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)
160 wait_queue_head_t *wq;
161 struct ext4_inode_info *ei = EXT4_I(inode);
163 #if (BITS_PER_LONG < 64)
164 DEFINE_WAIT_BIT(wait, &ei->i_state_flags,
165 EXT4_STATE_FC_COMMITTING);
166 wq = bit_waitqueue(&ei->i_state_flags,
167 EXT4_STATE_FC_COMMITTING);
169 DEFINE_WAIT_BIT(wait, &ei->i_flags,
170 EXT4_STATE_FC_COMMITTING);
171 wq = bit_waitqueue(&ei->i_flags,
172 EXT4_STATE_FC_COMMITTING);
174 lockdep_assert_held(&EXT4_SB(inode->i_sb)->s_fc_lock);
175 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
176 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
178 finish_wait(wq, &wait.wq_entry);
182 * Inform Ext4's fast about start of an inode update
184 * This function is called by the high level call VFS callbacks before
185 * performing any inode update. This function blocks if there's an ongoing
186 * fast commit on the inode in question.
188 void ext4_fc_start_update(struct inode *inode)
190 struct ext4_inode_info *ei = EXT4_I(inode);
192 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
193 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
197 spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
198 if (list_empty(&ei->i_fc_list))
201 if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
202 ext4_fc_wait_committing_inode(inode);
206 atomic_inc(&ei->i_fc_updates);
207 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
211 * Stop inode update and wake up waiting fast commits if any.
213 void ext4_fc_stop_update(struct inode *inode)
215 struct ext4_inode_info *ei = EXT4_I(inode);
217 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
218 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
221 if (atomic_dec_and_test(&ei->i_fc_updates))
222 wake_up_all(&ei->i_fc_wait);
226 * Remove inode from fast commit list. If the inode is being committed
227 * we wait until inode commit is done.
229 void ext4_fc_del(struct inode *inode)
231 struct ext4_inode_info *ei = EXT4_I(inode);
233 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
234 (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY))
238 spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
239 if (list_empty(&ei->i_fc_list)) {
240 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
244 if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
245 ext4_fc_wait_committing_inode(inode);
248 list_del_init(&ei->i_fc_list);
249 spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
253 * Mark file system as fast commit ineligible. This means that next commit
254 * operation would result in a full jbd2 commit.
256 void ext4_fc_mark_ineligible(struct super_block *sb, int reason)
258 struct ext4_sb_info *sbi = EXT4_SB(sb);
260 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
261 (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
264 sbi->s_mount_flags |= EXT4_MF_FC_INELIGIBLE;
265 WARN_ON(reason >= EXT4_FC_REASON_MAX);
266 sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;
270 * Start a fast commit ineligible update. Any commits that happen while
271 * such an operation is in progress fall back to full commits.
273 void ext4_fc_start_ineligible(struct super_block *sb, int reason)
275 struct ext4_sb_info *sbi = EXT4_SB(sb);
277 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
278 (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
281 WARN_ON(reason >= EXT4_FC_REASON_MAX);
282 sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;
283 atomic_inc(&sbi->s_fc_ineligible_updates);
287 * Stop a fast commit ineligible update. We set EXT4_MF_FC_INELIGIBLE flag here
288 * to ensure that after stopping the ineligible update, at least one full
289 * commit takes place.
291 void ext4_fc_stop_ineligible(struct super_block *sb)
293 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
294 (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
297 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FC_INELIGIBLE;
298 atomic_dec(&EXT4_SB(sb)->s_fc_ineligible_updates);
301 static inline int ext4_fc_is_ineligible(struct super_block *sb)
303 return (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FC_INELIGIBLE) ||
304 atomic_read(&EXT4_SB(sb)->s_fc_ineligible_updates);
308 * Generic fast commit tracking function. If this is the first time this we are
309 * called after a full commit, we initialize fast commit fields and then call
310 * __fc_track_fn() with update = 0. If we have already been called after a full
311 * commit, we pass update = 1. Based on that, the track function can determine
312 * if it needs to track a field for the first time or if it needs to just
313 * update the previously tracked value.
315 * If enqueue is set, this function enqueues the inode in fast commit list.
317 static int ext4_fc_track_template(
318 handle_t *handle, struct inode *inode,
319 int (*__fc_track_fn)(struct inode *, void *, bool),
320 void *args, int enqueue)
323 struct ext4_inode_info *ei = EXT4_I(inode);
324 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
328 if (!test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT) ||
329 (sbi->s_mount_state & EXT4_FC_REPLAY))
332 if (ext4_fc_is_ineligible(inode->i_sb))
335 tid = handle->h_transaction->t_tid;
336 mutex_lock(&ei->i_fc_lock);
337 if (tid == ei->i_sync_tid) {
340 ext4_fc_reset_inode(inode);
341 ei->i_sync_tid = tid;
343 ret = __fc_track_fn(inode, args, update);
344 mutex_unlock(&ei->i_fc_lock);
349 spin_lock(&sbi->s_fc_lock);
350 if (list_empty(&EXT4_I(inode)->i_fc_list))
351 list_add_tail(&EXT4_I(inode)->i_fc_list,
352 (sbi->s_mount_flags & EXT4_MF_FC_COMMITTING) ?
353 &sbi->s_fc_q[FC_Q_STAGING] :
354 &sbi->s_fc_q[FC_Q_MAIN]);
355 spin_unlock(&sbi->s_fc_lock);
360 struct __track_dentry_update_args {
361 struct dentry *dentry;
365 /* __track_fn for directory entry updates. Called with ei->i_fc_lock. */
366 static int __track_dentry_update(struct inode *inode, void *arg, bool update)
368 struct ext4_fc_dentry_update *node;
369 struct ext4_inode_info *ei = EXT4_I(inode);
370 struct __track_dentry_update_args *dentry_update =
371 (struct __track_dentry_update_args *)arg;
372 struct dentry *dentry = dentry_update->dentry;
373 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
375 mutex_unlock(&ei->i_fc_lock);
376 node = kmem_cache_alloc(ext4_fc_dentry_cachep, GFP_NOFS);
378 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM);
379 mutex_lock(&ei->i_fc_lock);
383 node->fcd_op = dentry_update->op;
384 node->fcd_parent = dentry->d_parent->d_inode->i_ino;
385 node->fcd_ino = inode->i_ino;
386 if (dentry->d_name.len > DNAME_INLINE_LEN) {
387 node->fcd_name.name = kmalloc(dentry->d_name.len, GFP_NOFS);
388 if (!node->fcd_name.name) {
389 kmem_cache_free(ext4_fc_dentry_cachep, node);
390 ext4_fc_mark_ineligible(inode->i_sb,
391 EXT4_FC_REASON_NOMEM);
392 mutex_lock(&ei->i_fc_lock);
395 memcpy((u8 *)node->fcd_name.name, dentry->d_name.name,
398 memcpy(node->fcd_iname, dentry->d_name.name,
400 node->fcd_name.name = node->fcd_iname;
402 node->fcd_name.len = dentry->d_name.len;
404 spin_lock(&sbi->s_fc_lock);
405 if (sbi->s_mount_flags & EXT4_MF_FC_COMMITTING)
406 list_add_tail(&node->fcd_list,
407 &sbi->s_fc_dentry_q[FC_Q_STAGING]);
409 list_add_tail(&node->fcd_list, &sbi->s_fc_dentry_q[FC_Q_MAIN]);
410 spin_unlock(&sbi->s_fc_lock);
411 mutex_lock(&ei->i_fc_lock);
416 void __ext4_fc_track_unlink(handle_t *handle,
417 struct inode *inode, struct dentry *dentry)
419 struct __track_dentry_update_args args;
422 args.dentry = dentry;
423 args.op = EXT4_FC_TAG_UNLINK;
425 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
427 trace_ext4_fc_track_unlink(inode, dentry, ret);
430 void ext4_fc_track_unlink(handle_t *handle, struct dentry *dentry)
432 __ext4_fc_track_unlink(handle, d_inode(dentry), dentry);
435 void __ext4_fc_track_link(handle_t *handle,
436 struct inode *inode, struct dentry *dentry)
438 struct __track_dentry_update_args args;
441 args.dentry = dentry;
442 args.op = EXT4_FC_TAG_LINK;
444 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
446 trace_ext4_fc_track_link(inode, dentry, ret);
449 void ext4_fc_track_link(handle_t *handle, struct dentry *dentry)
451 __ext4_fc_track_link(handle, d_inode(dentry), dentry);
454 void ext4_fc_track_create(handle_t *handle, struct dentry *dentry)
456 struct __track_dentry_update_args args;
457 struct inode *inode = d_inode(dentry);
460 args.dentry = dentry;
461 args.op = EXT4_FC_TAG_CREAT;
463 ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
465 trace_ext4_fc_track_create(inode, dentry, ret);
468 /* __track_fn for inode tracking */
469 static int __track_inode(struct inode *inode, void *arg, bool update)
474 EXT4_I(inode)->i_fc_lblk_len = 0;
479 void ext4_fc_track_inode(handle_t *handle, struct inode *inode)
483 if (S_ISDIR(inode->i_mode))
486 if (ext4_should_journal_data(inode)) {
487 ext4_fc_mark_ineligible(inode->i_sb,
488 EXT4_FC_REASON_INODE_JOURNAL_DATA);
492 ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);
493 trace_ext4_fc_track_inode(inode, ret);
496 struct __track_range_args {
497 ext4_lblk_t start, end;
500 /* __track_fn for tracking data updates */
501 static int __track_range(struct inode *inode, void *arg, bool update)
503 struct ext4_inode_info *ei = EXT4_I(inode);
504 ext4_lblk_t oldstart;
505 struct __track_range_args *__arg =
506 (struct __track_range_args *)arg;
508 if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {
509 ext4_debug("Special inode %ld being modified\n", inode->i_ino);
513 oldstart = ei->i_fc_lblk_start;
515 if (update && ei->i_fc_lblk_len > 0) {
516 ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);
518 max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -
519 ei->i_fc_lblk_start + 1;
521 ei->i_fc_lblk_start = __arg->start;
522 ei->i_fc_lblk_len = __arg->end - __arg->start + 1;
528 void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,
531 struct __track_range_args args;
534 if (S_ISDIR(inode->i_mode))
540 ret = ext4_fc_track_template(handle, inode, __track_range, &args, 1);
542 trace_ext4_fc_track_range(inode, start, end, ret);
545 static void ext4_fc_submit_bh(struct super_block *sb)
547 int write_flags = REQ_SYNC;
548 struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
550 /* TODO: REQ_FUA | REQ_PREFLUSH is unnecessarily expensive. */
551 if (test_opt(sb, BARRIER))
552 write_flags |= REQ_FUA | REQ_PREFLUSH;
554 set_buffer_dirty(bh);
555 set_buffer_uptodate(bh);
556 bh->b_end_io = ext4_end_buffer_io_sync;
557 submit_bh(REQ_OP_WRITE, write_flags, bh);
558 EXT4_SB(sb)->s_fc_bh = NULL;
561 /* Ext4 commit path routines */
563 /* memzero and update CRC */
564 static void *ext4_fc_memzero(struct super_block *sb, void *dst, int len,
569 ret = memset(dst, 0, len);
571 *crc = ext4_chksum(EXT4_SB(sb), *crc, dst, len);
576 * Allocate len bytes on a fast commit buffer.
578 * During the commit time this function is used to manage fast commit
579 * block space. We don't split a fast commit log onto different
580 * blocks. So this function makes sure that if there's not enough space
581 * on the current block, the remaining space in the current block is
582 * marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,
583 * new block is from jbd2 and CRC is updated to reflect the padding
586 static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
588 struct ext4_fc_tl *tl;
589 struct ext4_sb_info *sbi = EXT4_SB(sb);
590 struct buffer_head *bh;
591 int bsize = sbi->s_journal->j_blocksize;
592 int ret, off = sbi->s_fc_bytes % bsize;
596 * After allocating len, we should have space at least for a 0 byte
599 if (len + sizeof(struct ext4_fc_tl) > bsize)
602 if (bsize - off - 1 > len + sizeof(struct ext4_fc_tl)) {
604 * Only allocate from current buffer if we have enough space for
605 * this request AND we have space to add a zero byte padding.
608 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
613 sbi->s_fc_bytes += len;
614 return sbi->s_fc_bh->b_data + off;
616 /* Need to add PAD tag */
617 tl = (struct ext4_fc_tl *)(sbi->s_fc_bh->b_data + off);
618 tl->fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);
619 pad_len = bsize - off - 1 - sizeof(struct ext4_fc_tl);
620 tl->fc_len = cpu_to_le16(pad_len);
622 *crc = ext4_chksum(sbi, *crc, tl, sizeof(*tl));
624 ext4_fc_memzero(sb, tl + 1, pad_len, crc);
625 ext4_fc_submit_bh(sb);
627 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
631 sbi->s_fc_bytes = (sbi->s_fc_bytes / bsize + 1) * bsize + len;
632 return sbi->s_fc_bh->b_data;
635 /* memcpy to fc reserved space and update CRC */
636 static void *ext4_fc_memcpy(struct super_block *sb, void *dst, const void *src,
640 *crc = ext4_chksum(EXT4_SB(sb), *crc, src, len);
641 return memcpy(dst, src, len);
645 * Complete a fast commit by writing tail tag.
647 * Writing tail tag marks the end of a fast commit. In order to guarantee
648 * atomicity, after writing tail tag, even if there's space remaining
649 * in the block, next commit shouldn't use it. That's why tail tag
650 * has the length as that of the remaining space on the block.
652 static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
654 struct ext4_sb_info *sbi = EXT4_SB(sb);
655 struct ext4_fc_tl tl;
656 struct ext4_fc_tail tail;
657 int off, bsize = sbi->s_journal->j_blocksize;
661 * ext4_fc_reserve_space takes care of allocating an extra block if
662 * there's no enough space on this block for accommodating this tail.
664 dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(tail), &crc);
668 off = sbi->s_fc_bytes % bsize;
670 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);
671 tl.fc_len = cpu_to_le16(bsize - off - 1 + sizeof(struct ext4_fc_tail));
672 sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);
674 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), &crc);
676 tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);
677 ext4_fc_memcpy(sb, dst, &tail.fc_tid, sizeof(tail.fc_tid), &crc);
678 dst += sizeof(tail.fc_tid);
679 tail.fc_crc = cpu_to_le32(crc);
680 ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
682 ext4_fc_submit_bh(sb);
688 * Adds tag, length, value and updates CRC. Returns true if tlv was added.
689 * Returns false if there's not enough space.
691 static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,
694 struct ext4_fc_tl tl;
697 dst = ext4_fc_reserve_space(sb, sizeof(tl) + len, crc);
701 tl.fc_tag = cpu_to_le16(tag);
702 tl.fc_len = cpu_to_le16(len);
704 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
705 ext4_fc_memcpy(sb, dst + sizeof(tl), val, len, crc);
710 /* Same as above, but adds dentry tlv. */
711 static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u16 tag,
712 int parent_ino, int ino, int dlen,
713 const unsigned char *dname,
716 struct ext4_fc_dentry_info fcd;
717 struct ext4_fc_tl tl;
718 u8 *dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(fcd) + dlen,
724 fcd.fc_parent_ino = cpu_to_le32(parent_ino);
725 fcd.fc_ino = cpu_to_le32(ino);
726 tl.fc_tag = cpu_to_le16(tag);
727 tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);
728 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
730 ext4_fc_memcpy(sb, dst, &fcd, sizeof(fcd), crc);
732 ext4_fc_memcpy(sb, dst, dname, dlen, crc);
739 * Writes inode in the fast commit space under TLV with tag @tag.
740 * Returns 0 on success, error on failure.
742 static int ext4_fc_write_inode(struct inode *inode, u32 *crc)
744 struct ext4_inode_info *ei = EXT4_I(inode);
745 int inode_len = EXT4_GOOD_OLD_INODE_SIZE;
747 struct ext4_iloc iloc;
748 struct ext4_fc_inode fc_inode;
749 struct ext4_fc_tl tl;
752 ret = ext4_get_inode_loc(inode, &iloc);
756 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)
757 inode_len += ei->i_extra_isize;
759 fc_inode.fc_ino = cpu_to_le32(inode->i_ino);
760 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);
761 tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));
763 dst = ext4_fc_reserve_space(inode->i_sb,
764 sizeof(tl) + inode_len + sizeof(fc_inode.fc_ino), crc);
768 if (!ext4_fc_memcpy(inode->i_sb, dst, &tl, sizeof(tl), crc))
771 if (!ext4_fc_memcpy(inode->i_sb, dst, &fc_inode, sizeof(fc_inode), crc))
773 dst += sizeof(fc_inode);
774 if (!ext4_fc_memcpy(inode->i_sb, dst, (u8 *)ext4_raw_inode(&iloc),
782 * Writes updated data ranges for the inode in question. Updates CRC.
783 * Returns 0 on success, error otherwise.
785 static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)
787 ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;
788 struct ext4_inode_info *ei = EXT4_I(inode);
789 struct ext4_map_blocks map;
790 struct ext4_fc_add_range fc_ext;
791 struct ext4_fc_del_range lrange;
792 struct ext4_extent *ex;
795 mutex_lock(&ei->i_fc_lock);
796 if (ei->i_fc_lblk_len == 0) {
797 mutex_unlock(&ei->i_fc_lock);
800 old_blk_size = ei->i_fc_lblk_start;
801 new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;
802 ei->i_fc_lblk_len = 0;
803 mutex_unlock(&ei->i_fc_lock);
805 cur_lblk_off = old_blk_size;
806 jbd_debug(1, "%s: will try writing %d to %d for inode %ld\n",
807 __func__, cur_lblk_off, new_blk_size, inode->i_ino);
809 while (cur_lblk_off <= new_blk_size) {
810 map.m_lblk = cur_lblk_off;
811 map.m_len = new_blk_size - cur_lblk_off + 1;
812 ret = ext4_map_blocks(NULL, inode, &map, 0);
816 if (map.m_len == 0) {
822 lrange.fc_ino = cpu_to_le32(inode->i_ino);
823 lrange.fc_lblk = cpu_to_le32(map.m_lblk);
824 lrange.fc_len = cpu_to_le32(map.m_len);
825 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,
826 sizeof(lrange), (u8 *)&lrange, crc))
829 fc_ext.fc_ino = cpu_to_le32(inode->i_ino);
830 ex = (struct ext4_extent *)&fc_ext.fc_ex;
831 ex->ee_block = cpu_to_le32(map.m_lblk);
832 ex->ee_len = cpu_to_le16(map.m_len);
833 ext4_ext_store_pblock(ex, map.m_pblk);
834 if (map.m_flags & EXT4_MAP_UNWRITTEN)
835 ext4_ext_mark_unwritten(ex);
837 ext4_ext_mark_initialized(ex);
838 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,
839 sizeof(fc_ext), (u8 *)&fc_ext, crc))
843 cur_lblk_off += map.m_len;
850 /* Submit data for all the fast commit inodes */
851 static int ext4_fc_submit_inode_data_all(journal_t *journal)
853 struct super_block *sb = (struct super_block *)(journal->j_private);
854 struct ext4_sb_info *sbi = EXT4_SB(sb);
855 struct ext4_inode_info *ei;
856 struct list_head *pos;
859 spin_lock(&sbi->s_fc_lock);
860 sbi->s_mount_flags |= EXT4_MF_FC_COMMITTING;
861 list_for_each(pos, &sbi->s_fc_q[FC_Q_MAIN]) {
862 ei = list_entry(pos, struct ext4_inode_info, i_fc_list);
863 ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
864 while (atomic_read(&ei->i_fc_updates)) {
867 prepare_to_wait(&ei->i_fc_wait, &wait,
868 TASK_UNINTERRUPTIBLE);
869 if (atomic_read(&ei->i_fc_updates)) {
870 spin_unlock(&sbi->s_fc_lock);
872 spin_lock(&sbi->s_fc_lock);
874 finish_wait(&ei->i_fc_wait, &wait);
876 spin_unlock(&sbi->s_fc_lock);
877 ret = jbd2_submit_inode_data(ei->jinode);
880 spin_lock(&sbi->s_fc_lock);
882 spin_unlock(&sbi->s_fc_lock);
887 /* Wait for completion of data for all the fast commit inodes */
888 static int ext4_fc_wait_inode_data_all(journal_t *journal)
890 struct super_block *sb = (struct super_block *)(journal->j_private);
891 struct ext4_sb_info *sbi = EXT4_SB(sb);
892 struct ext4_inode_info *pos, *n;
895 spin_lock(&sbi->s_fc_lock);
896 list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
897 if (!ext4_test_inode_state(&pos->vfs_inode,
898 EXT4_STATE_FC_COMMITTING))
900 spin_unlock(&sbi->s_fc_lock);
902 ret = jbd2_wait_inode_data(journal, pos->jinode);
905 spin_lock(&sbi->s_fc_lock);
907 spin_unlock(&sbi->s_fc_lock);
912 /* Commit all the directory entry updates */
913 static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)
915 struct super_block *sb = (struct super_block *)(journal->j_private);
916 struct ext4_sb_info *sbi = EXT4_SB(sb);
917 struct ext4_fc_dentry_update *fc_dentry;
919 struct list_head *pos, *n, *fcd_pos, *fcd_n;
920 struct ext4_inode_info *ei;
923 if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))
925 list_for_each_safe(fcd_pos, fcd_n, &sbi->s_fc_dentry_q[FC_Q_MAIN]) {
926 fc_dentry = list_entry(fcd_pos, struct ext4_fc_dentry_update,
928 if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {
929 spin_unlock(&sbi->s_fc_lock);
930 if (!ext4_fc_add_dentry_tlv(
931 sb, fc_dentry->fcd_op,
932 fc_dentry->fcd_parent, fc_dentry->fcd_ino,
933 fc_dentry->fcd_name.len,
934 fc_dentry->fcd_name.name, crc)) {
938 spin_lock(&sbi->s_fc_lock);
943 list_for_each_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN]) {
944 ei = list_entry(pos, struct ext4_inode_info, i_fc_list);
945 if (ei->vfs_inode.i_ino == fc_dentry->fcd_ino) {
946 inode = &ei->vfs_inode;
951 * If we don't find inode in our list, then it was deleted,
952 * in which case, we don't need to record it's create tag.
956 spin_unlock(&sbi->s_fc_lock);
959 * We first write the inode and then the create dirent. This
960 * allows the recovery code to create an unnamed inode first
961 * and then link it to a directory entry. This allows us
962 * to use namei.c routines almost as is and simplifies
965 ret = ext4_fc_write_inode(inode, crc);
969 ret = ext4_fc_write_inode_data(inode, crc);
973 if (!ext4_fc_add_dentry_tlv(
974 sb, fc_dentry->fcd_op,
975 fc_dentry->fcd_parent, fc_dentry->fcd_ino,
976 fc_dentry->fcd_name.len,
977 fc_dentry->fcd_name.name, crc)) {
982 spin_lock(&sbi->s_fc_lock);
986 spin_lock(&sbi->s_fc_lock);
990 static int ext4_fc_perform_commit(journal_t *journal)
992 struct super_block *sb = (struct super_block *)(journal->j_private);
993 struct ext4_sb_info *sbi = EXT4_SB(sb);
994 struct ext4_inode_info *iter;
995 struct ext4_fc_head head;
996 struct list_head *pos;
998 struct blk_plug plug;
1002 ret = ext4_fc_submit_inode_data_all(journal);
1006 ret = ext4_fc_wait_inode_data_all(journal);
1010 blk_start_plug(&plug);
1011 if (sbi->s_fc_bytes == 0) {
1013 * Add a head tag only if this is the first fast commit
1016 head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);
1017 head.fc_tid = cpu_to_le32(
1018 sbi->s_journal->j_running_transaction->t_tid);
1019 if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),
1024 spin_lock(&sbi->s_fc_lock);
1025 ret = ext4_fc_commit_dentry_updates(journal, &crc);
1027 spin_unlock(&sbi->s_fc_lock);
1031 list_for_each(pos, &sbi->s_fc_q[FC_Q_MAIN]) {
1032 iter = list_entry(pos, struct ext4_inode_info, i_fc_list);
1033 inode = &iter->vfs_inode;
1034 if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))
1037 spin_unlock(&sbi->s_fc_lock);
1038 ret = ext4_fc_write_inode_data(inode, &crc);
1041 ret = ext4_fc_write_inode(inode, &crc);
1044 spin_lock(&sbi->s_fc_lock);
1046 spin_unlock(&sbi->s_fc_lock);
1048 ret = ext4_fc_write_tail(sb, crc);
1051 blk_finish_plug(&plug);
1056 * The main commit entry point. Performs a fast commit for transaction
1057 * commit_tid if needed. If it's not possible to perform a fast commit
1058 * due to various reasons, we fall back to full commit. Returns 0
1059 * on success, error otherwise.
1061 int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
1063 struct super_block *sb = (struct super_block *)(journal->j_private);
1064 struct ext4_sb_info *sbi = EXT4_SB(sb);
1065 int nblks = 0, ret, bsize = journal->j_blocksize;
1066 int subtid = atomic_read(&sbi->s_fc_subtid);
1067 int reason = EXT4_FC_REASON_OK, fc_bufs_before = 0;
1068 ktime_t start_time, commit_time;
1070 trace_ext4_fc_commit_start(sb);
1072 start_time = ktime_get();
1074 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
1075 (ext4_fc_is_ineligible(sb))) {
1076 reason = EXT4_FC_REASON_INELIGIBLE;
1081 ret = jbd2_fc_begin_commit(journal, commit_tid);
1082 if (ret == -EALREADY) {
1083 /* There was an ongoing commit, check if we need to restart */
1084 if (atomic_read(&sbi->s_fc_subtid) <= subtid &&
1085 commit_tid > journal->j_commit_sequence)
1087 reason = EXT4_FC_REASON_ALREADY_COMMITTED;
1090 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1091 reason = EXT4_FC_REASON_FC_START_FAILED;
1095 fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
1096 ret = ext4_fc_perform_commit(journal);
1098 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1099 reason = EXT4_FC_REASON_FC_FAILED;
1102 nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;
1103 ret = jbd2_fc_wait_bufs(journal, nblks);
1105 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1106 reason = EXT4_FC_REASON_FC_FAILED;
1109 atomic_inc(&sbi->s_fc_subtid);
1110 jbd2_fc_end_commit(journal);
1112 /* Has any ineligible update happened since we started? */
1113 if (reason == EXT4_FC_REASON_OK && ext4_fc_is_ineligible(sb)) {
1114 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1115 reason = EXT4_FC_REASON_INELIGIBLE;
1118 spin_lock(&sbi->s_fc_lock);
1119 if (reason != EXT4_FC_REASON_OK &&
1120 reason != EXT4_FC_REASON_ALREADY_COMMITTED) {
1121 sbi->s_fc_stats.fc_ineligible_commits++;
1123 sbi->s_fc_stats.fc_num_commits++;
1124 sbi->s_fc_stats.fc_numblks += nblks;
1126 spin_unlock(&sbi->s_fc_lock);
1127 nblks = (reason == EXT4_FC_REASON_OK) ? nblks : 0;
1128 trace_ext4_fc_commit_stop(sb, nblks, reason);
1129 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1131 * weight the commit time higher than the average time so we don't
1132 * react too strongly to vast changes in the commit time
1134 if (likely(sbi->s_fc_avg_commit_time))
1135 sbi->s_fc_avg_commit_time = (commit_time +
1136 sbi->s_fc_avg_commit_time * 3) / 4;
1138 sbi->s_fc_avg_commit_time = commit_time;
1140 "Fast commit ended with blks = %d, reason = %d, subtid - %d",
1141 nblks, reason, subtid);
1142 if (reason == EXT4_FC_REASON_FC_FAILED)
1143 return jbd2_fc_end_commit_fallback(journal);
1144 if (reason == EXT4_FC_REASON_FC_START_FAILED ||
1145 reason == EXT4_FC_REASON_INELIGIBLE)
1146 return jbd2_complete_transaction(journal, commit_tid);
1151 * Fast commit cleanup routine. This is called after every fast commit and
1152 * full commit. full is true if we are called after a full commit.
1154 static void ext4_fc_cleanup(journal_t *journal, int full)
1156 struct super_block *sb = journal->j_private;
1157 struct ext4_sb_info *sbi = EXT4_SB(sb);
1158 struct ext4_inode_info *iter;
1159 struct ext4_fc_dentry_update *fc_dentry;
1160 struct list_head *pos, *n;
1162 if (full && sbi->s_fc_bh)
1163 sbi->s_fc_bh = NULL;
1165 jbd2_fc_release_bufs(journal);
1167 spin_lock(&sbi->s_fc_lock);
1168 list_for_each_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN]) {
1169 iter = list_entry(pos, struct ext4_inode_info, i_fc_list);
1170 list_del_init(&iter->i_fc_list);
1171 ext4_clear_inode_state(&iter->vfs_inode,
1172 EXT4_STATE_FC_COMMITTING);
1173 ext4_fc_reset_inode(&iter->vfs_inode);
1174 /* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
1176 #if (BITS_PER_LONG < 64)
1177 wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);
1179 wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);
1183 while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {
1184 fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],
1185 struct ext4_fc_dentry_update,
1187 list_del_init(&fc_dentry->fcd_list);
1188 spin_unlock(&sbi->s_fc_lock);
1190 if (fc_dentry->fcd_name.name &&
1191 fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
1192 kfree(fc_dentry->fcd_name.name);
1193 kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
1194 spin_lock(&sbi->s_fc_lock);
1197 list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],
1198 &sbi->s_fc_dentry_q[FC_Q_MAIN]);
1199 list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
1200 &sbi->s_fc_q[FC_Q_STAGING]);
1202 sbi->s_mount_flags &= ~EXT4_MF_FC_COMMITTING;
1203 sbi->s_mount_flags &= ~EXT4_MF_FC_INELIGIBLE;
1206 sbi->s_fc_bytes = 0;
1207 spin_unlock(&sbi->s_fc_lock);
1208 trace_ext4_fc_stats(sb);
1211 /* Ext4 Replay Path Routines */
1213 /* Get length of a particular tlv */
1214 static inline int ext4_fc_tag_len(struct ext4_fc_tl *tl)
1216 return le16_to_cpu(tl->fc_len);
1219 /* Get a pointer to "value" of a tlv */
1220 static inline u8 *ext4_fc_tag_val(struct ext4_fc_tl *tl)
1222 return (u8 *)tl + sizeof(*tl);
1225 /* Helper struct for dentry replay routines */
1226 struct dentry_info_args {
1227 int parent_ino, dname_len, ino, inode_len;
1231 static inline void tl_to_darg(struct dentry_info_args *darg,
1232 struct ext4_fc_tl *tl)
1234 struct ext4_fc_dentry_info *fcd;
1236 fcd = (struct ext4_fc_dentry_info *)ext4_fc_tag_val(tl);
1238 darg->parent_ino = le32_to_cpu(fcd->fc_parent_ino);
1239 darg->ino = le32_to_cpu(fcd->fc_ino);
1240 darg->dname = fcd->fc_dname;
1241 darg->dname_len = ext4_fc_tag_len(tl) -
1242 sizeof(struct ext4_fc_dentry_info);
1245 /* Unlink replay function */
1246 static int ext4_fc_replay_unlink(struct super_block *sb, struct ext4_fc_tl *tl)
1248 struct inode *inode, *old_parent;
1250 struct dentry_info_args darg;
1253 tl_to_darg(&darg, tl);
1255 trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,
1256 darg.parent_ino, darg.dname_len);
1258 entry.name = darg.dname;
1259 entry.len = darg.dname_len;
1260 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1262 if (IS_ERR_OR_NULL(inode)) {
1263 jbd_debug(1, "Inode %d not found", darg.ino);
1267 old_parent = ext4_iget(sb, darg.parent_ino,
1269 if (IS_ERR_OR_NULL(old_parent)) {
1270 jbd_debug(1, "Dir with inode %d not found", darg.parent_ino);
1275 ret = __ext4_unlink(NULL, old_parent, &entry, inode);
1276 /* -ENOENT ok coz it might not exist anymore. */
1284 static int ext4_fc_replay_link_internal(struct super_block *sb,
1285 struct dentry_info_args *darg,
1286 struct inode *inode)
1288 struct inode *dir = NULL;
1289 struct dentry *dentry_dir = NULL, *dentry_inode = NULL;
1290 struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);
1293 dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);
1295 jbd_debug(1, "Dir with inode %d not found.", darg->parent_ino);
1300 dentry_dir = d_obtain_alias(dir);
1301 if (IS_ERR(dentry_dir)) {
1302 jbd_debug(1, "Failed to obtain dentry");
1307 dentry_inode = d_alloc(dentry_dir, &qstr_dname);
1308 if (!dentry_inode) {
1309 jbd_debug(1, "Inode dentry not created.");
1314 ret = __ext4_link(dir, inode, dentry_inode);
1316 * It's possible that link already existed since data blocks
1317 * for the dir in question got persisted before we crashed OR
1318 * we replayed this tag and crashed before the entire replay
1321 if (ret && ret != -EEXIST) {
1322 jbd_debug(1, "Failed to link\n");
1335 d_drop(dentry_inode);
1342 /* Link replay function */
1343 static int ext4_fc_replay_link(struct super_block *sb, struct ext4_fc_tl *tl)
1345 struct inode *inode;
1346 struct dentry_info_args darg;
1349 tl_to_darg(&darg, tl);
1350 trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,
1351 darg.parent_ino, darg.dname_len);
1353 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1354 if (IS_ERR_OR_NULL(inode)) {
1355 jbd_debug(1, "Inode not found.");
1359 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1365 * Record all the modified inodes during replay. We use this later to setup
1366 * block bitmaps correctly.
1368 static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)
1370 struct ext4_fc_replay_state *state;
1373 state = &EXT4_SB(sb)->s_fc_replay_state;
1374 for (i = 0; i < state->fc_modified_inodes_used; i++)
1375 if (state->fc_modified_inodes[i] == ino)
1377 if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {
1378 state->fc_modified_inodes_size +=
1379 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1380 state->fc_modified_inodes = krealloc(
1381 state->fc_modified_inodes, sizeof(int) *
1382 state->fc_modified_inodes_size,
1384 if (!state->fc_modified_inodes)
1387 state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;
1392 * Inode replay function
1394 static int ext4_fc_replay_inode(struct super_block *sb, struct ext4_fc_tl *tl)
1396 struct ext4_fc_inode *fc_inode;
1397 struct ext4_inode *raw_inode;
1398 struct ext4_inode *raw_fc_inode;
1399 struct inode *inode = NULL;
1400 struct ext4_iloc iloc;
1401 int inode_len, ino, ret, tag = le16_to_cpu(tl->fc_tag);
1402 struct ext4_extent_header *eh;
1404 fc_inode = (struct ext4_fc_inode *)ext4_fc_tag_val(tl);
1406 ino = le32_to_cpu(fc_inode->fc_ino);
1407 trace_ext4_fc_replay(sb, tag, ino, 0, 0);
1409 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1410 if (!IS_ERR_OR_NULL(inode)) {
1411 ext4_ext_clear_bb(inode);
1415 ext4_fc_record_modified_inode(sb, ino);
1417 raw_fc_inode = (struct ext4_inode *)fc_inode->fc_raw_inode;
1418 ret = ext4_get_fc_inode_loc(sb, ino, &iloc);
1422 inode_len = ext4_fc_tag_len(tl) - sizeof(struct ext4_fc_inode);
1423 raw_inode = ext4_raw_inode(&iloc);
1425 memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));
1426 memcpy(&raw_inode->i_generation, &raw_fc_inode->i_generation,
1427 inode_len - offsetof(struct ext4_inode, i_generation));
1428 if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {
1429 eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);
1430 if (eh->eh_magic != EXT4_EXT_MAGIC) {
1431 memset(eh, 0, sizeof(*eh));
1432 eh->eh_magic = EXT4_EXT_MAGIC;
1433 eh->eh_max = cpu_to_le16(
1434 (sizeof(raw_inode->i_block) -
1435 sizeof(struct ext4_extent_header))
1436 / sizeof(struct ext4_extent));
1438 } else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {
1439 memcpy(raw_inode->i_block, raw_fc_inode->i_block,
1440 sizeof(raw_inode->i_block));
1443 /* Immediately update the inode on disk. */
1444 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1447 ret = sync_dirty_buffer(iloc.bh);
1450 ret = ext4_mark_inode_used(sb, ino);
1454 /* Given that we just wrote the inode on disk, this SHOULD succeed. */
1455 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1456 if (IS_ERR_OR_NULL(inode)) {
1457 jbd_debug(1, "Inode not found.");
1458 return -EFSCORRUPTED;
1462 * Our allocator could have made different decisions than before
1463 * crashing. This should be fixed but until then, we calculate
1464 * the number of blocks the inode.
1466 ext4_ext_replay_set_iblocks(inode);
1468 inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);
1469 ext4_reset_inode_seed(inode);
1471 ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));
1472 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1473 sync_dirty_buffer(iloc.bh);
1478 blkdev_issue_flush(sb->s_bdev, GFP_KERNEL);
1484 * Dentry create replay function.
1486 * EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the
1487 * inode for which we are trying to create a dentry here, should already have
1488 * been replayed before we start here.
1490 static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl)
1493 struct inode *inode = NULL;
1494 struct inode *dir = NULL;
1495 struct dentry_info_args darg;
1497 tl_to_darg(&darg, tl);
1499 trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,
1500 darg.parent_ino, darg.dname_len);
1502 /* This takes care of update group descriptor and other metadata */
1503 ret = ext4_mark_inode_used(sb, darg.ino);
1507 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1508 if (IS_ERR_OR_NULL(inode)) {
1509 jbd_debug(1, "inode %d not found.", darg.ino);
1515 if (S_ISDIR(inode->i_mode)) {
1517 * If we are creating a directory, we need to make sure that the
1518 * dot and dot dot dirents are setup properly.
1520 dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);
1521 if (IS_ERR_OR_NULL(dir)) {
1522 jbd_debug(1, "Dir %d not found.", darg.ino);
1525 ret = ext4_init_new_dir(NULL, dir, inode);
1532 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1535 set_nlink(inode, 1);
1536 ext4_mark_inode_dirty(NULL, inode);
1544 * Record physical disk regions which are in use as per fast commit area. Our
1545 * simple replay phase allocator excludes these regions from allocation.
1547 static int ext4_fc_record_regions(struct super_block *sb, int ino,
1548 ext4_lblk_t lblk, ext4_fsblk_t pblk, int len)
1550 struct ext4_fc_replay_state *state;
1551 struct ext4_fc_alloc_region *region;
1553 state = &EXT4_SB(sb)->s_fc_replay_state;
1554 if (state->fc_regions_used == state->fc_regions_size) {
1555 state->fc_regions_size +=
1556 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1557 state->fc_regions = krealloc(
1559 state->fc_regions_size *
1560 sizeof(struct ext4_fc_alloc_region),
1562 if (!state->fc_regions)
1565 region = &state->fc_regions[state->fc_regions_used++];
1567 region->lblk = lblk;
1568 region->pblk = pblk;
1574 /* Replay add range tag */
1575 static int ext4_fc_replay_add_range(struct super_block *sb,
1576 struct ext4_fc_tl *tl)
1578 struct ext4_fc_add_range *fc_add_ex;
1579 struct ext4_extent newex, *ex;
1580 struct inode *inode;
1581 ext4_lblk_t start, cur;
1583 ext4_fsblk_t start_pblk;
1584 struct ext4_map_blocks map;
1585 struct ext4_ext_path *path = NULL;
1588 fc_add_ex = (struct ext4_fc_add_range *)ext4_fc_tag_val(tl);
1589 ex = (struct ext4_extent *)&fc_add_ex->fc_ex;
1591 trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,
1592 le32_to_cpu(fc_add_ex->fc_ino), le32_to_cpu(ex->ee_block),
1593 ext4_ext_get_actual_len(ex));
1595 inode = ext4_iget(sb, le32_to_cpu(fc_add_ex->fc_ino),
1597 if (IS_ERR_OR_NULL(inode)) {
1598 jbd_debug(1, "Inode not found.");
1602 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1604 start = le32_to_cpu(ex->ee_block);
1605 start_pblk = ext4_ext_pblock(ex);
1606 len = ext4_ext_get_actual_len(ex);
1610 jbd_debug(1, "ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",
1611 start, start_pblk, len, ext4_ext_is_unwritten(ex),
1614 while (remaining > 0) {
1616 map.m_len = remaining;
1618 ret = ext4_map_blocks(NULL, inode, &map, 0);
1626 /* Range is not mapped */
1627 path = ext4_find_extent(inode, cur, NULL, 0);
1632 memset(&newex, 0, sizeof(newex));
1633 newex.ee_block = cpu_to_le32(cur);
1634 ext4_ext_store_pblock(
1635 &newex, start_pblk + cur - start);
1636 newex.ee_len = cpu_to_le16(map.m_len);
1637 if (ext4_ext_is_unwritten(ex))
1638 ext4_ext_mark_unwritten(&newex);
1639 down_write(&EXT4_I(inode)->i_data_sem);
1640 ret = ext4_ext_insert_extent(
1641 NULL, inode, &path, &newex, 0);
1642 up_write((&EXT4_I(inode)->i_data_sem));
1643 ext4_ext_drop_refs(path);
1652 if (start_pblk + cur - start != map.m_pblk) {
1654 * Logical to physical mapping changed. This can happen
1655 * if this range was removed and then reallocated to
1656 * map to new physical blocks during a fast commit.
1658 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1659 ext4_ext_is_unwritten(ex),
1660 start_pblk + cur - start);
1666 * Mark the old blocks as free since they aren't used
1667 * anymore. We maintain an array of all the modified
1668 * inodes. In case these blocks are still used at either
1669 * a different logical range in the same inode or in
1670 * some different inode, we will mark them as allocated
1671 * at the end of the FC replay using our array of
1674 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1678 /* Range is mapped and needs a state change */
1679 jbd_debug(1, "Converting from %d to %d %lld",
1680 map.m_flags & EXT4_MAP_UNWRITTEN,
1681 ext4_ext_is_unwritten(ex), map.m_pblk);
1682 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1683 ext4_ext_is_unwritten(ex), map.m_pblk);
1689 * We may have split the extent tree while toggling the state.
1690 * Try to shrink the extent tree now.
1692 ext4_ext_replay_shrink_inode(inode, start + len);
1695 remaining -= map.m_len;
1697 ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>
1698 sb->s_blocksize_bits);
1703 /* Replay DEL_RANGE tag */
1705 ext4_fc_replay_del_range(struct super_block *sb, struct ext4_fc_tl *tl)
1707 struct inode *inode;
1708 struct ext4_fc_del_range *lrange;
1709 struct ext4_map_blocks map;
1710 ext4_lblk_t cur, remaining;
1713 lrange = (struct ext4_fc_del_range *)ext4_fc_tag_val(tl);
1714 cur = le32_to_cpu(lrange->fc_lblk);
1715 remaining = le32_to_cpu(lrange->fc_len);
1717 trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,
1718 le32_to_cpu(lrange->fc_ino), cur, remaining);
1720 inode = ext4_iget(sb, le32_to_cpu(lrange->fc_ino), EXT4_IGET_NORMAL);
1721 if (IS_ERR_OR_NULL(inode)) {
1722 jbd_debug(1, "Inode %d not found", le32_to_cpu(lrange->fc_ino));
1726 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1728 jbd_debug(1, "DEL_RANGE, inode %ld, lblk %d, len %d\n",
1729 inode->i_ino, le32_to_cpu(lrange->fc_lblk),
1730 le32_to_cpu(lrange->fc_len));
1731 while (remaining > 0) {
1733 map.m_len = remaining;
1735 ret = ext4_map_blocks(NULL, inode, &map, 0);
1743 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1745 remaining -= map.m_len;
1750 ret = ext4_punch_hole(inode,
1751 le32_to_cpu(lrange->fc_lblk) << sb->s_blocksize_bits,
1752 le32_to_cpu(lrange->fc_len) << sb->s_blocksize_bits);
1754 jbd_debug(1, "ext4_punch_hole returned %d", ret);
1755 ext4_ext_replay_shrink_inode(inode,
1756 i_size_read(inode) >> sb->s_blocksize_bits);
1757 ext4_mark_inode_dirty(NULL, inode);
1763 static inline const char *tag2str(u16 tag)
1766 case EXT4_FC_TAG_LINK:
1767 return "TAG_ADD_ENTRY";
1768 case EXT4_FC_TAG_UNLINK:
1769 return "TAG_DEL_ENTRY";
1770 case EXT4_FC_TAG_ADD_RANGE:
1771 return "TAG_ADD_RANGE";
1772 case EXT4_FC_TAG_CREAT:
1773 return "TAG_CREAT_DENTRY";
1774 case EXT4_FC_TAG_DEL_RANGE:
1775 return "TAG_DEL_RANGE";
1776 case EXT4_FC_TAG_INODE:
1778 case EXT4_FC_TAG_PAD:
1780 case EXT4_FC_TAG_TAIL:
1782 case EXT4_FC_TAG_HEAD:
1789 static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)
1791 struct ext4_fc_replay_state *state;
1792 struct inode *inode;
1793 struct ext4_ext_path *path = NULL;
1794 struct ext4_map_blocks map;
1796 ext4_lblk_t cur, end;
1798 state = &EXT4_SB(sb)->s_fc_replay_state;
1799 for (i = 0; i < state->fc_modified_inodes_used; i++) {
1800 inode = ext4_iget(sb, state->fc_modified_inodes[i],
1802 if (IS_ERR_OR_NULL(inode)) {
1803 jbd_debug(1, "Inode %d not found.",
1804 state->fc_modified_inodes[i]);
1808 end = EXT_MAX_BLOCKS;
1811 map.m_len = end - cur;
1813 ret = ext4_map_blocks(NULL, inode, &map, 0);
1818 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
1819 if (!IS_ERR_OR_NULL(path)) {
1820 for (j = 0; j < path->p_depth; j++)
1821 ext4_mb_mark_bb(inode->i_sb,
1822 path[j].p_block, 1, 1);
1823 ext4_ext_drop_refs(path);
1827 ext4_mb_mark_bb(inode->i_sb, map.m_pblk,
1830 cur = cur + (map.m_len ? map.m_len : 1);
1838 * Check if block is in excluded regions for block allocation. The simple
1839 * allocator that runs during replay phase is calls this function to see
1840 * if it is okay to use a block.
1842 bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)
1845 struct ext4_fc_replay_state *state;
1847 state = &EXT4_SB(sb)->s_fc_replay_state;
1848 for (i = 0; i < state->fc_regions_valid; i++) {
1849 if (state->fc_regions[i].ino == 0 ||
1850 state->fc_regions[i].len == 0)
1852 if (blk >= state->fc_regions[i].pblk &&
1853 blk < state->fc_regions[i].pblk + state->fc_regions[i].len)
1859 /* Cleanup function called after replay */
1860 void ext4_fc_replay_cleanup(struct super_block *sb)
1862 struct ext4_sb_info *sbi = EXT4_SB(sb);
1864 sbi->s_mount_state &= ~EXT4_FC_REPLAY;
1865 kfree(sbi->s_fc_replay_state.fc_regions);
1866 kfree(sbi->s_fc_replay_state.fc_modified_inodes);
1870 * Recovery Scan phase handler
1872 * This function is called during the scan phase and is responsible
1873 * for doing following things:
1874 * - Make sure the fast commit area has valid tags for replay
1875 * - Count number of tags that need to be replayed by the replay handler
1877 * - Create a list of excluded blocks for allocation during replay phase
1879 * This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is
1880 * incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP
1881 * to indicate that scan has finished and JBD2 can now start replay phase.
1882 * It returns a negative error to indicate that there was an error. At the end
1883 * of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set
1884 * to indicate the number of tags that need to replayed during the replay phase.
1886 static int ext4_fc_replay_scan(journal_t *journal,
1887 struct buffer_head *bh, int off,
1890 struct super_block *sb = journal->j_private;
1891 struct ext4_sb_info *sbi = EXT4_SB(sb);
1892 struct ext4_fc_replay_state *state;
1893 int ret = JBD2_FC_REPLAY_CONTINUE;
1894 struct ext4_fc_add_range *ext;
1895 struct ext4_fc_tl *tl;
1896 struct ext4_fc_tail *tail;
1898 struct ext4_fc_head *head;
1899 struct ext4_extent *ex;
1901 state = &sbi->s_fc_replay_state;
1903 start = (u8 *)bh->b_data;
1904 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
1906 if (state->fc_replay_expected_off == 0) {
1907 state->fc_cur_tag = 0;
1908 state->fc_replay_num_tags = 0;
1910 state->fc_regions = NULL;
1911 state->fc_regions_valid = state->fc_regions_used =
1912 state->fc_regions_size = 0;
1913 /* Check if we can stop early */
1914 if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
1915 != EXT4_FC_TAG_HEAD)
1919 if (off != state->fc_replay_expected_off) {
1920 ret = -EFSCORRUPTED;
1924 state->fc_replay_expected_off++;
1925 fc_for_each_tl(start, end, tl) {
1926 jbd_debug(3, "Scan phase, tag:%s, blk %lld\n",
1927 tag2str(le16_to_cpu(tl->fc_tag)), bh->b_blocknr);
1928 switch (le16_to_cpu(tl->fc_tag)) {
1929 case EXT4_FC_TAG_ADD_RANGE:
1930 ext = (struct ext4_fc_add_range *)ext4_fc_tag_val(tl);
1931 ex = (struct ext4_extent *)&ext->fc_ex;
1932 ret = ext4_fc_record_regions(sb,
1933 le32_to_cpu(ext->fc_ino),
1934 le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
1935 ext4_ext_get_actual_len(ex));
1938 ret = JBD2_FC_REPLAY_CONTINUE;
1940 case EXT4_FC_TAG_DEL_RANGE:
1941 case EXT4_FC_TAG_LINK:
1942 case EXT4_FC_TAG_UNLINK:
1943 case EXT4_FC_TAG_CREAT:
1944 case EXT4_FC_TAG_INODE:
1945 case EXT4_FC_TAG_PAD:
1946 state->fc_cur_tag++;
1947 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1948 sizeof(*tl) + ext4_fc_tag_len(tl));
1950 case EXT4_FC_TAG_TAIL:
1951 state->fc_cur_tag++;
1952 tail = (struct ext4_fc_tail *)ext4_fc_tag_val(tl);
1953 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1955 offsetof(struct ext4_fc_tail,
1957 if (le32_to_cpu(tail->fc_tid) == expected_tid &&
1958 le32_to_cpu(tail->fc_crc) == state->fc_crc) {
1959 state->fc_replay_num_tags = state->fc_cur_tag;
1960 state->fc_regions_valid =
1961 state->fc_regions_used;
1963 ret = state->fc_replay_num_tags ?
1964 JBD2_FC_REPLAY_STOP : -EFSBADCRC;
1968 case EXT4_FC_TAG_HEAD:
1969 head = (struct ext4_fc_head *)ext4_fc_tag_val(tl);
1970 if (le32_to_cpu(head->fc_features) &
1971 ~EXT4_FC_SUPPORTED_FEATURES) {
1975 if (le32_to_cpu(head->fc_tid) != expected_tid) {
1976 ret = JBD2_FC_REPLAY_STOP;
1979 state->fc_cur_tag++;
1980 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1981 sizeof(*tl) + ext4_fc_tag_len(tl));
1984 ret = state->fc_replay_num_tags ?
1985 JBD2_FC_REPLAY_STOP : -ECANCELED;
1987 if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
1992 trace_ext4_fc_replay_scan(sb, ret, off);
1997 * Main recovery path entry point.
1998 * The meaning of return codes is similar as above.
2000 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
2001 enum passtype pass, int off, tid_t expected_tid)
2003 struct super_block *sb = journal->j_private;
2004 struct ext4_sb_info *sbi = EXT4_SB(sb);
2005 struct ext4_fc_tl *tl;
2007 int ret = JBD2_FC_REPLAY_CONTINUE;
2008 struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;
2009 struct ext4_fc_tail *tail;
2011 if (pass == PASS_SCAN) {
2012 state->fc_current_pass = PASS_SCAN;
2013 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
2016 if (state->fc_current_pass != pass) {
2017 state->fc_current_pass = pass;
2018 sbi->s_mount_state |= EXT4_FC_REPLAY;
2020 if (!sbi->s_fc_replay_state.fc_replay_num_tags) {
2021 jbd_debug(1, "Replay stops\n");
2022 ext4_fc_set_bitmaps_and_counters(sb);
2026 #ifdef CONFIG_EXT4_DEBUG
2027 if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {
2028 pr_warn("Dropping fc block %d because max_replay set\n", off);
2029 return JBD2_FC_REPLAY_STOP;
2033 start = (u8 *)bh->b_data;
2034 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
2036 fc_for_each_tl(start, end, tl) {
2037 if (state->fc_replay_num_tags == 0) {
2038 ret = JBD2_FC_REPLAY_STOP;
2039 ext4_fc_set_bitmaps_and_counters(sb);
2042 jbd_debug(3, "Replay phase, tag:%s\n",
2043 tag2str(le16_to_cpu(tl->fc_tag)));
2044 state->fc_replay_num_tags--;
2045 switch (le16_to_cpu(tl->fc_tag)) {
2046 case EXT4_FC_TAG_LINK:
2047 ret = ext4_fc_replay_link(sb, tl);
2049 case EXT4_FC_TAG_UNLINK:
2050 ret = ext4_fc_replay_unlink(sb, tl);
2052 case EXT4_FC_TAG_ADD_RANGE:
2053 ret = ext4_fc_replay_add_range(sb, tl);
2055 case EXT4_FC_TAG_CREAT:
2056 ret = ext4_fc_replay_create(sb, tl);
2058 case EXT4_FC_TAG_DEL_RANGE:
2059 ret = ext4_fc_replay_del_range(sb, tl);
2061 case EXT4_FC_TAG_INODE:
2062 ret = ext4_fc_replay_inode(sb, tl);
2064 case EXT4_FC_TAG_PAD:
2065 trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,
2066 ext4_fc_tag_len(tl), 0);
2068 case EXT4_FC_TAG_TAIL:
2069 trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL, 0,
2070 ext4_fc_tag_len(tl), 0);
2071 tail = (struct ext4_fc_tail *)ext4_fc_tag_val(tl);
2072 WARN_ON(le32_to_cpu(tail->fc_tid) != expected_tid);
2074 case EXT4_FC_TAG_HEAD:
2077 trace_ext4_fc_replay(sb, le16_to_cpu(tl->fc_tag), 0,
2078 ext4_fc_tag_len(tl), 0);
2084 ret = JBD2_FC_REPLAY_CONTINUE;
2089 void ext4_fc_init(struct super_block *sb, journal_t *journal)
2092 * We set replay callback even if fast commit disabled because we may
2093 * could still have fast commit blocks that need to be replayed even if
2094 * fast commit has now been turned off.
2096 journal->j_fc_replay_callback = ext4_fc_replay;
2097 if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
2099 journal->j_fc_cleanup_callback = ext4_fc_cleanup;
2102 const char *fc_ineligible_reasons[] = {
2103 "Extended attributes changed",
2105 "Journal flag changed",
2106 "Insufficient memory",
2115 int ext4_fc_info_show(struct seq_file *seq, void *v)
2117 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);
2118 struct ext4_fc_stats *stats = &sbi->s_fc_stats;
2121 if (v != SEQ_START_TOKEN)
2125 "fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",
2126 stats->fc_num_commits, stats->fc_ineligible_commits,
2128 div_u64(sbi->s_fc_avg_commit_time, 1000));
2129 seq_puts(seq, "Ineligible reasons:\n");
2130 for (i = 0; i < EXT4_FC_REASON_MAX; i++)
2131 seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],
2132 stats->fc_ineligible_reason_count[i]);
2137 int __init ext4_fc_init_dentry_cache(void)
2139 ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,
2140 SLAB_RECLAIM_ACCOUNT);
2142 if (ext4_fc_dentry_cachep == NULL)