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"
36 #define BTRFS_ROOT_TRANS_TAG 0
38 static unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
39 [TRANS_STATE_RUNNING] = 0U,
40 [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE |
42 [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE |
45 [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE |
49 [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE |
54 [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE |
61 void btrfs_put_transaction(struct btrfs_transaction *transaction)
63 WARN_ON(atomic_read(&transaction->use_count) == 0);
64 if (atomic_dec_and_test(&transaction->use_count)) {
65 BUG_ON(!list_empty(&transaction->list));
66 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
67 while (!list_empty(&transaction->pending_chunks)) {
68 struct extent_map *em;
70 em = list_first_entry(&transaction->pending_chunks,
71 struct extent_map, list);
72 list_del_init(&em->list);
75 kmem_cache_free(btrfs_transaction_cachep, transaction);
79 static void clear_btree_io_tree(struct extent_io_tree *tree)
81 spin_lock(&tree->lock);
82 while (!RB_EMPTY_ROOT(&tree->state)) {
84 struct extent_state *state;
86 node = rb_first(&tree->state);
87 state = rb_entry(node, struct extent_state, rb_node);
88 rb_erase(&state->rb_node, &tree->state);
89 RB_CLEAR_NODE(&state->rb_node);
91 * btree io trees aren't supposed to have tasks waiting for
92 * changes in the flags of extent states ever.
94 ASSERT(!waitqueue_active(&state->wq));
95 free_extent_state(state);
97 spin_unlock(&tree->lock);
99 spin_lock(&tree->lock);
102 spin_unlock(&tree->lock);
105 static noinline void switch_commit_roots(struct btrfs_transaction *trans,
106 struct btrfs_fs_info *fs_info)
108 struct btrfs_root *root, *tmp;
110 down_write(&fs_info->commit_root_sem);
111 list_for_each_entry_safe(root, tmp, &trans->switch_commits,
113 list_del_init(&root->dirty_list);
114 free_extent_buffer(root->commit_root);
115 root->commit_root = btrfs_root_node(root);
116 if (is_fstree(root->objectid))
117 btrfs_unpin_free_ino(root);
118 clear_btree_io_tree(&root->dirty_log_pages);
120 up_write(&fs_info->commit_root_sem);
123 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
126 if (type & TRANS_EXTWRITERS)
127 atomic_inc(&trans->num_extwriters);
130 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
133 if (type & TRANS_EXTWRITERS)
134 atomic_dec(&trans->num_extwriters);
137 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
140 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
143 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
145 return atomic_read(&trans->num_extwriters);
149 * either allocate a new transaction or hop into the existing one
151 static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
153 struct btrfs_transaction *cur_trans;
154 struct btrfs_fs_info *fs_info = root->fs_info;
156 spin_lock(&fs_info->trans_lock);
158 /* The file system has been taken offline. No new transactions. */
159 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
160 spin_unlock(&fs_info->trans_lock);
164 cur_trans = fs_info->running_transaction;
166 if (cur_trans->aborted) {
167 spin_unlock(&fs_info->trans_lock);
168 return cur_trans->aborted;
170 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
171 spin_unlock(&fs_info->trans_lock);
174 atomic_inc(&cur_trans->use_count);
175 atomic_inc(&cur_trans->num_writers);
176 extwriter_counter_inc(cur_trans, type);
177 spin_unlock(&fs_info->trans_lock);
180 spin_unlock(&fs_info->trans_lock);
183 * If we are ATTACH, we just want to catch the current transaction,
184 * and commit it. If there is no transaction, just return ENOENT.
186 if (type == TRANS_ATTACH)
190 * JOIN_NOLOCK only happens during the transaction commit, so
191 * it is impossible that ->running_transaction is NULL
193 BUG_ON(type == TRANS_JOIN_NOLOCK);
195 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
199 spin_lock(&fs_info->trans_lock);
200 if (fs_info->running_transaction) {
202 * someone started a transaction after we unlocked. Make sure
203 * to redo the checks above
205 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
207 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
208 spin_unlock(&fs_info->trans_lock);
209 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
213 atomic_set(&cur_trans->num_writers, 1);
214 extwriter_counter_init(cur_trans, type);
215 init_waitqueue_head(&cur_trans->writer_wait);
216 init_waitqueue_head(&cur_trans->commit_wait);
217 cur_trans->state = TRANS_STATE_RUNNING;
219 * One for this trans handle, one so it will live on until we
220 * commit the transaction.
222 atomic_set(&cur_trans->use_count, 2);
223 cur_trans->start_time = get_seconds();
225 cur_trans->delayed_refs.href_root = RB_ROOT;
226 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
227 cur_trans->delayed_refs.num_heads_ready = 0;
228 cur_trans->delayed_refs.num_heads = 0;
229 cur_trans->delayed_refs.flushing = 0;
230 cur_trans->delayed_refs.run_delayed_start = 0;
233 * although the tree mod log is per file system and not per transaction,
234 * the log must never go across transaction boundaries.
237 if (!list_empty(&fs_info->tree_mod_seq_list))
238 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
239 "creating a fresh transaction\n");
240 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
241 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
242 "creating a fresh transaction\n");
243 atomic64_set(&fs_info->tree_mod_seq, 0);
245 spin_lock_init(&cur_trans->delayed_refs.lock);
247 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
248 INIT_LIST_HEAD(&cur_trans->pending_chunks);
249 INIT_LIST_HEAD(&cur_trans->switch_commits);
250 INIT_LIST_HEAD(&cur_trans->pending_ordered);
251 list_add_tail(&cur_trans->list, &fs_info->trans_list);
252 extent_io_tree_init(&cur_trans->dirty_pages,
253 fs_info->btree_inode->i_mapping);
254 fs_info->generation++;
255 cur_trans->transid = fs_info->generation;
256 fs_info->running_transaction = cur_trans;
257 cur_trans->aborted = 0;
258 spin_unlock(&fs_info->trans_lock);
264 * this does all the record keeping required to make sure that a reference
265 * counted root is properly recorded in a given transaction. This is required
266 * to make sure the old root from before we joined the transaction is deleted
267 * when the transaction commits
269 static int record_root_in_trans(struct btrfs_trans_handle *trans,
270 struct btrfs_root *root)
272 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
273 root->last_trans < trans->transid) {
274 WARN_ON(root == root->fs_info->extent_root);
275 WARN_ON(root->commit_root != root->node);
278 * see below for IN_TRANS_SETUP usage rules
279 * we have the reloc mutex held now, so there
280 * is only one writer in this function
282 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
284 /* make sure readers find IN_TRANS_SETUP before
285 * they find our root->last_trans update
289 spin_lock(&root->fs_info->fs_roots_radix_lock);
290 if (root->last_trans == trans->transid) {
291 spin_unlock(&root->fs_info->fs_roots_radix_lock);
294 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
295 (unsigned long)root->root_key.objectid,
296 BTRFS_ROOT_TRANS_TAG);
297 spin_unlock(&root->fs_info->fs_roots_radix_lock);
298 root->last_trans = trans->transid;
300 /* this is pretty tricky. We don't want to
301 * take the relocation lock in btrfs_record_root_in_trans
302 * unless we're really doing the first setup for this root in
305 * Normally we'd use root->last_trans as a flag to decide
306 * if we want to take the expensive mutex.
308 * But, we have to set root->last_trans before we
309 * init the relocation root, otherwise, we trip over warnings
310 * in ctree.c. The solution used here is to flag ourselves
311 * with root IN_TRANS_SETUP. When this is 1, we're still
312 * fixing up the reloc trees and everyone must wait.
314 * When this is zero, they can trust root->last_trans and fly
315 * through btrfs_record_root_in_trans without having to take the
316 * lock. smp_wmb() makes sure that all the writes above are
317 * done before we pop in the zero below
319 btrfs_init_reloc_root(trans, root);
320 smp_mb__before_atomic();
321 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
327 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
328 struct btrfs_root *root)
330 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
334 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
338 if (root->last_trans == trans->transid &&
339 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
342 mutex_lock(&root->fs_info->reloc_mutex);
343 record_root_in_trans(trans, root);
344 mutex_unlock(&root->fs_info->reloc_mutex);
349 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
351 return (trans->state >= TRANS_STATE_BLOCKED &&
352 trans->state < TRANS_STATE_UNBLOCKED &&
356 /* wait for commit against the current transaction to become unblocked
357 * when this is done, it is safe to start a new transaction, but the current
358 * transaction might not be fully on disk.
360 static void wait_current_trans(struct btrfs_root *root)
362 struct btrfs_transaction *cur_trans;
364 spin_lock(&root->fs_info->trans_lock);
365 cur_trans = root->fs_info->running_transaction;
366 if (cur_trans && is_transaction_blocked(cur_trans)) {
367 atomic_inc(&cur_trans->use_count);
368 spin_unlock(&root->fs_info->trans_lock);
370 wait_event(root->fs_info->transaction_wait,
371 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
373 btrfs_put_transaction(cur_trans);
375 spin_unlock(&root->fs_info->trans_lock);
379 static int may_wait_transaction(struct btrfs_root *root, int type)
381 if (root->fs_info->log_root_recovering)
384 if (type == TRANS_USERSPACE)
387 if (type == TRANS_START &&
388 !atomic_read(&root->fs_info->open_ioctl_trans))
394 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
396 if (!root->fs_info->reloc_ctl ||
397 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
398 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
405 static struct btrfs_trans_handle *
406 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
407 enum btrfs_reserve_flush_enum flush)
409 struct btrfs_trans_handle *h;
410 struct btrfs_transaction *cur_trans;
412 u64 qgroup_reserved = 0;
413 bool reloc_reserved = false;
416 /* Send isn't supposed to start transactions. */
417 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
419 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
420 return ERR_PTR(-EROFS);
422 if (current->journal_info) {
423 WARN_ON(type & TRANS_EXTWRITERS);
424 h = current->journal_info;
426 WARN_ON(h->use_count > 2);
427 h->orig_rsv = h->block_rsv;
433 * Do the reservation before we join the transaction so we can do all
434 * the appropriate flushing if need be.
436 if (num_items > 0 && root != root->fs_info->chunk_root) {
437 if (root->fs_info->quota_enabled &&
438 is_fstree(root->root_key.objectid)) {
439 qgroup_reserved = num_items * root->nodesize;
440 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
445 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
447 * Do the reservation for the relocation root creation
449 if (need_reserve_reloc_root(root)) {
450 num_bytes += root->nodesize;
451 reloc_reserved = true;
454 ret = btrfs_block_rsv_add(root,
455 &root->fs_info->trans_block_rsv,
461 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
468 * If we are JOIN_NOLOCK we're already committing a transaction and
469 * waiting on this guy, so we don't need to do the sb_start_intwrite
470 * because we're already holding a ref. We need this because we could
471 * have raced in and did an fsync() on a file which can kick a commit
472 * and then we deadlock with somebody doing a freeze.
474 * If we are ATTACH, it means we just want to catch the current
475 * transaction and commit it, so we needn't do sb_start_intwrite().
477 if (type & __TRANS_FREEZABLE)
478 sb_start_intwrite(root->fs_info->sb);
480 if (may_wait_transaction(root, type))
481 wait_current_trans(root);
484 ret = join_transaction(root, type);
486 wait_current_trans(root);
487 if (unlikely(type == TRANS_ATTACH))
490 } while (ret == -EBUSY);
493 /* We must get the transaction if we are JOIN_NOLOCK. */
494 BUG_ON(type == TRANS_JOIN_NOLOCK);
498 cur_trans = root->fs_info->running_transaction;
500 h->transid = cur_trans->transid;
501 h->transaction = cur_trans;
503 h->bytes_reserved = 0;
505 h->delayed_ref_updates = 0;
511 h->qgroup_reserved = 0;
512 h->delayed_ref_elem.seq = 0;
514 h->allocating_chunk = false;
515 h->reloc_reserved = false;
517 INIT_LIST_HEAD(&h->qgroup_ref_list);
518 INIT_LIST_HEAD(&h->new_bgs);
519 INIT_LIST_HEAD(&h->ordered);
522 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
523 may_wait_transaction(root, type)) {
524 current->journal_info = h;
525 btrfs_commit_transaction(h, root);
530 trace_btrfs_space_reservation(root->fs_info, "transaction",
531 h->transid, num_bytes, 1);
532 h->block_rsv = &root->fs_info->trans_block_rsv;
533 h->bytes_reserved = num_bytes;
534 h->reloc_reserved = reloc_reserved;
536 h->qgroup_reserved = qgroup_reserved;
539 btrfs_record_root_in_trans(h, root);
541 if (!current->journal_info && type != TRANS_USERSPACE)
542 current->journal_info = h;
546 if (type & __TRANS_FREEZABLE)
547 sb_end_intwrite(root->fs_info->sb);
548 kmem_cache_free(btrfs_trans_handle_cachep, h);
551 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
555 btrfs_qgroup_free(root, qgroup_reserved);
559 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
562 return start_transaction(root, num_items, TRANS_START,
563 BTRFS_RESERVE_FLUSH_ALL);
566 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
567 struct btrfs_root *root, int num_items)
569 return start_transaction(root, num_items, TRANS_START,
570 BTRFS_RESERVE_FLUSH_LIMIT);
573 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
575 return start_transaction(root, 0, TRANS_JOIN, 0);
578 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
580 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
583 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
585 return start_transaction(root, 0, TRANS_USERSPACE, 0);
589 * btrfs_attach_transaction() - catch the running transaction
591 * It is used when we want to commit the current the transaction, but
592 * don't want to start a new one.
594 * Note: If this function return -ENOENT, it just means there is no
595 * running transaction. But it is possible that the inactive transaction
596 * is still in the memory, not fully on disk. If you hope there is no
597 * inactive transaction in the fs when -ENOENT is returned, you should
599 * btrfs_attach_transaction_barrier()
601 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
603 return start_transaction(root, 0, TRANS_ATTACH, 0);
607 * btrfs_attach_transaction_barrier() - catch the running transaction
609 * It is similar to the above function, the differentia is this one
610 * will wait for all the inactive transactions until they fully
613 struct btrfs_trans_handle *
614 btrfs_attach_transaction_barrier(struct btrfs_root *root)
616 struct btrfs_trans_handle *trans;
618 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
619 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
620 btrfs_wait_for_commit(root, 0);
625 /* wait for a transaction commit to be fully complete */
626 static noinline void wait_for_commit(struct btrfs_root *root,
627 struct btrfs_transaction *commit)
629 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
632 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
634 struct btrfs_transaction *cur_trans = NULL, *t;
638 if (transid <= root->fs_info->last_trans_committed)
641 /* find specified transaction */
642 spin_lock(&root->fs_info->trans_lock);
643 list_for_each_entry(t, &root->fs_info->trans_list, list) {
644 if (t->transid == transid) {
646 atomic_inc(&cur_trans->use_count);
650 if (t->transid > transid) {
655 spin_unlock(&root->fs_info->trans_lock);
658 * The specified transaction doesn't exist, or we
659 * raced with btrfs_commit_transaction
662 if (transid > root->fs_info->last_trans_committed)
667 /* find newest transaction that is committing | committed */
668 spin_lock(&root->fs_info->trans_lock);
669 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
671 if (t->state >= TRANS_STATE_COMMIT_START) {
672 if (t->state == TRANS_STATE_COMPLETED)
675 atomic_inc(&cur_trans->use_count);
679 spin_unlock(&root->fs_info->trans_lock);
681 goto out; /* nothing committing|committed */
684 wait_for_commit(root, cur_trans);
685 btrfs_put_transaction(cur_trans);
690 void btrfs_throttle(struct btrfs_root *root)
692 if (!atomic_read(&root->fs_info->open_ioctl_trans))
693 wait_current_trans(root);
696 static int should_end_transaction(struct btrfs_trans_handle *trans,
697 struct btrfs_root *root)
699 if (root->fs_info->global_block_rsv.space_info->full &&
700 btrfs_check_space_for_delayed_refs(trans, root))
703 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
706 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
707 struct btrfs_root *root)
709 struct btrfs_transaction *cur_trans = trans->transaction;
714 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
715 cur_trans->delayed_refs.flushing)
718 updates = trans->delayed_ref_updates;
719 trans->delayed_ref_updates = 0;
721 err = btrfs_run_delayed_refs(trans, root, updates);
722 if (err) /* Error code will also eval true */
726 return should_end_transaction(trans, root);
729 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
730 struct btrfs_root *root, int throttle)
732 struct btrfs_transaction *cur_trans = trans->transaction;
733 struct btrfs_fs_info *info = root->fs_info;
734 unsigned long cur = trans->delayed_ref_updates;
735 int lock = (trans->type != TRANS_JOIN_NOLOCK);
737 int must_run_delayed_refs = 0;
739 if (trans->use_count > 1) {
741 trans->block_rsv = trans->orig_rsv;
745 btrfs_trans_release_metadata(trans, root);
746 trans->block_rsv = NULL;
748 if (!list_empty(&trans->new_bgs))
749 btrfs_create_pending_block_groups(trans, root);
751 if (!list_empty(&trans->ordered)) {
752 spin_lock(&info->trans_lock);
753 list_splice(&trans->ordered, &cur_trans->pending_ordered);
754 spin_unlock(&info->trans_lock);
757 trans->delayed_ref_updates = 0;
759 must_run_delayed_refs =
760 btrfs_should_throttle_delayed_refs(trans, root);
761 cur = max_t(unsigned long, cur, 32);
764 * don't make the caller wait if they are from a NOLOCK
765 * or ATTACH transaction, it will deadlock with commit
767 if (must_run_delayed_refs == 1 &&
768 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
769 must_run_delayed_refs = 2;
772 if (trans->qgroup_reserved) {
774 * the same root has to be passed here between start_transaction
775 * and end_transaction. Subvolume quota depends on this.
777 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
778 trans->qgroup_reserved = 0;
781 btrfs_trans_release_metadata(trans, root);
782 trans->block_rsv = NULL;
784 if (!list_empty(&trans->new_bgs))
785 btrfs_create_pending_block_groups(trans, root);
787 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
788 should_end_transaction(trans, root) &&
789 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
790 spin_lock(&info->trans_lock);
791 if (cur_trans->state == TRANS_STATE_RUNNING)
792 cur_trans->state = TRANS_STATE_BLOCKED;
793 spin_unlock(&info->trans_lock);
796 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
798 return btrfs_commit_transaction(trans, root);
800 wake_up_process(info->transaction_kthread);
803 if (trans->type & __TRANS_FREEZABLE)
804 sb_end_intwrite(root->fs_info->sb);
806 WARN_ON(cur_trans != info->running_transaction);
807 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
808 atomic_dec(&cur_trans->num_writers);
809 extwriter_counter_dec(cur_trans, trans->type);
812 if (waitqueue_active(&cur_trans->writer_wait))
813 wake_up(&cur_trans->writer_wait);
814 btrfs_put_transaction(cur_trans);
816 if (current->journal_info == trans)
817 current->journal_info = NULL;
820 btrfs_run_delayed_iputs(root);
822 if (trans->aborted ||
823 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
824 wake_up_process(info->transaction_kthread);
827 assert_qgroups_uptodate(trans);
829 kmem_cache_free(btrfs_trans_handle_cachep, trans);
830 if (must_run_delayed_refs) {
831 btrfs_async_run_delayed_refs(root, cur,
832 must_run_delayed_refs == 1);
837 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
838 struct btrfs_root *root)
840 return __btrfs_end_transaction(trans, root, 0);
843 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
844 struct btrfs_root *root)
846 return __btrfs_end_transaction(trans, root, 1);
850 * when btree blocks are allocated, they have some corresponding bits set for
851 * them in one of two extent_io trees. This is used to make sure all of
852 * those extents are sent to disk but does not wait on them
854 int btrfs_write_marked_extents(struct btrfs_root *root,
855 struct extent_io_tree *dirty_pages, int mark)
859 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
860 struct extent_state *cached_state = NULL;
864 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
865 mark, &cached_state)) {
866 bool wait_writeback = false;
868 err = convert_extent_bit(dirty_pages, start, end,
870 mark, &cached_state, GFP_NOFS);
872 * convert_extent_bit can return -ENOMEM, which is most of the
873 * time a temporary error. So when it happens, ignore the error
874 * and wait for writeback of this range to finish - because we
875 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
876 * to btrfs_wait_marked_extents() would not know that writeback
877 * for this range started and therefore wouldn't wait for it to
878 * finish - we don't want to commit a superblock that points to
879 * btree nodes/leafs for which writeback hasn't finished yet
880 * (and without errors).
881 * We cleanup any entries left in the io tree when committing
882 * the transaction (through clear_btree_io_tree()).
884 if (err == -ENOMEM) {
886 wait_writeback = true;
889 err = filemap_fdatawrite_range(mapping, start, end);
892 else if (wait_writeback)
893 werr = filemap_fdatawait_range(mapping, start, end);
894 free_extent_state(cached_state);
903 * when btree blocks are allocated, they have some corresponding bits set for
904 * them in one of two extent_io trees. This is used to make sure all of
905 * those extents are on disk for transaction or log commit. We wait
906 * on all the pages and clear them from the dirty pages state tree
908 int btrfs_wait_marked_extents(struct btrfs_root *root,
909 struct extent_io_tree *dirty_pages, int mark)
913 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
914 struct extent_state *cached_state = NULL;
917 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
920 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
921 EXTENT_NEED_WAIT, &cached_state)) {
923 * Ignore -ENOMEM errors returned by clear_extent_bit().
924 * When committing the transaction, we'll remove any entries
925 * left in the io tree. For a log commit, we don't remove them
926 * after committing the log because the tree can be accessed
927 * concurrently - we do it only at transaction commit time when
928 * it's safe to do it (through clear_btree_io_tree()).
930 err = clear_extent_bit(dirty_pages, start, end,
932 0, 0, &cached_state, GFP_NOFS);
936 err = filemap_fdatawait_range(mapping, start, end);
939 free_extent_state(cached_state);
947 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
948 if ((mark & EXTENT_DIRTY) &&
949 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
950 &btree_ino->runtime_flags))
953 if ((mark & EXTENT_NEW) &&
954 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
955 &btree_ino->runtime_flags))
958 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
959 &btree_ino->runtime_flags))
970 * when btree blocks are allocated, they have some corresponding bits set for
971 * them in one of two extent_io trees. This is used to make sure all of
972 * those extents are on disk for transaction or log commit
974 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
975 struct extent_io_tree *dirty_pages, int mark)
979 struct blk_plug plug;
981 blk_start_plug(&plug);
982 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
983 blk_finish_plug(&plug);
984 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
993 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
994 struct btrfs_root *root)
998 ret = btrfs_write_and_wait_marked_extents(root,
999 &trans->transaction->dirty_pages,
1001 clear_btree_io_tree(&trans->transaction->dirty_pages);
1007 * this is used to update the root pointer in the tree of tree roots.
1009 * But, in the case of the extent allocation tree, updating the root
1010 * pointer may allocate blocks which may change the root of the extent
1013 * So, this loops and repeats and makes sure the cowonly root didn't
1014 * change while the root pointer was being updated in the metadata.
1016 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1017 struct btrfs_root *root)
1020 u64 old_root_bytenr;
1022 struct btrfs_root *tree_root = root->fs_info->tree_root;
1024 old_root_used = btrfs_root_used(&root->root_item);
1025 btrfs_write_dirty_block_groups(trans, root);
1028 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1029 if (old_root_bytenr == root->node->start &&
1030 old_root_used == btrfs_root_used(&root->root_item))
1033 btrfs_set_root_node(&root->root_item, root->node);
1034 ret = btrfs_update_root(trans, tree_root,
1040 old_root_used = btrfs_root_used(&root->root_item);
1041 ret = btrfs_write_dirty_block_groups(trans, root);
1050 * update all the cowonly tree roots on disk
1052 * The error handling in this function may not be obvious. Any of the
1053 * failures will cause the file system to go offline. We still need
1054 * to clean up the delayed refs.
1056 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1057 struct btrfs_root *root)
1059 struct btrfs_fs_info *fs_info = root->fs_info;
1060 struct list_head *next;
1061 struct extent_buffer *eb;
1064 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1068 eb = btrfs_lock_root_node(fs_info->tree_root);
1069 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1071 btrfs_tree_unlock(eb);
1072 free_extent_buffer(eb);
1077 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1081 ret = btrfs_run_dev_stats(trans, root->fs_info);
1084 ret = btrfs_run_dev_replace(trans, root->fs_info);
1087 ret = btrfs_run_qgroups(trans, root->fs_info);
1091 /* run_qgroups might have added some more refs */
1092 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1096 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1097 next = fs_info->dirty_cowonly_roots.next;
1098 list_del_init(next);
1099 root = list_entry(next, struct btrfs_root, dirty_list);
1101 if (root != fs_info->extent_root)
1102 list_add_tail(&root->dirty_list,
1103 &trans->transaction->switch_commits);
1104 ret = update_cowonly_root(trans, root);
1109 list_add_tail(&fs_info->extent_root->dirty_list,
1110 &trans->transaction->switch_commits);
1111 btrfs_after_dev_replace_commit(fs_info);
1117 * dead roots are old snapshots that need to be deleted. This allocates
1118 * a dirty root struct and adds it into the list of dead roots that need to
1121 void btrfs_add_dead_root(struct btrfs_root *root)
1123 spin_lock(&root->fs_info->trans_lock);
1124 if (list_empty(&root->root_list))
1125 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1126 spin_unlock(&root->fs_info->trans_lock);
1130 * update all the cowonly tree roots on disk
1132 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1133 struct btrfs_root *root)
1135 struct btrfs_root *gang[8];
1136 struct btrfs_fs_info *fs_info = root->fs_info;
1141 spin_lock(&fs_info->fs_roots_radix_lock);
1143 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1146 BTRFS_ROOT_TRANS_TAG);
1149 for (i = 0; i < ret; i++) {
1151 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1152 (unsigned long)root->root_key.objectid,
1153 BTRFS_ROOT_TRANS_TAG);
1154 spin_unlock(&fs_info->fs_roots_radix_lock);
1156 btrfs_free_log(trans, root);
1157 btrfs_update_reloc_root(trans, root);
1158 btrfs_orphan_commit_root(trans, root);
1160 btrfs_save_ino_cache(root, trans);
1162 /* see comments in should_cow_block() */
1163 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1164 smp_mb__after_atomic();
1166 if (root->commit_root != root->node) {
1167 list_add_tail(&root->dirty_list,
1168 &trans->transaction->switch_commits);
1169 btrfs_set_root_node(&root->root_item,
1173 err = btrfs_update_root(trans, fs_info->tree_root,
1176 spin_lock(&fs_info->fs_roots_radix_lock);
1181 spin_unlock(&fs_info->fs_roots_radix_lock);
1186 * defrag a given btree.
1187 * Every leaf in the btree is read and defragged.
1189 int btrfs_defrag_root(struct btrfs_root *root)
1191 struct btrfs_fs_info *info = root->fs_info;
1192 struct btrfs_trans_handle *trans;
1195 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1199 trans = btrfs_start_transaction(root, 0);
1201 return PTR_ERR(trans);
1203 ret = btrfs_defrag_leaves(trans, root);
1205 btrfs_end_transaction(trans, root);
1206 btrfs_btree_balance_dirty(info->tree_root);
1209 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1212 if (btrfs_defrag_cancelled(root->fs_info)) {
1213 pr_debug("BTRFS: defrag_root cancelled\n");
1218 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1223 * new snapshots need to be created at a very specific time in the
1224 * transaction commit. This does the actual creation.
1227 * If the error which may affect the commitment of the current transaction
1228 * happens, we should return the error number. If the error which just affect
1229 * the creation of the pending snapshots, just return 0.
1231 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1232 struct btrfs_fs_info *fs_info,
1233 struct btrfs_pending_snapshot *pending)
1235 struct btrfs_key key;
1236 struct btrfs_root_item *new_root_item;
1237 struct btrfs_root *tree_root = fs_info->tree_root;
1238 struct btrfs_root *root = pending->root;
1239 struct btrfs_root *parent_root;
1240 struct btrfs_block_rsv *rsv;
1241 struct inode *parent_inode;
1242 struct btrfs_path *path;
1243 struct btrfs_dir_item *dir_item;
1244 struct dentry *dentry;
1245 struct extent_buffer *tmp;
1246 struct extent_buffer *old;
1247 struct timespec cur_time = CURRENT_TIME;
1255 path = btrfs_alloc_path();
1257 pending->error = -ENOMEM;
1261 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1262 if (!new_root_item) {
1263 pending->error = -ENOMEM;
1264 goto root_item_alloc_fail;
1267 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1269 goto no_free_objectid;
1271 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1273 if (to_reserve > 0) {
1274 pending->error = btrfs_block_rsv_add(root,
1275 &pending->block_rsv,
1277 BTRFS_RESERVE_NO_FLUSH);
1279 goto no_free_objectid;
1282 key.objectid = objectid;
1283 key.offset = (u64)-1;
1284 key.type = BTRFS_ROOT_ITEM_KEY;
1286 rsv = trans->block_rsv;
1287 trans->block_rsv = &pending->block_rsv;
1288 trans->bytes_reserved = trans->block_rsv->reserved;
1290 dentry = pending->dentry;
1291 parent_inode = pending->dir;
1292 parent_root = BTRFS_I(parent_inode)->root;
1293 record_root_in_trans(trans, parent_root);
1296 * insert the directory item
1298 ret = btrfs_set_inode_index(parent_inode, &index);
1299 BUG_ON(ret); /* -ENOMEM */
1301 /* check if there is a file/dir which has the same name. */
1302 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1303 btrfs_ino(parent_inode),
1304 dentry->d_name.name,
1305 dentry->d_name.len, 0);
1306 if (dir_item != NULL && !IS_ERR(dir_item)) {
1307 pending->error = -EEXIST;
1308 goto dir_item_existed;
1309 } else if (IS_ERR(dir_item)) {
1310 ret = PTR_ERR(dir_item);
1311 btrfs_abort_transaction(trans, root, ret);
1314 btrfs_release_path(path);
1317 * pull in the delayed directory update
1318 * and the delayed inode item
1319 * otherwise we corrupt the FS during
1322 ret = btrfs_run_delayed_items(trans, root);
1323 if (ret) { /* Transaction aborted */
1324 btrfs_abort_transaction(trans, root, ret);
1328 record_root_in_trans(trans, root);
1329 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1330 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1331 btrfs_check_and_init_root_item(new_root_item);
1333 root_flags = btrfs_root_flags(new_root_item);
1334 if (pending->readonly)
1335 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1337 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1338 btrfs_set_root_flags(new_root_item, root_flags);
1340 btrfs_set_root_generation_v2(new_root_item,
1342 uuid_le_gen(&new_uuid);
1343 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1344 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1346 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1347 memset(new_root_item->received_uuid, 0,
1348 sizeof(new_root_item->received_uuid));
1349 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1350 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1351 btrfs_set_root_stransid(new_root_item, 0);
1352 btrfs_set_root_rtransid(new_root_item, 0);
1354 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1355 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1356 btrfs_set_root_otransid(new_root_item, trans->transid);
1358 old = btrfs_lock_root_node(root);
1359 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1361 btrfs_tree_unlock(old);
1362 free_extent_buffer(old);
1363 btrfs_abort_transaction(trans, root, ret);
1367 btrfs_set_lock_blocking(old);
1369 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1370 /* clean up in any case */
1371 btrfs_tree_unlock(old);
1372 free_extent_buffer(old);
1374 btrfs_abort_transaction(trans, root, ret);
1379 * We need to flush delayed refs in order to make sure all of our quota
1380 * operations have been done before we call btrfs_qgroup_inherit.
1382 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1384 btrfs_abort_transaction(trans, root, ret);
1388 ret = btrfs_qgroup_inherit(trans, fs_info,
1389 root->root_key.objectid,
1390 objectid, pending->inherit);
1392 btrfs_abort_transaction(trans, root, ret);
1396 /* see comments in should_cow_block() */
1397 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1400 btrfs_set_root_node(new_root_item, tmp);
1401 /* record when the snapshot was created in key.offset */
1402 key.offset = trans->transid;
1403 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1404 btrfs_tree_unlock(tmp);
1405 free_extent_buffer(tmp);
1407 btrfs_abort_transaction(trans, root, ret);
1412 * insert root back/forward references
1414 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1415 parent_root->root_key.objectid,
1416 btrfs_ino(parent_inode), index,
1417 dentry->d_name.name, dentry->d_name.len);
1419 btrfs_abort_transaction(trans, root, ret);
1423 key.offset = (u64)-1;
1424 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1425 if (IS_ERR(pending->snap)) {
1426 ret = PTR_ERR(pending->snap);
1427 btrfs_abort_transaction(trans, root, ret);
1431 ret = btrfs_reloc_post_snapshot(trans, pending);
1433 btrfs_abort_transaction(trans, root, ret);
1437 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1439 btrfs_abort_transaction(trans, root, ret);
1443 ret = btrfs_insert_dir_item(trans, parent_root,
1444 dentry->d_name.name, dentry->d_name.len,
1446 BTRFS_FT_DIR, index);
1447 /* We have check then name at the beginning, so it is impossible. */
1448 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1450 btrfs_abort_transaction(trans, root, ret);
1454 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1455 dentry->d_name.len * 2);
1456 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1457 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1459 btrfs_abort_transaction(trans, root, ret);
1462 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1463 BTRFS_UUID_KEY_SUBVOL, objectid);
1465 btrfs_abort_transaction(trans, root, ret);
1468 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1469 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1470 new_root_item->received_uuid,
1471 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1473 if (ret && ret != -EEXIST) {
1474 btrfs_abort_transaction(trans, root, ret);
1479 pending->error = ret;
1481 trans->block_rsv = rsv;
1482 trans->bytes_reserved = 0;
1484 kfree(new_root_item);
1485 root_item_alloc_fail:
1486 btrfs_free_path(path);
1491 * create all the snapshots we've scheduled for creation
1493 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1494 struct btrfs_fs_info *fs_info)
1496 struct btrfs_pending_snapshot *pending, *next;
1497 struct list_head *head = &trans->transaction->pending_snapshots;
1500 list_for_each_entry_safe(pending, next, head, list) {
1501 list_del(&pending->list);
1502 ret = create_pending_snapshot(trans, fs_info, pending);
1509 static void update_super_roots(struct btrfs_root *root)
1511 struct btrfs_root_item *root_item;
1512 struct btrfs_super_block *super;
1514 super = root->fs_info->super_copy;
1516 root_item = &root->fs_info->chunk_root->root_item;
1517 super->chunk_root = root_item->bytenr;
1518 super->chunk_root_generation = root_item->generation;
1519 super->chunk_root_level = root_item->level;
1521 root_item = &root->fs_info->tree_root->root_item;
1522 super->root = root_item->bytenr;
1523 super->generation = root_item->generation;
1524 super->root_level = root_item->level;
1525 if (btrfs_test_opt(root, SPACE_CACHE))
1526 super->cache_generation = root_item->generation;
1527 if (root->fs_info->update_uuid_tree_gen)
1528 super->uuid_tree_generation = root_item->generation;
1531 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1533 struct btrfs_transaction *trans;
1536 spin_lock(&info->trans_lock);
1537 trans = info->running_transaction;
1539 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1540 spin_unlock(&info->trans_lock);
1544 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1546 struct btrfs_transaction *trans;
1549 spin_lock(&info->trans_lock);
1550 trans = info->running_transaction;
1552 ret = is_transaction_blocked(trans);
1553 spin_unlock(&info->trans_lock);
1558 * wait for the current transaction commit to start and block subsequent
1561 static void wait_current_trans_commit_start(struct btrfs_root *root,
1562 struct btrfs_transaction *trans)
1564 wait_event(root->fs_info->transaction_blocked_wait,
1565 trans->state >= TRANS_STATE_COMMIT_START ||
1570 * wait for the current transaction to start and then become unblocked.
1573 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1574 struct btrfs_transaction *trans)
1576 wait_event(root->fs_info->transaction_wait,
1577 trans->state >= TRANS_STATE_UNBLOCKED ||
1582 * commit transactions asynchronously. once btrfs_commit_transaction_async
1583 * returns, any subsequent transaction will not be allowed to join.
1585 struct btrfs_async_commit {
1586 struct btrfs_trans_handle *newtrans;
1587 struct btrfs_root *root;
1588 struct work_struct work;
1591 static void do_async_commit(struct work_struct *work)
1593 struct btrfs_async_commit *ac =
1594 container_of(work, struct btrfs_async_commit, work);
1597 * We've got freeze protection passed with the transaction.
1598 * Tell lockdep about it.
1600 if (ac->newtrans->type & __TRANS_FREEZABLE)
1602 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1605 current->journal_info = ac->newtrans;
1607 btrfs_commit_transaction(ac->newtrans, ac->root);
1611 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1612 struct btrfs_root *root,
1613 int wait_for_unblock)
1615 struct btrfs_async_commit *ac;
1616 struct btrfs_transaction *cur_trans;
1618 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1622 INIT_WORK(&ac->work, do_async_commit);
1624 ac->newtrans = btrfs_join_transaction(root);
1625 if (IS_ERR(ac->newtrans)) {
1626 int err = PTR_ERR(ac->newtrans);
1631 /* take transaction reference */
1632 cur_trans = trans->transaction;
1633 atomic_inc(&cur_trans->use_count);
1635 btrfs_end_transaction(trans, root);
1638 * Tell lockdep we've released the freeze rwsem, since the
1639 * async commit thread will be the one to unlock it.
1641 if (ac->newtrans->type & __TRANS_FREEZABLE)
1643 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1646 schedule_work(&ac->work);
1648 /* wait for transaction to start and unblock */
1649 if (wait_for_unblock)
1650 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1652 wait_current_trans_commit_start(root, cur_trans);
1654 if (current->journal_info == trans)
1655 current->journal_info = NULL;
1657 btrfs_put_transaction(cur_trans);
1662 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1663 struct btrfs_root *root, int err)
1665 struct btrfs_transaction *cur_trans = trans->transaction;
1668 WARN_ON(trans->use_count > 1);
1670 btrfs_abort_transaction(trans, root, err);
1672 spin_lock(&root->fs_info->trans_lock);
1675 * If the transaction is removed from the list, it means this
1676 * transaction has been committed successfully, so it is impossible
1677 * to call the cleanup function.
1679 BUG_ON(list_empty(&cur_trans->list));
1681 list_del_init(&cur_trans->list);
1682 if (cur_trans == root->fs_info->running_transaction) {
1683 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1684 spin_unlock(&root->fs_info->trans_lock);
1685 wait_event(cur_trans->writer_wait,
1686 atomic_read(&cur_trans->num_writers) == 1);
1688 spin_lock(&root->fs_info->trans_lock);
1690 spin_unlock(&root->fs_info->trans_lock);
1692 btrfs_cleanup_one_transaction(trans->transaction, root);
1694 spin_lock(&root->fs_info->trans_lock);
1695 if (cur_trans == root->fs_info->running_transaction)
1696 root->fs_info->running_transaction = NULL;
1697 spin_unlock(&root->fs_info->trans_lock);
1699 if (trans->type & __TRANS_FREEZABLE)
1700 sb_end_intwrite(root->fs_info->sb);
1701 btrfs_put_transaction(cur_trans);
1702 btrfs_put_transaction(cur_trans);
1704 trace_btrfs_transaction_commit(root);
1706 if (current->journal_info == trans)
1707 current->journal_info = NULL;
1708 btrfs_scrub_cancel(root->fs_info);
1710 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1713 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1715 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1716 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1720 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1722 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1723 btrfs_wait_ordered_roots(fs_info, -1);
1727 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans,
1728 struct btrfs_fs_info *fs_info)
1730 struct btrfs_ordered_extent *ordered;
1732 spin_lock(&fs_info->trans_lock);
1733 while (!list_empty(&cur_trans->pending_ordered)) {
1734 ordered = list_first_entry(&cur_trans->pending_ordered,
1735 struct btrfs_ordered_extent,
1737 list_del_init(&ordered->trans_list);
1738 spin_unlock(&fs_info->trans_lock);
1740 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE,
1742 btrfs_put_ordered_extent(ordered);
1743 spin_lock(&fs_info->trans_lock);
1745 spin_unlock(&fs_info->trans_lock);
1748 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1749 struct btrfs_root *root)
1751 struct btrfs_transaction *cur_trans = trans->transaction;
1752 struct btrfs_transaction *prev_trans = NULL;
1753 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
1756 /* Stop the commit early if ->aborted is set */
1757 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1758 ret = cur_trans->aborted;
1759 btrfs_end_transaction(trans, root);
1763 /* make a pass through all the delayed refs we have so far
1764 * any runnings procs may add more while we are here
1766 ret = btrfs_run_delayed_refs(trans, root, 0);
1768 btrfs_end_transaction(trans, root);
1772 btrfs_trans_release_metadata(trans, root);
1773 trans->block_rsv = NULL;
1774 if (trans->qgroup_reserved) {
1775 btrfs_qgroup_free(root, trans->qgroup_reserved);
1776 trans->qgroup_reserved = 0;
1779 cur_trans = trans->transaction;
1782 * set the flushing flag so procs in this transaction have to
1783 * start sending their work down.
1785 cur_trans->delayed_refs.flushing = 1;
1788 if (!list_empty(&trans->new_bgs))
1789 btrfs_create_pending_block_groups(trans, root);
1791 ret = btrfs_run_delayed_refs(trans, root, 0);
1793 btrfs_end_transaction(trans, root);
1797 spin_lock(&root->fs_info->trans_lock);
1798 list_splice(&trans->ordered, &cur_trans->pending_ordered);
1799 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1800 spin_unlock(&root->fs_info->trans_lock);
1801 atomic_inc(&cur_trans->use_count);
1802 ret = btrfs_end_transaction(trans, root);
1804 wait_for_commit(root, cur_trans);
1806 btrfs_put_transaction(cur_trans);
1811 cur_trans->state = TRANS_STATE_COMMIT_START;
1812 wake_up(&root->fs_info->transaction_blocked_wait);
1814 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1815 prev_trans = list_entry(cur_trans->list.prev,
1816 struct btrfs_transaction, list);
1817 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1818 atomic_inc(&prev_trans->use_count);
1819 spin_unlock(&root->fs_info->trans_lock);
1821 wait_for_commit(root, prev_trans);
1823 btrfs_put_transaction(prev_trans);
1825 spin_unlock(&root->fs_info->trans_lock);
1828 spin_unlock(&root->fs_info->trans_lock);
1831 extwriter_counter_dec(cur_trans, trans->type);
1833 ret = btrfs_start_delalloc_flush(root->fs_info);
1835 goto cleanup_transaction;
1837 ret = btrfs_run_delayed_items(trans, root);
1839 goto cleanup_transaction;
1841 wait_event(cur_trans->writer_wait,
1842 extwriter_counter_read(cur_trans) == 0);
1844 /* some pending stuffs might be added after the previous flush. */
1845 ret = btrfs_run_delayed_items(trans, root);
1847 goto cleanup_transaction;
1849 btrfs_wait_delalloc_flush(root->fs_info);
1851 btrfs_wait_pending_ordered(cur_trans, root->fs_info);
1853 btrfs_scrub_pause(root);
1855 * Ok now we need to make sure to block out any other joins while we
1856 * commit the transaction. We could have started a join before setting
1857 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1859 spin_lock(&root->fs_info->trans_lock);
1860 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1861 spin_unlock(&root->fs_info->trans_lock);
1862 wait_event(cur_trans->writer_wait,
1863 atomic_read(&cur_trans->num_writers) == 1);
1865 /* ->aborted might be set after the previous check, so check it */
1866 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1867 ret = cur_trans->aborted;
1868 goto scrub_continue;
1871 * the reloc mutex makes sure that we stop
1872 * the balancing code from coming in and moving
1873 * extents around in the middle of the commit
1875 mutex_lock(&root->fs_info->reloc_mutex);
1878 * We needn't worry about the delayed items because we will
1879 * deal with them in create_pending_snapshot(), which is the
1880 * core function of the snapshot creation.
1882 ret = create_pending_snapshots(trans, root->fs_info);
1884 mutex_unlock(&root->fs_info->reloc_mutex);
1885 goto scrub_continue;
1889 * We insert the dir indexes of the snapshots and update the inode
1890 * of the snapshots' parents after the snapshot creation, so there
1891 * are some delayed items which are not dealt with. Now deal with
1894 * We needn't worry that this operation will corrupt the snapshots,
1895 * because all the tree which are snapshoted will be forced to COW
1896 * the nodes and leaves.
1898 ret = btrfs_run_delayed_items(trans, root);
1900 mutex_unlock(&root->fs_info->reloc_mutex);
1901 goto scrub_continue;
1904 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1906 mutex_unlock(&root->fs_info->reloc_mutex);
1907 goto scrub_continue;
1911 * make sure none of the code above managed to slip in a
1914 btrfs_assert_delayed_root_empty(root);
1916 WARN_ON(cur_trans != trans->transaction);
1918 /* btrfs_commit_tree_roots is responsible for getting the
1919 * various roots consistent with each other. Every pointer
1920 * in the tree of tree roots has to point to the most up to date
1921 * root for every subvolume and other tree. So, we have to keep
1922 * the tree logging code from jumping in and changing any
1925 * At this point in the commit, there can't be any tree-log
1926 * writers, but a little lower down we drop the trans mutex
1927 * and let new people in. By holding the tree_log_mutex
1928 * from now until after the super is written, we avoid races
1929 * with the tree-log code.
1931 mutex_lock(&root->fs_info->tree_log_mutex);
1933 ret = commit_fs_roots(trans, root);
1935 mutex_unlock(&root->fs_info->tree_log_mutex);
1936 mutex_unlock(&root->fs_info->reloc_mutex);
1937 goto scrub_continue;
1941 * Since the transaction is done, we can apply the pending changes
1942 * before the next transaction.
1944 btrfs_apply_pending_changes(root->fs_info);
1946 /* commit_fs_roots gets rid of all the tree log roots, it is now
1947 * safe to free the root of tree log roots
1949 btrfs_free_log_root_tree(trans, root->fs_info);
1951 ret = commit_cowonly_roots(trans, root);
1953 mutex_unlock(&root->fs_info->tree_log_mutex);
1954 mutex_unlock(&root->fs_info->reloc_mutex);
1955 goto scrub_continue;
1959 * The tasks which save the space cache and inode cache may also
1960 * update ->aborted, check it.
1962 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1963 ret = cur_trans->aborted;
1964 mutex_unlock(&root->fs_info->tree_log_mutex);
1965 mutex_unlock(&root->fs_info->reloc_mutex);
1966 goto scrub_continue;
1969 btrfs_prepare_extent_commit(trans, root);
1971 cur_trans = root->fs_info->running_transaction;
1973 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1974 root->fs_info->tree_root->node);
1975 list_add_tail(&root->fs_info->tree_root->dirty_list,
1976 &cur_trans->switch_commits);
1978 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1979 root->fs_info->chunk_root->node);
1980 list_add_tail(&root->fs_info->chunk_root->dirty_list,
1981 &cur_trans->switch_commits);
1983 switch_commit_roots(cur_trans, root->fs_info);
1985 assert_qgroups_uptodate(trans);
1986 update_super_roots(root);
1988 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1989 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1990 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1991 sizeof(*root->fs_info->super_copy));
1993 btrfs_update_commit_device_size(root->fs_info);
1994 btrfs_update_commit_device_bytes_used(root, cur_trans);
1996 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
1997 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
1999 spin_lock(&root->fs_info->trans_lock);
2000 cur_trans->state = TRANS_STATE_UNBLOCKED;
2001 root->fs_info->running_transaction = NULL;
2002 spin_unlock(&root->fs_info->trans_lock);
2003 mutex_unlock(&root->fs_info->reloc_mutex);
2005 wake_up(&root->fs_info->transaction_wait);
2007 ret = btrfs_write_and_wait_transaction(trans, root);
2009 btrfs_error(root->fs_info, ret,
2010 "Error while writing out transaction");
2011 mutex_unlock(&root->fs_info->tree_log_mutex);
2012 goto scrub_continue;
2015 ret = write_ctree_super(trans, root, 0);
2017 mutex_unlock(&root->fs_info->tree_log_mutex);
2018 goto scrub_continue;
2022 * the super is written, we can safely allow the tree-loggers
2023 * to go about their business
2025 mutex_unlock(&root->fs_info->tree_log_mutex);
2027 btrfs_finish_extent_commit(trans, root);
2029 root->fs_info->last_trans_committed = cur_trans->transid;
2031 * We needn't acquire the lock here because there is no other task
2032 * which can change it.
2034 cur_trans->state = TRANS_STATE_COMPLETED;
2035 wake_up(&cur_trans->commit_wait);
2037 spin_lock(&root->fs_info->trans_lock);
2038 list_del_init(&cur_trans->list);
2039 spin_unlock(&root->fs_info->trans_lock);
2041 btrfs_put_transaction(cur_trans);
2042 btrfs_put_transaction(cur_trans);
2044 if (trans->type & __TRANS_FREEZABLE)
2045 sb_end_intwrite(root->fs_info->sb);
2047 trace_btrfs_transaction_commit(root);
2049 btrfs_scrub_continue(root);
2051 if (current->journal_info == trans)
2052 current->journal_info = NULL;
2054 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2056 if (current != root->fs_info->transaction_kthread)
2057 btrfs_run_delayed_iputs(root);
2062 btrfs_scrub_continue(root);
2063 cleanup_transaction:
2064 btrfs_trans_release_metadata(trans, root);
2065 trans->block_rsv = NULL;
2066 if (trans->qgroup_reserved) {
2067 btrfs_qgroup_free(root, trans->qgroup_reserved);
2068 trans->qgroup_reserved = 0;
2070 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
2071 if (current->journal_info == trans)
2072 current->journal_info = NULL;
2073 cleanup_transaction(trans, root, ret);
2079 * return < 0 if error
2080 * 0 if there are no more dead_roots at the time of call
2081 * 1 there are more to be processed, call me again
2083 * The return value indicates there are certainly more snapshots to delete, but
2084 * if there comes a new one during processing, it may return 0. We don't mind,
2085 * because btrfs_commit_super will poke cleaner thread and it will process it a
2086 * few seconds later.
2088 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2091 struct btrfs_fs_info *fs_info = root->fs_info;
2093 spin_lock(&fs_info->trans_lock);
2094 if (list_empty(&fs_info->dead_roots)) {
2095 spin_unlock(&fs_info->trans_lock);
2098 root = list_first_entry(&fs_info->dead_roots,
2099 struct btrfs_root, root_list);
2100 list_del_init(&root->root_list);
2101 spin_unlock(&fs_info->trans_lock);
2103 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2105 btrfs_kill_all_delayed_nodes(root);
2107 if (btrfs_header_backref_rev(root->node) <
2108 BTRFS_MIXED_BACKREF_REV)
2109 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2111 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2113 return (ret < 0) ? 0 : 1;
2116 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2121 prev = xchg(&fs_info->pending_changes, 0);
2125 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2127 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2130 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2132 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2135 bit = 1 << BTRFS_PENDING_COMMIT;
2137 btrfs_debug(fs_info, "pending commit done");
2142 "unknown pending changes left 0x%lx, ignoring", prev);