2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/slab.h>
21 #include <linux/sched.h>
22 #include <linux/writeback.h>
23 #include <linux/pagemap.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
28 #include "transaction.h"
31 #include "inode-map.h"
33 #include "dev-replace.h"
36 #define BTRFS_ROOT_TRANS_TAG 0
38 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
39 [TRANS_STATE_RUNNING] = 0U,
40 [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE |
42 [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE |
45 [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE |
49 [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE |
54 [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE |
61 void btrfs_put_transaction(struct btrfs_transaction *transaction)
63 WARN_ON(refcount_read(&transaction->use_count) == 0);
64 if (refcount_dec_and_test(&transaction->use_count)) {
65 BUG_ON(!list_empty(&transaction->list));
66 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
67 if (transaction->delayed_refs.pending_csums)
68 btrfs_err(transaction->fs_info,
69 "pending csums is %llu",
70 transaction->delayed_refs.pending_csums);
71 while (!list_empty(&transaction->pending_chunks)) {
72 struct extent_map *em;
74 em = list_first_entry(&transaction->pending_chunks,
75 struct extent_map, list);
76 list_del_init(&em->list);
80 * If any block groups are found in ->deleted_bgs then it's
81 * because the transaction was aborted and a commit did not
82 * happen (things failed before writing the new superblock
83 * and calling btrfs_finish_extent_commit()), so we can not
84 * discard the physical locations of the block groups.
86 while (!list_empty(&transaction->deleted_bgs)) {
87 struct btrfs_block_group_cache *cache;
89 cache = list_first_entry(&transaction->deleted_bgs,
90 struct btrfs_block_group_cache,
92 list_del_init(&cache->bg_list);
93 btrfs_put_block_group_trimming(cache);
94 btrfs_put_block_group(cache);
100 static void clear_btree_io_tree(struct extent_io_tree *tree)
102 spin_lock(&tree->lock);
104 * Do a single barrier for the waitqueue_active check here, the state
105 * of the waitqueue should not change once clear_btree_io_tree is
109 while (!RB_EMPTY_ROOT(&tree->state)) {
110 struct rb_node *node;
111 struct extent_state *state;
113 node = rb_first(&tree->state);
114 state = rb_entry(node, struct extent_state, rb_node);
115 rb_erase(&state->rb_node, &tree->state);
116 RB_CLEAR_NODE(&state->rb_node);
118 * btree io trees aren't supposed to have tasks waiting for
119 * changes in the flags of extent states ever.
121 ASSERT(!waitqueue_active(&state->wq));
122 free_extent_state(state);
124 cond_resched_lock(&tree->lock);
126 spin_unlock(&tree->lock);
129 static noinline void switch_commit_roots(struct btrfs_transaction *trans,
130 struct btrfs_fs_info *fs_info)
132 struct btrfs_root *root, *tmp;
134 down_write(&fs_info->commit_root_sem);
135 list_for_each_entry_safe(root, tmp, &trans->switch_commits,
137 list_del_init(&root->dirty_list);
138 free_extent_buffer(root->commit_root);
139 root->commit_root = btrfs_root_node(root);
140 if (is_fstree(root->objectid))
141 btrfs_unpin_free_ino(root);
142 clear_btree_io_tree(&root->dirty_log_pages);
145 /* We can free old roots now. */
146 spin_lock(&trans->dropped_roots_lock);
147 while (!list_empty(&trans->dropped_roots)) {
148 root = list_first_entry(&trans->dropped_roots,
149 struct btrfs_root, root_list);
150 list_del_init(&root->root_list);
151 spin_unlock(&trans->dropped_roots_lock);
152 btrfs_drop_and_free_fs_root(fs_info, root);
153 spin_lock(&trans->dropped_roots_lock);
155 spin_unlock(&trans->dropped_roots_lock);
156 up_write(&fs_info->commit_root_sem);
159 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
162 if (type & TRANS_EXTWRITERS)
163 atomic_inc(&trans->num_extwriters);
166 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
169 if (type & TRANS_EXTWRITERS)
170 atomic_dec(&trans->num_extwriters);
173 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
176 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
179 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
181 return atomic_read(&trans->num_extwriters);
185 * either allocate a new transaction or hop into the existing one
187 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
190 struct btrfs_transaction *cur_trans;
192 spin_lock(&fs_info->trans_lock);
194 /* The file system has been taken offline. No new transactions. */
195 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
196 spin_unlock(&fs_info->trans_lock);
200 cur_trans = fs_info->running_transaction;
202 if (cur_trans->aborted) {
203 spin_unlock(&fs_info->trans_lock);
204 return cur_trans->aborted;
206 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
207 spin_unlock(&fs_info->trans_lock);
210 refcount_inc(&cur_trans->use_count);
211 atomic_inc(&cur_trans->num_writers);
212 extwriter_counter_inc(cur_trans, type);
213 spin_unlock(&fs_info->trans_lock);
216 spin_unlock(&fs_info->trans_lock);
219 * If we are ATTACH, we just want to catch the current transaction,
220 * and commit it. If there is no transaction, just return ENOENT.
222 if (type == TRANS_ATTACH)
226 * JOIN_NOLOCK only happens during the transaction commit, so
227 * it is impossible that ->running_transaction is NULL
229 BUG_ON(type == TRANS_JOIN_NOLOCK);
231 cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
235 spin_lock(&fs_info->trans_lock);
236 if (fs_info->running_transaction) {
238 * someone started a transaction after we unlocked. Make sure
239 * to redo the checks above
243 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
244 spin_unlock(&fs_info->trans_lock);
249 cur_trans->fs_info = fs_info;
250 atomic_set(&cur_trans->num_writers, 1);
251 extwriter_counter_init(cur_trans, type);
252 init_waitqueue_head(&cur_trans->writer_wait);
253 init_waitqueue_head(&cur_trans->commit_wait);
254 init_waitqueue_head(&cur_trans->pending_wait);
255 cur_trans->state = TRANS_STATE_RUNNING;
257 * One for this trans handle, one so it will live on until we
258 * commit the transaction.
260 refcount_set(&cur_trans->use_count, 2);
261 atomic_set(&cur_trans->pending_ordered, 0);
262 cur_trans->flags = 0;
263 cur_trans->start_time = get_seconds();
265 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
267 cur_trans->delayed_refs.href_root = RB_ROOT;
268 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
269 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
272 * although the tree mod log is per file system and not per transaction,
273 * the log must never go across transaction boundaries.
276 if (!list_empty(&fs_info->tree_mod_seq_list))
277 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
278 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
279 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
280 atomic64_set(&fs_info->tree_mod_seq, 0);
282 spin_lock_init(&cur_trans->delayed_refs.lock);
284 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
285 INIT_LIST_HEAD(&cur_trans->pending_chunks);
286 INIT_LIST_HEAD(&cur_trans->switch_commits);
287 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
288 INIT_LIST_HEAD(&cur_trans->io_bgs);
289 INIT_LIST_HEAD(&cur_trans->dropped_roots);
290 mutex_init(&cur_trans->cache_write_mutex);
291 cur_trans->num_dirty_bgs = 0;
292 spin_lock_init(&cur_trans->dirty_bgs_lock);
293 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
294 spin_lock_init(&cur_trans->dropped_roots_lock);
295 list_add_tail(&cur_trans->list, &fs_info->trans_list);
296 extent_io_tree_init(&cur_trans->dirty_pages,
297 fs_info->btree_inode);
298 fs_info->generation++;
299 cur_trans->transid = fs_info->generation;
300 fs_info->running_transaction = cur_trans;
301 cur_trans->aborted = 0;
302 spin_unlock(&fs_info->trans_lock);
308 * this does all the record keeping required to make sure that a reference
309 * counted root is properly recorded in a given transaction. This is required
310 * to make sure the old root from before we joined the transaction is deleted
311 * when the transaction commits
313 static int record_root_in_trans(struct btrfs_trans_handle *trans,
314 struct btrfs_root *root,
317 struct btrfs_fs_info *fs_info = root->fs_info;
319 if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
320 root->last_trans < trans->transid) || force) {
321 WARN_ON(root == fs_info->extent_root);
322 WARN_ON(!force && root->commit_root != root->node);
325 * see below for IN_TRANS_SETUP usage rules
326 * we have the reloc mutex held now, so there
327 * is only one writer in this function
329 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
331 /* make sure readers find IN_TRANS_SETUP before
332 * they find our root->last_trans update
336 spin_lock(&fs_info->fs_roots_radix_lock);
337 if (root->last_trans == trans->transid && !force) {
338 spin_unlock(&fs_info->fs_roots_radix_lock);
341 radix_tree_tag_set(&fs_info->fs_roots_radix,
342 (unsigned long)root->root_key.objectid,
343 BTRFS_ROOT_TRANS_TAG);
344 spin_unlock(&fs_info->fs_roots_radix_lock);
345 root->last_trans = trans->transid;
347 /* this is pretty tricky. We don't want to
348 * take the relocation lock in btrfs_record_root_in_trans
349 * unless we're really doing the first setup for this root in
352 * Normally we'd use root->last_trans as a flag to decide
353 * if we want to take the expensive mutex.
355 * But, we have to set root->last_trans before we
356 * init the relocation root, otherwise, we trip over warnings
357 * in ctree.c. The solution used here is to flag ourselves
358 * with root IN_TRANS_SETUP. When this is 1, we're still
359 * fixing up the reloc trees and everyone must wait.
361 * When this is zero, they can trust root->last_trans and fly
362 * through btrfs_record_root_in_trans without having to take the
363 * lock. smp_wmb() makes sure that all the writes above are
364 * done before we pop in the zero below
366 btrfs_init_reloc_root(trans, root);
367 smp_mb__before_atomic();
368 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
374 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
375 struct btrfs_root *root)
377 struct btrfs_fs_info *fs_info = root->fs_info;
378 struct btrfs_transaction *cur_trans = trans->transaction;
380 /* Add ourselves to the transaction dropped list */
381 spin_lock(&cur_trans->dropped_roots_lock);
382 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
383 spin_unlock(&cur_trans->dropped_roots_lock);
385 /* Make sure we don't try to update the root at commit time */
386 spin_lock(&fs_info->fs_roots_radix_lock);
387 radix_tree_tag_clear(&fs_info->fs_roots_radix,
388 (unsigned long)root->root_key.objectid,
389 BTRFS_ROOT_TRANS_TAG);
390 spin_unlock(&fs_info->fs_roots_radix_lock);
393 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
394 struct btrfs_root *root)
396 struct btrfs_fs_info *fs_info = root->fs_info;
398 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
402 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
406 if (root->last_trans == trans->transid &&
407 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
410 mutex_lock(&fs_info->reloc_mutex);
411 record_root_in_trans(trans, root, 0);
412 mutex_unlock(&fs_info->reloc_mutex);
417 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
419 return (trans->state >= TRANS_STATE_BLOCKED &&
420 trans->state < TRANS_STATE_UNBLOCKED &&
424 /* wait for commit against the current transaction to become unblocked
425 * when this is done, it is safe to start a new transaction, but the current
426 * transaction might not be fully on disk.
428 static void wait_current_trans(struct btrfs_fs_info *fs_info)
430 struct btrfs_transaction *cur_trans;
432 spin_lock(&fs_info->trans_lock);
433 cur_trans = fs_info->running_transaction;
434 if (cur_trans && is_transaction_blocked(cur_trans)) {
435 refcount_inc(&cur_trans->use_count);
436 spin_unlock(&fs_info->trans_lock);
438 wait_event(fs_info->transaction_wait,
439 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
441 btrfs_put_transaction(cur_trans);
443 spin_unlock(&fs_info->trans_lock);
447 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
449 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
452 if (type == TRANS_USERSPACE)
455 if (type == TRANS_START &&
456 !atomic_read(&fs_info->open_ioctl_trans))
462 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
464 struct btrfs_fs_info *fs_info = root->fs_info;
466 if (!fs_info->reloc_ctl ||
467 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
468 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
475 static struct btrfs_trans_handle *
476 start_transaction(struct btrfs_root *root, unsigned int num_items,
477 unsigned int type, enum btrfs_reserve_flush_enum flush,
478 bool enforce_qgroups)
480 struct btrfs_fs_info *fs_info = root->fs_info;
482 struct btrfs_trans_handle *h;
483 struct btrfs_transaction *cur_trans;
485 u64 qgroup_reserved = 0;
486 bool reloc_reserved = false;
489 /* Send isn't supposed to start transactions. */
490 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
492 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
493 return ERR_PTR(-EROFS);
495 if (current->journal_info) {
496 WARN_ON(type & TRANS_EXTWRITERS);
497 h = current->journal_info;
499 WARN_ON(h->use_count > 2);
500 h->orig_rsv = h->block_rsv;
506 * Do the reservation before we join the transaction so we can do all
507 * the appropriate flushing if need be.
509 if (num_items && root != fs_info->chunk_root) {
510 qgroup_reserved = num_items * fs_info->nodesize;
511 ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved,
516 num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items);
518 * Do the reservation for the relocation root creation
520 if (need_reserve_reloc_root(root)) {
521 num_bytes += fs_info->nodesize;
522 reloc_reserved = true;
525 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
531 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
538 * If we are JOIN_NOLOCK we're already committing a transaction and
539 * waiting on this guy, so we don't need to do the sb_start_intwrite
540 * because we're already holding a ref. We need this because we could
541 * have raced in and did an fsync() on a file which can kick a commit
542 * and then we deadlock with somebody doing a freeze.
544 * If we are ATTACH, it means we just want to catch the current
545 * transaction and commit it, so we needn't do sb_start_intwrite().
547 if (type & __TRANS_FREEZABLE)
548 sb_start_intwrite(fs_info->sb);
550 if (may_wait_transaction(fs_info, type))
551 wait_current_trans(fs_info);
554 ret = join_transaction(fs_info, type);
556 wait_current_trans(fs_info);
557 if (unlikely(type == TRANS_ATTACH))
560 } while (ret == -EBUSY);
565 cur_trans = fs_info->running_transaction;
567 h->transid = cur_trans->transid;
568 h->transaction = cur_trans;
571 h->fs_info = root->fs_info;
574 h->can_flush_pending_bgs = true;
575 INIT_LIST_HEAD(&h->new_bgs);
578 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
579 may_wait_transaction(fs_info, type)) {
580 current->journal_info = h;
581 btrfs_commit_transaction(h);
586 trace_btrfs_space_reservation(fs_info, "transaction",
587 h->transid, num_bytes, 1);
588 h->block_rsv = &fs_info->trans_block_rsv;
589 h->bytes_reserved = num_bytes;
590 h->reloc_reserved = reloc_reserved;
594 btrfs_record_root_in_trans(h, root);
596 if (!current->journal_info && type != TRANS_USERSPACE)
597 current->journal_info = h;
601 if (type & __TRANS_FREEZABLE)
602 sb_end_intwrite(fs_info->sb);
603 kmem_cache_free(btrfs_trans_handle_cachep, h);
606 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
609 btrfs_qgroup_free_meta(root, qgroup_reserved);
613 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
614 unsigned int num_items)
616 return start_transaction(root, num_items, TRANS_START,
617 BTRFS_RESERVE_FLUSH_ALL, true);
620 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
621 struct btrfs_root *root,
622 unsigned int num_items,
625 struct btrfs_fs_info *fs_info = root->fs_info;
626 struct btrfs_trans_handle *trans;
631 * We have two callers: unlink and block group removal. The
632 * former should succeed even if we will temporarily exceed
633 * quota and the latter operates on the extent root so
634 * qgroup enforcement is ignored anyway.
636 trans = start_transaction(root, num_items, TRANS_START,
637 BTRFS_RESERVE_FLUSH_ALL, false);
638 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
641 trans = btrfs_start_transaction(root, 0);
645 num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items);
646 ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv,
647 num_bytes, min_factor);
649 btrfs_end_transaction(trans);
653 trans->block_rsv = &fs_info->trans_block_rsv;
654 trans->bytes_reserved = num_bytes;
655 trace_btrfs_space_reservation(fs_info, "transaction",
656 trans->transid, num_bytes, 1);
661 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
662 struct btrfs_root *root,
663 unsigned int num_items)
665 return start_transaction(root, num_items, TRANS_START,
666 BTRFS_RESERVE_FLUSH_LIMIT, true);
669 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
671 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
675 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
677 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
678 BTRFS_RESERVE_NO_FLUSH, true);
681 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
683 return start_transaction(root, 0, TRANS_USERSPACE,
684 BTRFS_RESERVE_NO_FLUSH, true);
688 * btrfs_attach_transaction() - catch the running transaction
690 * It is used when we want to commit the current the transaction, but
691 * don't want to start a new one.
693 * Note: If this function return -ENOENT, it just means there is no
694 * running transaction. But it is possible that the inactive transaction
695 * is still in the memory, not fully on disk. If you hope there is no
696 * inactive transaction in the fs when -ENOENT is returned, you should
698 * btrfs_attach_transaction_barrier()
700 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
702 return start_transaction(root, 0, TRANS_ATTACH,
703 BTRFS_RESERVE_NO_FLUSH, true);
707 * btrfs_attach_transaction_barrier() - catch the running transaction
709 * It is similar to the above function, the differentia is this one
710 * will wait for all the inactive transactions until they fully
713 struct btrfs_trans_handle *
714 btrfs_attach_transaction_barrier(struct btrfs_root *root)
716 struct btrfs_trans_handle *trans;
718 trans = start_transaction(root, 0, TRANS_ATTACH,
719 BTRFS_RESERVE_NO_FLUSH, true);
720 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
721 btrfs_wait_for_commit(root->fs_info, 0);
726 /* wait for a transaction commit to be fully complete */
727 static noinline void wait_for_commit(struct btrfs_transaction *commit)
729 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
732 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
734 struct btrfs_transaction *cur_trans = NULL, *t;
738 if (transid <= fs_info->last_trans_committed)
741 /* find specified transaction */
742 spin_lock(&fs_info->trans_lock);
743 list_for_each_entry(t, &fs_info->trans_list, list) {
744 if (t->transid == transid) {
746 refcount_inc(&cur_trans->use_count);
750 if (t->transid > transid) {
755 spin_unlock(&fs_info->trans_lock);
758 * The specified transaction doesn't exist, or we
759 * raced with btrfs_commit_transaction
762 if (transid > fs_info->last_trans_committed)
767 /* find newest transaction that is committing | committed */
768 spin_lock(&fs_info->trans_lock);
769 list_for_each_entry_reverse(t, &fs_info->trans_list,
771 if (t->state >= TRANS_STATE_COMMIT_START) {
772 if (t->state == TRANS_STATE_COMPLETED)
775 refcount_inc(&cur_trans->use_count);
779 spin_unlock(&fs_info->trans_lock);
781 goto out; /* nothing committing|committed */
784 wait_for_commit(cur_trans);
785 btrfs_put_transaction(cur_trans);
790 void btrfs_throttle(struct btrfs_fs_info *fs_info)
792 if (!atomic_read(&fs_info->open_ioctl_trans))
793 wait_current_trans(fs_info);
796 static int should_end_transaction(struct btrfs_trans_handle *trans)
798 struct btrfs_fs_info *fs_info = trans->fs_info;
800 if (fs_info->global_block_rsv.space_info->full &&
801 btrfs_check_space_for_delayed_refs(trans, fs_info))
804 return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
807 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
809 struct btrfs_transaction *cur_trans = trans->transaction;
810 struct btrfs_fs_info *fs_info = trans->fs_info;
815 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
816 cur_trans->delayed_refs.flushing)
819 updates = trans->delayed_ref_updates;
820 trans->delayed_ref_updates = 0;
822 err = btrfs_run_delayed_refs(trans, fs_info, updates * 2);
823 if (err) /* Error code will also eval true */
827 return should_end_transaction(trans);
830 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
833 struct btrfs_fs_info *info = trans->fs_info;
834 struct btrfs_transaction *cur_trans = trans->transaction;
835 u64 transid = trans->transid;
836 unsigned long cur = trans->delayed_ref_updates;
837 int lock = (trans->type != TRANS_JOIN_NOLOCK);
839 int must_run_delayed_refs = 0;
841 if (trans->use_count > 1) {
843 trans->block_rsv = trans->orig_rsv;
847 btrfs_trans_release_metadata(trans, info);
848 trans->block_rsv = NULL;
850 if (!list_empty(&trans->new_bgs))
851 btrfs_create_pending_block_groups(trans, info);
853 trans->delayed_ref_updates = 0;
855 must_run_delayed_refs =
856 btrfs_should_throttle_delayed_refs(trans, info);
857 cur = max_t(unsigned long, cur, 32);
860 * don't make the caller wait if they are from a NOLOCK
861 * or ATTACH transaction, it will deadlock with commit
863 if (must_run_delayed_refs == 1 &&
864 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
865 must_run_delayed_refs = 2;
868 btrfs_trans_release_metadata(trans, info);
869 trans->block_rsv = NULL;
871 if (!list_empty(&trans->new_bgs))
872 btrfs_create_pending_block_groups(trans, info);
874 btrfs_trans_release_chunk_metadata(trans);
876 if (lock && !atomic_read(&info->open_ioctl_trans) &&
877 should_end_transaction(trans) &&
878 READ_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
879 spin_lock(&info->trans_lock);
880 if (cur_trans->state == TRANS_STATE_RUNNING)
881 cur_trans->state = TRANS_STATE_BLOCKED;
882 spin_unlock(&info->trans_lock);
885 if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
887 return btrfs_commit_transaction(trans);
889 wake_up_process(info->transaction_kthread);
892 if (trans->type & __TRANS_FREEZABLE)
893 sb_end_intwrite(info->sb);
895 WARN_ON(cur_trans != info->running_transaction);
896 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
897 atomic_dec(&cur_trans->num_writers);
898 extwriter_counter_dec(cur_trans, trans->type);
901 * Make sure counter is updated before we wake up waiters.
904 if (waitqueue_active(&cur_trans->writer_wait))
905 wake_up(&cur_trans->writer_wait);
906 btrfs_put_transaction(cur_trans);
908 if (current->journal_info == trans)
909 current->journal_info = NULL;
912 btrfs_run_delayed_iputs(info);
914 if (trans->aborted ||
915 test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
916 wake_up_process(info->transaction_kthread);
920 kmem_cache_free(btrfs_trans_handle_cachep, trans);
921 if (must_run_delayed_refs) {
922 btrfs_async_run_delayed_refs(info, cur, transid,
923 must_run_delayed_refs == 1);
928 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
930 return __btrfs_end_transaction(trans, 0);
933 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
935 return __btrfs_end_transaction(trans, 1);
939 * when btree blocks are allocated, they have some corresponding bits set for
940 * them in one of two extent_io trees. This is used to make sure all of
941 * those extents are sent to disk but does not wait on them
943 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
944 struct extent_io_tree *dirty_pages, int mark)
948 struct address_space *mapping = fs_info->btree_inode->i_mapping;
949 struct extent_state *cached_state = NULL;
953 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
954 mark, &cached_state)) {
955 bool wait_writeback = false;
957 err = convert_extent_bit(dirty_pages, start, end,
959 mark, &cached_state);
961 * convert_extent_bit can return -ENOMEM, which is most of the
962 * time a temporary error. So when it happens, ignore the error
963 * and wait for writeback of this range to finish - because we
964 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
965 * to __btrfs_wait_marked_extents() would not know that
966 * writeback for this range started and therefore wouldn't
967 * wait for it to finish - we don't want to commit a
968 * superblock that points to btree nodes/leafs for which
969 * writeback hasn't finished yet (and without errors).
970 * We cleanup any entries left in the io tree when committing
971 * the transaction (through clear_btree_io_tree()).
973 if (err == -ENOMEM) {
975 wait_writeback = true;
978 err = filemap_fdatawrite_range(mapping, start, end);
981 else if (wait_writeback)
982 werr = filemap_fdatawait_range(mapping, start, end);
983 free_extent_state(cached_state);
992 * when btree blocks are allocated, they have some corresponding bits set for
993 * them in one of two extent_io trees. This is used to make sure all of
994 * those extents are on disk for transaction or log commit. We wait
995 * on all the pages and clear them from the dirty pages state tree
997 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
998 struct extent_io_tree *dirty_pages)
1002 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1003 struct extent_state *cached_state = NULL;
1007 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1008 EXTENT_NEED_WAIT, &cached_state)) {
1010 * Ignore -ENOMEM errors returned by clear_extent_bit().
1011 * When committing the transaction, we'll remove any entries
1012 * left in the io tree. For a log commit, we don't remove them
1013 * after committing the log because the tree can be accessed
1014 * concurrently - we do it only at transaction commit time when
1015 * it's safe to do it (through clear_btree_io_tree()).
1017 err = clear_extent_bit(dirty_pages, start, end,
1019 0, 0, &cached_state, GFP_NOFS);
1023 err = filemap_fdatawait_range(mapping, start, end);
1026 free_extent_state(cached_state);
1027 cached_state = NULL;
1036 int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1037 struct extent_io_tree *dirty_pages)
1039 bool errors = false;
1042 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1043 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1051 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1053 struct btrfs_fs_info *fs_info = log_root->fs_info;
1054 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1055 bool errors = false;
1058 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1060 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1061 if ((mark & EXTENT_DIRTY) &&
1062 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1065 if ((mark & EXTENT_NEW) &&
1066 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1075 * when btree blocks are allocated, they have some corresponding bits set for
1076 * them in one of two extent_io trees. This is used to make sure all of
1077 * those extents are on disk for transaction or log commit
1079 static int btrfs_write_and_wait_marked_extents(struct btrfs_fs_info *fs_info,
1080 struct extent_io_tree *dirty_pages, int mark)
1084 struct blk_plug plug;
1086 blk_start_plug(&plug);
1087 ret = btrfs_write_marked_extents(fs_info, dirty_pages, mark);
1088 blk_finish_plug(&plug);
1089 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1098 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
1099 struct btrfs_fs_info *fs_info)
1103 ret = btrfs_write_and_wait_marked_extents(fs_info,
1104 &trans->transaction->dirty_pages,
1106 clear_btree_io_tree(&trans->transaction->dirty_pages);
1112 * this is used to update the root pointer in the tree of tree roots.
1114 * But, in the case of the extent allocation tree, updating the root
1115 * pointer may allocate blocks which may change the root of the extent
1118 * So, this loops and repeats and makes sure the cowonly root didn't
1119 * change while the root pointer was being updated in the metadata.
1121 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1122 struct btrfs_root *root)
1125 u64 old_root_bytenr;
1127 struct btrfs_fs_info *fs_info = root->fs_info;
1128 struct btrfs_root *tree_root = fs_info->tree_root;
1130 old_root_used = btrfs_root_used(&root->root_item);
1133 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1134 if (old_root_bytenr == root->node->start &&
1135 old_root_used == btrfs_root_used(&root->root_item))
1138 btrfs_set_root_node(&root->root_item, root->node);
1139 ret = btrfs_update_root(trans, tree_root,
1145 old_root_used = btrfs_root_used(&root->root_item);
1152 * update all the cowonly tree roots on disk
1154 * The error handling in this function may not be obvious. Any of the
1155 * failures will cause the file system to go offline. We still need
1156 * to clean up the delayed refs.
1158 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1159 struct btrfs_fs_info *fs_info)
1161 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1162 struct list_head *io_bgs = &trans->transaction->io_bgs;
1163 struct list_head *next;
1164 struct extent_buffer *eb;
1167 eb = btrfs_lock_root_node(fs_info->tree_root);
1168 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1170 btrfs_tree_unlock(eb);
1171 free_extent_buffer(eb);
1176 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1180 ret = btrfs_run_dev_stats(trans, fs_info);
1183 ret = btrfs_run_dev_replace(trans, fs_info);
1186 ret = btrfs_run_qgroups(trans, fs_info);
1190 ret = btrfs_setup_space_cache(trans, fs_info);
1194 /* run_qgroups might have added some more refs */
1195 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1199 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1200 struct btrfs_root *root;
1201 next = fs_info->dirty_cowonly_roots.next;
1202 list_del_init(next);
1203 root = list_entry(next, struct btrfs_root, dirty_list);
1204 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1206 if (root != fs_info->extent_root)
1207 list_add_tail(&root->dirty_list,
1208 &trans->transaction->switch_commits);
1209 ret = update_cowonly_root(trans, root);
1212 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1217 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1218 ret = btrfs_write_dirty_block_groups(trans, fs_info);
1221 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1226 if (!list_empty(&fs_info->dirty_cowonly_roots))
1229 list_add_tail(&fs_info->extent_root->dirty_list,
1230 &trans->transaction->switch_commits);
1231 btrfs_after_dev_replace_commit(fs_info);
1237 * dead roots are old snapshots that need to be deleted. This allocates
1238 * a dirty root struct and adds it into the list of dead roots that need to
1241 void btrfs_add_dead_root(struct btrfs_root *root)
1243 struct btrfs_fs_info *fs_info = root->fs_info;
1245 spin_lock(&fs_info->trans_lock);
1246 if (list_empty(&root->root_list))
1247 list_add_tail(&root->root_list, &fs_info->dead_roots);
1248 spin_unlock(&fs_info->trans_lock);
1252 * update all the cowonly tree roots on disk
1254 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1255 struct btrfs_fs_info *fs_info)
1257 struct btrfs_root *gang[8];
1262 spin_lock(&fs_info->fs_roots_radix_lock);
1264 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1267 BTRFS_ROOT_TRANS_TAG);
1270 for (i = 0; i < ret; i++) {
1271 struct btrfs_root *root = gang[i];
1272 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1273 (unsigned long)root->root_key.objectid,
1274 BTRFS_ROOT_TRANS_TAG);
1275 spin_unlock(&fs_info->fs_roots_radix_lock);
1277 btrfs_free_log(trans, root);
1278 btrfs_update_reloc_root(trans, root);
1279 btrfs_orphan_commit_root(trans, root);
1281 btrfs_save_ino_cache(root, trans);
1283 /* see comments in should_cow_block() */
1284 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1285 smp_mb__after_atomic();
1287 if (root->commit_root != root->node) {
1288 list_add_tail(&root->dirty_list,
1289 &trans->transaction->switch_commits);
1290 btrfs_set_root_node(&root->root_item,
1294 err = btrfs_update_root(trans, fs_info->tree_root,
1297 spin_lock(&fs_info->fs_roots_radix_lock);
1300 btrfs_qgroup_free_meta_all(root);
1303 spin_unlock(&fs_info->fs_roots_radix_lock);
1308 * defrag a given btree.
1309 * Every leaf in the btree is read and defragged.
1311 int btrfs_defrag_root(struct btrfs_root *root)
1313 struct btrfs_fs_info *info = root->fs_info;
1314 struct btrfs_trans_handle *trans;
1317 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1321 trans = btrfs_start_transaction(root, 0);
1322 if (IS_ERR(trans)) {
1323 ret = PTR_ERR(trans);
1327 ret = btrfs_defrag_leaves(trans, root);
1329 btrfs_end_transaction(trans);
1330 btrfs_btree_balance_dirty(info);
1333 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1336 if (btrfs_defrag_cancelled(info)) {
1337 btrfs_debug(info, "defrag_root cancelled");
1342 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1347 * Do all special snapshot related qgroup dirty hack.
1349 * Will do all needed qgroup inherit and dirty hack like switch commit
1350 * roots inside one transaction and write all btree into disk, to make
1353 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1354 struct btrfs_root *src,
1355 struct btrfs_root *parent,
1356 struct btrfs_qgroup_inherit *inherit,
1359 struct btrfs_fs_info *fs_info = src->fs_info;
1363 * Save some performance in the case that qgroups are not
1364 * enabled. If this check races with the ioctl, rescan will
1367 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1371 * Ensure dirty @src will be commited. Or, after comming
1372 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1373 * recorded root will never be updated again, causing an outdated root
1376 record_root_in_trans(trans, src, 1);
1379 * We are going to commit transaction, see btrfs_commit_transaction()
1380 * comment for reason locking tree_log_mutex
1382 mutex_lock(&fs_info->tree_log_mutex);
1384 ret = commit_fs_roots(trans, fs_info);
1387 ret = btrfs_qgroup_account_extents(trans, fs_info);
1391 /* Now qgroup are all updated, we can inherit it to new qgroups */
1392 ret = btrfs_qgroup_inherit(trans, fs_info,
1393 src->root_key.objectid, dst_objectid,
1399 * Now we do a simplified commit transaction, which will:
1400 * 1) commit all subvolume and extent tree
1401 * To ensure all subvolume and extent tree have a valid
1402 * commit_root to accounting later insert_dir_item()
1403 * 2) write all btree blocks onto disk
1404 * This is to make sure later btree modification will be cowed
1405 * Or commit_root can be populated and cause wrong qgroup numbers
1406 * In this simplified commit, we don't really care about other trees
1407 * like chunk and root tree, as they won't affect qgroup.
1408 * And we don't write super to avoid half committed status.
1410 ret = commit_cowonly_roots(trans, fs_info);
1413 switch_commit_roots(trans->transaction, fs_info);
1414 ret = btrfs_write_and_wait_transaction(trans, fs_info);
1416 btrfs_handle_fs_error(fs_info, ret,
1417 "Error while writing out transaction for qgroup");
1420 mutex_unlock(&fs_info->tree_log_mutex);
1423 * Force parent root to be updated, as we recorded it before so its
1424 * last_trans == cur_transid.
1425 * Or it won't be committed again onto disk after later
1429 record_root_in_trans(trans, parent, 1);
1434 * new snapshots need to be created at a very specific time in the
1435 * transaction commit. This does the actual creation.
1438 * If the error which may affect the commitment of the current transaction
1439 * happens, we should return the error number. If the error which just affect
1440 * the creation of the pending snapshots, just return 0.
1442 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1443 struct btrfs_fs_info *fs_info,
1444 struct btrfs_pending_snapshot *pending)
1446 struct btrfs_key key;
1447 struct btrfs_root_item *new_root_item;
1448 struct btrfs_root *tree_root = fs_info->tree_root;
1449 struct btrfs_root *root = pending->root;
1450 struct btrfs_root *parent_root;
1451 struct btrfs_block_rsv *rsv;
1452 struct inode *parent_inode;
1453 struct btrfs_path *path;
1454 struct btrfs_dir_item *dir_item;
1455 struct dentry *dentry;
1456 struct extent_buffer *tmp;
1457 struct extent_buffer *old;
1458 struct timespec cur_time;
1466 ASSERT(pending->path);
1467 path = pending->path;
1469 ASSERT(pending->root_item);
1470 new_root_item = pending->root_item;
1472 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1474 goto no_free_objectid;
1477 * Make qgroup to skip current new snapshot's qgroupid, as it is
1478 * accounted by later btrfs_qgroup_inherit().
1480 btrfs_set_skip_qgroup(trans, objectid);
1482 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1484 if (to_reserve > 0) {
1485 pending->error = btrfs_block_rsv_add(root,
1486 &pending->block_rsv,
1488 BTRFS_RESERVE_NO_FLUSH);
1490 goto clear_skip_qgroup;
1493 key.objectid = objectid;
1494 key.offset = (u64)-1;
1495 key.type = BTRFS_ROOT_ITEM_KEY;
1497 rsv = trans->block_rsv;
1498 trans->block_rsv = &pending->block_rsv;
1499 trans->bytes_reserved = trans->block_rsv->reserved;
1500 trace_btrfs_space_reservation(fs_info, "transaction",
1502 trans->bytes_reserved, 1);
1503 dentry = pending->dentry;
1504 parent_inode = pending->dir;
1505 parent_root = BTRFS_I(parent_inode)->root;
1506 record_root_in_trans(trans, parent_root, 0);
1508 cur_time = current_time(parent_inode);
1511 * insert the directory item
1513 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1514 BUG_ON(ret); /* -ENOMEM */
1516 /* check if there is a file/dir which has the same name. */
1517 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1518 btrfs_ino(BTRFS_I(parent_inode)),
1519 dentry->d_name.name,
1520 dentry->d_name.len, 0);
1521 if (dir_item != NULL && !IS_ERR(dir_item)) {
1522 pending->error = -EEXIST;
1523 goto dir_item_existed;
1524 } else if (IS_ERR(dir_item)) {
1525 ret = PTR_ERR(dir_item);
1526 btrfs_abort_transaction(trans, ret);
1529 btrfs_release_path(path);
1532 * pull in the delayed directory update
1533 * and the delayed inode item
1534 * otherwise we corrupt the FS during
1537 ret = btrfs_run_delayed_items(trans, fs_info);
1538 if (ret) { /* Transaction aborted */
1539 btrfs_abort_transaction(trans, ret);
1543 record_root_in_trans(trans, root, 0);
1544 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1545 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1546 btrfs_check_and_init_root_item(new_root_item);
1548 root_flags = btrfs_root_flags(new_root_item);
1549 if (pending->readonly)
1550 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1552 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1553 btrfs_set_root_flags(new_root_item, root_flags);
1555 btrfs_set_root_generation_v2(new_root_item,
1557 uuid_le_gen(&new_uuid);
1558 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1559 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1561 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1562 memset(new_root_item->received_uuid, 0,
1563 sizeof(new_root_item->received_uuid));
1564 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1565 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1566 btrfs_set_root_stransid(new_root_item, 0);
1567 btrfs_set_root_rtransid(new_root_item, 0);
1569 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1570 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1571 btrfs_set_root_otransid(new_root_item, trans->transid);
1573 old = btrfs_lock_root_node(root);
1574 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1576 btrfs_tree_unlock(old);
1577 free_extent_buffer(old);
1578 btrfs_abort_transaction(trans, ret);
1582 btrfs_set_lock_blocking(old);
1584 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1585 /* clean up in any case */
1586 btrfs_tree_unlock(old);
1587 free_extent_buffer(old);
1589 btrfs_abort_transaction(trans, ret);
1592 /* see comments in should_cow_block() */
1593 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1596 btrfs_set_root_node(new_root_item, tmp);
1597 /* record when the snapshot was created in key.offset */
1598 key.offset = trans->transid;
1599 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1600 btrfs_tree_unlock(tmp);
1601 free_extent_buffer(tmp);
1603 btrfs_abort_transaction(trans, ret);
1608 * insert root back/forward references
1610 ret = btrfs_add_root_ref(trans, fs_info, objectid,
1611 parent_root->root_key.objectid,
1612 btrfs_ino(BTRFS_I(parent_inode)), index,
1613 dentry->d_name.name, dentry->d_name.len);
1615 btrfs_abort_transaction(trans, ret);
1619 key.offset = (u64)-1;
1620 pending->snap = btrfs_read_fs_root_no_name(fs_info, &key);
1621 if (IS_ERR(pending->snap)) {
1622 ret = PTR_ERR(pending->snap);
1623 btrfs_abort_transaction(trans, ret);
1627 ret = btrfs_reloc_post_snapshot(trans, pending);
1629 btrfs_abort_transaction(trans, ret);
1633 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1635 btrfs_abort_transaction(trans, ret);
1640 * Do special qgroup accounting for snapshot, as we do some qgroup
1641 * snapshot hack to do fast snapshot.
1642 * To co-operate with that hack, we do hack again.
1643 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1645 ret = qgroup_account_snapshot(trans, root, parent_root,
1646 pending->inherit, objectid);
1650 ret = btrfs_insert_dir_item(trans, parent_root,
1651 dentry->d_name.name, dentry->d_name.len,
1652 BTRFS_I(parent_inode), &key,
1653 BTRFS_FT_DIR, index);
1654 /* We have check then name at the beginning, so it is impossible. */
1655 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1657 btrfs_abort_transaction(trans, ret);
1661 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1662 dentry->d_name.len * 2);
1663 parent_inode->i_mtime = parent_inode->i_ctime =
1664 current_time(parent_inode);
1665 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1667 btrfs_abort_transaction(trans, ret);
1670 ret = btrfs_uuid_tree_add(trans, fs_info, new_uuid.b,
1671 BTRFS_UUID_KEY_SUBVOL, objectid);
1673 btrfs_abort_transaction(trans, ret);
1676 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1677 ret = btrfs_uuid_tree_add(trans, fs_info,
1678 new_root_item->received_uuid,
1679 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1681 if (ret && ret != -EEXIST) {
1682 btrfs_abort_transaction(trans, ret);
1687 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1689 btrfs_abort_transaction(trans, ret);
1694 pending->error = ret;
1696 trans->block_rsv = rsv;
1697 trans->bytes_reserved = 0;
1699 btrfs_clear_skip_qgroup(trans);
1701 kfree(new_root_item);
1702 pending->root_item = NULL;
1703 btrfs_free_path(path);
1704 pending->path = NULL;
1710 * create all the snapshots we've scheduled for creation
1712 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1713 struct btrfs_fs_info *fs_info)
1715 struct btrfs_pending_snapshot *pending, *next;
1716 struct list_head *head = &trans->transaction->pending_snapshots;
1719 list_for_each_entry_safe(pending, next, head, list) {
1720 list_del(&pending->list);
1721 ret = create_pending_snapshot(trans, fs_info, pending);
1728 static void update_super_roots(struct btrfs_fs_info *fs_info)
1730 struct btrfs_root_item *root_item;
1731 struct btrfs_super_block *super;
1733 super = fs_info->super_copy;
1735 root_item = &fs_info->chunk_root->root_item;
1736 super->chunk_root = root_item->bytenr;
1737 super->chunk_root_generation = root_item->generation;
1738 super->chunk_root_level = root_item->level;
1740 root_item = &fs_info->tree_root->root_item;
1741 super->root = root_item->bytenr;
1742 super->generation = root_item->generation;
1743 super->root_level = root_item->level;
1744 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1745 super->cache_generation = root_item->generation;
1746 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1747 super->uuid_tree_generation = root_item->generation;
1750 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1752 struct btrfs_transaction *trans;
1755 spin_lock(&info->trans_lock);
1756 trans = info->running_transaction;
1758 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1759 spin_unlock(&info->trans_lock);
1763 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1765 struct btrfs_transaction *trans;
1768 spin_lock(&info->trans_lock);
1769 trans = info->running_transaction;
1771 ret = is_transaction_blocked(trans);
1772 spin_unlock(&info->trans_lock);
1777 * wait for the current transaction commit to start and block subsequent
1780 static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info,
1781 struct btrfs_transaction *trans)
1783 wait_event(fs_info->transaction_blocked_wait,
1784 trans->state >= TRANS_STATE_COMMIT_START || trans->aborted);
1788 * wait for the current transaction to start and then become unblocked.
1791 static void wait_current_trans_commit_start_and_unblock(
1792 struct btrfs_fs_info *fs_info,
1793 struct btrfs_transaction *trans)
1795 wait_event(fs_info->transaction_wait,
1796 trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted);
1800 * commit transactions asynchronously. once btrfs_commit_transaction_async
1801 * returns, any subsequent transaction will not be allowed to join.
1803 struct btrfs_async_commit {
1804 struct btrfs_trans_handle *newtrans;
1805 struct work_struct work;
1808 static void do_async_commit(struct work_struct *work)
1810 struct btrfs_async_commit *ac =
1811 container_of(work, struct btrfs_async_commit, work);
1814 * We've got freeze protection passed with the transaction.
1815 * Tell lockdep about it.
1817 if (ac->newtrans->type & __TRANS_FREEZABLE)
1818 __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1820 current->journal_info = ac->newtrans;
1822 btrfs_commit_transaction(ac->newtrans);
1826 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1827 int wait_for_unblock)
1829 struct btrfs_fs_info *fs_info = trans->fs_info;
1830 struct btrfs_async_commit *ac;
1831 struct btrfs_transaction *cur_trans;
1833 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1837 INIT_WORK(&ac->work, do_async_commit);
1838 ac->newtrans = btrfs_join_transaction(trans->root);
1839 if (IS_ERR(ac->newtrans)) {
1840 int err = PTR_ERR(ac->newtrans);
1845 /* take transaction reference */
1846 cur_trans = trans->transaction;
1847 refcount_inc(&cur_trans->use_count);
1849 btrfs_end_transaction(trans);
1852 * Tell lockdep we've released the freeze rwsem, since the
1853 * async commit thread will be the one to unlock it.
1855 if (ac->newtrans->type & __TRANS_FREEZABLE)
1856 __sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1858 schedule_work(&ac->work);
1860 /* wait for transaction to start and unblock */
1861 if (wait_for_unblock)
1862 wait_current_trans_commit_start_and_unblock(fs_info, cur_trans);
1864 wait_current_trans_commit_start(fs_info, cur_trans);
1866 if (current->journal_info == trans)
1867 current->journal_info = NULL;
1869 btrfs_put_transaction(cur_trans);
1874 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1875 struct btrfs_root *root, int err)
1877 struct btrfs_fs_info *fs_info = root->fs_info;
1878 struct btrfs_transaction *cur_trans = trans->transaction;
1881 WARN_ON(trans->use_count > 1);
1883 btrfs_abort_transaction(trans, err);
1885 spin_lock(&fs_info->trans_lock);
1888 * If the transaction is removed from the list, it means this
1889 * transaction has been committed successfully, so it is impossible
1890 * to call the cleanup function.
1892 BUG_ON(list_empty(&cur_trans->list));
1894 list_del_init(&cur_trans->list);
1895 if (cur_trans == fs_info->running_transaction) {
1896 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1897 spin_unlock(&fs_info->trans_lock);
1898 wait_event(cur_trans->writer_wait,
1899 atomic_read(&cur_trans->num_writers) == 1);
1901 spin_lock(&fs_info->trans_lock);
1903 spin_unlock(&fs_info->trans_lock);
1905 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
1907 spin_lock(&fs_info->trans_lock);
1908 if (cur_trans == fs_info->running_transaction)
1909 fs_info->running_transaction = NULL;
1910 spin_unlock(&fs_info->trans_lock);
1912 if (trans->type & __TRANS_FREEZABLE)
1913 sb_end_intwrite(fs_info->sb);
1914 btrfs_put_transaction(cur_trans);
1915 btrfs_put_transaction(cur_trans);
1917 trace_btrfs_transaction_commit(root);
1919 if (current->journal_info == trans)
1920 current->journal_info = NULL;
1921 btrfs_scrub_cancel(fs_info);
1923 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1926 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1928 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
1929 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1933 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1935 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
1936 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1940 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans)
1942 wait_event(cur_trans->pending_wait,
1943 atomic_read(&cur_trans->pending_ordered) == 0);
1946 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
1948 struct btrfs_fs_info *fs_info = trans->fs_info;
1949 struct btrfs_transaction *cur_trans = trans->transaction;
1950 struct btrfs_transaction *prev_trans = NULL;
1954 * Some places just start a transaction to commit it. We need to make
1955 * sure that if this commit fails that the abort code actually marks the
1956 * transaction as failed, so set trans->dirty to make the abort code do
1959 trans->dirty = true;
1961 /* Stop the commit early if ->aborted is set */
1962 if (unlikely(READ_ONCE(cur_trans->aborted))) {
1963 ret = cur_trans->aborted;
1964 btrfs_end_transaction(trans);
1968 btrfs_trans_release_metadata(trans, fs_info);
1969 trans->block_rsv = NULL;
1971 /* make a pass through all the delayed refs we have so far
1972 * any runnings procs may add more while we are here
1974 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
1976 btrfs_end_transaction(trans);
1980 cur_trans = trans->transaction;
1983 * set the flushing flag so procs in this transaction have to
1984 * start sending their work down.
1986 cur_trans->delayed_refs.flushing = 1;
1989 if (!list_empty(&trans->new_bgs))
1990 btrfs_create_pending_block_groups(trans, fs_info);
1992 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
1994 btrfs_end_transaction(trans);
1998 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2001 /* this mutex is also taken before trying to set
2002 * block groups readonly. We need to make sure
2003 * that nobody has set a block group readonly
2004 * after a extents from that block group have been
2005 * allocated for cache files. btrfs_set_block_group_ro
2006 * will wait for the transaction to commit if it
2007 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2009 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2010 * only one process starts all the block group IO. It wouldn't
2011 * hurt to have more than one go through, but there's no
2012 * real advantage to it either.
2014 mutex_lock(&fs_info->ro_block_group_mutex);
2015 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2018 mutex_unlock(&fs_info->ro_block_group_mutex);
2021 ret = btrfs_start_dirty_block_groups(trans, fs_info);
2024 btrfs_end_transaction(trans);
2028 spin_lock(&fs_info->trans_lock);
2029 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2030 spin_unlock(&fs_info->trans_lock);
2031 refcount_inc(&cur_trans->use_count);
2032 ret = btrfs_end_transaction(trans);
2034 wait_for_commit(cur_trans);
2036 if (unlikely(cur_trans->aborted))
2037 ret = cur_trans->aborted;
2039 btrfs_put_transaction(cur_trans);
2044 cur_trans->state = TRANS_STATE_COMMIT_START;
2045 wake_up(&fs_info->transaction_blocked_wait);
2047 if (cur_trans->list.prev != &fs_info->trans_list) {
2048 prev_trans = list_entry(cur_trans->list.prev,
2049 struct btrfs_transaction, list);
2050 if (prev_trans->state != TRANS_STATE_COMPLETED) {
2051 refcount_inc(&prev_trans->use_count);
2052 spin_unlock(&fs_info->trans_lock);
2054 wait_for_commit(prev_trans);
2055 ret = prev_trans->aborted;
2057 btrfs_put_transaction(prev_trans);
2059 goto cleanup_transaction;
2061 spin_unlock(&fs_info->trans_lock);
2064 spin_unlock(&fs_info->trans_lock);
2066 * The previous transaction was aborted and was already removed
2067 * from the list of transactions at fs_info->trans_list. So we
2068 * abort to prevent writing a new superblock that reflects a
2069 * corrupt state (pointing to trees with unwritten nodes/leafs).
2071 if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) {
2073 goto cleanup_transaction;
2077 extwriter_counter_dec(cur_trans, trans->type);
2079 ret = btrfs_start_delalloc_flush(fs_info);
2081 goto cleanup_transaction;
2083 ret = btrfs_run_delayed_items(trans, fs_info);
2085 goto cleanup_transaction;
2087 wait_event(cur_trans->writer_wait,
2088 extwriter_counter_read(cur_trans) == 0);
2090 /* some pending stuffs might be added after the previous flush. */
2091 ret = btrfs_run_delayed_items(trans, fs_info);
2093 goto cleanup_transaction;
2095 btrfs_wait_delalloc_flush(fs_info);
2097 btrfs_wait_pending_ordered(cur_trans);
2099 btrfs_scrub_pause(fs_info);
2101 * Ok now we need to make sure to block out any other joins while we
2102 * commit the transaction. We could have started a join before setting
2103 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2105 spin_lock(&fs_info->trans_lock);
2106 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2107 spin_unlock(&fs_info->trans_lock);
2108 wait_event(cur_trans->writer_wait,
2109 atomic_read(&cur_trans->num_writers) == 1);
2111 /* ->aborted might be set after the previous check, so check it */
2112 if (unlikely(READ_ONCE(cur_trans->aborted))) {
2113 ret = cur_trans->aborted;
2114 goto scrub_continue;
2117 * the reloc mutex makes sure that we stop
2118 * the balancing code from coming in and moving
2119 * extents around in the middle of the commit
2121 mutex_lock(&fs_info->reloc_mutex);
2124 * We needn't worry about the delayed items because we will
2125 * deal with them in create_pending_snapshot(), which is the
2126 * core function of the snapshot creation.
2128 ret = create_pending_snapshots(trans, fs_info);
2130 mutex_unlock(&fs_info->reloc_mutex);
2131 goto scrub_continue;
2135 * We insert the dir indexes of the snapshots and update the inode
2136 * of the snapshots' parents after the snapshot creation, so there
2137 * are some delayed items which are not dealt with. Now deal with
2140 * We needn't worry that this operation will corrupt the snapshots,
2141 * because all the tree which are snapshoted will be forced to COW
2142 * the nodes and leaves.
2144 ret = btrfs_run_delayed_items(trans, fs_info);
2146 mutex_unlock(&fs_info->reloc_mutex);
2147 goto scrub_continue;
2150 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
2152 mutex_unlock(&fs_info->reloc_mutex);
2153 goto scrub_continue;
2157 * make sure none of the code above managed to slip in a
2160 btrfs_assert_delayed_root_empty(fs_info);
2162 WARN_ON(cur_trans != trans->transaction);
2164 /* btrfs_commit_tree_roots is responsible for getting the
2165 * various roots consistent with each other. Every pointer
2166 * in the tree of tree roots has to point to the most up to date
2167 * root for every subvolume and other tree. So, we have to keep
2168 * the tree logging code from jumping in and changing any
2171 * At this point in the commit, there can't be any tree-log
2172 * writers, but a little lower down we drop the trans mutex
2173 * and let new people in. By holding the tree_log_mutex
2174 * from now until after the super is written, we avoid races
2175 * with the tree-log code.
2177 mutex_lock(&fs_info->tree_log_mutex);
2179 ret = commit_fs_roots(trans, fs_info);
2181 mutex_unlock(&fs_info->tree_log_mutex);
2182 mutex_unlock(&fs_info->reloc_mutex);
2183 goto scrub_continue;
2187 * Since the transaction is done, we can apply the pending changes
2188 * before the next transaction.
2190 btrfs_apply_pending_changes(fs_info);
2192 /* commit_fs_roots gets rid of all the tree log roots, it is now
2193 * safe to free the root of tree log roots
2195 btrfs_free_log_root_tree(trans, fs_info);
2198 * commit_fs_roots() can call btrfs_save_ino_cache(), which generates
2199 * new delayed refs. Must handle them or qgroup can be wrong.
2201 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
2203 mutex_unlock(&fs_info->tree_log_mutex);
2204 mutex_unlock(&fs_info->reloc_mutex);
2205 goto scrub_continue;
2209 * Since fs roots are all committed, we can get a quite accurate
2210 * new_roots. So let's do quota accounting.
2212 ret = btrfs_qgroup_account_extents(trans, fs_info);
2214 mutex_unlock(&fs_info->tree_log_mutex);
2215 mutex_unlock(&fs_info->reloc_mutex);
2216 goto scrub_continue;
2219 ret = commit_cowonly_roots(trans, fs_info);
2221 mutex_unlock(&fs_info->tree_log_mutex);
2222 mutex_unlock(&fs_info->reloc_mutex);
2223 goto scrub_continue;
2227 * The tasks which save the space cache and inode cache may also
2228 * update ->aborted, check it.
2230 if (unlikely(READ_ONCE(cur_trans->aborted))) {
2231 ret = cur_trans->aborted;
2232 mutex_unlock(&fs_info->tree_log_mutex);
2233 mutex_unlock(&fs_info->reloc_mutex);
2234 goto scrub_continue;
2237 btrfs_prepare_extent_commit(fs_info);
2239 cur_trans = fs_info->running_transaction;
2241 btrfs_set_root_node(&fs_info->tree_root->root_item,
2242 fs_info->tree_root->node);
2243 list_add_tail(&fs_info->tree_root->dirty_list,
2244 &cur_trans->switch_commits);
2246 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2247 fs_info->chunk_root->node);
2248 list_add_tail(&fs_info->chunk_root->dirty_list,
2249 &cur_trans->switch_commits);
2251 switch_commit_roots(cur_trans, fs_info);
2253 ASSERT(list_empty(&cur_trans->dirty_bgs));
2254 ASSERT(list_empty(&cur_trans->io_bgs));
2255 update_super_roots(fs_info);
2257 btrfs_set_super_log_root(fs_info->super_copy, 0);
2258 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2259 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2260 sizeof(*fs_info->super_copy));
2262 btrfs_update_commit_device_size(fs_info);
2263 btrfs_update_commit_device_bytes_used(fs_info, cur_trans);
2265 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2266 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2268 btrfs_trans_release_chunk_metadata(trans);
2270 spin_lock(&fs_info->trans_lock);
2271 cur_trans->state = TRANS_STATE_UNBLOCKED;
2272 fs_info->running_transaction = NULL;
2273 spin_unlock(&fs_info->trans_lock);
2274 mutex_unlock(&fs_info->reloc_mutex);
2276 wake_up(&fs_info->transaction_wait);
2278 ret = btrfs_write_and_wait_transaction(trans, fs_info);
2280 btrfs_handle_fs_error(fs_info, ret,
2281 "Error while writing out transaction");
2282 mutex_unlock(&fs_info->tree_log_mutex);
2283 goto scrub_continue;
2286 ret = write_all_supers(fs_info, 0);
2288 mutex_unlock(&fs_info->tree_log_mutex);
2289 goto scrub_continue;
2293 * the super is written, we can safely allow the tree-loggers
2294 * to go about their business
2296 mutex_unlock(&fs_info->tree_log_mutex);
2298 btrfs_finish_extent_commit(trans, fs_info);
2300 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2301 btrfs_clear_space_info_full(fs_info);
2303 fs_info->last_trans_committed = cur_trans->transid;
2305 * We needn't acquire the lock here because there is no other task
2306 * which can change it.
2308 cur_trans->state = TRANS_STATE_COMPLETED;
2309 wake_up(&cur_trans->commit_wait);
2311 spin_lock(&fs_info->trans_lock);
2312 list_del_init(&cur_trans->list);
2313 spin_unlock(&fs_info->trans_lock);
2315 btrfs_put_transaction(cur_trans);
2316 btrfs_put_transaction(cur_trans);
2318 if (trans->type & __TRANS_FREEZABLE)
2319 sb_end_intwrite(fs_info->sb);
2321 trace_btrfs_transaction_commit(trans->root);
2323 btrfs_scrub_continue(fs_info);
2325 if (current->journal_info == trans)
2326 current->journal_info = NULL;
2328 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2333 btrfs_scrub_continue(fs_info);
2334 cleanup_transaction:
2335 btrfs_trans_release_metadata(trans, fs_info);
2336 btrfs_trans_release_chunk_metadata(trans);
2337 trans->block_rsv = NULL;
2338 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2339 if (current->journal_info == trans)
2340 current->journal_info = NULL;
2341 cleanup_transaction(trans, trans->root, ret);
2347 * return < 0 if error
2348 * 0 if there are no more dead_roots at the time of call
2349 * 1 there are more to be processed, call me again
2351 * The return value indicates there are certainly more snapshots to delete, but
2352 * if there comes a new one during processing, it may return 0. We don't mind,
2353 * because btrfs_commit_super will poke cleaner thread and it will process it a
2354 * few seconds later.
2356 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2359 struct btrfs_fs_info *fs_info = root->fs_info;
2361 spin_lock(&fs_info->trans_lock);
2362 if (list_empty(&fs_info->dead_roots)) {
2363 spin_unlock(&fs_info->trans_lock);
2366 root = list_first_entry(&fs_info->dead_roots,
2367 struct btrfs_root, root_list);
2368 list_del_init(&root->root_list);
2369 spin_unlock(&fs_info->trans_lock);
2371 btrfs_debug(fs_info, "cleaner removing %llu", root->objectid);
2373 btrfs_kill_all_delayed_nodes(root);
2375 if (btrfs_header_backref_rev(root->node) <
2376 BTRFS_MIXED_BACKREF_REV)
2377 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2379 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2381 return (ret < 0) ? 0 : 1;
2384 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2389 prev = xchg(&fs_info->pending_changes, 0);
2393 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2395 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2398 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2400 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2403 bit = 1 << BTRFS_PENDING_COMMIT;
2405 btrfs_debug(fs_info, "pending commit done");
2410 "unknown pending changes left 0x%lx, ignoring", prev);