1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/writeback.h>
10 #include <linux/pagemap.h>
11 #include <linux/blkdev.h>
12 #include <linux/uuid.h>
16 #include "transaction.h"
20 #include "dev-replace.h"
22 #include "block-group.h"
23 #include "space-info.h"
26 #define BTRFS_ROOT_TRANS_TAG 0
29 * Transaction states and transitions
31 * No running transaction (fs tree blocks are not modified)
34 * | Call start_transaction() variants. Except btrfs_join_transaction_nostart().
36 * Transaction N [[TRANS_STATE_RUNNING]]
38 * | New trans handles can be attached to transaction N by calling all
39 * | start_transaction() variants.
42 * | Call btrfs_commit_transaction() on any trans handle attached to
45 * Transaction N [[TRANS_STATE_COMMIT_START]]
47 * | Will wait for previous running transaction to completely finish if there
50 * | Then one of the following happes:
51 * | - Wait for all other trans handle holders to release.
52 * | The btrfs_commit_transaction() caller will do the commit work.
53 * | - Wait for current transaction to be committed by others.
54 * | Other btrfs_commit_transaction() caller will do the commit work.
56 * | At this stage, only btrfs_join_transaction*() variants can attach
57 * | to this running transaction.
58 * | All other variants will wait for current one to finish and attach to
62 * | Caller is chosen to commit transaction N, and all other trans handle
63 * | haven been released.
65 * Transaction N [[TRANS_STATE_COMMIT_DOING]]
67 * | The heavy lifting transaction work is started.
68 * | From running delayed refs (modifying extent tree) to creating pending
69 * | snapshots, running qgroups.
70 * | In short, modify supporting trees to reflect modifications of subvolume
73 * | At this stage, all start_transaction() calls will wait for this
74 * | transaction to finish and attach to transaction N+1.
77 * | Until all supporting trees are updated.
79 * Transaction N [[TRANS_STATE_UNBLOCKED]]
81 * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
82 * | need to write them back to disk and update |
85 * | At this stage, new transaction is allowed to |
87 * | All new start_transaction() calls will be |
88 * | attached to transid N+1. |
91 * | Until all tree blocks are super blocks are |
92 * | written to block devices |
94 * Transaction N [[TRANS_STATE_COMPLETED]] V
95 * All tree blocks and super blocks are written. Transaction N+1
96 * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]]
97 * data structures will be cleaned up. | Life goes on
99 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
100 [TRANS_STATE_RUNNING] = 0U,
101 [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH),
102 [TRANS_STATE_COMMIT_DOING] = (__TRANS_START |
105 __TRANS_JOIN_NOSTART),
106 [TRANS_STATE_UNBLOCKED] = (__TRANS_START |
109 __TRANS_JOIN_NOLOCK |
110 __TRANS_JOIN_NOSTART),
111 [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START |
114 __TRANS_JOIN_NOLOCK |
115 __TRANS_JOIN_NOSTART),
116 [TRANS_STATE_COMPLETED] = (__TRANS_START |
119 __TRANS_JOIN_NOLOCK |
120 __TRANS_JOIN_NOSTART),
123 void btrfs_put_transaction(struct btrfs_transaction *transaction)
125 WARN_ON(refcount_read(&transaction->use_count) == 0);
126 if (refcount_dec_and_test(&transaction->use_count)) {
127 BUG_ON(!list_empty(&transaction->list));
128 WARN_ON(!RB_EMPTY_ROOT(
129 &transaction->delayed_refs.href_root.rb_root));
130 WARN_ON(!RB_EMPTY_ROOT(
131 &transaction->delayed_refs.dirty_extent_root));
132 if (transaction->delayed_refs.pending_csums)
133 btrfs_err(transaction->fs_info,
134 "pending csums is %llu",
135 transaction->delayed_refs.pending_csums);
137 * If any block groups are found in ->deleted_bgs then it's
138 * because the transaction was aborted and a commit did not
139 * happen (things failed before writing the new superblock
140 * and calling btrfs_finish_extent_commit()), so we can not
141 * discard the physical locations of the block groups.
143 while (!list_empty(&transaction->deleted_bgs)) {
144 struct btrfs_block_group *cache;
146 cache = list_first_entry(&transaction->deleted_bgs,
147 struct btrfs_block_group,
149 list_del_init(&cache->bg_list);
150 btrfs_unfreeze_block_group(cache);
151 btrfs_put_block_group(cache);
153 WARN_ON(!list_empty(&transaction->dev_update_list));
158 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
160 struct btrfs_transaction *cur_trans = trans->transaction;
161 struct btrfs_fs_info *fs_info = trans->fs_info;
162 struct btrfs_root *root, *tmp;
163 struct btrfs_caching_control *caching_ctl, *next;
165 down_write(&fs_info->commit_root_sem);
166 list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
168 list_del_init(&root->dirty_list);
169 free_extent_buffer(root->commit_root);
170 root->commit_root = btrfs_root_node(root);
171 extent_io_tree_release(&root->dirty_log_pages);
172 btrfs_qgroup_clean_swapped_blocks(root);
175 /* We can free old roots now. */
176 spin_lock(&cur_trans->dropped_roots_lock);
177 while (!list_empty(&cur_trans->dropped_roots)) {
178 root = list_first_entry(&cur_trans->dropped_roots,
179 struct btrfs_root, root_list);
180 list_del_init(&root->root_list);
181 spin_unlock(&cur_trans->dropped_roots_lock);
182 btrfs_free_log(trans, root);
183 btrfs_drop_and_free_fs_root(fs_info, root);
184 spin_lock(&cur_trans->dropped_roots_lock);
186 spin_unlock(&cur_trans->dropped_roots_lock);
189 * We have to update the last_byte_to_unpin under the commit_root_sem,
190 * at the same time we swap out the commit roots.
192 * This is because we must have a real view of the last spot the caching
193 * kthreads were while caching. Consider the following views of the
194 * extent tree for a block group
197 * +----+----+----+----+----+----+----+
198 * |\\\\| |\\\\|\\\\| |\\\\|\\\\|
199 * +----+----+----+----+----+----+----+
203 * +----+----+----+----+----+----+----+
204 * | | | |\\\\| | |\\\\|
205 * +----+----+----+----+----+----+----+
208 * If the cache_ctl->progress was at 3, then we are only allowed to
209 * unpin [0,1) and [2,3], because the caching thread has already
210 * processed those extents. We are not allowed to unpin [5,6), because
211 * the caching thread will re-start it's search from 3, and thus find
212 * the hole from [4,6) to add to the free space cache.
214 spin_lock(&fs_info->block_group_cache_lock);
215 list_for_each_entry_safe(caching_ctl, next,
216 &fs_info->caching_block_groups, list) {
217 struct btrfs_block_group *cache = caching_ctl->block_group;
219 if (btrfs_block_group_done(cache)) {
220 cache->last_byte_to_unpin = (u64)-1;
221 list_del_init(&caching_ctl->list);
222 btrfs_put_caching_control(caching_ctl);
224 cache->last_byte_to_unpin = caching_ctl->progress;
227 spin_unlock(&fs_info->block_group_cache_lock);
228 up_write(&fs_info->commit_root_sem);
231 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
234 if (type & TRANS_EXTWRITERS)
235 atomic_inc(&trans->num_extwriters);
238 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
241 if (type & TRANS_EXTWRITERS)
242 atomic_dec(&trans->num_extwriters);
245 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
248 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
251 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
253 return atomic_read(&trans->num_extwriters);
257 * To be called after all the new block groups attached to the transaction
258 * handle have been created (btrfs_create_pending_block_groups()).
260 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
262 struct btrfs_fs_info *fs_info = trans->fs_info;
263 struct btrfs_transaction *cur_trans = trans->transaction;
265 if (!trans->chunk_bytes_reserved)
268 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
270 btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
271 trans->chunk_bytes_reserved, NULL);
272 atomic64_sub(trans->chunk_bytes_reserved, &cur_trans->chunk_bytes_reserved);
273 cond_wake_up(&cur_trans->chunk_reserve_wait);
274 trans->chunk_bytes_reserved = 0;
278 * either allocate a new transaction or hop into the existing one
280 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
283 struct btrfs_transaction *cur_trans;
285 spin_lock(&fs_info->trans_lock);
287 /* The file system has been taken offline. No new transactions. */
288 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
289 spin_unlock(&fs_info->trans_lock);
293 cur_trans = fs_info->running_transaction;
295 if (TRANS_ABORTED(cur_trans)) {
296 spin_unlock(&fs_info->trans_lock);
297 return cur_trans->aborted;
299 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
300 spin_unlock(&fs_info->trans_lock);
303 refcount_inc(&cur_trans->use_count);
304 atomic_inc(&cur_trans->num_writers);
305 extwriter_counter_inc(cur_trans, type);
306 spin_unlock(&fs_info->trans_lock);
309 spin_unlock(&fs_info->trans_lock);
312 * If we are ATTACH, we just want to catch the current transaction,
313 * and commit it. If there is no transaction, just return ENOENT.
315 if (type == TRANS_ATTACH)
319 * JOIN_NOLOCK only happens during the transaction commit, so
320 * it is impossible that ->running_transaction is NULL
322 BUG_ON(type == TRANS_JOIN_NOLOCK);
324 cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
328 spin_lock(&fs_info->trans_lock);
329 if (fs_info->running_transaction) {
331 * someone started a transaction after we unlocked. Make sure
332 * to redo the checks above
336 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
337 spin_unlock(&fs_info->trans_lock);
342 cur_trans->fs_info = fs_info;
343 atomic_set(&cur_trans->pending_ordered, 0);
344 init_waitqueue_head(&cur_trans->pending_wait);
345 atomic_set(&cur_trans->num_writers, 1);
346 extwriter_counter_init(cur_trans, type);
347 init_waitqueue_head(&cur_trans->writer_wait);
348 init_waitqueue_head(&cur_trans->commit_wait);
349 cur_trans->state = TRANS_STATE_RUNNING;
351 * One for this trans handle, one so it will live on until we
352 * commit the transaction.
354 refcount_set(&cur_trans->use_count, 2);
355 cur_trans->flags = 0;
356 cur_trans->start_time = ktime_get_seconds();
358 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
360 cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
361 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
362 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
365 * although the tree mod log is per file system and not per transaction,
366 * the log must never go across transaction boundaries.
369 if (!list_empty(&fs_info->tree_mod_seq_list))
370 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
371 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
372 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
373 atomic64_set(&fs_info->tree_mod_seq, 0);
375 spin_lock_init(&cur_trans->delayed_refs.lock);
377 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
378 INIT_LIST_HEAD(&cur_trans->dev_update_list);
379 INIT_LIST_HEAD(&cur_trans->switch_commits);
380 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
381 INIT_LIST_HEAD(&cur_trans->io_bgs);
382 INIT_LIST_HEAD(&cur_trans->dropped_roots);
383 mutex_init(&cur_trans->cache_write_mutex);
384 spin_lock_init(&cur_trans->dirty_bgs_lock);
385 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
386 spin_lock_init(&cur_trans->dropped_roots_lock);
387 INIT_LIST_HEAD(&cur_trans->releasing_ebs);
388 spin_lock_init(&cur_trans->releasing_ebs_lock);
389 atomic64_set(&cur_trans->chunk_bytes_reserved, 0);
390 init_waitqueue_head(&cur_trans->chunk_reserve_wait);
391 list_add_tail(&cur_trans->list, &fs_info->trans_list);
392 extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
393 IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode);
394 extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
395 IO_TREE_FS_PINNED_EXTENTS, NULL);
396 fs_info->generation++;
397 cur_trans->transid = fs_info->generation;
398 fs_info->running_transaction = cur_trans;
399 cur_trans->aborted = 0;
400 spin_unlock(&fs_info->trans_lock);
406 * This does all the record keeping required to make sure that a shareable root
407 * is properly recorded in a given transaction. This is required to make sure
408 * the old root from before we joined the transaction is deleted when the
409 * transaction commits.
411 static int record_root_in_trans(struct btrfs_trans_handle *trans,
412 struct btrfs_root *root,
415 struct btrfs_fs_info *fs_info = root->fs_info;
417 if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
418 root->last_trans < trans->transid) || force) {
419 WARN_ON(root == fs_info->extent_root);
420 WARN_ON(!force && root->commit_root != root->node);
423 * see below for IN_TRANS_SETUP usage rules
424 * we have the reloc mutex held now, so there
425 * is only one writer in this function
427 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
429 /* make sure readers find IN_TRANS_SETUP before
430 * they find our root->last_trans update
434 spin_lock(&fs_info->fs_roots_radix_lock);
435 if (root->last_trans == trans->transid && !force) {
436 spin_unlock(&fs_info->fs_roots_radix_lock);
439 radix_tree_tag_set(&fs_info->fs_roots_radix,
440 (unsigned long)root->root_key.objectid,
441 BTRFS_ROOT_TRANS_TAG);
442 spin_unlock(&fs_info->fs_roots_radix_lock);
443 root->last_trans = trans->transid;
445 /* this is pretty tricky. We don't want to
446 * take the relocation lock in btrfs_record_root_in_trans
447 * unless we're really doing the first setup for this root in
450 * Normally we'd use root->last_trans as a flag to decide
451 * if we want to take the expensive mutex.
453 * But, we have to set root->last_trans before we
454 * init the relocation root, otherwise, we trip over warnings
455 * in ctree.c. The solution used here is to flag ourselves
456 * with root IN_TRANS_SETUP. When this is 1, we're still
457 * fixing up the reloc trees and everyone must wait.
459 * When this is zero, they can trust root->last_trans and fly
460 * through btrfs_record_root_in_trans without having to take the
461 * lock. smp_wmb() makes sure that all the writes above are
462 * done before we pop in the zero below
464 btrfs_init_reloc_root(trans, root);
465 smp_mb__before_atomic();
466 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
472 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
473 struct btrfs_root *root)
475 struct btrfs_fs_info *fs_info = root->fs_info;
476 struct btrfs_transaction *cur_trans = trans->transaction;
478 /* Add ourselves to the transaction dropped list */
479 spin_lock(&cur_trans->dropped_roots_lock);
480 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
481 spin_unlock(&cur_trans->dropped_roots_lock);
483 /* Make sure we don't try to update the root at commit time */
484 spin_lock(&fs_info->fs_roots_radix_lock);
485 radix_tree_tag_clear(&fs_info->fs_roots_radix,
486 (unsigned long)root->root_key.objectid,
487 BTRFS_ROOT_TRANS_TAG);
488 spin_unlock(&fs_info->fs_roots_radix_lock);
491 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
492 struct btrfs_root *root)
494 struct btrfs_fs_info *fs_info = root->fs_info;
496 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
500 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
504 if (root->last_trans == trans->transid &&
505 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
508 mutex_lock(&fs_info->reloc_mutex);
509 record_root_in_trans(trans, root, 0);
510 mutex_unlock(&fs_info->reloc_mutex);
515 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
517 return (trans->state >= TRANS_STATE_COMMIT_START &&
518 trans->state < TRANS_STATE_UNBLOCKED &&
519 !TRANS_ABORTED(trans));
522 /* wait for commit against the current transaction to become unblocked
523 * when this is done, it is safe to start a new transaction, but the current
524 * transaction might not be fully on disk.
526 static void wait_current_trans(struct btrfs_fs_info *fs_info)
528 struct btrfs_transaction *cur_trans;
530 spin_lock(&fs_info->trans_lock);
531 cur_trans = fs_info->running_transaction;
532 if (cur_trans && is_transaction_blocked(cur_trans)) {
533 refcount_inc(&cur_trans->use_count);
534 spin_unlock(&fs_info->trans_lock);
536 wait_event(fs_info->transaction_wait,
537 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
538 TRANS_ABORTED(cur_trans));
539 btrfs_put_transaction(cur_trans);
541 spin_unlock(&fs_info->trans_lock);
545 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
547 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
550 if (type == TRANS_START)
556 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
558 struct btrfs_fs_info *fs_info = root->fs_info;
560 if (!fs_info->reloc_ctl ||
561 !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
562 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
569 static struct btrfs_trans_handle *
570 start_transaction(struct btrfs_root *root, unsigned int num_items,
571 unsigned int type, enum btrfs_reserve_flush_enum flush,
572 bool enforce_qgroups)
574 struct btrfs_fs_info *fs_info = root->fs_info;
575 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
576 struct btrfs_trans_handle *h;
577 struct btrfs_transaction *cur_trans;
579 u64 qgroup_reserved = 0;
580 bool reloc_reserved = false;
581 bool do_chunk_alloc = false;
584 /* Send isn't supposed to start transactions. */
585 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
587 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
588 return ERR_PTR(-EROFS);
590 if (current->journal_info) {
591 WARN_ON(type & TRANS_EXTWRITERS);
592 h = current->journal_info;
593 refcount_inc(&h->use_count);
594 WARN_ON(refcount_read(&h->use_count) > 2);
595 h->orig_rsv = h->block_rsv;
601 * Do the reservation before we join the transaction so we can do all
602 * the appropriate flushing if need be.
604 if (num_items && root != fs_info->chunk_root) {
605 struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
606 u64 delayed_refs_bytes = 0;
608 qgroup_reserved = num_items * fs_info->nodesize;
609 ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
615 * We want to reserve all the bytes we may need all at once, so
616 * we only do 1 enospc flushing cycle per transaction start. We
617 * accomplish this by simply assuming we'll do 2 x num_items
618 * worth of delayed refs updates in this trans handle, and
619 * refill that amount for whatever is missing in the reserve.
621 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
622 if (flush == BTRFS_RESERVE_FLUSH_ALL &&
623 delayed_refs_rsv->full == 0) {
624 delayed_refs_bytes = num_bytes;
629 * Do the reservation for the relocation root creation
631 if (need_reserve_reloc_root(root)) {
632 num_bytes += fs_info->nodesize;
633 reloc_reserved = true;
636 ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
639 if (delayed_refs_bytes) {
640 btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
642 num_bytes -= delayed_refs_bytes;
645 if (rsv->space_info->force_alloc)
646 do_chunk_alloc = true;
647 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
648 !delayed_refs_rsv->full) {
650 * Some people call with btrfs_start_transaction(root, 0)
651 * because they can be throttled, but have some other mechanism
652 * for reserving space. We still want these guys to refill the
653 * delayed block_rsv so just add 1 items worth of reservation
656 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
661 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
668 * If we are JOIN_NOLOCK we're already committing a transaction and
669 * waiting on this guy, so we don't need to do the sb_start_intwrite
670 * because we're already holding a ref. We need this because we could
671 * have raced in and did an fsync() on a file which can kick a commit
672 * and then we deadlock with somebody doing a freeze.
674 * If we are ATTACH, it means we just want to catch the current
675 * transaction and commit it, so we needn't do sb_start_intwrite().
677 if (type & __TRANS_FREEZABLE)
678 sb_start_intwrite(fs_info->sb);
680 if (may_wait_transaction(fs_info, type))
681 wait_current_trans(fs_info);
684 ret = join_transaction(fs_info, type);
686 wait_current_trans(fs_info);
687 if (unlikely(type == TRANS_ATTACH ||
688 type == TRANS_JOIN_NOSTART))
691 } while (ret == -EBUSY);
696 cur_trans = fs_info->running_transaction;
698 h->transid = cur_trans->transid;
699 h->transaction = cur_trans;
701 refcount_set(&h->use_count, 1);
702 h->fs_info = root->fs_info;
705 h->can_flush_pending_bgs = true;
706 INIT_LIST_HEAD(&h->new_bgs);
709 if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
710 may_wait_transaction(fs_info, type)) {
711 current->journal_info = h;
712 btrfs_commit_transaction(h);
717 trace_btrfs_space_reservation(fs_info, "transaction",
718 h->transid, num_bytes, 1);
719 h->block_rsv = &fs_info->trans_block_rsv;
720 h->bytes_reserved = num_bytes;
721 h->reloc_reserved = reloc_reserved;
725 if (!current->journal_info)
726 current->journal_info = h;
729 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
730 * ALLOC_FORCE the first run through, and then we won't allocate for
731 * anybody else who races in later. We don't care about the return
734 if (do_chunk_alloc && num_bytes) {
735 u64 flags = h->block_rsv->space_info->flags;
737 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
738 CHUNK_ALLOC_NO_FORCE);
742 * btrfs_record_root_in_trans() needs to alloc new extents, and may
743 * call btrfs_join_transaction() while we're also starting a
746 * Thus it need to be called after current->journal_info initialized,
747 * or we can deadlock.
749 ret = btrfs_record_root_in_trans(h, root);
752 * The transaction handle is fully initialized and linked with
753 * other structures so it needs to be ended in case of errors,
756 btrfs_end_transaction(h);
763 if (type & __TRANS_FREEZABLE)
764 sb_end_intwrite(fs_info->sb);
765 kmem_cache_free(btrfs_trans_handle_cachep, h);
768 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
771 btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
775 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
776 unsigned int num_items)
778 return start_transaction(root, num_items, TRANS_START,
779 BTRFS_RESERVE_FLUSH_ALL, true);
782 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
783 struct btrfs_root *root,
784 unsigned int num_items)
786 return start_transaction(root, num_items, TRANS_START,
787 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
790 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
792 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
796 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
798 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
799 BTRFS_RESERVE_NO_FLUSH, true);
803 * Similar to regular join but it never starts a transaction when none is
804 * running or after waiting for the current one to finish.
806 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
808 return start_transaction(root, 0, TRANS_JOIN_NOSTART,
809 BTRFS_RESERVE_NO_FLUSH, true);
813 * btrfs_attach_transaction() - catch the running transaction
815 * It is used when we want to commit the current the transaction, but
816 * don't want to start a new one.
818 * Note: If this function return -ENOENT, it just means there is no
819 * running transaction. But it is possible that the inactive transaction
820 * is still in the memory, not fully on disk. If you hope there is no
821 * inactive transaction in the fs when -ENOENT is returned, you should
823 * btrfs_attach_transaction_barrier()
825 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
827 return start_transaction(root, 0, TRANS_ATTACH,
828 BTRFS_RESERVE_NO_FLUSH, true);
832 * btrfs_attach_transaction_barrier() - catch the running transaction
834 * It is similar to the above function, the difference is this one
835 * will wait for all the inactive transactions until they fully
838 struct btrfs_trans_handle *
839 btrfs_attach_transaction_barrier(struct btrfs_root *root)
841 struct btrfs_trans_handle *trans;
843 trans = start_transaction(root, 0, TRANS_ATTACH,
844 BTRFS_RESERVE_NO_FLUSH, true);
845 if (trans == ERR_PTR(-ENOENT))
846 btrfs_wait_for_commit(root->fs_info, 0);
851 /* Wait for a transaction commit to reach at least the given state. */
852 static noinline void wait_for_commit(struct btrfs_transaction *commit,
853 const enum btrfs_trans_state min_state)
855 wait_event(commit->commit_wait, commit->state >= min_state);
858 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
860 struct btrfs_transaction *cur_trans = NULL, *t;
864 if (transid <= fs_info->last_trans_committed)
867 /* find specified transaction */
868 spin_lock(&fs_info->trans_lock);
869 list_for_each_entry(t, &fs_info->trans_list, list) {
870 if (t->transid == transid) {
872 refcount_inc(&cur_trans->use_count);
876 if (t->transid > transid) {
881 spin_unlock(&fs_info->trans_lock);
884 * The specified transaction doesn't exist, or we
885 * raced with btrfs_commit_transaction
888 if (transid > fs_info->last_trans_committed)
893 /* find newest transaction that is committing | committed */
894 spin_lock(&fs_info->trans_lock);
895 list_for_each_entry_reverse(t, &fs_info->trans_list,
897 if (t->state >= TRANS_STATE_COMMIT_START) {
898 if (t->state == TRANS_STATE_COMPLETED)
901 refcount_inc(&cur_trans->use_count);
905 spin_unlock(&fs_info->trans_lock);
907 goto out; /* nothing committing|committed */
910 wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
911 btrfs_put_transaction(cur_trans);
916 void btrfs_throttle(struct btrfs_fs_info *fs_info)
918 wait_current_trans(fs_info);
921 static bool should_end_transaction(struct btrfs_trans_handle *trans)
923 struct btrfs_fs_info *fs_info = trans->fs_info;
925 if (btrfs_check_space_for_delayed_refs(fs_info))
928 return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
931 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
933 struct btrfs_transaction *cur_trans = trans->transaction;
935 if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
936 test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
939 return should_end_transaction(trans);
942 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
945 struct btrfs_fs_info *fs_info = trans->fs_info;
947 if (!trans->block_rsv) {
948 ASSERT(!trans->bytes_reserved);
952 if (!trans->bytes_reserved)
955 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
956 trace_btrfs_space_reservation(fs_info, "transaction",
957 trans->transid, trans->bytes_reserved, 0);
958 btrfs_block_rsv_release(fs_info, trans->block_rsv,
959 trans->bytes_reserved, NULL);
960 trans->bytes_reserved = 0;
963 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
966 struct btrfs_fs_info *info = trans->fs_info;
967 struct btrfs_transaction *cur_trans = trans->transaction;
970 if (refcount_read(&trans->use_count) > 1) {
971 refcount_dec(&trans->use_count);
972 trans->block_rsv = trans->orig_rsv;
976 btrfs_trans_release_metadata(trans);
977 trans->block_rsv = NULL;
979 btrfs_create_pending_block_groups(trans);
981 btrfs_trans_release_chunk_metadata(trans);
983 if (trans->type & __TRANS_FREEZABLE)
984 sb_end_intwrite(info->sb);
986 WARN_ON(cur_trans != info->running_transaction);
987 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
988 atomic_dec(&cur_trans->num_writers);
989 extwriter_counter_dec(cur_trans, trans->type);
991 cond_wake_up(&cur_trans->writer_wait);
992 btrfs_put_transaction(cur_trans);
994 if (current->journal_info == trans)
995 current->journal_info = NULL;
998 btrfs_run_delayed_iputs(info);
1000 if (TRANS_ABORTED(trans) ||
1001 test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
1002 wake_up_process(info->transaction_kthread);
1003 if (TRANS_ABORTED(trans))
1004 err = trans->aborted;
1009 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1013 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1015 return __btrfs_end_transaction(trans, 0);
1018 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1020 return __btrfs_end_transaction(trans, 1);
1024 * when btree blocks are allocated, they have some corresponding bits set for
1025 * them in one of two extent_io trees. This is used to make sure all of
1026 * those extents are sent to disk but does not wait on them
1028 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1029 struct extent_io_tree *dirty_pages, int mark)
1033 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1034 struct extent_state *cached_state = NULL;
1038 atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1039 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1040 mark, &cached_state)) {
1041 bool wait_writeback = false;
1043 err = convert_extent_bit(dirty_pages, start, end,
1045 mark, &cached_state);
1047 * convert_extent_bit can return -ENOMEM, which is most of the
1048 * time a temporary error. So when it happens, ignore the error
1049 * and wait for writeback of this range to finish - because we
1050 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1051 * to __btrfs_wait_marked_extents() would not know that
1052 * writeback for this range started and therefore wouldn't
1053 * wait for it to finish - we don't want to commit a
1054 * superblock that points to btree nodes/leafs for which
1055 * writeback hasn't finished yet (and without errors).
1056 * We cleanup any entries left in the io tree when committing
1057 * the transaction (through extent_io_tree_release()).
1059 if (err == -ENOMEM) {
1061 wait_writeback = true;
1064 err = filemap_fdatawrite_range(mapping, start, end);
1067 else if (wait_writeback)
1068 werr = filemap_fdatawait_range(mapping, start, end);
1069 free_extent_state(cached_state);
1070 cached_state = NULL;
1074 atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1079 * when btree blocks are allocated, they have some corresponding bits set for
1080 * them in one of two extent_io trees. This is used to make sure all of
1081 * those extents are on disk for transaction or log commit. We wait
1082 * on all the pages and clear them from the dirty pages state tree
1084 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1085 struct extent_io_tree *dirty_pages)
1089 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1090 struct extent_state *cached_state = NULL;
1094 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1095 EXTENT_NEED_WAIT, &cached_state)) {
1097 * Ignore -ENOMEM errors returned by clear_extent_bit().
1098 * When committing the transaction, we'll remove any entries
1099 * left in the io tree. For a log commit, we don't remove them
1100 * after committing the log because the tree can be accessed
1101 * concurrently - we do it only at transaction commit time when
1102 * it's safe to do it (through extent_io_tree_release()).
1104 err = clear_extent_bit(dirty_pages, start, end,
1105 EXTENT_NEED_WAIT, 0, 0, &cached_state);
1109 err = filemap_fdatawait_range(mapping, start, end);
1112 free_extent_state(cached_state);
1113 cached_state = NULL;
1122 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1123 struct extent_io_tree *dirty_pages)
1125 bool errors = false;
1128 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1129 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1137 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1139 struct btrfs_fs_info *fs_info = log_root->fs_info;
1140 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1141 bool errors = false;
1144 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1146 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1147 if ((mark & EXTENT_DIRTY) &&
1148 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1151 if ((mark & EXTENT_NEW) &&
1152 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1161 * When btree blocks are allocated the corresponding extents are marked dirty.
1162 * This function ensures such extents are persisted on disk for transaction or
1165 * @trans: transaction whose dirty pages we'd like to write
1167 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1171 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1172 struct btrfs_fs_info *fs_info = trans->fs_info;
1173 struct blk_plug plug;
1175 blk_start_plug(&plug);
1176 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1177 blk_finish_plug(&plug);
1178 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1180 extent_io_tree_release(&trans->transaction->dirty_pages);
1191 * this is used to update the root pointer in the tree of tree roots.
1193 * But, in the case of the extent allocation tree, updating the root
1194 * pointer may allocate blocks which may change the root of the extent
1197 * So, this loops and repeats and makes sure the cowonly root didn't
1198 * change while the root pointer was being updated in the metadata.
1200 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1201 struct btrfs_root *root)
1204 u64 old_root_bytenr;
1206 struct btrfs_fs_info *fs_info = root->fs_info;
1207 struct btrfs_root *tree_root = fs_info->tree_root;
1209 old_root_used = btrfs_root_used(&root->root_item);
1212 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1213 if (old_root_bytenr == root->node->start &&
1214 old_root_used == btrfs_root_used(&root->root_item))
1217 btrfs_set_root_node(&root->root_item, root->node);
1218 ret = btrfs_update_root(trans, tree_root,
1224 old_root_used = btrfs_root_used(&root->root_item);
1231 * update all the cowonly tree roots on disk
1233 * The error handling in this function may not be obvious. Any of the
1234 * failures will cause the file system to go offline. We still need
1235 * to clean up the delayed refs.
1237 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1239 struct btrfs_fs_info *fs_info = trans->fs_info;
1240 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1241 struct list_head *io_bgs = &trans->transaction->io_bgs;
1242 struct list_head *next;
1243 struct extent_buffer *eb;
1246 eb = btrfs_lock_root_node(fs_info->tree_root);
1247 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1248 0, &eb, BTRFS_NESTING_COW);
1249 btrfs_tree_unlock(eb);
1250 free_extent_buffer(eb);
1255 ret = btrfs_run_dev_stats(trans);
1258 ret = btrfs_run_dev_replace(trans);
1261 ret = btrfs_run_qgroups(trans);
1265 ret = btrfs_setup_space_cache(trans);
1270 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1271 struct btrfs_root *root;
1272 next = fs_info->dirty_cowonly_roots.next;
1273 list_del_init(next);
1274 root = list_entry(next, struct btrfs_root, dirty_list);
1275 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1277 if (root != fs_info->extent_root)
1278 list_add_tail(&root->dirty_list,
1279 &trans->transaction->switch_commits);
1280 ret = update_cowonly_root(trans, root);
1285 /* Now flush any delayed refs generated by updating all of the roots */
1286 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1290 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1291 ret = btrfs_write_dirty_block_groups(trans);
1296 * We're writing the dirty block groups, which could generate
1297 * delayed refs, which could generate more dirty block groups,
1298 * so we want to keep this flushing in this loop to make sure
1299 * everything gets run.
1301 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1306 if (!list_empty(&fs_info->dirty_cowonly_roots))
1309 list_add_tail(&fs_info->extent_root->dirty_list,
1310 &trans->transaction->switch_commits);
1312 /* Update dev-replace pointer once everything is committed */
1313 fs_info->dev_replace.committed_cursor_left =
1314 fs_info->dev_replace.cursor_left_last_write_of_item;
1320 * dead roots are old snapshots that need to be deleted. This allocates
1321 * a dirty root struct and adds it into the list of dead roots that need to
1324 void btrfs_add_dead_root(struct btrfs_root *root)
1326 struct btrfs_fs_info *fs_info = root->fs_info;
1328 spin_lock(&fs_info->trans_lock);
1329 if (list_empty(&root->root_list)) {
1330 btrfs_grab_root(root);
1331 list_add_tail(&root->root_list, &fs_info->dead_roots);
1333 spin_unlock(&fs_info->trans_lock);
1337 * update all the cowonly tree roots on disk
1339 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1341 struct btrfs_fs_info *fs_info = trans->fs_info;
1342 struct btrfs_root *gang[8];
1346 spin_lock(&fs_info->fs_roots_radix_lock);
1348 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1351 BTRFS_ROOT_TRANS_TAG);
1354 for (i = 0; i < ret; i++) {
1355 struct btrfs_root *root = gang[i];
1358 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1359 (unsigned long)root->root_key.objectid,
1360 BTRFS_ROOT_TRANS_TAG);
1361 spin_unlock(&fs_info->fs_roots_radix_lock);
1363 btrfs_free_log(trans, root);
1364 btrfs_update_reloc_root(trans, root);
1366 /* see comments in should_cow_block() */
1367 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1368 smp_mb__after_atomic();
1370 if (root->commit_root != root->node) {
1371 list_add_tail(&root->dirty_list,
1372 &trans->transaction->switch_commits);
1373 btrfs_set_root_node(&root->root_item,
1377 ret2 = btrfs_update_root(trans, fs_info->tree_root,
1382 spin_lock(&fs_info->fs_roots_radix_lock);
1383 btrfs_qgroup_free_meta_all_pertrans(root);
1386 spin_unlock(&fs_info->fs_roots_radix_lock);
1391 * defrag a given btree.
1392 * Every leaf in the btree is read and defragged.
1394 int btrfs_defrag_root(struct btrfs_root *root)
1396 struct btrfs_fs_info *info = root->fs_info;
1397 struct btrfs_trans_handle *trans;
1400 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1404 trans = btrfs_start_transaction(root, 0);
1406 return PTR_ERR(trans);
1408 ret = btrfs_defrag_leaves(trans, root);
1410 btrfs_end_transaction(trans);
1411 btrfs_btree_balance_dirty(info);
1414 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1417 if (btrfs_defrag_cancelled(info)) {
1418 btrfs_debug(info, "defrag_root cancelled");
1423 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1428 * Do all special snapshot related qgroup dirty hack.
1430 * Will do all needed qgroup inherit and dirty hack like switch commit
1431 * roots inside one transaction and write all btree into disk, to make
1434 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1435 struct btrfs_root *src,
1436 struct btrfs_root *parent,
1437 struct btrfs_qgroup_inherit *inherit,
1440 struct btrfs_fs_info *fs_info = src->fs_info;
1444 * Save some performance in the case that qgroups are not
1445 * enabled. If this check races with the ioctl, rescan will
1448 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1452 * Ensure dirty @src will be committed. Or, after coming
1453 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1454 * recorded root will never be updated again, causing an outdated root
1457 ret = record_root_in_trans(trans, src, 1);
1462 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1463 * src root, so we must run the delayed refs here.
1465 * However this isn't particularly fool proof, because there's no
1466 * synchronization keeping us from changing the tree after this point
1467 * before we do the qgroup_inherit, or even from making changes while
1468 * we're doing the qgroup_inherit. But that's a problem for the future,
1469 * for now flush the delayed refs to narrow the race window where the
1470 * qgroup counters could end up wrong.
1472 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1474 btrfs_abort_transaction(trans, ret);
1479 * We are going to commit transaction, see btrfs_commit_transaction()
1480 * comment for reason locking tree_log_mutex
1482 mutex_lock(&fs_info->tree_log_mutex);
1484 ret = commit_fs_roots(trans);
1487 ret = btrfs_qgroup_account_extents(trans);
1491 /* Now qgroup are all updated, we can inherit it to new qgroups */
1492 ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1498 * Now we do a simplified commit transaction, which will:
1499 * 1) commit all subvolume and extent tree
1500 * To ensure all subvolume and extent tree have a valid
1501 * commit_root to accounting later insert_dir_item()
1502 * 2) write all btree blocks onto disk
1503 * This is to make sure later btree modification will be cowed
1504 * Or commit_root can be populated and cause wrong qgroup numbers
1505 * In this simplified commit, we don't really care about other trees
1506 * like chunk and root tree, as they won't affect qgroup.
1507 * And we don't write super to avoid half committed status.
1509 ret = commit_cowonly_roots(trans);
1512 switch_commit_roots(trans);
1513 ret = btrfs_write_and_wait_transaction(trans);
1515 btrfs_handle_fs_error(fs_info, ret,
1516 "Error while writing out transaction for qgroup");
1519 mutex_unlock(&fs_info->tree_log_mutex);
1522 * Force parent root to be updated, as we recorded it before so its
1523 * last_trans == cur_transid.
1524 * Or it won't be committed again onto disk after later
1528 ret = record_root_in_trans(trans, parent, 1);
1533 * new snapshots need to be created at a very specific time in the
1534 * transaction commit. This does the actual creation.
1537 * If the error which may affect the commitment of the current transaction
1538 * happens, we should return the error number. If the error which just affect
1539 * the creation of the pending snapshots, just return 0.
1541 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1542 struct btrfs_pending_snapshot *pending)
1545 struct btrfs_fs_info *fs_info = trans->fs_info;
1546 struct btrfs_key key;
1547 struct btrfs_root_item *new_root_item;
1548 struct btrfs_root *tree_root = fs_info->tree_root;
1549 struct btrfs_root *root = pending->root;
1550 struct btrfs_root *parent_root;
1551 struct btrfs_block_rsv *rsv;
1552 struct inode *parent_inode;
1553 struct btrfs_path *path;
1554 struct btrfs_dir_item *dir_item;
1555 struct dentry *dentry;
1556 struct extent_buffer *tmp;
1557 struct extent_buffer *old;
1558 struct timespec64 cur_time;
1565 ASSERT(pending->path);
1566 path = pending->path;
1568 ASSERT(pending->root_item);
1569 new_root_item = pending->root_item;
1571 pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1573 goto no_free_objectid;
1576 * Make qgroup to skip current new snapshot's qgroupid, as it is
1577 * accounted by later btrfs_qgroup_inherit().
1579 btrfs_set_skip_qgroup(trans, objectid);
1581 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1583 if (to_reserve > 0) {
1584 pending->error = btrfs_block_rsv_add(root,
1585 &pending->block_rsv,
1587 BTRFS_RESERVE_NO_FLUSH);
1589 goto clear_skip_qgroup;
1592 key.objectid = objectid;
1593 key.offset = (u64)-1;
1594 key.type = BTRFS_ROOT_ITEM_KEY;
1596 rsv = trans->block_rsv;
1597 trans->block_rsv = &pending->block_rsv;
1598 trans->bytes_reserved = trans->block_rsv->reserved;
1599 trace_btrfs_space_reservation(fs_info, "transaction",
1601 trans->bytes_reserved, 1);
1602 dentry = pending->dentry;
1603 parent_inode = pending->dir;
1604 parent_root = BTRFS_I(parent_inode)->root;
1605 record_root_in_trans(trans, parent_root, 0);
1607 cur_time = current_time(parent_inode);
1610 * insert the directory item
1612 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1613 BUG_ON(ret); /* -ENOMEM */
1615 /* check if there is a file/dir which has the same name. */
1616 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1617 btrfs_ino(BTRFS_I(parent_inode)),
1618 dentry->d_name.name,
1619 dentry->d_name.len, 0);
1620 if (dir_item != NULL && !IS_ERR(dir_item)) {
1621 pending->error = -EEXIST;
1622 goto dir_item_existed;
1623 } else if (IS_ERR(dir_item)) {
1624 ret = PTR_ERR(dir_item);
1625 btrfs_abort_transaction(trans, ret);
1628 btrfs_release_path(path);
1631 * pull in the delayed directory update
1632 * and the delayed inode item
1633 * otherwise we corrupt the FS during
1636 ret = btrfs_run_delayed_items(trans);
1637 if (ret) { /* Transaction aborted */
1638 btrfs_abort_transaction(trans, ret);
1642 record_root_in_trans(trans, root, 0);
1643 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1644 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1645 btrfs_check_and_init_root_item(new_root_item);
1647 root_flags = btrfs_root_flags(new_root_item);
1648 if (pending->readonly)
1649 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1651 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1652 btrfs_set_root_flags(new_root_item, root_flags);
1654 btrfs_set_root_generation_v2(new_root_item,
1656 generate_random_guid(new_root_item->uuid);
1657 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1659 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1660 memset(new_root_item->received_uuid, 0,
1661 sizeof(new_root_item->received_uuid));
1662 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1663 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1664 btrfs_set_root_stransid(new_root_item, 0);
1665 btrfs_set_root_rtransid(new_root_item, 0);
1667 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1668 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1669 btrfs_set_root_otransid(new_root_item, trans->transid);
1671 old = btrfs_lock_root_node(root);
1672 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1675 btrfs_tree_unlock(old);
1676 free_extent_buffer(old);
1677 btrfs_abort_transaction(trans, ret);
1681 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1682 /* clean up in any case */
1683 btrfs_tree_unlock(old);
1684 free_extent_buffer(old);
1686 btrfs_abort_transaction(trans, ret);
1689 /* see comments in should_cow_block() */
1690 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1693 btrfs_set_root_node(new_root_item, tmp);
1694 /* record when the snapshot was created in key.offset */
1695 key.offset = trans->transid;
1696 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1697 btrfs_tree_unlock(tmp);
1698 free_extent_buffer(tmp);
1700 btrfs_abort_transaction(trans, ret);
1705 * insert root back/forward references
1707 ret = btrfs_add_root_ref(trans, objectid,
1708 parent_root->root_key.objectid,
1709 btrfs_ino(BTRFS_I(parent_inode)), index,
1710 dentry->d_name.name, dentry->d_name.len);
1712 btrfs_abort_transaction(trans, ret);
1716 key.offset = (u64)-1;
1717 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1718 if (IS_ERR(pending->snap)) {
1719 ret = PTR_ERR(pending->snap);
1720 pending->snap = NULL;
1721 btrfs_abort_transaction(trans, ret);
1725 ret = btrfs_reloc_post_snapshot(trans, pending);
1727 btrfs_abort_transaction(trans, ret);
1732 * Do special qgroup accounting for snapshot, as we do some qgroup
1733 * snapshot hack to do fast snapshot.
1734 * To co-operate with that hack, we do hack again.
1735 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1737 ret = qgroup_account_snapshot(trans, root, parent_root,
1738 pending->inherit, objectid);
1742 ret = btrfs_insert_dir_item(trans, dentry->d_name.name,
1743 dentry->d_name.len, BTRFS_I(parent_inode),
1744 &key, BTRFS_FT_DIR, index);
1745 /* We have check then name at the beginning, so it is impossible. */
1746 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1748 btrfs_abort_transaction(trans, ret);
1752 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1753 dentry->d_name.len * 2);
1754 parent_inode->i_mtime = parent_inode->i_ctime =
1755 current_time(parent_inode);
1756 ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1758 btrfs_abort_transaction(trans, ret);
1761 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1762 BTRFS_UUID_KEY_SUBVOL,
1765 btrfs_abort_transaction(trans, ret);
1768 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1769 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1770 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1772 if (ret && ret != -EEXIST) {
1773 btrfs_abort_transaction(trans, ret);
1779 pending->error = ret;
1781 trans->block_rsv = rsv;
1782 trans->bytes_reserved = 0;
1784 btrfs_clear_skip_qgroup(trans);
1786 kfree(new_root_item);
1787 pending->root_item = NULL;
1788 btrfs_free_path(path);
1789 pending->path = NULL;
1795 * create all the snapshots we've scheduled for creation
1797 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1799 struct btrfs_pending_snapshot *pending, *next;
1800 struct list_head *head = &trans->transaction->pending_snapshots;
1803 list_for_each_entry_safe(pending, next, head, list) {
1804 list_del(&pending->list);
1805 ret = create_pending_snapshot(trans, pending);
1812 static void update_super_roots(struct btrfs_fs_info *fs_info)
1814 struct btrfs_root_item *root_item;
1815 struct btrfs_super_block *super;
1817 super = fs_info->super_copy;
1819 root_item = &fs_info->chunk_root->root_item;
1820 super->chunk_root = root_item->bytenr;
1821 super->chunk_root_generation = root_item->generation;
1822 super->chunk_root_level = root_item->level;
1824 root_item = &fs_info->tree_root->root_item;
1825 super->root = root_item->bytenr;
1826 super->generation = root_item->generation;
1827 super->root_level = root_item->level;
1828 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1829 super->cache_generation = root_item->generation;
1830 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1831 super->cache_generation = 0;
1832 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1833 super->uuid_tree_generation = root_item->generation;
1836 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1838 struct btrfs_transaction *trans;
1841 spin_lock(&info->trans_lock);
1842 trans = info->running_transaction;
1844 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1845 spin_unlock(&info->trans_lock);
1849 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1851 struct btrfs_transaction *trans;
1854 spin_lock(&info->trans_lock);
1855 trans = info->running_transaction;
1857 ret = is_transaction_blocked(trans);
1858 spin_unlock(&info->trans_lock);
1863 * wait for the current transaction commit to start and block subsequent
1866 static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info,
1867 struct btrfs_transaction *trans)
1869 wait_event(fs_info->transaction_blocked_wait,
1870 trans->state >= TRANS_STATE_COMMIT_START ||
1871 TRANS_ABORTED(trans));
1875 * wait for the current transaction to start and then become unblocked.
1878 static void wait_current_trans_commit_start_and_unblock(
1879 struct btrfs_fs_info *fs_info,
1880 struct btrfs_transaction *trans)
1882 wait_event(fs_info->transaction_wait,
1883 trans->state >= TRANS_STATE_UNBLOCKED ||
1884 TRANS_ABORTED(trans));
1888 * commit transactions asynchronously. once btrfs_commit_transaction_async
1889 * returns, any subsequent transaction will not be allowed to join.
1891 struct btrfs_async_commit {
1892 struct btrfs_trans_handle *newtrans;
1893 struct work_struct work;
1896 static void do_async_commit(struct work_struct *work)
1898 struct btrfs_async_commit *ac =
1899 container_of(work, struct btrfs_async_commit, work);
1902 * We've got freeze protection passed with the transaction.
1903 * Tell lockdep about it.
1905 if (ac->newtrans->type & __TRANS_FREEZABLE)
1906 __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1908 current->journal_info = ac->newtrans;
1910 btrfs_commit_transaction(ac->newtrans);
1914 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1915 int wait_for_unblock)
1917 struct btrfs_fs_info *fs_info = trans->fs_info;
1918 struct btrfs_async_commit *ac;
1919 struct btrfs_transaction *cur_trans;
1921 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1925 INIT_WORK(&ac->work, do_async_commit);
1926 ac->newtrans = btrfs_join_transaction(trans->root);
1927 if (IS_ERR(ac->newtrans)) {
1928 int err = PTR_ERR(ac->newtrans);
1933 /* take transaction reference */
1934 cur_trans = trans->transaction;
1935 refcount_inc(&cur_trans->use_count);
1937 btrfs_end_transaction(trans);
1940 * Tell lockdep we've released the freeze rwsem, since the
1941 * async commit thread will be the one to unlock it.
1943 if (ac->newtrans->type & __TRANS_FREEZABLE)
1944 __sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1946 schedule_work(&ac->work);
1948 /* wait for transaction to start and unblock */
1949 if (wait_for_unblock)
1950 wait_current_trans_commit_start_and_unblock(fs_info, cur_trans);
1952 wait_current_trans_commit_start(fs_info, cur_trans);
1954 if (current->journal_info == trans)
1955 current->journal_info = NULL;
1957 btrfs_put_transaction(cur_trans);
1962 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1964 struct btrfs_fs_info *fs_info = trans->fs_info;
1965 struct btrfs_transaction *cur_trans = trans->transaction;
1967 WARN_ON(refcount_read(&trans->use_count) > 1);
1969 btrfs_abort_transaction(trans, err);
1971 spin_lock(&fs_info->trans_lock);
1974 * If the transaction is removed from the list, it means this
1975 * transaction has been committed successfully, so it is impossible
1976 * to call the cleanup function.
1978 BUG_ON(list_empty(&cur_trans->list));
1980 list_del_init(&cur_trans->list);
1981 if (cur_trans == fs_info->running_transaction) {
1982 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1983 spin_unlock(&fs_info->trans_lock);
1984 wait_event(cur_trans->writer_wait,
1985 atomic_read(&cur_trans->num_writers) == 1);
1987 spin_lock(&fs_info->trans_lock);
1989 spin_unlock(&fs_info->trans_lock);
1991 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
1993 spin_lock(&fs_info->trans_lock);
1994 if (cur_trans == fs_info->running_transaction)
1995 fs_info->running_transaction = NULL;
1996 spin_unlock(&fs_info->trans_lock);
1998 if (trans->type & __TRANS_FREEZABLE)
1999 sb_end_intwrite(fs_info->sb);
2000 btrfs_put_transaction(cur_trans);
2001 btrfs_put_transaction(cur_trans);
2003 trace_btrfs_transaction_commit(trans->root);
2005 if (current->journal_info == trans)
2006 current->journal_info = NULL;
2007 btrfs_scrub_cancel(fs_info);
2009 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2013 * Release reserved delayed ref space of all pending block groups of the
2014 * transaction and remove them from the list
2016 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2018 struct btrfs_fs_info *fs_info = trans->fs_info;
2019 struct btrfs_block_group *block_group, *tmp;
2021 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2022 btrfs_delayed_refs_rsv_release(fs_info, 1);
2023 list_del_init(&block_group->bg_list);
2027 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2030 * We use writeback_inodes_sb here because if we used
2031 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2032 * Currently are holding the fs freeze lock, if we do an async flush
2033 * we'll do btrfs_join_transaction() and deadlock because we need to
2034 * wait for the fs freeze lock. Using the direct flushing we benefit
2035 * from already being in a transaction and our join_transaction doesn't
2036 * have to re-take the fs freeze lock.
2038 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2039 writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2043 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2045 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2046 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2049 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2051 struct btrfs_fs_info *fs_info = trans->fs_info;
2052 struct btrfs_transaction *cur_trans = trans->transaction;
2053 struct btrfs_transaction *prev_trans = NULL;
2056 ASSERT(refcount_read(&trans->use_count) == 1);
2059 * Some places just start a transaction to commit it. We need to make
2060 * sure that if this commit fails that the abort code actually marks the
2061 * transaction as failed, so set trans->dirty to make the abort code do
2064 trans->dirty = true;
2066 /* Stop the commit early if ->aborted is set */
2067 if (TRANS_ABORTED(cur_trans)) {
2068 ret = cur_trans->aborted;
2069 btrfs_end_transaction(trans);
2073 btrfs_trans_release_metadata(trans);
2074 trans->block_rsv = NULL;
2077 * We only want one transaction commit doing the flushing so we do not
2078 * waste a bunch of time on lock contention on the extent root node.
2080 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2081 &cur_trans->delayed_refs.flags)) {
2083 * Make a pass through all the delayed refs we have so far.
2084 * Any running threads may add more while we are here.
2086 ret = btrfs_run_delayed_refs(trans, 0);
2088 btrfs_end_transaction(trans);
2093 btrfs_create_pending_block_groups(trans);
2095 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2098 /* this mutex is also taken before trying to set
2099 * block groups readonly. We need to make sure
2100 * that nobody has set a block group readonly
2101 * after a extents from that block group have been
2102 * allocated for cache files. btrfs_set_block_group_ro
2103 * will wait for the transaction to commit if it
2104 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2106 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2107 * only one process starts all the block group IO. It wouldn't
2108 * hurt to have more than one go through, but there's no
2109 * real advantage to it either.
2111 mutex_lock(&fs_info->ro_block_group_mutex);
2112 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2115 mutex_unlock(&fs_info->ro_block_group_mutex);
2118 ret = btrfs_start_dirty_block_groups(trans);
2120 btrfs_end_transaction(trans);
2126 spin_lock(&fs_info->trans_lock);
2127 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2128 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2130 spin_unlock(&fs_info->trans_lock);
2131 refcount_inc(&cur_trans->use_count);
2133 if (trans->in_fsync)
2134 want_state = TRANS_STATE_SUPER_COMMITTED;
2135 ret = btrfs_end_transaction(trans);
2136 wait_for_commit(cur_trans, want_state);
2138 if (TRANS_ABORTED(cur_trans))
2139 ret = cur_trans->aborted;
2141 btrfs_put_transaction(cur_trans);
2146 cur_trans->state = TRANS_STATE_COMMIT_START;
2147 wake_up(&fs_info->transaction_blocked_wait);
2149 if (cur_trans->list.prev != &fs_info->trans_list) {
2150 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2152 if (trans->in_fsync)
2153 want_state = TRANS_STATE_SUPER_COMMITTED;
2155 prev_trans = list_entry(cur_trans->list.prev,
2156 struct btrfs_transaction, list);
2157 if (prev_trans->state < want_state) {
2158 refcount_inc(&prev_trans->use_count);
2159 spin_unlock(&fs_info->trans_lock);
2161 wait_for_commit(prev_trans, want_state);
2163 ret = READ_ONCE(prev_trans->aborted);
2165 btrfs_put_transaction(prev_trans);
2167 goto cleanup_transaction;
2169 spin_unlock(&fs_info->trans_lock);
2172 spin_unlock(&fs_info->trans_lock);
2174 * The previous transaction was aborted and was already removed
2175 * from the list of transactions at fs_info->trans_list. So we
2176 * abort to prevent writing a new superblock that reflects a
2177 * corrupt state (pointing to trees with unwritten nodes/leafs).
2179 if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) {
2181 goto cleanup_transaction;
2185 extwriter_counter_dec(cur_trans, trans->type);
2187 ret = btrfs_start_delalloc_flush(fs_info);
2189 goto cleanup_transaction;
2191 ret = btrfs_run_delayed_items(trans);
2193 goto cleanup_transaction;
2195 wait_event(cur_trans->writer_wait,
2196 extwriter_counter_read(cur_trans) == 0);
2198 /* some pending stuffs might be added after the previous flush. */
2199 ret = btrfs_run_delayed_items(trans);
2201 goto cleanup_transaction;
2203 btrfs_wait_delalloc_flush(fs_info);
2206 * Wait for all ordered extents started by a fast fsync that joined this
2207 * transaction. Otherwise if this transaction commits before the ordered
2208 * extents complete we lose logged data after a power failure.
2210 wait_event(cur_trans->pending_wait,
2211 atomic_read(&cur_trans->pending_ordered) == 0);
2213 btrfs_scrub_pause(fs_info);
2215 * Ok now we need to make sure to block out any other joins while we
2216 * commit the transaction. We could have started a join before setting
2217 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2219 spin_lock(&fs_info->trans_lock);
2220 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2221 spin_unlock(&fs_info->trans_lock);
2222 wait_event(cur_trans->writer_wait,
2223 atomic_read(&cur_trans->num_writers) == 1);
2225 if (TRANS_ABORTED(cur_trans)) {
2226 ret = cur_trans->aborted;
2227 goto scrub_continue;
2230 * the reloc mutex makes sure that we stop
2231 * the balancing code from coming in and moving
2232 * extents around in the middle of the commit
2234 mutex_lock(&fs_info->reloc_mutex);
2237 * We needn't worry about the delayed items because we will
2238 * deal with them in create_pending_snapshot(), which is the
2239 * core function of the snapshot creation.
2241 ret = create_pending_snapshots(trans);
2246 * We insert the dir indexes of the snapshots and update the inode
2247 * of the snapshots' parents after the snapshot creation, so there
2248 * are some delayed items which are not dealt with. Now deal with
2251 * We needn't worry that this operation will corrupt the snapshots,
2252 * because all the tree which are snapshoted will be forced to COW
2253 * the nodes and leaves.
2255 ret = btrfs_run_delayed_items(trans);
2259 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2264 * make sure none of the code above managed to slip in a
2267 btrfs_assert_delayed_root_empty(fs_info);
2269 WARN_ON(cur_trans != trans->transaction);
2271 /* btrfs_commit_tree_roots is responsible for getting the
2272 * various roots consistent with each other. Every pointer
2273 * in the tree of tree roots has to point to the most up to date
2274 * root for every subvolume and other tree. So, we have to keep
2275 * the tree logging code from jumping in and changing any
2278 * At this point in the commit, there can't be any tree-log
2279 * writers, but a little lower down we drop the trans mutex
2280 * and let new people in. By holding the tree_log_mutex
2281 * from now until after the super is written, we avoid races
2282 * with the tree-log code.
2284 mutex_lock(&fs_info->tree_log_mutex);
2286 ret = commit_fs_roots(trans);
2288 goto unlock_tree_log;
2291 * Since the transaction is done, we can apply the pending changes
2292 * before the next transaction.
2294 btrfs_apply_pending_changes(fs_info);
2296 /* commit_fs_roots gets rid of all the tree log roots, it is now
2297 * safe to free the root of tree log roots
2299 btrfs_free_log_root_tree(trans, fs_info);
2302 * Since fs roots are all committed, we can get a quite accurate
2303 * new_roots. So let's do quota accounting.
2305 ret = btrfs_qgroup_account_extents(trans);
2307 goto unlock_tree_log;
2309 ret = commit_cowonly_roots(trans);
2311 goto unlock_tree_log;
2314 * The tasks which save the space cache and inode cache may also
2315 * update ->aborted, check it.
2317 if (TRANS_ABORTED(cur_trans)) {
2318 ret = cur_trans->aborted;
2319 goto unlock_tree_log;
2322 cur_trans = fs_info->running_transaction;
2324 btrfs_set_root_node(&fs_info->tree_root->root_item,
2325 fs_info->tree_root->node);
2326 list_add_tail(&fs_info->tree_root->dirty_list,
2327 &cur_trans->switch_commits);
2329 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2330 fs_info->chunk_root->node);
2331 list_add_tail(&fs_info->chunk_root->dirty_list,
2332 &cur_trans->switch_commits);
2334 switch_commit_roots(trans);
2336 ASSERT(list_empty(&cur_trans->dirty_bgs));
2337 ASSERT(list_empty(&cur_trans->io_bgs));
2338 update_super_roots(fs_info);
2340 btrfs_set_super_log_root(fs_info->super_copy, 0);
2341 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2342 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2343 sizeof(*fs_info->super_copy));
2345 btrfs_commit_device_sizes(cur_trans);
2347 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2348 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2350 btrfs_trans_release_chunk_metadata(trans);
2352 spin_lock(&fs_info->trans_lock);
2353 cur_trans->state = TRANS_STATE_UNBLOCKED;
2354 fs_info->running_transaction = NULL;
2355 spin_unlock(&fs_info->trans_lock);
2356 mutex_unlock(&fs_info->reloc_mutex);
2358 wake_up(&fs_info->transaction_wait);
2360 ret = btrfs_write_and_wait_transaction(trans);
2362 btrfs_handle_fs_error(fs_info, ret,
2363 "Error while writing out transaction");
2365 * reloc_mutex has been unlocked, tree_log_mutex is still held
2366 * but we can't jump to unlock_tree_log causing double unlock
2368 mutex_unlock(&fs_info->tree_log_mutex);
2369 goto scrub_continue;
2373 * At this point, we should have written all the tree blocks allocated
2374 * in this transaction. So it's now safe to free the redirtyied extent
2377 btrfs_free_redirty_list(cur_trans);
2379 ret = write_all_supers(fs_info, 0);
2381 * the super is written, we can safely allow the tree-loggers
2382 * to go about their business
2384 mutex_unlock(&fs_info->tree_log_mutex);
2386 goto scrub_continue;
2389 * We needn't acquire the lock here because there is no other task
2390 * which can change it.
2392 cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2393 wake_up(&cur_trans->commit_wait);
2395 btrfs_finish_extent_commit(trans);
2397 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2398 btrfs_clear_space_info_full(fs_info);
2400 fs_info->last_trans_committed = cur_trans->transid;
2402 * We needn't acquire the lock here because there is no other task
2403 * which can change it.
2405 cur_trans->state = TRANS_STATE_COMPLETED;
2406 wake_up(&cur_trans->commit_wait);
2408 spin_lock(&fs_info->trans_lock);
2409 list_del_init(&cur_trans->list);
2410 spin_unlock(&fs_info->trans_lock);
2412 btrfs_put_transaction(cur_trans);
2413 btrfs_put_transaction(cur_trans);
2415 if (trans->type & __TRANS_FREEZABLE)
2416 sb_end_intwrite(fs_info->sb);
2418 trace_btrfs_transaction_commit(trans->root);
2420 btrfs_scrub_continue(fs_info);
2422 if (current->journal_info == trans)
2423 current->journal_info = NULL;
2425 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2430 mutex_unlock(&fs_info->tree_log_mutex);
2432 mutex_unlock(&fs_info->reloc_mutex);
2434 btrfs_scrub_continue(fs_info);
2435 cleanup_transaction:
2436 btrfs_trans_release_metadata(trans);
2437 btrfs_cleanup_pending_block_groups(trans);
2438 btrfs_trans_release_chunk_metadata(trans);
2439 trans->block_rsv = NULL;
2440 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2441 if (current->journal_info == trans)
2442 current->journal_info = NULL;
2443 cleanup_transaction(trans, ret);
2449 * return < 0 if error
2450 * 0 if there are no more dead_roots at the time of call
2451 * 1 there are more to be processed, call me again
2453 * The return value indicates there are certainly more snapshots to delete, but
2454 * if there comes a new one during processing, it may return 0. We don't mind,
2455 * because btrfs_commit_super will poke cleaner thread and it will process it a
2456 * few seconds later.
2458 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2461 struct btrfs_fs_info *fs_info = root->fs_info;
2463 spin_lock(&fs_info->trans_lock);
2464 if (list_empty(&fs_info->dead_roots)) {
2465 spin_unlock(&fs_info->trans_lock);
2468 root = list_first_entry(&fs_info->dead_roots,
2469 struct btrfs_root, root_list);
2470 list_del_init(&root->root_list);
2471 spin_unlock(&fs_info->trans_lock);
2473 btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2475 btrfs_kill_all_delayed_nodes(root);
2477 if (btrfs_header_backref_rev(root->node) <
2478 BTRFS_MIXED_BACKREF_REV)
2479 ret = btrfs_drop_snapshot(root, 0, 0);
2481 ret = btrfs_drop_snapshot(root, 1, 0);
2483 btrfs_put_root(root);
2484 return (ret < 0) ? 0 : 1;
2487 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2492 prev = xchg(&fs_info->pending_changes, 0);
2496 bit = 1 << BTRFS_PENDING_COMMIT;
2498 btrfs_debug(fs_info, "pending commit done");
2503 "unknown pending changes left 0x%lx, ignoring", prev);