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);
155 ei->i_fc_committed_subtid = 0;
158 /* This function must be called with sbi->s_fc_lock held. */
159 static void ext4_fc_wait_committing_inode(struct inode *inode)
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 sbi->s_mount_flags |= 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_SB(sb)->s_mount_flags |= 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_SB(sb)->s_mount_flags & 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 (sbi->s_mount_flags & 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 (sbi->s_mount_flags & 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 dentry *dentry)
457 struct __track_dentry_update_args args;
458 struct inode *inode = d_inode(dentry);
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 /* __track_fn for inode tracking */
470 static int __track_inode(struct inode *inode, void *arg, bool update)
475 EXT4_I(inode)->i_fc_lblk_len = 0;
480 void ext4_fc_track_inode(handle_t *handle, struct inode *inode)
484 if (S_ISDIR(inode->i_mode))
487 ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);
488 trace_ext4_fc_track_inode(inode, ret);
491 struct __track_range_args {
492 ext4_lblk_t start, end;
495 /* __track_fn for tracking data updates */
496 static int __track_range(struct inode *inode, void *arg, bool update)
498 struct ext4_inode_info *ei = EXT4_I(inode);
499 ext4_lblk_t oldstart;
500 struct __track_range_args *__arg =
501 (struct __track_range_args *)arg;
503 if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {
504 ext4_debug("Special inode %ld being modified\n", inode->i_ino);
508 oldstart = ei->i_fc_lblk_start;
510 if (update && ei->i_fc_lblk_len > 0) {
511 ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);
513 max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -
514 ei->i_fc_lblk_start + 1;
516 ei->i_fc_lblk_start = __arg->start;
517 ei->i_fc_lblk_len = __arg->end - __arg->start + 1;
523 void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,
526 struct __track_range_args args;
529 if (S_ISDIR(inode->i_mode))
535 ret = ext4_fc_track_template(handle, inode, __track_range, &args, 1);
537 trace_ext4_fc_track_range(inode, start, end, ret);
540 static void ext4_fc_submit_bh(struct super_block *sb)
542 int write_flags = REQ_SYNC;
543 struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
545 /* TODO: REQ_FUA | REQ_PREFLUSH is unnecessarily expensive. */
546 if (test_opt(sb, BARRIER))
547 write_flags |= REQ_FUA | REQ_PREFLUSH;
549 clear_buffer_dirty(bh);
550 set_buffer_uptodate(bh);
551 bh->b_end_io = ext4_end_buffer_io_sync;
552 submit_bh(REQ_OP_WRITE, write_flags, bh);
553 EXT4_SB(sb)->s_fc_bh = NULL;
556 /* Ext4 commit path routines */
558 /* memzero and update CRC */
559 static void *ext4_fc_memzero(struct super_block *sb, void *dst, int len,
564 ret = memset(dst, 0, len);
566 *crc = ext4_chksum(EXT4_SB(sb), *crc, dst, len);
571 * Allocate len bytes on a fast commit buffer.
573 * During the commit time this function is used to manage fast commit
574 * block space. We don't split a fast commit log onto different
575 * blocks. So this function makes sure that if there's not enough space
576 * on the current block, the remaining space in the current block is
577 * marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,
578 * new block is from jbd2 and CRC is updated to reflect the padding
581 static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
583 struct ext4_fc_tl *tl;
584 struct ext4_sb_info *sbi = EXT4_SB(sb);
585 struct buffer_head *bh;
586 int bsize = sbi->s_journal->j_blocksize;
587 int ret, off = sbi->s_fc_bytes % bsize;
591 * After allocating len, we should have space at least for a 0 byte
594 if (len + sizeof(struct ext4_fc_tl) > bsize)
597 if (bsize - off - 1 > len + sizeof(struct ext4_fc_tl)) {
599 * Only allocate from current buffer if we have enough space for
600 * this request AND we have space to add a zero byte padding.
603 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
608 sbi->s_fc_bytes += len;
609 return sbi->s_fc_bh->b_data + off;
611 /* Need to add PAD tag */
612 tl = (struct ext4_fc_tl *)(sbi->s_fc_bh->b_data + off);
613 tl->fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);
614 pad_len = bsize - off - 1 - sizeof(struct ext4_fc_tl);
615 tl->fc_len = cpu_to_le16(pad_len);
617 *crc = ext4_chksum(sbi, *crc, tl, sizeof(*tl));
619 ext4_fc_memzero(sb, tl + 1, pad_len, crc);
620 ext4_fc_submit_bh(sb);
622 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
626 sbi->s_fc_bytes = (sbi->s_fc_bytes / bsize + 1) * bsize + len;
627 return sbi->s_fc_bh->b_data;
630 /* memcpy to fc reserved space and update CRC */
631 static void *ext4_fc_memcpy(struct super_block *sb, void *dst, const void *src,
635 *crc = ext4_chksum(EXT4_SB(sb), *crc, src, len);
636 return memcpy(dst, src, len);
640 * Complete a fast commit by writing tail tag.
642 * Writing tail tag marks the end of a fast commit. In order to guarantee
643 * atomicity, after writing tail tag, even if there's space remaining
644 * in the block, next commit shouldn't use it. That's why tail tag
645 * has the length as that of the remaining space on the block.
647 static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
649 struct ext4_sb_info *sbi = EXT4_SB(sb);
650 struct ext4_fc_tl tl;
651 struct ext4_fc_tail tail;
652 int off, bsize = sbi->s_journal->j_blocksize;
656 * ext4_fc_reserve_space takes care of allocating an extra block if
657 * there's no enough space on this block for accommodating this tail.
659 dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(tail), &crc);
663 off = sbi->s_fc_bytes % bsize;
665 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);
666 tl.fc_len = cpu_to_le16(bsize - off - 1 + sizeof(struct ext4_fc_tail));
667 sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);
669 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), &crc);
671 tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);
672 ext4_fc_memcpy(sb, dst, &tail.fc_tid, sizeof(tail.fc_tid), &crc);
673 dst += sizeof(tail.fc_tid);
674 tail.fc_crc = cpu_to_le32(crc);
675 ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
677 ext4_fc_submit_bh(sb);
683 * Adds tag, length, value and updates CRC. Returns true if tlv was added.
684 * Returns false if there's not enough space.
686 static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,
689 struct ext4_fc_tl tl;
692 dst = ext4_fc_reserve_space(sb, sizeof(tl) + len, crc);
696 tl.fc_tag = cpu_to_le16(tag);
697 tl.fc_len = cpu_to_le16(len);
699 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
700 ext4_fc_memcpy(sb, dst + sizeof(tl), val, len, crc);
705 /* Same as above, but adds dentry tlv. */
706 static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u16 tag,
707 int parent_ino, int ino, int dlen,
708 const unsigned char *dname,
711 struct ext4_fc_dentry_info fcd;
712 struct ext4_fc_tl tl;
713 u8 *dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(fcd) + dlen,
719 fcd.fc_parent_ino = cpu_to_le32(parent_ino);
720 fcd.fc_ino = cpu_to_le32(ino);
721 tl.fc_tag = cpu_to_le16(tag);
722 tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);
723 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
725 ext4_fc_memcpy(sb, dst, &fcd, sizeof(fcd), crc);
727 ext4_fc_memcpy(sb, dst, dname, dlen, crc);
734 * Writes inode in the fast commit space under TLV with tag @tag.
735 * Returns 0 on success, error on failure.
737 static int ext4_fc_write_inode(struct inode *inode, u32 *crc)
739 struct ext4_inode_info *ei = EXT4_I(inode);
740 int inode_len = EXT4_GOOD_OLD_INODE_SIZE;
742 struct ext4_iloc iloc;
743 struct ext4_fc_inode fc_inode;
744 struct ext4_fc_tl tl;
747 ret = ext4_get_inode_loc(inode, &iloc);
751 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)
752 inode_len += ei->i_extra_isize;
754 fc_inode.fc_ino = cpu_to_le32(inode->i_ino);
755 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);
756 tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));
758 dst = ext4_fc_reserve_space(inode->i_sb,
759 sizeof(tl) + inode_len + sizeof(fc_inode.fc_ino), crc);
763 if (!ext4_fc_memcpy(inode->i_sb, dst, &tl, sizeof(tl), crc))
766 if (!ext4_fc_memcpy(inode->i_sb, dst, &fc_inode, sizeof(fc_inode), crc))
768 dst += sizeof(fc_inode);
769 if (!ext4_fc_memcpy(inode->i_sb, dst, (u8 *)ext4_raw_inode(&iloc),
777 * Writes updated data ranges for the inode in question. Updates CRC.
778 * Returns 0 on success, error otherwise.
780 static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)
782 ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;
783 struct ext4_inode_info *ei = EXT4_I(inode);
784 struct ext4_map_blocks map;
785 struct ext4_fc_add_range fc_ext;
786 struct ext4_fc_del_range lrange;
787 struct ext4_extent *ex;
790 mutex_lock(&ei->i_fc_lock);
791 if (ei->i_fc_lblk_len == 0) {
792 mutex_unlock(&ei->i_fc_lock);
795 old_blk_size = ei->i_fc_lblk_start;
796 new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;
797 ei->i_fc_lblk_len = 0;
798 mutex_unlock(&ei->i_fc_lock);
800 cur_lblk_off = old_blk_size;
801 jbd_debug(1, "%s: will try writing %d to %d for inode %ld\n",
802 __func__, cur_lblk_off, new_blk_size, inode->i_ino);
804 while (cur_lblk_off <= new_blk_size) {
805 map.m_lblk = cur_lblk_off;
806 map.m_len = new_blk_size - cur_lblk_off + 1;
807 ret = ext4_map_blocks(NULL, inode, &map, 0);
811 if (map.m_len == 0) {
817 lrange.fc_ino = cpu_to_le32(inode->i_ino);
818 lrange.fc_lblk = cpu_to_le32(map.m_lblk);
819 lrange.fc_len = cpu_to_le32(map.m_len);
820 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,
821 sizeof(lrange), (u8 *)&lrange, crc))
824 fc_ext.fc_ino = cpu_to_le32(inode->i_ino);
825 ex = (struct ext4_extent *)&fc_ext.fc_ex;
826 ex->ee_block = cpu_to_le32(map.m_lblk);
827 ex->ee_len = cpu_to_le16(map.m_len);
828 ext4_ext_store_pblock(ex, map.m_pblk);
829 if (map.m_flags & EXT4_MAP_UNWRITTEN)
830 ext4_ext_mark_unwritten(ex);
832 ext4_ext_mark_initialized(ex);
833 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,
834 sizeof(fc_ext), (u8 *)&fc_ext, crc))
838 cur_lblk_off += map.m_len;
845 /* Submit data for all the fast commit inodes */
846 static int ext4_fc_submit_inode_data_all(journal_t *journal)
848 struct super_block *sb = (struct super_block *)(journal->j_private);
849 struct ext4_sb_info *sbi = EXT4_SB(sb);
850 struct ext4_inode_info *ei;
851 struct list_head *pos;
854 spin_lock(&sbi->s_fc_lock);
855 sbi->s_mount_flags |= EXT4_MF_FC_COMMITTING;
856 list_for_each(pos, &sbi->s_fc_q[FC_Q_MAIN]) {
857 ei = list_entry(pos, struct ext4_inode_info, i_fc_list);
858 ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
859 while (atomic_read(&ei->i_fc_updates)) {
862 prepare_to_wait(&ei->i_fc_wait, &wait,
863 TASK_UNINTERRUPTIBLE);
864 if (atomic_read(&ei->i_fc_updates)) {
865 spin_unlock(&sbi->s_fc_lock);
867 spin_lock(&sbi->s_fc_lock);
869 finish_wait(&ei->i_fc_wait, &wait);
871 spin_unlock(&sbi->s_fc_lock);
872 ret = jbd2_submit_inode_data(ei->jinode);
875 spin_lock(&sbi->s_fc_lock);
877 spin_unlock(&sbi->s_fc_lock);
882 /* Wait for completion of data for all the fast commit inodes */
883 static int ext4_fc_wait_inode_data_all(journal_t *journal)
885 struct super_block *sb = (struct super_block *)(journal->j_private);
886 struct ext4_sb_info *sbi = EXT4_SB(sb);
887 struct ext4_inode_info *pos, *n;
890 spin_lock(&sbi->s_fc_lock);
891 list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
892 if (!ext4_test_inode_state(&pos->vfs_inode,
893 EXT4_STATE_FC_COMMITTING))
895 spin_unlock(&sbi->s_fc_lock);
897 ret = jbd2_wait_inode_data(journal, pos->jinode);
900 spin_lock(&sbi->s_fc_lock);
902 spin_unlock(&sbi->s_fc_lock);
907 /* Commit all the directory entry updates */
908 static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)
910 struct super_block *sb = (struct super_block *)(journal->j_private);
911 struct ext4_sb_info *sbi = EXT4_SB(sb);
912 struct ext4_fc_dentry_update *fc_dentry;
914 struct list_head *pos, *n, *fcd_pos, *fcd_n;
915 struct ext4_inode_info *ei;
918 if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))
920 list_for_each_safe(fcd_pos, fcd_n, &sbi->s_fc_dentry_q[FC_Q_MAIN]) {
921 fc_dentry = list_entry(fcd_pos, struct ext4_fc_dentry_update,
923 if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {
924 spin_unlock(&sbi->s_fc_lock);
925 if (!ext4_fc_add_dentry_tlv(
926 sb, fc_dentry->fcd_op,
927 fc_dentry->fcd_parent, fc_dentry->fcd_ino,
928 fc_dentry->fcd_name.len,
929 fc_dentry->fcd_name.name, crc)) {
933 spin_lock(&sbi->s_fc_lock);
938 list_for_each_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN]) {
939 ei = list_entry(pos, struct ext4_inode_info, i_fc_list);
940 if (ei->vfs_inode.i_ino == fc_dentry->fcd_ino) {
941 inode = &ei->vfs_inode;
946 * If we don't find inode in our list, then it was deleted,
947 * in which case, we don't need to record it's create tag.
951 spin_unlock(&sbi->s_fc_lock);
954 * We first write the inode and then the create dirent. This
955 * allows the recovery code to create an unnamed inode first
956 * and then link it to a directory entry. This allows us
957 * to use namei.c routines almost as is and simplifies
960 ret = ext4_fc_write_inode(inode, crc);
964 ret = ext4_fc_write_inode_data(inode, crc);
968 if (!ext4_fc_add_dentry_tlv(
969 sb, fc_dentry->fcd_op,
970 fc_dentry->fcd_parent, fc_dentry->fcd_ino,
971 fc_dentry->fcd_name.len,
972 fc_dentry->fcd_name.name, crc)) {
977 spin_lock(&sbi->s_fc_lock);
981 spin_lock(&sbi->s_fc_lock);
985 static int ext4_fc_perform_commit(journal_t *journal)
987 struct super_block *sb = (struct super_block *)(journal->j_private);
988 struct ext4_sb_info *sbi = EXT4_SB(sb);
989 struct ext4_inode_info *iter;
990 struct ext4_fc_head head;
991 struct list_head *pos;
993 struct blk_plug plug;
997 ret = ext4_fc_submit_inode_data_all(journal);
1001 ret = ext4_fc_wait_inode_data_all(journal);
1005 blk_start_plug(&plug);
1006 if (sbi->s_fc_bytes == 0) {
1008 * Add a head tag only if this is the first fast commit
1011 head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);
1012 head.fc_tid = cpu_to_le32(
1013 sbi->s_journal->j_running_transaction->t_tid);
1014 if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),
1019 spin_lock(&sbi->s_fc_lock);
1020 ret = ext4_fc_commit_dentry_updates(journal, &crc);
1022 spin_unlock(&sbi->s_fc_lock);
1026 list_for_each(pos, &sbi->s_fc_q[FC_Q_MAIN]) {
1027 iter = list_entry(pos, struct ext4_inode_info, i_fc_list);
1028 inode = &iter->vfs_inode;
1029 if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))
1032 spin_unlock(&sbi->s_fc_lock);
1033 ret = ext4_fc_write_inode_data(inode, &crc);
1036 ret = ext4_fc_write_inode(inode, &crc);
1039 spin_lock(&sbi->s_fc_lock);
1040 EXT4_I(inode)->i_fc_committed_subtid =
1041 atomic_read(&sbi->s_fc_subtid);
1043 spin_unlock(&sbi->s_fc_lock);
1045 ret = ext4_fc_write_tail(sb, crc);
1048 blk_finish_plug(&plug);
1053 * The main commit entry point. Performs a fast commit for transaction
1054 * commit_tid if needed. If it's not possible to perform a fast commit
1055 * due to various reasons, we fall back to full commit. Returns 0
1056 * on success, error otherwise.
1058 int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
1060 struct super_block *sb = (struct super_block *)(journal->j_private);
1061 struct ext4_sb_info *sbi = EXT4_SB(sb);
1062 int nblks = 0, ret, bsize = journal->j_blocksize;
1063 int subtid = atomic_read(&sbi->s_fc_subtid);
1064 int reason = EXT4_FC_REASON_OK, fc_bufs_before = 0;
1065 ktime_t start_time, commit_time;
1067 trace_ext4_fc_commit_start(sb);
1069 start_time = ktime_get();
1071 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
1072 (ext4_fc_is_ineligible(sb))) {
1073 reason = EXT4_FC_REASON_INELIGIBLE;
1078 ret = jbd2_fc_begin_commit(journal, commit_tid);
1079 if (ret == -EALREADY) {
1080 /* There was an ongoing commit, check if we need to restart */
1081 if (atomic_read(&sbi->s_fc_subtid) <= subtid &&
1082 commit_tid > journal->j_commit_sequence)
1084 reason = EXT4_FC_REASON_ALREADY_COMMITTED;
1087 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1088 reason = EXT4_FC_REASON_FC_START_FAILED;
1092 fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
1093 ret = ext4_fc_perform_commit(journal);
1095 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1096 reason = EXT4_FC_REASON_FC_FAILED;
1099 nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;
1100 ret = jbd2_fc_wait_bufs(journal, nblks);
1102 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1103 reason = EXT4_FC_REASON_FC_FAILED;
1106 atomic_inc(&sbi->s_fc_subtid);
1107 jbd2_fc_end_commit(journal);
1109 /* Has any ineligible update happened since we started? */
1110 if (reason == EXT4_FC_REASON_OK && ext4_fc_is_ineligible(sb)) {
1111 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1112 reason = EXT4_FC_REASON_INELIGIBLE;
1115 spin_lock(&sbi->s_fc_lock);
1116 if (reason != EXT4_FC_REASON_OK &&
1117 reason != EXT4_FC_REASON_ALREADY_COMMITTED) {
1118 sbi->s_fc_stats.fc_ineligible_commits++;
1120 sbi->s_fc_stats.fc_num_commits++;
1121 sbi->s_fc_stats.fc_numblks += nblks;
1123 spin_unlock(&sbi->s_fc_lock);
1124 nblks = (reason == EXT4_FC_REASON_OK) ? nblks : 0;
1125 trace_ext4_fc_commit_stop(sb, nblks, reason);
1126 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1128 * weight the commit time higher than the average time so we don't
1129 * react too strongly to vast changes in the commit time
1131 if (likely(sbi->s_fc_avg_commit_time))
1132 sbi->s_fc_avg_commit_time = (commit_time +
1133 sbi->s_fc_avg_commit_time * 3) / 4;
1135 sbi->s_fc_avg_commit_time = commit_time;
1137 "Fast commit ended with blks = %d, reason = %d, subtid - %d",
1138 nblks, reason, subtid);
1139 if (reason == EXT4_FC_REASON_FC_FAILED)
1140 return jbd2_fc_end_commit_fallback(journal);
1141 if (reason == EXT4_FC_REASON_FC_START_FAILED ||
1142 reason == EXT4_FC_REASON_INELIGIBLE)
1143 return jbd2_complete_transaction(journal, commit_tid);
1148 * Fast commit cleanup routine. This is called after every fast commit and
1149 * full commit. full is true if we are called after a full commit.
1151 static void ext4_fc_cleanup(journal_t *journal, int full)
1153 struct super_block *sb = journal->j_private;
1154 struct ext4_sb_info *sbi = EXT4_SB(sb);
1155 struct ext4_inode_info *iter;
1156 struct ext4_fc_dentry_update *fc_dentry;
1157 struct list_head *pos, *n;
1159 if (full && sbi->s_fc_bh)
1160 sbi->s_fc_bh = NULL;
1162 jbd2_fc_release_bufs(journal);
1164 spin_lock(&sbi->s_fc_lock);
1165 list_for_each_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN]) {
1166 iter = list_entry(pos, struct ext4_inode_info, i_fc_list);
1167 list_del_init(&iter->i_fc_list);
1168 ext4_clear_inode_state(&iter->vfs_inode,
1169 EXT4_STATE_FC_COMMITTING);
1170 ext4_fc_reset_inode(&iter->vfs_inode);
1171 /* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
1173 #if (BITS_PER_LONG < 64)
1174 wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);
1176 wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);
1180 while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {
1181 fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],
1182 struct ext4_fc_dentry_update,
1184 list_del_init(&fc_dentry->fcd_list);
1185 spin_unlock(&sbi->s_fc_lock);
1187 if (fc_dentry->fcd_name.name &&
1188 fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
1189 kfree(fc_dentry->fcd_name.name);
1190 kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
1191 spin_lock(&sbi->s_fc_lock);
1194 list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],
1195 &sbi->s_fc_dentry_q[FC_Q_MAIN]);
1196 list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
1197 &sbi->s_fc_q[FC_Q_STAGING]);
1199 sbi->s_mount_flags &= ~EXT4_MF_FC_COMMITTING;
1200 sbi->s_mount_flags &= ~EXT4_MF_FC_INELIGIBLE;
1203 sbi->s_fc_bytes = 0;
1204 spin_unlock(&sbi->s_fc_lock);
1205 trace_ext4_fc_stats(sb);
1208 /* Ext4 Replay Path Routines */
1210 /* Get length of a particular tlv */
1211 static inline int ext4_fc_tag_len(struct ext4_fc_tl *tl)
1213 return le16_to_cpu(tl->fc_len);
1216 /* Get a pointer to "value" of a tlv */
1217 static inline u8 *ext4_fc_tag_val(struct ext4_fc_tl *tl)
1219 return (u8 *)tl + sizeof(*tl);
1222 /* Helper struct for dentry replay routines */
1223 struct dentry_info_args {
1224 int parent_ino, dname_len, ino, inode_len;
1228 static inline void tl_to_darg(struct dentry_info_args *darg,
1229 struct ext4_fc_tl *tl)
1231 struct ext4_fc_dentry_info *fcd;
1233 fcd = (struct ext4_fc_dentry_info *)ext4_fc_tag_val(tl);
1235 darg->parent_ino = le32_to_cpu(fcd->fc_parent_ino);
1236 darg->ino = le32_to_cpu(fcd->fc_ino);
1237 darg->dname = fcd->fc_dname;
1238 darg->dname_len = ext4_fc_tag_len(tl) -
1239 sizeof(struct ext4_fc_dentry_info);
1242 /* Unlink replay function */
1243 static int ext4_fc_replay_unlink(struct super_block *sb, struct ext4_fc_tl *tl)
1245 struct inode *inode, *old_parent;
1247 struct dentry_info_args darg;
1250 tl_to_darg(&darg, tl);
1252 trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,
1253 darg.parent_ino, darg.dname_len);
1255 entry.name = darg.dname;
1256 entry.len = darg.dname_len;
1257 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1259 if (IS_ERR_OR_NULL(inode)) {
1260 jbd_debug(1, "Inode %d not found", darg.ino);
1264 old_parent = ext4_iget(sb, darg.parent_ino,
1266 if (IS_ERR_OR_NULL(old_parent)) {
1267 jbd_debug(1, "Dir with inode %d not found", darg.parent_ino);
1272 ret = __ext4_unlink(NULL, old_parent, &entry, inode);
1273 /* -ENOENT ok coz it might not exist anymore. */
1281 static int ext4_fc_replay_link_internal(struct super_block *sb,
1282 struct dentry_info_args *darg,
1283 struct inode *inode)
1285 struct inode *dir = NULL;
1286 struct dentry *dentry_dir = NULL, *dentry_inode = NULL;
1287 struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);
1290 dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);
1292 jbd_debug(1, "Dir with inode %d not found.", darg->parent_ino);
1297 dentry_dir = d_obtain_alias(dir);
1298 if (IS_ERR(dentry_dir)) {
1299 jbd_debug(1, "Failed to obtain dentry");
1304 dentry_inode = d_alloc(dentry_dir, &qstr_dname);
1305 if (!dentry_inode) {
1306 jbd_debug(1, "Inode dentry not created.");
1311 ret = __ext4_link(dir, inode, dentry_inode);
1313 * It's possible that link already existed since data blocks
1314 * for the dir in question got persisted before we crashed OR
1315 * we replayed this tag and crashed before the entire replay
1318 if (ret && ret != -EEXIST) {
1319 jbd_debug(1, "Failed to link\n");
1332 d_drop(dentry_inode);
1339 /* Link replay function */
1340 static int ext4_fc_replay_link(struct super_block *sb, struct ext4_fc_tl *tl)
1342 struct inode *inode;
1343 struct dentry_info_args darg;
1346 tl_to_darg(&darg, tl);
1347 trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,
1348 darg.parent_ino, darg.dname_len);
1350 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1351 if (IS_ERR_OR_NULL(inode)) {
1352 jbd_debug(1, "Inode not found.");
1356 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1362 * Record all the modified inodes during replay. We use this later to setup
1363 * block bitmaps correctly.
1365 static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)
1367 struct ext4_fc_replay_state *state;
1370 state = &EXT4_SB(sb)->s_fc_replay_state;
1371 for (i = 0; i < state->fc_modified_inodes_used; i++)
1372 if (state->fc_modified_inodes[i] == ino)
1374 if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {
1375 state->fc_modified_inodes_size +=
1376 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1377 state->fc_modified_inodes = krealloc(
1378 state->fc_modified_inodes, sizeof(int) *
1379 state->fc_modified_inodes_size,
1381 if (!state->fc_modified_inodes)
1384 state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;
1389 * Inode replay function
1391 static int ext4_fc_replay_inode(struct super_block *sb, struct ext4_fc_tl *tl)
1393 struct ext4_fc_inode *fc_inode;
1394 struct ext4_inode *raw_inode;
1395 struct ext4_inode *raw_fc_inode;
1396 struct inode *inode = NULL;
1397 struct ext4_iloc iloc;
1398 int inode_len, ino, ret, tag = le16_to_cpu(tl->fc_tag);
1399 struct ext4_extent_header *eh;
1401 fc_inode = (struct ext4_fc_inode *)ext4_fc_tag_val(tl);
1403 ino = le32_to_cpu(fc_inode->fc_ino);
1404 trace_ext4_fc_replay(sb, tag, ino, 0, 0);
1406 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1407 if (!IS_ERR_OR_NULL(inode)) {
1408 ext4_ext_clear_bb(inode);
1412 ext4_fc_record_modified_inode(sb, ino);
1414 raw_fc_inode = (struct ext4_inode *)fc_inode->fc_raw_inode;
1415 ret = ext4_get_fc_inode_loc(sb, ino, &iloc);
1419 inode_len = ext4_fc_tag_len(tl) - sizeof(struct ext4_fc_inode);
1420 raw_inode = ext4_raw_inode(&iloc);
1422 memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));
1423 memcpy(&raw_inode->i_generation, &raw_fc_inode->i_generation,
1424 inode_len - offsetof(struct ext4_inode, i_generation));
1425 if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {
1426 eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);
1427 if (eh->eh_magic != EXT4_EXT_MAGIC) {
1428 memset(eh, 0, sizeof(*eh));
1429 eh->eh_magic = EXT4_EXT_MAGIC;
1430 eh->eh_max = cpu_to_le16(
1431 (sizeof(raw_inode->i_block) -
1432 sizeof(struct ext4_extent_header))
1433 / sizeof(struct ext4_extent));
1435 } else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {
1436 memcpy(raw_inode->i_block, raw_fc_inode->i_block,
1437 sizeof(raw_inode->i_block));
1440 /* Immediately update the inode on disk. */
1441 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1444 ret = sync_dirty_buffer(iloc.bh);
1447 ret = ext4_mark_inode_used(sb, ino);
1451 /* Given that we just wrote the inode on disk, this SHOULD succeed. */
1452 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1453 if (IS_ERR_OR_NULL(inode)) {
1454 jbd_debug(1, "Inode not found.");
1455 return -EFSCORRUPTED;
1459 * Our allocator could have made different decisions than before
1460 * crashing. This should be fixed but until then, we calculate
1461 * the number of blocks the inode.
1463 ext4_ext_replay_set_iblocks(inode);
1465 inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);
1466 ext4_reset_inode_seed(inode);
1468 ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));
1469 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1470 sync_dirty_buffer(iloc.bh);
1475 blkdev_issue_flush(sb->s_bdev, GFP_KERNEL);
1481 * Dentry create replay function.
1483 * EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the
1484 * inode for which we are trying to create a dentry here, should already have
1485 * been replayed before we start here.
1487 static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl)
1490 struct inode *inode = NULL;
1491 struct inode *dir = NULL;
1492 struct dentry_info_args darg;
1494 tl_to_darg(&darg, tl);
1496 trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,
1497 darg.parent_ino, darg.dname_len);
1499 /* This takes care of update group descriptor and other metadata */
1500 ret = ext4_mark_inode_used(sb, darg.ino);
1504 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1505 if (IS_ERR_OR_NULL(inode)) {
1506 jbd_debug(1, "inode %d not found.", darg.ino);
1512 if (S_ISDIR(inode->i_mode)) {
1514 * If we are creating a directory, we need to make sure that the
1515 * dot and dot dot dirents are setup properly.
1517 dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);
1518 if (IS_ERR_OR_NULL(dir)) {
1519 jbd_debug(1, "Dir %d not found.", darg.ino);
1522 ret = ext4_init_new_dir(NULL, dir, inode);
1529 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1532 set_nlink(inode, 1);
1533 ext4_mark_inode_dirty(NULL, inode);
1541 * Record physical disk regions which are in use as per fast commit area. Our
1542 * simple replay phase allocator excludes these regions from allocation.
1544 static int ext4_fc_record_regions(struct super_block *sb, int ino,
1545 ext4_lblk_t lblk, ext4_fsblk_t pblk, int len)
1547 struct ext4_fc_replay_state *state;
1548 struct ext4_fc_alloc_region *region;
1550 state = &EXT4_SB(sb)->s_fc_replay_state;
1551 if (state->fc_regions_used == state->fc_regions_size) {
1552 state->fc_regions_size +=
1553 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1554 state->fc_regions = krealloc(
1556 state->fc_regions_size *
1557 sizeof(struct ext4_fc_alloc_region),
1559 if (!state->fc_regions)
1562 region = &state->fc_regions[state->fc_regions_used++];
1564 region->lblk = lblk;
1565 region->pblk = pblk;
1571 /* Replay add range tag */
1572 static int ext4_fc_replay_add_range(struct super_block *sb,
1573 struct ext4_fc_tl *tl)
1575 struct ext4_fc_add_range *fc_add_ex;
1576 struct ext4_extent newex, *ex;
1577 struct inode *inode;
1578 ext4_lblk_t start, cur;
1580 ext4_fsblk_t start_pblk;
1581 struct ext4_map_blocks map;
1582 struct ext4_ext_path *path = NULL;
1585 fc_add_ex = (struct ext4_fc_add_range *)ext4_fc_tag_val(tl);
1586 ex = (struct ext4_extent *)&fc_add_ex->fc_ex;
1588 trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,
1589 le32_to_cpu(fc_add_ex->fc_ino), le32_to_cpu(ex->ee_block),
1590 ext4_ext_get_actual_len(ex));
1592 inode = ext4_iget(sb, le32_to_cpu(fc_add_ex->fc_ino),
1594 if (IS_ERR_OR_NULL(inode)) {
1595 jbd_debug(1, "Inode not found.");
1599 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1601 start = le32_to_cpu(ex->ee_block);
1602 start_pblk = ext4_ext_pblock(ex);
1603 len = ext4_ext_get_actual_len(ex);
1607 jbd_debug(1, "ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",
1608 start, start_pblk, len, ext4_ext_is_unwritten(ex),
1611 while (remaining > 0) {
1613 map.m_len = remaining;
1615 ret = ext4_map_blocks(NULL, inode, &map, 0);
1623 /* Range is not mapped */
1624 path = ext4_find_extent(inode, cur, NULL, 0);
1629 memset(&newex, 0, sizeof(newex));
1630 newex.ee_block = cpu_to_le32(cur);
1631 ext4_ext_store_pblock(
1632 &newex, start_pblk + cur - start);
1633 newex.ee_len = cpu_to_le16(map.m_len);
1634 if (ext4_ext_is_unwritten(ex))
1635 ext4_ext_mark_unwritten(&newex);
1636 down_write(&EXT4_I(inode)->i_data_sem);
1637 ret = ext4_ext_insert_extent(
1638 NULL, inode, &path, &newex, 0);
1639 up_write((&EXT4_I(inode)->i_data_sem));
1640 ext4_ext_drop_refs(path);
1649 if (start_pblk + cur - start != map.m_pblk) {
1651 * Logical to physical mapping changed. This can happen
1652 * if this range was removed and then reallocated to
1653 * map to new physical blocks during a fast commit.
1655 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1656 ext4_ext_is_unwritten(ex),
1657 start_pblk + cur - start);
1663 * Mark the old blocks as free since they aren't used
1664 * anymore. We maintain an array of all the modified
1665 * inodes. In case these blocks are still used at either
1666 * a different logical range in the same inode or in
1667 * some different inode, we will mark them as allocated
1668 * at the end of the FC replay using our array of
1671 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1675 /* Range is mapped and needs a state change */
1676 jbd_debug(1, "Converting from %d to %d %lld",
1677 map.m_flags & EXT4_MAP_UNWRITTEN,
1678 ext4_ext_is_unwritten(ex), map.m_pblk);
1679 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1680 ext4_ext_is_unwritten(ex), map.m_pblk);
1686 * We may have split the extent tree while toggling the state.
1687 * Try to shrink the extent tree now.
1689 ext4_ext_replay_shrink_inode(inode, start + len);
1692 remaining -= map.m_len;
1694 ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>
1695 sb->s_blocksize_bits);
1700 /* Replay DEL_RANGE tag */
1702 ext4_fc_replay_del_range(struct super_block *sb, struct ext4_fc_tl *tl)
1704 struct inode *inode;
1705 struct ext4_fc_del_range *lrange;
1706 struct ext4_map_blocks map;
1707 ext4_lblk_t cur, remaining;
1710 lrange = (struct ext4_fc_del_range *)ext4_fc_tag_val(tl);
1711 cur = le32_to_cpu(lrange->fc_lblk);
1712 remaining = le32_to_cpu(lrange->fc_len);
1714 trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,
1715 le32_to_cpu(lrange->fc_ino), cur, remaining);
1717 inode = ext4_iget(sb, le32_to_cpu(lrange->fc_ino), EXT4_IGET_NORMAL);
1718 if (IS_ERR_OR_NULL(inode)) {
1719 jbd_debug(1, "Inode %d not found", le32_to_cpu(lrange->fc_ino));
1723 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1725 jbd_debug(1, "DEL_RANGE, inode %ld, lblk %d, len %d\n",
1726 inode->i_ino, le32_to_cpu(lrange->fc_lblk),
1727 le32_to_cpu(lrange->fc_len));
1728 while (remaining > 0) {
1730 map.m_len = remaining;
1732 ret = ext4_map_blocks(NULL, inode, &map, 0);
1740 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1742 remaining -= map.m_len;
1747 ret = ext4_punch_hole(inode,
1748 le32_to_cpu(lrange->fc_lblk) << sb->s_blocksize_bits,
1749 le32_to_cpu(lrange->fc_len) << sb->s_blocksize_bits);
1751 jbd_debug(1, "ext4_punch_hole returned %d", ret);
1752 ext4_ext_replay_shrink_inode(inode,
1753 i_size_read(inode) >> sb->s_blocksize_bits);
1754 ext4_mark_inode_dirty(NULL, inode);
1760 static inline const char *tag2str(u16 tag)
1763 case EXT4_FC_TAG_LINK:
1764 return "TAG_ADD_ENTRY";
1765 case EXT4_FC_TAG_UNLINK:
1766 return "TAG_DEL_ENTRY";
1767 case EXT4_FC_TAG_ADD_RANGE:
1768 return "TAG_ADD_RANGE";
1769 case EXT4_FC_TAG_CREAT:
1770 return "TAG_CREAT_DENTRY";
1771 case EXT4_FC_TAG_DEL_RANGE:
1772 return "TAG_DEL_RANGE";
1773 case EXT4_FC_TAG_INODE:
1775 case EXT4_FC_TAG_PAD:
1777 case EXT4_FC_TAG_TAIL:
1779 case EXT4_FC_TAG_HEAD:
1786 static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)
1788 struct ext4_fc_replay_state *state;
1789 struct inode *inode;
1790 struct ext4_ext_path *path = NULL;
1791 struct ext4_map_blocks map;
1793 ext4_lblk_t cur, end;
1795 state = &EXT4_SB(sb)->s_fc_replay_state;
1796 for (i = 0; i < state->fc_modified_inodes_used; i++) {
1797 inode = ext4_iget(sb, state->fc_modified_inodes[i],
1799 if (IS_ERR_OR_NULL(inode)) {
1800 jbd_debug(1, "Inode %d not found.",
1801 state->fc_modified_inodes[i]);
1805 end = EXT_MAX_BLOCKS;
1808 map.m_len = end - cur;
1810 ret = ext4_map_blocks(NULL, inode, &map, 0);
1815 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
1816 if (!IS_ERR_OR_NULL(path)) {
1817 for (j = 0; j < path->p_depth; j++)
1818 ext4_mb_mark_bb(inode->i_sb,
1819 path[j].p_block, 1, 1);
1820 ext4_ext_drop_refs(path);
1824 ext4_mb_mark_bb(inode->i_sb, map.m_pblk,
1827 cur = cur + (map.m_len ? map.m_len : 1);
1835 * Check if block is in excluded regions for block allocation. The simple
1836 * allocator that runs during replay phase is calls this function to see
1837 * if it is okay to use a block.
1839 bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)
1842 struct ext4_fc_replay_state *state;
1844 state = &EXT4_SB(sb)->s_fc_replay_state;
1845 for (i = 0; i < state->fc_regions_valid; i++) {
1846 if (state->fc_regions[i].ino == 0 ||
1847 state->fc_regions[i].len == 0)
1849 if (blk >= state->fc_regions[i].pblk &&
1850 blk < state->fc_regions[i].pblk + state->fc_regions[i].len)
1856 /* Cleanup function called after replay */
1857 void ext4_fc_replay_cleanup(struct super_block *sb)
1859 struct ext4_sb_info *sbi = EXT4_SB(sb);
1861 sbi->s_mount_state &= ~EXT4_FC_REPLAY;
1862 kfree(sbi->s_fc_replay_state.fc_regions);
1863 kfree(sbi->s_fc_replay_state.fc_modified_inodes);
1867 * Recovery Scan phase handler
1869 * This function is called during the scan phase and is responsible
1870 * for doing following things:
1871 * - Make sure the fast commit area has valid tags for replay
1872 * - Count number of tags that need to be replayed by the replay handler
1874 * - Create a list of excluded blocks for allocation during replay phase
1876 * This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is
1877 * incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP
1878 * to indicate that scan has finished and JBD2 can now start replay phase.
1879 * It returns a negative error to indicate that there was an error. At the end
1880 * of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set
1881 * to indicate the number of tags that need to replayed during the replay phase.
1883 static int ext4_fc_replay_scan(journal_t *journal,
1884 struct buffer_head *bh, int off,
1887 struct super_block *sb = journal->j_private;
1888 struct ext4_sb_info *sbi = EXT4_SB(sb);
1889 struct ext4_fc_replay_state *state;
1890 int ret = JBD2_FC_REPLAY_CONTINUE;
1891 struct ext4_fc_add_range *ext;
1892 struct ext4_fc_tl *tl;
1893 struct ext4_fc_tail *tail;
1895 struct ext4_fc_head *head;
1896 struct ext4_extent *ex;
1898 state = &sbi->s_fc_replay_state;
1900 start = (u8 *)bh->b_data;
1901 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
1903 if (state->fc_replay_expected_off == 0) {
1904 state->fc_cur_tag = 0;
1905 state->fc_replay_num_tags = 0;
1907 state->fc_regions = NULL;
1908 state->fc_regions_valid = state->fc_regions_used =
1909 state->fc_regions_size = 0;
1910 /* Check if we can stop early */
1911 if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
1912 != EXT4_FC_TAG_HEAD)
1916 if (off != state->fc_replay_expected_off) {
1917 ret = -EFSCORRUPTED;
1921 state->fc_replay_expected_off++;
1922 fc_for_each_tl(start, end, tl) {
1923 jbd_debug(3, "Scan phase, tag:%s, blk %lld\n",
1924 tag2str(le16_to_cpu(tl->fc_tag)), bh->b_blocknr);
1925 switch (le16_to_cpu(tl->fc_tag)) {
1926 case EXT4_FC_TAG_ADD_RANGE:
1927 ext = (struct ext4_fc_add_range *)ext4_fc_tag_val(tl);
1928 ex = (struct ext4_extent *)&ext->fc_ex;
1929 ret = ext4_fc_record_regions(sb,
1930 le32_to_cpu(ext->fc_ino),
1931 le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
1932 ext4_ext_get_actual_len(ex));
1935 ret = JBD2_FC_REPLAY_CONTINUE;
1937 case EXT4_FC_TAG_DEL_RANGE:
1938 case EXT4_FC_TAG_LINK:
1939 case EXT4_FC_TAG_UNLINK:
1940 case EXT4_FC_TAG_CREAT:
1941 case EXT4_FC_TAG_INODE:
1942 case EXT4_FC_TAG_PAD:
1943 state->fc_cur_tag++;
1944 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1945 sizeof(*tl) + ext4_fc_tag_len(tl));
1947 case EXT4_FC_TAG_TAIL:
1948 state->fc_cur_tag++;
1949 tail = (struct ext4_fc_tail *)ext4_fc_tag_val(tl);
1950 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1952 offsetof(struct ext4_fc_tail,
1954 if (le32_to_cpu(tail->fc_tid) == expected_tid &&
1955 le32_to_cpu(tail->fc_crc) == state->fc_crc) {
1956 state->fc_replay_num_tags = state->fc_cur_tag;
1957 state->fc_regions_valid =
1958 state->fc_regions_used;
1960 ret = state->fc_replay_num_tags ?
1961 JBD2_FC_REPLAY_STOP : -EFSBADCRC;
1965 case EXT4_FC_TAG_HEAD:
1966 head = (struct ext4_fc_head *)ext4_fc_tag_val(tl);
1967 if (le32_to_cpu(head->fc_features) &
1968 ~EXT4_FC_SUPPORTED_FEATURES) {
1972 if (le32_to_cpu(head->fc_tid) != expected_tid) {
1973 ret = JBD2_FC_REPLAY_STOP;
1976 state->fc_cur_tag++;
1977 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1978 sizeof(*tl) + ext4_fc_tag_len(tl));
1981 ret = state->fc_replay_num_tags ?
1982 JBD2_FC_REPLAY_STOP : -ECANCELED;
1984 if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
1989 trace_ext4_fc_replay_scan(sb, ret, off);
1994 * Main recovery path entry point.
1995 * The meaning of return codes is similar as above.
1997 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
1998 enum passtype pass, int off, tid_t expected_tid)
2000 struct super_block *sb = journal->j_private;
2001 struct ext4_sb_info *sbi = EXT4_SB(sb);
2002 struct ext4_fc_tl *tl;
2004 int ret = JBD2_FC_REPLAY_CONTINUE;
2005 struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;
2006 struct ext4_fc_tail *tail;
2008 if (pass == PASS_SCAN) {
2009 state->fc_current_pass = PASS_SCAN;
2010 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
2013 if (state->fc_current_pass != pass) {
2014 state->fc_current_pass = pass;
2015 sbi->s_mount_state |= EXT4_FC_REPLAY;
2017 if (!sbi->s_fc_replay_state.fc_replay_num_tags) {
2018 jbd_debug(1, "Replay stops\n");
2019 ext4_fc_set_bitmaps_and_counters(sb);
2023 #ifdef CONFIG_EXT4_DEBUG
2024 if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {
2025 pr_warn("Dropping fc block %d because max_replay set\n", off);
2026 return JBD2_FC_REPLAY_STOP;
2030 start = (u8 *)bh->b_data;
2031 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
2033 fc_for_each_tl(start, end, tl) {
2034 if (state->fc_replay_num_tags == 0) {
2035 ret = JBD2_FC_REPLAY_STOP;
2036 ext4_fc_set_bitmaps_and_counters(sb);
2039 jbd_debug(3, "Replay phase, tag:%s\n",
2040 tag2str(le16_to_cpu(tl->fc_tag)));
2041 state->fc_replay_num_tags--;
2042 switch (le16_to_cpu(tl->fc_tag)) {
2043 case EXT4_FC_TAG_LINK:
2044 ret = ext4_fc_replay_link(sb, tl);
2046 case EXT4_FC_TAG_UNLINK:
2047 ret = ext4_fc_replay_unlink(sb, tl);
2049 case EXT4_FC_TAG_ADD_RANGE:
2050 ret = ext4_fc_replay_add_range(sb, tl);
2052 case EXT4_FC_TAG_CREAT:
2053 ret = ext4_fc_replay_create(sb, tl);
2055 case EXT4_FC_TAG_DEL_RANGE:
2056 ret = ext4_fc_replay_del_range(sb, tl);
2058 case EXT4_FC_TAG_INODE:
2059 ret = ext4_fc_replay_inode(sb, tl);
2061 case EXT4_FC_TAG_PAD:
2062 trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,
2063 ext4_fc_tag_len(tl), 0);
2065 case EXT4_FC_TAG_TAIL:
2066 trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL, 0,
2067 ext4_fc_tag_len(tl), 0);
2068 tail = (struct ext4_fc_tail *)ext4_fc_tag_val(tl);
2069 WARN_ON(le32_to_cpu(tail->fc_tid) != expected_tid);
2071 case EXT4_FC_TAG_HEAD:
2074 trace_ext4_fc_replay(sb, le16_to_cpu(tl->fc_tag), 0,
2075 ext4_fc_tag_len(tl), 0);
2081 ret = JBD2_FC_REPLAY_CONTINUE;
2086 void ext4_fc_init(struct super_block *sb, journal_t *journal)
2089 * We set replay callback even if fast commit disabled because we may
2090 * could still have fast commit blocks that need to be replayed even if
2091 * fast commit has now been turned off.
2093 journal->j_fc_replay_callback = ext4_fc_replay;
2094 if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
2096 journal->j_fc_cleanup_callback = ext4_fc_cleanup;
2099 const char *fc_ineligible_reasons[] = {
2100 "Extended attributes changed",
2102 "Journal flag changed",
2103 "Insufficient memory",
2111 int ext4_fc_info_show(struct seq_file *seq, void *v)
2113 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);
2114 struct ext4_fc_stats *stats = &sbi->s_fc_stats;
2117 if (v != SEQ_START_TOKEN)
2121 "fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",
2122 stats->fc_num_commits, stats->fc_ineligible_commits,
2124 div_u64(sbi->s_fc_avg_commit_time, 1000));
2125 seq_puts(seq, "Ineligible reasons:\n");
2126 for (i = 0; i < EXT4_FC_REASON_MAX; i++)
2127 seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],
2128 stats->fc_ineligible_reason_count[i]);
2133 int __init ext4_fc_init_dentry_cache(void)
2135 ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,
2136 SLAB_RECLAIM_ACCOUNT);
2138 if (ext4_fc_dentry_cachep == NULL)