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 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
252 spin_lock_init(&cur_trans->dirty_bgs_lock);
253 list_add_tail(&cur_trans->list, &fs_info->trans_list);
254 extent_io_tree_init(&cur_trans->dirty_pages,
255 fs_info->btree_inode->i_mapping);
256 fs_info->generation++;
257 cur_trans->transid = fs_info->generation;
258 fs_info->running_transaction = cur_trans;
259 cur_trans->aborted = 0;
260 spin_unlock(&fs_info->trans_lock);
266 * this does all the record keeping required to make sure that a reference
267 * counted root is properly recorded in a given transaction. This is required
268 * to make sure the old root from before we joined the transaction is deleted
269 * when the transaction commits
271 static int record_root_in_trans(struct btrfs_trans_handle *trans,
272 struct btrfs_root *root)
274 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
275 root->last_trans < trans->transid) {
276 WARN_ON(root == root->fs_info->extent_root);
277 WARN_ON(root->commit_root != root->node);
280 * see below for IN_TRANS_SETUP usage rules
281 * we have the reloc mutex held now, so there
282 * is only one writer in this function
284 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
286 /* make sure readers find IN_TRANS_SETUP before
287 * they find our root->last_trans update
291 spin_lock(&root->fs_info->fs_roots_radix_lock);
292 if (root->last_trans == trans->transid) {
293 spin_unlock(&root->fs_info->fs_roots_radix_lock);
296 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
297 (unsigned long)root->root_key.objectid,
298 BTRFS_ROOT_TRANS_TAG);
299 spin_unlock(&root->fs_info->fs_roots_radix_lock);
300 root->last_trans = trans->transid;
302 /* this is pretty tricky. We don't want to
303 * take the relocation lock in btrfs_record_root_in_trans
304 * unless we're really doing the first setup for this root in
307 * Normally we'd use root->last_trans as a flag to decide
308 * if we want to take the expensive mutex.
310 * But, we have to set root->last_trans before we
311 * init the relocation root, otherwise, we trip over warnings
312 * in ctree.c. The solution used here is to flag ourselves
313 * with root IN_TRANS_SETUP. When this is 1, we're still
314 * fixing up the reloc trees and everyone must wait.
316 * When this is zero, they can trust root->last_trans and fly
317 * through btrfs_record_root_in_trans without having to take the
318 * lock. smp_wmb() makes sure that all the writes above are
319 * done before we pop in the zero below
321 btrfs_init_reloc_root(trans, root);
322 smp_mb__before_atomic();
323 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
329 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
330 struct btrfs_root *root)
332 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
336 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
340 if (root->last_trans == trans->transid &&
341 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
344 mutex_lock(&root->fs_info->reloc_mutex);
345 record_root_in_trans(trans, root);
346 mutex_unlock(&root->fs_info->reloc_mutex);
351 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
353 return (trans->state >= TRANS_STATE_BLOCKED &&
354 trans->state < TRANS_STATE_UNBLOCKED &&
358 /* wait for commit against the current transaction to become unblocked
359 * when this is done, it is safe to start a new transaction, but the current
360 * transaction might not be fully on disk.
362 static void wait_current_trans(struct btrfs_root *root)
364 struct btrfs_transaction *cur_trans;
366 spin_lock(&root->fs_info->trans_lock);
367 cur_trans = root->fs_info->running_transaction;
368 if (cur_trans && is_transaction_blocked(cur_trans)) {
369 atomic_inc(&cur_trans->use_count);
370 spin_unlock(&root->fs_info->trans_lock);
372 wait_event(root->fs_info->transaction_wait,
373 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
375 btrfs_put_transaction(cur_trans);
377 spin_unlock(&root->fs_info->trans_lock);
381 static int may_wait_transaction(struct btrfs_root *root, int type)
383 if (root->fs_info->log_root_recovering)
386 if (type == TRANS_USERSPACE)
389 if (type == TRANS_START &&
390 !atomic_read(&root->fs_info->open_ioctl_trans))
396 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
398 if (!root->fs_info->reloc_ctl ||
399 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
400 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
407 static struct btrfs_trans_handle *
408 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
409 enum btrfs_reserve_flush_enum flush)
411 struct btrfs_trans_handle *h;
412 struct btrfs_transaction *cur_trans;
414 u64 qgroup_reserved = 0;
415 bool reloc_reserved = false;
418 /* Send isn't supposed to start transactions. */
419 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
421 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
422 return ERR_PTR(-EROFS);
424 if (current->journal_info) {
425 WARN_ON(type & TRANS_EXTWRITERS);
426 h = current->journal_info;
428 WARN_ON(h->use_count > 2);
429 h->orig_rsv = h->block_rsv;
435 * Do the reservation before we join the transaction so we can do all
436 * the appropriate flushing if need be.
438 if (num_items > 0 && root != root->fs_info->chunk_root) {
439 if (root->fs_info->quota_enabled &&
440 is_fstree(root->root_key.objectid)) {
441 qgroup_reserved = num_items * root->nodesize;
442 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
447 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
449 * Do the reservation for the relocation root creation
451 if (need_reserve_reloc_root(root)) {
452 num_bytes += root->nodesize;
453 reloc_reserved = true;
456 ret = btrfs_block_rsv_add(root,
457 &root->fs_info->trans_block_rsv,
463 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
470 * If we are JOIN_NOLOCK we're already committing a transaction and
471 * waiting on this guy, so we don't need to do the sb_start_intwrite
472 * because we're already holding a ref. We need this because we could
473 * have raced in and did an fsync() on a file which can kick a commit
474 * and then we deadlock with somebody doing a freeze.
476 * If we are ATTACH, it means we just want to catch the current
477 * transaction and commit it, so we needn't do sb_start_intwrite().
479 if (type & __TRANS_FREEZABLE)
480 sb_start_intwrite(root->fs_info->sb);
482 if (may_wait_transaction(root, type))
483 wait_current_trans(root);
486 ret = join_transaction(root, type);
488 wait_current_trans(root);
489 if (unlikely(type == TRANS_ATTACH))
492 } while (ret == -EBUSY);
495 /* We must get the transaction if we are JOIN_NOLOCK. */
496 BUG_ON(type == TRANS_JOIN_NOLOCK);
500 cur_trans = root->fs_info->running_transaction;
502 h->transid = cur_trans->transid;
503 h->transaction = cur_trans;
505 h->bytes_reserved = 0;
507 h->delayed_ref_updates = 0;
513 h->qgroup_reserved = 0;
514 h->delayed_ref_elem.seq = 0;
516 h->allocating_chunk = false;
517 h->reloc_reserved = false;
519 INIT_LIST_HEAD(&h->qgroup_ref_list);
520 INIT_LIST_HEAD(&h->new_bgs);
521 INIT_LIST_HEAD(&h->ordered);
524 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
525 may_wait_transaction(root, type)) {
526 current->journal_info = h;
527 btrfs_commit_transaction(h, root);
532 trace_btrfs_space_reservation(root->fs_info, "transaction",
533 h->transid, num_bytes, 1);
534 h->block_rsv = &root->fs_info->trans_block_rsv;
535 h->bytes_reserved = num_bytes;
536 h->reloc_reserved = reloc_reserved;
538 h->qgroup_reserved = qgroup_reserved;
541 btrfs_record_root_in_trans(h, root);
543 if (!current->journal_info && type != TRANS_USERSPACE)
544 current->journal_info = h;
548 if (type & __TRANS_FREEZABLE)
549 sb_end_intwrite(root->fs_info->sb);
550 kmem_cache_free(btrfs_trans_handle_cachep, h);
553 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
557 btrfs_qgroup_free(root, qgroup_reserved);
561 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
564 return start_transaction(root, num_items, TRANS_START,
565 BTRFS_RESERVE_FLUSH_ALL);
568 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
569 struct btrfs_root *root, int num_items)
571 return start_transaction(root, num_items, TRANS_START,
572 BTRFS_RESERVE_FLUSH_LIMIT);
575 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
577 return start_transaction(root, 0, TRANS_JOIN, 0);
580 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
582 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
585 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
587 return start_transaction(root, 0, TRANS_USERSPACE, 0);
591 * btrfs_attach_transaction() - catch the running transaction
593 * It is used when we want to commit the current the transaction, but
594 * don't want to start a new one.
596 * Note: If this function return -ENOENT, it just means there is no
597 * running transaction. But it is possible that the inactive transaction
598 * is still in the memory, not fully on disk. If you hope there is no
599 * inactive transaction in the fs when -ENOENT is returned, you should
601 * btrfs_attach_transaction_barrier()
603 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
605 return start_transaction(root, 0, TRANS_ATTACH, 0);
609 * btrfs_attach_transaction_barrier() - catch the running transaction
611 * It is similar to the above function, the differentia is this one
612 * will wait for all the inactive transactions until they fully
615 struct btrfs_trans_handle *
616 btrfs_attach_transaction_barrier(struct btrfs_root *root)
618 struct btrfs_trans_handle *trans;
620 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
621 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
622 btrfs_wait_for_commit(root, 0);
627 /* wait for a transaction commit to be fully complete */
628 static noinline void wait_for_commit(struct btrfs_root *root,
629 struct btrfs_transaction *commit)
631 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
634 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
636 struct btrfs_transaction *cur_trans = NULL, *t;
640 if (transid <= root->fs_info->last_trans_committed)
643 /* find specified transaction */
644 spin_lock(&root->fs_info->trans_lock);
645 list_for_each_entry(t, &root->fs_info->trans_list, list) {
646 if (t->transid == transid) {
648 atomic_inc(&cur_trans->use_count);
652 if (t->transid > transid) {
657 spin_unlock(&root->fs_info->trans_lock);
660 * The specified transaction doesn't exist, or we
661 * raced with btrfs_commit_transaction
664 if (transid > root->fs_info->last_trans_committed)
669 /* find newest transaction that is committing | committed */
670 spin_lock(&root->fs_info->trans_lock);
671 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
673 if (t->state >= TRANS_STATE_COMMIT_START) {
674 if (t->state == TRANS_STATE_COMPLETED)
677 atomic_inc(&cur_trans->use_count);
681 spin_unlock(&root->fs_info->trans_lock);
683 goto out; /* nothing committing|committed */
686 wait_for_commit(root, cur_trans);
687 btrfs_put_transaction(cur_trans);
692 void btrfs_throttle(struct btrfs_root *root)
694 if (!atomic_read(&root->fs_info->open_ioctl_trans))
695 wait_current_trans(root);
698 static int should_end_transaction(struct btrfs_trans_handle *trans,
699 struct btrfs_root *root)
701 if (root->fs_info->global_block_rsv.space_info->full &&
702 btrfs_check_space_for_delayed_refs(trans, root))
705 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
708 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
709 struct btrfs_root *root)
711 struct btrfs_transaction *cur_trans = trans->transaction;
716 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
717 cur_trans->delayed_refs.flushing)
720 updates = trans->delayed_ref_updates;
721 trans->delayed_ref_updates = 0;
723 err = btrfs_run_delayed_refs(trans, root, updates);
724 if (err) /* Error code will also eval true */
728 return should_end_transaction(trans, root);
731 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
732 struct btrfs_root *root, int throttle)
734 struct btrfs_transaction *cur_trans = trans->transaction;
735 struct btrfs_fs_info *info = root->fs_info;
736 unsigned long cur = trans->delayed_ref_updates;
737 int lock = (trans->type != TRANS_JOIN_NOLOCK);
739 int must_run_delayed_refs = 0;
741 if (trans->use_count > 1) {
743 trans->block_rsv = trans->orig_rsv;
747 btrfs_trans_release_metadata(trans, root);
748 trans->block_rsv = NULL;
750 if (!list_empty(&trans->new_bgs))
751 btrfs_create_pending_block_groups(trans, root);
753 if (!list_empty(&trans->ordered)) {
754 spin_lock(&info->trans_lock);
755 list_splice(&trans->ordered, &cur_trans->pending_ordered);
756 spin_unlock(&info->trans_lock);
759 trans->delayed_ref_updates = 0;
761 must_run_delayed_refs =
762 btrfs_should_throttle_delayed_refs(trans, root);
763 cur = max_t(unsigned long, cur, 32);
766 * don't make the caller wait if they are from a NOLOCK
767 * or ATTACH transaction, it will deadlock with commit
769 if (must_run_delayed_refs == 1 &&
770 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
771 must_run_delayed_refs = 2;
774 if (trans->qgroup_reserved) {
776 * the same root has to be passed here between start_transaction
777 * and end_transaction. Subvolume quota depends on this.
779 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
780 trans->qgroup_reserved = 0;
783 btrfs_trans_release_metadata(trans, root);
784 trans->block_rsv = NULL;
786 if (!list_empty(&trans->new_bgs))
787 btrfs_create_pending_block_groups(trans, root);
789 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
790 should_end_transaction(trans, root) &&
791 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
792 spin_lock(&info->trans_lock);
793 if (cur_trans->state == TRANS_STATE_RUNNING)
794 cur_trans->state = TRANS_STATE_BLOCKED;
795 spin_unlock(&info->trans_lock);
798 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
800 return btrfs_commit_transaction(trans, root);
802 wake_up_process(info->transaction_kthread);
805 if (trans->type & __TRANS_FREEZABLE)
806 sb_end_intwrite(root->fs_info->sb);
808 WARN_ON(cur_trans != info->running_transaction);
809 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
810 atomic_dec(&cur_trans->num_writers);
811 extwriter_counter_dec(cur_trans, trans->type);
814 if (waitqueue_active(&cur_trans->writer_wait))
815 wake_up(&cur_trans->writer_wait);
816 btrfs_put_transaction(cur_trans);
818 if (current->journal_info == trans)
819 current->journal_info = NULL;
822 btrfs_run_delayed_iputs(root);
824 if (trans->aborted ||
825 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
826 wake_up_process(info->transaction_kthread);
829 assert_qgroups_uptodate(trans);
831 kmem_cache_free(btrfs_trans_handle_cachep, trans);
832 if (must_run_delayed_refs) {
833 btrfs_async_run_delayed_refs(root, cur,
834 must_run_delayed_refs == 1);
839 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
840 struct btrfs_root *root)
842 return __btrfs_end_transaction(trans, root, 0);
845 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
846 struct btrfs_root *root)
848 return __btrfs_end_transaction(trans, root, 1);
852 * when btree blocks are allocated, they have some corresponding bits set for
853 * them in one of two extent_io trees. This is used to make sure all of
854 * those extents are sent to disk but does not wait on them
856 int btrfs_write_marked_extents(struct btrfs_root *root,
857 struct extent_io_tree *dirty_pages, int mark)
861 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
862 struct extent_state *cached_state = NULL;
866 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
867 mark, &cached_state)) {
868 bool wait_writeback = false;
870 err = convert_extent_bit(dirty_pages, start, end,
872 mark, &cached_state, GFP_NOFS);
874 * convert_extent_bit can return -ENOMEM, which is most of the
875 * time a temporary error. So when it happens, ignore the error
876 * and wait for writeback of this range to finish - because we
877 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
878 * to btrfs_wait_marked_extents() would not know that writeback
879 * for this range started and therefore wouldn't wait for it to
880 * finish - we don't want to commit a superblock that points to
881 * btree nodes/leafs for which writeback hasn't finished yet
882 * (and without errors).
883 * We cleanup any entries left in the io tree when committing
884 * the transaction (through clear_btree_io_tree()).
886 if (err == -ENOMEM) {
888 wait_writeback = true;
891 err = filemap_fdatawrite_range(mapping, start, end);
894 else if (wait_writeback)
895 werr = filemap_fdatawait_range(mapping, start, end);
896 free_extent_state(cached_state);
905 * when btree blocks are allocated, they have some corresponding bits set for
906 * them in one of two extent_io trees. This is used to make sure all of
907 * those extents are on disk for transaction or log commit. We wait
908 * on all the pages and clear them from the dirty pages state tree
910 int btrfs_wait_marked_extents(struct btrfs_root *root,
911 struct extent_io_tree *dirty_pages, int mark)
915 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
916 struct extent_state *cached_state = NULL;
919 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
922 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
923 EXTENT_NEED_WAIT, &cached_state)) {
925 * Ignore -ENOMEM errors returned by clear_extent_bit().
926 * When committing the transaction, we'll remove any entries
927 * left in the io tree. For a log commit, we don't remove them
928 * after committing the log because the tree can be accessed
929 * concurrently - we do it only at transaction commit time when
930 * it's safe to do it (through clear_btree_io_tree()).
932 err = clear_extent_bit(dirty_pages, start, end,
934 0, 0, &cached_state, GFP_NOFS);
938 err = filemap_fdatawait_range(mapping, start, end);
941 free_extent_state(cached_state);
949 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
950 if ((mark & EXTENT_DIRTY) &&
951 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
952 &btree_ino->runtime_flags))
955 if ((mark & EXTENT_NEW) &&
956 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
957 &btree_ino->runtime_flags))
960 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
961 &btree_ino->runtime_flags))
972 * when btree blocks are allocated, they have some corresponding bits set for
973 * them in one of two extent_io trees. This is used to make sure all of
974 * those extents are on disk for transaction or log commit
976 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
977 struct extent_io_tree *dirty_pages, int mark)
981 struct blk_plug plug;
983 blk_start_plug(&plug);
984 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
985 blk_finish_plug(&plug);
986 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
995 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
996 struct btrfs_root *root)
1000 ret = btrfs_write_and_wait_marked_extents(root,
1001 &trans->transaction->dirty_pages,
1003 clear_btree_io_tree(&trans->transaction->dirty_pages);
1009 * this is used to update the root pointer in the tree of tree roots.
1011 * But, in the case of the extent allocation tree, updating the root
1012 * pointer may allocate blocks which may change the root of the extent
1015 * So, this loops and repeats and makes sure the cowonly root didn't
1016 * change while the root pointer was being updated in the metadata.
1018 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1019 struct btrfs_root *root)
1022 u64 old_root_bytenr;
1024 struct btrfs_root *tree_root = root->fs_info->tree_root;
1025 bool extent_root = (root->objectid == BTRFS_EXTENT_TREE_OBJECTID);
1027 old_root_used = btrfs_root_used(&root->root_item);
1028 btrfs_write_dirty_block_groups(trans, root);
1031 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1032 if (old_root_bytenr == root->node->start &&
1033 old_root_used == btrfs_root_used(&root->root_item) &&
1035 list_empty(&trans->transaction->dirty_bgs)))
1038 btrfs_set_root_node(&root->root_item, root->node);
1039 ret = btrfs_update_root(trans, tree_root,
1045 old_root_used = btrfs_root_used(&root->root_item);
1047 ret = btrfs_write_dirty_block_groups(trans, root);
1051 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1054 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1063 * update all the cowonly tree roots on disk
1065 * The error handling in this function may not be obvious. Any of the
1066 * failures will cause the file system to go offline. We still need
1067 * to clean up the delayed refs.
1069 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1070 struct btrfs_root *root)
1072 struct btrfs_fs_info *fs_info = root->fs_info;
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 /* run_qgroups might have added some more refs */
1101 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1105 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1106 next = fs_info->dirty_cowonly_roots.next;
1107 list_del_init(next);
1108 root = list_entry(next, struct btrfs_root, dirty_list);
1109 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1111 if (root != fs_info->extent_root)
1112 list_add_tail(&root->dirty_list,
1113 &trans->transaction->switch_commits);
1114 ret = update_cowonly_root(trans, root);
1119 list_add_tail(&fs_info->extent_root->dirty_list,
1120 &trans->transaction->switch_commits);
1121 btrfs_after_dev_replace_commit(fs_info);
1127 * dead roots are old snapshots that need to be deleted. This allocates
1128 * a dirty root struct and adds it into the list of dead roots that need to
1131 void btrfs_add_dead_root(struct btrfs_root *root)
1133 spin_lock(&root->fs_info->trans_lock);
1134 if (list_empty(&root->root_list))
1135 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1136 spin_unlock(&root->fs_info->trans_lock);
1140 * update all the cowonly tree roots on disk
1142 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1143 struct btrfs_root *root)
1145 struct btrfs_root *gang[8];
1146 struct btrfs_fs_info *fs_info = root->fs_info;
1151 spin_lock(&fs_info->fs_roots_radix_lock);
1153 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1156 BTRFS_ROOT_TRANS_TAG);
1159 for (i = 0; i < ret; i++) {
1161 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1162 (unsigned long)root->root_key.objectid,
1163 BTRFS_ROOT_TRANS_TAG);
1164 spin_unlock(&fs_info->fs_roots_radix_lock);
1166 btrfs_free_log(trans, root);
1167 btrfs_update_reloc_root(trans, root);
1168 btrfs_orphan_commit_root(trans, root);
1170 btrfs_save_ino_cache(root, trans);
1172 /* see comments in should_cow_block() */
1173 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1174 smp_mb__after_atomic();
1176 if (root->commit_root != root->node) {
1177 list_add_tail(&root->dirty_list,
1178 &trans->transaction->switch_commits);
1179 btrfs_set_root_node(&root->root_item,
1183 err = btrfs_update_root(trans, fs_info->tree_root,
1186 spin_lock(&fs_info->fs_roots_radix_lock);
1191 spin_unlock(&fs_info->fs_roots_radix_lock);
1196 * defrag a given btree.
1197 * Every leaf in the btree is read and defragged.
1199 int btrfs_defrag_root(struct btrfs_root *root)
1201 struct btrfs_fs_info *info = root->fs_info;
1202 struct btrfs_trans_handle *trans;
1205 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1209 trans = btrfs_start_transaction(root, 0);
1211 return PTR_ERR(trans);
1213 ret = btrfs_defrag_leaves(trans, root);
1215 btrfs_end_transaction(trans, root);
1216 btrfs_btree_balance_dirty(info->tree_root);
1219 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1222 if (btrfs_defrag_cancelled(root->fs_info)) {
1223 pr_debug("BTRFS: defrag_root cancelled\n");
1228 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1233 * new snapshots need to be created at a very specific time in the
1234 * transaction commit. This does the actual creation.
1237 * If the error which may affect the commitment of the current transaction
1238 * happens, we should return the error number. If the error which just affect
1239 * the creation of the pending snapshots, just return 0.
1241 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1242 struct btrfs_fs_info *fs_info,
1243 struct btrfs_pending_snapshot *pending)
1245 struct btrfs_key key;
1246 struct btrfs_root_item *new_root_item;
1247 struct btrfs_root *tree_root = fs_info->tree_root;
1248 struct btrfs_root *root = pending->root;
1249 struct btrfs_root *parent_root;
1250 struct btrfs_block_rsv *rsv;
1251 struct inode *parent_inode;
1252 struct btrfs_path *path;
1253 struct btrfs_dir_item *dir_item;
1254 struct dentry *dentry;
1255 struct extent_buffer *tmp;
1256 struct extent_buffer *old;
1257 struct timespec cur_time = CURRENT_TIME;
1265 path = btrfs_alloc_path();
1267 pending->error = -ENOMEM;
1271 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1272 if (!new_root_item) {
1273 pending->error = -ENOMEM;
1274 goto root_item_alloc_fail;
1277 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1279 goto no_free_objectid;
1281 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1283 if (to_reserve > 0) {
1284 pending->error = btrfs_block_rsv_add(root,
1285 &pending->block_rsv,
1287 BTRFS_RESERVE_NO_FLUSH);
1289 goto no_free_objectid;
1292 key.objectid = objectid;
1293 key.offset = (u64)-1;
1294 key.type = BTRFS_ROOT_ITEM_KEY;
1296 rsv = trans->block_rsv;
1297 trans->block_rsv = &pending->block_rsv;
1298 trans->bytes_reserved = trans->block_rsv->reserved;
1300 dentry = pending->dentry;
1301 parent_inode = pending->dir;
1302 parent_root = BTRFS_I(parent_inode)->root;
1303 record_root_in_trans(trans, parent_root);
1306 * insert the directory item
1308 ret = btrfs_set_inode_index(parent_inode, &index);
1309 BUG_ON(ret); /* -ENOMEM */
1311 /* check if there is a file/dir which has the same name. */
1312 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1313 btrfs_ino(parent_inode),
1314 dentry->d_name.name,
1315 dentry->d_name.len, 0);
1316 if (dir_item != NULL && !IS_ERR(dir_item)) {
1317 pending->error = -EEXIST;
1318 goto dir_item_existed;
1319 } else if (IS_ERR(dir_item)) {
1320 ret = PTR_ERR(dir_item);
1321 btrfs_abort_transaction(trans, root, ret);
1324 btrfs_release_path(path);
1327 * pull in the delayed directory update
1328 * and the delayed inode item
1329 * otherwise we corrupt the FS during
1332 ret = btrfs_run_delayed_items(trans, root);
1333 if (ret) { /* Transaction aborted */
1334 btrfs_abort_transaction(trans, root, ret);
1338 record_root_in_trans(trans, root);
1339 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1340 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1341 btrfs_check_and_init_root_item(new_root_item);
1343 root_flags = btrfs_root_flags(new_root_item);
1344 if (pending->readonly)
1345 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1347 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1348 btrfs_set_root_flags(new_root_item, root_flags);
1350 btrfs_set_root_generation_v2(new_root_item,
1352 uuid_le_gen(&new_uuid);
1353 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1354 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1356 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1357 memset(new_root_item->received_uuid, 0,
1358 sizeof(new_root_item->received_uuid));
1359 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1360 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1361 btrfs_set_root_stransid(new_root_item, 0);
1362 btrfs_set_root_rtransid(new_root_item, 0);
1364 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1365 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1366 btrfs_set_root_otransid(new_root_item, trans->transid);
1368 old = btrfs_lock_root_node(root);
1369 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1371 btrfs_tree_unlock(old);
1372 free_extent_buffer(old);
1373 btrfs_abort_transaction(trans, root, ret);
1377 btrfs_set_lock_blocking(old);
1379 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1380 /* clean up in any case */
1381 btrfs_tree_unlock(old);
1382 free_extent_buffer(old);
1384 btrfs_abort_transaction(trans, root, ret);
1389 * We need to flush delayed refs in order to make sure all of our quota
1390 * operations have been done before we call btrfs_qgroup_inherit.
1392 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1394 btrfs_abort_transaction(trans, root, ret);
1398 ret = btrfs_qgroup_inherit(trans, fs_info,
1399 root->root_key.objectid,
1400 objectid, pending->inherit);
1402 btrfs_abort_transaction(trans, root, ret);
1406 /* see comments in should_cow_block() */
1407 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1410 btrfs_set_root_node(new_root_item, tmp);
1411 /* record when the snapshot was created in key.offset */
1412 key.offset = trans->transid;
1413 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1414 btrfs_tree_unlock(tmp);
1415 free_extent_buffer(tmp);
1417 btrfs_abort_transaction(trans, root, ret);
1422 * insert root back/forward references
1424 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1425 parent_root->root_key.objectid,
1426 btrfs_ino(parent_inode), index,
1427 dentry->d_name.name, dentry->d_name.len);
1429 btrfs_abort_transaction(trans, root, ret);
1433 key.offset = (u64)-1;
1434 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1435 if (IS_ERR(pending->snap)) {
1436 ret = PTR_ERR(pending->snap);
1437 btrfs_abort_transaction(trans, root, ret);
1441 ret = btrfs_reloc_post_snapshot(trans, pending);
1443 btrfs_abort_transaction(trans, root, ret);
1447 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1449 btrfs_abort_transaction(trans, root, ret);
1453 ret = btrfs_insert_dir_item(trans, parent_root,
1454 dentry->d_name.name, dentry->d_name.len,
1456 BTRFS_FT_DIR, index);
1457 /* We have check then name at the beginning, so it is impossible. */
1458 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1460 btrfs_abort_transaction(trans, root, ret);
1464 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1465 dentry->d_name.len * 2);
1466 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1467 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1469 btrfs_abort_transaction(trans, root, ret);
1472 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1473 BTRFS_UUID_KEY_SUBVOL, objectid);
1475 btrfs_abort_transaction(trans, root, ret);
1478 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1479 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1480 new_root_item->received_uuid,
1481 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1483 if (ret && ret != -EEXIST) {
1484 btrfs_abort_transaction(trans, root, ret);
1489 pending->error = ret;
1491 trans->block_rsv = rsv;
1492 trans->bytes_reserved = 0;
1494 kfree(new_root_item);
1495 root_item_alloc_fail:
1496 btrfs_free_path(path);
1501 * create all the snapshots we've scheduled for creation
1503 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1504 struct btrfs_fs_info *fs_info)
1506 struct btrfs_pending_snapshot *pending, *next;
1507 struct list_head *head = &trans->transaction->pending_snapshots;
1510 list_for_each_entry_safe(pending, next, head, list) {
1511 list_del(&pending->list);
1512 ret = create_pending_snapshot(trans, fs_info, pending);
1519 static void update_super_roots(struct btrfs_root *root)
1521 struct btrfs_root_item *root_item;
1522 struct btrfs_super_block *super;
1524 super = root->fs_info->super_copy;
1526 root_item = &root->fs_info->chunk_root->root_item;
1527 super->chunk_root = root_item->bytenr;
1528 super->chunk_root_generation = root_item->generation;
1529 super->chunk_root_level = root_item->level;
1531 root_item = &root->fs_info->tree_root->root_item;
1532 super->root = root_item->bytenr;
1533 super->generation = root_item->generation;
1534 super->root_level = root_item->level;
1535 if (btrfs_test_opt(root, SPACE_CACHE))
1536 super->cache_generation = root_item->generation;
1537 if (root->fs_info->update_uuid_tree_gen)
1538 super->uuid_tree_generation = root_item->generation;
1541 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1543 struct btrfs_transaction *trans;
1546 spin_lock(&info->trans_lock);
1547 trans = info->running_transaction;
1549 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1550 spin_unlock(&info->trans_lock);
1554 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1556 struct btrfs_transaction *trans;
1559 spin_lock(&info->trans_lock);
1560 trans = info->running_transaction;
1562 ret = is_transaction_blocked(trans);
1563 spin_unlock(&info->trans_lock);
1568 * wait for the current transaction commit to start and block subsequent
1571 static void wait_current_trans_commit_start(struct btrfs_root *root,
1572 struct btrfs_transaction *trans)
1574 wait_event(root->fs_info->transaction_blocked_wait,
1575 trans->state >= TRANS_STATE_COMMIT_START ||
1580 * wait for the current transaction to start and then become unblocked.
1583 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1584 struct btrfs_transaction *trans)
1586 wait_event(root->fs_info->transaction_wait,
1587 trans->state >= TRANS_STATE_UNBLOCKED ||
1592 * commit transactions asynchronously. once btrfs_commit_transaction_async
1593 * returns, any subsequent transaction will not be allowed to join.
1595 struct btrfs_async_commit {
1596 struct btrfs_trans_handle *newtrans;
1597 struct btrfs_root *root;
1598 struct work_struct work;
1601 static void do_async_commit(struct work_struct *work)
1603 struct btrfs_async_commit *ac =
1604 container_of(work, struct btrfs_async_commit, work);
1607 * We've got freeze protection passed with the transaction.
1608 * Tell lockdep about it.
1610 if (ac->newtrans->type & __TRANS_FREEZABLE)
1612 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1615 current->journal_info = ac->newtrans;
1617 btrfs_commit_transaction(ac->newtrans, ac->root);
1621 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1622 struct btrfs_root *root,
1623 int wait_for_unblock)
1625 struct btrfs_async_commit *ac;
1626 struct btrfs_transaction *cur_trans;
1628 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1632 INIT_WORK(&ac->work, do_async_commit);
1634 ac->newtrans = btrfs_join_transaction(root);
1635 if (IS_ERR(ac->newtrans)) {
1636 int err = PTR_ERR(ac->newtrans);
1641 /* take transaction reference */
1642 cur_trans = trans->transaction;
1643 atomic_inc(&cur_trans->use_count);
1645 btrfs_end_transaction(trans, root);
1648 * Tell lockdep we've released the freeze rwsem, since the
1649 * async commit thread will be the one to unlock it.
1651 if (ac->newtrans->type & __TRANS_FREEZABLE)
1653 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1656 schedule_work(&ac->work);
1658 /* wait for transaction to start and unblock */
1659 if (wait_for_unblock)
1660 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1662 wait_current_trans_commit_start(root, cur_trans);
1664 if (current->journal_info == trans)
1665 current->journal_info = NULL;
1667 btrfs_put_transaction(cur_trans);
1672 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1673 struct btrfs_root *root, int err)
1675 struct btrfs_transaction *cur_trans = trans->transaction;
1678 WARN_ON(trans->use_count > 1);
1680 btrfs_abort_transaction(trans, root, err);
1682 spin_lock(&root->fs_info->trans_lock);
1685 * If the transaction is removed from the list, it means this
1686 * transaction has been committed successfully, so it is impossible
1687 * to call the cleanup function.
1689 BUG_ON(list_empty(&cur_trans->list));
1691 list_del_init(&cur_trans->list);
1692 if (cur_trans == root->fs_info->running_transaction) {
1693 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1694 spin_unlock(&root->fs_info->trans_lock);
1695 wait_event(cur_trans->writer_wait,
1696 atomic_read(&cur_trans->num_writers) == 1);
1698 spin_lock(&root->fs_info->trans_lock);
1700 spin_unlock(&root->fs_info->trans_lock);
1702 btrfs_cleanup_one_transaction(trans->transaction, root);
1704 spin_lock(&root->fs_info->trans_lock);
1705 if (cur_trans == root->fs_info->running_transaction)
1706 root->fs_info->running_transaction = NULL;
1707 spin_unlock(&root->fs_info->trans_lock);
1709 if (trans->type & __TRANS_FREEZABLE)
1710 sb_end_intwrite(root->fs_info->sb);
1711 btrfs_put_transaction(cur_trans);
1712 btrfs_put_transaction(cur_trans);
1714 trace_btrfs_transaction_commit(root);
1716 if (current->journal_info == trans)
1717 current->journal_info = NULL;
1718 btrfs_scrub_cancel(root->fs_info);
1720 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1723 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1725 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1726 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1730 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1732 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1733 btrfs_wait_ordered_roots(fs_info, -1);
1737 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans,
1738 struct btrfs_fs_info *fs_info)
1740 struct btrfs_ordered_extent *ordered;
1742 spin_lock(&fs_info->trans_lock);
1743 while (!list_empty(&cur_trans->pending_ordered)) {
1744 ordered = list_first_entry(&cur_trans->pending_ordered,
1745 struct btrfs_ordered_extent,
1747 list_del_init(&ordered->trans_list);
1748 spin_unlock(&fs_info->trans_lock);
1750 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE,
1752 btrfs_put_ordered_extent(ordered);
1753 spin_lock(&fs_info->trans_lock);
1755 spin_unlock(&fs_info->trans_lock);
1758 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1759 struct btrfs_root *root)
1761 struct btrfs_transaction *cur_trans = trans->transaction;
1762 struct btrfs_transaction *prev_trans = NULL;
1763 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
1766 /* Stop the commit early if ->aborted is set */
1767 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1768 ret = cur_trans->aborted;
1769 btrfs_end_transaction(trans, root);
1773 /* make a pass through all the delayed refs we have so far
1774 * any runnings procs may add more while we are here
1776 ret = btrfs_run_delayed_refs(trans, root, 0);
1778 btrfs_end_transaction(trans, root);
1782 btrfs_trans_release_metadata(trans, root);
1783 trans->block_rsv = NULL;
1784 if (trans->qgroup_reserved) {
1785 btrfs_qgroup_free(root, trans->qgroup_reserved);
1786 trans->qgroup_reserved = 0;
1789 cur_trans = trans->transaction;
1792 * set the flushing flag so procs in this transaction have to
1793 * start sending their work down.
1795 cur_trans->delayed_refs.flushing = 1;
1798 if (!list_empty(&trans->new_bgs))
1799 btrfs_create_pending_block_groups(trans, root);
1801 ret = btrfs_run_delayed_refs(trans, root, 0);
1803 btrfs_end_transaction(trans, root);
1807 spin_lock(&root->fs_info->trans_lock);
1808 list_splice(&trans->ordered, &cur_trans->pending_ordered);
1809 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1810 spin_unlock(&root->fs_info->trans_lock);
1811 atomic_inc(&cur_trans->use_count);
1812 ret = btrfs_end_transaction(trans, root);
1814 wait_for_commit(root, cur_trans);
1816 btrfs_put_transaction(cur_trans);
1821 cur_trans->state = TRANS_STATE_COMMIT_START;
1822 wake_up(&root->fs_info->transaction_blocked_wait);
1824 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1825 prev_trans = list_entry(cur_trans->list.prev,
1826 struct btrfs_transaction, list);
1827 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1828 atomic_inc(&prev_trans->use_count);
1829 spin_unlock(&root->fs_info->trans_lock);
1831 wait_for_commit(root, prev_trans);
1833 btrfs_put_transaction(prev_trans);
1835 spin_unlock(&root->fs_info->trans_lock);
1838 spin_unlock(&root->fs_info->trans_lock);
1841 extwriter_counter_dec(cur_trans, trans->type);
1843 ret = btrfs_start_delalloc_flush(root->fs_info);
1845 goto cleanup_transaction;
1847 ret = btrfs_run_delayed_items(trans, root);
1849 goto cleanup_transaction;
1851 wait_event(cur_trans->writer_wait,
1852 extwriter_counter_read(cur_trans) == 0);
1854 /* some pending stuffs might be added after the previous flush. */
1855 ret = btrfs_run_delayed_items(trans, root);
1857 goto cleanup_transaction;
1859 btrfs_wait_delalloc_flush(root->fs_info);
1861 btrfs_wait_pending_ordered(cur_trans, root->fs_info);
1863 btrfs_scrub_pause(root);
1865 * Ok now we need to make sure to block out any other joins while we
1866 * commit the transaction. We could have started a join before setting
1867 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1869 spin_lock(&root->fs_info->trans_lock);
1870 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1871 spin_unlock(&root->fs_info->trans_lock);
1872 wait_event(cur_trans->writer_wait,
1873 atomic_read(&cur_trans->num_writers) == 1);
1875 /* ->aborted might be set after the previous check, so check it */
1876 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1877 ret = cur_trans->aborted;
1878 goto scrub_continue;
1881 * the reloc mutex makes sure that we stop
1882 * the balancing code from coming in and moving
1883 * extents around in the middle of the commit
1885 mutex_lock(&root->fs_info->reloc_mutex);
1888 * We needn't worry about the delayed items because we will
1889 * deal with them in create_pending_snapshot(), which is the
1890 * core function of the snapshot creation.
1892 ret = create_pending_snapshots(trans, root->fs_info);
1894 mutex_unlock(&root->fs_info->reloc_mutex);
1895 goto scrub_continue;
1899 * We insert the dir indexes of the snapshots and update the inode
1900 * of the snapshots' parents after the snapshot creation, so there
1901 * are some delayed items which are not dealt with. Now deal with
1904 * We needn't worry that this operation will corrupt the snapshots,
1905 * because all the tree which are snapshoted will be forced to COW
1906 * the nodes and leaves.
1908 ret = btrfs_run_delayed_items(trans, root);
1910 mutex_unlock(&root->fs_info->reloc_mutex);
1911 goto scrub_continue;
1914 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1916 mutex_unlock(&root->fs_info->reloc_mutex);
1917 goto scrub_continue;
1921 * make sure none of the code above managed to slip in a
1924 btrfs_assert_delayed_root_empty(root);
1926 WARN_ON(cur_trans != trans->transaction);
1928 /* btrfs_commit_tree_roots is responsible for getting the
1929 * various roots consistent with each other. Every pointer
1930 * in the tree of tree roots has to point to the most up to date
1931 * root for every subvolume and other tree. So, we have to keep
1932 * the tree logging code from jumping in and changing any
1935 * At this point in the commit, there can't be any tree-log
1936 * writers, but a little lower down we drop the trans mutex
1937 * and let new people in. By holding the tree_log_mutex
1938 * from now until after the super is written, we avoid races
1939 * with the tree-log code.
1941 mutex_lock(&root->fs_info->tree_log_mutex);
1943 ret = commit_fs_roots(trans, root);
1945 mutex_unlock(&root->fs_info->tree_log_mutex);
1946 mutex_unlock(&root->fs_info->reloc_mutex);
1947 goto scrub_continue;
1951 * Since the transaction is done, we can apply the pending changes
1952 * before the next transaction.
1954 btrfs_apply_pending_changes(root->fs_info);
1956 /* commit_fs_roots gets rid of all the tree log roots, it is now
1957 * safe to free the root of tree log roots
1959 btrfs_free_log_root_tree(trans, root->fs_info);
1961 ret = commit_cowonly_roots(trans, root);
1963 mutex_unlock(&root->fs_info->tree_log_mutex);
1964 mutex_unlock(&root->fs_info->reloc_mutex);
1965 goto scrub_continue;
1969 * The tasks which save the space cache and inode cache may also
1970 * update ->aborted, check it.
1972 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1973 ret = cur_trans->aborted;
1974 mutex_unlock(&root->fs_info->tree_log_mutex);
1975 mutex_unlock(&root->fs_info->reloc_mutex);
1976 goto scrub_continue;
1979 btrfs_prepare_extent_commit(trans, root);
1981 cur_trans = root->fs_info->running_transaction;
1983 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1984 root->fs_info->tree_root->node);
1985 list_add_tail(&root->fs_info->tree_root->dirty_list,
1986 &cur_trans->switch_commits);
1988 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1989 root->fs_info->chunk_root->node);
1990 list_add_tail(&root->fs_info->chunk_root->dirty_list,
1991 &cur_trans->switch_commits);
1993 switch_commit_roots(cur_trans, root->fs_info);
1995 assert_qgroups_uptodate(trans);
1996 ASSERT(list_empty(&cur_trans->dirty_bgs));
1997 update_super_roots(root);
1999 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2000 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
2001 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2002 sizeof(*root->fs_info->super_copy));
2004 btrfs_update_commit_device_size(root->fs_info);
2005 btrfs_update_commit_device_bytes_used(root, cur_trans);
2007 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
2008 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
2010 spin_lock(&root->fs_info->trans_lock);
2011 cur_trans->state = TRANS_STATE_UNBLOCKED;
2012 root->fs_info->running_transaction = NULL;
2013 spin_unlock(&root->fs_info->trans_lock);
2014 mutex_unlock(&root->fs_info->reloc_mutex);
2016 wake_up(&root->fs_info->transaction_wait);
2018 ret = btrfs_write_and_wait_transaction(trans, root);
2020 btrfs_error(root->fs_info, ret,
2021 "Error while writing out transaction");
2022 mutex_unlock(&root->fs_info->tree_log_mutex);
2023 goto scrub_continue;
2026 ret = write_ctree_super(trans, root, 0);
2028 mutex_unlock(&root->fs_info->tree_log_mutex);
2029 goto scrub_continue;
2033 * the super is written, we can safely allow the tree-loggers
2034 * to go about their business
2036 mutex_unlock(&root->fs_info->tree_log_mutex);
2038 btrfs_finish_extent_commit(trans, root);
2040 root->fs_info->last_trans_committed = cur_trans->transid;
2042 * We needn't acquire the lock here because there is no other task
2043 * which can change it.
2045 cur_trans->state = TRANS_STATE_COMPLETED;
2046 wake_up(&cur_trans->commit_wait);
2048 spin_lock(&root->fs_info->trans_lock);
2049 list_del_init(&cur_trans->list);
2050 spin_unlock(&root->fs_info->trans_lock);
2052 btrfs_put_transaction(cur_trans);
2053 btrfs_put_transaction(cur_trans);
2055 if (trans->type & __TRANS_FREEZABLE)
2056 sb_end_intwrite(root->fs_info->sb);
2058 trace_btrfs_transaction_commit(root);
2060 btrfs_scrub_continue(root);
2062 if (current->journal_info == trans)
2063 current->journal_info = NULL;
2065 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2067 if (current != root->fs_info->transaction_kthread)
2068 btrfs_run_delayed_iputs(root);
2073 btrfs_scrub_continue(root);
2074 cleanup_transaction:
2075 btrfs_trans_release_metadata(trans, root);
2076 trans->block_rsv = NULL;
2077 if (trans->qgroup_reserved) {
2078 btrfs_qgroup_free(root, trans->qgroup_reserved);
2079 trans->qgroup_reserved = 0;
2081 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
2082 if (current->journal_info == trans)
2083 current->journal_info = NULL;
2084 cleanup_transaction(trans, root, ret);
2090 * return < 0 if error
2091 * 0 if there are no more dead_roots at the time of call
2092 * 1 there are more to be processed, call me again
2094 * The return value indicates there are certainly more snapshots to delete, but
2095 * if there comes a new one during processing, it may return 0. We don't mind,
2096 * because btrfs_commit_super will poke cleaner thread and it will process it a
2097 * few seconds later.
2099 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2102 struct btrfs_fs_info *fs_info = root->fs_info;
2104 spin_lock(&fs_info->trans_lock);
2105 if (list_empty(&fs_info->dead_roots)) {
2106 spin_unlock(&fs_info->trans_lock);
2109 root = list_first_entry(&fs_info->dead_roots,
2110 struct btrfs_root, root_list);
2111 list_del_init(&root->root_list);
2112 spin_unlock(&fs_info->trans_lock);
2114 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2116 btrfs_kill_all_delayed_nodes(root);
2118 if (btrfs_header_backref_rev(root->node) <
2119 BTRFS_MIXED_BACKREF_REV)
2120 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2122 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2124 return (ret < 0) ? 0 : 1;
2127 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2132 prev = cmpxchg(&fs_info->pending_changes, 0, 0);
2136 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2138 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2141 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2143 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2146 bit = 1 << BTRFS_PENDING_COMMIT;
2148 btrfs_debug(fs_info, "pending commit done");
2153 "unknown pending changes left 0x%lx, ignoring", prev);