1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/writeback.h>
10 #include <linux/pagemap.h>
11 #include <linux/blkdev.h>
12 #include <linux/uuid.h>
16 #include "transaction.h"
20 #include "dev-replace.h"
22 #include "block-group.h"
23 #include "space-info.h"
26 #define BTRFS_ROOT_TRANS_TAG 0
29 * Transaction states and transitions
31 * No running transaction (fs tree blocks are not modified)
34 * | Call start_transaction() variants. Except btrfs_join_transaction_nostart().
36 * Transaction N [[TRANS_STATE_RUNNING]]
38 * | New trans handles can be attached to transaction N by calling all
39 * | start_transaction() variants.
42 * | Call btrfs_commit_transaction() on any trans handle attached to
45 * Transaction N [[TRANS_STATE_COMMIT_START]]
47 * | Will wait for previous running transaction to completely finish if there
50 * | Then one of the following happes:
51 * | - Wait for all other trans handle holders to release.
52 * | The btrfs_commit_transaction() caller will do the commit work.
53 * | - Wait for current transaction to be committed by others.
54 * | Other btrfs_commit_transaction() caller will do the commit work.
56 * | At this stage, only btrfs_join_transaction*() variants can attach
57 * | to this running transaction.
58 * | All other variants will wait for current one to finish and attach to
62 * | Caller is chosen to commit transaction N, and all other trans handle
63 * | haven been released.
65 * Transaction N [[TRANS_STATE_COMMIT_DOING]]
67 * | The heavy lifting transaction work is started.
68 * | From running delayed refs (modifying extent tree) to creating pending
69 * | snapshots, running qgroups.
70 * | In short, modify supporting trees to reflect modifications of subvolume
73 * | At this stage, all start_transaction() calls will wait for this
74 * | transaction to finish and attach to transaction N+1.
77 * | Until all supporting trees are updated.
79 * Transaction N [[TRANS_STATE_UNBLOCKED]]
81 * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
82 * | need to write them back to disk and update |
85 * | At this stage, new transaction is allowed to |
87 * | All new start_transaction() calls will be |
88 * | attached to transid N+1. |
91 * | Until all tree blocks are super blocks are |
92 * | written to block devices |
94 * Transaction N [[TRANS_STATE_COMPLETED]] V
95 * All tree blocks and super blocks are written. Transaction N+1
96 * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]]
97 * data structures will be cleaned up. | Life goes on
99 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
100 [TRANS_STATE_RUNNING] = 0U,
101 [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH),
102 [TRANS_STATE_COMMIT_DOING] = (__TRANS_START |
105 __TRANS_JOIN_NOSTART),
106 [TRANS_STATE_UNBLOCKED] = (__TRANS_START |
109 __TRANS_JOIN_NOLOCK |
110 __TRANS_JOIN_NOSTART),
111 [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START |
114 __TRANS_JOIN_NOLOCK |
115 __TRANS_JOIN_NOSTART),
116 [TRANS_STATE_COMPLETED] = (__TRANS_START |
119 __TRANS_JOIN_NOLOCK |
120 __TRANS_JOIN_NOSTART),
123 void btrfs_put_transaction(struct btrfs_transaction *transaction)
125 WARN_ON(refcount_read(&transaction->use_count) == 0);
126 if (refcount_dec_and_test(&transaction->use_count)) {
127 BUG_ON(!list_empty(&transaction->list));
128 WARN_ON(!RB_EMPTY_ROOT(
129 &transaction->delayed_refs.href_root.rb_root));
130 WARN_ON(!RB_EMPTY_ROOT(
131 &transaction->delayed_refs.dirty_extent_root));
132 if (transaction->delayed_refs.pending_csums)
133 btrfs_err(transaction->fs_info,
134 "pending csums is %llu",
135 transaction->delayed_refs.pending_csums);
137 * If any block groups are found in ->deleted_bgs then it's
138 * because the transaction was aborted and a commit did not
139 * happen (things failed before writing the new superblock
140 * and calling btrfs_finish_extent_commit()), so we can not
141 * discard the physical locations of the block groups.
143 while (!list_empty(&transaction->deleted_bgs)) {
144 struct btrfs_block_group *cache;
146 cache = list_first_entry(&transaction->deleted_bgs,
147 struct btrfs_block_group,
149 list_del_init(&cache->bg_list);
150 btrfs_unfreeze_block_group(cache);
151 btrfs_put_block_group(cache);
153 WARN_ON(!list_empty(&transaction->dev_update_list));
158 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
160 struct btrfs_transaction *cur_trans = trans->transaction;
161 struct btrfs_fs_info *fs_info = trans->fs_info;
162 struct btrfs_root *root, *tmp;
163 struct btrfs_caching_control *caching_ctl, *next;
165 down_write(&fs_info->commit_root_sem);
166 list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
168 list_del_init(&root->dirty_list);
169 free_extent_buffer(root->commit_root);
170 root->commit_root = btrfs_root_node(root);
171 extent_io_tree_release(&root->dirty_log_pages);
172 btrfs_qgroup_clean_swapped_blocks(root);
175 /* We can free old roots now. */
176 spin_lock(&cur_trans->dropped_roots_lock);
177 while (!list_empty(&cur_trans->dropped_roots)) {
178 root = list_first_entry(&cur_trans->dropped_roots,
179 struct btrfs_root, root_list);
180 list_del_init(&root->root_list);
181 spin_unlock(&cur_trans->dropped_roots_lock);
182 btrfs_free_log(trans, root);
183 btrfs_drop_and_free_fs_root(fs_info, root);
184 spin_lock(&cur_trans->dropped_roots_lock);
186 spin_unlock(&cur_trans->dropped_roots_lock);
189 * We have to update the last_byte_to_unpin under the commit_root_sem,
190 * at the same time we swap out the commit roots.
192 * This is because we must have a real view of the last spot the caching
193 * kthreads were while caching. Consider the following views of the
194 * extent tree for a block group
197 * +----+----+----+----+----+----+----+
198 * |\\\\| |\\\\|\\\\| |\\\\|\\\\|
199 * +----+----+----+----+----+----+----+
203 * +----+----+----+----+----+----+----+
204 * | | | |\\\\| | |\\\\|
205 * +----+----+----+----+----+----+----+
208 * If the cache_ctl->progress was at 3, then we are only allowed to
209 * unpin [0,1) and [2,3], because the caching thread has already
210 * processed those extents. We are not allowed to unpin [5,6), because
211 * the caching thread will re-start it's search from 3, and thus find
212 * the hole from [4,6) to add to the free space cache.
214 spin_lock(&fs_info->block_group_cache_lock);
215 list_for_each_entry_safe(caching_ctl, next,
216 &fs_info->caching_block_groups, list) {
217 struct btrfs_block_group *cache = caching_ctl->block_group;
219 if (btrfs_block_group_done(cache)) {
220 cache->last_byte_to_unpin = (u64)-1;
221 list_del_init(&caching_ctl->list);
222 btrfs_put_caching_control(caching_ctl);
224 cache->last_byte_to_unpin = caching_ctl->progress;
227 spin_unlock(&fs_info->block_group_cache_lock);
228 up_write(&fs_info->commit_root_sem);
231 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
234 if (type & TRANS_EXTWRITERS)
235 atomic_inc(&trans->num_extwriters);
238 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
241 if (type & TRANS_EXTWRITERS)
242 atomic_dec(&trans->num_extwriters);
245 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
248 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
251 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
253 return atomic_read(&trans->num_extwriters);
257 * To be called after doing the chunk btree updates right after allocating a new
258 * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
259 * chunk after all chunk btree updates and after finishing the second phase of
260 * chunk allocation (btrfs_create_pending_block_groups()) in case some block
261 * group had its chunk item insertion delayed to the second phase.
263 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
265 struct btrfs_fs_info *fs_info = trans->fs_info;
267 if (!trans->chunk_bytes_reserved)
270 btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
271 trans->chunk_bytes_reserved, NULL);
272 trans->chunk_bytes_reserved = 0;
276 * either allocate a new transaction or hop into the existing one
278 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
281 struct btrfs_transaction *cur_trans;
283 spin_lock(&fs_info->trans_lock);
285 /* The file system has been taken offline. No new transactions. */
286 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
287 spin_unlock(&fs_info->trans_lock);
291 cur_trans = fs_info->running_transaction;
293 if (TRANS_ABORTED(cur_trans)) {
294 spin_unlock(&fs_info->trans_lock);
295 return cur_trans->aborted;
297 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
298 spin_unlock(&fs_info->trans_lock);
301 refcount_inc(&cur_trans->use_count);
302 atomic_inc(&cur_trans->num_writers);
303 extwriter_counter_inc(cur_trans, type);
304 spin_unlock(&fs_info->trans_lock);
307 spin_unlock(&fs_info->trans_lock);
310 * If we are ATTACH, we just want to catch the current transaction,
311 * and commit it. If there is no transaction, just return ENOENT.
313 if (type == TRANS_ATTACH)
317 * JOIN_NOLOCK only happens during the transaction commit, so
318 * it is impossible that ->running_transaction is NULL
320 BUG_ON(type == TRANS_JOIN_NOLOCK);
322 cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
326 spin_lock(&fs_info->trans_lock);
327 if (fs_info->running_transaction) {
329 * someone started a transaction after we unlocked. Make sure
330 * to redo the checks above
334 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
335 spin_unlock(&fs_info->trans_lock);
340 cur_trans->fs_info = fs_info;
341 atomic_set(&cur_trans->pending_ordered, 0);
342 init_waitqueue_head(&cur_trans->pending_wait);
343 atomic_set(&cur_trans->num_writers, 1);
344 extwriter_counter_init(cur_trans, type);
345 init_waitqueue_head(&cur_trans->writer_wait);
346 init_waitqueue_head(&cur_trans->commit_wait);
347 cur_trans->state = TRANS_STATE_RUNNING;
349 * One for this trans handle, one so it will live on until we
350 * commit the transaction.
352 refcount_set(&cur_trans->use_count, 2);
353 cur_trans->flags = 0;
354 cur_trans->start_time = ktime_get_seconds();
356 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
358 cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
359 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
360 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
363 * although the tree mod log is per file system and not per transaction,
364 * the log must never go across transaction boundaries.
367 if (!list_empty(&fs_info->tree_mod_seq_list))
368 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
369 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
370 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
371 atomic64_set(&fs_info->tree_mod_seq, 0);
373 spin_lock_init(&cur_trans->delayed_refs.lock);
375 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
376 INIT_LIST_HEAD(&cur_trans->dev_update_list);
377 INIT_LIST_HEAD(&cur_trans->switch_commits);
378 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
379 INIT_LIST_HEAD(&cur_trans->io_bgs);
380 INIT_LIST_HEAD(&cur_trans->dropped_roots);
381 mutex_init(&cur_trans->cache_write_mutex);
382 spin_lock_init(&cur_trans->dirty_bgs_lock);
383 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
384 spin_lock_init(&cur_trans->dropped_roots_lock);
385 INIT_LIST_HEAD(&cur_trans->releasing_ebs);
386 spin_lock_init(&cur_trans->releasing_ebs_lock);
387 list_add_tail(&cur_trans->list, &fs_info->trans_list);
388 extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
389 IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode);
390 extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
391 IO_TREE_FS_PINNED_EXTENTS, NULL);
392 fs_info->generation++;
393 cur_trans->transid = fs_info->generation;
394 fs_info->running_transaction = cur_trans;
395 cur_trans->aborted = 0;
396 spin_unlock(&fs_info->trans_lock);
402 * This does all the record keeping required to make sure that a shareable root
403 * is properly recorded in a given transaction. This is required to make sure
404 * the old root from before we joined the transaction is deleted when the
405 * transaction commits.
407 static int record_root_in_trans(struct btrfs_trans_handle *trans,
408 struct btrfs_root *root,
411 struct btrfs_fs_info *fs_info = root->fs_info;
414 if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
415 root->last_trans < trans->transid) || force) {
416 WARN_ON(root == fs_info->extent_root);
417 WARN_ON(!force && root->commit_root != root->node);
420 * see below for IN_TRANS_SETUP usage rules
421 * we have the reloc mutex held now, so there
422 * is only one writer in this function
424 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
426 /* make sure readers find IN_TRANS_SETUP before
427 * they find our root->last_trans update
431 spin_lock(&fs_info->fs_roots_radix_lock);
432 if (root->last_trans == trans->transid && !force) {
433 spin_unlock(&fs_info->fs_roots_radix_lock);
436 radix_tree_tag_set(&fs_info->fs_roots_radix,
437 (unsigned long)root->root_key.objectid,
438 BTRFS_ROOT_TRANS_TAG);
439 spin_unlock(&fs_info->fs_roots_radix_lock);
440 root->last_trans = trans->transid;
442 /* this is pretty tricky. We don't want to
443 * take the relocation lock in btrfs_record_root_in_trans
444 * unless we're really doing the first setup for this root in
447 * Normally we'd use root->last_trans as a flag to decide
448 * if we want to take the expensive mutex.
450 * But, we have to set root->last_trans before we
451 * init the relocation root, otherwise, we trip over warnings
452 * in ctree.c. The solution used here is to flag ourselves
453 * with root IN_TRANS_SETUP. When this is 1, we're still
454 * fixing up the reloc trees and everyone must wait.
456 * When this is zero, they can trust root->last_trans and fly
457 * through btrfs_record_root_in_trans without having to take the
458 * lock. smp_wmb() makes sure that all the writes above are
459 * done before we pop in the zero below
461 ret = btrfs_init_reloc_root(trans, root);
462 smp_mb__before_atomic();
463 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
469 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
470 struct btrfs_root *root)
472 struct btrfs_fs_info *fs_info = root->fs_info;
473 struct btrfs_transaction *cur_trans = trans->transaction;
475 /* Add ourselves to the transaction dropped list */
476 spin_lock(&cur_trans->dropped_roots_lock);
477 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
478 spin_unlock(&cur_trans->dropped_roots_lock);
480 /* Make sure we don't try to update the root at commit time */
481 spin_lock(&fs_info->fs_roots_radix_lock);
482 radix_tree_tag_clear(&fs_info->fs_roots_radix,
483 (unsigned long)root->root_key.objectid,
484 BTRFS_ROOT_TRANS_TAG);
485 spin_unlock(&fs_info->fs_roots_radix_lock);
488 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
489 struct btrfs_root *root)
491 struct btrfs_fs_info *fs_info = root->fs_info;
494 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
498 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
502 if (root->last_trans == trans->transid &&
503 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
506 mutex_lock(&fs_info->reloc_mutex);
507 ret = record_root_in_trans(trans, root, 0);
508 mutex_unlock(&fs_info->reloc_mutex);
513 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
515 return (trans->state >= TRANS_STATE_COMMIT_START &&
516 trans->state < TRANS_STATE_UNBLOCKED &&
517 !TRANS_ABORTED(trans));
520 /* wait for commit against the current transaction to become unblocked
521 * when this is done, it is safe to start a new transaction, but the current
522 * transaction might not be fully on disk.
524 static void wait_current_trans(struct btrfs_fs_info *fs_info)
526 struct btrfs_transaction *cur_trans;
528 spin_lock(&fs_info->trans_lock);
529 cur_trans = fs_info->running_transaction;
530 if (cur_trans && is_transaction_blocked(cur_trans)) {
531 refcount_inc(&cur_trans->use_count);
532 spin_unlock(&fs_info->trans_lock);
534 wait_event(fs_info->transaction_wait,
535 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
536 TRANS_ABORTED(cur_trans));
537 btrfs_put_transaction(cur_trans);
539 spin_unlock(&fs_info->trans_lock);
543 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
545 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
548 if (type == TRANS_START)
554 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
556 struct btrfs_fs_info *fs_info = root->fs_info;
558 if (!fs_info->reloc_ctl ||
559 !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
560 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
567 static struct btrfs_trans_handle *
568 start_transaction(struct btrfs_root *root, unsigned int num_items,
569 unsigned int type, enum btrfs_reserve_flush_enum flush,
570 bool enforce_qgroups)
572 struct btrfs_fs_info *fs_info = root->fs_info;
573 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
574 struct btrfs_trans_handle *h;
575 struct btrfs_transaction *cur_trans;
577 u64 qgroup_reserved = 0;
578 bool reloc_reserved = false;
579 bool do_chunk_alloc = false;
582 /* Send isn't supposed to start transactions. */
583 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
585 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
586 return ERR_PTR(-EROFS);
588 if (current->journal_info) {
589 WARN_ON(type & TRANS_EXTWRITERS);
590 h = current->journal_info;
591 refcount_inc(&h->use_count);
592 WARN_ON(refcount_read(&h->use_count) > 2);
593 h->orig_rsv = h->block_rsv;
599 * Do the reservation before we join the transaction so we can do all
600 * the appropriate flushing if need be.
602 if (num_items && root != fs_info->chunk_root) {
603 struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
604 u64 delayed_refs_bytes = 0;
606 qgroup_reserved = num_items * fs_info->nodesize;
607 ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
613 * We want to reserve all the bytes we may need all at once, so
614 * we only do 1 enospc flushing cycle per transaction start. We
615 * accomplish this by simply assuming we'll do 2 x num_items
616 * worth of delayed refs updates in this trans handle, and
617 * refill that amount for whatever is missing in the reserve.
619 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
620 if (flush == BTRFS_RESERVE_FLUSH_ALL &&
621 delayed_refs_rsv->full == 0) {
622 delayed_refs_bytes = num_bytes;
627 * Do the reservation for the relocation root creation
629 if (need_reserve_reloc_root(root)) {
630 num_bytes += fs_info->nodesize;
631 reloc_reserved = true;
634 ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
637 if (delayed_refs_bytes) {
638 btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
640 num_bytes -= delayed_refs_bytes;
643 if (rsv->space_info->force_alloc)
644 do_chunk_alloc = true;
645 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
646 !delayed_refs_rsv->full) {
648 * Some people call with btrfs_start_transaction(root, 0)
649 * because they can be throttled, but have some other mechanism
650 * for reserving space. We still want these guys to refill the
651 * delayed block_rsv so just add 1 items worth of reservation
654 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
659 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
666 * If we are JOIN_NOLOCK we're already committing a transaction and
667 * waiting on this guy, so we don't need to do the sb_start_intwrite
668 * because we're already holding a ref. We need this because we could
669 * have raced in and did an fsync() on a file which can kick a commit
670 * and then we deadlock with somebody doing a freeze.
672 * If we are ATTACH, it means we just want to catch the current
673 * transaction and commit it, so we needn't do sb_start_intwrite().
675 if (type & __TRANS_FREEZABLE)
676 sb_start_intwrite(fs_info->sb);
678 if (may_wait_transaction(fs_info, type))
679 wait_current_trans(fs_info);
682 ret = join_transaction(fs_info, type);
684 wait_current_trans(fs_info);
685 if (unlikely(type == TRANS_ATTACH ||
686 type == TRANS_JOIN_NOSTART))
689 } while (ret == -EBUSY);
694 cur_trans = fs_info->running_transaction;
696 h->transid = cur_trans->transid;
697 h->transaction = cur_trans;
699 refcount_set(&h->use_count, 1);
700 h->fs_info = root->fs_info;
703 INIT_LIST_HEAD(&h->new_bgs);
706 if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
707 may_wait_transaction(fs_info, type)) {
708 current->journal_info = h;
709 btrfs_commit_transaction(h);
714 trace_btrfs_space_reservation(fs_info, "transaction",
715 h->transid, num_bytes, 1);
716 h->block_rsv = &fs_info->trans_block_rsv;
717 h->bytes_reserved = num_bytes;
718 h->reloc_reserved = reloc_reserved;
722 if (!current->journal_info)
723 current->journal_info = h;
726 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
727 * ALLOC_FORCE the first run through, and then we won't allocate for
728 * anybody else who races in later. We don't care about the return
731 if (do_chunk_alloc && num_bytes) {
732 u64 flags = h->block_rsv->space_info->flags;
734 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
735 CHUNK_ALLOC_NO_FORCE);
739 * btrfs_record_root_in_trans() needs to alloc new extents, and may
740 * call btrfs_join_transaction() while we're also starting a
743 * Thus it need to be called after current->journal_info initialized,
744 * or we can deadlock.
746 ret = btrfs_record_root_in_trans(h, root);
749 * The transaction handle is fully initialized and linked with
750 * other structures so it needs to be ended in case of errors,
753 btrfs_end_transaction(h);
760 if (type & __TRANS_FREEZABLE)
761 sb_end_intwrite(fs_info->sb);
762 kmem_cache_free(btrfs_trans_handle_cachep, h);
765 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
768 btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
772 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
773 unsigned int num_items)
775 return start_transaction(root, num_items, TRANS_START,
776 BTRFS_RESERVE_FLUSH_ALL, true);
779 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
780 struct btrfs_root *root,
781 unsigned int num_items)
783 return start_transaction(root, num_items, TRANS_START,
784 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
787 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
789 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
793 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
795 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
796 BTRFS_RESERVE_NO_FLUSH, true);
800 * Similar to regular join but it never starts a transaction when none is
801 * running or after waiting for the current one to finish.
803 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
805 return start_transaction(root, 0, TRANS_JOIN_NOSTART,
806 BTRFS_RESERVE_NO_FLUSH, true);
810 * btrfs_attach_transaction() - catch the running transaction
812 * It is used when we want to commit the current the transaction, but
813 * don't want to start a new one.
815 * Note: If this function return -ENOENT, it just means there is no
816 * running transaction. But it is possible that the inactive transaction
817 * is still in the memory, not fully on disk. If you hope there is no
818 * inactive transaction in the fs when -ENOENT is returned, you should
820 * btrfs_attach_transaction_barrier()
822 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
824 return start_transaction(root, 0, TRANS_ATTACH,
825 BTRFS_RESERVE_NO_FLUSH, true);
829 * btrfs_attach_transaction_barrier() - catch the running transaction
831 * It is similar to the above function, the difference is this one
832 * will wait for all the inactive transactions until they fully
835 struct btrfs_trans_handle *
836 btrfs_attach_transaction_barrier(struct btrfs_root *root)
838 struct btrfs_trans_handle *trans;
840 trans = start_transaction(root, 0, TRANS_ATTACH,
841 BTRFS_RESERVE_NO_FLUSH, true);
842 if (trans == ERR_PTR(-ENOENT))
843 btrfs_wait_for_commit(root->fs_info, 0);
848 /* Wait for a transaction commit to reach at least the given state. */
849 static noinline void wait_for_commit(struct btrfs_transaction *commit,
850 const enum btrfs_trans_state min_state)
852 wait_event(commit->commit_wait, commit->state >= min_state);
855 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
857 struct btrfs_transaction *cur_trans = NULL, *t;
861 if (transid <= fs_info->last_trans_committed)
864 /* find specified transaction */
865 spin_lock(&fs_info->trans_lock);
866 list_for_each_entry(t, &fs_info->trans_list, list) {
867 if (t->transid == transid) {
869 refcount_inc(&cur_trans->use_count);
873 if (t->transid > transid) {
878 spin_unlock(&fs_info->trans_lock);
881 * The specified transaction doesn't exist, or we
882 * raced with btrfs_commit_transaction
885 if (transid > fs_info->last_trans_committed)
890 /* find newest transaction that is committing | committed */
891 spin_lock(&fs_info->trans_lock);
892 list_for_each_entry_reverse(t, &fs_info->trans_list,
894 if (t->state >= TRANS_STATE_COMMIT_START) {
895 if (t->state == TRANS_STATE_COMPLETED)
898 refcount_inc(&cur_trans->use_count);
902 spin_unlock(&fs_info->trans_lock);
904 goto out; /* nothing committing|committed */
907 wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
908 btrfs_put_transaction(cur_trans);
913 void btrfs_throttle(struct btrfs_fs_info *fs_info)
915 wait_current_trans(fs_info);
918 static bool should_end_transaction(struct btrfs_trans_handle *trans)
920 struct btrfs_fs_info *fs_info = trans->fs_info;
922 if (btrfs_check_space_for_delayed_refs(fs_info))
925 return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
928 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
930 struct btrfs_transaction *cur_trans = trans->transaction;
932 if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
933 test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
936 return should_end_transaction(trans);
939 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
942 struct btrfs_fs_info *fs_info = trans->fs_info;
944 if (!trans->block_rsv) {
945 ASSERT(!trans->bytes_reserved);
949 if (!trans->bytes_reserved)
952 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
953 trace_btrfs_space_reservation(fs_info, "transaction",
954 trans->transid, trans->bytes_reserved, 0);
955 btrfs_block_rsv_release(fs_info, trans->block_rsv,
956 trans->bytes_reserved, NULL);
957 trans->bytes_reserved = 0;
960 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
963 struct btrfs_fs_info *info = trans->fs_info;
964 struct btrfs_transaction *cur_trans = trans->transaction;
967 if (refcount_read(&trans->use_count) > 1) {
968 refcount_dec(&trans->use_count);
969 trans->block_rsv = trans->orig_rsv;
973 btrfs_trans_release_metadata(trans);
974 trans->block_rsv = NULL;
976 btrfs_create_pending_block_groups(trans);
978 btrfs_trans_release_chunk_metadata(trans);
980 if (trans->type & __TRANS_FREEZABLE)
981 sb_end_intwrite(info->sb);
983 WARN_ON(cur_trans != info->running_transaction);
984 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
985 atomic_dec(&cur_trans->num_writers);
986 extwriter_counter_dec(cur_trans, trans->type);
988 cond_wake_up(&cur_trans->writer_wait);
989 btrfs_put_transaction(cur_trans);
991 if (current->journal_info == trans)
992 current->journal_info = NULL;
995 btrfs_run_delayed_iputs(info);
997 if (TRANS_ABORTED(trans) ||
998 test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
999 wake_up_process(info->transaction_kthread);
1000 if (TRANS_ABORTED(trans))
1001 err = trans->aborted;
1006 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1010 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1012 return __btrfs_end_transaction(trans, 0);
1015 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1017 return __btrfs_end_transaction(trans, 1);
1021 * when btree blocks are allocated, they have some corresponding bits set for
1022 * them in one of two extent_io trees. This is used to make sure all of
1023 * those extents are sent to disk but does not wait on them
1025 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1026 struct extent_io_tree *dirty_pages, int mark)
1030 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1031 struct extent_state *cached_state = NULL;
1035 atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1036 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1037 mark, &cached_state)) {
1038 bool wait_writeback = false;
1040 err = convert_extent_bit(dirty_pages, start, end,
1042 mark, &cached_state);
1044 * convert_extent_bit can return -ENOMEM, which is most of the
1045 * time a temporary error. So when it happens, ignore the error
1046 * and wait for writeback of this range to finish - because we
1047 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1048 * to __btrfs_wait_marked_extents() would not know that
1049 * writeback for this range started and therefore wouldn't
1050 * wait for it to finish - we don't want to commit a
1051 * superblock that points to btree nodes/leafs for which
1052 * writeback hasn't finished yet (and without errors).
1053 * We cleanup any entries left in the io tree when committing
1054 * the transaction (through extent_io_tree_release()).
1056 if (err == -ENOMEM) {
1058 wait_writeback = true;
1061 err = filemap_fdatawrite_range(mapping, start, end);
1064 else if (wait_writeback)
1065 werr = filemap_fdatawait_range(mapping, start, end);
1066 free_extent_state(cached_state);
1067 cached_state = NULL;
1071 atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1076 * when btree blocks are allocated, they have some corresponding bits set for
1077 * them in one of two extent_io trees. This is used to make sure all of
1078 * those extents are on disk for transaction or log commit. We wait
1079 * on all the pages and clear them from the dirty pages state tree
1081 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1082 struct extent_io_tree *dirty_pages)
1086 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1087 struct extent_state *cached_state = NULL;
1091 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1092 EXTENT_NEED_WAIT, &cached_state)) {
1094 * Ignore -ENOMEM errors returned by clear_extent_bit().
1095 * When committing the transaction, we'll remove any entries
1096 * left in the io tree. For a log commit, we don't remove them
1097 * after committing the log because the tree can be accessed
1098 * concurrently - we do it only at transaction commit time when
1099 * it's safe to do it (through extent_io_tree_release()).
1101 err = clear_extent_bit(dirty_pages, start, end,
1102 EXTENT_NEED_WAIT, 0, 0, &cached_state);
1106 err = filemap_fdatawait_range(mapping, start, end);
1109 free_extent_state(cached_state);
1110 cached_state = NULL;
1119 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1120 struct extent_io_tree *dirty_pages)
1122 bool errors = false;
1125 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1126 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1134 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1136 struct btrfs_fs_info *fs_info = log_root->fs_info;
1137 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1138 bool errors = false;
1141 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1143 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1144 if ((mark & EXTENT_DIRTY) &&
1145 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1148 if ((mark & EXTENT_NEW) &&
1149 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1158 * When btree blocks are allocated the corresponding extents are marked dirty.
1159 * This function ensures such extents are persisted on disk for transaction or
1162 * @trans: transaction whose dirty pages we'd like to write
1164 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1168 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1169 struct btrfs_fs_info *fs_info = trans->fs_info;
1170 struct blk_plug plug;
1172 blk_start_plug(&plug);
1173 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1174 blk_finish_plug(&plug);
1175 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1177 extent_io_tree_release(&trans->transaction->dirty_pages);
1188 * this is used to update the root pointer in the tree of tree roots.
1190 * But, in the case of the extent allocation tree, updating the root
1191 * pointer may allocate blocks which may change the root of the extent
1194 * So, this loops and repeats and makes sure the cowonly root didn't
1195 * change while the root pointer was being updated in the metadata.
1197 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1198 struct btrfs_root *root)
1201 u64 old_root_bytenr;
1203 struct btrfs_fs_info *fs_info = root->fs_info;
1204 struct btrfs_root *tree_root = fs_info->tree_root;
1206 old_root_used = btrfs_root_used(&root->root_item);
1209 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1210 if (old_root_bytenr == root->node->start &&
1211 old_root_used == btrfs_root_used(&root->root_item))
1214 btrfs_set_root_node(&root->root_item, root->node);
1215 ret = btrfs_update_root(trans, tree_root,
1221 old_root_used = btrfs_root_used(&root->root_item);
1228 * update all the cowonly tree roots on disk
1230 * The error handling in this function may not be obvious. Any of the
1231 * failures will cause the file system to go offline. We still need
1232 * to clean up the delayed refs.
1234 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1236 struct btrfs_fs_info *fs_info = trans->fs_info;
1237 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1238 struct list_head *io_bgs = &trans->transaction->io_bgs;
1239 struct list_head *next;
1240 struct extent_buffer *eb;
1243 eb = btrfs_lock_root_node(fs_info->tree_root);
1244 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1245 0, &eb, BTRFS_NESTING_COW);
1246 btrfs_tree_unlock(eb);
1247 free_extent_buffer(eb);
1252 ret = btrfs_run_dev_stats(trans);
1255 ret = btrfs_run_dev_replace(trans);
1258 ret = btrfs_run_qgroups(trans);
1262 ret = btrfs_setup_space_cache(trans);
1267 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1268 struct btrfs_root *root;
1269 next = fs_info->dirty_cowonly_roots.next;
1270 list_del_init(next);
1271 root = list_entry(next, struct btrfs_root, dirty_list);
1272 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1274 if (root != fs_info->extent_root)
1275 list_add_tail(&root->dirty_list,
1276 &trans->transaction->switch_commits);
1277 ret = update_cowonly_root(trans, root);
1282 /* Now flush any delayed refs generated by updating all of the roots */
1283 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1287 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1288 ret = btrfs_write_dirty_block_groups(trans);
1293 * We're writing the dirty block groups, which could generate
1294 * delayed refs, which could generate more dirty block groups,
1295 * so we want to keep this flushing in this loop to make sure
1296 * everything gets run.
1298 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1303 if (!list_empty(&fs_info->dirty_cowonly_roots))
1306 list_add_tail(&fs_info->extent_root->dirty_list,
1307 &trans->transaction->switch_commits);
1309 /* Update dev-replace pointer once everything is committed */
1310 fs_info->dev_replace.committed_cursor_left =
1311 fs_info->dev_replace.cursor_left_last_write_of_item;
1317 * dead roots are old snapshots that need to be deleted. This allocates
1318 * a dirty root struct and adds it into the list of dead roots that need to
1321 void btrfs_add_dead_root(struct btrfs_root *root)
1323 struct btrfs_fs_info *fs_info = root->fs_info;
1325 spin_lock(&fs_info->trans_lock);
1326 if (list_empty(&root->root_list)) {
1327 btrfs_grab_root(root);
1328 list_add_tail(&root->root_list, &fs_info->dead_roots);
1330 spin_unlock(&fs_info->trans_lock);
1334 * update all the cowonly tree roots on disk
1336 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1338 struct btrfs_fs_info *fs_info = trans->fs_info;
1339 struct btrfs_root *gang[8];
1343 spin_lock(&fs_info->fs_roots_radix_lock);
1345 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1348 BTRFS_ROOT_TRANS_TAG);
1351 for (i = 0; i < ret; i++) {
1352 struct btrfs_root *root = gang[i];
1355 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1356 (unsigned long)root->root_key.objectid,
1357 BTRFS_ROOT_TRANS_TAG);
1358 spin_unlock(&fs_info->fs_roots_radix_lock);
1360 btrfs_free_log(trans, root);
1361 ret2 = btrfs_update_reloc_root(trans, root);
1365 /* see comments in should_cow_block() */
1366 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1367 smp_mb__after_atomic();
1369 if (root->commit_root != root->node) {
1370 list_add_tail(&root->dirty_list,
1371 &trans->transaction->switch_commits);
1372 btrfs_set_root_node(&root->root_item,
1376 ret2 = btrfs_update_root(trans, fs_info->tree_root,
1381 spin_lock(&fs_info->fs_roots_radix_lock);
1382 btrfs_qgroup_free_meta_all_pertrans(root);
1385 spin_unlock(&fs_info->fs_roots_radix_lock);
1390 * defrag a given btree.
1391 * Every leaf in the btree is read and defragged.
1393 int btrfs_defrag_root(struct btrfs_root *root)
1395 struct btrfs_fs_info *info = root->fs_info;
1396 struct btrfs_trans_handle *trans;
1399 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1403 trans = btrfs_start_transaction(root, 0);
1404 if (IS_ERR(trans)) {
1405 ret = PTR_ERR(trans);
1409 ret = btrfs_defrag_leaves(trans, root);
1411 btrfs_end_transaction(trans);
1412 btrfs_btree_balance_dirty(info);
1415 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1418 if (btrfs_defrag_cancelled(info)) {
1419 btrfs_debug(info, "defrag_root cancelled");
1424 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1429 * Do all special snapshot related qgroup dirty hack.
1431 * Will do all needed qgroup inherit and dirty hack like switch commit
1432 * roots inside one transaction and write all btree into disk, to make
1435 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1436 struct btrfs_root *src,
1437 struct btrfs_root *parent,
1438 struct btrfs_qgroup_inherit *inherit,
1441 struct btrfs_fs_info *fs_info = src->fs_info;
1445 * Save some performance in the case that qgroups are not
1446 * enabled. If this check races with the ioctl, rescan will
1449 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1453 * Ensure dirty @src will be committed. Or, after coming
1454 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1455 * recorded root will never be updated again, causing an outdated root
1458 ret = record_root_in_trans(trans, src, 1);
1463 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1464 * src root, so we must run the delayed refs here.
1466 * However this isn't particularly fool proof, because there's no
1467 * synchronization keeping us from changing the tree after this point
1468 * before we do the qgroup_inherit, or even from making changes while
1469 * we're doing the qgroup_inherit. But that's a problem for the future,
1470 * for now flush the delayed refs to narrow the race window where the
1471 * qgroup counters could end up wrong.
1473 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1475 btrfs_abort_transaction(trans, ret);
1480 * We are going to commit transaction, see btrfs_commit_transaction()
1481 * comment for reason locking tree_log_mutex
1483 mutex_lock(&fs_info->tree_log_mutex);
1485 ret = commit_fs_roots(trans);
1488 ret = btrfs_qgroup_account_extents(trans);
1492 /* Now qgroup are all updated, we can inherit it to new qgroups */
1493 ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1499 * Now we do a simplified commit transaction, which will:
1500 * 1) commit all subvolume and extent tree
1501 * To ensure all subvolume and extent tree have a valid
1502 * commit_root to accounting later insert_dir_item()
1503 * 2) write all btree blocks onto disk
1504 * This is to make sure later btree modification will be cowed
1505 * Or commit_root can be populated and cause wrong qgroup numbers
1506 * In this simplified commit, we don't really care about other trees
1507 * like chunk and root tree, as they won't affect qgroup.
1508 * And we don't write super to avoid half committed status.
1510 ret = commit_cowonly_roots(trans);
1513 switch_commit_roots(trans);
1514 ret = btrfs_write_and_wait_transaction(trans);
1516 btrfs_handle_fs_error(fs_info, ret,
1517 "Error while writing out transaction for qgroup");
1520 mutex_unlock(&fs_info->tree_log_mutex);
1523 * Force parent root to be updated, as we recorded it before so its
1524 * last_trans == cur_transid.
1525 * Or it won't be committed again onto disk after later
1529 ret = record_root_in_trans(trans, parent, 1);
1534 * new snapshots need to be created at a very specific time in the
1535 * transaction commit. This does the actual creation.
1538 * If the error which may affect the commitment of the current transaction
1539 * happens, we should return the error number. If the error which just affect
1540 * the creation of the pending snapshots, just return 0.
1542 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1543 struct btrfs_pending_snapshot *pending)
1546 struct btrfs_fs_info *fs_info = trans->fs_info;
1547 struct btrfs_key key;
1548 struct btrfs_root_item *new_root_item;
1549 struct btrfs_root *tree_root = fs_info->tree_root;
1550 struct btrfs_root *root = pending->root;
1551 struct btrfs_root *parent_root;
1552 struct btrfs_block_rsv *rsv;
1553 struct inode *parent_inode;
1554 struct btrfs_path *path;
1555 struct btrfs_dir_item *dir_item;
1556 struct dentry *dentry;
1557 struct extent_buffer *tmp;
1558 struct extent_buffer *old;
1559 struct timespec64 cur_time;
1566 ASSERT(pending->path);
1567 path = pending->path;
1569 ASSERT(pending->root_item);
1570 new_root_item = pending->root_item;
1572 pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1574 goto no_free_objectid;
1577 * Make qgroup to skip current new snapshot's qgroupid, as it is
1578 * accounted by later btrfs_qgroup_inherit().
1580 btrfs_set_skip_qgroup(trans, objectid);
1582 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1584 if (to_reserve > 0) {
1585 pending->error = btrfs_block_rsv_add(root,
1586 &pending->block_rsv,
1588 BTRFS_RESERVE_NO_FLUSH);
1590 goto clear_skip_qgroup;
1593 key.objectid = objectid;
1594 key.offset = (u64)-1;
1595 key.type = BTRFS_ROOT_ITEM_KEY;
1597 rsv = trans->block_rsv;
1598 trans->block_rsv = &pending->block_rsv;
1599 trans->bytes_reserved = trans->block_rsv->reserved;
1600 trace_btrfs_space_reservation(fs_info, "transaction",
1602 trans->bytes_reserved, 1);
1603 dentry = pending->dentry;
1604 parent_inode = pending->dir;
1605 parent_root = BTRFS_I(parent_inode)->root;
1606 ret = record_root_in_trans(trans, parent_root, 0);
1609 cur_time = current_time(parent_inode);
1612 * insert the directory item
1614 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1615 BUG_ON(ret); /* -ENOMEM */
1617 /* check if there is a file/dir which has the same name. */
1618 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1619 btrfs_ino(BTRFS_I(parent_inode)),
1620 dentry->d_name.name,
1621 dentry->d_name.len, 0);
1622 if (dir_item != NULL && !IS_ERR(dir_item)) {
1623 pending->error = -EEXIST;
1624 goto dir_item_existed;
1625 } else if (IS_ERR(dir_item)) {
1626 ret = PTR_ERR(dir_item);
1627 btrfs_abort_transaction(trans, ret);
1630 btrfs_release_path(path);
1633 * pull in the delayed directory update
1634 * and the delayed inode item
1635 * otherwise we corrupt the FS during
1638 ret = btrfs_run_delayed_items(trans);
1639 if (ret) { /* Transaction aborted */
1640 btrfs_abort_transaction(trans, ret);
1644 ret = record_root_in_trans(trans, root, 0);
1646 btrfs_abort_transaction(trans, ret);
1649 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1650 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1651 btrfs_check_and_init_root_item(new_root_item);
1653 root_flags = btrfs_root_flags(new_root_item);
1654 if (pending->readonly)
1655 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1657 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1658 btrfs_set_root_flags(new_root_item, root_flags);
1660 btrfs_set_root_generation_v2(new_root_item,
1662 generate_random_guid(new_root_item->uuid);
1663 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1665 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1666 memset(new_root_item->received_uuid, 0,
1667 sizeof(new_root_item->received_uuid));
1668 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1669 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1670 btrfs_set_root_stransid(new_root_item, 0);
1671 btrfs_set_root_rtransid(new_root_item, 0);
1673 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1674 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1675 btrfs_set_root_otransid(new_root_item, trans->transid);
1677 old = btrfs_lock_root_node(root);
1678 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1681 btrfs_tree_unlock(old);
1682 free_extent_buffer(old);
1683 btrfs_abort_transaction(trans, ret);
1687 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1688 /* clean up in any case */
1689 btrfs_tree_unlock(old);
1690 free_extent_buffer(old);
1692 btrfs_abort_transaction(trans, ret);
1695 /* see comments in should_cow_block() */
1696 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1699 btrfs_set_root_node(new_root_item, tmp);
1700 /* record when the snapshot was created in key.offset */
1701 key.offset = trans->transid;
1702 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1703 btrfs_tree_unlock(tmp);
1704 free_extent_buffer(tmp);
1706 btrfs_abort_transaction(trans, ret);
1711 * insert root back/forward references
1713 ret = btrfs_add_root_ref(trans, objectid,
1714 parent_root->root_key.objectid,
1715 btrfs_ino(BTRFS_I(parent_inode)), index,
1716 dentry->d_name.name, dentry->d_name.len);
1718 btrfs_abort_transaction(trans, ret);
1722 key.offset = (u64)-1;
1723 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1724 if (IS_ERR(pending->snap)) {
1725 ret = PTR_ERR(pending->snap);
1726 pending->snap = NULL;
1727 btrfs_abort_transaction(trans, ret);
1731 ret = btrfs_reloc_post_snapshot(trans, pending);
1733 btrfs_abort_transaction(trans, ret);
1738 * Do special qgroup accounting for snapshot, as we do some qgroup
1739 * snapshot hack to do fast snapshot.
1740 * To co-operate with that hack, we do hack again.
1741 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1743 ret = qgroup_account_snapshot(trans, root, parent_root,
1744 pending->inherit, objectid);
1748 ret = btrfs_insert_dir_item(trans, dentry->d_name.name,
1749 dentry->d_name.len, BTRFS_I(parent_inode),
1750 &key, BTRFS_FT_DIR, index);
1751 /* We have check then name at the beginning, so it is impossible. */
1752 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1754 btrfs_abort_transaction(trans, ret);
1758 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1759 dentry->d_name.len * 2);
1760 parent_inode->i_mtime = parent_inode->i_ctime =
1761 current_time(parent_inode);
1762 ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1764 btrfs_abort_transaction(trans, ret);
1767 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1768 BTRFS_UUID_KEY_SUBVOL,
1771 btrfs_abort_transaction(trans, ret);
1774 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1775 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1776 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1778 if (ret && ret != -EEXIST) {
1779 btrfs_abort_transaction(trans, ret);
1785 pending->error = ret;
1787 trans->block_rsv = rsv;
1788 trans->bytes_reserved = 0;
1790 btrfs_clear_skip_qgroup(trans);
1792 kfree(new_root_item);
1793 pending->root_item = NULL;
1794 btrfs_free_path(path);
1795 pending->path = NULL;
1801 * create all the snapshots we've scheduled for creation
1803 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1805 struct btrfs_pending_snapshot *pending, *next;
1806 struct list_head *head = &trans->transaction->pending_snapshots;
1809 list_for_each_entry_safe(pending, next, head, list) {
1810 list_del(&pending->list);
1811 ret = create_pending_snapshot(trans, pending);
1818 static void update_super_roots(struct btrfs_fs_info *fs_info)
1820 struct btrfs_root_item *root_item;
1821 struct btrfs_super_block *super;
1823 super = fs_info->super_copy;
1825 root_item = &fs_info->chunk_root->root_item;
1826 super->chunk_root = root_item->bytenr;
1827 super->chunk_root_generation = root_item->generation;
1828 super->chunk_root_level = root_item->level;
1830 root_item = &fs_info->tree_root->root_item;
1831 super->root = root_item->bytenr;
1832 super->generation = root_item->generation;
1833 super->root_level = root_item->level;
1834 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1835 super->cache_generation = root_item->generation;
1836 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1837 super->cache_generation = 0;
1838 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1839 super->uuid_tree_generation = root_item->generation;
1842 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1844 struct btrfs_transaction *trans;
1847 spin_lock(&info->trans_lock);
1848 trans = info->running_transaction;
1850 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1851 spin_unlock(&info->trans_lock);
1855 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1857 struct btrfs_transaction *trans;
1860 spin_lock(&info->trans_lock);
1861 trans = info->running_transaction;
1863 ret = is_transaction_blocked(trans);
1864 spin_unlock(&info->trans_lock);
1869 * wait for the current transaction commit to start and block subsequent
1872 static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info,
1873 struct btrfs_transaction *trans)
1875 wait_event(fs_info->transaction_blocked_wait,
1876 trans->state >= TRANS_STATE_COMMIT_START ||
1877 TRANS_ABORTED(trans));
1881 * wait for the current transaction to start and then become unblocked.
1884 static void wait_current_trans_commit_start_and_unblock(
1885 struct btrfs_fs_info *fs_info,
1886 struct btrfs_transaction *trans)
1888 wait_event(fs_info->transaction_wait,
1889 trans->state >= TRANS_STATE_UNBLOCKED ||
1890 TRANS_ABORTED(trans));
1894 * commit transactions asynchronously. once btrfs_commit_transaction_async
1895 * returns, any subsequent transaction will not be allowed to join.
1897 struct btrfs_async_commit {
1898 struct btrfs_trans_handle *newtrans;
1899 struct work_struct work;
1902 static void do_async_commit(struct work_struct *work)
1904 struct btrfs_async_commit *ac =
1905 container_of(work, struct btrfs_async_commit, work);
1908 * We've got freeze protection passed with the transaction.
1909 * Tell lockdep about it.
1911 if (ac->newtrans->type & __TRANS_FREEZABLE)
1912 __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1914 current->journal_info = ac->newtrans;
1916 btrfs_commit_transaction(ac->newtrans);
1920 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1921 int wait_for_unblock)
1923 struct btrfs_fs_info *fs_info = trans->fs_info;
1924 struct btrfs_async_commit *ac;
1925 struct btrfs_transaction *cur_trans;
1927 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1931 INIT_WORK(&ac->work, do_async_commit);
1932 ac->newtrans = btrfs_join_transaction(trans->root);
1933 if (IS_ERR(ac->newtrans)) {
1934 int err = PTR_ERR(ac->newtrans);
1939 /* take transaction reference */
1940 cur_trans = trans->transaction;
1941 refcount_inc(&cur_trans->use_count);
1943 btrfs_end_transaction(trans);
1946 * Tell lockdep we've released the freeze rwsem, since the
1947 * async commit thread will be the one to unlock it.
1949 if (ac->newtrans->type & __TRANS_FREEZABLE)
1950 __sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1952 schedule_work(&ac->work);
1954 /* wait for transaction to start and unblock */
1955 if (wait_for_unblock)
1956 wait_current_trans_commit_start_and_unblock(fs_info, cur_trans);
1958 wait_current_trans_commit_start(fs_info, cur_trans);
1960 if (current->journal_info == trans)
1961 current->journal_info = NULL;
1963 btrfs_put_transaction(cur_trans);
1968 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1970 struct btrfs_fs_info *fs_info = trans->fs_info;
1971 struct btrfs_transaction *cur_trans = trans->transaction;
1973 WARN_ON(refcount_read(&trans->use_count) > 1);
1975 btrfs_abort_transaction(trans, err);
1977 spin_lock(&fs_info->trans_lock);
1980 * If the transaction is removed from the list, it means this
1981 * transaction has been committed successfully, so it is impossible
1982 * to call the cleanup function.
1984 BUG_ON(list_empty(&cur_trans->list));
1986 if (cur_trans == fs_info->running_transaction) {
1987 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1988 spin_unlock(&fs_info->trans_lock);
1989 wait_event(cur_trans->writer_wait,
1990 atomic_read(&cur_trans->num_writers) == 1);
1992 spin_lock(&fs_info->trans_lock);
1996 * Now that we know no one else is still using the transaction we can
1997 * remove the transaction from the list of transactions. This avoids
1998 * the transaction kthread from cleaning up the transaction while some
1999 * other task is still using it, which could result in a use-after-free
2000 * on things like log trees, as it forces the transaction kthread to
2001 * wait for this transaction to be cleaned up by us.
2003 list_del_init(&cur_trans->list);
2005 spin_unlock(&fs_info->trans_lock);
2007 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
2009 spin_lock(&fs_info->trans_lock);
2010 if (cur_trans == fs_info->running_transaction)
2011 fs_info->running_transaction = NULL;
2012 spin_unlock(&fs_info->trans_lock);
2014 if (trans->type & __TRANS_FREEZABLE)
2015 sb_end_intwrite(fs_info->sb);
2016 btrfs_put_transaction(cur_trans);
2017 btrfs_put_transaction(cur_trans);
2019 trace_btrfs_transaction_commit(trans->root);
2021 if (current->journal_info == trans)
2022 current->journal_info = NULL;
2023 btrfs_scrub_cancel(fs_info);
2025 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2029 * Release reserved delayed ref space of all pending block groups of the
2030 * transaction and remove them from the list
2032 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2034 struct btrfs_fs_info *fs_info = trans->fs_info;
2035 struct btrfs_block_group *block_group, *tmp;
2037 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2038 btrfs_delayed_refs_rsv_release(fs_info, 1);
2039 list_del_init(&block_group->bg_list);
2043 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2046 * We use writeback_inodes_sb here because if we used
2047 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2048 * Currently are holding the fs freeze lock, if we do an async flush
2049 * we'll do btrfs_join_transaction() and deadlock because we need to
2050 * wait for the fs freeze lock. Using the direct flushing we benefit
2051 * from already being in a transaction and our join_transaction doesn't
2052 * have to re-take the fs freeze lock.
2054 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2055 writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2059 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2061 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2062 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2065 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2067 struct btrfs_fs_info *fs_info = trans->fs_info;
2068 struct btrfs_transaction *cur_trans = trans->transaction;
2069 struct btrfs_transaction *prev_trans = NULL;
2072 ASSERT(refcount_read(&trans->use_count) == 1);
2074 /* Stop the commit early if ->aborted is set */
2075 if (TRANS_ABORTED(cur_trans)) {
2076 ret = cur_trans->aborted;
2077 btrfs_end_transaction(trans);
2081 btrfs_trans_release_metadata(trans);
2082 trans->block_rsv = NULL;
2085 * We only want one transaction commit doing the flushing so we do not
2086 * waste a bunch of time on lock contention on the extent root node.
2088 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2089 &cur_trans->delayed_refs.flags)) {
2091 * Make a pass through all the delayed refs we have so far.
2092 * Any running threads may add more while we are here.
2094 ret = btrfs_run_delayed_refs(trans, 0);
2096 btrfs_end_transaction(trans);
2101 btrfs_create_pending_block_groups(trans);
2103 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2106 /* this mutex is also taken before trying to set
2107 * block groups readonly. We need to make sure
2108 * that nobody has set a block group readonly
2109 * after a extents from that block group have been
2110 * allocated for cache files. btrfs_set_block_group_ro
2111 * will wait for the transaction to commit if it
2112 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2114 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2115 * only one process starts all the block group IO. It wouldn't
2116 * hurt to have more than one go through, but there's no
2117 * real advantage to it either.
2119 mutex_lock(&fs_info->ro_block_group_mutex);
2120 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2123 mutex_unlock(&fs_info->ro_block_group_mutex);
2126 ret = btrfs_start_dirty_block_groups(trans);
2128 btrfs_end_transaction(trans);
2134 spin_lock(&fs_info->trans_lock);
2135 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2136 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2138 spin_unlock(&fs_info->trans_lock);
2139 refcount_inc(&cur_trans->use_count);
2141 if (trans->in_fsync)
2142 want_state = TRANS_STATE_SUPER_COMMITTED;
2143 ret = btrfs_end_transaction(trans);
2144 wait_for_commit(cur_trans, want_state);
2146 if (TRANS_ABORTED(cur_trans))
2147 ret = cur_trans->aborted;
2149 btrfs_put_transaction(cur_trans);
2154 cur_trans->state = TRANS_STATE_COMMIT_START;
2155 wake_up(&fs_info->transaction_blocked_wait);
2157 if (cur_trans->list.prev != &fs_info->trans_list) {
2158 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2160 if (trans->in_fsync)
2161 want_state = TRANS_STATE_SUPER_COMMITTED;
2163 prev_trans = list_entry(cur_trans->list.prev,
2164 struct btrfs_transaction, list);
2165 if (prev_trans->state < want_state) {
2166 refcount_inc(&prev_trans->use_count);
2167 spin_unlock(&fs_info->trans_lock);
2169 wait_for_commit(prev_trans, want_state);
2171 ret = READ_ONCE(prev_trans->aborted);
2173 btrfs_put_transaction(prev_trans);
2175 goto cleanup_transaction;
2177 spin_unlock(&fs_info->trans_lock);
2180 spin_unlock(&fs_info->trans_lock);
2182 * The previous transaction was aborted and was already removed
2183 * from the list of transactions at fs_info->trans_list. So we
2184 * abort to prevent writing a new superblock that reflects a
2185 * corrupt state (pointing to trees with unwritten nodes/leafs).
2187 if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) {
2189 goto cleanup_transaction;
2193 extwriter_counter_dec(cur_trans, trans->type);
2195 ret = btrfs_start_delalloc_flush(fs_info);
2197 goto cleanup_transaction;
2199 ret = btrfs_run_delayed_items(trans);
2201 goto cleanup_transaction;
2203 wait_event(cur_trans->writer_wait,
2204 extwriter_counter_read(cur_trans) == 0);
2206 /* some pending stuffs might be added after the previous flush. */
2207 ret = btrfs_run_delayed_items(trans);
2209 goto cleanup_transaction;
2211 btrfs_wait_delalloc_flush(fs_info);
2214 * Wait for all ordered extents started by a fast fsync that joined this
2215 * transaction. Otherwise if this transaction commits before the ordered
2216 * extents complete we lose logged data after a power failure.
2218 wait_event(cur_trans->pending_wait,
2219 atomic_read(&cur_trans->pending_ordered) == 0);
2221 btrfs_scrub_pause(fs_info);
2223 * Ok now we need to make sure to block out any other joins while we
2224 * commit the transaction. We could have started a join before setting
2225 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2227 spin_lock(&fs_info->trans_lock);
2228 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2229 spin_unlock(&fs_info->trans_lock);
2230 wait_event(cur_trans->writer_wait,
2231 atomic_read(&cur_trans->num_writers) == 1);
2233 if (TRANS_ABORTED(cur_trans)) {
2234 ret = cur_trans->aborted;
2235 goto scrub_continue;
2238 * the reloc mutex makes sure that we stop
2239 * the balancing code from coming in and moving
2240 * extents around in the middle of the commit
2242 mutex_lock(&fs_info->reloc_mutex);
2245 * We needn't worry about the delayed items because we will
2246 * deal with them in create_pending_snapshot(), which is the
2247 * core function of the snapshot creation.
2249 ret = create_pending_snapshots(trans);
2254 * We insert the dir indexes of the snapshots and update the inode
2255 * of the snapshots' parents after the snapshot creation, so there
2256 * are some delayed items which are not dealt with. Now deal with
2259 * We needn't worry that this operation will corrupt the snapshots,
2260 * because all the tree which are snapshoted will be forced to COW
2261 * the nodes and leaves.
2263 ret = btrfs_run_delayed_items(trans);
2267 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2272 * make sure none of the code above managed to slip in a
2275 btrfs_assert_delayed_root_empty(fs_info);
2277 WARN_ON(cur_trans != trans->transaction);
2279 /* btrfs_commit_tree_roots is responsible for getting the
2280 * various roots consistent with each other. Every pointer
2281 * in the tree of tree roots has to point to the most up to date
2282 * root for every subvolume and other tree. So, we have to keep
2283 * the tree logging code from jumping in and changing any
2286 * At this point in the commit, there can't be any tree-log
2287 * writers, but a little lower down we drop the trans mutex
2288 * and let new people in. By holding the tree_log_mutex
2289 * from now until after the super is written, we avoid races
2290 * with the tree-log code.
2292 mutex_lock(&fs_info->tree_log_mutex);
2294 ret = commit_fs_roots(trans);
2296 goto unlock_tree_log;
2299 * Since the transaction is done, we can apply the pending changes
2300 * before the next transaction.
2302 btrfs_apply_pending_changes(fs_info);
2304 /* commit_fs_roots gets rid of all the tree log roots, it is now
2305 * safe to free the root of tree log roots
2307 btrfs_free_log_root_tree(trans, fs_info);
2310 * Since fs roots are all committed, we can get a quite accurate
2311 * new_roots. So let's do quota accounting.
2313 ret = btrfs_qgroup_account_extents(trans);
2315 goto unlock_tree_log;
2317 ret = commit_cowonly_roots(trans);
2319 goto unlock_tree_log;
2322 * The tasks which save the space cache and inode cache may also
2323 * update ->aborted, check it.
2325 if (TRANS_ABORTED(cur_trans)) {
2326 ret = cur_trans->aborted;
2327 goto unlock_tree_log;
2330 cur_trans = fs_info->running_transaction;
2332 btrfs_set_root_node(&fs_info->tree_root->root_item,
2333 fs_info->tree_root->node);
2334 list_add_tail(&fs_info->tree_root->dirty_list,
2335 &cur_trans->switch_commits);
2337 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2338 fs_info->chunk_root->node);
2339 list_add_tail(&fs_info->chunk_root->dirty_list,
2340 &cur_trans->switch_commits);
2342 switch_commit_roots(trans);
2344 ASSERT(list_empty(&cur_trans->dirty_bgs));
2345 ASSERT(list_empty(&cur_trans->io_bgs));
2346 update_super_roots(fs_info);
2348 btrfs_set_super_log_root(fs_info->super_copy, 0);
2349 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2350 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2351 sizeof(*fs_info->super_copy));
2353 btrfs_commit_device_sizes(cur_trans);
2355 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2356 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2358 btrfs_trans_release_chunk_metadata(trans);
2360 spin_lock(&fs_info->trans_lock);
2361 cur_trans->state = TRANS_STATE_UNBLOCKED;
2362 fs_info->running_transaction = NULL;
2363 spin_unlock(&fs_info->trans_lock);
2364 mutex_unlock(&fs_info->reloc_mutex);
2366 wake_up(&fs_info->transaction_wait);
2368 ret = btrfs_write_and_wait_transaction(trans);
2370 btrfs_handle_fs_error(fs_info, ret,
2371 "Error while writing out transaction");
2373 * reloc_mutex has been unlocked, tree_log_mutex is still held
2374 * but we can't jump to unlock_tree_log causing double unlock
2376 mutex_unlock(&fs_info->tree_log_mutex);
2377 goto scrub_continue;
2381 * At this point, we should have written all the tree blocks allocated
2382 * in this transaction. So it's now safe to free the redirtyied extent
2385 btrfs_free_redirty_list(cur_trans);
2387 ret = write_all_supers(fs_info, 0);
2389 * the super is written, we can safely allow the tree-loggers
2390 * to go about their business
2392 mutex_unlock(&fs_info->tree_log_mutex);
2394 goto scrub_continue;
2397 * We needn't acquire the lock here because there is no other task
2398 * which can change it.
2400 cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2401 wake_up(&cur_trans->commit_wait);
2403 btrfs_finish_extent_commit(trans);
2405 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2406 btrfs_clear_space_info_full(fs_info);
2408 fs_info->last_trans_committed = cur_trans->transid;
2410 * We needn't acquire the lock here because there is no other task
2411 * which can change it.
2413 cur_trans->state = TRANS_STATE_COMPLETED;
2414 wake_up(&cur_trans->commit_wait);
2416 spin_lock(&fs_info->trans_lock);
2417 list_del_init(&cur_trans->list);
2418 spin_unlock(&fs_info->trans_lock);
2420 btrfs_put_transaction(cur_trans);
2421 btrfs_put_transaction(cur_trans);
2423 if (trans->type & __TRANS_FREEZABLE)
2424 sb_end_intwrite(fs_info->sb);
2426 trace_btrfs_transaction_commit(trans->root);
2428 btrfs_scrub_continue(fs_info);
2430 if (current->journal_info == trans)
2431 current->journal_info = NULL;
2433 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2438 mutex_unlock(&fs_info->tree_log_mutex);
2440 mutex_unlock(&fs_info->reloc_mutex);
2442 btrfs_scrub_continue(fs_info);
2443 cleanup_transaction:
2444 btrfs_trans_release_metadata(trans);
2445 btrfs_cleanup_pending_block_groups(trans);
2446 btrfs_trans_release_chunk_metadata(trans);
2447 trans->block_rsv = NULL;
2448 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2449 if (current->journal_info == trans)
2450 current->journal_info = NULL;
2451 cleanup_transaction(trans, ret);
2457 * return < 0 if error
2458 * 0 if there are no more dead_roots at the time of call
2459 * 1 there are more to be processed, call me again
2461 * The return value indicates there are certainly more snapshots to delete, but
2462 * if there comes a new one during processing, it may return 0. We don't mind,
2463 * because btrfs_commit_super will poke cleaner thread and it will process it a
2464 * few seconds later.
2466 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2469 struct btrfs_fs_info *fs_info = root->fs_info;
2471 spin_lock(&fs_info->trans_lock);
2472 if (list_empty(&fs_info->dead_roots)) {
2473 spin_unlock(&fs_info->trans_lock);
2476 root = list_first_entry(&fs_info->dead_roots,
2477 struct btrfs_root, root_list);
2478 list_del_init(&root->root_list);
2479 spin_unlock(&fs_info->trans_lock);
2481 btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2483 btrfs_kill_all_delayed_nodes(root);
2485 if (btrfs_header_backref_rev(root->node) <
2486 BTRFS_MIXED_BACKREF_REV)
2487 ret = btrfs_drop_snapshot(root, 0, 0);
2489 ret = btrfs_drop_snapshot(root, 1, 0);
2491 btrfs_put_root(root);
2492 return (ret < 0) ? 0 : 1;
2495 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2500 prev = xchg(&fs_info->pending_changes, 0);
2504 bit = 1 << BTRFS_PENDING_COMMIT;
2506 btrfs_debug(fs_info, "pending commit done");
2511 "unknown pending changes left 0x%lx, ignoring", prev);