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/sched/mm.h>
10 #include <linux/writeback.h>
11 #include <linux/pagemap.h>
12 #include <linux/blkdev.h>
13 #include <linux/uuid.h>
14 #include <linux/timekeeping.h>
18 #include "transaction.h"
22 #include "dev-replace.h"
24 #include "block-group.h"
25 #include "space-info.h"
27 #include "accessors.h"
28 #include "extent-tree.h"
29 #include "root-tree.h"
31 #include "uuid-tree.h"
33 #include "relocation.h"
36 static struct kmem_cache *btrfs_trans_handle_cachep;
39 * Transaction states and transitions
41 * No running transaction (fs tree blocks are not modified)
44 * | Call start_transaction() variants. Except btrfs_join_transaction_nostart().
46 * Transaction N [[TRANS_STATE_RUNNING]]
48 * | New trans handles can be attached to transaction N by calling all
49 * | start_transaction() variants.
52 * | Call btrfs_commit_transaction() on any trans handle attached to
55 * Transaction N [[TRANS_STATE_COMMIT_PREP]]
57 * | If there are simultaneous calls to btrfs_commit_transaction() one will win
58 * | the race and the rest will wait for the winner to commit the transaction.
60 * | The winner will wait for previous running transaction to completely finish
63 * Transaction N [[TRANS_STATE_COMMIT_START]]
65 * | Then one of the following happens:
66 * | - Wait for all other trans handle holders to release.
67 * | The btrfs_commit_transaction() caller will do the commit work.
68 * | - Wait for current transaction to be committed by others.
69 * | Other btrfs_commit_transaction() caller will do the commit work.
71 * | At this stage, only btrfs_join_transaction*() variants can attach
72 * | to this running transaction.
73 * | All other variants will wait for current one to finish and attach to
77 * | Caller is chosen to commit transaction N, and all other trans handle
78 * | haven been released.
80 * Transaction N [[TRANS_STATE_COMMIT_DOING]]
82 * | The heavy lifting transaction work is started.
83 * | From running delayed refs (modifying extent tree) to creating pending
84 * | snapshots, running qgroups.
85 * | In short, modify supporting trees to reflect modifications of subvolume
88 * | At this stage, all start_transaction() calls will wait for this
89 * | transaction to finish and attach to transaction N+1.
92 * | Until all supporting trees are updated.
94 * Transaction N [[TRANS_STATE_UNBLOCKED]]
96 * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
97 * | need to write them back to disk and update |
100 * | At this stage, new transaction is allowed to |
102 * | All new start_transaction() calls will be |
103 * | attached to transid N+1. |
106 * | Until all tree blocks are super blocks are |
107 * | written to block devices |
109 * Transaction N [[TRANS_STATE_COMPLETED]] V
110 * All tree blocks and super blocks are written. Transaction N+1
111 * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]]
112 * data structures will be cleaned up. | Life goes on
114 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
115 [TRANS_STATE_RUNNING] = 0U,
116 [TRANS_STATE_COMMIT_PREP] = 0U,
117 [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH),
118 [TRANS_STATE_COMMIT_DOING] = (__TRANS_START |
121 __TRANS_JOIN_NOSTART),
122 [TRANS_STATE_UNBLOCKED] = (__TRANS_START |
125 __TRANS_JOIN_NOLOCK |
126 __TRANS_JOIN_NOSTART),
127 [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START |
130 __TRANS_JOIN_NOLOCK |
131 __TRANS_JOIN_NOSTART),
132 [TRANS_STATE_COMPLETED] = (__TRANS_START |
135 __TRANS_JOIN_NOLOCK |
136 __TRANS_JOIN_NOSTART),
139 void btrfs_put_transaction(struct btrfs_transaction *transaction)
141 WARN_ON(refcount_read(&transaction->use_count) == 0);
142 if (refcount_dec_and_test(&transaction->use_count)) {
143 BUG_ON(!list_empty(&transaction->list));
144 WARN_ON(!RB_EMPTY_ROOT(
145 &transaction->delayed_refs.href_root.rb_root));
146 WARN_ON(!RB_EMPTY_ROOT(
147 &transaction->delayed_refs.dirty_extent_root));
148 if (transaction->delayed_refs.pending_csums)
149 btrfs_err(transaction->fs_info,
150 "pending csums is %llu",
151 transaction->delayed_refs.pending_csums);
153 * If any block groups are found in ->deleted_bgs then it's
154 * because the transaction was aborted and a commit did not
155 * happen (things failed before writing the new superblock
156 * and calling btrfs_finish_extent_commit()), so we can not
157 * discard the physical locations of the block groups.
159 while (!list_empty(&transaction->deleted_bgs)) {
160 struct btrfs_block_group *cache;
162 cache = list_first_entry(&transaction->deleted_bgs,
163 struct btrfs_block_group,
165 list_del_init(&cache->bg_list);
166 btrfs_unfreeze_block_group(cache);
167 btrfs_put_block_group(cache);
169 WARN_ON(!list_empty(&transaction->dev_update_list));
174 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
176 struct btrfs_transaction *cur_trans = trans->transaction;
177 struct btrfs_fs_info *fs_info = trans->fs_info;
178 struct btrfs_root *root, *tmp;
181 * At this point no one can be using this transaction to modify any tree
182 * and no one can start another transaction to modify any tree either.
184 ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING);
186 down_write(&fs_info->commit_root_sem);
188 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
189 fs_info->last_reloc_trans = trans->transid;
191 list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
193 list_del_init(&root->dirty_list);
194 free_extent_buffer(root->commit_root);
195 root->commit_root = btrfs_root_node(root);
196 extent_io_tree_release(&root->dirty_log_pages);
197 btrfs_qgroup_clean_swapped_blocks(root);
200 /* We can free old roots now. */
201 spin_lock(&cur_trans->dropped_roots_lock);
202 while (!list_empty(&cur_trans->dropped_roots)) {
203 root = list_first_entry(&cur_trans->dropped_roots,
204 struct btrfs_root, root_list);
205 list_del_init(&root->root_list);
206 spin_unlock(&cur_trans->dropped_roots_lock);
207 btrfs_free_log(trans, root);
208 btrfs_drop_and_free_fs_root(fs_info, root);
209 spin_lock(&cur_trans->dropped_roots_lock);
211 spin_unlock(&cur_trans->dropped_roots_lock);
213 up_write(&fs_info->commit_root_sem);
216 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
219 if (type & TRANS_EXTWRITERS)
220 atomic_inc(&trans->num_extwriters);
223 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
226 if (type & TRANS_EXTWRITERS)
227 atomic_dec(&trans->num_extwriters);
230 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
233 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
236 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
238 return atomic_read(&trans->num_extwriters);
242 * To be called after doing the chunk btree updates right after allocating a new
243 * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
244 * chunk after all chunk btree updates and after finishing the second phase of
245 * chunk allocation (btrfs_create_pending_block_groups()) in case some block
246 * group had its chunk item insertion delayed to the second phase.
248 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
250 struct btrfs_fs_info *fs_info = trans->fs_info;
252 if (!trans->chunk_bytes_reserved)
255 btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
256 trans->chunk_bytes_reserved, NULL);
257 trans->chunk_bytes_reserved = 0;
261 * either allocate a new transaction or hop into the existing one
263 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
266 struct btrfs_transaction *cur_trans;
268 spin_lock(&fs_info->trans_lock);
270 /* The file system has been taken offline. No new transactions. */
271 if (BTRFS_FS_ERROR(fs_info)) {
272 spin_unlock(&fs_info->trans_lock);
276 cur_trans = fs_info->running_transaction;
278 if (TRANS_ABORTED(cur_trans)) {
279 spin_unlock(&fs_info->trans_lock);
280 return cur_trans->aborted;
282 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
283 spin_unlock(&fs_info->trans_lock);
286 refcount_inc(&cur_trans->use_count);
287 atomic_inc(&cur_trans->num_writers);
288 extwriter_counter_inc(cur_trans, type);
289 spin_unlock(&fs_info->trans_lock);
290 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
291 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
294 spin_unlock(&fs_info->trans_lock);
297 * If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the
298 * current transaction, and commit it. If there is no transaction, just
301 if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART)
305 * JOIN_NOLOCK only happens during the transaction commit, so
306 * it is impossible that ->running_transaction is NULL
308 BUG_ON(type == TRANS_JOIN_NOLOCK);
310 cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
314 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
315 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
317 spin_lock(&fs_info->trans_lock);
318 if (fs_info->running_transaction) {
320 * someone started a transaction after we unlocked. Make sure
321 * to redo the checks above
323 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
324 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
327 } else if (BTRFS_FS_ERROR(fs_info)) {
328 spin_unlock(&fs_info->trans_lock);
329 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
330 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
335 cur_trans->fs_info = fs_info;
336 atomic_set(&cur_trans->pending_ordered, 0);
337 init_waitqueue_head(&cur_trans->pending_wait);
338 atomic_set(&cur_trans->num_writers, 1);
339 extwriter_counter_init(cur_trans, type);
340 init_waitqueue_head(&cur_trans->writer_wait);
341 init_waitqueue_head(&cur_trans->commit_wait);
342 cur_trans->state = TRANS_STATE_RUNNING;
344 * One for this trans handle, one so it will live on until we
345 * commit the transaction.
347 refcount_set(&cur_trans->use_count, 2);
348 cur_trans->flags = 0;
349 cur_trans->start_time = ktime_get_seconds();
351 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
353 cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
354 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
355 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
358 * although the tree mod log is per file system and not per transaction,
359 * the log must never go across transaction boundaries.
362 if (!list_empty(&fs_info->tree_mod_seq_list))
363 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
364 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
365 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
366 atomic64_set(&fs_info->tree_mod_seq, 0);
368 spin_lock_init(&cur_trans->delayed_refs.lock);
370 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
371 INIT_LIST_HEAD(&cur_trans->dev_update_list);
372 INIT_LIST_HEAD(&cur_trans->switch_commits);
373 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
374 INIT_LIST_HEAD(&cur_trans->io_bgs);
375 INIT_LIST_HEAD(&cur_trans->dropped_roots);
376 mutex_init(&cur_trans->cache_write_mutex);
377 spin_lock_init(&cur_trans->dirty_bgs_lock);
378 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
379 spin_lock_init(&cur_trans->dropped_roots_lock);
380 list_add_tail(&cur_trans->list, &fs_info->trans_list);
381 extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
382 IO_TREE_TRANS_DIRTY_PAGES);
383 extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
384 IO_TREE_FS_PINNED_EXTENTS);
385 btrfs_set_fs_generation(fs_info, fs_info->generation + 1);
386 cur_trans->transid = fs_info->generation;
387 fs_info->running_transaction = cur_trans;
388 cur_trans->aborted = 0;
389 spin_unlock(&fs_info->trans_lock);
395 * This does all the record keeping required to make sure that a shareable root
396 * is properly recorded in a given transaction. This is required to make sure
397 * the old root from before we joined the transaction is deleted when the
398 * transaction commits.
400 static int record_root_in_trans(struct btrfs_trans_handle *trans,
401 struct btrfs_root *root,
404 struct btrfs_fs_info *fs_info = root->fs_info;
407 if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
408 root->last_trans < trans->transid) || force) {
409 WARN_ON(!force && root->commit_root != root->node);
412 * see below for IN_TRANS_SETUP usage rules
413 * we have the reloc mutex held now, so there
414 * is only one writer in this function
416 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
418 /* make sure readers find IN_TRANS_SETUP before
419 * they find our root->last_trans update
423 spin_lock(&fs_info->fs_roots_radix_lock);
424 if (root->last_trans == trans->transid && !force) {
425 spin_unlock(&fs_info->fs_roots_radix_lock);
428 radix_tree_tag_set(&fs_info->fs_roots_radix,
429 (unsigned long)btrfs_root_id(root),
430 BTRFS_ROOT_TRANS_TAG);
431 spin_unlock(&fs_info->fs_roots_radix_lock);
432 root->last_trans = trans->transid;
434 /* this is pretty tricky. We don't want to
435 * take the relocation lock in btrfs_record_root_in_trans
436 * unless we're really doing the first setup for this root in
439 * Normally we'd use root->last_trans as a flag to decide
440 * if we want to take the expensive mutex.
442 * But, we have to set root->last_trans before we
443 * init the relocation root, otherwise, we trip over warnings
444 * in ctree.c. The solution used here is to flag ourselves
445 * with root IN_TRANS_SETUP. When this is 1, we're still
446 * fixing up the reloc trees and everyone must wait.
448 * When this is zero, they can trust root->last_trans and fly
449 * through btrfs_record_root_in_trans without having to take the
450 * lock. smp_wmb() makes sure that all the writes above are
451 * done before we pop in the zero below
453 ret = btrfs_init_reloc_root(trans, root);
454 smp_mb__before_atomic();
455 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
461 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
462 struct btrfs_root *root)
464 struct btrfs_fs_info *fs_info = root->fs_info;
465 struct btrfs_transaction *cur_trans = trans->transaction;
467 /* Add ourselves to the transaction dropped list */
468 spin_lock(&cur_trans->dropped_roots_lock);
469 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
470 spin_unlock(&cur_trans->dropped_roots_lock);
472 /* Make sure we don't try to update the root at commit time */
473 spin_lock(&fs_info->fs_roots_radix_lock);
474 radix_tree_tag_clear(&fs_info->fs_roots_radix,
475 (unsigned long)btrfs_root_id(root),
476 BTRFS_ROOT_TRANS_TAG);
477 spin_unlock(&fs_info->fs_roots_radix_lock);
480 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
481 struct btrfs_root *root)
483 struct btrfs_fs_info *fs_info = root->fs_info;
486 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
490 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
494 if (root->last_trans == trans->transid &&
495 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
498 mutex_lock(&fs_info->reloc_mutex);
499 ret = record_root_in_trans(trans, root, 0);
500 mutex_unlock(&fs_info->reloc_mutex);
505 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
507 return (trans->state >= TRANS_STATE_COMMIT_START &&
508 trans->state < TRANS_STATE_UNBLOCKED &&
509 !TRANS_ABORTED(trans));
512 /* wait for commit against the current transaction to become unblocked
513 * when this is done, it is safe to start a new transaction, but the current
514 * transaction might not be fully on disk.
516 static void wait_current_trans(struct btrfs_fs_info *fs_info)
518 struct btrfs_transaction *cur_trans;
520 spin_lock(&fs_info->trans_lock);
521 cur_trans = fs_info->running_transaction;
522 if (cur_trans && is_transaction_blocked(cur_trans)) {
523 refcount_inc(&cur_trans->use_count);
524 spin_unlock(&fs_info->trans_lock);
526 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
527 wait_event(fs_info->transaction_wait,
528 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
529 TRANS_ABORTED(cur_trans));
530 btrfs_put_transaction(cur_trans);
532 spin_unlock(&fs_info->trans_lock);
536 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
538 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
541 if (type == TRANS_START)
547 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
549 struct btrfs_fs_info *fs_info = root->fs_info;
551 if (!fs_info->reloc_ctl ||
552 !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
553 btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID ||
560 static int btrfs_reserve_trans_metadata(struct btrfs_fs_info *fs_info,
561 enum btrfs_reserve_flush_enum flush,
563 u64 *delayed_refs_bytes)
565 struct btrfs_space_info *si = fs_info->trans_block_rsv.space_info;
566 u64 bytes = num_bytes + *delayed_refs_bytes;
570 * We want to reserve all the bytes we may need all at once, so we only
571 * do 1 enospc flushing cycle per transaction start.
573 ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush);
576 * If we are an emergency flush, which can steal from the global block
577 * reserve, then attempt to not reserve space for the delayed refs, as
578 * we will consume space for them from the global block reserve.
580 if (ret && flush == BTRFS_RESERVE_FLUSH_ALL_STEAL) {
581 bytes -= *delayed_refs_bytes;
582 *delayed_refs_bytes = 0;
583 ret = btrfs_reserve_metadata_bytes(fs_info, si, bytes, flush);
589 static struct btrfs_trans_handle *
590 start_transaction(struct btrfs_root *root, unsigned int num_items,
591 unsigned int type, enum btrfs_reserve_flush_enum flush,
592 bool enforce_qgroups)
594 struct btrfs_fs_info *fs_info = root->fs_info;
595 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
596 struct btrfs_block_rsv *trans_rsv = &fs_info->trans_block_rsv;
597 struct btrfs_trans_handle *h;
598 struct btrfs_transaction *cur_trans;
600 u64 qgroup_reserved = 0;
601 u64 delayed_refs_bytes = 0;
602 bool reloc_reserved = false;
603 bool do_chunk_alloc = false;
606 if (BTRFS_FS_ERROR(fs_info))
607 return ERR_PTR(-EROFS);
609 if (current->journal_info) {
610 WARN_ON(type & TRANS_EXTWRITERS);
611 h = current->journal_info;
612 refcount_inc(&h->use_count);
613 WARN_ON(refcount_read(&h->use_count) > 2);
614 h->orig_rsv = h->block_rsv;
620 * Do the reservation before we join the transaction so we can do all
621 * the appropriate flushing if need be.
623 if (num_items && root != fs_info->chunk_root) {
624 qgroup_reserved = num_items * fs_info->nodesize;
626 * Use prealloc for now, as there might be a currently running
627 * transaction that could free this reserved space prematurely
630 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserved,
631 enforce_qgroups, false);
635 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
637 * If we plan to insert/update/delete "num_items" from a btree,
638 * we will also generate delayed refs for extent buffers in the
639 * respective btree paths, so reserve space for the delayed refs
640 * that will be generated by the caller as it modifies btrees.
641 * Try to reserve them to avoid excessive use of the global
644 delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, num_items);
647 * Do the reservation for the relocation root creation
649 if (need_reserve_reloc_root(root)) {
650 num_bytes += fs_info->nodesize;
651 reloc_reserved = true;
654 ret = btrfs_reserve_trans_metadata(fs_info, flush, num_bytes,
655 &delayed_refs_bytes);
659 btrfs_block_rsv_add_bytes(trans_rsv, num_bytes, true);
661 if (trans_rsv->space_info->force_alloc)
662 do_chunk_alloc = true;
663 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
664 !btrfs_block_rsv_full(delayed_refs_rsv)) {
666 * Some people call with btrfs_start_transaction(root, 0)
667 * because they can be throttled, but have some other mechanism
668 * for reserving space. We still want these guys to refill the
669 * delayed block_rsv so just add 1 items worth of reservation
672 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
677 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
684 * If we are JOIN_NOLOCK we're already committing a transaction and
685 * waiting on this guy, so we don't need to do the sb_start_intwrite
686 * because we're already holding a ref. We need this because we could
687 * have raced in and did an fsync() on a file which can kick a commit
688 * and then we deadlock with somebody doing a freeze.
690 * If we are ATTACH, it means we just want to catch the current
691 * transaction and commit it, so we needn't do sb_start_intwrite().
693 if (type & __TRANS_FREEZABLE)
694 sb_start_intwrite(fs_info->sb);
696 if (may_wait_transaction(fs_info, type))
697 wait_current_trans(fs_info);
700 ret = join_transaction(fs_info, type);
702 wait_current_trans(fs_info);
703 if (unlikely(type == TRANS_ATTACH ||
704 type == TRANS_JOIN_NOSTART))
707 } while (ret == -EBUSY);
712 cur_trans = fs_info->running_transaction;
714 h->transid = cur_trans->transid;
715 h->transaction = cur_trans;
716 refcount_set(&h->use_count, 1);
717 h->fs_info = root->fs_info;
720 INIT_LIST_HEAD(&h->new_bgs);
721 btrfs_init_metadata_block_rsv(fs_info, &h->delayed_rsv, BTRFS_BLOCK_RSV_DELOPS);
724 if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
725 may_wait_transaction(fs_info, type)) {
726 current->journal_info = h;
727 btrfs_commit_transaction(h);
732 trace_btrfs_space_reservation(fs_info, "transaction",
733 h->transid, num_bytes, 1);
734 h->block_rsv = trans_rsv;
735 h->bytes_reserved = num_bytes;
736 if (delayed_refs_bytes > 0) {
737 trace_btrfs_space_reservation(fs_info,
738 "local_delayed_refs_rsv",
740 delayed_refs_bytes, 1);
741 h->delayed_refs_bytes_reserved = delayed_refs_bytes;
742 btrfs_block_rsv_add_bytes(&h->delayed_rsv, delayed_refs_bytes, true);
743 delayed_refs_bytes = 0;
745 h->reloc_reserved = reloc_reserved;
749 if (!current->journal_info)
750 current->journal_info = h;
753 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
754 * ALLOC_FORCE the first run through, and then we won't allocate for
755 * anybody else who races in later. We don't care about the return
758 if (do_chunk_alloc && num_bytes) {
759 u64 flags = h->block_rsv->space_info->flags;
761 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
762 CHUNK_ALLOC_NO_FORCE);
766 * btrfs_record_root_in_trans() needs to alloc new extents, and may
767 * call btrfs_join_transaction() while we're also starting a
770 * Thus it need to be called after current->journal_info initialized,
771 * or we can deadlock.
773 ret = btrfs_record_root_in_trans(h, root);
776 * The transaction handle is fully initialized and linked with
777 * other structures so it needs to be ended in case of errors,
780 btrfs_end_transaction(h);
784 * Now that we have found a transaction to be a part of, convert the
785 * qgroup reservation from prealloc to pertrans. A different transaction
786 * can't race in and free our pertrans out from under us.
789 btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
794 if (type & __TRANS_FREEZABLE)
795 sb_end_intwrite(fs_info->sb);
796 kmem_cache_free(btrfs_trans_handle_cachep, h);
799 btrfs_block_rsv_release(fs_info, trans_rsv, num_bytes, NULL);
800 if (delayed_refs_bytes)
801 btrfs_space_info_free_bytes_may_use(fs_info, trans_rsv->space_info,
804 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
808 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
809 unsigned int num_items)
811 return start_transaction(root, num_items, TRANS_START,
812 BTRFS_RESERVE_FLUSH_ALL, true);
815 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
816 struct btrfs_root *root,
817 unsigned int num_items)
819 return start_transaction(root, num_items, TRANS_START,
820 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
823 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
825 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
829 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
831 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
832 BTRFS_RESERVE_NO_FLUSH, true);
836 * Similar to regular join but it never starts a transaction when none is
837 * running or when there's a running one at a state >= TRANS_STATE_UNBLOCKED.
838 * This is similar to btrfs_attach_transaction() but it allows the join to
839 * happen if the transaction commit already started but it's not yet in the
840 * "doing" phase (the state is < TRANS_STATE_COMMIT_DOING).
842 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
844 return start_transaction(root, 0, TRANS_JOIN_NOSTART,
845 BTRFS_RESERVE_NO_FLUSH, true);
849 * Catch the running transaction.
851 * It is used when we want to commit the current the transaction, but
852 * don't want to start a new one.
854 * Note: If this function return -ENOENT, it just means there is no
855 * running transaction. But it is possible that the inactive transaction
856 * is still in the memory, not fully on disk. If you hope there is no
857 * inactive transaction in the fs when -ENOENT is returned, you should
859 * btrfs_attach_transaction_barrier()
861 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
863 return start_transaction(root, 0, TRANS_ATTACH,
864 BTRFS_RESERVE_NO_FLUSH, true);
868 * Catch the running transaction.
870 * It is similar to the above function, the difference is this one
871 * will wait for all the inactive transactions until they fully
874 struct btrfs_trans_handle *
875 btrfs_attach_transaction_barrier(struct btrfs_root *root)
877 struct btrfs_trans_handle *trans;
879 trans = start_transaction(root, 0, TRANS_ATTACH,
880 BTRFS_RESERVE_NO_FLUSH, true);
881 if (trans == ERR_PTR(-ENOENT)) {
884 ret = btrfs_wait_for_commit(root->fs_info, 0);
892 /* Wait for a transaction commit to reach at least the given state. */
893 static noinline void wait_for_commit(struct btrfs_transaction *commit,
894 const enum btrfs_trans_state min_state)
896 struct btrfs_fs_info *fs_info = commit->fs_info;
897 u64 transid = commit->transid;
901 * At the moment this function is called with min_state either being
902 * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED.
904 if (min_state == TRANS_STATE_COMPLETED)
905 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
907 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
910 wait_event(commit->commit_wait, commit->state >= min_state);
912 btrfs_put_transaction(commit);
914 if (min_state < TRANS_STATE_COMPLETED)
918 * A transaction isn't really completed until all of the
919 * previous transactions are completed, but with fsync we can
920 * end up with SUPER_COMMITTED transactions before a COMPLETED
921 * transaction. Wait for those.
924 spin_lock(&fs_info->trans_lock);
925 commit = list_first_entry_or_null(&fs_info->trans_list,
926 struct btrfs_transaction,
928 if (!commit || commit->transid > transid) {
929 spin_unlock(&fs_info->trans_lock);
932 refcount_inc(&commit->use_count);
934 spin_unlock(&fs_info->trans_lock);
938 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
940 struct btrfs_transaction *cur_trans = NULL, *t;
944 if (transid <= btrfs_get_last_trans_committed(fs_info))
947 /* find specified transaction */
948 spin_lock(&fs_info->trans_lock);
949 list_for_each_entry(t, &fs_info->trans_list, list) {
950 if (t->transid == transid) {
952 refcount_inc(&cur_trans->use_count);
956 if (t->transid > transid) {
961 spin_unlock(&fs_info->trans_lock);
964 * The specified transaction doesn't exist, or we
965 * raced with btrfs_commit_transaction
968 if (transid > btrfs_get_last_trans_committed(fs_info))
973 /* find newest transaction that is committing | committed */
974 spin_lock(&fs_info->trans_lock);
975 list_for_each_entry_reverse(t, &fs_info->trans_list,
977 if (t->state >= TRANS_STATE_COMMIT_START) {
978 if (t->state == TRANS_STATE_COMPLETED)
981 refcount_inc(&cur_trans->use_count);
985 spin_unlock(&fs_info->trans_lock);
987 goto out; /* nothing committing|committed */
990 wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
991 ret = cur_trans->aborted;
992 btrfs_put_transaction(cur_trans);
997 void btrfs_throttle(struct btrfs_fs_info *fs_info)
999 wait_current_trans(fs_info);
1002 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
1004 struct btrfs_transaction *cur_trans = trans->transaction;
1006 if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
1007 test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
1010 if (btrfs_check_space_for_delayed_refs(trans->fs_info))
1013 return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50);
1016 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
1019 struct btrfs_fs_info *fs_info = trans->fs_info;
1021 if (!trans->block_rsv) {
1022 ASSERT(!trans->bytes_reserved);
1023 ASSERT(!trans->delayed_refs_bytes_reserved);
1027 if (!trans->bytes_reserved) {
1028 ASSERT(!trans->delayed_refs_bytes_reserved);
1032 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
1033 trace_btrfs_space_reservation(fs_info, "transaction",
1034 trans->transid, trans->bytes_reserved, 0);
1035 btrfs_block_rsv_release(fs_info, trans->block_rsv,
1036 trans->bytes_reserved, NULL);
1037 trans->bytes_reserved = 0;
1039 if (!trans->delayed_refs_bytes_reserved)
1042 trace_btrfs_space_reservation(fs_info, "local_delayed_refs_rsv",
1044 trans->delayed_refs_bytes_reserved, 0);
1045 btrfs_block_rsv_release(fs_info, &trans->delayed_rsv,
1046 trans->delayed_refs_bytes_reserved, NULL);
1047 trans->delayed_refs_bytes_reserved = 0;
1050 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
1053 struct btrfs_fs_info *info = trans->fs_info;
1054 struct btrfs_transaction *cur_trans = trans->transaction;
1057 if (refcount_read(&trans->use_count) > 1) {
1058 refcount_dec(&trans->use_count);
1059 trans->block_rsv = trans->orig_rsv;
1063 btrfs_trans_release_metadata(trans);
1064 trans->block_rsv = NULL;
1066 btrfs_create_pending_block_groups(trans);
1068 btrfs_trans_release_chunk_metadata(trans);
1070 if (trans->type & __TRANS_FREEZABLE)
1071 sb_end_intwrite(info->sb);
1073 WARN_ON(cur_trans != info->running_transaction);
1074 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
1075 atomic_dec(&cur_trans->num_writers);
1076 extwriter_counter_dec(cur_trans, trans->type);
1078 cond_wake_up(&cur_trans->writer_wait);
1080 btrfs_lockdep_release(info, btrfs_trans_num_extwriters);
1081 btrfs_lockdep_release(info, btrfs_trans_num_writers);
1083 btrfs_put_transaction(cur_trans);
1085 if (current->journal_info == trans)
1086 current->journal_info = NULL;
1089 btrfs_run_delayed_iputs(info);
1091 if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) {
1092 wake_up_process(info->transaction_kthread);
1093 if (TRANS_ABORTED(trans))
1094 ret = trans->aborted;
1099 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1103 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1105 return __btrfs_end_transaction(trans, 0);
1108 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1110 return __btrfs_end_transaction(trans, 1);
1114 * when btree blocks are allocated, they have some corresponding bits set for
1115 * them in one of two extent_io trees. This is used to make sure all of
1116 * those extents are sent to disk but does not wait on them
1118 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1119 struct extent_io_tree *dirty_pages, int mark)
1122 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1123 struct extent_state *cached_state = NULL;
1127 while (find_first_extent_bit(dirty_pages, start, &start, &end,
1128 mark, &cached_state)) {
1129 bool wait_writeback = false;
1131 ret = convert_extent_bit(dirty_pages, start, end,
1133 mark, &cached_state);
1135 * convert_extent_bit can return -ENOMEM, which is most of the
1136 * time a temporary error. So when it happens, ignore the error
1137 * and wait for writeback of this range to finish - because we
1138 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1139 * to __btrfs_wait_marked_extents() would not know that
1140 * writeback for this range started and therefore wouldn't
1141 * wait for it to finish - we don't want to commit a
1142 * superblock that points to btree nodes/leafs for which
1143 * writeback hasn't finished yet (and without errors).
1144 * We cleanup any entries left in the io tree when committing
1145 * the transaction (through extent_io_tree_release()).
1147 if (ret == -ENOMEM) {
1149 wait_writeback = true;
1152 ret = filemap_fdatawrite_range(mapping, start, end);
1153 if (!ret && wait_writeback)
1154 ret = filemap_fdatawait_range(mapping, start, end);
1155 free_extent_state(cached_state);
1158 cached_state = NULL;
1166 * when btree blocks are allocated, they have some corresponding bits set for
1167 * them in one of two extent_io trees. This is used to make sure all of
1168 * those extents are on disk for transaction or log commit. We wait
1169 * on all the pages and clear them from the dirty pages state tree
1171 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1172 struct extent_io_tree *dirty_pages)
1174 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1175 struct extent_state *cached_state = NULL;
1180 while (find_first_extent_bit(dirty_pages, start, &start, &end,
1181 EXTENT_NEED_WAIT, &cached_state)) {
1183 * Ignore -ENOMEM errors returned by clear_extent_bit().
1184 * When committing the transaction, we'll remove any entries
1185 * left in the io tree. For a log commit, we don't remove them
1186 * after committing the log because the tree can be accessed
1187 * concurrently - we do it only at transaction commit time when
1188 * it's safe to do it (through extent_io_tree_release()).
1190 ret = clear_extent_bit(dirty_pages, start, end,
1191 EXTENT_NEED_WAIT, &cached_state);
1195 ret = filemap_fdatawait_range(mapping, start, end);
1196 free_extent_state(cached_state);
1199 cached_state = NULL;
1206 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1207 struct extent_io_tree *dirty_pages)
1209 bool errors = false;
1212 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1213 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1221 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1223 struct btrfs_fs_info *fs_info = log_root->fs_info;
1224 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1225 bool errors = false;
1228 ASSERT(btrfs_root_id(log_root) == BTRFS_TREE_LOG_OBJECTID);
1230 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1231 if ((mark & EXTENT_DIRTY) &&
1232 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1235 if ((mark & EXTENT_NEW) &&
1236 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1245 * When btree blocks are allocated the corresponding extents are marked dirty.
1246 * This function ensures such extents are persisted on disk for transaction or
1249 * @trans: transaction whose dirty pages we'd like to write
1251 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1255 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1256 struct btrfs_fs_info *fs_info = trans->fs_info;
1257 struct blk_plug plug;
1259 blk_start_plug(&plug);
1260 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1261 blk_finish_plug(&plug);
1262 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1264 extent_io_tree_release(&trans->transaction->dirty_pages);
1275 * this is used to update the root pointer in the tree of tree roots.
1277 * But, in the case of the extent allocation tree, updating the root
1278 * pointer may allocate blocks which may change the root of the extent
1281 * So, this loops and repeats and makes sure the cowonly root didn't
1282 * change while the root pointer was being updated in the metadata.
1284 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1285 struct btrfs_root *root)
1288 u64 old_root_bytenr;
1290 struct btrfs_fs_info *fs_info = root->fs_info;
1291 struct btrfs_root *tree_root = fs_info->tree_root;
1293 old_root_used = btrfs_root_used(&root->root_item);
1296 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1297 if (old_root_bytenr == root->node->start &&
1298 old_root_used == btrfs_root_used(&root->root_item))
1301 btrfs_set_root_node(&root->root_item, root->node);
1302 ret = btrfs_update_root(trans, tree_root,
1308 old_root_used = btrfs_root_used(&root->root_item);
1315 * update all the cowonly tree roots on disk
1317 * The error handling in this function may not be obvious. Any of the
1318 * failures will cause the file system to go offline. We still need
1319 * to clean up the delayed refs.
1321 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1323 struct btrfs_fs_info *fs_info = trans->fs_info;
1324 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1325 struct list_head *io_bgs = &trans->transaction->io_bgs;
1326 struct list_head *next;
1327 struct extent_buffer *eb;
1331 * At this point no one can be using this transaction to modify any tree
1332 * and no one can start another transaction to modify any tree either.
1334 ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1336 eb = btrfs_lock_root_node(fs_info->tree_root);
1337 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1338 0, &eb, BTRFS_NESTING_COW);
1339 btrfs_tree_unlock(eb);
1340 free_extent_buffer(eb);
1345 ret = btrfs_run_dev_stats(trans);
1348 ret = btrfs_run_dev_replace(trans);
1351 ret = btrfs_run_qgroups(trans);
1355 ret = btrfs_setup_space_cache(trans);
1360 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1361 struct btrfs_root *root;
1362 next = fs_info->dirty_cowonly_roots.next;
1363 list_del_init(next);
1364 root = list_entry(next, struct btrfs_root, dirty_list);
1365 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1367 list_add_tail(&root->dirty_list,
1368 &trans->transaction->switch_commits);
1369 ret = update_cowonly_root(trans, root);
1374 /* Now flush any delayed refs generated by updating all of the roots */
1375 ret = btrfs_run_delayed_refs(trans, U64_MAX);
1379 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1380 ret = btrfs_write_dirty_block_groups(trans);
1385 * We're writing the dirty block groups, which could generate
1386 * delayed refs, which could generate more dirty block groups,
1387 * so we want to keep this flushing in this loop to make sure
1388 * everything gets run.
1390 ret = btrfs_run_delayed_refs(trans, U64_MAX);
1395 if (!list_empty(&fs_info->dirty_cowonly_roots))
1398 /* Update dev-replace pointer once everything is committed */
1399 fs_info->dev_replace.committed_cursor_left =
1400 fs_info->dev_replace.cursor_left_last_write_of_item;
1406 * If we had a pending drop we need to see if there are any others left in our
1407 * dead roots list, and if not clear our bit and wake any waiters.
1409 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
1412 * We put the drop in progress roots at the front of the list, so if the
1413 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
1416 spin_lock(&fs_info->trans_lock);
1417 if (!list_empty(&fs_info->dead_roots)) {
1418 struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
1421 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
1422 spin_unlock(&fs_info->trans_lock);
1426 spin_unlock(&fs_info->trans_lock);
1428 btrfs_wake_unfinished_drop(fs_info);
1432 * dead roots are old snapshots that need to be deleted. This allocates
1433 * a dirty root struct and adds it into the list of dead roots that need to
1436 void btrfs_add_dead_root(struct btrfs_root *root)
1438 struct btrfs_fs_info *fs_info = root->fs_info;
1440 spin_lock(&fs_info->trans_lock);
1441 if (list_empty(&root->root_list)) {
1442 btrfs_grab_root(root);
1444 /* We want to process the partially complete drops first. */
1445 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
1446 list_add(&root->root_list, &fs_info->dead_roots);
1448 list_add_tail(&root->root_list, &fs_info->dead_roots);
1450 spin_unlock(&fs_info->trans_lock);
1454 * Update each subvolume root and its relocation root, if it exists, in the tree
1455 * of tree roots. Also free log roots if they exist.
1457 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1459 struct btrfs_fs_info *fs_info = trans->fs_info;
1460 struct btrfs_root *gang[8];
1465 * At this point no one can be using this transaction to modify any tree
1466 * and no one can start another transaction to modify any tree either.
1468 ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1470 spin_lock(&fs_info->fs_roots_radix_lock);
1472 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1475 BTRFS_ROOT_TRANS_TAG);
1478 for (i = 0; i < ret; i++) {
1479 struct btrfs_root *root = gang[i];
1483 * At this point we can neither have tasks logging inodes
1484 * from a root nor trying to commit a log tree.
1486 ASSERT(atomic_read(&root->log_writers) == 0);
1487 ASSERT(atomic_read(&root->log_commit[0]) == 0);
1488 ASSERT(atomic_read(&root->log_commit[1]) == 0);
1490 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1491 (unsigned long)btrfs_root_id(root),
1492 BTRFS_ROOT_TRANS_TAG);
1493 btrfs_qgroup_free_meta_all_pertrans(root);
1494 spin_unlock(&fs_info->fs_roots_radix_lock);
1496 btrfs_free_log(trans, root);
1497 ret2 = btrfs_update_reloc_root(trans, root);
1501 /* see comments in should_cow_block() */
1502 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1503 smp_mb__after_atomic();
1505 if (root->commit_root != root->node) {
1506 list_add_tail(&root->dirty_list,
1507 &trans->transaction->switch_commits);
1508 btrfs_set_root_node(&root->root_item,
1512 ret2 = btrfs_update_root(trans, fs_info->tree_root,
1517 spin_lock(&fs_info->fs_roots_radix_lock);
1520 spin_unlock(&fs_info->fs_roots_radix_lock);
1525 * Do all special snapshot related qgroup dirty hack.
1527 * Will do all needed qgroup inherit and dirty hack like switch commit
1528 * roots inside one transaction and write all btree into disk, to make
1531 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1532 struct btrfs_root *src,
1533 struct btrfs_root *parent,
1534 struct btrfs_qgroup_inherit *inherit,
1537 struct btrfs_fs_info *fs_info = src->fs_info;
1541 * Save some performance in the case that qgroups are not enabled. If
1542 * this check races with the ioctl, rescan will kick in anyway.
1544 if (!btrfs_qgroup_full_accounting(fs_info))
1548 * Ensure dirty @src will be committed. Or, after coming
1549 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1550 * recorded root will never be updated again, causing an outdated root
1553 ret = record_root_in_trans(trans, src, 1);
1558 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1559 * src root, so we must run the delayed refs here.
1561 * However this isn't particularly fool proof, because there's no
1562 * synchronization keeping us from changing the tree after this point
1563 * before we do the qgroup_inherit, or even from making changes while
1564 * we're doing the qgroup_inherit. But that's a problem for the future,
1565 * for now flush the delayed refs to narrow the race window where the
1566 * qgroup counters could end up wrong.
1568 ret = btrfs_run_delayed_refs(trans, U64_MAX);
1570 btrfs_abort_transaction(trans, ret);
1574 ret = commit_fs_roots(trans);
1577 ret = btrfs_qgroup_account_extents(trans);
1581 /* Now qgroup are all updated, we can inherit it to new qgroups */
1582 ret = btrfs_qgroup_inherit(trans, btrfs_root_id(src), dst_objectid,
1583 btrfs_root_id(parent), inherit);
1588 * Now we do a simplified commit transaction, which will:
1589 * 1) commit all subvolume and extent tree
1590 * To ensure all subvolume and extent tree have a valid
1591 * commit_root to accounting later insert_dir_item()
1592 * 2) write all btree blocks onto disk
1593 * This is to make sure later btree modification will be cowed
1594 * Or commit_root can be populated and cause wrong qgroup numbers
1595 * In this simplified commit, we don't really care about other trees
1596 * like chunk and root tree, as they won't affect qgroup.
1597 * And we don't write super to avoid half committed status.
1599 ret = commit_cowonly_roots(trans);
1602 switch_commit_roots(trans);
1603 ret = btrfs_write_and_wait_transaction(trans);
1605 btrfs_handle_fs_error(fs_info, ret,
1606 "Error while writing out transaction for qgroup");
1610 * Force parent root to be updated, as we recorded it before so its
1611 * last_trans == cur_transid.
1612 * Or it won't be committed again onto disk after later
1616 ret = record_root_in_trans(trans, parent, 1);
1621 * new snapshots need to be created at a very specific time in the
1622 * transaction commit. This does the actual creation.
1625 * If the error which may affect the commitment of the current transaction
1626 * happens, we should return the error number. If the error which just affect
1627 * the creation of the pending snapshots, just return 0.
1629 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1630 struct btrfs_pending_snapshot *pending)
1633 struct btrfs_fs_info *fs_info = trans->fs_info;
1634 struct btrfs_key key;
1635 struct btrfs_root_item *new_root_item;
1636 struct btrfs_root *tree_root = fs_info->tree_root;
1637 struct btrfs_root *root = pending->root;
1638 struct btrfs_root *parent_root;
1639 struct btrfs_block_rsv *rsv;
1640 struct inode *parent_inode = pending->dir;
1641 struct btrfs_path *path;
1642 struct btrfs_dir_item *dir_item;
1643 struct extent_buffer *tmp;
1644 struct extent_buffer *old;
1645 struct timespec64 cur_time;
1651 unsigned int nofs_flags;
1652 struct fscrypt_name fname;
1654 ASSERT(pending->path);
1655 path = pending->path;
1657 ASSERT(pending->root_item);
1658 new_root_item = pending->root_item;
1661 * We're inside a transaction and must make sure that any potential
1662 * allocations with GFP_KERNEL in fscrypt won't recurse back to
1665 nofs_flags = memalloc_nofs_save();
1666 pending->error = fscrypt_setup_filename(parent_inode,
1667 &pending->dentry->d_name, 0,
1669 memalloc_nofs_restore(nofs_flags);
1673 pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1678 * Make qgroup to skip current new snapshot's qgroupid, as it is
1679 * accounted by later btrfs_qgroup_inherit().
1681 btrfs_set_skip_qgroup(trans, objectid);
1683 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1685 if (to_reserve > 0) {
1686 pending->error = btrfs_block_rsv_add(fs_info,
1687 &pending->block_rsv,
1689 BTRFS_RESERVE_NO_FLUSH);
1691 goto clear_skip_qgroup;
1694 key.objectid = objectid;
1695 key.offset = (u64)-1;
1696 key.type = BTRFS_ROOT_ITEM_KEY;
1698 rsv = trans->block_rsv;
1699 trans->block_rsv = &pending->block_rsv;
1700 trans->bytes_reserved = trans->block_rsv->reserved;
1701 trace_btrfs_space_reservation(fs_info, "transaction",
1703 trans->bytes_reserved, 1);
1704 parent_root = BTRFS_I(parent_inode)->root;
1705 ret = record_root_in_trans(trans, parent_root, 0);
1708 cur_time = current_time(parent_inode);
1711 * insert the directory item
1713 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1715 btrfs_abort_transaction(trans, ret);
1719 /* check if there is a file/dir which has the same name. */
1720 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1721 btrfs_ino(BTRFS_I(parent_inode)),
1722 &fname.disk_name, 0);
1723 if (dir_item != NULL && !IS_ERR(dir_item)) {
1724 pending->error = -EEXIST;
1725 goto dir_item_existed;
1726 } else if (IS_ERR(dir_item)) {
1727 ret = PTR_ERR(dir_item);
1728 btrfs_abort_transaction(trans, ret);
1731 btrfs_release_path(path);
1733 ret = btrfs_create_qgroup(trans, objectid);
1734 if (ret && ret != -EEXIST) {
1735 btrfs_abort_transaction(trans, ret);
1740 * pull in the delayed directory update
1741 * and the delayed inode item
1742 * otherwise we corrupt the FS during
1745 ret = btrfs_run_delayed_items(trans);
1746 if (ret) { /* Transaction aborted */
1747 btrfs_abort_transaction(trans, ret);
1751 ret = record_root_in_trans(trans, root, 0);
1753 btrfs_abort_transaction(trans, ret);
1756 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1757 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1758 btrfs_check_and_init_root_item(new_root_item);
1760 root_flags = btrfs_root_flags(new_root_item);
1761 if (pending->readonly)
1762 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1764 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1765 btrfs_set_root_flags(new_root_item, root_flags);
1767 btrfs_set_root_generation_v2(new_root_item,
1769 generate_random_guid(new_root_item->uuid);
1770 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1772 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1773 memset(new_root_item->received_uuid, 0,
1774 sizeof(new_root_item->received_uuid));
1775 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1776 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1777 btrfs_set_root_stransid(new_root_item, 0);
1778 btrfs_set_root_rtransid(new_root_item, 0);
1780 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1781 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1782 btrfs_set_root_otransid(new_root_item, trans->transid);
1784 old = btrfs_lock_root_node(root);
1785 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1788 btrfs_tree_unlock(old);
1789 free_extent_buffer(old);
1790 btrfs_abort_transaction(trans, ret);
1794 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1795 /* clean up in any case */
1796 btrfs_tree_unlock(old);
1797 free_extent_buffer(old);
1799 btrfs_abort_transaction(trans, ret);
1802 /* see comments in should_cow_block() */
1803 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1806 btrfs_set_root_node(new_root_item, tmp);
1807 /* record when the snapshot was created in key.offset */
1808 key.offset = trans->transid;
1809 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1810 btrfs_tree_unlock(tmp);
1811 free_extent_buffer(tmp);
1813 btrfs_abort_transaction(trans, ret);
1818 * insert root back/forward references
1820 ret = btrfs_add_root_ref(trans, objectid,
1821 btrfs_root_id(parent_root),
1822 btrfs_ino(BTRFS_I(parent_inode)), index,
1825 btrfs_abort_transaction(trans, ret);
1829 key.offset = (u64)-1;
1830 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, &pending->anon_dev);
1831 if (IS_ERR(pending->snap)) {
1832 ret = PTR_ERR(pending->snap);
1833 pending->snap = NULL;
1834 btrfs_abort_transaction(trans, ret);
1838 ret = btrfs_reloc_post_snapshot(trans, pending);
1840 btrfs_abort_transaction(trans, ret);
1845 * Do special qgroup accounting for snapshot, as we do some qgroup
1846 * snapshot hack to do fast snapshot.
1847 * To co-operate with that hack, we do hack again.
1848 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1850 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL)
1851 ret = qgroup_account_snapshot(trans, root, parent_root,
1852 pending->inherit, objectid);
1853 else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
1854 ret = btrfs_qgroup_inherit(trans, btrfs_root_id(root), objectid,
1855 btrfs_root_id(parent_root), pending->inherit);
1859 ret = btrfs_insert_dir_item(trans, &fname.disk_name,
1860 BTRFS_I(parent_inode), &key, BTRFS_FT_DIR,
1863 btrfs_abort_transaction(trans, ret);
1867 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1868 fname.disk_name.len * 2);
1869 inode_set_mtime_to_ts(parent_inode,
1870 inode_set_ctime_current(parent_inode));
1871 ret = btrfs_update_inode_fallback(trans, BTRFS_I(parent_inode));
1873 btrfs_abort_transaction(trans, ret);
1876 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1877 BTRFS_UUID_KEY_SUBVOL,
1880 btrfs_abort_transaction(trans, ret);
1883 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1884 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1885 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1887 if (ret && ret != -EEXIST) {
1888 btrfs_abort_transaction(trans, ret);
1894 pending->error = ret;
1896 trans->block_rsv = rsv;
1897 trans->bytes_reserved = 0;
1899 btrfs_clear_skip_qgroup(trans);
1901 fscrypt_free_filename(&fname);
1903 kfree(new_root_item);
1904 pending->root_item = NULL;
1905 btrfs_free_path(path);
1906 pending->path = NULL;
1912 * create all the snapshots we've scheduled for creation
1914 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1916 struct btrfs_pending_snapshot *pending, *next;
1917 struct list_head *head = &trans->transaction->pending_snapshots;
1920 list_for_each_entry_safe(pending, next, head, list) {
1921 list_del(&pending->list);
1922 ret = create_pending_snapshot(trans, pending);
1929 static void update_super_roots(struct btrfs_fs_info *fs_info)
1931 struct btrfs_root_item *root_item;
1932 struct btrfs_super_block *super;
1934 super = fs_info->super_copy;
1936 root_item = &fs_info->chunk_root->root_item;
1937 super->chunk_root = root_item->bytenr;
1938 super->chunk_root_generation = root_item->generation;
1939 super->chunk_root_level = root_item->level;
1941 root_item = &fs_info->tree_root->root_item;
1942 super->root = root_item->bytenr;
1943 super->generation = root_item->generation;
1944 super->root_level = root_item->level;
1945 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1946 super->cache_generation = root_item->generation;
1947 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1948 super->cache_generation = 0;
1949 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1950 super->uuid_tree_generation = root_item->generation;
1953 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1955 struct btrfs_transaction *trans;
1958 spin_lock(&info->trans_lock);
1959 trans = info->running_transaction;
1961 ret = is_transaction_blocked(trans);
1962 spin_unlock(&info->trans_lock);
1966 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1968 struct btrfs_fs_info *fs_info = trans->fs_info;
1969 struct btrfs_transaction *cur_trans;
1971 /* Kick the transaction kthread. */
1972 set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
1973 wake_up_process(fs_info->transaction_kthread);
1975 /* take transaction reference */
1976 cur_trans = trans->transaction;
1977 refcount_inc(&cur_trans->use_count);
1979 btrfs_end_transaction(trans);
1982 * Wait for the current transaction commit to start and block
1983 * subsequent transaction joins
1985 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
1986 wait_event(fs_info->transaction_blocked_wait,
1987 cur_trans->state >= TRANS_STATE_COMMIT_START ||
1988 TRANS_ABORTED(cur_trans));
1989 btrfs_put_transaction(cur_trans);
1992 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1994 struct btrfs_fs_info *fs_info = trans->fs_info;
1995 struct btrfs_transaction *cur_trans = trans->transaction;
1997 WARN_ON(refcount_read(&trans->use_count) > 1);
1999 btrfs_abort_transaction(trans, err);
2001 spin_lock(&fs_info->trans_lock);
2004 * If the transaction is removed from the list, it means this
2005 * transaction has been committed successfully, so it is impossible
2006 * to call the cleanup function.
2008 BUG_ON(list_empty(&cur_trans->list));
2010 if (cur_trans == fs_info->running_transaction) {
2011 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2012 spin_unlock(&fs_info->trans_lock);
2015 * The thread has already released the lockdep map as reader
2016 * already in btrfs_commit_transaction().
2018 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2019 wait_event(cur_trans->writer_wait,
2020 atomic_read(&cur_trans->num_writers) == 1);
2022 spin_lock(&fs_info->trans_lock);
2026 * Now that we know no one else is still using the transaction we can
2027 * remove the transaction from the list of transactions. This avoids
2028 * the transaction kthread from cleaning up the transaction while some
2029 * other task is still using it, which could result in a use-after-free
2030 * on things like log trees, as it forces the transaction kthread to
2031 * wait for this transaction to be cleaned up by us.
2033 list_del_init(&cur_trans->list);
2035 spin_unlock(&fs_info->trans_lock);
2037 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
2039 spin_lock(&fs_info->trans_lock);
2040 if (cur_trans == fs_info->running_transaction)
2041 fs_info->running_transaction = NULL;
2042 spin_unlock(&fs_info->trans_lock);
2044 if (trans->type & __TRANS_FREEZABLE)
2045 sb_end_intwrite(fs_info->sb);
2046 btrfs_put_transaction(cur_trans);
2047 btrfs_put_transaction(cur_trans);
2049 trace_btrfs_transaction_commit(fs_info);
2051 if (current->journal_info == trans)
2052 current->journal_info = NULL;
2055 * If relocation is running, we can't cancel scrub because that will
2056 * result in a deadlock. Before relocating a block group, relocation
2057 * pauses scrub, then starts and commits a transaction before unpausing
2058 * scrub. If the transaction commit is being done by the relocation
2059 * task or triggered by another task and the relocation task is waiting
2060 * for the commit, and we end up here due to an error in the commit
2061 * path, then calling btrfs_scrub_cancel() will deadlock, as we are
2062 * asking for scrub to stop while having it asked to be paused higher
2063 * above in relocation code.
2065 if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
2066 btrfs_scrub_cancel(fs_info);
2068 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2072 * Release reserved delayed ref space of all pending block groups of the
2073 * transaction and remove them from the list
2075 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2077 struct btrfs_fs_info *fs_info = trans->fs_info;
2078 struct btrfs_block_group *block_group, *tmp;
2080 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2081 btrfs_dec_delayed_refs_rsv_bg_inserts(fs_info);
2082 list_del_init(&block_group->bg_list);
2086 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2089 * We use try_to_writeback_inodes_sb() here because if we used
2090 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2091 * Currently are holding the fs freeze lock, if we do an async flush
2092 * we'll do btrfs_join_transaction() and deadlock because we need to
2093 * wait for the fs freeze lock. Using the direct flushing we benefit
2094 * from already being in a transaction and our join_transaction doesn't
2095 * have to re-take the fs freeze lock.
2097 * Note that try_to_writeback_inodes_sb() will only trigger writeback
2098 * if it can read lock sb->s_umount. It will always be able to lock it,
2099 * except when the filesystem is being unmounted or being frozen, but in
2100 * those cases sync_filesystem() is called, which results in calling
2101 * writeback_inodes_sb() while holding a write lock on sb->s_umount.
2102 * Note that we don't call writeback_inodes_sb() directly, because it
2103 * will emit a warning if sb->s_umount is not locked.
2105 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2106 try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2110 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2112 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2113 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2117 * Add a pending snapshot associated with the given transaction handle to the
2118 * respective handle. This must be called after the transaction commit started
2119 * and while holding fs_info->trans_lock.
2120 * This serves to guarantee a caller of btrfs_commit_transaction() that it can
2121 * safely free the pending snapshot pointer in case btrfs_commit_transaction()
2124 static void add_pending_snapshot(struct btrfs_trans_handle *trans)
2126 struct btrfs_transaction *cur_trans = trans->transaction;
2128 if (!trans->pending_snapshot)
2131 lockdep_assert_held(&trans->fs_info->trans_lock);
2132 ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_PREP);
2134 list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots);
2137 static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval)
2139 fs_info->commit_stats.commit_count++;
2140 fs_info->commit_stats.last_commit_dur = interval;
2141 fs_info->commit_stats.max_commit_dur =
2142 max_t(u64, fs_info->commit_stats.max_commit_dur, interval);
2143 fs_info->commit_stats.total_commit_dur += interval;
2146 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2148 struct btrfs_fs_info *fs_info = trans->fs_info;
2149 struct btrfs_transaction *cur_trans = trans->transaction;
2150 struct btrfs_transaction *prev_trans = NULL;
2155 ASSERT(refcount_read(&trans->use_count) == 1);
2156 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2158 clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags);
2160 /* Stop the commit early if ->aborted is set */
2161 if (TRANS_ABORTED(cur_trans)) {
2162 ret = cur_trans->aborted;
2163 goto lockdep_trans_commit_start_release;
2166 btrfs_trans_release_metadata(trans);
2167 trans->block_rsv = NULL;
2170 * We only want one transaction commit doing the flushing so we do not
2171 * waste a bunch of time on lock contention on the extent root node.
2173 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2174 &cur_trans->delayed_refs.flags)) {
2176 * Make a pass through all the delayed refs we have so far.
2177 * Any running threads may add more while we are here.
2179 ret = btrfs_run_delayed_refs(trans, 0);
2181 goto lockdep_trans_commit_start_release;
2184 btrfs_create_pending_block_groups(trans);
2186 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2189 /* this mutex is also taken before trying to set
2190 * block groups readonly. We need to make sure
2191 * that nobody has set a block group readonly
2192 * after a extents from that block group have been
2193 * allocated for cache files. btrfs_set_block_group_ro
2194 * will wait for the transaction to commit if it
2195 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2197 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2198 * only one process starts all the block group IO. It wouldn't
2199 * hurt to have more than one go through, but there's no
2200 * real advantage to it either.
2202 mutex_lock(&fs_info->ro_block_group_mutex);
2203 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2206 mutex_unlock(&fs_info->ro_block_group_mutex);
2209 ret = btrfs_start_dirty_block_groups(trans);
2211 goto lockdep_trans_commit_start_release;
2215 spin_lock(&fs_info->trans_lock);
2216 if (cur_trans->state >= TRANS_STATE_COMMIT_PREP) {
2217 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2219 add_pending_snapshot(trans);
2221 spin_unlock(&fs_info->trans_lock);
2222 refcount_inc(&cur_trans->use_count);
2224 if (trans->in_fsync)
2225 want_state = TRANS_STATE_SUPER_COMMITTED;
2227 btrfs_trans_state_lockdep_release(fs_info,
2228 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2229 ret = btrfs_end_transaction(trans);
2230 wait_for_commit(cur_trans, want_state);
2232 if (TRANS_ABORTED(cur_trans))
2233 ret = cur_trans->aborted;
2235 btrfs_put_transaction(cur_trans);
2240 cur_trans->state = TRANS_STATE_COMMIT_PREP;
2241 wake_up(&fs_info->transaction_blocked_wait);
2242 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2244 if (cur_trans->list.prev != &fs_info->trans_list) {
2245 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2247 if (trans->in_fsync)
2248 want_state = TRANS_STATE_SUPER_COMMITTED;
2250 prev_trans = list_entry(cur_trans->list.prev,
2251 struct btrfs_transaction, list);
2252 if (prev_trans->state < want_state) {
2253 refcount_inc(&prev_trans->use_count);
2254 spin_unlock(&fs_info->trans_lock);
2256 wait_for_commit(prev_trans, want_state);
2258 ret = READ_ONCE(prev_trans->aborted);
2260 btrfs_put_transaction(prev_trans);
2262 goto lockdep_release;
2263 spin_lock(&fs_info->trans_lock);
2267 * The previous transaction was aborted and was already removed
2268 * from the list of transactions at fs_info->trans_list. So we
2269 * abort to prevent writing a new superblock that reflects a
2270 * corrupt state (pointing to trees with unwritten nodes/leafs).
2272 if (BTRFS_FS_ERROR(fs_info)) {
2273 spin_unlock(&fs_info->trans_lock);
2275 goto lockdep_release;
2279 cur_trans->state = TRANS_STATE_COMMIT_START;
2280 wake_up(&fs_info->transaction_blocked_wait);
2281 spin_unlock(&fs_info->trans_lock);
2284 * Get the time spent on the work done by the commit thread and not
2285 * the time spent waiting on a previous commit
2287 start_time = ktime_get_ns();
2289 extwriter_counter_dec(cur_trans, trans->type);
2291 ret = btrfs_start_delalloc_flush(fs_info);
2293 goto lockdep_release;
2295 ret = btrfs_run_delayed_items(trans);
2297 goto lockdep_release;
2300 * The thread has started/joined the transaction thus it holds the
2301 * lockdep map as a reader. It has to release it before acquiring the
2302 * lockdep map as a writer.
2304 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2305 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters);
2306 wait_event(cur_trans->writer_wait,
2307 extwriter_counter_read(cur_trans) == 0);
2309 /* some pending stuffs might be added after the previous flush. */
2310 ret = btrfs_run_delayed_items(trans);
2312 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2313 goto cleanup_transaction;
2316 btrfs_wait_delalloc_flush(fs_info);
2319 * Wait for all ordered extents started by a fast fsync that joined this
2320 * transaction. Otherwise if this transaction commits before the ordered
2321 * extents complete we lose logged data after a power failure.
2323 btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered);
2324 wait_event(cur_trans->pending_wait,
2325 atomic_read(&cur_trans->pending_ordered) == 0);
2327 btrfs_scrub_pause(fs_info);
2329 * Ok now we need to make sure to block out any other joins while we
2330 * commit the transaction. We could have started a join before setting
2331 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2333 spin_lock(&fs_info->trans_lock);
2334 add_pending_snapshot(trans);
2335 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2336 spin_unlock(&fs_info->trans_lock);
2339 * The thread has started/joined the transaction thus it holds the
2340 * lockdep map as a reader. It has to release it before acquiring the
2341 * lockdep map as a writer.
2343 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2344 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2345 wait_event(cur_trans->writer_wait,
2346 atomic_read(&cur_trans->num_writers) == 1);
2349 * Make lockdep happy by acquiring the state locks after
2350 * btrfs_trans_num_writers is released. If we acquired the state locks
2351 * before releasing the btrfs_trans_num_writers lock then lockdep would
2352 * complain because we did not follow the reverse order unlocking rule.
2354 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2355 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2356 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2359 * We've started the commit, clear the flag in case we were triggered to
2360 * do an async commit but somebody else started before the transaction
2361 * kthread could do the work.
2363 clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
2365 if (TRANS_ABORTED(cur_trans)) {
2366 ret = cur_trans->aborted;
2367 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2368 goto scrub_continue;
2371 * the reloc mutex makes sure that we stop
2372 * the balancing code from coming in and moving
2373 * extents around in the middle of the commit
2375 mutex_lock(&fs_info->reloc_mutex);
2378 * We needn't worry about the delayed items because we will
2379 * deal with them in create_pending_snapshot(), which is the
2380 * core function of the snapshot creation.
2382 ret = create_pending_snapshots(trans);
2387 * We insert the dir indexes of the snapshots and update the inode
2388 * of the snapshots' parents after the snapshot creation, so there
2389 * are some delayed items which are not dealt with. Now deal with
2392 * We needn't worry that this operation will corrupt the snapshots,
2393 * because all the tree which are snapshoted will be forced to COW
2394 * the nodes and leaves.
2396 ret = btrfs_run_delayed_items(trans);
2400 ret = btrfs_run_delayed_refs(trans, U64_MAX);
2405 * make sure none of the code above managed to slip in a
2408 btrfs_assert_delayed_root_empty(fs_info);
2410 WARN_ON(cur_trans != trans->transaction);
2412 ret = commit_fs_roots(trans);
2416 /* commit_fs_roots gets rid of all the tree log roots, it is now
2417 * safe to free the root of tree log roots
2419 btrfs_free_log_root_tree(trans, fs_info);
2422 * Since fs roots are all committed, we can get a quite accurate
2423 * new_roots. So let's do quota accounting.
2425 ret = btrfs_qgroup_account_extents(trans);
2429 ret = commit_cowonly_roots(trans);
2434 * The tasks which save the space cache and inode cache may also
2435 * update ->aborted, check it.
2437 if (TRANS_ABORTED(cur_trans)) {
2438 ret = cur_trans->aborted;
2442 cur_trans = fs_info->running_transaction;
2444 btrfs_set_root_node(&fs_info->tree_root->root_item,
2445 fs_info->tree_root->node);
2446 list_add_tail(&fs_info->tree_root->dirty_list,
2447 &cur_trans->switch_commits);
2449 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2450 fs_info->chunk_root->node);
2451 list_add_tail(&fs_info->chunk_root->dirty_list,
2452 &cur_trans->switch_commits);
2454 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2455 btrfs_set_root_node(&fs_info->block_group_root->root_item,
2456 fs_info->block_group_root->node);
2457 list_add_tail(&fs_info->block_group_root->dirty_list,
2458 &cur_trans->switch_commits);
2461 switch_commit_roots(trans);
2463 ASSERT(list_empty(&cur_trans->dirty_bgs));
2464 ASSERT(list_empty(&cur_trans->io_bgs));
2465 update_super_roots(fs_info);
2467 btrfs_set_super_log_root(fs_info->super_copy, 0);
2468 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2469 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2470 sizeof(*fs_info->super_copy));
2472 btrfs_commit_device_sizes(cur_trans);
2474 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2475 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2477 btrfs_trans_release_chunk_metadata(trans);
2480 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and
2481 * setting fs_info->running_transaction to NULL, lock tree_log_mutex to
2482 * make sure that before we commit our superblock, no other task can
2483 * start a new transaction and commit a log tree before we commit our
2484 * superblock. Anyone trying to commit a log tree locks this mutex before
2485 * writing its superblock.
2487 mutex_lock(&fs_info->tree_log_mutex);
2489 spin_lock(&fs_info->trans_lock);
2490 cur_trans->state = TRANS_STATE_UNBLOCKED;
2491 fs_info->running_transaction = NULL;
2492 spin_unlock(&fs_info->trans_lock);
2493 mutex_unlock(&fs_info->reloc_mutex);
2495 wake_up(&fs_info->transaction_wait);
2496 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2498 /* If we have features changed, wake up the cleaner to update sysfs. */
2499 if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) &&
2500 fs_info->cleaner_kthread)
2501 wake_up_process(fs_info->cleaner_kthread);
2503 ret = btrfs_write_and_wait_transaction(trans);
2505 btrfs_handle_fs_error(fs_info, ret,
2506 "Error while writing out transaction");
2507 mutex_unlock(&fs_info->tree_log_mutex);
2508 goto scrub_continue;
2511 ret = write_all_supers(fs_info, 0);
2513 * the super is written, we can safely allow the tree-loggers
2514 * to go about their business
2516 mutex_unlock(&fs_info->tree_log_mutex);
2518 goto scrub_continue;
2521 * We needn't acquire the lock here because there is no other task
2522 * which can change it.
2524 cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2525 wake_up(&cur_trans->commit_wait);
2526 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2528 btrfs_finish_extent_commit(trans);
2530 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2531 btrfs_clear_space_info_full(fs_info);
2533 btrfs_set_last_trans_committed(fs_info, cur_trans->transid);
2535 * We needn't acquire the lock here because there is no other task
2536 * which can change it.
2538 cur_trans->state = TRANS_STATE_COMPLETED;
2539 wake_up(&cur_trans->commit_wait);
2540 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2542 spin_lock(&fs_info->trans_lock);
2543 list_del_init(&cur_trans->list);
2544 spin_unlock(&fs_info->trans_lock);
2546 btrfs_put_transaction(cur_trans);
2547 btrfs_put_transaction(cur_trans);
2549 if (trans->type & __TRANS_FREEZABLE)
2550 sb_end_intwrite(fs_info->sb);
2552 trace_btrfs_transaction_commit(fs_info);
2554 interval = ktime_get_ns() - start_time;
2556 btrfs_scrub_continue(fs_info);
2558 if (current->journal_info == trans)
2559 current->journal_info = NULL;
2561 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2563 update_commit_stats(fs_info, interval);
2568 mutex_unlock(&fs_info->reloc_mutex);
2569 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2571 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2572 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2573 btrfs_scrub_continue(fs_info);
2574 cleanup_transaction:
2575 btrfs_trans_release_metadata(trans);
2576 btrfs_cleanup_pending_block_groups(trans);
2577 btrfs_trans_release_chunk_metadata(trans);
2578 trans->block_rsv = NULL;
2579 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2580 if (current->journal_info == trans)
2581 current->journal_info = NULL;
2582 cleanup_transaction(trans, ret);
2587 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2588 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2589 goto cleanup_transaction;
2591 lockdep_trans_commit_start_release:
2592 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2593 btrfs_end_transaction(trans);
2598 * return < 0 if error
2599 * 0 if there are no more dead_roots at the time of call
2600 * 1 there are more to be processed, call me again
2602 * The return value indicates there are certainly more snapshots to delete, but
2603 * if there comes a new one during processing, it may return 0. We don't mind,
2604 * because btrfs_commit_super will poke cleaner thread and it will process it a
2605 * few seconds later.
2607 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
2609 struct btrfs_root *root;
2612 spin_lock(&fs_info->trans_lock);
2613 if (list_empty(&fs_info->dead_roots)) {
2614 spin_unlock(&fs_info->trans_lock);
2617 root = list_first_entry(&fs_info->dead_roots,
2618 struct btrfs_root, root_list);
2619 list_del_init(&root->root_list);
2620 spin_unlock(&fs_info->trans_lock);
2622 btrfs_debug(fs_info, "cleaner removing %llu", btrfs_root_id(root));
2624 btrfs_kill_all_delayed_nodes(root);
2626 if (btrfs_header_backref_rev(root->node) <
2627 BTRFS_MIXED_BACKREF_REV)
2628 ret = btrfs_drop_snapshot(root, 0, 0);
2630 ret = btrfs_drop_snapshot(root, 1, 0);
2632 btrfs_put_root(root);
2633 return (ret < 0) ? 0 : 1;
2637 * We only mark the transaction aborted and then set the file system read-only.
2638 * This will prevent new transactions from starting or trying to join this
2641 * This means that error recovery at the call site is limited to freeing
2642 * any local memory allocations and passing the error code up without
2643 * further cleanup. The transaction should complete as it normally would
2644 * in the call path but will return -EIO.
2646 * We'll complete the cleanup in btrfs_end_transaction and
2647 * btrfs_commit_transaction.
2649 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
2650 const char *function,
2651 unsigned int line, int error, bool first_hit)
2653 struct btrfs_fs_info *fs_info = trans->fs_info;
2655 WRITE_ONCE(trans->aborted, error);
2656 WRITE_ONCE(trans->transaction->aborted, error);
2657 if (first_hit && error == -ENOSPC)
2658 btrfs_dump_space_info_for_trans_abort(fs_info);
2659 /* Wake up anybody who may be waiting on this transaction */
2660 wake_up(&fs_info->transaction_wait);
2661 wake_up(&fs_info->transaction_blocked_wait);
2662 __btrfs_handle_fs_error(fs_info, function, line, error, NULL);
2665 int __init btrfs_transaction_init(void)
2667 btrfs_trans_handle_cachep = KMEM_CACHE(btrfs_trans_handle, SLAB_TEMPORARY);
2668 if (!btrfs_trans_handle_cachep)
2673 void __cold btrfs_transaction_exit(void)
2675 kmem_cache_destroy(btrfs_trans_handle_cachep);