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;
418 if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
419 root->last_trans < trans->transid) || force) {
420 WARN_ON(root == fs_info->extent_root);
421 WARN_ON(!force && root->commit_root != root->node);
424 * see below for IN_TRANS_SETUP usage rules
425 * we have the reloc mutex held now, so there
426 * is only one writer in this function
428 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
430 /* make sure readers find IN_TRANS_SETUP before
431 * they find our root->last_trans update
435 spin_lock(&fs_info->fs_roots_radix_lock);
436 if (root->last_trans == trans->transid && !force) {
437 spin_unlock(&fs_info->fs_roots_radix_lock);
440 radix_tree_tag_set(&fs_info->fs_roots_radix,
441 (unsigned long)root->root_key.objectid,
442 BTRFS_ROOT_TRANS_TAG);
443 spin_unlock(&fs_info->fs_roots_radix_lock);
444 root->last_trans = trans->transid;
446 /* this is pretty tricky. We don't want to
447 * take the relocation lock in btrfs_record_root_in_trans
448 * unless we're really doing the first setup for this root in
451 * Normally we'd use root->last_trans as a flag to decide
452 * if we want to take the expensive mutex.
454 * But, we have to set root->last_trans before we
455 * init the relocation root, otherwise, we trip over warnings
456 * in ctree.c. The solution used here is to flag ourselves
457 * with root IN_TRANS_SETUP. When this is 1, we're still
458 * fixing up the reloc trees and everyone must wait.
460 * When this is zero, they can trust root->last_trans and fly
461 * through btrfs_record_root_in_trans without having to take the
462 * lock. smp_wmb() makes sure that all the writes above are
463 * done before we pop in the zero below
465 ret = btrfs_init_reloc_root(trans, root);
466 smp_mb__before_atomic();
467 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
473 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
474 struct btrfs_root *root)
476 struct btrfs_fs_info *fs_info = root->fs_info;
477 struct btrfs_transaction *cur_trans = trans->transaction;
479 /* Add ourselves to the transaction dropped list */
480 spin_lock(&cur_trans->dropped_roots_lock);
481 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
482 spin_unlock(&cur_trans->dropped_roots_lock);
484 /* Make sure we don't try to update the root at commit time */
485 spin_lock(&fs_info->fs_roots_radix_lock);
486 radix_tree_tag_clear(&fs_info->fs_roots_radix,
487 (unsigned long)root->root_key.objectid,
488 BTRFS_ROOT_TRANS_TAG);
489 spin_unlock(&fs_info->fs_roots_radix_lock);
492 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
493 struct btrfs_root *root)
495 struct btrfs_fs_info *fs_info = root->fs_info;
498 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
502 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
506 if (root->last_trans == trans->transid &&
507 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
510 mutex_lock(&fs_info->reloc_mutex);
511 ret = record_root_in_trans(trans, root, 0);
512 mutex_unlock(&fs_info->reloc_mutex);
517 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
519 return (trans->state >= TRANS_STATE_COMMIT_START &&
520 trans->state < TRANS_STATE_UNBLOCKED &&
521 !TRANS_ABORTED(trans));
524 /* wait for commit against the current transaction to become unblocked
525 * when this is done, it is safe to start a new transaction, but the current
526 * transaction might not be fully on disk.
528 static void wait_current_trans(struct btrfs_fs_info *fs_info)
530 struct btrfs_transaction *cur_trans;
532 spin_lock(&fs_info->trans_lock);
533 cur_trans = fs_info->running_transaction;
534 if (cur_trans && is_transaction_blocked(cur_trans)) {
535 refcount_inc(&cur_trans->use_count);
536 spin_unlock(&fs_info->trans_lock);
538 wait_event(fs_info->transaction_wait,
539 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
540 TRANS_ABORTED(cur_trans));
541 btrfs_put_transaction(cur_trans);
543 spin_unlock(&fs_info->trans_lock);
547 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
549 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
552 if (type == TRANS_START)
558 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
560 struct btrfs_fs_info *fs_info = root->fs_info;
562 if (!fs_info->reloc_ctl ||
563 !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
564 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
571 static struct btrfs_trans_handle *
572 start_transaction(struct btrfs_root *root, unsigned int num_items,
573 unsigned int type, enum btrfs_reserve_flush_enum flush,
574 bool enforce_qgroups)
576 struct btrfs_fs_info *fs_info = root->fs_info;
577 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
578 struct btrfs_trans_handle *h;
579 struct btrfs_transaction *cur_trans;
581 u64 qgroup_reserved = 0;
582 bool reloc_reserved = false;
583 bool do_chunk_alloc = false;
586 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
587 return ERR_PTR(-EROFS);
589 if (current->journal_info) {
590 WARN_ON(type & TRANS_EXTWRITERS);
591 h = current->journal_info;
592 refcount_inc(&h->use_count);
593 WARN_ON(refcount_read(&h->use_count) > 2);
594 h->orig_rsv = h->block_rsv;
600 * Do the reservation before we join the transaction so we can do all
601 * the appropriate flushing if need be.
603 if (num_items && root != fs_info->chunk_root) {
604 struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
605 u64 delayed_refs_bytes = 0;
607 qgroup_reserved = num_items * fs_info->nodesize;
608 ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
614 * We want to reserve all the bytes we may need all at once, so
615 * we only do 1 enospc flushing cycle per transaction start. We
616 * accomplish this by simply assuming we'll do 2 x num_items
617 * worth of delayed refs updates in this trans handle, and
618 * refill that amount for whatever is missing in the reserve.
620 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
621 if (flush == BTRFS_RESERVE_FLUSH_ALL &&
622 delayed_refs_rsv->full == 0) {
623 delayed_refs_bytes = num_bytes;
628 * Do the reservation for the relocation root creation
630 if (need_reserve_reloc_root(root)) {
631 num_bytes += fs_info->nodesize;
632 reloc_reserved = true;
635 ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
638 if (delayed_refs_bytes) {
639 btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
641 num_bytes -= delayed_refs_bytes;
644 if (rsv->space_info->force_alloc)
645 do_chunk_alloc = true;
646 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
647 !delayed_refs_rsv->full) {
649 * Some people call with btrfs_start_transaction(root, 0)
650 * because they can be throttled, but have some other mechanism
651 * for reserving space. We still want these guys to refill the
652 * delayed block_rsv so just add 1 items worth of reservation
655 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
660 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
667 * If we are JOIN_NOLOCK we're already committing a transaction and
668 * waiting on this guy, so we don't need to do the sb_start_intwrite
669 * because we're already holding a ref. We need this because we could
670 * have raced in and did an fsync() on a file which can kick a commit
671 * and then we deadlock with somebody doing a freeze.
673 * If we are ATTACH, it means we just want to catch the current
674 * transaction and commit it, so we needn't do sb_start_intwrite().
676 if (type & __TRANS_FREEZABLE)
677 sb_start_intwrite(fs_info->sb);
679 if (may_wait_transaction(fs_info, type))
680 wait_current_trans(fs_info);
683 ret = join_transaction(fs_info, type);
685 wait_current_trans(fs_info);
686 if (unlikely(type == TRANS_ATTACH ||
687 type == TRANS_JOIN_NOSTART))
690 } while (ret == -EBUSY);
695 cur_trans = fs_info->running_transaction;
697 h->transid = cur_trans->transid;
698 h->transaction = cur_trans;
700 refcount_set(&h->use_count, 1);
701 h->fs_info = root->fs_info;
704 h->can_flush_pending_bgs = true;
705 INIT_LIST_HEAD(&h->new_bgs);
708 if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
709 may_wait_transaction(fs_info, type)) {
710 current->journal_info = h;
711 btrfs_commit_transaction(h);
716 trace_btrfs_space_reservation(fs_info, "transaction",
717 h->transid, num_bytes, 1);
718 h->block_rsv = &fs_info->trans_block_rsv;
719 h->bytes_reserved = num_bytes;
720 h->reloc_reserved = reloc_reserved;
724 if (!current->journal_info)
725 current->journal_info = h;
728 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
729 * ALLOC_FORCE the first run through, and then we won't allocate for
730 * anybody else who races in later. We don't care about the return
733 if (do_chunk_alloc && num_bytes) {
734 u64 flags = h->block_rsv->space_info->flags;
736 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
737 CHUNK_ALLOC_NO_FORCE);
741 * btrfs_record_root_in_trans() needs to alloc new extents, and may
742 * call btrfs_join_transaction() while we're also starting a
745 * Thus it need to be called after current->journal_info initialized,
746 * or we can deadlock.
748 ret = btrfs_record_root_in_trans(h, root);
751 * The transaction handle is fully initialized and linked with
752 * other structures so it needs to be ended in case of errors,
755 btrfs_end_transaction(h);
762 if (type & __TRANS_FREEZABLE)
763 sb_end_intwrite(fs_info->sb);
764 kmem_cache_free(btrfs_trans_handle_cachep, h);
767 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
770 btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
774 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
775 unsigned int num_items)
777 return start_transaction(root, num_items, TRANS_START,
778 BTRFS_RESERVE_FLUSH_ALL, true);
781 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
782 struct btrfs_root *root,
783 unsigned int num_items)
785 return start_transaction(root, num_items, TRANS_START,
786 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
789 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
791 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
795 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
797 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
798 BTRFS_RESERVE_NO_FLUSH, true);
802 * Similar to regular join but it never starts a transaction when none is
803 * running or after waiting for the current one to finish.
805 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
807 return start_transaction(root, 0, TRANS_JOIN_NOSTART,
808 BTRFS_RESERVE_NO_FLUSH, true);
812 * btrfs_attach_transaction() - catch the running transaction
814 * It is used when we want to commit the current the transaction, but
815 * don't want to start a new one.
817 * Note: If this function return -ENOENT, it just means there is no
818 * running transaction. But it is possible that the inactive transaction
819 * is still in the memory, not fully on disk. If you hope there is no
820 * inactive transaction in the fs when -ENOENT is returned, you should
822 * btrfs_attach_transaction_barrier()
824 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
826 return start_transaction(root, 0, TRANS_ATTACH,
827 BTRFS_RESERVE_NO_FLUSH, true);
831 * btrfs_attach_transaction_barrier() - catch the running transaction
833 * It is similar to the above function, the difference is this one
834 * will wait for all the inactive transactions until they fully
837 struct btrfs_trans_handle *
838 btrfs_attach_transaction_barrier(struct btrfs_root *root)
840 struct btrfs_trans_handle *trans;
842 trans = start_transaction(root, 0, TRANS_ATTACH,
843 BTRFS_RESERVE_NO_FLUSH, true);
844 if (trans == ERR_PTR(-ENOENT))
845 btrfs_wait_for_commit(root->fs_info, 0);
850 /* Wait for a transaction commit to reach at least the given state. */
851 static noinline void wait_for_commit(struct btrfs_transaction *commit,
852 const enum btrfs_trans_state min_state)
854 wait_event(commit->commit_wait, commit->state >= min_state);
857 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
859 struct btrfs_transaction *cur_trans = NULL, *t;
863 if (transid <= fs_info->last_trans_committed)
866 /* find specified transaction */
867 spin_lock(&fs_info->trans_lock);
868 list_for_each_entry(t, &fs_info->trans_list, list) {
869 if (t->transid == transid) {
871 refcount_inc(&cur_trans->use_count);
875 if (t->transid > transid) {
880 spin_unlock(&fs_info->trans_lock);
883 * The specified transaction doesn't exist, or we
884 * raced with btrfs_commit_transaction
887 if (transid > fs_info->last_trans_committed)
892 /* find newest transaction that is committing | committed */
893 spin_lock(&fs_info->trans_lock);
894 list_for_each_entry_reverse(t, &fs_info->trans_list,
896 if (t->state >= TRANS_STATE_COMMIT_START) {
897 if (t->state == TRANS_STATE_COMPLETED)
900 refcount_inc(&cur_trans->use_count);
904 spin_unlock(&fs_info->trans_lock);
906 goto out; /* nothing committing|committed */
909 wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
910 btrfs_put_transaction(cur_trans);
915 void btrfs_throttle(struct btrfs_fs_info *fs_info)
917 wait_current_trans(fs_info);
920 static bool should_end_transaction(struct btrfs_trans_handle *trans)
922 struct btrfs_fs_info *fs_info = trans->fs_info;
924 if (btrfs_check_space_for_delayed_refs(fs_info))
927 return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
930 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
932 struct btrfs_transaction *cur_trans = trans->transaction;
934 if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
935 test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
938 return should_end_transaction(trans);
941 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
944 struct btrfs_fs_info *fs_info = trans->fs_info;
946 if (!trans->block_rsv) {
947 ASSERT(!trans->bytes_reserved);
951 if (!trans->bytes_reserved)
954 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
955 trace_btrfs_space_reservation(fs_info, "transaction",
956 trans->transid, trans->bytes_reserved, 0);
957 btrfs_block_rsv_release(fs_info, trans->block_rsv,
958 trans->bytes_reserved, NULL);
959 trans->bytes_reserved = 0;
962 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
965 struct btrfs_fs_info *info = trans->fs_info;
966 struct btrfs_transaction *cur_trans = trans->transaction;
969 if (refcount_read(&trans->use_count) > 1) {
970 refcount_dec(&trans->use_count);
971 trans->block_rsv = trans->orig_rsv;
975 btrfs_trans_release_metadata(trans);
976 trans->block_rsv = NULL;
978 btrfs_create_pending_block_groups(trans);
980 btrfs_trans_release_chunk_metadata(trans);
982 if (trans->type & __TRANS_FREEZABLE)
983 sb_end_intwrite(info->sb);
985 WARN_ON(cur_trans != info->running_transaction);
986 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
987 atomic_dec(&cur_trans->num_writers);
988 extwriter_counter_dec(cur_trans, trans->type);
990 cond_wake_up(&cur_trans->writer_wait);
991 btrfs_put_transaction(cur_trans);
993 if (current->journal_info == trans)
994 current->journal_info = NULL;
997 btrfs_run_delayed_iputs(info);
999 if (TRANS_ABORTED(trans) ||
1000 test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
1001 wake_up_process(info->transaction_kthread);
1002 if (TRANS_ABORTED(trans))
1003 err = trans->aborted;
1008 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1012 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1014 return __btrfs_end_transaction(trans, 0);
1017 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1019 return __btrfs_end_transaction(trans, 1);
1023 * when btree blocks are allocated, they have some corresponding bits set for
1024 * them in one of two extent_io trees. This is used to make sure all of
1025 * those extents are sent to disk but does not wait on them
1027 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1028 struct extent_io_tree *dirty_pages, int mark)
1032 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1033 struct extent_state *cached_state = NULL;
1037 atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1038 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1039 mark, &cached_state)) {
1040 bool wait_writeback = false;
1042 err = convert_extent_bit(dirty_pages, start, end,
1044 mark, &cached_state);
1046 * convert_extent_bit can return -ENOMEM, which is most of the
1047 * time a temporary error. So when it happens, ignore the error
1048 * and wait for writeback of this range to finish - because we
1049 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1050 * to __btrfs_wait_marked_extents() would not know that
1051 * writeback for this range started and therefore wouldn't
1052 * wait for it to finish - we don't want to commit a
1053 * superblock that points to btree nodes/leafs for which
1054 * writeback hasn't finished yet (and without errors).
1055 * We cleanup any entries left in the io tree when committing
1056 * the transaction (through extent_io_tree_release()).
1058 if (err == -ENOMEM) {
1060 wait_writeback = true;
1063 err = filemap_fdatawrite_range(mapping, start, end);
1066 else if (wait_writeback)
1067 werr = filemap_fdatawait_range(mapping, start, end);
1068 free_extent_state(cached_state);
1069 cached_state = NULL;
1073 atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1078 * when btree blocks are allocated, they have some corresponding bits set for
1079 * them in one of two extent_io trees. This is used to make sure all of
1080 * those extents are on disk for transaction or log commit. We wait
1081 * on all the pages and clear them from the dirty pages state tree
1083 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1084 struct extent_io_tree *dirty_pages)
1088 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1089 struct extent_state *cached_state = NULL;
1093 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1094 EXTENT_NEED_WAIT, &cached_state)) {
1096 * Ignore -ENOMEM errors returned by clear_extent_bit().
1097 * When committing the transaction, we'll remove any entries
1098 * left in the io tree. For a log commit, we don't remove them
1099 * after committing the log because the tree can be accessed
1100 * concurrently - we do it only at transaction commit time when
1101 * it's safe to do it (through extent_io_tree_release()).
1103 err = clear_extent_bit(dirty_pages, start, end,
1104 EXTENT_NEED_WAIT, 0, 0, &cached_state);
1108 err = filemap_fdatawait_range(mapping, start, end);
1111 free_extent_state(cached_state);
1112 cached_state = NULL;
1121 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1122 struct extent_io_tree *dirty_pages)
1124 bool errors = false;
1127 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1128 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1136 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1138 struct btrfs_fs_info *fs_info = log_root->fs_info;
1139 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1140 bool errors = false;
1143 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1145 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1146 if ((mark & EXTENT_DIRTY) &&
1147 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1150 if ((mark & EXTENT_NEW) &&
1151 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1160 * When btree blocks are allocated the corresponding extents are marked dirty.
1161 * This function ensures such extents are persisted on disk for transaction or
1164 * @trans: transaction whose dirty pages we'd like to write
1166 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1170 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1171 struct btrfs_fs_info *fs_info = trans->fs_info;
1172 struct blk_plug plug;
1174 blk_start_plug(&plug);
1175 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1176 blk_finish_plug(&plug);
1177 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1179 extent_io_tree_release(&trans->transaction->dirty_pages);
1190 * this is used to update the root pointer in the tree of tree roots.
1192 * But, in the case of the extent allocation tree, updating the root
1193 * pointer may allocate blocks which may change the root of the extent
1196 * So, this loops and repeats and makes sure the cowonly root didn't
1197 * change while the root pointer was being updated in the metadata.
1199 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1200 struct btrfs_root *root)
1203 u64 old_root_bytenr;
1205 struct btrfs_fs_info *fs_info = root->fs_info;
1206 struct btrfs_root *tree_root = fs_info->tree_root;
1208 old_root_used = btrfs_root_used(&root->root_item);
1211 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1212 if (old_root_bytenr == root->node->start &&
1213 old_root_used == btrfs_root_used(&root->root_item))
1216 btrfs_set_root_node(&root->root_item, root->node);
1217 ret = btrfs_update_root(trans, tree_root,
1223 old_root_used = btrfs_root_used(&root->root_item);
1230 * update all the cowonly tree roots on disk
1232 * The error handling in this function may not be obvious. Any of the
1233 * failures will cause the file system to go offline. We still need
1234 * to clean up the delayed refs.
1236 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1238 struct btrfs_fs_info *fs_info = trans->fs_info;
1239 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1240 struct list_head *io_bgs = &trans->transaction->io_bgs;
1241 struct list_head *next;
1242 struct extent_buffer *eb;
1245 eb = btrfs_lock_root_node(fs_info->tree_root);
1246 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1247 0, &eb, BTRFS_NESTING_COW);
1248 btrfs_tree_unlock(eb);
1249 free_extent_buffer(eb);
1254 ret = btrfs_run_dev_stats(trans);
1257 ret = btrfs_run_dev_replace(trans);
1260 ret = btrfs_run_qgroups(trans);
1264 ret = btrfs_setup_space_cache(trans);
1269 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1270 struct btrfs_root *root;
1271 next = fs_info->dirty_cowonly_roots.next;
1272 list_del_init(next);
1273 root = list_entry(next, struct btrfs_root, dirty_list);
1274 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1276 if (root != fs_info->extent_root)
1277 list_add_tail(&root->dirty_list,
1278 &trans->transaction->switch_commits);
1279 ret = update_cowonly_root(trans, root);
1284 /* Now flush any delayed refs generated by updating all of the roots */
1285 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1289 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1290 ret = btrfs_write_dirty_block_groups(trans);
1295 * We're writing the dirty block groups, which could generate
1296 * delayed refs, which could generate more dirty block groups,
1297 * so we want to keep this flushing in this loop to make sure
1298 * everything gets run.
1300 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1305 if (!list_empty(&fs_info->dirty_cowonly_roots))
1308 list_add_tail(&fs_info->extent_root->dirty_list,
1309 &trans->transaction->switch_commits);
1311 /* Update dev-replace pointer once everything is committed */
1312 fs_info->dev_replace.committed_cursor_left =
1313 fs_info->dev_replace.cursor_left_last_write_of_item;
1319 * dead roots are old snapshots that need to be deleted. This allocates
1320 * a dirty root struct and adds it into the list of dead roots that need to
1323 void btrfs_add_dead_root(struct btrfs_root *root)
1325 struct btrfs_fs_info *fs_info = root->fs_info;
1327 spin_lock(&fs_info->trans_lock);
1328 if (list_empty(&root->root_list)) {
1329 btrfs_grab_root(root);
1330 list_add_tail(&root->root_list, &fs_info->dead_roots);
1332 spin_unlock(&fs_info->trans_lock);
1336 * update all the cowonly tree roots on disk
1338 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1340 struct btrfs_fs_info *fs_info = trans->fs_info;
1341 struct btrfs_root *gang[8];
1345 spin_lock(&fs_info->fs_roots_radix_lock);
1347 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1350 BTRFS_ROOT_TRANS_TAG);
1353 for (i = 0; i < ret; i++) {
1354 struct btrfs_root *root = gang[i];
1357 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1358 (unsigned long)root->root_key.objectid,
1359 BTRFS_ROOT_TRANS_TAG);
1360 spin_unlock(&fs_info->fs_roots_radix_lock);
1362 btrfs_free_log(trans, root);
1363 ret2 = btrfs_update_reloc_root(trans, root);
1367 /* see comments in should_cow_block() */
1368 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1369 smp_mb__after_atomic();
1371 if (root->commit_root != root->node) {
1372 list_add_tail(&root->dirty_list,
1373 &trans->transaction->switch_commits);
1374 btrfs_set_root_node(&root->root_item,
1378 ret2 = btrfs_update_root(trans, fs_info->tree_root,
1383 spin_lock(&fs_info->fs_roots_radix_lock);
1384 btrfs_qgroup_free_meta_all_pertrans(root);
1387 spin_unlock(&fs_info->fs_roots_radix_lock);
1392 * defrag a given btree.
1393 * Every leaf in the btree is read and defragged.
1395 int btrfs_defrag_root(struct btrfs_root *root)
1397 struct btrfs_fs_info *info = root->fs_info;
1398 struct btrfs_trans_handle *trans;
1401 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1405 trans = btrfs_start_transaction(root, 0);
1406 if (IS_ERR(trans)) {
1407 ret = PTR_ERR(trans);
1411 ret = btrfs_defrag_leaves(trans, root);
1413 btrfs_end_transaction(trans);
1414 btrfs_btree_balance_dirty(info);
1417 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1420 if (btrfs_defrag_cancelled(info)) {
1421 btrfs_debug(info, "defrag_root cancelled");
1426 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1431 * Do all special snapshot related qgroup dirty hack.
1433 * Will do all needed qgroup inherit and dirty hack like switch commit
1434 * roots inside one transaction and write all btree into disk, to make
1437 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1438 struct btrfs_root *src,
1439 struct btrfs_root *parent,
1440 struct btrfs_qgroup_inherit *inherit,
1443 struct btrfs_fs_info *fs_info = src->fs_info;
1447 * Save some performance in the case that qgroups are not
1448 * enabled. If this check races with the ioctl, rescan will
1451 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1455 * Ensure dirty @src will be committed. Or, after coming
1456 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1457 * recorded root will never be updated again, causing an outdated root
1460 ret = record_root_in_trans(trans, src, 1);
1465 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1466 * src root, so we must run the delayed refs here.
1468 * However this isn't particularly fool proof, because there's no
1469 * synchronization keeping us from changing the tree after this point
1470 * before we do the qgroup_inherit, or even from making changes while
1471 * we're doing the qgroup_inherit. But that's a problem for the future,
1472 * for now flush the delayed refs to narrow the race window where the
1473 * qgroup counters could end up wrong.
1475 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1477 btrfs_abort_transaction(trans, ret);
1482 * We are going to commit transaction, see btrfs_commit_transaction()
1483 * comment for reason locking tree_log_mutex
1485 mutex_lock(&fs_info->tree_log_mutex);
1487 ret = commit_fs_roots(trans);
1490 ret = btrfs_qgroup_account_extents(trans);
1494 /* Now qgroup are all updated, we can inherit it to new qgroups */
1495 ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1501 * Now we do a simplified commit transaction, which will:
1502 * 1) commit all subvolume and extent tree
1503 * To ensure all subvolume and extent tree have a valid
1504 * commit_root to accounting later insert_dir_item()
1505 * 2) write all btree blocks onto disk
1506 * This is to make sure later btree modification will be cowed
1507 * Or commit_root can be populated and cause wrong qgroup numbers
1508 * In this simplified commit, we don't really care about other trees
1509 * like chunk and root tree, as they won't affect qgroup.
1510 * And we don't write super to avoid half committed status.
1512 ret = commit_cowonly_roots(trans);
1515 switch_commit_roots(trans);
1516 ret = btrfs_write_and_wait_transaction(trans);
1518 btrfs_handle_fs_error(fs_info, ret,
1519 "Error while writing out transaction for qgroup");
1522 mutex_unlock(&fs_info->tree_log_mutex);
1525 * Force parent root to be updated, as we recorded it before so its
1526 * last_trans == cur_transid.
1527 * Or it won't be committed again onto disk after later
1531 ret = record_root_in_trans(trans, parent, 1);
1536 * new snapshots need to be created at a very specific time in the
1537 * transaction commit. This does the actual creation.
1540 * If the error which may affect the commitment of the current transaction
1541 * happens, we should return the error number. If the error which just affect
1542 * the creation of the pending snapshots, just return 0.
1544 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1545 struct btrfs_pending_snapshot *pending)
1548 struct btrfs_fs_info *fs_info = trans->fs_info;
1549 struct btrfs_key key;
1550 struct btrfs_root_item *new_root_item;
1551 struct btrfs_root *tree_root = fs_info->tree_root;
1552 struct btrfs_root *root = pending->root;
1553 struct btrfs_root *parent_root;
1554 struct btrfs_block_rsv *rsv;
1555 struct inode *parent_inode;
1556 struct btrfs_path *path;
1557 struct btrfs_dir_item *dir_item;
1558 struct dentry *dentry;
1559 struct extent_buffer *tmp;
1560 struct extent_buffer *old;
1561 struct timespec64 cur_time;
1568 ASSERT(pending->path);
1569 path = pending->path;
1571 ASSERT(pending->root_item);
1572 new_root_item = pending->root_item;
1574 pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1576 goto no_free_objectid;
1579 * Make qgroup to skip current new snapshot's qgroupid, as it is
1580 * accounted by later btrfs_qgroup_inherit().
1582 btrfs_set_skip_qgroup(trans, objectid);
1584 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1586 if (to_reserve > 0) {
1587 pending->error = btrfs_block_rsv_add(root,
1588 &pending->block_rsv,
1590 BTRFS_RESERVE_NO_FLUSH);
1592 goto clear_skip_qgroup;
1595 key.objectid = objectid;
1596 key.offset = (u64)-1;
1597 key.type = BTRFS_ROOT_ITEM_KEY;
1599 rsv = trans->block_rsv;
1600 trans->block_rsv = &pending->block_rsv;
1601 trans->bytes_reserved = trans->block_rsv->reserved;
1602 trace_btrfs_space_reservation(fs_info, "transaction",
1604 trans->bytes_reserved, 1);
1605 dentry = pending->dentry;
1606 parent_inode = pending->dir;
1607 parent_root = BTRFS_I(parent_inode)->root;
1608 ret = record_root_in_trans(trans, parent_root, 0);
1611 cur_time = current_time(parent_inode);
1614 * insert the directory item
1616 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1617 BUG_ON(ret); /* -ENOMEM */
1619 /* check if there is a file/dir which has the same name. */
1620 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1621 btrfs_ino(BTRFS_I(parent_inode)),
1622 dentry->d_name.name,
1623 dentry->d_name.len, 0);
1624 if (dir_item != NULL && !IS_ERR(dir_item)) {
1625 pending->error = -EEXIST;
1626 goto dir_item_existed;
1627 } else if (IS_ERR(dir_item)) {
1628 ret = PTR_ERR(dir_item);
1629 btrfs_abort_transaction(trans, ret);
1632 btrfs_release_path(path);
1635 * pull in the delayed directory update
1636 * and the delayed inode item
1637 * otherwise we corrupt the FS during
1640 ret = btrfs_run_delayed_items(trans);
1641 if (ret) { /* Transaction aborted */
1642 btrfs_abort_transaction(trans, ret);
1646 ret = record_root_in_trans(trans, root, 0);
1648 btrfs_abort_transaction(trans, ret);
1651 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1652 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1653 btrfs_check_and_init_root_item(new_root_item);
1655 root_flags = btrfs_root_flags(new_root_item);
1656 if (pending->readonly)
1657 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1659 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1660 btrfs_set_root_flags(new_root_item, root_flags);
1662 btrfs_set_root_generation_v2(new_root_item,
1664 generate_random_guid(new_root_item->uuid);
1665 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1667 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1668 memset(new_root_item->received_uuid, 0,
1669 sizeof(new_root_item->received_uuid));
1670 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1671 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1672 btrfs_set_root_stransid(new_root_item, 0);
1673 btrfs_set_root_rtransid(new_root_item, 0);
1675 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1676 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1677 btrfs_set_root_otransid(new_root_item, trans->transid);
1679 old = btrfs_lock_root_node(root);
1680 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1683 btrfs_tree_unlock(old);
1684 free_extent_buffer(old);
1685 btrfs_abort_transaction(trans, ret);
1689 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1690 /* clean up in any case */
1691 btrfs_tree_unlock(old);
1692 free_extent_buffer(old);
1694 btrfs_abort_transaction(trans, ret);
1697 /* see comments in should_cow_block() */
1698 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1701 btrfs_set_root_node(new_root_item, tmp);
1702 /* record when the snapshot was created in key.offset */
1703 key.offset = trans->transid;
1704 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1705 btrfs_tree_unlock(tmp);
1706 free_extent_buffer(tmp);
1708 btrfs_abort_transaction(trans, ret);
1713 * insert root back/forward references
1715 ret = btrfs_add_root_ref(trans, objectid,
1716 parent_root->root_key.objectid,
1717 btrfs_ino(BTRFS_I(parent_inode)), index,
1718 dentry->d_name.name, dentry->d_name.len);
1720 btrfs_abort_transaction(trans, ret);
1724 key.offset = (u64)-1;
1725 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1726 if (IS_ERR(pending->snap)) {
1727 ret = PTR_ERR(pending->snap);
1728 pending->snap = NULL;
1729 btrfs_abort_transaction(trans, ret);
1733 ret = btrfs_reloc_post_snapshot(trans, pending);
1735 btrfs_abort_transaction(trans, ret);
1740 * Do special qgroup accounting for snapshot, as we do some qgroup
1741 * snapshot hack to do fast snapshot.
1742 * To co-operate with that hack, we do hack again.
1743 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1745 ret = qgroup_account_snapshot(trans, root, parent_root,
1746 pending->inherit, objectid);
1750 ret = btrfs_insert_dir_item(trans, dentry->d_name.name,
1751 dentry->d_name.len, BTRFS_I(parent_inode),
1752 &key, BTRFS_FT_DIR, index);
1753 /* We have check then name at the beginning, so it is impossible. */
1754 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1756 btrfs_abort_transaction(trans, ret);
1760 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1761 dentry->d_name.len * 2);
1762 parent_inode->i_mtime = parent_inode->i_ctime =
1763 current_time(parent_inode);
1764 ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1766 btrfs_abort_transaction(trans, ret);
1769 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1770 BTRFS_UUID_KEY_SUBVOL,
1773 btrfs_abort_transaction(trans, ret);
1776 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1777 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1778 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1780 if (ret && ret != -EEXIST) {
1781 btrfs_abort_transaction(trans, ret);
1787 pending->error = ret;
1789 trans->block_rsv = rsv;
1790 trans->bytes_reserved = 0;
1792 btrfs_clear_skip_qgroup(trans);
1794 kfree(new_root_item);
1795 pending->root_item = NULL;
1796 btrfs_free_path(path);
1797 pending->path = NULL;
1803 * create all the snapshots we've scheduled for creation
1805 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1807 struct btrfs_pending_snapshot *pending, *next;
1808 struct list_head *head = &trans->transaction->pending_snapshots;
1811 list_for_each_entry_safe(pending, next, head, list) {
1812 list_del(&pending->list);
1813 ret = create_pending_snapshot(trans, pending);
1820 static void update_super_roots(struct btrfs_fs_info *fs_info)
1822 struct btrfs_root_item *root_item;
1823 struct btrfs_super_block *super;
1825 super = fs_info->super_copy;
1827 root_item = &fs_info->chunk_root->root_item;
1828 super->chunk_root = root_item->bytenr;
1829 super->chunk_root_generation = root_item->generation;
1830 super->chunk_root_level = root_item->level;
1832 root_item = &fs_info->tree_root->root_item;
1833 super->root = root_item->bytenr;
1834 super->generation = root_item->generation;
1835 super->root_level = root_item->level;
1836 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1837 super->cache_generation = root_item->generation;
1838 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1839 super->cache_generation = 0;
1840 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1841 super->uuid_tree_generation = root_item->generation;
1844 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1846 struct btrfs_transaction *trans;
1849 spin_lock(&info->trans_lock);
1850 trans = info->running_transaction;
1852 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1853 spin_unlock(&info->trans_lock);
1857 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1859 struct btrfs_transaction *trans;
1862 spin_lock(&info->trans_lock);
1863 trans = info->running_transaction;
1865 ret = is_transaction_blocked(trans);
1866 spin_unlock(&info->trans_lock);
1871 * commit transactions asynchronously. once btrfs_commit_transaction_async
1872 * returns, any subsequent transaction will not be allowed to join.
1874 struct btrfs_async_commit {
1875 struct btrfs_trans_handle *newtrans;
1876 struct work_struct work;
1879 static void do_async_commit(struct work_struct *work)
1881 struct btrfs_async_commit *ac =
1882 container_of(work, struct btrfs_async_commit, work);
1885 * We've got freeze protection passed with the transaction.
1886 * Tell lockdep about it.
1888 if (ac->newtrans->type & __TRANS_FREEZABLE)
1889 __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1891 current->journal_info = ac->newtrans;
1893 btrfs_commit_transaction(ac->newtrans);
1897 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1899 struct btrfs_fs_info *fs_info = trans->fs_info;
1900 struct btrfs_async_commit *ac;
1901 struct btrfs_transaction *cur_trans;
1903 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1907 INIT_WORK(&ac->work, do_async_commit);
1908 ac->newtrans = btrfs_join_transaction(trans->root);
1909 if (IS_ERR(ac->newtrans)) {
1910 int err = PTR_ERR(ac->newtrans);
1915 /* take transaction reference */
1916 cur_trans = trans->transaction;
1917 refcount_inc(&cur_trans->use_count);
1919 btrfs_end_transaction(trans);
1922 * Tell lockdep we've released the freeze rwsem, since the
1923 * async commit thread will be the one to unlock it.
1925 if (ac->newtrans->type & __TRANS_FREEZABLE)
1926 __sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1928 schedule_work(&ac->work);
1930 * Wait for the current transaction commit to start and block
1931 * subsequent transaction joins
1933 wait_event(fs_info->transaction_blocked_wait,
1934 cur_trans->state >= TRANS_STATE_COMMIT_START ||
1935 TRANS_ABORTED(cur_trans));
1936 if (current->journal_info == trans)
1937 current->journal_info = NULL;
1939 btrfs_put_transaction(cur_trans);
1944 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1946 struct btrfs_fs_info *fs_info = trans->fs_info;
1947 struct btrfs_transaction *cur_trans = trans->transaction;
1949 WARN_ON(refcount_read(&trans->use_count) > 1);
1951 btrfs_abort_transaction(trans, err);
1953 spin_lock(&fs_info->trans_lock);
1956 * If the transaction is removed from the list, it means this
1957 * transaction has been committed successfully, so it is impossible
1958 * to call the cleanup function.
1960 BUG_ON(list_empty(&cur_trans->list));
1962 if (cur_trans == fs_info->running_transaction) {
1963 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1964 spin_unlock(&fs_info->trans_lock);
1965 wait_event(cur_trans->writer_wait,
1966 atomic_read(&cur_trans->num_writers) == 1);
1968 spin_lock(&fs_info->trans_lock);
1972 * Now that we know no one else is still using the transaction we can
1973 * remove the transaction from the list of transactions. This avoids
1974 * the transaction kthread from cleaning up the transaction while some
1975 * other task is still using it, which could result in a use-after-free
1976 * on things like log trees, as it forces the transaction kthread to
1977 * wait for this transaction to be cleaned up by us.
1979 list_del_init(&cur_trans->list);
1981 spin_unlock(&fs_info->trans_lock);
1983 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
1985 spin_lock(&fs_info->trans_lock);
1986 if (cur_trans == fs_info->running_transaction)
1987 fs_info->running_transaction = NULL;
1988 spin_unlock(&fs_info->trans_lock);
1990 if (trans->type & __TRANS_FREEZABLE)
1991 sb_end_intwrite(fs_info->sb);
1992 btrfs_put_transaction(cur_trans);
1993 btrfs_put_transaction(cur_trans);
1995 trace_btrfs_transaction_commit(trans->root);
1997 if (current->journal_info == trans)
1998 current->journal_info = NULL;
1999 btrfs_scrub_cancel(fs_info);
2001 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2005 * Release reserved delayed ref space of all pending block groups of the
2006 * transaction and remove them from the list
2008 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2010 struct btrfs_fs_info *fs_info = trans->fs_info;
2011 struct btrfs_block_group *block_group, *tmp;
2013 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2014 btrfs_delayed_refs_rsv_release(fs_info, 1);
2015 list_del_init(&block_group->bg_list);
2019 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2022 * We use writeback_inodes_sb here because if we used
2023 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2024 * Currently are holding the fs freeze lock, if we do an async flush
2025 * we'll do btrfs_join_transaction() and deadlock because we need to
2026 * wait for the fs freeze lock. Using the direct flushing we benefit
2027 * from already being in a transaction and our join_transaction doesn't
2028 * have to re-take the fs freeze lock.
2030 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2031 writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2035 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2037 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2038 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2041 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2043 struct btrfs_fs_info *fs_info = trans->fs_info;
2044 struct btrfs_transaction *cur_trans = trans->transaction;
2045 struct btrfs_transaction *prev_trans = NULL;
2048 ASSERT(refcount_read(&trans->use_count) == 1);
2050 /* Stop the commit early if ->aborted is set */
2051 if (TRANS_ABORTED(cur_trans)) {
2052 ret = cur_trans->aborted;
2053 btrfs_end_transaction(trans);
2057 btrfs_trans_release_metadata(trans);
2058 trans->block_rsv = NULL;
2061 * We only want one transaction commit doing the flushing so we do not
2062 * waste a bunch of time on lock contention on the extent root node.
2064 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2065 &cur_trans->delayed_refs.flags)) {
2067 * Make a pass through all the delayed refs we have so far.
2068 * Any running threads may add more while we are here.
2070 ret = btrfs_run_delayed_refs(trans, 0);
2072 btrfs_end_transaction(trans);
2077 btrfs_create_pending_block_groups(trans);
2079 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2082 /* this mutex is also taken before trying to set
2083 * block groups readonly. We need to make sure
2084 * that nobody has set a block group readonly
2085 * after a extents from that block group have been
2086 * allocated for cache files. btrfs_set_block_group_ro
2087 * will wait for the transaction to commit if it
2088 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2090 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2091 * only one process starts all the block group IO. It wouldn't
2092 * hurt to have more than one go through, but there's no
2093 * real advantage to it either.
2095 mutex_lock(&fs_info->ro_block_group_mutex);
2096 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2099 mutex_unlock(&fs_info->ro_block_group_mutex);
2102 ret = btrfs_start_dirty_block_groups(trans);
2104 btrfs_end_transaction(trans);
2110 spin_lock(&fs_info->trans_lock);
2111 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2112 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2114 spin_unlock(&fs_info->trans_lock);
2115 refcount_inc(&cur_trans->use_count);
2117 if (trans->in_fsync)
2118 want_state = TRANS_STATE_SUPER_COMMITTED;
2119 ret = btrfs_end_transaction(trans);
2120 wait_for_commit(cur_trans, want_state);
2122 if (TRANS_ABORTED(cur_trans))
2123 ret = cur_trans->aborted;
2125 btrfs_put_transaction(cur_trans);
2130 cur_trans->state = TRANS_STATE_COMMIT_START;
2131 wake_up(&fs_info->transaction_blocked_wait);
2133 if (cur_trans->list.prev != &fs_info->trans_list) {
2134 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2136 if (trans->in_fsync)
2137 want_state = TRANS_STATE_SUPER_COMMITTED;
2139 prev_trans = list_entry(cur_trans->list.prev,
2140 struct btrfs_transaction, list);
2141 if (prev_trans->state < want_state) {
2142 refcount_inc(&prev_trans->use_count);
2143 spin_unlock(&fs_info->trans_lock);
2145 wait_for_commit(prev_trans, want_state);
2147 ret = READ_ONCE(prev_trans->aborted);
2149 btrfs_put_transaction(prev_trans);
2151 goto cleanup_transaction;
2153 spin_unlock(&fs_info->trans_lock);
2156 spin_unlock(&fs_info->trans_lock);
2158 * The previous transaction was aborted and was already removed
2159 * from the list of transactions at fs_info->trans_list. So we
2160 * abort to prevent writing a new superblock that reflects a
2161 * corrupt state (pointing to trees with unwritten nodes/leafs).
2163 if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) {
2165 goto cleanup_transaction;
2169 extwriter_counter_dec(cur_trans, trans->type);
2171 ret = btrfs_start_delalloc_flush(fs_info);
2173 goto cleanup_transaction;
2175 ret = btrfs_run_delayed_items(trans);
2177 goto cleanup_transaction;
2179 wait_event(cur_trans->writer_wait,
2180 extwriter_counter_read(cur_trans) == 0);
2182 /* some pending stuffs might be added after the previous flush. */
2183 ret = btrfs_run_delayed_items(trans);
2185 goto cleanup_transaction;
2187 btrfs_wait_delalloc_flush(fs_info);
2190 * Wait for all ordered extents started by a fast fsync that joined this
2191 * transaction. Otherwise if this transaction commits before the ordered
2192 * extents complete we lose logged data after a power failure.
2194 wait_event(cur_trans->pending_wait,
2195 atomic_read(&cur_trans->pending_ordered) == 0);
2197 btrfs_scrub_pause(fs_info);
2199 * Ok now we need to make sure to block out any other joins while we
2200 * commit the transaction. We could have started a join before setting
2201 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2203 spin_lock(&fs_info->trans_lock);
2204 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2205 spin_unlock(&fs_info->trans_lock);
2206 wait_event(cur_trans->writer_wait,
2207 atomic_read(&cur_trans->num_writers) == 1);
2209 if (TRANS_ABORTED(cur_trans)) {
2210 ret = cur_trans->aborted;
2211 goto scrub_continue;
2214 * the reloc mutex makes sure that we stop
2215 * the balancing code from coming in and moving
2216 * extents around in the middle of the commit
2218 mutex_lock(&fs_info->reloc_mutex);
2221 * We needn't worry about the delayed items because we will
2222 * deal with them in create_pending_snapshot(), which is the
2223 * core function of the snapshot creation.
2225 ret = create_pending_snapshots(trans);
2230 * We insert the dir indexes of the snapshots and update the inode
2231 * of the snapshots' parents after the snapshot creation, so there
2232 * are some delayed items which are not dealt with. Now deal with
2235 * We needn't worry that this operation will corrupt the snapshots,
2236 * because all the tree which are snapshoted will be forced to COW
2237 * the nodes and leaves.
2239 ret = btrfs_run_delayed_items(trans);
2243 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2248 * make sure none of the code above managed to slip in a
2251 btrfs_assert_delayed_root_empty(fs_info);
2253 WARN_ON(cur_trans != trans->transaction);
2255 /* btrfs_commit_tree_roots is responsible for getting the
2256 * various roots consistent with each other. Every pointer
2257 * in the tree of tree roots has to point to the most up to date
2258 * root for every subvolume and other tree. So, we have to keep
2259 * the tree logging code from jumping in and changing any
2262 * At this point in the commit, there can't be any tree-log
2263 * writers, but a little lower down we drop the trans mutex
2264 * and let new people in. By holding the tree_log_mutex
2265 * from now until after the super is written, we avoid races
2266 * with the tree-log code.
2268 mutex_lock(&fs_info->tree_log_mutex);
2270 ret = commit_fs_roots(trans);
2272 goto unlock_tree_log;
2275 * Since the transaction is done, we can apply the pending changes
2276 * before the next transaction.
2278 btrfs_apply_pending_changes(fs_info);
2280 /* commit_fs_roots gets rid of all the tree log roots, it is now
2281 * safe to free the root of tree log roots
2283 btrfs_free_log_root_tree(trans, fs_info);
2286 * Since fs roots are all committed, we can get a quite accurate
2287 * new_roots. So let's do quota accounting.
2289 ret = btrfs_qgroup_account_extents(trans);
2291 goto unlock_tree_log;
2293 ret = commit_cowonly_roots(trans);
2295 goto unlock_tree_log;
2298 * The tasks which save the space cache and inode cache may also
2299 * update ->aborted, check it.
2301 if (TRANS_ABORTED(cur_trans)) {
2302 ret = cur_trans->aborted;
2303 goto unlock_tree_log;
2306 cur_trans = fs_info->running_transaction;
2308 btrfs_set_root_node(&fs_info->tree_root->root_item,
2309 fs_info->tree_root->node);
2310 list_add_tail(&fs_info->tree_root->dirty_list,
2311 &cur_trans->switch_commits);
2313 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2314 fs_info->chunk_root->node);
2315 list_add_tail(&fs_info->chunk_root->dirty_list,
2316 &cur_trans->switch_commits);
2318 switch_commit_roots(trans);
2320 ASSERT(list_empty(&cur_trans->dirty_bgs));
2321 ASSERT(list_empty(&cur_trans->io_bgs));
2322 update_super_roots(fs_info);
2324 btrfs_set_super_log_root(fs_info->super_copy, 0);
2325 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2326 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2327 sizeof(*fs_info->super_copy));
2329 btrfs_commit_device_sizes(cur_trans);
2331 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2332 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2334 btrfs_trans_release_chunk_metadata(trans);
2336 spin_lock(&fs_info->trans_lock);
2337 cur_trans->state = TRANS_STATE_UNBLOCKED;
2338 fs_info->running_transaction = NULL;
2339 spin_unlock(&fs_info->trans_lock);
2340 mutex_unlock(&fs_info->reloc_mutex);
2342 wake_up(&fs_info->transaction_wait);
2344 ret = btrfs_write_and_wait_transaction(trans);
2346 btrfs_handle_fs_error(fs_info, ret,
2347 "Error while writing out transaction");
2349 * reloc_mutex has been unlocked, tree_log_mutex is still held
2350 * but we can't jump to unlock_tree_log causing double unlock
2352 mutex_unlock(&fs_info->tree_log_mutex);
2353 goto scrub_continue;
2357 * At this point, we should have written all the tree blocks allocated
2358 * in this transaction. So it's now safe to free the redirtyied extent
2361 btrfs_free_redirty_list(cur_trans);
2363 ret = write_all_supers(fs_info, 0);
2365 * the super is written, we can safely allow the tree-loggers
2366 * to go about their business
2368 mutex_unlock(&fs_info->tree_log_mutex);
2370 goto scrub_continue;
2373 * We needn't acquire the lock here because there is no other task
2374 * which can change it.
2376 cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2377 wake_up(&cur_trans->commit_wait);
2379 btrfs_finish_extent_commit(trans);
2381 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2382 btrfs_clear_space_info_full(fs_info);
2384 fs_info->last_trans_committed = cur_trans->transid;
2386 * We needn't acquire the lock here because there is no other task
2387 * which can change it.
2389 cur_trans->state = TRANS_STATE_COMPLETED;
2390 wake_up(&cur_trans->commit_wait);
2392 spin_lock(&fs_info->trans_lock);
2393 list_del_init(&cur_trans->list);
2394 spin_unlock(&fs_info->trans_lock);
2396 btrfs_put_transaction(cur_trans);
2397 btrfs_put_transaction(cur_trans);
2399 if (trans->type & __TRANS_FREEZABLE)
2400 sb_end_intwrite(fs_info->sb);
2402 trace_btrfs_transaction_commit(trans->root);
2404 btrfs_scrub_continue(fs_info);
2406 if (current->journal_info == trans)
2407 current->journal_info = NULL;
2409 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2414 mutex_unlock(&fs_info->tree_log_mutex);
2416 mutex_unlock(&fs_info->reloc_mutex);
2418 btrfs_scrub_continue(fs_info);
2419 cleanup_transaction:
2420 btrfs_trans_release_metadata(trans);
2421 btrfs_cleanup_pending_block_groups(trans);
2422 btrfs_trans_release_chunk_metadata(trans);
2423 trans->block_rsv = NULL;
2424 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2425 if (current->journal_info == trans)
2426 current->journal_info = NULL;
2427 cleanup_transaction(trans, ret);
2433 * return < 0 if error
2434 * 0 if there are no more dead_roots at the time of call
2435 * 1 there are more to be processed, call me again
2437 * The return value indicates there are certainly more snapshots to delete, but
2438 * if there comes a new one during processing, it may return 0. We don't mind,
2439 * because btrfs_commit_super will poke cleaner thread and it will process it a
2440 * few seconds later.
2442 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2445 struct btrfs_fs_info *fs_info = root->fs_info;
2447 spin_lock(&fs_info->trans_lock);
2448 if (list_empty(&fs_info->dead_roots)) {
2449 spin_unlock(&fs_info->trans_lock);
2452 root = list_first_entry(&fs_info->dead_roots,
2453 struct btrfs_root, root_list);
2454 list_del_init(&root->root_list);
2455 spin_unlock(&fs_info->trans_lock);
2457 btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2459 btrfs_kill_all_delayed_nodes(root);
2461 if (btrfs_header_backref_rev(root->node) <
2462 BTRFS_MIXED_BACKREF_REV)
2463 ret = btrfs_drop_snapshot(root, 0, 0);
2465 ret = btrfs_drop_snapshot(root, 1, 0);
2467 btrfs_put_root(root);
2468 return (ret < 0) ? 0 : 1;
2471 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2476 prev = xchg(&fs_info->pending_changes, 0);
2480 bit = 1 << BTRFS_PENDING_COMMIT;
2482 btrfs_debug(fs_info, "pending commit done");
2487 "unknown pending changes left 0x%lx, ignoring", prev);