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 * -------------------------
68 * Not all operations are supported by fast commits today (e.g extended
69 * attributes). Fast commit ineligiblity is marked by calling one of the
70 * two following functions:
72 * - ext4_fc_mark_ineligible(): This makes next fast commit operation to fall
73 * back to full commit. This is useful in case of transient errors.
75 * - ext4_fc_start_ineligible() and ext4_fc_stop_ineligible() - This makes all
76 * the fast commits happening between ext4_fc_start_ineligible() and
77 * ext4_fc_stop_ineligible() and one fast commit after the call to
78 * ext4_fc_stop_ineligible() to fall back to full commits. It is important to
79 * make one more fast commit to fall back to full commit after stop call so
80 * that it guaranteed that the fast commit ineligible operation contained
81 * within ext4_fc_start_ineligible() and ext4_fc_stop_ineligible() is
82 * followed by at least 1 full commit.
84 * Atomicity of commits
85 * --------------------
86 * In order to guarantee atomicity during the commit operation, fast commit
87 * uses "EXT4_FC_TAG_TAIL" tag that marks a fast commit as complete. Tail
88 * tag contains CRC of the contents and TID of the transaction after which
89 * this fast commit should be applied. Recovery code replays fast commit
90 * logs only if there's at least 1 valid tail present. For every fast commit
91 * operation, there is 1 tail. This means, we may end up with multiple tails
92 * in the fast commit space. Here's an example:
94 * - Create a new file A and remove existing file B
96 * - Append contents to file A
100 * The fast commit space at the end of above operations would look like this:
101 * [HEAD] [CREAT A] [UNLINK B] [TAIL] [ADD_RANGE A] [DEL_RANGE A] [TAIL]
102 * |<--- Fast Commit 1 --->|<--- Fast Commit 2 ---->|
104 * Replay code should thus check for all the valid tails in the FC area.
108 * 1) Make fast commit atomic updates more fine grained. Today, a fast commit
109 * eligible update must be protected within ext4_fc_start_update() and
110 * ext4_fc_stop_update(). These routines are called at much higher
111 * routines. This can be made more fine grained by combining with
112 * ext4_journal_start().
114 * 2) Same above for ext4_fc_start_ineligible() and ext4_fc_stop_ineligible()
116 * 3) Handle more ineligible cases.
119 #include <trace/events/ext4.h>
120 static struct kmem_cache *ext4_fc_dentry_cachep;
122 static void ext4_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
124 BUFFER_TRACE(bh, "");
126 ext4_debug("%s: Block %lld up-to-date",
127 __func__, bh->b_blocknr);
128 set_buffer_uptodate(bh);
130 ext4_debug("%s: Block %lld not up-to-date",
131 __func__, bh->b_blocknr);
132 clear_buffer_uptodate(bh);
138 static inline void ext4_fc_reset_inode(struct inode *inode)
140 struct ext4_inode_info *ei = EXT4_I(inode);
142 ei->i_fc_lblk_start = 0;
143 ei->i_fc_lblk_len = 0;
146 void ext4_fc_init_inode(struct inode *inode)
148 struct ext4_inode_info *ei = EXT4_I(inode);
150 ext4_fc_reset_inode(inode);
151 ext4_clear_inode_state(inode, EXT4_STATE_FC_COMMITTING);
152 INIT_LIST_HEAD(&ei->i_fc_list);
153 init_waitqueue_head(&ei->i_fc_wait);
154 atomic_set(&ei->i_fc_updates, 0);
157 /* This function must be called with sbi->s_fc_lock held. */
158 static void ext4_fc_wait_committing_inode(struct inode *inode)
159 __releases(&EXT4_SB(inode->i_sb)->s_fc_lock)
161 wait_queue_head_t *wq;
162 struct ext4_inode_info *ei = EXT4_I(inode);
164 #if (BITS_PER_LONG < 64)
165 DEFINE_WAIT_BIT(wait, &ei->i_state_flags,
166 EXT4_STATE_FC_COMMITTING);
167 wq = bit_waitqueue(&ei->i_state_flags,
168 EXT4_STATE_FC_COMMITTING);
170 DEFINE_WAIT_BIT(wait, &ei->i_flags,
171 EXT4_STATE_FC_COMMITTING);
172 wq = bit_waitqueue(&ei->i_flags,
173 EXT4_STATE_FC_COMMITTING);
175 lockdep_assert_held(&EXT4_SB(inode->i_sb)->s_fc_lock);
176 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
177 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
179 finish_wait(wq, &wait.wq_entry);
183 * Inform Ext4's fast about start of an inode update
185 * This function is called by the high level call VFS callbacks before
186 * performing any inode update. This function blocks if there's an ongoing
187 * fast commit on the inode in question.
189 void ext4_fc_start_update(struct inode *inode)
191 struct ext4_inode_info *ei = EXT4_I(inode);
193 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
194 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
198 spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
199 if (list_empty(&ei->i_fc_list))
202 if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
203 ext4_fc_wait_committing_inode(inode);
207 atomic_inc(&ei->i_fc_updates);
208 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
212 * Stop inode update and wake up waiting fast commits if any.
214 void ext4_fc_stop_update(struct inode *inode)
216 struct ext4_inode_info *ei = EXT4_I(inode);
218 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
219 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
222 if (atomic_dec_and_test(&ei->i_fc_updates))
223 wake_up_all(&ei->i_fc_wait);
227 * Remove inode from fast commit list. If the inode is being committed
228 * we wait until inode commit is done.
230 void ext4_fc_del(struct inode *inode)
232 struct ext4_inode_info *ei = EXT4_I(inode);
234 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
235 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
239 spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
240 if (list_empty(&ei->i_fc_list)) {
241 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
245 if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
246 ext4_fc_wait_committing_inode(inode);
249 list_del_init(&ei->i_fc_list);
250 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
254 * Mark file system as fast commit ineligible. This means that next commit
255 * operation would result in a full jbd2 commit.
257 void ext4_fc_mark_ineligible(struct super_block *sb, int reason)
259 struct ext4_sb_info *sbi = EXT4_SB(sb);
261 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
262 (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
265 ext4_set_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
266 WARN_ON(reason >= EXT4_FC_REASON_MAX);
267 sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;
271 * Start a fast commit ineligible update. Any commits that happen while
272 * such an operation is in progress fall back to full commits.
274 void ext4_fc_start_ineligible(struct super_block *sb, int reason)
276 struct ext4_sb_info *sbi = EXT4_SB(sb);
278 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
279 (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
282 WARN_ON(reason >= EXT4_FC_REASON_MAX);
283 sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;
284 atomic_inc(&sbi->s_fc_ineligible_updates);
288 * Stop a fast commit ineligible update. We set EXT4_MF_FC_INELIGIBLE flag here
289 * to ensure that after stopping the ineligible update, at least one full
290 * commit takes place.
292 void ext4_fc_stop_ineligible(struct super_block *sb)
294 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
295 (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
298 ext4_set_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
299 atomic_dec(&EXT4_SB(sb)->s_fc_ineligible_updates);
302 static inline int ext4_fc_is_ineligible(struct super_block *sb)
304 return (ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE) ||
305 atomic_read(&EXT4_SB(sb)->s_fc_ineligible_updates));
309 * Generic fast commit tracking function. If this is the first time this we are
310 * called after a full commit, we initialize fast commit fields and then call
311 * __fc_track_fn() with update = 0. If we have already been called after a full
312 * commit, we pass update = 1. Based on that, the track function can determine
313 * if it needs to track a field for the first time or if it needs to just
314 * update the previously tracked value.
316 * If enqueue is set, this function enqueues the inode in fast commit list.
318 static int ext4_fc_track_template(
319 handle_t *handle, struct inode *inode,
320 int (*__fc_track_fn)(struct inode *, void *, bool),
321 void *args, int enqueue)
324 struct ext4_inode_info *ei = EXT4_I(inode);
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
329 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
330 (sbi->s_mount_state & EXT4_FC_REPLAY))
333 if (ext4_fc_is_ineligible(inode->i_sb))
336 tid = handle->h_transaction->t_tid;
337 mutex_lock(&ei->i_fc_lock);
338 if (tid == ei->i_sync_tid) {
341 ext4_fc_reset_inode(inode);
342 ei->i_sync_tid = tid;
344 ret = __fc_track_fn(inode, args, update);
345 mutex_unlock(&ei->i_fc_lock);
350 spin_lock(&sbi->s_fc_lock);
351 if (list_empty(&EXT4_I(inode)->i_fc_list))
352 list_add_tail(&EXT4_I(inode)->i_fc_list,
353 (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_COMMITTING)) ?
354 &sbi->s_fc_q[FC_Q_STAGING] :
355 &sbi->s_fc_q[FC_Q_MAIN]);
356 spin_unlock(&sbi->s_fc_lock);
361 struct __track_dentry_update_args {
362 struct dentry *dentry;
366 /* __track_fn for directory entry updates. Called with ei->i_fc_lock. */
367 static int __track_dentry_update(struct inode *inode, void *arg, bool update)
369 struct ext4_fc_dentry_update *node;
370 struct ext4_inode_info *ei = EXT4_I(inode);
371 struct __track_dentry_update_args *dentry_update =
372 (struct __track_dentry_update_args *)arg;
373 struct dentry *dentry = dentry_update->dentry;
374 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
376 mutex_unlock(&ei->i_fc_lock);
377 node = kmem_cache_alloc(ext4_fc_dentry_cachep, GFP_NOFS);
379 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM);
380 mutex_lock(&ei->i_fc_lock);
384 node->fcd_op = dentry_update->op;
385 node->fcd_parent = dentry->d_parent->d_inode->i_ino;
386 node->fcd_ino = inode->i_ino;
387 if (dentry->d_name.len > DNAME_INLINE_LEN) {
388 node->fcd_name.name = kmalloc(dentry->d_name.len, GFP_NOFS);
389 if (!node->fcd_name.name) {
390 kmem_cache_free(ext4_fc_dentry_cachep, node);
391 ext4_fc_mark_ineligible(inode->i_sb,
392 EXT4_FC_REASON_NOMEM);
393 mutex_lock(&ei->i_fc_lock);
396 memcpy((u8 *)node->fcd_name.name, dentry->d_name.name,
399 memcpy(node->fcd_iname, dentry->d_name.name,
401 node->fcd_name.name = node->fcd_iname;
403 node->fcd_name.len = dentry->d_name.len;
405 spin_lock(&sbi->s_fc_lock);
406 if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_COMMITTING))
407 list_add_tail(&node->fcd_list,
408 &sbi->s_fc_dentry_q[FC_Q_STAGING]);
410 list_add_tail(&node->fcd_list, &sbi->s_fc_dentry_q[FC_Q_MAIN]);
411 spin_unlock(&sbi->s_fc_lock);
412 mutex_lock(&ei->i_fc_lock);
417 void __ext4_fc_track_unlink(handle_t *handle,
418 struct inode *inode, struct dentry *dentry)
420 struct __track_dentry_update_args args;
423 args.dentry = dentry;
424 args.op = EXT4_FC_TAG_UNLINK;
426 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
428 trace_ext4_fc_track_unlink(inode, dentry, ret);
431 void ext4_fc_track_unlink(handle_t *handle, struct dentry *dentry)
433 __ext4_fc_track_unlink(handle, d_inode(dentry), dentry);
436 void __ext4_fc_track_link(handle_t *handle,
437 struct inode *inode, struct dentry *dentry)
439 struct __track_dentry_update_args args;
442 args.dentry = dentry;
443 args.op = EXT4_FC_TAG_LINK;
445 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
447 trace_ext4_fc_track_link(inode, dentry, ret);
450 void ext4_fc_track_link(handle_t *handle, struct dentry *dentry)
452 __ext4_fc_track_link(handle, d_inode(dentry), dentry);
455 void __ext4_fc_track_create(handle_t *handle, struct inode *inode,
456 struct dentry *dentry)
458 struct __track_dentry_update_args args;
461 args.dentry = dentry;
462 args.op = EXT4_FC_TAG_CREAT;
464 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
466 trace_ext4_fc_track_create(inode, dentry, ret);
469 void ext4_fc_track_create(handle_t *handle, struct dentry *dentry)
471 __ext4_fc_track_create(handle, d_inode(dentry), dentry);
474 /* __track_fn for inode tracking */
475 static int __track_inode(struct inode *inode, void *arg, bool update)
480 EXT4_I(inode)->i_fc_lblk_len = 0;
485 void ext4_fc_track_inode(handle_t *handle, struct inode *inode)
489 if (S_ISDIR(inode->i_mode))
492 if (ext4_should_journal_data(inode)) {
493 ext4_fc_mark_ineligible(inode->i_sb,
494 EXT4_FC_REASON_INODE_JOURNAL_DATA);
498 ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);
499 trace_ext4_fc_track_inode(inode, ret);
502 struct __track_range_args {
503 ext4_lblk_t start, end;
506 /* __track_fn for tracking data updates */
507 static int __track_range(struct inode *inode, void *arg, bool update)
509 struct ext4_inode_info *ei = EXT4_I(inode);
510 ext4_lblk_t oldstart;
511 struct __track_range_args *__arg =
512 (struct __track_range_args *)arg;
514 if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {
515 ext4_debug("Special inode %ld being modified\n", inode->i_ino);
519 oldstart = ei->i_fc_lblk_start;
521 if (update && ei->i_fc_lblk_len > 0) {
522 ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);
524 max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -
525 ei->i_fc_lblk_start + 1;
527 ei->i_fc_lblk_start = __arg->start;
528 ei->i_fc_lblk_len = __arg->end - __arg->start + 1;
534 void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,
537 struct __track_range_args args;
540 if (S_ISDIR(inode->i_mode))
546 ret = ext4_fc_track_template(handle, inode, __track_range, &args, 1);
548 trace_ext4_fc_track_range(inode, start, end, ret);
551 static void ext4_fc_submit_bh(struct super_block *sb)
553 int write_flags = REQ_SYNC;
554 struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
556 /* TODO: REQ_FUA | REQ_PREFLUSH is unnecessarily expensive. */
557 if (test_opt(sb, BARRIER))
558 write_flags |= REQ_FUA | REQ_PREFLUSH;
560 set_buffer_dirty(bh);
561 set_buffer_uptodate(bh);
562 bh->b_end_io = ext4_end_buffer_io_sync;
563 submit_bh(REQ_OP_WRITE, write_flags, bh);
564 EXT4_SB(sb)->s_fc_bh = NULL;
567 /* Ext4 commit path routines */
569 /* memzero and update CRC */
570 static void *ext4_fc_memzero(struct super_block *sb, void *dst, int len,
575 ret = memset(dst, 0, len);
577 *crc = ext4_chksum(EXT4_SB(sb), *crc, dst, len);
582 * Allocate len bytes on a fast commit buffer.
584 * During the commit time this function is used to manage fast commit
585 * block space. We don't split a fast commit log onto different
586 * blocks. So this function makes sure that if there's not enough space
587 * on the current block, the remaining space in the current block is
588 * marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,
589 * new block is from jbd2 and CRC is updated to reflect the padding
592 static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
594 struct ext4_fc_tl *tl;
595 struct ext4_sb_info *sbi = EXT4_SB(sb);
596 struct buffer_head *bh;
597 int bsize = sbi->s_journal->j_blocksize;
598 int ret, off = sbi->s_fc_bytes % bsize;
602 * After allocating len, we should have space at least for a 0 byte
605 if (len + sizeof(struct ext4_fc_tl) > bsize)
608 if (bsize - off - 1 > len + sizeof(struct ext4_fc_tl)) {
610 * Only allocate from current buffer if we have enough space for
611 * this request AND we have space to add a zero byte padding.
614 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
619 sbi->s_fc_bytes += len;
620 return sbi->s_fc_bh->b_data + off;
622 /* Need to add PAD tag */
623 tl = (struct ext4_fc_tl *)(sbi->s_fc_bh->b_data + off);
624 tl->fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);
625 pad_len = bsize - off - 1 - sizeof(struct ext4_fc_tl);
626 tl->fc_len = cpu_to_le16(pad_len);
628 *crc = ext4_chksum(sbi, *crc, tl, sizeof(*tl));
630 ext4_fc_memzero(sb, tl + 1, pad_len, crc);
631 ext4_fc_submit_bh(sb);
633 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
637 sbi->s_fc_bytes = (sbi->s_fc_bytes / bsize + 1) * bsize + len;
638 return sbi->s_fc_bh->b_data;
641 /* memcpy to fc reserved space and update CRC */
642 static void *ext4_fc_memcpy(struct super_block *sb, void *dst, const void *src,
646 *crc = ext4_chksum(EXT4_SB(sb), *crc, src, len);
647 return memcpy(dst, src, len);
651 * Complete a fast commit by writing tail tag.
653 * Writing tail tag marks the end of a fast commit. In order to guarantee
654 * atomicity, after writing tail tag, even if there's space remaining
655 * in the block, next commit shouldn't use it. That's why tail tag
656 * has the length as that of the remaining space on the block.
658 static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
660 struct ext4_sb_info *sbi = EXT4_SB(sb);
661 struct ext4_fc_tl tl;
662 struct ext4_fc_tail tail;
663 int off, bsize = sbi->s_journal->j_blocksize;
667 * ext4_fc_reserve_space takes care of allocating an extra block if
668 * there's no enough space on this block for accommodating this tail.
670 dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(tail), &crc);
674 off = sbi->s_fc_bytes % bsize;
676 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);
677 tl.fc_len = cpu_to_le16(bsize - off - 1 + sizeof(struct ext4_fc_tail));
678 sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);
680 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), &crc);
682 tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);
683 ext4_fc_memcpy(sb, dst, &tail.fc_tid, sizeof(tail.fc_tid), &crc);
684 dst += sizeof(tail.fc_tid);
685 tail.fc_crc = cpu_to_le32(crc);
686 ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
688 ext4_fc_submit_bh(sb);
694 * Adds tag, length, value and updates CRC. Returns true if tlv was added.
695 * Returns false if there's not enough space.
697 static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,
700 struct ext4_fc_tl tl;
703 dst = ext4_fc_reserve_space(sb, sizeof(tl) + len, crc);
707 tl.fc_tag = cpu_to_le16(tag);
708 tl.fc_len = cpu_to_le16(len);
710 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
711 ext4_fc_memcpy(sb, dst + sizeof(tl), val, len, crc);
716 /* Same as above, but adds dentry tlv. */
717 static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u16 tag,
718 int parent_ino, int ino, int dlen,
719 const unsigned char *dname,
722 struct ext4_fc_dentry_info fcd;
723 struct ext4_fc_tl tl;
724 u8 *dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(fcd) + dlen,
730 fcd.fc_parent_ino = cpu_to_le32(parent_ino);
731 fcd.fc_ino = cpu_to_le32(ino);
732 tl.fc_tag = cpu_to_le16(tag);
733 tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);
734 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
736 ext4_fc_memcpy(sb, dst, &fcd, sizeof(fcd), crc);
738 ext4_fc_memcpy(sb, dst, dname, dlen, crc);
745 * Writes inode in the fast commit space under TLV with tag @tag.
746 * Returns 0 on success, error on failure.
748 static int ext4_fc_write_inode(struct inode *inode, u32 *crc)
750 struct ext4_inode_info *ei = EXT4_I(inode);
751 int inode_len = EXT4_GOOD_OLD_INODE_SIZE;
753 struct ext4_iloc iloc;
754 struct ext4_fc_inode fc_inode;
755 struct ext4_fc_tl tl;
758 ret = ext4_get_inode_loc(inode, &iloc);
762 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)
763 inode_len += ei->i_extra_isize;
765 fc_inode.fc_ino = cpu_to_le32(inode->i_ino);
766 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);
767 tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));
770 dst = ext4_fc_reserve_space(inode->i_sb,
771 sizeof(tl) + inode_len + sizeof(fc_inode.fc_ino), crc);
775 if (!ext4_fc_memcpy(inode->i_sb, dst, &tl, sizeof(tl), crc))
778 if (!ext4_fc_memcpy(inode->i_sb, dst, &fc_inode, sizeof(fc_inode), crc))
780 dst += sizeof(fc_inode);
781 if (!ext4_fc_memcpy(inode->i_sb, dst, (u8 *)ext4_raw_inode(&iloc),
791 * Writes updated data ranges for the inode in question. Updates CRC.
792 * Returns 0 on success, error otherwise.
794 static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)
796 ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;
797 struct ext4_inode_info *ei = EXT4_I(inode);
798 struct ext4_map_blocks map;
799 struct ext4_fc_add_range fc_ext;
800 struct ext4_fc_del_range lrange;
801 struct ext4_extent *ex;
804 mutex_lock(&ei->i_fc_lock);
805 if (ei->i_fc_lblk_len == 0) {
806 mutex_unlock(&ei->i_fc_lock);
809 old_blk_size = ei->i_fc_lblk_start;
810 new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;
811 ei->i_fc_lblk_len = 0;
812 mutex_unlock(&ei->i_fc_lock);
814 cur_lblk_off = old_blk_size;
815 jbd_debug(1, "%s: will try writing %d to %d for inode %ld\n",
816 __func__, cur_lblk_off, new_blk_size, inode->i_ino);
818 while (cur_lblk_off <= new_blk_size) {
819 map.m_lblk = cur_lblk_off;
820 map.m_len = new_blk_size - cur_lblk_off + 1;
821 ret = ext4_map_blocks(NULL, inode, &map, 0);
825 if (map.m_len == 0) {
831 lrange.fc_ino = cpu_to_le32(inode->i_ino);
832 lrange.fc_lblk = cpu_to_le32(map.m_lblk);
833 lrange.fc_len = cpu_to_le32(map.m_len);
834 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,
835 sizeof(lrange), (u8 *)&lrange, crc))
838 unsigned int max = (map.m_flags & EXT4_MAP_UNWRITTEN) ?
839 EXT_UNWRITTEN_MAX_LEN : EXT_INIT_MAX_LEN;
841 /* Limit the number of blocks in one extent */
842 map.m_len = min(max, map.m_len);
844 fc_ext.fc_ino = cpu_to_le32(inode->i_ino);
845 ex = (struct ext4_extent *)&fc_ext.fc_ex;
846 ex->ee_block = cpu_to_le32(map.m_lblk);
847 ex->ee_len = cpu_to_le16(map.m_len);
848 ext4_ext_store_pblock(ex, map.m_pblk);
849 if (map.m_flags & EXT4_MAP_UNWRITTEN)
850 ext4_ext_mark_unwritten(ex);
852 ext4_ext_mark_initialized(ex);
853 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,
854 sizeof(fc_ext), (u8 *)&fc_ext, crc))
858 cur_lblk_off += map.m_len;
865 /* Submit data for all the fast commit inodes */
866 static int ext4_fc_submit_inode_data_all(journal_t *journal)
868 struct super_block *sb = (struct super_block *)(journal->j_private);
869 struct ext4_sb_info *sbi = EXT4_SB(sb);
870 struct ext4_inode_info *ei;
871 struct list_head *pos;
874 spin_lock(&sbi->s_fc_lock);
875 ext4_set_mount_flag(sb, EXT4_MF_FC_COMMITTING);
876 list_for_each(pos, &sbi->s_fc_q[FC_Q_MAIN]) {
877 ei = list_entry(pos, struct ext4_inode_info, i_fc_list);
878 ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
879 while (atomic_read(&ei->i_fc_updates)) {
882 prepare_to_wait(&ei->i_fc_wait, &wait,
883 TASK_UNINTERRUPTIBLE);
884 if (atomic_read(&ei->i_fc_updates)) {
885 spin_unlock(&sbi->s_fc_lock);
887 spin_lock(&sbi->s_fc_lock);
889 finish_wait(&ei->i_fc_wait, &wait);
891 spin_unlock(&sbi->s_fc_lock);
892 ret = jbd2_submit_inode_data(ei->jinode);
895 spin_lock(&sbi->s_fc_lock);
897 spin_unlock(&sbi->s_fc_lock);
902 /* Wait for completion of data for all the fast commit inodes */
903 static int ext4_fc_wait_inode_data_all(journal_t *journal)
905 struct super_block *sb = (struct super_block *)(journal->j_private);
906 struct ext4_sb_info *sbi = EXT4_SB(sb);
907 struct ext4_inode_info *pos, *n;
910 spin_lock(&sbi->s_fc_lock);
911 list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
912 if (!ext4_test_inode_state(&pos->vfs_inode,
913 EXT4_STATE_FC_COMMITTING))
915 spin_unlock(&sbi->s_fc_lock);
917 ret = jbd2_wait_inode_data(journal, pos->jinode);
920 spin_lock(&sbi->s_fc_lock);
922 spin_unlock(&sbi->s_fc_lock);
927 /* Commit all the directory entry updates */
928 static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)
929 __acquires(&sbi->s_fc_lock)
930 __releases(&sbi->s_fc_lock)
932 struct super_block *sb = (struct super_block *)(journal->j_private);
933 struct ext4_sb_info *sbi = EXT4_SB(sb);
934 struct ext4_fc_dentry_update *fc_dentry;
936 struct list_head *pos, *n, *fcd_pos, *fcd_n;
937 struct ext4_inode_info *ei;
940 if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))
942 list_for_each_safe(fcd_pos, fcd_n, &sbi->s_fc_dentry_q[FC_Q_MAIN]) {
943 fc_dentry = list_entry(fcd_pos, struct ext4_fc_dentry_update,
945 if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {
946 spin_unlock(&sbi->s_fc_lock);
947 if (!ext4_fc_add_dentry_tlv(
948 sb, fc_dentry->fcd_op,
949 fc_dentry->fcd_parent, fc_dentry->fcd_ino,
950 fc_dentry->fcd_name.len,
951 fc_dentry->fcd_name.name, crc)) {
955 spin_lock(&sbi->s_fc_lock);
960 list_for_each_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN]) {
961 ei = list_entry(pos, struct ext4_inode_info, i_fc_list);
962 if (ei->vfs_inode.i_ino == fc_dentry->fcd_ino) {
963 inode = &ei->vfs_inode;
968 * If we don't find inode in our list, then it was deleted,
969 * in which case, we don't need to record it's create tag.
973 spin_unlock(&sbi->s_fc_lock);
976 * We first write the inode and then the create dirent. This
977 * allows the recovery code to create an unnamed inode first
978 * and then link it to a directory entry. This allows us
979 * to use namei.c routines almost as is and simplifies
982 ret = ext4_fc_write_inode(inode, crc);
986 ret = ext4_fc_write_inode_data(inode, crc);
990 if (!ext4_fc_add_dentry_tlv(
991 sb, fc_dentry->fcd_op,
992 fc_dentry->fcd_parent, fc_dentry->fcd_ino,
993 fc_dentry->fcd_name.len,
994 fc_dentry->fcd_name.name, crc)) {
999 spin_lock(&sbi->s_fc_lock);
1003 spin_lock(&sbi->s_fc_lock);
1007 static int ext4_fc_perform_commit(journal_t *journal)
1009 struct super_block *sb = (struct super_block *)(journal->j_private);
1010 struct ext4_sb_info *sbi = EXT4_SB(sb);
1011 struct ext4_inode_info *iter;
1012 struct ext4_fc_head head;
1013 struct list_head *pos;
1014 struct inode *inode;
1015 struct blk_plug plug;
1019 ret = ext4_fc_submit_inode_data_all(journal);
1023 ret = ext4_fc_wait_inode_data_all(journal);
1028 * If file system device is different from journal device, issue a cache
1029 * flush before we start writing fast commit blocks.
1031 if (journal->j_fs_dev != journal->j_dev)
1032 blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS);
1034 blk_start_plug(&plug);
1035 if (sbi->s_fc_bytes == 0) {
1037 * Add a head tag only if this is the first fast commit
1040 head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);
1041 head.fc_tid = cpu_to_le32(
1042 sbi->s_journal->j_running_transaction->t_tid);
1043 if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),
1044 (u8 *)&head, &crc)) {
1050 spin_lock(&sbi->s_fc_lock);
1051 ret = ext4_fc_commit_dentry_updates(journal, &crc);
1053 spin_unlock(&sbi->s_fc_lock);
1057 list_for_each(pos, &sbi->s_fc_q[FC_Q_MAIN]) {
1058 iter = list_entry(pos, struct ext4_inode_info, i_fc_list);
1059 inode = &iter->vfs_inode;
1060 if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))
1063 spin_unlock(&sbi->s_fc_lock);
1064 ret = ext4_fc_write_inode_data(inode, &crc);
1067 ret = ext4_fc_write_inode(inode, &crc);
1070 spin_lock(&sbi->s_fc_lock);
1072 spin_unlock(&sbi->s_fc_lock);
1074 ret = ext4_fc_write_tail(sb, crc);
1077 blk_finish_plug(&plug);
1082 * The main commit entry point. Performs a fast commit for transaction
1083 * commit_tid if needed. If it's not possible to perform a fast commit
1084 * due to various reasons, we fall back to full commit. Returns 0
1085 * on success, error otherwise.
1087 int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
1089 struct super_block *sb = (struct super_block *)(journal->j_private);
1090 struct ext4_sb_info *sbi = EXT4_SB(sb);
1091 int nblks = 0, ret, bsize = journal->j_blocksize;
1092 int subtid = atomic_read(&sbi->s_fc_subtid);
1093 int reason = EXT4_FC_REASON_OK, fc_bufs_before = 0;
1094 ktime_t start_time, commit_time;
1096 trace_ext4_fc_commit_start(sb);
1098 start_time = ktime_get();
1100 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
1101 (ext4_fc_is_ineligible(sb))) {
1102 reason = EXT4_FC_REASON_INELIGIBLE;
1107 ret = jbd2_fc_begin_commit(journal, commit_tid);
1108 if (ret == -EALREADY) {
1109 /* There was an ongoing commit, check if we need to restart */
1110 if (atomic_read(&sbi->s_fc_subtid) <= subtid &&
1111 commit_tid > journal->j_commit_sequence)
1113 reason = EXT4_FC_REASON_ALREADY_COMMITTED;
1116 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1117 reason = EXT4_FC_REASON_FC_START_FAILED;
1121 fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
1122 ret = ext4_fc_perform_commit(journal);
1124 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1125 reason = EXT4_FC_REASON_FC_FAILED;
1128 nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;
1129 ret = jbd2_fc_wait_bufs(journal, nblks);
1131 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1132 reason = EXT4_FC_REASON_FC_FAILED;
1135 atomic_inc(&sbi->s_fc_subtid);
1136 jbd2_fc_end_commit(journal);
1138 /* Has any ineligible update happened since we started? */
1139 if (reason == EXT4_FC_REASON_OK && ext4_fc_is_ineligible(sb)) {
1140 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1141 reason = EXT4_FC_REASON_INELIGIBLE;
1144 spin_lock(&sbi->s_fc_lock);
1145 if (reason != EXT4_FC_REASON_OK &&
1146 reason != EXT4_FC_REASON_ALREADY_COMMITTED) {
1147 sbi->s_fc_stats.fc_ineligible_commits++;
1149 sbi->s_fc_stats.fc_num_commits++;
1150 sbi->s_fc_stats.fc_numblks += nblks;
1152 spin_unlock(&sbi->s_fc_lock);
1153 nblks = (reason == EXT4_FC_REASON_OK) ? nblks : 0;
1154 trace_ext4_fc_commit_stop(sb, nblks, reason);
1155 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1157 * weight the commit time higher than the average time so we don't
1158 * react too strongly to vast changes in the commit time
1160 if (likely(sbi->s_fc_avg_commit_time))
1161 sbi->s_fc_avg_commit_time = (commit_time +
1162 sbi->s_fc_avg_commit_time * 3) / 4;
1164 sbi->s_fc_avg_commit_time = commit_time;
1166 "Fast commit ended with blks = %d, reason = %d, subtid - %d",
1167 nblks, reason, subtid);
1168 if (reason == EXT4_FC_REASON_FC_FAILED)
1169 return jbd2_fc_end_commit_fallback(journal);
1170 if (reason == EXT4_FC_REASON_FC_START_FAILED ||
1171 reason == EXT4_FC_REASON_INELIGIBLE)
1172 return jbd2_complete_transaction(journal, commit_tid);
1177 * Fast commit cleanup routine. This is called after every fast commit and
1178 * full commit. full is true if we are called after a full commit.
1180 static void ext4_fc_cleanup(journal_t *journal, int full)
1182 struct super_block *sb = journal->j_private;
1183 struct ext4_sb_info *sbi = EXT4_SB(sb);
1184 struct ext4_inode_info *iter;
1185 struct ext4_fc_dentry_update *fc_dentry;
1186 struct list_head *pos, *n;
1188 if (full && sbi->s_fc_bh)
1189 sbi->s_fc_bh = NULL;
1191 jbd2_fc_release_bufs(journal);
1193 spin_lock(&sbi->s_fc_lock);
1194 list_for_each_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN]) {
1195 iter = list_entry(pos, struct ext4_inode_info, i_fc_list);
1196 list_del_init(&iter->i_fc_list);
1197 ext4_clear_inode_state(&iter->vfs_inode,
1198 EXT4_STATE_FC_COMMITTING);
1199 ext4_fc_reset_inode(&iter->vfs_inode);
1200 /* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
1202 #if (BITS_PER_LONG < 64)
1203 wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);
1205 wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);
1209 while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {
1210 fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],
1211 struct ext4_fc_dentry_update,
1213 list_del_init(&fc_dentry->fcd_list);
1214 spin_unlock(&sbi->s_fc_lock);
1216 if (fc_dentry->fcd_name.name &&
1217 fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
1218 kfree(fc_dentry->fcd_name.name);
1219 kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
1220 spin_lock(&sbi->s_fc_lock);
1223 list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],
1224 &sbi->s_fc_dentry_q[FC_Q_MAIN]);
1225 list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
1226 &sbi->s_fc_q[FC_Q_MAIN]);
1228 ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING);
1229 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
1232 sbi->s_fc_bytes = 0;
1233 spin_unlock(&sbi->s_fc_lock);
1234 trace_ext4_fc_stats(sb);
1237 /* Ext4 Replay Path Routines */
1239 /* Helper struct for dentry replay routines */
1240 struct dentry_info_args {
1241 int parent_ino, dname_len, ino, inode_len;
1245 static inline void tl_to_darg(struct dentry_info_args *darg,
1246 struct ext4_fc_tl *tl, u8 *val)
1248 struct ext4_fc_dentry_info fcd;
1250 memcpy(&fcd, val, sizeof(fcd));
1252 darg->parent_ino = le32_to_cpu(fcd.fc_parent_ino);
1253 darg->ino = le32_to_cpu(fcd.fc_ino);
1254 darg->dname = val + offsetof(struct ext4_fc_dentry_info, fc_dname);
1255 darg->dname_len = le16_to_cpu(tl->fc_len) -
1256 sizeof(struct ext4_fc_dentry_info);
1259 /* Unlink replay function */
1260 static int ext4_fc_replay_unlink(struct super_block *sb, struct ext4_fc_tl *tl,
1263 struct inode *inode, *old_parent;
1265 struct dentry_info_args darg;
1268 tl_to_darg(&darg, tl, val);
1270 trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,
1271 darg.parent_ino, darg.dname_len);
1273 entry.name = darg.dname;
1274 entry.len = darg.dname_len;
1275 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1277 if (IS_ERR(inode)) {
1278 jbd_debug(1, "Inode %d not found", darg.ino);
1282 old_parent = ext4_iget(sb, darg.parent_ino,
1284 if (IS_ERR(old_parent)) {
1285 jbd_debug(1, "Dir with inode %d not found", darg.parent_ino);
1290 ret = __ext4_unlink(NULL, old_parent, &entry, inode);
1291 /* -ENOENT ok coz it might not exist anymore. */
1299 static int ext4_fc_replay_link_internal(struct super_block *sb,
1300 struct dentry_info_args *darg,
1301 struct inode *inode)
1303 struct inode *dir = NULL;
1304 struct dentry *dentry_dir = NULL, *dentry_inode = NULL;
1305 struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);
1308 dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);
1310 jbd_debug(1, "Dir with inode %d not found.", darg->parent_ino);
1315 dentry_dir = d_obtain_alias(dir);
1316 if (IS_ERR(dentry_dir)) {
1317 jbd_debug(1, "Failed to obtain dentry");
1322 dentry_inode = d_alloc(dentry_dir, &qstr_dname);
1323 if (!dentry_inode) {
1324 jbd_debug(1, "Inode dentry not created.");
1329 ret = __ext4_link(dir, inode, dentry_inode);
1331 * It's possible that link already existed since data blocks
1332 * for the dir in question got persisted before we crashed OR
1333 * we replayed this tag and crashed before the entire replay
1336 if (ret && ret != -EEXIST) {
1337 jbd_debug(1, "Failed to link\n");
1350 d_drop(dentry_inode);
1357 /* Link replay function */
1358 static int ext4_fc_replay_link(struct super_block *sb, struct ext4_fc_tl *tl,
1361 struct inode *inode;
1362 struct dentry_info_args darg;
1365 tl_to_darg(&darg, tl, val);
1366 trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,
1367 darg.parent_ino, darg.dname_len);
1369 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1370 if (IS_ERR(inode)) {
1371 jbd_debug(1, "Inode not found.");
1375 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1381 * Record all the modified inodes during replay. We use this later to setup
1382 * block bitmaps correctly.
1384 static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)
1386 struct ext4_fc_replay_state *state;
1389 state = &EXT4_SB(sb)->s_fc_replay_state;
1390 for (i = 0; i < state->fc_modified_inodes_used; i++)
1391 if (state->fc_modified_inodes[i] == ino)
1393 if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {
1394 int *fc_modified_inodes;
1396 fc_modified_inodes = krealloc(state->fc_modified_inodes,
1397 sizeof(int) * (state->fc_modified_inodes_size +
1398 EXT4_FC_REPLAY_REALLOC_INCREMENT),
1400 if (!fc_modified_inodes)
1402 state->fc_modified_inodes = fc_modified_inodes;
1403 state->fc_modified_inodes_size +=
1404 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1406 state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;
1411 * Inode replay function
1413 static int ext4_fc_replay_inode(struct super_block *sb, struct ext4_fc_tl *tl,
1416 struct ext4_fc_inode fc_inode;
1417 struct ext4_inode *raw_inode;
1418 struct ext4_inode *raw_fc_inode;
1419 struct inode *inode = NULL;
1420 struct ext4_iloc iloc;
1421 int inode_len, ino, ret, tag = le16_to_cpu(tl->fc_tag);
1422 struct ext4_extent_header *eh;
1424 memcpy(&fc_inode, val, sizeof(fc_inode));
1426 ino = le32_to_cpu(fc_inode.fc_ino);
1427 trace_ext4_fc_replay(sb, tag, ino, 0, 0);
1429 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1430 if (!IS_ERR(inode)) {
1431 ext4_ext_clear_bb(inode);
1436 ret = ext4_fc_record_modified_inode(sb, ino);
1440 raw_fc_inode = (struct ext4_inode *)
1441 (val + offsetof(struct ext4_fc_inode, fc_raw_inode));
1442 ret = ext4_get_fc_inode_loc(sb, ino, &iloc);
1446 inode_len = le16_to_cpu(tl->fc_len) - sizeof(struct ext4_fc_inode);
1447 raw_inode = ext4_raw_inode(&iloc);
1449 memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));
1450 memcpy(&raw_inode->i_generation, &raw_fc_inode->i_generation,
1451 inode_len - offsetof(struct ext4_inode, i_generation));
1452 if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {
1453 eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);
1454 if (eh->eh_magic != EXT4_EXT_MAGIC) {
1455 memset(eh, 0, sizeof(*eh));
1456 eh->eh_magic = EXT4_EXT_MAGIC;
1457 eh->eh_max = cpu_to_le16(
1458 (sizeof(raw_inode->i_block) -
1459 sizeof(struct ext4_extent_header))
1460 / sizeof(struct ext4_extent));
1462 } else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {
1463 memcpy(raw_inode->i_block, raw_fc_inode->i_block,
1464 sizeof(raw_inode->i_block));
1467 /* Immediately update the inode on disk. */
1468 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1471 ret = sync_dirty_buffer(iloc.bh);
1474 ret = ext4_mark_inode_used(sb, ino);
1478 /* Given that we just wrote the inode on disk, this SHOULD succeed. */
1479 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1480 if (IS_ERR(inode)) {
1481 jbd_debug(1, "Inode not found.");
1482 return -EFSCORRUPTED;
1486 * Our allocator could have made different decisions than before
1487 * crashing. This should be fixed but until then, we calculate
1488 * the number of blocks the inode.
1490 ext4_ext_replay_set_iblocks(inode);
1492 inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);
1493 ext4_reset_inode_seed(inode);
1495 ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));
1496 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1497 sync_dirty_buffer(iloc.bh);
1502 blkdev_issue_flush(sb->s_bdev, GFP_KERNEL);
1508 * Dentry create replay function.
1510 * EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the
1511 * inode for which we are trying to create a dentry here, should already have
1512 * been replayed before we start here.
1514 static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
1518 struct inode *inode = NULL;
1519 struct inode *dir = NULL;
1520 struct dentry_info_args darg;
1522 tl_to_darg(&darg, tl, val);
1524 trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,
1525 darg.parent_ino, darg.dname_len);
1527 /* This takes care of update group descriptor and other metadata */
1528 ret = ext4_mark_inode_used(sb, darg.ino);
1532 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1533 if (IS_ERR(inode)) {
1534 jbd_debug(1, "inode %d not found.", darg.ino);
1540 if (S_ISDIR(inode->i_mode)) {
1542 * If we are creating a directory, we need to make sure that the
1543 * dot and dot dot dirents are setup properly.
1545 dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);
1547 jbd_debug(1, "Dir %d not found.", darg.ino);
1550 ret = ext4_init_new_dir(NULL, dir, inode);
1557 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1560 set_nlink(inode, 1);
1561 ext4_mark_inode_dirty(NULL, inode);
1569 * Record physical disk regions which are in use as per fast commit area,
1570 * and used by inodes during replay phase. Our simple replay phase
1571 * allocator excludes these regions from allocation.
1573 int ext4_fc_record_regions(struct super_block *sb, int ino,
1574 ext4_lblk_t lblk, ext4_fsblk_t pblk, int len, int replay)
1576 struct ext4_fc_replay_state *state;
1577 struct ext4_fc_alloc_region *region;
1579 state = &EXT4_SB(sb)->s_fc_replay_state;
1581 * during replay phase, the fc_regions_valid may not same as
1582 * fc_regions_used, update it when do new additions.
1584 if (replay && state->fc_regions_used != state->fc_regions_valid)
1585 state->fc_regions_used = state->fc_regions_valid;
1586 if (state->fc_regions_used == state->fc_regions_size) {
1587 struct ext4_fc_alloc_region *fc_regions;
1589 fc_regions = krealloc(state->fc_regions,
1590 sizeof(struct ext4_fc_alloc_region) *
1591 (state->fc_regions_size +
1592 EXT4_FC_REPLAY_REALLOC_INCREMENT),
1596 state->fc_regions_size +=
1597 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1598 state->fc_regions = fc_regions;
1600 region = &state->fc_regions[state->fc_regions_used++];
1602 region->lblk = lblk;
1603 region->pblk = pblk;
1607 state->fc_regions_valid++;
1612 /* Replay add range tag */
1613 static int ext4_fc_replay_add_range(struct super_block *sb,
1614 struct ext4_fc_tl *tl, u8 *val)
1616 struct ext4_fc_add_range fc_add_ex;
1617 struct ext4_extent newex, *ex;
1618 struct inode *inode;
1619 ext4_lblk_t start, cur;
1621 ext4_fsblk_t start_pblk;
1622 struct ext4_map_blocks map;
1623 struct ext4_ext_path *path = NULL;
1626 memcpy(&fc_add_ex, val, sizeof(fc_add_ex));
1627 ex = (struct ext4_extent *)&fc_add_ex.fc_ex;
1629 trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,
1630 le32_to_cpu(fc_add_ex.fc_ino), le32_to_cpu(ex->ee_block),
1631 ext4_ext_get_actual_len(ex));
1633 inode = ext4_iget(sb, le32_to_cpu(fc_add_ex.fc_ino), EXT4_IGET_NORMAL);
1634 if (IS_ERR(inode)) {
1635 jbd_debug(1, "Inode not found.");
1639 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1643 start = le32_to_cpu(ex->ee_block);
1644 start_pblk = ext4_ext_pblock(ex);
1645 len = ext4_ext_get_actual_len(ex);
1649 jbd_debug(1, "ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",
1650 start, start_pblk, len, ext4_ext_is_unwritten(ex),
1653 while (remaining > 0) {
1655 map.m_len = remaining;
1657 ret = ext4_map_blocks(NULL, inode, &map, 0);
1663 /* Range is not mapped */
1664 path = ext4_find_extent(inode, cur, NULL, 0);
1667 memset(&newex, 0, sizeof(newex));
1668 newex.ee_block = cpu_to_le32(cur);
1669 ext4_ext_store_pblock(
1670 &newex, start_pblk + cur - start);
1671 newex.ee_len = cpu_to_le16(map.m_len);
1672 if (ext4_ext_is_unwritten(ex))
1673 ext4_ext_mark_unwritten(&newex);
1674 down_write(&EXT4_I(inode)->i_data_sem);
1675 ret = ext4_ext_insert_extent(
1676 NULL, inode, &path, &newex, 0);
1677 up_write((&EXT4_I(inode)->i_data_sem));
1678 ext4_ext_drop_refs(path);
1685 if (start_pblk + cur - start != map.m_pblk) {
1687 * Logical to physical mapping changed. This can happen
1688 * if this range was removed and then reallocated to
1689 * map to new physical blocks during a fast commit.
1691 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1692 ext4_ext_is_unwritten(ex),
1693 start_pblk + cur - start);
1697 * Mark the old blocks as free since they aren't used
1698 * anymore. We maintain an array of all the modified
1699 * inodes. In case these blocks are still used at either
1700 * a different logical range in the same inode or in
1701 * some different inode, we will mark them as allocated
1702 * at the end of the FC replay using our array of
1705 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1709 /* Range is mapped and needs a state change */
1710 jbd_debug(1, "Converting from %ld to %d %lld",
1711 map.m_flags & EXT4_MAP_UNWRITTEN,
1712 ext4_ext_is_unwritten(ex), map.m_pblk);
1713 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1714 ext4_ext_is_unwritten(ex), map.m_pblk);
1718 * We may have split the extent tree while toggling the state.
1719 * Try to shrink the extent tree now.
1721 ext4_ext_replay_shrink_inode(inode, start + len);
1724 remaining -= map.m_len;
1726 ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>
1727 sb->s_blocksize_bits);
1733 /* Replay DEL_RANGE tag */
1735 ext4_fc_replay_del_range(struct super_block *sb, struct ext4_fc_tl *tl,
1738 struct inode *inode;
1739 struct ext4_fc_del_range lrange;
1740 struct ext4_map_blocks map;
1741 ext4_lblk_t cur, remaining;
1744 memcpy(&lrange, val, sizeof(lrange));
1745 cur = le32_to_cpu(lrange.fc_lblk);
1746 remaining = le32_to_cpu(lrange.fc_len);
1748 trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,
1749 le32_to_cpu(lrange.fc_ino), cur, remaining);
1751 inode = ext4_iget(sb, le32_to_cpu(lrange.fc_ino), EXT4_IGET_NORMAL);
1752 if (IS_ERR(inode)) {
1753 jbd_debug(1, "Inode %d not found", le32_to_cpu(lrange.fc_ino));
1757 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1761 jbd_debug(1, "DEL_RANGE, inode %ld, lblk %d, len %d\n",
1762 inode->i_ino, le32_to_cpu(lrange.fc_lblk),
1763 le32_to_cpu(lrange.fc_len));
1764 while (remaining > 0) {
1766 map.m_len = remaining;
1768 ret = ext4_map_blocks(NULL, inode, &map, 0);
1774 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1776 remaining -= map.m_len;
1781 down_write(&EXT4_I(inode)->i_data_sem);
1782 ret = ext4_ext_remove_space(inode, le32_to_cpu(lrange.fc_lblk),
1783 le32_to_cpu(lrange.fc_lblk) +
1784 le32_to_cpu(lrange.fc_len) - 1);
1785 up_write(&EXT4_I(inode)->i_data_sem);
1788 ext4_ext_replay_shrink_inode(inode,
1789 i_size_read(inode) >> sb->s_blocksize_bits);
1790 ext4_mark_inode_dirty(NULL, inode);
1796 static inline const char *tag2str(u16 tag)
1799 case EXT4_FC_TAG_LINK:
1800 return "TAG_ADD_ENTRY";
1801 case EXT4_FC_TAG_UNLINK:
1802 return "TAG_DEL_ENTRY";
1803 case EXT4_FC_TAG_ADD_RANGE:
1804 return "TAG_ADD_RANGE";
1805 case EXT4_FC_TAG_CREAT:
1806 return "TAG_CREAT_DENTRY";
1807 case EXT4_FC_TAG_DEL_RANGE:
1808 return "TAG_DEL_RANGE";
1809 case EXT4_FC_TAG_INODE:
1811 case EXT4_FC_TAG_PAD:
1813 case EXT4_FC_TAG_TAIL:
1815 case EXT4_FC_TAG_HEAD:
1822 static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)
1824 struct ext4_fc_replay_state *state;
1825 struct inode *inode;
1826 struct ext4_ext_path *path = NULL;
1827 struct ext4_map_blocks map;
1829 ext4_lblk_t cur, end;
1831 state = &EXT4_SB(sb)->s_fc_replay_state;
1832 for (i = 0; i < state->fc_modified_inodes_used; i++) {
1833 inode = ext4_iget(sb, state->fc_modified_inodes[i],
1835 if (IS_ERR(inode)) {
1836 jbd_debug(1, "Inode %d not found.",
1837 state->fc_modified_inodes[i]);
1841 end = EXT_MAX_BLOCKS;
1844 map.m_len = end - cur;
1846 ret = ext4_map_blocks(NULL, inode, &map, 0);
1851 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
1852 if (!IS_ERR(path)) {
1853 for (j = 0; j < path->p_depth; j++)
1854 ext4_mb_mark_bb(inode->i_sb,
1855 path[j].p_block, 1, 1);
1856 ext4_ext_drop_refs(path);
1860 ext4_mb_mark_bb(inode->i_sb, map.m_pblk,
1863 cur = cur + (map.m_len ? map.m_len : 1);
1871 * Check if block is in excluded regions for block allocation. The simple
1872 * allocator that runs during replay phase is calls this function to see
1873 * if it is okay to use a block.
1875 bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)
1878 struct ext4_fc_replay_state *state;
1880 state = &EXT4_SB(sb)->s_fc_replay_state;
1881 for (i = 0; i < state->fc_regions_valid; i++) {
1882 if (state->fc_regions[i].ino == 0 ||
1883 state->fc_regions[i].len == 0)
1885 if (blk >= state->fc_regions[i].pblk &&
1886 blk < state->fc_regions[i].pblk + state->fc_regions[i].len)
1892 /* Cleanup function called after replay */
1893 void ext4_fc_replay_cleanup(struct super_block *sb)
1895 struct ext4_sb_info *sbi = EXT4_SB(sb);
1897 sbi->s_mount_state &= ~EXT4_FC_REPLAY;
1898 kfree(sbi->s_fc_replay_state.fc_regions);
1899 kfree(sbi->s_fc_replay_state.fc_modified_inodes);
1903 * Recovery Scan phase handler
1905 * This function is called during the scan phase and is responsible
1906 * for doing following things:
1907 * - Make sure the fast commit area has valid tags for replay
1908 * - Count number of tags that need to be replayed by the replay handler
1910 * - Create a list of excluded blocks for allocation during replay phase
1912 * This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is
1913 * incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP
1914 * to indicate that scan has finished and JBD2 can now start replay phase.
1915 * It returns a negative error to indicate that there was an error. At the end
1916 * of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set
1917 * to indicate the number of tags that need to replayed during the replay phase.
1919 static int ext4_fc_replay_scan(journal_t *journal,
1920 struct buffer_head *bh, int off,
1923 struct super_block *sb = journal->j_private;
1924 struct ext4_sb_info *sbi = EXT4_SB(sb);
1925 struct ext4_fc_replay_state *state;
1926 int ret = JBD2_FC_REPLAY_CONTINUE;
1927 struct ext4_fc_add_range ext;
1928 struct ext4_fc_tl tl;
1929 struct ext4_fc_tail tail;
1930 __u8 *start, *end, *cur, *val;
1931 struct ext4_fc_head head;
1932 struct ext4_extent *ex;
1934 state = &sbi->s_fc_replay_state;
1936 start = (u8 *)bh->b_data;
1937 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
1939 if (state->fc_replay_expected_off == 0) {
1940 state->fc_cur_tag = 0;
1941 state->fc_replay_num_tags = 0;
1943 state->fc_regions = NULL;
1944 state->fc_regions_valid = state->fc_regions_used =
1945 state->fc_regions_size = 0;
1946 /* Check if we can stop early */
1947 if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
1948 != EXT4_FC_TAG_HEAD)
1952 if (off != state->fc_replay_expected_off) {
1953 ret = -EFSCORRUPTED;
1957 state->fc_replay_expected_off++;
1958 for (cur = start; cur < end; cur = cur + sizeof(tl) + le16_to_cpu(tl.fc_len)) {
1959 memcpy(&tl, cur, sizeof(tl));
1960 val = cur + sizeof(tl);
1961 jbd_debug(3, "Scan phase, tag:%s, blk %lld\n",
1962 tag2str(le16_to_cpu(tl.fc_tag)), bh->b_blocknr);
1963 switch (le16_to_cpu(tl.fc_tag)) {
1964 case EXT4_FC_TAG_ADD_RANGE:
1965 memcpy(&ext, val, sizeof(ext));
1966 ex = (struct ext4_extent *)&ext.fc_ex;
1967 ret = ext4_fc_record_regions(sb,
1968 le32_to_cpu(ext.fc_ino),
1969 le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
1970 ext4_ext_get_actual_len(ex), 0);
1973 ret = JBD2_FC_REPLAY_CONTINUE;
1975 case EXT4_FC_TAG_DEL_RANGE:
1976 case EXT4_FC_TAG_LINK:
1977 case EXT4_FC_TAG_UNLINK:
1978 case EXT4_FC_TAG_CREAT:
1979 case EXT4_FC_TAG_INODE:
1980 case EXT4_FC_TAG_PAD:
1981 state->fc_cur_tag++;
1982 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
1983 sizeof(tl) + le16_to_cpu(tl.fc_len));
1985 case EXT4_FC_TAG_TAIL:
1986 state->fc_cur_tag++;
1987 memcpy(&tail, val, sizeof(tail));
1988 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
1990 offsetof(struct ext4_fc_tail,
1992 if (le32_to_cpu(tail.fc_tid) == expected_tid &&
1993 le32_to_cpu(tail.fc_crc) == state->fc_crc) {
1994 state->fc_replay_num_tags = state->fc_cur_tag;
1995 state->fc_regions_valid =
1996 state->fc_regions_used;
1998 ret = state->fc_replay_num_tags ?
1999 JBD2_FC_REPLAY_STOP : -EFSBADCRC;
2003 case EXT4_FC_TAG_HEAD:
2004 memcpy(&head, val, sizeof(head));
2005 if (le32_to_cpu(head.fc_features) &
2006 ~EXT4_FC_SUPPORTED_FEATURES) {
2010 if (le32_to_cpu(head.fc_tid) != expected_tid) {
2011 ret = JBD2_FC_REPLAY_STOP;
2014 state->fc_cur_tag++;
2015 state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
2016 sizeof(tl) + le16_to_cpu(tl.fc_len));
2019 ret = state->fc_replay_num_tags ?
2020 JBD2_FC_REPLAY_STOP : -ECANCELED;
2022 if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
2027 trace_ext4_fc_replay_scan(sb, ret, off);
2032 * Main recovery path entry point.
2033 * The meaning of return codes is similar as above.
2035 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
2036 enum passtype pass, int off, tid_t expected_tid)
2038 struct super_block *sb = journal->j_private;
2039 struct ext4_sb_info *sbi = EXT4_SB(sb);
2040 struct ext4_fc_tl tl;
2041 __u8 *start, *end, *cur, *val;
2042 int ret = JBD2_FC_REPLAY_CONTINUE;
2043 struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;
2044 struct ext4_fc_tail tail;
2046 if (pass == PASS_SCAN) {
2047 state->fc_current_pass = PASS_SCAN;
2048 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
2051 if (state->fc_current_pass != pass) {
2052 state->fc_current_pass = pass;
2053 sbi->s_mount_state |= EXT4_FC_REPLAY;
2055 if (!sbi->s_fc_replay_state.fc_replay_num_tags) {
2056 jbd_debug(1, "Replay stops\n");
2057 ext4_fc_set_bitmaps_and_counters(sb);
2061 #ifdef CONFIG_EXT4_DEBUG
2062 if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {
2063 pr_warn("Dropping fc block %d because max_replay set\n", off);
2064 return JBD2_FC_REPLAY_STOP;
2068 start = (u8 *)bh->b_data;
2069 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
2071 for (cur = start; cur < end; cur = cur + sizeof(tl) + le16_to_cpu(tl.fc_len)) {
2072 memcpy(&tl, cur, sizeof(tl));
2073 val = cur + sizeof(tl);
2075 if (state->fc_replay_num_tags == 0) {
2076 ret = JBD2_FC_REPLAY_STOP;
2077 ext4_fc_set_bitmaps_and_counters(sb);
2080 jbd_debug(3, "Replay phase, tag:%s\n",
2081 tag2str(le16_to_cpu(tl.fc_tag)));
2082 state->fc_replay_num_tags--;
2083 switch (le16_to_cpu(tl.fc_tag)) {
2084 case EXT4_FC_TAG_LINK:
2085 ret = ext4_fc_replay_link(sb, &tl, val);
2087 case EXT4_FC_TAG_UNLINK:
2088 ret = ext4_fc_replay_unlink(sb, &tl, val);
2090 case EXT4_FC_TAG_ADD_RANGE:
2091 ret = ext4_fc_replay_add_range(sb, &tl, val);
2093 case EXT4_FC_TAG_CREAT:
2094 ret = ext4_fc_replay_create(sb, &tl, val);
2096 case EXT4_FC_TAG_DEL_RANGE:
2097 ret = ext4_fc_replay_del_range(sb, &tl, val);
2099 case EXT4_FC_TAG_INODE:
2100 ret = ext4_fc_replay_inode(sb, &tl, val);
2102 case EXT4_FC_TAG_PAD:
2103 trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,
2104 le16_to_cpu(tl.fc_len), 0);
2106 case EXT4_FC_TAG_TAIL:
2107 trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL, 0,
2108 le16_to_cpu(tl.fc_len), 0);
2109 memcpy(&tail, val, sizeof(tail));
2110 WARN_ON(le32_to_cpu(tail.fc_tid) != expected_tid);
2112 case EXT4_FC_TAG_HEAD:
2115 trace_ext4_fc_replay(sb, le16_to_cpu(tl.fc_tag), 0,
2116 le16_to_cpu(tl.fc_len), 0);
2122 ret = JBD2_FC_REPLAY_CONTINUE;
2127 void ext4_fc_init(struct super_block *sb, journal_t *journal)
2130 * We set replay callback even if fast commit disabled because we may
2131 * could still have fast commit blocks that need to be replayed even if
2132 * fast commit has now been turned off.
2134 journal->j_fc_replay_callback = ext4_fc_replay;
2135 if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
2137 journal->j_fc_cleanup_callback = ext4_fc_cleanup;
2140 static const char *fc_ineligible_reasons[] = {
2141 "Extended attributes changed",
2143 "Journal flag changed",
2144 "Insufficient memory",
2153 int ext4_fc_info_show(struct seq_file *seq, void *v)
2155 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);
2156 struct ext4_fc_stats *stats = &sbi->s_fc_stats;
2159 if (v != SEQ_START_TOKEN)
2163 "fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",
2164 stats->fc_num_commits, stats->fc_ineligible_commits,
2166 div_u64(sbi->s_fc_avg_commit_time, 1000));
2167 seq_puts(seq, "Ineligible reasons:\n");
2168 for (i = 0; i < EXT4_FC_REASON_MAX; i++)
2169 seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],
2170 stats->fc_ineligible_reason_count[i]);
2175 int __init ext4_fc_init_dentry_cache(void)
2177 ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,
2178 SLAB_RECLAIM_ACCOUNT);
2180 if (ext4_fc_dentry_cachep == NULL)
2186 void ext4_fc_destroy_dentry_cache(void)
2188 kmem_cache_destroy(ext4_fc_dentry_cachep);