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 ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);
487 trace_ext4_fc_track_inode(inode, ret);
490 struct __track_range_args {
491 ext4_lblk_t start, end;
494 /* __track_fn for tracking data updates */
495 static int __track_range(struct inode *inode, void *arg, bool update)
497 struct ext4_inode_info *ei = EXT4_I(inode);
498 ext4_lblk_t oldstart;
499 struct __track_range_args *__arg =
500 (struct __track_range_args *)arg;
502 if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {
503 ext4_debug("Special inode %ld being modified\n", inode->i_ino);
507 oldstart = ei->i_fc_lblk_start;
509 if (update && ei->i_fc_lblk_len > 0) {
510 ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);
512 max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -
513 ei->i_fc_lblk_start + 1;
515 ei->i_fc_lblk_start = __arg->start;
516 ei->i_fc_lblk_len = __arg->end - __arg->start + 1;
522 void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,
525 struct __track_range_args args;
528 if (S_ISDIR(inode->i_mode))
534 ret = ext4_fc_track_template(handle, inode, __track_range, &args, 1);
536 trace_ext4_fc_track_range(inode, start, end, ret);
539 static void ext4_fc_submit_bh(struct super_block *sb)
541 int write_flags = REQ_SYNC;
542 struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
544 /* TODO: REQ_FUA | REQ_PREFLUSH is unnecessarily expensive. */
545 if (test_opt(sb, BARRIER))
546 write_flags |= REQ_FUA | REQ_PREFLUSH;
548 set_buffer_dirty(bh);
549 set_buffer_uptodate(bh);
550 bh->b_end_io = ext4_end_buffer_io_sync;
551 submit_bh(REQ_OP_WRITE, write_flags, bh);
552 EXT4_SB(sb)->s_fc_bh = NULL;
555 /* Ext4 commit path routines */
557 /* memzero and update CRC */
558 static void *ext4_fc_memzero(struct super_block *sb, void *dst, int len,
563 ret = memset(dst, 0, len);
565 *crc = ext4_chksum(EXT4_SB(sb), *crc, dst, len);
570 * Allocate len bytes on a fast commit buffer.
572 * During the commit time this function is used to manage fast commit
573 * block space. We don't split a fast commit log onto different
574 * blocks. So this function makes sure that if there's not enough space
575 * on the current block, the remaining space in the current block is
576 * marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,
577 * new block is from jbd2 and CRC is updated to reflect the padding
580 static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
582 struct ext4_fc_tl *tl;
583 struct ext4_sb_info *sbi = EXT4_SB(sb);
584 struct buffer_head *bh;
585 int bsize = sbi->s_journal->j_blocksize;
586 int ret, off = sbi->s_fc_bytes % bsize;
590 * After allocating len, we should have space at least for a 0 byte
593 if (len + sizeof(struct ext4_fc_tl) > bsize)
596 if (bsize - off - 1 > len + sizeof(struct ext4_fc_tl)) {
598 * Only allocate from current buffer if we have enough space for
599 * this request AND we have space to add a zero byte padding.
602 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
607 sbi->s_fc_bytes += len;
608 return sbi->s_fc_bh->b_data + off;
610 /* Need to add PAD tag */
611 tl = (struct ext4_fc_tl *)(sbi->s_fc_bh->b_data + off);
612 tl->fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);
613 pad_len = bsize - off - 1 - sizeof(struct ext4_fc_tl);
614 tl->fc_len = cpu_to_le16(pad_len);
616 *crc = ext4_chksum(sbi, *crc, tl, sizeof(*tl));
618 ext4_fc_memzero(sb, tl + 1, pad_len, crc);
619 ext4_fc_submit_bh(sb);
621 ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
625 sbi->s_fc_bytes = (sbi->s_fc_bytes / bsize + 1) * bsize + len;
626 return sbi->s_fc_bh->b_data;
629 /* memcpy to fc reserved space and update CRC */
630 static void *ext4_fc_memcpy(struct super_block *sb, void *dst, const void *src,
634 *crc = ext4_chksum(EXT4_SB(sb), *crc, src, len);
635 return memcpy(dst, src, len);
639 * Complete a fast commit by writing tail tag.
641 * Writing tail tag marks the end of a fast commit. In order to guarantee
642 * atomicity, after writing tail tag, even if there's space remaining
643 * in the block, next commit shouldn't use it. That's why tail tag
644 * has the length as that of the remaining space on the block.
646 static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
648 struct ext4_sb_info *sbi = EXT4_SB(sb);
649 struct ext4_fc_tl tl;
650 struct ext4_fc_tail tail;
651 int off, bsize = sbi->s_journal->j_blocksize;
655 * ext4_fc_reserve_space takes care of allocating an extra block if
656 * there's no enough space on this block for accommodating this tail.
658 dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(tail), &crc);
662 off = sbi->s_fc_bytes % bsize;
664 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);
665 tl.fc_len = cpu_to_le16(bsize - off - 1 + sizeof(struct ext4_fc_tail));
666 sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);
668 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), &crc);
670 tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);
671 ext4_fc_memcpy(sb, dst, &tail.fc_tid, sizeof(tail.fc_tid), &crc);
672 dst += sizeof(tail.fc_tid);
673 tail.fc_crc = cpu_to_le32(crc);
674 ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
676 ext4_fc_submit_bh(sb);
682 * Adds tag, length, value and updates CRC. Returns true if tlv was added.
683 * Returns false if there's not enough space.
685 static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,
688 struct ext4_fc_tl tl;
691 dst = ext4_fc_reserve_space(sb, sizeof(tl) + len, crc);
695 tl.fc_tag = cpu_to_le16(tag);
696 tl.fc_len = cpu_to_le16(len);
698 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
699 ext4_fc_memcpy(sb, dst + sizeof(tl), val, len, crc);
704 /* Same as above, but adds dentry tlv. */
705 static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u16 tag,
706 int parent_ino, int ino, int dlen,
707 const unsigned char *dname,
710 struct ext4_fc_dentry_info fcd;
711 struct ext4_fc_tl tl;
712 u8 *dst = ext4_fc_reserve_space(sb, sizeof(tl) + sizeof(fcd) + dlen,
718 fcd.fc_parent_ino = cpu_to_le32(parent_ino);
719 fcd.fc_ino = cpu_to_le32(ino);
720 tl.fc_tag = cpu_to_le16(tag);
721 tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);
722 ext4_fc_memcpy(sb, dst, &tl, sizeof(tl), crc);
724 ext4_fc_memcpy(sb, dst, &fcd, sizeof(fcd), crc);
726 ext4_fc_memcpy(sb, dst, dname, dlen, crc);
733 * Writes inode in the fast commit space under TLV with tag @tag.
734 * Returns 0 on success, error on failure.
736 static int ext4_fc_write_inode(struct inode *inode, u32 *crc)
738 struct ext4_inode_info *ei = EXT4_I(inode);
739 int inode_len = EXT4_GOOD_OLD_INODE_SIZE;
741 struct ext4_iloc iloc;
742 struct ext4_fc_inode fc_inode;
743 struct ext4_fc_tl tl;
746 ret = ext4_get_inode_loc(inode, &iloc);
750 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)
751 inode_len += ei->i_extra_isize;
753 fc_inode.fc_ino = cpu_to_le32(inode->i_ino);
754 tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);
755 tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));
757 dst = ext4_fc_reserve_space(inode->i_sb,
758 sizeof(tl) + inode_len + sizeof(fc_inode.fc_ino), crc);
762 if (!ext4_fc_memcpy(inode->i_sb, dst, &tl, sizeof(tl), crc))
765 if (!ext4_fc_memcpy(inode->i_sb, dst, &fc_inode, sizeof(fc_inode), crc))
767 dst += sizeof(fc_inode);
768 if (!ext4_fc_memcpy(inode->i_sb, dst, (u8 *)ext4_raw_inode(&iloc),
776 * Writes updated data ranges for the inode in question. Updates CRC.
777 * Returns 0 on success, error otherwise.
779 static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)
781 ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;
782 struct ext4_inode_info *ei = EXT4_I(inode);
783 struct ext4_map_blocks map;
784 struct ext4_fc_add_range fc_ext;
785 struct ext4_fc_del_range lrange;
786 struct ext4_extent *ex;
789 mutex_lock(&ei->i_fc_lock);
790 if (ei->i_fc_lblk_len == 0) {
791 mutex_unlock(&ei->i_fc_lock);
794 old_blk_size = ei->i_fc_lblk_start;
795 new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;
796 ei->i_fc_lblk_len = 0;
797 mutex_unlock(&ei->i_fc_lock);
799 cur_lblk_off = old_blk_size;
800 jbd_debug(1, "%s: will try writing %d to %d for inode %ld\n",
801 __func__, cur_lblk_off, new_blk_size, inode->i_ino);
803 while (cur_lblk_off <= new_blk_size) {
804 map.m_lblk = cur_lblk_off;
805 map.m_len = new_blk_size - cur_lblk_off + 1;
806 ret = ext4_map_blocks(NULL, inode, &map, 0);
810 if (map.m_len == 0) {
816 lrange.fc_ino = cpu_to_le32(inode->i_ino);
817 lrange.fc_lblk = cpu_to_le32(map.m_lblk);
818 lrange.fc_len = cpu_to_le32(map.m_len);
819 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,
820 sizeof(lrange), (u8 *)&lrange, crc))
823 fc_ext.fc_ino = cpu_to_le32(inode->i_ino);
824 ex = (struct ext4_extent *)&fc_ext.fc_ex;
825 ex->ee_block = cpu_to_le32(map.m_lblk);
826 ex->ee_len = cpu_to_le16(map.m_len);
827 ext4_ext_store_pblock(ex, map.m_pblk);
828 if (map.m_flags & EXT4_MAP_UNWRITTEN)
829 ext4_ext_mark_unwritten(ex);
831 ext4_ext_mark_initialized(ex);
832 if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,
833 sizeof(fc_ext), (u8 *)&fc_ext, crc))
837 cur_lblk_off += map.m_len;
844 /* Submit data for all the fast commit inodes */
845 static int ext4_fc_submit_inode_data_all(journal_t *journal)
847 struct super_block *sb = (struct super_block *)(journal->j_private);
848 struct ext4_sb_info *sbi = EXT4_SB(sb);
849 struct ext4_inode_info *ei;
850 struct list_head *pos;
853 spin_lock(&sbi->s_fc_lock);
854 sbi->s_mount_flags |= EXT4_MF_FC_COMMITTING;
855 list_for_each(pos, &sbi->s_fc_q[FC_Q_MAIN]) {
856 ei = list_entry(pos, struct ext4_inode_info, i_fc_list);
857 ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
858 while (atomic_read(&ei->i_fc_updates)) {
861 prepare_to_wait(&ei->i_fc_wait, &wait,
862 TASK_UNINTERRUPTIBLE);
863 if (atomic_read(&ei->i_fc_updates)) {
864 spin_unlock(&sbi->s_fc_lock);
866 spin_lock(&sbi->s_fc_lock);
868 finish_wait(&ei->i_fc_wait, &wait);
870 spin_unlock(&sbi->s_fc_lock);
871 ret = jbd2_submit_inode_data(ei->jinode);
874 spin_lock(&sbi->s_fc_lock);
876 spin_unlock(&sbi->s_fc_lock);
881 /* Wait for completion of data for all the fast commit inodes */
882 static int ext4_fc_wait_inode_data_all(journal_t *journal)
884 struct super_block *sb = (struct super_block *)(journal->j_private);
885 struct ext4_sb_info *sbi = EXT4_SB(sb);
886 struct ext4_inode_info *pos, *n;
889 spin_lock(&sbi->s_fc_lock);
890 list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
891 if (!ext4_test_inode_state(&pos->vfs_inode,
892 EXT4_STATE_FC_COMMITTING))
894 spin_unlock(&sbi->s_fc_lock);
896 ret = jbd2_wait_inode_data(journal, pos->jinode);
899 spin_lock(&sbi->s_fc_lock);
901 spin_unlock(&sbi->s_fc_lock);
906 /* Commit all the directory entry updates */
907 static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)
909 struct super_block *sb = (struct super_block *)(journal->j_private);
910 struct ext4_sb_info *sbi = EXT4_SB(sb);
911 struct ext4_fc_dentry_update *fc_dentry;
913 struct list_head *pos, *n, *fcd_pos, *fcd_n;
914 struct ext4_inode_info *ei;
917 if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))
919 list_for_each_safe(fcd_pos, fcd_n, &sbi->s_fc_dentry_q[FC_Q_MAIN]) {
920 fc_dentry = list_entry(fcd_pos, struct ext4_fc_dentry_update,
922 if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {
923 spin_unlock(&sbi->s_fc_lock);
924 if (!ext4_fc_add_dentry_tlv(
925 sb, fc_dentry->fcd_op,
926 fc_dentry->fcd_parent, fc_dentry->fcd_ino,
927 fc_dentry->fcd_name.len,
928 fc_dentry->fcd_name.name, crc)) {
932 spin_lock(&sbi->s_fc_lock);
937 list_for_each_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN]) {
938 ei = list_entry(pos, struct ext4_inode_info, i_fc_list);
939 if (ei->vfs_inode.i_ino == fc_dentry->fcd_ino) {
940 inode = &ei->vfs_inode;
945 * If we don't find inode in our list, then it was deleted,
946 * in which case, we don't need to record it's create tag.
950 spin_unlock(&sbi->s_fc_lock);
953 * We first write the inode and then the create dirent. This
954 * allows the recovery code to create an unnamed inode first
955 * and then link it to a directory entry. This allows us
956 * to use namei.c routines almost as is and simplifies
959 ret = ext4_fc_write_inode(inode, crc);
963 ret = ext4_fc_write_inode_data(inode, crc);
967 if (!ext4_fc_add_dentry_tlv(
968 sb, fc_dentry->fcd_op,
969 fc_dentry->fcd_parent, fc_dentry->fcd_ino,
970 fc_dentry->fcd_name.len,
971 fc_dentry->fcd_name.name, crc)) {
976 spin_lock(&sbi->s_fc_lock);
980 spin_lock(&sbi->s_fc_lock);
984 static int ext4_fc_perform_commit(journal_t *journal)
986 struct super_block *sb = (struct super_block *)(journal->j_private);
987 struct ext4_sb_info *sbi = EXT4_SB(sb);
988 struct ext4_inode_info *iter;
989 struct ext4_fc_head head;
990 struct list_head *pos;
992 struct blk_plug plug;
996 ret = ext4_fc_submit_inode_data_all(journal);
1000 ret = ext4_fc_wait_inode_data_all(journal);
1004 blk_start_plug(&plug);
1005 if (sbi->s_fc_bytes == 0) {
1007 * Add a head tag only if this is the first fast commit
1010 head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);
1011 head.fc_tid = cpu_to_le32(
1012 sbi->s_journal->j_running_transaction->t_tid);
1013 if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),
1018 spin_lock(&sbi->s_fc_lock);
1019 ret = ext4_fc_commit_dentry_updates(journal, &crc);
1021 spin_unlock(&sbi->s_fc_lock);
1025 list_for_each(pos, &sbi->s_fc_q[FC_Q_MAIN]) {
1026 iter = list_entry(pos, struct ext4_inode_info, i_fc_list);
1027 inode = &iter->vfs_inode;
1028 if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))
1031 spin_unlock(&sbi->s_fc_lock);
1032 ret = ext4_fc_write_inode_data(inode, &crc);
1035 ret = ext4_fc_write_inode(inode, &crc);
1038 spin_lock(&sbi->s_fc_lock);
1040 spin_unlock(&sbi->s_fc_lock);
1042 ret = ext4_fc_write_tail(sb, crc);
1045 blk_finish_plug(&plug);
1050 * The main commit entry point. Performs a fast commit for transaction
1051 * commit_tid if needed. If it's not possible to perform a fast commit
1052 * due to various reasons, we fall back to full commit. Returns 0
1053 * on success, error otherwise.
1055 int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
1057 struct super_block *sb = (struct super_block *)(journal->j_private);
1058 struct ext4_sb_info *sbi = EXT4_SB(sb);
1059 int nblks = 0, ret, bsize = journal->j_blocksize;
1060 int subtid = atomic_read(&sbi->s_fc_subtid);
1061 int reason = EXT4_FC_REASON_OK, fc_bufs_before = 0;
1062 ktime_t start_time, commit_time;
1064 trace_ext4_fc_commit_start(sb);
1066 start_time = ktime_get();
1068 if (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
1069 (ext4_fc_is_ineligible(sb))) {
1070 reason = EXT4_FC_REASON_INELIGIBLE;
1075 ret = jbd2_fc_begin_commit(journal, commit_tid);
1076 if (ret == -EALREADY) {
1077 /* There was an ongoing commit, check if we need to restart */
1078 if (atomic_read(&sbi->s_fc_subtid) <= subtid &&
1079 commit_tid > journal->j_commit_sequence)
1081 reason = EXT4_FC_REASON_ALREADY_COMMITTED;
1084 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1085 reason = EXT4_FC_REASON_FC_START_FAILED;
1089 fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
1090 ret = ext4_fc_perform_commit(journal);
1092 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1093 reason = EXT4_FC_REASON_FC_FAILED;
1096 nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;
1097 ret = jbd2_fc_wait_bufs(journal, nblks);
1099 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1100 reason = EXT4_FC_REASON_FC_FAILED;
1103 atomic_inc(&sbi->s_fc_subtid);
1104 jbd2_fc_end_commit(journal);
1106 /* Has any ineligible update happened since we started? */
1107 if (reason == EXT4_FC_REASON_OK && ext4_fc_is_ineligible(sb)) {
1108 sbi->s_fc_stats.fc_ineligible_reason_count[EXT4_FC_COMMIT_FAILED]++;
1109 reason = EXT4_FC_REASON_INELIGIBLE;
1112 spin_lock(&sbi->s_fc_lock);
1113 if (reason != EXT4_FC_REASON_OK &&
1114 reason != EXT4_FC_REASON_ALREADY_COMMITTED) {
1115 sbi->s_fc_stats.fc_ineligible_commits++;
1117 sbi->s_fc_stats.fc_num_commits++;
1118 sbi->s_fc_stats.fc_numblks += nblks;
1120 spin_unlock(&sbi->s_fc_lock);
1121 nblks = (reason == EXT4_FC_REASON_OK) ? nblks : 0;
1122 trace_ext4_fc_commit_stop(sb, nblks, reason);
1123 commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1125 * weight the commit time higher than the average time so we don't
1126 * react too strongly to vast changes in the commit time
1128 if (likely(sbi->s_fc_avg_commit_time))
1129 sbi->s_fc_avg_commit_time = (commit_time +
1130 sbi->s_fc_avg_commit_time * 3) / 4;
1132 sbi->s_fc_avg_commit_time = commit_time;
1134 "Fast commit ended with blks = %d, reason = %d, subtid - %d",
1135 nblks, reason, subtid);
1136 if (reason == EXT4_FC_REASON_FC_FAILED)
1137 return jbd2_fc_end_commit_fallback(journal);
1138 if (reason == EXT4_FC_REASON_FC_START_FAILED ||
1139 reason == EXT4_FC_REASON_INELIGIBLE)
1140 return jbd2_complete_transaction(journal, commit_tid);
1145 * Fast commit cleanup routine. This is called after every fast commit and
1146 * full commit. full is true if we are called after a full commit.
1148 static void ext4_fc_cleanup(journal_t *journal, int full)
1150 struct super_block *sb = journal->j_private;
1151 struct ext4_sb_info *sbi = EXT4_SB(sb);
1152 struct ext4_inode_info *iter;
1153 struct ext4_fc_dentry_update *fc_dentry;
1154 struct list_head *pos, *n;
1156 if (full && sbi->s_fc_bh)
1157 sbi->s_fc_bh = NULL;
1159 jbd2_fc_release_bufs(journal);
1161 spin_lock(&sbi->s_fc_lock);
1162 list_for_each_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN]) {
1163 iter = list_entry(pos, struct ext4_inode_info, i_fc_list);
1164 list_del_init(&iter->i_fc_list);
1165 ext4_clear_inode_state(&iter->vfs_inode,
1166 EXT4_STATE_FC_COMMITTING);
1167 ext4_fc_reset_inode(&iter->vfs_inode);
1168 /* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
1170 #if (BITS_PER_LONG < 64)
1171 wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);
1173 wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);
1177 while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {
1178 fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],
1179 struct ext4_fc_dentry_update,
1181 list_del_init(&fc_dentry->fcd_list);
1182 spin_unlock(&sbi->s_fc_lock);
1184 if (fc_dentry->fcd_name.name &&
1185 fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
1186 kfree(fc_dentry->fcd_name.name);
1187 kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
1188 spin_lock(&sbi->s_fc_lock);
1191 list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],
1192 &sbi->s_fc_dentry_q[FC_Q_MAIN]);
1193 list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
1194 &sbi->s_fc_q[FC_Q_STAGING]);
1196 sbi->s_mount_flags &= ~EXT4_MF_FC_COMMITTING;
1197 sbi->s_mount_flags &= ~EXT4_MF_FC_INELIGIBLE;
1200 sbi->s_fc_bytes = 0;
1201 spin_unlock(&sbi->s_fc_lock);
1202 trace_ext4_fc_stats(sb);
1205 /* Ext4 Replay Path Routines */
1207 /* Get length of a particular tlv */
1208 static inline int ext4_fc_tag_len(struct ext4_fc_tl *tl)
1210 return le16_to_cpu(tl->fc_len);
1213 /* Get a pointer to "value" of a tlv */
1214 static inline u8 *ext4_fc_tag_val(struct ext4_fc_tl *tl)
1216 return (u8 *)tl + sizeof(*tl);
1219 /* Helper struct for dentry replay routines */
1220 struct dentry_info_args {
1221 int parent_ino, dname_len, ino, inode_len;
1225 static inline void tl_to_darg(struct dentry_info_args *darg,
1226 struct ext4_fc_tl *tl)
1228 struct ext4_fc_dentry_info *fcd;
1230 fcd = (struct ext4_fc_dentry_info *)ext4_fc_tag_val(tl);
1232 darg->parent_ino = le32_to_cpu(fcd->fc_parent_ino);
1233 darg->ino = le32_to_cpu(fcd->fc_ino);
1234 darg->dname = fcd->fc_dname;
1235 darg->dname_len = ext4_fc_tag_len(tl) -
1236 sizeof(struct ext4_fc_dentry_info);
1239 /* Unlink replay function */
1240 static int ext4_fc_replay_unlink(struct super_block *sb, struct ext4_fc_tl *tl)
1242 struct inode *inode, *old_parent;
1244 struct dentry_info_args darg;
1247 tl_to_darg(&darg, tl);
1249 trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,
1250 darg.parent_ino, darg.dname_len);
1252 entry.name = darg.dname;
1253 entry.len = darg.dname_len;
1254 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1256 if (IS_ERR_OR_NULL(inode)) {
1257 jbd_debug(1, "Inode %d not found", darg.ino);
1261 old_parent = ext4_iget(sb, darg.parent_ino,
1263 if (IS_ERR_OR_NULL(old_parent)) {
1264 jbd_debug(1, "Dir with inode %d not found", darg.parent_ino);
1269 ret = __ext4_unlink(NULL, old_parent, &entry, inode);
1270 /* -ENOENT ok coz it might not exist anymore. */
1278 static int ext4_fc_replay_link_internal(struct super_block *sb,
1279 struct dentry_info_args *darg,
1280 struct inode *inode)
1282 struct inode *dir = NULL;
1283 struct dentry *dentry_dir = NULL, *dentry_inode = NULL;
1284 struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);
1287 dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);
1289 jbd_debug(1, "Dir with inode %d not found.", darg->parent_ino);
1294 dentry_dir = d_obtain_alias(dir);
1295 if (IS_ERR(dentry_dir)) {
1296 jbd_debug(1, "Failed to obtain dentry");
1301 dentry_inode = d_alloc(dentry_dir, &qstr_dname);
1302 if (!dentry_inode) {
1303 jbd_debug(1, "Inode dentry not created.");
1308 ret = __ext4_link(dir, inode, dentry_inode);
1310 * It's possible that link already existed since data blocks
1311 * for the dir in question got persisted before we crashed OR
1312 * we replayed this tag and crashed before the entire replay
1315 if (ret && ret != -EEXIST) {
1316 jbd_debug(1, "Failed to link\n");
1329 d_drop(dentry_inode);
1336 /* Link replay function */
1337 static int ext4_fc_replay_link(struct super_block *sb, struct ext4_fc_tl *tl)
1339 struct inode *inode;
1340 struct dentry_info_args darg;
1343 tl_to_darg(&darg, tl);
1344 trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,
1345 darg.parent_ino, darg.dname_len);
1347 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1348 if (IS_ERR_OR_NULL(inode)) {
1349 jbd_debug(1, "Inode not found.");
1353 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1359 * Record all the modified inodes during replay. We use this later to setup
1360 * block bitmaps correctly.
1362 static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)
1364 struct ext4_fc_replay_state *state;
1367 state = &EXT4_SB(sb)->s_fc_replay_state;
1368 for (i = 0; i < state->fc_modified_inodes_used; i++)
1369 if (state->fc_modified_inodes[i] == ino)
1371 if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {
1372 state->fc_modified_inodes_size +=
1373 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1374 state->fc_modified_inodes = krealloc(
1375 state->fc_modified_inodes, sizeof(int) *
1376 state->fc_modified_inodes_size,
1378 if (!state->fc_modified_inodes)
1381 state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;
1386 * Inode replay function
1388 static int ext4_fc_replay_inode(struct super_block *sb, struct ext4_fc_tl *tl)
1390 struct ext4_fc_inode *fc_inode;
1391 struct ext4_inode *raw_inode;
1392 struct ext4_inode *raw_fc_inode;
1393 struct inode *inode = NULL;
1394 struct ext4_iloc iloc;
1395 int inode_len, ino, ret, tag = le16_to_cpu(tl->fc_tag);
1396 struct ext4_extent_header *eh;
1398 fc_inode = (struct ext4_fc_inode *)ext4_fc_tag_val(tl);
1400 ino = le32_to_cpu(fc_inode->fc_ino);
1401 trace_ext4_fc_replay(sb, tag, ino, 0, 0);
1403 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1404 if (!IS_ERR_OR_NULL(inode)) {
1405 ext4_ext_clear_bb(inode);
1409 ext4_fc_record_modified_inode(sb, ino);
1411 raw_fc_inode = (struct ext4_inode *)fc_inode->fc_raw_inode;
1412 ret = ext4_get_fc_inode_loc(sb, ino, &iloc);
1416 inode_len = ext4_fc_tag_len(tl) - sizeof(struct ext4_fc_inode);
1417 raw_inode = ext4_raw_inode(&iloc);
1419 memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));
1420 memcpy(&raw_inode->i_generation, &raw_fc_inode->i_generation,
1421 inode_len - offsetof(struct ext4_inode, i_generation));
1422 if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {
1423 eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);
1424 if (eh->eh_magic != EXT4_EXT_MAGIC) {
1425 memset(eh, 0, sizeof(*eh));
1426 eh->eh_magic = EXT4_EXT_MAGIC;
1427 eh->eh_max = cpu_to_le16(
1428 (sizeof(raw_inode->i_block) -
1429 sizeof(struct ext4_extent_header))
1430 / sizeof(struct ext4_extent));
1432 } else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {
1433 memcpy(raw_inode->i_block, raw_fc_inode->i_block,
1434 sizeof(raw_inode->i_block));
1437 /* Immediately update the inode on disk. */
1438 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1441 ret = sync_dirty_buffer(iloc.bh);
1444 ret = ext4_mark_inode_used(sb, ino);
1448 /* Given that we just wrote the inode on disk, this SHOULD succeed. */
1449 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1450 if (IS_ERR_OR_NULL(inode)) {
1451 jbd_debug(1, "Inode not found.");
1452 return -EFSCORRUPTED;
1456 * Our allocator could have made different decisions than before
1457 * crashing. This should be fixed but until then, we calculate
1458 * the number of blocks the inode.
1460 ext4_ext_replay_set_iblocks(inode);
1462 inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);
1463 ext4_reset_inode_seed(inode);
1465 ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));
1466 ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1467 sync_dirty_buffer(iloc.bh);
1472 blkdev_issue_flush(sb->s_bdev, GFP_KERNEL);
1478 * Dentry create replay function.
1480 * EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the
1481 * inode for which we are trying to create a dentry here, should already have
1482 * been replayed before we start here.
1484 static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl)
1487 struct inode *inode = NULL;
1488 struct inode *dir = NULL;
1489 struct dentry_info_args darg;
1491 tl_to_darg(&darg, tl);
1493 trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,
1494 darg.parent_ino, darg.dname_len);
1496 /* This takes care of update group descriptor and other metadata */
1497 ret = ext4_mark_inode_used(sb, darg.ino);
1501 inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1502 if (IS_ERR_OR_NULL(inode)) {
1503 jbd_debug(1, "inode %d not found.", darg.ino);
1509 if (S_ISDIR(inode->i_mode)) {
1511 * If we are creating a directory, we need to make sure that the
1512 * dot and dot dot dirents are setup properly.
1514 dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);
1515 if (IS_ERR_OR_NULL(dir)) {
1516 jbd_debug(1, "Dir %d not found.", darg.ino);
1519 ret = ext4_init_new_dir(NULL, dir, inode);
1526 ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1529 set_nlink(inode, 1);
1530 ext4_mark_inode_dirty(NULL, inode);
1538 * Record physical disk regions which are in use as per fast commit area. Our
1539 * simple replay phase allocator excludes these regions from allocation.
1541 static int ext4_fc_record_regions(struct super_block *sb, int ino,
1542 ext4_lblk_t lblk, ext4_fsblk_t pblk, int len)
1544 struct ext4_fc_replay_state *state;
1545 struct ext4_fc_alloc_region *region;
1547 state = &EXT4_SB(sb)->s_fc_replay_state;
1548 if (state->fc_regions_used == state->fc_regions_size) {
1549 state->fc_regions_size +=
1550 EXT4_FC_REPLAY_REALLOC_INCREMENT;
1551 state->fc_regions = krealloc(
1553 state->fc_regions_size *
1554 sizeof(struct ext4_fc_alloc_region),
1556 if (!state->fc_regions)
1559 region = &state->fc_regions[state->fc_regions_used++];
1561 region->lblk = lblk;
1562 region->pblk = pblk;
1568 /* Replay add range tag */
1569 static int ext4_fc_replay_add_range(struct super_block *sb,
1570 struct ext4_fc_tl *tl)
1572 struct ext4_fc_add_range *fc_add_ex;
1573 struct ext4_extent newex, *ex;
1574 struct inode *inode;
1575 ext4_lblk_t start, cur;
1577 ext4_fsblk_t start_pblk;
1578 struct ext4_map_blocks map;
1579 struct ext4_ext_path *path = NULL;
1582 fc_add_ex = (struct ext4_fc_add_range *)ext4_fc_tag_val(tl);
1583 ex = (struct ext4_extent *)&fc_add_ex->fc_ex;
1585 trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,
1586 le32_to_cpu(fc_add_ex->fc_ino), le32_to_cpu(ex->ee_block),
1587 ext4_ext_get_actual_len(ex));
1589 inode = ext4_iget(sb, le32_to_cpu(fc_add_ex->fc_ino),
1591 if (IS_ERR_OR_NULL(inode)) {
1592 jbd_debug(1, "Inode not found.");
1596 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1598 start = le32_to_cpu(ex->ee_block);
1599 start_pblk = ext4_ext_pblock(ex);
1600 len = ext4_ext_get_actual_len(ex);
1604 jbd_debug(1, "ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",
1605 start, start_pblk, len, ext4_ext_is_unwritten(ex),
1608 while (remaining > 0) {
1610 map.m_len = remaining;
1612 ret = ext4_map_blocks(NULL, inode, &map, 0);
1620 /* Range is not mapped */
1621 path = ext4_find_extent(inode, cur, NULL, 0);
1626 memset(&newex, 0, sizeof(newex));
1627 newex.ee_block = cpu_to_le32(cur);
1628 ext4_ext_store_pblock(
1629 &newex, start_pblk + cur - start);
1630 newex.ee_len = cpu_to_le16(map.m_len);
1631 if (ext4_ext_is_unwritten(ex))
1632 ext4_ext_mark_unwritten(&newex);
1633 down_write(&EXT4_I(inode)->i_data_sem);
1634 ret = ext4_ext_insert_extent(
1635 NULL, inode, &path, &newex, 0);
1636 up_write((&EXT4_I(inode)->i_data_sem));
1637 ext4_ext_drop_refs(path);
1646 if (start_pblk + cur - start != map.m_pblk) {
1648 * Logical to physical mapping changed. This can happen
1649 * if this range was removed and then reallocated to
1650 * map to new physical blocks during a fast commit.
1652 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1653 ext4_ext_is_unwritten(ex),
1654 start_pblk + cur - start);
1660 * Mark the old blocks as free since they aren't used
1661 * anymore. We maintain an array of all the modified
1662 * inodes. In case these blocks are still used at either
1663 * a different logical range in the same inode or in
1664 * some different inode, we will mark them as allocated
1665 * at the end of the FC replay using our array of
1668 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1672 /* Range is mapped and needs a state change */
1673 jbd_debug(1, "Converting from %d to %d %lld",
1674 map.m_flags & EXT4_MAP_UNWRITTEN,
1675 ext4_ext_is_unwritten(ex), map.m_pblk);
1676 ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1677 ext4_ext_is_unwritten(ex), map.m_pblk);
1683 * We may have split the extent tree while toggling the state.
1684 * Try to shrink the extent tree now.
1686 ext4_ext_replay_shrink_inode(inode, start + len);
1689 remaining -= map.m_len;
1691 ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>
1692 sb->s_blocksize_bits);
1697 /* Replay DEL_RANGE tag */
1699 ext4_fc_replay_del_range(struct super_block *sb, struct ext4_fc_tl *tl)
1701 struct inode *inode;
1702 struct ext4_fc_del_range *lrange;
1703 struct ext4_map_blocks map;
1704 ext4_lblk_t cur, remaining;
1707 lrange = (struct ext4_fc_del_range *)ext4_fc_tag_val(tl);
1708 cur = le32_to_cpu(lrange->fc_lblk);
1709 remaining = le32_to_cpu(lrange->fc_len);
1711 trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,
1712 le32_to_cpu(lrange->fc_ino), cur, remaining);
1714 inode = ext4_iget(sb, le32_to_cpu(lrange->fc_ino), EXT4_IGET_NORMAL);
1715 if (IS_ERR_OR_NULL(inode)) {
1716 jbd_debug(1, "Inode %d not found", le32_to_cpu(lrange->fc_ino));
1720 ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1722 jbd_debug(1, "DEL_RANGE, inode %ld, lblk %d, len %d\n",
1723 inode->i_ino, le32_to_cpu(lrange->fc_lblk),
1724 le32_to_cpu(lrange->fc_len));
1725 while (remaining > 0) {
1727 map.m_len = remaining;
1729 ret = ext4_map_blocks(NULL, inode, &map, 0);
1737 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1739 remaining -= map.m_len;
1744 ret = ext4_punch_hole(inode,
1745 le32_to_cpu(lrange->fc_lblk) << sb->s_blocksize_bits,
1746 le32_to_cpu(lrange->fc_len) << sb->s_blocksize_bits);
1748 jbd_debug(1, "ext4_punch_hole returned %d", ret);
1749 ext4_ext_replay_shrink_inode(inode,
1750 i_size_read(inode) >> sb->s_blocksize_bits);
1751 ext4_mark_inode_dirty(NULL, inode);
1757 static inline const char *tag2str(u16 tag)
1760 case EXT4_FC_TAG_LINK:
1761 return "TAG_ADD_ENTRY";
1762 case EXT4_FC_TAG_UNLINK:
1763 return "TAG_DEL_ENTRY";
1764 case EXT4_FC_TAG_ADD_RANGE:
1765 return "TAG_ADD_RANGE";
1766 case EXT4_FC_TAG_CREAT:
1767 return "TAG_CREAT_DENTRY";
1768 case EXT4_FC_TAG_DEL_RANGE:
1769 return "TAG_DEL_RANGE";
1770 case EXT4_FC_TAG_INODE:
1772 case EXT4_FC_TAG_PAD:
1774 case EXT4_FC_TAG_TAIL:
1776 case EXT4_FC_TAG_HEAD:
1783 static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)
1785 struct ext4_fc_replay_state *state;
1786 struct inode *inode;
1787 struct ext4_ext_path *path = NULL;
1788 struct ext4_map_blocks map;
1790 ext4_lblk_t cur, end;
1792 state = &EXT4_SB(sb)->s_fc_replay_state;
1793 for (i = 0; i < state->fc_modified_inodes_used; i++) {
1794 inode = ext4_iget(sb, state->fc_modified_inodes[i],
1796 if (IS_ERR_OR_NULL(inode)) {
1797 jbd_debug(1, "Inode %d not found.",
1798 state->fc_modified_inodes[i]);
1802 end = EXT_MAX_BLOCKS;
1805 map.m_len = end - cur;
1807 ret = ext4_map_blocks(NULL, inode, &map, 0);
1812 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
1813 if (!IS_ERR_OR_NULL(path)) {
1814 for (j = 0; j < path->p_depth; j++)
1815 ext4_mb_mark_bb(inode->i_sb,
1816 path[j].p_block, 1, 1);
1817 ext4_ext_drop_refs(path);
1821 ext4_mb_mark_bb(inode->i_sb, map.m_pblk,
1824 cur = cur + (map.m_len ? map.m_len : 1);
1832 * Check if block is in excluded regions for block allocation. The simple
1833 * allocator that runs during replay phase is calls this function to see
1834 * if it is okay to use a block.
1836 bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)
1839 struct ext4_fc_replay_state *state;
1841 state = &EXT4_SB(sb)->s_fc_replay_state;
1842 for (i = 0; i < state->fc_regions_valid; i++) {
1843 if (state->fc_regions[i].ino == 0 ||
1844 state->fc_regions[i].len == 0)
1846 if (blk >= state->fc_regions[i].pblk &&
1847 blk < state->fc_regions[i].pblk + state->fc_regions[i].len)
1853 /* Cleanup function called after replay */
1854 void ext4_fc_replay_cleanup(struct super_block *sb)
1856 struct ext4_sb_info *sbi = EXT4_SB(sb);
1858 sbi->s_mount_state &= ~EXT4_FC_REPLAY;
1859 kfree(sbi->s_fc_replay_state.fc_regions);
1860 kfree(sbi->s_fc_replay_state.fc_modified_inodes);
1864 * Recovery Scan phase handler
1866 * This function is called during the scan phase and is responsible
1867 * for doing following things:
1868 * - Make sure the fast commit area has valid tags for replay
1869 * - Count number of tags that need to be replayed by the replay handler
1871 * - Create a list of excluded blocks for allocation during replay phase
1873 * This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is
1874 * incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP
1875 * to indicate that scan has finished and JBD2 can now start replay phase.
1876 * It returns a negative error to indicate that there was an error. At the end
1877 * of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set
1878 * to indicate the number of tags that need to replayed during the replay phase.
1880 static int ext4_fc_replay_scan(journal_t *journal,
1881 struct buffer_head *bh, int off,
1884 struct super_block *sb = journal->j_private;
1885 struct ext4_sb_info *sbi = EXT4_SB(sb);
1886 struct ext4_fc_replay_state *state;
1887 int ret = JBD2_FC_REPLAY_CONTINUE;
1888 struct ext4_fc_add_range *ext;
1889 struct ext4_fc_tl *tl;
1890 struct ext4_fc_tail *tail;
1892 struct ext4_fc_head *head;
1893 struct ext4_extent *ex;
1895 state = &sbi->s_fc_replay_state;
1897 start = (u8 *)bh->b_data;
1898 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
1900 if (state->fc_replay_expected_off == 0) {
1901 state->fc_cur_tag = 0;
1902 state->fc_replay_num_tags = 0;
1904 state->fc_regions = NULL;
1905 state->fc_regions_valid = state->fc_regions_used =
1906 state->fc_regions_size = 0;
1907 /* Check if we can stop early */
1908 if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
1909 != EXT4_FC_TAG_HEAD)
1913 if (off != state->fc_replay_expected_off) {
1914 ret = -EFSCORRUPTED;
1918 state->fc_replay_expected_off++;
1919 fc_for_each_tl(start, end, tl) {
1920 jbd_debug(3, "Scan phase, tag:%s, blk %lld\n",
1921 tag2str(le16_to_cpu(tl->fc_tag)), bh->b_blocknr);
1922 switch (le16_to_cpu(tl->fc_tag)) {
1923 case EXT4_FC_TAG_ADD_RANGE:
1924 ext = (struct ext4_fc_add_range *)ext4_fc_tag_val(tl);
1925 ex = (struct ext4_extent *)&ext->fc_ex;
1926 ret = ext4_fc_record_regions(sb,
1927 le32_to_cpu(ext->fc_ino),
1928 le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
1929 ext4_ext_get_actual_len(ex));
1932 ret = JBD2_FC_REPLAY_CONTINUE;
1934 case EXT4_FC_TAG_DEL_RANGE:
1935 case EXT4_FC_TAG_LINK:
1936 case EXT4_FC_TAG_UNLINK:
1937 case EXT4_FC_TAG_CREAT:
1938 case EXT4_FC_TAG_INODE:
1939 case EXT4_FC_TAG_PAD:
1940 state->fc_cur_tag++;
1941 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1942 sizeof(*tl) + ext4_fc_tag_len(tl));
1944 case EXT4_FC_TAG_TAIL:
1945 state->fc_cur_tag++;
1946 tail = (struct ext4_fc_tail *)ext4_fc_tag_val(tl);
1947 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1949 offsetof(struct ext4_fc_tail,
1951 if (le32_to_cpu(tail->fc_tid) == expected_tid &&
1952 le32_to_cpu(tail->fc_crc) == state->fc_crc) {
1953 state->fc_replay_num_tags = state->fc_cur_tag;
1954 state->fc_regions_valid =
1955 state->fc_regions_used;
1957 ret = state->fc_replay_num_tags ?
1958 JBD2_FC_REPLAY_STOP : -EFSBADCRC;
1962 case EXT4_FC_TAG_HEAD:
1963 head = (struct ext4_fc_head *)ext4_fc_tag_val(tl);
1964 if (le32_to_cpu(head->fc_features) &
1965 ~EXT4_FC_SUPPORTED_FEATURES) {
1969 if (le32_to_cpu(head->fc_tid) != expected_tid) {
1970 ret = JBD2_FC_REPLAY_STOP;
1973 state->fc_cur_tag++;
1974 state->fc_crc = ext4_chksum(sbi, state->fc_crc, tl,
1975 sizeof(*tl) + ext4_fc_tag_len(tl));
1978 ret = state->fc_replay_num_tags ?
1979 JBD2_FC_REPLAY_STOP : -ECANCELED;
1981 if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
1986 trace_ext4_fc_replay_scan(sb, ret, off);
1991 * Main recovery path entry point.
1992 * The meaning of return codes is similar as above.
1994 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
1995 enum passtype pass, int off, tid_t expected_tid)
1997 struct super_block *sb = journal->j_private;
1998 struct ext4_sb_info *sbi = EXT4_SB(sb);
1999 struct ext4_fc_tl *tl;
2001 int ret = JBD2_FC_REPLAY_CONTINUE;
2002 struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;
2003 struct ext4_fc_tail *tail;
2005 if (pass == PASS_SCAN) {
2006 state->fc_current_pass = PASS_SCAN;
2007 return ext4_fc_replay_scan(journal, bh, off, expected_tid);
2010 if (state->fc_current_pass != pass) {
2011 state->fc_current_pass = pass;
2012 sbi->s_mount_state |= EXT4_FC_REPLAY;
2014 if (!sbi->s_fc_replay_state.fc_replay_num_tags) {
2015 jbd_debug(1, "Replay stops\n");
2016 ext4_fc_set_bitmaps_and_counters(sb);
2020 #ifdef CONFIG_EXT4_DEBUG
2021 if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {
2022 pr_warn("Dropping fc block %d because max_replay set\n", off);
2023 return JBD2_FC_REPLAY_STOP;
2027 start = (u8 *)bh->b_data;
2028 end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
2030 fc_for_each_tl(start, end, tl) {
2031 if (state->fc_replay_num_tags == 0) {
2032 ret = JBD2_FC_REPLAY_STOP;
2033 ext4_fc_set_bitmaps_and_counters(sb);
2036 jbd_debug(3, "Replay phase, tag:%s\n",
2037 tag2str(le16_to_cpu(tl->fc_tag)));
2038 state->fc_replay_num_tags--;
2039 switch (le16_to_cpu(tl->fc_tag)) {
2040 case EXT4_FC_TAG_LINK:
2041 ret = ext4_fc_replay_link(sb, tl);
2043 case EXT4_FC_TAG_UNLINK:
2044 ret = ext4_fc_replay_unlink(sb, tl);
2046 case EXT4_FC_TAG_ADD_RANGE:
2047 ret = ext4_fc_replay_add_range(sb, tl);
2049 case EXT4_FC_TAG_CREAT:
2050 ret = ext4_fc_replay_create(sb, tl);
2052 case EXT4_FC_TAG_DEL_RANGE:
2053 ret = ext4_fc_replay_del_range(sb, tl);
2055 case EXT4_FC_TAG_INODE:
2056 ret = ext4_fc_replay_inode(sb, tl);
2058 case EXT4_FC_TAG_PAD:
2059 trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,
2060 ext4_fc_tag_len(tl), 0);
2062 case EXT4_FC_TAG_TAIL:
2063 trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL, 0,
2064 ext4_fc_tag_len(tl), 0);
2065 tail = (struct ext4_fc_tail *)ext4_fc_tag_val(tl);
2066 WARN_ON(le32_to_cpu(tail->fc_tid) != expected_tid);
2068 case EXT4_FC_TAG_HEAD:
2071 trace_ext4_fc_replay(sb, le16_to_cpu(tl->fc_tag), 0,
2072 ext4_fc_tag_len(tl), 0);
2078 ret = JBD2_FC_REPLAY_CONTINUE;
2083 void ext4_fc_init(struct super_block *sb, journal_t *journal)
2086 * We set replay callback even if fast commit disabled because we may
2087 * could still have fast commit blocks that need to be replayed even if
2088 * fast commit has now been turned off.
2090 journal->j_fc_replay_callback = ext4_fc_replay;
2091 if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
2093 journal->j_fc_cleanup_callback = ext4_fc_cleanup;
2096 const char *fc_ineligible_reasons[] = {
2097 "Extended attributes changed",
2099 "Journal flag changed",
2100 "Insufficient memory",
2108 int ext4_fc_info_show(struct seq_file *seq, void *v)
2110 struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);
2111 struct ext4_fc_stats *stats = &sbi->s_fc_stats;
2114 if (v != SEQ_START_TOKEN)
2118 "fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",
2119 stats->fc_num_commits, stats->fc_ineligible_commits,
2121 div_u64(sbi->s_fc_avg_commit_time, 1000));
2122 seq_puts(seq, "Ineligible reasons:\n");
2123 for (i = 0; i < EXT4_FC_REASON_MAX; i++)
2124 seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],
2125 stats->fc_ineligible_reason_count[i]);
2130 int __init ext4_fc_init_dentry_cache(void)
2132 ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,
2133 SLAB_RECLAIM_ACCOUNT);
2135 if (ext4_fc_dentry_cachep == NULL)