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 const 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 if (transaction->delayed_refs.pending_csums)
68 printk(KERN_ERR "pending csums is %llu\n",
69 transaction->delayed_refs.pending_csums);
70 while (!list_empty(&transaction->pending_chunks)) {
71 struct extent_map *em;
73 em = list_first_entry(&transaction->pending_chunks,
74 struct extent_map, list);
75 list_del_init(&em->list);
78 kmem_cache_free(btrfs_transaction_cachep, transaction);
82 static void clear_btree_io_tree(struct extent_io_tree *tree)
84 spin_lock(&tree->lock);
85 while (!RB_EMPTY_ROOT(&tree->state)) {
87 struct extent_state *state;
89 node = rb_first(&tree->state);
90 state = rb_entry(node, struct extent_state, rb_node);
91 rb_erase(&state->rb_node, &tree->state);
92 RB_CLEAR_NODE(&state->rb_node);
94 * btree io trees aren't supposed to have tasks waiting for
95 * changes in the flags of extent states ever.
97 ASSERT(!waitqueue_active(&state->wq));
98 free_extent_state(state);
100 cond_resched_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->have_free_bgs = 0;
224 cur_trans->start_time = get_seconds();
225 cur_trans->dirty_bg_run = 0;
227 cur_trans->delayed_refs.href_root = RB_ROOT;
228 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
229 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
230 cur_trans->delayed_refs.num_heads_ready = 0;
231 cur_trans->delayed_refs.pending_csums = 0;
232 cur_trans->delayed_refs.num_heads = 0;
233 cur_trans->delayed_refs.flushing = 0;
234 cur_trans->delayed_refs.run_delayed_start = 0;
235 cur_trans->delayed_refs.qgroup_to_skip = 0;
238 * although the tree mod log is per file system and not per transaction,
239 * the log must never go across transaction boundaries.
242 if (!list_empty(&fs_info->tree_mod_seq_list))
243 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
244 "creating a fresh transaction\n");
245 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
246 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
247 "creating a fresh transaction\n");
248 atomic64_set(&fs_info->tree_mod_seq, 0);
250 spin_lock_init(&cur_trans->delayed_refs.lock);
252 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
253 INIT_LIST_HEAD(&cur_trans->pending_chunks);
254 INIT_LIST_HEAD(&cur_trans->switch_commits);
255 INIT_LIST_HEAD(&cur_trans->pending_ordered);
256 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
257 INIT_LIST_HEAD(&cur_trans->io_bgs);
258 mutex_init(&cur_trans->cache_write_mutex);
259 cur_trans->num_dirty_bgs = 0;
260 spin_lock_init(&cur_trans->dirty_bgs_lock);
261 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
262 spin_lock_init(&cur_trans->deleted_bgs_lock);
263 list_add_tail(&cur_trans->list, &fs_info->trans_list);
264 extent_io_tree_init(&cur_trans->dirty_pages,
265 fs_info->btree_inode->i_mapping);
266 fs_info->generation++;
267 cur_trans->transid = fs_info->generation;
268 fs_info->running_transaction = cur_trans;
269 cur_trans->aborted = 0;
270 spin_unlock(&fs_info->trans_lock);
276 * this does all the record keeping required to make sure that a reference
277 * counted root is properly recorded in a given transaction. This is required
278 * to make sure the old root from before we joined the transaction is deleted
279 * when the transaction commits
281 static int record_root_in_trans(struct btrfs_trans_handle *trans,
282 struct btrfs_root *root)
284 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
285 root->last_trans < trans->transid) {
286 WARN_ON(root == root->fs_info->extent_root);
287 WARN_ON(root->commit_root != root->node);
290 * see below for IN_TRANS_SETUP usage rules
291 * we have the reloc mutex held now, so there
292 * is only one writer in this function
294 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
296 /* make sure readers find IN_TRANS_SETUP before
297 * they find our root->last_trans update
301 spin_lock(&root->fs_info->fs_roots_radix_lock);
302 if (root->last_trans == trans->transid) {
303 spin_unlock(&root->fs_info->fs_roots_radix_lock);
306 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
307 (unsigned long)root->root_key.objectid,
308 BTRFS_ROOT_TRANS_TAG);
309 spin_unlock(&root->fs_info->fs_roots_radix_lock);
310 root->last_trans = trans->transid;
312 /* this is pretty tricky. We don't want to
313 * take the relocation lock in btrfs_record_root_in_trans
314 * unless we're really doing the first setup for this root in
317 * Normally we'd use root->last_trans as a flag to decide
318 * if we want to take the expensive mutex.
320 * But, we have to set root->last_trans before we
321 * init the relocation root, otherwise, we trip over warnings
322 * in ctree.c. The solution used here is to flag ourselves
323 * with root IN_TRANS_SETUP. When this is 1, we're still
324 * fixing up the reloc trees and everyone must wait.
326 * When this is zero, they can trust root->last_trans and fly
327 * through btrfs_record_root_in_trans without having to take the
328 * lock. smp_wmb() makes sure that all the writes above are
329 * done before we pop in the zero below
331 btrfs_init_reloc_root(trans, root);
332 smp_mb__before_atomic();
333 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
339 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
340 struct btrfs_root *root)
342 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
346 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
350 if (root->last_trans == trans->transid &&
351 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
354 mutex_lock(&root->fs_info->reloc_mutex);
355 record_root_in_trans(trans, root);
356 mutex_unlock(&root->fs_info->reloc_mutex);
361 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
363 return (trans->state >= TRANS_STATE_BLOCKED &&
364 trans->state < TRANS_STATE_UNBLOCKED &&
368 /* wait for commit against the current transaction to become unblocked
369 * when this is done, it is safe to start a new transaction, but the current
370 * transaction might not be fully on disk.
372 static void wait_current_trans(struct btrfs_root *root)
374 struct btrfs_transaction *cur_trans;
376 spin_lock(&root->fs_info->trans_lock);
377 cur_trans = root->fs_info->running_transaction;
378 if (cur_trans && is_transaction_blocked(cur_trans)) {
379 atomic_inc(&cur_trans->use_count);
380 spin_unlock(&root->fs_info->trans_lock);
382 wait_event(root->fs_info->transaction_wait,
383 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
385 btrfs_put_transaction(cur_trans);
387 spin_unlock(&root->fs_info->trans_lock);
391 static int may_wait_transaction(struct btrfs_root *root, int type)
393 if (root->fs_info->log_root_recovering)
396 if (type == TRANS_USERSPACE)
399 if (type == TRANS_START &&
400 !atomic_read(&root->fs_info->open_ioctl_trans))
406 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
408 if (!root->fs_info->reloc_ctl ||
409 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
410 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
417 static struct btrfs_trans_handle *
418 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
419 enum btrfs_reserve_flush_enum flush)
421 struct btrfs_trans_handle *h;
422 struct btrfs_transaction *cur_trans;
424 u64 qgroup_reserved = 0;
425 bool reloc_reserved = false;
428 /* Send isn't supposed to start transactions. */
429 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
431 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
432 return ERR_PTR(-EROFS);
434 if (current->journal_info) {
435 WARN_ON(type & TRANS_EXTWRITERS);
436 h = current->journal_info;
438 WARN_ON(h->use_count > 2);
439 h->orig_rsv = h->block_rsv;
445 * Do the reservation before we join the transaction so we can do all
446 * the appropriate flushing if need be.
448 if (num_items > 0 && root != root->fs_info->chunk_root) {
449 if (root->fs_info->quota_enabled &&
450 is_fstree(root->root_key.objectid)) {
451 qgroup_reserved = num_items * root->nodesize;
452 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
457 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
459 * Do the reservation for the relocation root creation
461 if (need_reserve_reloc_root(root)) {
462 num_bytes += root->nodesize;
463 reloc_reserved = true;
466 ret = btrfs_block_rsv_add(root,
467 &root->fs_info->trans_block_rsv,
473 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
480 * If we are JOIN_NOLOCK we're already committing a transaction and
481 * waiting on this guy, so we don't need to do the sb_start_intwrite
482 * because we're already holding a ref. We need this because we could
483 * have raced in and did an fsync() on a file which can kick a commit
484 * and then we deadlock with somebody doing a freeze.
486 * If we are ATTACH, it means we just want to catch the current
487 * transaction and commit it, so we needn't do sb_start_intwrite().
489 if (type & __TRANS_FREEZABLE)
490 sb_start_intwrite(root->fs_info->sb);
492 if (may_wait_transaction(root, type))
493 wait_current_trans(root);
496 ret = join_transaction(root, type);
498 wait_current_trans(root);
499 if (unlikely(type == TRANS_ATTACH))
502 } while (ret == -EBUSY);
505 /* We must get the transaction if we are JOIN_NOLOCK. */
506 BUG_ON(type == TRANS_JOIN_NOLOCK);
510 cur_trans = root->fs_info->running_transaction;
512 h->transid = cur_trans->transid;
513 h->transaction = cur_trans;
515 h->bytes_reserved = 0;
516 h->chunk_bytes_reserved = 0;
518 h->delayed_ref_updates = 0;
524 h->qgroup_reserved = 0;
525 h->delayed_ref_elem.seq = 0;
527 h->allocating_chunk = false;
528 h->reloc_reserved = false;
530 INIT_LIST_HEAD(&h->qgroup_ref_list);
531 INIT_LIST_HEAD(&h->new_bgs);
532 INIT_LIST_HEAD(&h->ordered);
535 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
536 may_wait_transaction(root, type)) {
537 current->journal_info = h;
538 btrfs_commit_transaction(h, root);
543 trace_btrfs_space_reservation(root->fs_info, "transaction",
544 h->transid, num_bytes, 1);
545 h->block_rsv = &root->fs_info->trans_block_rsv;
546 h->bytes_reserved = num_bytes;
547 h->reloc_reserved = reloc_reserved;
549 h->qgroup_reserved = qgroup_reserved;
552 btrfs_record_root_in_trans(h, root);
554 if (!current->journal_info && type != TRANS_USERSPACE)
555 current->journal_info = h;
559 if (type & __TRANS_FREEZABLE)
560 sb_end_intwrite(root->fs_info->sb);
561 kmem_cache_free(btrfs_trans_handle_cachep, h);
564 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
568 btrfs_qgroup_free(root, qgroup_reserved);
572 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
575 return start_transaction(root, num_items, TRANS_START,
576 BTRFS_RESERVE_FLUSH_ALL);
579 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
580 struct btrfs_root *root, int num_items)
582 return start_transaction(root, num_items, TRANS_START,
583 BTRFS_RESERVE_FLUSH_LIMIT);
586 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
588 return start_transaction(root, 0, TRANS_JOIN, 0);
591 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
593 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
596 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
598 return start_transaction(root, 0, TRANS_USERSPACE, 0);
602 * btrfs_attach_transaction() - catch the running transaction
604 * It is used when we want to commit the current the transaction, but
605 * don't want to start a new one.
607 * Note: If this function return -ENOENT, it just means there is no
608 * running transaction. But it is possible that the inactive transaction
609 * is still in the memory, not fully on disk. If you hope there is no
610 * inactive transaction in the fs when -ENOENT is returned, you should
612 * btrfs_attach_transaction_barrier()
614 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
616 return start_transaction(root, 0, TRANS_ATTACH, 0);
620 * btrfs_attach_transaction_barrier() - catch the running transaction
622 * It is similar to the above function, the differentia is this one
623 * will wait for all the inactive transactions until they fully
626 struct btrfs_trans_handle *
627 btrfs_attach_transaction_barrier(struct btrfs_root *root)
629 struct btrfs_trans_handle *trans;
631 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
632 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
633 btrfs_wait_for_commit(root, 0);
638 /* wait for a transaction commit to be fully complete */
639 static noinline void wait_for_commit(struct btrfs_root *root,
640 struct btrfs_transaction *commit)
642 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
645 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
647 struct btrfs_transaction *cur_trans = NULL, *t;
651 if (transid <= root->fs_info->last_trans_committed)
654 /* find specified transaction */
655 spin_lock(&root->fs_info->trans_lock);
656 list_for_each_entry(t, &root->fs_info->trans_list, list) {
657 if (t->transid == transid) {
659 atomic_inc(&cur_trans->use_count);
663 if (t->transid > transid) {
668 spin_unlock(&root->fs_info->trans_lock);
671 * The specified transaction doesn't exist, or we
672 * raced with btrfs_commit_transaction
675 if (transid > root->fs_info->last_trans_committed)
680 /* find newest transaction that is committing | committed */
681 spin_lock(&root->fs_info->trans_lock);
682 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
684 if (t->state >= TRANS_STATE_COMMIT_START) {
685 if (t->state == TRANS_STATE_COMPLETED)
688 atomic_inc(&cur_trans->use_count);
692 spin_unlock(&root->fs_info->trans_lock);
694 goto out; /* nothing committing|committed */
697 wait_for_commit(root, cur_trans);
698 btrfs_put_transaction(cur_trans);
703 void btrfs_throttle(struct btrfs_root *root)
705 if (!atomic_read(&root->fs_info->open_ioctl_trans))
706 wait_current_trans(root);
709 static int should_end_transaction(struct btrfs_trans_handle *trans,
710 struct btrfs_root *root)
712 if (root->fs_info->global_block_rsv.space_info->full &&
713 btrfs_check_space_for_delayed_refs(trans, root))
716 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
719 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
720 struct btrfs_root *root)
722 struct btrfs_transaction *cur_trans = trans->transaction;
727 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
728 cur_trans->delayed_refs.flushing)
731 updates = trans->delayed_ref_updates;
732 trans->delayed_ref_updates = 0;
734 err = btrfs_run_delayed_refs(trans, root, updates * 2);
735 if (err) /* Error code will also eval true */
739 return should_end_transaction(trans, root);
742 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, int throttle)
745 struct btrfs_transaction *cur_trans = trans->transaction;
746 struct btrfs_fs_info *info = root->fs_info;
747 unsigned long cur = trans->delayed_ref_updates;
748 int lock = (trans->type != TRANS_JOIN_NOLOCK);
750 int must_run_delayed_refs = 0;
752 if (trans->use_count > 1) {
754 trans->block_rsv = trans->orig_rsv;
758 btrfs_trans_release_metadata(trans, root);
759 trans->block_rsv = NULL;
761 if (!list_empty(&trans->new_bgs))
762 btrfs_create_pending_block_groups(trans, root);
764 if (!list_empty(&trans->ordered)) {
765 spin_lock(&info->trans_lock);
766 list_splice_init(&trans->ordered, &cur_trans->pending_ordered);
767 spin_unlock(&info->trans_lock);
770 trans->delayed_ref_updates = 0;
772 must_run_delayed_refs =
773 btrfs_should_throttle_delayed_refs(trans, root);
774 cur = max_t(unsigned long, cur, 32);
777 * don't make the caller wait if they are from a NOLOCK
778 * or ATTACH transaction, it will deadlock with commit
780 if (must_run_delayed_refs == 1 &&
781 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
782 must_run_delayed_refs = 2;
785 if (trans->qgroup_reserved) {
787 * the same root has to be passed here between start_transaction
788 * and end_transaction. Subvolume quota depends on this.
790 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
791 trans->qgroup_reserved = 0;
794 btrfs_trans_release_metadata(trans, root);
795 trans->block_rsv = NULL;
797 if (!list_empty(&trans->new_bgs))
798 btrfs_create_pending_block_groups(trans, root);
800 btrfs_trans_release_chunk_metadata(trans);
802 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
803 should_end_transaction(trans, root) &&
804 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
805 spin_lock(&info->trans_lock);
806 if (cur_trans->state == TRANS_STATE_RUNNING)
807 cur_trans->state = TRANS_STATE_BLOCKED;
808 spin_unlock(&info->trans_lock);
811 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
813 return btrfs_commit_transaction(trans, root);
815 wake_up_process(info->transaction_kthread);
818 if (trans->type & __TRANS_FREEZABLE)
819 sb_end_intwrite(root->fs_info->sb);
821 WARN_ON(cur_trans != info->running_transaction);
822 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
823 atomic_dec(&cur_trans->num_writers);
824 extwriter_counter_dec(cur_trans, trans->type);
827 if (waitqueue_active(&cur_trans->writer_wait))
828 wake_up(&cur_trans->writer_wait);
829 btrfs_put_transaction(cur_trans);
831 if (current->journal_info == trans)
832 current->journal_info = NULL;
835 btrfs_run_delayed_iputs(root);
837 if (trans->aborted ||
838 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
839 wake_up_process(info->transaction_kthread);
842 assert_qgroups_uptodate(trans);
844 kmem_cache_free(btrfs_trans_handle_cachep, trans);
845 if (must_run_delayed_refs) {
846 btrfs_async_run_delayed_refs(root, cur,
847 must_run_delayed_refs == 1);
852 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
853 struct btrfs_root *root)
855 return __btrfs_end_transaction(trans, root, 0);
858 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
859 struct btrfs_root *root)
861 return __btrfs_end_transaction(trans, root, 1);
865 * when btree blocks are allocated, they have some corresponding bits set for
866 * them in one of two extent_io trees. This is used to make sure all of
867 * those extents are sent to disk but does not wait on them
869 int btrfs_write_marked_extents(struct btrfs_root *root,
870 struct extent_io_tree *dirty_pages, int mark)
874 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
875 struct extent_state *cached_state = NULL;
879 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
880 mark, &cached_state)) {
881 bool wait_writeback = false;
883 err = convert_extent_bit(dirty_pages, start, end,
885 mark, &cached_state, GFP_NOFS);
887 * convert_extent_bit can return -ENOMEM, which is most of the
888 * time a temporary error. So when it happens, ignore the error
889 * and wait for writeback of this range to finish - because we
890 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
891 * to btrfs_wait_marked_extents() would not know that writeback
892 * for this range started and therefore wouldn't wait for it to
893 * finish - we don't want to commit a superblock that points to
894 * btree nodes/leafs for which writeback hasn't finished yet
895 * (and without errors).
896 * We cleanup any entries left in the io tree when committing
897 * the transaction (through clear_btree_io_tree()).
899 if (err == -ENOMEM) {
901 wait_writeback = true;
904 err = filemap_fdatawrite_range(mapping, start, end);
907 else if (wait_writeback)
908 werr = filemap_fdatawait_range(mapping, start, end);
909 free_extent_state(cached_state);
918 * when btree blocks are allocated, they have some corresponding bits set for
919 * them in one of two extent_io trees. This is used to make sure all of
920 * those extents are on disk for transaction or log commit. We wait
921 * on all the pages and clear them from the dirty pages state tree
923 int btrfs_wait_marked_extents(struct btrfs_root *root,
924 struct extent_io_tree *dirty_pages, int mark)
928 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
929 struct extent_state *cached_state = NULL;
932 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
935 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
936 EXTENT_NEED_WAIT, &cached_state)) {
938 * Ignore -ENOMEM errors returned by clear_extent_bit().
939 * When committing the transaction, we'll remove any entries
940 * left in the io tree. For a log commit, we don't remove them
941 * after committing the log because the tree can be accessed
942 * concurrently - we do it only at transaction commit time when
943 * it's safe to do it (through clear_btree_io_tree()).
945 err = clear_extent_bit(dirty_pages, start, end,
947 0, 0, &cached_state, GFP_NOFS);
951 err = filemap_fdatawait_range(mapping, start, end);
954 free_extent_state(cached_state);
962 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
963 if ((mark & EXTENT_DIRTY) &&
964 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
965 &btree_ino->runtime_flags))
968 if ((mark & EXTENT_NEW) &&
969 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
970 &btree_ino->runtime_flags))
973 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
974 &btree_ino->runtime_flags))
985 * when btree blocks are allocated, they have some corresponding bits set for
986 * them in one of two extent_io trees. This is used to make sure all of
987 * those extents are on disk for transaction or log commit
989 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
990 struct extent_io_tree *dirty_pages, int mark)
994 struct blk_plug plug;
996 blk_start_plug(&plug);
997 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
998 blk_finish_plug(&plug);
999 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
1008 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
1009 struct btrfs_root *root)
1013 ret = btrfs_write_and_wait_marked_extents(root,
1014 &trans->transaction->dirty_pages,
1016 clear_btree_io_tree(&trans->transaction->dirty_pages);
1022 * this is used to update the root pointer in the tree of tree roots.
1024 * But, in the case of the extent allocation tree, updating the root
1025 * pointer may allocate blocks which may change the root of the extent
1028 * So, this loops and repeats and makes sure the cowonly root didn't
1029 * change while the root pointer was being updated in the metadata.
1031 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1032 struct btrfs_root *root)
1035 u64 old_root_bytenr;
1037 struct btrfs_root *tree_root = root->fs_info->tree_root;
1039 old_root_used = btrfs_root_used(&root->root_item);
1042 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1043 if (old_root_bytenr == root->node->start &&
1044 old_root_used == btrfs_root_used(&root->root_item))
1047 btrfs_set_root_node(&root->root_item, root->node);
1048 ret = btrfs_update_root(trans, tree_root,
1054 old_root_used = btrfs_root_used(&root->root_item);
1061 * update all the cowonly tree roots on disk
1063 * The error handling in this function may not be obvious. Any of the
1064 * failures will cause the file system to go offline. We still need
1065 * to clean up the delayed refs.
1067 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1068 struct btrfs_root *root)
1070 struct btrfs_fs_info *fs_info = root->fs_info;
1071 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1072 struct list_head *io_bgs = &trans->transaction->io_bgs;
1073 struct list_head *next;
1074 struct extent_buffer *eb;
1077 eb = btrfs_lock_root_node(fs_info->tree_root);
1078 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1080 btrfs_tree_unlock(eb);
1081 free_extent_buffer(eb);
1086 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1090 ret = btrfs_run_dev_stats(trans, root->fs_info);
1093 ret = btrfs_run_dev_replace(trans, root->fs_info);
1096 ret = btrfs_run_qgroups(trans, root->fs_info);
1100 ret = btrfs_setup_space_cache(trans, root);
1104 /* run_qgroups might have added some more refs */
1105 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1109 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1110 next = fs_info->dirty_cowonly_roots.next;
1111 list_del_init(next);
1112 root = list_entry(next, struct btrfs_root, dirty_list);
1113 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1115 if (root != fs_info->extent_root)
1116 list_add_tail(&root->dirty_list,
1117 &trans->transaction->switch_commits);
1118 ret = update_cowonly_root(trans, root);
1121 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1126 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1127 ret = btrfs_write_dirty_block_groups(trans, root);
1130 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1135 if (!list_empty(&fs_info->dirty_cowonly_roots))
1138 list_add_tail(&fs_info->extent_root->dirty_list,
1139 &trans->transaction->switch_commits);
1140 btrfs_after_dev_replace_commit(fs_info);
1146 * dead roots are old snapshots that need to be deleted. This allocates
1147 * a dirty root struct and adds it into the list of dead roots that need to
1150 void btrfs_add_dead_root(struct btrfs_root *root)
1152 spin_lock(&root->fs_info->trans_lock);
1153 if (list_empty(&root->root_list))
1154 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1155 spin_unlock(&root->fs_info->trans_lock);
1159 * update all the cowonly tree roots on disk
1161 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1162 struct btrfs_root *root)
1164 struct btrfs_root *gang[8];
1165 struct btrfs_fs_info *fs_info = root->fs_info;
1170 spin_lock(&fs_info->fs_roots_radix_lock);
1172 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1175 BTRFS_ROOT_TRANS_TAG);
1178 for (i = 0; i < ret; i++) {
1180 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1181 (unsigned long)root->root_key.objectid,
1182 BTRFS_ROOT_TRANS_TAG);
1183 spin_unlock(&fs_info->fs_roots_radix_lock);
1185 btrfs_free_log(trans, root);
1186 btrfs_update_reloc_root(trans, root);
1187 btrfs_orphan_commit_root(trans, root);
1189 btrfs_save_ino_cache(root, trans);
1191 /* see comments in should_cow_block() */
1192 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1193 smp_mb__after_atomic();
1195 if (root->commit_root != root->node) {
1196 list_add_tail(&root->dirty_list,
1197 &trans->transaction->switch_commits);
1198 btrfs_set_root_node(&root->root_item,
1202 err = btrfs_update_root(trans, fs_info->tree_root,
1205 spin_lock(&fs_info->fs_roots_radix_lock);
1210 spin_unlock(&fs_info->fs_roots_radix_lock);
1215 * defrag a given btree.
1216 * Every leaf in the btree is read and defragged.
1218 int btrfs_defrag_root(struct btrfs_root *root)
1220 struct btrfs_fs_info *info = root->fs_info;
1221 struct btrfs_trans_handle *trans;
1224 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1228 trans = btrfs_start_transaction(root, 0);
1230 return PTR_ERR(trans);
1232 ret = btrfs_defrag_leaves(trans, root);
1234 btrfs_end_transaction(trans, root);
1235 btrfs_btree_balance_dirty(info->tree_root);
1238 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1241 if (btrfs_defrag_cancelled(root->fs_info)) {
1242 pr_debug("BTRFS: defrag_root cancelled\n");
1247 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1252 * new snapshots need to be created at a very specific time in the
1253 * transaction commit. This does the actual creation.
1256 * If the error which may affect the commitment of the current transaction
1257 * happens, we should return the error number. If the error which just affect
1258 * the creation of the pending snapshots, just return 0.
1260 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1261 struct btrfs_fs_info *fs_info,
1262 struct btrfs_pending_snapshot *pending)
1264 struct btrfs_key key;
1265 struct btrfs_root_item *new_root_item;
1266 struct btrfs_root *tree_root = fs_info->tree_root;
1267 struct btrfs_root *root = pending->root;
1268 struct btrfs_root *parent_root;
1269 struct btrfs_block_rsv *rsv;
1270 struct inode *parent_inode;
1271 struct btrfs_path *path;
1272 struct btrfs_dir_item *dir_item;
1273 struct dentry *dentry;
1274 struct extent_buffer *tmp;
1275 struct extent_buffer *old;
1276 struct timespec cur_time = CURRENT_TIME;
1284 path = btrfs_alloc_path();
1286 pending->error = -ENOMEM;
1290 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1291 if (!new_root_item) {
1292 pending->error = -ENOMEM;
1293 goto root_item_alloc_fail;
1296 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1298 goto no_free_objectid;
1301 * Make qgroup to skip current new snapshot's qgroupid, as it is
1302 * accounted by later btrfs_qgroup_inherit().
1304 btrfs_set_skip_qgroup(trans, objectid);
1306 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1308 if (to_reserve > 0) {
1309 pending->error = btrfs_block_rsv_add(root,
1310 &pending->block_rsv,
1312 BTRFS_RESERVE_NO_FLUSH);
1314 goto clear_skip_qgroup;
1317 key.objectid = objectid;
1318 key.offset = (u64)-1;
1319 key.type = BTRFS_ROOT_ITEM_KEY;
1321 rsv = trans->block_rsv;
1322 trans->block_rsv = &pending->block_rsv;
1323 trans->bytes_reserved = trans->block_rsv->reserved;
1325 dentry = pending->dentry;
1326 parent_inode = pending->dir;
1327 parent_root = BTRFS_I(parent_inode)->root;
1328 record_root_in_trans(trans, parent_root);
1331 * insert the directory item
1333 ret = btrfs_set_inode_index(parent_inode, &index);
1334 BUG_ON(ret); /* -ENOMEM */
1336 /* check if there is a file/dir which has the same name. */
1337 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1338 btrfs_ino(parent_inode),
1339 dentry->d_name.name,
1340 dentry->d_name.len, 0);
1341 if (dir_item != NULL && !IS_ERR(dir_item)) {
1342 pending->error = -EEXIST;
1343 goto dir_item_existed;
1344 } else if (IS_ERR(dir_item)) {
1345 ret = PTR_ERR(dir_item);
1346 btrfs_abort_transaction(trans, root, ret);
1349 btrfs_release_path(path);
1352 * pull in the delayed directory update
1353 * and the delayed inode item
1354 * otherwise we corrupt the FS during
1357 ret = btrfs_run_delayed_items(trans, root);
1358 if (ret) { /* Transaction aborted */
1359 btrfs_abort_transaction(trans, root, ret);
1363 record_root_in_trans(trans, root);
1364 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1365 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1366 btrfs_check_and_init_root_item(new_root_item);
1368 root_flags = btrfs_root_flags(new_root_item);
1369 if (pending->readonly)
1370 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1372 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1373 btrfs_set_root_flags(new_root_item, root_flags);
1375 btrfs_set_root_generation_v2(new_root_item,
1377 uuid_le_gen(&new_uuid);
1378 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1379 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1381 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1382 memset(new_root_item->received_uuid, 0,
1383 sizeof(new_root_item->received_uuid));
1384 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1385 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1386 btrfs_set_root_stransid(new_root_item, 0);
1387 btrfs_set_root_rtransid(new_root_item, 0);
1389 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1390 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1391 btrfs_set_root_otransid(new_root_item, trans->transid);
1393 old = btrfs_lock_root_node(root);
1394 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1396 btrfs_tree_unlock(old);
1397 free_extent_buffer(old);
1398 btrfs_abort_transaction(trans, root, ret);
1402 btrfs_set_lock_blocking(old);
1404 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1405 /* clean up in any case */
1406 btrfs_tree_unlock(old);
1407 free_extent_buffer(old);
1409 btrfs_abort_transaction(trans, root, ret);
1412 /* see comments in should_cow_block() */
1413 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1416 btrfs_set_root_node(new_root_item, tmp);
1417 /* record when the snapshot was created in key.offset */
1418 key.offset = trans->transid;
1419 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1420 btrfs_tree_unlock(tmp);
1421 free_extent_buffer(tmp);
1423 btrfs_abort_transaction(trans, root, ret);
1428 * insert root back/forward references
1430 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1431 parent_root->root_key.objectid,
1432 btrfs_ino(parent_inode), index,
1433 dentry->d_name.name, dentry->d_name.len);
1435 btrfs_abort_transaction(trans, root, ret);
1439 key.offset = (u64)-1;
1440 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1441 if (IS_ERR(pending->snap)) {
1442 ret = PTR_ERR(pending->snap);
1443 btrfs_abort_transaction(trans, root, ret);
1447 ret = btrfs_reloc_post_snapshot(trans, pending);
1449 btrfs_abort_transaction(trans, root, ret);
1453 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1455 btrfs_abort_transaction(trans, root, ret);
1459 ret = btrfs_insert_dir_item(trans, parent_root,
1460 dentry->d_name.name, dentry->d_name.len,
1462 BTRFS_FT_DIR, index);
1463 /* We have check then name at the beginning, so it is impossible. */
1464 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1466 btrfs_abort_transaction(trans, root, ret);
1470 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1471 dentry->d_name.len * 2);
1472 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1473 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1475 btrfs_abort_transaction(trans, root, ret);
1478 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1479 BTRFS_UUID_KEY_SUBVOL, objectid);
1481 btrfs_abort_transaction(trans, root, ret);
1484 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1485 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1486 new_root_item->received_uuid,
1487 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1489 if (ret && ret != -EEXIST) {
1490 btrfs_abort_transaction(trans, root, ret);
1495 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1497 btrfs_abort_transaction(trans, root, ret);
1502 * account qgroup counters before qgroup_inherit()
1504 ret = btrfs_qgroup_prepare_account_extents(trans, fs_info);
1507 ret = btrfs_qgroup_account_extents(trans, fs_info);
1510 ret = btrfs_qgroup_inherit(trans, fs_info,
1511 root->root_key.objectid,
1512 objectid, pending->inherit);
1514 btrfs_abort_transaction(trans, root, ret);
1519 pending->error = ret;
1521 trans->block_rsv = rsv;
1522 trans->bytes_reserved = 0;
1524 btrfs_clear_skip_qgroup(trans);
1526 kfree(new_root_item);
1527 root_item_alloc_fail:
1528 btrfs_free_path(path);
1533 * create all the snapshots we've scheduled for creation
1535 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1536 struct btrfs_fs_info *fs_info)
1538 struct btrfs_pending_snapshot *pending, *next;
1539 struct list_head *head = &trans->transaction->pending_snapshots;
1542 list_for_each_entry_safe(pending, next, head, list) {
1543 list_del(&pending->list);
1544 ret = create_pending_snapshot(trans, fs_info, pending);
1551 static void update_super_roots(struct btrfs_root *root)
1553 struct btrfs_root_item *root_item;
1554 struct btrfs_super_block *super;
1556 super = root->fs_info->super_copy;
1558 root_item = &root->fs_info->chunk_root->root_item;
1559 super->chunk_root = root_item->bytenr;
1560 super->chunk_root_generation = root_item->generation;
1561 super->chunk_root_level = root_item->level;
1563 root_item = &root->fs_info->tree_root->root_item;
1564 super->root = root_item->bytenr;
1565 super->generation = root_item->generation;
1566 super->root_level = root_item->level;
1567 if (btrfs_test_opt(root, SPACE_CACHE))
1568 super->cache_generation = root_item->generation;
1569 if (root->fs_info->update_uuid_tree_gen)
1570 super->uuid_tree_generation = root_item->generation;
1573 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1575 struct btrfs_transaction *trans;
1578 spin_lock(&info->trans_lock);
1579 trans = info->running_transaction;
1581 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1582 spin_unlock(&info->trans_lock);
1586 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1588 struct btrfs_transaction *trans;
1591 spin_lock(&info->trans_lock);
1592 trans = info->running_transaction;
1594 ret = is_transaction_blocked(trans);
1595 spin_unlock(&info->trans_lock);
1600 * wait for the current transaction commit to start and block subsequent
1603 static void wait_current_trans_commit_start(struct btrfs_root *root,
1604 struct btrfs_transaction *trans)
1606 wait_event(root->fs_info->transaction_blocked_wait,
1607 trans->state >= TRANS_STATE_COMMIT_START ||
1612 * wait for the current transaction to start and then become unblocked.
1615 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1616 struct btrfs_transaction *trans)
1618 wait_event(root->fs_info->transaction_wait,
1619 trans->state >= TRANS_STATE_UNBLOCKED ||
1624 * commit transactions asynchronously. once btrfs_commit_transaction_async
1625 * returns, any subsequent transaction will not be allowed to join.
1627 struct btrfs_async_commit {
1628 struct btrfs_trans_handle *newtrans;
1629 struct btrfs_root *root;
1630 struct work_struct work;
1633 static void do_async_commit(struct work_struct *work)
1635 struct btrfs_async_commit *ac =
1636 container_of(work, struct btrfs_async_commit, work);
1639 * We've got freeze protection passed with the transaction.
1640 * Tell lockdep about it.
1642 if (ac->newtrans->type & __TRANS_FREEZABLE)
1643 __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS);
1645 current->journal_info = ac->newtrans;
1647 btrfs_commit_transaction(ac->newtrans, ac->root);
1651 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1652 struct btrfs_root *root,
1653 int wait_for_unblock)
1655 struct btrfs_async_commit *ac;
1656 struct btrfs_transaction *cur_trans;
1658 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1662 INIT_WORK(&ac->work, do_async_commit);
1664 ac->newtrans = btrfs_join_transaction(root);
1665 if (IS_ERR(ac->newtrans)) {
1666 int err = PTR_ERR(ac->newtrans);
1671 /* take transaction reference */
1672 cur_trans = trans->transaction;
1673 atomic_inc(&cur_trans->use_count);
1675 btrfs_end_transaction(trans, root);
1678 * Tell lockdep we've released the freeze rwsem, since the
1679 * async commit thread will be the one to unlock it.
1681 if (ac->newtrans->type & __TRANS_FREEZABLE)
1682 __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS);
1684 schedule_work(&ac->work);
1686 /* wait for transaction to start and unblock */
1687 if (wait_for_unblock)
1688 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1690 wait_current_trans_commit_start(root, cur_trans);
1692 if (current->journal_info == trans)
1693 current->journal_info = NULL;
1695 btrfs_put_transaction(cur_trans);
1700 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1701 struct btrfs_root *root, int err)
1703 struct btrfs_transaction *cur_trans = trans->transaction;
1706 WARN_ON(trans->use_count > 1);
1708 btrfs_abort_transaction(trans, root, err);
1710 spin_lock(&root->fs_info->trans_lock);
1713 * If the transaction is removed from the list, it means this
1714 * transaction has been committed successfully, so it is impossible
1715 * to call the cleanup function.
1717 BUG_ON(list_empty(&cur_trans->list));
1719 list_del_init(&cur_trans->list);
1720 if (cur_trans == root->fs_info->running_transaction) {
1721 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1722 spin_unlock(&root->fs_info->trans_lock);
1723 wait_event(cur_trans->writer_wait,
1724 atomic_read(&cur_trans->num_writers) == 1);
1726 spin_lock(&root->fs_info->trans_lock);
1728 spin_unlock(&root->fs_info->trans_lock);
1730 btrfs_cleanup_one_transaction(trans->transaction, root);
1732 spin_lock(&root->fs_info->trans_lock);
1733 if (cur_trans == root->fs_info->running_transaction)
1734 root->fs_info->running_transaction = NULL;
1735 spin_unlock(&root->fs_info->trans_lock);
1737 if (trans->type & __TRANS_FREEZABLE)
1738 sb_end_intwrite(root->fs_info->sb);
1739 btrfs_put_transaction(cur_trans);
1740 btrfs_put_transaction(cur_trans);
1742 trace_btrfs_transaction_commit(root);
1744 if (current->journal_info == trans)
1745 current->journal_info = NULL;
1746 btrfs_scrub_cancel(root->fs_info);
1748 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1751 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1753 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1754 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1758 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1760 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1761 btrfs_wait_ordered_roots(fs_info, -1);
1765 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans,
1766 struct btrfs_fs_info *fs_info)
1768 struct btrfs_ordered_extent *ordered;
1770 spin_lock(&fs_info->trans_lock);
1771 while (!list_empty(&cur_trans->pending_ordered)) {
1772 ordered = list_first_entry(&cur_trans->pending_ordered,
1773 struct btrfs_ordered_extent,
1775 list_del_init(&ordered->trans_list);
1776 spin_unlock(&fs_info->trans_lock);
1778 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE,
1780 btrfs_put_ordered_extent(ordered);
1781 spin_lock(&fs_info->trans_lock);
1783 spin_unlock(&fs_info->trans_lock);
1786 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1787 struct btrfs_root *root)
1789 struct btrfs_transaction *cur_trans = trans->transaction;
1790 struct btrfs_transaction *prev_trans = NULL;
1791 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
1794 /* Stop the commit early if ->aborted is set */
1795 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1796 ret = cur_trans->aborted;
1797 btrfs_end_transaction(trans, root);
1801 /* make a pass through all the delayed refs we have so far
1802 * any runnings procs may add more while we are here
1804 ret = btrfs_run_delayed_refs(trans, root, 0);
1806 btrfs_end_transaction(trans, root);
1810 btrfs_trans_release_metadata(trans, root);
1811 trans->block_rsv = NULL;
1812 if (trans->qgroup_reserved) {
1813 btrfs_qgroup_free(root, trans->qgroup_reserved);
1814 trans->qgroup_reserved = 0;
1817 cur_trans = trans->transaction;
1820 * set the flushing flag so procs in this transaction have to
1821 * start sending their work down.
1823 cur_trans->delayed_refs.flushing = 1;
1826 if (!list_empty(&trans->new_bgs))
1827 btrfs_create_pending_block_groups(trans, root);
1829 ret = btrfs_run_delayed_refs(trans, root, 0);
1831 btrfs_end_transaction(trans, root);
1835 if (!cur_trans->dirty_bg_run) {
1838 /* this mutex is also taken before trying to set
1839 * block groups readonly. We need to make sure
1840 * that nobody has set a block group readonly
1841 * after a extents from that block group have been
1842 * allocated for cache files. btrfs_set_block_group_ro
1843 * will wait for the transaction to commit if it
1844 * finds dirty_bg_run = 1
1846 * The dirty_bg_run flag is also used to make sure only
1847 * one process starts all the block group IO. It wouldn't
1848 * hurt to have more than one go through, but there's no
1849 * real advantage to it either.
1851 mutex_lock(&root->fs_info->ro_block_group_mutex);
1852 if (!cur_trans->dirty_bg_run) {
1854 cur_trans->dirty_bg_run = 1;
1856 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1859 ret = btrfs_start_dirty_block_groups(trans, root);
1862 btrfs_end_transaction(trans, root);
1866 spin_lock(&root->fs_info->trans_lock);
1867 list_splice_init(&trans->ordered, &cur_trans->pending_ordered);
1868 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1869 spin_unlock(&root->fs_info->trans_lock);
1870 atomic_inc(&cur_trans->use_count);
1871 ret = btrfs_end_transaction(trans, root);
1873 wait_for_commit(root, cur_trans);
1875 if (unlikely(cur_trans->aborted))
1876 ret = cur_trans->aborted;
1878 btrfs_put_transaction(cur_trans);
1883 cur_trans->state = TRANS_STATE_COMMIT_START;
1884 wake_up(&root->fs_info->transaction_blocked_wait);
1886 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1887 prev_trans = list_entry(cur_trans->list.prev,
1888 struct btrfs_transaction, list);
1889 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1890 atomic_inc(&prev_trans->use_count);
1891 spin_unlock(&root->fs_info->trans_lock);
1893 wait_for_commit(root, prev_trans);
1894 ret = prev_trans->aborted;
1896 btrfs_put_transaction(prev_trans);
1898 goto cleanup_transaction;
1900 spin_unlock(&root->fs_info->trans_lock);
1903 spin_unlock(&root->fs_info->trans_lock);
1906 extwriter_counter_dec(cur_trans, trans->type);
1908 ret = btrfs_start_delalloc_flush(root->fs_info);
1910 goto cleanup_transaction;
1912 ret = btrfs_run_delayed_items(trans, root);
1914 goto cleanup_transaction;
1916 wait_event(cur_trans->writer_wait,
1917 extwriter_counter_read(cur_trans) == 0);
1919 /* some pending stuffs might be added after the previous flush. */
1920 ret = btrfs_run_delayed_items(trans, root);
1922 goto cleanup_transaction;
1924 btrfs_wait_delalloc_flush(root->fs_info);
1926 btrfs_wait_pending_ordered(cur_trans, root->fs_info);
1928 btrfs_scrub_pause(root);
1930 * Ok now we need to make sure to block out any other joins while we
1931 * commit the transaction. We could have started a join before setting
1932 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1934 spin_lock(&root->fs_info->trans_lock);
1935 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1936 spin_unlock(&root->fs_info->trans_lock);
1937 wait_event(cur_trans->writer_wait,
1938 atomic_read(&cur_trans->num_writers) == 1);
1940 /* ->aborted might be set after the previous check, so check it */
1941 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1942 ret = cur_trans->aborted;
1943 goto scrub_continue;
1946 * the reloc mutex makes sure that we stop
1947 * the balancing code from coming in and moving
1948 * extents around in the middle of the commit
1950 mutex_lock(&root->fs_info->reloc_mutex);
1953 * We needn't worry about the delayed items because we will
1954 * deal with them in create_pending_snapshot(), which is the
1955 * core function of the snapshot creation.
1957 ret = create_pending_snapshots(trans, root->fs_info);
1959 mutex_unlock(&root->fs_info->reloc_mutex);
1960 goto scrub_continue;
1964 * We insert the dir indexes of the snapshots and update the inode
1965 * of the snapshots' parents after the snapshot creation, so there
1966 * are some delayed items which are not dealt with. Now deal with
1969 * We needn't worry that this operation will corrupt the snapshots,
1970 * because all the tree which are snapshoted will be forced to COW
1971 * the nodes and leaves.
1973 ret = btrfs_run_delayed_items(trans, root);
1975 mutex_unlock(&root->fs_info->reloc_mutex);
1976 goto scrub_continue;
1979 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1981 mutex_unlock(&root->fs_info->reloc_mutex);
1982 goto scrub_continue;
1985 /* Reocrd old roots for later qgroup accounting */
1986 ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info);
1988 mutex_unlock(&root->fs_info->reloc_mutex);
1989 goto scrub_continue;
1993 * make sure none of the code above managed to slip in a
1996 btrfs_assert_delayed_root_empty(root);
1998 WARN_ON(cur_trans != trans->transaction);
2000 /* btrfs_commit_tree_roots is responsible for getting the
2001 * various roots consistent with each other. Every pointer
2002 * in the tree of tree roots has to point to the most up to date
2003 * root for every subvolume and other tree. So, we have to keep
2004 * the tree logging code from jumping in and changing any
2007 * At this point in the commit, there can't be any tree-log
2008 * writers, but a little lower down we drop the trans mutex
2009 * and let new people in. By holding the tree_log_mutex
2010 * from now until after the super is written, we avoid races
2011 * with the tree-log code.
2013 mutex_lock(&root->fs_info->tree_log_mutex);
2015 ret = commit_fs_roots(trans, root);
2017 mutex_unlock(&root->fs_info->tree_log_mutex);
2018 mutex_unlock(&root->fs_info->reloc_mutex);
2019 goto scrub_continue;
2023 * Since the transaction is done, we can apply the pending changes
2024 * before the next transaction.
2026 btrfs_apply_pending_changes(root->fs_info);
2028 /* commit_fs_roots gets rid of all the tree log roots, it is now
2029 * safe to free the root of tree log roots
2031 btrfs_free_log_root_tree(trans, root->fs_info);
2034 * Since fs roots are all committed, we can get a quite accurate
2035 * new_roots. So let's do quota accounting.
2037 ret = btrfs_qgroup_account_extents(trans, root->fs_info);
2039 mutex_unlock(&root->fs_info->tree_log_mutex);
2040 mutex_unlock(&root->fs_info->reloc_mutex);
2041 goto scrub_continue;
2044 ret = commit_cowonly_roots(trans, root);
2046 mutex_unlock(&root->fs_info->tree_log_mutex);
2047 mutex_unlock(&root->fs_info->reloc_mutex);
2048 goto scrub_continue;
2052 * The tasks which save the space cache and inode cache may also
2053 * update ->aborted, check it.
2055 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2056 ret = cur_trans->aborted;
2057 mutex_unlock(&root->fs_info->tree_log_mutex);
2058 mutex_unlock(&root->fs_info->reloc_mutex);
2059 goto scrub_continue;
2062 btrfs_prepare_extent_commit(trans, root);
2064 cur_trans = root->fs_info->running_transaction;
2066 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2067 root->fs_info->tree_root->node);
2068 list_add_tail(&root->fs_info->tree_root->dirty_list,
2069 &cur_trans->switch_commits);
2071 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2072 root->fs_info->chunk_root->node);
2073 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2074 &cur_trans->switch_commits);
2076 switch_commit_roots(cur_trans, root->fs_info);
2078 assert_qgroups_uptodate(trans);
2079 ASSERT(list_empty(&cur_trans->dirty_bgs));
2080 ASSERT(list_empty(&cur_trans->io_bgs));
2081 update_super_roots(root);
2083 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2084 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
2085 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2086 sizeof(*root->fs_info->super_copy));
2088 btrfs_update_commit_device_size(root->fs_info);
2089 btrfs_update_commit_device_bytes_used(root, cur_trans);
2091 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
2092 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
2094 btrfs_trans_release_chunk_metadata(trans);
2096 spin_lock(&root->fs_info->trans_lock);
2097 cur_trans->state = TRANS_STATE_UNBLOCKED;
2098 root->fs_info->running_transaction = NULL;
2099 spin_unlock(&root->fs_info->trans_lock);
2100 mutex_unlock(&root->fs_info->reloc_mutex);
2102 wake_up(&root->fs_info->transaction_wait);
2104 ret = btrfs_write_and_wait_transaction(trans, root);
2106 btrfs_error(root->fs_info, ret,
2107 "Error while writing out transaction");
2108 mutex_unlock(&root->fs_info->tree_log_mutex);
2109 goto scrub_continue;
2112 ret = write_ctree_super(trans, root, 0);
2114 mutex_unlock(&root->fs_info->tree_log_mutex);
2115 goto scrub_continue;
2119 * the super is written, we can safely allow the tree-loggers
2120 * to go about their business
2122 mutex_unlock(&root->fs_info->tree_log_mutex);
2124 btrfs_finish_extent_commit(trans, root);
2126 if (cur_trans->have_free_bgs)
2127 btrfs_clear_space_info_full(root->fs_info);
2129 root->fs_info->last_trans_committed = cur_trans->transid;
2131 * We needn't acquire the lock here because there is no other task
2132 * which can change it.
2134 cur_trans->state = TRANS_STATE_COMPLETED;
2135 wake_up(&cur_trans->commit_wait);
2137 spin_lock(&root->fs_info->trans_lock);
2138 list_del_init(&cur_trans->list);
2139 spin_unlock(&root->fs_info->trans_lock);
2141 btrfs_put_transaction(cur_trans);
2142 btrfs_put_transaction(cur_trans);
2144 if (trans->type & __TRANS_FREEZABLE)
2145 sb_end_intwrite(root->fs_info->sb);
2147 trace_btrfs_transaction_commit(root);
2149 btrfs_scrub_continue(root);
2151 if (current->journal_info == trans)
2152 current->journal_info = NULL;
2154 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2156 if (current != root->fs_info->transaction_kthread &&
2157 current != root->fs_info->cleaner_kthread)
2158 btrfs_run_delayed_iputs(root);
2163 btrfs_scrub_continue(root);
2164 cleanup_transaction:
2165 btrfs_trans_release_metadata(trans, root);
2166 btrfs_trans_release_chunk_metadata(trans);
2167 trans->block_rsv = NULL;
2168 if (trans->qgroup_reserved) {
2169 btrfs_qgroup_free(root, trans->qgroup_reserved);
2170 trans->qgroup_reserved = 0;
2172 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
2173 if (current->journal_info == trans)
2174 current->journal_info = NULL;
2175 cleanup_transaction(trans, root, ret);
2181 * return < 0 if error
2182 * 0 if there are no more dead_roots at the time of call
2183 * 1 there are more to be processed, call me again
2185 * The return value indicates there are certainly more snapshots to delete, but
2186 * if there comes a new one during processing, it may return 0. We don't mind,
2187 * because btrfs_commit_super will poke cleaner thread and it will process it a
2188 * few seconds later.
2190 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2193 struct btrfs_fs_info *fs_info = root->fs_info;
2195 spin_lock(&fs_info->trans_lock);
2196 if (list_empty(&fs_info->dead_roots)) {
2197 spin_unlock(&fs_info->trans_lock);
2200 root = list_first_entry(&fs_info->dead_roots,
2201 struct btrfs_root, root_list);
2202 list_del_init(&root->root_list);
2203 spin_unlock(&fs_info->trans_lock);
2205 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2207 btrfs_kill_all_delayed_nodes(root);
2209 if (btrfs_header_backref_rev(root->node) <
2210 BTRFS_MIXED_BACKREF_REV)
2211 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2213 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2215 return (ret < 0) ? 0 : 1;
2218 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2223 prev = xchg(&fs_info->pending_changes, 0);
2227 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2229 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2232 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2234 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2237 bit = 1 << BTRFS_PENDING_COMMIT;
2239 btrfs_debug(fs_info, "pending commit done");
2244 "unknown pending changes left 0x%lx, ignoring", prev);