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 (BTRFS_FS_ERROR(fs_info)) {
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 (BTRFS_FS_ERROR(fs_info)) {
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 if (BTRFS_FS_ERROR(fs_info))
583 return ERR_PTR(-EROFS);
585 if (current->journal_info) {
586 WARN_ON(type & TRANS_EXTWRITERS);
587 h = current->journal_info;
588 refcount_inc(&h->use_count);
589 WARN_ON(refcount_read(&h->use_count) > 2);
590 h->orig_rsv = h->block_rsv;
596 * Do the reservation before we join the transaction so we can do all
597 * the appropriate flushing if need be.
599 if (num_items && root != fs_info->chunk_root) {
600 struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
601 u64 delayed_refs_bytes = 0;
603 qgroup_reserved = num_items * fs_info->nodesize;
604 ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
610 * We want to reserve all the bytes we may need all at once, so
611 * we only do 1 enospc flushing cycle per transaction start. We
612 * accomplish this by simply assuming we'll do 2 x num_items
613 * worth of delayed refs updates in this trans handle, and
614 * refill that amount for whatever is missing in the reserve.
616 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
617 if (flush == BTRFS_RESERVE_FLUSH_ALL &&
618 delayed_refs_rsv->full == 0) {
619 delayed_refs_bytes = num_bytes;
624 * Do the reservation for the relocation root creation
626 if (need_reserve_reloc_root(root)) {
627 num_bytes += fs_info->nodesize;
628 reloc_reserved = true;
631 ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
634 if (delayed_refs_bytes) {
635 btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
637 num_bytes -= delayed_refs_bytes;
640 if (rsv->space_info->force_alloc)
641 do_chunk_alloc = true;
642 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
643 !delayed_refs_rsv->full) {
645 * Some people call with btrfs_start_transaction(root, 0)
646 * because they can be throttled, but have some other mechanism
647 * for reserving space. We still want these guys to refill the
648 * delayed block_rsv so just add 1 items worth of reservation
651 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
656 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
663 * If we are JOIN_NOLOCK we're already committing a transaction and
664 * waiting on this guy, so we don't need to do the sb_start_intwrite
665 * because we're already holding a ref. We need this because we could
666 * have raced in and did an fsync() on a file which can kick a commit
667 * and then we deadlock with somebody doing a freeze.
669 * If we are ATTACH, it means we just want to catch the current
670 * transaction and commit it, so we needn't do sb_start_intwrite().
672 if (type & __TRANS_FREEZABLE)
673 sb_start_intwrite(fs_info->sb);
675 if (may_wait_transaction(fs_info, type))
676 wait_current_trans(fs_info);
679 ret = join_transaction(fs_info, type);
681 wait_current_trans(fs_info);
682 if (unlikely(type == TRANS_ATTACH ||
683 type == TRANS_JOIN_NOSTART))
686 } while (ret == -EBUSY);
691 cur_trans = fs_info->running_transaction;
693 h->transid = cur_trans->transid;
694 h->transaction = cur_trans;
696 refcount_set(&h->use_count, 1);
697 h->fs_info = root->fs_info;
700 INIT_LIST_HEAD(&h->new_bgs);
703 if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
704 may_wait_transaction(fs_info, type)) {
705 current->journal_info = h;
706 btrfs_commit_transaction(h);
711 trace_btrfs_space_reservation(fs_info, "transaction",
712 h->transid, num_bytes, 1);
713 h->block_rsv = &fs_info->trans_block_rsv;
714 h->bytes_reserved = num_bytes;
715 h->reloc_reserved = reloc_reserved;
719 if (!current->journal_info)
720 current->journal_info = h;
723 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
724 * ALLOC_FORCE the first run through, and then we won't allocate for
725 * anybody else who races in later. We don't care about the return
728 if (do_chunk_alloc && num_bytes) {
729 u64 flags = h->block_rsv->space_info->flags;
731 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
732 CHUNK_ALLOC_NO_FORCE);
736 * btrfs_record_root_in_trans() needs to alloc new extents, and may
737 * call btrfs_join_transaction() while we're also starting a
740 * Thus it need to be called after current->journal_info initialized,
741 * or we can deadlock.
743 ret = btrfs_record_root_in_trans(h, root);
746 * The transaction handle is fully initialized and linked with
747 * other structures so it needs to be ended in case of errors,
750 btrfs_end_transaction(h);
757 if (type & __TRANS_FREEZABLE)
758 sb_end_intwrite(fs_info->sb);
759 kmem_cache_free(btrfs_trans_handle_cachep, h);
762 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
765 btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
769 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
770 unsigned int num_items)
772 return start_transaction(root, num_items, TRANS_START,
773 BTRFS_RESERVE_FLUSH_ALL, true);
776 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
777 struct btrfs_root *root,
778 unsigned int num_items)
780 return start_transaction(root, num_items, TRANS_START,
781 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
784 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
786 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
790 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
792 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
793 BTRFS_RESERVE_NO_FLUSH, true);
797 * Similar to regular join but it never starts a transaction when none is
798 * running or after waiting for the current one to finish.
800 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
802 return start_transaction(root, 0, TRANS_JOIN_NOSTART,
803 BTRFS_RESERVE_NO_FLUSH, true);
807 * btrfs_attach_transaction() - catch the running transaction
809 * It is used when we want to commit the current the transaction, but
810 * don't want to start a new one.
812 * Note: If this function return -ENOENT, it just means there is no
813 * running transaction. But it is possible that the inactive transaction
814 * is still in the memory, not fully on disk. If you hope there is no
815 * inactive transaction in the fs when -ENOENT is returned, you should
817 * btrfs_attach_transaction_barrier()
819 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
821 return start_transaction(root, 0, TRANS_ATTACH,
822 BTRFS_RESERVE_NO_FLUSH, true);
826 * btrfs_attach_transaction_barrier() - catch the running transaction
828 * It is similar to the above function, the difference is this one
829 * will wait for all the inactive transactions until they fully
832 struct btrfs_trans_handle *
833 btrfs_attach_transaction_barrier(struct btrfs_root *root)
835 struct btrfs_trans_handle *trans;
837 trans = start_transaction(root, 0, TRANS_ATTACH,
838 BTRFS_RESERVE_NO_FLUSH, true);
839 if (trans == ERR_PTR(-ENOENT))
840 btrfs_wait_for_commit(root->fs_info, 0);
845 /* Wait for a transaction commit to reach at least the given state. */
846 static noinline void wait_for_commit(struct btrfs_transaction *commit,
847 const enum btrfs_trans_state min_state)
849 wait_event(commit->commit_wait, commit->state >= min_state);
852 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
854 struct btrfs_transaction *cur_trans = NULL, *t;
858 if (transid <= fs_info->last_trans_committed)
861 /* find specified transaction */
862 spin_lock(&fs_info->trans_lock);
863 list_for_each_entry(t, &fs_info->trans_list, list) {
864 if (t->transid == transid) {
866 refcount_inc(&cur_trans->use_count);
870 if (t->transid > transid) {
875 spin_unlock(&fs_info->trans_lock);
878 * The specified transaction doesn't exist, or we
879 * raced with btrfs_commit_transaction
882 if (transid > fs_info->last_trans_committed)
887 /* find newest transaction that is committing | committed */
888 spin_lock(&fs_info->trans_lock);
889 list_for_each_entry_reverse(t, &fs_info->trans_list,
891 if (t->state >= TRANS_STATE_COMMIT_START) {
892 if (t->state == TRANS_STATE_COMPLETED)
895 refcount_inc(&cur_trans->use_count);
899 spin_unlock(&fs_info->trans_lock);
901 goto out; /* nothing committing|committed */
904 wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
905 btrfs_put_transaction(cur_trans);
910 void btrfs_throttle(struct btrfs_fs_info *fs_info)
912 wait_current_trans(fs_info);
915 static bool should_end_transaction(struct btrfs_trans_handle *trans)
917 struct btrfs_fs_info *fs_info = trans->fs_info;
919 if (btrfs_check_space_for_delayed_refs(fs_info))
922 return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
925 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
927 struct btrfs_transaction *cur_trans = trans->transaction;
929 if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
930 test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
933 return should_end_transaction(trans);
936 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
939 struct btrfs_fs_info *fs_info = trans->fs_info;
941 if (!trans->block_rsv) {
942 ASSERT(!trans->bytes_reserved);
946 if (!trans->bytes_reserved)
949 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
950 trace_btrfs_space_reservation(fs_info, "transaction",
951 trans->transid, trans->bytes_reserved, 0);
952 btrfs_block_rsv_release(fs_info, trans->block_rsv,
953 trans->bytes_reserved, NULL);
954 trans->bytes_reserved = 0;
957 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
960 struct btrfs_fs_info *info = trans->fs_info;
961 struct btrfs_transaction *cur_trans = trans->transaction;
964 if (refcount_read(&trans->use_count) > 1) {
965 refcount_dec(&trans->use_count);
966 trans->block_rsv = trans->orig_rsv;
970 btrfs_trans_release_metadata(trans);
971 trans->block_rsv = NULL;
973 btrfs_create_pending_block_groups(trans);
975 btrfs_trans_release_chunk_metadata(trans);
977 if (trans->type & __TRANS_FREEZABLE)
978 sb_end_intwrite(info->sb);
980 WARN_ON(cur_trans != info->running_transaction);
981 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
982 atomic_dec(&cur_trans->num_writers);
983 extwriter_counter_dec(cur_trans, trans->type);
985 cond_wake_up(&cur_trans->writer_wait);
986 btrfs_put_transaction(cur_trans);
988 if (current->journal_info == trans)
989 current->journal_info = NULL;
992 btrfs_run_delayed_iputs(info);
994 if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) {
995 wake_up_process(info->transaction_kthread);
996 if (TRANS_ABORTED(trans))
997 err = trans->aborted;
1002 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1006 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1008 return __btrfs_end_transaction(trans, 0);
1011 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1013 return __btrfs_end_transaction(trans, 1);
1017 * when btree blocks are allocated, they have some corresponding bits set for
1018 * them in one of two extent_io trees. This is used to make sure all of
1019 * those extents are sent to disk but does not wait on them
1021 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1022 struct extent_io_tree *dirty_pages, int mark)
1026 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1027 struct extent_state *cached_state = NULL;
1031 atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1032 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1033 mark, &cached_state)) {
1034 bool wait_writeback = false;
1036 err = convert_extent_bit(dirty_pages, start, end,
1038 mark, &cached_state);
1040 * convert_extent_bit can return -ENOMEM, which is most of the
1041 * time a temporary error. So when it happens, ignore the error
1042 * and wait for writeback of this range to finish - because we
1043 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1044 * to __btrfs_wait_marked_extents() would not know that
1045 * writeback for this range started and therefore wouldn't
1046 * wait for it to finish - we don't want to commit a
1047 * superblock that points to btree nodes/leafs for which
1048 * writeback hasn't finished yet (and without errors).
1049 * We cleanup any entries left in the io tree when committing
1050 * the transaction (through extent_io_tree_release()).
1052 if (err == -ENOMEM) {
1054 wait_writeback = true;
1057 err = filemap_fdatawrite_range(mapping, start, end);
1060 else if (wait_writeback)
1061 werr = filemap_fdatawait_range(mapping, start, end);
1062 free_extent_state(cached_state);
1063 cached_state = NULL;
1067 atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1072 * when btree blocks are allocated, they have some corresponding bits set for
1073 * them in one of two extent_io trees. This is used to make sure all of
1074 * those extents are on disk for transaction or log commit. We wait
1075 * on all the pages and clear them from the dirty pages state tree
1077 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1078 struct extent_io_tree *dirty_pages)
1082 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1083 struct extent_state *cached_state = NULL;
1087 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1088 EXTENT_NEED_WAIT, &cached_state)) {
1090 * Ignore -ENOMEM errors returned by clear_extent_bit().
1091 * When committing the transaction, we'll remove any entries
1092 * left in the io tree. For a log commit, we don't remove them
1093 * after committing the log because the tree can be accessed
1094 * concurrently - we do it only at transaction commit time when
1095 * it's safe to do it (through extent_io_tree_release()).
1097 err = clear_extent_bit(dirty_pages, start, end,
1098 EXTENT_NEED_WAIT, 0, 0, &cached_state);
1102 err = filemap_fdatawait_range(mapping, start, end);
1105 free_extent_state(cached_state);
1106 cached_state = NULL;
1115 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1116 struct extent_io_tree *dirty_pages)
1118 bool errors = false;
1121 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1122 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1130 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1132 struct btrfs_fs_info *fs_info = log_root->fs_info;
1133 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1134 bool errors = false;
1137 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1139 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1140 if ((mark & EXTENT_DIRTY) &&
1141 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1144 if ((mark & EXTENT_NEW) &&
1145 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1154 * When btree blocks are allocated the corresponding extents are marked dirty.
1155 * This function ensures such extents are persisted on disk for transaction or
1158 * @trans: transaction whose dirty pages we'd like to write
1160 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1164 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1165 struct btrfs_fs_info *fs_info = trans->fs_info;
1166 struct blk_plug plug;
1168 blk_start_plug(&plug);
1169 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1170 blk_finish_plug(&plug);
1171 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1173 extent_io_tree_release(&trans->transaction->dirty_pages);
1184 * this is used to update the root pointer in the tree of tree roots.
1186 * But, in the case of the extent allocation tree, updating the root
1187 * pointer may allocate blocks which may change the root of the extent
1190 * So, this loops and repeats and makes sure the cowonly root didn't
1191 * change while the root pointer was being updated in the metadata.
1193 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1194 struct btrfs_root *root)
1197 u64 old_root_bytenr;
1199 struct btrfs_fs_info *fs_info = root->fs_info;
1200 struct btrfs_root *tree_root = fs_info->tree_root;
1202 old_root_used = btrfs_root_used(&root->root_item);
1205 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1206 if (old_root_bytenr == root->node->start &&
1207 old_root_used == btrfs_root_used(&root->root_item))
1210 btrfs_set_root_node(&root->root_item, root->node);
1211 ret = btrfs_update_root(trans, tree_root,
1217 old_root_used = btrfs_root_used(&root->root_item);
1224 * update all the cowonly tree roots on disk
1226 * The error handling in this function may not be obvious. Any of the
1227 * failures will cause the file system to go offline. We still need
1228 * to clean up the delayed refs.
1230 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1232 struct btrfs_fs_info *fs_info = trans->fs_info;
1233 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1234 struct list_head *io_bgs = &trans->transaction->io_bgs;
1235 struct list_head *next;
1236 struct extent_buffer *eb;
1239 eb = btrfs_lock_root_node(fs_info->tree_root);
1240 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1241 0, &eb, BTRFS_NESTING_COW);
1242 btrfs_tree_unlock(eb);
1243 free_extent_buffer(eb);
1248 ret = btrfs_run_dev_stats(trans);
1251 ret = btrfs_run_dev_replace(trans);
1254 ret = btrfs_run_qgroups(trans);
1258 ret = btrfs_setup_space_cache(trans);
1263 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1264 struct btrfs_root *root;
1265 next = fs_info->dirty_cowonly_roots.next;
1266 list_del_init(next);
1267 root = list_entry(next, struct btrfs_root, dirty_list);
1268 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1270 if (root != fs_info->extent_root)
1271 list_add_tail(&root->dirty_list,
1272 &trans->transaction->switch_commits);
1273 ret = update_cowonly_root(trans, root);
1278 /* Now flush any delayed refs generated by updating all of the roots */
1279 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1283 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1284 ret = btrfs_write_dirty_block_groups(trans);
1289 * We're writing the dirty block groups, which could generate
1290 * delayed refs, which could generate more dirty block groups,
1291 * so we want to keep this flushing in this loop to make sure
1292 * everything gets run.
1294 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1299 if (!list_empty(&fs_info->dirty_cowonly_roots))
1302 list_add_tail(&fs_info->extent_root->dirty_list,
1303 &trans->transaction->switch_commits);
1305 /* Update dev-replace pointer once everything is committed */
1306 fs_info->dev_replace.committed_cursor_left =
1307 fs_info->dev_replace.cursor_left_last_write_of_item;
1313 * dead roots are old snapshots that need to be deleted. This allocates
1314 * a dirty root struct and adds it into the list of dead roots that need to
1317 void btrfs_add_dead_root(struct btrfs_root *root)
1319 struct btrfs_fs_info *fs_info = root->fs_info;
1321 spin_lock(&fs_info->trans_lock);
1322 if (list_empty(&root->root_list)) {
1323 btrfs_grab_root(root);
1324 list_add_tail(&root->root_list, &fs_info->dead_roots);
1326 spin_unlock(&fs_info->trans_lock);
1330 * update all the cowonly tree roots on disk
1332 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1334 struct btrfs_fs_info *fs_info = trans->fs_info;
1335 struct btrfs_root *gang[8];
1339 spin_lock(&fs_info->fs_roots_radix_lock);
1341 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1344 BTRFS_ROOT_TRANS_TAG);
1347 for (i = 0; i < ret; i++) {
1348 struct btrfs_root *root = gang[i];
1351 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1352 (unsigned long)root->root_key.objectid,
1353 BTRFS_ROOT_TRANS_TAG);
1354 spin_unlock(&fs_info->fs_roots_radix_lock);
1356 btrfs_free_log(trans, root);
1357 ret2 = btrfs_update_reloc_root(trans, root);
1361 /* see comments in should_cow_block() */
1362 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1363 smp_mb__after_atomic();
1365 if (root->commit_root != root->node) {
1366 list_add_tail(&root->dirty_list,
1367 &trans->transaction->switch_commits);
1368 btrfs_set_root_node(&root->root_item,
1372 ret2 = btrfs_update_root(trans, fs_info->tree_root,
1377 spin_lock(&fs_info->fs_roots_radix_lock);
1378 btrfs_qgroup_free_meta_all_pertrans(root);
1381 spin_unlock(&fs_info->fs_roots_radix_lock);
1386 * defrag a given btree.
1387 * Every leaf in the btree is read and defragged.
1389 int btrfs_defrag_root(struct btrfs_root *root)
1391 struct btrfs_fs_info *info = root->fs_info;
1392 struct btrfs_trans_handle *trans;
1395 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1399 trans = btrfs_start_transaction(root, 0);
1400 if (IS_ERR(trans)) {
1401 ret = PTR_ERR(trans);
1405 ret = btrfs_defrag_leaves(trans, root);
1407 btrfs_end_transaction(trans);
1408 btrfs_btree_balance_dirty(info);
1411 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1414 if (btrfs_defrag_cancelled(info)) {
1415 btrfs_debug(info, "defrag_root cancelled");
1420 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1425 * Do all special snapshot related qgroup dirty hack.
1427 * Will do all needed qgroup inherit and dirty hack like switch commit
1428 * roots inside one transaction and write all btree into disk, to make
1431 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1432 struct btrfs_root *src,
1433 struct btrfs_root *parent,
1434 struct btrfs_qgroup_inherit *inherit,
1437 struct btrfs_fs_info *fs_info = src->fs_info;
1441 * Save some performance in the case that qgroups are not
1442 * enabled. If this check races with the ioctl, rescan will
1445 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1449 * Ensure dirty @src will be committed. Or, after coming
1450 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1451 * recorded root will never be updated again, causing an outdated root
1454 ret = record_root_in_trans(trans, src, 1);
1459 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1460 * src root, so we must run the delayed refs here.
1462 * However this isn't particularly fool proof, because there's no
1463 * synchronization keeping us from changing the tree after this point
1464 * before we do the qgroup_inherit, or even from making changes while
1465 * we're doing the qgroup_inherit. But that's a problem for the future,
1466 * for now flush the delayed refs to narrow the race window where the
1467 * qgroup counters could end up wrong.
1469 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1471 btrfs_abort_transaction(trans, ret);
1476 * We are going to commit transaction, see btrfs_commit_transaction()
1477 * comment for reason locking tree_log_mutex
1479 mutex_lock(&fs_info->tree_log_mutex);
1481 ret = commit_fs_roots(trans);
1484 ret = btrfs_qgroup_account_extents(trans);
1488 /* Now qgroup are all updated, we can inherit it to new qgroups */
1489 ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1495 * Now we do a simplified commit transaction, which will:
1496 * 1) commit all subvolume and extent tree
1497 * To ensure all subvolume and extent tree have a valid
1498 * commit_root to accounting later insert_dir_item()
1499 * 2) write all btree blocks onto disk
1500 * This is to make sure later btree modification will be cowed
1501 * Or commit_root can be populated and cause wrong qgroup numbers
1502 * In this simplified commit, we don't really care about other trees
1503 * like chunk and root tree, as they won't affect qgroup.
1504 * And we don't write super to avoid half committed status.
1506 ret = commit_cowonly_roots(trans);
1509 switch_commit_roots(trans);
1510 ret = btrfs_write_and_wait_transaction(trans);
1512 btrfs_handle_fs_error(fs_info, ret,
1513 "Error while writing out transaction for qgroup");
1516 mutex_unlock(&fs_info->tree_log_mutex);
1519 * Force parent root to be updated, as we recorded it before so its
1520 * last_trans == cur_transid.
1521 * Or it won't be committed again onto disk after later
1525 ret = record_root_in_trans(trans, parent, 1);
1530 * new snapshots need to be created at a very specific time in the
1531 * transaction commit. This does the actual creation.
1534 * If the error which may affect the commitment of the current transaction
1535 * happens, we should return the error number. If the error which just affect
1536 * the creation of the pending snapshots, just return 0.
1538 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1539 struct btrfs_pending_snapshot *pending)
1542 struct btrfs_fs_info *fs_info = trans->fs_info;
1543 struct btrfs_key key;
1544 struct btrfs_root_item *new_root_item;
1545 struct btrfs_root *tree_root = fs_info->tree_root;
1546 struct btrfs_root *root = pending->root;
1547 struct btrfs_root *parent_root;
1548 struct btrfs_block_rsv *rsv;
1549 struct inode *parent_inode;
1550 struct btrfs_path *path;
1551 struct btrfs_dir_item *dir_item;
1552 struct dentry *dentry;
1553 struct extent_buffer *tmp;
1554 struct extent_buffer *old;
1555 struct timespec64 cur_time;
1562 ASSERT(pending->path);
1563 path = pending->path;
1565 ASSERT(pending->root_item);
1566 new_root_item = pending->root_item;
1568 pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1570 goto no_free_objectid;
1573 * Make qgroup to skip current new snapshot's qgroupid, as it is
1574 * accounted by later btrfs_qgroup_inherit().
1576 btrfs_set_skip_qgroup(trans, objectid);
1578 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1580 if (to_reserve > 0) {
1581 pending->error = btrfs_block_rsv_add(root,
1582 &pending->block_rsv,
1584 BTRFS_RESERVE_NO_FLUSH);
1586 goto clear_skip_qgroup;
1589 key.objectid = objectid;
1590 key.offset = (u64)-1;
1591 key.type = BTRFS_ROOT_ITEM_KEY;
1593 rsv = trans->block_rsv;
1594 trans->block_rsv = &pending->block_rsv;
1595 trans->bytes_reserved = trans->block_rsv->reserved;
1596 trace_btrfs_space_reservation(fs_info, "transaction",
1598 trans->bytes_reserved, 1);
1599 dentry = pending->dentry;
1600 parent_inode = pending->dir;
1601 parent_root = BTRFS_I(parent_inode)->root;
1602 ret = record_root_in_trans(trans, parent_root, 0);
1605 cur_time = current_time(parent_inode);
1608 * insert the directory item
1610 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1611 BUG_ON(ret); /* -ENOMEM */
1613 /* check if there is a file/dir which has the same name. */
1614 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1615 btrfs_ino(BTRFS_I(parent_inode)),
1616 dentry->d_name.name,
1617 dentry->d_name.len, 0);
1618 if (dir_item != NULL && !IS_ERR(dir_item)) {
1619 pending->error = -EEXIST;
1620 goto dir_item_existed;
1621 } else if (IS_ERR(dir_item)) {
1622 ret = PTR_ERR(dir_item);
1623 btrfs_abort_transaction(trans, ret);
1626 btrfs_release_path(path);
1629 * pull in the delayed directory update
1630 * and the delayed inode item
1631 * otherwise we corrupt the FS during
1634 ret = btrfs_run_delayed_items(trans);
1635 if (ret) { /* Transaction aborted */
1636 btrfs_abort_transaction(trans, ret);
1640 ret = record_root_in_trans(trans, root, 0);
1642 btrfs_abort_transaction(trans, ret);
1645 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1646 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1647 btrfs_check_and_init_root_item(new_root_item);
1649 root_flags = btrfs_root_flags(new_root_item);
1650 if (pending->readonly)
1651 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1653 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1654 btrfs_set_root_flags(new_root_item, root_flags);
1656 btrfs_set_root_generation_v2(new_root_item,
1658 generate_random_guid(new_root_item->uuid);
1659 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1661 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1662 memset(new_root_item->received_uuid, 0,
1663 sizeof(new_root_item->received_uuid));
1664 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1665 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1666 btrfs_set_root_stransid(new_root_item, 0);
1667 btrfs_set_root_rtransid(new_root_item, 0);
1669 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1670 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1671 btrfs_set_root_otransid(new_root_item, trans->transid);
1673 old = btrfs_lock_root_node(root);
1674 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1677 btrfs_tree_unlock(old);
1678 free_extent_buffer(old);
1679 btrfs_abort_transaction(trans, ret);
1683 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1684 /* clean up in any case */
1685 btrfs_tree_unlock(old);
1686 free_extent_buffer(old);
1688 btrfs_abort_transaction(trans, ret);
1691 /* see comments in should_cow_block() */
1692 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1695 btrfs_set_root_node(new_root_item, tmp);
1696 /* record when the snapshot was created in key.offset */
1697 key.offset = trans->transid;
1698 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1699 btrfs_tree_unlock(tmp);
1700 free_extent_buffer(tmp);
1702 btrfs_abort_transaction(trans, ret);
1707 * insert root back/forward references
1709 ret = btrfs_add_root_ref(trans, objectid,
1710 parent_root->root_key.objectid,
1711 btrfs_ino(BTRFS_I(parent_inode)), index,
1712 dentry->d_name.name, dentry->d_name.len);
1714 btrfs_abort_transaction(trans, ret);
1718 key.offset = (u64)-1;
1719 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1720 if (IS_ERR(pending->snap)) {
1721 ret = PTR_ERR(pending->snap);
1722 pending->snap = NULL;
1723 btrfs_abort_transaction(trans, ret);
1727 ret = btrfs_reloc_post_snapshot(trans, pending);
1729 btrfs_abort_transaction(trans, ret);
1734 * Do special qgroup accounting for snapshot, as we do some qgroup
1735 * snapshot hack to do fast snapshot.
1736 * To co-operate with that hack, we do hack again.
1737 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1739 ret = qgroup_account_snapshot(trans, root, parent_root,
1740 pending->inherit, objectid);
1744 ret = btrfs_insert_dir_item(trans, dentry->d_name.name,
1745 dentry->d_name.len, BTRFS_I(parent_inode),
1746 &key, BTRFS_FT_DIR, index);
1747 /* We have check then name at the beginning, so it is impossible. */
1748 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1750 btrfs_abort_transaction(trans, ret);
1754 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1755 dentry->d_name.len * 2);
1756 parent_inode->i_mtime = parent_inode->i_ctime =
1757 current_time(parent_inode);
1758 ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1760 btrfs_abort_transaction(trans, ret);
1763 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1764 BTRFS_UUID_KEY_SUBVOL,
1767 btrfs_abort_transaction(trans, ret);
1770 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1771 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1772 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1774 if (ret && ret != -EEXIST) {
1775 btrfs_abort_transaction(trans, ret);
1781 pending->error = ret;
1783 trans->block_rsv = rsv;
1784 trans->bytes_reserved = 0;
1786 btrfs_clear_skip_qgroup(trans);
1788 kfree(new_root_item);
1789 pending->root_item = NULL;
1790 btrfs_free_path(path);
1791 pending->path = NULL;
1797 * create all the snapshots we've scheduled for creation
1799 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1801 struct btrfs_pending_snapshot *pending, *next;
1802 struct list_head *head = &trans->transaction->pending_snapshots;
1805 list_for_each_entry_safe(pending, next, head, list) {
1806 list_del(&pending->list);
1807 ret = create_pending_snapshot(trans, pending);
1814 static void update_super_roots(struct btrfs_fs_info *fs_info)
1816 struct btrfs_root_item *root_item;
1817 struct btrfs_super_block *super;
1819 super = fs_info->super_copy;
1821 root_item = &fs_info->chunk_root->root_item;
1822 super->chunk_root = root_item->bytenr;
1823 super->chunk_root_generation = root_item->generation;
1824 super->chunk_root_level = root_item->level;
1826 root_item = &fs_info->tree_root->root_item;
1827 super->root = root_item->bytenr;
1828 super->generation = root_item->generation;
1829 super->root_level = root_item->level;
1830 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1831 super->cache_generation = root_item->generation;
1832 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1833 super->cache_generation = 0;
1834 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1835 super->uuid_tree_generation = root_item->generation;
1838 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1840 struct btrfs_transaction *trans;
1843 spin_lock(&info->trans_lock);
1844 trans = info->running_transaction;
1846 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1847 spin_unlock(&info->trans_lock);
1851 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1853 struct btrfs_transaction *trans;
1856 spin_lock(&info->trans_lock);
1857 trans = info->running_transaction;
1859 ret = is_transaction_blocked(trans);
1860 spin_unlock(&info->trans_lock);
1865 * commit transactions asynchronously. once btrfs_commit_transaction_async
1866 * returns, any subsequent transaction will not be allowed to join.
1868 struct btrfs_async_commit {
1869 struct btrfs_trans_handle *newtrans;
1870 struct work_struct work;
1873 static void do_async_commit(struct work_struct *work)
1875 struct btrfs_async_commit *ac =
1876 container_of(work, struct btrfs_async_commit, work);
1879 * We've got freeze protection passed with the transaction.
1880 * Tell lockdep about it.
1882 if (ac->newtrans->type & __TRANS_FREEZABLE)
1883 __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1885 current->journal_info = ac->newtrans;
1887 btrfs_commit_transaction(ac->newtrans);
1891 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1893 struct btrfs_fs_info *fs_info = trans->fs_info;
1894 struct btrfs_async_commit *ac;
1895 struct btrfs_transaction *cur_trans;
1897 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1901 INIT_WORK(&ac->work, do_async_commit);
1902 ac->newtrans = btrfs_join_transaction(trans->root);
1903 if (IS_ERR(ac->newtrans)) {
1904 int err = PTR_ERR(ac->newtrans);
1909 /* take transaction reference */
1910 cur_trans = trans->transaction;
1911 refcount_inc(&cur_trans->use_count);
1913 btrfs_end_transaction(trans);
1916 * Tell lockdep we've released the freeze rwsem, since the
1917 * async commit thread will be the one to unlock it.
1919 if (ac->newtrans->type & __TRANS_FREEZABLE)
1920 __sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1922 schedule_work(&ac->work);
1924 * Wait for the current transaction commit to start and block
1925 * subsequent transaction joins
1927 wait_event(fs_info->transaction_blocked_wait,
1928 cur_trans->state >= TRANS_STATE_COMMIT_START ||
1929 TRANS_ABORTED(cur_trans));
1930 if (current->journal_info == trans)
1931 current->journal_info = NULL;
1933 btrfs_put_transaction(cur_trans);
1938 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1940 struct btrfs_fs_info *fs_info = trans->fs_info;
1941 struct btrfs_transaction *cur_trans = trans->transaction;
1943 WARN_ON(refcount_read(&trans->use_count) > 1);
1945 btrfs_abort_transaction(trans, err);
1947 spin_lock(&fs_info->trans_lock);
1950 * If the transaction is removed from the list, it means this
1951 * transaction has been committed successfully, so it is impossible
1952 * to call the cleanup function.
1954 BUG_ON(list_empty(&cur_trans->list));
1956 if (cur_trans == fs_info->running_transaction) {
1957 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1958 spin_unlock(&fs_info->trans_lock);
1959 wait_event(cur_trans->writer_wait,
1960 atomic_read(&cur_trans->num_writers) == 1);
1962 spin_lock(&fs_info->trans_lock);
1966 * Now that we know no one else is still using the transaction we can
1967 * remove the transaction from the list of transactions. This avoids
1968 * the transaction kthread from cleaning up the transaction while some
1969 * other task is still using it, which could result in a use-after-free
1970 * on things like log trees, as it forces the transaction kthread to
1971 * wait for this transaction to be cleaned up by us.
1973 list_del_init(&cur_trans->list);
1975 spin_unlock(&fs_info->trans_lock);
1977 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
1979 spin_lock(&fs_info->trans_lock);
1980 if (cur_trans == fs_info->running_transaction)
1981 fs_info->running_transaction = NULL;
1982 spin_unlock(&fs_info->trans_lock);
1984 if (trans->type & __TRANS_FREEZABLE)
1985 sb_end_intwrite(fs_info->sb);
1986 btrfs_put_transaction(cur_trans);
1987 btrfs_put_transaction(cur_trans);
1989 trace_btrfs_transaction_commit(trans->root);
1991 if (current->journal_info == trans)
1992 current->journal_info = NULL;
1993 btrfs_scrub_cancel(fs_info);
1995 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1999 * Release reserved delayed ref space of all pending block groups of the
2000 * transaction and remove them from the list
2002 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2004 struct btrfs_fs_info *fs_info = trans->fs_info;
2005 struct btrfs_block_group *block_group, *tmp;
2007 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2008 btrfs_delayed_refs_rsv_release(fs_info, 1);
2009 list_del_init(&block_group->bg_list);
2013 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2016 * We use writeback_inodes_sb here because if we used
2017 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2018 * Currently are holding the fs freeze lock, if we do an async flush
2019 * we'll do btrfs_join_transaction() and deadlock because we need to
2020 * wait for the fs freeze lock. Using the direct flushing we benefit
2021 * from already being in a transaction and our join_transaction doesn't
2022 * have to re-take the fs freeze lock.
2024 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2025 writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2029 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2031 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2032 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2035 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2037 struct btrfs_fs_info *fs_info = trans->fs_info;
2038 struct btrfs_transaction *cur_trans = trans->transaction;
2039 struct btrfs_transaction *prev_trans = NULL;
2042 ASSERT(refcount_read(&trans->use_count) == 1);
2044 /* Stop the commit early if ->aborted is set */
2045 if (TRANS_ABORTED(cur_trans)) {
2046 ret = cur_trans->aborted;
2047 btrfs_end_transaction(trans);
2051 btrfs_trans_release_metadata(trans);
2052 trans->block_rsv = NULL;
2055 * We only want one transaction commit doing the flushing so we do not
2056 * waste a bunch of time on lock contention on the extent root node.
2058 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2059 &cur_trans->delayed_refs.flags)) {
2061 * Make a pass through all the delayed refs we have so far.
2062 * Any running threads may add more while we are here.
2064 ret = btrfs_run_delayed_refs(trans, 0);
2066 btrfs_end_transaction(trans);
2071 btrfs_create_pending_block_groups(trans);
2073 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2076 /* this mutex is also taken before trying to set
2077 * block groups readonly. We need to make sure
2078 * that nobody has set a block group readonly
2079 * after a extents from that block group have been
2080 * allocated for cache files. btrfs_set_block_group_ro
2081 * will wait for the transaction to commit if it
2082 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2084 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2085 * only one process starts all the block group IO. It wouldn't
2086 * hurt to have more than one go through, but there's no
2087 * real advantage to it either.
2089 mutex_lock(&fs_info->ro_block_group_mutex);
2090 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2093 mutex_unlock(&fs_info->ro_block_group_mutex);
2096 ret = btrfs_start_dirty_block_groups(trans);
2098 btrfs_end_transaction(trans);
2104 spin_lock(&fs_info->trans_lock);
2105 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2106 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2108 spin_unlock(&fs_info->trans_lock);
2109 refcount_inc(&cur_trans->use_count);
2111 if (trans->in_fsync)
2112 want_state = TRANS_STATE_SUPER_COMMITTED;
2113 ret = btrfs_end_transaction(trans);
2114 wait_for_commit(cur_trans, want_state);
2116 if (TRANS_ABORTED(cur_trans))
2117 ret = cur_trans->aborted;
2119 btrfs_put_transaction(cur_trans);
2124 cur_trans->state = TRANS_STATE_COMMIT_START;
2125 wake_up(&fs_info->transaction_blocked_wait);
2127 if (cur_trans->list.prev != &fs_info->trans_list) {
2128 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2130 if (trans->in_fsync)
2131 want_state = TRANS_STATE_SUPER_COMMITTED;
2133 prev_trans = list_entry(cur_trans->list.prev,
2134 struct btrfs_transaction, list);
2135 if (prev_trans->state < want_state) {
2136 refcount_inc(&prev_trans->use_count);
2137 spin_unlock(&fs_info->trans_lock);
2139 wait_for_commit(prev_trans, want_state);
2141 ret = READ_ONCE(prev_trans->aborted);
2143 btrfs_put_transaction(prev_trans);
2145 goto cleanup_transaction;
2147 spin_unlock(&fs_info->trans_lock);
2150 spin_unlock(&fs_info->trans_lock);
2152 * The previous transaction was aborted and was already removed
2153 * from the list of transactions at fs_info->trans_list. So we
2154 * abort to prevent writing a new superblock that reflects a
2155 * corrupt state (pointing to trees with unwritten nodes/leafs).
2157 if (BTRFS_FS_ERROR(fs_info)) {
2159 goto cleanup_transaction;
2163 extwriter_counter_dec(cur_trans, trans->type);
2165 ret = btrfs_start_delalloc_flush(fs_info);
2167 goto cleanup_transaction;
2169 ret = btrfs_run_delayed_items(trans);
2171 goto cleanup_transaction;
2173 wait_event(cur_trans->writer_wait,
2174 extwriter_counter_read(cur_trans) == 0);
2176 /* some pending stuffs might be added after the previous flush. */
2177 ret = btrfs_run_delayed_items(trans);
2179 goto cleanup_transaction;
2181 btrfs_wait_delalloc_flush(fs_info);
2184 * Wait for all ordered extents started by a fast fsync that joined this
2185 * transaction. Otherwise if this transaction commits before the ordered
2186 * extents complete we lose logged data after a power failure.
2188 wait_event(cur_trans->pending_wait,
2189 atomic_read(&cur_trans->pending_ordered) == 0);
2191 btrfs_scrub_pause(fs_info);
2193 * Ok now we need to make sure to block out any other joins while we
2194 * commit the transaction. We could have started a join before setting
2195 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2197 spin_lock(&fs_info->trans_lock);
2198 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2199 spin_unlock(&fs_info->trans_lock);
2200 wait_event(cur_trans->writer_wait,
2201 atomic_read(&cur_trans->num_writers) == 1);
2203 if (TRANS_ABORTED(cur_trans)) {
2204 ret = cur_trans->aborted;
2205 goto scrub_continue;
2208 * the reloc mutex makes sure that we stop
2209 * the balancing code from coming in and moving
2210 * extents around in the middle of the commit
2212 mutex_lock(&fs_info->reloc_mutex);
2215 * We needn't worry about the delayed items because we will
2216 * deal with them in create_pending_snapshot(), which is the
2217 * core function of the snapshot creation.
2219 ret = create_pending_snapshots(trans);
2224 * We insert the dir indexes of the snapshots and update the inode
2225 * of the snapshots' parents after the snapshot creation, so there
2226 * are some delayed items which are not dealt with. Now deal with
2229 * We needn't worry that this operation will corrupt the snapshots,
2230 * because all the tree which are snapshoted will be forced to COW
2231 * the nodes and leaves.
2233 ret = btrfs_run_delayed_items(trans);
2237 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2242 * make sure none of the code above managed to slip in a
2245 btrfs_assert_delayed_root_empty(fs_info);
2247 WARN_ON(cur_trans != trans->transaction);
2249 /* btrfs_commit_tree_roots is responsible for getting the
2250 * various roots consistent with each other. Every pointer
2251 * in the tree of tree roots has to point to the most up to date
2252 * root for every subvolume and other tree. So, we have to keep
2253 * the tree logging code from jumping in and changing any
2256 * At this point in the commit, there can't be any tree-log
2257 * writers, but a little lower down we drop the trans mutex
2258 * and let new people in. By holding the tree_log_mutex
2259 * from now until after the super is written, we avoid races
2260 * with the tree-log code.
2262 mutex_lock(&fs_info->tree_log_mutex);
2264 ret = commit_fs_roots(trans);
2266 goto unlock_tree_log;
2269 * Since the transaction is done, we can apply the pending changes
2270 * before the next transaction.
2272 btrfs_apply_pending_changes(fs_info);
2274 /* commit_fs_roots gets rid of all the tree log roots, it is now
2275 * safe to free the root of tree log roots
2277 btrfs_free_log_root_tree(trans, fs_info);
2280 * Since fs roots are all committed, we can get a quite accurate
2281 * new_roots. So let's do quota accounting.
2283 ret = btrfs_qgroup_account_extents(trans);
2285 goto unlock_tree_log;
2287 ret = commit_cowonly_roots(trans);
2289 goto unlock_tree_log;
2292 * The tasks which save the space cache and inode cache may also
2293 * update ->aborted, check it.
2295 if (TRANS_ABORTED(cur_trans)) {
2296 ret = cur_trans->aborted;
2297 goto unlock_tree_log;
2300 cur_trans = fs_info->running_transaction;
2302 btrfs_set_root_node(&fs_info->tree_root->root_item,
2303 fs_info->tree_root->node);
2304 list_add_tail(&fs_info->tree_root->dirty_list,
2305 &cur_trans->switch_commits);
2307 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2308 fs_info->chunk_root->node);
2309 list_add_tail(&fs_info->chunk_root->dirty_list,
2310 &cur_trans->switch_commits);
2312 switch_commit_roots(trans);
2314 ASSERT(list_empty(&cur_trans->dirty_bgs));
2315 ASSERT(list_empty(&cur_trans->io_bgs));
2316 update_super_roots(fs_info);
2318 btrfs_set_super_log_root(fs_info->super_copy, 0);
2319 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2320 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2321 sizeof(*fs_info->super_copy));
2323 btrfs_commit_device_sizes(cur_trans);
2325 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2326 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2328 btrfs_trans_release_chunk_metadata(trans);
2330 spin_lock(&fs_info->trans_lock);
2331 cur_trans->state = TRANS_STATE_UNBLOCKED;
2332 fs_info->running_transaction = NULL;
2333 spin_unlock(&fs_info->trans_lock);
2334 mutex_unlock(&fs_info->reloc_mutex);
2336 wake_up(&fs_info->transaction_wait);
2338 ret = btrfs_write_and_wait_transaction(trans);
2340 btrfs_handle_fs_error(fs_info, ret,
2341 "Error while writing out transaction");
2343 * reloc_mutex has been unlocked, tree_log_mutex is still held
2344 * but we can't jump to unlock_tree_log causing double unlock
2346 mutex_unlock(&fs_info->tree_log_mutex);
2347 goto scrub_continue;
2351 * At this point, we should have written all the tree blocks allocated
2352 * in this transaction. So it's now safe to free the redirtyied extent
2355 btrfs_free_redirty_list(cur_trans);
2357 ret = write_all_supers(fs_info, 0);
2359 * the super is written, we can safely allow the tree-loggers
2360 * to go about their business
2362 mutex_unlock(&fs_info->tree_log_mutex);
2364 goto scrub_continue;
2367 * We needn't acquire the lock here because there is no other task
2368 * which can change it.
2370 cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2371 wake_up(&cur_trans->commit_wait);
2373 btrfs_finish_extent_commit(trans);
2375 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2376 btrfs_clear_space_info_full(fs_info);
2378 fs_info->last_trans_committed = cur_trans->transid;
2380 * We needn't acquire the lock here because there is no other task
2381 * which can change it.
2383 cur_trans->state = TRANS_STATE_COMPLETED;
2384 wake_up(&cur_trans->commit_wait);
2386 spin_lock(&fs_info->trans_lock);
2387 list_del_init(&cur_trans->list);
2388 spin_unlock(&fs_info->trans_lock);
2390 btrfs_put_transaction(cur_trans);
2391 btrfs_put_transaction(cur_trans);
2393 if (trans->type & __TRANS_FREEZABLE)
2394 sb_end_intwrite(fs_info->sb);
2396 trace_btrfs_transaction_commit(trans->root);
2398 btrfs_scrub_continue(fs_info);
2400 if (current->journal_info == trans)
2401 current->journal_info = NULL;
2403 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2408 mutex_unlock(&fs_info->tree_log_mutex);
2410 mutex_unlock(&fs_info->reloc_mutex);
2412 btrfs_scrub_continue(fs_info);
2413 cleanup_transaction:
2414 btrfs_trans_release_metadata(trans);
2415 btrfs_cleanup_pending_block_groups(trans);
2416 btrfs_trans_release_chunk_metadata(trans);
2417 trans->block_rsv = NULL;
2418 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2419 if (current->journal_info == trans)
2420 current->journal_info = NULL;
2421 cleanup_transaction(trans, ret);
2427 * return < 0 if error
2428 * 0 if there are no more dead_roots at the time of call
2429 * 1 there are more to be processed, call me again
2431 * The return value indicates there are certainly more snapshots to delete, but
2432 * if there comes a new one during processing, it may return 0. We don't mind,
2433 * because btrfs_commit_super will poke cleaner thread and it will process it a
2434 * few seconds later.
2436 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2439 struct btrfs_fs_info *fs_info = root->fs_info;
2441 spin_lock(&fs_info->trans_lock);
2442 if (list_empty(&fs_info->dead_roots)) {
2443 spin_unlock(&fs_info->trans_lock);
2446 root = list_first_entry(&fs_info->dead_roots,
2447 struct btrfs_root, root_list);
2448 list_del_init(&root->root_list);
2449 spin_unlock(&fs_info->trans_lock);
2451 btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2453 btrfs_kill_all_delayed_nodes(root);
2455 if (btrfs_header_backref_rev(root->node) <
2456 BTRFS_MIXED_BACKREF_REV)
2457 ret = btrfs_drop_snapshot(root, 0, 0);
2459 ret = btrfs_drop_snapshot(root, 1, 0);
2461 btrfs_put_root(root);
2462 return (ret < 0) ? 0 : 1;
2465 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2470 prev = xchg(&fs_info->pending_changes, 0);
2474 bit = 1 << BTRFS_PENDING_COMMIT;
2476 btrfs_debug(fs_info, "pending commit done");
2481 "unknown pending changes left 0x%lx, ignoring", prev);