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->have_free_bgs = 0;
224 cur_trans->start_time = get_seconds();
226 cur_trans->delayed_refs.href_root = RB_ROOT;
227 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
228 cur_trans->delayed_refs.num_heads_ready = 0;
229 cur_trans->delayed_refs.num_heads = 0;
230 cur_trans->delayed_refs.flushing = 0;
231 cur_trans->delayed_refs.run_delayed_start = 0;
234 * although the tree mod log is per file system and not per transaction,
235 * the log must never go across transaction boundaries.
238 if (!list_empty(&fs_info->tree_mod_seq_list))
239 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
240 "creating a fresh transaction\n");
241 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
242 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
243 "creating a fresh transaction\n");
244 atomic64_set(&fs_info->tree_mod_seq, 0);
246 spin_lock_init(&cur_trans->delayed_refs.lock);
248 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
249 INIT_LIST_HEAD(&cur_trans->pending_chunks);
250 INIT_LIST_HEAD(&cur_trans->switch_commits);
251 INIT_LIST_HEAD(&cur_trans->pending_ordered);
252 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
253 spin_lock_init(&cur_trans->dirty_bgs_lock);
254 list_add_tail(&cur_trans->list, &fs_info->trans_list);
255 extent_io_tree_init(&cur_trans->dirty_pages,
256 fs_info->btree_inode->i_mapping);
257 fs_info->generation++;
258 cur_trans->transid = fs_info->generation;
259 fs_info->running_transaction = cur_trans;
260 cur_trans->aborted = 0;
261 spin_unlock(&fs_info->trans_lock);
267 * this does all the record keeping required to make sure that a reference
268 * counted root is properly recorded in a given transaction. This is required
269 * to make sure the old root from before we joined the transaction is deleted
270 * when the transaction commits
272 static int record_root_in_trans(struct btrfs_trans_handle *trans,
273 struct btrfs_root *root)
275 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
276 root->last_trans < trans->transid) {
277 WARN_ON(root == root->fs_info->extent_root);
278 WARN_ON(root->commit_root != root->node);
281 * see below for IN_TRANS_SETUP usage rules
282 * we have the reloc mutex held now, so there
283 * is only one writer in this function
285 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
287 /* make sure readers find IN_TRANS_SETUP before
288 * they find our root->last_trans update
292 spin_lock(&root->fs_info->fs_roots_radix_lock);
293 if (root->last_trans == trans->transid) {
294 spin_unlock(&root->fs_info->fs_roots_radix_lock);
297 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
298 (unsigned long)root->root_key.objectid,
299 BTRFS_ROOT_TRANS_TAG);
300 spin_unlock(&root->fs_info->fs_roots_radix_lock);
301 root->last_trans = trans->transid;
303 /* this is pretty tricky. We don't want to
304 * take the relocation lock in btrfs_record_root_in_trans
305 * unless we're really doing the first setup for this root in
308 * Normally we'd use root->last_trans as a flag to decide
309 * if we want to take the expensive mutex.
311 * But, we have to set root->last_trans before we
312 * init the relocation root, otherwise, we trip over warnings
313 * in ctree.c. The solution used here is to flag ourselves
314 * with root IN_TRANS_SETUP. When this is 1, we're still
315 * fixing up the reloc trees and everyone must wait.
317 * When this is zero, they can trust root->last_trans and fly
318 * through btrfs_record_root_in_trans without having to take the
319 * lock. smp_wmb() makes sure that all the writes above are
320 * done before we pop in the zero below
322 btrfs_init_reloc_root(trans, root);
323 smp_mb__before_atomic();
324 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
330 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
331 struct btrfs_root *root)
333 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
337 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
341 if (root->last_trans == trans->transid &&
342 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
345 mutex_lock(&root->fs_info->reloc_mutex);
346 record_root_in_trans(trans, root);
347 mutex_unlock(&root->fs_info->reloc_mutex);
352 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
354 return (trans->state >= TRANS_STATE_BLOCKED &&
355 trans->state < TRANS_STATE_UNBLOCKED &&
359 /* wait for commit against the current transaction to become unblocked
360 * when this is done, it is safe to start a new transaction, but the current
361 * transaction might not be fully on disk.
363 static void wait_current_trans(struct btrfs_root *root)
365 struct btrfs_transaction *cur_trans;
367 spin_lock(&root->fs_info->trans_lock);
368 cur_trans = root->fs_info->running_transaction;
369 if (cur_trans && is_transaction_blocked(cur_trans)) {
370 atomic_inc(&cur_trans->use_count);
371 spin_unlock(&root->fs_info->trans_lock);
373 wait_event(root->fs_info->transaction_wait,
374 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
376 btrfs_put_transaction(cur_trans);
378 spin_unlock(&root->fs_info->trans_lock);
382 static int may_wait_transaction(struct btrfs_root *root, int type)
384 if (root->fs_info->log_root_recovering)
387 if (type == TRANS_USERSPACE)
390 if (type == TRANS_START &&
391 !atomic_read(&root->fs_info->open_ioctl_trans))
397 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
399 if (!root->fs_info->reloc_ctl ||
400 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
401 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
408 static struct btrfs_trans_handle *
409 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
410 enum btrfs_reserve_flush_enum flush)
412 struct btrfs_trans_handle *h;
413 struct btrfs_transaction *cur_trans;
415 u64 qgroup_reserved = 0;
416 bool reloc_reserved = false;
419 /* Send isn't supposed to start transactions. */
420 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
422 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
423 return ERR_PTR(-EROFS);
425 if (current->journal_info) {
426 WARN_ON(type & TRANS_EXTWRITERS);
427 h = current->journal_info;
429 WARN_ON(h->use_count > 2);
430 h->orig_rsv = h->block_rsv;
436 * Do the reservation before we join the transaction so we can do all
437 * the appropriate flushing if need be.
439 if (num_items > 0 && root != root->fs_info->chunk_root) {
440 if (root->fs_info->quota_enabled &&
441 is_fstree(root->root_key.objectid)) {
442 qgroup_reserved = num_items * root->nodesize;
443 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
448 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
450 * Do the reservation for the relocation root creation
452 if (need_reserve_reloc_root(root)) {
453 num_bytes += root->nodesize;
454 reloc_reserved = true;
457 ret = btrfs_block_rsv_add(root,
458 &root->fs_info->trans_block_rsv,
464 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
471 * If we are JOIN_NOLOCK we're already committing a transaction and
472 * waiting on this guy, so we don't need to do the sb_start_intwrite
473 * because we're already holding a ref. We need this because we could
474 * have raced in and did an fsync() on a file which can kick a commit
475 * and then we deadlock with somebody doing a freeze.
477 * If we are ATTACH, it means we just want to catch the current
478 * transaction and commit it, so we needn't do sb_start_intwrite().
480 if (type & __TRANS_FREEZABLE)
481 sb_start_intwrite(root->fs_info->sb);
483 if (may_wait_transaction(root, type))
484 wait_current_trans(root);
487 ret = join_transaction(root, type);
489 wait_current_trans(root);
490 if (unlikely(type == TRANS_ATTACH))
493 } while (ret == -EBUSY);
496 /* We must get the transaction if we are JOIN_NOLOCK. */
497 BUG_ON(type == TRANS_JOIN_NOLOCK);
501 cur_trans = root->fs_info->running_transaction;
503 h->transid = cur_trans->transid;
504 h->transaction = cur_trans;
506 h->bytes_reserved = 0;
508 h->delayed_ref_updates = 0;
514 h->qgroup_reserved = 0;
515 h->delayed_ref_elem.seq = 0;
517 h->allocating_chunk = false;
518 h->reloc_reserved = false;
520 INIT_LIST_HEAD(&h->qgroup_ref_list);
521 INIT_LIST_HEAD(&h->new_bgs);
522 INIT_LIST_HEAD(&h->ordered);
525 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
526 may_wait_transaction(root, type)) {
527 current->journal_info = h;
528 btrfs_commit_transaction(h, root);
533 trace_btrfs_space_reservation(root->fs_info, "transaction",
534 h->transid, num_bytes, 1);
535 h->block_rsv = &root->fs_info->trans_block_rsv;
536 h->bytes_reserved = num_bytes;
537 h->reloc_reserved = reloc_reserved;
539 h->qgroup_reserved = qgroup_reserved;
542 btrfs_record_root_in_trans(h, root);
544 if (!current->journal_info && type != TRANS_USERSPACE)
545 current->journal_info = h;
549 if (type & __TRANS_FREEZABLE)
550 sb_end_intwrite(root->fs_info->sb);
551 kmem_cache_free(btrfs_trans_handle_cachep, h);
554 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
558 btrfs_qgroup_free(root, qgroup_reserved);
562 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
565 return start_transaction(root, num_items, TRANS_START,
566 BTRFS_RESERVE_FLUSH_ALL);
569 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
570 struct btrfs_root *root, int num_items)
572 return start_transaction(root, num_items, TRANS_START,
573 BTRFS_RESERVE_FLUSH_LIMIT);
576 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
578 return start_transaction(root, 0, TRANS_JOIN, 0);
581 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
583 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
586 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
588 return start_transaction(root, 0, TRANS_USERSPACE, 0);
592 * btrfs_attach_transaction() - catch the running transaction
594 * It is used when we want to commit the current the transaction, but
595 * don't want to start a new one.
597 * Note: If this function return -ENOENT, it just means there is no
598 * running transaction. But it is possible that the inactive transaction
599 * is still in the memory, not fully on disk. If you hope there is no
600 * inactive transaction in the fs when -ENOENT is returned, you should
602 * btrfs_attach_transaction_barrier()
604 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
606 return start_transaction(root, 0, TRANS_ATTACH, 0);
610 * btrfs_attach_transaction_barrier() - catch the running transaction
612 * It is similar to the above function, the differentia is this one
613 * will wait for all the inactive transactions until they fully
616 struct btrfs_trans_handle *
617 btrfs_attach_transaction_barrier(struct btrfs_root *root)
619 struct btrfs_trans_handle *trans;
621 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
622 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
623 btrfs_wait_for_commit(root, 0);
628 /* wait for a transaction commit to be fully complete */
629 static noinline void wait_for_commit(struct btrfs_root *root,
630 struct btrfs_transaction *commit)
632 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
635 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
637 struct btrfs_transaction *cur_trans = NULL, *t;
641 if (transid <= root->fs_info->last_trans_committed)
644 /* find specified transaction */
645 spin_lock(&root->fs_info->trans_lock);
646 list_for_each_entry(t, &root->fs_info->trans_list, list) {
647 if (t->transid == transid) {
649 atomic_inc(&cur_trans->use_count);
653 if (t->transid > transid) {
658 spin_unlock(&root->fs_info->trans_lock);
661 * The specified transaction doesn't exist, or we
662 * raced with btrfs_commit_transaction
665 if (transid > root->fs_info->last_trans_committed)
670 /* find newest transaction that is committing | committed */
671 spin_lock(&root->fs_info->trans_lock);
672 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
674 if (t->state >= TRANS_STATE_COMMIT_START) {
675 if (t->state == TRANS_STATE_COMPLETED)
678 atomic_inc(&cur_trans->use_count);
682 spin_unlock(&root->fs_info->trans_lock);
684 goto out; /* nothing committing|committed */
687 wait_for_commit(root, cur_trans);
688 btrfs_put_transaction(cur_trans);
693 void btrfs_throttle(struct btrfs_root *root)
695 if (!atomic_read(&root->fs_info->open_ioctl_trans))
696 wait_current_trans(root);
699 static int should_end_transaction(struct btrfs_trans_handle *trans,
700 struct btrfs_root *root)
702 if (root->fs_info->global_block_rsv.space_info->full &&
703 btrfs_check_space_for_delayed_refs(trans, root))
706 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
709 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
710 struct btrfs_root *root)
712 struct btrfs_transaction *cur_trans = trans->transaction;
717 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
718 cur_trans->delayed_refs.flushing)
721 updates = trans->delayed_ref_updates;
722 trans->delayed_ref_updates = 0;
724 err = btrfs_run_delayed_refs(trans, root, updates);
725 if (err) /* Error code will also eval true */
729 return should_end_transaction(trans, root);
732 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
733 struct btrfs_root *root, int throttle)
735 struct btrfs_transaction *cur_trans = trans->transaction;
736 struct btrfs_fs_info *info = root->fs_info;
737 unsigned long cur = trans->delayed_ref_updates;
738 int lock = (trans->type != TRANS_JOIN_NOLOCK);
740 int must_run_delayed_refs = 0;
742 if (trans->use_count > 1) {
744 trans->block_rsv = trans->orig_rsv;
748 btrfs_trans_release_metadata(trans, root);
749 trans->block_rsv = NULL;
751 if (!list_empty(&trans->new_bgs))
752 btrfs_create_pending_block_groups(trans, root);
754 if (!list_empty(&trans->ordered)) {
755 spin_lock(&info->trans_lock);
756 list_splice(&trans->ordered, &cur_trans->pending_ordered);
757 spin_unlock(&info->trans_lock);
760 trans->delayed_ref_updates = 0;
762 must_run_delayed_refs =
763 btrfs_should_throttle_delayed_refs(trans, root);
764 cur = max_t(unsigned long, cur, 32);
767 * don't make the caller wait if they are from a NOLOCK
768 * or ATTACH transaction, it will deadlock with commit
770 if (must_run_delayed_refs == 1 &&
771 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
772 must_run_delayed_refs = 2;
775 if (trans->qgroup_reserved) {
777 * the same root has to be passed here between start_transaction
778 * and end_transaction. Subvolume quota depends on this.
780 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
781 trans->qgroup_reserved = 0;
784 btrfs_trans_release_metadata(trans, root);
785 trans->block_rsv = NULL;
787 if (!list_empty(&trans->new_bgs))
788 btrfs_create_pending_block_groups(trans, root);
790 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
791 should_end_transaction(trans, root) &&
792 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
793 spin_lock(&info->trans_lock);
794 if (cur_trans->state == TRANS_STATE_RUNNING)
795 cur_trans->state = TRANS_STATE_BLOCKED;
796 spin_unlock(&info->trans_lock);
799 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
801 return btrfs_commit_transaction(trans, root);
803 wake_up_process(info->transaction_kthread);
806 if (trans->type & __TRANS_FREEZABLE)
807 sb_end_intwrite(root->fs_info->sb);
809 WARN_ON(cur_trans != info->running_transaction);
810 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
811 atomic_dec(&cur_trans->num_writers);
812 extwriter_counter_dec(cur_trans, trans->type);
815 if (waitqueue_active(&cur_trans->writer_wait))
816 wake_up(&cur_trans->writer_wait);
817 btrfs_put_transaction(cur_trans);
819 if (current->journal_info == trans)
820 current->journal_info = NULL;
823 btrfs_run_delayed_iputs(root);
825 if (trans->aborted ||
826 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
827 wake_up_process(info->transaction_kthread);
830 assert_qgroups_uptodate(trans);
832 kmem_cache_free(btrfs_trans_handle_cachep, trans);
833 if (must_run_delayed_refs) {
834 btrfs_async_run_delayed_refs(root, cur,
835 must_run_delayed_refs == 1);
840 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
841 struct btrfs_root *root)
843 return __btrfs_end_transaction(trans, root, 0);
846 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
847 struct btrfs_root *root)
849 return __btrfs_end_transaction(trans, root, 1);
853 * when btree blocks are allocated, they have some corresponding bits set for
854 * them in one of two extent_io trees. This is used to make sure all of
855 * those extents are sent to disk but does not wait on them
857 int btrfs_write_marked_extents(struct btrfs_root *root,
858 struct extent_io_tree *dirty_pages, int mark)
862 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
863 struct extent_state *cached_state = NULL;
867 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
868 mark, &cached_state)) {
869 bool wait_writeback = false;
871 err = convert_extent_bit(dirty_pages, start, end,
873 mark, &cached_state, GFP_NOFS);
875 * convert_extent_bit can return -ENOMEM, which is most of the
876 * time a temporary error. So when it happens, ignore the error
877 * and wait for writeback of this range to finish - because we
878 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
879 * to btrfs_wait_marked_extents() would not know that writeback
880 * for this range started and therefore wouldn't wait for it to
881 * finish - we don't want to commit a superblock that points to
882 * btree nodes/leafs for which writeback hasn't finished yet
883 * (and without errors).
884 * We cleanup any entries left in the io tree when committing
885 * the transaction (through clear_btree_io_tree()).
887 if (err == -ENOMEM) {
889 wait_writeback = true;
892 err = filemap_fdatawrite_range(mapping, start, end);
895 else if (wait_writeback)
896 werr = filemap_fdatawait_range(mapping, start, end);
897 free_extent_state(cached_state);
906 * when btree blocks are allocated, they have some corresponding bits set for
907 * them in one of two extent_io trees. This is used to make sure all of
908 * those extents are on disk for transaction or log commit. We wait
909 * on all the pages and clear them from the dirty pages state tree
911 int btrfs_wait_marked_extents(struct btrfs_root *root,
912 struct extent_io_tree *dirty_pages, int mark)
916 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
917 struct extent_state *cached_state = NULL;
920 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
923 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
924 EXTENT_NEED_WAIT, &cached_state)) {
926 * Ignore -ENOMEM errors returned by clear_extent_bit().
927 * When committing the transaction, we'll remove any entries
928 * left in the io tree. For a log commit, we don't remove them
929 * after committing the log because the tree can be accessed
930 * concurrently - we do it only at transaction commit time when
931 * it's safe to do it (through clear_btree_io_tree()).
933 err = clear_extent_bit(dirty_pages, start, end,
935 0, 0, &cached_state, GFP_NOFS);
939 err = filemap_fdatawait_range(mapping, start, end);
942 free_extent_state(cached_state);
950 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
951 if ((mark & EXTENT_DIRTY) &&
952 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
953 &btree_ino->runtime_flags))
956 if ((mark & EXTENT_NEW) &&
957 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
958 &btree_ino->runtime_flags))
961 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
962 &btree_ino->runtime_flags))
973 * when btree blocks are allocated, they have some corresponding bits set for
974 * them in one of two extent_io trees. This is used to make sure all of
975 * those extents are on disk for transaction or log commit
977 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
978 struct extent_io_tree *dirty_pages, int mark)
982 struct blk_plug plug;
984 blk_start_plug(&plug);
985 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
986 blk_finish_plug(&plug);
987 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
996 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
997 struct btrfs_root *root)
1001 ret = btrfs_write_and_wait_marked_extents(root,
1002 &trans->transaction->dirty_pages,
1004 clear_btree_io_tree(&trans->transaction->dirty_pages);
1010 * this is used to update the root pointer in the tree of tree roots.
1012 * But, in the case of the extent allocation tree, updating the root
1013 * pointer may allocate blocks which may change the root of the extent
1016 * So, this loops and repeats and makes sure the cowonly root didn't
1017 * change while the root pointer was being updated in the metadata.
1019 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1020 struct btrfs_root *root)
1023 u64 old_root_bytenr;
1025 struct btrfs_root *tree_root = root->fs_info->tree_root;
1027 old_root_used = btrfs_root_used(&root->root_item);
1030 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1031 if (old_root_bytenr == root->node->start &&
1032 old_root_used == btrfs_root_used(&root->root_item))
1035 btrfs_set_root_node(&root->root_item, root->node);
1036 ret = btrfs_update_root(trans, tree_root,
1042 old_root_used = btrfs_root_used(&root->root_item);
1049 * update all the cowonly tree roots on disk
1051 * The error handling in this function may not be obvious. Any of the
1052 * failures will cause the file system to go offline. We still need
1053 * to clean up the delayed refs.
1055 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1056 struct btrfs_root *root)
1058 struct btrfs_fs_info *fs_info = root->fs_info;
1059 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1060 struct list_head *next;
1061 struct extent_buffer *eb;
1064 eb = btrfs_lock_root_node(fs_info->tree_root);
1065 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1067 btrfs_tree_unlock(eb);
1068 free_extent_buffer(eb);
1073 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1077 ret = btrfs_run_dev_stats(trans, root->fs_info);
1080 ret = btrfs_run_dev_replace(trans, root->fs_info);
1083 ret = btrfs_run_qgroups(trans, root->fs_info);
1087 ret = btrfs_setup_space_cache(trans, root);
1091 /* run_qgroups might have added some more refs */
1092 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1096 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1097 next = fs_info->dirty_cowonly_roots.next;
1098 list_del_init(next);
1099 root = list_entry(next, struct btrfs_root, dirty_list);
1100 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1102 if (root != fs_info->extent_root)
1103 list_add_tail(&root->dirty_list,
1104 &trans->transaction->switch_commits);
1105 ret = update_cowonly_root(trans, root);
1108 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1113 while (!list_empty(dirty_bgs)) {
1114 ret = btrfs_write_dirty_block_groups(trans, root);
1117 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1122 if (!list_empty(&fs_info->dirty_cowonly_roots))
1125 list_add_tail(&fs_info->extent_root->dirty_list,
1126 &trans->transaction->switch_commits);
1127 btrfs_after_dev_replace_commit(fs_info);
1133 * dead roots are old snapshots that need to be deleted. This allocates
1134 * a dirty root struct and adds it into the list of dead roots that need to
1137 void btrfs_add_dead_root(struct btrfs_root *root)
1139 spin_lock(&root->fs_info->trans_lock);
1140 if (list_empty(&root->root_list))
1141 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1142 spin_unlock(&root->fs_info->trans_lock);
1146 * update all the cowonly tree roots on disk
1148 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1149 struct btrfs_root *root)
1151 struct btrfs_root *gang[8];
1152 struct btrfs_fs_info *fs_info = root->fs_info;
1157 spin_lock(&fs_info->fs_roots_radix_lock);
1159 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1162 BTRFS_ROOT_TRANS_TAG);
1165 for (i = 0; i < ret; i++) {
1167 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1168 (unsigned long)root->root_key.objectid,
1169 BTRFS_ROOT_TRANS_TAG);
1170 spin_unlock(&fs_info->fs_roots_radix_lock);
1172 btrfs_free_log(trans, root);
1173 btrfs_update_reloc_root(trans, root);
1174 btrfs_orphan_commit_root(trans, root);
1176 btrfs_save_ino_cache(root, trans);
1178 /* see comments in should_cow_block() */
1179 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1180 smp_mb__after_atomic();
1182 if (root->commit_root != root->node) {
1183 list_add_tail(&root->dirty_list,
1184 &trans->transaction->switch_commits);
1185 btrfs_set_root_node(&root->root_item,
1189 err = btrfs_update_root(trans, fs_info->tree_root,
1192 spin_lock(&fs_info->fs_roots_radix_lock);
1197 spin_unlock(&fs_info->fs_roots_radix_lock);
1202 * defrag a given btree.
1203 * Every leaf in the btree is read and defragged.
1205 int btrfs_defrag_root(struct btrfs_root *root)
1207 struct btrfs_fs_info *info = root->fs_info;
1208 struct btrfs_trans_handle *trans;
1211 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1215 trans = btrfs_start_transaction(root, 0);
1217 return PTR_ERR(trans);
1219 ret = btrfs_defrag_leaves(trans, root);
1221 btrfs_end_transaction(trans, root);
1222 btrfs_btree_balance_dirty(info->tree_root);
1225 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1228 if (btrfs_defrag_cancelled(root->fs_info)) {
1229 pr_debug("BTRFS: defrag_root cancelled\n");
1234 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1239 * new snapshots need to be created at a very specific time in the
1240 * transaction commit. This does the actual creation.
1243 * If the error which may affect the commitment of the current transaction
1244 * happens, we should return the error number. If the error which just affect
1245 * the creation of the pending snapshots, just return 0.
1247 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1248 struct btrfs_fs_info *fs_info,
1249 struct btrfs_pending_snapshot *pending)
1251 struct btrfs_key key;
1252 struct btrfs_root_item *new_root_item;
1253 struct btrfs_root *tree_root = fs_info->tree_root;
1254 struct btrfs_root *root = pending->root;
1255 struct btrfs_root *parent_root;
1256 struct btrfs_block_rsv *rsv;
1257 struct inode *parent_inode;
1258 struct btrfs_path *path;
1259 struct btrfs_dir_item *dir_item;
1260 struct dentry *dentry;
1261 struct extent_buffer *tmp;
1262 struct extent_buffer *old;
1263 struct timespec cur_time = CURRENT_TIME;
1271 path = btrfs_alloc_path();
1273 pending->error = -ENOMEM;
1277 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1278 if (!new_root_item) {
1279 pending->error = -ENOMEM;
1280 goto root_item_alloc_fail;
1283 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1285 goto no_free_objectid;
1287 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1289 if (to_reserve > 0) {
1290 pending->error = btrfs_block_rsv_add(root,
1291 &pending->block_rsv,
1293 BTRFS_RESERVE_NO_FLUSH);
1295 goto no_free_objectid;
1298 key.objectid = objectid;
1299 key.offset = (u64)-1;
1300 key.type = BTRFS_ROOT_ITEM_KEY;
1302 rsv = trans->block_rsv;
1303 trans->block_rsv = &pending->block_rsv;
1304 trans->bytes_reserved = trans->block_rsv->reserved;
1306 dentry = pending->dentry;
1307 parent_inode = pending->dir;
1308 parent_root = BTRFS_I(parent_inode)->root;
1309 record_root_in_trans(trans, parent_root);
1312 * insert the directory item
1314 ret = btrfs_set_inode_index(parent_inode, &index);
1315 BUG_ON(ret); /* -ENOMEM */
1317 /* check if there is a file/dir which has the same name. */
1318 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1319 btrfs_ino(parent_inode),
1320 dentry->d_name.name,
1321 dentry->d_name.len, 0);
1322 if (dir_item != NULL && !IS_ERR(dir_item)) {
1323 pending->error = -EEXIST;
1324 goto dir_item_existed;
1325 } else if (IS_ERR(dir_item)) {
1326 ret = PTR_ERR(dir_item);
1327 btrfs_abort_transaction(trans, root, ret);
1330 btrfs_release_path(path);
1333 * pull in the delayed directory update
1334 * and the delayed inode item
1335 * otherwise we corrupt the FS during
1338 ret = btrfs_run_delayed_items(trans, root);
1339 if (ret) { /* Transaction aborted */
1340 btrfs_abort_transaction(trans, root, ret);
1344 record_root_in_trans(trans, root);
1345 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1346 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1347 btrfs_check_and_init_root_item(new_root_item);
1349 root_flags = btrfs_root_flags(new_root_item);
1350 if (pending->readonly)
1351 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1353 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1354 btrfs_set_root_flags(new_root_item, root_flags);
1356 btrfs_set_root_generation_v2(new_root_item,
1358 uuid_le_gen(&new_uuid);
1359 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1360 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1362 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1363 memset(new_root_item->received_uuid, 0,
1364 sizeof(new_root_item->received_uuid));
1365 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1366 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1367 btrfs_set_root_stransid(new_root_item, 0);
1368 btrfs_set_root_rtransid(new_root_item, 0);
1370 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1371 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1372 btrfs_set_root_otransid(new_root_item, trans->transid);
1374 old = btrfs_lock_root_node(root);
1375 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1377 btrfs_tree_unlock(old);
1378 free_extent_buffer(old);
1379 btrfs_abort_transaction(trans, root, ret);
1383 btrfs_set_lock_blocking(old);
1385 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1386 /* clean up in any case */
1387 btrfs_tree_unlock(old);
1388 free_extent_buffer(old);
1390 btrfs_abort_transaction(trans, root, ret);
1395 * We need to flush delayed refs in order to make sure all of our quota
1396 * operations have been done before we call btrfs_qgroup_inherit.
1398 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1400 btrfs_abort_transaction(trans, root, ret);
1404 ret = btrfs_qgroup_inherit(trans, fs_info,
1405 root->root_key.objectid,
1406 objectid, pending->inherit);
1408 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 pending->error = ret;
1497 trans->block_rsv = rsv;
1498 trans->bytes_reserved = 0;
1500 kfree(new_root_item);
1501 root_item_alloc_fail:
1502 btrfs_free_path(path);
1507 * create all the snapshots we've scheduled for creation
1509 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1510 struct btrfs_fs_info *fs_info)
1512 struct btrfs_pending_snapshot *pending, *next;
1513 struct list_head *head = &trans->transaction->pending_snapshots;
1516 list_for_each_entry_safe(pending, next, head, list) {
1517 list_del(&pending->list);
1518 ret = create_pending_snapshot(trans, fs_info, pending);
1525 static void update_super_roots(struct btrfs_root *root)
1527 struct btrfs_root_item *root_item;
1528 struct btrfs_super_block *super;
1530 super = root->fs_info->super_copy;
1532 root_item = &root->fs_info->chunk_root->root_item;
1533 super->chunk_root = root_item->bytenr;
1534 super->chunk_root_generation = root_item->generation;
1535 super->chunk_root_level = root_item->level;
1537 root_item = &root->fs_info->tree_root->root_item;
1538 super->root = root_item->bytenr;
1539 super->generation = root_item->generation;
1540 super->root_level = root_item->level;
1541 if (btrfs_test_opt(root, SPACE_CACHE))
1542 super->cache_generation = root_item->generation;
1543 if (root->fs_info->update_uuid_tree_gen)
1544 super->uuid_tree_generation = root_item->generation;
1547 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1549 struct btrfs_transaction *trans;
1552 spin_lock(&info->trans_lock);
1553 trans = info->running_transaction;
1555 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1556 spin_unlock(&info->trans_lock);
1560 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1562 struct btrfs_transaction *trans;
1565 spin_lock(&info->trans_lock);
1566 trans = info->running_transaction;
1568 ret = is_transaction_blocked(trans);
1569 spin_unlock(&info->trans_lock);
1574 * wait for the current transaction commit to start and block subsequent
1577 static void wait_current_trans_commit_start(struct btrfs_root *root,
1578 struct btrfs_transaction *trans)
1580 wait_event(root->fs_info->transaction_blocked_wait,
1581 trans->state >= TRANS_STATE_COMMIT_START ||
1586 * wait for the current transaction to start and then become unblocked.
1589 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1590 struct btrfs_transaction *trans)
1592 wait_event(root->fs_info->transaction_wait,
1593 trans->state >= TRANS_STATE_UNBLOCKED ||
1598 * commit transactions asynchronously. once btrfs_commit_transaction_async
1599 * returns, any subsequent transaction will not be allowed to join.
1601 struct btrfs_async_commit {
1602 struct btrfs_trans_handle *newtrans;
1603 struct btrfs_root *root;
1604 struct work_struct work;
1607 static void do_async_commit(struct work_struct *work)
1609 struct btrfs_async_commit *ac =
1610 container_of(work, struct btrfs_async_commit, work);
1613 * We've got freeze protection passed with the transaction.
1614 * Tell lockdep about it.
1616 if (ac->newtrans->type & __TRANS_FREEZABLE)
1618 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1621 current->journal_info = ac->newtrans;
1623 btrfs_commit_transaction(ac->newtrans, ac->root);
1627 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1628 struct btrfs_root *root,
1629 int wait_for_unblock)
1631 struct btrfs_async_commit *ac;
1632 struct btrfs_transaction *cur_trans;
1634 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1638 INIT_WORK(&ac->work, do_async_commit);
1640 ac->newtrans = btrfs_join_transaction(root);
1641 if (IS_ERR(ac->newtrans)) {
1642 int err = PTR_ERR(ac->newtrans);
1647 /* take transaction reference */
1648 cur_trans = trans->transaction;
1649 atomic_inc(&cur_trans->use_count);
1651 btrfs_end_transaction(trans, root);
1654 * Tell lockdep we've released the freeze rwsem, since the
1655 * async commit thread will be the one to unlock it.
1657 if (ac->newtrans->type & __TRANS_FREEZABLE)
1659 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1662 schedule_work(&ac->work);
1664 /* wait for transaction to start and unblock */
1665 if (wait_for_unblock)
1666 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1668 wait_current_trans_commit_start(root, cur_trans);
1670 if (current->journal_info == trans)
1671 current->journal_info = NULL;
1673 btrfs_put_transaction(cur_trans);
1678 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1679 struct btrfs_root *root, int err)
1681 struct btrfs_transaction *cur_trans = trans->transaction;
1684 WARN_ON(trans->use_count > 1);
1686 btrfs_abort_transaction(trans, root, err);
1688 spin_lock(&root->fs_info->trans_lock);
1691 * If the transaction is removed from the list, it means this
1692 * transaction has been committed successfully, so it is impossible
1693 * to call the cleanup function.
1695 BUG_ON(list_empty(&cur_trans->list));
1697 list_del_init(&cur_trans->list);
1698 if (cur_trans == root->fs_info->running_transaction) {
1699 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1700 spin_unlock(&root->fs_info->trans_lock);
1701 wait_event(cur_trans->writer_wait,
1702 atomic_read(&cur_trans->num_writers) == 1);
1704 spin_lock(&root->fs_info->trans_lock);
1706 spin_unlock(&root->fs_info->trans_lock);
1708 btrfs_cleanup_one_transaction(trans->transaction, root);
1710 spin_lock(&root->fs_info->trans_lock);
1711 if (cur_trans == root->fs_info->running_transaction)
1712 root->fs_info->running_transaction = NULL;
1713 spin_unlock(&root->fs_info->trans_lock);
1715 if (trans->type & __TRANS_FREEZABLE)
1716 sb_end_intwrite(root->fs_info->sb);
1717 btrfs_put_transaction(cur_trans);
1718 btrfs_put_transaction(cur_trans);
1720 trace_btrfs_transaction_commit(root);
1722 if (current->journal_info == trans)
1723 current->journal_info = NULL;
1724 btrfs_scrub_cancel(root->fs_info);
1726 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1729 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1731 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1732 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1736 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1738 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1739 btrfs_wait_ordered_roots(fs_info, -1);
1743 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans,
1744 struct btrfs_fs_info *fs_info)
1746 struct btrfs_ordered_extent *ordered;
1748 spin_lock(&fs_info->trans_lock);
1749 while (!list_empty(&cur_trans->pending_ordered)) {
1750 ordered = list_first_entry(&cur_trans->pending_ordered,
1751 struct btrfs_ordered_extent,
1753 list_del_init(&ordered->trans_list);
1754 spin_unlock(&fs_info->trans_lock);
1756 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE,
1758 btrfs_put_ordered_extent(ordered);
1759 spin_lock(&fs_info->trans_lock);
1761 spin_unlock(&fs_info->trans_lock);
1764 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1765 struct btrfs_root *root)
1767 struct btrfs_transaction *cur_trans = trans->transaction;
1768 struct btrfs_transaction *prev_trans = NULL;
1769 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
1772 /* Stop the commit early if ->aborted is set */
1773 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1774 ret = cur_trans->aborted;
1775 btrfs_end_transaction(trans, root);
1779 /* make a pass through all the delayed refs we have so far
1780 * any runnings procs may add more while we are here
1782 ret = btrfs_run_delayed_refs(trans, root, 0);
1784 btrfs_end_transaction(trans, root);
1788 btrfs_trans_release_metadata(trans, root);
1789 trans->block_rsv = NULL;
1790 if (trans->qgroup_reserved) {
1791 btrfs_qgroup_free(root, trans->qgroup_reserved);
1792 trans->qgroup_reserved = 0;
1795 cur_trans = trans->transaction;
1798 * set the flushing flag so procs in this transaction have to
1799 * start sending their work down.
1801 cur_trans->delayed_refs.flushing = 1;
1804 if (!list_empty(&trans->new_bgs))
1805 btrfs_create_pending_block_groups(trans, root);
1807 ret = btrfs_run_delayed_refs(trans, root, 0);
1809 btrfs_end_transaction(trans, root);
1813 spin_lock(&root->fs_info->trans_lock);
1814 list_splice(&trans->ordered, &cur_trans->pending_ordered);
1815 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1816 spin_unlock(&root->fs_info->trans_lock);
1817 atomic_inc(&cur_trans->use_count);
1818 ret = btrfs_end_transaction(trans, root);
1820 wait_for_commit(root, cur_trans);
1822 if (unlikely(cur_trans->aborted))
1823 ret = cur_trans->aborted;
1825 btrfs_put_transaction(cur_trans);
1830 cur_trans->state = TRANS_STATE_COMMIT_START;
1831 wake_up(&root->fs_info->transaction_blocked_wait);
1833 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1834 prev_trans = list_entry(cur_trans->list.prev,
1835 struct btrfs_transaction, list);
1836 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1837 atomic_inc(&prev_trans->use_count);
1838 spin_unlock(&root->fs_info->trans_lock);
1840 wait_for_commit(root, prev_trans);
1842 btrfs_put_transaction(prev_trans);
1844 spin_unlock(&root->fs_info->trans_lock);
1847 spin_unlock(&root->fs_info->trans_lock);
1850 extwriter_counter_dec(cur_trans, trans->type);
1852 ret = btrfs_start_delalloc_flush(root->fs_info);
1854 goto cleanup_transaction;
1856 ret = btrfs_run_delayed_items(trans, root);
1858 goto cleanup_transaction;
1860 wait_event(cur_trans->writer_wait,
1861 extwriter_counter_read(cur_trans) == 0);
1863 /* some pending stuffs might be added after the previous flush. */
1864 ret = btrfs_run_delayed_items(trans, root);
1866 goto cleanup_transaction;
1868 btrfs_wait_delalloc_flush(root->fs_info);
1870 btrfs_wait_pending_ordered(cur_trans, root->fs_info);
1872 btrfs_scrub_pause(root);
1874 * Ok now we need to make sure to block out any other joins while we
1875 * commit the transaction. We could have started a join before setting
1876 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1878 spin_lock(&root->fs_info->trans_lock);
1879 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1880 spin_unlock(&root->fs_info->trans_lock);
1881 wait_event(cur_trans->writer_wait,
1882 atomic_read(&cur_trans->num_writers) == 1);
1884 /* ->aborted might be set after the previous check, so check it */
1885 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1886 ret = cur_trans->aborted;
1887 goto scrub_continue;
1890 * the reloc mutex makes sure that we stop
1891 * the balancing code from coming in and moving
1892 * extents around in the middle of the commit
1894 mutex_lock(&root->fs_info->reloc_mutex);
1897 * We needn't worry about the delayed items because we will
1898 * deal with them in create_pending_snapshot(), which is the
1899 * core function of the snapshot creation.
1901 ret = create_pending_snapshots(trans, root->fs_info);
1903 mutex_unlock(&root->fs_info->reloc_mutex);
1904 goto scrub_continue;
1908 * We insert the dir indexes of the snapshots and update the inode
1909 * of the snapshots' parents after the snapshot creation, so there
1910 * are some delayed items which are not dealt with. Now deal with
1913 * We needn't worry that this operation will corrupt the snapshots,
1914 * because all the tree which are snapshoted will be forced to COW
1915 * the nodes and leaves.
1917 ret = btrfs_run_delayed_items(trans, root);
1919 mutex_unlock(&root->fs_info->reloc_mutex);
1920 goto scrub_continue;
1923 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1925 mutex_unlock(&root->fs_info->reloc_mutex);
1926 goto scrub_continue;
1930 * make sure none of the code above managed to slip in a
1933 btrfs_assert_delayed_root_empty(root);
1935 WARN_ON(cur_trans != trans->transaction);
1937 /* btrfs_commit_tree_roots is responsible for getting the
1938 * various roots consistent with each other. Every pointer
1939 * in the tree of tree roots has to point to the most up to date
1940 * root for every subvolume and other tree. So, we have to keep
1941 * the tree logging code from jumping in and changing any
1944 * At this point in the commit, there can't be any tree-log
1945 * writers, but a little lower down we drop the trans mutex
1946 * and let new people in. By holding the tree_log_mutex
1947 * from now until after the super is written, we avoid races
1948 * with the tree-log code.
1950 mutex_lock(&root->fs_info->tree_log_mutex);
1952 ret = commit_fs_roots(trans, root);
1954 mutex_unlock(&root->fs_info->tree_log_mutex);
1955 mutex_unlock(&root->fs_info->reloc_mutex);
1956 goto scrub_continue;
1960 * Since the transaction is done, we can apply the pending changes
1961 * before the next transaction.
1963 btrfs_apply_pending_changes(root->fs_info);
1965 /* commit_fs_roots gets rid of all the tree log roots, it is now
1966 * safe to free the root of tree log roots
1968 btrfs_free_log_root_tree(trans, root->fs_info);
1970 ret = commit_cowonly_roots(trans, root);
1972 mutex_unlock(&root->fs_info->tree_log_mutex);
1973 mutex_unlock(&root->fs_info->reloc_mutex);
1974 goto scrub_continue;
1978 * The tasks which save the space cache and inode cache may also
1979 * update ->aborted, check it.
1981 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1982 ret = cur_trans->aborted;
1983 mutex_unlock(&root->fs_info->tree_log_mutex);
1984 mutex_unlock(&root->fs_info->reloc_mutex);
1985 goto scrub_continue;
1988 btrfs_prepare_extent_commit(trans, root);
1990 cur_trans = root->fs_info->running_transaction;
1992 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1993 root->fs_info->tree_root->node);
1994 list_add_tail(&root->fs_info->tree_root->dirty_list,
1995 &cur_trans->switch_commits);
1997 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1998 root->fs_info->chunk_root->node);
1999 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2000 &cur_trans->switch_commits);
2002 switch_commit_roots(cur_trans, root->fs_info);
2004 assert_qgroups_uptodate(trans);
2005 ASSERT(list_empty(&cur_trans->dirty_bgs));
2006 update_super_roots(root);
2008 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2009 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
2010 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2011 sizeof(*root->fs_info->super_copy));
2013 btrfs_update_commit_device_size(root->fs_info);
2014 btrfs_update_commit_device_bytes_used(root, cur_trans);
2016 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
2017 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
2019 spin_lock(&root->fs_info->trans_lock);
2020 cur_trans->state = TRANS_STATE_UNBLOCKED;
2021 root->fs_info->running_transaction = NULL;
2022 spin_unlock(&root->fs_info->trans_lock);
2023 mutex_unlock(&root->fs_info->reloc_mutex);
2025 wake_up(&root->fs_info->transaction_wait);
2027 ret = btrfs_write_and_wait_transaction(trans, root);
2029 btrfs_error(root->fs_info, ret,
2030 "Error while writing out transaction");
2031 mutex_unlock(&root->fs_info->tree_log_mutex);
2032 goto scrub_continue;
2035 ret = write_ctree_super(trans, root, 0);
2037 mutex_unlock(&root->fs_info->tree_log_mutex);
2038 goto scrub_continue;
2042 * the super is written, we can safely allow the tree-loggers
2043 * to go about their business
2045 mutex_unlock(&root->fs_info->tree_log_mutex);
2047 btrfs_finish_extent_commit(trans, root);
2049 if (cur_trans->have_free_bgs)
2050 btrfs_clear_space_info_full(root->fs_info);
2052 root->fs_info->last_trans_committed = cur_trans->transid;
2054 * We needn't acquire the lock here because there is no other task
2055 * which can change it.
2057 cur_trans->state = TRANS_STATE_COMPLETED;
2058 wake_up(&cur_trans->commit_wait);
2060 spin_lock(&root->fs_info->trans_lock);
2061 list_del_init(&cur_trans->list);
2062 spin_unlock(&root->fs_info->trans_lock);
2064 btrfs_put_transaction(cur_trans);
2065 btrfs_put_transaction(cur_trans);
2067 if (trans->type & __TRANS_FREEZABLE)
2068 sb_end_intwrite(root->fs_info->sb);
2070 trace_btrfs_transaction_commit(root);
2072 btrfs_scrub_continue(root);
2074 if (current->journal_info == trans)
2075 current->journal_info = NULL;
2077 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2079 if (current != root->fs_info->transaction_kthread)
2080 btrfs_run_delayed_iputs(root);
2085 btrfs_scrub_continue(root);
2086 cleanup_transaction:
2087 btrfs_trans_release_metadata(trans, root);
2088 trans->block_rsv = NULL;
2089 if (trans->qgroup_reserved) {
2090 btrfs_qgroup_free(root, trans->qgroup_reserved);
2091 trans->qgroup_reserved = 0;
2093 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
2094 if (current->journal_info == trans)
2095 current->journal_info = NULL;
2096 cleanup_transaction(trans, root, ret);
2102 * return < 0 if error
2103 * 0 if there are no more dead_roots at the time of call
2104 * 1 there are more to be processed, call me again
2106 * The return value indicates there are certainly more snapshots to delete, but
2107 * if there comes a new one during processing, it may return 0. We don't mind,
2108 * because btrfs_commit_super will poke cleaner thread and it will process it a
2109 * few seconds later.
2111 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2114 struct btrfs_fs_info *fs_info = root->fs_info;
2116 spin_lock(&fs_info->trans_lock);
2117 if (list_empty(&fs_info->dead_roots)) {
2118 spin_unlock(&fs_info->trans_lock);
2121 root = list_first_entry(&fs_info->dead_roots,
2122 struct btrfs_root, root_list);
2123 list_del_init(&root->root_list);
2124 spin_unlock(&fs_info->trans_lock);
2126 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2128 btrfs_kill_all_delayed_nodes(root);
2130 if (btrfs_header_backref_rev(root->node) <
2131 BTRFS_MIXED_BACKREF_REV)
2132 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2134 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2136 return (ret < 0) ? 0 : 1;
2139 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2144 prev = xchg(&fs_info->pending_changes, 0);
2148 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2150 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2153 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2155 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2158 bit = 1 << BTRFS_PENDING_COMMIT;
2160 btrfs_debug(fs_info, "pending commit done");
2165 "unknown pending changes left 0x%lx, ignoring", prev);