blob: 19d74d86d99cc630aec2c0ca8f7bf4ef619e3aa3 [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. Wongdb9ee222014-08-27 18:40:07 -0400127 if (!jbd2_journal_has_csum_v2or3(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
Theodore Ts'o47def822010-07-27 11:56:05 -0400374retry_alloc:
375 new_bh = alloc_buffer_head(GFP_NOFS);
376 if (!new_bh) {
377 /*
378 * Failure is not an option, but __GFP_NOFAIL is going
379 * away; so we retry ourselves here.
380 */
381 congestion_wait(BLK_RW_ASYNC, HZ/50);
382 goto retry_alloc;
383 }
384
dingdinghua96577c42009-07-13 17:55:35 -0400385 /* keep subsequent assertions sane */
dingdinghua96577c42009-07-13 17:55:35 -0400386 atomic_set(&new_bh->b_count, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700387
Jan Karaf5113ef2013-06-04 12:01:45 -0400388 jbd_lock_bh_state(bh_in);
389repeat:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700390 /*
391 * If a new transaction has already done a buffer copy-out, then
392 * we use that version of the data for the commit.
393 */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700394 if (jh_in->b_frozen_data) {
395 done_copy_out = 1;
396 new_page = virt_to_page(jh_in->b_frozen_data);
397 new_offset = offset_in_page(jh_in->b_frozen_data);
398 } else {
399 new_page = jh2bh(jh_in)->b_page;
400 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
401 }
402
Cong Wang303a8f22011-11-25 23:14:31 +0800403 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700404 /*
Jan Kara13ceef02010-07-14 07:56:33 +0200405 * Fire data frozen trigger if data already wasn't frozen. Do this
406 * before checking for escaping, as the trigger may modify the magic
407 * offset. If a copy-out happens afterwards, it will have the correct
408 * data in the buffer.
Joel Beckere06c8222008-09-11 15:35:47 -0700409 */
Jan Kara13ceef02010-07-14 07:56:33 +0200410 if (!done_copy_out)
411 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
412 jh_in->b_triggers);
Joel Beckere06c8222008-09-11 15:35:47 -0700413
414 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700415 * Check for escaping
416 */
417 if (*((__be32 *)(mapped_data + new_offset)) ==
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700418 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700419 need_copy_out = 1;
420 do_escape = 1;
421 }
Cong Wang303a8f22011-11-25 23:14:31 +0800422 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700423
424 /*
425 * Do we need to do a data copy?
426 */
427 if (need_copy_out && !done_copy_out) {
428 char *tmp;
429
430 jbd_unlock_bh_state(bh_in);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400431 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500432 if (!tmp) {
Jan Karaf5113ef2013-06-04 12:01:45 -0400433 brelse(new_bh);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500434 return -ENOMEM;
435 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700436 jbd_lock_bh_state(bh_in);
437 if (jh_in->b_frozen_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400438 jbd2_free(tmp, bh_in->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700439 goto repeat;
440 }
441
442 jh_in->b_frozen_data = tmp;
Cong Wang303a8f22011-11-25 23:14:31 +0800443 mapped_data = kmap_atomic(new_page);
Jan Karaf5113ef2013-06-04 12:01:45 -0400444 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
Cong Wang303a8f22011-11-25 23:14:31 +0800445 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700446
447 new_page = virt_to_page(tmp);
448 new_offset = offset_in_page(tmp);
449 done_copy_out = 1;
Joel Beckere06c8222008-09-11 15:35:47 -0700450
451 /*
452 * This isn't strictly necessary, as we're using frozen
453 * data for the escaping, but it keeps consistency with
454 * b_frozen_data usage.
455 */
456 jh_in->b_frozen_triggers = jh_in->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700457 }
458
459 /*
460 * Did we need to do an escaping? Now we've done all the
461 * copying, we can finally do so.
462 */
463 if (do_escape) {
Cong Wang303a8f22011-11-25 23:14:31 +0800464 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700465 *((unsigned int *)(mapped_data + new_offset)) = 0;
Cong Wang303a8f22011-11-25 23:14:31 +0800466 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700467 }
468
Dave Kleikamp470decc2006-10-11 01:20:57 -0700469 set_bh_page(new_bh, new_page, new_offset);
Jan Karaf5113ef2013-06-04 12:01:45 -0400470 new_bh->b_size = bh_in->b_size;
471 new_bh->b_bdev = journal->j_dev;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700472 new_bh->b_blocknr = blocknr;
Jan Karab34090e2013-06-04 12:08:56 -0400473 new_bh->b_private = bh_in;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700474 set_buffer_mapped(new_bh);
475 set_buffer_dirty(new_bh);
476
Jan Karaf5113ef2013-06-04 12:01:45 -0400477 *bh_out = new_bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700478
479 /*
480 * The to-be-written buffer needs to get moved to the io queue,
481 * and the original buffer whose contents we are shadowing or
482 * copying is moved to the transaction's shadow queue.
483 */
484 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghua96577c42009-07-13 17:55:35 -0400485 spin_lock(&journal->j_list_lock);
486 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
487 spin_unlock(&journal->j_list_lock);
Jan Karab34090e2013-06-04 12:08:56 -0400488 set_buffer_shadow(bh_in);
dingdinghua96577c42009-07-13 17:55:35 -0400489 jbd_unlock_bh_state(bh_in);
490
Dave Kleikamp470decc2006-10-11 01:20:57 -0700491 return do_escape | (done_copy_out << 1);
492}
493
494/*
495 * Allocation code for the journal file. Manage the space left in the
496 * journal, so that we can begin checkpointing when appropriate.
497 */
498
499/*
Theodore Ts'oe4471832011-02-12 08:18:24 -0500500 * Called with j_state_lock locked for writing.
501 * Returns true if a transaction commit was started.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700502 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700503int __jbd2_log_start_commit(journal_t *journal, tid_t target)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700504{
Eric Sandeene7b04ac2013-01-30 00:39:28 -0500505 /* Return if the txn has already requested to be committed */
506 if (journal->j_commit_request == target)
507 return 0;
508
Dave Kleikamp470decc2006-10-11 01:20:57 -0700509 /*
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400510 * The only transaction we can possibly wait upon is the
511 * currently running transaction (if it exists). Otherwise,
512 * the target tid must be an old one.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700513 */
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400514 if (journal->j_running_transaction &&
515 journal->j_running_transaction->t_tid == target) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700516 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200517 * We want a new commit: OK, mark the request and wakeup the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700518 * commit thread. We do _not_ do the commit ourselves.
519 */
520
521 journal->j_commit_request = target;
Eryu Guanf2a44522011-11-01 19:09:18 -0400522 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700523 journal->j_commit_request,
524 journal->j_commit_sequence);
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500525 journal->j_running_transaction->t_requested = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700526 wake_up(&journal->j_wait_commit);
527 return 1;
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400528 } else if (!tid_geq(journal->j_commit_request, target))
529 /* This should never happen, but if it does, preserve
530 the evidence before kjournald goes into a loop and
531 increments j_commit_sequence beyond all recognition. */
Eryu Guanf2a44522011-11-01 19:09:18 -0400532 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
Theodore Ts'o1be2add2011-05-08 19:37:54 -0400533 journal->j_commit_request,
534 journal->j_commit_sequence,
535 target, journal->j_running_transaction ?
536 journal->j_running_transaction->t_tid : 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700537 return 0;
538}
539
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700540int jbd2_log_start_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700541{
542 int ret;
543
Theodore Ts'oa931da62010-08-03 21:35:12 -0400544 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700545 ret = __jbd2_log_start_commit(journal, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400546 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700547 return ret;
548}
549
550/*
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400551 * Force and wait any uncommitted transactions. We can only force the running
552 * transaction if we don't have an active handle, otherwise, we will deadlock.
553 * Returns: <0 in case of error,
554 * 0 if nothing to commit,
555 * 1 if transaction was successfully committed.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700556 */
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400557static int __jbd2_journal_force_commit(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700558{
559 transaction_t *transaction = NULL;
560 tid_t tid;
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400561 int need_to_start = 0, ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700562
Theodore Ts'oa931da62010-08-03 21:35:12 -0400563 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700564 if (journal->j_running_transaction && !current->journal_info) {
565 transaction = journal->j_running_transaction;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500566 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
567 need_to_start = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700568 } else if (journal->j_committing_transaction)
569 transaction = journal->j_committing_transaction;
570
571 if (!transaction) {
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400572 /* Nothing to commit */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400573 read_unlock(&journal->j_state_lock);
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400574 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700575 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700576 tid = transaction->t_tid;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400577 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500578 if (need_to_start)
579 jbd2_log_start_commit(journal, tid);
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400580 ret = jbd2_log_wait_commit(journal, tid);
581 if (!ret)
582 ret = 1;
583
584 return ret;
585}
586
587/**
588 * Force and wait upon a commit if the calling process is not within
589 * transaction. This is used for forcing out undo-protected data which contains
590 * bitmaps, when the fs is running out of space.
591 *
592 * @journal: journal to force
593 * Returns true if progress was made.
594 */
595int jbd2_journal_force_commit_nested(journal_t *journal)
596{
597 int ret;
598
599 ret = __jbd2_journal_force_commit(journal);
600 return ret > 0;
601}
602
603/**
604 * int journal_force_commit() - force any uncommitted transactions
605 * @journal: journal to force
606 *
607 * Caller want unconditional commit. We can only force the running transaction
608 * if we don't have an active handle, otherwise, we will deadlock.
609 */
610int jbd2_journal_force_commit(journal_t *journal)
611{
612 int ret;
613
614 J_ASSERT(!current->journal_info);
615 ret = __jbd2_journal_force_commit(journal);
616 if (ret > 0)
617 ret = 0;
618 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700619}
620
621/*
622 * Start a commit of the current running transaction (if any). Returns true
Jan Karac88ccea2009-02-10 11:27:46 -0500623 * if a transaction is going to be committed (or is currently already
624 * committing), and fills its tid in at *ptid
Dave Kleikamp470decc2006-10-11 01:20:57 -0700625 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700626int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700627{
628 int ret = 0;
629
Theodore Ts'oa931da62010-08-03 21:35:12 -0400630 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700631 if (journal->j_running_transaction) {
632 tid_t tid = journal->j_running_transaction->t_tid;
633
Jan Karac88ccea2009-02-10 11:27:46 -0500634 __jbd2_log_start_commit(journal, tid);
635 /* There's a running transaction and we've just made sure
636 * it's commit has been scheduled. */
637 if (ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700638 *ptid = tid;
Jan Karac88ccea2009-02-10 11:27:46 -0500639 ret = 1;
640 } else if (journal->j_committing_transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700641 /*
Artem Bityutskiy12810ad2012-07-25 18:12:07 +0300642 * If commit has been started, then we have to wait for
643 * completion of that transaction.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700644 */
Jan Karac88ccea2009-02-10 11:27:46 -0500645 if (ptid)
646 *ptid = journal->j_committing_transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700647 ret = 1;
648 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400649 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700650 return ret;
651}
652
653/*
Jan Karabbd2be32011-05-24 11:59:18 -0400654 * Return 1 if a given transaction has not yet sent barrier request
655 * connected with a transaction commit. If 0 is returned, transaction
656 * may or may not have sent the barrier. Used to avoid sending barrier
657 * twice in common cases.
658 */
659int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
660{
661 int ret = 0;
662 transaction_t *commit_trans;
663
664 if (!(journal->j_flags & JBD2_BARRIER))
665 return 0;
666 read_lock(&journal->j_state_lock);
667 /* Transaction already committed? */
668 if (tid_geq(journal->j_commit_sequence, tid))
669 goto out;
670 commit_trans = journal->j_committing_transaction;
671 if (!commit_trans || commit_trans->t_tid != tid) {
672 ret = 1;
673 goto out;
674 }
675 /*
676 * Transaction is being committed and we already proceeded to
677 * submitting a flush to fs partition?
678 */
679 if (journal->j_fs_dev != journal->j_dev) {
680 if (!commit_trans->t_need_data_flush ||
681 commit_trans->t_state >= T_COMMIT_DFLUSH)
682 goto out;
683 } else {
684 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
685 goto out;
686 }
687 ret = 1;
688out:
689 read_unlock(&journal->j_state_lock);
690 return ret;
691}
692EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
693
694/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700695 * Wait for a specified commit to complete.
696 * The caller may not hold the journal lock.
697 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700698int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700699{
700 int err = 0;
701
Theodore Ts'oa931da62010-08-03 21:35:12 -0400702 read_lock(&journal->j_state_lock);
Jose R. Santose23291b2007-07-18 08:57:06 -0400703#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -0700704 if (!tid_geq(journal->j_commit_request, tid)) {
Jan Kara75685072013-12-08 21:13:59 -0500705 printk(KERN_ERR
Dave Kleikamp470decc2006-10-11 01:20:57 -0700706 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400707 __func__, journal->j_commit_request, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700708 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700709#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -0700710 while (tid_gt(tid, journal->j_commit_sequence)) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400711 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700712 tid, journal->j_commit_sequence);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400713 read_unlock(&journal->j_state_lock);
Theodore Ts'o3469a322014-03-08 19:11:36 -0500714 wake_up(&journal->j_wait_commit);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700715 wait_event(journal->j_wait_done_commit,
716 !tid_gt(tid, journal->j_commit_sequence));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400717 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700718 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400719 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700720
Jan Kara75685072013-12-08 21:13:59 -0500721 if (unlikely(is_journal_aborted(journal)))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700722 err = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700723 return err;
724}
725
726/*
Theodore Ts'od76a3a772013-04-03 22:02:52 -0400727 * When this function returns the transaction corresponding to tid
728 * will be completed. If the transaction has currently running, start
729 * committing that transaction before waiting for it to complete. If
730 * the transaction id is stale, it is by definition already completed,
731 * so just return SUCCESS.
732 */
733int jbd2_complete_transaction(journal_t *journal, tid_t tid)
734{
735 int need_to_wait = 1;
736
737 read_lock(&journal->j_state_lock);
738 if (journal->j_running_transaction &&
739 journal->j_running_transaction->t_tid == tid) {
740 if (journal->j_commit_request != tid) {
741 /* transaction not yet started, so request it */
742 read_unlock(&journal->j_state_lock);
743 jbd2_log_start_commit(journal, tid);
744 goto wait_commit;
745 }
746 } else if (!(journal->j_committing_transaction &&
747 journal->j_committing_transaction->t_tid == tid))
748 need_to_wait = 0;
749 read_unlock(&journal->j_state_lock);
750 if (!need_to_wait)
751 return 0;
752wait_commit:
753 return jbd2_log_wait_commit(journal, tid);
754}
755EXPORT_SYMBOL(jbd2_complete_transaction);
756
757/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700758 * Log buffer allocation routines:
759 */
760
Mingming Cao18eba7a2006-10-11 01:21:13 -0700761int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700762{
763 unsigned long blocknr;
764
Theodore Ts'oa931da62010-08-03 21:35:12 -0400765 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700766 J_ASSERT(journal->j_free > 1);
767
768 blocknr = journal->j_head;
769 journal->j_head++;
770 journal->j_free--;
771 if (journal->j_head == journal->j_last)
772 journal->j_head = journal->j_first;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400773 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700774 return jbd2_journal_bmap(journal, blocknr, retp);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700775}
776
777/*
778 * Conversion of logical to physical block numbers for the journal
779 *
780 * On external journals the journal blocks are identity-mapped, so
781 * this is a no-op. If needed, we can use j_blk_offset - everything is
782 * ready.
783 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700784int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700785 unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700786{
787 int err = 0;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700788 unsigned long long ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700789
790 if (journal->j_inode) {
791 ret = bmap(journal->j_inode, blocknr);
792 if (ret)
793 *retp = ret;
794 else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700795 printk(KERN_ALERT "%s: journal block not found "
796 "at offset %lu on %s\n",
Theodore Ts'o05496762008-09-16 14:36:17 -0400797 __func__, blocknr, journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700798 err = -EIO;
799 __journal_abort_soft(journal, err);
800 }
801 } else {
802 *retp = blocknr; /* +journal->j_blk_offset */
803 }
804 return err;
805}
806
807/*
808 * We play buffer_head aliasing tricks to write data/metadata blocks to
809 * the journal without copying their contents, but for journal
810 * descriptor blocks we do need to generate bona fide buffers.
811 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700812 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
Dave Kleikamp470decc2006-10-11 01:20:57 -0700813 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
814 * But we don't bother doing that, so there will be coherency problems with
815 * mmaps of blockdevs which hold live JBD-controlled filesystems.
816 */
Jan Karae5a120a2013-06-04 12:06:01 -0400817struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700818{
819 struct buffer_head *bh;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700820 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700821 int err;
822
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700823 err = jbd2_journal_next_log_block(journal, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700824
825 if (err)
826 return NULL;
827
828 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500829 if (!bh)
830 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700831 lock_buffer(bh);
832 memset(bh->b_data, 0, journal->j_blocksize);
833 set_buffer_uptodate(bh);
834 unlock_buffer(bh);
835 BUFFER_TRACE(bh, "return this buffer");
Jan Karae5a120a2013-06-04 12:06:01 -0400836 return bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700837}
838
Jan Kara79feb522012-03-13 22:22:54 -0400839/*
840 * Return tid of the oldest transaction in the journal and block in the journal
841 * where the transaction starts.
842 *
843 * If the journal is now empty, return which will be the next transaction ID
844 * we will write and where will that transaction start.
845 *
846 * The return value is 0 if journal tail cannot be pushed any further, 1 if
847 * it can.
848 */
849int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
850 unsigned long *block)
851{
852 transaction_t *transaction;
853 int ret;
854
855 read_lock(&journal->j_state_lock);
856 spin_lock(&journal->j_list_lock);
857 transaction = journal->j_checkpoint_transactions;
858 if (transaction) {
859 *tid = transaction->t_tid;
860 *block = transaction->t_log_start;
861 } else if ((transaction = journal->j_committing_transaction) != NULL) {
862 *tid = transaction->t_tid;
863 *block = transaction->t_log_start;
864 } else if ((transaction = journal->j_running_transaction) != NULL) {
865 *tid = transaction->t_tid;
866 *block = journal->j_head;
867 } else {
868 *tid = journal->j_transaction_sequence;
869 *block = journal->j_head;
870 }
871 ret = tid_gt(*tid, journal->j_tail_sequence);
872 spin_unlock(&journal->j_list_lock);
873 read_unlock(&journal->j_state_lock);
874
875 return ret;
876}
877
878/*
879 * Update information in journal structure and in on disk journal superblock
880 * about log tail. This function does not check whether information passed in
881 * really pushes log tail further. It's responsibility of the caller to make
882 * sure provided log tail information is valid (e.g. by holding
883 * j_checkpoint_mutex all the time between computing log tail and calling this
884 * function as is the case with jbd2_cleanup_journal_tail()).
885 *
886 * Requires j_checkpoint_mutex
887 */
888void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
889{
890 unsigned long freed;
891
892 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
893
894 /*
895 * We cannot afford for write to remain in drive's caches since as
896 * soon as we update j_tail, next transaction can start reusing journal
897 * space and if we lose sb update during power failure we'd replay
898 * old transaction with possibly newly overwritten data.
899 */
900 jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
901 write_lock(&journal->j_state_lock);
902 freed = block - journal->j_tail;
903 if (block < journal->j_tail)
904 freed += journal->j_last - journal->j_first;
905
906 trace_jbd2_update_log_tail(journal, tid, block, freed);
907 jbd_debug(1,
908 "Cleaning journal tail from %d to %d (offset %lu), "
909 "freeing %lu\n",
910 journal->j_tail_sequence, tid, block, freed);
911
912 journal->j_free += freed;
913 journal->j_tail_sequence = tid;
914 journal->j_tail = block;
915 write_unlock(&journal->j_state_lock);
916}
917
Jan Kara33395782012-03-13 22:45:38 -0400918/*
919 * This is a variaon of __jbd2_update_log_tail which checks for validity of
920 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
921 * with other threads updating log tail.
922 */
923void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
924{
925 mutex_lock(&journal->j_checkpoint_mutex);
926 if (tid_gt(tid, journal->j_tail_sequence))
927 __jbd2_update_log_tail(journal, tid, block);
928 mutex_unlock(&journal->j_checkpoint_mutex);
929}
930
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500931struct jbd2_stats_proc_session {
932 journal_t *journal;
933 struct transaction_stats_s *stats;
934 int start;
935 int max;
936};
937
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500938static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
939{
940 return *pos ? NULL : SEQ_START_TOKEN;
941}
942
943static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
944{
945 return NULL;
946}
947
948static int jbd2_seq_info_show(struct seq_file *seq, void *v)
949{
950 struct jbd2_stats_proc_session *s = seq->private;
951
952 if (v != SEQ_START_TOKEN)
953 return 0;
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500954 seq_printf(seq, "%lu transactions (%lu requested), "
955 "each up to %u blocks\n",
956 s->stats->ts_tid, s->stats->ts_requested,
957 s->journal->j_max_transaction_buffers);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500958 if (s->stats->ts_tid == 0)
959 return 0;
960 seq_printf(seq, "average: \n %ums waiting for transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400961 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500962 seq_printf(seq, " %ums request delay\n",
963 (s->stats->ts_requested == 0) ? 0 :
964 jiffies_to_msecs(s->stats->run.rs_request_delay /
965 s->stats->ts_requested));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500966 seq_printf(seq, " %ums running transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400967 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500968 seq_printf(seq, " %ums transaction was being locked\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400969 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500970 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400971 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500972 seq_printf(seq, " %ums logging transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400973 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -0500974 seq_printf(seq, " %lluus average transaction commit time\n",
975 div_u64(s->journal->j_average_commit_time, 1000));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500976 seq_printf(seq, " %lu handles per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400977 s->stats->run.rs_handle_count / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500978 seq_printf(seq, " %lu blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400979 s->stats->run.rs_blocks / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500980 seq_printf(seq, " %lu logged blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400981 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500982 return 0;
983}
984
985static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
986{
987}
988
James Morris88e9d342009-09-22 16:43:43 -0700989static const struct seq_operations jbd2_seq_info_ops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500990 .start = jbd2_seq_info_start,
991 .next = jbd2_seq_info_next,
992 .stop = jbd2_seq_info_stop,
993 .show = jbd2_seq_info_show,
994};
995
996static int jbd2_seq_info_open(struct inode *inode, struct file *file)
997{
Al Virod9dda782013-03-31 18:16:14 -0400998 journal_t *journal = PDE_DATA(inode);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500999 struct jbd2_stats_proc_session *s;
1000 int rc, size;
1001
1002 s = kmalloc(sizeof(*s), GFP_KERNEL);
1003 if (s == NULL)
1004 return -ENOMEM;
1005 size = sizeof(struct transaction_stats_s);
1006 s->stats = kmalloc(size, GFP_KERNEL);
1007 if (s->stats == NULL) {
1008 kfree(s);
1009 return -ENOMEM;
1010 }
1011 spin_lock(&journal->j_history_lock);
1012 memcpy(s->stats, &journal->j_stats, size);
1013 s->journal = journal;
1014 spin_unlock(&journal->j_history_lock);
1015
1016 rc = seq_open(file, &jbd2_seq_info_ops);
1017 if (rc == 0) {
1018 struct seq_file *m = file->private_data;
1019 m->private = s;
1020 } else {
1021 kfree(s->stats);
1022 kfree(s);
1023 }
1024 return rc;
1025
1026}
1027
1028static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1029{
1030 struct seq_file *seq = file->private_data;
1031 struct jbd2_stats_proc_session *s = seq->private;
1032 kfree(s->stats);
1033 kfree(s);
1034 return seq_release(inode, file);
1035}
1036
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001037static const struct file_operations jbd2_seq_info_fops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001038 .owner = THIS_MODULE,
1039 .open = jbd2_seq_info_open,
1040 .read = seq_read,
1041 .llseek = seq_lseek,
1042 .release = jbd2_seq_info_release,
1043};
1044
1045static struct proc_dir_entry *proc_jbd2_stats;
1046
1047static void jbd2_stats_proc_init(journal_t *journal)
1048{
Theodore Ts'o05496762008-09-16 14:36:17 -04001049 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001050 if (journal->j_proc_entry) {
Denis V. Lunev79da3662008-04-29 01:02:11 -07001051 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1052 &jbd2_seq_info_fops, journal);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001053 }
1054}
1055
1056static void jbd2_stats_proc_exit(journal_t *journal)
1057{
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001058 remove_proc_entry("info", journal->j_proc_entry);
Theodore Ts'o05496762008-09-16 14:36:17 -04001059 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001060}
1061
Dave Kleikamp470decc2006-10-11 01:20:57 -07001062/*
1063 * Management for journal control blocks: functions to create and
1064 * destroy journal_t structures, and to initialise and read existing
1065 * journal blocks from disk. */
1066
1067/* First: create and setup a journal_t object in memory. We initialise
1068 * very few fields yet: that has to wait until we have created the
1069 * journal structures from from scratch, or loaded them from disk. */
1070
1071static journal_t * journal_init_common (void)
1072{
1073 journal_t *journal;
1074 int err;
1075
Andrew Morton3ebfdf82009-12-23 08:05:15 -05001076 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001077 if (!journal)
Theodore Ts'ob7271b02010-12-18 13:39:38 -05001078 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001079
1080 init_waitqueue_head(&journal->j_wait_transaction_locked);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001081 init_waitqueue_head(&journal->j_wait_done_commit);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001082 init_waitqueue_head(&journal->j_wait_commit);
1083 init_waitqueue_head(&journal->j_wait_updates);
Jan Kara8f7d89f2013-06-04 12:35:11 -04001084 init_waitqueue_head(&journal->j_wait_reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001085 mutex_init(&journal->j_barrier);
1086 mutex_init(&journal->j_checkpoint_mutex);
1087 spin_lock_init(&journal->j_revoke_lock);
1088 spin_lock_init(&journal->j_list_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001089 rwlock_init(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001090
Mingming Caocd02ff02007-10-16 18:38:25 -04001091 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
Theodore Ts'o30773842009-01-03 20:27:38 -05001092 journal->j_min_batch_time = 0;
1093 journal->j_max_batch_time = 15000; /* 15ms */
Jan Kara8f7d89f2013-06-04 12:35:11 -04001094 atomic_set(&journal->j_reserved_credits, 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001095
1096 /* The journal is marked for error until we succeed with recovery! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001097 journal->j_flags = JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001098
1099 /* Set up a default-sized revoke table for the new mount. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001100 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001101 if (err) {
1102 kfree(journal);
Theodore Ts'ob7271b02010-12-18 13:39:38 -05001103 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001104 }
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001105
Theodore Ts'obf699322009-09-30 00:32:06 -04001106 spin_lock_init(&journal->j_history_lock);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001107
Dave Kleikamp470decc2006-10-11 01:20:57 -07001108 return journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001109}
1110
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001111/* jbd2_journal_init_dev and jbd2_journal_init_inode:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001112 *
1113 * Create a journal structure assigned some fixed set of disk blocks to
1114 * the journal. We don't actually touch those disk blocks yet, but we
1115 * need to set up all of the mapping information to tell the journaling
1116 * system where the journal blocks are.
1117 *
1118 */
1119
1120/**
Randy Dunlap5648ba52008-04-17 10:38:59 -04001121 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
Dave Kleikamp470decc2006-10-11 01:20:57 -07001122 * @bdev: Block device on which to create the journal
1123 * @fs_dev: Device which hold journalled filesystem for this journal.
1124 * @start: Block nr Start of journal.
1125 * @len: Length of the journal in blocks.
1126 * @blocksize: blocksize of journalling device
Randy Dunlap5648ba52008-04-17 10:38:59 -04001127 *
1128 * Returns: a newly created journal_t *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001129 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001130 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
Dave Kleikamp470decc2006-10-11 01:20:57 -07001131 * range of blocks on an arbitrary block device.
1132 *
1133 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001134journal_t * jbd2_journal_init_dev(struct block_device *bdev,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001135 struct block_device *fs_dev,
Mingming Cao18eba7a2006-10-11 01:21:13 -07001136 unsigned long long start, int len, int blocksize)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001137{
1138 journal_t *journal = journal_init_common();
1139 struct buffer_head *bh;
Theodore Ts'o05496762008-09-16 14:36:17 -04001140 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001141 int n;
1142
1143 if (!journal)
1144 return NULL;
1145
1146 /* journal descriptor can store up to n blocks -bzzz */
1147 journal->j_blocksize = blocksize;
yangsheng0587aa32010-11-17 21:46:26 -05001148 journal->j_dev = bdev;
1149 journal->j_fs_dev = fs_dev;
1150 journal->j_blk_offset = start;
1151 journal->j_maxlen = len;
1152 bdevname(journal->j_dev, journal->j_devname);
1153 p = journal->j_devname;
1154 while ((p = strchr(p, '/')))
1155 *p = '!';
Jan Kara4b905672009-01-06 14:53:35 -05001156 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001157 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1158 journal->j_wbufsize = n;
1159 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1160 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001161 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001162 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001163 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001164 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001165
1166 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001167 if (!bh) {
1168 printk(KERN_ERR
1169 "%s: Cannot get buffer for journal superblock\n",
1170 __func__);
1171 goto out_err;
1172 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001173 journal->j_sb_buffer = bh;
1174 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Kara4b905672009-01-06 14:53:35 -05001175
Dave Kleikamp470decc2006-10-11 01:20:57 -07001176 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001177out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001178 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001179 jbd2_stats_proc_exit(journal);
1180 kfree(journal);
1181 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001182}
1183
1184/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001185 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001186 * @inode: An inode to create the journal in
1187 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001188 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
Dave Kleikamp470decc2006-10-11 01:20:57 -07001189 * the journal. The inode must exist already, must support bmap() and
1190 * must have all data blocks preallocated.
1191 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001192journal_t * jbd2_journal_init_inode (struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001193{
1194 struct buffer_head *bh;
1195 journal_t *journal = journal_init_common();
Theodore Ts'o05496762008-09-16 14:36:17 -04001196 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001197 int err;
1198 int n;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001199 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001200
1201 if (!journal)
1202 return NULL;
1203
1204 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1205 journal->j_inode = inode;
Theodore Ts'o05496762008-09-16 14:36:17 -04001206 bdevname(journal->j_dev, journal->j_devname);
1207 p = journal->j_devname;
1208 while ((p = strchr(p, '/')))
1209 *p = '!';
1210 p = journal->j_devname + strlen(journal->j_devname);
Theodore Ts'o90576c02009-09-29 15:51:30 -04001211 sprintf(p, "-%lu", journal->j_inode->i_ino);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001212 jbd_debug(1,
1213 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1214 journal, inode->i_sb->s_id, inode->i_ino,
1215 (long long) inode->i_size,
1216 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1217
1218 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1219 journal->j_blocksize = inode->i_sb->s_blocksize;
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001220 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001221
1222 /* journal descriptor can store up to n blocks -bzzz */
1223 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1224 journal->j_wbufsize = n;
1225 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1226 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001227 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\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
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001232 err = jbd2_journal_bmap(journal, 0, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001233 /* If that failed, give up */
1234 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -08001235 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001236 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001237 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001238 }
1239
1240 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001241 if (!bh) {
1242 printk(KERN_ERR
1243 "%s: Cannot get buffer for journal superblock\n",
1244 __func__);
1245 goto out_err;
1246 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001247 journal->j_sb_buffer = bh;
1248 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1249
1250 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001251out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001252 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001253 jbd2_stats_proc_exit(journal);
1254 kfree(journal);
1255 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001256}
1257
1258/*
1259 * If the journal init or create aborts, we need to mark the journal
1260 * superblock as being NULL to prevent the journal destroy from writing
1261 * back a bogus superblock.
1262 */
1263static void journal_fail_superblock (journal_t *journal)
1264{
1265 struct buffer_head *bh = journal->j_sb_buffer;
1266 brelse(bh);
1267 journal->j_sb_buffer = NULL;
1268}
1269
1270/*
1271 * Given a journal_t structure, initialise the various fields for
1272 * startup of a new journaling session. We use this both when creating
1273 * a journal, and after recovering an old journal to reset it for
1274 * subsequent use.
1275 */
1276
1277static int journal_reset(journal_t *journal)
1278{
1279 journal_superblock_t *sb = journal->j_superblock;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001280 unsigned long long first, last;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001281
1282 first = be32_to_cpu(sb->s_first);
1283 last = be32_to_cpu(sb->s_maxlen);
Jan Karaf6f50e22009-07-17 10:40:01 -04001284 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001285 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
Jan Karaf6f50e22009-07-17 10:40:01 -04001286 first, last);
1287 journal_fail_superblock(journal);
1288 return -EINVAL;
1289 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001290
1291 journal->j_first = first;
1292 journal->j_last = last;
1293
1294 journal->j_head = first;
1295 journal->j_tail = first;
1296 journal->j_free = last - first;
1297
1298 journal->j_tail_sequence = journal->j_transaction_sequence;
1299 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1300 journal->j_commit_request = journal->j_commit_sequence;
1301
1302 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1303
Dave Kleikamp470decc2006-10-11 01:20:57 -07001304 /*
1305 * As a special case, if the on-disk copy is already marked as needing
Jan Kara24bcc892012-03-13 15:41:04 -04001306 * no recovery (s_start == 0), then we can safely defer the superblock
1307 * update until the next commit by setting JBD2_FLUSHED. This avoids
Dave Kleikamp470decc2006-10-11 01:20:57 -07001308 * attempting a write to a potential-readonly device.
1309 */
Jan Kara24bcc892012-03-13 15:41:04 -04001310 if (sb->s_start == 0) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001311 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
Dave Kleikamp470decc2006-10-11 01:20:57 -07001312 "(start %ld, seq %d, errno %d)\n",
1313 journal->j_tail, journal->j_tail_sequence,
1314 journal->j_errno);
Jan Kara24bcc892012-03-13 15:41:04 -04001315 journal->j_flags |= JBD2_FLUSHED;
1316 } else {
Jan Karaa78bb112012-03-13 15:43:04 -04001317 /* Lock here to make assertions happy... */
1318 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara79feb522012-03-13 22:22:54 -04001319 /*
1320 * Update log tail information. We use WRITE_FUA since new
1321 * transaction will start reusing journal space and so we
1322 * must make sure information about current log tail is on
1323 * disk before that.
1324 */
1325 jbd2_journal_update_sb_log_tail(journal,
1326 journal->j_tail_sequence,
1327 journal->j_tail,
1328 WRITE_FUA);
Jan Karaa78bb112012-03-13 15:43:04 -04001329 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001330 }
Jan Kara24bcc892012-03-13 15:41:04 -04001331 return jbd2_journal_start_thread(journal);
1332}
Dave Kleikamp470decc2006-10-11 01:20:57 -07001333
Jan Kara79feb522012-03-13 22:22:54 -04001334static void jbd2_write_superblock(journal_t *journal, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001335{
1336 struct buffer_head *bh = journal->j_sb_buffer;
Theodore Ts'ofe52d172013-07-01 08:12:38 -04001337 journal_superblock_t *sb = journal->j_superblock;
Jan Kara79feb522012-03-13 22:22:54 -04001338 int ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001339
Jan Kara79feb522012-03-13 22:22:54 -04001340 trace_jbd2_write_superblock(journal, write_op);
1341 if (!(journal->j_flags & JBD2_BARRIER))
1342 write_op &= ~(REQ_FUA | REQ_FLUSH);
1343 lock_buffer(bh);
Theodore Ts'o914258b2008-10-06 21:35:40 -04001344 if (buffer_write_io_error(bh)) {
1345 /*
1346 * Oh, dear. A previous attempt to write the journal
1347 * superblock failed. This could happen because the
1348 * USB device was yanked out. Or it could happen to
1349 * be a transient write error and maybe the block will
1350 * be remapped. Nothing we can do but to retry the
1351 * write and hope for the best.
1352 */
1353 printk(KERN_ERR "JBD2: previous I/O error detected "
1354 "for journal superblock update for %s.\n",
1355 journal->j_devname);
1356 clear_buffer_write_io_error(bh);
1357 set_buffer_uptodate(bh);
1358 }
Theodore Ts'ofe52d172013-07-01 08:12:38 -04001359 jbd2_superblock_csum_set(journal, sb);
Jan Kara79feb522012-03-13 22:22:54 -04001360 get_bh(bh);
1361 bh->b_end_io = end_buffer_write_sync;
1362 ret = submit_bh(write_op, bh);
1363 wait_on_buffer(bh);
Jan Kara24bcc892012-03-13 15:41:04 -04001364 if (buffer_write_io_error(bh)) {
Jan Kara24bcc892012-03-13 15:41:04 -04001365 clear_buffer_write_io_error(bh);
1366 set_buffer_uptodate(bh);
Jan Kara79feb522012-03-13 22:22:54 -04001367 ret = -EIO;
1368 }
1369 if (ret) {
1370 printk(KERN_ERR "JBD2: Error %d detected when updating "
1371 "journal superblock for %s.\n", ret,
1372 journal->j_devname);
Jan Kara24bcc892012-03-13 15:41:04 -04001373 }
1374}
Theodore Ts'o914258b2008-10-06 21:35:40 -04001375
Jan Kara24bcc892012-03-13 15:41:04 -04001376/**
1377 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1378 * @journal: The journal to update.
Jan Kara79feb522012-03-13 22:22:54 -04001379 * @tail_tid: TID of the new transaction at the tail of the log
1380 * @tail_block: The first block of the transaction at the tail of the log
1381 * @write_op: With which operation should we write the journal sb
Jan Kara24bcc892012-03-13 15:41:04 -04001382 *
1383 * Update a journal's superblock information about log tail and write it to
1384 * disk, waiting for the IO to complete.
1385 */
Jan Kara79feb522012-03-13 22:22:54 -04001386void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1387 unsigned long tail_block, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001388{
1389 journal_superblock_t *sb = journal->j_superblock;
1390
Jan Karaa78bb112012-03-13 15:43:04 -04001391 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Jan Kara79feb522012-03-13 22:22:54 -04001392 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1393 tail_block, tail_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001394
Jan Kara79feb522012-03-13 22:22:54 -04001395 sb->s_sequence = cpu_to_be32(tail_tid);
1396 sb->s_start = cpu_to_be32(tail_block);
Jan Kara24bcc892012-03-13 15:41:04 -04001397
Jan Kara79feb522012-03-13 22:22:54 -04001398 jbd2_write_superblock(journal, write_op);
Jan Kara24bcc892012-03-13 15:41:04 -04001399
1400 /* Log is no longer empty */
1401 write_lock(&journal->j_state_lock);
1402 WARN_ON(!sb->s_sequence);
1403 journal->j_flags &= ~JBD2_FLUSHED;
1404 write_unlock(&journal->j_state_lock);
1405}
1406
1407/**
1408 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1409 * @journal: The journal to update.
1410 *
1411 * Update a journal's dynamic superblock fields to show that journal is empty.
1412 * Write updated superblock to disk waiting for IO to complete.
1413 */
1414static void jbd2_mark_journal_empty(journal_t *journal)
1415{
1416 journal_superblock_t *sb = journal->j_superblock;
1417
Jan Karaa78bb112012-03-13 15:43:04 -04001418 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001419 read_lock(&journal->j_state_lock);
Eric Sandeeneeecef02012-08-18 22:29:40 -04001420 /* Is it already empty? */
1421 if (sb->s_start == 0) {
1422 read_unlock(&journal->j_state_lock);
1423 return;
1424 }
Jan Kara24bcc892012-03-13 15:41:04 -04001425 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1426 journal->j_tail_sequence);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001427
1428 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
Jan Kara24bcc892012-03-13 15:41:04 -04001429 sb->s_start = cpu_to_be32(0);
1430 read_unlock(&journal->j_state_lock);
1431
Jan Kara79feb522012-03-13 22:22:54 -04001432 jbd2_write_superblock(journal, WRITE_FUA);
Jan Kara24bcc892012-03-13 15:41:04 -04001433
1434 /* Log is no longer empty */
1435 write_lock(&journal->j_state_lock);
1436 journal->j_flags |= JBD2_FLUSHED;
1437 write_unlock(&journal->j_state_lock);
1438}
1439
1440
1441/**
1442 * jbd2_journal_update_sb_errno() - Update error in the journal.
1443 * @journal: The journal to update.
1444 *
1445 * Update a journal's errno. Write updated superblock to disk waiting for IO
1446 * to complete.
1447 */
Theodore Ts'od796c522012-08-05 19:04:57 -04001448void jbd2_journal_update_sb_errno(journal_t *journal)
Jan Kara24bcc892012-03-13 15:41:04 -04001449{
1450 journal_superblock_t *sb = journal->j_superblock;
1451
1452 read_lock(&journal->j_state_lock);
1453 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1454 journal->j_errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001455 sb->s_errno = cpu_to_be32(journal->j_errno);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001456 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001457
Jan Kara79feb522012-03-13 22:22:54 -04001458 jbd2_write_superblock(journal, WRITE_SYNC);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001459}
Theodore Ts'od796c522012-08-05 19:04:57 -04001460EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001461
1462/*
1463 * Read the superblock for a given journal, performing initial
1464 * validation of the format.
1465 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001466static int journal_get_superblock(journal_t *journal)
1467{
1468 struct buffer_head *bh;
1469 journal_superblock_t *sb;
1470 int err = -EIO;
1471
1472 bh = journal->j_sb_buffer;
1473
1474 J_ASSERT(bh != NULL);
1475 if (!buffer_uptodate(bh)) {
1476 ll_rw_block(READ, 1, &bh);
1477 wait_on_buffer(bh);
1478 if (!buffer_uptodate(bh)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001479 printk(KERN_ERR
1480 "JBD2: IO error reading journal superblock\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001481 goto out;
1482 }
1483 }
1484
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001485 if (buffer_verified(bh))
1486 return 0;
1487
Dave Kleikamp470decc2006-10-11 01:20:57 -07001488 sb = journal->j_superblock;
1489
1490 err = -EINVAL;
1491
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001492 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001493 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001494 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001495 goto out;
1496 }
1497
1498 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001499 case JBD2_SUPERBLOCK_V1:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001500 journal->j_format_version = 1;
1501 break;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001502 case JBD2_SUPERBLOCK_V2:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001503 journal->j_format_version = 2;
1504 break;
1505 default:
Eryu Guanf2a44522011-11-01 19:09:18 -04001506 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001507 goto out;
1508 }
1509
1510 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1511 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1512 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001513 printk(KERN_WARNING "JBD2: journal file too short\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001514 goto out;
1515 }
1516
Eryu Guan87622022011-11-01 19:04:59 -04001517 if (be32_to_cpu(sb->s_first) == 0 ||
1518 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1519 printk(KERN_WARNING
1520 "JBD2: Invalid start block of journal: %u\n",
1521 be32_to_cpu(sb->s_first));
1522 goto out;
1523 }
1524
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001525 if (jbd2_journal_has_csum_v2or3(journal) &&
1526 JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM)) {
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001527 /* Can't have checksum v1 and v2 on at the same time! */
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001528 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2 "
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001529 "at the same time!\n");
1530 goto out;
1531 }
1532
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001533 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2) &&
1534 JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1535 /* Can't have checksum v2 and v3 at the same time! */
1536 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1537 "at the same time!\n");
1538 goto out;
1539 }
1540
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001541 if (!jbd2_verify_csum_type(journal, sb)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001542 printk(KERN_ERR "JBD2: Unknown checksum type\n");
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001543 goto out;
1544 }
1545
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001546 /* Load the checksum driver */
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001547 if (jbd2_journal_has_csum_v2or3(journal)) {
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001548 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1549 if (IS_ERR(journal->j_chksum_driver)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001550 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001551 err = PTR_ERR(journal->j_chksum_driver);
1552 journal->j_chksum_driver = NULL;
1553 goto out;
1554 }
1555 }
1556
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001557 /* Check superblock checksum */
1558 if (!jbd2_superblock_csum_verify(journal, sb)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001559 printk(KERN_ERR "JBD2: journal checksum error\n");
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001560 goto out;
1561 }
1562
1563 /* Precompute checksum seed for all metadata */
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001564 if (jbd2_journal_has_csum_v2or3(journal))
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001565 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1566 sizeof(sb->s_uuid));
1567
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001568 set_buffer_verified(bh);
1569
Dave Kleikamp470decc2006-10-11 01:20:57 -07001570 return 0;
1571
1572out:
1573 journal_fail_superblock(journal);
1574 return err;
1575}
1576
1577/*
1578 * Load the on-disk journal superblock and read the key fields into the
1579 * journal_t.
1580 */
1581
1582static int load_superblock(journal_t *journal)
1583{
1584 int err;
1585 journal_superblock_t *sb;
1586
1587 err = journal_get_superblock(journal);
1588 if (err)
1589 return err;
1590
1591 sb = journal->j_superblock;
1592
1593 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1594 journal->j_tail = be32_to_cpu(sb->s_start);
1595 journal->j_first = be32_to_cpu(sb->s_first);
1596 journal->j_last = be32_to_cpu(sb->s_maxlen);
1597 journal->j_errno = be32_to_cpu(sb->s_errno);
1598
1599 return 0;
1600}
1601
1602
1603/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001604 * int jbd2_journal_load() - Read journal from disk.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001605 * @journal: Journal to act on.
1606 *
1607 * Given a journal_t structure which tells us which disk blocks contain
1608 * a journal, read the journal from disk to initialise the in-memory
1609 * structures.
1610 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001611int jbd2_journal_load(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001612{
1613 int err;
1614 journal_superblock_t *sb;
1615
1616 err = load_superblock(journal);
1617 if (err)
1618 return err;
1619
1620 sb = journal->j_superblock;
1621 /* If this is a V2 superblock, then we have to check the
1622 * features flags on it. */
1623
1624 if (journal->j_format_version >= 2) {
1625 if ((sb->s_feature_ro_compat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001626 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001627 (sb->s_feature_incompat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001628 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001629 printk(KERN_WARNING
1630 "JBD2: Unrecognised features on journal\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001631 return -EINVAL;
1632 }
1633 }
1634
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001635 /*
1636 * Create a slab for this blocksize
1637 */
1638 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1639 if (err)
1640 return err;
1641
Dave Kleikamp470decc2006-10-11 01:20:57 -07001642 /* Let the recovery code check whether it needs to recover any
1643 * data from the journal. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001644 if (jbd2_journal_recover(journal))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001645 goto recovery_error;
1646
Theodore Ts'oe6a47422009-11-15 15:31:37 -05001647 if (journal->j_failed_commit) {
1648 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1649 "is corrupt.\n", journal->j_failed_commit,
1650 journal->j_devname);
1651 return -EIO;
1652 }
1653
Dave Kleikamp470decc2006-10-11 01:20:57 -07001654 /* OK, we've finished with the dynamic journal bits:
1655 * reinitialise the dynamic contents of the superblock in memory
1656 * and reset them on disk. */
1657 if (journal_reset(journal))
1658 goto recovery_error;
1659
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001660 journal->j_flags &= ~JBD2_ABORT;
1661 journal->j_flags |= JBD2_LOADED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001662 return 0;
1663
1664recovery_error:
Eryu Guanf2a44522011-11-01 19:09:18 -04001665 printk(KERN_WARNING "JBD2: recovery failed\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001666 return -EIO;
1667}
1668
1669/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001670 * void jbd2_journal_destroy() - Release a journal_t structure.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001671 * @journal: Journal to act on.
1672 *
1673 * Release a journal_t structure once it is no longer in use by the
1674 * journaled object.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001675 * Return <0 if we couldn't clean up the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001676 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001677int jbd2_journal_destroy(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001678{
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001679 int err = 0;
1680
Dave Kleikamp470decc2006-10-11 01:20:57 -07001681 /* Wait for the commit thread to wake up and die. */
1682 journal_kill_thread(journal);
1683
1684 /* Force a final log commit */
1685 if (journal->j_running_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001686 jbd2_journal_commit_transaction(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001687
1688 /* Force any old transactions to disk */
1689
1690 /* Totally anal locking here... */
1691 spin_lock(&journal->j_list_lock);
1692 while (journal->j_checkpoint_transactions != NULL) {
1693 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001694 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001695 jbd2_log_do_checkpoint(journal);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001696 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001697 spin_lock(&journal->j_list_lock);
1698 }
1699
1700 J_ASSERT(journal->j_running_transaction == NULL);
1701 J_ASSERT(journal->j_committing_transaction == NULL);
1702 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1703 spin_unlock(&journal->j_list_lock);
1704
Dave Kleikamp470decc2006-10-11 01:20:57 -07001705 if (journal->j_sb_buffer) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001706 if (!is_journal_aborted(journal)) {
Jan Karaa78bb112012-03-13 15:43:04 -04001707 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara24bcc892012-03-13 15:41:04 -04001708 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04001709 mutex_unlock(&journal->j_checkpoint_mutex);
1710 } else
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001711 err = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001712 brelse(journal->j_sb_buffer);
1713 }
1714
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001715 if (journal->j_proc_entry)
1716 jbd2_stats_proc_exit(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001717 if (journal->j_inode)
1718 iput(journal->j_inode);
1719 if (journal->j_revoke)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001720 jbd2_journal_destroy_revoke(journal);
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001721 if (journal->j_chksum_driver)
1722 crypto_free_shash(journal->j_chksum_driver);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001723 kfree(journal->j_wbuf);
1724 kfree(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001725
1726 return err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001727}
1728
1729
1730/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001731 *int jbd2_journal_check_used_features () - Check if features specified are used.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001732 * @journal: Journal to check.
1733 * @compat: bitmask of compatible features
1734 * @ro: bitmask of features that force read-only mount
1735 * @incompat: bitmask of incompatible features
1736 *
1737 * Check whether the journal uses all of a given set of
1738 * features. Return true (non-zero) if it does.
1739 **/
1740
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001741int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001742 unsigned long ro, unsigned long incompat)
1743{
1744 journal_superblock_t *sb;
1745
1746 if (!compat && !ro && !incompat)
1747 return 1;
Patrick J. LoPresti1113e1b2010-07-22 15:04:16 -07001748 /* Load journal superblock if it is not loaded yet. */
1749 if (journal->j_format_version == 0 &&
1750 journal_get_superblock(journal) != 0)
1751 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001752 if (journal->j_format_version == 1)
1753 return 0;
1754
1755 sb = journal->j_superblock;
1756
1757 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1758 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1759 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1760 return 1;
1761
1762 return 0;
1763}
1764
1765/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001766 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001767 * @journal: Journal to check.
1768 * @compat: bitmask of compatible features
1769 * @ro: bitmask of features that force read-only mount
1770 * @incompat: bitmask of incompatible features
1771 *
1772 * Check whether the journaling code supports the use of
1773 * all of a given set of features on this journal. Return true
1774 * (non-zero) if it can. */
1775
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001776int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001777 unsigned long ro, unsigned long incompat)
1778{
Dave Kleikamp470decc2006-10-11 01:20:57 -07001779 if (!compat && !ro && !incompat)
1780 return 1;
1781
Dave Kleikamp470decc2006-10-11 01:20:57 -07001782 /* We can support any known requested features iff the
1783 * superblock is in version 2. Otherwise we fail to support any
1784 * extended sb features. */
1785
1786 if (journal->j_format_version != 2)
1787 return 0;
1788
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001789 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1790 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1791 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001792 return 1;
1793
1794 return 0;
1795}
1796
1797/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001798 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
Dave Kleikamp470decc2006-10-11 01:20:57 -07001799 * @journal: Journal to act on.
1800 * @compat: bitmask of compatible features
1801 * @ro: bitmask of features that force read-only mount
1802 * @incompat: bitmask of incompatible features
1803 *
1804 * Mark a given journal feature as present on the
1805 * superblock. Returns true if the requested features could be set.
1806 *
1807 */
1808
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001809int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001810 unsigned long ro, unsigned long incompat)
1811{
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001812#define INCOMPAT_FEATURE_ON(f) \
1813 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1814#define COMPAT_FEATURE_ON(f) \
1815 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001816 journal_superblock_t *sb;
1817
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001818 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001819 return 1;
1820
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001821 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001822 return 0;
1823
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001824 /* If enabling v2 checksums, turn on v3 instead */
1825 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1826 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1827 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1828 }
1829
1830 /* Asking for checksumming v3 and v1? Only give them v3. */
1831 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001832 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1833 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1834
Dave Kleikamp470decc2006-10-11 01:20:57 -07001835 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1836 compat, ro, incompat);
1837
1838 sb = journal->j_superblock;
1839
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001840 /* If enabling v3 checksums, update superblock */
1841 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001842 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1843 sb->s_feature_compat &=
1844 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001845
1846 /* Load the checksum driver */
1847 if (journal->j_chksum_driver == NULL) {
1848 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1849 0, 0);
1850 if (IS_ERR(journal->j_chksum_driver)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001851 printk(KERN_ERR "JBD2: Cannot load crc32c "
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001852 "driver.\n");
1853 journal->j_chksum_driver = NULL;
1854 return 0;
1855 }
1856 }
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001857
1858 /* Precompute checksum seed for all metadata */
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001859 if (jbd2_journal_has_csum_v2or3(journal))
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001860 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1861 sb->s_uuid,
1862 sizeof(sb->s_uuid));
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001863 }
1864
1865 /* If enabling v1 checksums, downgrade superblock */
1866 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1867 sb->s_feature_incompat &=
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001868 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1869 JBD2_FEATURE_INCOMPAT_CSUM_V3);
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001870
Dave Kleikamp470decc2006-10-11 01:20:57 -07001871 sb->s_feature_compat |= cpu_to_be32(compat);
1872 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1873 sb->s_feature_incompat |= cpu_to_be32(incompat);
1874
1875 return 1;
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001876#undef COMPAT_FEATURE_ON
1877#undef INCOMPAT_FEATURE_ON
Dave Kleikamp470decc2006-10-11 01:20:57 -07001878}
1879
Girish Shilamkar818d2762008-01-28 23:58:27 -05001880/*
1881 * jbd2_journal_clear_features () - Clear a given journal feature in the
1882 * superblock
1883 * @journal: Journal to act on.
1884 * @compat: bitmask of compatible features
1885 * @ro: bitmask of features that force read-only mount
1886 * @incompat: bitmask of incompatible features
1887 *
1888 * Clear a given journal feature as present on the
1889 * superblock.
1890 */
1891void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1892 unsigned long ro, unsigned long incompat)
1893{
1894 journal_superblock_t *sb;
1895
1896 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1897 compat, ro, incompat);
1898
1899 sb = journal->j_superblock;
1900
1901 sb->s_feature_compat &= ~cpu_to_be32(compat);
1902 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1903 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1904}
1905EXPORT_SYMBOL(jbd2_journal_clear_features);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001906
1907/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001908 * int jbd2_journal_flush () - Flush journal
Dave Kleikamp470decc2006-10-11 01:20:57 -07001909 * @journal: Journal to act on.
1910 *
1911 * Flush all data for a given journal to disk and empty the journal.
1912 * Filesystems can use this when remounting readonly to ensure that
1913 * recovery does not need to happen on remount.
1914 */
1915
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001916int jbd2_journal_flush(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001917{
1918 int err = 0;
1919 transaction_t *transaction = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001920
Theodore Ts'oa931da62010-08-03 21:35:12 -04001921 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001922
1923 /* Force everything buffered to the log... */
1924 if (journal->j_running_transaction) {
1925 transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001926 __jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001927 } else if (journal->j_committing_transaction)
1928 transaction = journal->j_committing_transaction;
1929
1930 /* Wait for the log commit to complete... */
1931 if (transaction) {
1932 tid_t tid = transaction->t_tid;
1933
Theodore Ts'oa931da62010-08-03 21:35:12 -04001934 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001935 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001936 } else {
Theodore Ts'oa931da62010-08-03 21:35:12 -04001937 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001938 }
1939
1940 /* ...and flush everything in the log out to disk. */
1941 spin_lock(&journal->j_list_lock);
1942 while (!err && journal->j_checkpoint_transactions != NULL) {
1943 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001944 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001945 err = jbd2_log_do_checkpoint(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001946 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001947 spin_lock(&journal->j_list_lock);
1948 }
1949 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001950
1951 if (is_journal_aborted(journal))
1952 return -EIO;
1953
Jan Karaa78bb112012-03-13 15:43:04 -04001954 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001955 jbd2_cleanup_journal_tail(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001956
1957 /* Finally, mark the journal as really needing no recovery.
1958 * This sets s_start==0 in the underlying superblock, which is
1959 * the magic code for a fully-recovered superblock. Any future
1960 * commits of data to the journal will restore the current
1961 * s_start value. */
Jan Kara24bcc892012-03-13 15:41:04 -04001962 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04001963 mutex_unlock(&journal->j_checkpoint_mutex);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001964 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001965 J_ASSERT(!journal->j_running_transaction);
1966 J_ASSERT(!journal->j_committing_transaction);
1967 J_ASSERT(!journal->j_checkpoint_transactions);
1968 J_ASSERT(journal->j_head == journal->j_tail);
1969 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001970 write_unlock(&journal->j_state_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001971 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001972}
1973
1974/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001975 * int jbd2_journal_wipe() - Wipe journal contents
Dave Kleikamp470decc2006-10-11 01:20:57 -07001976 * @journal: Journal to act on.
1977 * @write: flag (see below)
1978 *
1979 * Wipe out all of the contents of a journal, safely. This will produce
1980 * a warning if the journal contains any valid recovery information.
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001981 * Must be called between journal_init_*() and jbd2_journal_load().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001982 *
1983 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1984 * we merely suppress recovery.
1985 */
1986
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001987int jbd2_journal_wipe(journal_t *journal, int write)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001988{
Dave Kleikamp470decc2006-10-11 01:20:57 -07001989 int err = 0;
1990
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001991 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001992
1993 err = load_superblock(journal);
1994 if (err)
1995 return err;
1996
Dave Kleikamp470decc2006-10-11 01:20:57 -07001997 if (!journal->j_tail)
1998 goto no_recovery;
1999
Eryu Guanf2a44522011-11-01 19:09:18 -04002000 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002001 write ? "Clearing" : "Ignoring");
2002
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002003 err = jbd2_journal_skip_recovery(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04002004 if (write) {
2005 /* Lock to make assertions happy... */
2006 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara24bcc892012-03-13 15:41:04 -04002007 jbd2_mark_journal_empty(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04002008 mutex_unlock(&journal->j_checkpoint_mutex);
2009 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002010
2011 no_recovery:
2012 return err;
2013}
2014
2015/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07002016 * Journal abort has very specific semantics, which we describe
2017 * for journal abort.
2018 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002019 * Two internal functions, which provide abort to the jbd layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07002020 * itself are here.
2021 */
2022
2023/*
2024 * Quick version for internal journal use (doesn't lock the journal).
2025 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2026 * and don't attempt to make any other journal updates.
2027 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002028void __jbd2_journal_abort_hard(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002029{
2030 transaction_t *transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002031
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002032 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002033 return;
2034
2035 printk(KERN_ERR "Aborting journal on device %s.\n",
Theodore Ts'o05496762008-09-16 14:36:17 -04002036 journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002037
Theodore Ts'oa931da62010-08-03 21:35:12 -04002038 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002039 journal->j_flags |= JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002040 transaction = journal->j_running_transaction;
2041 if (transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002042 __jbd2_log_start_commit(journal, transaction->t_tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002043 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002044}
2045
2046/* Soft abort: record the abort error status in the journal superblock,
2047 * but don't do any other IO. */
2048static void __journal_abort_soft (journal_t *journal, int errno)
2049{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002050 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002051 return;
2052
2053 if (!journal->j_errno)
2054 journal->j_errno = errno;
2055
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002056 __jbd2_journal_abort_hard(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002057
2058 if (errno)
Jan Kara24bcc892012-03-13 15:41:04 -04002059 jbd2_journal_update_sb_errno(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002060}
2061
2062/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002063 * void jbd2_journal_abort () - Shutdown the journal immediately.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002064 * @journal: the journal to shutdown.
2065 * @errno: an error number to record in the journal indicating
2066 * the reason for the shutdown.
2067 *
2068 * Perform a complete, immediate shutdown of the ENTIRE
2069 * journal (not of a single transaction). This operation cannot be
2070 * undone without closing and reopening the journal.
2071 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002072 * The jbd2_journal_abort function is intended to support higher level error
Dave Kleikamp470decc2006-10-11 01:20:57 -07002073 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2074 * mode.
2075 *
2076 * Journal abort has very specific semantics. Any existing dirty,
2077 * unjournaled buffers in the main filesystem will still be written to
2078 * disk by bdflush, but the journaling mechanism will be suspended
2079 * immediately and no further transaction commits will be honoured.
2080 *
2081 * Any dirty, journaled buffers will be written back to disk without
2082 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2083 * filesystem, but we _do_ attempt to leave as much data as possible
2084 * behind for fsck to use for cleanup.
2085 *
2086 * Any attempt to get a new transaction handle on a journal which is in
2087 * ABORT state will just result in an -EROFS error return. A
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002088 * jbd2_journal_stop on an existing handle will return -EIO if we have
Dave Kleikamp470decc2006-10-11 01:20:57 -07002089 * entered abort state during the update.
2090 *
2091 * Recursive transactions are not disturbed by journal abort until the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002092 * final jbd2_journal_stop, which will receive the -EIO error.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002093 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002094 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
Dave Kleikamp470decc2006-10-11 01:20:57 -07002095 * which will be recorded (if possible) in the journal superblock. This
2096 * allows a client to record failure conditions in the middle of a
2097 * transaction without having to complete the transaction to record the
2098 * failure to disk. ext3_error, for example, now uses this
2099 * functionality.
2100 *
2101 * Errors which originate from within the journaling layer will NOT
2102 * supply an errno; a null errno implies that absolutely no further
2103 * writes are done to the journal (unless there are any already in
2104 * progress).
2105 *
2106 */
2107
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002108void jbd2_journal_abort(journal_t *journal, int errno)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002109{
2110 __journal_abort_soft(journal, errno);
2111}
2112
2113/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002114 * int jbd2_journal_errno () - returns the journal's error state.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002115 * @journal: journal to examine.
2116 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002117 * This is the errno number set with jbd2_journal_abort(), the last
Dave Kleikamp470decc2006-10-11 01:20:57 -07002118 * time the journal was mounted - if the journal was stopped
2119 * without calling abort this will be 0.
2120 *
2121 * If the journal has been aborted on this mount time -EROFS will
2122 * be returned.
2123 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002124int jbd2_journal_errno(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002125{
2126 int err;
2127
Theodore Ts'oa931da62010-08-03 21:35:12 -04002128 read_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002129 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002130 err = -EROFS;
2131 else
2132 err = journal->j_errno;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002133 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002134 return err;
2135}
2136
2137/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002138 * int jbd2_journal_clear_err () - clears the journal's error state
Dave Kleikamp470decc2006-10-11 01:20:57 -07002139 * @journal: journal to act on.
2140 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002141 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07002142 * mode.
2143 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002144int jbd2_journal_clear_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002145{
2146 int err = 0;
2147
Theodore Ts'oa931da62010-08-03 21:35:12 -04002148 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002149 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002150 err = -EROFS;
2151 else
2152 journal->j_errno = 0;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002153 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002154 return err;
2155}
2156
2157/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002158 * void jbd2_journal_ack_err() - Ack journal err.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002159 * @journal: journal to act on.
2160 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002161 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07002162 * mode.
2163 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002164void jbd2_journal_ack_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002165{
Theodore Ts'oa931da62010-08-03 21:35:12 -04002166 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002167 if (journal->j_errno)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002168 journal->j_flags |= JBD2_ACK_ERR;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002169 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002170}
2171
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002172int jbd2_journal_blocks_per_page(struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002173{
2174 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2175}
2176
2177/*
Zach Brownb517bea2006-10-11 01:21:08 -07002178 * helper functions to deal with 32 or 64bit block numbers.
2179 */
2180size_t journal_tag_bytes(journal_t *journal)
2181{
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002182 size_t sz;
2183
2184 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3))
2185 return sizeof(journal_block_tag3_t);
2186
2187 sz = sizeof(journal_block_tag_t);
Darrick J. Wongc3900872012-05-27 08:12:12 -04002188
2189 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002190 sz += sizeof(__u16);
Darrick J. Wongc3900872012-05-27 08:12:12 -04002191
Zach Brownb517bea2006-10-11 01:21:08 -07002192 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002193 return sz;
Zach Brownb517bea2006-10-11 01:21:08 -07002194 else
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002195 return sz - sizeof(__u32);
Zach Brownb517bea2006-10-11 01:21:08 -07002196}
2197
2198/*
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002199 * JBD memory management
2200 *
2201 * These functions are used to allocate block-sized chunks of memory
2202 * used for making copies of buffer_head data. Very often it will be
2203 * page-sized chunks of data, but sometimes it will be in
2204 * sub-page-size chunks. (For example, 16k pages on Power systems
2205 * with a 4k block file system.) For blocks smaller than a page, we
2206 * use a SLAB allocator. There are slab caches for each block size,
2207 * which are allocated at mount time, if necessary, and we only free
2208 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2209 * this reason we don't need to a mutex to protect access to
2210 * jbd2_slab[] allocating or releasing memory; only in
2211 * jbd2_journal_create_slab().
2212 */
2213#define JBD2_MAX_SLABS 8
2214static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002215
2216static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2217 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2218 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2219};
2220
2221
2222static void jbd2_journal_destroy_slabs(void)
2223{
2224 int i;
2225
2226 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2227 if (jbd2_slab[i])
2228 kmem_cache_destroy(jbd2_slab[i]);
2229 jbd2_slab[i] = NULL;
2230 }
2231}
2232
2233static int jbd2_journal_create_slab(size_t size)
2234{
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002235 static DEFINE_MUTEX(jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002236 int i = order_base_2(size) - 10;
2237 size_t slab_size;
2238
2239 if (size == PAGE_SIZE)
2240 return 0;
2241
2242 if (i >= JBD2_MAX_SLABS)
2243 return -EINVAL;
2244
2245 if (unlikely(i < 0))
2246 i = 0;
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002247 mutex_lock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002248 if (jbd2_slab[i]) {
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002249 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002250 return 0; /* Already created */
2251 }
2252
2253 slab_size = 1 << (i+10);
2254 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2255 slab_size, 0, NULL);
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002256 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002257 if (!jbd2_slab[i]) {
2258 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2259 return -ENOMEM;
2260 }
2261 return 0;
2262}
2263
2264static struct kmem_cache *get_slab(size_t size)
2265{
2266 int i = order_base_2(size) - 10;
2267
2268 BUG_ON(i >= JBD2_MAX_SLABS);
2269 if (unlikely(i < 0))
2270 i = 0;
Bill Pemberton8ac97b72010-04-30 09:34:31 -04002271 BUG_ON(jbd2_slab[i] == NULL);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002272 return jbd2_slab[i];
2273}
2274
2275void *jbd2_alloc(size_t size, gfp_t flags)
2276{
2277 void *ptr;
2278
2279 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2280
2281 flags |= __GFP_REPEAT;
2282 if (size == PAGE_SIZE)
2283 ptr = (void *)__get_free_pages(flags, 0);
2284 else if (size > PAGE_SIZE) {
2285 int order = get_order(size);
2286
2287 if (order < 3)
2288 ptr = (void *)__get_free_pages(flags, order);
2289 else
2290 ptr = vmalloc(size);
2291 } else
2292 ptr = kmem_cache_alloc(get_slab(size), flags);
2293
2294 /* Check alignment; SLUB has gotten this wrong in the past,
2295 * and this can lead to user data corruption! */
2296 BUG_ON(((unsigned long) ptr) & (size-1));
2297
2298 return ptr;
2299}
2300
2301void jbd2_free(void *ptr, size_t size)
2302{
2303 if (size == PAGE_SIZE) {
2304 free_pages((unsigned long)ptr, 0);
2305 return;
2306 }
2307 if (size > PAGE_SIZE) {
2308 int order = get_order(size);
2309
2310 if (order < 3)
2311 free_pages((unsigned long)ptr, order);
2312 else
2313 vfree(ptr);
2314 return;
2315 }
2316 kmem_cache_free(get_slab(size), ptr);
2317};
2318
2319/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07002320 * Journal_head storage management
2321 */
Christoph Lametere18b8902006-12-06 20:33:20 -08002322static struct kmem_cache *jbd2_journal_head_cache;
Jose R. Santose23291b2007-07-18 08:57:06 -04002323#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002324static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2325#endif
2326
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002327static int jbd2_journal_init_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002328{
2329 int retval;
2330
Al Viro1076d172008-03-29 03:07:18 +00002331 J_ASSERT(jbd2_journal_head_cache == NULL);
Johann Lombardia920e942006-10-11 01:21:00 -07002332 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002333 sizeof(struct journal_head),
2334 0, /* offset */
Mingming Cao77160952008-01-28 23:58:27 -05002335 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09002336 NULL); /* ctor */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002337 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00002338 if (!jbd2_journal_head_cache) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002339 retval = -ENOMEM;
Eryu Guanf2a44522011-11-01 19:09:18 -04002340 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002341 }
2342 return retval;
2343}
2344
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002345static void jbd2_journal_destroy_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002346{
Duane Griffin8a9362e2008-04-17 10:38:59 -04002347 if (jbd2_journal_head_cache) {
2348 kmem_cache_destroy(jbd2_journal_head_cache);
2349 jbd2_journal_head_cache = NULL;
2350 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002351}
2352
2353/*
2354 * journal_head splicing and dicing
2355 */
2356static struct journal_head *journal_alloc_journal_head(void)
2357{
2358 struct journal_head *ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002359
Jose R. Santose23291b2007-07-18 08:57:06 -04002360#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002361 atomic_inc(&nr_journal_heads);
2362#endif
Zheng Liu5d9cf9c2013-05-28 07:27:11 -04002363 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
Al Viro1076d172008-03-29 03:07:18 +00002364 if (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002365 jbd_debug(1, "out of memory for journal_head\n");
Theodore Ts'o670be5a2010-12-17 10:44:16 -05002366 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
Al Viro1076d172008-03-29 03:07:18 +00002367 while (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002368 yield();
Zheng Liu5d9cf9c2013-05-28 07:27:11 -04002369 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002370 }
2371 }
2372 return ret;
2373}
2374
2375static void journal_free_journal_head(struct journal_head *jh)
2376{
Jose R. Santose23291b2007-07-18 08:57:06 -04002377#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002378 atomic_dec(&nr_journal_heads);
Mingming Caocd02ff02007-10-16 18:38:25 -04002379 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -07002380#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002381 kmem_cache_free(jbd2_journal_head_cache, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002382}
2383
2384/*
2385 * A journal_head is attached to a buffer_head whenever JBD has an
2386 * interest in the buffer.
2387 *
2388 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2389 * is set. This bit is tested in core kernel code where we need to take
2390 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2391 * there.
2392 *
2393 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2394 *
2395 * When a buffer has its BH_JBD bit set it is immune from being released by
2396 * core kernel code, mainly via ->b_count.
2397 *
Jan Karade1b7942011-06-13 15:38:22 -04002398 * A journal_head is detached from its buffer_head when the journal_head's
2399 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2400 * transaction (b_cp_transaction) hold their references to b_jcount.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002401 *
2402 * Various places in the kernel want to attach a journal_head to a buffer_head
2403 * _before_ attaching the journal_head to a transaction. To protect the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002404 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
Dave Kleikamp470decc2006-10-11 01:20:57 -07002405 * journal_head's b_jcount refcount by one. The caller must call
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002406 * jbd2_journal_put_journal_head() to undo this.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002407 *
2408 * So the typical usage would be:
2409 *
2410 * (Attach a journal_head if needed. Increments b_jcount)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002411 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002412 * ...
Jan Karade1b7942011-06-13 15:38:22 -04002413 * (Get another reference for transaction)
2414 * jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002415 * jh->b_transaction = xxx;
Jan Karade1b7942011-06-13 15:38:22 -04002416 * (Put original reference)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002417 * jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002418 */
2419
2420/*
2421 * Give a buffer_head a journal_head.
2422 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07002423 * May sleep.
2424 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002425struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002426{
2427 struct journal_head *jh;
2428 struct journal_head *new_jh = NULL;
2429
2430repeat:
Zheng Liu5d9cf9c2013-05-28 07:27:11 -04002431 if (!buffer_jbd(bh))
Dave Kleikamp470decc2006-10-11 01:20:57 -07002432 new_jh = journal_alloc_journal_head();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002433
2434 jbd_lock_bh_journal_head(bh);
2435 if (buffer_jbd(bh)) {
2436 jh = bh2jh(bh);
2437 } else {
2438 J_ASSERT_BH(bh,
2439 (atomic_read(&bh->b_count) > 0) ||
2440 (bh->b_page && bh->b_page->mapping));
2441
2442 if (!new_jh) {
2443 jbd_unlock_bh_journal_head(bh);
2444 goto repeat;
2445 }
2446
2447 jh = new_jh;
2448 new_jh = NULL; /* We consumed it */
2449 set_buffer_jbd(bh);
2450 bh->b_private = jh;
2451 jh->b_bh = bh;
2452 get_bh(bh);
2453 BUFFER_TRACE(bh, "added journal_head");
2454 }
2455 jh->b_jcount++;
2456 jbd_unlock_bh_journal_head(bh);
2457 if (new_jh)
2458 journal_free_journal_head(new_jh);
2459 return bh->b_private;
2460}
2461
2462/*
2463 * Grab a ref against this buffer_head's journal_head. If it ended up not
2464 * having a journal_head, return NULL
2465 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002466struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002467{
2468 struct journal_head *jh = NULL;
2469
2470 jbd_lock_bh_journal_head(bh);
2471 if (buffer_jbd(bh)) {
2472 jh = bh2jh(bh);
2473 jh->b_jcount++;
2474 }
2475 jbd_unlock_bh_journal_head(bh);
2476 return jh;
2477}
2478
2479static void __journal_remove_journal_head(struct buffer_head *bh)
2480{
2481 struct journal_head *jh = bh2jh(bh);
2482
2483 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karade1b7942011-06-13 15:38:22 -04002484 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2485 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2486 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2487 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2488 J_ASSERT_BH(bh, buffer_jbd(bh));
2489 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2490 BUFFER_TRACE(bh, "remove journal_head");
2491 if (jh->b_frozen_data) {
2492 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2493 jbd2_free(jh->b_frozen_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002494 }
Jan Karade1b7942011-06-13 15:38:22 -04002495 if (jh->b_committed_data) {
2496 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2497 jbd2_free(jh->b_committed_data, bh->b_size);
2498 }
2499 bh->b_private = NULL;
2500 jh->b_bh = NULL; /* debug, really */
2501 clear_buffer_jbd(bh);
2502 journal_free_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002503}
2504
2505/*
Jan Karade1b7942011-06-13 15:38:22 -04002506 * Drop a reference on the passed journal_head. If it fell to zero then
Dave Kleikamp470decc2006-10-11 01:20:57 -07002507 * release the journal_head from the buffer_head.
2508 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002509void jbd2_journal_put_journal_head(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002510{
2511 struct buffer_head *bh = jh2bh(jh);
2512
2513 jbd_lock_bh_journal_head(bh);
2514 J_ASSERT_JH(jh, jh->b_jcount > 0);
2515 --jh->b_jcount;
Jan Karade1b7942011-06-13 15:38:22 -04002516 if (!jh->b_jcount) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002517 __journal_remove_journal_head(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002518 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002519 __brelse(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002520 } else
2521 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002522}
2523
2524/*
Jan Karac851ed52008-07-11 19:27:31 -04002525 * Initialize jbd inode head
2526 */
2527void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2528{
2529 jinode->i_transaction = NULL;
2530 jinode->i_next_transaction = NULL;
2531 jinode->i_vfs_inode = inode;
2532 jinode->i_flags = 0;
2533 INIT_LIST_HEAD(&jinode->i_list);
2534}
2535
2536/*
2537 * Function to be called before we start removing inode from memory (i.e.,
2538 * clear_inode() is a fine place to be called from). It removes inode from
2539 * transaction's lists.
2540 */
2541void jbd2_journal_release_jbd_inode(journal_t *journal,
2542 struct jbd2_inode *jinode)
2543{
Jan Karac851ed52008-07-11 19:27:31 -04002544 if (!journal)
2545 return;
2546restart:
2547 spin_lock(&journal->j_list_lock);
2548 /* Is commit writing out inode - we have to wait */
Brian King39e3ac22010-10-27 21:25:12 -04002549 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
Jan Karac851ed52008-07-11 19:27:31 -04002550 wait_queue_head_t *wq;
2551 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2552 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2553 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2554 spin_unlock(&journal->j_list_lock);
2555 schedule();
2556 finish_wait(wq, &wait.wait);
2557 goto restart;
2558 }
2559
Jan Karac851ed52008-07-11 19:27:31 -04002560 if (jinode->i_transaction) {
2561 list_del(&jinode->i_list);
2562 jinode->i_transaction = NULL;
2563 }
2564 spin_unlock(&journal->j_list_lock);
2565}
2566
Dave Kleikamp470decc2006-10-11 01:20:57 -07002567
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002568#ifdef CONFIG_PROC_FS
2569
2570#define JBD2_STATS_PROC_NAME "fs/jbd2"
2571
2572static void __init jbd2_create_jbd_stats_proc_entry(void)
2573{
2574 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2575}
2576
2577static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2578{
2579 if (proc_jbd2_stats)
2580 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2581}
2582
2583#else
2584
2585#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2586#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2587
2588#endif
2589
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002590struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002591
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002592static int __init jbd2_journal_init_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002593{
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002594 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002595 if (jbd2_handle_cache == NULL) {
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002596 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2597 return -ENOMEM;
2598 }
2599 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2600 if (jbd2_inode_cache == NULL) {
2601 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2602 kmem_cache_destroy(jbd2_handle_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002603 return -ENOMEM;
2604 }
2605 return 0;
2606}
2607
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002608static void jbd2_journal_destroy_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002609{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002610 if (jbd2_handle_cache)
2611 kmem_cache_destroy(jbd2_handle_cache);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002612 if (jbd2_inode_cache)
2613 kmem_cache_destroy(jbd2_inode_cache);
2614
Dave Kleikamp470decc2006-10-11 01:20:57 -07002615}
2616
2617/*
2618 * Module startup and shutdown
2619 */
2620
2621static int __init journal_init_caches(void)
2622{
2623 int ret;
2624
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002625 ret = jbd2_journal_init_revoke_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002626 if (ret == 0)
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002627 ret = jbd2_journal_init_journal_head_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002628 if (ret == 0)
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002629 ret = jbd2_journal_init_handle_cache();
Yongqiang Yang0c2022e2012-02-20 17:53:02 -05002630 if (ret == 0)
2631 ret = jbd2_journal_init_transaction_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002632 return ret;
2633}
2634
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002635static void jbd2_journal_destroy_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002636{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002637 jbd2_journal_destroy_revoke_caches();
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002638 jbd2_journal_destroy_journal_head_cache();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002639 jbd2_journal_destroy_handle_cache();
Yongqiang Yang0c2022e2012-02-20 17:53:02 -05002640 jbd2_journal_destroy_transaction_cache();
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002641 jbd2_journal_destroy_slabs();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002642}
2643
2644static int __init journal_init(void)
2645{
2646 int ret;
2647
2648 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2649
2650 ret = journal_init_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002651 if (ret == 0) {
Duane Griffin620de4e2008-04-29 22:02:47 -04002652 jbd2_create_jbd_stats_proc_entry();
2653 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002654 jbd2_journal_destroy_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002655 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002656 return ret;
2657}
2658
2659static void __exit journal_exit(void)
2660{
Jose R. Santose23291b2007-07-18 08:57:06 -04002661#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002662 int n = atomic_read(&nr_journal_heads);
2663 if (n)
Jan Kara75685072013-12-08 21:13:59 -05002664 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002665#endif
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002666 jbd2_remove_jbd_stats_proc_entry();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002667 jbd2_journal_destroy_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002668}
2669
Dave Kleikamp470decc2006-10-11 01:20:57 -07002670MODULE_LICENSE("GPL");
2671module_init(journal_init);
2672module_exit(journal_exit);
2673