blob: b31852f76f46585137021df6266a18ba743b8528 [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 Kara41617e12016-04-24 00:56:07 -040097EXPORT_SYMBOL(jbd2_journal_inode_add_write);
98EXPORT_SYMBOL(jbd2_journal_inode_add_wait);
Jan Karac851ed52008-07-11 19:27:31 -040099EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
100EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
101EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500102EXPORT_SYMBOL(jbd2_inode_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700103
Dave Kleikamp470decc2006-10-11 01:20:57 -0700104static void __journal_abort_soft (journal_t *journal, int errno);
Theodore Ts'od2eecb02009-12-07 10:36:20 -0500105static int jbd2_journal_create_slab(size_t slab_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700106
Paul Gortmaker169f1a22013-06-12 23:04:04 -0400107#ifdef CONFIG_JBD2_DEBUG
108void __jbd2_debug(int level, const char *file, const char *func,
109 unsigned int line, const char *fmt, ...)
110{
111 struct va_format vaf;
112 va_list args;
113
114 if (level > jbd2_journal_enable_debug)
115 return;
116 va_start(args, fmt);
117 vaf.fmt = fmt;
118 vaf.va = &args;
119 printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
120 va_end(args);
121}
122EXPORT_SYMBOL(__jbd2_debug);
123#endif
124
Darrick J. Wong25ed6e82012-05-27 07:48:56 -0400125/* Checksumming functions */
Rashika Kheria7747e6d2014-02-17 20:49:04 -0500126static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
Darrick J. Wong25ed6e82012-05-27 07:48:56 -0400127{
Darrick J. Wong85957982015-10-15 10:30:36 -0400128 if (!jbd2_journal_has_csum_v2or3_feature(j))
Darrick J. Wong25ed6e82012-05-27 07:48:56 -0400129 return 1;
130
131 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
132}
133
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400134static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400135{
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400136 __u32 csum;
137 __be32 old_csum;
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400138
139 old_csum = sb->s_checksum;
140 sb->s_checksum = 0;
141 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
142 sb->s_checksum = old_csum;
143
144 return cpu_to_be32(csum);
145}
146
Rashika Kheria7747e6d2014-02-17 20:49:04 -0500147static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400148{
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400149 if (!jbd2_journal_has_csum_v2or3(j))
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400150 return 1;
151
152 return sb->s_checksum == jbd2_superblock_csum(j, sb);
153}
154
Rashika Kheria7747e6d2014-02-17 20:49:04 -0500155static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400156{
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400157 if (!jbd2_journal_has_csum_v2or3(j))
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -0400158 return;
159
160 sb->s_checksum = jbd2_superblock_csum(j, sb);
161}
162
Dave Kleikamp470decc2006-10-11 01:20:57 -0700163/*
164 * Helper function used to manage commit timeouts
165 */
166
167static void commit_timeout(unsigned long __data)
168{
169 struct task_struct * p = (struct task_struct *) __data;
170
171 wake_up_process(p);
172}
173
174/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700175 * kjournald2: The main thread function used to manage a logging device
Dave Kleikamp470decc2006-10-11 01:20:57 -0700176 * journal.
177 *
178 * This kernel thread is responsible for two things:
179 *
180 * 1) COMMIT: Every so often we need to commit the current state of the
181 * filesystem to disk. The journal thread is responsible for writing
182 * all of the metadata buffers to disk.
183 *
184 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
185 * of the data in that part of the log has been rewritten elsewhere on
186 * the disk. Flushing these old buffers to reclaim space in the log is
187 * known as checkpointing, and this thread is responsible for that job.
188 */
189
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700190static int kjournald2(void *arg)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700191{
192 journal_t *journal = arg;
193 transaction_t *transaction;
194
195 /*
196 * Set up an interval timer which can be used to trigger a commit wakeup
197 * after the commit interval expires
198 */
199 setup_timer(&journal->j_commit_timer, commit_timeout,
200 (unsigned long)current);
201
Nigel Cunningham35c80422012-02-03 19:59:41 +1100202 set_freezable();
203
Dave Kleikamp470decc2006-10-11 01:20:57 -0700204 /* Record that the journal thread is running */
205 journal->j_task = current;
206 wake_up(&journal->j_wait_done_commit);
207
Dave Kleikamp470decc2006-10-11 01:20:57 -0700208 /*
209 * And now, wait forever for commit wakeup events.
210 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400211 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700212
213loop:
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700214 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700215 goto end_loop;
216
217 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
218 journal->j_commit_sequence, journal->j_commit_request);
219
220 if (journal->j_commit_sequence != journal->j_commit_request) {
221 jbd_debug(1, "OK, requests differ\n");
Theodore Ts'oa931da62010-08-03 21:35:12 -0400222 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700223 del_timer_sync(&journal->j_commit_timer);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700224 jbd2_journal_commit_transaction(journal);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400225 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700226 goto loop;
227 }
228
229 wake_up(&journal->j_wait_done_commit);
230 if (freezing(current)) {
231 /*
232 * The simpler the better. Flushing journal isn't a
233 * good idea, because that depends on threads that may
234 * be already stopped.
235 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700236 jbd_debug(1, "Now suspending kjournald2\n");
Theodore Ts'oa931da62010-08-03 21:35:12 -0400237 write_unlock(&journal->j_state_lock);
Tejun Heoa0acae02011-11-21 12:32:22 -0800238 try_to_freeze();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400239 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700240 } else {
241 /*
242 * We assume on resume that commits are already there,
243 * so we don't sleep
244 */
245 DEFINE_WAIT(wait);
246 int should_sleep = 1;
247
248 prepare_to_wait(&journal->j_wait_commit, &wait,
249 TASK_INTERRUPTIBLE);
250 if (journal->j_commit_sequence != journal->j_commit_request)
251 should_sleep = 0;
252 transaction = journal->j_running_transaction;
253 if (transaction && time_after_eq(jiffies,
254 transaction->t_expires))
255 should_sleep = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700256 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700257 should_sleep = 0;
258 if (should_sleep) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400259 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700260 schedule();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400261 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700262 }
263 finish_wait(&journal->j_wait_commit, &wait);
264 }
265
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700266 jbd_debug(1, "kjournald2 wakes\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700267
268 /*
269 * Were we woken up by a commit wakeup event?
270 */
271 transaction = journal->j_running_transaction;
272 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
273 journal->j_commit_request = transaction->t_tid;
274 jbd_debug(1, "woke because of timeout\n");
275 }
276 goto loop;
277
278end_loop:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400279 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700280 del_timer_sync(&journal->j_commit_timer);
281 journal->j_task = NULL;
282 wake_up(&journal->j_wait_done_commit);
283 jbd_debug(1, "Journal thread exiting.\n");
284 return 0;
285}
286
Pavel Emelianov97f06782007-05-08 00:30:42 -0700287static int jbd2_journal_start_thread(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700288{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700289 struct task_struct *t;
290
Theodore Ts'o90576c02009-09-29 15:51:30 -0400291 t = kthread_run(kjournald2, journal, "jbd2/%s",
292 journal->j_devname);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700293 if (IS_ERR(t))
294 return PTR_ERR(t);
295
Al Viro1076d172008-03-29 03:07:18 +0000296 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700297 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700298}
299
300static void journal_kill_thread(journal_t *journal)
301{
Theodore Ts'oa931da62010-08-03 21:35:12 -0400302 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700303 journal->j_flags |= JBD2_UNMOUNT;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700304
305 while (journal->j_task) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400306 write_unlock(&journal->j_state_lock);
Theodore Ts'o3469a322014-03-08 19:11:36 -0500307 wake_up(&journal->j_wait_commit);
Al Viro1076d172008-03-29 03:07:18 +0000308 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400309 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700310 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400311 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700312}
313
314/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700315 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700316 *
317 * Writes a metadata buffer to a given disk block. The actual IO is not
318 * performed but a new buffer_head is constructed which labels the data
319 * to be written with the correct destination disk block.
320 *
321 * Any magic-number escaping which needs to be done will cause a
322 * copy-out here. If the buffer happens to start with the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700323 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700324 * magic number is only written to the log for descripter blocks. In
325 * this case, we copy the data and replace the first word with 0, and we
326 * return a result code which indicates that this buffer needs to be
327 * marked as an escaped buffer in the corresponding log descriptor
328 * block. The missing word can then be restored when the block is read
329 * during recovery.
330 *
331 * If the source buffer has already been modified by a new transaction
332 * since we took the last commit snapshot, we use the frozen copy of
Jan Karaf5113ef2013-06-04 12:01:45 -0400333 * that data for IO. If we end up using the existing buffer_head's data
334 * for the write, then we have to make sure nobody modifies it while the
335 * IO is in progress. do_get_write_access() handles this.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700336 *
Jan Karaf5113ef2013-06-04 12:01:45 -0400337 * The function returns a pointer to the buffer_head to be used for IO.
338 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700339 *
340 * Return value:
341 * <0: Error
342 * >=0: Finished OK
343 *
344 * On success:
345 * Bit 0 set == escape performed on the data
346 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
347 */
348
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700349int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700350 struct journal_head *jh_in,
Jan Karaf5113ef2013-06-04 12:01:45 -0400351 struct buffer_head **bh_out,
352 sector_t blocknr)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700353{
354 int need_copy_out = 0;
355 int done_copy_out = 0;
356 int do_escape = 0;
357 char *mapped_data;
358 struct buffer_head *new_bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700359 struct page *new_page;
360 unsigned int new_offset;
361 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghua96577c42009-07-13 17:55:35 -0400362 journal_t *journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700363
364 /*
365 * The buffer really shouldn't be locked: only the current committing
366 * transaction is allowed to write it, so nobody else is allowed
367 * to do any IO.
368 *
369 * akpm: except if we're journalling data, and write() output is
370 * also part of a shared mapping, and another thread has
371 * decided to launch a writepage() against this buffer.
372 */
373 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
374
Michal Hocko6ccaf3e2015-06-08 10:53:10 -0400375 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
Theodore Ts'o47def822010-07-27 11:56:05 -0400376
dingdinghua96577c42009-07-13 17:55:35 -0400377 /* keep subsequent assertions sane */
dingdinghua96577c42009-07-13 17:55:35 -0400378 atomic_set(&new_bh->b_count, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700379
Jan Karaf5113ef2013-06-04 12:01:45 -0400380 jbd_lock_bh_state(bh_in);
381repeat:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700382 /*
383 * If a new transaction has already done a buffer copy-out, then
384 * we use that version of the data for the commit.
385 */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700386 if (jh_in->b_frozen_data) {
387 done_copy_out = 1;
388 new_page = virt_to_page(jh_in->b_frozen_data);
389 new_offset = offset_in_page(jh_in->b_frozen_data);
390 } else {
391 new_page = jh2bh(jh_in)->b_page;
392 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
393 }
394
Cong Wang303a8f22011-11-25 23:14:31 +0800395 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700396 /*
Jan Kara13ceef02010-07-14 07:56:33 +0200397 * Fire data frozen trigger if data already wasn't frozen. Do this
398 * before checking for escaping, as the trigger may modify the magic
399 * offset. If a copy-out happens afterwards, it will have the correct
400 * data in the buffer.
Joel Beckere06c8222008-09-11 15:35:47 -0700401 */
Jan Kara13ceef02010-07-14 07:56:33 +0200402 if (!done_copy_out)
403 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
404 jh_in->b_triggers);
Joel Beckere06c8222008-09-11 15:35:47 -0700405
406 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700407 * Check for escaping
408 */
409 if (*((__be32 *)(mapped_data + new_offset)) ==
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700410 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700411 need_copy_out = 1;
412 do_escape = 1;
413 }
Cong Wang303a8f22011-11-25 23:14:31 +0800414 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700415
416 /*
417 * Do we need to do a data copy?
418 */
419 if (need_copy_out && !done_copy_out) {
420 char *tmp;
421
422 jbd_unlock_bh_state(bh_in);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400423 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500424 if (!tmp) {
Jan Karaf5113ef2013-06-04 12:01:45 -0400425 brelse(new_bh);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500426 return -ENOMEM;
427 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700428 jbd_lock_bh_state(bh_in);
429 if (jh_in->b_frozen_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400430 jbd2_free(tmp, bh_in->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700431 goto repeat;
432 }
433
434 jh_in->b_frozen_data = tmp;
Cong Wang303a8f22011-11-25 23:14:31 +0800435 mapped_data = kmap_atomic(new_page);
Jan Karaf5113ef2013-06-04 12:01:45 -0400436 memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
Cong Wang303a8f22011-11-25 23:14:31 +0800437 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700438
439 new_page = virt_to_page(tmp);
440 new_offset = offset_in_page(tmp);
441 done_copy_out = 1;
Joel Beckere06c8222008-09-11 15:35:47 -0700442
443 /*
444 * This isn't strictly necessary, as we're using frozen
445 * data for the escaping, but it keeps consistency with
446 * b_frozen_data usage.
447 */
448 jh_in->b_frozen_triggers = jh_in->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700449 }
450
451 /*
452 * Did we need to do an escaping? Now we've done all the
453 * copying, we can finally do so.
454 */
455 if (do_escape) {
Cong Wang303a8f22011-11-25 23:14:31 +0800456 mapped_data = kmap_atomic(new_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700457 *((unsigned int *)(mapped_data + new_offset)) = 0;
Cong Wang303a8f22011-11-25 23:14:31 +0800458 kunmap_atomic(mapped_data);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700459 }
460
Dave Kleikamp470decc2006-10-11 01:20:57 -0700461 set_bh_page(new_bh, new_page, new_offset);
Jan Karaf5113ef2013-06-04 12:01:45 -0400462 new_bh->b_size = bh_in->b_size;
463 new_bh->b_bdev = journal->j_dev;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700464 new_bh->b_blocknr = blocknr;
Jan Karab34090e2013-06-04 12:08:56 -0400465 new_bh->b_private = bh_in;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700466 set_buffer_mapped(new_bh);
467 set_buffer_dirty(new_bh);
468
Jan Karaf5113ef2013-06-04 12:01:45 -0400469 *bh_out = new_bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700470
471 /*
472 * The to-be-written buffer needs to get moved to the io queue,
473 * and the original buffer whose contents we are shadowing or
474 * copying is moved to the transaction's shadow queue.
475 */
476 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghua96577c42009-07-13 17:55:35 -0400477 spin_lock(&journal->j_list_lock);
478 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
479 spin_unlock(&journal->j_list_lock);
Jan Karab34090e2013-06-04 12:08:56 -0400480 set_buffer_shadow(bh_in);
dingdinghua96577c42009-07-13 17:55:35 -0400481 jbd_unlock_bh_state(bh_in);
482
Dave Kleikamp470decc2006-10-11 01:20:57 -0700483 return do_escape | (done_copy_out << 1);
484}
485
486/*
487 * Allocation code for the journal file. Manage the space left in the
488 * journal, so that we can begin checkpointing when appropriate.
489 */
490
491/*
Theodore Ts'oe4471832011-02-12 08:18:24 -0500492 * Called with j_state_lock locked for writing.
493 * Returns true if a transaction commit was started.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700494 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700495int __jbd2_log_start_commit(journal_t *journal, tid_t target)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700496{
Eric Sandeene7b04ac2013-01-30 00:39:28 -0500497 /* Return if the txn has already requested to be committed */
498 if (journal->j_commit_request == target)
499 return 0;
500
Dave Kleikamp470decc2006-10-11 01:20:57 -0700501 /*
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400502 * The only transaction we can possibly wait upon is the
503 * currently running transaction (if it exists). Otherwise,
504 * the target tid must be an old one.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700505 */
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400506 if (journal->j_running_transaction &&
507 journal->j_running_transaction->t_tid == target) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700508 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200509 * We want a new commit: OK, mark the request and wakeup the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700510 * commit thread. We do _not_ do the commit ourselves.
511 */
512
513 journal->j_commit_request = target;
Eryu Guanf2a44522011-11-01 19:09:18 -0400514 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700515 journal->j_commit_request,
516 journal->j_commit_sequence);
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500517 journal->j_running_transaction->t_requested = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700518 wake_up(&journal->j_wait_commit);
519 return 1;
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400520 } else if (!tid_geq(journal->j_commit_request, target))
521 /* This should never happen, but if it does, preserve
522 the evidence before kjournald goes into a loop and
523 increments j_commit_sequence beyond all recognition. */
Eryu Guanf2a44522011-11-01 19:09:18 -0400524 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
Theodore Ts'o1be2add2011-05-08 19:37:54 -0400525 journal->j_commit_request,
526 journal->j_commit_sequence,
527 target, journal->j_running_transaction ?
528 journal->j_running_transaction->t_tid : 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700529 return 0;
530}
531
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700532int jbd2_log_start_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700533{
534 int ret;
535
Theodore Ts'oa931da62010-08-03 21:35:12 -0400536 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700537 ret = __jbd2_log_start_commit(journal, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400538 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700539 return ret;
540}
541
542/*
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400543 * Force and wait any uncommitted transactions. We can only force the running
544 * transaction if we don't have an active handle, otherwise, we will deadlock.
545 * Returns: <0 in case of error,
546 * 0 if nothing to commit,
547 * 1 if transaction was successfully committed.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700548 */
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400549static int __jbd2_journal_force_commit(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700550{
551 transaction_t *transaction = NULL;
552 tid_t tid;
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400553 int need_to_start = 0, ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700554
Theodore Ts'oa931da62010-08-03 21:35:12 -0400555 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700556 if (journal->j_running_transaction && !current->journal_info) {
557 transaction = journal->j_running_transaction;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500558 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
559 need_to_start = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700560 } else if (journal->j_committing_transaction)
561 transaction = journal->j_committing_transaction;
562
563 if (!transaction) {
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400564 /* Nothing to commit */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400565 read_unlock(&journal->j_state_lock);
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400566 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700567 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700568 tid = transaction->t_tid;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400569 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500570 if (need_to_start)
571 jbd2_log_start_commit(journal, tid);
Dmitry Monakhov9ff86442013-06-12 22:24:07 -0400572 ret = jbd2_log_wait_commit(journal, tid);
573 if (!ret)
574 ret = 1;
575
576 return ret;
577}
578
579/**
580 * Force and wait upon a commit if the calling process is not within
581 * transaction. This is used for forcing out undo-protected data which contains
582 * bitmaps, when the fs is running out of space.
583 *
584 * @journal: journal to force
585 * Returns true if progress was made.
586 */
587int jbd2_journal_force_commit_nested(journal_t *journal)
588{
589 int ret;
590
591 ret = __jbd2_journal_force_commit(journal);
592 return ret > 0;
593}
594
595/**
596 * int journal_force_commit() - force any uncommitted transactions
597 * @journal: journal to force
598 *
599 * Caller want unconditional commit. We can only force the running transaction
600 * if we don't have an active handle, otherwise, we will deadlock.
601 */
602int jbd2_journal_force_commit(journal_t *journal)
603{
604 int ret;
605
606 J_ASSERT(!current->journal_info);
607 ret = __jbd2_journal_force_commit(journal);
608 if (ret > 0)
609 ret = 0;
610 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700611}
612
613/*
614 * Start a commit of the current running transaction (if any). Returns true
Jan Karac88ccea2009-02-10 11:27:46 -0500615 * if a transaction is going to be committed (or is currently already
616 * committing), and fills its tid in at *ptid
Dave Kleikamp470decc2006-10-11 01:20:57 -0700617 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700618int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700619{
620 int ret = 0;
621
Theodore Ts'oa931da62010-08-03 21:35:12 -0400622 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700623 if (journal->j_running_transaction) {
624 tid_t tid = journal->j_running_transaction->t_tid;
625
Jan Karac88ccea2009-02-10 11:27:46 -0500626 __jbd2_log_start_commit(journal, tid);
627 /* There's a running transaction and we've just made sure
628 * it's commit has been scheduled. */
629 if (ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700630 *ptid = tid;
Jan Karac88ccea2009-02-10 11:27:46 -0500631 ret = 1;
632 } else if (journal->j_committing_transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700633 /*
Artem Bityutskiy12810ad2012-07-25 18:12:07 +0300634 * If commit has been started, then we have to wait for
635 * completion of that transaction.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700636 */
Jan Karac88ccea2009-02-10 11:27:46 -0500637 if (ptid)
638 *ptid = journal->j_committing_transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700639 ret = 1;
640 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400641 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700642 return ret;
643}
644
645/*
Jan Karabbd2be32011-05-24 11:59:18 -0400646 * Return 1 if a given transaction has not yet sent barrier request
647 * connected with a transaction commit. If 0 is returned, transaction
648 * may or may not have sent the barrier. Used to avoid sending barrier
649 * twice in common cases.
650 */
651int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
652{
653 int ret = 0;
654 transaction_t *commit_trans;
655
656 if (!(journal->j_flags & JBD2_BARRIER))
657 return 0;
658 read_lock(&journal->j_state_lock);
659 /* Transaction already committed? */
660 if (tid_geq(journal->j_commit_sequence, tid))
661 goto out;
662 commit_trans = journal->j_committing_transaction;
663 if (!commit_trans || commit_trans->t_tid != tid) {
664 ret = 1;
665 goto out;
666 }
667 /*
668 * Transaction is being committed and we already proceeded to
669 * submitting a flush to fs partition?
670 */
671 if (journal->j_fs_dev != journal->j_dev) {
672 if (!commit_trans->t_need_data_flush ||
673 commit_trans->t_state >= T_COMMIT_DFLUSH)
674 goto out;
675 } else {
676 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
677 goto out;
678 }
679 ret = 1;
680out:
681 read_unlock(&journal->j_state_lock);
682 return ret;
683}
684EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
685
686/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700687 * Wait for a specified commit to complete.
688 * The caller may not hold the journal lock.
689 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700690int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700691{
692 int err = 0;
693
Theodore Ts'oa931da62010-08-03 21:35:12 -0400694 read_lock(&journal->j_state_lock);
Jose R. Santose23291b2007-07-18 08:57:06 -0400695#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -0700696 if (!tid_geq(journal->j_commit_request, tid)) {
Jan Kara75685072013-12-08 21:13:59 -0500697 printk(KERN_ERR
Dave Kleikamp470decc2006-10-11 01:20:57 -0700698 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400699 __func__, journal->j_commit_request, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700700 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700701#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -0700702 while (tid_gt(tid, journal->j_commit_sequence)) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400703 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700704 tid, journal->j_commit_sequence);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400705 read_unlock(&journal->j_state_lock);
Theodore Ts'o3469a322014-03-08 19:11:36 -0500706 wake_up(&journal->j_wait_commit);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700707 wait_event(journal->j_wait_done_commit,
708 !tid_gt(tid, journal->j_commit_sequence));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400709 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700710 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400711 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700712
Jan Kara75685072013-12-08 21:13:59 -0500713 if (unlikely(is_journal_aborted(journal)))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700714 err = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700715 return err;
716}
717
718/*
Theodore Ts'od76a3a772013-04-03 22:02:52 -0400719 * When this function returns the transaction corresponding to tid
720 * will be completed. If the transaction has currently running, start
721 * committing that transaction before waiting for it to complete. If
722 * the transaction id is stale, it is by definition already completed,
723 * so just return SUCCESS.
724 */
725int jbd2_complete_transaction(journal_t *journal, tid_t tid)
726{
727 int need_to_wait = 1;
728
729 read_lock(&journal->j_state_lock);
730 if (journal->j_running_transaction &&
731 journal->j_running_transaction->t_tid == tid) {
732 if (journal->j_commit_request != tid) {
733 /* transaction not yet started, so request it */
734 read_unlock(&journal->j_state_lock);
735 jbd2_log_start_commit(journal, tid);
736 goto wait_commit;
737 }
738 } else if (!(journal->j_committing_transaction &&
739 journal->j_committing_transaction->t_tid == tid))
740 need_to_wait = 0;
741 read_unlock(&journal->j_state_lock);
742 if (!need_to_wait)
743 return 0;
744wait_commit:
745 return jbd2_log_wait_commit(journal, tid);
746}
747EXPORT_SYMBOL(jbd2_complete_transaction);
748
749/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700750 * Log buffer allocation routines:
751 */
752
Mingming Cao18eba7a2006-10-11 01:21:13 -0700753int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700754{
755 unsigned long blocknr;
756
Theodore Ts'oa931da62010-08-03 21:35:12 -0400757 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700758 J_ASSERT(journal->j_free > 1);
759
760 blocknr = journal->j_head;
761 journal->j_head++;
762 journal->j_free--;
763 if (journal->j_head == journal->j_last)
764 journal->j_head = journal->j_first;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400765 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700766 return jbd2_journal_bmap(journal, blocknr, retp);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700767}
768
769/*
770 * Conversion of logical to physical block numbers for the journal
771 *
772 * On external journals the journal blocks are identity-mapped, so
773 * this is a no-op. If needed, we can use j_blk_offset - everything is
774 * ready.
775 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700776int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700777 unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700778{
779 int err = 0;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700780 unsigned long long ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700781
782 if (journal->j_inode) {
783 ret = bmap(journal->j_inode, blocknr);
784 if (ret)
785 *retp = ret;
786 else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700787 printk(KERN_ALERT "%s: journal block not found "
788 "at offset %lu on %s\n",
Theodore Ts'o05496762008-09-16 14:36:17 -0400789 __func__, blocknr, journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700790 err = -EIO;
791 __journal_abort_soft(journal, err);
792 }
793 } else {
794 *retp = blocknr; /* +journal->j_blk_offset */
795 }
796 return err;
797}
798
799/*
800 * We play buffer_head aliasing tricks to write data/metadata blocks to
801 * the journal without copying their contents, but for journal
802 * descriptor blocks we do need to generate bona fide buffers.
803 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700804 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
Dave Kleikamp470decc2006-10-11 01:20:57 -0700805 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
806 * But we don't bother doing that, so there will be coherency problems with
807 * mmaps of blockdevs which hold live JBD-controlled filesystems.
808 */
Jan Kara32ab6712016-02-22 23:17:15 -0500809struct buffer_head *
810jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700811{
Jan Kara32ab6712016-02-22 23:17:15 -0500812 journal_t *journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700813 struct buffer_head *bh;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700814 unsigned long long blocknr;
Jan Kara32ab6712016-02-22 23:17:15 -0500815 journal_header_t *header;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700816 int err;
817
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700818 err = jbd2_journal_next_log_block(journal, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700819
820 if (err)
821 return NULL;
822
823 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500824 if (!bh)
825 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700826 lock_buffer(bh);
827 memset(bh->b_data, 0, journal->j_blocksize);
Jan Kara32ab6712016-02-22 23:17:15 -0500828 header = (journal_header_t *)bh->b_data;
829 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
830 header->h_blocktype = cpu_to_be32(type);
831 header->h_sequence = cpu_to_be32(transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700832 set_buffer_uptodate(bh);
833 unlock_buffer(bh);
834 BUFFER_TRACE(bh, "return this buffer");
Jan Karae5a120a2013-06-04 12:06:01 -0400835 return bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700836}
837
Jan Kara1101cd42016-02-22 23:19:09 -0500838void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
839{
840 struct jbd2_journal_block_tail *tail;
841 __u32 csum;
842
843 if (!jbd2_journal_has_csum_v2or3(j))
844 return;
845
846 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
847 sizeof(struct jbd2_journal_block_tail));
848 tail->t_checksum = 0;
849 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
850 tail->t_checksum = cpu_to_be32(csum);
851}
852
Jan Kara79feb522012-03-13 22:22:54 -0400853/*
854 * Return tid of the oldest transaction in the journal and block in the journal
855 * where the transaction starts.
856 *
857 * If the journal is now empty, return which will be the next transaction ID
858 * we will write and where will that transaction start.
859 *
860 * The return value is 0 if journal tail cannot be pushed any further, 1 if
861 * it can.
862 */
863int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
864 unsigned long *block)
865{
866 transaction_t *transaction;
867 int ret;
868
869 read_lock(&journal->j_state_lock);
870 spin_lock(&journal->j_list_lock);
871 transaction = journal->j_checkpoint_transactions;
872 if (transaction) {
873 *tid = transaction->t_tid;
874 *block = transaction->t_log_start;
875 } else if ((transaction = journal->j_committing_transaction) != NULL) {
876 *tid = transaction->t_tid;
877 *block = transaction->t_log_start;
878 } else if ((transaction = journal->j_running_transaction) != NULL) {
879 *tid = transaction->t_tid;
880 *block = journal->j_head;
881 } else {
882 *tid = journal->j_transaction_sequence;
883 *block = journal->j_head;
884 }
885 ret = tid_gt(*tid, journal->j_tail_sequence);
886 spin_unlock(&journal->j_list_lock);
887 read_unlock(&journal->j_state_lock);
888
889 return ret;
890}
891
892/*
893 * Update information in journal structure and in on disk journal superblock
894 * about log tail. This function does not check whether information passed in
895 * really pushes log tail further. It's responsibility of the caller to make
896 * sure provided log tail information is valid (e.g. by holding
897 * j_checkpoint_mutex all the time between computing log tail and calling this
898 * function as is the case with jbd2_cleanup_journal_tail()).
899 *
900 * Requires j_checkpoint_mutex
901 */
Joseph Qi6f6a6fd2015-06-15 14:36:01 -0400902int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
Jan Kara79feb522012-03-13 22:22:54 -0400903{
904 unsigned long freed;
Joseph Qi6f6a6fd2015-06-15 14:36:01 -0400905 int ret;
Jan Kara79feb522012-03-13 22:22:54 -0400906
907 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
908
909 /*
910 * We cannot afford for write to remain in drive's caches since as
911 * soon as we update j_tail, next transaction can start reusing journal
912 * space and if we lose sb update during power failure we'd replay
913 * old transaction with possibly newly overwritten data.
914 */
Joseph Qi6f6a6fd2015-06-15 14:36:01 -0400915 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
916 if (ret)
917 goto out;
918
Jan Kara79feb522012-03-13 22:22:54 -0400919 write_lock(&journal->j_state_lock);
920 freed = block - journal->j_tail;
921 if (block < journal->j_tail)
922 freed += journal->j_last - journal->j_first;
923
924 trace_jbd2_update_log_tail(journal, tid, block, freed);
925 jbd_debug(1,
926 "Cleaning journal tail from %d to %d (offset %lu), "
927 "freeing %lu\n",
928 journal->j_tail_sequence, tid, block, freed);
929
930 journal->j_free += freed;
931 journal->j_tail_sequence = tid;
932 journal->j_tail = block;
933 write_unlock(&journal->j_state_lock);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -0400934
935out:
936 return ret;
Jan Kara79feb522012-03-13 22:22:54 -0400937}
938
Jan Kara33395782012-03-13 22:45:38 -0400939/*
940 * This is a variaon of __jbd2_update_log_tail which checks for validity of
941 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
942 * with other threads updating log tail.
943 */
944void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
945{
946 mutex_lock(&journal->j_checkpoint_mutex);
947 if (tid_gt(tid, journal->j_tail_sequence))
948 __jbd2_update_log_tail(journal, tid, block);
949 mutex_unlock(&journal->j_checkpoint_mutex);
950}
951
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500952struct jbd2_stats_proc_session {
953 journal_t *journal;
954 struct transaction_stats_s *stats;
955 int start;
956 int max;
957};
958
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500959static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
960{
961 return *pos ? NULL : SEQ_START_TOKEN;
962}
963
964static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
965{
966 return NULL;
967}
968
969static int jbd2_seq_info_show(struct seq_file *seq, void *v)
970{
971 struct jbd2_stats_proc_session *s = seq->private;
972
973 if (v != SEQ_START_TOKEN)
974 return 0;
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500975 seq_printf(seq, "%lu transactions (%lu requested), "
976 "each up to %u blocks\n",
977 s->stats->ts_tid, s->stats->ts_requested,
978 s->journal->j_max_transaction_buffers);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500979 if (s->stats->ts_tid == 0)
980 return 0;
981 seq_printf(seq, "average: \n %ums waiting for transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400982 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500983 seq_printf(seq, " %ums request delay\n",
984 (s->stats->ts_requested == 0) ? 0 :
985 jiffies_to_msecs(s->stats->run.rs_request_delay /
986 s->stats->ts_requested));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500987 seq_printf(seq, " %ums running transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400988 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500989 seq_printf(seq, " %ums transaction was being locked\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400990 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500991 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400992 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500993 seq_printf(seq, " %ums logging transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400994 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -0500995 seq_printf(seq, " %lluus average transaction commit time\n",
996 div_u64(s->journal->j_average_commit_time, 1000));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500997 seq_printf(seq, " %lu handles per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400998 s->stats->run.rs_handle_count / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500999 seq_printf(seq, " %lu blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -04001000 s->stats->run.rs_blocks / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001001 seq_printf(seq, " %lu logged blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -04001002 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001003 return 0;
1004}
1005
1006static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1007{
1008}
1009
James Morris88e9d342009-09-22 16:43:43 -07001010static const struct seq_operations jbd2_seq_info_ops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001011 .start = jbd2_seq_info_start,
1012 .next = jbd2_seq_info_next,
1013 .stop = jbd2_seq_info_stop,
1014 .show = jbd2_seq_info_show,
1015};
1016
1017static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1018{
Al Virod9dda782013-03-31 18:16:14 -04001019 journal_t *journal = PDE_DATA(inode);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001020 struct jbd2_stats_proc_session *s;
1021 int rc, size;
1022
1023 s = kmalloc(sizeof(*s), GFP_KERNEL);
1024 if (s == NULL)
1025 return -ENOMEM;
1026 size = sizeof(struct transaction_stats_s);
1027 s->stats = kmalloc(size, GFP_KERNEL);
1028 if (s->stats == NULL) {
1029 kfree(s);
1030 return -ENOMEM;
1031 }
1032 spin_lock(&journal->j_history_lock);
1033 memcpy(s->stats, &journal->j_stats, size);
1034 s->journal = journal;
1035 spin_unlock(&journal->j_history_lock);
1036
1037 rc = seq_open(file, &jbd2_seq_info_ops);
1038 if (rc == 0) {
1039 struct seq_file *m = file->private_data;
1040 m->private = s;
1041 } else {
1042 kfree(s->stats);
1043 kfree(s);
1044 }
1045 return rc;
1046
1047}
1048
1049static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1050{
1051 struct seq_file *seq = file->private_data;
1052 struct jbd2_stats_proc_session *s = seq->private;
1053 kfree(s->stats);
1054 kfree(s);
1055 return seq_release(inode, file);
1056}
1057
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001058static const struct file_operations jbd2_seq_info_fops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001059 .owner = THIS_MODULE,
1060 .open = jbd2_seq_info_open,
1061 .read = seq_read,
1062 .llseek = seq_lseek,
1063 .release = jbd2_seq_info_release,
1064};
1065
1066static struct proc_dir_entry *proc_jbd2_stats;
1067
1068static void jbd2_stats_proc_init(journal_t *journal)
1069{
Theodore Ts'o05496762008-09-16 14:36:17 -04001070 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001071 if (journal->j_proc_entry) {
Denis V. Lunev79da3662008-04-29 01:02:11 -07001072 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1073 &jbd2_seq_info_fops, journal);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001074 }
1075}
1076
1077static void jbd2_stats_proc_exit(journal_t *journal)
1078{
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001079 remove_proc_entry("info", journal->j_proc_entry);
Theodore Ts'o05496762008-09-16 14:36:17 -04001080 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001081}
1082
Dave Kleikamp470decc2006-10-11 01:20:57 -07001083/*
1084 * Management for journal control blocks: functions to create and
1085 * destroy journal_t structures, and to initialise and read existing
1086 * journal blocks from disk. */
1087
1088/* First: create and setup a journal_t object in memory. We initialise
1089 * very few fields yet: that has to wait until we have created the
1090 * journal structures from from scratch, or loaded them from disk. */
1091
1092static journal_t * journal_init_common (void)
1093{
1094 journal_t *journal;
1095 int err;
1096
Andrew Morton3ebfdf82009-12-23 08:05:15 -05001097 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001098 if (!journal)
Theodore Ts'ob7271b02010-12-18 13:39:38 -05001099 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001100
1101 init_waitqueue_head(&journal->j_wait_transaction_locked);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001102 init_waitqueue_head(&journal->j_wait_done_commit);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001103 init_waitqueue_head(&journal->j_wait_commit);
1104 init_waitqueue_head(&journal->j_wait_updates);
Jan Kara8f7d89f2013-06-04 12:35:11 -04001105 init_waitqueue_head(&journal->j_wait_reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001106 mutex_init(&journal->j_barrier);
1107 mutex_init(&journal->j_checkpoint_mutex);
1108 spin_lock_init(&journal->j_revoke_lock);
1109 spin_lock_init(&journal->j_list_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001110 rwlock_init(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001111
Mingming Caocd02ff02007-10-16 18:38:25 -04001112 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
Theodore Ts'o30773842009-01-03 20:27:38 -05001113 journal->j_min_batch_time = 0;
1114 journal->j_max_batch_time = 15000; /* 15ms */
Jan Kara8f7d89f2013-06-04 12:35:11 -04001115 atomic_set(&journal->j_reserved_credits, 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001116
1117 /* The journal is marked for error until we succeed with recovery! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001118 journal->j_flags = JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001119
1120 /* Set up a default-sized revoke table for the new mount. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001121 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001122 if (err) {
1123 kfree(journal);
Theodore Ts'ob7271b02010-12-18 13:39:38 -05001124 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001125 }
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001126
Theodore Ts'obf699322009-09-30 00:32:06 -04001127 spin_lock_init(&journal->j_history_lock);
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001128
Dave Kleikamp470decc2006-10-11 01:20:57 -07001129 return journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001130}
1131
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001132/* jbd2_journal_init_dev and jbd2_journal_init_inode:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001133 *
1134 * Create a journal structure assigned some fixed set of disk blocks to
1135 * the journal. We don't actually touch those disk blocks yet, but we
1136 * need to set up all of the mapping information to tell the journaling
1137 * system where the journal blocks are.
1138 *
1139 */
1140
1141/**
Randy Dunlap5648ba52008-04-17 10:38:59 -04001142 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
Dave Kleikamp470decc2006-10-11 01:20:57 -07001143 * @bdev: Block device on which to create the journal
1144 * @fs_dev: Device which hold journalled filesystem for this journal.
1145 * @start: Block nr Start of journal.
1146 * @len: Length of the journal in blocks.
1147 * @blocksize: blocksize of journalling device
Randy Dunlap5648ba52008-04-17 10:38:59 -04001148 *
1149 * Returns: a newly created journal_t *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001150 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001151 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
Dave Kleikamp470decc2006-10-11 01:20:57 -07001152 * range of blocks on an arbitrary block device.
1153 *
1154 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001155journal_t * jbd2_journal_init_dev(struct block_device *bdev,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001156 struct block_device *fs_dev,
Mingming Cao18eba7a2006-10-11 01:21:13 -07001157 unsigned long long start, int len, int blocksize)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001158{
1159 journal_t *journal = journal_init_common();
1160 struct buffer_head *bh;
1161 int n;
1162
1163 if (!journal)
1164 return NULL;
1165
1166 /* journal descriptor can store up to n blocks -bzzz */
1167 journal->j_blocksize = blocksize;
yangsheng0587aa32010-11-17 21:46:26 -05001168 journal->j_dev = bdev;
1169 journal->j_fs_dev = fs_dev;
1170 journal->j_blk_offset = start;
1171 journal->j_maxlen = len;
1172 bdevname(journal->j_dev, journal->j_devname);
Rasmus Villemoes81ae3942015-06-25 15:02:38 -07001173 strreplace(journal->j_devname, '/', '!');
Jan Kara4b905672009-01-06 14:53:35 -05001174 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001175 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1176 journal->j_wbufsize = n;
1177 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1178 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001179 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001180 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001181 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001182 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001183
1184 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001185 if (!bh) {
1186 printk(KERN_ERR
1187 "%s: Cannot get buffer for journal superblock\n",
1188 __func__);
1189 goto out_err;
1190 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001191 journal->j_sb_buffer = bh;
1192 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Kara4b905672009-01-06 14:53:35 -05001193
Dave Kleikamp470decc2006-10-11 01:20:57 -07001194 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001195out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001196 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001197 jbd2_stats_proc_exit(journal);
1198 kfree(journal);
1199 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001200}
1201
1202/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001203 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001204 * @inode: An inode to create the journal in
1205 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001206 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
Dave Kleikamp470decc2006-10-11 01:20:57 -07001207 * the journal. The inode must exist already, must support bmap() and
1208 * must have all data blocks preallocated.
1209 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001210journal_t * jbd2_journal_init_inode (struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001211{
1212 struct buffer_head *bh;
1213 journal_t *journal = journal_init_common();
Theodore Ts'o05496762008-09-16 14:36:17 -04001214 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001215 int err;
1216 int n;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001217 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001218
1219 if (!journal)
1220 return NULL;
1221
1222 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1223 journal->j_inode = inode;
Theodore Ts'o05496762008-09-16 14:36:17 -04001224 bdevname(journal->j_dev, journal->j_devname);
Rasmus Villemoes81ae3942015-06-25 15:02:38 -07001225 p = strreplace(journal->j_devname, '/', '!');
Theodore Ts'o90576c02009-09-29 15:51:30 -04001226 sprintf(p, "-%lu", journal->j_inode->i_ino);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001227 jbd_debug(1,
1228 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1229 journal, inode->i_sb->s_id, inode->i_ino,
1230 (long long) inode->i_size,
1231 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1232
1233 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1234 journal->j_blocksize = inode->i_sb->s_blocksize;
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001235 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001236
1237 /* journal descriptor can store up to n blocks -bzzz */
1238 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1239 journal->j_wbufsize = n;
1240 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1241 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001242 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001243 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001244 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001245 }
1246
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001247 err = jbd2_journal_bmap(journal, 0, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001248 /* If that failed, give up */
1249 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -08001250 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001251 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001252 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001253 }
1254
Gioh Kima49058f2014-09-04 22:36:35 -04001255 bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001256 if (!bh) {
1257 printk(KERN_ERR
1258 "%s: Cannot get buffer for journal superblock\n",
1259 __func__);
1260 goto out_err;
1261 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001262 journal->j_sb_buffer = bh;
1263 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1264
1265 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001266out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001267 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001268 jbd2_stats_proc_exit(journal);
1269 kfree(journal);
1270 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001271}
1272
1273/*
1274 * If the journal init or create aborts, we need to mark the journal
1275 * superblock as being NULL to prevent the journal destroy from writing
1276 * back a bogus superblock.
1277 */
1278static void journal_fail_superblock (journal_t *journal)
1279{
1280 struct buffer_head *bh = journal->j_sb_buffer;
1281 brelse(bh);
1282 journal->j_sb_buffer = NULL;
1283}
1284
1285/*
1286 * Given a journal_t structure, initialise the various fields for
1287 * startup of a new journaling session. We use this both when creating
1288 * a journal, and after recovering an old journal to reset it for
1289 * subsequent use.
1290 */
1291
1292static int journal_reset(journal_t *journal)
1293{
1294 journal_superblock_t *sb = journal->j_superblock;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001295 unsigned long long first, last;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001296
1297 first = be32_to_cpu(sb->s_first);
1298 last = be32_to_cpu(sb->s_maxlen);
Jan Karaf6f50e22009-07-17 10:40:01 -04001299 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001300 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
Jan Karaf6f50e22009-07-17 10:40:01 -04001301 first, last);
1302 journal_fail_superblock(journal);
1303 return -EINVAL;
1304 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001305
1306 journal->j_first = first;
1307 journal->j_last = last;
1308
1309 journal->j_head = first;
1310 journal->j_tail = first;
1311 journal->j_free = last - first;
1312
1313 journal->j_tail_sequence = journal->j_transaction_sequence;
1314 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1315 journal->j_commit_request = journal->j_commit_sequence;
1316
1317 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1318
Dave Kleikamp470decc2006-10-11 01:20:57 -07001319 /*
1320 * As a special case, if the on-disk copy is already marked as needing
Jan Kara24bcc892012-03-13 15:41:04 -04001321 * no recovery (s_start == 0), then we can safely defer the superblock
1322 * update until the next commit by setting JBD2_FLUSHED. This avoids
Dave Kleikamp470decc2006-10-11 01:20:57 -07001323 * attempting a write to a potential-readonly device.
1324 */
Jan Kara24bcc892012-03-13 15:41:04 -04001325 if (sb->s_start == 0) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001326 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
Dave Kleikamp470decc2006-10-11 01:20:57 -07001327 "(start %ld, seq %d, errno %d)\n",
1328 journal->j_tail, journal->j_tail_sequence,
1329 journal->j_errno);
Jan Kara24bcc892012-03-13 15:41:04 -04001330 journal->j_flags |= JBD2_FLUSHED;
1331 } else {
Jan Karaa78bb112012-03-13 15:43:04 -04001332 /* Lock here to make assertions happy... */
1333 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara79feb522012-03-13 22:22:54 -04001334 /*
1335 * Update log tail information. We use WRITE_FUA since new
1336 * transaction will start reusing journal space and so we
1337 * must make sure information about current log tail is on
1338 * disk before that.
1339 */
1340 jbd2_journal_update_sb_log_tail(journal,
1341 journal->j_tail_sequence,
1342 journal->j_tail,
1343 WRITE_FUA);
Jan Karaa78bb112012-03-13 15:43:04 -04001344 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001345 }
Jan Kara24bcc892012-03-13 15:41:04 -04001346 return jbd2_journal_start_thread(journal);
1347}
Dave Kleikamp470decc2006-10-11 01:20:57 -07001348
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001349static int jbd2_write_superblock(journal_t *journal, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001350{
1351 struct buffer_head *bh = journal->j_sb_buffer;
Theodore Ts'ofe52d172013-07-01 08:12:38 -04001352 journal_superblock_t *sb = journal->j_superblock;
Jan Kara79feb522012-03-13 22:22:54 -04001353 int ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001354
Jan Kara79feb522012-03-13 22:22:54 -04001355 trace_jbd2_write_superblock(journal, write_op);
1356 if (!(journal->j_flags & JBD2_BARRIER))
1357 write_op &= ~(REQ_FUA | REQ_FLUSH);
1358 lock_buffer(bh);
Theodore Ts'o914258b2008-10-06 21:35:40 -04001359 if (buffer_write_io_error(bh)) {
1360 /*
1361 * Oh, dear. A previous attempt to write the journal
1362 * superblock failed. This could happen because the
1363 * USB device was yanked out. Or it could happen to
1364 * be a transient write error and maybe the block will
1365 * be remapped. Nothing we can do but to retry the
1366 * write and hope for the best.
1367 */
1368 printk(KERN_ERR "JBD2: previous I/O error detected "
1369 "for journal superblock update for %s.\n",
1370 journal->j_devname);
1371 clear_buffer_write_io_error(bh);
1372 set_buffer_uptodate(bh);
1373 }
Theodore Ts'ofe52d172013-07-01 08:12:38 -04001374 jbd2_superblock_csum_set(journal, sb);
Jan Kara79feb522012-03-13 22:22:54 -04001375 get_bh(bh);
1376 bh->b_end_io = end_buffer_write_sync;
1377 ret = submit_bh(write_op, bh);
1378 wait_on_buffer(bh);
Jan Kara24bcc892012-03-13 15:41:04 -04001379 if (buffer_write_io_error(bh)) {
Jan Kara24bcc892012-03-13 15:41:04 -04001380 clear_buffer_write_io_error(bh);
1381 set_buffer_uptodate(bh);
Jan Kara79feb522012-03-13 22:22:54 -04001382 ret = -EIO;
1383 }
1384 if (ret) {
1385 printk(KERN_ERR "JBD2: Error %d detected when updating "
1386 "journal superblock for %s.\n", ret,
1387 journal->j_devname);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001388 jbd2_journal_abort(journal, ret);
Jan Kara24bcc892012-03-13 15:41:04 -04001389 }
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001390
1391 return ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001392}
Theodore Ts'o914258b2008-10-06 21:35:40 -04001393
Jan Kara24bcc892012-03-13 15:41:04 -04001394/**
1395 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1396 * @journal: The journal to update.
Jan Kara79feb522012-03-13 22:22:54 -04001397 * @tail_tid: TID of the new transaction at the tail of the log
1398 * @tail_block: The first block of the transaction at the tail of the log
1399 * @write_op: With which operation should we write the journal sb
Jan Kara24bcc892012-03-13 15:41:04 -04001400 *
1401 * Update a journal's superblock information about log tail and write it to
1402 * disk, waiting for the IO to complete.
1403 */
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001404int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
Jan Kara79feb522012-03-13 22:22:54 -04001405 unsigned long tail_block, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001406{
1407 journal_superblock_t *sb = journal->j_superblock;
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001408 int ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001409
Jan Karaa78bb112012-03-13 15:43:04 -04001410 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Jan Kara79feb522012-03-13 22:22:54 -04001411 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1412 tail_block, tail_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001413
Jan Kara79feb522012-03-13 22:22:54 -04001414 sb->s_sequence = cpu_to_be32(tail_tid);
1415 sb->s_start = cpu_to_be32(tail_block);
Jan Kara24bcc892012-03-13 15:41:04 -04001416
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001417 ret = jbd2_write_superblock(journal, write_op);
1418 if (ret)
1419 goto out;
Jan Kara24bcc892012-03-13 15:41:04 -04001420
1421 /* Log is no longer empty */
1422 write_lock(&journal->j_state_lock);
1423 WARN_ON(!sb->s_sequence);
1424 journal->j_flags &= ~JBD2_FLUSHED;
1425 write_unlock(&journal->j_state_lock);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001426
1427out:
1428 return ret;
Jan Kara24bcc892012-03-13 15:41:04 -04001429}
1430
1431/**
1432 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1433 * @journal: The journal to update.
OGAWA Hirofumic0a2ad92016-03-09 23:47:25 -05001434 * @write_op: With which operation should we write the journal sb
Jan Kara24bcc892012-03-13 15:41:04 -04001435 *
1436 * Update a journal's dynamic superblock fields to show that journal is empty.
1437 * Write updated superblock to disk waiting for IO to complete.
1438 */
OGAWA Hirofumic0a2ad92016-03-09 23:47:25 -05001439static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
Jan Kara24bcc892012-03-13 15:41:04 -04001440{
1441 journal_superblock_t *sb = journal->j_superblock;
1442
Jan Karaa78bb112012-03-13 15:43:04 -04001443 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001444 read_lock(&journal->j_state_lock);
Eric Sandeeneeecef02012-08-18 22:29:40 -04001445 /* Is it already empty? */
1446 if (sb->s_start == 0) {
1447 read_unlock(&journal->j_state_lock);
1448 return;
1449 }
Jan Kara24bcc892012-03-13 15:41:04 -04001450 jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1451 journal->j_tail_sequence);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001452
1453 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
Jan Kara24bcc892012-03-13 15:41:04 -04001454 sb->s_start = cpu_to_be32(0);
1455 read_unlock(&journal->j_state_lock);
1456
OGAWA Hirofumic0a2ad92016-03-09 23:47:25 -05001457 jbd2_write_superblock(journal, write_op);
Jan Kara24bcc892012-03-13 15:41:04 -04001458
1459 /* Log is no longer empty */
1460 write_lock(&journal->j_state_lock);
1461 journal->j_flags |= JBD2_FLUSHED;
1462 write_unlock(&journal->j_state_lock);
1463}
1464
1465
1466/**
1467 * jbd2_journal_update_sb_errno() - Update error in the journal.
1468 * @journal: The journal to update.
1469 *
1470 * Update a journal's errno. Write updated superblock to disk waiting for IO
1471 * to complete.
1472 */
Theodore Ts'od796c522012-08-05 19:04:57 -04001473void jbd2_journal_update_sb_errno(journal_t *journal)
Jan Kara24bcc892012-03-13 15:41:04 -04001474{
1475 journal_superblock_t *sb = journal->j_superblock;
1476
1477 read_lock(&journal->j_state_lock);
1478 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1479 journal->j_errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001480 sb->s_errno = cpu_to_be32(journal->j_errno);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001481 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001482
Daeho Jeong564bc402015-07-23 09:46:11 -04001483 jbd2_write_superblock(journal, WRITE_FUA);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001484}
Theodore Ts'od796c522012-08-05 19:04:57 -04001485EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001486
1487/*
1488 * Read the superblock for a given journal, performing initial
1489 * validation of the format.
1490 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001491static int journal_get_superblock(journal_t *journal)
1492{
1493 struct buffer_head *bh;
1494 journal_superblock_t *sb;
1495 int err = -EIO;
1496
1497 bh = journal->j_sb_buffer;
1498
1499 J_ASSERT(bh != NULL);
1500 if (!buffer_uptodate(bh)) {
1501 ll_rw_block(READ, 1, &bh);
1502 wait_on_buffer(bh);
1503 if (!buffer_uptodate(bh)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001504 printk(KERN_ERR
1505 "JBD2: IO error reading journal superblock\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001506 goto out;
1507 }
1508 }
1509
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001510 if (buffer_verified(bh))
1511 return 0;
1512
Dave Kleikamp470decc2006-10-11 01:20:57 -07001513 sb = journal->j_superblock;
1514
1515 err = -EINVAL;
1516
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001517 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001518 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001519 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001520 goto out;
1521 }
1522
1523 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001524 case JBD2_SUPERBLOCK_V1:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001525 journal->j_format_version = 1;
1526 break;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001527 case JBD2_SUPERBLOCK_V2:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001528 journal->j_format_version = 2;
1529 break;
1530 default:
Eryu Guanf2a44522011-11-01 19:09:18 -04001531 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001532 goto out;
1533 }
1534
1535 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1536 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1537 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001538 printk(KERN_WARNING "JBD2: journal file too short\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001539 goto out;
1540 }
1541
Eryu Guan87622022011-11-01 19:04:59 -04001542 if (be32_to_cpu(sb->s_first) == 0 ||
1543 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1544 printk(KERN_WARNING
1545 "JBD2: Invalid start block of journal: %u\n",
1546 be32_to_cpu(sb->s_first));
1547 goto out;
1548 }
1549
Darrick J. Wong56316a02015-10-17 16:18:45 -04001550 if (jbd2_has_feature_csum2(journal) &&
1551 jbd2_has_feature_csum3(journal)) {
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001552 /* Can't have checksum v2 and v3 at the same time! */
1553 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1554 "at the same time!\n");
1555 goto out;
1556 }
1557
Darrick J. Wong85957982015-10-15 10:30:36 -04001558 if (jbd2_journal_has_csum_v2or3_feature(journal) &&
Darrick J. Wong56316a02015-10-17 16:18:45 -04001559 jbd2_has_feature_checksum(journal)) {
Darrick J. Wongfeb8c6d2014-09-11 11:38:21 -04001560 /* Can't have checksum v1 and v2 on at the same time! */
1561 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1562 "at the same time!\n");
1563 goto out;
1564 }
1565
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001566 if (!jbd2_verify_csum_type(journal, sb)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001567 printk(KERN_ERR "JBD2: Unknown checksum type\n");
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001568 goto out;
1569 }
1570
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001571 /* Load the checksum driver */
Darrick J. Wong85957982015-10-15 10:30:36 -04001572 if (jbd2_journal_has_csum_v2or3_feature(journal)) {
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001573 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1574 if (IS_ERR(journal->j_chksum_driver)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001575 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001576 err = PTR_ERR(journal->j_chksum_driver);
1577 journal->j_chksum_driver = NULL;
1578 goto out;
1579 }
1580 }
1581
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001582 /* Check superblock checksum */
1583 if (!jbd2_superblock_csum_verify(journal, sb)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001584 printk(KERN_ERR "JBD2: journal checksum error\n");
Darrick J. Wong6a797d22015-10-17 16:16:04 -04001585 err = -EFSBADCRC;
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001586 goto out;
1587 }
1588
1589 /* Precompute checksum seed for all metadata */
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001590 if (jbd2_journal_has_csum_v2or3(journal))
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001591 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1592 sizeof(sb->s_uuid));
1593
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001594 set_buffer_verified(bh);
1595
Dave Kleikamp470decc2006-10-11 01:20:57 -07001596 return 0;
1597
1598out:
1599 journal_fail_superblock(journal);
1600 return err;
1601}
1602
1603/*
1604 * Load the on-disk journal superblock and read the key fields into the
1605 * journal_t.
1606 */
1607
1608static int load_superblock(journal_t *journal)
1609{
1610 int err;
1611 journal_superblock_t *sb;
1612
1613 err = journal_get_superblock(journal);
1614 if (err)
1615 return err;
1616
1617 sb = journal->j_superblock;
1618
1619 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1620 journal->j_tail = be32_to_cpu(sb->s_start);
1621 journal->j_first = be32_to_cpu(sb->s_first);
1622 journal->j_last = be32_to_cpu(sb->s_maxlen);
1623 journal->j_errno = be32_to_cpu(sb->s_errno);
1624
1625 return 0;
1626}
1627
1628
1629/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001630 * int jbd2_journal_load() - Read journal from disk.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001631 * @journal: Journal to act on.
1632 *
1633 * Given a journal_t structure which tells us which disk blocks contain
1634 * a journal, read the journal from disk to initialise the in-memory
1635 * structures.
1636 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001637int jbd2_journal_load(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001638{
1639 int err;
1640 journal_superblock_t *sb;
1641
1642 err = load_superblock(journal);
1643 if (err)
1644 return err;
1645
1646 sb = journal->j_superblock;
1647 /* If this is a V2 superblock, then we have to check the
1648 * features flags on it. */
1649
1650 if (journal->j_format_version >= 2) {
1651 if ((sb->s_feature_ro_compat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001652 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001653 (sb->s_feature_incompat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001654 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001655 printk(KERN_WARNING
1656 "JBD2: Unrecognised features on journal\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001657 return -EINVAL;
1658 }
1659 }
1660
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001661 /*
1662 * Create a slab for this blocksize
1663 */
1664 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1665 if (err)
1666 return err;
1667
Dave Kleikamp470decc2006-10-11 01:20:57 -07001668 /* Let the recovery code check whether it needs to recover any
1669 * data from the journal. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001670 if (jbd2_journal_recover(journal))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001671 goto recovery_error;
1672
Theodore Ts'oe6a47422009-11-15 15:31:37 -05001673 if (journal->j_failed_commit) {
1674 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1675 "is corrupt.\n", journal->j_failed_commit,
1676 journal->j_devname);
Darrick J. Wong6a797d22015-10-17 16:16:04 -04001677 return -EFSCORRUPTED;
Theodore Ts'oe6a47422009-11-15 15:31:37 -05001678 }
1679
Dave Kleikamp470decc2006-10-11 01:20:57 -07001680 /* OK, we've finished with the dynamic journal bits:
1681 * reinitialise the dynamic contents of the superblock in memory
1682 * and reset them on disk. */
1683 if (journal_reset(journal))
1684 goto recovery_error;
1685
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001686 journal->j_flags &= ~JBD2_ABORT;
1687 journal->j_flags |= JBD2_LOADED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001688 return 0;
1689
1690recovery_error:
Eryu Guanf2a44522011-11-01 19:09:18 -04001691 printk(KERN_WARNING "JBD2: recovery failed\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001692 return -EIO;
1693}
1694
1695/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001696 * void jbd2_journal_destroy() - Release a journal_t structure.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001697 * @journal: Journal to act on.
1698 *
1699 * Release a journal_t structure once it is no longer in use by the
1700 * journaled object.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001701 * Return <0 if we couldn't clean up the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001702 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001703int jbd2_journal_destroy(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001704{
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001705 int err = 0;
1706
Dave Kleikamp470decc2006-10-11 01:20:57 -07001707 /* Wait for the commit thread to wake up and die. */
1708 journal_kill_thread(journal);
1709
1710 /* Force a final log commit */
1711 if (journal->j_running_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001712 jbd2_journal_commit_transaction(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001713
1714 /* Force any old transactions to disk */
1715
1716 /* Totally anal locking here... */
1717 spin_lock(&journal->j_list_lock);
1718 while (journal->j_checkpoint_transactions != NULL) {
1719 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001720 mutex_lock(&journal->j_checkpoint_mutex);
Jan Kara841df7df2015-07-28 14:57:14 -04001721 err = jbd2_log_do_checkpoint(journal);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001722 mutex_unlock(&journal->j_checkpoint_mutex);
Jan Kara841df7df2015-07-28 14:57:14 -04001723 /*
1724 * If checkpointing failed, just free the buffers to avoid
1725 * looping forever
1726 */
1727 if (err) {
1728 jbd2_journal_destroy_checkpoint(journal);
1729 spin_lock(&journal->j_list_lock);
1730 break;
1731 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001732 spin_lock(&journal->j_list_lock);
1733 }
1734
1735 J_ASSERT(journal->j_running_transaction == NULL);
1736 J_ASSERT(journal->j_committing_transaction == NULL);
1737 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1738 spin_unlock(&journal->j_list_lock);
1739
Dave Kleikamp470decc2006-10-11 01:20:57 -07001740 if (journal->j_sb_buffer) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001741 if (!is_journal_aborted(journal)) {
Jan Karaa78bb112012-03-13 15:43:04 -04001742 mutex_lock(&journal->j_checkpoint_mutex);
OGAWA Hirofumic0a2ad92016-03-09 23:47:25 -05001743
1744 write_lock(&journal->j_state_lock);
1745 journal->j_tail_sequence =
1746 ++journal->j_transaction_sequence;
1747 write_unlock(&journal->j_state_lock);
1748
1749 jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA);
Jan Karaa78bb112012-03-13 15:43:04 -04001750 mutex_unlock(&journal->j_checkpoint_mutex);
1751 } else
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001752 err = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001753 brelse(journal->j_sb_buffer);
1754 }
1755
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001756 if (journal->j_proc_entry)
1757 jbd2_stats_proc_exit(journal);
Theodore Ts'od9f39d12014-11-25 20:02:37 -05001758 iput(journal->j_inode);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001759 if (journal->j_revoke)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001760 jbd2_journal_destroy_revoke(journal);
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001761 if (journal->j_chksum_driver)
1762 crypto_free_shash(journal->j_chksum_driver);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001763 kfree(journal->j_wbuf);
1764 kfree(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001765
1766 return err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001767}
1768
1769
1770/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001771 *int jbd2_journal_check_used_features () - Check if features specified are used.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001772 * @journal: Journal to check.
1773 * @compat: bitmask of compatible features
1774 * @ro: bitmask of features that force read-only mount
1775 * @incompat: bitmask of incompatible features
1776 *
1777 * Check whether the journal uses all of a given set of
1778 * features. Return true (non-zero) if it does.
1779 **/
1780
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001781int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001782 unsigned long ro, unsigned long incompat)
1783{
1784 journal_superblock_t *sb;
1785
1786 if (!compat && !ro && !incompat)
1787 return 1;
Patrick J. LoPresti1113e1b2010-07-22 15:04:16 -07001788 /* Load journal superblock if it is not loaded yet. */
1789 if (journal->j_format_version == 0 &&
1790 journal_get_superblock(journal) != 0)
1791 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001792 if (journal->j_format_version == 1)
1793 return 0;
1794
1795 sb = journal->j_superblock;
1796
1797 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1798 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1799 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1800 return 1;
1801
1802 return 0;
1803}
1804
1805/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001806 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001807 * @journal: Journal to check.
1808 * @compat: bitmask of compatible features
1809 * @ro: bitmask of features that force read-only mount
1810 * @incompat: bitmask of incompatible features
1811 *
1812 * Check whether the journaling code supports the use of
1813 * all of a given set of features on this journal. Return true
1814 * (non-zero) if it can. */
1815
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001816int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001817 unsigned long ro, unsigned long incompat)
1818{
Dave Kleikamp470decc2006-10-11 01:20:57 -07001819 if (!compat && !ro && !incompat)
1820 return 1;
1821
Dave Kleikamp470decc2006-10-11 01:20:57 -07001822 /* We can support any known requested features iff the
1823 * superblock is in version 2. Otherwise we fail to support any
1824 * extended sb features. */
1825
1826 if (journal->j_format_version != 2)
1827 return 0;
1828
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001829 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1830 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1831 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001832 return 1;
1833
1834 return 0;
1835}
1836
1837/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001838 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
Dave Kleikamp470decc2006-10-11 01:20:57 -07001839 * @journal: Journal to act on.
1840 * @compat: bitmask of compatible features
1841 * @ro: bitmask of features that force read-only mount
1842 * @incompat: bitmask of incompatible features
1843 *
1844 * Mark a given journal feature as present on the
1845 * superblock. Returns true if the requested features could be set.
1846 *
1847 */
1848
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001849int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001850 unsigned long ro, unsigned long incompat)
1851{
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001852#define INCOMPAT_FEATURE_ON(f) \
1853 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1854#define COMPAT_FEATURE_ON(f) \
1855 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001856 journal_superblock_t *sb;
1857
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001858 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001859 return 1;
1860
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001861 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001862 return 0;
1863
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001864 /* If enabling v2 checksums, turn on v3 instead */
1865 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1866 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1867 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1868 }
1869
1870 /* Asking for checksumming v3 and v1? Only give them v3. */
1871 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001872 compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1873 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1874
Dave Kleikamp470decc2006-10-11 01:20:57 -07001875 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1876 compat, ro, incompat);
1877
1878 sb = journal->j_superblock;
1879
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001880 /* If enabling v3 checksums, update superblock */
1881 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001882 sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1883 sb->s_feature_compat &=
1884 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001885
1886 /* Load the checksum driver */
1887 if (journal->j_chksum_driver == NULL) {
1888 journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1889 0, 0);
1890 if (IS_ERR(journal->j_chksum_driver)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001891 printk(KERN_ERR "JBD2: Cannot load crc32c "
Darrick J. Wong01b5adc2012-05-27 07:50:56 -04001892 "driver.\n");
1893 journal->j_chksum_driver = NULL;
1894 return 0;
1895 }
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001896
Darrick J. Wong32f38692014-12-01 16:22:23 -08001897 /* Precompute checksum seed for all metadata */
Darrick J. Wong4fd5ea42012-05-27 08:08:22 -04001898 journal->j_csum_seed = jbd2_chksum(journal, ~0,
1899 sb->s_uuid,
1900 sizeof(sb->s_uuid));
Darrick J. Wong32f38692014-12-01 16:22:23 -08001901 }
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001902 }
1903
1904 /* If enabling v1 checksums, downgrade superblock */
1905 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1906 sb->s_feature_incompat &=
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04001907 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1908 JBD2_FEATURE_INCOMPAT_CSUM_V3);
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001909
Dave Kleikamp470decc2006-10-11 01:20:57 -07001910 sb->s_feature_compat |= cpu_to_be32(compat);
1911 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1912 sb->s_feature_incompat |= cpu_to_be32(incompat);
1913
1914 return 1;
Darrick J. Wong25ed6e82012-05-27 07:48:56 -04001915#undef COMPAT_FEATURE_ON
1916#undef INCOMPAT_FEATURE_ON
Dave Kleikamp470decc2006-10-11 01:20:57 -07001917}
1918
Girish Shilamkar818d2762008-01-28 23:58:27 -05001919/*
1920 * jbd2_journal_clear_features () - Clear a given journal feature in the
1921 * superblock
1922 * @journal: Journal to act on.
1923 * @compat: bitmask of compatible features
1924 * @ro: bitmask of features that force read-only mount
1925 * @incompat: bitmask of incompatible features
1926 *
1927 * Clear a given journal feature as present on the
1928 * superblock.
1929 */
1930void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1931 unsigned long ro, unsigned long incompat)
1932{
1933 journal_superblock_t *sb;
1934
1935 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1936 compat, ro, incompat);
1937
1938 sb = journal->j_superblock;
1939
1940 sb->s_feature_compat &= ~cpu_to_be32(compat);
1941 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1942 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1943}
1944EXPORT_SYMBOL(jbd2_journal_clear_features);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001945
1946/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001947 * int jbd2_journal_flush () - Flush journal
Dave Kleikamp470decc2006-10-11 01:20:57 -07001948 * @journal: Journal to act on.
1949 *
1950 * Flush all data for a given journal to disk and empty the journal.
1951 * Filesystems can use this when remounting readonly to ensure that
1952 * recovery does not need to happen on remount.
1953 */
1954
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001955int jbd2_journal_flush(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001956{
1957 int err = 0;
1958 transaction_t *transaction = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001959
Theodore Ts'oa931da62010-08-03 21:35:12 -04001960 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001961
1962 /* Force everything buffered to the log... */
1963 if (journal->j_running_transaction) {
1964 transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001965 __jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001966 } else if (journal->j_committing_transaction)
1967 transaction = journal->j_committing_transaction;
1968
1969 /* Wait for the log commit to complete... */
1970 if (transaction) {
1971 tid_t tid = transaction->t_tid;
1972
Theodore Ts'oa931da62010-08-03 21:35:12 -04001973 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001974 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001975 } else {
Theodore Ts'oa931da62010-08-03 21:35:12 -04001976 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001977 }
1978
1979 /* ...and flush everything in the log out to disk. */
1980 spin_lock(&journal->j_list_lock);
1981 while (!err && journal->j_checkpoint_transactions != NULL) {
1982 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001983 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001984 err = jbd2_log_do_checkpoint(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001985 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001986 spin_lock(&journal->j_list_lock);
1987 }
1988 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001989
1990 if (is_journal_aborted(journal))
1991 return -EIO;
1992
Jan Karaa78bb112012-03-13 15:43:04 -04001993 mutex_lock(&journal->j_checkpoint_mutex);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04001994 if (!err) {
1995 err = jbd2_cleanup_journal_tail(journal);
1996 if (err < 0) {
1997 mutex_unlock(&journal->j_checkpoint_mutex);
1998 goto out;
1999 }
2000 err = 0;
2001 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002002
2003 /* Finally, mark the journal as really needing no recovery.
2004 * This sets s_start==0 in the underlying superblock, which is
2005 * the magic code for a fully-recovered superblock. Any future
2006 * commits of data to the journal will restore the current
2007 * s_start value. */
OGAWA Hirofumic0a2ad92016-03-09 23:47:25 -05002008 jbd2_mark_journal_empty(journal, WRITE_FUA);
Jan Karaa78bb112012-03-13 15:43:04 -04002009 mutex_unlock(&journal->j_checkpoint_mutex);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002010 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002011 J_ASSERT(!journal->j_running_transaction);
2012 J_ASSERT(!journal->j_committing_transaction);
2013 J_ASSERT(!journal->j_checkpoint_transactions);
2014 J_ASSERT(journal->j_head == journal->j_tail);
2015 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002016 write_unlock(&journal->j_state_lock);
Joseph Qi6f6a6fd2015-06-15 14:36:01 -04002017out:
2018 return err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002019}
2020
2021/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002022 * int jbd2_journal_wipe() - Wipe journal contents
Dave Kleikamp470decc2006-10-11 01:20:57 -07002023 * @journal: Journal to act on.
2024 * @write: flag (see below)
2025 *
2026 * Wipe out all of the contents of a journal, safely. This will produce
2027 * a warning if the journal contains any valid recovery information.
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002028 * Must be called between journal_init_*() and jbd2_journal_load().
Dave Kleikamp470decc2006-10-11 01:20:57 -07002029 *
2030 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2031 * we merely suppress recovery.
2032 */
2033
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002034int jbd2_journal_wipe(journal_t *journal, int write)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002035{
Dave Kleikamp470decc2006-10-11 01:20:57 -07002036 int err = 0;
2037
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002038 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
Dave Kleikamp470decc2006-10-11 01:20:57 -07002039
2040 err = load_superblock(journal);
2041 if (err)
2042 return err;
2043
Dave Kleikamp470decc2006-10-11 01:20:57 -07002044 if (!journal->j_tail)
2045 goto no_recovery;
2046
Eryu Guanf2a44522011-11-01 19:09:18 -04002047 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002048 write ? "Clearing" : "Ignoring");
2049
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002050 err = jbd2_journal_skip_recovery(journal);
Jan Karaa78bb112012-03-13 15:43:04 -04002051 if (write) {
2052 /* Lock to make assertions happy... */
2053 mutex_lock(&journal->j_checkpoint_mutex);
OGAWA Hirofumic0a2ad92016-03-09 23:47:25 -05002054 jbd2_mark_journal_empty(journal, WRITE_FUA);
Jan Karaa78bb112012-03-13 15:43:04 -04002055 mutex_unlock(&journal->j_checkpoint_mutex);
2056 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002057
2058 no_recovery:
2059 return err;
2060}
2061
2062/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07002063 * Journal abort has very specific semantics, which we describe
2064 * for journal abort.
2065 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002066 * Two internal functions, which provide abort to the jbd layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07002067 * itself are here.
2068 */
2069
2070/*
2071 * Quick version for internal journal use (doesn't lock the journal).
2072 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2073 * and don't attempt to make any other journal updates.
2074 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002075void __jbd2_journal_abort_hard(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002076{
2077 transaction_t *transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002078
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002079 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002080 return;
2081
2082 printk(KERN_ERR "Aborting journal on device %s.\n",
Theodore Ts'o05496762008-09-16 14:36:17 -04002083 journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002084
Theodore Ts'oa931da62010-08-03 21:35:12 -04002085 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002086 journal->j_flags |= JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002087 transaction = journal->j_running_transaction;
2088 if (transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002089 __jbd2_log_start_commit(journal, transaction->t_tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002090 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002091}
2092
2093/* Soft abort: record the abort error status in the journal superblock,
2094 * but don't do any other IO. */
2095static void __journal_abort_soft (journal_t *journal, int errno)
2096{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002097 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002098 return;
2099
2100 if (!journal->j_errno)
2101 journal->j_errno = errno;
2102
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002103 __jbd2_journal_abort_hard(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002104
Daeho Jeong4327ba52015-10-18 17:02:56 -04002105 if (errno) {
Jan Kara24bcc892012-03-13 15:41:04 -04002106 jbd2_journal_update_sb_errno(journal);
Daeho Jeong4327ba52015-10-18 17:02:56 -04002107 write_lock(&journal->j_state_lock);
2108 journal->j_flags |= JBD2_REC_ERR;
2109 write_unlock(&journal->j_state_lock);
2110 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002111}
2112
2113/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002114 * void jbd2_journal_abort () - Shutdown the journal immediately.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002115 * @journal: the journal to shutdown.
2116 * @errno: an error number to record in the journal indicating
2117 * the reason for the shutdown.
2118 *
2119 * Perform a complete, immediate shutdown of the ENTIRE
2120 * journal (not of a single transaction). This operation cannot be
2121 * undone without closing and reopening the journal.
2122 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002123 * The jbd2_journal_abort function is intended to support higher level error
Dave Kleikamp470decc2006-10-11 01:20:57 -07002124 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2125 * mode.
2126 *
2127 * Journal abort has very specific semantics. Any existing dirty,
2128 * unjournaled buffers in the main filesystem will still be written to
2129 * disk by bdflush, but the journaling mechanism will be suspended
2130 * immediately and no further transaction commits will be honoured.
2131 *
2132 * Any dirty, journaled buffers will be written back to disk without
2133 * hitting the journal. Atomicity cannot be guaranteed on an aborted
2134 * filesystem, but we _do_ attempt to leave as much data as possible
2135 * behind for fsck to use for cleanup.
2136 *
2137 * Any attempt to get a new transaction handle on a journal which is in
2138 * ABORT state will just result in an -EROFS error return. A
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002139 * jbd2_journal_stop on an existing handle will return -EIO if we have
Dave Kleikamp470decc2006-10-11 01:20:57 -07002140 * entered abort state during the update.
2141 *
2142 * Recursive transactions are not disturbed by journal abort until the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002143 * final jbd2_journal_stop, which will receive the -EIO error.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002144 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002145 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
Dave Kleikamp470decc2006-10-11 01:20:57 -07002146 * which will be recorded (if possible) in the journal superblock. This
2147 * allows a client to record failure conditions in the middle of a
2148 * transaction without having to complete the transaction to record the
2149 * failure to disk. ext3_error, for example, now uses this
2150 * functionality.
2151 *
2152 * Errors which originate from within the journaling layer will NOT
2153 * supply an errno; a null errno implies that absolutely no further
2154 * writes are done to the journal (unless there are any already in
2155 * progress).
2156 *
2157 */
2158
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002159void jbd2_journal_abort(journal_t *journal, int errno)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002160{
2161 __journal_abort_soft(journal, errno);
2162}
2163
2164/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002165 * int jbd2_journal_errno () - returns the journal's error state.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002166 * @journal: journal to examine.
2167 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002168 * This is the errno number set with jbd2_journal_abort(), the last
Dave Kleikamp470decc2006-10-11 01:20:57 -07002169 * time the journal was mounted - if the journal was stopped
2170 * without calling abort this will be 0.
2171 *
2172 * If the journal has been aborted on this mount time -EROFS will
2173 * be returned.
2174 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002175int jbd2_journal_errno(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002176{
2177 int err;
2178
Theodore Ts'oa931da62010-08-03 21:35:12 -04002179 read_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002180 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002181 err = -EROFS;
2182 else
2183 err = journal->j_errno;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002184 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002185 return err;
2186}
2187
2188/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002189 * int jbd2_journal_clear_err () - clears the journal's error state
Dave Kleikamp470decc2006-10-11 01:20:57 -07002190 * @journal: journal to act on.
2191 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002192 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07002193 * mode.
2194 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002195int jbd2_journal_clear_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002196{
2197 int err = 0;
2198
Theodore Ts'oa931da62010-08-03 21:35:12 -04002199 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002200 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002201 err = -EROFS;
2202 else
2203 journal->j_errno = 0;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002204 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002205 return err;
2206}
2207
2208/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002209 * void jbd2_journal_ack_err() - Ack journal err.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002210 * @journal: journal to act on.
2211 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04002212 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07002213 * mode.
2214 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002215void jbd2_journal_ack_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002216{
Theodore Ts'oa931da62010-08-03 21:35:12 -04002217 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002218 if (journal->j_errno)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002219 journal->j_flags |= JBD2_ACK_ERR;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002220 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002221}
2222
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002223int jbd2_journal_blocks_per_page(struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002224{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002225 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002226}
2227
2228/*
Zach Brownb517bea2006-10-11 01:21:08 -07002229 * helper functions to deal with 32 or 64bit block numbers.
2230 */
2231size_t journal_tag_bytes(journal_t *journal)
2232{
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002233 size_t sz;
2234
Darrick J. Wong56316a02015-10-17 16:18:45 -04002235 if (jbd2_has_feature_csum3(journal))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002236 return sizeof(journal_block_tag3_t);
2237
2238 sz = sizeof(journal_block_tag_t);
Darrick J. Wongc3900872012-05-27 08:12:12 -04002239
Darrick J. Wong56316a02015-10-17 16:18:45 -04002240 if (jbd2_has_feature_csum2(journal))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002241 sz += sizeof(__u16);
Darrick J. Wongc3900872012-05-27 08:12:12 -04002242
Darrick J. Wong56316a02015-10-17 16:18:45 -04002243 if (jbd2_has_feature_64bit(journal))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002244 return sz;
Zach Brownb517bea2006-10-11 01:21:08 -07002245 else
Darrick J. Wongdb9ee222014-08-27 18:40:07 -04002246 return sz - sizeof(__u32);
Zach Brownb517bea2006-10-11 01:21:08 -07002247}
2248
2249/*
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002250 * JBD memory management
2251 *
2252 * These functions are used to allocate block-sized chunks of memory
2253 * used for making copies of buffer_head data. Very often it will be
2254 * page-sized chunks of data, but sometimes it will be in
2255 * sub-page-size chunks. (For example, 16k pages on Power systems
2256 * with a 4k block file system.) For blocks smaller than a page, we
2257 * use a SLAB allocator. There are slab caches for each block size,
2258 * which are allocated at mount time, if necessary, and we only free
2259 * (all of) the slab caches when/if the jbd2 module is unloaded. For
2260 * this reason we don't need to a mutex to protect access to
2261 * jbd2_slab[] allocating or releasing memory; only in
2262 * jbd2_journal_create_slab().
2263 */
2264#define JBD2_MAX_SLABS 8
2265static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002266
2267static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2268 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2269 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2270};
2271
2272
2273static void jbd2_journal_destroy_slabs(void)
2274{
2275 int i;
2276
2277 for (i = 0; i < JBD2_MAX_SLABS; i++) {
2278 if (jbd2_slab[i])
2279 kmem_cache_destroy(jbd2_slab[i]);
2280 jbd2_slab[i] = NULL;
2281 }
2282}
2283
2284static int jbd2_journal_create_slab(size_t size)
2285{
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002286 static DEFINE_MUTEX(jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002287 int i = order_base_2(size) - 10;
2288 size_t slab_size;
2289
2290 if (size == PAGE_SIZE)
2291 return 0;
2292
2293 if (i >= JBD2_MAX_SLABS)
2294 return -EINVAL;
2295
2296 if (unlikely(i < 0))
2297 i = 0;
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002298 mutex_lock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002299 if (jbd2_slab[i]) {
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002300 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002301 return 0; /* Already created */
2302 }
2303
2304 slab_size = 1 << (i+10);
2305 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2306 slab_size, 0, NULL);
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02002307 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002308 if (!jbd2_slab[i]) {
2309 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2310 return -ENOMEM;
2311 }
2312 return 0;
2313}
2314
2315static struct kmem_cache *get_slab(size_t size)
2316{
2317 int i = order_base_2(size) - 10;
2318
2319 BUG_ON(i >= JBD2_MAX_SLABS);
2320 if (unlikely(i < 0))
2321 i = 0;
Bill Pemberton8ac97b72010-04-30 09:34:31 -04002322 BUG_ON(jbd2_slab[i] == NULL);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002323 return jbd2_slab[i];
2324}
2325
2326void *jbd2_alloc(size_t size, gfp_t flags)
2327{
2328 void *ptr;
2329
2330 BUG_ON(size & (size-1)); /* Must be a power of 2 */
2331
2332 flags |= __GFP_REPEAT;
2333 if (size == PAGE_SIZE)
2334 ptr = (void *)__get_free_pages(flags, 0);
2335 else if (size > PAGE_SIZE) {
2336 int order = get_order(size);
2337
2338 if (order < 3)
2339 ptr = (void *)__get_free_pages(flags, order);
2340 else
2341 ptr = vmalloc(size);
2342 } else
2343 ptr = kmem_cache_alloc(get_slab(size), flags);
2344
2345 /* Check alignment; SLUB has gotten this wrong in the past,
2346 * and this can lead to user data corruption! */
2347 BUG_ON(((unsigned long) ptr) & (size-1));
2348
2349 return ptr;
2350}
2351
2352void jbd2_free(void *ptr, size_t size)
2353{
2354 if (size == PAGE_SIZE) {
2355 free_pages((unsigned long)ptr, 0);
2356 return;
2357 }
2358 if (size > PAGE_SIZE) {
2359 int order = get_order(size);
2360
2361 if (order < 3)
2362 free_pages((unsigned long)ptr, order);
2363 else
2364 vfree(ptr);
2365 return;
2366 }
2367 kmem_cache_free(get_slab(size), ptr);
2368};
2369
2370/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07002371 * Journal_head storage management
2372 */
Christoph Lametere18b8902006-12-06 20:33:20 -08002373static struct kmem_cache *jbd2_journal_head_cache;
Jose R. Santose23291b2007-07-18 08:57:06 -04002374#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002375static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2376#endif
2377
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002378static int jbd2_journal_init_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002379{
2380 int retval;
2381
Al Viro1076d172008-03-29 03:07:18 +00002382 J_ASSERT(jbd2_journal_head_cache == NULL);
Johann Lombardia920e942006-10-11 01:21:00 -07002383 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002384 sizeof(struct journal_head),
2385 0, /* offset */
Jan Karade92c8c2015-06-08 12:46:37 -04002386 SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
Paul Mundt20c2df82007-07-20 10:11:58 +09002387 NULL); /* ctor */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002388 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00002389 if (!jbd2_journal_head_cache) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002390 retval = -ENOMEM;
Eryu Guanf2a44522011-11-01 19:09:18 -04002391 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002392 }
2393 return retval;
2394}
2395
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002396static void jbd2_journal_destroy_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002397{
Duane Griffin8a9362e2008-04-17 10:38:59 -04002398 if (jbd2_journal_head_cache) {
2399 kmem_cache_destroy(jbd2_journal_head_cache);
2400 jbd2_journal_head_cache = NULL;
2401 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002402}
2403
2404/*
2405 * journal_head splicing and dicing
2406 */
2407static struct journal_head *journal_alloc_journal_head(void)
2408{
2409 struct journal_head *ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002410
Jose R. Santose23291b2007-07-18 08:57:06 -04002411#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002412 atomic_inc(&nr_journal_heads);
2413#endif
Zheng Liu5d9cf9c2013-05-28 07:27:11 -04002414 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
Al Viro1076d172008-03-29 03:07:18 +00002415 if (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002416 jbd_debug(1, "out of memory for journal_head\n");
Theodore Ts'o670be5a2010-12-17 10:44:16 -05002417 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
Michal Hocko7b506b12015-06-15 15:45:58 -04002418 ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2419 GFP_NOFS | __GFP_NOFAIL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002420 }
2421 return ret;
2422}
2423
2424static void journal_free_journal_head(struct journal_head *jh)
2425{
Jose R. Santose23291b2007-07-18 08:57:06 -04002426#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002427 atomic_dec(&nr_journal_heads);
Mingming Caocd02ff02007-10-16 18:38:25 -04002428 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -07002429#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002430 kmem_cache_free(jbd2_journal_head_cache, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002431}
2432
2433/*
2434 * A journal_head is attached to a buffer_head whenever JBD has an
2435 * interest in the buffer.
2436 *
2437 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2438 * is set. This bit is tested in core kernel code where we need to take
2439 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2440 * there.
2441 *
2442 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2443 *
2444 * When a buffer has its BH_JBD bit set it is immune from being released by
2445 * core kernel code, mainly via ->b_count.
2446 *
Jan Karade1b7942011-06-13 15:38:22 -04002447 * A journal_head is detached from its buffer_head when the journal_head's
2448 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2449 * transaction (b_cp_transaction) hold their references to b_jcount.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002450 *
2451 * Various places in the kernel want to attach a journal_head to a buffer_head
2452 * _before_ attaching the journal_head to a transaction. To protect the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002453 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
Dave Kleikamp470decc2006-10-11 01:20:57 -07002454 * journal_head's b_jcount refcount by one. The caller must call
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002455 * jbd2_journal_put_journal_head() to undo this.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002456 *
2457 * So the typical usage would be:
2458 *
2459 * (Attach a journal_head if needed. Increments b_jcount)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002460 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002461 * ...
Jan Karade1b7942011-06-13 15:38:22 -04002462 * (Get another reference for transaction)
2463 * jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002464 * jh->b_transaction = xxx;
Jan Karade1b7942011-06-13 15:38:22 -04002465 * (Put original reference)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002466 * jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002467 */
2468
2469/*
2470 * Give a buffer_head a journal_head.
2471 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07002472 * May sleep.
2473 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002474struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002475{
2476 struct journal_head *jh;
2477 struct journal_head *new_jh = NULL;
2478
2479repeat:
Zheng Liu5d9cf9c2013-05-28 07:27:11 -04002480 if (!buffer_jbd(bh))
Dave Kleikamp470decc2006-10-11 01:20:57 -07002481 new_jh = journal_alloc_journal_head();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002482
2483 jbd_lock_bh_journal_head(bh);
2484 if (buffer_jbd(bh)) {
2485 jh = bh2jh(bh);
2486 } else {
2487 J_ASSERT_BH(bh,
2488 (atomic_read(&bh->b_count) > 0) ||
2489 (bh->b_page && bh->b_page->mapping));
2490
2491 if (!new_jh) {
2492 jbd_unlock_bh_journal_head(bh);
2493 goto repeat;
2494 }
2495
2496 jh = new_jh;
2497 new_jh = NULL; /* We consumed it */
2498 set_buffer_jbd(bh);
2499 bh->b_private = jh;
2500 jh->b_bh = bh;
2501 get_bh(bh);
2502 BUFFER_TRACE(bh, "added journal_head");
2503 }
2504 jh->b_jcount++;
2505 jbd_unlock_bh_journal_head(bh);
2506 if (new_jh)
2507 journal_free_journal_head(new_jh);
2508 return bh->b_private;
2509}
2510
2511/*
2512 * Grab a ref against this buffer_head's journal_head. If it ended up not
2513 * having a journal_head, return NULL
2514 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002515struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002516{
2517 struct journal_head *jh = NULL;
2518
2519 jbd_lock_bh_journal_head(bh);
2520 if (buffer_jbd(bh)) {
2521 jh = bh2jh(bh);
2522 jh->b_jcount++;
2523 }
2524 jbd_unlock_bh_journal_head(bh);
2525 return jh;
2526}
2527
2528static void __journal_remove_journal_head(struct buffer_head *bh)
2529{
2530 struct journal_head *jh = bh2jh(bh);
2531
2532 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karade1b7942011-06-13 15:38:22 -04002533 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2534 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2535 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2536 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2537 J_ASSERT_BH(bh, buffer_jbd(bh));
2538 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2539 BUFFER_TRACE(bh, "remove journal_head");
2540 if (jh->b_frozen_data) {
2541 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2542 jbd2_free(jh->b_frozen_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002543 }
Jan Karade1b7942011-06-13 15:38:22 -04002544 if (jh->b_committed_data) {
2545 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2546 jbd2_free(jh->b_committed_data, bh->b_size);
2547 }
2548 bh->b_private = NULL;
2549 jh->b_bh = NULL; /* debug, really */
2550 clear_buffer_jbd(bh);
2551 journal_free_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002552}
2553
2554/*
Jan Karade1b7942011-06-13 15:38:22 -04002555 * Drop a reference on the passed journal_head. If it fell to zero then
Dave Kleikamp470decc2006-10-11 01:20:57 -07002556 * release the journal_head from the buffer_head.
2557 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002558void jbd2_journal_put_journal_head(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002559{
2560 struct buffer_head *bh = jh2bh(jh);
2561
2562 jbd_lock_bh_journal_head(bh);
2563 J_ASSERT_JH(jh, jh->b_jcount > 0);
2564 --jh->b_jcount;
Jan Karade1b7942011-06-13 15:38:22 -04002565 if (!jh->b_jcount) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002566 __journal_remove_journal_head(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002567 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002568 __brelse(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002569 } else
2570 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002571}
2572
2573/*
Jan Karac851ed52008-07-11 19:27:31 -04002574 * Initialize jbd inode head
2575 */
2576void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2577{
2578 jinode->i_transaction = NULL;
2579 jinode->i_next_transaction = NULL;
2580 jinode->i_vfs_inode = inode;
2581 jinode->i_flags = 0;
2582 INIT_LIST_HEAD(&jinode->i_list);
2583}
2584
2585/*
2586 * Function to be called before we start removing inode from memory (i.e.,
2587 * clear_inode() is a fine place to be called from). It removes inode from
2588 * transaction's lists.
2589 */
2590void jbd2_journal_release_jbd_inode(journal_t *journal,
2591 struct jbd2_inode *jinode)
2592{
Jan Karac851ed52008-07-11 19:27:31 -04002593 if (!journal)
2594 return;
2595restart:
2596 spin_lock(&journal->j_list_lock);
2597 /* Is commit writing out inode - we have to wait */
Jan Karacb0d9d42016-02-22 23:20:30 -05002598 if (jinode->i_flags & JI_COMMIT_RUNNING) {
Jan Karac851ed52008-07-11 19:27:31 -04002599 wait_queue_head_t *wq;
2600 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2601 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2602 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2603 spin_unlock(&journal->j_list_lock);
2604 schedule();
2605 finish_wait(wq, &wait.wait);
2606 goto restart;
2607 }
2608
Jan Karac851ed52008-07-11 19:27:31 -04002609 if (jinode->i_transaction) {
2610 list_del(&jinode->i_list);
2611 jinode->i_transaction = NULL;
2612 }
2613 spin_unlock(&journal->j_list_lock);
2614}
2615
Dave Kleikamp470decc2006-10-11 01:20:57 -07002616
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002617#ifdef CONFIG_PROC_FS
2618
2619#define JBD2_STATS_PROC_NAME "fs/jbd2"
2620
2621static void __init jbd2_create_jbd_stats_proc_entry(void)
2622{
2623 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2624}
2625
2626static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2627{
2628 if (proc_jbd2_stats)
2629 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2630}
2631
2632#else
2633
2634#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2635#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2636
2637#endif
2638
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002639struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002640
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002641static int __init jbd2_journal_init_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002642{
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002643 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002644 if (jbd2_handle_cache == NULL) {
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002645 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2646 return -ENOMEM;
2647 }
2648 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2649 if (jbd2_inode_cache == NULL) {
2650 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2651 kmem_cache_destroy(jbd2_handle_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002652 return -ENOMEM;
2653 }
2654 return 0;
2655}
2656
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002657static void jbd2_journal_destroy_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002658{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002659 if (jbd2_handle_cache)
2660 kmem_cache_destroy(jbd2_handle_cache);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002661 if (jbd2_inode_cache)
2662 kmem_cache_destroy(jbd2_inode_cache);
2663
Dave Kleikamp470decc2006-10-11 01:20:57 -07002664}
2665
2666/*
2667 * Module startup and shutdown
2668 */
2669
2670static int __init journal_init_caches(void)
2671{
2672 int ret;
2673
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002674 ret = jbd2_journal_init_revoke_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002675 if (ret == 0)
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002676 ret = jbd2_journal_init_journal_head_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002677 if (ret == 0)
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002678 ret = jbd2_journal_init_handle_cache();
Yongqiang Yang0c2022e2012-02-20 17:53:02 -05002679 if (ret == 0)
2680 ret = jbd2_journal_init_transaction_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002681 return ret;
2682}
2683
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002684static void jbd2_journal_destroy_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002685{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002686 jbd2_journal_destroy_revoke_caches();
Yongqiang Yang4185a2a2012-02-20 17:53:03 -05002687 jbd2_journal_destroy_journal_head_cache();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002688 jbd2_journal_destroy_handle_cache();
Yongqiang Yang0c2022e2012-02-20 17:53:02 -05002689 jbd2_journal_destroy_transaction_cache();
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002690 jbd2_journal_destroy_slabs();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002691}
2692
2693static int __init journal_init(void)
2694{
2695 int ret;
2696
2697 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2698
2699 ret = journal_init_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002700 if (ret == 0) {
Duane Griffin620de4e2008-04-29 22:02:47 -04002701 jbd2_create_jbd_stats_proc_entry();
2702 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002703 jbd2_journal_destroy_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002704 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002705 return ret;
2706}
2707
2708static void __exit journal_exit(void)
2709{
Jose R. Santose23291b2007-07-18 08:57:06 -04002710#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002711 int n = atomic_read(&nr_journal_heads);
2712 if (n)
Jan Kara75685072013-12-08 21:13:59 -05002713 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002714#endif
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002715 jbd2_remove_jbd_stats_proc_entry();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002716 jbd2_journal_destroy_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002717}
2718
Dave Kleikamp470decc2006-10-11 01:20:57 -07002719MODULE_LICENSE("GPL");
2720module_init(journal_init);
2721module_exit(journal_exit);
2722