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(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 (btrfs_check_space_for_delayed_refs(trans, fs_info))
803 return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
806 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
808 struct btrfs_transaction *cur_trans = trans->transaction;
809 struct btrfs_fs_info *fs_info = trans->fs_info;
814 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
815 cur_trans->delayed_refs.flushing)
818 updates = trans->delayed_ref_updates;
819 trans->delayed_ref_updates = 0;
821 err = btrfs_run_delayed_refs(trans, fs_info, updates * 2);
822 if (err) /* Error code will also eval true */
826 return should_end_transaction(trans);
829 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
832 struct btrfs_fs_info *info = trans->fs_info;
833 struct btrfs_transaction *cur_trans = trans->transaction;
834 u64 transid = trans->transid;
835 unsigned long cur = trans->delayed_ref_updates;
836 int lock = (trans->type != TRANS_JOIN_NOLOCK);
838 int must_run_delayed_refs = 0;
840 if (trans->use_count > 1) {
842 trans->block_rsv = trans->orig_rsv;
846 btrfs_trans_release_metadata(trans, info);
847 trans->block_rsv = NULL;
849 if (!list_empty(&trans->new_bgs))
850 btrfs_create_pending_block_groups(trans, info);
852 trans->delayed_ref_updates = 0;
854 must_run_delayed_refs =
855 btrfs_should_throttle_delayed_refs(trans, info);
856 cur = max_t(unsigned long, cur, 32);
859 * don't make the caller wait if they are from a NOLOCK
860 * or ATTACH transaction, it will deadlock with commit
862 if (must_run_delayed_refs == 1 &&
863 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
864 must_run_delayed_refs = 2;
867 btrfs_trans_release_metadata(trans, info);
868 trans->block_rsv = NULL;
870 if (!list_empty(&trans->new_bgs))
871 btrfs_create_pending_block_groups(trans, info);
873 btrfs_trans_release_chunk_metadata(trans);
875 if (lock && !atomic_read(&info->open_ioctl_trans) &&
876 should_end_transaction(trans) &&
877 READ_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
878 spin_lock(&info->trans_lock);
879 if (cur_trans->state == TRANS_STATE_RUNNING)
880 cur_trans->state = TRANS_STATE_BLOCKED;
881 spin_unlock(&info->trans_lock);
884 if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
886 return btrfs_commit_transaction(trans);
888 wake_up_process(info->transaction_kthread);
891 if (trans->type & __TRANS_FREEZABLE)
892 sb_end_intwrite(info->sb);
894 WARN_ON(cur_trans != info->running_transaction);
895 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
896 atomic_dec(&cur_trans->num_writers);
897 extwriter_counter_dec(cur_trans, trans->type);
900 * Make sure counter is updated before we wake up waiters.
903 if (waitqueue_active(&cur_trans->writer_wait))
904 wake_up(&cur_trans->writer_wait);
905 btrfs_put_transaction(cur_trans);
907 if (current->journal_info == trans)
908 current->journal_info = NULL;
911 btrfs_run_delayed_iputs(info);
913 if (trans->aborted ||
914 test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
915 wake_up_process(info->transaction_kthread);
919 kmem_cache_free(btrfs_trans_handle_cachep, trans);
920 if (must_run_delayed_refs) {
921 btrfs_async_run_delayed_refs(info, cur, transid,
922 must_run_delayed_refs == 1);
927 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
929 return __btrfs_end_transaction(trans, 0);
932 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
934 return __btrfs_end_transaction(trans, 1);
938 * when btree blocks are allocated, they have some corresponding bits set for
939 * them in one of two extent_io trees. This is used to make sure all of
940 * those extents are sent to disk but does not wait on them
942 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
943 struct extent_io_tree *dirty_pages, int mark)
947 struct address_space *mapping = fs_info->btree_inode->i_mapping;
948 struct extent_state *cached_state = NULL;
952 atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
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);
988 atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
993 * when btree blocks are allocated, they have some corresponding bits set for
994 * them in one of two extent_io trees. This is used to make sure all of
995 * those extents are on disk for transaction or log commit. We wait
996 * on all the pages and clear them from the dirty pages state tree
998 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
999 struct extent_io_tree *dirty_pages)
1003 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1004 struct extent_state *cached_state = NULL;
1008 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1009 EXTENT_NEED_WAIT, &cached_state)) {
1011 * Ignore -ENOMEM errors returned by clear_extent_bit().
1012 * When committing the transaction, we'll remove any entries
1013 * left in the io tree. For a log commit, we don't remove them
1014 * after committing the log because the tree can be accessed
1015 * concurrently - we do it only at transaction commit time when
1016 * it's safe to do it (through clear_btree_io_tree()).
1018 err = clear_extent_bit(dirty_pages, start, end,
1020 0, 0, &cached_state, GFP_NOFS);
1024 err = filemap_fdatawait_range(mapping, start, end);
1027 free_extent_state(cached_state);
1028 cached_state = NULL;
1037 int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1038 struct extent_io_tree *dirty_pages)
1040 bool errors = false;
1043 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1044 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1052 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1054 struct btrfs_fs_info *fs_info = log_root->fs_info;
1055 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1056 bool errors = false;
1059 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1061 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1062 if ((mark & EXTENT_DIRTY) &&
1063 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1066 if ((mark & EXTENT_NEW) &&
1067 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1076 * when btree blocks are allocated, they have some corresponding bits set for
1077 * them in one of two extent_io trees. This is used to make sure all of
1078 * those extents are on disk for transaction or log commit
1080 static int btrfs_write_and_wait_marked_extents(struct btrfs_fs_info *fs_info,
1081 struct extent_io_tree *dirty_pages, int mark)
1085 struct blk_plug plug;
1087 blk_start_plug(&plug);
1088 ret = btrfs_write_marked_extents(fs_info, dirty_pages, mark);
1089 blk_finish_plug(&plug);
1090 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1099 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
1100 struct btrfs_fs_info *fs_info)
1104 ret = btrfs_write_and_wait_marked_extents(fs_info,
1105 &trans->transaction->dirty_pages,
1107 clear_btree_io_tree(&trans->transaction->dirty_pages);
1113 * this is used to update the root pointer in the tree of tree roots.
1115 * But, in the case of the extent allocation tree, updating the root
1116 * pointer may allocate blocks which may change the root of the extent
1119 * So, this loops and repeats and makes sure the cowonly root didn't
1120 * change while the root pointer was being updated in the metadata.
1122 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1123 struct btrfs_root *root)
1126 u64 old_root_bytenr;
1128 struct btrfs_fs_info *fs_info = root->fs_info;
1129 struct btrfs_root *tree_root = fs_info->tree_root;
1131 old_root_used = btrfs_root_used(&root->root_item);
1134 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1135 if (old_root_bytenr == root->node->start &&
1136 old_root_used == btrfs_root_used(&root->root_item))
1139 btrfs_set_root_node(&root->root_item, root->node);
1140 ret = btrfs_update_root(trans, tree_root,
1146 old_root_used = btrfs_root_used(&root->root_item);
1153 * update all the cowonly tree roots on disk
1155 * The error handling in this function may not be obvious. Any of the
1156 * failures will cause the file system to go offline. We still need
1157 * to clean up the delayed refs.
1159 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1160 struct btrfs_fs_info *fs_info)
1162 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1163 struct list_head *io_bgs = &trans->transaction->io_bgs;
1164 struct list_head *next;
1165 struct extent_buffer *eb;
1168 eb = btrfs_lock_root_node(fs_info->tree_root);
1169 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1171 btrfs_tree_unlock(eb);
1172 free_extent_buffer(eb);
1177 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1181 ret = btrfs_run_dev_stats(trans, fs_info);
1184 ret = btrfs_run_dev_replace(trans, fs_info);
1187 ret = btrfs_run_qgroups(trans, fs_info);
1191 ret = btrfs_setup_space_cache(trans, fs_info);
1195 /* run_qgroups might have added some more refs */
1196 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1200 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1201 struct btrfs_root *root;
1202 next = fs_info->dirty_cowonly_roots.next;
1203 list_del_init(next);
1204 root = list_entry(next, struct btrfs_root, dirty_list);
1205 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1207 if (root != fs_info->extent_root)
1208 list_add_tail(&root->dirty_list,
1209 &trans->transaction->switch_commits);
1210 ret = update_cowonly_root(trans, root);
1213 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1218 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1219 ret = btrfs_write_dirty_block_groups(trans, fs_info);
1222 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1227 if (!list_empty(&fs_info->dirty_cowonly_roots))
1230 list_add_tail(&fs_info->extent_root->dirty_list,
1231 &trans->transaction->switch_commits);
1232 btrfs_after_dev_replace_commit(fs_info);
1238 * dead roots are old snapshots that need to be deleted. This allocates
1239 * a dirty root struct and adds it into the list of dead roots that need to
1242 void btrfs_add_dead_root(struct btrfs_root *root)
1244 struct btrfs_fs_info *fs_info = root->fs_info;
1246 spin_lock(&fs_info->trans_lock);
1247 if (list_empty(&root->root_list))
1248 list_add_tail(&root->root_list, &fs_info->dead_roots);
1249 spin_unlock(&fs_info->trans_lock);
1253 * update all the cowonly tree roots on disk
1255 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1256 struct btrfs_fs_info *fs_info)
1258 struct btrfs_root *gang[8];
1263 spin_lock(&fs_info->fs_roots_radix_lock);
1265 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1268 BTRFS_ROOT_TRANS_TAG);
1271 for (i = 0; i < ret; i++) {
1272 struct btrfs_root *root = gang[i];
1273 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1274 (unsigned long)root->root_key.objectid,
1275 BTRFS_ROOT_TRANS_TAG);
1276 spin_unlock(&fs_info->fs_roots_radix_lock);
1278 btrfs_free_log(trans, root);
1279 btrfs_update_reloc_root(trans, root);
1280 btrfs_orphan_commit_root(trans, root);
1282 btrfs_save_ino_cache(root, trans);
1284 /* see comments in should_cow_block() */
1285 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1286 smp_mb__after_atomic();
1288 if (root->commit_root != root->node) {
1289 list_add_tail(&root->dirty_list,
1290 &trans->transaction->switch_commits);
1291 btrfs_set_root_node(&root->root_item,
1295 err = btrfs_update_root(trans, fs_info->tree_root,
1298 spin_lock(&fs_info->fs_roots_radix_lock);
1301 btrfs_qgroup_free_meta_all(root);
1304 spin_unlock(&fs_info->fs_roots_radix_lock);
1309 * defrag a given btree.
1310 * Every leaf in the btree is read and defragged.
1312 int btrfs_defrag_root(struct btrfs_root *root)
1314 struct btrfs_fs_info *info = root->fs_info;
1315 struct btrfs_trans_handle *trans;
1318 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1322 trans = btrfs_start_transaction(root, 0);
1324 return PTR_ERR(trans);
1326 ret = btrfs_defrag_leaves(trans, root);
1328 btrfs_end_transaction(trans);
1329 btrfs_btree_balance_dirty(info);
1332 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1335 if (btrfs_defrag_cancelled(info)) {
1336 btrfs_debug(info, "defrag_root cancelled");
1341 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1346 * Do all special snapshot related qgroup dirty hack.
1348 * Will do all needed qgroup inherit and dirty hack like switch commit
1349 * roots inside one transaction and write all btree into disk, to make
1352 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1353 struct btrfs_root *src,
1354 struct btrfs_root *parent,
1355 struct btrfs_qgroup_inherit *inherit,
1358 struct btrfs_fs_info *fs_info = src->fs_info;
1362 * Save some performance in the case that qgroups are not
1363 * enabled. If this check races with the ioctl, rescan will
1366 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1370 * We are going to commit transaction, see btrfs_commit_transaction()
1371 * comment for reason locking tree_log_mutex
1373 mutex_lock(&fs_info->tree_log_mutex);
1375 ret = commit_fs_roots(trans, fs_info);
1378 ret = btrfs_qgroup_account_extents(trans, fs_info);
1382 /* Now qgroup are all updated, we can inherit it to new qgroups */
1383 ret = btrfs_qgroup_inherit(trans, fs_info,
1384 src->root_key.objectid, dst_objectid,
1390 * Now we do a simplified commit transaction, which will:
1391 * 1) commit all subvolume and extent tree
1392 * To ensure all subvolume and extent tree have a valid
1393 * commit_root to accounting later insert_dir_item()
1394 * 2) write all btree blocks onto disk
1395 * This is to make sure later btree modification will be cowed
1396 * Or commit_root can be populated and cause wrong qgroup numbers
1397 * In this simplified commit, we don't really care about other trees
1398 * like chunk and root tree, as they won't affect qgroup.
1399 * And we don't write super to avoid half committed status.
1401 ret = commit_cowonly_roots(trans, fs_info);
1404 switch_commit_roots(trans->transaction, fs_info);
1405 ret = btrfs_write_and_wait_transaction(trans, fs_info);
1407 btrfs_handle_fs_error(fs_info, ret,
1408 "Error while writing out transaction for qgroup");
1411 mutex_unlock(&fs_info->tree_log_mutex);
1414 * Force parent root to be updated, as we recorded it before so its
1415 * last_trans == cur_transid.
1416 * Or it won't be committed again onto disk after later
1420 record_root_in_trans(trans, parent, 1);
1425 * new snapshots need to be created at a very specific time in the
1426 * transaction commit. This does the actual creation.
1429 * If the error which may affect the commitment of the current transaction
1430 * happens, we should return the error number. If the error which just affect
1431 * the creation of the pending snapshots, just return 0.
1433 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1434 struct btrfs_fs_info *fs_info,
1435 struct btrfs_pending_snapshot *pending)
1437 struct btrfs_key key;
1438 struct btrfs_root_item *new_root_item;
1439 struct btrfs_root *tree_root = fs_info->tree_root;
1440 struct btrfs_root *root = pending->root;
1441 struct btrfs_root *parent_root;
1442 struct btrfs_block_rsv *rsv;
1443 struct inode *parent_inode;
1444 struct btrfs_path *path;
1445 struct btrfs_dir_item *dir_item;
1446 struct dentry *dentry;
1447 struct extent_buffer *tmp;
1448 struct extent_buffer *old;
1449 struct timespec cur_time;
1457 ASSERT(pending->path);
1458 path = pending->path;
1460 ASSERT(pending->root_item);
1461 new_root_item = pending->root_item;
1463 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1465 goto no_free_objectid;
1468 * Make qgroup to skip current new snapshot's qgroupid, as it is
1469 * accounted by later btrfs_qgroup_inherit().
1471 btrfs_set_skip_qgroup(trans, objectid);
1473 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1475 if (to_reserve > 0) {
1476 pending->error = btrfs_block_rsv_add(root,
1477 &pending->block_rsv,
1479 BTRFS_RESERVE_NO_FLUSH);
1481 goto clear_skip_qgroup;
1484 key.objectid = objectid;
1485 key.offset = (u64)-1;
1486 key.type = BTRFS_ROOT_ITEM_KEY;
1488 rsv = trans->block_rsv;
1489 trans->block_rsv = &pending->block_rsv;
1490 trans->bytes_reserved = trans->block_rsv->reserved;
1491 trace_btrfs_space_reservation(fs_info, "transaction",
1493 trans->bytes_reserved, 1);
1494 dentry = pending->dentry;
1495 parent_inode = pending->dir;
1496 parent_root = BTRFS_I(parent_inode)->root;
1497 record_root_in_trans(trans, parent_root, 0);
1499 cur_time = current_time(parent_inode);
1502 * insert the directory item
1504 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1505 BUG_ON(ret); /* -ENOMEM */
1507 /* check if there is a file/dir which has the same name. */
1508 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1509 btrfs_ino(BTRFS_I(parent_inode)),
1510 dentry->d_name.name,
1511 dentry->d_name.len, 0);
1512 if (dir_item != NULL && !IS_ERR(dir_item)) {
1513 pending->error = -EEXIST;
1514 goto dir_item_existed;
1515 } else if (IS_ERR(dir_item)) {
1516 ret = PTR_ERR(dir_item);
1517 btrfs_abort_transaction(trans, ret);
1520 btrfs_release_path(path);
1523 * pull in the delayed directory update
1524 * and the delayed inode item
1525 * otherwise we corrupt the FS during
1528 ret = btrfs_run_delayed_items(trans, fs_info);
1529 if (ret) { /* Transaction aborted */
1530 btrfs_abort_transaction(trans, ret);
1534 record_root_in_trans(trans, root, 0);
1535 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1536 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1537 btrfs_check_and_init_root_item(new_root_item);
1539 root_flags = btrfs_root_flags(new_root_item);
1540 if (pending->readonly)
1541 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1543 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1544 btrfs_set_root_flags(new_root_item, root_flags);
1546 btrfs_set_root_generation_v2(new_root_item,
1548 uuid_le_gen(&new_uuid);
1549 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1550 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1552 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1553 memset(new_root_item->received_uuid, 0,
1554 sizeof(new_root_item->received_uuid));
1555 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1556 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1557 btrfs_set_root_stransid(new_root_item, 0);
1558 btrfs_set_root_rtransid(new_root_item, 0);
1560 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1561 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1562 btrfs_set_root_otransid(new_root_item, trans->transid);
1564 old = btrfs_lock_root_node(root);
1565 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1567 btrfs_tree_unlock(old);
1568 free_extent_buffer(old);
1569 btrfs_abort_transaction(trans, ret);
1573 btrfs_set_lock_blocking(old);
1575 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1576 /* clean up in any case */
1577 btrfs_tree_unlock(old);
1578 free_extent_buffer(old);
1580 btrfs_abort_transaction(trans, ret);
1583 /* see comments in should_cow_block() */
1584 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1587 btrfs_set_root_node(new_root_item, tmp);
1588 /* record when the snapshot was created in key.offset */
1589 key.offset = trans->transid;
1590 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1591 btrfs_tree_unlock(tmp);
1592 free_extent_buffer(tmp);
1594 btrfs_abort_transaction(trans, ret);
1599 * insert root back/forward references
1601 ret = btrfs_add_root_ref(trans, fs_info, objectid,
1602 parent_root->root_key.objectid,
1603 btrfs_ino(BTRFS_I(parent_inode)), index,
1604 dentry->d_name.name, dentry->d_name.len);
1606 btrfs_abort_transaction(trans, ret);
1610 key.offset = (u64)-1;
1611 pending->snap = btrfs_read_fs_root_no_name(fs_info, &key);
1612 if (IS_ERR(pending->snap)) {
1613 ret = PTR_ERR(pending->snap);
1614 btrfs_abort_transaction(trans, ret);
1618 ret = btrfs_reloc_post_snapshot(trans, pending);
1620 btrfs_abort_transaction(trans, ret);
1624 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1626 btrfs_abort_transaction(trans, ret);
1631 * Do special qgroup accounting for snapshot, as we do some qgroup
1632 * snapshot hack to do fast snapshot.
1633 * To co-operate with that hack, we do hack again.
1634 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1636 ret = qgroup_account_snapshot(trans, root, parent_root,
1637 pending->inherit, objectid);
1641 ret = btrfs_insert_dir_item(trans, parent_root,
1642 dentry->d_name.name, dentry->d_name.len,
1643 BTRFS_I(parent_inode), &key,
1644 BTRFS_FT_DIR, index);
1645 /* We have check then name at the beginning, so it is impossible. */
1646 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1648 btrfs_abort_transaction(trans, ret);
1652 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1653 dentry->d_name.len * 2);
1654 parent_inode->i_mtime = parent_inode->i_ctime =
1655 current_time(parent_inode);
1656 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1658 btrfs_abort_transaction(trans, ret);
1661 ret = btrfs_uuid_tree_add(trans, fs_info, new_uuid.b,
1662 BTRFS_UUID_KEY_SUBVOL, objectid);
1664 btrfs_abort_transaction(trans, ret);
1667 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1668 ret = btrfs_uuid_tree_add(trans, fs_info,
1669 new_root_item->received_uuid,
1670 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1672 if (ret && ret != -EEXIST) {
1673 btrfs_abort_transaction(trans, ret);
1678 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
1680 btrfs_abort_transaction(trans, ret);
1685 pending->error = ret;
1687 trans->block_rsv = rsv;
1688 trans->bytes_reserved = 0;
1690 btrfs_clear_skip_qgroup(trans);
1692 kfree(new_root_item);
1693 pending->root_item = NULL;
1694 btrfs_free_path(path);
1695 pending->path = NULL;
1701 * create all the snapshots we've scheduled for creation
1703 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1704 struct btrfs_fs_info *fs_info)
1706 struct btrfs_pending_snapshot *pending, *next;
1707 struct list_head *head = &trans->transaction->pending_snapshots;
1710 list_for_each_entry_safe(pending, next, head, list) {
1711 list_del(&pending->list);
1712 ret = create_pending_snapshot(trans, fs_info, pending);
1719 static void update_super_roots(struct btrfs_fs_info *fs_info)
1721 struct btrfs_root_item *root_item;
1722 struct btrfs_super_block *super;
1724 super = fs_info->super_copy;
1726 root_item = &fs_info->chunk_root->root_item;
1727 super->chunk_root = root_item->bytenr;
1728 super->chunk_root_generation = root_item->generation;
1729 super->chunk_root_level = root_item->level;
1731 root_item = &fs_info->tree_root->root_item;
1732 super->root = root_item->bytenr;
1733 super->generation = root_item->generation;
1734 super->root_level = root_item->level;
1735 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1736 super->cache_generation = root_item->generation;
1737 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1738 super->uuid_tree_generation = root_item->generation;
1741 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1743 struct btrfs_transaction *trans;
1746 spin_lock(&info->trans_lock);
1747 trans = info->running_transaction;
1749 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1750 spin_unlock(&info->trans_lock);
1754 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1756 struct btrfs_transaction *trans;
1759 spin_lock(&info->trans_lock);
1760 trans = info->running_transaction;
1762 ret = is_transaction_blocked(trans);
1763 spin_unlock(&info->trans_lock);
1768 * wait for the current transaction commit to start and block subsequent
1771 static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info,
1772 struct btrfs_transaction *trans)
1774 wait_event(fs_info->transaction_blocked_wait,
1775 trans->state >= TRANS_STATE_COMMIT_START || trans->aborted);
1779 * wait for the current transaction to start and then become unblocked.
1782 static void wait_current_trans_commit_start_and_unblock(
1783 struct btrfs_fs_info *fs_info,
1784 struct btrfs_transaction *trans)
1786 wait_event(fs_info->transaction_wait,
1787 trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted);
1791 * commit transactions asynchronously. once btrfs_commit_transaction_async
1792 * returns, any subsequent transaction will not be allowed to join.
1794 struct btrfs_async_commit {
1795 struct btrfs_trans_handle *newtrans;
1796 struct work_struct work;
1799 static void do_async_commit(struct work_struct *work)
1801 struct btrfs_async_commit *ac =
1802 container_of(work, struct btrfs_async_commit, work);
1805 * We've got freeze protection passed with the transaction.
1806 * Tell lockdep about it.
1808 if (ac->newtrans->type & __TRANS_FREEZABLE)
1809 __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1811 current->journal_info = ac->newtrans;
1813 btrfs_commit_transaction(ac->newtrans);
1817 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1818 int wait_for_unblock)
1820 struct btrfs_fs_info *fs_info = trans->fs_info;
1821 struct btrfs_async_commit *ac;
1822 struct btrfs_transaction *cur_trans;
1824 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1828 INIT_WORK(&ac->work, do_async_commit);
1829 ac->newtrans = btrfs_join_transaction(trans->root);
1830 if (IS_ERR(ac->newtrans)) {
1831 int err = PTR_ERR(ac->newtrans);
1836 /* take transaction reference */
1837 cur_trans = trans->transaction;
1838 refcount_inc(&cur_trans->use_count);
1840 btrfs_end_transaction(trans);
1843 * Tell lockdep we've released the freeze rwsem, since the
1844 * async commit thread will be the one to unlock it.
1846 if (ac->newtrans->type & __TRANS_FREEZABLE)
1847 __sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1849 schedule_work(&ac->work);
1851 /* wait for transaction to start and unblock */
1852 if (wait_for_unblock)
1853 wait_current_trans_commit_start_and_unblock(fs_info, cur_trans);
1855 wait_current_trans_commit_start(fs_info, cur_trans);
1857 if (current->journal_info == trans)
1858 current->journal_info = NULL;
1860 btrfs_put_transaction(cur_trans);
1865 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1866 struct btrfs_root *root, int err)
1868 struct btrfs_fs_info *fs_info = root->fs_info;
1869 struct btrfs_transaction *cur_trans = trans->transaction;
1872 WARN_ON(trans->use_count > 1);
1874 btrfs_abort_transaction(trans, err);
1876 spin_lock(&fs_info->trans_lock);
1879 * If the transaction is removed from the list, it means this
1880 * transaction has been committed successfully, so it is impossible
1881 * to call the cleanup function.
1883 BUG_ON(list_empty(&cur_trans->list));
1885 list_del_init(&cur_trans->list);
1886 if (cur_trans == fs_info->running_transaction) {
1887 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1888 spin_unlock(&fs_info->trans_lock);
1889 wait_event(cur_trans->writer_wait,
1890 atomic_read(&cur_trans->num_writers) == 1);
1892 spin_lock(&fs_info->trans_lock);
1894 spin_unlock(&fs_info->trans_lock);
1896 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
1898 spin_lock(&fs_info->trans_lock);
1899 if (cur_trans == fs_info->running_transaction)
1900 fs_info->running_transaction = NULL;
1901 spin_unlock(&fs_info->trans_lock);
1903 if (trans->type & __TRANS_FREEZABLE)
1904 sb_end_intwrite(fs_info->sb);
1905 btrfs_put_transaction(cur_trans);
1906 btrfs_put_transaction(cur_trans);
1908 trace_btrfs_transaction_commit(root);
1910 if (current->journal_info == trans)
1911 current->journal_info = NULL;
1912 btrfs_scrub_cancel(fs_info);
1914 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1917 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1920 * We use writeback_inodes_sb here because if we used
1921 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
1922 * Currently are holding the fs freeze lock, if we do an async flush
1923 * we'll do btrfs_join_transaction() and deadlock because we need to
1924 * wait for the fs freeze lock. Using the direct flushing we benefit
1925 * from already being in a transaction and our join_transaction doesn't
1926 * have to re-take the fs freeze lock.
1928 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
1929 writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
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;
1953 /* Stop the commit early if ->aborted is set */
1954 if (unlikely(READ_ONCE(cur_trans->aborted))) {
1955 ret = cur_trans->aborted;
1956 btrfs_end_transaction(trans);
1960 /* make a pass through all the delayed refs we have so far
1961 * any runnings procs may add more while we are here
1963 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
1965 btrfs_end_transaction(trans);
1969 btrfs_trans_release_metadata(trans, fs_info);
1970 trans->block_rsv = NULL;
1972 cur_trans = trans->transaction;
1975 * set the flushing flag so procs in this transaction have to
1976 * start sending their work down.
1978 cur_trans->delayed_refs.flushing = 1;
1981 if (!list_empty(&trans->new_bgs))
1982 btrfs_create_pending_block_groups(trans, fs_info);
1984 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
1986 btrfs_end_transaction(trans);
1990 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
1993 /* this mutex is also taken before trying to set
1994 * block groups readonly. We need to make sure
1995 * that nobody has set a block group readonly
1996 * after a extents from that block group have been
1997 * allocated for cache files. btrfs_set_block_group_ro
1998 * will wait for the transaction to commit if it
1999 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2001 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2002 * only one process starts all the block group IO. It wouldn't
2003 * hurt to have more than one go through, but there's no
2004 * real advantage to it either.
2006 mutex_lock(&fs_info->ro_block_group_mutex);
2007 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2010 mutex_unlock(&fs_info->ro_block_group_mutex);
2013 ret = btrfs_start_dirty_block_groups(trans, fs_info);
2016 btrfs_end_transaction(trans);
2020 spin_lock(&fs_info->trans_lock);
2021 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2022 spin_unlock(&fs_info->trans_lock);
2023 refcount_inc(&cur_trans->use_count);
2024 ret = btrfs_end_transaction(trans);
2026 wait_for_commit(cur_trans);
2028 if (unlikely(cur_trans->aborted))
2029 ret = cur_trans->aborted;
2031 btrfs_put_transaction(cur_trans);
2036 cur_trans->state = TRANS_STATE_COMMIT_START;
2037 wake_up(&fs_info->transaction_blocked_wait);
2039 if (cur_trans->list.prev != &fs_info->trans_list) {
2040 prev_trans = list_entry(cur_trans->list.prev,
2041 struct btrfs_transaction, list);
2042 if (prev_trans->state != TRANS_STATE_COMPLETED) {
2043 refcount_inc(&prev_trans->use_count);
2044 spin_unlock(&fs_info->trans_lock);
2046 wait_for_commit(prev_trans);
2047 ret = prev_trans->aborted;
2049 btrfs_put_transaction(prev_trans);
2051 goto cleanup_transaction;
2053 spin_unlock(&fs_info->trans_lock);
2056 spin_unlock(&fs_info->trans_lock);
2059 extwriter_counter_dec(cur_trans, trans->type);
2061 ret = btrfs_start_delalloc_flush(fs_info);
2063 goto cleanup_transaction;
2065 ret = btrfs_run_delayed_items(trans, fs_info);
2067 goto cleanup_transaction;
2069 wait_event(cur_trans->writer_wait,
2070 extwriter_counter_read(cur_trans) == 0);
2072 /* some pending stuffs might be added after the previous flush. */
2073 ret = btrfs_run_delayed_items(trans, fs_info);
2075 goto cleanup_transaction;
2077 btrfs_wait_delalloc_flush(fs_info);
2079 btrfs_wait_pending_ordered(cur_trans);
2081 btrfs_scrub_pause(fs_info);
2083 * Ok now we need to make sure to block out any other joins while we
2084 * commit the transaction. We could have started a join before setting
2085 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2087 spin_lock(&fs_info->trans_lock);
2088 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2089 spin_unlock(&fs_info->trans_lock);
2090 wait_event(cur_trans->writer_wait,
2091 atomic_read(&cur_trans->num_writers) == 1);
2093 /* ->aborted might be set after the previous check, so check it */
2094 if (unlikely(READ_ONCE(cur_trans->aborted))) {
2095 ret = cur_trans->aborted;
2096 goto scrub_continue;
2099 * the reloc mutex makes sure that we stop
2100 * the balancing code from coming in and moving
2101 * extents around in the middle of the commit
2103 mutex_lock(&fs_info->reloc_mutex);
2106 * We needn't worry about the delayed items because we will
2107 * deal with them in create_pending_snapshot(), which is the
2108 * core function of the snapshot creation.
2110 ret = create_pending_snapshots(trans, fs_info);
2112 mutex_unlock(&fs_info->reloc_mutex);
2113 goto scrub_continue;
2117 * We insert the dir indexes of the snapshots and update the inode
2118 * of the snapshots' parents after the snapshot creation, so there
2119 * are some delayed items which are not dealt with. Now deal with
2122 * We needn't worry that this operation will corrupt the snapshots,
2123 * because all the tree which are snapshoted will be forced to COW
2124 * the nodes and leaves.
2126 ret = btrfs_run_delayed_items(trans, fs_info);
2128 mutex_unlock(&fs_info->reloc_mutex);
2129 goto scrub_continue;
2132 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
2134 mutex_unlock(&fs_info->reloc_mutex);
2135 goto scrub_continue;
2139 * make sure none of the code above managed to slip in a
2142 btrfs_assert_delayed_root_empty(fs_info);
2144 WARN_ON(cur_trans != trans->transaction);
2146 /* btrfs_commit_tree_roots is responsible for getting the
2147 * various roots consistent with each other. Every pointer
2148 * in the tree of tree roots has to point to the most up to date
2149 * root for every subvolume and other tree. So, we have to keep
2150 * the tree logging code from jumping in and changing any
2153 * At this point in the commit, there can't be any tree-log
2154 * writers, but a little lower down we drop the trans mutex
2155 * and let new people in. By holding the tree_log_mutex
2156 * from now until after the super is written, we avoid races
2157 * with the tree-log code.
2159 mutex_lock(&fs_info->tree_log_mutex);
2161 ret = commit_fs_roots(trans, fs_info);
2163 mutex_unlock(&fs_info->tree_log_mutex);
2164 mutex_unlock(&fs_info->reloc_mutex);
2165 goto scrub_continue;
2169 * Since the transaction is done, we can apply the pending changes
2170 * before the next transaction.
2172 btrfs_apply_pending_changes(fs_info);
2174 /* commit_fs_roots gets rid of all the tree log roots, it is now
2175 * safe to free the root of tree log roots
2177 btrfs_free_log_root_tree(trans, fs_info);
2180 * commit_fs_roots() can call btrfs_save_ino_cache(), which generates
2181 * new delayed refs. Must handle them or qgroup can be wrong.
2183 ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1);
2185 mutex_unlock(&fs_info->tree_log_mutex);
2186 mutex_unlock(&fs_info->reloc_mutex);
2187 goto scrub_continue;
2191 * Since fs roots are all committed, we can get a quite accurate
2192 * new_roots. So let's do quota accounting.
2194 ret = btrfs_qgroup_account_extents(trans, fs_info);
2196 mutex_unlock(&fs_info->tree_log_mutex);
2197 mutex_unlock(&fs_info->reloc_mutex);
2198 goto scrub_continue;
2201 ret = commit_cowonly_roots(trans, fs_info);
2203 mutex_unlock(&fs_info->tree_log_mutex);
2204 mutex_unlock(&fs_info->reloc_mutex);
2205 goto scrub_continue;
2209 * The tasks which save the space cache and inode cache may also
2210 * update ->aborted, check it.
2212 if (unlikely(READ_ONCE(cur_trans->aborted))) {
2213 ret = cur_trans->aborted;
2214 mutex_unlock(&fs_info->tree_log_mutex);
2215 mutex_unlock(&fs_info->reloc_mutex);
2216 goto scrub_continue;
2219 btrfs_prepare_extent_commit(fs_info);
2221 cur_trans = fs_info->running_transaction;
2223 btrfs_set_root_node(&fs_info->tree_root->root_item,
2224 fs_info->tree_root->node);
2225 list_add_tail(&fs_info->tree_root->dirty_list,
2226 &cur_trans->switch_commits);
2228 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2229 fs_info->chunk_root->node);
2230 list_add_tail(&fs_info->chunk_root->dirty_list,
2231 &cur_trans->switch_commits);
2233 switch_commit_roots(cur_trans, fs_info);
2235 ASSERT(list_empty(&cur_trans->dirty_bgs));
2236 ASSERT(list_empty(&cur_trans->io_bgs));
2237 update_super_roots(fs_info);
2239 btrfs_set_super_log_root(fs_info->super_copy, 0);
2240 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2241 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2242 sizeof(*fs_info->super_copy));
2244 btrfs_update_commit_device_size(fs_info);
2245 btrfs_update_commit_device_bytes_used(fs_info, cur_trans);
2247 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2248 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2250 btrfs_trans_release_chunk_metadata(trans);
2252 spin_lock(&fs_info->trans_lock);
2253 cur_trans->state = TRANS_STATE_UNBLOCKED;
2254 fs_info->running_transaction = NULL;
2255 spin_unlock(&fs_info->trans_lock);
2256 mutex_unlock(&fs_info->reloc_mutex);
2258 wake_up(&fs_info->transaction_wait);
2260 ret = btrfs_write_and_wait_transaction(trans, fs_info);
2262 btrfs_handle_fs_error(fs_info, ret,
2263 "Error while writing out transaction");
2264 mutex_unlock(&fs_info->tree_log_mutex);
2265 goto scrub_continue;
2268 ret = write_all_supers(fs_info, 0);
2270 mutex_unlock(&fs_info->tree_log_mutex);
2271 goto scrub_continue;
2275 * the super is written, we can safely allow the tree-loggers
2276 * to go about their business
2278 mutex_unlock(&fs_info->tree_log_mutex);
2280 btrfs_finish_extent_commit(trans, fs_info);
2282 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2283 btrfs_clear_space_info_full(fs_info);
2285 fs_info->last_trans_committed = cur_trans->transid;
2287 * We needn't acquire the lock here because there is no other task
2288 * which can change it.
2290 cur_trans->state = TRANS_STATE_COMPLETED;
2291 wake_up(&cur_trans->commit_wait);
2293 spin_lock(&fs_info->trans_lock);
2294 list_del_init(&cur_trans->list);
2295 spin_unlock(&fs_info->trans_lock);
2297 btrfs_put_transaction(cur_trans);
2298 btrfs_put_transaction(cur_trans);
2300 if (trans->type & __TRANS_FREEZABLE)
2301 sb_end_intwrite(fs_info->sb);
2303 trace_btrfs_transaction_commit(trans->root);
2305 btrfs_scrub_continue(fs_info);
2307 if (current->journal_info == trans)
2308 current->journal_info = NULL;
2310 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2313 * If fs has been frozen, we can not handle delayed iputs, otherwise
2314 * it'll result in deadlock about SB_FREEZE_FS.
2316 if (current != fs_info->transaction_kthread &&
2317 current != fs_info->cleaner_kthread &&
2318 !test_bit(BTRFS_FS_FROZEN, &fs_info->flags))
2319 btrfs_run_delayed_iputs(fs_info);
2324 btrfs_scrub_continue(fs_info);
2325 cleanup_transaction:
2326 btrfs_trans_release_metadata(trans, fs_info);
2327 btrfs_trans_release_chunk_metadata(trans);
2328 trans->block_rsv = NULL;
2329 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2330 if (current->journal_info == trans)
2331 current->journal_info = NULL;
2332 cleanup_transaction(trans, trans->root, ret);
2338 * return < 0 if error
2339 * 0 if there are no more dead_roots at the time of call
2340 * 1 there are more to be processed, call me again
2342 * The return value indicates there are certainly more snapshots to delete, but
2343 * if there comes a new one during processing, it may return 0. We don't mind,
2344 * because btrfs_commit_super will poke cleaner thread and it will process it a
2345 * few seconds later.
2347 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2350 struct btrfs_fs_info *fs_info = root->fs_info;
2352 spin_lock(&fs_info->trans_lock);
2353 if (list_empty(&fs_info->dead_roots)) {
2354 spin_unlock(&fs_info->trans_lock);
2357 root = list_first_entry(&fs_info->dead_roots,
2358 struct btrfs_root, root_list);
2359 list_del_init(&root->root_list);
2360 spin_unlock(&fs_info->trans_lock);
2362 btrfs_debug(fs_info, "cleaner removing %llu", root->objectid);
2364 btrfs_kill_all_delayed_nodes(root);
2366 if (btrfs_header_backref_rev(root->node) <
2367 BTRFS_MIXED_BACKREF_REV)
2368 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2370 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2372 return (ret < 0) ? 0 : 1;
2375 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2380 prev = xchg(&fs_info->pending_changes, 0);
2384 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2386 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2389 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2391 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2394 bit = 1 << BTRFS_PENDING_COMMIT;
2396 btrfs_debug(fs_info, "pending commit done");
2401 "unknown pending changes left 0x%lx, ignoring", prev);