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 kmem_cache_free(btrfs_transaction_cachep, transaction);
70 static noinline void switch_commit_root(struct btrfs_root *root)
72 free_extent_buffer(root->commit_root);
73 root->commit_root = btrfs_root_node(root);
76 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
79 if (type & TRANS_EXTWRITERS)
80 atomic_inc(&trans->num_extwriters);
83 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
86 if (type & TRANS_EXTWRITERS)
87 atomic_dec(&trans->num_extwriters);
90 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
93 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
96 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
98 return atomic_read(&trans->num_extwriters);
102 * either allocate a new transaction or hop into the existing one
104 static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
106 struct btrfs_transaction *cur_trans;
107 struct btrfs_fs_info *fs_info = root->fs_info;
109 spin_lock(&fs_info->trans_lock);
111 /* The file system has been taken offline. No new transactions. */
112 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
113 spin_unlock(&fs_info->trans_lock);
117 cur_trans = fs_info->running_transaction;
119 if (cur_trans->aborted) {
120 spin_unlock(&fs_info->trans_lock);
121 return cur_trans->aborted;
123 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
124 spin_unlock(&fs_info->trans_lock);
127 atomic_inc(&cur_trans->use_count);
128 atomic_inc(&cur_trans->num_writers);
129 extwriter_counter_inc(cur_trans, type);
130 spin_unlock(&fs_info->trans_lock);
133 spin_unlock(&fs_info->trans_lock);
136 * If we are ATTACH, we just want to catch the current transaction,
137 * and commit it. If there is no transaction, just return ENOENT.
139 if (type == TRANS_ATTACH)
143 * JOIN_NOLOCK only happens during the transaction commit, so
144 * it is impossible that ->running_transaction is NULL
146 BUG_ON(type == TRANS_JOIN_NOLOCK);
148 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
152 spin_lock(&fs_info->trans_lock);
153 if (fs_info->running_transaction) {
155 * someone started a transaction after we unlocked. Make sure
156 * to redo the checks above
158 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
160 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
161 spin_unlock(&fs_info->trans_lock);
162 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
166 atomic_set(&cur_trans->num_writers, 1);
167 extwriter_counter_init(cur_trans, type);
168 init_waitqueue_head(&cur_trans->writer_wait);
169 init_waitqueue_head(&cur_trans->commit_wait);
170 cur_trans->state = TRANS_STATE_RUNNING;
172 * One for this trans handle, one so it will live on until we
173 * commit the transaction.
175 atomic_set(&cur_trans->use_count, 2);
176 cur_trans->start_time = get_seconds();
178 cur_trans->delayed_refs.root = RB_ROOT;
179 cur_trans->delayed_refs.num_entries = 0;
180 cur_trans->delayed_refs.num_heads_ready = 0;
181 cur_trans->delayed_refs.num_heads = 0;
182 cur_trans->delayed_refs.flushing = 0;
183 cur_trans->delayed_refs.run_delayed_start = 0;
186 * although the tree mod log is per file system and not per transaction,
187 * the log must never go across transaction boundaries.
190 if (!list_empty(&fs_info->tree_mod_seq_list))
191 WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when "
192 "creating a fresh transaction\n");
193 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
194 WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
195 "creating a fresh transaction\n");
196 atomic64_set(&fs_info->tree_mod_seq, 0);
198 spin_lock_init(&cur_trans->delayed_refs.lock);
199 atomic_set(&cur_trans->delayed_refs.procs_running_refs, 0);
200 atomic_set(&cur_trans->delayed_refs.ref_seq, 0);
201 init_waitqueue_head(&cur_trans->delayed_refs.wait);
203 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
204 INIT_LIST_HEAD(&cur_trans->ordered_operations);
205 list_add_tail(&cur_trans->list, &fs_info->trans_list);
206 extent_io_tree_init(&cur_trans->dirty_pages,
207 fs_info->btree_inode->i_mapping);
208 fs_info->generation++;
209 cur_trans->transid = fs_info->generation;
210 fs_info->running_transaction = cur_trans;
211 cur_trans->aborted = 0;
212 spin_unlock(&fs_info->trans_lock);
218 * this does all the record keeping required to make sure that a reference
219 * counted root is properly recorded in a given transaction. This is required
220 * to make sure the old root from before we joined the transaction is deleted
221 * when the transaction commits
223 static int record_root_in_trans(struct btrfs_trans_handle *trans,
224 struct btrfs_root *root)
226 if (root->ref_cows && root->last_trans < trans->transid) {
227 WARN_ON(root == root->fs_info->extent_root);
228 WARN_ON(root->commit_root != root->node);
231 * see below for in_trans_setup usage rules
232 * we have the reloc mutex held now, so there
233 * is only one writer in this function
235 root->in_trans_setup = 1;
237 /* make sure readers find in_trans_setup before
238 * they find our root->last_trans update
242 spin_lock(&root->fs_info->fs_roots_radix_lock);
243 if (root->last_trans == trans->transid) {
244 spin_unlock(&root->fs_info->fs_roots_radix_lock);
247 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
248 (unsigned long)root->root_key.objectid,
249 BTRFS_ROOT_TRANS_TAG);
250 spin_unlock(&root->fs_info->fs_roots_radix_lock);
251 root->last_trans = trans->transid;
253 /* this is pretty tricky. We don't want to
254 * take the relocation lock in btrfs_record_root_in_trans
255 * unless we're really doing the first setup for this root in
258 * Normally we'd use root->last_trans as a flag to decide
259 * if we want to take the expensive mutex.
261 * But, we have to set root->last_trans before we
262 * init the relocation root, otherwise, we trip over warnings
263 * in ctree.c. The solution used here is to flag ourselves
264 * with root->in_trans_setup. When this is 1, we're still
265 * fixing up the reloc trees and everyone must wait.
267 * When this is zero, they can trust root->last_trans and fly
268 * through btrfs_record_root_in_trans without having to take the
269 * lock. smp_wmb() makes sure that all the writes above are
270 * done before we pop in the zero below
272 btrfs_init_reloc_root(trans, root);
274 root->in_trans_setup = 0;
280 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
281 struct btrfs_root *root)
287 * see record_root_in_trans for comments about in_trans_setup usage
291 if (root->last_trans == trans->transid &&
292 !root->in_trans_setup)
295 mutex_lock(&root->fs_info->reloc_mutex);
296 record_root_in_trans(trans, root);
297 mutex_unlock(&root->fs_info->reloc_mutex);
302 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
304 return (trans->state >= TRANS_STATE_BLOCKED &&
305 trans->state < TRANS_STATE_UNBLOCKED &&
309 /* wait for commit against the current transaction to become unblocked
310 * when this is done, it is safe to start a new transaction, but the current
311 * transaction might not be fully on disk.
313 static void wait_current_trans(struct btrfs_root *root)
315 struct btrfs_transaction *cur_trans;
317 spin_lock(&root->fs_info->trans_lock);
318 cur_trans = root->fs_info->running_transaction;
319 if (cur_trans && is_transaction_blocked(cur_trans)) {
320 atomic_inc(&cur_trans->use_count);
321 spin_unlock(&root->fs_info->trans_lock);
323 wait_event(root->fs_info->transaction_wait,
324 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
326 put_transaction(cur_trans);
328 spin_unlock(&root->fs_info->trans_lock);
332 static int may_wait_transaction(struct btrfs_root *root, int type)
334 if (root->fs_info->log_root_recovering)
337 if (type == TRANS_USERSPACE)
340 if (type == TRANS_START &&
341 !atomic_read(&root->fs_info->open_ioctl_trans))
347 static struct btrfs_trans_handle *
348 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
349 enum btrfs_reserve_flush_enum flush)
351 struct btrfs_trans_handle *h;
352 struct btrfs_transaction *cur_trans;
355 u64 qgroup_reserved = 0;
357 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
358 return ERR_PTR(-EROFS);
360 if (current->journal_info) {
361 WARN_ON(type & TRANS_EXTWRITERS);
362 h = current->journal_info;
364 WARN_ON(h->use_count > 2);
365 h->orig_rsv = h->block_rsv;
371 * Do the reservation before we join the transaction so we can do all
372 * the appropriate flushing if need be.
374 if (num_items > 0 && root != root->fs_info->chunk_root) {
375 if (root->fs_info->quota_enabled &&
376 is_fstree(root->root_key.objectid)) {
377 qgroup_reserved = num_items * root->leafsize;
378 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
383 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
384 ret = btrfs_block_rsv_add(root,
385 &root->fs_info->trans_block_rsv,
391 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
398 * If we are JOIN_NOLOCK we're already committing a transaction and
399 * waiting on this guy, so we don't need to do the sb_start_intwrite
400 * because we're already holding a ref. We need this because we could
401 * have raced in and did an fsync() on a file which can kick a commit
402 * and then we deadlock with somebody doing a freeze.
404 * If we are ATTACH, it means we just want to catch the current
405 * transaction and commit it, so we needn't do sb_start_intwrite().
407 if (type & __TRANS_FREEZABLE)
408 sb_start_intwrite(root->fs_info->sb);
410 if (may_wait_transaction(root, type))
411 wait_current_trans(root);
414 ret = join_transaction(root, type);
416 wait_current_trans(root);
417 if (unlikely(type == TRANS_ATTACH))
420 } while (ret == -EBUSY);
423 /* We must get the transaction if we are JOIN_NOLOCK. */
424 BUG_ON(type == TRANS_JOIN_NOLOCK);
428 cur_trans = root->fs_info->running_transaction;
430 h->transid = cur_trans->transid;
431 h->transaction = cur_trans;
433 h->bytes_reserved = 0;
435 h->delayed_ref_updates = 0;
441 h->qgroup_reserved = 0;
442 h->delayed_ref_elem.seq = 0;
444 h->allocating_chunk = false;
445 INIT_LIST_HEAD(&h->qgroup_ref_list);
446 INIT_LIST_HEAD(&h->new_bgs);
449 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
450 may_wait_transaction(root, type)) {
451 btrfs_commit_transaction(h, root);
456 trace_btrfs_space_reservation(root->fs_info, "transaction",
457 h->transid, num_bytes, 1);
458 h->block_rsv = &root->fs_info->trans_block_rsv;
459 h->bytes_reserved = num_bytes;
461 h->qgroup_reserved = qgroup_reserved;
464 btrfs_record_root_in_trans(h, root);
466 if (!current->journal_info && type != TRANS_USERSPACE)
467 current->journal_info = h;
471 if (type & __TRANS_FREEZABLE)
472 sb_end_intwrite(root->fs_info->sb);
473 kmem_cache_free(btrfs_trans_handle_cachep, h);
476 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
480 btrfs_qgroup_free(root, qgroup_reserved);
484 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
487 return start_transaction(root, num_items, TRANS_START,
488 BTRFS_RESERVE_FLUSH_ALL);
491 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
492 struct btrfs_root *root, int num_items)
494 return start_transaction(root, num_items, TRANS_START,
495 BTRFS_RESERVE_FLUSH_LIMIT);
498 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
500 return start_transaction(root, 0, TRANS_JOIN, 0);
503 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
505 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
508 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
510 return start_transaction(root, 0, TRANS_USERSPACE, 0);
514 * btrfs_attach_transaction() - catch the running transaction
516 * It is used when we want to commit the current the transaction, but
517 * don't want to start a new one.
519 * Note: If this function return -ENOENT, it just means there is no
520 * running transaction. But it is possible that the inactive transaction
521 * is still in the memory, not fully on disk. If you hope there is no
522 * inactive transaction in the fs when -ENOENT is returned, you should
524 * btrfs_attach_transaction_barrier()
526 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
528 return start_transaction(root, 0, TRANS_ATTACH, 0);
532 * btrfs_attach_transaction_barrier() - catch the running transaction
534 * It is similar to the above function, the differentia is this one
535 * will wait for all the inactive transactions until they fully
538 struct btrfs_trans_handle *
539 btrfs_attach_transaction_barrier(struct btrfs_root *root)
541 struct btrfs_trans_handle *trans;
543 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
544 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
545 btrfs_wait_for_commit(root, 0);
550 /* wait for a transaction commit to be fully complete */
551 static noinline void wait_for_commit(struct btrfs_root *root,
552 struct btrfs_transaction *commit)
554 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
557 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
559 struct btrfs_transaction *cur_trans = NULL, *t;
563 if (transid <= root->fs_info->last_trans_committed)
567 /* find specified transaction */
568 spin_lock(&root->fs_info->trans_lock);
569 list_for_each_entry(t, &root->fs_info->trans_list, list) {
570 if (t->transid == transid) {
572 atomic_inc(&cur_trans->use_count);
576 if (t->transid > transid) {
581 spin_unlock(&root->fs_info->trans_lock);
582 /* The specified transaction doesn't exist */
586 /* find newest transaction that is committing | committed */
587 spin_lock(&root->fs_info->trans_lock);
588 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
590 if (t->state >= TRANS_STATE_COMMIT_START) {
591 if (t->state == TRANS_STATE_COMPLETED)
594 atomic_inc(&cur_trans->use_count);
598 spin_unlock(&root->fs_info->trans_lock);
600 goto out; /* nothing committing|committed */
603 wait_for_commit(root, cur_trans);
604 put_transaction(cur_trans);
609 void btrfs_throttle(struct btrfs_root *root)
611 if (!atomic_read(&root->fs_info->open_ioctl_trans))
612 wait_current_trans(root);
615 static int should_end_transaction(struct btrfs_trans_handle *trans,
616 struct btrfs_root *root)
618 if (root->fs_info->global_block_rsv.space_info->full &&
619 btrfs_should_throttle_delayed_refs(trans, root))
622 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
625 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
626 struct btrfs_root *root)
628 struct btrfs_transaction *cur_trans = trans->transaction;
633 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
634 cur_trans->delayed_refs.flushing)
637 updates = trans->delayed_ref_updates;
638 trans->delayed_ref_updates = 0;
640 err = btrfs_run_delayed_refs(trans, root, updates);
641 if (err) /* Error code will also eval true */
645 return should_end_transaction(trans, root);
648 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
649 struct btrfs_root *root, int throttle)
651 struct btrfs_transaction *cur_trans = trans->transaction;
652 struct btrfs_fs_info *info = root->fs_info;
653 unsigned long cur = trans->delayed_ref_updates;
654 int lock = (trans->type != TRANS_JOIN_NOLOCK);
657 if (--trans->use_count) {
658 trans->block_rsv = trans->orig_rsv;
663 * do the qgroup accounting as early as possible
665 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
667 btrfs_trans_release_metadata(trans, root);
668 trans->block_rsv = NULL;
670 if (trans->qgroup_reserved) {
672 * the same root has to be passed here between start_transaction
673 * and end_transaction. Subvolume quota depends on this.
675 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
676 trans->qgroup_reserved = 0;
679 if (!list_empty(&trans->new_bgs))
680 btrfs_create_pending_block_groups(trans, root);
682 trans->delayed_ref_updates = 0;
683 if (btrfs_should_throttle_delayed_refs(trans, root)) {
684 cur = max_t(unsigned long, cur, 1);
685 trans->delayed_ref_updates = 0;
686 btrfs_run_delayed_refs(trans, root, cur);
689 btrfs_trans_release_metadata(trans, root);
690 trans->block_rsv = NULL;
692 if (!list_empty(&trans->new_bgs))
693 btrfs_create_pending_block_groups(trans, root);
695 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
696 should_end_transaction(trans, root) &&
697 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
698 spin_lock(&info->trans_lock);
699 if (cur_trans->state == TRANS_STATE_RUNNING)
700 cur_trans->state = TRANS_STATE_BLOCKED;
701 spin_unlock(&info->trans_lock);
704 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
707 * We may race with somebody else here so end up having
708 * to call end_transaction on ourselves again, so inc
712 return btrfs_commit_transaction(trans, root);
714 wake_up_process(info->transaction_kthread);
718 if (trans->type & __TRANS_FREEZABLE)
719 sb_end_intwrite(root->fs_info->sb);
721 WARN_ON(cur_trans != info->running_transaction);
722 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
723 atomic_dec(&cur_trans->num_writers);
724 extwriter_counter_dec(cur_trans, trans->type);
727 if (waitqueue_active(&cur_trans->writer_wait))
728 wake_up(&cur_trans->writer_wait);
729 put_transaction(cur_trans);
731 if (current->journal_info == trans)
732 current->journal_info = NULL;
735 btrfs_run_delayed_iputs(root);
737 if (trans->aborted ||
738 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
740 assert_qgroups_uptodate(trans);
742 kmem_cache_free(btrfs_trans_handle_cachep, trans);
746 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
747 struct btrfs_root *root)
749 return __btrfs_end_transaction(trans, root, 0);
752 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
753 struct btrfs_root *root)
755 return __btrfs_end_transaction(trans, root, 1);
758 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
759 struct btrfs_root *root)
761 return __btrfs_end_transaction(trans, root, 1);
765 * when btree blocks are allocated, they have some corresponding bits set for
766 * them in one of two extent_io trees. This is used to make sure all of
767 * those extents are sent to disk but does not wait on them
769 int btrfs_write_marked_extents(struct btrfs_root *root,
770 struct extent_io_tree *dirty_pages, int mark)
774 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
775 struct extent_state *cached_state = NULL;
779 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
780 mark, &cached_state)) {
781 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
782 mark, &cached_state, GFP_NOFS);
784 err = filemap_fdatawrite_range(mapping, start, end);
796 * when btree blocks are allocated, they have some corresponding bits set for
797 * them in one of two extent_io trees. This is used to make sure all of
798 * those extents are on disk for transaction or log commit. We wait
799 * on all the pages and clear them from the dirty pages state tree
801 int btrfs_wait_marked_extents(struct btrfs_root *root,
802 struct extent_io_tree *dirty_pages, int mark)
806 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
807 struct extent_state *cached_state = NULL;
811 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
812 EXTENT_NEED_WAIT, &cached_state)) {
813 clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
814 0, 0, &cached_state, GFP_NOFS);
815 err = filemap_fdatawait_range(mapping, start, end);
827 * when btree blocks are allocated, they have some corresponding bits set for
828 * them in one of two extent_io trees. This is used to make sure all of
829 * those extents are on disk for transaction or log commit
831 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
832 struct extent_io_tree *dirty_pages, int mark)
836 struct blk_plug plug;
838 blk_start_plug(&plug);
839 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
840 blk_finish_plug(&plug);
841 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
850 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
851 struct btrfs_root *root)
853 if (!trans || !trans->transaction) {
854 struct inode *btree_inode;
855 btree_inode = root->fs_info->btree_inode;
856 return filemap_write_and_wait(btree_inode->i_mapping);
858 return btrfs_write_and_wait_marked_extents(root,
859 &trans->transaction->dirty_pages,
864 * this is used to update the root pointer in the tree of tree roots.
866 * But, in the case of the extent allocation tree, updating the root
867 * pointer may allocate blocks which may change the root of the extent
870 * So, this loops and repeats and makes sure the cowonly root didn't
871 * change while the root pointer was being updated in the metadata.
873 static int update_cowonly_root(struct btrfs_trans_handle *trans,
874 struct btrfs_root *root)
879 struct btrfs_root *tree_root = root->fs_info->tree_root;
881 old_root_used = btrfs_root_used(&root->root_item);
882 btrfs_write_dirty_block_groups(trans, root);
885 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
886 if (old_root_bytenr == root->node->start &&
887 old_root_used == btrfs_root_used(&root->root_item))
890 btrfs_set_root_node(&root->root_item, root->node);
891 ret = btrfs_update_root(trans, tree_root,
897 old_root_used = btrfs_root_used(&root->root_item);
898 ret = btrfs_write_dirty_block_groups(trans, root);
903 if (root != root->fs_info->extent_root)
904 switch_commit_root(root);
910 * update all the cowonly tree roots on disk
912 * The error handling in this function may not be obvious. Any of the
913 * failures will cause the file system to go offline. We still need
914 * to clean up the delayed refs.
916 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
917 struct btrfs_root *root)
919 struct btrfs_fs_info *fs_info = root->fs_info;
920 struct list_head *next;
921 struct extent_buffer *eb;
924 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
928 eb = btrfs_lock_root_node(fs_info->tree_root);
929 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
931 btrfs_tree_unlock(eb);
932 free_extent_buffer(eb);
937 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
941 ret = btrfs_run_dev_stats(trans, root->fs_info);
943 ret = btrfs_run_dev_replace(trans, root->fs_info);
946 ret = btrfs_run_qgroups(trans, root->fs_info);
949 /* run_qgroups might have added some more refs */
950 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
953 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
954 next = fs_info->dirty_cowonly_roots.next;
956 root = list_entry(next, struct btrfs_root, dirty_list);
958 ret = update_cowonly_root(trans, root);
963 down_write(&fs_info->extent_commit_sem);
964 switch_commit_root(fs_info->extent_root);
965 up_write(&fs_info->extent_commit_sem);
967 btrfs_after_dev_replace_commit(fs_info);
973 * dead roots are old snapshots that need to be deleted. This allocates
974 * a dirty root struct and adds it into the list of dead roots that need to
977 int btrfs_add_dead_root(struct btrfs_root *root)
979 spin_lock(&root->fs_info->trans_lock);
980 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
981 spin_unlock(&root->fs_info->trans_lock);
986 * update all the cowonly tree roots on disk
988 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
989 struct btrfs_root *root)
991 struct btrfs_root *gang[8];
992 struct btrfs_fs_info *fs_info = root->fs_info;
997 spin_lock(&fs_info->fs_roots_radix_lock);
999 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1002 BTRFS_ROOT_TRANS_TAG);
1005 for (i = 0; i < ret; i++) {
1007 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1008 (unsigned long)root->root_key.objectid,
1009 BTRFS_ROOT_TRANS_TAG);
1010 spin_unlock(&fs_info->fs_roots_radix_lock);
1012 btrfs_free_log(trans, root);
1013 btrfs_update_reloc_root(trans, root);
1014 btrfs_orphan_commit_root(trans, root);
1016 btrfs_save_ino_cache(root, trans);
1018 /* see comments in should_cow_block() */
1019 root->force_cow = 0;
1022 if (root->commit_root != root->node) {
1023 mutex_lock(&root->fs_commit_mutex);
1024 switch_commit_root(root);
1025 btrfs_unpin_free_ino(root);
1026 mutex_unlock(&root->fs_commit_mutex);
1028 btrfs_set_root_node(&root->root_item,
1032 err = btrfs_update_root(trans, fs_info->tree_root,
1035 spin_lock(&fs_info->fs_roots_radix_lock);
1040 spin_unlock(&fs_info->fs_roots_radix_lock);
1045 * defrag a given btree.
1046 * Every leaf in the btree is read and defragged.
1048 int btrfs_defrag_root(struct btrfs_root *root)
1050 struct btrfs_fs_info *info = root->fs_info;
1051 struct btrfs_trans_handle *trans;
1054 if (xchg(&root->defrag_running, 1))
1058 trans = btrfs_start_transaction(root, 0);
1060 return PTR_ERR(trans);
1062 ret = btrfs_defrag_leaves(trans, root);
1064 btrfs_end_transaction(trans, root);
1065 btrfs_btree_balance_dirty(info->tree_root);
1068 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1071 if (btrfs_defrag_cancelled(root->fs_info)) {
1072 printk(KERN_DEBUG "btrfs: defrag_root cancelled\n");
1077 root->defrag_running = 0;
1082 * new snapshots need to be created at a very specific time in the
1083 * transaction commit. This does the actual creation.
1086 * If the error which may affect the commitment of the current transaction
1087 * happens, we should return the error number. If the error which just affect
1088 * the creation of the pending snapshots, just return 0.
1090 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1091 struct btrfs_fs_info *fs_info,
1092 struct btrfs_pending_snapshot *pending)
1094 struct btrfs_key key;
1095 struct btrfs_root_item *new_root_item;
1096 struct btrfs_root *tree_root = fs_info->tree_root;
1097 struct btrfs_root *root = pending->root;
1098 struct btrfs_root *parent_root;
1099 struct btrfs_block_rsv *rsv;
1100 struct inode *parent_inode;
1101 struct btrfs_path *path;
1102 struct btrfs_dir_item *dir_item;
1103 struct dentry *dentry;
1104 struct extent_buffer *tmp;
1105 struct extent_buffer *old;
1106 struct timespec cur_time = CURRENT_TIME;
1114 path = btrfs_alloc_path();
1116 pending->error = -ENOMEM;
1120 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1121 if (!new_root_item) {
1122 pending->error = -ENOMEM;
1123 goto root_item_alloc_fail;
1126 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1128 goto no_free_objectid;
1130 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1132 if (to_reserve > 0) {
1133 pending->error = btrfs_block_rsv_add(root,
1134 &pending->block_rsv,
1136 BTRFS_RESERVE_NO_FLUSH);
1138 goto no_free_objectid;
1141 pending->error = btrfs_qgroup_inherit(trans, fs_info,
1142 root->root_key.objectid,
1143 objectid, pending->inherit);
1145 goto no_free_objectid;
1147 key.objectid = objectid;
1148 key.offset = (u64)-1;
1149 key.type = BTRFS_ROOT_ITEM_KEY;
1151 rsv = trans->block_rsv;
1152 trans->block_rsv = &pending->block_rsv;
1153 trans->bytes_reserved = trans->block_rsv->reserved;
1155 dentry = pending->dentry;
1156 parent_inode = pending->dir;
1157 parent_root = BTRFS_I(parent_inode)->root;
1158 record_root_in_trans(trans, parent_root);
1161 * insert the directory item
1163 ret = btrfs_set_inode_index(parent_inode, &index);
1164 BUG_ON(ret); /* -ENOMEM */
1166 /* check if there is a file/dir which has the same name. */
1167 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1168 btrfs_ino(parent_inode),
1169 dentry->d_name.name,
1170 dentry->d_name.len, 0);
1171 if (dir_item != NULL && !IS_ERR(dir_item)) {
1172 pending->error = -EEXIST;
1173 goto dir_item_existed;
1174 } else if (IS_ERR(dir_item)) {
1175 ret = PTR_ERR(dir_item);
1176 btrfs_abort_transaction(trans, root, ret);
1179 btrfs_release_path(path);
1182 * pull in the delayed directory update
1183 * and the delayed inode item
1184 * otherwise we corrupt the FS during
1187 ret = btrfs_run_delayed_items(trans, root);
1188 if (ret) { /* Transaction aborted */
1189 btrfs_abort_transaction(trans, root, ret);
1193 record_root_in_trans(trans, root);
1194 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1195 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1196 btrfs_check_and_init_root_item(new_root_item);
1198 root_flags = btrfs_root_flags(new_root_item);
1199 if (pending->readonly)
1200 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1202 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1203 btrfs_set_root_flags(new_root_item, root_flags);
1205 btrfs_set_root_generation_v2(new_root_item,
1207 uuid_le_gen(&new_uuid);
1208 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1209 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1211 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1212 memset(new_root_item->received_uuid, 0,
1213 sizeof(new_root_item->received_uuid));
1214 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1215 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1216 btrfs_set_root_stransid(new_root_item, 0);
1217 btrfs_set_root_rtransid(new_root_item, 0);
1219 new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1220 new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
1221 btrfs_set_root_otransid(new_root_item, trans->transid);
1223 old = btrfs_lock_root_node(root);
1224 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1226 btrfs_tree_unlock(old);
1227 free_extent_buffer(old);
1228 btrfs_abort_transaction(trans, root, ret);
1232 btrfs_set_lock_blocking(old);
1234 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1235 /* clean up in any case */
1236 btrfs_tree_unlock(old);
1237 free_extent_buffer(old);
1239 btrfs_abort_transaction(trans, root, ret);
1243 /* see comments in should_cow_block() */
1244 root->force_cow = 1;
1247 btrfs_set_root_node(new_root_item, tmp);
1248 /* record when the snapshot was created in key.offset */
1249 key.offset = trans->transid;
1250 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1251 btrfs_tree_unlock(tmp);
1252 free_extent_buffer(tmp);
1254 btrfs_abort_transaction(trans, root, ret);
1259 * insert root back/forward references
1261 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1262 parent_root->root_key.objectid,
1263 btrfs_ino(parent_inode), index,
1264 dentry->d_name.name, dentry->d_name.len);
1266 btrfs_abort_transaction(trans, root, ret);
1270 key.offset = (u64)-1;
1271 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1272 if (IS_ERR(pending->snap)) {
1273 ret = PTR_ERR(pending->snap);
1274 btrfs_abort_transaction(trans, root, ret);
1278 ret = btrfs_reloc_post_snapshot(trans, pending);
1280 btrfs_abort_transaction(trans, root, ret);
1284 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1286 btrfs_abort_transaction(trans, root, ret);
1290 ret = btrfs_insert_dir_item(trans, parent_root,
1291 dentry->d_name.name, dentry->d_name.len,
1293 BTRFS_FT_DIR, index);
1294 /* We have check then name at the beginning, so it is impossible. */
1295 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1297 btrfs_abort_transaction(trans, root, ret);
1301 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1302 dentry->d_name.len * 2);
1303 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1304 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1306 btrfs_abort_transaction(trans, root, ret);
1308 pending->error = ret;
1310 trans->block_rsv = rsv;
1311 trans->bytes_reserved = 0;
1313 kfree(new_root_item);
1314 root_item_alloc_fail:
1315 btrfs_free_path(path);
1320 * create all the snapshots we've scheduled for creation
1322 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1323 struct btrfs_fs_info *fs_info)
1325 struct btrfs_pending_snapshot *pending, *next;
1326 struct list_head *head = &trans->transaction->pending_snapshots;
1329 list_for_each_entry_safe(pending, next, head, list) {
1330 list_del(&pending->list);
1331 ret = create_pending_snapshot(trans, fs_info, pending);
1338 static void update_super_roots(struct btrfs_root *root)
1340 struct btrfs_root_item *root_item;
1341 struct btrfs_super_block *super;
1343 super = root->fs_info->super_copy;
1345 root_item = &root->fs_info->chunk_root->root_item;
1346 super->chunk_root = root_item->bytenr;
1347 super->chunk_root_generation = root_item->generation;
1348 super->chunk_root_level = root_item->level;
1350 root_item = &root->fs_info->tree_root->root_item;
1351 super->root = root_item->bytenr;
1352 super->generation = root_item->generation;
1353 super->root_level = root_item->level;
1354 if (btrfs_test_opt(root, SPACE_CACHE))
1355 super->cache_generation = root_item->generation;
1358 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1360 struct btrfs_transaction *trans;
1363 spin_lock(&info->trans_lock);
1364 trans = info->running_transaction;
1366 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1367 spin_unlock(&info->trans_lock);
1371 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1373 struct btrfs_transaction *trans;
1376 spin_lock(&info->trans_lock);
1377 trans = info->running_transaction;
1379 ret = is_transaction_blocked(trans);
1380 spin_unlock(&info->trans_lock);
1385 * wait for the current transaction commit to start and block subsequent
1388 static void wait_current_trans_commit_start(struct btrfs_root *root,
1389 struct btrfs_transaction *trans)
1391 wait_event(root->fs_info->transaction_blocked_wait,
1392 trans->state >= TRANS_STATE_COMMIT_START ||
1397 * wait for the current transaction to start and then become unblocked.
1400 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1401 struct btrfs_transaction *trans)
1403 wait_event(root->fs_info->transaction_wait,
1404 trans->state >= TRANS_STATE_UNBLOCKED ||
1409 * commit transactions asynchronously. once btrfs_commit_transaction_async
1410 * returns, any subsequent transaction will not be allowed to join.
1412 struct btrfs_async_commit {
1413 struct btrfs_trans_handle *newtrans;
1414 struct btrfs_root *root;
1415 struct work_struct work;
1418 static void do_async_commit(struct work_struct *work)
1420 struct btrfs_async_commit *ac =
1421 container_of(work, struct btrfs_async_commit, work);
1424 * We've got freeze protection passed with the transaction.
1425 * Tell lockdep about it.
1427 if (ac->newtrans->type < TRANS_JOIN_NOLOCK)
1429 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1432 current->journal_info = ac->newtrans;
1434 btrfs_commit_transaction(ac->newtrans, ac->root);
1438 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1439 struct btrfs_root *root,
1440 int wait_for_unblock)
1442 struct btrfs_async_commit *ac;
1443 struct btrfs_transaction *cur_trans;
1445 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1449 INIT_WORK(&ac->work, do_async_commit);
1451 ac->newtrans = btrfs_join_transaction(root);
1452 if (IS_ERR(ac->newtrans)) {
1453 int err = PTR_ERR(ac->newtrans);
1458 /* take transaction reference */
1459 cur_trans = trans->transaction;
1460 atomic_inc(&cur_trans->use_count);
1462 btrfs_end_transaction(trans, root);
1465 * Tell lockdep we've released the freeze rwsem, since the
1466 * async commit thread will be the one to unlock it.
1468 if (trans->type < TRANS_JOIN_NOLOCK)
1470 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1473 schedule_work(&ac->work);
1475 /* wait for transaction to start and unblock */
1476 if (wait_for_unblock)
1477 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1479 wait_current_trans_commit_start(root, cur_trans);
1481 if (current->journal_info == trans)
1482 current->journal_info = NULL;
1484 put_transaction(cur_trans);
1489 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1490 struct btrfs_root *root, int err)
1492 struct btrfs_transaction *cur_trans = trans->transaction;
1495 WARN_ON(trans->use_count > 1);
1497 btrfs_abort_transaction(trans, root, err);
1499 spin_lock(&root->fs_info->trans_lock);
1502 * If the transaction is removed from the list, it means this
1503 * transaction has been committed successfully, so it is impossible
1504 * to call the cleanup function.
1506 BUG_ON(list_empty(&cur_trans->list));
1508 list_del_init(&cur_trans->list);
1509 if (cur_trans == root->fs_info->running_transaction) {
1510 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1511 spin_unlock(&root->fs_info->trans_lock);
1512 wait_event(cur_trans->writer_wait,
1513 atomic_read(&cur_trans->num_writers) == 1);
1515 spin_lock(&root->fs_info->trans_lock);
1517 spin_unlock(&root->fs_info->trans_lock);
1519 btrfs_cleanup_one_transaction(trans->transaction, root);
1521 spin_lock(&root->fs_info->trans_lock);
1522 if (cur_trans == root->fs_info->running_transaction)
1523 root->fs_info->running_transaction = NULL;
1524 spin_unlock(&root->fs_info->trans_lock);
1526 put_transaction(cur_trans);
1527 put_transaction(cur_trans);
1529 trace_btrfs_transaction_commit(root);
1531 btrfs_scrub_continue(root);
1533 if (current->journal_info == trans)
1534 current->journal_info = NULL;
1536 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1539 static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans,
1540 struct btrfs_root *root)
1544 ret = btrfs_run_delayed_items(trans, root);
1549 * running the delayed items may have added new refs. account
1550 * them now so that they hinder processing of more delayed refs
1551 * as little as possible.
1553 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1556 * rename don't use btrfs_join_transaction, so, once we
1557 * set the transaction to blocked above, we aren't going
1558 * to get any new ordered operations. We can safely run
1559 * it here and no for sure that nothing new will be added
1562 ret = btrfs_run_ordered_operations(trans, root, 1);
1567 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1569 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1570 return btrfs_start_all_delalloc_inodes(fs_info, 1);
1574 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1576 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1577 btrfs_wait_all_ordered_extents(fs_info, 1);
1580 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1581 struct btrfs_root *root)
1583 struct btrfs_transaction *cur_trans = trans->transaction;
1584 struct btrfs_transaction *prev_trans = NULL;
1587 ret = btrfs_run_ordered_operations(trans, root, 0);
1589 btrfs_abort_transaction(trans, root, ret);
1590 btrfs_end_transaction(trans, root);
1594 /* Stop the commit early if ->aborted is set */
1595 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1596 ret = cur_trans->aborted;
1597 btrfs_end_transaction(trans, root);
1601 /* make a pass through all the delayed refs we have so far
1602 * any runnings procs may add more while we are here
1604 ret = btrfs_run_delayed_refs(trans, root, 0);
1606 btrfs_end_transaction(trans, root);
1610 btrfs_trans_release_metadata(trans, root);
1611 trans->block_rsv = NULL;
1612 if (trans->qgroup_reserved) {
1613 btrfs_qgroup_free(root, trans->qgroup_reserved);
1614 trans->qgroup_reserved = 0;
1617 cur_trans = trans->transaction;
1620 * set the flushing flag so procs in this transaction have to
1621 * start sending their work down.
1623 cur_trans->delayed_refs.flushing = 1;
1626 if (!list_empty(&trans->new_bgs))
1627 btrfs_create_pending_block_groups(trans, root);
1629 ret = btrfs_run_delayed_refs(trans, root, 0);
1631 btrfs_end_transaction(trans, root);
1635 spin_lock(&root->fs_info->trans_lock);
1636 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1637 spin_unlock(&root->fs_info->trans_lock);
1638 atomic_inc(&cur_trans->use_count);
1639 ret = btrfs_end_transaction(trans, root);
1641 wait_for_commit(root, cur_trans);
1643 put_transaction(cur_trans);
1648 cur_trans->state = TRANS_STATE_COMMIT_START;
1649 wake_up(&root->fs_info->transaction_blocked_wait);
1651 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1652 prev_trans = list_entry(cur_trans->list.prev,
1653 struct btrfs_transaction, list);
1654 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1655 atomic_inc(&prev_trans->use_count);
1656 spin_unlock(&root->fs_info->trans_lock);
1658 wait_for_commit(root, prev_trans);
1660 put_transaction(prev_trans);
1662 spin_unlock(&root->fs_info->trans_lock);
1665 spin_unlock(&root->fs_info->trans_lock);
1668 extwriter_counter_dec(cur_trans, trans->type);
1670 ret = btrfs_start_delalloc_flush(root->fs_info);
1672 goto cleanup_transaction;
1674 ret = btrfs_flush_all_pending_stuffs(trans, root);
1676 goto cleanup_transaction;
1678 wait_event(cur_trans->writer_wait,
1679 extwriter_counter_read(cur_trans) == 0);
1681 /* some pending stuffs might be added after the previous flush. */
1682 ret = btrfs_flush_all_pending_stuffs(trans, root);
1684 goto cleanup_transaction;
1686 btrfs_wait_delalloc_flush(root->fs_info);
1688 * Ok now we need to make sure to block out any other joins while we
1689 * commit the transaction. We could have started a join before setting
1690 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1692 spin_lock(&root->fs_info->trans_lock);
1693 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1694 spin_unlock(&root->fs_info->trans_lock);
1695 wait_event(cur_trans->writer_wait,
1696 atomic_read(&cur_trans->num_writers) == 1);
1698 /* ->aborted might be set after the previous check, so check it */
1699 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1700 ret = cur_trans->aborted;
1701 goto cleanup_transaction;
1704 * the reloc mutex makes sure that we stop
1705 * the balancing code from coming in and moving
1706 * extents around in the middle of the commit
1708 mutex_lock(&root->fs_info->reloc_mutex);
1711 * We needn't worry about the delayed items because we will
1712 * deal with them in create_pending_snapshot(), which is the
1713 * core function of the snapshot creation.
1715 ret = create_pending_snapshots(trans, root->fs_info);
1717 mutex_unlock(&root->fs_info->reloc_mutex);
1718 goto cleanup_transaction;
1722 * We insert the dir indexes of the snapshots and update the inode
1723 * of the snapshots' parents after the snapshot creation, so there
1724 * are some delayed items which are not dealt with. Now deal with
1727 * We needn't worry that this operation will corrupt the snapshots,
1728 * because all the tree which are snapshoted will be forced to COW
1729 * the nodes and leaves.
1731 ret = btrfs_run_delayed_items(trans, root);
1733 mutex_unlock(&root->fs_info->reloc_mutex);
1734 goto cleanup_transaction;
1737 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1739 mutex_unlock(&root->fs_info->reloc_mutex);
1740 goto cleanup_transaction;
1744 * make sure none of the code above managed to slip in a
1747 btrfs_assert_delayed_root_empty(root);
1749 WARN_ON(cur_trans != trans->transaction);
1751 btrfs_scrub_pause(root);
1752 /* btrfs_commit_tree_roots is responsible for getting the
1753 * various roots consistent with each other. Every pointer
1754 * in the tree of tree roots has to point to the most up to date
1755 * root for every subvolume and other tree. So, we have to keep
1756 * the tree logging code from jumping in and changing any
1759 * At this point in the commit, there can't be any tree-log
1760 * writers, but a little lower down we drop the trans mutex
1761 * and let new people in. By holding the tree_log_mutex
1762 * from now until after the super is written, we avoid races
1763 * with the tree-log code.
1765 mutex_lock(&root->fs_info->tree_log_mutex);
1767 ret = commit_fs_roots(trans, root);
1769 mutex_unlock(&root->fs_info->tree_log_mutex);
1770 mutex_unlock(&root->fs_info->reloc_mutex);
1771 goto cleanup_transaction;
1774 /* commit_fs_roots gets rid of all the tree log roots, it is now
1775 * safe to free the root of tree log roots
1777 btrfs_free_log_root_tree(trans, root->fs_info);
1779 ret = commit_cowonly_roots(trans, root);
1781 mutex_unlock(&root->fs_info->tree_log_mutex);
1782 mutex_unlock(&root->fs_info->reloc_mutex);
1783 goto cleanup_transaction;
1787 * The tasks which save the space cache and inode cache may also
1788 * update ->aborted, check it.
1790 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1791 ret = cur_trans->aborted;
1792 mutex_unlock(&root->fs_info->tree_log_mutex);
1793 mutex_unlock(&root->fs_info->reloc_mutex);
1794 goto cleanup_transaction;
1797 btrfs_prepare_extent_commit(trans, root);
1799 cur_trans = root->fs_info->running_transaction;
1801 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1802 root->fs_info->tree_root->node);
1803 switch_commit_root(root->fs_info->tree_root);
1805 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1806 root->fs_info->chunk_root->node);
1807 switch_commit_root(root->fs_info->chunk_root);
1809 assert_qgroups_uptodate(trans);
1810 update_super_roots(root);
1812 if (!root->fs_info->log_root_recovering) {
1813 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1814 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1817 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1818 sizeof(*root->fs_info->super_copy));
1820 spin_lock(&root->fs_info->trans_lock);
1821 cur_trans->state = TRANS_STATE_UNBLOCKED;
1822 root->fs_info->running_transaction = NULL;
1823 spin_unlock(&root->fs_info->trans_lock);
1824 mutex_unlock(&root->fs_info->reloc_mutex);
1826 wake_up(&root->fs_info->transaction_wait);
1828 ret = btrfs_write_and_wait_transaction(trans, root);
1830 btrfs_error(root->fs_info, ret,
1831 "Error while writing out transaction");
1832 mutex_unlock(&root->fs_info->tree_log_mutex);
1833 goto cleanup_transaction;
1836 ret = write_ctree_super(trans, root, 0);
1838 mutex_unlock(&root->fs_info->tree_log_mutex);
1839 goto cleanup_transaction;
1843 * the super is written, we can safely allow the tree-loggers
1844 * to go about their business
1846 mutex_unlock(&root->fs_info->tree_log_mutex);
1848 btrfs_finish_extent_commit(trans, root);
1850 root->fs_info->last_trans_committed = cur_trans->transid;
1852 * We needn't acquire the lock here because there is no other task
1853 * which can change it.
1855 cur_trans->state = TRANS_STATE_COMPLETED;
1856 wake_up(&cur_trans->commit_wait);
1858 spin_lock(&root->fs_info->trans_lock);
1859 list_del_init(&cur_trans->list);
1860 spin_unlock(&root->fs_info->trans_lock);
1862 put_transaction(cur_trans);
1863 put_transaction(cur_trans);
1865 if (trans->type & __TRANS_FREEZABLE)
1866 sb_end_intwrite(root->fs_info->sb);
1868 trace_btrfs_transaction_commit(root);
1870 btrfs_scrub_continue(root);
1872 if (current->journal_info == trans)
1873 current->journal_info = NULL;
1875 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1877 if (current != root->fs_info->transaction_kthread)
1878 btrfs_run_delayed_iputs(root);
1882 cleanup_transaction:
1883 btrfs_trans_release_metadata(trans, root);
1884 trans->block_rsv = NULL;
1885 if (trans->qgroup_reserved) {
1886 btrfs_qgroup_free(root, trans->qgroup_reserved);
1887 trans->qgroup_reserved = 0;
1889 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
1890 if (current->journal_info == trans)
1891 current->journal_info = NULL;
1892 cleanup_transaction(trans, root, ret);
1898 * return < 0 if error
1899 * 0 if there are no more dead_roots at the time of call
1900 * 1 there are more to be processed, call me again
1902 * The return value indicates there are certainly more snapshots to delete, but
1903 * if there comes a new one during processing, it may return 0. We don't mind,
1904 * because btrfs_commit_super will poke cleaner thread and it will process it a
1905 * few seconds later.
1907 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
1910 struct btrfs_fs_info *fs_info = root->fs_info;
1912 spin_lock(&fs_info->trans_lock);
1913 if (list_empty(&fs_info->dead_roots)) {
1914 spin_unlock(&fs_info->trans_lock);
1917 root = list_first_entry(&fs_info->dead_roots,
1918 struct btrfs_root, root_list);
1919 list_del(&root->root_list);
1920 spin_unlock(&fs_info->trans_lock);
1922 pr_debug("btrfs: cleaner removing %llu\n",
1923 (unsigned long long)root->objectid);
1925 btrfs_kill_all_delayed_nodes(root);
1927 if (btrfs_header_backref_rev(root->node) <
1928 BTRFS_MIXED_BACKREF_REV)
1929 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1931 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
1933 * If we encounter a transaction abort during snapshot cleaning, we
1934 * don't want to crash here
1936 BUG_ON(ret < 0 && ret != -EAGAIN && ret != -EROFS);