 0;
499
Dave Kleikamp470decc2006-10-11 01:20:57 -0700500 /*
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400501 * The only transaction we can possibly wait upon is the
502 * currently running transaction (if it exists). Otherwise,
503 * the target tid must be an old one.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700504 */
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400505 if (journal->j_running_transaction &&
506 journal->j_running_transaction->t_tid == target) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700507 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200508 * We want a new commit: OK, mark the request and wakeup the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700509 * commit thread. We do _not_ do the commit ourselves.
510 */
511
512 journal->j_commit_request = target;
Eryu Guanf2a44522011-11-01 19:09:18 -0400513 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700514 journal->j_commit_request,
515 journal->j_commit_sequence);
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500516 journal->j_running_transaction->t_requested = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700517 wake_up(&journal->j_wait_commit);
518 return 1;
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400519 } else if (!tid_geq(journal->j_commit_request, target))
520 /* This should never happen, but if it does, preserve
521 the evidence before kjournald goes into a loop and
522 increments j_commit_sequence beyond all recognition. */
Eryu Guanf2a44522011-11-01 19:09:18 -0400523 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
Theodore Ts'o1be2add2011-05-08 19:37:54 -0400524 journal->j_commit_request,
525 journal->j_commit_sequence,
526 target, journal->j_running_transaction ?
527 journal->j_running_transaction->t_tid : 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700528 return 0;
529}
530
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700531int jbd2_log_start_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700532{
533 int ret;
534
Theodore Ts'oa931da62010-08-03 21:35:12 -0400535 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700536 ret = __jbd2_log_start_commit(journal, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400537 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700538 return ret;
539}
540
541/*
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400542 * Force and wait any uncommitted transactions. We can only force the running
543 * transaction if we don't have an active handle, otherwise, we will deadlock.
544 * Returns: <0 in case of error,
545 * 0 if nothing to commit,
546 * 1 if transaction was successfully committed.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700547 */
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400548static int __jbd2_journal_force_commit(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700549{
550 transaction_t *transaction = NULL;
551 tid_t tid;
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400552 int need_to_start = 0, ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700553
Theodore Ts'oa931da62010-08-03 21:35:12 -0400554 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700555 if (journal->j_running_transaction && !current->journal_info) {
556 transaction = journal->j_running_transaction;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500557 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
558 need_to_start = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700559 } else if (journal->j_committing_transaction)
560 transaction = journal->j_committing_transaction;
561
562 if (!transaction) {
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400563 /* Nothing to commit */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400564 read_unlock(&journal->j_state_lock);
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400565 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700566 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700567 tid = transaction->t_tid;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400568 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500569 if (need_to_start)
570 jbd2_log_start_commit(journal, tid);
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400571 ret = jbd2_log_wait_commit(journal, tid);
572 if (!ret)
573 ret = 1;
574
575 return ret;
576}
577
578/**
579 * Force and wait upon a commit if the calling process is not within
580 * transaction. This is used for forcing out undo-protected data which contains
581 * bitmaps, when the fs is running out of space.
582 *
583 * @journal: journal to force
584 * Returns true if progress was made.
585 */
586int jbd2_journal_force_commit_nested(journal_t *journal)
587{
588 int ret;
589
590 ret = __jbd2_journal_force_commit(journal);
591 return ret > 0;
592}
593
594/**
595 * int journal_force_commit() - force any uncommitted transactions
596 * @journal: journal to force
597 *
598 * Caller want unconditional commit. We can only force the running transaction
599 * if we don't have an active handle, otherwise, we will deadlock.
600 */
601int jbd2_journal_force_commit(journal_t *journal)
602{
603 int ret;
604
605 J_ASSERT(!current->journal_info);
606 ret = __jbd2_journal_force_commit(journal);
607 if (ret > 0)
608 ret = 0;
609 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700610}
611
612/*
613 * Start a commit of the current running transaction (if any). Returns true
Jan Karac88ccea2009-02-10 11:27:46 -0500614 * if a transaction is going to be committed (or is currently already
615 * committing), and fills its tid in at *ptid
Dave Kleikamp470decc2006-10-11 01:20:57 -0700616 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700617int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700618{
619 int ret = 0;
620
Theodore Ts'oa931da62010-08-03 21:35:12 -0400621 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700622 if (journal->j_running_transaction) {
623 tid_t tid = journal->j_running_transaction->t_tid;
624
Jan Karac88ccea2009-02-10 11:27:46 -0500625 __jbd2_log_start_commit(journal, tid);
626 /* There's a running transaction and we've just made sure
627 * it's commit has been scheduled. */
628 if (ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700629 *ptid = tid;
Jan Karac88ccea2009-02-10 11:27:46 -0500630 ret = 1;
631 } else if (journal->j_committing_transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700632 /*
Artem Bityutskiy12810ad2012-07-25 18:12:07 +0300633 * If commit has been started, then we have to wait for
634 * completion of that transaction.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700635 */
Jan Karac88ccea2009-02-10 11:27:46 -0500636 if (ptid)
637 *ptid = journal->j_committing_transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700638 ret = 1;
639 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400640 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700641 return ret;
642}
643
644/*
Jan Karabbd2be32011-05-24 11:59:18 -0400645 * Return 1 if a given transaction has not yet sent barrier request
646 * connected with a transaction commit. If 0 is returned, transaction
647 * may or may not have sent the barrier. Used to avoid sending barrier
648 * twice in common cases.
649 */
650int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
651{
652 int ret = 0;
653 transaction_t *commit_trans;
654
655 if (!(journal->j_flags & JBD2_BARRIER))
656 return 0;
657 read_lock(&journal->j_state_lock);
658 /* Transaction already committed? */
659 if (tid_geq(journal->j_commit_sequence, tid))
660 goto out;
661 commit_trans = journal->j_committing_transaction;
662 if (!commit_trans || commit_trans->t_tid != tid) {
663 ret = 1;
664 goto out;
665 }
666 /*
667 * Transaction is being committed and we already proceeded to
668 * submitting a flush to fs partition?
669 */
670 if (journal->j_fs_dev != journal->j_dev) {
671 if (!commit_trans->t_need_data_flush ||
672 commit_trans->t_state >= T_COMMIT_DFLUSH)
673 goto out;
674 } else {
675 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
676 goto out;
677 }
678 ret = 1;
679out:
680 read_unlock(&journal->j_state_lock);
681 return ret;
682}
683EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
684
685/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700686 * Wait for a specified commit to complete.
687 * The caller may not hold the journal lock.
688 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700689int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700690{
691 int err = 0;
692
Theodore Ts'oa931da62010-08-03 21:35:12 -0400693 read_lock(&journal->j_state_lock);
Jose R. Santose23291b2007-07-18 08:57:06 -0400694#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -0700695 if (!tid_geq(journal->j_commit_request, tid)) {
Jan Kara75685072013-12-08 21:13:59 -0500696 printk(KERN_ERR
Dave Kleikamp470decc2006-10-11 01:20:57 -0700697 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400698 __func__, journal->j_commit_request, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700699 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700700#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -0700701 while (tid_gt(tid, journal->j_commit_sequence)) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400702 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700703 tid, journal->j_commit_sequence);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400704 read_unlock(&journal->j_state_lock);
Theodore Ts'o3469a322014-03-08 19:11:36 -0500705 wake_up(&journal->j_wait_commit);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700706 wait_event(journal->j_wait_done_commit,
707 !tid_gt(tid, journal->j_commit_sequence));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400708 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700709 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400710 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700711
Jan Kara75685072013-12-08 21:13:59 -0500712 if (unlikely(is_journal_aborted(journal)))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700713 err = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700714 return err;
715}
716
717/*
Theodore Ts'od76a3a772013-04-03 22:02:52 -0400718 * When this function returns the transaction corresponding to tid
719 * will be completed. If the transaction has currently running, start
720 * committing that transaction before waiting for it to complete. If
721 * the transaction id is stale, it is by definition already completed,
722 * so just return SUCCESS.
723 */
724int jbd2_complete_transaction(journal_t *journal, tid_t tid)
725{
726 int need_to_wait = 1;
727
728 read_lock(&journal->j_state_lock);
729 if (journal->j_running_transaction &&
730 journal->j_running_transaction->t_tid == tid) {
731 if (journal->j_commit_request != tid) {
732 /* transaction not yet started, so request it */
733 read_unlock(&journal->j_state_lock);
734 jbd2_log_start_commit(journal, tid);
735 goto wait_commit;
736 }
737 } else if (!(journal->j_committing_transaction &&
738 journal->j_committing_transaction->t_tid == tid))
739 need_to_wait = 0;
740 read_unlock(&journal->j_state_lock);
741 if (!need_to_wait)
742 return 0;
743wait_commit:
744 return jbd2_log_wait_commit(journal, tid);
745}
746EXPORT_SYMBOL(jbd2_complete_transaction);
747
748/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700749 * Log buffer allocation routines:
750 */
751
Mingming Cao18eba7a2006-10-11 01:21:13 -0700752int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700753{
754 unsigned long blocknr;
755
Theodore Ts'oa931da62010-08-03 21:35:12 -0400756 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700757 J_ASSERT(journal->j_free > 1);
758
759 blocknr = journal->j_head;
760 journal->j_head++;
761 journal->j_free--;
762 if (journal->j_head == journal->j_last)
763 journal->j_head = journal->j_first;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400764 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700765 return jbd2_journal_bmap(journal, blocknr, retp);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700766}
767
768/*
769 * Conversion of logical to physical block numbers for the journal
770 *
771 * On external journals the journal blocks are identity-mapped, so
772 * this is a no-op. If needed, we can use j_blk_offset - everything is
773 * ready.
774 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700775int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700776 unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700777{
778 int err = 0;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700779 unsigned long long ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700780
781 if (journal->j_inode) {
782 ret = bmap(journal->j_inode, blocknr);
783 if (ret)
784 *retp = ret;
785 else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700786 printk(KERN_ALERT "%s: journal block not found "
787 "at offset %lu on %s\n",
Theodore Ts'o05496762008-09-16 14:36:17 -0400788 __func__, blocknr, journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700789 err = -EIO;
790 __journal_abort_soft(journal, err);
791 }
792 } else {
793 *retp = blocknr; /* +journal->j_blk_offset */
794 }
795 return err;
796}
797
798/*
799 * We play buffer_head aliasing tricks to write data/metadata blocks to
800 * the journal without copying their contents, but for journal
801 * descriptor blocks we do need to generate bona fide buffers.
802 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700803 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
Dave Kleikamp470decc2006-10-11 01:20:57 -0700804 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
805 * But we don't bother doing that, so there will be coherency problems with
806 * mmaps of blockdevs which hold live JBD-controlled filesystems.
807 */
Jan Karae5a120a2013-06-04 12:06:01 -0400808struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700809{
810 struct buffer_head *bh;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700811 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700812 int err;
813
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700814 err = jbd2_journal_next_log_block(journal, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700815
816 if (err)
817 return NULL;
818
819 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500820 if (!bh)
821 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700822 lock_buffer(bh);
823 memset(bh->b_data, 0, journal->j_blocksize);
824 set_buffer_uptodate(bh);
825 unlock_buffer(bh);
826 BUFFER_TRACE(bh, "return this buffer");
Jan Karae5a120a2013-06-04 12:06:01 -0400827 return bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700828}
829
Jan Kara79feb522012-03-13 22:22:54 -0400830/*
831 * Return tid of the oldest transaction in the journal and block in the journal
832 * where the transaction starts.
833 *
834 * If the journal is now empty, return which will be the next transaction ID
835 * we will write and where will that transaction start.
836 *
837 * The return value is 0 if journal tail cannot be pushed any further, 1 if
838 * it can.
839 */
840int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
841 unsigned long *block)
842{
843 transaction_t *transaction;
844 int ret;
845
846 read_lock(&journal->j_state_lock);
847 spin_lock(&journal->j_list_lock);
848 transaction = journal->j_checkpoint_transactions;
849 if (transaction) {
850 *tid = transaction->t_tid;
851 *block = transaction->t_log_start;
852 } else if ((transaction = journal->j_committing_transaction) != NULL) {
853 *tid = transaction->t_tid;
854 *block = transaction->t_log_start;
855 } else if ((transaction = journal->j_running_transaction) != NULL) {
856 *tid = transaction->t_tid;
857 *block = journal->j_head;
858 } else {
859 *tid = journal->j_transaction_sequence;
860 *block = journal->j_head;
861 }
862 ret = tid_gt(*tid, journal->j_tail_sequence);
863 spin_unlock(&journal->j_list_lock);
864 read_unlock(&journal->j_state_lock);
865
866 return ret;
867}
868
869/*
870 * Update information in journal structure and in on disk journal superblock
871 * about log tail. This function does not check whether information passed in
872 * really pushes log tail further. It's responsibility of the caller to make
873 * sure provided log tail information is valid (e.g. by holding
874 * j_checkpoint_mutex all the time between computing log tail and calling this
875 * function as is the case with jbd2_cleanup_journal_tail()).
876 *
877 * Requires j_checkpoint_mutex
878 */
Joseph Qi6f6a6fd2015-06-15 14:36:01 -0400879int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
Jan Kara79feb522012-03-13 22:22:54 -0400880{
881 unsigned long freed;
Joseph Qi6f6a6fd2015-06-15 14:36:01 -0400882 int ret;
Jan Kara79feb522012-03-13 22:22:54 -0400883
884 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
885
886 /*
887 * We cannot afford for write to remain in drive's caches since as
888 * soon as we update j_tail, next transaction can start reusing journal
889 * space and if we lose sb update during power failure we'd replay
890 * old transaction with possibly newly overwritten data.
891 */
Joseph Qi6f6a6fd2015-06-15 14:36:01 -0400892 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
893 if (ret)
894 goto out;
895
Jan Kara79feb522012-03-13 22:22:54 -0400896 write_lock(&journal->j_state_lock);
897 freed = block - journal->j_tail;
898 if (block < journal->j_tail)
899 freed += journal->j_last - journal->j_first;
900
901 trace_jbd2_update_log_tail(journal, tid, block, freed);
902 jbd_debug(1,
903 "Cleaning journal tail from %d to %d (offset %lu), "
904 "freeing %lu\n",
905 journal->j_tail_sequence, tid, block, freed);
906
907 journal->j_free += freed;
908 journal->j_tail_sequence = tid;
909 journal->j_tail = block;
910 write_unlock(&journal->j_state_lock);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -0400911
912out:
913 return ret;
Jan Kara79feb522012-03-13 22:22:54 -0400914}
915
Jan Kara33395782012-03-13 22:45:38 -0400916/*
917 * This is a variaon of __jbd2_update_log_tail which checks for validity of
918 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
919 * with other threads updating log tail.
920 */
921void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
922{
923 mutex_lock(&journal->j_checkpoint_mutex);
924 if (tid_gt(tid, journal->j_tail_sequence))
925 __jbd2_update_log_tail(journal, tid, block);
926 mutex_unlock(&journal->j_checkpoint_mutex);
927}
928
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500929struct jbd2_stats_proc_session {
930 journal_t *journal;
931 struct transaction_stats_s *stats;
932 int start;
933 int max;
934};
935
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500936static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
937{
938 return *pos ? NULL : SEQ_START_TOKEN;
939}
940
941static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
942{
943 return NULL;
944}
945
946static int jbd2_seq_info_show(struct seq_file *seq, void *v)
947{
948 struct jbd2_stats_proc_session *s = seq->private;
949
950 if (v != SEQ_START_TOKEN)
951 return 0;
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500952 seq_printf(seq, "%lu transactions (%lu requested), "
953 "each up to %u blocks\n",
954 s->stats->ts_tid, s->stats->ts_requested,
955 s->journal->j_max_transaction_buffers);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500956 if (s->stats->ts_tid == 0)
957 return 0;
958 seq_printf(seq, "average: \n %ums waiting for transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400959 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500960 seq_printf(seq, " %ums request delay\n",
961 (s->stats->ts_requested == 0) ? 0 :
962 jiffies_to_msecs(s->stats->run.rs_request_delay /
963 s->stats->ts_requested));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500964 seq_printf(seq, " %ums running transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400965 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500966 seq_printf(seq, " %ums transaction was being locked\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400967 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500968 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400969 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500970 seq_printf(seq, " %ums logging transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400971 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -0500972 seq_printf(seq, " %lluus average transaction commit time\n",
973 div_u64(s->journal->j_average_commit_time, 1000));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500974 seq_printf(seq, " %lu handles per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400975 s->stats->run.rs_handle_count / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500976 seq_printf(seq, " %lu blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400977 s->stats->run.rs_blocks / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500978 seq_printf(seq, " %lu logged blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400979 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500980 return 0;
981}
982
983static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
984{
985}
986
James Morris88e9d342009-09-22 16:43:43 -0700987static const struct seq_operations jbd2_seq_info_ops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500988 .start = jbd2_seq_info_start,
989 .next = jbd2_seq_info_next,
990 .stop = jbd2_seq_info_stop,
991 .show = jbd2_seq_info_show,
992};
993
994static int jbd2_seq_info_open(struct inode *inode, struct file *file)
995{
Al Virod9dda782013-03-31 18:16:14 -0400996 journal_t *journal = PDE_DATA(inode);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500997 struct jbd2_stats_proc_session *s;
998 int rc, size;
999
1000 s = kmalloc(sizeof(*s), GFP_KERNEL);
1001 if (s == NULL)
1002 return -ENOMEM;
1003 size = sizeof(struct transaction_stats_s);
1004 s->stats = kmalloc(size, GFP_KERNEL);
1005 if (s->stats == NULL) {
1006 kfree(s);
1007 return -ENOMEM;
1008 }
1009 spin_lock(&journal->j_history_lock);
1010 memcpy(s->stats, &journal->j_stats, size);
1011 s->journal = journal;
1012 spin_unlock(&journal->j_history_lock);
1013
1014 rc = seq_open(file, &jbd2_seq_info_ops);
1015 if (rc == 0) {
1016 struct seq_file *m = file->private_data;
1017 m->private = s;
1018 } else {
1019 kfree(s->stats);
1020 kfree(s);
1021 }
1022 return rc;
1023
1024}
1025
1026static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1027{
1028 struct seq_file *seq = file->private_data;
1029 struct jbd2_stats_proc_session *s = seq->private;
1030 kfree(s->stats);
1031 kfree(s);
1032 return seq_release(inode, file);
1033}
1034
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001035static const struct file_operations jbd2_seq_info_fops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001036 .owner = THIS_MODULE,
1037 .open = jbd2_seq_info_open,
1038 .read = seq_read,
1039 .llseek = seq_lseek,
1040 .release = jbd2_seq_info_release,
1041};
1042
1043static struct proc_dir_entry *proc_jbd2_stats;
1044
1045static void jbd2_stats_proc_init(journal_t *journal)
1046{
Theodore Ts'o05496762008-09-16 14:36:17 -04001047 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001048 if (journal->j_proc_entry) {
Denis V. Lunev79da3662008-04-29 01:02:11 -07001049 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1050 &jbd2_seq_info_fops, journal);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001051 }
1052}
1053
1054static void jbd2_stats_proc_exit(journal_t *journal)
1055{
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001056 remove_proc_entry("info", journal->j_proc_entry);
Theodore Ts'o05496762008-09-16 14:36:17 -04001057 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001058}
1059
Dave Kleikamp470decc2006-10-11 01:20:57 -07001060/*
1061 * Management for journal control blocks: functions to create and
1062 * destroy journal_t structures, and to initialise and read existing
1063 * journal blocks from disk. */
1064
1065/* First: create and setup a journal_t object in memory. We initialise
1066 * very few fields yet: that has to wait until we have created the
1067 * journal structures from from scratch, or loaded them from disk. */
1068
1069static journal_t * journal_init_common (void)
1070{
1071 journal_t *journal;
1072 int err;
1073
Andrew Morton3ebfdf82009-12-23 08:05:15 -05001074 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001075 if (!journal)
Theodore Ts'ob7271b02010-12-18 13:39:38 -05001076 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001077
1078 init_waitqueue_head(&journal->j_wait_transaction_locked);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001079 init_waitqueue_head(&journal->j_wait_done_commit);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001080 init_waitqueue_head(&journal->j_wait_commit);
1081 init_waitqueue_head(&journal->j_wait_updates);
Jan Kara8f7d89f2013-06-04 12:35:11 -04001082 init_waitqueue_head(&journal->j_wait_reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001083 mutex_init(&journal->j_barrier);
1084 mutex_init(&journal->j_checkpoint_mutex);
1085 spin_lock_init(&journal->j_revoke_lock);
1086 spin_lock_init(&journal->j_list_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001087 rwlock_init(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001088
Mingming Caocd02ff02007-10-16 18:38:25 -04001089 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
Theodore Ts'o30773842009-01-03 20:27:38 -05001090 journal->j_min_batch_time = 0;
1091 journal->j_max_batch_time = 15000; /* 15ms */
Jan Kara8f7d89f2013-06-04 12:35:11 -04001092 atomic_set(&journal->j_reserved_credits, 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001093
1094 /* The journal is marked for error until we succeed with recovery! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001095 journal->j_flags = JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001096
1097 /* Set up a default-sized revoke table for the new mount. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001098 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001099 if (err) {
1100 kfree(journal);
Theodore Ts'ob7271b02010-12-18 13:39:38 -05001101 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001102 }
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001103
Theodore Ts'obf699322009-09-30 00:32:06 -04001104 spin_lock_init(&journal->j_history_lock);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001105
Dave Kleikamp470decc2006-10-11 01:20:57 -07001106 return journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001107}
1108
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001109/* jbd2_journal_init_dev and jbd2_journal_init_inode:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001110 *
1111 * Create a journal structure assigned some fixed set of disk blocks to
1112 * the journal. We don't actually touch those disk blocks yet, but we
1113 * need to set up all of the mapping information to tell the journaling
1114 * system where the journal blocks are.
1115 *
1116 */
1117
1118/**
Randy Dunlap5648ba52008-04-17 10:38:59 -04001119 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
Dave Kleikamp470decc2006-10-11 01:20:57 -07001120 * @bdev: Block device on which to create the journal
1121 * @fs_dev: Device which hold journalled filesystem for this journal.
1122 * @start: Block nr Start of journal.
1123 * @len: Length of the journal in blocks.
1124 * @blocksize: blocksize of journalling device
Randy Dunlap5648ba52008-04-17 10:38:59 -04001125 *
1126 * Returns: a newly created journal_t *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001127 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001128 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
Dave Kleikamp470decc2006-10-11 01:20:57 -07001129 * range of blocks on an arbitrary block device.
1130 *
1131 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001132journal_t * jbd2_journal_init_dev(struct block_device *bdev,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001133 struct block_device *fs_dev,
Mingming Cao18eba7a2006-10-11 01:21:13 -07001134 unsigned long long start, int len, int blocksize)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001135{
1136 journal_t *journal = journal_init_common();
1137 struct buffer_head *bh;
1138 int n;
1139
1140 if (!journal)
1141 return NULL;
1142
1143 /* journal descriptor can store up to n blocks -bzzz */
1144 journal->j_blocksize = blocksize;
yangsheng0587aa32010-11-17 21:46:26 -05001145 journal->j_dev = bdev;
1146 journal->j_fs_dev = fs_dev;
1147 journal->j_blk_offset = start;
1148 journal->j_maxlen = len;
1149 bdevname(journal->j_dev, journal->j_devname);
Rasmus Villemoes81ae3942015-06-25 15:02:38 -07001150 strreplace(journal->j_devname, '/', '!');
Jan Kara4b905672009-01-06 14:53:35 -05001151 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001152 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1153 journal->j_wbufsize = n;
1154 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1155 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001156 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001157 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001158 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001159 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001160
1161 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001162 if (!bh) {
1163 printk(KERN_ERR
1164 "%s: Cannot get buffer for journal superblock\n",
1165 __func__);
1166 goto out_err;
1167 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001168 journal->j_sb_buffer = bh;
1169 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Kara4b905672009-01-06 14:53:35 -05001170
Dave Kleikamp470decc2006-10-11 01:20:57 -07001171 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001172out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001173 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001174 jbd2_stats_proc_exit(journal);
1175 kfree(journal);
1176 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001177}
1178
1179/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001180 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001181 * @inode: An inode to create the journal in
1182 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001183 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
Dave Kleikamp470decc2006-10-11 01:20:57 -07001184 * the journal. The inode must exist already, must support bmap() and
1185 * must have all data blocks preallocated.
1186 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001187journal_t * jbd2_journal_init_inode (struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001188{
1189 struct buffer_head *bh;
1190 journal_t *journal = journal_init_common();
Theodore Ts'o05496762008-09-16 14:36:17 -04001191 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001192 int err;
1193 int n;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001194 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001195
1196 if (!journal)
1197 return NULL;
1198
1199 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1200 journal->j_inode = inode;
Theodore Ts'o05496762008-09-16 14:36:17 -04001201 bdevname(journal->j_dev, journal->j_devname);
Rasmus Villemoes81ae3942015-06-25 15:02:38 -07001202 p = strreplace(journal->j_devname, '/', '!');
Theodore Ts'o90576c02009-09-29 15:51:30 -04001203 sprintf(p, "-%lu", journal->j_inode->i_ino);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001204 jbd_debug(1,
1205 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1206 journal, inode->i_sb->s_id, inode->i_ino,
1207 (long long) inode->i_size,
1208 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1209
1210 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1211 journal->j_blocksize = inode->i_sb->s_blocksize;
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001212 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001213
1214 /* journal descriptor can store up to n blocks -bzzz */
1215 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1216 journal->j_wbufsize = n;
1217 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1218 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001219 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001220 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001221 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001222 }
1223
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001224 err = jbd2_journal_bmap(journal, 0, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001225 /* If that failed, give up */
1226 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -08001227 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001228 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001229 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001230 }
1231
Gioh Kima49058f2014-09-04 22:36:35 -04001232 bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001233 if (!bh) {
1234 printk(KERN_ERR
1235 "%s: Cannot get buffer for journal superblock\n",
1236 __func__);
1237 goto out_err;
1238 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001239 journal->j_sb_buffer = bh;
1240 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1241
1242 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001243out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001244 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001245 jbd2_stats_proc_exit(journal);
1246 kfree(journal);
1247 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001248}
1249
1250/*
1251 * If the journal init or create aborts, we need to mark the journal
1252 * superblock as being NULL to prevent the journal destroy from writing
1253 * back a bogus superblock.
1254 */
1255static void journal_fail_superblock (journal_t *journal)
1256{
1257 struct buffer_head *bh = journal->j_sb_buffer;
1258 brelse(bh);
1259 journal->j_sb_buffer = NULL;
1260}
1261
1262/*
1263 * Given a journal_t structure, initialise the various fields for
1264 * startup of a new journaling session. We use this both when creating
1265 * a journal, and after recovering an old journal to reset it for
1266 * subsequent use.
1267 */
1268
1269static int journal_reset(journal_t *journal)
1270{
1271 journal_superblock_t *sb = journal->j_superblock;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001272 unsigned long long first, last;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001273
1274 first = be32_to_cpu(sb->s_first);
1275 last = be32_to_cpu(sb->s_maxlen);
Jan Karaf6f50e22009-07-17 10:40:01 -04001276 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001277 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
Jan Karaf6f50e22009-07-17 10:40:01 -04001278 first, last);
1279 journal_fail_superblock(journal);
1280 return -EINVAL;
1281 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001282
1283 journal->j_first = first;
1284 journal->j_last = last;
1285
1286 journal->j_head = first;
1287 journal->j_tail = first;
1288 journal->j_free = last - first;
1289
1290 journal->j_tail_sequence = journal->j_transaction_sequence;
1291 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1292 journal->j_commit_request = journal->j_commit_sequence;
1293
1294 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1295
Dave Kleikamp470decc2006-10-11 01:20:57 -07001296 /*
1297 * As a special case, if the on-disk copy is already marked as needing
Jan Kara24bcc892012-03-13 15:41:04 -04001298 * no recovery (s_start == 0), then we can safely defer the superblock
1299 * update until the next commit by setting JBD2_FLUSHED. This avoids
Dave Kleikamp470decc2006-10-11 01:20:57 -07001300 * attempting a write to a potential-readonly device.
1301 */
Jan Kara24bcc892012-03-13 15:41:04 -04001302 if (sb->s_start == 0) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001303 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
Dave Kleikamp470decc2006-10-11 01:20:57 -07001304 "(start %ld, seq %d, errno %d)\n",
1305 journal->j_tail, journal->j_tail_sequence,
1306 journal->j_errno);
Jan Kara24bcc892012-03-13 15:41:04 -04001307 journal->j_flags |= JBD2_FLUSHED;
1308 } else {
Jan Karaa78bb112012-03-13 15:43:04 -04001309 /* Lock here to make assertions happy... */
1310 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara79feb522012-03-13 22:22:54 -04001311 /*
1312 * Update log tail information. We use WRITE_FUA since new
1313 * transaction will start reusing journal space and so we
1314 * must make sure information about current log tail is on
1315 * disk before that.
1316 */
1317 jbd2_journal_update_sb_log_tail(journal,
1318 journal->j_tail_sequence,
1319 journal->j_tail,
1320 WRITE_FUA);
Jan Karaa78bb112012-03-13 15:43:04 -04001321 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001322 }
Jan Kara24bcc892012-03-13 15:41:04 -04001323 return jbd2_journal_start_thread(journal);
1324}
Dave Kleikamp470decc2006-10-11 01:20:57 -07001325
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001326static int jbd2_write_superblock(journal_t *journal, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001327{
1328 struct buffer_head *bh = journal->j_sb_buffer;
Theodore Ts'ofe52d172013-07-01 08:12:38 -04001329 journal_superblock_t *sb = journal->j_superblock;
Jan Kara79feb522012-03-13 22:22:54 -04001330 int ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001331
Jan Kara79feb522012-03-13 22:22:54 -04001332 trace_jbd2_write_superblock(journal, write_op);
1333 if (!(journal->j_flags & JBD2_BARRIER))
1334 write_op &= ~(REQ_FUA | REQ_FLUSH);
1335 lock_buffer(bh);
Theodore Ts'o914258b2008-10-06 21:35:40 -04001336 if (buffer_write_io_error(bh)) {
1337 /*
1338 * Oh, dear. A previous attempt to write the journal
1339 * superblock failed. This could happen because the
1340 * USB device was yanked out. Or it could happen to
1341 * be a transient write error and maybe the block will
1342 * be remapped. Nothing we can do but to retry the
1343 * write and hope for the best.
1344 */
1345 printk(KERN_ERR "JBD2: previous I/O error detected "
1346 "for journal superblock update for %s.\n",
1347 journal->j_devname);
1348 clear_buffer_write_io_error(bh);
1349 set_buffer_uptodate(bh);
1350 }
Theodore Ts'ofe52d172013-07-01 08:12:38 -04001351 jbd2_superblock_csum_set(journal, sb);
Jan Kara79feb522012-03-13 22:22:54 -04001352 get_bh(bh);
1353 bh->b_end_io = end_buffer_write_sync;
1354 ret = submit_bh(write_op, bh);
1355 wait_on_buffer(bh);
Jan Kara24bcc892012-03-13 15:41:04 -04001356 if (buffer_write_io_error(bh)) {
Jan Kara24bcc892012-03-13 15:41:04 -04001357 clear_buffer_write_io_error(bh);
1358 set_buffer_uptodate(bh);
Jan Kara79feb522012-03-13 22:22:54 -04001359 ret = -EIO;
1360 }
1361 if (ret) {
1362 printk(KERN_ERR "JBD2: Error %d detected when updating "
1363 "journal superblock for %s.\n", ret,
1364 journal->j_devname);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001365 jbd2_journal_abort(journal, ret);
Jan Kara24bcc892012-03-13 15:41:04 -04001366 }
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001367
1368 return ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001369}
Theodore Ts'o914258b2008-10-06 21:35:40 -04001370
Jan Kara24bcc892012-03-13 15:41:04 -04001371/**
1372 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1373 * @journal: The journal to update.
Jan Kara79feb522012-03-13 22:22:54 -04001374 * @tail_tid: TID of the new transaction at the tail of the log
1375 * @tail_block: The first block of the transaction at the tail of the log
1376 * @write_op: With which operation should we write the journal sb
Jan Kara24bcc892012-03-13 15:41:04 -04001377 *
1378 * Update a journal's superblock information about log tail and write it to
1379 * disk, waiting for the IO to complete.
1380 */
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001381int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
Jan Kara79feb522012-03-13 22:22:54 -04001382 unsigned long tail_block, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001383{
1384 journal_superblock_t *sb = journal->j_superblock;
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001385 int ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001386
Jan Karaa78bb112012-03-13 15:43:04 -04001387 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Jan Kara79feb522012-03-13 22:22:54 -04001388 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1389 tail_block, tail_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001390
Jan Kara79feb522012-03-13 22:22:54 -04001391 sb->s_sequence = cpu_to_be32(tail_tid);
1392 sb->s_start = cpu_to_be32(tail_block);
Jan Kara24bcc892012-03-13 15:41:04 -04001393
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001394 ret = jbd2_write_superblock(journal, write_op);
1395 if (ret)
1396 goto out;
Jan Kara24bcc892012-03-13 15:41:04 -04001397
1398 /* Log is no longer empty */
1399 write_lock(&journal->j_state_lock);
1400 WARN_ON(!sb->s_sequence);
1401 journal->j_flags &= ~JBD2_FLUSHED;
1402 write_unlock(&journal->j_state_lock);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001403
1404out:
1405 return ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001406}
1407
1408/**
1409 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1410 * @journal: The journal to update.
1411 *
1412 * Update a journal's dynamic superblock fields to show that journal is empty.
1413 * Write updated superblock to disk waiting for IO to complete.
1414 */
1415static void jbd2_mark_journal_empty(journal_t *journal)
1416{
1417 journal_superblock_t *sb = journal->j_superblock;
1418
Jan Karaa78bb112012-03-13 15:43:04 -04001419 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001420 read_lock(&journal->j_state_lock);
Eric Sandeeneeecef02012-08-18 22:29:40 -04001421 /* Is it already empty? */
1422 if (sb->s_start == 0) {
1423 read_unlock(&journal->j_state_lock);
1424 return;
1425 }
Jan Kara24bcc892012-03-13 15:41:04 -04001426 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1427 journal->j_tail_sequence);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001428
1429 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
Jan Kara24bcc892012-03-13 15:41:04 -04001430 sb->s_start = cpu_to_be32(0);
1431 read_unlock(&journal->j_state_lock);
1432
Jan Kara79feb522012-03-13 22:22:54 -04001433 jbd2_write_superblock(journal, WRITE_FUA);
Jan Kara24bcc892012-03-13 15:41:04 -04001434
1435 /* Log is no longer empty */
1436 write_lock(&journal->j_state_lock);
1437 journal->j_flags |= JBD2_FLUSHED;
1438 write_unlock(&journal->j_state_lock);
1439}
1440
1441
1442/**
1443 * jbd2_journal_update_sb_errno() - Update error in the journal.
1444 * @journal: The journal to update.
1445 *
1446 * Update a journal's errno. Write updated superblock to disk waiting for IO
1447 * to complete.
1448 */
Theodore Ts'od796c522012-08-05 19:04:57 -04001449void jbd2_journal_update_sb_errno(journal_t *journal)
Jan Kara24bcc892012-03-13 15:41:04 -04001450{
1451 journal_superblock_t *sb = journal->j_superblock;
1452
1453 read_lock(&journal->j_state_lock);
1454 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1455 journal->j_errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001456 sb->s_errno = cpu_to_be32(journal->j_errno);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001457 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001458
Daeho Jeong564bc402015-07-23 09:46:11 -04001459 jbd2_write_superblock(journal, WRITE_FUA);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001460}
Theodore Ts'od796c522012-08-05 19:04:57 -04001461EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001462
1463/*
1464 * Read the superblock for a given journal, performing initial
1465 * validation of the format.
1466 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001467static int journal_get_superblock(journal_t *journal)
1468{
1469 struct buffer_head *bh;
1470 journal_superblock_t *sb;
1471 int err = -EIO;
1472
1473 bh = journal->j_sb_buffer;
1474
1475 J_ASSERT(bh != NULL);
1476 if (!buffer_uptodate(bh)) {
1477 ll_rw_block(READ, 1, &bh);
1478 wait_on_buffer(bh);
1479 if (!buffer_uptodate(bh)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001480 printk(KERN_ERR
1481 "JBD2: IO error reading journal superblock\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001482 goto out;
1483 }
1484 }
1485
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001486 if (buffer_verified(bh))
1487 return 0;
1488
Dave Kleikamp470decc2006-10-11 01:20:57 -07001489 sb = journal->j_superblock;
1490
1491 err = -EINVAL;
1492
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001493 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001494 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001495 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001496 goto out;
1497 }
1498
1499 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001500 case JBD2_SUPERBLOCK_V1:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001501 journal->j_format_version = 1;
1502 break;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001503 case JBD2_SUPERBLOCK_V2:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001504 journal->j_format_version = 2;
1505 break;
1506 default:
Eryu Guanf2a44522011-11-01 19:09:18 -04001507 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001508 goto out;
1509 }
1510
1511 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1512 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1513 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001514 printk(KERN_WARNING "JBD2: journal file too short\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001515 goto out;
1516 }
1517
Eryu Guan87622022011-11-01 19:04:59 -04001518 if (be32_to_cpu(sb->s_first) == 0 ||
1519 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1520 printk(KERN_WARNING
1521 "JBD2: Invalid start block of journal: %u\n",
1522 be32_to_cpu(sb->s_first));
1523 goto out;
1524 }
1525
Darrick J. Wong56316a02015-10-17 16:18:45 -04001526 if (jbd2_has_feature_csum2(journal) &&
1527 jbd2_has_feature_csum3(journal)) {
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001528 /* Can't have checksum v2 and v3 at the same time! */
1529 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1530 "at the same time!\n");
1531 goto out;
1532 }
1533
Darrick J. Wong85957982015-10-15 10:30:36 -04001534 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
Darrick J. Wong56316a02015-10-17 16:18:45 -04001535 jbd2_has_feature_checksum(journal)) {
Darrick J. Wongfeb8c6d2014-09-11 11:38:21 -04001536 /* Can't have checksum v1 and v2 on at the same time! */
1537 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1538 "at the same time!\n");
1539 goto out;
1540 }
1541
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001542 if (!jbd2_verify_csum_type(journal, sb)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001543 printk(KERN_ERR "JBD2: Unknown checksum type\n");
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001544 goto out;
1545 }
1546
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001547 /* Load the checksum driver */
Darrick J. Wong85957982015-10-15 10:30:36 -04001548 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001549 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1550 if (IS_ERR(journal->j_chksum_driver)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001551 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001552 err = PTR_ERR(journal->j_chksum_driver);
1553 journal->j_chksum_driver = NULL;
1554 goto out;
1555 }
1556 }
1557
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001558 /* Check superblock checksum */
1559 if (!jbd2_superblock_csum_verify(journal, sb)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001560 printk(KERN_ERR "JBD2: journal checksum error\n");
Darrick J. Wong6a797d22015-10-17 16:16:04 -04001561 err = -EFSBADCRC;
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001562 goto out;
1563 }
1564
1565 /* Precompute checksum seed for all metadata */
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001566 if (jbd2_journal_has_csum_v2or3(journal))
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001567 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1568 sizeof(sb->s_uuid));
1569
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001570 set_buffer_verified(bh);
1571
Dave Kleikamp470decc2006-10-11 01:20:57 -07001572 return 0;
1573
1574out:
1575 journal_fail_superblock(journal);
1576 return err;
1577}
1578
1579/*
1580 * Load the on-disk journal superblock and read the key fields into the
1581 * journal_t.
1582 */
1583
1584static int load_superblock(journal_t *journal)
1585{
1586 int err;
1587 journal_superblock_t *sb;
1588
1589 err = journal_get_superblock(journal);
1590 if (err)
1591 return err;
1592
1593 sb = journal->j_superblock;
1594
1595 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1596 journal->j_tail = be32_to_cpu(sb->s_start);
1597 journal->j_first = be32_to_cpu(sb->s_first);
1598 journal->j_last = be32_to_cpu(sb->s_maxlen);
1599 journal->j_errno = be32_to_cpu(sb->s_errno);
1600
1601 return 0;
1602}
1603
1604
1605/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001606 * int jbd2_journal_load() - Read journal from disk.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001607 * @journal: Journal to act on.
1608 *
1609 * Given a journal_t structure which tells us which disk blocks contain
1610 * a journal, read the journal from disk to initialise the in-memory
1611 * structures.
1612 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001613int jbd2_journal_load(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001614{
1615 int err;
1616 journal_superblock_t *sb;
1617
1618 err = load_superblock(journal);
1619 if (err)
1620 return err;
1621
1622 sb = journal->j_superblock;
1623 /* If this is a V2 superblock, then we have to check the
1624 * features flags on it. */
1625
1626 if (journal->j_format_version >= 2) {
1627 if ((sb->s_feature_ro_compat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001628 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001629 (sb->s_feature_incompat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001630 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001631 printk(KERN_WARNING
1632 "JBD2: Unrecognised features on journal\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001633 return -EINVAL;
1634 }
1635 }
1636
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001637 /*
1638 * Create a slab for this blocksize
1639 */
1640 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1641 if (err)
1642 return err;
1643
Dave Kleikamp470decc2006-10-11 01:20:57 -07001644 /* Let the recovery code check whether it needs to recover any
1645 * data from the journal. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001646 if (jbd2_journal_recover(journal))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001647 goto recovery_error;
1648
Theodore Ts'oe6a47422009-11-15 15:31:37 -05001649 if (journal->j_failed_commit) {
1650 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1651 "is corrupt.\n", journal->j_failed_commit,
1652 journal->j_devname);
Darrick J. Wong6a797d22015-10-17 16:16:04 -04001653 return -EFSCORRUPTED;
Theodore Ts'oe6a47422009-11-15 15:31:37 -05001654 }
1655
Dave Kleikamp470decc2006-10-11 01:20:57 -07001656 /* OK, we've finished with the dynamic journal bits:
1657 * reinitialise the dynamic contents of the superblock in memory
1658 * and reset them on disk. */
1659 if (journal_reset(journal))
1660 goto recovery_error;
1661
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001662 journal->j_flags &= ~JBD2_ABORT;
1663 journal->j_flags |= JBD2_LOADED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001664 return 0;
1665
1666recovery_error:
Eryu Guanf2a44522011-11-01 19:09:18 -04001667 printk(KERN_WARNING "JBD2: recovery failed\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001668 return -EIO;
1669}
1670
1671/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001672 * void jbd2_journal_destroy() - Release a journal_t structure.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001673 * @journal: Journal to act on.
1674 *
1675 * Release a journal_t structure once it is no longer in use by the
1676 * journaled object.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001677 * Return <0 if we couldn't clean up the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001678 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001679int jbd2_journal_destroy(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001680{
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001681 int err = 0;
1682
Dave Kleikamp470decc2006-10-11 01:20:57 -07001683 /* Wait for the commit thread to wake up and die. */
1684 journal_kill_thread(journal);
1685
1686 /* Force a final log commit */
1687 if (journal->j_running_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001688 jbd2_journal_commit_transaction(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001689
1690 /* Force any old transactions to disk */
1691
1692 /* Totally anal locking here... */
1693 spin_lock(&journal->j_list_lock);
1694 while (journal->j_checkpoint_transactions != NULL) {
1695 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001696 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara841df7df2015-07-28 14:57:14 -04001697 err = jbd2_log_do_checkpoint(journal);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001698 mutex_unlock(&journal->j_checkpoint_mutex);
Jan Kara841df7df2015-07-28 14:57:14 -04001699 /*
1700 * If checkpointing failed, just free the buffers to avoid
1701 * looping forever
1702 */
1703 if (err) {
1704 jbd2_journal_destroy_checkpoint(journal);
1705 spin_lock(&journal->j_list_lock);
1706 break;
1707 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001708 spin_lock(&journal->j_list_lock);
1709 }
1710
1711 J_ASSERT(journal->j_running_transaction == NULL);
1712 J_ASSERT(journal->j_committing_transaction == NULL);
1713 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1714 spin_unlock(&journal->j_list_lock);
1715
Dave Kleikamp470decc2006-10-11 01:20:57 -07001716 if (journal->j_sb_buffer) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001717 if (!is_journal_aborted(journal)) {
Jan Karaa78bb112012-03-13 15:43:04 -04001718 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara24bcc892012-03-13 15:41:04 -04001719 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04001720 mutex_unlock(&journal->j_checkpoint_mutex);
1721 } else
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001722 err = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001723 brelse(journal->j_sb_buffer);
1724 }
1725
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001726 if (journal->j_proc_entry)
1727 jbd2_stats_proc_exit(journal);
Theodore Ts'od9f39d12014-11-25 20:02:37 -05001728 iput(journal->j_inode);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001729 if (journal->j_revoke)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001730 jbd2_journal_destroy_revoke(journal);
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001731 if (journal->j_chksum_driver)
1732 crypto_free_shash(journal->j_chksum_driver);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001733 kfree(journal->j_wbuf);
1734 kfree(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001735
1736 return err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001737}
1738
1739
1740/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001741 *int jbd2_journal_check_used_features () - Check if features specified are used.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001742 * @journal: Journal to check.
1743 * @compat: bitmask of compatible features
1744 * @ro: bitmask of features that force read-only mount
1745 * @incompat: bitmask of incompatible features
1746 *
1747 * Check whether the journal uses all of a given set of
1748 * features. Return true (non-zero) if it does.
1749 **/
1750
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001751int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001752 unsigned long ro, unsigned long incompat)
1753{
1754 journal_superblock_t *sb;
1755
1756 if (!compat && !ro && !incompat)
1757 return 1;
Patrick J. LoPresti1113e1b2010-07-22 15:04:16 -07001758 /* Load journal superblock if it is not loaded yet. */
1759 if (journal->j_format_version == 0 &&
1760 journal_get_superblock(journal) != 0)
1761 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001762 if (journal->j_format_version == 1)
1763 return 0;
1764
1765 sb = journal->j_superblock;
1766
1767 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1768 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1769 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1770 return 1;
1771
1772 return 0;
1773}
1774
1775/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001776 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001777 * @journal: Journal to check.
1778 * @compat: bitmask of compatible features
1779 * @ro: bitmask of features that force read-only mount
1780 * @incompat: bitmask of incompatible features
1781 *
1782 * Check whether the journaling code supports the use of
1783 * all of a given set of features on this journal. Return true
1784 * (non-zero) if it can. */
1785
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001786int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001787 unsigned long ro, unsigned long incompat)
1788{
Dave Kleikamp470decc2006-10-11 01:20:57 -07001789 if (!compat && !ro && !incompat)
1790 return 1;
1791
Dave Kleikamp470decc2006-10-11 01:20:57 -07001792 /* We can support any known requested features iff the
1793 * superblock is in version 2. Otherwise we fail to support any
1794 * extended sb features. */
1795
1796 if (journal->j_format_version != 2)
1797 return 0;
1798
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001799 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1800 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1801 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001802 return 1;
1803
1804 return 0;
1805}
1806
1807/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001808 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
Dave Kleikamp470decc2006-10-11 01:20:57 -07001809 * @journal: Journal to act on.
1810 * @compat: bitmask of compatible features
1811 * @ro: bitmask of features that force read-only mount
1812 * @incompat: bitmask of incompatible features
1813 *
1814 * Mark a given journal feature as present on the
1815 * superblock. Returns true if the requested features could be set.
1816 *
1817 */
1818
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001819int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001820 unsigned long ro, unsigned long incompat)
1821{
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001822#define INCOMPAT_FEATURE_ON(f) \
1823 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1824#define COMPAT_FEATURE_ON(f) \
1825 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001826 journal_superblock_t *sb;
1827
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001828 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001829 return 1;
1830
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001831 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001832 return 0;
1833
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001834 /* If enabling v2 checksums, turn on v3 instead */
1835 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1836 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1837 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1838 }
1839
1840 /* Asking for checksumming v3 and v1? Only give them v3. */
1841 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001842 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1843 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1844
Dave Kleikamp470decc2006-10-11 01:20:57 -07001845 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1846 compat, ro, incompat);
1847
1848 sb = journal->j_superblock;
1849
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001850 /* If enabling v3 checksums, update superblock */
1851 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001852 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1853 sb->s_feature_compat &=
1854 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001855
1856 /* Load the checksum driver */
1857 if (journal->j_chksum_driver == NULL) {
1858 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1859 0, 0);
1860 if (IS_ERR(journal->j_chksum_driver)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001861 printk(KERN_ERR "JBD2: Cannot load crc32c "
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001862 "driver.\n");
1863 journal->j_chksum_driver = NULL;
1864 return 0;
1865 }
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001866
Darrick J. Wong32f38692014-12-01 16:22:23 -08001867 /* Precompute checksum seed for all metadata */
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001868 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1869 sb->s_uuid,
1870 sizeof(sb->s_uuid));
Darrick J. Wong32f38692014-12-01 16:22:23 -08001871 }
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001872 }
1873
1874 /* If enabling v1 checksums, downgrade superblock */
1875 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1876 sb->s_feature_incompat &=
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001877 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1878 JBD2_FEATURE_INCOMPAT_CSUM_V3);
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001879
Dave Kleikamp470decc2006-10-11 01:20:57 -07001880 sb->s_feature_compat |= cpu_to_be32(compat);
1881 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1882 sb->s_feature_incompat |= cpu_to_be32(incompat);
1883
1884 return 1;
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001885#undef COMPAT_FEATURE_ON
1886#undef INCOMPAT_FEATURE_ON
Dave Kleikamp470decc2006-10-11 01:20:57 -07001887}
1888
Girish Shilamkar818d2762008-01-28 23:58:27 -05001889/*
1890 * jbd2_journal_clear_features () - Clear a given journal feature in the
1891 * superblock
1892 * @journal: Journal to act on.
1893 * @compat: bitmask of compatible features
1894 * @ro: bitmask of features that force read-only mount
1895 * @incompat: bitmask of incompatible features
1896 *
1897 * Clear a given journal feature as present on the
1898 * superblock.
1899 */
1900void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1901 unsigned long ro, unsigned long incompat)
1902{
1903 journal_superblock_t *sb;
1904
1905 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1906 compat, ro, incompat);
1907
1908 sb = journal->j_superblock;
1909
1910 sb->s_feature_compat &= ~cpu_to_be32(compat);
1911 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1912 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1913}
1914EXPORT_SYMBOL(jbd2_journal_clear_features);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001915
1916/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001917 * int jbd2_journal_flush () - Flush journal
Dave Kleikamp470decc2006-10-11 01:20:57 -07001918 * @journal: Journal to act on.
1919 *
1920 * Flush all data for a given journal to disk and empty the journal.
1921 * Filesystems can use this when remounting readonly to ensure that
1922 * recovery does not need to happen on remount.
1923 */
1924
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001925int jbd2_journal_flush(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001926{
1927 int err = 0;
1928 transaction_t *transaction = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001929
Theodore Ts'oa931da62010-08-03 21:35:12 -04001930 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001931
1932 /* Force everything buffered to the log... */
1933 if (journal->j_running_transaction) {
1934 transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001935 __jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001936 } else if (journal->j_committing_transaction)
1937 transaction = journal->j_committing_transaction;
1938
1939 /* Wait for the log commit to complete... */
1940 if (transaction) {
1941 tid_t tid = transaction->t_tid;
1942
Theodore Ts'oa931da62010-08-03 21:35:12 -04001943 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001944 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001945 } else {
Theodore Ts'oa931da62010-08-03 21:35:12 -04001946 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001947 }
1948
1949 /* ...and flush everything in the log out to disk. */
1950 spin_lock(&journal->j_list_lock);
1951 while (!err && journal->j_checkpoint_transactions != NULL) {
1952 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001953 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001954 err = jbd2_log_do_checkpoint(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001955 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001956 spin_lock(&journal->j_list_lock);
1957 }
1958 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001959
1960 if (is_journal_aborted(journal))
1961 return -EIO;
1962
Jan Karaa78bb112012-03-13 15:43:04 -04001963 mutex_lock(&journal->j_checkpoint_mutex);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001964 if (!err) {
1965 err = jbd2_cleanup_journal_tail(journal);
1966 if (err < 0) {
1967 mutex_unlock(&journal->j_checkpoint_mutex);
1968 goto out;
1969 }
1970 err = 0;
1971 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001972
1973 /* Finally, mark the journal as really needing no recovery.
1974 * This sets s_start==0 in the underlying superblock, which is
1975 * the magic code for a fully-recovered superblock. Any future
1976 * commits of data to the journal will restore the current
1977 * s_start value. */
Jan Kara24bcc892012-03-13 15:41:04 -04001978 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04001979 mutex_unlock(&journal->j_checkpoint_mutex);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001980 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001981 J_ASSERT(!journal->j_running_transaction);
1982 J_ASSERT(!journal->j_committing_transaction);
1983 J_ASSERT(!journal->j_checkpoint_transactions);
1984 J_ASSERT(journal->j_head == journal->j_tail);
1985 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001986 write_unlock(&journal->j_state_lock);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001987out:
1988 return err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001989}
1990
1991/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001992 * int jbd2_journal_wipe() - Wipe journal contents
Dave Kleikamp470decc2006-10-11 01:20:57 -07001993 * @journal: Journal to act on.
1994 * @write: flag (see below)
1995 *
1996 * Wipe out all of the contents of a journal, safely. This will produce
1997 * a warning if the journal contains any valid recovery information.
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001998 * Must be called between journal_init_*() and jbd2_journal_load().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001999 *
2000 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2001 * we merely suppress recovery.
2002 */
2003
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002004int jbd2_journal_wipe(journal_t *journal, int write)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002005{
Dave Kleikamp470decc2006-10-11 01:20:57 -07002006 int err = 0;
2007
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002008 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
Dave Kleikamp470decc2006-10-11 01:20:57 -07002009
2010 err = load_superblock(journal);
2011 if (err)
2012 return err;
2013
Dave Kleikamp470decc2006-10-11 01:20:57 -07002014 if (!journal->j_tail)
2015 goto no_recovery;
2016
Eryu Guanf2a44522011-11-01 19:09:18 -04002017 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002018 write ? "Clearing" : "Ignoring");
2019
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002020 err = jbd2_journal_skip_recovery(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04002021 if (write) {
2022 /* Lock to make assertions happy... */
2023 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara24bcc892012-03-13 15:41:04 -04002024 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04002025 mutex_unlock(&journal->j_checkpoint_mutex);
2026 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002027
2028 no_recovery:
2029 return err;
2030}
2031
2032/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07002033 * Journal abort has very specific semantics, which we describe
2034 * for journal abort.
2035 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002036 * Two internal functions, which provide abort to the jbd layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07002037 * itself are here.
2038 */
2039
2040/*
2041 * Quick version for internal journal use (doesn't lock the journal).
2042 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2043 * and don't attempt to make any other journal updates.
2044 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002045void __jbd2_journal_abort_hard(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002046{
2047 transaction_t *transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002048
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002049 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002050 return;
2051
2052 printk(KERN_ERR "Aborting journal on device %s.\n",
Theodore Ts'o05496762008-09-16 14:36:17 -04002053 journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002054
Theodore Ts'oa931da62010-08-03 21:35:12 -04002055 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002056 journal->j_flags |= JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002057 transaction = journal->j_running_transaction;
2058 if (transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002059 __jbd2_log_start_commit(journal, transaction->t_tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002060 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002061}
2062
2063/* Soft abort: record the abort error status in the journal superblock,
2064 * but don't do any other IO. */
2065static void __journal_abort_soft (journal_t *journal, int errno)
2066{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002067 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002068 return;
2069
2070 if (!journal->j_errno)
2071 journal->j_errno = errno;
2072
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002073 __jbd2_journal_abort_hard(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002074
Daeho Jeong4327ba52015-10-18 17:02:56 -04002075 if (errno) {
Jan Kara24bcc892012-03-13 15:41:04 -04002076 jbd2_journal_update_sb_errno(journal);
Daeho Jeong4327ba52015-10-18 17:02:56 -04002077 write_lock(&journal->j_state_lock);
2078 journal->j_flags |= JBD2_REC_ERR;
2079 write_unlock(&journal->j_state_lock);
2080 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002081}
2082
2083/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002084 * void jbd2_journal_abort () - Shutdown the journal immediately.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002085 * @journal: the journal to shutdown.
2086 * @errno: an error number to record in the journal indicating
2087 * the reason for the shutdown.
2088 *
2089 * Perform a complete, immediate shutdown of the ENTIRE
2090 * journal (not of a single transaction). This operation cannot be
2091 * undone without closing and reopening the journal.
2092 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002093 * The jbd2_journal_abort function is intended to support higher level error
Dave Kleikamp470decc2006-10-11 01:20:57 -07002094 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2095 * mode.
2096 *
2097 * Journal abort has very specific semantics. Any existing dirty,
2098 * unjournaled buffers in the main filesystem will still be written to
2099 * disk by bdflush, but the journaling mechanism will be suspended
2100 * immediately and no further transaction commits will be honoured.
2101 *
2102 * Any dirty, journaled buffers will be written back to disk without
2103 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2104 * filesystem, but we _do_ attempt to leave as much data as possible
2105 * behind for fsck to use for cleanup.
2106 *
2107 * Any attempt to get a new transaction handle on a journal which is in
2108 * ABORT state will just result in an -EROFS error return. A
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002109 * jbd2_journal_stop on an existing handle will return -EIO if we have
Dave Kleikamp470decc2006-10-11 01:20:57 -07002110 * entered abort state during the update.
2111 *
2112 * Recursive transactions are not disturbed by journal abort until the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002113 * final jbd2_journal_stop, which will receive the -EIO error.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002114 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002115 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
Dave Kleikamp470decc2006-10-11 01:20:57 -07002116 * which will be recorded (if possible) in the journal superblock. This
2117 * allows a client to record failure conditions in the middle of a
2118 * transaction without having to complete the transaction to record the
2119 * failure to disk. ext3_error, for example, now uses this
2120 * functionality.
2121 *
2122 * Errors which originate from within the journaling layer will NOT
2123 * supply an errno; a null errno implies that absolutely no further
2124 * writes are done to the journal (unless there are any already in
2125 * progress).
2126 *
2127 */
2128
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002129void jbd2_journal_abort(journal_t *journal, int errno)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002130{
2131 __journal_abort_soft(journal, errno);
2132}
2133
2134/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002135 * int jbd2_journal_errno () - returns the journal's error state.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002136 * @journal: journal to examine.
2137 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002138 * This is the errno number set with jbd2_journal_abort(), the last
Dave Kleikamp470decc2006-10-11 01:20:57 -07002139 * time the journal was mounted - if the journal was stopped
2140 * without calling abort this will be 0.
2141 *
2142 * If the journal has been aborted on this mount time -EROFS will
2143 * be returned.
2144 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002145int jbd2_journal_errno(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002146{
2147 int err;
2148
Theodore Ts'oa931da62010-08-03 21:35:12 -04002149 read_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002150 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002151 err = -EROFS;
2152 else
2153 err = journal->j_errno;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002154 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002155 return err;
2156}
2157
2158/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002159 * int jbd2_journal_clear_err () - clears the journal's error state
Dave Kleikamp470decc2006-10-11 01:20:57 -07002160 * @journal: journal to act on.
2161 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002162 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07002163 * mode.
2164 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002165int jbd2_journal_clear_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002166{
2167 int err = 0;
2168
Theodore Ts'oa931da62010-08-03 21:35:12 -04002169 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002170 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002171 err = -EROFS;
2172 else
2173 journal->j_errno = 0;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002174 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002175 return err;
2176}
2177
2178/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002179 * void jbd2_journal_ack_err() - Ack journal err.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002180 * @journal: journal to act on.
2181 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002182 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07002183 * mode.
2184 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002185void jbd2_journal_ack_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002186{
Theodore Ts'oa931da62010-08-03 21:35:12 -04002187 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002188 if (journal->j_errno)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002189 journal->j_flags |= JBD2_ACK_ERR;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002190 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002191}
2192
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002193int jbd2_journal_blocks_per_page(struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002194{
2195 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2196}
2197
2198/*
Zach Brownb517bea2006-10-11 01:21:08 -07002199 * helper functions to deal with 32 or 64bit block numbers.
2200 */
2201size_t journal_tag_bytes(journal_t *journal)
2202{
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002203 size_t sz;
2204
Darrick J. Wong56316a02015-10-17 16:18:45 -04002205 if (jbd2_has_feature_csum3(journal))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002206 return sizeof(journal_block_tag3_t);
2207
2208 sz = sizeof(journal_block_tag_t);
Darrick J. Wongc3900872012-05-27 08:12:12 -04002209
Darrick J. Wong56316a02015-10-17 16:18:45 -04002210 if (jbd2_has_feature_csum2(journal))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002211 sz += sizeof(__u16);
Darrick J. Wongc3900872012-05-27 08:12:12 -04002212
Darrick J. Wong56316a02015-10-17 16:18:45 -04002213 if (jbd2_has_feature_64bit(journal))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002214 return sz;
Zach Brownb517bea2006-10-11 01:21:08 -07002215 else
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002216 return sz - sizeof(__u32);
Zach Brownb517bea2006-10-11 01:21:08 -07002217}
2218
2219/*
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002220 * JBD memory management
2221 *
2222 * These functions are used to allocate block-sized chunks of memory
2223 * used for making copies of buffer_head data. Very often it will be
2224 * page-sized chunks of data, but sometimes it will be in
2225 * sub-page-size chunks. (For example, 16k pages on Power systems
2226 * with a 4k block file system.) For blocks smaller than a page, we
2227 * use a SLAB allocator. There are slab caches for each block size,
2228 * which are allocated at mount time, if necessary, and we only free
2229 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2230 * this reason we don't need to a mutex to protect access to
2231 * jbd2_slab[] allocating or releasing memory; only in
2232 * jbd2_journal_create_slab().
2233 */
2234#define JBD2_MAX_SLABS 8
2235static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002236
2237static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2238 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2239 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2240};
2241
2242
2243static void jbd2_journal_destroy_slabs(void)
2244{
2245 int i;
2246
2247 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2248 if (jbd2_slab[i])
2249 kmem_cache_destroy(jbd2_slab[i]);
2250 jbd2_slab[i] = NULL;
2251 }
2252}
2253
2254static int jbd2_journal_create_slab(size_t size)
2255{
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002256 static DEFINE_MUTEX(jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002257 int i = order_base_2(size) - 10;
2258 size_t slab_size;
2259
2260 if (size == PAGE_SIZE)
2261 return 0;
2262
2263 if (i >= JBD2_MAX_SLABS)
2264 return -EINVAL;
2265
2266 if (unlikely(i < 0))
2267 i = 0;
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002268 mutex_lock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002269 if (jbd2_slab[i]) {
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002270 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002271 return 0; /* Already created */
2272 }
2273
2274 slab_size = 1 << (i+10);
2275 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2276 slab_size, 0, NULL);
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002277 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002278 if (!jbd2_slab[i]) {
2279 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2280 return -ENOMEM;
2281 }
2282 return 0;
2283}
2284
2285static struct kmem_cache *get_slab(size_t size)
2286{
2287 int i = order_base_2(size) - 10;
2288
2289 BUG_ON(i >= JBD2_MAX_SLABS);
2290 if (unlikely(i < 0))
2291 i = 0;
Bill Pemberton8ac97b72010-04-30 09:34:31 -04002292 BUG_ON(jbd2_slab[i] == NULL);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002293 return jbd2_slab[i];
2294}
2295
2296void *jbd2_alloc(size_t size, gfp_t flags)
2297{
2298 void *ptr;
2299
2300 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2301
2302 flags |= __GFP_REPEAT;
2303 if (size == PAGE_SIZE)
2304 ptr = (void *)__get_free_pages(flags, 0);
2305 else if (size > PAGE_SIZE) {
2306 int order = get_order(size);
2307
2308 if (order < 3)
2309 ptr = (void *)__get_free_pages(flags, order);
2310 else
2311 ptr = vmalloc(size);
2312 } else
2313 ptr = kmem_cache_alloc(get_slab(size), flags);
2314
2315 /* Check alignment; SLUB has gotten this wrong in the past,
2316 * and this can lead to user data corruption! */
2317 BUG_ON(((unsigned long) ptr) & (size-1));
2318
2319 return ptr;
2320}
2321
2322void jbd2_free(void *ptr, size_t size)
2323{
2324 if (size == PAGE_SIZE) {
2325 free_pages((unsigned long)ptr, 0);
2326 return;
2327 }
2328 if (size > PAGE_SIZE) {
2329 int order = get_order(size);
2330
2331 if (order < 3)
2332 free_pages((unsigned long)ptr, order);
2333 else
2334 vfree(ptr);
2335 return;
2336 }
2337 kmem_cache_free(get_slab(size), ptr);
2338};
2339
2340/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07002341 * Journal_head storage management
2342 */
Christoph Lametere18b8902006-12-06 20:33:20 -08002343static struct kmem_cache *jbd2_journal_head_cache;
Jose R. Santose23291b2007-07-18 08:57:06 -04002344#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002345static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2346#endif
2347
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002348static int jbd2_journal_init_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002349{
2350 int retval;
2351
Al Viro1076d172008-03-29 03:07:18 +00002352 J_ASSERT(jbd2_journal_head_cache == NULL);
Johann Lombardia920e942006-10-11 01:21:00 -07002353 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002354 sizeof(struct journal_head),
2355 0, /* offset */
Jan Karade92c8c2015-06-08 12:46:37 -04002356 SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
Paul Mundt20c2df82007-07-20 10:11:58 +09002357 NULL); /* ctor */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002358 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00002359 if (!jbd2_journal_head_cache) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002360 retval = -ENOMEM;
Eryu Guanf2a44522011-11-01 19:09:18 -04002361 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002362 }
2363 return retval;
2364}
2365
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002366static void jbd2_journal_destroy_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002367{
Duane Griffin8a9362e2008-04-17 10:38:59 -04002368 if (jbd2_journal_head_cache) {
2369 kmem_cache_destroy(jbd2_journal_head_cache);
2370 jbd2_journal_head_cache = NULL;
2371 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002372}
2373
2374/*
2375 * journal_head splicing and dicing
2376 */
2377static struct journal_head *journal_alloc_journal_head(void)
2378{
2379 struct journal_head *ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002380
Jose R. Santose23291b2007-07-18 08:57:06 -04002381#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002382 atomic_inc(&nr_journal_heads);
2383#endif
Zheng Liu5d9cf9c2013-05-28 07:27:11 -04002384 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
Al Viro1076d172008-03-29 03:07:18 +00002385 if (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002386 jbd_debug(1, "out of memory for journal_head\n");
Theodore Ts'o670be5a2010-12-17 10:44:16 -05002387 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
Michal Hocko7b506b12015-06-15 15:45:58 -04002388 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2389 GFP_NOFS | __GFP_NOFAIL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002390 }
2391 return ret;
2392}
2393
2394static void journal_free_journal_head(struct journal_head *jh)
2395{
Jose R. Santose23291b2007-07-18 08:57:06 -04002396#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002397 atomic_dec(&nr_journal_heads);
Mingming Caocd02ff02007-10-16 18:38:25 -04002398 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -07002399#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002400 kmem_cache_free(jbd2_journal_head_cache, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002401}
2402
2403/*
2404 * A journal_head is attached to a buffer_head whenever JBD has an
2405 * interest in the buffer.
2406 *
2407 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2408 * is set. This bit is tested in core kernel code where we need to take
2409 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2410 * there.
2411 *
2412 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2413 *
2414 * When a buffer has its BH_JBD bit set it is immune from being released by
2415 * core kernel code, mainly via ->b_count.
2416 *
Jan Karade1b7942011-06-13 15:38:22 -04002417 * A journal_head is detached from its buffer_head when the journal_head's
2418 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2419 * transaction (b_cp_transaction) hold their references to b_jcount.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002420 *
2421 * Various places in the kernel want to attach a journal_head to a buffer_head
2422 * _before_ attaching the journal_head to a transaction. To protect the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002423 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
Dave Kleikamp470decc2006-10-11 01:20:57 -07002424 * journal_head's b_jcount refcount by one. The caller must call
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002425 * jbd2_journal_put_journal_head() to undo this.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002426 *
2427 * So the typical usage would be:
2428 *
2429 * (Attach a journal_head if needed. Increments b_jcount)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002430 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002431 * ...
Jan Karade1b7942011-06-13 15:38:22 -04002432 * (Get another reference for transaction)
2433 * jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002434 * jh->b_transaction = xxx;
Jan Karade1b7942011-06-13 15:38:22 -04002435 * (Put original reference)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002436 * jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002437 */
2438
2439/*
2440 * Give a buffer_head a journal_head.
2441 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07002442 * May sleep.
2443 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002444struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002445{
2446 struct journal_head *jh;
2447 struct journal_head *new_jh = NULL;
2448
2449repeat:
Zheng Liu5d9cf9c2013-05-28 07:27:11 -04002450 if (!buffer_jbd(bh))
Dave Kleikamp470decc2006-10-11 01:20:57 -07002451 new_jh = journal_alloc_journal_head();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002452
2453 jbd_lock_bh_journal_head(bh);
2454 if (buffer_jbd(bh)) {
2455 jh = bh2jh(bh);
2456 } else {
2457 J_ASSERT_BH(bh,
2458 (atomic_read(&bh->b_count) > 0) ||
2459 (bh->b_page && bh->b_page->mapping));
2460
2461 if (!new_jh) {
2462 jbd_unlock_bh_journal_head(bh);
2463 goto repeat;
2464 }
2465
2466 jh = new_jh;
2467 new_jh = NULL; /* We consumed it */
2468 set_buffer_jbd(bh);
2469 bh->b_private = jh;
2470 jh->b_bh = bh;
2471 get_bh(bh);
2472 BUFFER_TRACE(bh, "added journal_head");
2473 }
2474 jh->b_jcount++;
2475 jbd_unlock_bh_journal_head(bh);
2476 if (new_jh)
2477 journal_free_journal_head(new_jh);
2478 return bh->b_private;
2479}
2480
2481/*
2482 * Grab a ref against this buffer_head's journal_head. If it ended up not
2483 * having a journal_head, return NULL
2484 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002485struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002486{
2487 struct journal_head *jh = NULL;
2488
2489 jbd_lock_bh_journal_head(bh);
2490 if (buffer_jbd(bh)) {
2491 jh = bh2jh(bh);
2492 jh->b_jcount++;
2493 }
2494 jbd_unlock_bh_journal_head(bh);
2495 return jh;
2496}
2497
2498static void __journal_remove_journal_head(struct buffer_head *bh)
2499{
2500 struct journal_head *jh = bh2jh(bh);
2501
2502 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karade1b7942011-06-13 15:38:22 -04002503 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2504 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2505 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2506 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2507 J_ASSERT_BH(bh, buffer_jbd(bh));
2508 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2509 BUFFER_TRACE(bh, "remove journal_head");
2510 if (jh->b_frozen_data) {
2511 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2512 jbd2_free(jh->b_frozen_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002513 }
Jan Karade1b7942011-06-13 15:38:22 -04002514 if (jh->b_committed_data) {
2515 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2516 jbd2_free(jh->b_committed_data, bh->b_size);
2517 }
2518 bh->b_private = NULL;
2519 jh->b_bh = NULL; /* debug, really */
2520 clear_buffer_jbd(bh);
2521 journal_free_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002522}
2523
2524/*
Jan Karade1b7942011-06-13 15:38:22 -04002525 * Drop a reference on the passed journal_head. If it fell to zero then
Dave Kleikamp470decc2006-10-11 01:20:57 -07002526 * release the journal_head from the buffer_head.
2527 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002528void jbd2_journal_put_journal_head(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002529{
2530 struct buffer_head *bh = jh2bh(jh);
2531
2532 jbd_lock_bh_journal_head(bh);
2533 J_ASSERT_JH(jh, jh->b_jcount > 0);
2534 --jh->b_jcount;
Jan Karade1b7942011-06-13 15:38:22 -04002535 if (!jh->b_jcount) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002536 __journal_remove_journal_head(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002537 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002538 __brelse(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002539 } else
2540 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002541}
2542
2543/*
Jan Karac851ed52008-07-11 19:27:31 -04002544 * Initialize jbd inode head
2545 */
2546void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2547{
2548 jinode->i_transaction = NULL;
2549 jinode->i_next_transaction = NULL;
2550 jinode->i_vfs_inode = inode;
2551 jinode->i_flags = 0;
2552 INIT_LIST_HEAD(&jinode->i_list);
2553}
2554
2555/*
2556 * Function to be called before we start removing inode from memory (i.e.,
2557 * clear_inode() is a fine place to be called from). It removes inode from
2558 * transaction's lists.
2559 */
2560void jbd2_journal_release_jbd_inode(journal_t *journal,
2561 struct jbd2_inode *jinode)
2562{
Jan Karac851ed52008-07-11 19:27:31 -04002563 if (!journal)
2564 return;
2565restart:
2566 spin_lock(&journal->j_list_lock);
2567 /* Is commit writing out inode - we have to wait */
Brian King39e3ac22010-10-27 21:25:12 -04002568 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
Jan Karac851ed52008-07-11 19:27:31 -04002569 wait_queue_head_t *wq;
2570 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2571 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2572 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2573 spin_unlock(&journal->j_list_lock);
2574 schedule();
2575 finish_wait(wq, &wait.wait);
2576 goto restart;
2577 }
2578
Jan Karac851ed52008-07-11 19:27:31 -04002579 if (jinode->i_transaction) {
2580 list_del(&jinode->i_list);
2581 jinode->i_transaction = NULL;
2582 }
2583 spin_unlock(&journal->j_list_lock);
2584}
2585
Dave Kleikamp470decc2006-10-11 01:20:57 -07002586
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002587#ifdef CONFIG_PROC_FS
2588
2589#define JBD2_STATS_PROC_NAME "fs/jbd2"
2590
2591static void __init jbd2_create_jbd_stats_proc_entry(void)
2592{
2593 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2594}
2595
2596static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2597{
2598 if (proc_jbd2_stats)
2599 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2600}
2601
2602#else
2603
2604#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2605#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2606
2607#endif
2608
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002609struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002610
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002611static int __init jbd2_journal_init_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002612{
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002613 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002614 if (jbd2_handle_cache == NULL) {
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002615 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2616 return -ENOMEM;
2617 }
2618 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2619 if (jbd2_inode_cache == NULL) {
2620 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2621 kmem_cache_destroy(jbd2_handle_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002622 return -ENOMEM;
2623 }
2624 return 0;
2625}
2626
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002627static void jbd2_journal_destroy_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002628{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002629 if (jbd2_handle_cache)
2630 kmem_cache_destroy(jbd2_handle_cache);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002631 if (jbd2_inode_cache)
2632 kmem_cache_destroy(jbd2_inode_cache);
2633
Dave Kleikamp470decc2006-10-11 01:20:57 -07002634}
2635
2636/*
2637 * Module startup and shutdown
2638 */
2639
2640static int __init journal_init_caches(void)
2641{
2642 int ret;
2643
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002644 ret = jbd2_journal_init_revoke_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002645 if (ret == 0)
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002646 ret = jbd2_journal_init_journal_head_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002647 if (ret == 0)
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002648 ret = jbd2_journal_init_handle_cache();
Yongqiang Yang0c2022e2012-02-20 17:53:02 -05002649 if (ret == 0)
2650 ret = jbd2_journal_init_transaction_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002651 return ret;
2652}
2653
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002654static void jbd2_journal_destroy_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002655{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002656 jbd2_journal_destroy_revoke_caches();
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002657 jbd2_journal_destroy_journal_head_cache();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002658 jbd2_journal_destroy_handle_cache();
Yongqiang Yang0c2022e2012-02-20 17:53:02 -05002659 jbd2_journal_destroy_transaction_cache();
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002660 jbd2_journal_destroy_slabs();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002661}
2662
2663static int __init journal_init(void)
2664{
2665 int ret;
2666
2667 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2668
2669 ret = journal_init_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002670 if (ret == 0) {
Duane Griffin620de4e2008-04-29 22:02:47 -04002671 jbd2_create_jbd_stats_proc_entry();
2672 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002673 jbd2_journal_destroy_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002674 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002675 return ret;
2676}
2677
2678static void __exit journal_exit(void)
2679{
Jose R. Santose23291b2007-07-18 08:57:06 -04002680#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002681 int n = atomic_read(&nr_journal_heads);
2682 if (n)
Jan Kara75685072013-12-08 21:13:59 -05002683 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002684#endif
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002685 jbd2_remove_jbd_stats_proc_entry();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002686 jbd2_journal_destroy_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002687}
2688
Dave Kleikamp470decc2006-10-11 01:20:57 -07002689MODULE_LICENSE("GPL");
2690module_init(journal_init);
2691module_exit(journal_exit);
2692