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 static void 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(transaction->delayed_refs.root.rb_node);
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.root = RB_ROOT;
187 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);
207 atomic_set(&cur_trans->delayed_refs.procs_running_refs, 0);
208 atomic_set(&cur_trans->delayed_refs.ref_seq, 0);
209 init_waitqueue_head(&cur_trans->delayed_refs.wait);
211 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
212 INIT_LIST_HEAD(&cur_trans->ordered_operations);
213 INIT_LIST_HEAD(&cur_trans->pending_chunks);
214 list_add_tail(&cur_trans->list, &fs_info->trans_list);
215 extent_io_tree_init(&cur_trans->dirty_pages,
216 fs_info->btree_inode->i_mapping);
217 fs_info->generation++;
218 cur_trans->transid = fs_info->generation;
219 fs_info->running_transaction = cur_trans;
220 cur_trans->aborted = 0;
221 spin_unlock(&fs_info->trans_lock);
227 * this does all the record keeping required to make sure that a reference
228 * counted root is properly recorded in a given transaction. This is required
229 * to make sure the old root from before we joined the transaction is deleted
230 * when the transaction commits
232 static int record_root_in_trans(struct btrfs_trans_handle *trans,
233 struct btrfs_root *root)
235 if (root->ref_cows && root->last_trans < trans->transid) {
236 WARN_ON(root == root->fs_info->extent_root);
237 WARN_ON(root->commit_root != root->node);
240 * see below for in_trans_setup usage rules
241 * we have the reloc mutex held now, so there
242 * is only one writer in this function
244 root->in_trans_setup = 1;
246 /* make sure readers find in_trans_setup before
247 * they find our root->last_trans update
251 spin_lock(&root->fs_info->fs_roots_radix_lock);
252 if (root->last_trans == trans->transid) {
253 spin_unlock(&root->fs_info->fs_roots_radix_lock);
256 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
257 (unsigned long)root->root_key.objectid,
258 BTRFS_ROOT_TRANS_TAG);
259 spin_unlock(&root->fs_info->fs_roots_radix_lock);
260 root->last_trans = trans->transid;
262 /* this is pretty tricky. We don't want to
263 * take the relocation lock in btrfs_record_root_in_trans
264 * unless we're really doing the first setup for this root in
267 * Normally we'd use root->last_trans as a flag to decide
268 * if we want to take the expensive mutex.
270 * But, we have to set root->last_trans before we
271 * init the relocation root, otherwise, we trip over warnings
272 * in ctree.c. The solution used here is to flag ourselves
273 * with root->in_trans_setup. When this is 1, we're still
274 * fixing up the reloc trees and everyone must wait.
276 * When this is zero, they can trust root->last_trans and fly
277 * through btrfs_record_root_in_trans without having to take the
278 * lock. smp_wmb() makes sure that all the writes above are
279 * done before we pop in the zero below
281 btrfs_init_reloc_root(trans, root);
283 root->in_trans_setup = 0;
289 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
290 struct btrfs_root *root)
296 * see record_root_in_trans for comments about in_trans_setup usage
300 if (root->last_trans == trans->transid &&
301 !root->in_trans_setup)
304 mutex_lock(&root->fs_info->reloc_mutex);
305 record_root_in_trans(trans, root);
306 mutex_unlock(&root->fs_info->reloc_mutex);
311 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
313 return (trans->state >= TRANS_STATE_BLOCKED &&
314 trans->state < TRANS_STATE_UNBLOCKED &&
318 /* wait for commit against the current transaction to become unblocked
319 * when this is done, it is safe to start a new transaction, but the current
320 * transaction might not be fully on disk.
322 static void wait_current_trans(struct btrfs_root *root)
324 struct btrfs_transaction *cur_trans;
326 spin_lock(&root->fs_info->trans_lock);
327 cur_trans = root->fs_info->running_transaction;
328 if (cur_trans && is_transaction_blocked(cur_trans)) {
329 atomic_inc(&cur_trans->use_count);
330 spin_unlock(&root->fs_info->trans_lock);
332 wait_event(root->fs_info->transaction_wait,
333 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
335 put_transaction(cur_trans);
337 spin_unlock(&root->fs_info->trans_lock);
341 static int may_wait_transaction(struct btrfs_root *root, int type)
343 if (root->fs_info->log_root_recovering)
346 if (type == TRANS_USERSPACE)
349 if (type == TRANS_START &&
350 !atomic_read(&root->fs_info->open_ioctl_trans))
356 static struct btrfs_trans_handle *
357 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
358 enum btrfs_reserve_flush_enum flush)
360 struct btrfs_trans_handle *h;
361 struct btrfs_transaction *cur_trans;
364 u64 qgroup_reserved = 0;
366 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
367 return ERR_PTR(-EROFS);
369 if (current->journal_info) {
370 WARN_ON(type & TRANS_EXTWRITERS);
371 h = current->journal_info;
373 WARN_ON(h->use_count > 2);
374 h->orig_rsv = h->block_rsv;
380 * Do the reservation before we join the transaction so we can do all
381 * the appropriate flushing if need be.
383 if (num_items > 0 && root != root->fs_info->chunk_root) {
384 if (root->fs_info->quota_enabled &&
385 is_fstree(root->root_key.objectid)) {
386 qgroup_reserved = num_items * root->leafsize;
387 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
392 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
393 ret = btrfs_block_rsv_add(root,
394 &root->fs_info->trans_block_rsv,
400 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
407 * If we are JOIN_NOLOCK we're already committing a transaction and
408 * waiting on this guy, so we don't need to do the sb_start_intwrite
409 * because we're already holding a ref. We need this because we could
410 * have raced in and did an fsync() on a file which can kick a commit
411 * and then we deadlock with somebody doing a freeze.
413 * If we are ATTACH, it means we just want to catch the current
414 * transaction and commit it, so we needn't do sb_start_intwrite().
416 if (type & __TRANS_FREEZABLE)
417 sb_start_intwrite(root->fs_info->sb);
419 if (may_wait_transaction(root, type))
420 wait_current_trans(root);
423 ret = join_transaction(root, type);
425 wait_current_trans(root);
426 if (unlikely(type == TRANS_ATTACH))
429 } while (ret == -EBUSY);
432 /* We must get the transaction if we are JOIN_NOLOCK. */
433 BUG_ON(type == TRANS_JOIN_NOLOCK);
437 cur_trans = root->fs_info->running_transaction;
439 h->transid = cur_trans->transid;
440 h->transaction = cur_trans;
442 h->bytes_reserved = 0;
444 h->delayed_ref_updates = 0;
450 h->qgroup_reserved = 0;
451 h->delayed_ref_elem.seq = 0;
453 h->allocating_chunk = false;
454 INIT_LIST_HEAD(&h->qgroup_ref_list);
455 INIT_LIST_HEAD(&h->new_bgs);
458 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
459 may_wait_transaction(root, type)) {
460 btrfs_commit_transaction(h, root);
465 trace_btrfs_space_reservation(root->fs_info, "transaction",
466 h->transid, num_bytes, 1);
467 h->block_rsv = &root->fs_info->trans_block_rsv;
468 h->bytes_reserved = num_bytes;
470 h->qgroup_reserved = qgroup_reserved;
473 btrfs_record_root_in_trans(h, root);
475 if (!current->journal_info && type != TRANS_USERSPACE)
476 current->journal_info = h;
480 if (type & __TRANS_FREEZABLE)
481 sb_end_intwrite(root->fs_info->sb);
482 kmem_cache_free(btrfs_trans_handle_cachep, h);
485 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
489 btrfs_qgroup_free(root, qgroup_reserved);
493 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
496 return start_transaction(root, num_items, TRANS_START,
497 BTRFS_RESERVE_FLUSH_ALL);
500 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
501 struct btrfs_root *root, int num_items)
503 return start_transaction(root, num_items, TRANS_START,
504 BTRFS_RESERVE_FLUSH_LIMIT);
507 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
509 return start_transaction(root, 0, TRANS_JOIN, 0);
512 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
514 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
517 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
519 return start_transaction(root, 0, TRANS_USERSPACE, 0);
523 * btrfs_attach_transaction() - catch the running transaction
525 * It is used when we want to commit the current the transaction, but
526 * don't want to start a new one.
528 * Note: If this function return -ENOENT, it just means there is no
529 * running transaction. But it is possible that the inactive transaction
530 * is still in the memory, not fully on disk. If you hope there is no
531 * inactive transaction in the fs when -ENOENT is returned, you should
533 * btrfs_attach_transaction_barrier()
535 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
537 return start_transaction(root, 0, TRANS_ATTACH, 0);
541 * btrfs_attach_transaction_barrier() - catch the running transaction
543 * It is similar to the above function, the differentia is this one
544 * will wait for all the inactive transactions until they fully
547 struct btrfs_trans_handle *
548 btrfs_attach_transaction_barrier(struct btrfs_root *root)
550 struct btrfs_trans_handle *trans;
552 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
553 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
554 btrfs_wait_for_commit(root, 0);
559 /* wait for a transaction commit to be fully complete */
560 static noinline void wait_for_commit(struct btrfs_root *root,
561 struct btrfs_transaction *commit)
563 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
566 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
568 struct btrfs_transaction *cur_trans = NULL, *t;
572 if (transid <= root->fs_info->last_trans_committed)
576 /* find specified transaction */
577 spin_lock(&root->fs_info->trans_lock);
578 list_for_each_entry(t, &root->fs_info->trans_list, list) {
579 if (t->transid == transid) {
581 atomic_inc(&cur_trans->use_count);
585 if (t->transid > transid) {
590 spin_unlock(&root->fs_info->trans_lock);
591 /* The specified transaction doesn't exist */
595 /* find newest transaction that is committing | committed */
596 spin_lock(&root->fs_info->trans_lock);
597 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
599 if (t->state >= TRANS_STATE_COMMIT_START) {
600 if (t->state == TRANS_STATE_COMPLETED)
603 atomic_inc(&cur_trans->use_count);
607 spin_unlock(&root->fs_info->trans_lock);
609 goto out; /* nothing committing|committed */
612 wait_for_commit(root, cur_trans);
613 put_transaction(cur_trans);
618 void btrfs_throttle(struct btrfs_root *root)
620 if (!atomic_read(&root->fs_info->open_ioctl_trans))
621 wait_current_trans(root);
624 static int should_end_transaction(struct btrfs_trans_handle *trans,
625 struct btrfs_root *root)
627 if (root->fs_info->global_block_rsv.space_info->full &&
628 btrfs_should_throttle_delayed_refs(trans, root))
631 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
634 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
635 struct btrfs_root *root)
637 struct btrfs_transaction *cur_trans = trans->transaction;
642 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
643 cur_trans->delayed_refs.flushing)
646 updates = trans->delayed_ref_updates;
647 trans->delayed_ref_updates = 0;
649 err = btrfs_run_delayed_refs(trans, root, updates);
650 if (err) /* Error code will also eval true */
654 return should_end_transaction(trans, root);
657 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
658 struct btrfs_root *root, int throttle)
660 struct btrfs_transaction *cur_trans = trans->transaction;
661 struct btrfs_fs_info *info = root->fs_info;
662 unsigned long cur = trans->delayed_ref_updates;
663 int lock = (trans->type != TRANS_JOIN_NOLOCK);
666 if (--trans->use_count) {
667 trans->block_rsv = trans->orig_rsv;
672 * do the qgroup accounting as early as possible
674 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
676 btrfs_trans_release_metadata(trans, root);
677 trans->block_rsv = NULL;
679 if (trans->qgroup_reserved) {
681 * the same root has to be passed here between start_transaction
682 * and end_transaction. Subvolume quota depends on this.
684 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
685 trans->qgroup_reserved = 0;
688 if (!list_empty(&trans->new_bgs))
689 btrfs_create_pending_block_groups(trans, root);
691 trans->delayed_ref_updates = 0;
692 if (btrfs_should_throttle_delayed_refs(trans, root)) {
693 cur = max_t(unsigned long, cur, 1);
694 trans->delayed_ref_updates = 0;
695 btrfs_run_delayed_refs(trans, root, cur);
698 btrfs_trans_release_metadata(trans, root);
699 trans->block_rsv = NULL;
701 if (!list_empty(&trans->new_bgs))
702 btrfs_create_pending_block_groups(trans, root);
704 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
705 should_end_transaction(trans, root) &&
706 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
707 spin_lock(&info->trans_lock);
708 if (cur_trans->state == TRANS_STATE_RUNNING)
709 cur_trans->state = TRANS_STATE_BLOCKED;
710 spin_unlock(&info->trans_lock);
713 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
716 * We may race with somebody else here so end up having
717 * to call end_transaction on ourselves again, so inc
721 return btrfs_commit_transaction(trans, root);
723 wake_up_process(info->transaction_kthread);
727 if (trans->type & __TRANS_FREEZABLE)
728 sb_end_intwrite(root->fs_info->sb);
730 WARN_ON(cur_trans != info->running_transaction);
731 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
732 atomic_dec(&cur_trans->num_writers);
733 extwriter_counter_dec(cur_trans, trans->type);
736 if (waitqueue_active(&cur_trans->writer_wait))
737 wake_up(&cur_trans->writer_wait);
738 put_transaction(cur_trans);
740 if (current->journal_info == trans)
741 current->journal_info = NULL;
744 btrfs_run_delayed_iputs(root);
746 if (trans->aborted ||
747 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
748 wake_up_process(info->transaction_kthread);
751 assert_qgroups_uptodate(trans);
753 kmem_cache_free(btrfs_trans_handle_cachep, trans);
757 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
758 struct btrfs_root *root)
760 return __btrfs_end_transaction(trans, root, 0);
763 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
764 struct btrfs_root *root)
766 return __btrfs_end_transaction(trans, root, 1);
769 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
770 struct btrfs_root *root)
772 return __btrfs_end_transaction(trans, root, 1);
776 * when btree blocks are allocated, they have some corresponding bits set for
777 * them in one of two extent_io trees. This is used to make sure all of
778 * those extents are sent to disk but does not wait on them
780 int btrfs_write_marked_extents(struct btrfs_root *root,
781 struct extent_io_tree *dirty_pages, int mark)
785 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
786 struct extent_state *cached_state = NULL;
790 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
791 mark, &cached_state)) {
792 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
793 mark, &cached_state, GFP_NOFS);
795 err = filemap_fdatawrite_range(mapping, start, end);
807 * when btree blocks are allocated, they have some corresponding bits set for
808 * them in one of two extent_io trees. This is used to make sure all of
809 * those extents are on disk for transaction or log commit. We wait
810 * on all the pages and clear them from the dirty pages state tree
812 int btrfs_wait_marked_extents(struct btrfs_root *root,
813 struct extent_io_tree *dirty_pages, int mark)
817 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
818 struct extent_state *cached_state = NULL;
822 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
823 EXTENT_NEED_WAIT, &cached_state)) {
824 clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
825 0, 0, &cached_state, GFP_NOFS);
826 err = filemap_fdatawait_range(mapping, start, end);
838 * when btree blocks are allocated, they have some corresponding bits set for
839 * them in one of two extent_io trees. This is used to make sure all of
840 * those extents are on disk for transaction or log commit
842 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
843 struct extent_io_tree *dirty_pages, int mark)
847 struct blk_plug plug;
849 blk_start_plug(&plug);
850 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
851 blk_finish_plug(&plug);
852 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
861 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
862 struct btrfs_root *root)
864 if (!trans || !trans->transaction) {
865 struct inode *btree_inode;
866 btree_inode = root->fs_info->btree_inode;
867 return filemap_write_and_wait(btree_inode->i_mapping);
869 return btrfs_write_and_wait_marked_extents(root,
870 &trans->transaction->dirty_pages,
875 * this is used to update the root pointer in the tree of tree roots.
877 * But, in the case of the extent allocation tree, updating the root
878 * pointer may allocate blocks which may change the root of the extent
881 * So, this loops and repeats and makes sure the cowonly root didn't
882 * change while the root pointer was being updated in the metadata.
884 static int update_cowonly_root(struct btrfs_trans_handle *trans,
885 struct btrfs_root *root)
890 struct btrfs_root *tree_root = root->fs_info->tree_root;
892 old_root_used = btrfs_root_used(&root->root_item);
893 btrfs_write_dirty_block_groups(trans, root);
896 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
897 if (old_root_bytenr == root->node->start &&
898 old_root_used == btrfs_root_used(&root->root_item))
901 btrfs_set_root_node(&root->root_item, root->node);
902 ret = btrfs_update_root(trans, tree_root,
908 old_root_used = btrfs_root_used(&root->root_item);
909 ret = btrfs_write_dirty_block_groups(trans, root);
914 if (root != root->fs_info->extent_root)
915 switch_commit_root(root);
921 * update all the cowonly tree roots on disk
923 * The error handling in this function may not be obvious. Any of the
924 * failures will cause the file system to go offline. We still need
925 * to clean up the delayed refs.
927 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
928 struct btrfs_root *root)
930 struct btrfs_fs_info *fs_info = root->fs_info;
931 struct list_head *next;
932 struct extent_buffer *eb;
935 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
939 eb = btrfs_lock_root_node(fs_info->tree_root);
940 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
942 btrfs_tree_unlock(eb);
943 free_extent_buffer(eb);
948 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
952 ret = btrfs_run_dev_stats(trans, root->fs_info);
954 ret = btrfs_run_dev_replace(trans, root->fs_info);
957 ret = btrfs_run_qgroups(trans, root->fs_info);
960 /* run_qgroups might have added some more refs */
961 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
964 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
965 next = fs_info->dirty_cowonly_roots.next;
967 root = list_entry(next, struct btrfs_root, dirty_list);
969 ret = update_cowonly_root(trans, root);
974 down_write(&fs_info->extent_commit_sem);
975 switch_commit_root(fs_info->extent_root);
976 up_write(&fs_info->extent_commit_sem);
978 btrfs_after_dev_replace_commit(fs_info);
984 * dead roots are old snapshots that need to be deleted. This allocates
985 * a dirty root struct and adds it into the list of dead roots that need to
988 void btrfs_add_dead_root(struct btrfs_root *root)
990 spin_lock(&root->fs_info->trans_lock);
991 if (list_empty(&root->root_list))
992 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
993 spin_unlock(&root->fs_info->trans_lock);
997 * update all the cowonly tree roots on disk
999 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1000 struct btrfs_root *root)
1002 struct btrfs_root *gang[8];
1003 struct btrfs_fs_info *fs_info = root->fs_info;
1008 spin_lock(&fs_info->fs_roots_radix_lock);
1010 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1013 BTRFS_ROOT_TRANS_TAG);
1016 for (i = 0; i < ret; i++) {
1018 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1019 (unsigned long)root->root_key.objectid,
1020 BTRFS_ROOT_TRANS_TAG);
1021 spin_unlock(&fs_info->fs_roots_radix_lock);
1023 btrfs_free_log(trans, root);
1024 btrfs_update_reloc_root(trans, root);
1025 btrfs_orphan_commit_root(trans, root);
1027 btrfs_save_ino_cache(root, trans);
1029 /* see comments in should_cow_block() */
1030 root->force_cow = 0;
1033 if (root->commit_root != root->node) {
1034 mutex_lock(&root->fs_commit_mutex);
1035 switch_commit_root(root);
1036 btrfs_unpin_free_ino(root);
1037 mutex_unlock(&root->fs_commit_mutex);
1039 btrfs_set_root_node(&root->root_item,
1043 err = btrfs_update_root(trans, fs_info->tree_root,
1046 spin_lock(&fs_info->fs_roots_radix_lock);
1051 spin_unlock(&fs_info->fs_roots_radix_lock);
1056 * defrag a given btree.
1057 * Every leaf in the btree is read and defragged.
1059 int btrfs_defrag_root(struct btrfs_root *root)
1061 struct btrfs_fs_info *info = root->fs_info;
1062 struct btrfs_trans_handle *trans;
1065 if (xchg(&root->defrag_running, 1))
1069 trans = btrfs_start_transaction(root, 0);
1071 return PTR_ERR(trans);
1073 ret = btrfs_defrag_leaves(trans, root);
1075 btrfs_end_transaction(trans, root);
1076 btrfs_btree_balance_dirty(info->tree_root);
1079 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1082 if (btrfs_defrag_cancelled(root->fs_info)) {
1083 printk(KERN_DEBUG "btrfs: defrag_root cancelled\n");
1088 root->defrag_running = 0;
1093 * new snapshots need to be created at a very specific time in the
1094 * transaction commit. This does the actual creation.
1097 * If the error which may affect the commitment of the current transaction
1098 * happens, we should return the error number. If the error which just affect
1099 * the creation of the pending snapshots, just return 0.
1101 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1102 struct btrfs_fs_info *fs_info,
1103 struct btrfs_pending_snapshot *pending)
1105 struct btrfs_key key;
1106 struct btrfs_root_item *new_root_item;
1107 struct btrfs_root *tree_root = fs_info->tree_root;
1108 struct btrfs_root *root = pending->root;
1109 struct btrfs_root *parent_root;
1110 struct btrfs_block_rsv *rsv;
1111 struct inode *parent_inode;
1112 struct btrfs_path *path;
1113 struct btrfs_dir_item *dir_item;
1114 struct dentry *dentry;
1115 struct extent_buffer *tmp;
1116 struct extent_buffer *old;
1117 struct timespec cur_time = CURRENT_TIME;
1125 path = btrfs_alloc_path();
1127 pending->error = -ENOMEM;
1131 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1132 if (!new_root_item) {
1133 pending->error = -ENOMEM;
1134 goto root_item_alloc_fail;
1137 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1139 goto no_free_objectid;
1141 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1143 if (to_reserve > 0) {
1144 pending->error = btrfs_block_rsv_add(root,
1145 &pending->block_rsv,
1147 BTRFS_RESERVE_NO_FLUSH);
1149 goto no_free_objectid;
1152 pending->error = btrfs_qgroup_inherit(trans, fs_info,
1153 root->root_key.objectid,
1154 objectid, pending->inherit);
1156 goto no_free_objectid;
1158 key.objectid = objectid;
1159 key.offset = (u64)-1;
1160 key.type = BTRFS_ROOT_ITEM_KEY;
1162 rsv = trans->block_rsv;
1163 trans->block_rsv = &pending->block_rsv;
1164 trans->bytes_reserved = trans->block_rsv->reserved;
1166 dentry = pending->dentry;
1167 parent_inode = pending->dir;
1168 parent_root = BTRFS_I(parent_inode)->root;
1169 record_root_in_trans(trans, parent_root);
1172 * insert the directory item
1174 ret = btrfs_set_inode_index(parent_inode, &index);
1175 BUG_ON(ret); /* -ENOMEM */
1177 /* check if there is a file/dir which has the same name. */
1178 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1179 btrfs_ino(parent_inode),
1180 dentry->d_name.name,
1181 dentry->d_name.len, 0);
1182 if (dir_item != NULL && !IS_ERR(dir_item)) {
1183 pending->error = -EEXIST;
1184 goto dir_item_existed;
1185 } else if (IS_ERR(dir_item)) {
1186 ret = PTR_ERR(dir_item);
1187 btrfs_abort_transaction(trans, root, ret);
1190 btrfs_release_path(path);
1193 * pull in the delayed directory update
1194 * and the delayed inode item
1195 * otherwise we corrupt the FS during
1198 ret = btrfs_run_delayed_items(trans, root);
1199 if (ret) { /* Transaction aborted */
1200 btrfs_abort_transaction(trans, root, ret);
1204 record_root_in_trans(trans, root);
1205 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1206 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1207 btrfs_check_and_init_root_item(new_root_item);
1209 root_flags = btrfs_root_flags(new_root_item);
1210 if (pending->readonly)
1211 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1213 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1214 btrfs_set_root_flags(new_root_item, root_flags);
1216 btrfs_set_root_generation_v2(new_root_item,
1218 uuid_le_gen(&new_uuid);
1219 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1220 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1222 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1223 memset(new_root_item->received_uuid, 0,
1224 sizeof(new_root_item->received_uuid));
1225 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1226 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1227 btrfs_set_root_stransid(new_root_item, 0);
1228 btrfs_set_root_rtransid(new_root_item, 0);
1230 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1231 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1232 btrfs_set_root_otransid(new_root_item, trans->transid);
1234 old = btrfs_lock_root_node(root);
1235 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1237 btrfs_tree_unlock(old);
1238 free_extent_buffer(old);
1239 btrfs_abort_transaction(trans, root, ret);
1243 btrfs_set_lock_blocking(old);
1245 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1246 /* clean up in any case */
1247 btrfs_tree_unlock(old);
1248 free_extent_buffer(old);
1250 btrfs_abort_transaction(trans, root, ret);
1254 /* see comments in should_cow_block() */
1255 root->force_cow = 1;
1258 btrfs_set_root_node(new_root_item, tmp);
1259 /* record when the snapshot was created in key.offset */
1260 key.offset = trans->transid;
1261 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1262 btrfs_tree_unlock(tmp);
1263 free_extent_buffer(tmp);
1265 btrfs_abort_transaction(trans, root, ret);
1270 * insert root back/forward references
1272 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1273 parent_root->root_key.objectid,
1274 btrfs_ino(parent_inode), index,
1275 dentry->d_name.name, dentry->d_name.len);
1277 btrfs_abort_transaction(trans, root, ret);
1281 key.offset = (u64)-1;
1282 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1283 if (IS_ERR(pending->snap)) {
1284 ret = PTR_ERR(pending->snap);
1285 btrfs_abort_transaction(trans, root, ret);
1289 ret = btrfs_reloc_post_snapshot(trans, pending);
1291 btrfs_abort_transaction(trans, root, ret);
1295 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1297 btrfs_abort_transaction(trans, root, ret);
1301 ret = btrfs_insert_dir_item(trans, parent_root,
1302 dentry->d_name.name, dentry->d_name.len,
1304 BTRFS_FT_DIR, index);
1305 /* We have check then name at the beginning, so it is impossible. */
1306 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1308 btrfs_abort_transaction(trans, root, ret);
1312 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1313 dentry->d_name.len * 2);
1314 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1315 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1317 btrfs_abort_transaction(trans, root, ret);
1320 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1321 BTRFS_UUID_KEY_SUBVOL, objectid);
1323 btrfs_abort_transaction(trans, root, ret);
1326 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1327 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1328 new_root_item->received_uuid,
1329 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1331 if (ret && ret != -EEXIST) {
1332 btrfs_abort_transaction(trans, root, ret);
1337 pending->error = ret;
1339 trans->block_rsv = rsv;
1340 trans->bytes_reserved = 0;
1342 kfree(new_root_item);
1343 root_item_alloc_fail:
1344 btrfs_free_path(path);
1349 * create all the snapshots we've scheduled for creation
1351 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1352 struct btrfs_fs_info *fs_info)
1354 struct btrfs_pending_snapshot *pending, *next;
1355 struct list_head *head = &trans->transaction->pending_snapshots;
1358 list_for_each_entry_safe(pending, next, head, list) {
1359 list_del(&pending->list);
1360 ret = create_pending_snapshot(trans, fs_info, pending);
1367 static void update_super_roots(struct btrfs_root *root)
1369 struct btrfs_root_item *root_item;
1370 struct btrfs_super_block *super;
1372 super = root->fs_info->super_copy;
1374 root_item = &root->fs_info->chunk_root->root_item;
1375 super->chunk_root = root_item->bytenr;
1376 super->chunk_root_generation = root_item->generation;
1377 super->chunk_root_level = root_item->level;
1379 root_item = &root->fs_info->tree_root->root_item;
1380 super->root = root_item->bytenr;
1381 super->generation = root_item->generation;
1382 super->root_level = root_item->level;
1383 if (btrfs_test_opt(root, SPACE_CACHE))
1384 super->cache_generation = root_item->generation;
1385 if (root->fs_info->update_uuid_tree_gen)
1386 super->uuid_tree_generation = root_item->generation;
1389 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1391 struct btrfs_transaction *trans;
1394 spin_lock(&info->trans_lock);
1395 trans = info->running_transaction;
1397 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1398 spin_unlock(&info->trans_lock);
1402 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1404 struct btrfs_transaction *trans;
1407 spin_lock(&info->trans_lock);
1408 trans = info->running_transaction;
1410 ret = is_transaction_blocked(trans);
1411 spin_unlock(&info->trans_lock);
1416 * wait for the current transaction commit to start and block subsequent
1419 static void wait_current_trans_commit_start(struct btrfs_root *root,
1420 struct btrfs_transaction *trans)
1422 wait_event(root->fs_info->transaction_blocked_wait,
1423 trans->state >= TRANS_STATE_COMMIT_START ||
1428 * wait for the current transaction to start and then become unblocked.
1431 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1432 struct btrfs_transaction *trans)
1434 wait_event(root->fs_info->transaction_wait,
1435 trans->state >= TRANS_STATE_UNBLOCKED ||
1440 * commit transactions asynchronously. once btrfs_commit_transaction_async
1441 * returns, any subsequent transaction will not be allowed to join.
1443 struct btrfs_async_commit {
1444 struct btrfs_trans_handle *newtrans;
1445 struct btrfs_root *root;
1446 struct work_struct work;
1449 static void do_async_commit(struct work_struct *work)
1451 struct btrfs_async_commit *ac =
1452 container_of(work, struct btrfs_async_commit, work);
1455 * We've got freeze protection passed with the transaction.
1456 * Tell lockdep about it.
1458 if (ac->newtrans->type < TRANS_JOIN_NOLOCK)
1460 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1463 current->journal_info = ac->newtrans;
1465 btrfs_commit_transaction(ac->newtrans, ac->root);
1469 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 int wait_for_unblock)
1473 struct btrfs_async_commit *ac;
1474 struct btrfs_transaction *cur_trans;
1476 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1480 INIT_WORK(&ac->work, do_async_commit);
1482 ac->newtrans = btrfs_join_transaction(root);
1483 if (IS_ERR(ac->newtrans)) {
1484 int err = PTR_ERR(ac->newtrans);
1489 /* take transaction reference */
1490 cur_trans = trans->transaction;
1491 atomic_inc(&cur_trans->use_count);
1493 btrfs_end_transaction(trans, root);
1496 * Tell lockdep we've released the freeze rwsem, since the
1497 * async commit thread will be the one to unlock it.
1499 if (trans->type < TRANS_JOIN_NOLOCK)
1501 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1504 schedule_work(&ac->work);
1506 /* wait for transaction to start and unblock */
1507 if (wait_for_unblock)
1508 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1510 wait_current_trans_commit_start(root, cur_trans);
1512 if (current->journal_info == trans)
1513 current->journal_info = NULL;
1515 put_transaction(cur_trans);
1520 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1521 struct btrfs_root *root, int err)
1523 struct btrfs_transaction *cur_trans = trans->transaction;
1526 WARN_ON(trans->use_count > 1);
1528 btrfs_abort_transaction(trans, root, err);
1530 spin_lock(&root->fs_info->trans_lock);
1533 * If the transaction is removed from the list, it means this
1534 * transaction has been committed successfully, so it is impossible
1535 * to call the cleanup function.
1537 BUG_ON(list_empty(&cur_trans->list));
1539 list_del_init(&cur_trans->list);
1540 if (cur_trans == root->fs_info->running_transaction) {
1541 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1542 spin_unlock(&root->fs_info->trans_lock);
1543 wait_event(cur_trans->writer_wait,
1544 atomic_read(&cur_trans->num_writers) == 1);
1546 spin_lock(&root->fs_info->trans_lock);
1548 spin_unlock(&root->fs_info->trans_lock);
1550 btrfs_cleanup_one_transaction(trans->transaction, root);
1552 spin_lock(&root->fs_info->trans_lock);
1553 if (cur_trans == root->fs_info->running_transaction)
1554 root->fs_info->running_transaction = NULL;
1555 spin_unlock(&root->fs_info->trans_lock);
1557 if (trans->type & __TRANS_FREEZABLE)
1558 sb_end_intwrite(root->fs_info->sb);
1559 put_transaction(cur_trans);
1560 put_transaction(cur_trans);
1562 trace_btrfs_transaction_commit(root);
1564 btrfs_scrub_continue(root);
1566 if (current->journal_info == trans)
1567 current->journal_info = NULL;
1569 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1572 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans,
1573 struct btrfs_root *root)
1577 ret = btrfs_run_delayed_items(trans, root);
1582 * running the delayed items may have added new refs. account
1583 * them now so that they hinder processing of more delayed refs
1584 * as little as possible.
1586 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1589 * rename don't use btrfs_join_transaction, so, once we
1590 * set the transaction to blocked above, we aren't going
1591 * to get any new ordered operations. We can safely run
1592 * it here and no for sure that nothing new will be added
1595 ret = btrfs_run_ordered_operations(trans, root, 1);
1600 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1602 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1603 return btrfs_start_all_delalloc_inodes(fs_info, 1);
1607 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1609 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1610 btrfs_wait_all_ordered_extents(fs_info);
1613 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1614 struct btrfs_root *root)
1616 struct btrfs_transaction *cur_trans = trans->transaction;
1617 struct btrfs_transaction *prev_trans = NULL;
1620 ret = btrfs_run_ordered_operations(trans, root, 0);
1622 btrfs_abort_transaction(trans, root, ret);
1623 btrfs_end_transaction(trans, root);
1627 /* Stop the commit early if ->aborted is set */
1628 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1629 ret = cur_trans->aborted;
1630 btrfs_end_transaction(trans, root);
1634 /* make a pass through all the delayed refs we have so far
1635 * any runnings procs may add more while we are here
1637 ret = btrfs_run_delayed_refs(trans, root, 0);
1639 btrfs_end_transaction(trans, root);
1643 btrfs_trans_release_metadata(trans, root);
1644 trans->block_rsv = NULL;
1645 if (trans->qgroup_reserved) {
1646 btrfs_qgroup_free(root, trans->qgroup_reserved);
1647 trans->qgroup_reserved = 0;
1650 cur_trans = trans->transaction;
1653 * set the flushing flag so procs in this transaction have to
1654 * start sending their work down.
1656 cur_trans->delayed_refs.flushing = 1;
1659 if (!list_empty(&trans->new_bgs))
1660 btrfs_create_pending_block_groups(trans, root);
1662 ret = btrfs_run_delayed_refs(trans, root, 0);
1664 btrfs_end_transaction(trans, root);
1668 spin_lock(&root->fs_info->trans_lock);
1669 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1670 spin_unlock(&root->fs_info->trans_lock);
1671 atomic_inc(&cur_trans->use_count);
1672 ret = btrfs_end_transaction(trans, root);
1674 wait_for_commit(root, cur_trans);
1676 put_transaction(cur_trans);
1681 cur_trans->state = TRANS_STATE_COMMIT_START;
1682 wake_up(&root->fs_info->transaction_blocked_wait);
1684 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1685 prev_trans = list_entry(cur_trans->list.prev,
1686 struct btrfs_transaction, list);
1687 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1688 atomic_inc(&prev_trans->use_count);
1689 spin_unlock(&root->fs_info->trans_lock);
1691 wait_for_commit(root, prev_trans);
1693 put_transaction(prev_trans);
1695 spin_unlock(&root->fs_info->trans_lock);
1698 spin_unlock(&root->fs_info->trans_lock);
1701 extwriter_counter_dec(cur_trans, trans->type);
1703 ret = btrfs_start_delalloc_flush(root->fs_info);
1705 goto cleanup_transaction;
1707 ret = btrfs_flush_all_pending_stuffs(trans, root);
1709 goto cleanup_transaction;
1711 wait_event(cur_trans->writer_wait,
1712 extwriter_counter_read(cur_trans) == 0);
1714 /* some pending stuffs might be added after the previous flush. */
1715 ret = btrfs_flush_all_pending_stuffs(trans, root);
1717 goto cleanup_transaction;
1719 btrfs_wait_delalloc_flush(root->fs_info);
1721 * Ok now we need to make sure to block out any other joins while we
1722 * commit the transaction. We could have started a join before setting
1723 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1725 spin_lock(&root->fs_info->trans_lock);
1726 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1727 spin_unlock(&root->fs_info->trans_lock);
1728 wait_event(cur_trans->writer_wait,
1729 atomic_read(&cur_trans->num_writers) == 1);
1731 /* ->aborted might be set after the previous check, so check it */
1732 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1733 ret = cur_trans->aborted;
1734 goto cleanup_transaction;
1737 * the reloc mutex makes sure that we stop
1738 * the balancing code from coming in and moving
1739 * extents around in the middle of the commit
1741 mutex_lock(&root->fs_info->reloc_mutex);
1744 * We needn't worry about the delayed items because we will
1745 * deal with them in create_pending_snapshot(), which is the
1746 * core function of the snapshot creation.
1748 ret = create_pending_snapshots(trans, root->fs_info);
1750 mutex_unlock(&root->fs_info->reloc_mutex);
1751 goto cleanup_transaction;
1755 * We insert the dir indexes of the snapshots and update the inode
1756 * of the snapshots' parents after the snapshot creation, so there
1757 * are some delayed items which are not dealt with. Now deal with
1760 * We needn't worry that this operation will corrupt the snapshots,
1761 * because all the tree which are snapshoted will be forced to COW
1762 * the nodes and leaves.
1764 ret = btrfs_run_delayed_items(trans, root);
1766 mutex_unlock(&root->fs_info->reloc_mutex);
1767 goto cleanup_transaction;
1770 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1772 mutex_unlock(&root->fs_info->reloc_mutex);
1773 goto cleanup_transaction;
1777 * make sure none of the code above managed to slip in a
1780 btrfs_assert_delayed_root_empty(root);
1782 WARN_ON(cur_trans != trans->transaction);
1784 btrfs_scrub_pause(root);
1785 /* btrfs_commit_tree_roots is responsible for getting the
1786 * various roots consistent with each other. Every pointer
1787 * in the tree of tree roots has to point to the most up to date
1788 * root for every subvolume and other tree. So, we have to keep
1789 * the tree logging code from jumping in and changing any
1792 * At this point in the commit, there can't be any tree-log
1793 * writers, but a little lower down we drop the trans mutex
1794 * and let new people in. By holding the tree_log_mutex
1795 * from now until after the super is written, we avoid races
1796 * with the tree-log code.
1798 mutex_lock(&root->fs_info->tree_log_mutex);
1800 ret = commit_fs_roots(trans, root);
1802 mutex_unlock(&root->fs_info->tree_log_mutex);
1803 mutex_unlock(&root->fs_info->reloc_mutex);
1804 goto cleanup_transaction;
1807 /* commit_fs_roots gets rid of all the tree log roots, it is now
1808 * safe to free the root of tree log roots
1810 btrfs_free_log_root_tree(trans, root->fs_info);
1812 ret = commit_cowonly_roots(trans, root);
1814 mutex_unlock(&root->fs_info->tree_log_mutex);
1815 mutex_unlock(&root->fs_info->reloc_mutex);
1816 goto cleanup_transaction;
1820 * The tasks which save the space cache and inode cache may also
1821 * update ->aborted, check it.
1823 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1824 ret = cur_trans->aborted;
1825 mutex_unlock(&root->fs_info->tree_log_mutex);
1826 mutex_unlock(&root->fs_info->reloc_mutex);
1827 goto cleanup_transaction;
1830 btrfs_prepare_extent_commit(trans, root);
1832 cur_trans = root->fs_info->running_transaction;
1834 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1835 root->fs_info->tree_root->node);
1836 switch_commit_root(root->fs_info->tree_root);
1838 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1839 root->fs_info->chunk_root->node);
1840 switch_commit_root(root->fs_info->chunk_root);
1842 assert_qgroups_uptodate(trans);
1843 update_super_roots(root);
1845 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1846 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1847 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1848 sizeof(*root->fs_info->super_copy));
1850 spin_lock(&root->fs_info->trans_lock);
1851 cur_trans->state = TRANS_STATE_UNBLOCKED;
1852 root->fs_info->running_transaction = NULL;
1853 spin_unlock(&root->fs_info->trans_lock);
1854 mutex_unlock(&root->fs_info->reloc_mutex);
1856 wake_up(&root->fs_info->transaction_wait);
1858 ret = btrfs_write_and_wait_transaction(trans, root);
1860 btrfs_error(root->fs_info, ret,
1861 "Error while writing out transaction");
1862 mutex_unlock(&root->fs_info->tree_log_mutex);
1863 goto cleanup_transaction;
1866 ret = write_ctree_super(trans, root, 0);
1868 mutex_unlock(&root->fs_info->tree_log_mutex);
1869 goto cleanup_transaction;
1873 * the super is written, we can safely allow the tree-loggers
1874 * to go about their business
1876 mutex_unlock(&root->fs_info->tree_log_mutex);
1878 btrfs_finish_extent_commit(trans, root);
1880 root->fs_info->last_trans_committed = cur_trans->transid;
1882 * We needn't acquire the lock here because there is no other task
1883 * which can change it.
1885 cur_trans->state = TRANS_STATE_COMPLETED;
1886 wake_up(&cur_trans->commit_wait);
1888 spin_lock(&root->fs_info->trans_lock);
1889 list_del_init(&cur_trans->list);
1890 spin_unlock(&root->fs_info->trans_lock);
1892 put_transaction(cur_trans);
1893 put_transaction(cur_trans);
1895 if (trans->type & __TRANS_FREEZABLE)
1896 sb_end_intwrite(root->fs_info->sb);
1898 trace_btrfs_transaction_commit(root);
1900 btrfs_scrub_continue(root);
1902 if (current->journal_info == trans)
1903 current->journal_info = NULL;
1905 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1907 if (current != root->fs_info->transaction_kthread)
1908 btrfs_run_delayed_iputs(root);
1912 cleanup_transaction:
1913 btrfs_trans_release_metadata(trans, root);
1914 trans->block_rsv = NULL;
1915 if (trans->qgroup_reserved) {
1916 btrfs_qgroup_free(root, trans->qgroup_reserved);
1917 trans->qgroup_reserved = 0;
1919 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
1920 if (current->journal_info == trans)
1921 current->journal_info = NULL;
1922 cleanup_transaction(trans, root, ret);
1928 * return < 0 if error
1929 * 0 if there are no more dead_roots at the time of call
1930 * 1 there are more to be processed, call me again
1932 * The return value indicates there are certainly more snapshots to delete, but
1933 * if there comes a new one during processing, it may return 0. We don't mind,
1934 * because btrfs_commit_super will poke cleaner thread and it will process it a
1935 * few seconds later.
1937 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
1940 struct btrfs_fs_info *fs_info = root->fs_info;
1942 spin_lock(&fs_info->trans_lock);
1943 if (list_empty(&fs_info->dead_roots)) {
1944 spin_unlock(&fs_info->trans_lock);
1947 root = list_first_entry(&fs_info->dead_roots,
1948 struct btrfs_root, root_list);
1949 list_del_init(&root->root_list);
1950 spin_unlock(&fs_info->trans_lock);
1952 pr_debug("btrfs: cleaner removing %llu\n", root->objectid);
1954 btrfs_kill_all_delayed_nodes(root);
1956 if (btrfs_header_backref_rev(root->node) <
1957 BTRFS_MIXED_BACKREF_REV)
1958 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1960 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
1962 * If we encounter a transaction abort during snapshot cleaning, we
1963 * don't want to crash here
1965 return (ret < 0) ? 0 : 1;