1 // SPDX-License-Identifier: GPL-2.0
4 * fs/ext4/fast_commit.c
6 * Written by Harshad Shirwadkar <harshadshirwadkar@gmail.com>
8 * Ext4 fast commits routines.
11 #include "ext4_jbd2.h"
12 #include "ext4_extents.h"
19 * Ext4 fast commits implement fine grained journalling for Ext4.
21 * Fast commits are organized as a log of tag-length-value (TLV) structs. (See
22 * struct ext4_fc_tl). Each TLV contains some delta that is replayed TLV by
23 * TLV during the recovery phase. For the scenarios for which we currently
24 * don't have replay code, fast commit falls back to full commits.
25 * Fast commits record delta in one of the following three categories.
27 * (A) Directory entry updates:
29 * - EXT4_FC_TAG_UNLINK - records directory entry unlink
30 * - EXT4_FC_TAG_LINK - records directory entry link
31 * - EXT4_FC_TAG_CREAT - records inode and directory entry creation
33 * (B) File specific data range updates:
35 * - EXT4_FC_TAG_ADD_RANGE - records addition of new blocks to an inode
36 * - EXT4_FC_TAG_DEL_RANGE - records deletion of blocks from an inode
38 * (C) Inode metadata (mtime / ctime etc):
40 * - EXT4_FC_TAG_INODE - record the inode that should be replayed
41 * during recovery. Note that iblocks field is
42 * not replayed and instead derived during
46 * With fast commits, we maintain all the directory entry operations in the
47 * order in which they are issued in an in-memory queue. This queue is flushed
48 * to disk during the commit operation. We also maintain a list of inodes
49 * that need to be committed during a fast commit in another in memory queue of
50 * inodes. During the commit operation, we commit in the following order:
52 * [1] Lock inodes for any further data updates by setting COMMITTING state
53 * [2] Submit data buffers of all the inodes
54 * [3] Wait for [2] to complete
55 * [4] Commit all the directory entry updates in the fast commit space
56 * [5] Commit all the changed inode structures
57 * [6] Write tail tag (this tag ensures the atomicity, please read the following
58 * section for more details).
59 * [7] Wait for [4], [5] and [6] to complete.
61 * All the inode updates must call ext4_fc_start_update() before starting an
62 * update. If such an ongoing update is present, fast commit waits for it to
63 * complete. The completion of such an update is marked by
64 * ext4_fc_stop_update().
66 * Fast Commit Ineligibility
67 * -------------------------
69 * Not all operations are supported by fast commits today (e.g extended
70 * attributes). Fast commit ineligibility is marked by calling
71 * ext4_fc_mark_ineligible(): This makes next fast commit operation to fall back
74 * Atomicity of commits
75 * --------------------
76 * In order to guarantee atomicity during the commit operation, fast commit
77 * uses "EXT4_FC_TAG_TAIL" tag that marks a fast commit as complete. Tail
78 * tag contains CRC of the contents and TID of the transaction after which
79 * this fast commit should be applied. Recovery code replays fast commit
80 * logs only if there's at least 1 valid tail present. For every fast commit
81 * operation, there is 1 tail. This means, we may end up with multiple tails
82 * in the fast commit space. Here's an example:
84 * - Create a new file A and remove existing file B
86 * - Append contents to file A
90 * The fast commit space at the end of above operations would look like this:
91 * [HEAD] [CREAT A] [UNLINK B] [TAIL] [ADD_RANGE A] [DEL_RANGE A] [TAIL]
92 * |<--- Fast Commit 1 --->|<--- Fast Commit 2 ---->|
94 * Replay code should thus check for all the valid tails in the FC area.
96 * Fast Commit Replay Idempotence
97 * ------------------------------
99 * Fast commits tags are idempotent in nature provided the recovery code follows
100 * certain rules. The guiding principle that the commit path follows while
101 * committing is that it stores the result of a particular operation instead of
102 * storing the procedure.
104 * Let's consider this rename operation: 'mv /a /b'. Let's assume dirent '/a'
105 * was associated with inode 10. During fast commit, instead of storing this
106 * operation as a procedure "rename a to b", we store the resulting file system
107 * state as a "series" of outcomes:
109 * - Link dirent b to inode 10
111 * - Inode <10> with valid refcount
113 * Now when recovery code runs, it needs "enforce" this state on the file
114 * system. This is what guarantees idempotence of fast commit replay.
116 * Let's take an example of a procedure that is not idempotent and see how fast
117 * commits make it idempotent. Consider following sequence of operations:
119 * rm A; mv B A; read A
122 * (x), (y) and (z) are the points at which we can crash. If we store this
123 * sequence of operations as is then the replay is not idempotent. Let's say
124 * while in replay, we crash at (z). During the second replay, file A (which was
125 * actually created as a result of "mv B A" operation) would get deleted. Thus,
126 * file named A would be absent when we try to read A. So, this sequence of
127 * operations is not idempotent. However, as mentioned above, instead of storing
128 * the procedure fast commits store the outcome of each procedure. Thus the fast
129 * commit log for above procedure would be as follows:
131 * (Let's assume dirent A was linked to inode 10 and dirent B was linked to
132 * inode 11 before the replay)
134 * [Unlink A] [Link A to inode 11] [Unlink B] [Inode 11]
137 * If we crash at (z), we will have file A linked to inode 11. During the second
138 * replay, we will remove file A (inode 11). But we will create it back and make
139 * it point to inode 11. We won't find B, so we'll just skip that step. At this
140 * point, the refcount for inode 11 is not reliable, but that gets fixed by the
141 * replay of last inode 11 tag. Crashes at points (w), (x) and (y) get handled
142 * similarly. Thus, by converting a non-idempotent procedure into a series of
143 * idempotent outcomes, fast commits ensured idempotence during the replay.
148 * 0) Fast commit replay path hardening: Fast commit replay code should use
149 * journal handles to make sure all the updates it does during the replay
150 * path are atomic. With that if we crash during fast commit replay, after
151 * trying to do recovery again, we will find a file system where fast commit
152 * area is invalid (because new full commit would be found). In order to deal
153 * with that, fast commit replay code should ensure that the "FC_REPLAY"
154 * superblock state is persisted before starting the replay, so that after
155 * the crash, fast commit recovery code can look at that flag and perform
156 * fast commit recovery even if that area is invalidated by later full
159 * 1) Fast commit's commit path locks the entire file system during fast
160 * commit. This has significant performance penalty. Instead of that, we
161 * should use ext4_fc_start/stop_update functions to start inode level
162 * updates from ext4_journal_start/stop. Once we do that we can drop file
163 * system locking during commit path.
165 * 2) Handle more ineligible cases.
168 #include <trace/events/ext4.h>
169 static struct kmem_cache *ext4_fc_dentry_cachep;
171 static void ext4_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
173 BUFFER_TRACE(bh, "");
175 ext4_debug("%s: Block %lld up-to-date",
176 __func__, bh->b_blocknr);
177 set_buffer_uptodate(bh);
179 ext4_debug("%s: Block %lld not up-to-date",
180 __func__, bh->b_blocknr);
181 clear_buffer_uptodate(bh);
187 static inline void ext4_fc_reset_inode(struct inode *inode)
189 struct ext4_inode_info *ei = EXT4_I(inode);
191 ei->i_fc_lblk_start = 0;
192 ei->i_fc_lblk_len = 0;
195 void ext4_fc_init_inode(struct inode *inode)
197 struct ext4_inode_info *ei = EXT4_I(inode);
199 ext4_fc_reset_inode(inode);
200 ext4_clear_inode_state(inode, EXT4_STATE_FC_COMMITTING);
201 INIT_LIST_HEAD(&ei->i_fc_list);
202 INIT_LIST_HEAD(&ei->i_fc_dilist);
203 init_waitqueue_head(&ei->i_fc_wait);
204 atomic_set(&ei->i_fc_updates, 0);
207 /* This function must be called with sbi->s_fc_lock held. */
208 static void ext4_fc_wait_committing_inode(struct inode *inode)
209 __releases(&EXT4_SB(inode->i_sb)->s_fc_lock)
211 wait_queue_head_t *wq;
212 struct ext4_inode_info *ei = EXT4_I(inode);
214 #if (BITS_PER_LONG < 64)
215 DEFINE_WAIT_BIT(wait, &ei->i_state_flags,
216 EXT4_STATE_FC_COMMITTING);
217 wq = bit_waitqueue(&ei->i_state_flags,
218 EXT4_STATE_FC_COMMITTING);
220 DEFINE_WAIT_BIT(wait, &ei->i_flags,
221 EXT4_STATE_FC_COMMITTING);
222 wq = bit_waitqueue(&ei->i_flags,
223 EXT4_STATE_FC_COMMITTING);
225 lockdep_assert_held(&EXT4_SB(inode->i_sb)->s_fc_lock);
226 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
227 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
229 finish_wait(wq, &wait.wq_entry);
232 static bool ext4_fc_disabled(struct super_block *sb)
234 return (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
235 (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY));
239 * Inform Ext4's fast about start of an inode update
241 * This function is called by the high level call VFS callbacks before
242 * performing any inode update. This function blocks if there's an ongoing
243 * fast commit on the inode in question.
245 void ext4_fc_start_update(struct inode *inode)
247 struct ext4_inode_info *ei = EXT4_I(inode);
249 if (ext4_fc_disabled(inode->i_sb))
253 spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
254 if (list_empty(&ei->i_fc_list))
257 if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
258 ext4_fc_wait_committing_inode(inode);
262 atomic_inc(&ei->i_fc_updates);
263 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
267 * Stop inode update and wake up waiting fast commits if any.
269 void ext4_fc_stop_update(struct inode *inode)
271 struct ext4_inode_info *ei = EXT4_I(inode);
273 if (ext4_fc_disabled(inode->i_sb))
276 if (atomic_dec_and_test(&ei->i_fc_updates))
277 wake_up_all(&ei->i_fc_wait);
281 * Remove inode from fast commit list. If the inode is being committed
282 * we wait until inode commit is done.
284 void ext4_fc_del(struct inode *inode)
286 struct ext4_inode_info *ei = EXT4_I(inode);
287 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
288 struct ext4_fc_dentry_update *fc_dentry;
290 if (ext4_fc_disabled(inode->i_sb))
294 spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
295 if (list_empty(&ei->i_fc_list) && list_empty(&ei->i_fc_dilist)) {
296 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
300 if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
301 ext4_fc_wait_committing_inode(inode);
305 if (!list_empty(&ei->i_fc_list))
306 list_del_init(&ei->i_fc_list);
309 * Since this inode is getting removed, let's also remove all FC
310 * dentry create references, since it is not needed to log it anyways.
312 if (list_empty(&ei->i_fc_dilist)) {
313 spin_unlock(&sbi->s_fc_lock);
317 fc_dentry = list_first_entry(&ei->i_fc_dilist, struct ext4_fc_dentry_update, fcd_dilist);
318 WARN_ON(fc_dentry->fcd_op != EXT4_FC_TAG_CREAT);
319 list_del_init(&fc_dentry->fcd_list);
320 list_del_init(&fc_dentry->fcd_dilist);
322 WARN_ON(!list_empty(&ei->i_fc_dilist));
323 spin_unlock(&sbi->s_fc_lock);
325 if (fc_dentry->fcd_name.name &&
326 fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
327 kfree(fc_dentry->fcd_name.name);
328 kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
334 * Mark file system as fast commit ineligible, and record latest
335 * ineligible transaction tid. This means until the recorded
336 * transaction, commit operation would result in a full jbd2 commit.
338 void ext4_fc_mark_ineligible(struct super_block *sb, int reason, handle_t *handle)
340 struct ext4_sb_info *sbi = EXT4_SB(sb);
343 if (ext4_fc_disabled(sb))
346 ext4_set_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
347 if (handle && !IS_ERR(handle))
348 tid = handle->h_transaction->t_tid;
350 read_lock(&sbi->s_journal->j_state_lock);
351 tid = sbi->s_journal->j_running_transaction ?
352 sbi->s_journal->j_running_transaction->t_tid : 0;
353 read_unlock(&sbi->s_journal->j_state_lock);
355 spin_lock(&sbi->s_fc_lock);
356 if (sbi->s_fc_ineligible_tid < tid)
357 sbi->s_fc_ineligible_tid = tid;
358 spin_unlock(&sbi->s_fc_lock);
359 WARN_ON(reason >= EXT4_FC_REASON_MAX);
360 sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;
364 * Generic fast commit tracking function. If this is the first time this we are
365 * called after a full commit, we initialize fast commit fields and then call
366 * __fc_track_fn() with update = 0. If we have already been called after a full
367 * commit, we pass update = 1. Based on that, the track function can determine
368 * if it needs to track a field for the first time or if it needs to just
369 * update the previously tracked value.
371 * If enqueue is set, this function enqueues the inode in fast commit list.
373 static int ext4_fc_track_template(
374 handle_t *handle, struct inode *inode,
375 int (*__fc_track_fn)(struct inode *, void *, bool),
376 void *args, int enqueue)
379 struct ext4_inode_info *ei = EXT4_I(inode);
380 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
384 tid = handle->h_transaction->t_tid;
385 mutex_lock(&ei->i_fc_lock);
386 if (tid == ei->i_sync_tid) {
389 ext4_fc_reset_inode(inode);
390 ei->i_sync_tid = tid;
392 ret = __fc_track_fn(inode, args, update);
393 mutex_unlock(&ei->i_fc_lock);
398 spin_lock(&sbi->s_fc_lock);
399 if (list_empty(&EXT4_I(inode)->i_fc_list))
400 list_add_tail(&EXT4_I(inode)->i_fc_list,
401 (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
402 sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING) ?
403 &sbi->s_fc_q[FC_Q_STAGING] :
404 &sbi->s_fc_q[FC_Q_MAIN]);
405 spin_unlock(&sbi->s_fc_lock);
410 struct __track_dentry_update_args {
411 struct dentry *dentry;
415 /* __track_fn for directory entry updates. Called with ei->i_fc_lock. */
416 static int __track_dentry_update(struct inode *inode, void *arg, bool update)
418 struct ext4_fc_dentry_update *node;
419 struct ext4_inode_info *ei = EXT4_I(inode);
420 struct __track_dentry_update_args *dentry_update =
421 (struct __track_dentry_update_args *)arg;
422 struct dentry *dentry = dentry_update->dentry;
423 struct inode *dir = dentry->d_parent->d_inode;
424 struct super_block *sb = inode->i_sb;
425 struct ext4_sb_info *sbi = EXT4_SB(sb);
427 mutex_unlock(&ei->i_fc_lock);
429 if (IS_ENCRYPTED(dir)) {
430 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_ENCRYPTED_FILENAME,
432 mutex_lock(&ei->i_fc_lock);
436 node = kmem_cache_alloc(ext4_fc_dentry_cachep, GFP_NOFS);
438 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_NOMEM, NULL);
439 mutex_lock(&ei->i_fc_lock);
443 node->fcd_op = dentry_update->op;
444 node->fcd_parent = dir->i_ino;
445 node->fcd_ino = inode->i_ino;
446 if (dentry->d_name.len > DNAME_INLINE_LEN) {
447 node->fcd_name.name = kmalloc(dentry->d_name.len, GFP_NOFS);
448 if (!node->fcd_name.name) {
449 kmem_cache_free(ext4_fc_dentry_cachep, node);
450 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_NOMEM, NULL);
451 mutex_lock(&ei->i_fc_lock);
454 memcpy((u8 *)node->fcd_name.name, dentry->d_name.name,
457 memcpy(node->fcd_iname, dentry->d_name.name,
459 node->fcd_name.name = node->fcd_iname;
461 node->fcd_name.len = dentry->d_name.len;
462 INIT_LIST_HEAD(&node->fcd_dilist);
463 spin_lock(&sbi->s_fc_lock);
464 if (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
465 sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING)
466 list_add_tail(&node->fcd_list,
467 &sbi->s_fc_dentry_q[FC_Q_STAGING]);
469 list_add_tail(&node->fcd_list, &sbi->s_fc_dentry_q[FC_Q_MAIN]);
472 * This helps us keep a track of all fc_dentry updates which is part of
473 * this ext4 inode. So in case the inode is getting unlinked, before
474 * even we get a chance to fsync, we could remove all fc_dentry
475 * references while evicting the inode in ext4_fc_del().
476 * Also with this, we don't need to loop over all the inodes in
477 * sbi->s_fc_q to get the corresponding inode in
478 * ext4_fc_commit_dentry_updates().
480 if (dentry_update->op == EXT4_FC_TAG_CREAT) {
481 WARN_ON(!list_empty(&ei->i_fc_dilist));
482 list_add_tail(&node->fcd_dilist, &ei->i_fc_dilist);
484 spin_unlock(&sbi->s_fc_lock);
485 mutex_lock(&ei->i_fc_lock);
490 void __ext4_fc_track_unlink(handle_t *handle,
491 struct inode *inode, struct dentry *dentry)
493 struct __track_dentry_update_args args;
496 args.dentry = dentry;
497 args.op = EXT4_FC_TAG_UNLINK;
499 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
501 trace_ext4_fc_track_unlink(handle, inode, dentry, ret);
504 void ext4_fc_track_unlink(handle_t *handle, struct dentry *dentry)
506 struct inode *inode = d_inode(dentry);
508 if (ext4_fc_disabled(inode->i_sb))
511 if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
514 __ext4_fc_track_unlink(handle, inode, dentry);
517 void __ext4_fc_track_link(handle_t *handle,
518 struct inode *inode, struct dentry *dentry)
520 struct __track_dentry_update_args args;
523 args.dentry = dentry;
524 args.op = EXT4_FC_TAG_LINK;
526 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
528 trace_ext4_fc_track_link(handle, inode, dentry, ret);
531 void ext4_fc_track_link(handle_t *handle, struct dentry *dentry)
533 struct inode *inode = d_inode(dentry);
535 if (ext4_fc_disabled(inode->i_sb))
538 if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
541 __ext4_fc_track_link(handle, inode, dentry);
544 void __ext4_fc_track_create(handle_t *handle, struct inode *inode,
545 struct dentry *dentry)
547 struct __track_dentry_update_args args;
550 args.dentry = dentry;
551 args.op = EXT4_FC_TAG_CREAT;
553 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
555 trace_ext4_fc_track_create(handle, inode, dentry, ret);
558 void ext4_fc_track_create(handle_t *handle, struct dentry *dentry)
560 struct inode *inode = d_inode(dentry);
562 if (ext4_fc_disabled(inode->i_sb))
565 if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
568 __ext4_fc_track_create(handle, inode, dentry);
571 /* __track_fn for inode tracking */
572 static int __track_inode(struct inode *inode, void *arg, bool update)
577 EXT4_I(inode)->i_fc_lblk_len = 0;
582 void ext4_fc_track_inode(handle_t *handle, struct inode *inode)
586 if (S_ISDIR(inode->i_mode))
589 if (ext4_fc_disabled(inode->i_sb))
592 if (ext4_should_journal_data(inode)) {
593 ext4_fc_mark_ineligible(inode->i_sb,
594 EXT4_FC_REASON_INODE_JOURNAL_DATA, handle);
598 if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
601 ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);
602 trace_ext4_fc_track_inode(handle, inode, ret);
605 struct __track_range_args {
606 ext4_lblk_t start, end;
609 /* __track_fn for tracking data updates */
610 static int __track_range(struct inode *inode, void *arg, bool update)
612 struct ext4_inode_info *ei = EXT4_I(inode);
613 ext4_lblk_t oldstart;
614 struct __track_range_args *__arg =
615 (struct __track_range_args *)arg;
617 if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {
618 ext4_debug("Special inode %ld being modified\n", inode->i_ino);
622 oldstart = ei->i_fc_lblk_start;
624 if (update && ei->i_fc_lblk_len > 0) {
625 ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);
627 max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -
628 ei->i_fc_lblk_start + 1;
630 ei->i_fc_lblk_start = __arg->start;
631 ei->i_fc_lblk_len = __arg->end - __arg->start + 1;
637 void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,
640 struct __track_range_args args;
643 if (S_ISDIR(inode->i_mode))
646 if (ext4_fc_disabled(inode->i_sb))
649 if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
655 ret = ext4_fc_track_template(handle, inode, __track_range, &args, 1);
657 trace_ext4_fc_track_range(handle, inode, start, end, ret);
660 static void ext4_fc_submit_bh(struct super_block *sb, bool is_tail)
662 blk_opf_t write_flags = REQ_SYNC;
663 struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
665 /* Add REQ_FUA | REQ_PREFLUSH only its tail */
666 if (test_opt(sb, BARRIER) && is_tail)
667 write_flags |= REQ_FUA | REQ_PREFLUSH;
669 set_buffer_dirty(bh);
670 set_buffer_uptodate(bh);
671 bh->b_end_io = ext4_end_buffer_io_sync;
672 submit_bh(REQ_OP_WRITE | write_flags, bh);
673 EXT4_SB(sb)->s_fc_bh = NULL;
676 /* Ext4 commit path routines */
678 /* memcpy to fc reserved space and update CRC */
679 static void *ext4_fc_memcpy(struct super_block *sb, void *dst, const void *src,
683 *crc = ext4_chksum(EXT4_SB(sb), *crc, src, len);
684 return memcpy(dst, src, len);
687 /* memzero and update CRC */
688 static void *ext4_fc_memzero(struct super_block *sb, void *dst, int len,
693 ret = memset(dst, 0, len);
695 *crc = ext4_chksum(EXT4_SB(sb), *crc, dst, len);
700 * Allocate len bytes on a fast commit buffer.
702 * During the commit time this function is used to manage fast commit
703 * block space. We don't split a fast commit log onto different
704 * blocks. So this function makes sure that if there's not enough space
705 * on the current block, the remaining space in the current block is
706 * marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,
707 * new block is from jbd2 and CRC is updated to reflect the padding
710 static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
712 struct ext4_fc_tl tl;
713 struct ext4_sb_info *sbi = EXT4_SB(sb);
714 struct buffer_head *bh;
715 int bsize = sbi->s_journal->j_blocksize;
716 int ret, off = sbi->s_fc_bytes % bsize;
721 * If 'len' is too long to fit in any block alongside a PAD tlv, then we
722 * cannot fulfill the request.
724 if (len > bsize - EXT4_FC_TAG_BASE_LEN)
728 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
733 dst = sbi->s_fc_bh->b_data + off;
736 * Allocate the bytes in the current block if we can do so while still
737 * leaving enough space for a PAD tlv.
739 remaining = bsize - EXT4_FC_TAG_BASE_LEN - off;
740 if (len <= remaining) {
741 sbi->s_fc_bytes += len;
746 * Else, terminate the current block with a PAD tlv, then allocate a new
747 * block and allocate the bytes at the start of that new block.
750 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);
751 tl.fc_len = cpu_to_le16(remaining);
752 ext4_fc_memcpy(sb, dst, &tl, EXT4_FC_TAG_BASE_LEN, crc);
753 ext4_fc_memzero(sb, dst + EXT4_FC_TAG_BASE_LEN, remaining, crc);
755 ext4_fc_submit_bh(sb, false);
757 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
761 sbi->s_fc_bytes += bsize - off + len;
762 return sbi->s_fc_bh->b_data;
766 * Complete a fast commit by writing tail tag.
768 * Writing tail tag marks the end of a fast commit. In order to guarantee
769 * atomicity, after writing tail tag, even if there's space remaining
770 * in the block, next commit shouldn't use it. That's why tail tag
771 * has the length as that of the remaining space on the block.
773 static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
775 struct ext4_sb_info *sbi = EXT4_SB(sb);
776 struct ext4_fc_tl tl;
777 struct ext4_fc_tail tail;
778 int off, bsize = sbi->s_journal->j_blocksize;
782 * ext4_fc_reserve_space takes care of allocating an extra block if
783 * there's no enough space on this block for accommodating this tail.
785 dst = ext4_fc_reserve_space(sb, EXT4_FC_TAG_BASE_LEN + sizeof(tail), &crc);
789 off = sbi->s_fc_bytes % bsize;
791 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);
792 tl.fc_len = cpu_to_le16(bsize - off + sizeof(struct ext4_fc_tail));
793 sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);
795 ext4_fc_memcpy(sb, dst, &tl, EXT4_FC_TAG_BASE_LEN, &crc);
796 dst += EXT4_FC_TAG_BASE_LEN;
797 tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);
798 ext4_fc_memcpy(sb, dst, &tail.fc_tid, sizeof(tail.fc_tid), &crc);
799 dst += sizeof(tail.fc_tid);
800 tail.fc_crc = cpu_to_le32(crc);
801 ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
802 dst += sizeof(tail.fc_crc);
803 memset(dst, 0, bsize - off); /* Don't leak uninitialized memory. */
805 ext4_fc_submit_bh(sb, true);
811 * Adds tag, length, value and updates CRC. Returns true if tlv was added.
812 * Returns false if there's not enough space.
814 static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,
817 struct ext4_fc_tl tl;
820 dst = ext4_fc_reserve_space(sb, EXT4_FC_TAG_BASE_LEN + len, crc);
824 tl.fc_tag = cpu_to_le16(tag);
825 tl.fc_len = cpu_to_le16(len);
827 ext4_fc_memcpy(sb, dst, &tl, EXT4_FC_TAG_BASE_LEN, crc);
828 ext4_fc_memcpy(sb, dst + EXT4_FC_TAG_BASE_LEN, val, len, crc);
833 /* Same as above, but adds dentry tlv. */
834 static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u32 *crc,
835 struct ext4_fc_dentry_update *fc_dentry)
837 struct ext4_fc_dentry_info fcd;
838 struct ext4_fc_tl tl;
839 int dlen = fc_dentry->fcd_name.len;
840 u8 *dst = ext4_fc_reserve_space(sb,
841 EXT4_FC_TAG_BASE_LEN + sizeof(fcd) + dlen, crc);
846 fcd.fc_parent_ino = cpu_to_le32(fc_dentry->fcd_parent);
847 fcd.fc_ino = cpu_to_le32(fc_dentry->fcd_ino);
848 tl.fc_tag = cpu_to_le16(fc_dentry->fcd_op);
849 tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);
850 ext4_fc_memcpy(sb, dst, &tl, EXT4_FC_TAG_BASE_LEN, crc);
851 dst += EXT4_FC_TAG_BASE_LEN;
852 ext4_fc_memcpy(sb, dst, &fcd, sizeof(fcd), crc);
854 ext4_fc_memcpy(sb, dst, fc_dentry->fcd_name.name, dlen, crc);
860 * Writes inode in the fast commit space under TLV with tag @tag.
861 * Returns 0 on success, error on failure.
863 static int ext4_fc_write_inode(struct inode *inode, u32 *crc)
865 struct ext4_inode_info *ei = EXT4_I(inode);
866 int inode_len = EXT4_GOOD_OLD_INODE_SIZE;
868 struct ext4_iloc iloc;
869 struct ext4_fc_inode fc_inode;
870 struct ext4_fc_tl tl;
873 ret = ext4_get_inode_loc(inode, &iloc);
877 if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
878 inode_len = EXT4_INODE_SIZE(inode->i_sb);
879 else if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)
880 inode_len += ei->i_extra_isize;
882 fc_inode.fc_ino = cpu_to_le32(inode->i_ino);
883 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);
884 tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));
887 dst = ext4_fc_reserve_space(inode->i_sb,
888 EXT4_FC_TAG_BASE_LEN + inode_len + sizeof(fc_inode.fc_ino), crc);
892 if (!ext4_fc_memcpy(inode->i_sb, dst, &tl, EXT4_FC_TAG_BASE_LEN, crc))
894 dst += EXT4_FC_TAG_BASE_LEN;
895 if (!ext4_fc_memcpy(inode->i_sb, dst, &fc_inode, sizeof(fc_inode), crc))
897 dst += sizeof(fc_inode);
898 if (!ext4_fc_memcpy(inode->i_sb, dst, (u8 *)ext4_raw_inode(&iloc),
908 * Writes updated data ranges for the inode in question. Updates CRC.
909 * Returns 0 on success, error otherwise.
911 static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)
913 ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;
914 struct ext4_inode_info *ei = EXT4_I(inode);
915 struct ext4_map_blocks map;
916 struct ext4_fc_add_range fc_ext;
917 struct ext4_fc_del_range lrange;
918 struct ext4_extent *ex;
921 mutex_lock(&ei->i_fc_lock);
922 if (ei->i_fc_lblk_len == 0) {
923 mutex_unlock(&ei->i_fc_lock);
926 old_blk_size = ei->i_fc_lblk_start;
927 new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;
928 ei->i_fc_lblk_len = 0;
929 mutex_unlock(&ei->i_fc_lock);
931 cur_lblk_off = old_blk_size;
932 ext4_debug("will try writing %d to %d for inode %ld\n",
933 cur_lblk_off, new_blk_size, inode->i_ino);
935 while (cur_lblk_off <= new_blk_size) {
936 map.m_lblk = cur_lblk_off;
937 map.m_len = new_blk_size - cur_lblk_off + 1;
938 ret = ext4_map_blocks(NULL, inode, &map, 0);
942 if (map.m_len == 0) {
948 lrange.fc_ino = cpu_to_le32(inode->i_ino);
949 lrange.fc_lblk = cpu_to_le32(map.m_lblk);
950 lrange.fc_len = cpu_to_le32(map.m_len);
951 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,
952 sizeof(lrange), (u8 *)&lrange, crc))
955 unsigned int max = (map.m_flags & EXT4_MAP_UNWRITTEN) ?
956 EXT_UNWRITTEN_MAX_LEN : EXT_INIT_MAX_LEN;
958 /* Limit the number of blocks in one extent */
959 map.m_len = min(max, map.m_len);
961 fc_ext.fc_ino = cpu_to_le32(inode->i_ino);
962 ex = (struct ext4_extent *)&fc_ext.fc_ex;
963 ex->ee_block = cpu_to_le32(map.m_lblk);
964 ex->ee_len = cpu_to_le16(map.m_len);
965 ext4_ext_store_pblock(ex, map.m_pblk);
966 if (map.m_flags & EXT4_MAP_UNWRITTEN)
967 ext4_ext_mark_unwritten(ex);
969 ext4_ext_mark_initialized(ex);
970 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,
971 sizeof(fc_ext), (u8 *)&fc_ext, crc))
975 cur_lblk_off += map.m_len;
982 /* Submit data for all the fast commit inodes */
983 static int ext4_fc_submit_inode_data_all(journal_t *journal)
985 struct super_block *sb = journal->j_private;
986 struct ext4_sb_info *sbi = EXT4_SB(sb);
987 struct ext4_inode_info *ei;
990 spin_lock(&sbi->s_fc_lock);
991 list_for_each_entry(ei, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
992 ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
993 while (atomic_read(&ei->i_fc_updates)) {
996 prepare_to_wait(&ei->i_fc_wait, &wait,
997 TASK_UNINTERRUPTIBLE);
998 if (atomic_read(&ei->i_fc_updates)) {
999 spin_unlock(&sbi->s_fc_lock);
1001 spin_lock(&sbi->s_fc_lock);
1003 finish_wait(&ei->i_fc_wait, &wait);
1005 spin_unlock(&sbi->s_fc_lock);
1006 ret = jbd2_submit_inode_data(ei->jinode);
1009 spin_lock(&sbi->s_fc_lock);
1011 spin_unlock(&sbi->s_fc_lock);
1016 /* Wait for completion of data for all the fast commit inodes */
1017 static int ext4_fc_wait_inode_data_all(journal_t *journal)
1019 struct super_block *sb = journal->j_private;
1020 struct ext4_sb_info *sbi = EXT4_SB(sb);
1021 struct ext4_inode_info *pos, *n;
1024 spin_lock(&sbi->s_fc_lock);
1025 list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
1026 if (!ext4_test_inode_state(&pos->vfs_inode,
1027 EXT4_STATE_FC_COMMITTING))
1029 spin_unlock(&sbi->s_fc_lock);
1031 ret = jbd2_wait_inode_data(journal, pos->jinode);
1034 spin_lock(&sbi->s_fc_lock);
1036 spin_unlock(&sbi->s_fc_lock);
1041 /* Commit all the directory entry updates */
1042 static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)
1043 __acquires(&sbi->s_fc_lock)
1044 __releases(&sbi->s_fc_lock)
1046 struct super_block *sb = journal->j_private;
1047 struct ext4_sb_info *sbi = EXT4_SB(sb);
1048 struct ext4_fc_dentry_update *fc_dentry, *fc_dentry_n;
1049 struct inode *inode;
1050 struct ext4_inode_info *ei;
1053 if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))
1055 list_for_each_entry_safe(fc_dentry, fc_dentry_n,
1056 &sbi->s_fc_dentry_q[FC_Q_MAIN], fcd_list) {
1057 if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {
1058 spin_unlock(&sbi->s_fc_lock);
1059 if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {
1063 spin_lock(&sbi->s_fc_lock);
1067 * With fcd_dilist we need not loop in sbi->s_fc_q to get the
1068 * corresponding inode pointer
1070 WARN_ON(list_empty(&fc_dentry->fcd_dilist));
1071 ei = list_first_entry(&fc_dentry->fcd_dilist,
1072 struct ext4_inode_info, i_fc_dilist);
1073 inode = &ei->vfs_inode;
1074 WARN_ON(inode->i_ino != fc_dentry->fcd_ino);
1076 spin_unlock(&sbi->s_fc_lock);
1079 * We first write the inode and then the create dirent. This
1080 * allows the recovery code to create an unnamed inode first
1081 * and then link it to a directory entry. This allows us
1082 * to use namei.c routines almost as is and simplifies
1083 * the recovery code.
1085 ret = ext4_fc_write_inode(inode, crc);
1089 ret = ext4_fc_write_inode_data(inode, crc);
1093 if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {
1098 spin_lock(&sbi->s_fc_lock);
1102 spin_lock(&sbi->s_fc_lock);
1106 static int ext4_fc_perform_commit(journal_t *journal)
1108 struct super_block *sb = journal->j_private;
1109 struct ext4_sb_info *sbi = EXT4_SB(sb);
1110 struct ext4_inode_info *iter;
1111 struct ext4_fc_head head;
1112 struct inode *inode;
1113 struct blk_plug plug;
1117 ret = ext4_fc_submit_inode_data_all(journal);
1121 ret = ext4_fc_wait_inode_data_all(journal);
1126 * If file system device is different from journal device, issue a cache
1127 * flush before we start writing fast commit blocks.
1129 if (journal->j_fs_dev != journal->j_dev)
1130 blkdev_issue_flush(journal->j_fs_dev);
1132 blk_start_plug(&plug);
1133 if (sbi->s_fc_bytes == 0) {
1135 * Add a head tag only if this is the first fast commit
1138 head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);
1139 head.fc_tid = cpu_to_le32(
1140 sbi->s_journal->j_running_transaction->t_tid);
1141 if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),
1142 (u8 *)&head, &crc)) {
1148 spin_lock(&sbi->s_fc_lock);
1149 ret = ext4_fc_commit_dentry_updates(journal, &crc);
1151 spin_unlock(&sbi->s_fc_lock);
1155 list_for_each_entry(iter, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
1156 inode = &iter->vfs_inode;
1157 if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))
1160 spin_unlock(&sbi->s_fc_lock);
1161 ret = ext4_fc_write_inode_data(inode, &crc);
1164 ret = ext4_fc_write_inode(inode, &crc);
1167 spin_lock(&sbi->s_fc_lock);
1169 spin_unlock(&sbi->s_fc_lock);
1171 ret = ext4_fc_write_tail(sb, crc);
1174 blk_finish_plug(&plug);
1178 static void ext4_fc_update_stats(struct super_block *sb, int status,
1179 u64 commit_time, int nblks, tid_t commit_tid)
1181 struct ext4_fc_stats *stats = &EXT4_SB(sb)->s_fc_stats;
1183 ext4_debug("Fast commit ended with status = %d for tid %u",
1184 status, commit_tid);
1185 if (status == EXT4_FC_STATUS_OK) {
1186 stats->fc_num_commits++;
1187 stats->fc_numblks += nblks;
1188 if (likely(stats->s_fc_avg_commit_time))
1189 stats->s_fc_avg_commit_time =
1191 stats->s_fc_avg_commit_time * 3) / 4;
1193 stats->s_fc_avg_commit_time = commit_time;
1194 } else if (status == EXT4_FC_STATUS_FAILED ||
1195 status == EXT4_FC_STATUS_INELIGIBLE) {
1196 if (status == EXT4_FC_STATUS_FAILED)
1197 stats->fc_failed_commits++;
1198 stats->fc_ineligible_commits++;
1200 stats->fc_skipped_commits++;
1202 trace_ext4_fc_commit_stop(sb, nblks, status, commit_tid);
1206 * The main commit entry point. Performs a fast commit for transaction
1207 * commit_tid if needed. If it's not possible to perform a fast commit
1208 * due to various reasons, we fall back to full commit. Returns 0
1209 * on success, error otherwise.
1211 int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
1213 struct super_block *sb = journal->j_private;
1214 struct ext4_sb_info *sbi = EXT4_SB(sb);
1215 int nblks = 0, ret, bsize = journal->j_blocksize;
1216 int subtid = atomic_read(&sbi->s_fc_subtid);
1217 int status = EXT4_FC_STATUS_OK, fc_bufs_before = 0;
1218 ktime_t start_time, commit_time;
1220 if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
1221 return jbd2_complete_transaction(journal, commit_tid);
1223 trace_ext4_fc_commit_start(sb, commit_tid);
1225 start_time = ktime_get();
1228 ret = jbd2_fc_begin_commit(journal, commit_tid);
1229 if (ret == -EALREADY) {
1230 /* There was an ongoing commit, check if we need to restart */
1231 if (atomic_read(&sbi->s_fc_subtid) <= subtid &&
1232 commit_tid > journal->j_commit_sequence)
1234 ext4_fc_update_stats(sb, EXT4_FC_STATUS_SKIPPED, 0, 0,
1239 * Commit couldn't start. Just update stats and perform a
1242 ext4_fc_update_stats(sb, EXT4_FC_STATUS_FAILED, 0, 0,
1244 return jbd2_complete_transaction(journal, commit_tid);
1248 * After establishing journal barrier via jbd2_fc_begin_commit(), check
1249 * if we are fast commit ineligible.
1251 if (ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE)) {
1252 status = EXT4_FC_STATUS_INELIGIBLE;
1256 fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
1257 ret = ext4_fc_perform_commit(journal);
1259 status = EXT4_FC_STATUS_FAILED;
1262 nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;
1263 ret = jbd2_fc_wait_bufs(journal, nblks);
1265 status = EXT4_FC_STATUS_FAILED;
1268 atomic_inc(&sbi->s_fc_subtid);
1269 ret = jbd2_fc_end_commit(journal);
1271 * weight the commit time higher than the average time so we
1272 * don't react too strongly to vast changes in the commit time
1274 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1275 ext4_fc_update_stats(sb, status, commit_time, nblks, commit_tid);
1279 ret = jbd2_fc_end_commit_fallback(journal);
1280 ext4_fc_update_stats(sb, status, 0, 0, commit_tid);
1285 * Fast commit cleanup routine. This is called after every fast commit and
1286 * full commit. full is true if we are called after a full commit.
1288 static void ext4_fc_cleanup(journal_t *journal, int full, tid_t tid)
1290 struct super_block *sb = journal->j_private;
1291 struct ext4_sb_info *sbi = EXT4_SB(sb);
1292 struct ext4_inode_info *iter, *iter_n;
1293 struct ext4_fc_dentry_update *fc_dentry;
1295 if (full && sbi->s_fc_bh)
1296 sbi->s_fc_bh = NULL;
1298 trace_ext4_fc_cleanup(journal, full, tid);
1299 jbd2_fc_release_bufs(journal);
1301 spin_lock(&sbi->s_fc_lock);
1302 list_for_each_entry_safe(iter, iter_n, &sbi->s_fc_q[FC_Q_MAIN],
1304 list_del_init(&iter->i_fc_list);
1305 ext4_clear_inode_state(&iter->vfs_inode,
1306 EXT4_STATE_FC_COMMITTING);
1307 if (iter->i_sync_tid <= tid)
1308 ext4_fc_reset_inode(&iter->vfs_inode);
1309 /* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
1311 #if (BITS_PER_LONG < 64)
1312 wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);
1314 wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);
1318 while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {
1319 fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],
1320 struct ext4_fc_dentry_update,
1322 list_del_init(&fc_dentry->fcd_list);
1323 list_del_init(&fc_dentry->fcd_dilist);
1324 spin_unlock(&sbi->s_fc_lock);
1326 if (fc_dentry->fcd_name.name &&
1327 fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
1328 kfree(fc_dentry->fcd_name.name);
1329 kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
1330 spin_lock(&sbi->s_fc_lock);
1333 list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],
1334 &sbi->s_fc_dentry_q[FC_Q_MAIN]);
1335 list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
1336 &sbi->s_fc_q[FC_Q_MAIN]);
1338 if (tid >= sbi->s_fc_ineligible_tid) {
1339 sbi->s_fc_ineligible_tid = 0;
1340 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
1344 sbi->s_fc_bytes = 0;
1345 spin_unlock(&sbi->s_fc_lock);
1346 trace_ext4_fc_stats(sb);
1349 /* Ext4 Replay Path Routines */
1351 /* Helper struct for dentry replay routines */
1352 struct dentry_info_args {
1353 int parent_ino, dname_len, ino, inode_len;
1357 /* Same as struct ext4_fc_tl, but uses native endianness fields */
1358 struct ext4_fc_tl_mem {
1363 static inline void tl_to_darg(struct dentry_info_args *darg,
1364 struct ext4_fc_tl_mem *tl, u8 *val)
1366 struct ext4_fc_dentry_info fcd;
1368 memcpy(&fcd, val, sizeof(fcd));
1370 darg->parent_ino = le32_to_cpu(fcd.fc_parent_ino);
1371 darg->ino = le32_to_cpu(fcd.fc_ino);
1372 darg->dname = val + offsetof(struct ext4_fc_dentry_info, fc_dname);
1373 darg->dname_len = tl->fc_len - sizeof(struct ext4_fc_dentry_info);
1376 static inline void ext4_fc_get_tl(struct ext4_fc_tl_mem *tl, u8 *val)
1378 struct ext4_fc_tl tl_disk;
1380 memcpy(&tl_disk, val, EXT4_FC_TAG_BASE_LEN);
1381 tl->fc_len = le16_to_cpu(tl_disk.fc_len);
1382 tl->fc_tag = le16_to_cpu(tl_disk.fc_tag);
1385 /* Unlink replay function */
1386 static int ext4_fc_replay_unlink(struct super_block *sb,
1387 struct ext4_fc_tl_mem *tl, u8 *val)
1389 struct inode *inode, *old_parent;
1391 struct dentry_info_args darg;
1394 tl_to_darg(&darg, tl, val);
1396 trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,
1397 darg.parent_ino, darg.dname_len);
1399 entry.name = darg.dname;
1400 entry.len = darg.dname_len;
1401 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1403 if (IS_ERR(inode)) {
1404 ext4_debug("Inode %d not found", darg.ino);
1408 old_parent = ext4_iget(sb, darg.parent_ino,
1410 if (IS_ERR(old_parent)) {
1411 ext4_debug("Dir with inode %d not found", darg.parent_ino);
1416 ret = __ext4_unlink(old_parent, &entry, inode, NULL);
1417 /* -ENOENT ok coz it might not exist anymore. */
1425 static int ext4_fc_replay_link_internal(struct super_block *sb,
1426 struct dentry_info_args *darg,
1427 struct inode *inode)
1429 struct inode *dir = NULL;
1430 struct dentry *dentry_dir = NULL, *dentry_inode = NULL;
1431 struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);
1434 dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);
1436 ext4_debug("Dir with inode %d not found.", darg->parent_ino);
1441 dentry_dir = d_obtain_alias(dir);
1442 if (IS_ERR(dentry_dir)) {
1443 ext4_debug("Failed to obtain dentry");
1448 dentry_inode = d_alloc(dentry_dir, &qstr_dname);
1449 if (!dentry_inode) {
1450 ext4_debug("Inode dentry not created.");
1455 ret = __ext4_link(dir, inode, dentry_inode);
1457 * It's possible that link already existed since data blocks
1458 * for the dir in question got persisted before we crashed OR
1459 * we replayed this tag and crashed before the entire replay
1462 if (ret && ret != -EEXIST) {
1463 ext4_debug("Failed to link\n");
1476 d_drop(dentry_inode);
1483 /* Link replay function */
1484 static int ext4_fc_replay_link(struct super_block *sb,
1485 struct ext4_fc_tl_mem *tl, u8 *val)
1487 struct inode *inode;
1488 struct dentry_info_args darg;
1491 tl_to_darg(&darg, tl, val);
1492 trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,
1493 darg.parent_ino, darg.dname_len);
1495 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1496 if (IS_ERR(inode)) {
1497 ext4_debug("Inode not found.");
1501 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1507 * Record all the modified inodes during replay. We use this later to setup
1508 * block bitmaps correctly.
1510 static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)
1512 struct ext4_fc_replay_state *state;
1515 state = &EXT4_SB(sb)->s_fc_replay_state;
1516 for (i = 0; i < state->fc_modified_inodes_used; i++)
1517 if (state->fc_modified_inodes[i] == ino)
1519 if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {
1520 int *fc_modified_inodes;
1522 fc_modified_inodes = krealloc(state->fc_modified_inodes,
1523 sizeof(int) * (state->fc_modified_inodes_size +
1524 EXT4_FC_REPLAY_REALLOC_INCREMENT),
1526 if (!fc_modified_inodes)
1528 state->fc_modified_inodes = fc_modified_inodes;
1529 state->fc_modified_inodes_size +=
1530 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1532 state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;
1537 * Inode replay function
1539 static int ext4_fc_replay_inode(struct super_block *sb,
1540 struct ext4_fc_tl_mem *tl, u8 *val)
1542 struct ext4_fc_inode fc_inode;
1543 struct ext4_inode *raw_inode;
1544 struct ext4_inode *raw_fc_inode;
1545 struct inode *inode = NULL;
1546 struct ext4_iloc iloc;
1547 int inode_len, ino, ret, tag = tl->fc_tag;
1548 struct ext4_extent_header *eh;
1549 size_t off_gen = offsetof(struct ext4_inode, i_generation);
1551 memcpy(&fc_inode, val, sizeof(fc_inode));
1553 ino = le32_to_cpu(fc_inode.fc_ino);
1554 trace_ext4_fc_replay(sb, tag, ino, 0, 0);
1556 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1557 if (!IS_ERR(inode)) {
1558 ext4_ext_clear_bb(inode);
1563 ret = ext4_fc_record_modified_inode(sb, ino);
1567 raw_fc_inode = (struct ext4_inode *)
1568 (val + offsetof(struct ext4_fc_inode, fc_raw_inode));
1569 ret = ext4_get_fc_inode_loc(sb, ino, &iloc);
1573 inode_len = tl->fc_len - sizeof(struct ext4_fc_inode);
1574 raw_inode = ext4_raw_inode(&iloc);
1576 memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));
1577 memcpy((u8 *)raw_inode + off_gen, (u8 *)raw_fc_inode + off_gen,
1578 inode_len - off_gen);
1579 if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {
1580 eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);
1581 if (eh->eh_magic != EXT4_EXT_MAGIC) {
1582 memset(eh, 0, sizeof(*eh));
1583 eh->eh_magic = EXT4_EXT_MAGIC;
1584 eh->eh_max = cpu_to_le16(
1585 (sizeof(raw_inode->i_block) -
1586 sizeof(struct ext4_extent_header))
1587 / sizeof(struct ext4_extent));
1589 } else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {
1590 memcpy(raw_inode->i_block, raw_fc_inode->i_block,
1591 sizeof(raw_inode->i_block));
1594 /* Immediately update the inode on disk. */
1595 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1598 ret = sync_dirty_buffer(iloc.bh);
1601 ret = ext4_mark_inode_used(sb, ino);
1605 /* Given that we just wrote the inode on disk, this SHOULD succeed. */
1606 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1607 if (IS_ERR(inode)) {
1608 ext4_debug("Inode not found.");
1609 return -EFSCORRUPTED;
1613 * Our allocator could have made different decisions than before
1614 * crashing. This should be fixed but until then, we calculate
1615 * the number of blocks the inode.
1617 if (!ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
1618 ext4_ext_replay_set_iblocks(inode);
1620 inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);
1621 ext4_reset_inode_seed(inode);
1623 ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));
1624 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1625 sync_dirty_buffer(iloc.bh);
1630 blkdev_issue_flush(sb->s_bdev);
1636 * Dentry create replay function.
1638 * EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the
1639 * inode for which we are trying to create a dentry here, should already have
1640 * been replayed before we start here.
1642 static int ext4_fc_replay_create(struct super_block *sb,
1643 struct ext4_fc_tl_mem *tl, u8 *val)
1646 struct inode *inode = NULL;
1647 struct inode *dir = NULL;
1648 struct dentry_info_args darg;
1650 tl_to_darg(&darg, tl, val);
1652 trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,
1653 darg.parent_ino, darg.dname_len);
1655 /* This takes care of update group descriptor and other metadata */
1656 ret = ext4_mark_inode_used(sb, darg.ino);
1660 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1661 if (IS_ERR(inode)) {
1662 ext4_debug("inode %d not found.", darg.ino);
1668 if (S_ISDIR(inode->i_mode)) {
1670 * If we are creating a directory, we need to make sure that the
1671 * dot and dot dot dirents are setup properly.
1673 dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);
1675 ext4_debug("Dir %d not found.", darg.ino);
1678 ret = ext4_init_new_dir(NULL, dir, inode);
1685 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1688 set_nlink(inode, 1);
1689 ext4_mark_inode_dirty(NULL, inode);
1696 * Record physical disk regions which are in use as per fast commit area,
1697 * and used by inodes during replay phase. Our simple replay phase
1698 * allocator excludes these regions from allocation.
1700 int ext4_fc_record_regions(struct super_block *sb, int ino,
1701 ext4_lblk_t lblk, ext4_fsblk_t pblk, int len, int replay)
1703 struct ext4_fc_replay_state *state;
1704 struct ext4_fc_alloc_region *region;
1706 state = &EXT4_SB(sb)->s_fc_replay_state;
1708 * during replay phase, the fc_regions_valid may not same as
1709 * fc_regions_used, update it when do new additions.
1711 if (replay && state->fc_regions_used != state->fc_regions_valid)
1712 state->fc_regions_used = state->fc_regions_valid;
1713 if (state->fc_regions_used == state->fc_regions_size) {
1714 struct ext4_fc_alloc_region *fc_regions;
1716 fc_regions = krealloc(state->fc_regions,
1717 sizeof(struct ext4_fc_alloc_region) *
1718 (state->fc_regions_size +
1719 EXT4_FC_REPLAY_REALLOC_INCREMENT),
1723 state->fc_regions_size +=
1724 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1725 state->fc_regions = fc_regions;
1727 region = &state->fc_regions[state->fc_regions_used++];
1729 region->lblk = lblk;
1730 region->pblk = pblk;
1734 state->fc_regions_valid++;
1739 /* Replay add range tag */
1740 static int ext4_fc_replay_add_range(struct super_block *sb,
1741 struct ext4_fc_tl_mem *tl, u8 *val)
1743 struct ext4_fc_add_range fc_add_ex;
1744 struct ext4_extent newex, *ex;
1745 struct inode *inode;
1746 ext4_lblk_t start, cur;
1748 ext4_fsblk_t start_pblk;
1749 struct ext4_map_blocks map;
1750 struct ext4_ext_path *path = NULL;
1753 memcpy(&fc_add_ex, val, sizeof(fc_add_ex));
1754 ex = (struct ext4_extent *)&fc_add_ex.fc_ex;
1756 trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,
1757 le32_to_cpu(fc_add_ex.fc_ino), le32_to_cpu(ex->ee_block),
1758 ext4_ext_get_actual_len(ex));
1760 inode = ext4_iget(sb, le32_to_cpu(fc_add_ex.fc_ino), EXT4_IGET_NORMAL);
1761 if (IS_ERR(inode)) {
1762 ext4_debug("Inode not found.");
1766 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1770 start = le32_to_cpu(ex->ee_block);
1771 start_pblk = ext4_ext_pblock(ex);
1772 len = ext4_ext_get_actual_len(ex);
1776 ext4_debug("ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",
1777 start, start_pblk, len, ext4_ext_is_unwritten(ex),
1780 while (remaining > 0) {
1782 map.m_len = remaining;
1784 ret = ext4_map_blocks(NULL, inode, &map, 0);
1790 /* Range is not mapped */
1791 path = ext4_find_extent(inode, cur, NULL, 0);
1794 memset(&newex, 0, sizeof(newex));
1795 newex.ee_block = cpu_to_le32(cur);
1796 ext4_ext_store_pblock(
1797 &newex, start_pblk + cur - start);
1798 newex.ee_len = cpu_to_le16(map.m_len);
1799 if (ext4_ext_is_unwritten(ex))
1800 ext4_ext_mark_unwritten(&newex);
1801 down_write(&EXT4_I(inode)->i_data_sem);
1802 ret = ext4_ext_insert_extent(
1803 NULL, inode, &path, &newex, 0);
1804 up_write((&EXT4_I(inode)->i_data_sem));
1805 ext4_free_ext_path(path);
1811 if (start_pblk + cur - start != map.m_pblk) {
1813 * Logical to physical mapping changed. This can happen
1814 * if this range was removed and then reallocated to
1815 * map to new physical blocks during a fast commit.
1817 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1818 ext4_ext_is_unwritten(ex),
1819 start_pblk + cur - start);
1823 * Mark the old blocks as free since they aren't used
1824 * anymore. We maintain an array of all the modified
1825 * inodes. In case these blocks are still used at either
1826 * a different logical range in the same inode or in
1827 * some different inode, we will mark them as allocated
1828 * at the end of the FC replay using our array of
1831 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1835 /* Range is mapped and needs a state change */
1836 ext4_debug("Converting from %ld to %d %lld",
1837 map.m_flags & EXT4_MAP_UNWRITTEN,
1838 ext4_ext_is_unwritten(ex), map.m_pblk);
1839 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1840 ext4_ext_is_unwritten(ex), map.m_pblk);
1844 * We may have split the extent tree while toggling the state.
1845 * Try to shrink the extent tree now.
1847 ext4_ext_replay_shrink_inode(inode, start + len);
1850 remaining -= map.m_len;
1852 ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>
1853 sb->s_blocksize_bits);
1859 /* Replay DEL_RANGE tag */
1861 ext4_fc_replay_del_range(struct super_block *sb,
1862 struct ext4_fc_tl_mem *tl, u8 *val)
1864 struct inode *inode;
1865 struct ext4_fc_del_range lrange;
1866 struct ext4_map_blocks map;
1867 ext4_lblk_t cur, remaining;
1870 memcpy(&lrange, val, sizeof(lrange));
1871 cur = le32_to_cpu(lrange.fc_lblk);
1872 remaining = le32_to_cpu(lrange.fc_len);
1874 trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,
1875 le32_to_cpu(lrange.fc_ino), cur, remaining);
1877 inode = ext4_iget(sb, le32_to_cpu(lrange.fc_ino), EXT4_IGET_NORMAL);
1878 if (IS_ERR(inode)) {
1879 ext4_debug("Inode %d not found", le32_to_cpu(lrange.fc_ino));
1883 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1887 ext4_debug("DEL_RANGE, inode %ld, lblk %d, len %d\n",
1888 inode->i_ino, le32_to_cpu(lrange.fc_lblk),
1889 le32_to_cpu(lrange.fc_len));
1890 while (remaining > 0) {
1892 map.m_len = remaining;
1894 ret = ext4_map_blocks(NULL, inode, &map, 0);
1900 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1902 remaining -= map.m_len;
1907 down_write(&EXT4_I(inode)->i_data_sem);
1908 ret = ext4_ext_remove_space(inode, le32_to_cpu(lrange.fc_lblk),
1909 le32_to_cpu(lrange.fc_lblk) +
1910 le32_to_cpu(lrange.fc_len) - 1);
1911 up_write(&EXT4_I(inode)->i_data_sem);
1914 ext4_ext_replay_shrink_inode(inode,
1915 i_size_read(inode) >> sb->s_blocksize_bits);
1916 ext4_mark_inode_dirty(NULL, inode);
1922 static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)
1924 struct ext4_fc_replay_state *state;
1925 struct inode *inode;
1926 struct ext4_ext_path *path = NULL;
1927 struct ext4_map_blocks map;
1929 ext4_lblk_t cur, end;
1931 state = &EXT4_SB(sb)->s_fc_replay_state;
1932 for (i = 0; i < state->fc_modified_inodes_used; i++) {
1933 inode = ext4_iget(sb, state->fc_modified_inodes[i],
1935 if (IS_ERR(inode)) {
1936 ext4_debug("Inode %d not found.",
1937 state->fc_modified_inodes[i]);
1941 end = EXT_MAX_BLOCKS;
1942 if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA)) {
1948 map.m_len = end - cur;
1950 ret = ext4_map_blocks(NULL, inode, &map, 0);
1955 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
1956 if (!IS_ERR(path)) {
1957 for (j = 0; j < path->p_depth; j++)
1958 ext4_mb_mark_bb(inode->i_sb,
1959 path[j].p_block, 1, 1);
1960 ext4_free_ext_path(path);
1963 ext4_mb_mark_bb(inode->i_sb, map.m_pblk,
1966 cur = cur + (map.m_len ? map.m_len : 1);
1974 * Check if block is in excluded regions for block allocation. The simple
1975 * allocator that runs during replay phase is calls this function to see
1976 * if it is okay to use a block.
1978 bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)
1981 struct ext4_fc_replay_state *state;
1983 state = &EXT4_SB(sb)->s_fc_replay_state;
1984 for (i = 0; i < state->fc_regions_valid; i++) {
1985 if (state->fc_regions[i].ino == 0 ||
1986 state->fc_regions[i].len == 0)
1988 if (in_range(blk, state->fc_regions[i].pblk,
1989 state->fc_regions[i].len))
1995 /* Cleanup function called after replay */
1996 void ext4_fc_replay_cleanup(struct super_block *sb)
1998 struct ext4_sb_info *sbi = EXT4_SB(sb);
2000 sbi->s_mount_state &= ~EXT4_FC_REPLAY;
2001 kfree(sbi->s_fc_replay_state.fc_regions);
2002 kfree(sbi->s_fc_replay_state.fc_modified_inodes);
2005 static bool ext4_fc_value_len_isvalid(struct ext4_sb_info *sbi,
2009 case EXT4_FC_TAG_ADD_RANGE:
2010 return len == sizeof(struct ext4_fc_add_range);
2011 case EXT4_FC_TAG_DEL_RANGE:
2012 return len == sizeof(struct ext4_fc_del_range);
2013 case EXT4_FC_TAG_CREAT:
2014 case EXT4_FC_TAG_LINK:
2015 case EXT4_FC_TAG_UNLINK:
2016 len -= sizeof(struct ext4_fc_dentry_info);
2017 return len >= 1 && len <= EXT4_NAME_LEN;
2018 case EXT4_FC_TAG_INODE:
2019 len -= sizeof(struct ext4_fc_inode);
2020 return len >= EXT4_GOOD_OLD_INODE_SIZE &&
2021 len <= sbi->s_inode_size;
2022 case EXT4_FC_TAG_PAD:
2023 return true; /* padding can have any length */
2024 case EXT4_FC_TAG_TAIL:
2025 return len >= sizeof(struct ext4_fc_tail);
2026 case EXT4_FC_TAG_HEAD:
2027 return len == sizeof(struct ext4_fc_head);
2033 * Recovery Scan phase handler
2035 * This function is called during the scan phase and is responsible
2036 * for doing following things:
2037 * - Make sure the fast commit area has valid tags for replay
2038 * - Count number of tags that need to be replayed by the replay handler
2040 * - Create a list of excluded blocks for allocation during replay phase
2042 * This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is
2043 * incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP
2044 * to indicate that scan has finished and JBD2 can now start replay phase.
2045 * It returns a negative error to indicate that there was an error. At the end
2046 * of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set
2047 * to indicate the number of tags that need to replayed during the replay phase.
2049 static int ext4_fc_replay_scan(journal_t *journal,
2050 struct buffer_head *bh, int off,
2053 struct super_block *sb = journal->j_private;
2054 struct ext4_sb_info *sbi = EXT4_SB(sb);
2055 struct ext4_fc_replay_state *state;
2056 int ret = JBD2_FC_REPLAY_CONTINUE;
2057 struct ext4_fc_add_range ext;
2058 struct ext4_fc_tl_mem tl;
2059 struct ext4_fc_tail tail;
2060 __u8 *start, *end, *cur, *val;
2061 struct ext4_fc_head head;
2062 struct ext4_extent *ex;
2064 state = &sbi->s_fc_replay_state;
2066 start = (u8 *)bh->b_data;
2067 end = start + journal->j_blocksize;
2069 if (state->fc_replay_expected_off == 0) {
2070 state->fc_cur_tag = 0;
2071 state->fc_replay_num_tags = 0;
2073 state->fc_regions = NULL;
2074 state->fc_regions_valid = state->fc_regions_used =
2075 state->fc_regions_size = 0;
2076 /* Check if we can stop early */
2077 if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
2078 != EXT4_FC_TAG_HEAD)
2082 if (off != state->fc_replay_expected_off) {
2083 ret = -EFSCORRUPTED;
2087 state->fc_replay_expected_off++;
2088 for (cur = start; cur <= end - EXT4_FC_TAG_BASE_LEN;
2089 cur = cur + EXT4_FC_TAG_BASE_LEN + tl.fc_len) {
2090 ext4_fc_get_tl(&tl, cur);
2091 val = cur + EXT4_FC_TAG_BASE_LEN;
2092 if (tl.fc_len > end - val ||
2093 !ext4_fc_value_len_isvalid(sbi, tl.fc_tag, tl.fc_len)) {
2094 ret = state->fc_replay_num_tags ?
2095 JBD2_FC_REPLAY_STOP : -ECANCELED;
2098 ext4_debug("Scan phase, tag:%s, blk %lld\n",
2099 tag2str(tl.fc_tag), bh->b_blocknr);
2100 switch (tl.fc_tag) {
2101 case EXT4_FC_TAG_ADD_RANGE:
2102 memcpy(&ext, val, sizeof(ext));
2103 ex = (struct ext4_extent *)&ext.fc_ex;
2104 ret = ext4_fc_record_regions(sb,
2105 le32_to_cpu(ext.fc_ino),
2106 le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
2107 ext4_ext_get_actual_len(ex), 0);
2110 ret = JBD2_FC_REPLAY_CONTINUE;
2112 case EXT4_FC_TAG_DEL_RANGE:
2113 case EXT4_FC_TAG_LINK:
2114 case EXT4_FC_TAG_UNLINK:
2115 case EXT4_FC_TAG_CREAT:
2116 case EXT4_FC_TAG_INODE:
2117 case EXT4_FC_TAG_PAD:
2118 state->fc_cur_tag++;
2119 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
2120 EXT4_FC_TAG_BASE_LEN + tl.fc_len);
2122 case EXT4_FC_TAG_TAIL:
2123 state->fc_cur_tag++;
2124 memcpy(&tail, val, sizeof(tail));
2125 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
2126 EXT4_FC_TAG_BASE_LEN +
2127 offsetof(struct ext4_fc_tail,
2129 if (le32_to_cpu(tail.fc_tid) == expected_tid &&
2130 le32_to_cpu(tail.fc_crc) == state->fc_crc) {
2131 state->fc_replay_num_tags = state->fc_cur_tag;
2132 state->fc_regions_valid =
2133 state->fc_regions_used;
2135 ret = state->fc_replay_num_tags ?
2136 JBD2_FC_REPLAY_STOP : -EFSBADCRC;
2140 case EXT4_FC_TAG_HEAD:
2141 memcpy(&head, val, sizeof(head));
2142 if (le32_to_cpu(head.fc_features) &
2143 ~EXT4_FC_SUPPORTED_FEATURES) {
2147 if (le32_to_cpu(head.fc_tid) != expected_tid) {
2148 ret = JBD2_FC_REPLAY_STOP;
2151 state->fc_cur_tag++;
2152 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
2153 EXT4_FC_TAG_BASE_LEN + tl.fc_len);
2156 ret = state->fc_replay_num_tags ?
2157 JBD2_FC_REPLAY_STOP : -ECANCELED;
2159 if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
2164 trace_ext4_fc_replay_scan(sb, ret, off);
2169 * Main recovery path entry point.
2170 * The meaning of return codes is similar as above.
2172 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
2173 enum passtype pass, int off, tid_t expected_tid)
2175 struct super_block *sb = journal->j_private;
2176 struct ext4_sb_info *sbi = EXT4_SB(sb);
2177 struct ext4_fc_tl_mem tl;
2178 __u8 *start, *end, *cur, *val;
2179 int ret = JBD2_FC_REPLAY_CONTINUE;
2180 struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;
2181 struct ext4_fc_tail tail;
2183 if (pass == PASS_SCAN) {
2184 state->fc_current_pass = PASS_SCAN;
2185 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
2188 if (state->fc_current_pass != pass) {
2189 state->fc_current_pass = pass;
2190 sbi->s_mount_state |= EXT4_FC_REPLAY;
2192 if (!sbi->s_fc_replay_state.fc_replay_num_tags) {
2193 ext4_debug("Replay stops\n");
2194 ext4_fc_set_bitmaps_and_counters(sb);
2198 #ifdef CONFIG_EXT4_DEBUG
2199 if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {
2200 pr_warn("Dropping fc block %d because max_replay set\n", off);
2201 return JBD2_FC_REPLAY_STOP;
2205 start = (u8 *)bh->b_data;
2206 end = start + journal->j_blocksize;
2208 for (cur = start; cur <= end - EXT4_FC_TAG_BASE_LEN;
2209 cur = cur + EXT4_FC_TAG_BASE_LEN + tl.fc_len) {
2210 ext4_fc_get_tl(&tl, cur);
2211 val = cur + EXT4_FC_TAG_BASE_LEN;
2213 if (state->fc_replay_num_tags == 0) {
2214 ret = JBD2_FC_REPLAY_STOP;
2215 ext4_fc_set_bitmaps_and_counters(sb);
2219 ext4_debug("Replay phase, tag:%s\n", tag2str(tl.fc_tag));
2220 state->fc_replay_num_tags--;
2221 switch (tl.fc_tag) {
2222 case EXT4_FC_TAG_LINK:
2223 ret = ext4_fc_replay_link(sb, &tl, val);
2225 case EXT4_FC_TAG_UNLINK:
2226 ret = ext4_fc_replay_unlink(sb, &tl, val);
2228 case EXT4_FC_TAG_ADD_RANGE:
2229 ret = ext4_fc_replay_add_range(sb, &tl, val);
2231 case EXT4_FC_TAG_CREAT:
2232 ret = ext4_fc_replay_create(sb, &tl, val);
2234 case EXT4_FC_TAG_DEL_RANGE:
2235 ret = ext4_fc_replay_del_range(sb, &tl, val);
2237 case EXT4_FC_TAG_INODE:
2238 ret = ext4_fc_replay_inode(sb, &tl, val);
2240 case EXT4_FC_TAG_PAD:
2241 trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,
2244 case EXT4_FC_TAG_TAIL:
2245 trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL,
2247 memcpy(&tail, val, sizeof(tail));
2248 WARN_ON(le32_to_cpu(tail.fc_tid) != expected_tid);
2250 case EXT4_FC_TAG_HEAD:
2253 trace_ext4_fc_replay(sb, tl.fc_tag, 0, tl.fc_len, 0);
2259 ret = JBD2_FC_REPLAY_CONTINUE;
2264 void ext4_fc_init(struct super_block *sb, journal_t *journal)
2267 * We set replay callback even if fast commit disabled because we may
2268 * could still have fast commit blocks that need to be replayed even if
2269 * fast commit has now been turned off.
2271 journal->j_fc_replay_callback = ext4_fc_replay;
2272 if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
2274 journal->j_fc_cleanup_callback = ext4_fc_cleanup;
2277 static const char * const fc_ineligible_reasons[] = {
2278 [EXT4_FC_REASON_XATTR] = "Extended attributes changed",
2279 [EXT4_FC_REASON_CROSS_RENAME] = "Cross rename",
2280 [EXT4_FC_REASON_JOURNAL_FLAG_CHANGE] = "Journal flag changed",
2281 [EXT4_FC_REASON_NOMEM] = "Insufficient memory",
2282 [EXT4_FC_REASON_SWAP_BOOT] = "Swap boot",
2283 [EXT4_FC_REASON_RESIZE] = "Resize",
2284 [EXT4_FC_REASON_RENAME_DIR] = "Dir renamed",
2285 [EXT4_FC_REASON_FALLOC_RANGE] = "Falloc range op",
2286 [EXT4_FC_REASON_INODE_JOURNAL_DATA] = "Data journalling",
2287 [EXT4_FC_REASON_ENCRYPTED_FILENAME] = "Encrypted filename",
2290 int ext4_fc_info_show(struct seq_file *seq, void *v)
2292 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);
2293 struct ext4_fc_stats *stats = &sbi->s_fc_stats;
2296 if (v != SEQ_START_TOKEN)
2300 "fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",
2301 stats->fc_num_commits, stats->fc_ineligible_commits,
2303 div_u64(stats->s_fc_avg_commit_time, 1000));
2304 seq_puts(seq, "Ineligible reasons:\n");
2305 for (i = 0; i < EXT4_FC_REASON_MAX; i++)
2306 seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],
2307 stats->fc_ineligible_reason_count[i]);
2312 int __init ext4_fc_init_dentry_cache(void)
2314 ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,
2315 SLAB_RECLAIM_ACCOUNT);
2317 if (ext4_fc_dentry_cachep == NULL)
2323 void ext4_fc_destroy_dentry_cache(void)
2325 kmem_cache_destroy(ext4_fc_dentry_cachep);