2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/slab.h>
21 #include <linux/sched.h>
22 #include <linux/writeback.h>
23 #include <linux/pagemap.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
28 #include "transaction.h"
31 #include "inode-map.h"
33 #include "dev-replace.h"
35 #define BTRFS_ROOT_TRANS_TAG 0
37 static unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
38 [TRANS_STATE_RUNNING] = 0U,
39 [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE |
41 [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE |
44 [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE |
48 [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE |
53 [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE |
60 void btrfs_put_transaction(struct btrfs_transaction *transaction)
62 WARN_ON(atomic_read(&transaction->use_count) == 0);
63 if (atomic_dec_and_test(&transaction->use_count)) {
64 BUG_ON(!list_empty(&transaction->list));
65 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
66 while (!list_empty(&transaction->pending_chunks)) {
67 struct extent_map *em;
69 em = list_first_entry(&transaction->pending_chunks,
70 struct extent_map, list);
71 list_del_init(&em->list);
74 kmem_cache_free(btrfs_transaction_cachep, transaction);
78 static noinline void switch_commit_root(struct btrfs_root *root)
80 free_extent_buffer(root->commit_root);
81 root->commit_root = btrfs_root_node(root);
84 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
87 if (type & TRANS_EXTWRITERS)
88 atomic_inc(&trans->num_extwriters);
91 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
94 if (type & TRANS_EXTWRITERS)
95 atomic_dec(&trans->num_extwriters);
98 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
101 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
104 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
106 return atomic_read(&trans->num_extwriters);
110 * either allocate a new transaction or hop into the existing one
112 static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
114 struct btrfs_transaction *cur_trans;
115 struct btrfs_fs_info *fs_info = root->fs_info;
117 spin_lock(&fs_info->trans_lock);
119 /* The file system has been taken offline. No new transactions. */
120 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
121 spin_unlock(&fs_info->trans_lock);
125 cur_trans = fs_info->running_transaction;
127 if (cur_trans->aborted) {
128 spin_unlock(&fs_info->trans_lock);
129 return cur_trans->aborted;
131 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
132 spin_unlock(&fs_info->trans_lock);
135 atomic_inc(&cur_trans->use_count);
136 atomic_inc(&cur_trans->num_writers);
137 extwriter_counter_inc(cur_trans, type);
138 spin_unlock(&fs_info->trans_lock);
141 spin_unlock(&fs_info->trans_lock);
144 * If we are ATTACH, we just want to catch the current transaction,
145 * and commit it. If there is no transaction, just return ENOENT.
147 if (type == TRANS_ATTACH)
151 * JOIN_NOLOCK only happens during the transaction commit, so
152 * it is impossible that ->running_transaction is NULL
154 BUG_ON(type == TRANS_JOIN_NOLOCK);
156 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
160 spin_lock(&fs_info->trans_lock);
161 if (fs_info->running_transaction) {
163 * someone started a transaction after we unlocked. Make sure
164 * to redo the checks above
166 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
168 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
169 spin_unlock(&fs_info->trans_lock);
170 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
174 atomic_set(&cur_trans->num_writers, 1);
175 extwriter_counter_init(cur_trans, type);
176 init_waitqueue_head(&cur_trans->writer_wait);
177 init_waitqueue_head(&cur_trans->commit_wait);
178 cur_trans->state = TRANS_STATE_RUNNING;
180 * One for this trans handle, one so it will live on until we
181 * commit the transaction.
183 atomic_set(&cur_trans->use_count, 2);
184 cur_trans->start_time = get_seconds();
186 cur_trans->delayed_refs.href_root = RB_ROOT;
187 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
188 cur_trans->delayed_refs.num_heads_ready = 0;
189 cur_trans->delayed_refs.num_heads = 0;
190 cur_trans->delayed_refs.flushing = 0;
191 cur_trans->delayed_refs.run_delayed_start = 0;
194 * although the tree mod log is per file system and not per transaction,
195 * the log must never go across transaction boundaries.
198 if (!list_empty(&fs_info->tree_mod_seq_list))
199 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
200 "creating a fresh transaction\n");
201 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
202 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
203 "creating a fresh transaction\n");
204 atomic64_set(&fs_info->tree_mod_seq, 0);
206 spin_lock_init(&cur_trans->delayed_refs.lock);
208 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
209 INIT_LIST_HEAD(&cur_trans->ordered_operations);
210 INIT_LIST_HEAD(&cur_trans->pending_chunks);
211 list_add_tail(&cur_trans->list, &fs_info->trans_list);
212 extent_io_tree_init(&cur_trans->dirty_pages,
213 fs_info->btree_inode->i_mapping);
214 fs_info->generation++;
215 cur_trans->transid = fs_info->generation;
216 fs_info->running_transaction = cur_trans;
217 cur_trans->aborted = 0;
218 spin_unlock(&fs_info->trans_lock);
224 * this does all the record keeping required to make sure that a reference
225 * counted root is properly recorded in a given transaction. This is required
226 * to make sure the old root from before we joined the transaction is deleted
227 * when the transaction commits
229 static int record_root_in_trans(struct btrfs_trans_handle *trans,
230 struct btrfs_root *root)
232 if (root->ref_cows && root->last_trans < trans->transid) {
233 WARN_ON(root == root->fs_info->extent_root);
234 WARN_ON(root->commit_root != root->node);
237 * see below for in_trans_setup usage rules
238 * we have the reloc mutex held now, so there
239 * is only one writer in this function
241 root->in_trans_setup = 1;
243 /* make sure readers find in_trans_setup before
244 * they find our root->last_trans update
248 spin_lock(&root->fs_info->fs_roots_radix_lock);
249 if (root->last_trans == trans->transid) {
250 spin_unlock(&root->fs_info->fs_roots_radix_lock);
253 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
254 (unsigned long)root->root_key.objectid,
255 BTRFS_ROOT_TRANS_TAG);
256 spin_unlock(&root->fs_info->fs_roots_radix_lock);
257 root->last_trans = trans->transid;
259 /* this is pretty tricky. We don't want to
260 * take the relocation lock in btrfs_record_root_in_trans
261 * unless we're really doing the first setup for this root in
264 * Normally we'd use root->last_trans as a flag to decide
265 * if we want to take the expensive mutex.
267 * But, we have to set root->last_trans before we
268 * init the relocation root, otherwise, we trip over warnings
269 * in ctree.c. The solution used here is to flag ourselves
270 * with root->in_trans_setup. When this is 1, we're still
271 * fixing up the reloc trees and everyone must wait.
273 * When this is zero, they can trust root->last_trans and fly
274 * through btrfs_record_root_in_trans without having to take the
275 * lock. smp_wmb() makes sure that all the writes above are
276 * done before we pop in the zero below
278 btrfs_init_reloc_root(trans, root);
280 root->in_trans_setup = 0;
286 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
287 struct btrfs_root *root)
293 * see record_root_in_trans for comments about in_trans_setup usage
297 if (root->last_trans == trans->transid &&
298 !root->in_trans_setup)
301 mutex_lock(&root->fs_info->reloc_mutex);
302 record_root_in_trans(trans, root);
303 mutex_unlock(&root->fs_info->reloc_mutex);
308 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
310 return (trans->state >= TRANS_STATE_BLOCKED &&
311 trans->state < TRANS_STATE_UNBLOCKED &&
315 /* wait for commit against the current transaction to become unblocked
316 * when this is done, it is safe to start a new transaction, but the current
317 * transaction might not be fully on disk.
319 static void wait_current_trans(struct btrfs_root *root)
321 struct btrfs_transaction *cur_trans;
323 spin_lock(&root->fs_info->trans_lock);
324 cur_trans = root->fs_info->running_transaction;
325 if (cur_trans && is_transaction_blocked(cur_trans)) {
326 atomic_inc(&cur_trans->use_count);
327 spin_unlock(&root->fs_info->trans_lock);
329 wait_event(root->fs_info->transaction_wait,
330 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
332 btrfs_put_transaction(cur_trans);
334 spin_unlock(&root->fs_info->trans_lock);
338 static int may_wait_transaction(struct btrfs_root *root, int type)
340 if (root->fs_info->log_root_recovering)
343 if (type == TRANS_USERSPACE)
346 if (type == TRANS_START &&
347 !atomic_read(&root->fs_info->open_ioctl_trans))
353 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
355 if (!root->fs_info->reloc_ctl ||
357 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
364 static struct btrfs_trans_handle *
365 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
366 enum btrfs_reserve_flush_enum flush)
368 struct btrfs_trans_handle *h;
369 struct btrfs_transaction *cur_trans;
371 u64 qgroup_reserved = 0;
372 bool reloc_reserved = false;
375 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
376 return ERR_PTR(-EROFS);
378 if (current->journal_info) {
379 WARN_ON(type & TRANS_EXTWRITERS);
380 h = current->journal_info;
382 WARN_ON(h->use_count > 2);
383 h->orig_rsv = h->block_rsv;
389 * Do the reservation before we join the transaction so we can do all
390 * the appropriate flushing if need be.
392 if (num_items > 0 && root != root->fs_info->chunk_root) {
393 if (root->fs_info->quota_enabled &&
394 is_fstree(root->root_key.objectid)) {
395 qgroup_reserved = num_items * root->leafsize;
396 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
401 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
403 * Do the reservation for the relocation root creation
405 if (unlikely(need_reserve_reloc_root(root))) {
406 num_bytes += root->nodesize;
407 reloc_reserved = true;
410 ret = btrfs_block_rsv_add(root,
411 &root->fs_info->trans_block_rsv,
417 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
424 * If we are JOIN_NOLOCK we're already committing a transaction and
425 * waiting on this guy, so we don't need to do the sb_start_intwrite
426 * because we're already holding a ref. We need this because we could
427 * have raced in and did an fsync() on a file which can kick a commit
428 * and then we deadlock with somebody doing a freeze.
430 * If we are ATTACH, it means we just want to catch the current
431 * transaction and commit it, so we needn't do sb_start_intwrite().
433 if (type & __TRANS_FREEZABLE)
434 sb_start_intwrite(root->fs_info->sb);
436 if (may_wait_transaction(root, type))
437 wait_current_trans(root);
440 ret = join_transaction(root, type);
442 wait_current_trans(root);
443 if (unlikely(type == TRANS_ATTACH))
446 } while (ret == -EBUSY);
449 /* We must get the transaction if we are JOIN_NOLOCK. */
450 BUG_ON(type == TRANS_JOIN_NOLOCK);
454 cur_trans = root->fs_info->running_transaction;
456 h->transid = cur_trans->transid;
457 h->transaction = cur_trans;
459 h->bytes_reserved = 0;
461 h->delayed_ref_updates = 0;
467 h->qgroup_reserved = 0;
468 h->delayed_ref_elem.seq = 0;
470 h->allocating_chunk = false;
471 h->reloc_reserved = false;
473 INIT_LIST_HEAD(&h->qgroup_ref_list);
474 INIT_LIST_HEAD(&h->new_bgs);
477 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
478 may_wait_transaction(root, type)) {
479 btrfs_commit_transaction(h, root);
484 trace_btrfs_space_reservation(root->fs_info, "transaction",
485 h->transid, num_bytes, 1);
486 h->block_rsv = &root->fs_info->trans_block_rsv;
487 h->bytes_reserved = num_bytes;
488 h->reloc_reserved = reloc_reserved;
490 h->qgroup_reserved = qgroup_reserved;
493 btrfs_record_root_in_trans(h, root);
495 if (!current->journal_info && type != TRANS_USERSPACE)
496 current->journal_info = h;
500 if (type & __TRANS_FREEZABLE)
501 sb_end_intwrite(root->fs_info->sb);
502 kmem_cache_free(btrfs_trans_handle_cachep, h);
505 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
509 btrfs_qgroup_free(root, qgroup_reserved);
513 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
516 return start_transaction(root, num_items, TRANS_START,
517 BTRFS_RESERVE_FLUSH_ALL);
520 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
521 struct btrfs_root *root, int num_items)
523 return start_transaction(root, num_items, TRANS_START,
524 BTRFS_RESERVE_FLUSH_LIMIT);
527 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
529 return start_transaction(root, 0, TRANS_JOIN, 0);
532 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
534 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
537 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
539 return start_transaction(root, 0, TRANS_USERSPACE, 0);
543 * btrfs_attach_transaction() - catch the running transaction
545 * It is used when we want to commit the current the transaction, but
546 * don't want to start a new one.
548 * Note: If this function return -ENOENT, it just means there is no
549 * running transaction. But it is possible that the inactive transaction
550 * is still in the memory, not fully on disk. If you hope there is no
551 * inactive transaction in the fs when -ENOENT is returned, you should
553 * btrfs_attach_transaction_barrier()
555 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
557 return start_transaction(root, 0, TRANS_ATTACH, 0);
561 * btrfs_attach_transaction_barrier() - catch the running transaction
563 * It is similar to the above function, the differentia is this one
564 * will wait for all the inactive transactions until they fully
567 struct btrfs_trans_handle *
568 btrfs_attach_transaction_barrier(struct btrfs_root *root)
570 struct btrfs_trans_handle *trans;
572 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
573 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
574 btrfs_wait_for_commit(root, 0);
579 /* wait for a transaction commit to be fully complete */
580 static noinline void wait_for_commit(struct btrfs_root *root,
581 struct btrfs_transaction *commit)
583 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
586 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
588 struct btrfs_transaction *cur_trans = NULL, *t;
592 if (transid <= root->fs_info->last_trans_committed)
596 /* find specified transaction */
597 spin_lock(&root->fs_info->trans_lock);
598 list_for_each_entry(t, &root->fs_info->trans_list, list) {
599 if (t->transid == transid) {
601 atomic_inc(&cur_trans->use_count);
605 if (t->transid > transid) {
610 spin_unlock(&root->fs_info->trans_lock);
611 /* The specified transaction doesn't exist */
615 /* find newest transaction that is committing | committed */
616 spin_lock(&root->fs_info->trans_lock);
617 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
619 if (t->state >= TRANS_STATE_COMMIT_START) {
620 if (t->state == TRANS_STATE_COMPLETED)
623 atomic_inc(&cur_trans->use_count);
627 spin_unlock(&root->fs_info->trans_lock);
629 goto out; /* nothing committing|committed */
632 wait_for_commit(root, cur_trans);
633 btrfs_put_transaction(cur_trans);
638 void btrfs_throttle(struct btrfs_root *root)
640 if (!atomic_read(&root->fs_info->open_ioctl_trans))
641 wait_current_trans(root);
644 static int should_end_transaction(struct btrfs_trans_handle *trans,
645 struct btrfs_root *root)
647 if (root->fs_info->global_block_rsv.space_info->full &&
648 btrfs_check_space_for_delayed_refs(trans, root))
651 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
654 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
655 struct btrfs_root *root)
657 struct btrfs_transaction *cur_trans = trans->transaction;
662 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
663 cur_trans->delayed_refs.flushing)
666 updates = trans->delayed_ref_updates;
667 trans->delayed_ref_updates = 0;
669 err = btrfs_run_delayed_refs(trans, root, updates);
670 if (err) /* Error code will also eval true */
674 return should_end_transaction(trans, root);
677 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
678 struct btrfs_root *root, int throttle)
680 struct btrfs_transaction *cur_trans = trans->transaction;
681 struct btrfs_fs_info *info = root->fs_info;
682 unsigned long cur = trans->delayed_ref_updates;
683 int lock = (trans->type != TRANS_JOIN_NOLOCK);
686 if (--trans->use_count) {
687 trans->block_rsv = trans->orig_rsv;
692 * do the qgroup accounting as early as possible
694 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
696 btrfs_trans_release_metadata(trans, root);
697 trans->block_rsv = NULL;
699 if (trans->qgroup_reserved) {
701 * the same root has to be passed here between start_transaction
702 * and end_transaction. Subvolume quota depends on this.
704 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
705 trans->qgroup_reserved = 0;
708 if (!list_empty(&trans->new_bgs))
709 btrfs_create_pending_block_groups(trans, root);
711 trans->delayed_ref_updates = 0;
712 if (!trans->sync && btrfs_should_throttle_delayed_refs(trans, root)) {
713 cur = max_t(unsigned long, cur, 32);
714 trans->delayed_ref_updates = 0;
715 btrfs_run_delayed_refs(trans, root, cur);
718 btrfs_trans_release_metadata(trans, root);
719 trans->block_rsv = NULL;
721 if (!list_empty(&trans->new_bgs))
722 btrfs_create_pending_block_groups(trans, root);
724 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
725 should_end_transaction(trans, root) &&
726 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
727 spin_lock(&info->trans_lock);
728 if (cur_trans->state == TRANS_STATE_RUNNING)
729 cur_trans->state = TRANS_STATE_BLOCKED;
730 spin_unlock(&info->trans_lock);
733 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
736 * We may race with somebody else here so end up having
737 * to call end_transaction on ourselves again, so inc
741 return btrfs_commit_transaction(trans, root);
743 wake_up_process(info->transaction_kthread);
747 if (trans->type & __TRANS_FREEZABLE)
748 sb_end_intwrite(root->fs_info->sb);
750 WARN_ON(cur_trans != info->running_transaction);
751 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
752 atomic_dec(&cur_trans->num_writers);
753 extwriter_counter_dec(cur_trans, trans->type);
756 if (waitqueue_active(&cur_trans->writer_wait))
757 wake_up(&cur_trans->writer_wait);
758 btrfs_put_transaction(cur_trans);
760 if (current->journal_info == trans)
761 current->journal_info = NULL;
764 btrfs_run_delayed_iputs(root);
766 if (trans->aborted ||
767 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
768 wake_up_process(info->transaction_kthread);
771 assert_qgroups_uptodate(trans);
773 kmem_cache_free(btrfs_trans_handle_cachep, trans);
777 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
778 struct btrfs_root *root)
780 return __btrfs_end_transaction(trans, root, 0);
783 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
784 struct btrfs_root *root)
786 return __btrfs_end_transaction(trans, root, 1);
790 * when btree blocks are allocated, they have some corresponding bits set for
791 * them in one of two extent_io trees. This is used to make sure all of
792 * those extents are sent to disk but does not wait on them
794 int btrfs_write_marked_extents(struct btrfs_root *root,
795 struct extent_io_tree *dirty_pages, int mark)
799 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
800 struct extent_state *cached_state = NULL;
804 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
805 mark, &cached_state)) {
806 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
807 mark, &cached_state, GFP_NOFS);
809 err = filemap_fdatawrite_range(mapping, start, end);
821 * when btree blocks are allocated, they have some corresponding bits set for
822 * them in one of two extent_io trees. This is used to make sure all of
823 * those extents are on disk for transaction or log commit. We wait
824 * on all the pages and clear them from the dirty pages state tree
826 int btrfs_wait_marked_extents(struct btrfs_root *root,
827 struct extent_io_tree *dirty_pages, int mark)
831 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
832 struct extent_state *cached_state = NULL;
836 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
837 EXTENT_NEED_WAIT, &cached_state)) {
838 clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
839 0, 0, &cached_state, GFP_NOFS);
840 err = filemap_fdatawait_range(mapping, start, end);
852 * when btree blocks are allocated, they have some corresponding bits set for
853 * them in one of two extent_io trees. This is used to make sure all of
854 * those extents are on disk for transaction or log commit
856 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
857 struct extent_io_tree *dirty_pages, int mark)
861 struct blk_plug plug;
863 blk_start_plug(&plug);
864 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
865 blk_finish_plug(&plug);
866 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
875 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
876 struct btrfs_root *root)
878 if (!trans || !trans->transaction) {
879 struct inode *btree_inode;
880 btree_inode = root->fs_info->btree_inode;
881 return filemap_write_and_wait(btree_inode->i_mapping);
883 return btrfs_write_and_wait_marked_extents(root,
884 &trans->transaction->dirty_pages,
889 * this is used to update the root pointer in the tree of tree roots.
891 * But, in the case of the extent allocation tree, updating the root
892 * pointer may allocate blocks which may change the root of the extent
895 * So, this loops and repeats and makes sure the cowonly root didn't
896 * change while the root pointer was being updated in the metadata.
898 static int update_cowonly_root(struct btrfs_trans_handle *trans,
899 struct btrfs_root *root)
904 struct btrfs_root *tree_root = root->fs_info->tree_root;
906 old_root_used = btrfs_root_used(&root->root_item);
907 btrfs_write_dirty_block_groups(trans, root);
910 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
911 if (old_root_bytenr == root->node->start &&
912 old_root_used == btrfs_root_used(&root->root_item))
915 btrfs_set_root_node(&root->root_item, root->node);
916 ret = btrfs_update_root(trans, tree_root,
922 old_root_used = btrfs_root_used(&root->root_item);
923 ret = btrfs_write_dirty_block_groups(trans, root);
928 if (root != root->fs_info->extent_root)
929 switch_commit_root(root);
935 * update all the cowonly tree roots on disk
937 * The error handling in this function may not be obvious. Any of the
938 * failures will cause the file system to go offline. We still need
939 * to clean up the delayed refs.
941 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
942 struct btrfs_root *root)
944 struct btrfs_fs_info *fs_info = root->fs_info;
945 struct list_head *next;
946 struct extent_buffer *eb;
949 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
953 eb = btrfs_lock_root_node(fs_info->tree_root);
954 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
956 btrfs_tree_unlock(eb);
957 free_extent_buffer(eb);
962 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
966 ret = btrfs_run_dev_stats(trans, root->fs_info);
969 ret = btrfs_run_dev_replace(trans, root->fs_info);
972 ret = btrfs_run_qgroups(trans, root->fs_info);
976 /* run_qgroups might have added some more refs */
977 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
981 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
982 next = fs_info->dirty_cowonly_roots.next;
984 root = list_entry(next, struct btrfs_root, dirty_list);
986 ret = update_cowonly_root(trans, root);
991 down_write(&fs_info->extent_commit_sem);
992 switch_commit_root(fs_info->extent_root);
993 up_write(&fs_info->extent_commit_sem);
995 btrfs_after_dev_replace_commit(fs_info);
1001 * dead roots are old snapshots that need to be deleted. This allocates
1002 * a dirty root struct and adds it into the list of dead roots that need to
1005 void btrfs_add_dead_root(struct btrfs_root *root)
1007 spin_lock(&root->fs_info->trans_lock);
1008 if (list_empty(&root->root_list))
1009 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1010 spin_unlock(&root->fs_info->trans_lock);
1014 * update all the cowonly tree roots on disk
1016 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1017 struct btrfs_root *root)
1019 struct btrfs_root *gang[8];
1020 struct btrfs_fs_info *fs_info = root->fs_info;
1025 spin_lock(&fs_info->fs_roots_radix_lock);
1027 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1030 BTRFS_ROOT_TRANS_TAG);
1033 for (i = 0; i < ret; i++) {
1035 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1036 (unsigned long)root->root_key.objectid,
1037 BTRFS_ROOT_TRANS_TAG);
1038 spin_unlock(&fs_info->fs_roots_radix_lock);
1040 btrfs_free_log(trans, root);
1041 btrfs_update_reloc_root(trans, root);
1042 btrfs_orphan_commit_root(trans, root);
1044 btrfs_save_ino_cache(root, trans);
1046 /* see comments in should_cow_block() */
1047 root->force_cow = 0;
1050 if (root->commit_root != root->node) {
1051 mutex_lock(&root->fs_commit_mutex);
1052 switch_commit_root(root);
1053 btrfs_unpin_free_ino(root);
1054 mutex_unlock(&root->fs_commit_mutex);
1056 btrfs_set_root_node(&root->root_item,
1060 err = btrfs_update_root(trans, fs_info->tree_root,
1063 spin_lock(&fs_info->fs_roots_radix_lock);
1068 spin_unlock(&fs_info->fs_roots_radix_lock);
1073 * defrag a given btree.
1074 * Every leaf in the btree is read and defragged.
1076 int btrfs_defrag_root(struct btrfs_root *root)
1078 struct btrfs_fs_info *info = root->fs_info;
1079 struct btrfs_trans_handle *trans;
1082 if (xchg(&root->defrag_running, 1))
1086 trans = btrfs_start_transaction(root, 0);
1088 return PTR_ERR(trans);
1090 ret = btrfs_defrag_leaves(trans, root);
1092 btrfs_end_transaction(trans, root);
1093 btrfs_btree_balance_dirty(info->tree_root);
1096 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1099 if (btrfs_defrag_cancelled(root->fs_info)) {
1100 pr_debug("BTRFS: defrag_root cancelled\n");
1105 root->defrag_running = 0;
1110 * new snapshots need to be created at a very specific time in the
1111 * transaction commit. This does the actual creation.
1114 * If the error which may affect the commitment of the current transaction
1115 * happens, we should return the error number. If the error which just affect
1116 * the creation of the pending snapshots, just return 0.
1118 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1119 struct btrfs_fs_info *fs_info,
1120 struct btrfs_pending_snapshot *pending)
1122 struct btrfs_key key;
1123 struct btrfs_root_item *new_root_item;
1124 struct btrfs_root *tree_root = fs_info->tree_root;
1125 struct btrfs_root *root = pending->root;
1126 struct btrfs_root *parent_root;
1127 struct btrfs_block_rsv *rsv;
1128 struct inode *parent_inode;
1129 struct btrfs_path *path;
1130 struct btrfs_dir_item *dir_item;
1131 struct dentry *dentry;
1132 struct extent_buffer *tmp;
1133 struct extent_buffer *old;
1134 struct timespec cur_time = CURRENT_TIME;
1142 path = btrfs_alloc_path();
1144 pending->error = -ENOMEM;
1148 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1149 if (!new_root_item) {
1150 pending->error = -ENOMEM;
1151 goto root_item_alloc_fail;
1154 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1156 goto no_free_objectid;
1158 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1160 if (to_reserve > 0) {
1161 pending->error = btrfs_block_rsv_add(root,
1162 &pending->block_rsv,
1164 BTRFS_RESERVE_NO_FLUSH);
1166 goto no_free_objectid;
1169 pending->error = btrfs_qgroup_inherit(trans, fs_info,
1170 root->root_key.objectid,
1171 objectid, pending->inherit);
1173 goto no_free_objectid;
1175 key.objectid = objectid;
1176 key.offset = (u64)-1;
1177 key.type = BTRFS_ROOT_ITEM_KEY;
1179 rsv = trans->block_rsv;
1180 trans->block_rsv = &pending->block_rsv;
1181 trans->bytes_reserved = trans->block_rsv->reserved;
1183 dentry = pending->dentry;
1184 parent_inode = pending->dir;
1185 parent_root = BTRFS_I(parent_inode)->root;
1186 record_root_in_trans(trans, parent_root);
1189 * insert the directory item
1191 ret = btrfs_set_inode_index(parent_inode, &index);
1192 BUG_ON(ret); /* -ENOMEM */
1194 /* check if there is a file/dir which has the same name. */
1195 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1196 btrfs_ino(parent_inode),
1197 dentry->d_name.name,
1198 dentry->d_name.len, 0);
1199 if (dir_item != NULL && !IS_ERR(dir_item)) {
1200 pending->error = -EEXIST;
1201 goto dir_item_existed;
1202 } else if (IS_ERR(dir_item)) {
1203 ret = PTR_ERR(dir_item);
1204 btrfs_abort_transaction(trans, root, ret);
1207 btrfs_release_path(path);
1210 * pull in the delayed directory update
1211 * and the delayed inode item
1212 * otherwise we corrupt the FS during
1215 ret = btrfs_run_delayed_items(trans, root);
1216 if (ret) { /* Transaction aborted */
1217 btrfs_abort_transaction(trans, root, ret);
1221 record_root_in_trans(trans, root);
1222 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1223 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1224 btrfs_check_and_init_root_item(new_root_item);
1226 root_flags = btrfs_root_flags(new_root_item);
1227 if (pending->readonly)
1228 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1230 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1231 btrfs_set_root_flags(new_root_item, root_flags);
1233 btrfs_set_root_generation_v2(new_root_item,
1235 uuid_le_gen(&new_uuid);
1236 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1237 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1239 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1240 memset(new_root_item->received_uuid, 0,
1241 sizeof(new_root_item->received_uuid));
1242 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1243 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1244 btrfs_set_root_stransid(new_root_item, 0);
1245 btrfs_set_root_rtransid(new_root_item, 0);
1247 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1248 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1249 btrfs_set_root_otransid(new_root_item, trans->transid);
1251 old = btrfs_lock_root_node(root);
1252 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1254 btrfs_tree_unlock(old);
1255 free_extent_buffer(old);
1256 btrfs_abort_transaction(trans, root, ret);
1260 btrfs_set_lock_blocking(old);
1262 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1263 /* clean up in any case */
1264 btrfs_tree_unlock(old);
1265 free_extent_buffer(old);
1267 btrfs_abort_transaction(trans, root, ret);
1271 /* see comments in should_cow_block() */
1272 root->force_cow = 1;
1275 btrfs_set_root_node(new_root_item, tmp);
1276 /* record when the snapshot was created in key.offset */
1277 key.offset = trans->transid;
1278 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1279 btrfs_tree_unlock(tmp);
1280 free_extent_buffer(tmp);
1282 btrfs_abort_transaction(trans, root, ret);
1287 * insert root back/forward references
1289 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1290 parent_root->root_key.objectid,
1291 btrfs_ino(parent_inode), index,
1292 dentry->d_name.name, dentry->d_name.len);
1294 btrfs_abort_transaction(trans, root, ret);
1298 key.offset = (u64)-1;
1299 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1300 if (IS_ERR(pending->snap)) {
1301 ret = PTR_ERR(pending->snap);
1302 btrfs_abort_transaction(trans, root, ret);
1306 ret = btrfs_reloc_post_snapshot(trans, pending);
1308 btrfs_abort_transaction(trans, root, ret);
1312 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1314 btrfs_abort_transaction(trans, root, ret);
1318 ret = btrfs_insert_dir_item(trans, parent_root,
1319 dentry->d_name.name, dentry->d_name.len,
1321 BTRFS_FT_DIR, index);
1322 /* We have check then name at the beginning, so it is impossible. */
1323 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1325 btrfs_abort_transaction(trans, root, ret);
1329 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1330 dentry->d_name.len * 2);
1331 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1332 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1334 btrfs_abort_transaction(trans, root, ret);
1337 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1338 BTRFS_UUID_KEY_SUBVOL, objectid);
1340 btrfs_abort_transaction(trans, root, ret);
1343 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1344 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1345 new_root_item->received_uuid,
1346 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1348 if (ret && ret != -EEXIST) {
1349 btrfs_abort_transaction(trans, root, ret);
1354 pending->error = ret;
1356 trans->block_rsv = rsv;
1357 trans->bytes_reserved = 0;
1359 kfree(new_root_item);
1360 root_item_alloc_fail:
1361 btrfs_free_path(path);
1366 * create all the snapshots we've scheduled for creation
1368 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1369 struct btrfs_fs_info *fs_info)
1371 struct btrfs_pending_snapshot *pending, *next;
1372 struct list_head *head = &trans->transaction->pending_snapshots;
1375 list_for_each_entry_safe(pending, next, head, list) {
1376 list_del(&pending->list);
1377 ret = create_pending_snapshot(trans, fs_info, pending);
1384 static void update_super_roots(struct btrfs_root *root)
1386 struct btrfs_root_item *root_item;
1387 struct btrfs_super_block *super;
1389 super = root->fs_info->super_copy;
1391 root_item = &root->fs_info->chunk_root->root_item;
1392 super->chunk_root = root_item->bytenr;
1393 super->chunk_root_generation = root_item->generation;
1394 super->chunk_root_level = root_item->level;
1396 root_item = &root->fs_info->tree_root->root_item;
1397 super->root = root_item->bytenr;
1398 super->generation = root_item->generation;
1399 super->root_level = root_item->level;
1400 if (btrfs_test_opt(root, SPACE_CACHE))
1401 super->cache_generation = root_item->generation;
1402 if (root->fs_info->update_uuid_tree_gen)
1403 super->uuid_tree_generation = root_item->generation;
1406 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1408 struct btrfs_transaction *trans;
1411 spin_lock(&info->trans_lock);
1412 trans = info->running_transaction;
1414 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1415 spin_unlock(&info->trans_lock);
1419 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1421 struct btrfs_transaction *trans;
1424 spin_lock(&info->trans_lock);
1425 trans = info->running_transaction;
1427 ret = is_transaction_blocked(trans);
1428 spin_unlock(&info->trans_lock);
1433 * wait for the current transaction commit to start and block subsequent
1436 static void wait_current_trans_commit_start(struct btrfs_root *root,
1437 struct btrfs_transaction *trans)
1439 wait_event(root->fs_info->transaction_blocked_wait,
1440 trans->state >= TRANS_STATE_COMMIT_START ||
1445 * wait for the current transaction to start and then become unblocked.
1448 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1449 struct btrfs_transaction *trans)
1451 wait_event(root->fs_info->transaction_wait,
1452 trans->state >= TRANS_STATE_UNBLOCKED ||
1457 * commit transactions asynchronously. once btrfs_commit_transaction_async
1458 * returns, any subsequent transaction will not be allowed to join.
1460 struct btrfs_async_commit {
1461 struct btrfs_trans_handle *newtrans;
1462 struct btrfs_root *root;
1463 struct work_struct work;
1466 static void do_async_commit(struct work_struct *work)
1468 struct btrfs_async_commit *ac =
1469 container_of(work, struct btrfs_async_commit, work);
1472 * We've got freeze protection passed with the transaction.
1473 * Tell lockdep about it.
1475 if (ac->newtrans->type & __TRANS_FREEZABLE)
1477 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1480 current->journal_info = ac->newtrans;
1482 btrfs_commit_transaction(ac->newtrans, ac->root);
1486 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1487 struct btrfs_root *root,
1488 int wait_for_unblock)
1490 struct btrfs_async_commit *ac;
1491 struct btrfs_transaction *cur_trans;
1493 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1497 INIT_WORK(&ac->work, do_async_commit);
1499 ac->newtrans = btrfs_join_transaction(root);
1500 if (IS_ERR(ac->newtrans)) {
1501 int err = PTR_ERR(ac->newtrans);
1506 /* take transaction reference */
1507 cur_trans = trans->transaction;
1508 atomic_inc(&cur_trans->use_count);
1510 btrfs_end_transaction(trans, root);
1513 * Tell lockdep we've released the freeze rwsem, since the
1514 * async commit thread will be the one to unlock it.
1516 if (ac->newtrans->type & __TRANS_FREEZABLE)
1518 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1521 schedule_work(&ac->work);
1523 /* wait for transaction to start and unblock */
1524 if (wait_for_unblock)
1525 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1527 wait_current_trans_commit_start(root, cur_trans);
1529 if (current->journal_info == trans)
1530 current->journal_info = NULL;
1532 btrfs_put_transaction(cur_trans);
1537 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1538 struct btrfs_root *root, int err)
1540 struct btrfs_transaction *cur_trans = trans->transaction;
1543 WARN_ON(trans->use_count > 1);
1545 btrfs_abort_transaction(trans, root, err);
1547 spin_lock(&root->fs_info->trans_lock);
1550 * If the transaction is removed from the list, it means this
1551 * transaction has been committed successfully, so it is impossible
1552 * to call the cleanup function.
1554 BUG_ON(list_empty(&cur_trans->list));
1556 list_del_init(&cur_trans->list);
1557 if (cur_trans == root->fs_info->running_transaction) {
1558 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1559 spin_unlock(&root->fs_info->trans_lock);
1560 wait_event(cur_trans->writer_wait,
1561 atomic_read(&cur_trans->num_writers) == 1);
1563 spin_lock(&root->fs_info->trans_lock);
1565 spin_unlock(&root->fs_info->trans_lock);
1567 btrfs_cleanup_one_transaction(trans->transaction, root);
1569 spin_lock(&root->fs_info->trans_lock);
1570 if (cur_trans == root->fs_info->running_transaction)
1571 root->fs_info->running_transaction = NULL;
1572 spin_unlock(&root->fs_info->trans_lock);
1574 if (trans->type & __TRANS_FREEZABLE)
1575 sb_end_intwrite(root->fs_info->sb);
1576 btrfs_put_transaction(cur_trans);
1577 btrfs_put_transaction(cur_trans);
1579 trace_btrfs_transaction_commit(root);
1581 if (current->journal_info == trans)
1582 current->journal_info = NULL;
1583 btrfs_scrub_cancel(root->fs_info);
1585 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1588 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans,
1589 struct btrfs_root *root)
1593 ret = btrfs_run_delayed_items(trans, root);
1595 * running the delayed items may have added new refs. account
1596 * them now so that they hinder processing of more delayed refs
1597 * as little as possible.
1600 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1604 ret = btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1609 * rename don't use btrfs_join_transaction, so, once we
1610 * set the transaction to blocked above, we aren't going
1611 * to get any new ordered operations. We can safely run
1612 * it here and no for sure that nothing new will be added
1615 ret = btrfs_run_ordered_operations(trans, root, 1);
1620 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1622 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1623 return btrfs_start_delalloc_roots(fs_info, 1);
1627 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1629 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1630 btrfs_wait_ordered_roots(fs_info, -1);
1633 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1634 struct btrfs_root *root)
1636 struct btrfs_transaction *cur_trans = trans->transaction;
1637 struct btrfs_transaction *prev_trans = NULL;
1640 ret = btrfs_run_ordered_operations(trans, root, 0);
1642 btrfs_abort_transaction(trans, root, ret);
1643 btrfs_end_transaction(trans, root);
1647 /* Stop the commit early if ->aborted is set */
1648 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1649 ret = cur_trans->aborted;
1650 btrfs_end_transaction(trans, root);
1654 /* make a pass through all the delayed refs we have so far
1655 * any runnings procs may add more while we are here
1657 ret = btrfs_run_delayed_refs(trans, root, 0);
1659 btrfs_end_transaction(trans, root);
1663 btrfs_trans_release_metadata(trans, root);
1664 trans->block_rsv = NULL;
1665 if (trans->qgroup_reserved) {
1666 btrfs_qgroup_free(root, trans->qgroup_reserved);
1667 trans->qgroup_reserved = 0;
1670 cur_trans = trans->transaction;
1673 * set the flushing flag so procs in this transaction have to
1674 * start sending their work down.
1676 cur_trans->delayed_refs.flushing = 1;
1679 if (!list_empty(&trans->new_bgs))
1680 btrfs_create_pending_block_groups(trans, root);
1682 ret = btrfs_run_delayed_refs(trans, root, 0);
1684 btrfs_end_transaction(trans, root);
1688 spin_lock(&root->fs_info->trans_lock);
1689 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1690 spin_unlock(&root->fs_info->trans_lock);
1691 atomic_inc(&cur_trans->use_count);
1692 ret = btrfs_end_transaction(trans, root);
1694 wait_for_commit(root, cur_trans);
1696 btrfs_put_transaction(cur_trans);
1701 cur_trans->state = TRANS_STATE_COMMIT_START;
1702 wake_up(&root->fs_info->transaction_blocked_wait);
1704 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1705 prev_trans = list_entry(cur_trans->list.prev,
1706 struct btrfs_transaction, list);
1707 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1708 atomic_inc(&prev_trans->use_count);
1709 spin_unlock(&root->fs_info->trans_lock);
1711 wait_for_commit(root, prev_trans);
1713 btrfs_put_transaction(prev_trans);
1715 spin_unlock(&root->fs_info->trans_lock);
1718 spin_unlock(&root->fs_info->trans_lock);
1721 extwriter_counter_dec(cur_trans, trans->type);
1723 ret = btrfs_start_delalloc_flush(root->fs_info);
1725 goto cleanup_transaction;
1727 ret = btrfs_flush_all_pending_stuffs(trans, root);
1729 goto cleanup_transaction;
1731 wait_event(cur_trans->writer_wait,
1732 extwriter_counter_read(cur_trans) == 0);
1734 /* some pending stuffs might be added after the previous flush. */
1735 ret = btrfs_flush_all_pending_stuffs(trans, root);
1737 goto cleanup_transaction;
1739 btrfs_wait_delalloc_flush(root->fs_info);
1741 btrfs_scrub_pause(root);
1743 * Ok now we need to make sure to block out any other joins while we
1744 * commit the transaction. We could have started a join before setting
1745 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1747 spin_lock(&root->fs_info->trans_lock);
1748 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1749 spin_unlock(&root->fs_info->trans_lock);
1750 wait_event(cur_trans->writer_wait,
1751 atomic_read(&cur_trans->num_writers) == 1);
1753 /* ->aborted might be set after the previous check, so check it */
1754 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1755 ret = cur_trans->aborted;
1756 goto scrub_continue;
1759 * the reloc mutex makes sure that we stop
1760 * the balancing code from coming in and moving
1761 * extents around in the middle of the commit
1763 mutex_lock(&root->fs_info->reloc_mutex);
1766 * We needn't worry about the delayed items because we will
1767 * deal with them in create_pending_snapshot(), which is the
1768 * core function of the snapshot creation.
1770 ret = create_pending_snapshots(trans, root->fs_info);
1772 mutex_unlock(&root->fs_info->reloc_mutex);
1773 goto scrub_continue;
1777 * We insert the dir indexes of the snapshots and update the inode
1778 * of the snapshots' parents after the snapshot creation, so there
1779 * are some delayed items which are not dealt with. Now deal with
1782 * We needn't worry that this operation will corrupt the snapshots,
1783 * because all the tree which are snapshoted will be forced to COW
1784 * the nodes and leaves.
1786 ret = btrfs_run_delayed_items(trans, root);
1788 mutex_unlock(&root->fs_info->reloc_mutex);
1789 goto scrub_continue;
1792 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1794 mutex_unlock(&root->fs_info->reloc_mutex);
1795 goto scrub_continue;
1799 * make sure none of the code above managed to slip in a
1802 btrfs_assert_delayed_root_empty(root);
1804 WARN_ON(cur_trans != trans->transaction);
1806 /* btrfs_commit_tree_roots is responsible for getting the
1807 * various roots consistent with each other. Every pointer
1808 * in the tree of tree roots has to point to the most up to date
1809 * root for every subvolume and other tree. So, we have to keep
1810 * the tree logging code from jumping in and changing any
1813 * At this point in the commit, there can't be any tree-log
1814 * writers, but a little lower down we drop the trans mutex
1815 * and let new people in. By holding the tree_log_mutex
1816 * from now until after the super is written, we avoid races
1817 * with the tree-log code.
1819 mutex_lock(&root->fs_info->tree_log_mutex);
1821 ret = commit_fs_roots(trans, root);
1823 mutex_unlock(&root->fs_info->tree_log_mutex);
1824 mutex_unlock(&root->fs_info->reloc_mutex);
1825 goto scrub_continue;
1829 * Since the transaction is done, we should set the inode map cache flag
1830 * before any other comming transaction.
1832 if (btrfs_test_opt(root, CHANGE_INODE_CACHE))
1833 btrfs_set_opt(root->fs_info->mount_opt, INODE_MAP_CACHE);
1835 btrfs_clear_opt(root->fs_info->mount_opt, INODE_MAP_CACHE);
1837 /* commit_fs_roots gets rid of all the tree log roots, it is now
1838 * safe to free the root of tree log roots
1840 btrfs_free_log_root_tree(trans, root->fs_info);
1842 ret = commit_cowonly_roots(trans, root);
1844 mutex_unlock(&root->fs_info->tree_log_mutex);
1845 mutex_unlock(&root->fs_info->reloc_mutex);
1846 goto scrub_continue;
1850 * The tasks which save the space cache and inode cache may also
1851 * update ->aborted, check it.
1853 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1854 ret = cur_trans->aborted;
1855 mutex_unlock(&root->fs_info->tree_log_mutex);
1856 mutex_unlock(&root->fs_info->reloc_mutex);
1857 goto scrub_continue;
1860 btrfs_prepare_extent_commit(trans, root);
1862 cur_trans = root->fs_info->running_transaction;
1864 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1865 root->fs_info->tree_root->node);
1866 switch_commit_root(root->fs_info->tree_root);
1868 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1869 root->fs_info->chunk_root->node);
1870 switch_commit_root(root->fs_info->chunk_root);
1872 assert_qgroups_uptodate(trans);
1873 update_super_roots(root);
1875 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1876 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1877 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1878 sizeof(*root->fs_info->super_copy));
1880 spin_lock(&root->fs_info->trans_lock);
1881 cur_trans->state = TRANS_STATE_UNBLOCKED;
1882 root->fs_info->running_transaction = NULL;
1883 spin_unlock(&root->fs_info->trans_lock);
1884 mutex_unlock(&root->fs_info->reloc_mutex);
1886 wake_up(&root->fs_info->transaction_wait);
1888 ret = btrfs_write_and_wait_transaction(trans, root);
1890 btrfs_error(root->fs_info, ret,
1891 "Error while writing out transaction");
1892 mutex_unlock(&root->fs_info->tree_log_mutex);
1893 goto scrub_continue;
1896 ret = write_ctree_super(trans, root, 0);
1898 mutex_unlock(&root->fs_info->tree_log_mutex);
1899 goto scrub_continue;
1903 * the super is written, we can safely allow the tree-loggers
1904 * to go about their business
1906 mutex_unlock(&root->fs_info->tree_log_mutex);
1908 btrfs_finish_extent_commit(trans, root);
1910 root->fs_info->last_trans_committed = cur_trans->transid;
1912 * We needn't acquire the lock here because there is no other task
1913 * which can change it.
1915 cur_trans->state = TRANS_STATE_COMPLETED;
1916 wake_up(&cur_trans->commit_wait);
1918 spin_lock(&root->fs_info->trans_lock);
1919 list_del_init(&cur_trans->list);
1920 spin_unlock(&root->fs_info->trans_lock);
1922 btrfs_put_transaction(cur_trans);
1923 btrfs_put_transaction(cur_trans);
1925 if (trans->type & __TRANS_FREEZABLE)
1926 sb_end_intwrite(root->fs_info->sb);
1928 trace_btrfs_transaction_commit(root);
1930 btrfs_scrub_continue(root);
1932 if (current->journal_info == trans)
1933 current->journal_info = NULL;
1935 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1937 if (current != root->fs_info->transaction_kthread)
1938 btrfs_run_delayed_iputs(root);
1943 btrfs_scrub_continue(root);
1944 cleanup_transaction:
1945 btrfs_trans_release_metadata(trans, root);
1946 trans->block_rsv = NULL;
1947 if (trans->qgroup_reserved) {
1948 btrfs_qgroup_free(root, trans->qgroup_reserved);
1949 trans->qgroup_reserved = 0;
1951 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
1952 if (current->journal_info == trans)
1953 current->journal_info = NULL;
1954 cleanup_transaction(trans, root, ret);
1960 * return < 0 if error
1961 * 0 if there are no more dead_roots at the time of call
1962 * 1 there are more to be processed, call me again
1964 * The return value indicates there are certainly more snapshots to delete, but
1965 * if there comes a new one during processing, it may return 0. We don't mind,
1966 * because btrfs_commit_super will poke cleaner thread and it will process it a
1967 * few seconds later.
1969 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
1972 struct btrfs_fs_info *fs_info = root->fs_info;
1974 spin_lock(&fs_info->trans_lock);
1975 if (list_empty(&fs_info->dead_roots)) {
1976 spin_unlock(&fs_info->trans_lock);
1979 root = list_first_entry(&fs_info->dead_roots,
1980 struct btrfs_root, root_list);
1982 * Make sure root is not involved in send,
1983 * if we fail with first root, we return
1984 * directly rather than continue.
1986 spin_lock(&root->root_item_lock);
1987 if (root->send_in_progress) {
1988 spin_unlock(&fs_info->trans_lock);
1989 spin_unlock(&root->root_item_lock);
1992 spin_unlock(&root->root_item_lock);
1994 list_del_init(&root->root_list);
1995 spin_unlock(&fs_info->trans_lock);
1997 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
1999 btrfs_kill_all_delayed_nodes(root);
2001 if (btrfs_header_backref_rev(root->node) <
2002 BTRFS_MIXED_BACKREF_REV)
2003 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2005 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2007 * If we encounter a transaction abort during snapshot cleaning, we
2008 * don't want to crash here
2010 return (ret < 0) ? 0 : 1;