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"
28 #include "accessors.h"
29 #include "extent-tree.h"
30 #include "root-tree.h"
33 #include "uuid-tree.h"
35 #include "relocation.h"
38 static struct kmem_cache *btrfs_trans_handle_cachep;
40 #define BTRFS_ROOT_TRANS_TAG 0
43 * Transaction states and transitions
45 * No running transaction (fs tree blocks are not modified)
48 * | Call start_transaction() variants. Except btrfs_join_transaction_nostart().
50 * Transaction N [[TRANS_STATE_RUNNING]]
52 * | New trans handles can be attached to transaction N by calling all
53 * | start_transaction() variants.
56 * | Call btrfs_commit_transaction() on any trans handle attached to
59 * Transaction N [[TRANS_STATE_COMMIT_START]]
61 * | Will wait for previous running transaction to completely finish if there
64 * | Then one of the following happes:
65 * | - Wait for all other trans handle holders to release.
66 * | The btrfs_commit_transaction() caller will do the commit work.
67 * | - Wait for current transaction to be committed by others.
68 * | Other btrfs_commit_transaction() caller will do the commit work.
70 * | At this stage, only btrfs_join_transaction*() variants can attach
71 * | to this running transaction.
72 * | All other variants will wait for current one to finish and attach to
76 * | Caller is chosen to commit transaction N, and all other trans handle
77 * | haven been released.
79 * Transaction N [[TRANS_STATE_COMMIT_DOING]]
81 * | The heavy lifting transaction work is started.
82 * | From running delayed refs (modifying extent tree) to creating pending
83 * | snapshots, running qgroups.
84 * | In short, modify supporting trees to reflect modifications of subvolume
87 * | At this stage, all start_transaction() calls will wait for this
88 * | transaction to finish and attach to transaction N+1.
91 * | Until all supporting trees are updated.
93 * Transaction N [[TRANS_STATE_UNBLOCKED]]
95 * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
96 * | need to write them back to disk and update |
99 * | At this stage, new transaction is allowed to |
101 * | All new start_transaction() calls will be |
102 * | attached to transid N+1. |
105 * | Until all tree blocks are super blocks are |
106 * | written to block devices |
108 * Transaction N [[TRANS_STATE_COMPLETED]] V
109 * All tree blocks and super blocks are written. Transaction N+1
110 * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]]
111 * data structures will be cleaned up. | Life goes on
113 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
114 [TRANS_STATE_RUNNING] = 0U,
115 [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH),
116 [TRANS_STATE_COMMIT_DOING] = (__TRANS_START |
119 __TRANS_JOIN_NOSTART),
120 [TRANS_STATE_UNBLOCKED] = (__TRANS_START |
123 __TRANS_JOIN_NOLOCK |
124 __TRANS_JOIN_NOSTART),
125 [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START |
128 __TRANS_JOIN_NOLOCK |
129 __TRANS_JOIN_NOSTART),
130 [TRANS_STATE_COMPLETED] = (__TRANS_START |
133 __TRANS_JOIN_NOLOCK |
134 __TRANS_JOIN_NOSTART),
137 void btrfs_put_transaction(struct btrfs_transaction *transaction)
139 WARN_ON(refcount_read(&transaction->use_count) == 0);
140 if (refcount_dec_and_test(&transaction->use_count)) {
141 BUG_ON(!list_empty(&transaction->list));
142 WARN_ON(!RB_EMPTY_ROOT(
143 &transaction->delayed_refs.href_root.rb_root));
144 WARN_ON(!RB_EMPTY_ROOT(
145 &transaction->delayed_refs.dirty_extent_root));
146 if (transaction->delayed_refs.pending_csums)
147 btrfs_err(transaction->fs_info,
148 "pending csums is %llu",
149 transaction->delayed_refs.pending_csums);
151 * If any block groups are found in ->deleted_bgs then it's
152 * because the transaction was aborted and a commit did not
153 * happen (things failed before writing the new superblock
154 * and calling btrfs_finish_extent_commit()), so we can not
155 * discard the physical locations of the block groups.
157 while (!list_empty(&transaction->deleted_bgs)) {
158 struct btrfs_block_group *cache;
160 cache = list_first_entry(&transaction->deleted_bgs,
161 struct btrfs_block_group,
163 list_del_init(&cache->bg_list);
164 btrfs_unfreeze_block_group(cache);
165 btrfs_put_block_group(cache);
167 WARN_ON(!list_empty(&transaction->dev_update_list));
172 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
174 struct btrfs_transaction *cur_trans = trans->transaction;
175 struct btrfs_fs_info *fs_info = trans->fs_info;
176 struct btrfs_root *root, *tmp;
179 * At this point no one can be using this transaction to modify any tree
180 * and no one can start another transaction to modify any tree either.
182 ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING);
184 down_write(&fs_info->commit_root_sem);
186 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
187 fs_info->last_reloc_trans = trans->transid;
189 list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
191 list_del_init(&root->dirty_list);
192 free_extent_buffer(root->commit_root);
193 root->commit_root = btrfs_root_node(root);
194 extent_io_tree_release(&root->dirty_log_pages);
195 btrfs_qgroup_clean_swapped_blocks(root);
198 /* We can free old roots now. */
199 spin_lock(&cur_trans->dropped_roots_lock);
200 while (!list_empty(&cur_trans->dropped_roots)) {
201 root = list_first_entry(&cur_trans->dropped_roots,
202 struct btrfs_root, root_list);
203 list_del_init(&root->root_list);
204 spin_unlock(&cur_trans->dropped_roots_lock);
205 btrfs_free_log(trans, root);
206 btrfs_drop_and_free_fs_root(fs_info, root);
207 spin_lock(&cur_trans->dropped_roots_lock);
209 spin_unlock(&cur_trans->dropped_roots_lock);
211 up_write(&fs_info->commit_root_sem);
214 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
217 if (type & TRANS_EXTWRITERS)
218 atomic_inc(&trans->num_extwriters);
221 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
224 if (type & TRANS_EXTWRITERS)
225 atomic_dec(&trans->num_extwriters);
228 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
231 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
234 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
236 return atomic_read(&trans->num_extwriters);
240 * To be called after doing the chunk btree updates right after allocating a new
241 * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
242 * chunk after all chunk btree updates and after finishing the second phase of
243 * chunk allocation (btrfs_create_pending_block_groups()) in case some block
244 * group had its chunk item insertion delayed to the second phase.
246 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
248 struct btrfs_fs_info *fs_info = trans->fs_info;
250 if (!trans->chunk_bytes_reserved)
253 btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
254 trans->chunk_bytes_reserved, NULL);
255 trans->chunk_bytes_reserved = 0;
259 * either allocate a new transaction or hop into the existing one
261 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
264 struct btrfs_transaction *cur_trans;
266 spin_lock(&fs_info->trans_lock);
268 /* The file system has been taken offline. No new transactions. */
269 if (BTRFS_FS_ERROR(fs_info)) {
270 spin_unlock(&fs_info->trans_lock);
274 cur_trans = fs_info->running_transaction;
276 if (TRANS_ABORTED(cur_trans)) {
277 spin_unlock(&fs_info->trans_lock);
278 return cur_trans->aborted;
280 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
281 spin_unlock(&fs_info->trans_lock);
284 refcount_inc(&cur_trans->use_count);
285 atomic_inc(&cur_trans->num_writers);
286 extwriter_counter_inc(cur_trans, type);
287 spin_unlock(&fs_info->trans_lock);
288 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
289 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
292 spin_unlock(&fs_info->trans_lock);
295 * If we are ATTACH, we just want to catch the current transaction,
296 * and commit it. If there is no transaction, just return ENOENT.
298 if (type == TRANS_ATTACH)
302 * JOIN_NOLOCK only happens during the transaction commit, so
303 * it is impossible that ->running_transaction is NULL
305 BUG_ON(type == TRANS_JOIN_NOLOCK);
307 cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
311 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers);
312 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters);
314 spin_lock(&fs_info->trans_lock);
315 if (fs_info->running_transaction) {
317 * someone started a transaction after we unlocked. Make sure
318 * to redo the checks above
320 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
321 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
324 } else if (BTRFS_FS_ERROR(fs_info)) {
325 spin_unlock(&fs_info->trans_lock);
326 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
327 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
332 cur_trans->fs_info = fs_info;
333 atomic_set(&cur_trans->pending_ordered, 0);
334 init_waitqueue_head(&cur_trans->pending_wait);
335 atomic_set(&cur_trans->num_writers, 1);
336 extwriter_counter_init(cur_trans, type);
337 init_waitqueue_head(&cur_trans->writer_wait);
338 init_waitqueue_head(&cur_trans->commit_wait);
339 cur_trans->state = TRANS_STATE_RUNNING;
341 * One for this trans handle, one so it will live on until we
342 * commit the transaction.
344 refcount_set(&cur_trans->use_count, 2);
345 cur_trans->flags = 0;
346 cur_trans->start_time = ktime_get_seconds();
348 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
350 cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
351 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
352 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
355 * although the tree mod log is per file system and not per transaction,
356 * the log must never go across transaction boundaries.
359 if (!list_empty(&fs_info->tree_mod_seq_list))
360 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
361 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
362 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
363 atomic64_set(&fs_info->tree_mod_seq, 0);
365 spin_lock_init(&cur_trans->delayed_refs.lock);
367 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
368 INIT_LIST_HEAD(&cur_trans->dev_update_list);
369 INIT_LIST_HEAD(&cur_trans->switch_commits);
370 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
371 INIT_LIST_HEAD(&cur_trans->io_bgs);
372 INIT_LIST_HEAD(&cur_trans->dropped_roots);
373 mutex_init(&cur_trans->cache_write_mutex);
374 spin_lock_init(&cur_trans->dirty_bgs_lock);
375 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
376 spin_lock_init(&cur_trans->dropped_roots_lock);
377 list_add_tail(&cur_trans->list, &fs_info->trans_list);
378 extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
379 IO_TREE_TRANS_DIRTY_PAGES);
380 extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
381 IO_TREE_FS_PINNED_EXTENTS);
382 fs_info->generation++;
383 cur_trans->transid = fs_info->generation;
384 fs_info->running_transaction = cur_trans;
385 cur_trans->aborted = 0;
386 spin_unlock(&fs_info->trans_lock);
392 * This does all the record keeping required to make sure that a shareable root
393 * is properly recorded in a given transaction. This is required to make sure
394 * the old root from before we joined the transaction is deleted when the
395 * transaction commits.
397 static int record_root_in_trans(struct btrfs_trans_handle *trans,
398 struct btrfs_root *root,
401 struct btrfs_fs_info *fs_info = root->fs_info;
404 if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
405 root->last_trans < trans->transid) || force) {
406 WARN_ON(!force && root->commit_root != root->node);
409 * see below for IN_TRANS_SETUP usage rules
410 * we have the reloc mutex held now, so there
411 * is only one writer in this function
413 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
415 /* make sure readers find IN_TRANS_SETUP before
416 * they find our root->last_trans update
420 spin_lock(&fs_info->fs_roots_radix_lock);
421 if (root->last_trans == trans->transid && !force) {
422 spin_unlock(&fs_info->fs_roots_radix_lock);
425 radix_tree_tag_set(&fs_info->fs_roots_radix,
426 (unsigned long)root->root_key.objectid,
427 BTRFS_ROOT_TRANS_TAG);
428 spin_unlock(&fs_info->fs_roots_radix_lock);
429 root->last_trans = trans->transid;
431 /* this is pretty tricky. We don't want to
432 * take the relocation lock in btrfs_record_root_in_trans
433 * unless we're really doing the first setup for this root in
436 * Normally we'd use root->last_trans as a flag to decide
437 * if we want to take the expensive mutex.
439 * But, we have to set root->last_trans before we
440 * init the relocation root, otherwise, we trip over warnings
441 * in ctree.c. The solution used here is to flag ourselves
442 * with root IN_TRANS_SETUP. When this is 1, we're still
443 * fixing up the reloc trees and everyone must wait.
445 * When this is zero, they can trust root->last_trans and fly
446 * through btrfs_record_root_in_trans without having to take the
447 * lock. smp_wmb() makes sure that all the writes above are
448 * done before we pop in the zero below
450 ret = btrfs_init_reloc_root(trans, root);
451 smp_mb__before_atomic();
452 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
458 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
459 struct btrfs_root *root)
461 struct btrfs_fs_info *fs_info = root->fs_info;
462 struct btrfs_transaction *cur_trans = trans->transaction;
464 /* Add ourselves to the transaction dropped list */
465 spin_lock(&cur_trans->dropped_roots_lock);
466 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
467 spin_unlock(&cur_trans->dropped_roots_lock);
469 /* Make sure we don't try to update the root at commit time */
470 spin_lock(&fs_info->fs_roots_radix_lock);
471 radix_tree_tag_clear(&fs_info->fs_roots_radix,
472 (unsigned long)root->root_key.objectid,
473 BTRFS_ROOT_TRANS_TAG);
474 spin_unlock(&fs_info->fs_roots_radix_lock);
477 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
478 struct btrfs_root *root)
480 struct btrfs_fs_info *fs_info = root->fs_info;
483 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
487 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
491 if (root->last_trans == trans->transid &&
492 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
495 mutex_lock(&fs_info->reloc_mutex);
496 ret = record_root_in_trans(trans, root, 0);
497 mutex_unlock(&fs_info->reloc_mutex);
502 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
504 return (trans->state >= TRANS_STATE_COMMIT_START &&
505 trans->state < TRANS_STATE_UNBLOCKED &&
506 !TRANS_ABORTED(trans));
509 /* wait for commit against the current transaction to become unblocked
510 * when this is done, it is safe to start a new transaction, but the current
511 * transaction might not be fully on disk.
513 static void wait_current_trans(struct btrfs_fs_info *fs_info)
515 struct btrfs_transaction *cur_trans;
517 spin_lock(&fs_info->trans_lock);
518 cur_trans = fs_info->running_transaction;
519 if (cur_trans && is_transaction_blocked(cur_trans)) {
520 refcount_inc(&cur_trans->use_count);
521 spin_unlock(&fs_info->trans_lock);
523 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
524 wait_event(fs_info->transaction_wait,
525 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
526 TRANS_ABORTED(cur_trans));
527 btrfs_put_transaction(cur_trans);
529 spin_unlock(&fs_info->trans_lock);
533 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
535 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
538 if (type == TRANS_START)
544 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
546 struct btrfs_fs_info *fs_info = root->fs_info;
548 if (!fs_info->reloc_ctl ||
549 !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
550 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
557 static struct btrfs_trans_handle *
558 start_transaction(struct btrfs_root *root, unsigned int num_items,
559 unsigned int type, enum btrfs_reserve_flush_enum flush,
560 bool enforce_qgroups)
562 struct btrfs_fs_info *fs_info = root->fs_info;
563 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
564 struct btrfs_trans_handle *h;
565 struct btrfs_transaction *cur_trans;
567 u64 qgroup_reserved = 0;
568 bool reloc_reserved = false;
569 bool do_chunk_alloc = false;
572 if (BTRFS_FS_ERROR(fs_info))
573 return ERR_PTR(-EROFS);
575 if (current->journal_info) {
576 WARN_ON(type & TRANS_EXTWRITERS);
577 h = current->journal_info;
578 refcount_inc(&h->use_count);
579 WARN_ON(refcount_read(&h->use_count) > 2);
580 h->orig_rsv = h->block_rsv;
586 * Do the reservation before we join the transaction so we can do all
587 * the appropriate flushing if need be.
589 if (num_items && root != fs_info->chunk_root) {
590 struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
591 u64 delayed_refs_bytes = 0;
593 qgroup_reserved = num_items * fs_info->nodesize;
594 ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
600 * We want to reserve all the bytes we may need all at once, so
601 * we only do 1 enospc flushing cycle per transaction start. We
602 * accomplish this by simply assuming we'll do num_items worth
603 * of delayed refs updates in this trans handle, and refill that
604 * amount for whatever is missing in the reserve.
606 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
607 if (flush == BTRFS_RESERVE_FLUSH_ALL &&
608 !btrfs_block_rsv_full(delayed_refs_rsv)) {
609 delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info,
611 num_bytes += delayed_refs_bytes;
615 * Do the reservation for the relocation root creation
617 if (need_reserve_reloc_root(root)) {
618 num_bytes += fs_info->nodesize;
619 reloc_reserved = true;
622 ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush);
625 if (delayed_refs_bytes) {
626 btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
628 num_bytes -= delayed_refs_bytes;
631 if (rsv->space_info->force_alloc)
632 do_chunk_alloc = true;
633 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
634 !btrfs_block_rsv_full(delayed_refs_rsv)) {
636 * Some people call with btrfs_start_transaction(root, 0)
637 * because they can be throttled, but have some other mechanism
638 * for reserving space. We still want these guys to refill the
639 * delayed block_rsv so just add 1 items worth of reservation
642 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
647 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
654 * If we are JOIN_NOLOCK we're already committing a transaction and
655 * waiting on this guy, so we don't need to do the sb_start_intwrite
656 * because we're already holding a ref. We need this because we could
657 * have raced in and did an fsync() on a file which can kick a commit
658 * and then we deadlock with somebody doing a freeze.
660 * If we are ATTACH, it means we just want to catch the current
661 * transaction and commit it, so we needn't do sb_start_intwrite().
663 if (type & __TRANS_FREEZABLE)
664 sb_start_intwrite(fs_info->sb);
666 if (may_wait_transaction(fs_info, type))
667 wait_current_trans(fs_info);
670 ret = join_transaction(fs_info, type);
672 wait_current_trans(fs_info);
673 if (unlikely(type == TRANS_ATTACH ||
674 type == TRANS_JOIN_NOSTART))
677 } while (ret == -EBUSY);
682 cur_trans = fs_info->running_transaction;
684 h->transid = cur_trans->transid;
685 h->transaction = cur_trans;
686 refcount_set(&h->use_count, 1);
687 h->fs_info = root->fs_info;
690 INIT_LIST_HEAD(&h->new_bgs);
693 if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
694 may_wait_transaction(fs_info, type)) {
695 current->journal_info = h;
696 btrfs_commit_transaction(h);
701 trace_btrfs_space_reservation(fs_info, "transaction",
702 h->transid, num_bytes, 1);
703 h->block_rsv = &fs_info->trans_block_rsv;
704 h->bytes_reserved = num_bytes;
705 h->reloc_reserved = reloc_reserved;
709 if (!current->journal_info)
710 current->journal_info = h;
713 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
714 * ALLOC_FORCE the first run through, and then we won't allocate for
715 * anybody else who races in later. We don't care about the return
718 if (do_chunk_alloc && num_bytes) {
719 u64 flags = h->block_rsv->space_info->flags;
721 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
722 CHUNK_ALLOC_NO_FORCE);
726 * btrfs_record_root_in_trans() needs to alloc new extents, and may
727 * call btrfs_join_transaction() while we're also starting a
730 * Thus it need to be called after current->journal_info initialized,
731 * or we can deadlock.
733 ret = btrfs_record_root_in_trans(h, root);
736 * The transaction handle is fully initialized and linked with
737 * other structures so it needs to be ended in case of errors,
740 btrfs_end_transaction(h);
747 if (type & __TRANS_FREEZABLE)
748 sb_end_intwrite(fs_info->sb);
749 kmem_cache_free(btrfs_trans_handle_cachep, h);
752 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
755 btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
759 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
760 unsigned int num_items)
762 return start_transaction(root, num_items, TRANS_START,
763 BTRFS_RESERVE_FLUSH_ALL, true);
766 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
767 struct btrfs_root *root,
768 unsigned int num_items)
770 return start_transaction(root, num_items, TRANS_START,
771 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
774 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
776 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
780 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
782 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
783 BTRFS_RESERVE_NO_FLUSH, true);
787 * Similar to regular join but it never starts a transaction when none is
788 * running or after waiting for the current one to finish.
790 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
792 return start_transaction(root, 0, TRANS_JOIN_NOSTART,
793 BTRFS_RESERVE_NO_FLUSH, true);
797 * btrfs_attach_transaction() - catch the running transaction
799 * It is used when we want to commit the current the transaction, but
800 * don't want to start a new one.
802 * Note: If this function return -ENOENT, it just means there is no
803 * running transaction. But it is possible that the inactive transaction
804 * is still in the memory, not fully on disk. If you hope there is no
805 * inactive transaction in the fs when -ENOENT is returned, you should
807 * btrfs_attach_transaction_barrier()
809 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
811 return start_transaction(root, 0, TRANS_ATTACH,
812 BTRFS_RESERVE_NO_FLUSH, true);
816 * btrfs_attach_transaction_barrier() - catch the running transaction
818 * It is similar to the above function, the difference is this one
819 * will wait for all the inactive transactions until they fully
822 struct btrfs_trans_handle *
823 btrfs_attach_transaction_barrier(struct btrfs_root *root)
825 struct btrfs_trans_handle *trans;
827 trans = start_transaction(root, 0, TRANS_ATTACH,
828 BTRFS_RESERVE_NO_FLUSH, true);
829 if (trans == ERR_PTR(-ENOENT))
830 btrfs_wait_for_commit(root->fs_info, 0);
835 /* Wait for a transaction commit to reach at least the given state. */
836 static noinline void wait_for_commit(struct btrfs_transaction *commit,
837 const enum btrfs_trans_state min_state)
839 struct btrfs_fs_info *fs_info = commit->fs_info;
840 u64 transid = commit->transid;
844 * At the moment this function is called with min_state either being
845 * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED.
847 if (min_state == TRANS_STATE_COMPLETED)
848 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
850 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
853 wait_event(commit->commit_wait, commit->state >= min_state);
855 btrfs_put_transaction(commit);
857 if (min_state < TRANS_STATE_COMPLETED)
861 * A transaction isn't really completed until all of the
862 * previous transactions are completed, but with fsync we can
863 * end up with SUPER_COMMITTED transactions before a COMPLETED
864 * transaction. Wait for those.
867 spin_lock(&fs_info->trans_lock);
868 commit = list_first_entry_or_null(&fs_info->trans_list,
869 struct btrfs_transaction,
871 if (!commit || commit->transid > transid) {
872 spin_unlock(&fs_info->trans_lock);
875 refcount_inc(&commit->use_count);
877 spin_unlock(&fs_info->trans_lock);
881 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
883 struct btrfs_transaction *cur_trans = NULL, *t;
887 if (transid <= fs_info->last_trans_committed)
890 /* find specified transaction */
891 spin_lock(&fs_info->trans_lock);
892 list_for_each_entry(t, &fs_info->trans_list, list) {
893 if (t->transid == transid) {
895 refcount_inc(&cur_trans->use_count);
899 if (t->transid > transid) {
904 spin_unlock(&fs_info->trans_lock);
907 * The specified transaction doesn't exist, or we
908 * raced with btrfs_commit_transaction
911 if (transid > fs_info->last_trans_committed)
916 /* find newest transaction that is committing | committed */
917 spin_lock(&fs_info->trans_lock);
918 list_for_each_entry_reverse(t, &fs_info->trans_list,
920 if (t->state >= TRANS_STATE_COMMIT_START) {
921 if (t->state == TRANS_STATE_COMPLETED)
924 refcount_inc(&cur_trans->use_count);
928 spin_unlock(&fs_info->trans_lock);
930 goto out; /* nothing committing|committed */
933 wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
934 ret = cur_trans->aborted;
935 btrfs_put_transaction(cur_trans);
940 void btrfs_throttle(struct btrfs_fs_info *fs_info)
942 wait_current_trans(fs_info);
945 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
947 struct btrfs_transaction *cur_trans = trans->transaction;
949 if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
950 test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
953 if (btrfs_check_space_for_delayed_refs(trans->fs_info))
956 return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50);
959 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
962 struct btrfs_fs_info *fs_info = trans->fs_info;
964 if (!trans->block_rsv) {
965 ASSERT(!trans->bytes_reserved);
969 if (!trans->bytes_reserved)
972 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
973 trace_btrfs_space_reservation(fs_info, "transaction",
974 trans->transid, trans->bytes_reserved, 0);
975 btrfs_block_rsv_release(fs_info, trans->block_rsv,
976 trans->bytes_reserved, NULL);
977 trans->bytes_reserved = 0;
980 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
983 struct btrfs_fs_info *info = trans->fs_info;
984 struct btrfs_transaction *cur_trans = trans->transaction;
987 if (refcount_read(&trans->use_count) > 1) {
988 refcount_dec(&trans->use_count);
989 trans->block_rsv = trans->orig_rsv;
993 btrfs_trans_release_metadata(trans);
994 trans->block_rsv = NULL;
996 btrfs_create_pending_block_groups(trans);
998 btrfs_trans_release_chunk_metadata(trans);
1000 if (trans->type & __TRANS_FREEZABLE)
1001 sb_end_intwrite(info->sb);
1003 WARN_ON(cur_trans != info->running_transaction);
1004 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
1005 atomic_dec(&cur_trans->num_writers);
1006 extwriter_counter_dec(cur_trans, trans->type);
1008 cond_wake_up(&cur_trans->writer_wait);
1010 btrfs_lockdep_release(info, btrfs_trans_num_extwriters);
1011 btrfs_lockdep_release(info, btrfs_trans_num_writers);
1013 btrfs_put_transaction(cur_trans);
1015 if (current->journal_info == trans)
1016 current->journal_info = NULL;
1019 btrfs_run_delayed_iputs(info);
1021 if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) {
1022 wake_up_process(info->transaction_kthread);
1023 if (TRANS_ABORTED(trans))
1024 err = trans->aborted;
1029 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1033 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1035 return __btrfs_end_transaction(trans, 0);
1038 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1040 return __btrfs_end_transaction(trans, 1);
1044 * when btree blocks are allocated, they have some corresponding bits set for
1045 * them in one of two extent_io trees. This is used to make sure all of
1046 * those extents are sent to disk but does not wait on them
1048 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1049 struct extent_io_tree *dirty_pages, int mark)
1053 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1054 struct extent_state *cached_state = NULL;
1058 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1059 mark, &cached_state)) {
1060 bool wait_writeback = false;
1062 err = convert_extent_bit(dirty_pages, start, end,
1064 mark, &cached_state);
1066 * convert_extent_bit can return -ENOMEM, which is most of the
1067 * time a temporary error. So when it happens, ignore the error
1068 * and wait for writeback of this range to finish - because we
1069 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1070 * to __btrfs_wait_marked_extents() would not know that
1071 * writeback for this range started and therefore wouldn't
1072 * wait for it to finish - we don't want to commit a
1073 * superblock that points to btree nodes/leafs for which
1074 * writeback hasn't finished yet (and without errors).
1075 * We cleanup any entries left in the io tree when committing
1076 * the transaction (through extent_io_tree_release()).
1078 if (err == -ENOMEM) {
1080 wait_writeback = true;
1083 err = filemap_fdatawrite_range(mapping, start, end);
1086 else if (wait_writeback)
1087 werr = filemap_fdatawait_range(mapping, start, end);
1088 free_extent_state(cached_state);
1089 cached_state = NULL;
1097 * when btree blocks are allocated, they have some corresponding bits set for
1098 * them in one of two extent_io trees. This is used to make sure all of
1099 * those extents are on disk for transaction or log commit. We wait
1100 * on all the pages and clear them from the dirty pages state tree
1102 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1103 struct extent_io_tree *dirty_pages)
1107 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1108 struct extent_state *cached_state = NULL;
1112 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1113 EXTENT_NEED_WAIT, &cached_state)) {
1115 * Ignore -ENOMEM errors returned by clear_extent_bit().
1116 * When committing the transaction, we'll remove any entries
1117 * left in the io tree. For a log commit, we don't remove them
1118 * after committing the log because the tree can be accessed
1119 * concurrently - we do it only at transaction commit time when
1120 * it's safe to do it (through extent_io_tree_release()).
1122 err = clear_extent_bit(dirty_pages, start, end,
1123 EXTENT_NEED_WAIT, &cached_state);
1127 err = filemap_fdatawait_range(mapping, start, end);
1130 free_extent_state(cached_state);
1131 cached_state = NULL;
1140 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1141 struct extent_io_tree *dirty_pages)
1143 bool errors = false;
1146 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1147 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1155 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1157 struct btrfs_fs_info *fs_info = log_root->fs_info;
1158 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1159 bool errors = false;
1162 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1164 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1165 if ((mark & EXTENT_DIRTY) &&
1166 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1169 if ((mark & EXTENT_NEW) &&
1170 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1179 * When btree blocks are allocated the corresponding extents are marked dirty.
1180 * This function ensures such extents are persisted on disk for transaction or
1183 * @trans: transaction whose dirty pages we'd like to write
1185 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1189 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1190 struct btrfs_fs_info *fs_info = trans->fs_info;
1191 struct blk_plug plug;
1193 blk_start_plug(&plug);
1194 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1195 blk_finish_plug(&plug);
1196 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1198 extent_io_tree_release(&trans->transaction->dirty_pages);
1209 * this is used to update the root pointer in the tree of tree roots.
1211 * But, in the case of the extent allocation tree, updating the root
1212 * pointer may allocate blocks which may change the root of the extent
1215 * So, this loops and repeats and makes sure the cowonly root didn't
1216 * change while the root pointer was being updated in the metadata.
1218 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1219 struct btrfs_root *root)
1222 u64 old_root_bytenr;
1224 struct btrfs_fs_info *fs_info = root->fs_info;
1225 struct btrfs_root *tree_root = fs_info->tree_root;
1227 old_root_used = btrfs_root_used(&root->root_item);
1230 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1231 if (old_root_bytenr == root->node->start &&
1232 old_root_used == btrfs_root_used(&root->root_item))
1235 btrfs_set_root_node(&root->root_item, root->node);
1236 ret = btrfs_update_root(trans, tree_root,
1242 old_root_used = btrfs_root_used(&root->root_item);
1249 * update all the cowonly tree roots on disk
1251 * The error handling in this function may not be obvious. Any of the
1252 * failures will cause the file system to go offline. We still need
1253 * to clean up the delayed refs.
1255 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1257 struct btrfs_fs_info *fs_info = trans->fs_info;
1258 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1259 struct list_head *io_bgs = &trans->transaction->io_bgs;
1260 struct list_head *next;
1261 struct extent_buffer *eb;
1265 * At this point no one can be using this transaction to modify any tree
1266 * and no one can start another transaction to modify any tree either.
1268 ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1270 eb = btrfs_lock_root_node(fs_info->tree_root);
1271 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1272 0, &eb, BTRFS_NESTING_COW);
1273 btrfs_tree_unlock(eb);
1274 free_extent_buffer(eb);
1279 ret = btrfs_run_dev_stats(trans);
1282 ret = btrfs_run_dev_replace(trans);
1285 ret = btrfs_run_qgroups(trans);
1289 ret = btrfs_setup_space_cache(trans);
1294 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1295 struct btrfs_root *root;
1296 next = fs_info->dirty_cowonly_roots.next;
1297 list_del_init(next);
1298 root = list_entry(next, struct btrfs_root, dirty_list);
1299 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1301 list_add_tail(&root->dirty_list,
1302 &trans->transaction->switch_commits);
1303 ret = update_cowonly_root(trans, root);
1308 /* Now flush any delayed refs generated by updating all of the roots */
1309 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1313 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1314 ret = btrfs_write_dirty_block_groups(trans);
1319 * We're writing the dirty block groups, which could generate
1320 * delayed refs, which could generate more dirty block groups,
1321 * so we want to keep this flushing in this loop to make sure
1322 * everything gets run.
1324 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1329 if (!list_empty(&fs_info->dirty_cowonly_roots))
1332 /* Update dev-replace pointer once everything is committed */
1333 fs_info->dev_replace.committed_cursor_left =
1334 fs_info->dev_replace.cursor_left_last_write_of_item;
1340 * If we had a pending drop we need to see if there are any others left in our
1341 * dead roots list, and if not clear our bit and wake any waiters.
1343 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
1346 * We put the drop in progress roots at the front of the list, so if the
1347 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
1350 spin_lock(&fs_info->trans_lock);
1351 if (!list_empty(&fs_info->dead_roots)) {
1352 struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
1355 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
1356 spin_unlock(&fs_info->trans_lock);
1360 spin_unlock(&fs_info->trans_lock);
1362 btrfs_wake_unfinished_drop(fs_info);
1366 * dead roots are old snapshots that need to be deleted. This allocates
1367 * a dirty root struct and adds it into the list of dead roots that need to
1370 void btrfs_add_dead_root(struct btrfs_root *root)
1372 struct btrfs_fs_info *fs_info = root->fs_info;
1374 spin_lock(&fs_info->trans_lock);
1375 if (list_empty(&root->root_list)) {
1376 btrfs_grab_root(root);
1378 /* We want to process the partially complete drops first. */
1379 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
1380 list_add(&root->root_list, &fs_info->dead_roots);
1382 list_add_tail(&root->root_list, &fs_info->dead_roots);
1384 spin_unlock(&fs_info->trans_lock);
1388 * Update each subvolume root and its relocation root, if it exists, in the tree
1389 * of tree roots. Also free log roots if they exist.
1391 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1393 struct btrfs_fs_info *fs_info = trans->fs_info;
1394 struct btrfs_root *gang[8];
1399 * At this point no one can be using this transaction to modify any tree
1400 * and no one can start another transaction to modify any tree either.
1402 ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING);
1404 spin_lock(&fs_info->fs_roots_radix_lock);
1406 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1409 BTRFS_ROOT_TRANS_TAG);
1412 for (i = 0; i < ret; i++) {
1413 struct btrfs_root *root = gang[i];
1417 * At this point we can neither have tasks logging inodes
1418 * from a root nor trying to commit a log tree.
1420 ASSERT(atomic_read(&root->log_writers) == 0);
1421 ASSERT(atomic_read(&root->log_commit[0]) == 0);
1422 ASSERT(atomic_read(&root->log_commit[1]) == 0);
1424 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1425 (unsigned long)root->root_key.objectid,
1426 BTRFS_ROOT_TRANS_TAG);
1427 spin_unlock(&fs_info->fs_roots_radix_lock);
1429 btrfs_free_log(trans, root);
1430 ret2 = btrfs_update_reloc_root(trans, root);
1434 /* see comments in should_cow_block() */
1435 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1436 smp_mb__after_atomic();
1438 if (root->commit_root != root->node) {
1439 list_add_tail(&root->dirty_list,
1440 &trans->transaction->switch_commits);
1441 btrfs_set_root_node(&root->root_item,
1445 ret2 = btrfs_update_root(trans, fs_info->tree_root,
1450 spin_lock(&fs_info->fs_roots_radix_lock);
1451 btrfs_qgroup_free_meta_all_pertrans(root);
1454 spin_unlock(&fs_info->fs_roots_radix_lock);
1459 * defrag a given btree.
1460 * Every leaf in the btree is read and defragged.
1462 int btrfs_defrag_root(struct btrfs_root *root)
1464 struct btrfs_fs_info *info = root->fs_info;
1465 struct btrfs_trans_handle *trans;
1468 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1472 trans = btrfs_start_transaction(root, 0);
1473 if (IS_ERR(trans)) {
1474 ret = PTR_ERR(trans);
1478 ret = btrfs_defrag_leaves(trans, root);
1480 btrfs_end_transaction(trans);
1481 btrfs_btree_balance_dirty(info);
1484 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1487 if (btrfs_defrag_cancelled(info)) {
1488 btrfs_debug(info, "defrag_root cancelled");
1493 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1498 * Do all special snapshot related qgroup dirty hack.
1500 * Will do all needed qgroup inherit and dirty hack like switch commit
1501 * roots inside one transaction and write all btree into disk, to make
1504 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1505 struct btrfs_root *src,
1506 struct btrfs_root *parent,
1507 struct btrfs_qgroup_inherit *inherit,
1510 struct btrfs_fs_info *fs_info = src->fs_info;
1514 * Save some performance in the case that qgroups are not
1515 * enabled. If this check races with the ioctl, rescan will
1518 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1522 * Ensure dirty @src will be committed. Or, after coming
1523 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1524 * recorded root will never be updated again, causing an outdated root
1527 ret = record_root_in_trans(trans, src, 1);
1532 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1533 * src root, so we must run the delayed refs here.
1535 * However this isn't particularly fool proof, because there's no
1536 * synchronization keeping us from changing the tree after this point
1537 * before we do the qgroup_inherit, or even from making changes while
1538 * we're doing the qgroup_inherit. But that's a problem for the future,
1539 * for now flush the delayed refs to narrow the race window where the
1540 * qgroup counters could end up wrong.
1542 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1544 btrfs_abort_transaction(trans, ret);
1548 ret = commit_fs_roots(trans);
1551 ret = btrfs_qgroup_account_extents(trans);
1555 /* Now qgroup are all updated, we can inherit it to new qgroups */
1556 ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1562 * Now we do a simplified commit transaction, which will:
1563 * 1) commit all subvolume and extent tree
1564 * To ensure all subvolume and extent tree have a valid
1565 * commit_root to accounting later insert_dir_item()
1566 * 2) write all btree blocks onto disk
1567 * This is to make sure later btree modification will be cowed
1568 * Or commit_root can be populated and cause wrong qgroup numbers
1569 * In this simplified commit, we don't really care about other trees
1570 * like chunk and root tree, as they won't affect qgroup.
1571 * And we don't write super to avoid half committed status.
1573 ret = commit_cowonly_roots(trans);
1576 switch_commit_roots(trans);
1577 ret = btrfs_write_and_wait_transaction(trans);
1579 btrfs_handle_fs_error(fs_info, ret,
1580 "Error while writing out transaction for qgroup");
1584 * Force parent root to be updated, as we recorded it before so its
1585 * last_trans == cur_transid.
1586 * Or it won't be committed again onto disk after later
1590 ret = record_root_in_trans(trans, parent, 1);
1595 * new snapshots need to be created at a very specific time in the
1596 * transaction commit. This does the actual creation.
1599 * If the error which may affect the commitment of the current transaction
1600 * happens, we should return the error number. If the error which just affect
1601 * the creation of the pending snapshots, just return 0.
1603 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1604 struct btrfs_pending_snapshot *pending)
1607 struct btrfs_fs_info *fs_info = trans->fs_info;
1608 struct btrfs_key key;
1609 struct btrfs_root_item *new_root_item;
1610 struct btrfs_root *tree_root = fs_info->tree_root;
1611 struct btrfs_root *root = pending->root;
1612 struct btrfs_root *parent_root;
1613 struct btrfs_block_rsv *rsv;
1614 struct inode *parent_inode = pending->dir;
1615 struct btrfs_path *path;
1616 struct btrfs_dir_item *dir_item;
1617 struct extent_buffer *tmp;
1618 struct extent_buffer *old;
1619 struct timespec64 cur_time;
1625 unsigned int nofs_flags;
1626 struct fscrypt_name fname;
1628 ASSERT(pending->path);
1629 path = pending->path;
1631 ASSERT(pending->root_item);
1632 new_root_item = pending->root_item;
1635 * We're inside a transaction and must make sure that any potential
1636 * allocations with GFP_KERNEL in fscrypt won't recurse back to
1639 nofs_flags = memalloc_nofs_save();
1640 pending->error = fscrypt_setup_filename(parent_inode,
1641 &pending->dentry->d_name, 0,
1643 memalloc_nofs_restore(nofs_flags);
1647 pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1652 * Make qgroup to skip current new snapshot's qgroupid, as it is
1653 * accounted by later btrfs_qgroup_inherit().
1655 btrfs_set_skip_qgroup(trans, objectid);
1657 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1659 if (to_reserve > 0) {
1660 pending->error = btrfs_block_rsv_add(fs_info,
1661 &pending->block_rsv,
1663 BTRFS_RESERVE_NO_FLUSH);
1665 goto clear_skip_qgroup;
1668 key.objectid = objectid;
1669 key.offset = (u64)-1;
1670 key.type = BTRFS_ROOT_ITEM_KEY;
1672 rsv = trans->block_rsv;
1673 trans->block_rsv = &pending->block_rsv;
1674 trans->bytes_reserved = trans->block_rsv->reserved;
1675 trace_btrfs_space_reservation(fs_info, "transaction",
1677 trans->bytes_reserved, 1);
1678 parent_root = BTRFS_I(parent_inode)->root;
1679 ret = record_root_in_trans(trans, parent_root, 0);
1682 cur_time = current_time(parent_inode);
1685 * insert the directory item
1687 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1689 btrfs_abort_transaction(trans, ret);
1693 /* check if there is a file/dir which has the same name. */
1694 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1695 btrfs_ino(BTRFS_I(parent_inode)),
1696 &fname.disk_name, 0);
1697 if (dir_item != NULL && !IS_ERR(dir_item)) {
1698 pending->error = -EEXIST;
1699 goto dir_item_existed;
1700 } else if (IS_ERR(dir_item)) {
1701 ret = PTR_ERR(dir_item);
1702 btrfs_abort_transaction(trans, ret);
1705 btrfs_release_path(path);
1708 * pull in the delayed directory update
1709 * and the delayed inode item
1710 * otherwise we corrupt the FS during
1713 ret = btrfs_run_delayed_items(trans);
1714 if (ret) { /* Transaction aborted */
1715 btrfs_abort_transaction(trans, ret);
1719 ret = record_root_in_trans(trans, root, 0);
1721 btrfs_abort_transaction(trans, ret);
1724 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1725 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1726 btrfs_check_and_init_root_item(new_root_item);
1728 root_flags = btrfs_root_flags(new_root_item);
1729 if (pending->readonly)
1730 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1732 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1733 btrfs_set_root_flags(new_root_item, root_flags);
1735 btrfs_set_root_generation_v2(new_root_item,
1737 generate_random_guid(new_root_item->uuid);
1738 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1740 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1741 memset(new_root_item->received_uuid, 0,
1742 sizeof(new_root_item->received_uuid));
1743 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1744 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1745 btrfs_set_root_stransid(new_root_item, 0);
1746 btrfs_set_root_rtransid(new_root_item, 0);
1748 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1749 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1750 btrfs_set_root_otransid(new_root_item, trans->transid);
1752 old = btrfs_lock_root_node(root);
1753 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1756 btrfs_tree_unlock(old);
1757 free_extent_buffer(old);
1758 btrfs_abort_transaction(trans, ret);
1762 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1763 /* clean up in any case */
1764 btrfs_tree_unlock(old);
1765 free_extent_buffer(old);
1767 btrfs_abort_transaction(trans, ret);
1770 /* see comments in should_cow_block() */
1771 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1774 btrfs_set_root_node(new_root_item, tmp);
1775 /* record when the snapshot was created in key.offset */
1776 key.offset = trans->transid;
1777 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1778 btrfs_tree_unlock(tmp);
1779 free_extent_buffer(tmp);
1781 btrfs_abort_transaction(trans, ret);
1786 * insert root back/forward references
1788 ret = btrfs_add_root_ref(trans, objectid,
1789 parent_root->root_key.objectid,
1790 btrfs_ino(BTRFS_I(parent_inode)), index,
1793 btrfs_abort_transaction(trans, ret);
1797 key.offset = (u64)-1;
1798 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1799 if (IS_ERR(pending->snap)) {
1800 ret = PTR_ERR(pending->snap);
1801 pending->snap = NULL;
1802 btrfs_abort_transaction(trans, ret);
1806 ret = btrfs_reloc_post_snapshot(trans, pending);
1808 btrfs_abort_transaction(trans, ret);
1813 * Do special qgroup accounting for snapshot, as we do some qgroup
1814 * snapshot hack to do fast snapshot.
1815 * To co-operate with that hack, we do hack again.
1816 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1818 ret = qgroup_account_snapshot(trans, root, parent_root,
1819 pending->inherit, objectid);
1823 ret = btrfs_insert_dir_item(trans, &fname.disk_name,
1824 BTRFS_I(parent_inode), &key, BTRFS_FT_DIR,
1826 /* We have check then name at the beginning, so it is impossible. */
1827 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1829 btrfs_abort_transaction(trans, ret);
1833 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1834 fname.disk_name.len * 2);
1835 parent_inode->i_mtime = current_time(parent_inode);
1836 parent_inode->i_ctime = parent_inode->i_mtime;
1837 ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1839 btrfs_abort_transaction(trans, ret);
1842 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1843 BTRFS_UUID_KEY_SUBVOL,
1846 btrfs_abort_transaction(trans, ret);
1849 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1850 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1851 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1853 if (ret && ret != -EEXIST) {
1854 btrfs_abort_transaction(trans, ret);
1860 pending->error = ret;
1862 trans->block_rsv = rsv;
1863 trans->bytes_reserved = 0;
1865 btrfs_clear_skip_qgroup(trans);
1867 fscrypt_free_filename(&fname);
1869 kfree(new_root_item);
1870 pending->root_item = NULL;
1871 btrfs_free_path(path);
1872 pending->path = NULL;
1878 * create all the snapshots we've scheduled for creation
1880 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1882 struct btrfs_pending_snapshot *pending, *next;
1883 struct list_head *head = &trans->transaction->pending_snapshots;
1886 list_for_each_entry_safe(pending, next, head, list) {
1887 list_del(&pending->list);
1888 ret = create_pending_snapshot(trans, pending);
1895 static void update_super_roots(struct btrfs_fs_info *fs_info)
1897 struct btrfs_root_item *root_item;
1898 struct btrfs_super_block *super;
1900 super = fs_info->super_copy;
1902 root_item = &fs_info->chunk_root->root_item;
1903 super->chunk_root = root_item->bytenr;
1904 super->chunk_root_generation = root_item->generation;
1905 super->chunk_root_level = root_item->level;
1907 root_item = &fs_info->tree_root->root_item;
1908 super->root = root_item->bytenr;
1909 super->generation = root_item->generation;
1910 super->root_level = root_item->level;
1911 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1912 super->cache_generation = root_item->generation;
1913 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1914 super->cache_generation = 0;
1915 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1916 super->uuid_tree_generation = root_item->generation;
1919 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1921 struct btrfs_transaction *trans;
1924 spin_lock(&info->trans_lock);
1925 trans = info->running_transaction;
1927 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1928 spin_unlock(&info->trans_lock);
1932 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1934 struct btrfs_transaction *trans;
1937 spin_lock(&info->trans_lock);
1938 trans = info->running_transaction;
1940 ret = is_transaction_blocked(trans);
1941 spin_unlock(&info->trans_lock);
1945 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1947 struct btrfs_fs_info *fs_info = trans->fs_info;
1948 struct btrfs_transaction *cur_trans;
1950 /* Kick the transaction kthread. */
1951 set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
1952 wake_up_process(fs_info->transaction_kthread);
1954 /* take transaction reference */
1955 cur_trans = trans->transaction;
1956 refcount_inc(&cur_trans->use_count);
1958 btrfs_end_transaction(trans);
1961 * Wait for the current transaction commit to start and block
1962 * subsequent transaction joins
1964 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
1965 wait_event(fs_info->transaction_blocked_wait,
1966 cur_trans->state >= TRANS_STATE_COMMIT_START ||
1967 TRANS_ABORTED(cur_trans));
1968 btrfs_put_transaction(cur_trans);
1971 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1973 struct btrfs_fs_info *fs_info = trans->fs_info;
1974 struct btrfs_transaction *cur_trans = trans->transaction;
1976 WARN_ON(refcount_read(&trans->use_count) > 1);
1978 btrfs_abort_transaction(trans, err);
1980 spin_lock(&fs_info->trans_lock);
1983 * If the transaction is removed from the list, it means this
1984 * transaction has been committed successfully, so it is impossible
1985 * to call the cleanup function.
1987 BUG_ON(list_empty(&cur_trans->list));
1989 if (cur_trans == fs_info->running_transaction) {
1990 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1991 spin_unlock(&fs_info->trans_lock);
1994 * The thread has already released the lockdep map as reader
1995 * already in btrfs_commit_transaction().
1997 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
1998 wait_event(cur_trans->writer_wait,
1999 atomic_read(&cur_trans->num_writers) == 1);
2001 spin_lock(&fs_info->trans_lock);
2005 * Now that we know no one else is still using the transaction we can
2006 * remove the transaction from the list of transactions. This avoids
2007 * the transaction kthread from cleaning up the transaction while some
2008 * other task is still using it, which could result in a use-after-free
2009 * on things like log trees, as it forces the transaction kthread to
2010 * wait for this transaction to be cleaned up by us.
2012 list_del_init(&cur_trans->list);
2014 spin_unlock(&fs_info->trans_lock);
2016 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
2018 spin_lock(&fs_info->trans_lock);
2019 if (cur_trans == fs_info->running_transaction)
2020 fs_info->running_transaction = NULL;
2021 spin_unlock(&fs_info->trans_lock);
2023 if (trans->type & __TRANS_FREEZABLE)
2024 sb_end_intwrite(fs_info->sb);
2025 btrfs_put_transaction(cur_trans);
2026 btrfs_put_transaction(cur_trans);
2028 trace_btrfs_transaction_commit(fs_info);
2030 if (current->journal_info == trans)
2031 current->journal_info = NULL;
2034 * If relocation is running, we can't cancel scrub because that will
2035 * result in a deadlock. Before relocating a block group, relocation
2036 * pauses scrub, then starts and commits a transaction before unpausing
2037 * scrub. If the transaction commit is being done by the relocation
2038 * task or triggered by another task and the relocation task is waiting
2039 * for the commit, and we end up here due to an error in the commit
2040 * path, then calling btrfs_scrub_cancel() will deadlock, as we are
2041 * asking for scrub to stop while having it asked to be paused higher
2042 * above in relocation code.
2044 if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
2045 btrfs_scrub_cancel(fs_info);
2047 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2051 * Release reserved delayed ref space of all pending block groups of the
2052 * transaction and remove them from the list
2054 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2056 struct btrfs_fs_info *fs_info = trans->fs_info;
2057 struct btrfs_block_group *block_group, *tmp;
2059 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2060 btrfs_delayed_refs_rsv_release(fs_info, 1);
2061 list_del_init(&block_group->bg_list);
2065 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2068 * We use try_to_writeback_inodes_sb() here because if we used
2069 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2070 * Currently are holding the fs freeze lock, if we do an async flush
2071 * we'll do btrfs_join_transaction() and deadlock because we need to
2072 * wait for the fs freeze lock. Using the direct flushing we benefit
2073 * from already being in a transaction and our join_transaction doesn't
2074 * have to re-take the fs freeze lock.
2076 * Note that try_to_writeback_inodes_sb() will only trigger writeback
2077 * if it can read lock sb->s_umount. It will always be able to lock it,
2078 * except when the filesystem is being unmounted or being frozen, but in
2079 * those cases sync_filesystem() is called, which results in calling
2080 * writeback_inodes_sb() while holding a write lock on sb->s_umount.
2081 * Note that we don't call writeback_inodes_sb() directly, because it
2082 * will emit a warning if sb->s_umount is not locked.
2084 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2085 try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2089 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2091 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2092 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2096 * Add a pending snapshot associated with the given transaction handle to the
2097 * respective handle. This must be called after the transaction commit started
2098 * and while holding fs_info->trans_lock.
2099 * This serves to guarantee a caller of btrfs_commit_transaction() that it can
2100 * safely free the pending snapshot pointer in case btrfs_commit_transaction()
2103 static void add_pending_snapshot(struct btrfs_trans_handle *trans)
2105 struct btrfs_transaction *cur_trans = trans->transaction;
2107 if (!trans->pending_snapshot)
2110 lockdep_assert_held(&trans->fs_info->trans_lock);
2111 ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START);
2113 list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots);
2116 static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval)
2118 fs_info->commit_stats.commit_count++;
2119 fs_info->commit_stats.last_commit_dur = interval;
2120 fs_info->commit_stats.max_commit_dur =
2121 max_t(u64, fs_info->commit_stats.max_commit_dur, interval);
2122 fs_info->commit_stats.total_commit_dur += interval;
2125 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2127 struct btrfs_fs_info *fs_info = trans->fs_info;
2128 struct btrfs_transaction *cur_trans = trans->transaction;
2129 struct btrfs_transaction *prev_trans = NULL;
2134 ASSERT(refcount_read(&trans->use_count) == 1);
2135 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2137 clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags);
2139 /* Stop the commit early if ->aborted is set */
2140 if (TRANS_ABORTED(cur_trans)) {
2141 ret = cur_trans->aborted;
2142 goto lockdep_trans_commit_start_release;
2145 btrfs_trans_release_metadata(trans);
2146 trans->block_rsv = NULL;
2149 * We only want one transaction commit doing the flushing so we do not
2150 * waste a bunch of time on lock contention on the extent root node.
2152 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2153 &cur_trans->delayed_refs.flags)) {
2155 * Make a pass through all the delayed refs we have so far.
2156 * Any running threads may add more while we are here.
2158 ret = btrfs_run_delayed_refs(trans, 0);
2160 goto lockdep_trans_commit_start_release;
2163 btrfs_create_pending_block_groups(trans);
2165 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2168 /* this mutex is also taken before trying to set
2169 * block groups readonly. We need to make sure
2170 * that nobody has set a block group readonly
2171 * after a extents from that block group have been
2172 * allocated for cache files. btrfs_set_block_group_ro
2173 * will wait for the transaction to commit if it
2174 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2176 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2177 * only one process starts all the block group IO. It wouldn't
2178 * hurt to have more than one go through, but there's no
2179 * real advantage to it either.
2181 mutex_lock(&fs_info->ro_block_group_mutex);
2182 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2185 mutex_unlock(&fs_info->ro_block_group_mutex);
2188 ret = btrfs_start_dirty_block_groups(trans);
2190 goto lockdep_trans_commit_start_release;
2194 spin_lock(&fs_info->trans_lock);
2195 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2196 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2198 add_pending_snapshot(trans);
2200 spin_unlock(&fs_info->trans_lock);
2201 refcount_inc(&cur_trans->use_count);
2203 if (trans->in_fsync)
2204 want_state = TRANS_STATE_SUPER_COMMITTED;
2206 btrfs_trans_state_lockdep_release(fs_info,
2207 BTRFS_LOCKDEP_TRANS_COMMIT_START);
2208 ret = btrfs_end_transaction(trans);
2209 wait_for_commit(cur_trans, want_state);
2211 if (TRANS_ABORTED(cur_trans))
2212 ret = cur_trans->aborted;
2214 btrfs_put_transaction(cur_trans);
2219 cur_trans->state = TRANS_STATE_COMMIT_START;
2220 wake_up(&fs_info->transaction_blocked_wait);
2221 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2223 if (cur_trans->list.prev != &fs_info->trans_list) {
2224 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2226 if (trans->in_fsync)
2227 want_state = TRANS_STATE_SUPER_COMMITTED;
2229 prev_trans = list_entry(cur_trans->list.prev,
2230 struct btrfs_transaction, list);
2231 if (prev_trans->state < want_state) {
2232 refcount_inc(&prev_trans->use_count);
2233 spin_unlock(&fs_info->trans_lock);
2235 wait_for_commit(prev_trans, want_state);
2237 ret = READ_ONCE(prev_trans->aborted);
2239 btrfs_put_transaction(prev_trans);
2241 goto lockdep_release;
2243 spin_unlock(&fs_info->trans_lock);
2246 spin_unlock(&fs_info->trans_lock);
2248 * The previous transaction was aborted and was already removed
2249 * from the list of transactions at fs_info->trans_list. So we
2250 * abort to prevent writing a new superblock that reflects a
2251 * corrupt state (pointing to trees with unwritten nodes/leafs).
2253 if (BTRFS_FS_ERROR(fs_info)) {
2255 goto lockdep_release;
2260 * Get the time spent on the work done by the commit thread and not
2261 * the time spent waiting on a previous commit
2263 start_time = ktime_get_ns();
2265 extwriter_counter_dec(cur_trans, trans->type);
2267 ret = btrfs_start_delalloc_flush(fs_info);
2269 goto lockdep_release;
2271 ret = btrfs_run_delayed_items(trans);
2273 goto lockdep_release;
2276 * The thread has started/joined the transaction thus it holds the
2277 * lockdep map as a reader. It has to release it before acquiring the
2278 * lockdep map as a writer.
2280 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2281 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters);
2282 wait_event(cur_trans->writer_wait,
2283 extwriter_counter_read(cur_trans) == 0);
2285 /* some pending stuffs might be added after the previous flush. */
2286 ret = btrfs_run_delayed_items(trans);
2288 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2289 goto cleanup_transaction;
2292 btrfs_wait_delalloc_flush(fs_info);
2295 * Wait for all ordered extents started by a fast fsync that joined this
2296 * transaction. Otherwise if this transaction commits before the ordered
2297 * extents complete we lose logged data after a power failure.
2299 btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered);
2300 wait_event(cur_trans->pending_wait,
2301 atomic_read(&cur_trans->pending_ordered) == 0);
2303 btrfs_scrub_pause(fs_info);
2305 * Ok now we need to make sure to block out any other joins while we
2306 * commit the transaction. We could have started a join before setting
2307 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2309 spin_lock(&fs_info->trans_lock);
2310 add_pending_snapshot(trans);
2311 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2312 spin_unlock(&fs_info->trans_lock);
2315 * The thread has started/joined the transaction thus it holds the
2316 * lockdep map as a reader. It has to release it before acquiring the
2317 * lockdep map as a writer.
2319 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2320 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers);
2321 wait_event(cur_trans->writer_wait,
2322 atomic_read(&cur_trans->num_writers) == 1);
2325 * Make lockdep happy by acquiring the state locks after
2326 * btrfs_trans_num_writers is released. If we acquired the state locks
2327 * before releasing the btrfs_trans_num_writers lock then lockdep would
2328 * complain because we did not follow the reverse order unlocking rule.
2330 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2331 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2332 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2335 * We've started the commit, clear the flag in case we were triggered to
2336 * do an async commit but somebody else started before the transaction
2337 * kthread could do the work.
2339 clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags);
2341 if (TRANS_ABORTED(cur_trans)) {
2342 ret = cur_trans->aborted;
2343 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2344 goto scrub_continue;
2347 * the reloc mutex makes sure that we stop
2348 * the balancing code from coming in and moving
2349 * extents around in the middle of the commit
2351 mutex_lock(&fs_info->reloc_mutex);
2354 * We needn't worry about the delayed items because we will
2355 * deal with them in create_pending_snapshot(), which is the
2356 * core function of the snapshot creation.
2358 ret = create_pending_snapshots(trans);
2363 * We insert the dir indexes of the snapshots and update the inode
2364 * of the snapshots' parents after the snapshot creation, so there
2365 * are some delayed items which are not dealt with. Now deal with
2368 * We needn't worry that this operation will corrupt the snapshots,
2369 * because all the tree which are snapshoted will be forced to COW
2370 * the nodes and leaves.
2372 ret = btrfs_run_delayed_items(trans);
2376 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2381 * make sure none of the code above managed to slip in a
2384 btrfs_assert_delayed_root_empty(fs_info);
2386 WARN_ON(cur_trans != trans->transaction);
2388 ret = commit_fs_roots(trans);
2392 /* commit_fs_roots gets rid of all the tree log roots, it is now
2393 * safe to free the root of tree log roots
2395 btrfs_free_log_root_tree(trans, fs_info);
2398 * Since fs roots are all committed, we can get a quite accurate
2399 * new_roots. So let's do quota accounting.
2401 ret = btrfs_qgroup_account_extents(trans);
2405 ret = commit_cowonly_roots(trans);
2410 * The tasks which save the space cache and inode cache may also
2411 * update ->aborted, check it.
2413 if (TRANS_ABORTED(cur_trans)) {
2414 ret = cur_trans->aborted;
2418 cur_trans = fs_info->running_transaction;
2420 btrfs_set_root_node(&fs_info->tree_root->root_item,
2421 fs_info->tree_root->node);
2422 list_add_tail(&fs_info->tree_root->dirty_list,
2423 &cur_trans->switch_commits);
2425 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2426 fs_info->chunk_root->node);
2427 list_add_tail(&fs_info->chunk_root->dirty_list,
2428 &cur_trans->switch_commits);
2430 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2431 btrfs_set_root_node(&fs_info->block_group_root->root_item,
2432 fs_info->block_group_root->node);
2433 list_add_tail(&fs_info->block_group_root->dirty_list,
2434 &cur_trans->switch_commits);
2437 switch_commit_roots(trans);
2439 ASSERT(list_empty(&cur_trans->dirty_bgs));
2440 ASSERT(list_empty(&cur_trans->io_bgs));
2441 update_super_roots(fs_info);
2443 btrfs_set_super_log_root(fs_info->super_copy, 0);
2444 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2445 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2446 sizeof(*fs_info->super_copy));
2448 btrfs_commit_device_sizes(cur_trans);
2450 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2451 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2453 btrfs_trans_release_chunk_metadata(trans);
2456 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and
2457 * setting fs_info->running_transaction to NULL, lock tree_log_mutex to
2458 * make sure that before we commit our superblock, no other task can
2459 * start a new transaction and commit a log tree before we commit our
2460 * superblock. Anyone trying to commit a log tree locks this mutex before
2461 * writing its superblock.
2463 mutex_lock(&fs_info->tree_log_mutex);
2465 spin_lock(&fs_info->trans_lock);
2466 cur_trans->state = TRANS_STATE_UNBLOCKED;
2467 fs_info->running_transaction = NULL;
2468 spin_unlock(&fs_info->trans_lock);
2469 mutex_unlock(&fs_info->reloc_mutex);
2471 wake_up(&fs_info->transaction_wait);
2472 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2474 /* If we have features changed, wake up the cleaner to update sysfs. */
2475 if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) &&
2476 fs_info->cleaner_kthread)
2477 wake_up_process(fs_info->cleaner_kthread);
2479 ret = btrfs_write_and_wait_transaction(trans);
2481 btrfs_handle_fs_error(fs_info, ret,
2482 "Error while writing out transaction");
2483 mutex_unlock(&fs_info->tree_log_mutex);
2484 goto scrub_continue;
2487 ret = write_all_supers(fs_info, 0);
2489 * the super is written, we can safely allow the tree-loggers
2490 * to go about their business
2492 mutex_unlock(&fs_info->tree_log_mutex);
2494 goto scrub_continue;
2497 * We needn't acquire the lock here because there is no other task
2498 * which can change it.
2500 cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2501 wake_up(&cur_trans->commit_wait);
2502 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2504 btrfs_finish_extent_commit(trans);
2506 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2507 btrfs_clear_space_info_full(fs_info);
2509 fs_info->last_trans_committed = cur_trans->transid;
2511 * We needn't acquire the lock here because there is no other task
2512 * which can change it.
2514 cur_trans->state = TRANS_STATE_COMPLETED;
2515 wake_up(&cur_trans->commit_wait);
2516 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2518 spin_lock(&fs_info->trans_lock);
2519 list_del_init(&cur_trans->list);
2520 spin_unlock(&fs_info->trans_lock);
2522 btrfs_put_transaction(cur_trans);
2523 btrfs_put_transaction(cur_trans);
2525 if (trans->type & __TRANS_FREEZABLE)
2526 sb_end_intwrite(fs_info->sb);
2528 trace_btrfs_transaction_commit(fs_info);
2530 interval = ktime_get_ns() - start_time;
2532 btrfs_scrub_continue(fs_info);
2534 if (current->journal_info == trans)
2535 current->journal_info = NULL;
2537 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2539 update_commit_stats(fs_info, interval);
2544 mutex_unlock(&fs_info->reloc_mutex);
2545 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2547 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2548 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED);
2549 btrfs_scrub_continue(fs_info);
2550 cleanup_transaction:
2551 btrfs_trans_release_metadata(trans);
2552 btrfs_cleanup_pending_block_groups(trans);
2553 btrfs_trans_release_chunk_metadata(trans);
2554 trans->block_rsv = NULL;
2555 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2556 if (current->journal_info == trans)
2557 current->journal_info = NULL;
2558 cleanup_transaction(trans, ret);
2563 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters);
2564 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers);
2565 goto cleanup_transaction;
2567 lockdep_trans_commit_start_release:
2568 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START);
2569 btrfs_end_transaction(trans);
2574 * return < 0 if error
2575 * 0 if there are no more dead_roots at the time of call
2576 * 1 there are more to be processed, call me again
2578 * The return value indicates there are certainly more snapshots to delete, but
2579 * if there comes a new one during processing, it may return 0. We don't mind,
2580 * because btrfs_commit_super will poke cleaner thread and it will process it a
2581 * few seconds later.
2583 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
2585 struct btrfs_root *root;
2588 spin_lock(&fs_info->trans_lock);
2589 if (list_empty(&fs_info->dead_roots)) {
2590 spin_unlock(&fs_info->trans_lock);
2593 root = list_first_entry(&fs_info->dead_roots,
2594 struct btrfs_root, root_list);
2595 list_del_init(&root->root_list);
2596 spin_unlock(&fs_info->trans_lock);
2598 btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2600 btrfs_kill_all_delayed_nodes(root);
2602 if (btrfs_header_backref_rev(root->node) <
2603 BTRFS_MIXED_BACKREF_REV)
2604 ret = btrfs_drop_snapshot(root, 0, 0);
2606 ret = btrfs_drop_snapshot(root, 1, 0);
2608 btrfs_put_root(root);
2609 return (ret < 0) ? 0 : 1;
2613 * We only mark the transaction aborted and then set the file system read-only.
2614 * This will prevent new transactions from starting or trying to join this
2617 * This means that error recovery at the call site is limited to freeing
2618 * any local memory allocations and passing the error code up without
2619 * further cleanup. The transaction should complete as it normally would
2620 * in the call path but will return -EIO.
2622 * We'll complete the cleanup in btrfs_end_transaction and
2623 * btrfs_commit_transaction.
2625 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
2626 const char *function,
2627 unsigned int line, int errno, bool first_hit)
2629 struct btrfs_fs_info *fs_info = trans->fs_info;
2631 WRITE_ONCE(trans->aborted, errno);
2632 WRITE_ONCE(trans->transaction->aborted, errno);
2633 if (first_hit && errno == -ENOSPC)
2634 btrfs_dump_space_info_for_trans_abort(fs_info);
2635 /* Wake up anybody who may be waiting on this transaction */
2636 wake_up(&fs_info->transaction_wait);
2637 wake_up(&fs_info->transaction_blocked_wait);
2638 __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
2641 int __init btrfs_transaction_init(void)
2643 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
2644 sizeof(struct btrfs_trans_handle), 0,
2645 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
2646 if (!btrfs_trans_handle_cachep)
2651 void __cold btrfs_transaction_exit(void)
2653 kmem_cache_destroy(btrfs_trans_handle_cachep);