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"
19 #include "inode-map.h"
21 #include "dev-replace.h"
23 #include "block-group.h"
24 #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_COMPLETED] = (__TRANS_START |
114 __TRANS_JOIN_NOLOCK |
115 __TRANS_JOIN_NOSTART),
118 void btrfs_put_transaction(struct btrfs_transaction *transaction)
120 WARN_ON(refcount_read(&transaction->use_count) == 0);
121 if (refcount_dec_and_test(&transaction->use_count)) {
122 BUG_ON(!list_empty(&transaction->list));
123 WARN_ON(!RB_EMPTY_ROOT(
124 &transaction->delayed_refs.href_root.rb_root));
125 WARN_ON(!RB_EMPTY_ROOT(
126 &transaction->delayed_refs.dirty_extent_root));
127 if (transaction->delayed_refs.pending_csums)
128 btrfs_err(transaction->fs_info,
129 "pending csums is %llu",
130 transaction->delayed_refs.pending_csums);
132 * If any block groups are found in ->deleted_bgs then it's
133 * because the transaction was aborted and a commit did not
134 * happen (things failed before writing the new superblock
135 * and calling btrfs_finish_extent_commit()), so we can not
136 * discard the physical locations of the block groups.
138 while (!list_empty(&transaction->deleted_bgs)) {
139 struct btrfs_block_group *cache;
141 cache = list_first_entry(&transaction->deleted_bgs,
142 struct btrfs_block_group,
144 list_del_init(&cache->bg_list);
145 btrfs_unfreeze_block_group(cache);
146 btrfs_put_block_group(cache);
148 WARN_ON(!list_empty(&transaction->dev_update_list));
153 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
155 struct btrfs_transaction *cur_trans = trans->transaction;
156 struct btrfs_fs_info *fs_info = trans->fs_info;
157 struct btrfs_root *root, *tmp;
158 struct btrfs_caching_control *caching_ctl, *next;
160 down_write(&fs_info->commit_root_sem);
161 list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
163 list_del_init(&root->dirty_list);
164 free_extent_buffer(root->commit_root);
165 root->commit_root = btrfs_root_node(root);
166 if (is_fstree(root->root_key.objectid))
167 btrfs_unpin_free_ino(root);
168 extent_io_tree_release(&root->dirty_log_pages);
169 btrfs_qgroup_clean_swapped_blocks(root);
172 /* We can free old roots now. */
173 spin_lock(&cur_trans->dropped_roots_lock);
174 while (!list_empty(&cur_trans->dropped_roots)) {
175 root = list_first_entry(&cur_trans->dropped_roots,
176 struct btrfs_root, root_list);
177 list_del_init(&root->root_list);
178 spin_unlock(&cur_trans->dropped_roots_lock);
179 btrfs_free_log(trans, root);
180 btrfs_drop_and_free_fs_root(fs_info, root);
181 spin_lock(&cur_trans->dropped_roots_lock);
183 spin_unlock(&cur_trans->dropped_roots_lock);
186 * We have to update the last_byte_to_unpin under the commit_root_sem,
187 * at the same time we swap out the commit roots.
189 * This is because we must have a real view of the last spot the caching
190 * kthreads were while caching. Consider the following views of the
191 * extent tree for a block group
194 * +----+----+----+----+----+----+----+
195 * |\\\\| |\\\\|\\\\| |\\\\|\\\\|
196 * +----+----+----+----+----+----+----+
200 * +----+----+----+----+----+----+----+
201 * | | | |\\\\| | |\\\\|
202 * +----+----+----+----+----+----+----+
205 * If the cache_ctl->progress was at 3, then we are only allowed to
206 * unpin [0,1) and [2,3], because the caching thread has already
207 * processed those extents. We are not allowed to unpin [5,6), because
208 * the caching thread will re-start it's search from 3, and thus find
209 * the hole from [4,6) to add to the free space cache.
211 list_for_each_entry_safe(caching_ctl, next,
212 &fs_info->caching_block_groups, list) {
213 struct btrfs_block_group *cache = caching_ctl->block_group;
215 if (btrfs_block_group_done(cache)) {
216 cache->last_byte_to_unpin = (u64)-1;
217 list_del_init(&caching_ctl->list);
218 btrfs_put_caching_control(caching_ctl);
220 cache->last_byte_to_unpin = caching_ctl->progress;
223 up_write(&fs_info->commit_root_sem);
226 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
229 if (type & TRANS_EXTWRITERS)
230 atomic_inc(&trans->num_extwriters);
233 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
236 if (type & TRANS_EXTWRITERS)
237 atomic_dec(&trans->num_extwriters);
240 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
243 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
246 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
248 return atomic_read(&trans->num_extwriters);
252 * To be called after all the new block groups attached to the transaction
253 * handle have been created (btrfs_create_pending_block_groups()).
255 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
257 struct btrfs_fs_info *fs_info = trans->fs_info;
259 if (!trans->chunk_bytes_reserved)
262 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
264 btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
265 trans->chunk_bytes_reserved, NULL);
266 trans->chunk_bytes_reserved = 0;
270 * either allocate a new transaction or hop into the existing one
272 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
275 struct btrfs_transaction *cur_trans;
277 spin_lock(&fs_info->trans_lock);
279 /* The file system has been taken offline. No new transactions. */
280 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
281 spin_unlock(&fs_info->trans_lock);
285 cur_trans = fs_info->running_transaction;
287 if (TRANS_ABORTED(cur_trans)) {
288 spin_unlock(&fs_info->trans_lock);
289 return cur_trans->aborted;
291 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
292 spin_unlock(&fs_info->trans_lock);
295 refcount_inc(&cur_trans->use_count);
296 atomic_inc(&cur_trans->num_writers);
297 extwriter_counter_inc(cur_trans, type);
298 spin_unlock(&fs_info->trans_lock);
301 spin_unlock(&fs_info->trans_lock);
304 * If we are ATTACH or TRANS_JOIN_NOSTART, we just want to catch the
305 * current transaction, and commit it. If there is no transaction, just
308 if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART)
312 * JOIN_NOLOCK only happens during the transaction commit, so
313 * it is impossible that ->running_transaction is NULL
315 BUG_ON(type == TRANS_JOIN_NOLOCK);
317 cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
321 spin_lock(&fs_info->trans_lock);
322 if (fs_info->running_transaction) {
324 * someone started a transaction after we unlocked. Make sure
325 * to redo the checks above
329 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
330 spin_unlock(&fs_info->trans_lock);
335 cur_trans->fs_info = fs_info;
336 atomic_set(&cur_trans->pending_ordered, 0);
337 init_waitqueue_head(&cur_trans->pending_wait);
338 atomic_set(&cur_trans->num_writers, 1);
339 extwriter_counter_init(cur_trans, type);
340 init_waitqueue_head(&cur_trans->writer_wait);
341 init_waitqueue_head(&cur_trans->commit_wait);
342 cur_trans->state = TRANS_STATE_RUNNING;
344 * One for this trans handle, one so it will live on until we
345 * commit the transaction.
347 refcount_set(&cur_trans->use_count, 2);
348 cur_trans->flags = 0;
349 cur_trans->start_time = ktime_get_seconds();
351 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
353 cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
354 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
355 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
358 * although the tree mod log is per file system and not per transaction,
359 * the log must never go across transaction boundaries.
362 if (!list_empty(&fs_info->tree_mod_seq_list))
363 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
364 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
365 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
366 atomic64_set(&fs_info->tree_mod_seq, 0);
368 spin_lock_init(&cur_trans->delayed_refs.lock);
370 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
371 INIT_LIST_HEAD(&cur_trans->dev_update_list);
372 INIT_LIST_HEAD(&cur_trans->switch_commits);
373 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
374 INIT_LIST_HEAD(&cur_trans->io_bgs);
375 INIT_LIST_HEAD(&cur_trans->dropped_roots);
376 mutex_init(&cur_trans->cache_write_mutex);
377 spin_lock_init(&cur_trans->dirty_bgs_lock);
378 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
379 spin_lock_init(&cur_trans->dropped_roots_lock);
380 list_add_tail(&cur_trans->list, &fs_info->trans_list);
381 extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
382 IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode);
383 extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
384 IO_TREE_FS_PINNED_EXTENTS, NULL);
385 fs_info->generation++;
386 cur_trans->transid = fs_info->generation;
387 fs_info->running_transaction = cur_trans;
388 cur_trans->aborted = 0;
389 spin_unlock(&fs_info->trans_lock);
395 * This does all the record keeping required to make sure that a shareable root
396 * is properly recorded in a given transaction. This is required to make sure
397 * the old root from before we joined the transaction is deleted when the
398 * transaction commits.
400 static int record_root_in_trans(struct btrfs_trans_handle *trans,
401 struct btrfs_root *root,
404 struct btrfs_fs_info *fs_info = root->fs_info;
406 if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
407 root->last_trans < trans->transid) || force) {
408 WARN_ON(root == fs_info->extent_root);
409 WARN_ON(!force && root->commit_root != root->node);
412 * see below for IN_TRANS_SETUP usage rules
413 * we have the reloc mutex held now, so there
414 * is only one writer in this function
416 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
418 /* make sure readers find IN_TRANS_SETUP before
419 * they find our root->last_trans update
423 spin_lock(&fs_info->fs_roots_radix_lock);
424 if (root->last_trans == trans->transid && !force) {
425 spin_unlock(&fs_info->fs_roots_radix_lock);
428 radix_tree_tag_set(&fs_info->fs_roots_radix,
429 (unsigned long)root->root_key.objectid,
430 BTRFS_ROOT_TRANS_TAG);
431 spin_unlock(&fs_info->fs_roots_radix_lock);
432 root->last_trans = trans->transid;
434 /* this is pretty tricky. We don't want to
435 * take the relocation lock in btrfs_record_root_in_trans
436 * unless we're really doing the first setup for this root in
439 * Normally we'd use root->last_trans as a flag to decide
440 * if we want to take the expensive mutex.
442 * But, we have to set root->last_trans before we
443 * init the relocation root, otherwise, we trip over warnings
444 * in ctree.c. The solution used here is to flag ourselves
445 * with root IN_TRANS_SETUP. When this is 1, we're still
446 * fixing up the reloc trees and everyone must wait.
448 * When this is zero, they can trust root->last_trans and fly
449 * through btrfs_record_root_in_trans without having to take the
450 * lock. smp_wmb() makes sure that all the writes above are
451 * done before we pop in the zero below
453 btrfs_init_reloc_root(trans, root);
454 smp_mb__before_atomic();
455 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
461 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
462 struct btrfs_root *root)
464 struct btrfs_fs_info *fs_info = root->fs_info;
465 struct btrfs_transaction *cur_trans = trans->transaction;
467 /* Add ourselves to the transaction dropped list */
468 spin_lock(&cur_trans->dropped_roots_lock);
469 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
470 spin_unlock(&cur_trans->dropped_roots_lock);
472 /* Make sure we don't try to update the root at commit time */
473 spin_lock(&fs_info->fs_roots_radix_lock);
474 radix_tree_tag_clear(&fs_info->fs_roots_radix,
475 (unsigned long)root->root_key.objectid,
476 BTRFS_ROOT_TRANS_TAG);
477 spin_unlock(&fs_info->fs_roots_radix_lock);
480 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
481 struct btrfs_root *root)
483 struct btrfs_fs_info *fs_info = root->fs_info;
485 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
489 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
493 if (root->last_trans == trans->transid &&
494 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
497 mutex_lock(&fs_info->reloc_mutex);
498 record_root_in_trans(trans, root, 0);
499 mutex_unlock(&fs_info->reloc_mutex);
504 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
506 return (trans->state >= TRANS_STATE_COMMIT_START &&
507 trans->state < TRANS_STATE_UNBLOCKED &&
508 !TRANS_ABORTED(trans));
511 /* wait for commit against the current transaction to become unblocked
512 * when this is done, it is safe to start a new transaction, but the current
513 * transaction might not be fully on disk.
515 static void wait_current_trans(struct btrfs_fs_info *fs_info)
517 struct btrfs_transaction *cur_trans;
519 spin_lock(&fs_info->trans_lock);
520 cur_trans = fs_info->running_transaction;
521 if (cur_trans && is_transaction_blocked(cur_trans)) {
522 refcount_inc(&cur_trans->use_count);
523 spin_unlock(&fs_info->trans_lock);
525 wait_event(fs_info->transaction_wait,
526 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
527 TRANS_ABORTED(cur_trans));
528 btrfs_put_transaction(cur_trans);
530 spin_unlock(&fs_info->trans_lock);
534 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
536 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
539 if (type == TRANS_START)
545 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
547 struct btrfs_fs_info *fs_info = root->fs_info;
549 if (!fs_info->reloc_ctl ||
550 !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
551 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
558 static struct btrfs_trans_handle *
559 start_transaction(struct btrfs_root *root, unsigned int num_items,
560 unsigned int type, enum btrfs_reserve_flush_enum flush,
561 bool enforce_qgroups)
563 struct btrfs_fs_info *fs_info = root->fs_info;
564 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
565 struct btrfs_trans_handle *h;
566 struct btrfs_transaction *cur_trans;
568 u64 qgroup_reserved = 0;
569 bool reloc_reserved = false;
570 bool do_chunk_alloc = false;
573 /* Send isn't supposed to start transactions. */
574 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
576 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
577 return ERR_PTR(-EROFS);
579 if (current->journal_info) {
580 WARN_ON(type & TRANS_EXTWRITERS);
581 h = current->journal_info;
582 refcount_inc(&h->use_count);
583 WARN_ON(refcount_read(&h->use_count) > 2);
584 h->orig_rsv = h->block_rsv;
590 * Do the reservation before we join the transaction so we can do all
591 * the appropriate flushing if need be.
593 if (num_items && root != fs_info->chunk_root) {
594 struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
595 u64 delayed_refs_bytes = 0;
597 qgroup_reserved = num_items * fs_info->nodesize;
598 ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
604 * We want to reserve all the bytes we may need all at once, so
605 * we only do 1 enospc flushing cycle per transaction start. We
606 * accomplish this by simply assuming we'll do 2 x num_items
607 * worth of delayed refs updates in this trans handle, and
608 * refill that amount for whatever is missing in the reserve.
610 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
611 if (flush == BTRFS_RESERVE_FLUSH_ALL &&
612 delayed_refs_rsv->full == 0) {
613 delayed_refs_bytes = num_bytes;
618 * Do the reservation for the relocation root creation
620 if (need_reserve_reloc_root(root)) {
621 num_bytes += fs_info->nodesize;
622 reloc_reserved = true;
625 ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
628 if (delayed_refs_bytes) {
629 btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
631 num_bytes -= delayed_refs_bytes;
634 if (rsv->space_info->force_alloc)
635 do_chunk_alloc = true;
636 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
637 !delayed_refs_rsv->full) {
639 * Some people call with btrfs_start_transaction(root, 0)
640 * because they can be throttled, but have some other mechanism
641 * for reserving space. We still want these guys to refill the
642 * delayed block_rsv so just add 1 items worth of reservation
645 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
650 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
657 * If we are JOIN_NOLOCK we're already committing a transaction and
658 * waiting on this guy, so we don't need to do the sb_start_intwrite
659 * because we're already holding a ref. We need this because we could
660 * have raced in and did an fsync() on a file which can kick a commit
661 * and then we deadlock with somebody doing a freeze.
663 * If we are ATTACH, it means we just want to catch the current
664 * transaction and commit it, so we needn't do sb_start_intwrite().
666 if (type & __TRANS_FREEZABLE)
667 sb_start_intwrite(fs_info->sb);
669 if (may_wait_transaction(fs_info, type))
670 wait_current_trans(fs_info);
673 ret = join_transaction(fs_info, type);
675 wait_current_trans(fs_info);
676 if (unlikely(type == TRANS_ATTACH ||
677 type == TRANS_JOIN_NOSTART))
680 } while (ret == -EBUSY);
685 cur_trans = fs_info->running_transaction;
687 h->transid = cur_trans->transid;
688 h->transaction = cur_trans;
690 refcount_set(&h->use_count, 1);
691 h->fs_info = root->fs_info;
694 h->can_flush_pending_bgs = true;
695 INIT_LIST_HEAD(&h->new_bgs);
698 if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
699 may_wait_transaction(fs_info, type)) {
700 current->journal_info = h;
701 btrfs_commit_transaction(h);
706 trace_btrfs_space_reservation(fs_info, "transaction",
707 h->transid, num_bytes, 1);
708 h->block_rsv = &fs_info->trans_block_rsv;
709 h->bytes_reserved = num_bytes;
710 h->reloc_reserved = reloc_reserved;
714 if (!current->journal_info)
715 current->journal_info = h;
718 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
719 * ALLOC_FORCE the first run through, and then we won't allocate for
720 * anybody else who races in later. We don't care about the return
723 if (do_chunk_alloc && num_bytes) {
724 u64 flags = h->block_rsv->space_info->flags;
726 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
727 CHUNK_ALLOC_NO_FORCE);
731 * btrfs_record_root_in_trans() needs to alloc new extents, and may
732 * call btrfs_join_transaction() while we're also starting a
735 * Thus it need to be called after current->journal_info initialized,
736 * or we can deadlock.
738 btrfs_record_root_in_trans(h, root);
743 if (type & __TRANS_FREEZABLE)
744 sb_end_intwrite(fs_info->sb);
745 kmem_cache_free(btrfs_trans_handle_cachep, h);
748 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
751 btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
755 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
756 unsigned int num_items)
758 return start_transaction(root, num_items, TRANS_START,
759 BTRFS_RESERVE_FLUSH_ALL, true);
762 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
763 struct btrfs_root *root,
764 unsigned int num_items)
766 return start_transaction(root, num_items, TRANS_START,
767 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
770 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
772 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
776 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
778 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
779 BTRFS_RESERVE_NO_FLUSH, true);
783 * Similar to regular join but it never starts a transaction when none is
784 * running or after waiting for the current one to finish.
786 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
788 return start_transaction(root, 0, TRANS_JOIN_NOSTART,
789 BTRFS_RESERVE_NO_FLUSH, true);
793 * btrfs_attach_transaction() - catch the running transaction
795 * It is used when we want to commit the current the transaction, but
796 * don't want to start a new one.
798 * Note: If this function return -ENOENT, it just means there is no
799 * running transaction. But it is possible that the inactive transaction
800 * is still in the memory, not fully on disk. If you hope there is no
801 * inactive transaction in the fs when -ENOENT is returned, you should
803 * btrfs_attach_transaction_barrier()
805 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
807 return start_transaction(root, 0, TRANS_ATTACH,
808 BTRFS_RESERVE_NO_FLUSH, true);
812 * btrfs_attach_transaction_barrier() - catch the running transaction
814 * It is similar to the above function, the difference is this one
815 * will wait for all the inactive transactions until they fully
818 struct btrfs_trans_handle *
819 btrfs_attach_transaction_barrier(struct btrfs_root *root)
821 struct btrfs_trans_handle *trans;
823 trans = start_transaction(root, 0, TRANS_ATTACH,
824 BTRFS_RESERVE_NO_FLUSH, true);
825 if (trans == ERR_PTR(-ENOENT)) {
828 ret = btrfs_wait_for_commit(root->fs_info, 0);
836 /* wait for a transaction commit to be fully complete */
837 static noinline void wait_for_commit(struct btrfs_transaction *commit)
839 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
842 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
844 struct btrfs_transaction *cur_trans = NULL, *t;
848 if (transid <= fs_info->last_trans_committed)
851 /* find specified transaction */
852 spin_lock(&fs_info->trans_lock);
853 list_for_each_entry(t, &fs_info->trans_list, list) {
854 if (t->transid == transid) {
856 refcount_inc(&cur_trans->use_count);
860 if (t->transid > transid) {
865 spin_unlock(&fs_info->trans_lock);
868 * The specified transaction doesn't exist, or we
869 * raced with btrfs_commit_transaction
872 if (transid > fs_info->last_trans_committed)
877 /* find newest transaction that is committing | committed */
878 spin_lock(&fs_info->trans_lock);
879 list_for_each_entry_reverse(t, &fs_info->trans_list,
881 if (t->state >= TRANS_STATE_COMMIT_START) {
882 if (t->state == TRANS_STATE_COMPLETED)
885 refcount_inc(&cur_trans->use_count);
889 spin_unlock(&fs_info->trans_lock);
891 goto out; /* nothing committing|committed */
894 wait_for_commit(cur_trans);
895 ret = cur_trans->aborted;
896 btrfs_put_transaction(cur_trans);
901 void btrfs_throttle(struct btrfs_fs_info *fs_info)
903 wait_current_trans(fs_info);
906 static int should_end_transaction(struct btrfs_trans_handle *trans)
908 struct btrfs_fs_info *fs_info = trans->fs_info;
910 if (btrfs_check_space_for_delayed_refs(fs_info))
913 return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
916 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
918 struct btrfs_transaction *cur_trans = trans->transaction;
921 if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
922 cur_trans->delayed_refs.flushing)
925 return should_end_transaction(trans);
928 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
931 struct btrfs_fs_info *fs_info = trans->fs_info;
933 if (!trans->block_rsv) {
934 ASSERT(!trans->bytes_reserved);
938 if (!trans->bytes_reserved)
941 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
942 trace_btrfs_space_reservation(fs_info, "transaction",
943 trans->transid, trans->bytes_reserved, 0);
944 btrfs_block_rsv_release(fs_info, trans->block_rsv,
945 trans->bytes_reserved, NULL);
946 trans->bytes_reserved = 0;
949 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
952 struct btrfs_fs_info *info = trans->fs_info;
953 struct btrfs_transaction *cur_trans = trans->transaction;
956 if (refcount_read(&trans->use_count) > 1) {
957 refcount_dec(&trans->use_count);
958 trans->block_rsv = trans->orig_rsv;
962 btrfs_trans_release_metadata(trans);
963 trans->block_rsv = NULL;
965 btrfs_create_pending_block_groups(trans);
967 btrfs_trans_release_chunk_metadata(trans);
969 if (trans->type & __TRANS_FREEZABLE)
970 sb_end_intwrite(info->sb);
972 WARN_ON(cur_trans != info->running_transaction);
973 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
974 atomic_dec(&cur_trans->num_writers);
975 extwriter_counter_dec(cur_trans, trans->type);
977 cond_wake_up(&cur_trans->writer_wait);
978 btrfs_put_transaction(cur_trans);
980 if (current->journal_info == trans)
981 current->journal_info = NULL;
984 btrfs_run_delayed_iputs(info);
986 if (TRANS_ABORTED(trans) ||
987 test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
988 wake_up_process(info->transaction_kthread);
989 if (TRANS_ABORTED(trans))
990 err = trans->aborted;
995 kmem_cache_free(btrfs_trans_handle_cachep, trans);
999 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1001 return __btrfs_end_transaction(trans, 0);
1004 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1006 return __btrfs_end_transaction(trans, 1);
1010 * when btree blocks are allocated, they have some corresponding bits set for
1011 * them in one of two extent_io trees. This is used to make sure all of
1012 * those extents are sent to disk but does not wait on them
1014 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1015 struct extent_io_tree *dirty_pages, int mark)
1019 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1020 struct extent_state *cached_state = NULL;
1024 atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1025 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1026 mark, &cached_state)) {
1027 bool wait_writeback = false;
1029 err = convert_extent_bit(dirty_pages, start, end,
1031 mark, &cached_state);
1033 * convert_extent_bit can return -ENOMEM, which is most of the
1034 * time a temporary error. So when it happens, ignore the error
1035 * and wait for writeback of this range to finish - because we
1036 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1037 * to __btrfs_wait_marked_extents() would not know that
1038 * writeback for this range started and therefore wouldn't
1039 * wait for it to finish - we don't want to commit a
1040 * superblock that points to btree nodes/leafs for which
1041 * writeback hasn't finished yet (and without errors).
1042 * We cleanup any entries left in the io tree when committing
1043 * the transaction (through extent_io_tree_release()).
1045 if (err == -ENOMEM) {
1047 wait_writeback = true;
1050 err = filemap_fdatawrite_range(mapping, start, end);
1053 else if (wait_writeback)
1054 werr = filemap_fdatawait_range(mapping, start, end);
1055 free_extent_state(cached_state);
1056 cached_state = NULL;
1060 atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1065 * when btree blocks are allocated, they have some corresponding bits set for
1066 * them in one of two extent_io trees. This is used to make sure all of
1067 * those extents are on disk for transaction or log commit. We wait
1068 * on all the pages and clear them from the dirty pages state tree
1070 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1071 struct extent_io_tree *dirty_pages)
1075 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1076 struct extent_state *cached_state = NULL;
1080 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1081 EXTENT_NEED_WAIT, &cached_state)) {
1083 * Ignore -ENOMEM errors returned by clear_extent_bit().
1084 * When committing the transaction, we'll remove any entries
1085 * left in the io tree. For a log commit, we don't remove them
1086 * after committing the log because the tree can be accessed
1087 * concurrently - we do it only at transaction commit time when
1088 * it's safe to do it (through extent_io_tree_release()).
1090 err = clear_extent_bit(dirty_pages, start, end,
1091 EXTENT_NEED_WAIT, 0, 0, &cached_state);
1095 err = filemap_fdatawait_range(mapping, start, end);
1098 free_extent_state(cached_state);
1099 cached_state = NULL;
1108 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1109 struct extent_io_tree *dirty_pages)
1111 bool errors = false;
1114 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1115 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1123 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1125 struct btrfs_fs_info *fs_info = log_root->fs_info;
1126 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1127 bool errors = false;
1130 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1132 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1133 if ((mark & EXTENT_DIRTY) &&
1134 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1137 if ((mark & EXTENT_NEW) &&
1138 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1147 * When btree blocks are allocated the corresponding extents are marked dirty.
1148 * This function ensures such extents are persisted on disk for transaction or
1151 * @trans: transaction whose dirty pages we'd like to write
1153 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1157 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1158 struct btrfs_fs_info *fs_info = trans->fs_info;
1159 struct blk_plug plug;
1161 blk_start_plug(&plug);
1162 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1163 blk_finish_plug(&plug);
1164 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1166 extent_io_tree_release(&trans->transaction->dirty_pages);
1177 * this is used to update the root pointer in the tree of tree roots.
1179 * But, in the case of the extent allocation tree, updating the root
1180 * pointer may allocate blocks which may change the root of the extent
1183 * So, this loops and repeats and makes sure the cowonly root didn't
1184 * change while the root pointer was being updated in the metadata.
1186 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1187 struct btrfs_root *root)
1190 u64 old_root_bytenr;
1192 struct btrfs_fs_info *fs_info = root->fs_info;
1193 struct btrfs_root *tree_root = fs_info->tree_root;
1195 old_root_used = btrfs_root_used(&root->root_item);
1198 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1199 if (old_root_bytenr == root->node->start &&
1200 old_root_used == btrfs_root_used(&root->root_item))
1203 btrfs_set_root_node(&root->root_item, root->node);
1204 ret = btrfs_update_root(trans, tree_root,
1210 old_root_used = btrfs_root_used(&root->root_item);
1217 * update all the cowonly tree roots on disk
1219 * The error handling in this function may not be obvious. Any of the
1220 * failures will cause the file system to go offline. We still need
1221 * to clean up the delayed refs.
1223 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1225 struct btrfs_fs_info *fs_info = trans->fs_info;
1226 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1227 struct list_head *io_bgs = &trans->transaction->io_bgs;
1228 struct list_head *next;
1229 struct extent_buffer *eb;
1232 eb = btrfs_lock_root_node(fs_info->tree_root);
1233 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1234 0, &eb, BTRFS_NESTING_COW);
1235 btrfs_tree_unlock(eb);
1236 free_extent_buffer(eb);
1241 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1245 ret = btrfs_run_dev_stats(trans);
1248 ret = btrfs_run_dev_replace(trans);
1251 ret = btrfs_run_qgroups(trans);
1255 ret = btrfs_setup_space_cache(trans);
1259 /* run_qgroups might have added some more refs */
1260 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1264 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1265 struct btrfs_root *root;
1266 next = fs_info->dirty_cowonly_roots.next;
1267 list_del_init(next);
1268 root = list_entry(next, struct btrfs_root, dirty_list);
1269 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1271 if (root != fs_info->extent_root)
1272 list_add_tail(&root->dirty_list,
1273 &trans->transaction->switch_commits);
1274 ret = update_cowonly_root(trans, root);
1277 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1282 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1283 ret = btrfs_write_dirty_block_groups(trans);
1286 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1291 if (!list_empty(&fs_info->dirty_cowonly_roots))
1294 list_add_tail(&fs_info->extent_root->dirty_list,
1295 &trans->transaction->switch_commits);
1297 /* Update dev-replace pointer once everything is committed */
1298 fs_info->dev_replace.committed_cursor_left =
1299 fs_info->dev_replace.cursor_left_last_write_of_item;
1305 * dead roots are old snapshots that need to be deleted. This allocates
1306 * a dirty root struct and adds it into the list of dead roots that need to
1309 void btrfs_add_dead_root(struct btrfs_root *root)
1311 struct btrfs_fs_info *fs_info = root->fs_info;
1313 spin_lock(&fs_info->trans_lock);
1314 if (list_empty(&root->root_list)) {
1315 btrfs_grab_root(root);
1316 list_add_tail(&root->root_list, &fs_info->dead_roots);
1318 spin_unlock(&fs_info->trans_lock);
1322 * update all the cowonly tree roots on disk
1324 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1326 struct btrfs_fs_info *fs_info = trans->fs_info;
1327 struct btrfs_root *gang[8];
1331 spin_lock(&fs_info->fs_roots_radix_lock);
1333 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1336 BTRFS_ROOT_TRANS_TAG);
1339 for (i = 0; i < ret; i++) {
1340 struct btrfs_root *root = gang[i];
1343 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1344 (unsigned long)root->root_key.objectid,
1345 BTRFS_ROOT_TRANS_TAG);
1346 btrfs_qgroup_free_meta_all_pertrans(root);
1347 spin_unlock(&fs_info->fs_roots_radix_lock);
1349 btrfs_free_log(trans, root);
1350 btrfs_update_reloc_root(trans, root);
1352 btrfs_save_ino_cache(root, trans);
1354 /* see comments in should_cow_block() */
1355 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1356 smp_mb__after_atomic();
1358 if (root->commit_root != root->node) {
1359 list_add_tail(&root->dirty_list,
1360 &trans->transaction->switch_commits);
1361 btrfs_set_root_node(&root->root_item,
1365 ret2 = btrfs_update_root(trans, fs_info->tree_root,
1370 spin_lock(&fs_info->fs_roots_radix_lock);
1373 spin_unlock(&fs_info->fs_roots_radix_lock);
1378 * defrag a given btree.
1379 * Every leaf in the btree is read and defragged.
1381 int btrfs_defrag_root(struct btrfs_root *root)
1383 struct btrfs_fs_info *info = root->fs_info;
1384 struct btrfs_trans_handle *trans;
1387 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1391 trans = btrfs_start_transaction(root, 0);
1392 if (IS_ERR(trans)) {
1393 ret = PTR_ERR(trans);
1397 ret = btrfs_defrag_leaves(trans, root);
1399 btrfs_end_transaction(trans);
1400 btrfs_btree_balance_dirty(info);
1403 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1406 if (btrfs_defrag_cancelled(info)) {
1407 btrfs_debug(info, "defrag_root cancelled");
1412 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1417 * Do all special snapshot related qgroup dirty hack.
1419 * Will do all needed qgroup inherit and dirty hack like switch commit
1420 * roots inside one transaction and write all btree into disk, to make
1423 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1424 struct btrfs_root *src,
1425 struct btrfs_root *parent,
1426 struct btrfs_qgroup_inherit *inherit,
1429 struct btrfs_fs_info *fs_info = src->fs_info;
1433 * Save some performance in the case that qgroups are not
1434 * enabled. If this check races with the ioctl, rescan will
1437 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1441 * Ensure dirty @src will be committed. Or, after coming
1442 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1443 * recorded root will never be updated again, causing an outdated root
1446 record_root_in_trans(trans, src, 1);
1449 * We are going to commit transaction, see btrfs_commit_transaction()
1450 * comment for reason locking tree_log_mutex
1452 mutex_lock(&fs_info->tree_log_mutex);
1454 ret = commit_fs_roots(trans);
1457 ret = btrfs_qgroup_account_extents(trans);
1461 /* Now qgroup are all updated, we can inherit it to new qgroups */
1462 ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1468 * Now we do a simplified commit transaction, which will:
1469 * 1) commit all subvolume and extent tree
1470 * To ensure all subvolume and extent tree have a valid
1471 * commit_root to accounting later insert_dir_item()
1472 * 2) write all btree blocks onto disk
1473 * This is to make sure later btree modification will be cowed
1474 * Or commit_root can be populated and cause wrong qgroup numbers
1475 * In this simplified commit, we don't really care about other trees
1476 * like chunk and root tree, as they won't affect qgroup.
1477 * And we don't write super to avoid half committed status.
1479 ret = commit_cowonly_roots(trans);
1482 switch_commit_roots(trans);
1483 ret = btrfs_write_and_wait_transaction(trans);
1485 btrfs_handle_fs_error(fs_info, ret,
1486 "Error while writing out transaction for qgroup");
1489 mutex_unlock(&fs_info->tree_log_mutex);
1492 * Force parent root to be updated, as we recorded it before so its
1493 * last_trans == cur_transid.
1494 * Or it won't be committed again onto disk after later
1498 record_root_in_trans(trans, parent, 1);
1503 * new snapshots need to be created at a very specific time in the
1504 * transaction commit. This does the actual creation.
1507 * If the error which may affect the commitment of the current transaction
1508 * happens, we should return the error number. If the error which just affect
1509 * the creation of the pending snapshots, just return 0.
1511 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1512 struct btrfs_pending_snapshot *pending)
1515 struct btrfs_fs_info *fs_info = trans->fs_info;
1516 struct btrfs_key key;
1517 struct btrfs_root_item *new_root_item;
1518 struct btrfs_root *tree_root = fs_info->tree_root;
1519 struct btrfs_root *root = pending->root;
1520 struct btrfs_root *parent_root;
1521 struct btrfs_block_rsv *rsv;
1522 struct inode *parent_inode;
1523 struct btrfs_path *path;
1524 struct btrfs_dir_item *dir_item;
1525 struct dentry *dentry;
1526 struct extent_buffer *tmp;
1527 struct extent_buffer *old;
1528 struct timespec64 cur_time;
1535 ASSERT(pending->path);
1536 path = pending->path;
1538 ASSERT(pending->root_item);
1539 new_root_item = pending->root_item;
1541 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1543 goto no_free_objectid;
1546 * Make qgroup to skip current new snapshot's qgroupid, as it is
1547 * accounted by later btrfs_qgroup_inherit().
1549 btrfs_set_skip_qgroup(trans, objectid);
1551 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1553 if (to_reserve > 0) {
1554 pending->error = btrfs_block_rsv_add(root,
1555 &pending->block_rsv,
1557 BTRFS_RESERVE_NO_FLUSH);
1559 goto clear_skip_qgroup;
1562 key.objectid = objectid;
1563 key.offset = (u64)-1;
1564 key.type = BTRFS_ROOT_ITEM_KEY;
1566 rsv = trans->block_rsv;
1567 trans->block_rsv = &pending->block_rsv;
1568 trans->bytes_reserved = trans->block_rsv->reserved;
1569 trace_btrfs_space_reservation(fs_info, "transaction",
1571 trans->bytes_reserved, 1);
1572 dentry = pending->dentry;
1573 parent_inode = pending->dir;
1574 parent_root = BTRFS_I(parent_inode)->root;
1575 record_root_in_trans(trans, parent_root, 0);
1577 cur_time = current_time(parent_inode);
1580 * insert the directory item
1582 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1583 BUG_ON(ret); /* -ENOMEM */
1585 /* check if there is a file/dir which has the same name. */
1586 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1587 btrfs_ino(BTRFS_I(parent_inode)),
1588 dentry->d_name.name,
1589 dentry->d_name.len, 0);
1590 if (dir_item != NULL && !IS_ERR(dir_item)) {
1591 pending->error = -EEXIST;
1592 goto dir_item_existed;
1593 } else if (IS_ERR(dir_item)) {
1594 ret = PTR_ERR(dir_item);
1595 btrfs_abort_transaction(trans, ret);
1598 btrfs_release_path(path);
1601 * pull in the delayed directory update
1602 * and the delayed inode item
1603 * otherwise we corrupt the FS during
1606 ret = btrfs_run_delayed_items(trans);
1607 if (ret) { /* Transaction aborted */
1608 btrfs_abort_transaction(trans, ret);
1612 record_root_in_trans(trans, root, 0);
1613 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1614 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1615 btrfs_check_and_init_root_item(new_root_item);
1617 root_flags = btrfs_root_flags(new_root_item);
1618 if (pending->readonly)
1619 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1621 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1622 btrfs_set_root_flags(new_root_item, root_flags);
1624 btrfs_set_root_generation_v2(new_root_item,
1626 generate_random_guid(new_root_item->uuid);
1627 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1629 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1630 memset(new_root_item->received_uuid, 0,
1631 sizeof(new_root_item->received_uuid));
1632 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1633 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1634 btrfs_set_root_stransid(new_root_item, 0);
1635 btrfs_set_root_rtransid(new_root_item, 0);
1637 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1638 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1639 btrfs_set_root_otransid(new_root_item, trans->transid);
1641 old = btrfs_lock_root_node(root);
1642 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1645 btrfs_tree_unlock(old);
1646 free_extent_buffer(old);
1647 btrfs_abort_transaction(trans, ret);
1651 btrfs_set_lock_blocking_write(old);
1653 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1654 /* clean up in any case */
1655 btrfs_tree_unlock(old);
1656 free_extent_buffer(old);
1658 btrfs_abort_transaction(trans, ret);
1661 /* see comments in should_cow_block() */
1662 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1665 btrfs_set_root_node(new_root_item, tmp);
1666 /* record when the snapshot was created in key.offset */
1667 key.offset = trans->transid;
1668 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1669 btrfs_tree_unlock(tmp);
1670 free_extent_buffer(tmp);
1672 btrfs_abort_transaction(trans, ret);
1677 * insert root back/forward references
1679 ret = btrfs_add_root_ref(trans, objectid,
1680 parent_root->root_key.objectid,
1681 btrfs_ino(BTRFS_I(parent_inode)), index,
1682 dentry->d_name.name, dentry->d_name.len);
1684 btrfs_abort_transaction(trans, ret);
1688 key.offset = (u64)-1;
1689 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1690 if (IS_ERR(pending->snap)) {
1691 ret = PTR_ERR(pending->snap);
1692 pending->snap = NULL;
1693 btrfs_abort_transaction(trans, ret);
1697 ret = btrfs_reloc_post_snapshot(trans, pending);
1699 btrfs_abort_transaction(trans, ret);
1703 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1705 btrfs_abort_transaction(trans, ret);
1710 * Do special qgroup accounting for snapshot, as we do some qgroup
1711 * snapshot hack to do fast snapshot.
1712 * To co-operate with that hack, we do hack again.
1713 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1715 ret = qgroup_account_snapshot(trans, root, parent_root,
1716 pending->inherit, objectid);
1720 ret = btrfs_insert_dir_item(trans, dentry->d_name.name,
1721 dentry->d_name.len, BTRFS_I(parent_inode),
1722 &key, BTRFS_FT_DIR, index);
1723 /* We have check then name at the beginning, so it is impossible. */
1724 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1726 btrfs_abort_transaction(trans, ret);
1730 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1731 dentry->d_name.len * 2);
1732 parent_inode->i_mtime = parent_inode->i_ctime =
1733 current_time(parent_inode);
1734 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1736 btrfs_abort_transaction(trans, ret);
1739 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1740 BTRFS_UUID_KEY_SUBVOL,
1743 btrfs_abort_transaction(trans, ret);
1746 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1747 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1748 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1750 if (ret && ret != -EEXIST) {
1751 btrfs_abort_transaction(trans, ret);
1756 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1758 btrfs_abort_transaction(trans, ret);
1763 pending->error = ret;
1765 trans->block_rsv = rsv;
1766 trans->bytes_reserved = 0;
1768 btrfs_clear_skip_qgroup(trans);
1770 kfree(new_root_item);
1771 pending->root_item = NULL;
1772 btrfs_free_path(path);
1773 pending->path = NULL;
1779 * create all the snapshots we've scheduled for creation
1781 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1783 struct btrfs_pending_snapshot *pending, *next;
1784 struct list_head *head = &trans->transaction->pending_snapshots;
1787 list_for_each_entry_safe(pending, next, head, list) {
1788 list_del(&pending->list);
1789 ret = create_pending_snapshot(trans, pending);
1796 static void update_super_roots(struct btrfs_fs_info *fs_info)
1798 struct btrfs_root_item *root_item;
1799 struct btrfs_super_block *super;
1801 super = fs_info->super_copy;
1803 root_item = &fs_info->chunk_root->root_item;
1804 super->chunk_root = root_item->bytenr;
1805 super->chunk_root_generation = root_item->generation;
1806 super->chunk_root_level = root_item->level;
1808 root_item = &fs_info->tree_root->root_item;
1809 super->root = root_item->bytenr;
1810 super->generation = root_item->generation;
1811 super->root_level = root_item->level;
1812 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1813 super->cache_generation = root_item->generation;
1814 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1815 super->uuid_tree_generation = root_item->generation;
1818 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1820 struct btrfs_transaction *trans;
1823 spin_lock(&info->trans_lock);
1824 trans = info->running_transaction;
1826 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1827 spin_unlock(&info->trans_lock);
1831 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1833 struct btrfs_transaction *trans;
1836 spin_lock(&info->trans_lock);
1837 trans = info->running_transaction;
1839 ret = is_transaction_blocked(trans);
1840 spin_unlock(&info->trans_lock);
1845 * wait for the current transaction commit to start and block subsequent
1848 static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info,
1849 struct btrfs_transaction *trans)
1851 wait_event(fs_info->transaction_blocked_wait,
1852 trans->state >= TRANS_STATE_COMMIT_START ||
1853 TRANS_ABORTED(trans));
1857 * wait for the current transaction to start and then become unblocked.
1860 static void wait_current_trans_commit_start_and_unblock(
1861 struct btrfs_fs_info *fs_info,
1862 struct btrfs_transaction *trans)
1864 wait_event(fs_info->transaction_wait,
1865 trans->state >= TRANS_STATE_UNBLOCKED ||
1866 TRANS_ABORTED(trans));
1870 * commit transactions asynchronously. once btrfs_commit_transaction_async
1871 * returns, any subsequent transaction will not be allowed to join.
1873 struct btrfs_async_commit {
1874 struct btrfs_trans_handle *newtrans;
1875 struct work_struct work;
1878 static void do_async_commit(struct work_struct *work)
1880 struct btrfs_async_commit *ac =
1881 container_of(work, struct btrfs_async_commit, work);
1884 * We've got freeze protection passed with the transaction.
1885 * Tell lockdep about it.
1887 if (ac->newtrans->type & __TRANS_FREEZABLE)
1888 __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1890 current->journal_info = ac->newtrans;
1892 btrfs_commit_transaction(ac->newtrans);
1896 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1897 int wait_for_unblock)
1899 struct btrfs_fs_info *fs_info = trans->fs_info;
1900 struct btrfs_async_commit *ac;
1901 struct btrfs_transaction *cur_trans;
1903 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1907 INIT_WORK(&ac->work, do_async_commit);
1908 ac->newtrans = btrfs_join_transaction(trans->root);
1909 if (IS_ERR(ac->newtrans)) {
1910 int err = PTR_ERR(ac->newtrans);
1915 /* take transaction reference */
1916 cur_trans = trans->transaction;
1917 refcount_inc(&cur_trans->use_count);
1919 btrfs_end_transaction(trans);
1922 * Tell lockdep we've released the freeze rwsem, since the
1923 * async commit thread will be the one to unlock it.
1925 if (ac->newtrans->type & __TRANS_FREEZABLE)
1926 __sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1928 schedule_work(&ac->work);
1930 /* wait for transaction to start and unblock */
1931 if (wait_for_unblock)
1932 wait_current_trans_commit_start_and_unblock(fs_info, cur_trans);
1934 wait_current_trans_commit_start(fs_info, cur_trans);
1936 if (current->journal_info == trans)
1937 current->journal_info = NULL;
1939 btrfs_put_transaction(cur_trans);
1944 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
1946 struct btrfs_fs_info *fs_info = trans->fs_info;
1947 struct btrfs_transaction *cur_trans = trans->transaction;
1949 WARN_ON(refcount_read(&trans->use_count) > 1);
1951 btrfs_abort_transaction(trans, err);
1953 spin_lock(&fs_info->trans_lock);
1956 * If the transaction is removed from the list, it means this
1957 * transaction has been committed successfully, so it is impossible
1958 * to call the cleanup function.
1960 BUG_ON(list_empty(&cur_trans->list));
1962 if (cur_trans == fs_info->running_transaction) {
1963 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1964 spin_unlock(&fs_info->trans_lock);
1965 wait_event(cur_trans->writer_wait,
1966 atomic_read(&cur_trans->num_writers) == 1);
1968 spin_lock(&fs_info->trans_lock);
1972 * Now that we know no one else is still using the transaction we can
1973 * remove the transaction from the list of transactions. This avoids
1974 * the transaction kthread from cleaning up the transaction while some
1975 * other task is still using it, which could result in a use-after-free
1976 * on things like log trees, as it forces the transaction kthread to
1977 * wait for this transaction to be cleaned up by us.
1979 list_del_init(&cur_trans->list);
1981 spin_unlock(&fs_info->trans_lock);
1983 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
1985 spin_lock(&fs_info->trans_lock);
1986 if (cur_trans == fs_info->running_transaction)
1987 fs_info->running_transaction = NULL;
1988 spin_unlock(&fs_info->trans_lock);
1990 if (trans->type & __TRANS_FREEZABLE)
1991 sb_end_intwrite(fs_info->sb);
1992 btrfs_put_transaction(cur_trans);
1993 btrfs_put_transaction(cur_trans);
1995 trace_btrfs_transaction_commit(trans->root);
1997 if (current->journal_info == trans)
1998 current->journal_info = NULL;
1999 btrfs_scrub_cancel(fs_info);
2001 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2005 * Release reserved delayed ref space of all pending block groups of the
2006 * transaction and remove them from the list
2008 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2010 struct btrfs_fs_info *fs_info = trans->fs_info;
2011 struct btrfs_block_group *block_group, *tmp;
2013 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2014 btrfs_delayed_refs_rsv_release(fs_info, 1);
2015 list_del_init(&block_group->bg_list);
2019 static inline int btrfs_start_delalloc_flush(struct btrfs_trans_handle *trans)
2021 struct btrfs_fs_info *fs_info = trans->fs_info;
2024 * We use writeback_inodes_sb here because if we used
2025 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2026 * Currently are holding the fs freeze lock, if we do an async flush
2027 * we'll do btrfs_join_transaction() and deadlock because we need to
2028 * wait for the fs freeze lock. Using the direct flushing we benefit
2029 * from already being in a transaction and our join_transaction doesn't
2030 * have to re-take the fs freeze lock.
2032 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) {
2033 writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2035 struct btrfs_pending_snapshot *pending;
2036 struct list_head *head = &trans->transaction->pending_snapshots;
2039 * Flush dellaloc for any root that is going to be snapshotted.
2040 * This is done to avoid a corrupted version of files, in the
2041 * snapshots, that had both buffered and direct IO writes (even
2042 * if they were done sequentially) due to an unordered update of
2043 * the inode's size on disk.
2045 list_for_each_entry(pending, head, list) {
2048 ret = btrfs_start_delalloc_snapshot(pending->root);
2056 static inline void btrfs_wait_delalloc_flush(struct btrfs_trans_handle *trans)
2058 struct btrfs_fs_info *fs_info = trans->fs_info;
2060 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) {
2061 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2063 struct btrfs_pending_snapshot *pending;
2064 struct list_head *head = &trans->transaction->pending_snapshots;
2067 * Wait for any dellaloc that we started previously for the roots
2068 * that are going to be snapshotted. This is to avoid a corrupted
2069 * version of files in the snapshots that had both buffered and
2070 * direct IO writes (even if they were done sequentially).
2072 list_for_each_entry(pending, head, list)
2073 btrfs_wait_ordered_extents(pending->root,
2074 U64_MAX, 0, U64_MAX);
2078 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2080 struct btrfs_fs_info *fs_info = trans->fs_info;
2081 struct btrfs_transaction *cur_trans = trans->transaction;
2082 struct btrfs_transaction *prev_trans = NULL;
2085 ASSERT(refcount_read(&trans->use_count) == 1);
2088 * Some places just start a transaction to commit it. We need to make
2089 * sure that if this commit fails that the abort code actually marks the
2090 * transaction as failed, so set trans->dirty to make the abort code do
2093 trans->dirty = true;
2095 /* Stop the commit early if ->aborted is set */
2096 if (TRANS_ABORTED(cur_trans)) {
2097 ret = cur_trans->aborted;
2098 btrfs_end_transaction(trans);
2102 btrfs_trans_release_metadata(trans);
2103 trans->block_rsv = NULL;
2105 /* make a pass through all the delayed refs we have so far
2106 * any runnings procs may add more while we are here
2108 ret = btrfs_run_delayed_refs(trans, 0);
2110 btrfs_end_transaction(trans);
2114 cur_trans = trans->transaction;
2117 * set the flushing flag so procs in this transaction have to
2118 * start sending their work down.
2120 cur_trans->delayed_refs.flushing = 1;
2123 btrfs_create_pending_block_groups(trans);
2125 ret = btrfs_run_delayed_refs(trans, 0);
2127 btrfs_end_transaction(trans);
2131 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2134 /* this mutex is also taken before trying to set
2135 * block groups readonly. We need to make sure
2136 * that nobody has set a block group readonly
2137 * after a extents from that block group have been
2138 * allocated for cache files. btrfs_set_block_group_ro
2139 * will wait for the transaction to commit if it
2140 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2142 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2143 * only one process starts all the block group IO. It wouldn't
2144 * hurt to have more than one go through, but there's no
2145 * real advantage to it either.
2147 mutex_lock(&fs_info->ro_block_group_mutex);
2148 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2151 mutex_unlock(&fs_info->ro_block_group_mutex);
2154 ret = btrfs_start_dirty_block_groups(trans);
2156 btrfs_end_transaction(trans);
2162 spin_lock(&fs_info->trans_lock);
2163 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2164 spin_unlock(&fs_info->trans_lock);
2165 refcount_inc(&cur_trans->use_count);
2166 ret = btrfs_end_transaction(trans);
2168 wait_for_commit(cur_trans);
2170 if (TRANS_ABORTED(cur_trans))
2171 ret = cur_trans->aborted;
2173 btrfs_put_transaction(cur_trans);
2178 cur_trans->state = TRANS_STATE_COMMIT_START;
2179 wake_up(&fs_info->transaction_blocked_wait);
2181 if (cur_trans->list.prev != &fs_info->trans_list) {
2182 prev_trans = list_entry(cur_trans->list.prev,
2183 struct btrfs_transaction, list);
2184 if (prev_trans->state != TRANS_STATE_COMPLETED) {
2185 refcount_inc(&prev_trans->use_count);
2186 spin_unlock(&fs_info->trans_lock);
2188 wait_for_commit(prev_trans);
2189 ret = READ_ONCE(prev_trans->aborted);
2191 btrfs_put_transaction(prev_trans);
2193 goto cleanup_transaction;
2195 spin_unlock(&fs_info->trans_lock);
2198 spin_unlock(&fs_info->trans_lock);
2200 * The previous transaction was aborted and was already removed
2201 * from the list of transactions at fs_info->trans_list. So we
2202 * abort to prevent writing a new superblock that reflects a
2203 * corrupt state (pointing to trees with unwritten nodes/leafs).
2205 if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) {
2207 goto cleanup_transaction;
2211 extwriter_counter_dec(cur_trans, trans->type);
2213 ret = btrfs_start_delalloc_flush(trans);
2215 goto cleanup_transaction;
2217 ret = btrfs_run_delayed_items(trans);
2219 goto cleanup_transaction;
2221 wait_event(cur_trans->writer_wait,
2222 extwriter_counter_read(cur_trans) == 0);
2224 /* some pending stuffs might be added after the previous flush. */
2225 ret = btrfs_run_delayed_items(trans);
2227 goto cleanup_transaction;
2229 btrfs_wait_delalloc_flush(trans);
2232 * Wait for all ordered extents started by a fast fsync that joined this
2233 * transaction. Otherwise if this transaction commits before the ordered
2234 * extents complete we lose logged data after a power failure.
2236 wait_event(cur_trans->pending_wait,
2237 atomic_read(&cur_trans->pending_ordered) == 0);
2239 btrfs_scrub_pause(fs_info);
2241 * Ok now we need to make sure to block out any other joins while we
2242 * commit the transaction. We could have started a join before setting
2243 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2245 spin_lock(&fs_info->trans_lock);
2246 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2247 spin_unlock(&fs_info->trans_lock);
2248 wait_event(cur_trans->writer_wait,
2249 atomic_read(&cur_trans->num_writers) == 1);
2251 if (TRANS_ABORTED(cur_trans)) {
2252 ret = cur_trans->aborted;
2253 goto scrub_continue;
2256 * the reloc mutex makes sure that we stop
2257 * the balancing code from coming in and moving
2258 * extents around in the middle of the commit
2260 mutex_lock(&fs_info->reloc_mutex);
2263 * We needn't worry about the delayed items because we will
2264 * deal with them in create_pending_snapshot(), which is the
2265 * core function of the snapshot creation.
2267 ret = create_pending_snapshots(trans);
2272 * We insert the dir indexes of the snapshots and update the inode
2273 * of the snapshots' parents after the snapshot creation, so there
2274 * are some delayed items which are not dealt with. Now deal with
2277 * We needn't worry that this operation will corrupt the snapshots,
2278 * because all the tree which are snapshoted will be forced to COW
2279 * the nodes and leaves.
2281 ret = btrfs_run_delayed_items(trans);
2285 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2290 * make sure none of the code above managed to slip in a
2293 btrfs_assert_delayed_root_empty(fs_info);
2295 WARN_ON(cur_trans != trans->transaction);
2297 /* btrfs_commit_tree_roots is responsible for getting the
2298 * various roots consistent with each other. Every pointer
2299 * in the tree of tree roots has to point to the most up to date
2300 * root for every subvolume and other tree. So, we have to keep
2301 * the tree logging code from jumping in and changing any
2304 * At this point in the commit, there can't be any tree-log
2305 * writers, but a little lower down we drop the trans mutex
2306 * and let new people in. By holding the tree_log_mutex
2307 * from now until after the super is written, we avoid races
2308 * with the tree-log code.
2310 mutex_lock(&fs_info->tree_log_mutex);
2312 ret = commit_fs_roots(trans);
2314 goto unlock_tree_log;
2317 * Since the transaction is done, we can apply the pending changes
2318 * before the next transaction.
2320 btrfs_apply_pending_changes(fs_info);
2322 /* commit_fs_roots gets rid of all the tree log roots, it is now
2323 * safe to free the root of tree log roots
2325 btrfs_free_log_root_tree(trans, fs_info);
2328 * commit_fs_roots() can call btrfs_save_ino_cache(), which generates
2329 * new delayed refs. Must handle them or qgroup can be wrong.
2331 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2333 goto unlock_tree_log;
2336 * Since fs roots are all committed, we can get a quite accurate
2337 * new_roots. So let's do quota accounting.
2339 ret = btrfs_qgroup_account_extents(trans);
2341 goto unlock_tree_log;
2343 ret = commit_cowonly_roots(trans);
2345 goto unlock_tree_log;
2348 * The tasks which save the space cache and inode cache may also
2349 * update ->aborted, check it.
2351 if (TRANS_ABORTED(cur_trans)) {
2352 ret = cur_trans->aborted;
2353 goto unlock_tree_log;
2356 cur_trans = fs_info->running_transaction;
2358 btrfs_set_root_node(&fs_info->tree_root->root_item,
2359 fs_info->tree_root->node);
2360 list_add_tail(&fs_info->tree_root->dirty_list,
2361 &cur_trans->switch_commits);
2363 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2364 fs_info->chunk_root->node);
2365 list_add_tail(&fs_info->chunk_root->dirty_list,
2366 &cur_trans->switch_commits);
2368 switch_commit_roots(trans);
2370 ASSERT(list_empty(&cur_trans->dirty_bgs));
2371 ASSERT(list_empty(&cur_trans->io_bgs));
2372 update_super_roots(fs_info);
2374 btrfs_set_super_log_root(fs_info->super_copy, 0);
2375 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2376 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2377 sizeof(*fs_info->super_copy));
2379 btrfs_commit_device_sizes(cur_trans);
2381 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2382 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2384 btrfs_trans_release_chunk_metadata(trans);
2386 spin_lock(&fs_info->trans_lock);
2387 cur_trans->state = TRANS_STATE_UNBLOCKED;
2388 fs_info->running_transaction = NULL;
2389 spin_unlock(&fs_info->trans_lock);
2390 mutex_unlock(&fs_info->reloc_mutex);
2392 wake_up(&fs_info->transaction_wait);
2394 ret = btrfs_write_and_wait_transaction(trans);
2396 btrfs_handle_fs_error(fs_info, ret,
2397 "Error while writing out transaction");
2399 * reloc_mutex has been unlocked, tree_log_mutex is still held
2400 * but we can't jump to unlock_tree_log causing double unlock
2402 mutex_unlock(&fs_info->tree_log_mutex);
2403 goto scrub_continue;
2406 ret = write_all_supers(fs_info, 0);
2408 * the super is written, we can safely allow the tree-loggers
2409 * to go about their business
2411 mutex_unlock(&fs_info->tree_log_mutex);
2413 goto scrub_continue;
2415 btrfs_finish_extent_commit(trans);
2417 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2418 btrfs_clear_space_info_full(fs_info);
2420 fs_info->last_trans_committed = cur_trans->transid;
2422 * We needn't acquire the lock here because there is no other task
2423 * which can change it.
2425 cur_trans->state = TRANS_STATE_COMPLETED;
2426 wake_up(&cur_trans->commit_wait);
2428 spin_lock(&fs_info->trans_lock);
2429 list_del_init(&cur_trans->list);
2430 spin_unlock(&fs_info->trans_lock);
2432 btrfs_put_transaction(cur_trans);
2433 btrfs_put_transaction(cur_trans);
2435 if (trans->type & __TRANS_FREEZABLE)
2436 sb_end_intwrite(fs_info->sb);
2438 trace_btrfs_transaction_commit(trans->root);
2440 btrfs_scrub_continue(fs_info);
2442 if (current->journal_info == trans)
2443 current->journal_info = NULL;
2445 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2450 mutex_unlock(&fs_info->tree_log_mutex);
2452 mutex_unlock(&fs_info->reloc_mutex);
2454 btrfs_scrub_continue(fs_info);
2455 cleanup_transaction:
2456 btrfs_trans_release_metadata(trans);
2457 btrfs_cleanup_pending_block_groups(trans);
2458 btrfs_trans_release_chunk_metadata(trans);
2459 trans->block_rsv = NULL;
2460 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2461 if (current->journal_info == trans)
2462 current->journal_info = NULL;
2463 cleanup_transaction(trans, ret);
2469 * return < 0 if error
2470 * 0 if there are no more dead_roots at the time of call
2471 * 1 there are more to be processed, call me again
2473 * The return value indicates there are certainly more snapshots to delete, but
2474 * if there comes a new one during processing, it may return 0. We don't mind,
2475 * because btrfs_commit_super will poke cleaner thread and it will process it a
2476 * few seconds later.
2478 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2481 struct btrfs_fs_info *fs_info = root->fs_info;
2483 spin_lock(&fs_info->trans_lock);
2484 if (list_empty(&fs_info->dead_roots)) {
2485 spin_unlock(&fs_info->trans_lock);
2488 root = list_first_entry(&fs_info->dead_roots,
2489 struct btrfs_root, root_list);
2490 list_del_init(&root->root_list);
2491 spin_unlock(&fs_info->trans_lock);
2493 btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2495 btrfs_kill_all_delayed_nodes(root);
2496 if (root->ino_cache_inode) {
2497 iput(root->ino_cache_inode);
2498 root->ino_cache_inode = NULL;
2501 if (btrfs_header_backref_rev(root->node) <
2502 BTRFS_MIXED_BACKREF_REV)
2503 ret = btrfs_drop_snapshot(root, 0, 0);
2505 ret = btrfs_drop_snapshot(root, 1, 0);
2507 btrfs_put_root(root);
2508 return (ret < 0) ? 0 : 1;
2511 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2516 prev = xchg(&fs_info->pending_changes, 0);
2520 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2522 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2525 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2527 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2530 bit = 1 << BTRFS_PENDING_COMMIT;
2532 btrfs_debug(fs_info, "pending commit done");
2537 "unknown pending changes left 0x%lx, ignoring", prev);