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_waitqueue_head(&ei->i_fc_wait);
203 atomic_set(&ei->i_fc_updates, 0);
206 /* This function must be called with sbi->s_fc_lock held. */
207 static void ext4_fc_wait_committing_inode(struct inode *inode)
208 __releases(&EXT4_SB(inode->i_sb)->s_fc_lock)
210 wait_queue_head_t *wq;
211 struct ext4_inode_info *ei = EXT4_I(inode);
213 #if (BITS_PER_LONG < 64)
214 DEFINE_WAIT_BIT(wait, &ei->i_state_flags,
215 EXT4_STATE_FC_COMMITTING);
216 wq = bit_waitqueue(&ei->i_state_flags,
217 EXT4_STATE_FC_COMMITTING);
219 DEFINE_WAIT_BIT(wait, &ei->i_flags,
220 EXT4_STATE_FC_COMMITTING);
221 wq = bit_waitqueue(&ei->i_flags,
222 EXT4_STATE_FC_COMMITTING);
224 lockdep_assert_held(&EXT4_SB(inode->i_sb)->s_fc_lock);
225 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
226 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
228 finish_wait(wq, &wait.wq_entry);
232 * Inform Ext4's fast about start of an inode update
234 * This function is called by the high level call VFS callbacks before
235 * performing any inode update. This function blocks if there's an ongoing
236 * fast commit on the inode in question.
238 void ext4_fc_start_update(struct inode *inode)
240 struct ext4_inode_info *ei = EXT4_I(inode);
242 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
243 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
247 spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
248 if (list_empty(&ei->i_fc_list))
251 if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
252 ext4_fc_wait_committing_inode(inode);
256 atomic_inc(&ei->i_fc_updates);
257 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
261 * Stop inode update and wake up waiting fast commits if any.
263 void ext4_fc_stop_update(struct inode *inode)
265 struct ext4_inode_info *ei = EXT4_I(inode);
267 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
268 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
271 if (atomic_dec_and_test(&ei->i_fc_updates))
272 wake_up_all(&ei->i_fc_wait);
276 * Remove inode from fast commit list. If the inode is being committed
277 * we wait until inode commit is done.
279 void ext4_fc_del(struct inode *inode)
281 struct ext4_inode_info *ei = EXT4_I(inode);
283 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
284 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
288 spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
289 if (list_empty(&ei->i_fc_list)) {
290 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
294 if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
295 ext4_fc_wait_committing_inode(inode);
298 list_del_init(&ei->i_fc_list);
299 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
303 * Mark file system as fast commit ineligible, and record latest
304 * ineligible transaction tid. This means until the recorded
305 * transaction, commit operation would result in a full jbd2 commit.
307 void ext4_fc_mark_ineligible(struct super_block *sb, int reason, handle_t *handle)
309 struct ext4_sb_info *sbi = EXT4_SB(sb);
312 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
313 (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
316 ext4_set_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
317 if (handle && !IS_ERR(handle))
318 tid = handle->h_transaction->t_tid;
320 read_lock(&sbi->s_journal->j_state_lock);
321 tid = sbi->s_journal->j_running_transaction ?
322 sbi->s_journal->j_running_transaction->t_tid : 0;
323 read_unlock(&sbi->s_journal->j_state_lock);
325 spin_lock(&sbi->s_fc_lock);
326 if (sbi->s_fc_ineligible_tid < tid)
327 sbi->s_fc_ineligible_tid = tid;
328 spin_unlock(&sbi->s_fc_lock);
329 WARN_ON(reason >= EXT4_FC_REASON_MAX);
330 sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;
334 * Generic fast commit tracking function. If this is the first time this we are
335 * called after a full commit, we initialize fast commit fields and then call
336 * __fc_track_fn() with update = 0. If we have already been called after a full
337 * commit, we pass update = 1. Based on that, the track function can determine
338 * if it needs to track a field for the first time or if it needs to just
339 * update the previously tracked value.
341 * If enqueue is set, this function enqueues the inode in fast commit list.
343 static int ext4_fc_track_template(
344 handle_t *handle, struct inode *inode,
345 int (*__fc_track_fn)(struct inode *, void *, bool),
346 void *args, int enqueue)
349 struct ext4_inode_info *ei = EXT4_I(inode);
350 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
354 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
355 (sbi->s_mount_state & EXT4_FC_REPLAY))
358 if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
361 tid = handle->h_transaction->t_tid;
362 mutex_lock(&ei->i_fc_lock);
363 if (tid == ei->i_sync_tid) {
366 ext4_fc_reset_inode(inode);
367 ei->i_sync_tid = tid;
369 ret = __fc_track_fn(inode, args, update);
370 mutex_unlock(&ei->i_fc_lock);
375 spin_lock(&sbi->s_fc_lock);
376 if (list_empty(&EXT4_I(inode)->i_fc_list))
377 list_add_tail(&EXT4_I(inode)->i_fc_list,
378 (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
379 sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING) ?
380 &sbi->s_fc_q[FC_Q_STAGING] :
381 &sbi->s_fc_q[FC_Q_MAIN]);
382 spin_unlock(&sbi->s_fc_lock);
387 struct __track_dentry_update_args {
388 struct dentry *dentry;
392 /* __track_fn for directory entry updates. Called with ei->i_fc_lock. */
393 static int __track_dentry_update(struct inode *inode, void *arg, bool update)
395 struct ext4_fc_dentry_update *node;
396 struct ext4_inode_info *ei = EXT4_I(inode);
397 struct __track_dentry_update_args *dentry_update =
398 (struct __track_dentry_update_args *)arg;
399 struct dentry *dentry = dentry_update->dentry;
400 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
402 mutex_unlock(&ei->i_fc_lock);
403 node = kmem_cache_alloc(ext4_fc_dentry_cachep, GFP_NOFS);
405 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
406 mutex_lock(&ei->i_fc_lock);
410 node->fcd_op = dentry_update->op;
411 node->fcd_parent = dentry->d_parent->d_inode->i_ino;
412 node->fcd_ino = inode->i_ino;
413 if (dentry->d_name.len > DNAME_INLINE_LEN) {
414 node->fcd_name.name = kmalloc(dentry->d_name.len, GFP_NOFS);
415 if (!node->fcd_name.name) {
416 kmem_cache_free(ext4_fc_dentry_cachep, node);
417 ext4_fc_mark_ineligible(inode->i_sb,
418 EXT4_FC_REASON_NOMEM, NULL);
419 mutex_lock(&ei->i_fc_lock);
422 memcpy((u8 *)node->fcd_name.name, dentry->d_name.name,
425 memcpy(node->fcd_iname, dentry->d_name.name,
427 node->fcd_name.name = node->fcd_iname;
429 node->fcd_name.len = dentry->d_name.len;
431 spin_lock(&sbi->s_fc_lock);
432 if (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
433 sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING)
434 list_add_tail(&node->fcd_list,
435 &sbi->s_fc_dentry_q[FC_Q_STAGING]);
437 list_add_tail(&node->fcd_list, &sbi->s_fc_dentry_q[FC_Q_MAIN]);
438 spin_unlock(&sbi->s_fc_lock);
439 mutex_lock(&ei->i_fc_lock);
444 void __ext4_fc_track_unlink(handle_t *handle,
445 struct inode *inode, struct dentry *dentry)
447 struct __track_dentry_update_args args;
450 args.dentry = dentry;
451 args.op = EXT4_FC_TAG_UNLINK;
453 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
455 trace_ext4_fc_track_unlink(inode, dentry, ret);
458 void ext4_fc_track_unlink(handle_t *handle, struct dentry *dentry)
460 __ext4_fc_track_unlink(handle, d_inode(dentry), dentry);
463 void __ext4_fc_track_link(handle_t *handle,
464 struct inode *inode, struct dentry *dentry)
466 struct __track_dentry_update_args args;
469 args.dentry = dentry;
470 args.op = EXT4_FC_TAG_LINK;
472 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
474 trace_ext4_fc_track_link(inode, dentry, ret);
477 void ext4_fc_track_link(handle_t *handle, struct dentry *dentry)
479 __ext4_fc_track_link(handle, d_inode(dentry), dentry);
482 void __ext4_fc_track_create(handle_t *handle, struct inode *inode,
483 struct dentry *dentry)
485 struct __track_dentry_update_args args;
488 args.dentry = dentry;
489 args.op = EXT4_FC_TAG_CREAT;
491 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
493 trace_ext4_fc_track_create(inode, dentry, ret);
496 void ext4_fc_track_create(handle_t *handle, struct dentry *dentry)
498 __ext4_fc_track_create(handle, d_inode(dentry), dentry);
501 /* __track_fn for inode tracking */
502 static int __track_inode(struct inode *inode, void *arg, bool update)
507 EXT4_I(inode)->i_fc_lblk_len = 0;
512 void ext4_fc_track_inode(handle_t *handle, struct inode *inode)
516 if (S_ISDIR(inode->i_mode))
519 if (ext4_should_journal_data(inode)) {
520 ext4_fc_mark_ineligible(inode->i_sb,
521 EXT4_FC_REASON_INODE_JOURNAL_DATA, handle);
525 ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);
526 trace_ext4_fc_track_inode(inode, ret);
529 struct __track_range_args {
530 ext4_lblk_t start, end;
533 /* __track_fn for tracking data updates */
534 static int __track_range(struct inode *inode, void *arg, bool update)
536 struct ext4_inode_info *ei = EXT4_I(inode);
537 ext4_lblk_t oldstart;
538 struct __track_range_args *__arg =
539 (struct __track_range_args *)arg;
541 if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {
542 ext4_debug("Special inode %ld being modified\n", inode->i_ino);
546 oldstart = ei->i_fc_lblk_start;
548 if (update && ei->i_fc_lblk_len > 0) {
549 ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);
551 max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -
552 ei->i_fc_lblk_start + 1;
554 ei->i_fc_lblk_start = __arg->start;
555 ei->i_fc_lblk_len = __arg->end - __arg->start + 1;
561 void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,
564 struct __track_range_args args;
567 if (S_ISDIR(inode->i_mode))
573 ret = ext4_fc_track_template(handle, inode, __track_range, &args, 1);
575 trace_ext4_fc_track_range(inode, start, end, ret);
578 static void ext4_fc_submit_bh(struct super_block *sb, bool is_tail)
580 int write_flags = REQ_SYNC;
581 struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
583 /* Add REQ_FUA | REQ_PREFLUSH only its tail */
584 if (test_opt(sb, BARRIER) && is_tail)
585 write_flags |= REQ_FUA | REQ_PREFLUSH;
587 set_buffer_dirty(bh);
588 set_buffer_uptodate(bh);
589 bh->b_end_io = ext4_end_buffer_io_sync;
590 submit_bh(REQ_OP_WRITE, write_flags, bh);
591 EXT4_SB(sb)->s_fc_bh = NULL;
594 /* Ext4 commit path routines */
596 /* memzero and update CRC */
597 static void *ext4_fc_memzero(struct super_block *sb, void *dst, int len,
602 ret = memset(dst, 0, len);
604 *crc = ext4_chksum(EXT4_SB(sb), *crc, dst, len);
609 * Allocate len bytes on a fast commit buffer.
611 * During the commit time this function is used to manage fast commit
612 * block space. We don't split a fast commit log onto different
613 * blocks. So this function makes sure that if there's not enough space
614 * on the current block, the remaining space in the current block is
615 * marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,
616 * new block is from jbd2 and CRC is updated to reflect the padding
619 static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
621 struct ext4_fc_tl *tl;
622 struct ext4_sb_info *sbi = EXT4_SB(sb);
623 struct buffer_head *bh;
624 int bsize = sbi->s_journal->j_blocksize;
625 int ret, off = sbi->s_fc_bytes % bsize;
629 * After allocating len, we should have space at least for a 0 byte
632 if (len + sizeof(struct ext4_fc_tl) > bsize)
635 if (bsize - off - 1 > len + sizeof(struct ext4_fc_tl)) {
637 * Only allocate from current buffer if we have enough space for
638 * this request AND we have space to add a zero byte padding.
641 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
646 sbi->s_fc_bytes += len;
647 return sbi->s_fc_bh->b_data + off;
649 /* Need to add PAD tag */
650 tl = (struct ext4_fc_tl *)(sbi->s_fc_bh->b_data + off);
651 tl->fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);
652 pad_len = bsize - off - 1 - sizeof(struct ext4_fc_tl);
653 tl->fc_len = cpu_to_le16(pad_len);
655 *crc = ext4_chksum(sbi, *crc, tl, sizeof(*tl));
657 ext4_fc_memzero(sb, tl + 1, pad_len, crc);
658 ext4_fc_submit_bh(sb, false);
660 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
664 sbi->s_fc_bytes = (sbi->s_fc_bytes / bsize + 1) * bsize + len;
665 return sbi->s_fc_bh->b_data;
668 /* memcpy to fc reserved space and update CRC */
669 static void *ext4_fc_memcpy(struct super_block *sb, void *dst, const void *src,
673 *crc = ext4_chksum(EXT4_SB(sb), *crc, src, len);
674 return memcpy(dst, src, len);
678 * Complete a fast commit by writing tail tag.
680 * Writing tail tag marks the end of a fast commit. In order to guarantee
681 * atomicity, after writing tail tag, even if there's space remaining
682 * in the block, next commit shouldn't use it. That's why tail tag
683 * has the length as that of the remaining space on the block.
685 static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
687 struct ext4_sb_info *sbi = EXT4_SB(sb);
688 struct ext4_fc_tl tl;
689 struct ext4_fc_tail tail;
690 int off, bsize = sbi->s_journal->j_blocksize;
694 * ext4_fc_reserve_space takes care of allocating an extra block if
695 * there's no enough space on this block for accommodating this tail.
697 dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(tail), &crc);
701 off = sbi->s_fc_bytes % bsize;
703 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);
704 tl.fc_len = cpu_to_le16(bsize - off - 1 + sizeof(struct ext4_fc_tail));
705 sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);
707 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), &crc);
709 tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);
710 ext4_fc_memcpy(sb, dst, &tail.fc_tid, sizeof(tail.fc_tid), &crc);
711 dst += sizeof(tail.fc_tid);
712 tail.fc_crc = cpu_to_le32(crc);
713 ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
715 ext4_fc_submit_bh(sb, true);
721 * Adds tag, length, value and updates CRC. Returns true if tlv was added.
722 * Returns false if there's not enough space.
724 static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,
727 struct ext4_fc_tl tl;
730 dst = ext4_fc_reserve_space(sb, sizeof(tl) + len, crc);
734 tl.fc_tag = cpu_to_le16(tag);
735 tl.fc_len = cpu_to_le16(len);
737 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
738 ext4_fc_memcpy(sb, dst + sizeof(tl), val, len, crc);
743 /* Same as above, but adds dentry tlv. */
744 static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u32 *crc,
745 struct ext4_fc_dentry_update *fc_dentry)
747 struct ext4_fc_dentry_info fcd;
748 struct ext4_fc_tl tl;
749 int dlen = fc_dentry->fcd_name.len;
750 u8 *dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(fcd) + dlen,
756 fcd.fc_parent_ino = cpu_to_le32(fc_dentry->fcd_parent);
757 fcd.fc_ino = cpu_to_le32(fc_dentry->fcd_ino);
758 tl.fc_tag = cpu_to_le16(fc_dentry->fcd_op);
759 tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);
760 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
762 ext4_fc_memcpy(sb, dst, &fcd, sizeof(fcd), crc);
764 ext4_fc_memcpy(sb, dst, fc_dentry->fcd_name.name, dlen, crc);
770 * Writes inode in the fast commit space under TLV with tag @tag.
771 * Returns 0 on success, error on failure.
773 static int ext4_fc_write_inode(struct inode *inode, u32 *crc)
775 struct ext4_inode_info *ei = EXT4_I(inode);
776 int inode_len = EXT4_GOOD_OLD_INODE_SIZE;
778 struct ext4_iloc iloc;
779 struct ext4_fc_inode fc_inode;
780 struct ext4_fc_tl tl;
783 ret = ext4_get_inode_loc(inode, &iloc);
787 if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
788 inode_len = EXT4_INODE_SIZE(inode->i_sb);
789 else if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)
790 inode_len += ei->i_extra_isize;
792 fc_inode.fc_ino = cpu_to_le32(inode->i_ino);
793 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);
794 tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));
796 dst = ext4_fc_reserve_space(inode->i_sb,
797 sizeof(tl) + inode_len + sizeof(fc_inode.fc_ino), crc);
801 if (!ext4_fc_memcpy(inode->i_sb, dst, &tl, sizeof(tl), crc))
804 if (!ext4_fc_memcpy(inode->i_sb, dst, &fc_inode, sizeof(fc_inode), crc))
806 dst += sizeof(fc_inode);
807 if (!ext4_fc_memcpy(inode->i_sb, dst, (u8 *)ext4_raw_inode(&iloc),
815 * Writes updated data ranges for the inode in question. Updates CRC.
816 * Returns 0 on success, error otherwise.
818 static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)
820 ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;
821 struct ext4_inode_info *ei = EXT4_I(inode);
822 struct ext4_map_blocks map;
823 struct ext4_fc_add_range fc_ext;
824 struct ext4_fc_del_range lrange;
825 struct ext4_extent *ex;
828 mutex_lock(&ei->i_fc_lock);
829 if (ei->i_fc_lblk_len == 0) {
830 mutex_unlock(&ei->i_fc_lock);
833 old_blk_size = ei->i_fc_lblk_start;
834 new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;
835 ei->i_fc_lblk_len = 0;
836 mutex_unlock(&ei->i_fc_lock);
838 cur_lblk_off = old_blk_size;
839 jbd_debug(1, "%s: will try writing %d to %d for inode %ld\n",
840 __func__, cur_lblk_off, new_blk_size, inode->i_ino);
842 while (cur_lblk_off <= new_blk_size) {
843 map.m_lblk = cur_lblk_off;
844 map.m_len = new_blk_size - cur_lblk_off + 1;
845 ret = ext4_map_blocks(NULL, inode, &map, 0);
849 if (map.m_len == 0) {
855 lrange.fc_ino = cpu_to_le32(inode->i_ino);
856 lrange.fc_lblk = cpu_to_le32(map.m_lblk);
857 lrange.fc_len = cpu_to_le32(map.m_len);
858 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,
859 sizeof(lrange), (u8 *)&lrange, crc))
862 unsigned int max = (map.m_flags & EXT4_MAP_UNWRITTEN) ?
863 EXT_UNWRITTEN_MAX_LEN : EXT_INIT_MAX_LEN;
865 /* Limit the number of blocks in one extent */
866 map.m_len = min(max, map.m_len);
868 fc_ext.fc_ino = cpu_to_le32(inode->i_ino);
869 ex = (struct ext4_extent *)&fc_ext.fc_ex;
870 ex->ee_block = cpu_to_le32(map.m_lblk);
871 ex->ee_len = cpu_to_le16(map.m_len);
872 ext4_ext_store_pblock(ex, map.m_pblk);
873 if (map.m_flags & EXT4_MAP_UNWRITTEN)
874 ext4_ext_mark_unwritten(ex);
876 ext4_ext_mark_initialized(ex);
877 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,
878 sizeof(fc_ext), (u8 *)&fc_ext, crc))
882 cur_lblk_off += map.m_len;
889 /* Submit data for all the fast commit inodes */
890 static int ext4_fc_submit_inode_data_all(journal_t *journal)
892 struct super_block *sb = (struct super_block *)(journal->j_private);
893 struct ext4_sb_info *sbi = EXT4_SB(sb);
894 struct ext4_inode_info *ei;
897 spin_lock(&sbi->s_fc_lock);
898 list_for_each_entry(ei, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
899 ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
900 while (atomic_read(&ei->i_fc_updates)) {
903 prepare_to_wait(&ei->i_fc_wait, &wait,
904 TASK_UNINTERRUPTIBLE);
905 if (atomic_read(&ei->i_fc_updates)) {
906 spin_unlock(&sbi->s_fc_lock);
908 spin_lock(&sbi->s_fc_lock);
910 finish_wait(&ei->i_fc_wait, &wait);
912 spin_unlock(&sbi->s_fc_lock);
913 ret = jbd2_submit_inode_data(ei->jinode);
916 spin_lock(&sbi->s_fc_lock);
918 spin_unlock(&sbi->s_fc_lock);
923 /* Wait for completion of data for all the fast commit inodes */
924 static int ext4_fc_wait_inode_data_all(journal_t *journal)
926 struct super_block *sb = (struct super_block *)(journal->j_private);
927 struct ext4_sb_info *sbi = EXT4_SB(sb);
928 struct ext4_inode_info *pos, *n;
931 spin_lock(&sbi->s_fc_lock);
932 list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
933 if (!ext4_test_inode_state(&pos->vfs_inode,
934 EXT4_STATE_FC_COMMITTING))
936 spin_unlock(&sbi->s_fc_lock);
938 ret = jbd2_wait_inode_data(journal, pos->jinode);
941 spin_lock(&sbi->s_fc_lock);
943 spin_unlock(&sbi->s_fc_lock);
948 /* Commit all the directory entry updates */
949 static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)
950 __acquires(&sbi->s_fc_lock)
951 __releases(&sbi->s_fc_lock)
953 struct super_block *sb = (struct super_block *)(journal->j_private);
954 struct ext4_sb_info *sbi = EXT4_SB(sb);
955 struct ext4_fc_dentry_update *fc_dentry, *fc_dentry_n;
957 struct ext4_inode_info *ei, *ei_n;
960 if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))
962 list_for_each_entry_safe(fc_dentry, fc_dentry_n,
963 &sbi->s_fc_dentry_q[FC_Q_MAIN], fcd_list) {
964 if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {
965 spin_unlock(&sbi->s_fc_lock);
966 if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {
970 spin_lock(&sbi->s_fc_lock);
975 list_for_each_entry_safe(ei, ei_n, &sbi->s_fc_q[FC_Q_MAIN],
977 if (ei->vfs_inode.i_ino == fc_dentry->fcd_ino) {
978 inode = &ei->vfs_inode;
983 * If we don't find inode in our list, then it was deleted,
984 * in which case, we don't need to record it's create tag.
988 spin_unlock(&sbi->s_fc_lock);
991 * We first write the inode and then the create dirent. This
992 * allows the recovery code to create an unnamed inode first
993 * and then link it to a directory entry. This allows us
994 * to use namei.c routines almost as is and simplifies
997 ret = ext4_fc_write_inode(inode, crc);
1001 ret = ext4_fc_write_inode_data(inode, crc);
1005 if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {
1010 spin_lock(&sbi->s_fc_lock);
1014 spin_lock(&sbi->s_fc_lock);
1018 static int ext4_fc_perform_commit(journal_t *journal)
1020 struct super_block *sb = (struct super_block *)(journal->j_private);
1021 struct ext4_sb_info *sbi = EXT4_SB(sb);
1022 struct ext4_inode_info *iter;
1023 struct ext4_fc_head head;
1024 struct inode *inode;
1025 struct blk_plug plug;
1029 ret = ext4_fc_submit_inode_data_all(journal);
1033 ret = ext4_fc_wait_inode_data_all(journal);
1038 * If file system device is different from journal device, issue a cache
1039 * flush before we start writing fast commit blocks.
1041 if (journal->j_fs_dev != journal->j_dev)
1042 blkdev_issue_flush(journal->j_fs_dev);
1044 blk_start_plug(&plug);
1045 if (sbi->s_fc_bytes == 0) {
1047 * Add a head tag only if this is the first fast commit
1050 head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);
1051 head.fc_tid = cpu_to_le32(
1052 sbi->s_journal->j_running_transaction->t_tid);
1053 if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),
1054 (u8 *)&head, &crc)) {
1060 spin_lock(&sbi->s_fc_lock);
1061 ret = ext4_fc_commit_dentry_updates(journal, &crc);
1063 spin_unlock(&sbi->s_fc_lock);
1067 list_for_each_entry(iter, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
1068 inode = &iter->vfs_inode;
1069 if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))
1072 spin_unlock(&sbi->s_fc_lock);
1073 ret = ext4_fc_write_inode_data(inode, &crc);
1076 ret = ext4_fc_write_inode(inode, &crc);
1079 spin_lock(&sbi->s_fc_lock);
1081 spin_unlock(&sbi->s_fc_lock);
1083 ret = ext4_fc_write_tail(sb, crc);
1086 blk_finish_plug(&plug);
1090 static void ext4_fc_update_stats(struct super_block *sb, int status,
1091 u64 commit_time, int nblks)
1093 struct ext4_fc_stats *stats = &EXT4_SB(sb)->s_fc_stats;
1095 jbd_debug(1, "Fast commit ended with status = %d", status);
1096 if (status == EXT4_FC_STATUS_OK) {
1097 stats->fc_num_commits++;
1098 stats->fc_numblks += nblks;
1099 if (likely(stats->s_fc_avg_commit_time))
1100 stats->s_fc_avg_commit_time =
1102 stats->s_fc_avg_commit_time * 3) / 4;
1104 stats->s_fc_avg_commit_time = commit_time;
1105 } else if (status == EXT4_FC_STATUS_FAILED ||
1106 status == EXT4_FC_STATUS_INELIGIBLE) {
1107 if (status == EXT4_FC_STATUS_FAILED)
1108 stats->fc_failed_commits++;
1109 stats->fc_ineligible_commits++;
1111 stats->fc_skipped_commits++;
1113 trace_ext4_fc_commit_stop(sb, nblks, status);
1117 * The main commit entry point. Performs a fast commit for transaction
1118 * commit_tid if needed. If it's not possible to perform a fast commit
1119 * due to various reasons, we fall back to full commit. Returns 0
1120 * on success, error otherwise.
1122 int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
1124 struct super_block *sb = (struct super_block *)(journal->j_private);
1125 struct ext4_sb_info *sbi = EXT4_SB(sb);
1126 int nblks = 0, ret, bsize = journal->j_blocksize;
1127 int subtid = atomic_read(&sbi->s_fc_subtid);
1128 int status = EXT4_FC_STATUS_OK, fc_bufs_before = 0;
1129 ktime_t start_time, commit_time;
1131 trace_ext4_fc_commit_start(sb);
1133 start_time = ktime_get();
1135 if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
1136 return jbd2_complete_transaction(journal, commit_tid);
1139 ret = jbd2_fc_begin_commit(journal, commit_tid);
1140 if (ret == -EALREADY) {
1141 /* There was an ongoing commit, check if we need to restart */
1142 if (atomic_read(&sbi->s_fc_subtid) <= subtid &&
1143 commit_tid > journal->j_commit_sequence)
1145 ext4_fc_update_stats(sb, EXT4_FC_STATUS_SKIPPED, 0, 0);
1149 * Commit couldn't start. Just update stats and perform a
1152 ext4_fc_update_stats(sb, EXT4_FC_STATUS_FAILED, 0, 0);
1153 return jbd2_complete_transaction(journal, commit_tid);
1157 * After establishing journal barrier via jbd2_fc_begin_commit(), check
1158 * if we are fast commit ineligible.
1160 if (ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE)) {
1161 status = EXT4_FC_STATUS_INELIGIBLE;
1165 fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
1166 ret = ext4_fc_perform_commit(journal);
1168 status = EXT4_FC_STATUS_FAILED;
1171 nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;
1172 ret = jbd2_fc_wait_bufs(journal, nblks);
1174 status = EXT4_FC_STATUS_FAILED;
1177 atomic_inc(&sbi->s_fc_subtid);
1178 ret = jbd2_fc_end_commit(journal);
1180 * weight the commit time higher than the average time so we
1181 * don't react too strongly to vast changes in the commit time
1183 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1184 ext4_fc_update_stats(sb, status, commit_time, nblks);
1188 ret = jbd2_fc_end_commit_fallback(journal);
1189 ext4_fc_update_stats(sb, status, 0, 0);
1194 * Fast commit cleanup routine. This is called after every fast commit and
1195 * full commit. full is true if we are called after a full commit.
1197 static void ext4_fc_cleanup(journal_t *journal, int full, tid_t tid)
1199 struct super_block *sb = journal->j_private;
1200 struct ext4_sb_info *sbi = EXT4_SB(sb);
1201 struct ext4_inode_info *iter, *iter_n;
1202 struct ext4_fc_dentry_update *fc_dentry;
1204 if (full && sbi->s_fc_bh)
1205 sbi->s_fc_bh = NULL;
1207 jbd2_fc_release_bufs(journal);
1209 spin_lock(&sbi->s_fc_lock);
1210 list_for_each_entry_safe(iter, iter_n, &sbi->s_fc_q[FC_Q_MAIN],
1212 list_del_init(&iter->i_fc_list);
1213 ext4_clear_inode_state(&iter->vfs_inode,
1214 EXT4_STATE_FC_COMMITTING);
1215 if (iter->i_sync_tid <= tid)
1216 ext4_fc_reset_inode(&iter->vfs_inode);
1217 /* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
1219 #if (BITS_PER_LONG < 64)
1220 wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);
1222 wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);
1226 while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {
1227 fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],
1228 struct ext4_fc_dentry_update,
1230 list_del_init(&fc_dentry->fcd_list);
1231 spin_unlock(&sbi->s_fc_lock);
1233 if (fc_dentry->fcd_name.name &&
1234 fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
1235 kfree(fc_dentry->fcd_name.name);
1236 kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
1237 spin_lock(&sbi->s_fc_lock);
1240 list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],
1241 &sbi->s_fc_dentry_q[FC_Q_MAIN]);
1242 list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
1243 &sbi->s_fc_q[FC_Q_MAIN]);
1245 if (tid >= sbi->s_fc_ineligible_tid) {
1246 sbi->s_fc_ineligible_tid = 0;
1247 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
1251 sbi->s_fc_bytes = 0;
1252 spin_unlock(&sbi->s_fc_lock);
1253 trace_ext4_fc_stats(sb);
1256 /* Ext4 Replay Path Routines */
1258 /* Helper struct for dentry replay routines */
1259 struct dentry_info_args {
1260 int parent_ino, dname_len, ino, inode_len;
1264 static inline void tl_to_darg(struct dentry_info_args *darg,
1265 struct ext4_fc_tl *tl, u8 *val)
1267 struct ext4_fc_dentry_info fcd;
1269 memcpy(&fcd, val, sizeof(fcd));
1271 darg->parent_ino = le32_to_cpu(fcd.fc_parent_ino);
1272 darg->ino = le32_to_cpu(fcd.fc_ino);
1273 darg->dname = val + offsetof(struct ext4_fc_dentry_info, fc_dname);
1274 darg->dname_len = le16_to_cpu(tl->fc_len) -
1275 sizeof(struct ext4_fc_dentry_info);
1278 /* Unlink replay function */
1279 static int ext4_fc_replay_unlink(struct super_block *sb, struct ext4_fc_tl *tl,
1282 struct inode *inode, *old_parent;
1284 struct dentry_info_args darg;
1287 tl_to_darg(&darg, tl, val);
1289 trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,
1290 darg.parent_ino, darg.dname_len);
1292 entry.name = darg.dname;
1293 entry.len = darg.dname_len;
1294 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1296 if (IS_ERR(inode)) {
1297 jbd_debug(1, "Inode %d not found", darg.ino);
1301 old_parent = ext4_iget(sb, darg.parent_ino,
1303 if (IS_ERR(old_parent)) {
1304 jbd_debug(1, "Dir with inode %d not found", darg.parent_ino);
1309 ret = __ext4_unlink(NULL, old_parent, &entry, inode);
1310 /* -ENOENT ok coz it might not exist anymore. */
1318 static int ext4_fc_replay_link_internal(struct super_block *sb,
1319 struct dentry_info_args *darg,
1320 struct inode *inode)
1322 struct inode *dir = NULL;
1323 struct dentry *dentry_dir = NULL, *dentry_inode = NULL;
1324 struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);
1327 dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);
1329 jbd_debug(1, "Dir with inode %d not found.", darg->parent_ino);
1334 dentry_dir = d_obtain_alias(dir);
1335 if (IS_ERR(dentry_dir)) {
1336 jbd_debug(1, "Failed to obtain dentry");
1341 dentry_inode = d_alloc(dentry_dir, &qstr_dname);
1342 if (!dentry_inode) {
1343 jbd_debug(1, "Inode dentry not created.");
1348 ret = __ext4_link(dir, inode, dentry_inode);
1350 * It's possible that link already existed since data blocks
1351 * for the dir in question got persisted before we crashed OR
1352 * we replayed this tag and crashed before the entire replay
1355 if (ret && ret != -EEXIST) {
1356 jbd_debug(1, "Failed to link\n");
1369 d_drop(dentry_inode);
1376 /* Link replay function */
1377 static int ext4_fc_replay_link(struct super_block *sb, struct ext4_fc_tl *tl,
1380 struct inode *inode;
1381 struct dentry_info_args darg;
1384 tl_to_darg(&darg, tl, val);
1385 trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,
1386 darg.parent_ino, darg.dname_len);
1388 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1389 if (IS_ERR(inode)) {
1390 jbd_debug(1, "Inode not found.");
1394 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1400 * Record all the modified inodes during replay. We use this later to setup
1401 * block bitmaps correctly.
1403 static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)
1405 struct ext4_fc_replay_state *state;
1408 state = &EXT4_SB(sb)->s_fc_replay_state;
1409 for (i = 0; i < state->fc_modified_inodes_used; i++)
1410 if (state->fc_modified_inodes[i] == ino)
1412 if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {
1413 state->fc_modified_inodes = krealloc(
1414 state->fc_modified_inodes,
1415 sizeof(int) * (state->fc_modified_inodes_size +
1416 EXT4_FC_REPLAY_REALLOC_INCREMENT),
1418 if (!state->fc_modified_inodes)
1420 state->fc_modified_inodes_size +=
1421 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1423 state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;
1428 * Inode replay function
1430 static int ext4_fc_replay_inode(struct super_block *sb, struct ext4_fc_tl *tl,
1433 struct ext4_fc_inode fc_inode;
1434 struct ext4_inode *raw_inode;
1435 struct ext4_inode *raw_fc_inode;
1436 struct inode *inode = NULL;
1437 struct ext4_iloc iloc;
1438 int inode_len, ino, ret, tag = le16_to_cpu(tl->fc_tag);
1439 struct ext4_extent_header *eh;
1441 memcpy(&fc_inode, val, sizeof(fc_inode));
1443 ino = le32_to_cpu(fc_inode.fc_ino);
1444 trace_ext4_fc_replay(sb, tag, ino, 0, 0);
1446 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1447 if (!IS_ERR(inode)) {
1448 ext4_ext_clear_bb(inode);
1453 ret = ext4_fc_record_modified_inode(sb, ino);
1457 raw_fc_inode = (struct ext4_inode *)
1458 (val + offsetof(struct ext4_fc_inode, fc_raw_inode));
1459 ret = ext4_get_fc_inode_loc(sb, ino, &iloc);
1463 inode_len = le16_to_cpu(tl->fc_len) - sizeof(struct ext4_fc_inode);
1464 raw_inode = ext4_raw_inode(&iloc);
1466 memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));
1467 memcpy(&raw_inode->i_generation, &raw_fc_inode->i_generation,
1468 inode_len - offsetof(struct ext4_inode, i_generation));
1469 if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {
1470 eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);
1471 if (eh->eh_magic != EXT4_EXT_MAGIC) {
1472 memset(eh, 0, sizeof(*eh));
1473 eh->eh_magic = EXT4_EXT_MAGIC;
1474 eh->eh_max = cpu_to_le16(
1475 (sizeof(raw_inode->i_block) -
1476 sizeof(struct ext4_extent_header))
1477 / sizeof(struct ext4_extent));
1479 } else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {
1480 memcpy(raw_inode->i_block, raw_fc_inode->i_block,
1481 sizeof(raw_inode->i_block));
1484 /* Immediately update the inode on disk. */
1485 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1488 ret = sync_dirty_buffer(iloc.bh);
1491 ret = ext4_mark_inode_used(sb, ino);
1495 /* Given that we just wrote the inode on disk, this SHOULD succeed. */
1496 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1497 if (IS_ERR(inode)) {
1498 jbd_debug(1, "Inode not found.");
1499 return -EFSCORRUPTED;
1503 * Our allocator could have made different decisions than before
1504 * crashing. This should be fixed but until then, we calculate
1505 * the number of blocks the inode.
1507 if (!ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
1508 ext4_ext_replay_set_iblocks(inode);
1510 inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);
1511 ext4_reset_inode_seed(inode);
1513 ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));
1514 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1515 sync_dirty_buffer(iloc.bh);
1520 blkdev_issue_flush(sb->s_bdev);
1526 * Dentry create replay function.
1528 * EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the
1529 * inode for which we are trying to create a dentry here, should already have
1530 * been replayed before we start here.
1532 static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
1536 struct inode *inode = NULL;
1537 struct inode *dir = NULL;
1538 struct dentry_info_args darg;
1540 tl_to_darg(&darg, tl, val);
1542 trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,
1543 darg.parent_ino, darg.dname_len);
1545 /* This takes care of update group descriptor and other metadata */
1546 ret = ext4_mark_inode_used(sb, darg.ino);
1550 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1551 if (IS_ERR(inode)) {
1552 jbd_debug(1, "inode %d not found.", darg.ino);
1558 if (S_ISDIR(inode->i_mode)) {
1560 * If we are creating a directory, we need to make sure that the
1561 * dot and dot dot dirents are setup properly.
1563 dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);
1565 jbd_debug(1, "Dir %d not found.", darg.ino);
1568 ret = ext4_init_new_dir(NULL, dir, inode);
1575 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1578 set_nlink(inode, 1);
1579 ext4_mark_inode_dirty(NULL, inode);
1587 * Record physical disk regions which are in use as per fast commit area,
1588 * and used by inodes during replay phase. Our simple replay phase
1589 * allocator excludes these regions from allocation.
1591 int ext4_fc_record_regions(struct super_block *sb, int ino,
1592 ext4_lblk_t lblk, ext4_fsblk_t pblk, int len, int replay)
1594 struct ext4_fc_replay_state *state;
1595 struct ext4_fc_alloc_region *region;
1597 state = &EXT4_SB(sb)->s_fc_replay_state;
1599 * during replay phase, the fc_regions_valid may not same as
1600 * fc_regions_used, update it when do new additions.
1602 if (replay && state->fc_regions_used != state->fc_regions_valid)
1603 state->fc_regions_used = state->fc_regions_valid;
1604 if (state->fc_regions_used == state->fc_regions_size) {
1605 state->fc_regions_size +=
1606 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1607 state->fc_regions = krealloc(
1609 state->fc_regions_size *
1610 sizeof(struct ext4_fc_alloc_region),
1612 if (!state->fc_regions)
1615 region = &state->fc_regions[state->fc_regions_used++];
1617 region->lblk = lblk;
1618 region->pblk = pblk;
1622 state->fc_regions_valid++;
1627 /* Replay add range tag */
1628 static int ext4_fc_replay_add_range(struct super_block *sb,
1629 struct ext4_fc_tl *tl, u8 *val)
1631 struct ext4_fc_add_range fc_add_ex;
1632 struct ext4_extent newex, *ex;
1633 struct inode *inode;
1634 ext4_lblk_t start, cur;
1636 ext4_fsblk_t start_pblk;
1637 struct ext4_map_blocks map;
1638 struct ext4_ext_path *path = NULL;
1641 memcpy(&fc_add_ex, val, sizeof(fc_add_ex));
1642 ex = (struct ext4_extent *)&fc_add_ex.fc_ex;
1644 trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,
1645 le32_to_cpu(fc_add_ex.fc_ino), le32_to_cpu(ex->ee_block),
1646 ext4_ext_get_actual_len(ex));
1648 inode = ext4_iget(sb, le32_to_cpu(fc_add_ex.fc_ino), EXT4_IGET_NORMAL);
1649 if (IS_ERR(inode)) {
1650 jbd_debug(1, "Inode not found.");
1654 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1658 start = le32_to_cpu(ex->ee_block);
1659 start_pblk = ext4_ext_pblock(ex);
1660 len = ext4_ext_get_actual_len(ex);
1664 jbd_debug(1, "ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",
1665 start, start_pblk, len, ext4_ext_is_unwritten(ex),
1668 while (remaining > 0) {
1670 map.m_len = remaining;
1672 ret = ext4_map_blocks(NULL, inode, &map, 0);
1678 /* Range is not mapped */
1679 path = ext4_find_extent(inode, cur, NULL, 0);
1682 memset(&newex, 0, sizeof(newex));
1683 newex.ee_block = cpu_to_le32(cur);
1684 ext4_ext_store_pblock(
1685 &newex, start_pblk + cur - start);
1686 newex.ee_len = cpu_to_le16(map.m_len);
1687 if (ext4_ext_is_unwritten(ex))
1688 ext4_ext_mark_unwritten(&newex);
1689 down_write(&EXT4_I(inode)->i_data_sem);
1690 ret = ext4_ext_insert_extent(
1691 NULL, inode, &path, &newex, 0);
1692 up_write((&EXT4_I(inode)->i_data_sem));
1693 ext4_ext_drop_refs(path);
1700 if (start_pblk + cur - start != map.m_pblk) {
1702 * Logical to physical mapping changed. This can happen
1703 * if this range was removed and then reallocated to
1704 * map to new physical blocks during a fast commit.
1706 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1707 ext4_ext_is_unwritten(ex),
1708 start_pblk + cur - start);
1712 * Mark the old blocks as free since they aren't used
1713 * anymore. We maintain an array of all the modified
1714 * inodes. In case these blocks are still used at either
1715 * a different logical range in the same inode or in
1716 * some different inode, we will mark them as allocated
1717 * at the end of the FC replay using our array of
1720 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1724 /* Range is mapped and needs a state change */
1725 jbd_debug(1, "Converting from %ld to %d %lld",
1726 map.m_flags & EXT4_MAP_UNWRITTEN,
1727 ext4_ext_is_unwritten(ex), map.m_pblk);
1728 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1729 ext4_ext_is_unwritten(ex), map.m_pblk);
1733 * We may have split the extent tree while toggling the state.
1734 * Try to shrink the extent tree now.
1736 ext4_ext_replay_shrink_inode(inode, start + len);
1739 remaining -= map.m_len;
1741 ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>
1742 sb->s_blocksize_bits);
1748 /* Replay DEL_RANGE tag */
1750 ext4_fc_replay_del_range(struct super_block *sb, struct ext4_fc_tl *tl,
1753 struct inode *inode;
1754 struct ext4_fc_del_range lrange;
1755 struct ext4_map_blocks map;
1756 ext4_lblk_t cur, remaining;
1759 memcpy(&lrange, val, sizeof(lrange));
1760 cur = le32_to_cpu(lrange.fc_lblk);
1761 remaining = le32_to_cpu(lrange.fc_len);
1763 trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,
1764 le32_to_cpu(lrange.fc_ino), cur, remaining);
1766 inode = ext4_iget(sb, le32_to_cpu(lrange.fc_ino), EXT4_IGET_NORMAL);
1767 if (IS_ERR(inode)) {
1768 jbd_debug(1, "Inode %d not found", le32_to_cpu(lrange.fc_ino));
1772 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1776 jbd_debug(1, "DEL_RANGE, inode %ld, lblk %d, len %d\n",
1777 inode->i_ino, le32_to_cpu(lrange.fc_lblk),
1778 le32_to_cpu(lrange.fc_len));
1779 while (remaining > 0) {
1781 map.m_len = remaining;
1783 ret = ext4_map_blocks(NULL, inode, &map, 0);
1789 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1791 remaining -= map.m_len;
1796 down_write(&EXT4_I(inode)->i_data_sem);
1797 ret = ext4_ext_remove_space(inode, le32_to_cpu(lrange.fc_lblk),
1798 le32_to_cpu(lrange.fc_lblk) +
1799 le32_to_cpu(lrange.fc_len) - 1);
1800 up_write(&EXT4_I(inode)->i_data_sem);
1803 ext4_ext_replay_shrink_inode(inode,
1804 i_size_read(inode) >> sb->s_blocksize_bits);
1805 ext4_mark_inode_dirty(NULL, inode);
1811 static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)
1813 struct ext4_fc_replay_state *state;
1814 struct inode *inode;
1815 struct ext4_ext_path *path = NULL;
1816 struct ext4_map_blocks map;
1818 ext4_lblk_t cur, end;
1820 state = &EXT4_SB(sb)->s_fc_replay_state;
1821 for (i = 0; i < state->fc_modified_inodes_used; i++) {
1822 inode = ext4_iget(sb, state->fc_modified_inodes[i],
1824 if (IS_ERR(inode)) {
1825 jbd_debug(1, "Inode %d not found.",
1826 state->fc_modified_inodes[i]);
1830 end = EXT_MAX_BLOCKS;
1831 if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA)) {
1837 map.m_len = end - cur;
1839 ret = ext4_map_blocks(NULL, inode, &map, 0);
1844 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
1845 if (!IS_ERR(path)) {
1846 for (j = 0; j < path->p_depth; j++)
1847 ext4_mb_mark_bb(inode->i_sb,
1848 path[j].p_block, 1, 1);
1849 ext4_ext_drop_refs(path);
1853 ext4_mb_mark_bb(inode->i_sb, map.m_pblk,
1856 cur = cur + (map.m_len ? map.m_len : 1);
1864 * Check if block is in excluded regions for block allocation. The simple
1865 * allocator that runs during replay phase is calls this function to see
1866 * if it is okay to use a block.
1868 bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)
1871 struct ext4_fc_replay_state *state;
1873 state = &EXT4_SB(sb)->s_fc_replay_state;
1874 for (i = 0; i < state->fc_regions_valid; i++) {
1875 if (state->fc_regions[i].ino == 0 ||
1876 state->fc_regions[i].len == 0)
1878 if (blk >= state->fc_regions[i].pblk &&
1879 blk < state->fc_regions[i].pblk + state->fc_regions[i].len)
1885 /* Cleanup function called after replay */
1886 void ext4_fc_replay_cleanup(struct super_block *sb)
1888 struct ext4_sb_info *sbi = EXT4_SB(sb);
1890 sbi->s_mount_state &= ~EXT4_FC_REPLAY;
1891 kfree(sbi->s_fc_replay_state.fc_regions);
1892 kfree(sbi->s_fc_replay_state.fc_modified_inodes);
1896 * Recovery Scan phase handler
1898 * This function is called during the scan phase and is responsible
1899 * for doing following things:
1900 * - Make sure the fast commit area has valid tags for replay
1901 * - Count number of tags that need to be replayed by the replay handler
1903 * - Create a list of excluded blocks for allocation during replay phase
1905 * This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is
1906 * incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP
1907 * to indicate that scan has finished and JBD2 can now start replay phase.
1908 * It returns a negative error to indicate that there was an error. At the end
1909 * of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set
1910 * to indicate the number of tags that need to replayed during the replay phase.
1912 static int ext4_fc_replay_scan(journal_t *journal,
1913 struct buffer_head *bh, int off,
1916 struct super_block *sb = journal->j_private;
1917 struct ext4_sb_info *sbi = EXT4_SB(sb);
1918 struct ext4_fc_replay_state *state;
1919 int ret = JBD2_FC_REPLAY_CONTINUE;
1920 struct ext4_fc_add_range ext;
1921 struct ext4_fc_tl tl;
1922 struct ext4_fc_tail tail;
1923 __u8 *start, *end, *cur, *val;
1924 struct ext4_fc_head head;
1925 struct ext4_extent *ex;
1927 state = &sbi->s_fc_replay_state;
1929 start = (u8 *)bh->b_data;
1930 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
1932 if (state->fc_replay_expected_off == 0) {
1933 state->fc_cur_tag = 0;
1934 state->fc_replay_num_tags = 0;
1936 state->fc_regions = NULL;
1937 state->fc_regions_valid = state->fc_regions_used =
1938 state->fc_regions_size = 0;
1939 /* Check if we can stop early */
1940 if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
1941 != EXT4_FC_TAG_HEAD)
1945 if (off != state->fc_replay_expected_off) {
1946 ret = -EFSCORRUPTED;
1950 state->fc_replay_expected_off++;
1951 for (cur = start; cur < end; cur = cur + sizeof(tl) + le16_to_cpu(tl.fc_len)) {
1952 memcpy(&tl, cur, sizeof(tl));
1953 val = cur + sizeof(tl);
1954 jbd_debug(3, "Scan phase, tag:%s, blk %lld\n",
1955 tag2str(le16_to_cpu(tl.fc_tag)), bh->b_blocknr);
1956 switch (le16_to_cpu(tl.fc_tag)) {
1957 case EXT4_FC_TAG_ADD_RANGE:
1958 memcpy(&ext, val, sizeof(ext));
1959 ex = (struct ext4_extent *)&ext.fc_ex;
1960 ret = ext4_fc_record_regions(sb,
1961 le32_to_cpu(ext.fc_ino),
1962 le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
1963 ext4_ext_get_actual_len(ex), 0);
1966 ret = JBD2_FC_REPLAY_CONTINUE;
1968 case EXT4_FC_TAG_DEL_RANGE:
1969 case EXT4_FC_TAG_LINK:
1970 case EXT4_FC_TAG_UNLINK:
1971 case EXT4_FC_TAG_CREAT:
1972 case EXT4_FC_TAG_INODE:
1973 case EXT4_FC_TAG_PAD:
1974 state->fc_cur_tag++;
1975 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
1976 sizeof(tl) + le16_to_cpu(tl.fc_len));
1978 case EXT4_FC_TAG_TAIL:
1979 state->fc_cur_tag++;
1980 memcpy(&tail, val, sizeof(tail));
1981 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
1983 offsetof(struct ext4_fc_tail,
1985 if (le32_to_cpu(tail.fc_tid) == expected_tid &&
1986 le32_to_cpu(tail.fc_crc) == state->fc_crc) {
1987 state->fc_replay_num_tags = state->fc_cur_tag;
1988 state->fc_regions_valid =
1989 state->fc_regions_used;
1991 ret = state->fc_replay_num_tags ?
1992 JBD2_FC_REPLAY_STOP : -EFSBADCRC;
1996 case EXT4_FC_TAG_HEAD:
1997 memcpy(&head, val, sizeof(head));
1998 if (le32_to_cpu(head.fc_features) &
1999 ~EXT4_FC_SUPPORTED_FEATURES) {
2003 if (le32_to_cpu(head.fc_tid) != expected_tid) {
2004 ret = JBD2_FC_REPLAY_STOP;
2007 state->fc_cur_tag++;
2008 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
2009 sizeof(tl) + le16_to_cpu(tl.fc_len));
2012 ret = state->fc_replay_num_tags ?
2013 JBD2_FC_REPLAY_STOP : -ECANCELED;
2015 if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
2020 trace_ext4_fc_replay_scan(sb, ret, off);
2025 * Main recovery path entry point.
2026 * The meaning of return codes is similar as above.
2028 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
2029 enum passtype pass, int off, tid_t expected_tid)
2031 struct super_block *sb = journal->j_private;
2032 struct ext4_sb_info *sbi = EXT4_SB(sb);
2033 struct ext4_fc_tl tl;
2034 __u8 *start, *end, *cur, *val;
2035 int ret = JBD2_FC_REPLAY_CONTINUE;
2036 struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;
2037 struct ext4_fc_tail tail;
2039 if (pass == PASS_SCAN) {
2040 state->fc_current_pass = PASS_SCAN;
2041 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
2044 if (state->fc_current_pass != pass) {
2045 state->fc_current_pass = pass;
2046 sbi->s_mount_state |= EXT4_FC_REPLAY;
2048 if (!sbi->s_fc_replay_state.fc_replay_num_tags) {
2049 jbd_debug(1, "Replay stops\n");
2050 ext4_fc_set_bitmaps_and_counters(sb);
2054 #ifdef CONFIG_EXT4_DEBUG
2055 if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {
2056 pr_warn("Dropping fc block %d because max_replay set\n", off);
2057 return JBD2_FC_REPLAY_STOP;
2061 start = (u8 *)bh->b_data;
2062 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
2064 for (cur = start; cur < end; cur = cur + sizeof(tl) + le16_to_cpu(tl.fc_len)) {
2065 memcpy(&tl, cur, sizeof(tl));
2066 val = cur + sizeof(tl);
2068 if (state->fc_replay_num_tags == 0) {
2069 ret = JBD2_FC_REPLAY_STOP;
2070 ext4_fc_set_bitmaps_and_counters(sb);
2073 jbd_debug(3, "Replay phase, tag:%s\n",
2074 tag2str(le16_to_cpu(tl.fc_tag)));
2075 state->fc_replay_num_tags--;
2076 switch (le16_to_cpu(tl.fc_tag)) {
2077 case EXT4_FC_TAG_LINK:
2078 ret = ext4_fc_replay_link(sb, &tl, val);
2080 case EXT4_FC_TAG_UNLINK:
2081 ret = ext4_fc_replay_unlink(sb, &tl, val);
2083 case EXT4_FC_TAG_ADD_RANGE:
2084 ret = ext4_fc_replay_add_range(sb, &tl, val);
2086 case EXT4_FC_TAG_CREAT:
2087 ret = ext4_fc_replay_create(sb, &tl, val);
2089 case EXT4_FC_TAG_DEL_RANGE:
2090 ret = ext4_fc_replay_del_range(sb, &tl, val);
2092 case EXT4_FC_TAG_INODE:
2093 ret = ext4_fc_replay_inode(sb, &tl, val);
2095 case EXT4_FC_TAG_PAD:
2096 trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,
2097 le16_to_cpu(tl.fc_len), 0);
2099 case EXT4_FC_TAG_TAIL:
2100 trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL, 0,
2101 le16_to_cpu(tl.fc_len), 0);
2102 memcpy(&tail, val, sizeof(tail));
2103 WARN_ON(le32_to_cpu(tail.fc_tid) != expected_tid);
2105 case EXT4_FC_TAG_HEAD:
2108 trace_ext4_fc_replay(sb, le16_to_cpu(tl.fc_tag), 0,
2109 le16_to_cpu(tl.fc_len), 0);
2115 ret = JBD2_FC_REPLAY_CONTINUE;
2120 void ext4_fc_init(struct super_block *sb, journal_t *journal)
2123 * We set replay callback even if fast commit disabled because we may
2124 * could still have fast commit blocks that need to be replayed even if
2125 * fast commit has now been turned off.
2127 journal->j_fc_replay_callback = ext4_fc_replay;
2128 if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
2130 journal->j_fc_cleanup_callback = ext4_fc_cleanup;
2133 static const char *fc_ineligible_reasons[] = {
2134 "Extended attributes changed",
2136 "Journal flag changed",
2137 "Insufficient memory",
2146 int ext4_fc_info_show(struct seq_file *seq, void *v)
2148 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);
2149 struct ext4_fc_stats *stats = &sbi->s_fc_stats;
2152 if (v != SEQ_START_TOKEN)
2156 "fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",
2157 stats->fc_num_commits, stats->fc_ineligible_commits,
2159 div_u64(stats->s_fc_avg_commit_time, 1000));
2160 seq_puts(seq, "Ineligible reasons:\n");
2161 for (i = 0; i < EXT4_FC_REASON_MAX; i++)
2162 seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],
2163 stats->fc_ineligible_reason_count[i]);
2168 int __init ext4_fc_init_dentry_cache(void)
2170 ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,
2171 SLAB_RECLAIM_ACCOUNT);
2173 if (ext4_fc_dentry_cachep == NULL)
2179 void ext4_fc_destroy_dentry_cache(void)
2181 kmem_cache_destroy(ext4_fc_dentry_cachep);