2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/slab.h>
21 #include <linux/sched.h>
22 #include <linux/writeback.h>
23 #include <linux/pagemap.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
28 #include "transaction.h"
31 #include "inode-map.h"
33 #include "dev-replace.h"
36 #define BTRFS_ROOT_TRANS_TAG 0
38 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
39 [TRANS_STATE_RUNNING] = 0U,
40 [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE |
42 [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE |
45 [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE |
49 [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE |
54 [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE |
61 void btrfs_put_transaction(struct btrfs_transaction *transaction)
63 WARN_ON(atomic_read(&transaction->use_count) == 0);
64 if (atomic_dec_and_test(&transaction->use_count)) {
65 BUG_ON(!list_empty(&transaction->list));
66 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
67 if (transaction->delayed_refs.pending_csums)
68 printk(KERN_ERR "pending csums is %llu\n",
69 transaction->delayed_refs.pending_csums);
70 while (!list_empty(&transaction->pending_chunks)) {
71 struct extent_map *em;
73 em = list_first_entry(&transaction->pending_chunks,
74 struct extent_map, list);
75 list_del_init(&em->list);
78 kmem_cache_free(btrfs_transaction_cachep, transaction);
82 static void clear_btree_io_tree(struct extent_io_tree *tree)
84 spin_lock(&tree->lock);
85 while (!RB_EMPTY_ROOT(&tree->state)) {
87 struct extent_state *state;
89 node = rb_first(&tree->state);
90 state = rb_entry(node, struct extent_state, rb_node);
91 rb_erase(&state->rb_node, &tree->state);
92 RB_CLEAR_NODE(&state->rb_node);
94 * btree io trees aren't supposed to have tasks waiting for
95 * changes in the flags of extent states ever.
97 ASSERT(!waitqueue_active(&state->wq));
98 free_extent_state(state);
100 cond_resched_lock(&tree->lock);
102 spin_unlock(&tree->lock);
105 static noinline void switch_commit_roots(struct btrfs_transaction *trans,
106 struct btrfs_fs_info *fs_info)
108 struct btrfs_root *root, *tmp;
110 down_write(&fs_info->commit_root_sem);
111 list_for_each_entry_safe(root, tmp, &trans->switch_commits,
113 list_del_init(&root->dirty_list);
114 free_extent_buffer(root->commit_root);
115 root->commit_root = btrfs_root_node(root);
116 if (is_fstree(root->objectid))
117 btrfs_unpin_free_ino(root);
118 clear_btree_io_tree(&root->dirty_log_pages);
120 up_write(&fs_info->commit_root_sem);
123 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
126 if (type & TRANS_EXTWRITERS)
127 atomic_inc(&trans->num_extwriters);
130 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
133 if (type & TRANS_EXTWRITERS)
134 atomic_dec(&trans->num_extwriters);
137 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
140 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
143 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
145 return atomic_read(&trans->num_extwriters);
149 * either allocate a new transaction or hop into the existing one
151 static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
153 struct btrfs_transaction *cur_trans;
154 struct btrfs_fs_info *fs_info = root->fs_info;
156 spin_lock(&fs_info->trans_lock);
158 /* The file system has been taken offline. No new transactions. */
159 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
160 spin_unlock(&fs_info->trans_lock);
164 cur_trans = fs_info->running_transaction;
166 if (cur_trans->aborted) {
167 spin_unlock(&fs_info->trans_lock);
168 return cur_trans->aborted;
170 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
171 spin_unlock(&fs_info->trans_lock);
174 atomic_inc(&cur_trans->use_count);
175 atomic_inc(&cur_trans->num_writers);
176 extwriter_counter_inc(cur_trans, type);
177 spin_unlock(&fs_info->trans_lock);
180 spin_unlock(&fs_info->trans_lock);
183 * If we are ATTACH, we just want to catch the current transaction,
184 * and commit it. If there is no transaction, just return ENOENT.
186 if (type == TRANS_ATTACH)
190 * JOIN_NOLOCK only happens during the transaction commit, so
191 * it is impossible that ->running_transaction is NULL
193 BUG_ON(type == TRANS_JOIN_NOLOCK);
195 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
199 spin_lock(&fs_info->trans_lock);
200 if (fs_info->running_transaction) {
202 * someone started a transaction after we unlocked. Make sure
203 * to redo the checks above
205 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
207 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
208 spin_unlock(&fs_info->trans_lock);
209 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
213 atomic_set(&cur_trans->num_writers, 1);
214 extwriter_counter_init(cur_trans, type);
215 init_waitqueue_head(&cur_trans->writer_wait);
216 init_waitqueue_head(&cur_trans->commit_wait);
217 cur_trans->state = TRANS_STATE_RUNNING;
219 * One for this trans handle, one so it will live on until we
220 * commit the transaction.
222 atomic_set(&cur_trans->use_count, 2);
223 cur_trans->have_free_bgs = 0;
224 cur_trans->start_time = get_seconds();
225 cur_trans->dirty_bg_run = 0;
227 cur_trans->delayed_refs.href_root = RB_ROOT;
228 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
229 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
230 cur_trans->delayed_refs.num_heads_ready = 0;
231 cur_trans->delayed_refs.pending_csums = 0;
232 cur_trans->delayed_refs.num_heads = 0;
233 cur_trans->delayed_refs.flushing = 0;
234 cur_trans->delayed_refs.run_delayed_start = 0;
235 cur_trans->delayed_refs.qgroup_to_skip = 0;
238 * although the tree mod log is per file system and not per transaction,
239 * the log must never go across transaction boundaries.
242 if (!list_empty(&fs_info->tree_mod_seq_list))
243 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
244 "creating a fresh transaction\n");
245 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
246 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
247 "creating a fresh transaction\n");
248 atomic64_set(&fs_info->tree_mod_seq, 0);
250 spin_lock_init(&cur_trans->delayed_refs.lock);
252 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
253 INIT_LIST_HEAD(&cur_trans->pending_chunks);
254 INIT_LIST_HEAD(&cur_trans->switch_commits);
255 INIT_LIST_HEAD(&cur_trans->pending_ordered);
256 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
257 INIT_LIST_HEAD(&cur_trans->io_bgs);
258 mutex_init(&cur_trans->cache_write_mutex);
259 cur_trans->num_dirty_bgs = 0;
260 spin_lock_init(&cur_trans->dirty_bgs_lock);
261 list_add_tail(&cur_trans->list, &fs_info->trans_list);
262 extent_io_tree_init(&cur_trans->dirty_pages,
263 fs_info->btree_inode->i_mapping);
264 fs_info->generation++;
265 cur_trans->transid = fs_info->generation;
266 fs_info->running_transaction = cur_trans;
267 cur_trans->aborted = 0;
268 spin_unlock(&fs_info->trans_lock);
274 * this does all the record keeping required to make sure that a reference
275 * counted root is properly recorded in a given transaction. This is required
276 * to make sure the old root from before we joined the transaction is deleted
277 * when the transaction commits
279 static int record_root_in_trans(struct btrfs_trans_handle *trans,
280 struct btrfs_root *root)
282 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
283 root->last_trans < trans->transid) {
284 WARN_ON(root == root->fs_info->extent_root);
285 WARN_ON(root->commit_root != root->node);
288 * see below for IN_TRANS_SETUP usage rules
289 * we have the reloc mutex held now, so there
290 * is only one writer in this function
292 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
294 /* make sure readers find IN_TRANS_SETUP before
295 * they find our root->last_trans update
299 spin_lock(&root->fs_info->fs_roots_radix_lock);
300 if (root->last_trans == trans->transid) {
301 spin_unlock(&root->fs_info->fs_roots_radix_lock);
304 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
305 (unsigned long)root->root_key.objectid,
306 BTRFS_ROOT_TRANS_TAG);
307 spin_unlock(&root->fs_info->fs_roots_radix_lock);
308 root->last_trans = trans->transid;
310 /* this is pretty tricky. We don't want to
311 * take the relocation lock in btrfs_record_root_in_trans
312 * unless we're really doing the first setup for this root in
315 * Normally we'd use root->last_trans as a flag to decide
316 * if we want to take the expensive mutex.
318 * But, we have to set root->last_trans before we
319 * init the relocation root, otherwise, we trip over warnings
320 * in ctree.c. The solution used here is to flag ourselves
321 * with root IN_TRANS_SETUP. When this is 1, we're still
322 * fixing up the reloc trees and everyone must wait.
324 * When this is zero, they can trust root->last_trans and fly
325 * through btrfs_record_root_in_trans without having to take the
326 * lock. smp_wmb() makes sure that all the writes above are
327 * done before we pop in the zero below
329 btrfs_init_reloc_root(trans, root);
330 smp_mb__before_atomic();
331 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
337 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
338 struct btrfs_root *root)
340 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
344 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
348 if (root->last_trans == trans->transid &&
349 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
352 mutex_lock(&root->fs_info->reloc_mutex);
353 record_root_in_trans(trans, root);
354 mutex_unlock(&root->fs_info->reloc_mutex);
359 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
361 return (trans->state >= TRANS_STATE_BLOCKED &&
362 trans->state < TRANS_STATE_UNBLOCKED &&
366 /* wait for commit against the current transaction to become unblocked
367 * when this is done, it is safe to start a new transaction, but the current
368 * transaction might not be fully on disk.
370 static void wait_current_trans(struct btrfs_root *root)
372 struct btrfs_transaction *cur_trans;
374 spin_lock(&root->fs_info->trans_lock);
375 cur_trans = root->fs_info->running_transaction;
376 if (cur_trans && is_transaction_blocked(cur_trans)) {
377 atomic_inc(&cur_trans->use_count);
378 spin_unlock(&root->fs_info->trans_lock);
380 wait_event(root->fs_info->transaction_wait,
381 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
383 btrfs_put_transaction(cur_trans);
385 spin_unlock(&root->fs_info->trans_lock);
389 static int may_wait_transaction(struct btrfs_root *root, int type)
391 if (root->fs_info->log_root_recovering)
394 if (type == TRANS_USERSPACE)
397 if (type == TRANS_START &&
398 !atomic_read(&root->fs_info->open_ioctl_trans))
404 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
406 if (!root->fs_info->reloc_ctl ||
407 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
408 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
415 static struct btrfs_trans_handle *
416 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
417 enum btrfs_reserve_flush_enum flush)
419 struct btrfs_trans_handle *h;
420 struct btrfs_transaction *cur_trans;
422 u64 qgroup_reserved = 0;
423 bool reloc_reserved = false;
426 /* Send isn't supposed to start transactions. */
427 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
429 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
430 return ERR_PTR(-EROFS);
432 if (current->journal_info) {
433 WARN_ON(type & TRANS_EXTWRITERS);
434 h = current->journal_info;
436 WARN_ON(h->use_count > 2);
437 h->orig_rsv = h->block_rsv;
443 * Do the reservation before we join the transaction so we can do all
444 * the appropriate flushing if need be.
446 if (num_items > 0 && root != root->fs_info->chunk_root) {
447 if (root->fs_info->quota_enabled &&
448 is_fstree(root->root_key.objectid)) {
449 qgroup_reserved = num_items * root->nodesize;
450 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
455 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
457 * Do the reservation for the relocation root creation
459 if (need_reserve_reloc_root(root)) {
460 num_bytes += root->nodesize;
461 reloc_reserved = true;
464 ret = btrfs_block_rsv_add(root,
465 &root->fs_info->trans_block_rsv,
471 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
478 * If we are JOIN_NOLOCK we're already committing a transaction and
479 * waiting on this guy, so we don't need to do the sb_start_intwrite
480 * because we're already holding a ref. We need this because we could
481 * have raced in and did an fsync() on a file which can kick a commit
482 * and then we deadlock with somebody doing a freeze.
484 * If we are ATTACH, it means we just want to catch the current
485 * transaction and commit it, so we needn't do sb_start_intwrite().
487 if (type & __TRANS_FREEZABLE)
488 sb_start_intwrite(root->fs_info->sb);
490 if (may_wait_transaction(root, type))
491 wait_current_trans(root);
494 ret = join_transaction(root, type);
496 wait_current_trans(root);
497 if (unlikely(type == TRANS_ATTACH))
500 } while (ret == -EBUSY);
503 /* We must get the transaction if we are JOIN_NOLOCK. */
504 BUG_ON(type == TRANS_JOIN_NOLOCK);
508 cur_trans = root->fs_info->running_transaction;
510 h->transid = cur_trans->transid;
511 h->transaction = cur_trans;
513 h->bytes_reserved = 0;
514 h->chunk_bytes_reserved = 0;
516 h->delayed_ref_updates = 0;
522 h->qgroup_reserved = 0;
523 h->delayed_ref_elem.seq = 0;
525 h->allocating_chunk = false;
526 h->reloc_reserved = false;
528 INIT_LIST_HEAD(&h->qgroup_ref_list);
529 INIT_LIST_HEAD(&h->new_bgs);
530 INIT_LIST_HEAD(&h->ordered);
533 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
534 may_wait_transaction(root, type)) {
535 current->journal_info = h;
536 btrfs_commit_transaction(h, root);
541 trace_btrfs_space_reservation(root->fs_info, "transaction",
542 h->transid, num_bytes, 1);
543 h->block_rsv = &root->fs_info->trans_block_rsv;
544 h->bytes_reserved = num_bytes;
545 h->reloc_reserved = reloc_reserved;
547 h->qgroup_reserved = qgroup_reserved;
550 btrfs_record_root_in_trans(h, root);
552 if (!current->journal_info && type != TRANS_USERSPACE)
553 current->journal_info = h;
557 if (type & __TRANS_FREEZABLE)
558 sb_end_intwrite(root->fs_info->sb);
559 kmem_cache_free(btrfs_trans_handle_cachep, h);
562 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
566 btrfs_qgroup_free(root, qgroup_reserved);
570 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
573 return start_transaction(root, num_items, TRANS_START,
574 BTRFS_RESERVE_FLUSH_ALL);
577 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
578 struct btrfs_root *root, int num_items)
580 return start_transaction(root, num_items, TRANS_START,
581 BTRFS_RESERVE_FLUSH_LIMIT);
584 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
586 return start_transaction(root, 0, TRANS_JOIN, 0);
589 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
591 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
594 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
596 return start_transaction(root, 0, TRANS_USERSPACE, 0);
600 * btrfs_attach_transaction() - catch the running transaction
602 * It is used when we want to commit the current the transaction, but
603 * don't want to start a new one.
605 * Note: If this function return -ENOENT, it just means there is no
606 * running transaction. But it is possible that the inactive transaction
607 * is still in the memory, not fully on disk. If you hope there is no
608 * inactive transaction in the fs when -ENOENT is returned, you should
610 * btrfs_attach_transaction_barrier()
612 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
614 return start_transaction(root, 0, TRANS_ATTACH, 0);
618 * btrfs_attach_transaction_barrier() - catch the running transaction
620 * It is similar to the above function, the differentia is this one
621 * will wait for all the inactive transactions until they fully
624 struct btrfs_trans_handle *
625 btrfs_attach_transaction_barrier(struct btrfs_root *root)
627 struct btrfs_trans_handle *trans;
629 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
630 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
631 btrfs_wait_for_commit(root, 0);
636 /* wait for a transaction commit to be fully complete */
637 static noinline void wait_for_commit(struct btrfs_root *root,
638 struct btrfs_transaction *commit)
640 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
643 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
645 struct btrfs_transaction *cur_trans = NULL, *t;
649 if (transid <= root->fs_info->last_trans_committed)
652 /* find specified transaction */
653 spin_lock(&root->fs_info->trans_lock);
654 list_for_each_entry(t, &root->fs_info->trans_list, list) {
655 if (t->transid == transid) {
657 atomic_inc(&cur_trans->use_count);
661 if (t->transid > transid) {
666 spin_unlock(&root->fs_info->trans_lock);
669 * The specified transaction doesn't exist, or we
670 * raced with btrfs_commit_transaction
673 if (transid > root->fs_info->last_trans_committed)
678 /* find newest transaction that is committing | committed */
679 spin_lock(&root->fs_info->trans_lock);
680 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
682 if (t->state >= TRANS_STATE_COMMIT_START) {
683 if (t->state == TRANS_STATE_COMPLETED)
686 atomic_inc(&cur_trans->use_count);
690 spin_unlock(&root->fs_info->trans_lock);
692 goto out; /* nothing committing|committed */
695 wait_for_commit(root, cur_trans);
696 btrfs_put_transaction(cur_trans);
701 void btrfs_throttle(struct btrfs_root *root)
703 if (!atomic_read(&root->fs_info->open_ioctl_trans))
704 wait_current_trans(root);
707 static int should_end_transaction(struct btrfs_trans_handle *trans,
708 struct btrfs_root *root)
710 if (root->fs_info->global_block_rsv.space_info->full &&
711 btrfs_check_space_for_delayed_refs(trans, root))
714 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
717 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
718 struct btrfs_root *root)
720 struct btrfs_transaction *cur_trans = trans->transaction;
725 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
726 cur_trans->delayed_refs.flushing)
729 updates = trans->delayed_ref_updates;
730 trans->delayed_ref_updates = 0;
732 err = btrfs_run_delayed_refs(trans, root, updates * 2);
733 if (err) /* Error code will also eval true */
737 return should_end_transaction(trans, root);
740 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
741 struct btrfs_root *root, int throttle)
743 struct btrfs_transaction *cur_trans = trans->transaction;
744 struct btrfs_fs_info *info = root->fs_info;
745 unsigned long cur = trans->delayed_ref_updates;
746 int lock = (trans->type != TRANS_JOIN_NOLOCK);
748 int must_run_delayed_refs = 0;
750 if (trans->use_count > 1) {
752 trans->block_rsv = trans->orig_rsv;
756 btrfs_trans_release_metadata(trans, root);
757 trans->block_rsv = NULL;
759 if (!list_empty(&trans->new_bgs))
760 btrfs_create_pending_block_groups(trans, root);
762 if (!list_empty(&trans->ordered)) {
763 spin_lock(&info->trans_lock);
764 list_splice(&trans->ordered, &cur_trans->pending_ordered);
765 spin_unlock(&info->trans_lock);
768 trans->delayed_ref_updates = 0;
770 must_run_delayed_refs =
771 btrfs_should_throttle_delayed_refs(trans, root);
772 cur = max_t(unsigned long, cur, 32);
775 * don't make the caller wait if they are from a NOLOCK
776 * or ATTACH transaction, it will deadlock with commit
778 if (must_run_delayed_refs == 1 &&
779 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
780 must_run_delayed_refs = 2;
783 if (trans->qgroup_reserved) {
785 * the same root has to be passed here between start_transaction
786 * and end_transaction. Subvolume quota depends on this.
788 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
789 trans->qgroup_reserved = 0;
792 btrfs_trans_release_metadata(trans, root);
793 trans->block_rsv = NULL;
795 if (!list_empty(&trans->new_bgs))
796 btrfs_create_pending_block_groups(trans, root);
798 btrfs_trans_release_chunk_metadata(trans);
800 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
801 should_end_transaction(trans, root) &&
802 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
803 spin_lock(&info->trans_lock);
804 if (cur_trans->state == TRANS_STATE_RUNNING)
805 cur_trans->state = TRANS_STATE_BLOCKED;
806 spin_unlock(&info->trans_lock);
809 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
811 return btrfs_commit_transaction(trans, root);
813 wake_up_process(info->transaction_kthread);
816 if (trans->type & __TRANS_FREEZABLE)
817 sb_end_intwrite(root->fs_info->sb);
819 WARN_ON(cur_trans != info->running_transaction);
820 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
821 atomic_dec(&cur_trans->num_writers);
822 extwriter_counter_dec(cur_trans, trans->type);
825 if (waitqueue_active(&cur_trans->writer_wait))
826 wake_up(&cur_trans->writer_wait);
827 btrfs_put_transaction(cur_trans);
829 if (current->journal_info == trans)
830 current->journal_info = NULL;
833 btrfs_run_delayed_iputs(root);
835 if (trans->aborted ||
836 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
837 wake_up_process(info->transaction_kthread);
840 assert_qgroups_uptodate(trans);
842 kmem_cache_free(btrfs_trans_handle_cachep, trans);
843 if (must_run_delayed_refs) {
844 btrfs_async_run_delayed_refs(root, cur,
845 must_run_delayed_refs == 1);
850 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
851 struct btrfs_root *root)
853 return __btrfs_end_transaction(trans, root, 0);
856 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
857 struct btrfs_root *root)
859 return __btrfs_end_transaction(trans, root, 1);
863 * when btree blocks are allocated, they have some corresponding bits set for
864 * them in one of two extent_io trees. This is used to make sure all of
865 * those extents are sent to disk but does not wait on them
867 int btrfs_write_marked_extents(struct btrfs_root *root,
868 struct extent_io_tree *dirty_pages, int mark)
872 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
873 struct extent_state *cached_state = NULL;
877 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
878 mark, &cached_state)) {
879 bool wait_writeback = false;
881 err = convert_extent_bit(dirty_pages, start, end,
883 mark, &cached_state, GFP_NOFS);
885 * convert_extent_bit can return -ENOMEM, which is most of the
886 * time a temporary error. So when it happens, ignore the error
887 * and wait for writeback of this range to finish - because we
888 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
889 * to btrfs_wait_marked_extents() would not know that writeback
890 * for this range started and therefore wouldn't wait for it to
891 * finish - we don't want to commit a superblock that points to
892 * btree nodes/leafs for which writeback hasn't finished yet
893 * (and without errors).
894 * We cleanup any entries left in the io tree when committing
895 * the transaction (through clear_btree_io_tree()).
897 if (err == -ENOMEM) {
899 wait_writeback = true;
902 err = filemap_fdatawrite_range(mapping, start, end);
905 else if (wait_writeback)
906 werr = filemap_fdatawait_range(mapping, start, end);
907 free_extent_state(cached_state);
916 * when btree blocks are allocated, they have some corresponding bits set for
917 * them in one of two extent_io trees. This is used to make sure all of
918 * those extents are on disk for transaction or log commit. We wait
919 * on all the pages and clear them from the dirty pages state tree
921 int btrfs_wait_marked_extents(struct btrfs_root *root,
922 struct extent_io_tree *dirty_pages, int mark)
926 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
927 struct extent_state *cached_state = NULL;
930 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
933 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
934 EXTENT_NEED_WAIT, &cached_state)) {
936 * Ignore -ENOMEM errors returned by clear_extent_bit().
937 * When committing the transaction, we'll remove any entries
938 * left in the io tree. For a log commit, we don't remove them
939 * after committing the log because the tree can be accessed
940 * concurrently - we do it only at transaction commit time when
941 * it's safe to do it (through clear_btree_io_tree()).
943 err = clear_extent_bit(dirty_pages, start, end,
945 0, 0, &cached_state, GFP_NOFS);
949 err = filemap_fdatawait_range(mapping, start, end);
952 free_extent_state(cached_state);
960 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
961 if ((mark & EXTENT_DIRTY) &&
962 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
963 &btree_ino->runtime_flags))
966 if ((mark & EXTENT_NEW) &&
967 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
968 &btree_ino->runtime_flags))
971 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
972 &btree_ino->runtime_flags))
983 * when btree blocks are allocated, they have some corresponding bits set for
984 * them in one of two extent_io trees. This is used to make sure all of
985 * those extents are on disk for transaction or log commit
987 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
988 struct extent_io_tree *dirty_pages, int mark)
992 struct blk_plug plug;
994 blk_start_plug(&plug);
995 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
996 blk_finish_plug(&plug);
997 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
1006 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
1007 struct btrfs_root *root)
1011 ret = btrfs_write_and_wait_marked_extents(root,
1012 &trans->transaction->dirty_pages,
1014 clear_btree_io_tree(&trans->transaction->dirty_pages);
1020 * this is used to update the root pointer in the tree of tree roots.
1022 * But, in the case of the extent allocation tree, updating the root
1023 * pointer may allocate blocks which may change the root of the extent
1026 * So, this loops and repeats and makes sure the cowonly root didn't
1027 * change while the root pointer was being updated in the metadata.
1029 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1030 struct btrfs_root *root)
1033 u64 old_root_bytenr;
1035 struct btrfs_root *tree_root = root->fs_info->tree_root;
1037 old_root_used = btrfs_root_used(&root->root_item);
1040 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1041 if (old_root_bytenr == root->node->start &&
1042 old_root_used == btrfs_root_used(&root->root_item))
1045 btrfs_set_root_node(&root->root_item, root->node);
1046 ret = btrfs_update_root(trans, tree_root,
1052 old_root_used = btrfs_root_used(&root->root_item);
1059 * update all the cowonly tree roots on disk
1061 * The error handling in this function may not be obvious. Any of the
1062 * failures will cause the file system to go offline. We still need
1063 * to clean up the delayed refs.
1065 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1066 struct btrfs_root *root)
1068 struct btrfs_fs_info *fs_info = root->fs_info;
1069 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1070 struct list_head *io_bgs = &trans->transaction->io_bgs;
1071 struct list_head *next;
1072 struct extent_buffer *eb;
1075 eb = btrfs_lock_root_node(fs_info->tree_root);
1076 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1078 btrfs_tree_unlock(eb);
1079 free_extent_buffer(eb);
1084 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1088 ret = btrfs_run_dev_stats(trans, root->fs_info);
1091 ret = btrfs_run_dev_replace(trans, root->fs_info);
1094 ret = btrfs_run_qgroups(trans, root->fs_info);
1098 ret = btrfs_setup_space_cache(trans, root);
1102 /* run_qgroups might have added some more refs */
1103 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1107 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1108 next = fs_info->dirty_cowonly_roots.next;
1109 list_del_init(next);
1110 root = list_entry(next, struct btrfs_root, dirty_list);
1111 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1113 if (root != fs_info->extent_root)
1114 list_add_tail(&root->dirty_list,
1115 &trans->transaction->switch_commits);
1116 ret = update_cowonly_root(trans, root);
1119 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1124 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1125 ret = btrfs_write_dirty_block_groups(trans, root);
1128 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1133 if (!list_empty(&fs_info->dirty_cowonly_roots))
1136 list_add_tail(&fs_info->extent_root->dirty_list,
1137 &trans->transaction->switch_commits);
1138 btrfs_after_dev_replace_commit(fs_info);
1144 * dead roots are old snapshots that need to be deleted. This allocates
1145 * a dirty root struct and adds it into the list of dead roots that need to
1148 void btrfs_add_dead_root(struct btrfs_root *root)
1150 spin_lock(&root->fs_info->trans_lock);
1151 if (list_empty(&root->root_list))
1152 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1153 spin_unlock(&root->fs_info->trans_lock);
1157 * update all the cowonly tree roots on disk
1159 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root)
1162 struct btrfs_root *gang[8];
1163 struct btrfs_fs_info *fs_info = root->fs_info;
1168 spin_lock(&fs_info->fs_roots_radix_lock);
1170 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1173 BTRFS_ROOT_TRANS_TAG);
1176 for (i = 0; i < ret; i++) {
1178 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1179 (unsigned long)root->root_key.objectid,
1180 BTRFS_ROOT_TRANS_TAG);
1181 spin_unlock(&fs_info->fs_roots_radix_lock);
1183 btrfs_free_log(trans, root);
1184 btrfs_update_reloc_root(trans, root);
1185 btrfs_orphan_commit_root(trans, root);
1187 btrfs_save_ino_cache(root, trans);
1189 /* see comments in should_cow_block() */
1190 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1191 smp_mb__after_atomic();
1193 if (root->commit_root != root->node) {
1194 list_add_tail(&root->dirty_list,
1195 &trans->transaction->switch_commits);
1196 btrfs_set_root_node(&root->root_item,
1200 err = btrfs_update_root(trans, fs_info->tree_root,
1203 spin_lock(&fs_info->fs_roots_radix_lock);
1208 spin_unlock(&fs_info->fs_roots_radix_lock);
1213 * defrag a given btree.
1214 * Every leaf in the btree is read and defragged.
1216 int btrfs_defrag_root(struct btrfs_root *root)
1218 struct btrfs_fs_info *info = root->fs_info;
1219 struct btrfs_trans_handle *trans;
1222 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1226 trans = btrfs_start_transaction(root, 0);
1228 return PTR_ERR(trans);
1230 ret = btrfs_defrag_leaves(trans, root);
1232 btrfs_end_transaction(trans, root);
1233 btrfs_btree_balance_dirty(info->tree_root);
1236 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1239 if (btrfs_defrag_cancelled(root->fs_info)) {
1240 pr_debug("BTRFS: defrag_root cancelled\n");
1245 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1250 * new snapshots need to be created at a very specific time in the
1251 * transaction commit. This does the actual creation.
1254 * If the error which may affect the commitment of the current transaction
1255 * happens, we should return the error number. If the error which just affect
1256 * the creation of the pending snapshots, just return 0.
1258 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1259 struct btrfs_fs_info *fs_info,
1260 struct btrfs_pending_snapshot *pending)
1262 struct btrfs_key key;
1263 struct btrfs_root_item *new_root_item;
1264 struct btrfs_root *tree_root = fs_info->tree_root;
1265 struct btrfs_root *root = pending->root;
1266 struct btrfs_root *parent_root;
1267 struct btrfs_block_rsv *rsv;
1268 struct inode *parent_inode;
1269 struct btrfs_path *path;
1270 struct btrfs_dir_item *dir_item;
1271 struct dentry *dentry;
1272 struct extent_buffer *tmp;
1273 struct extent_buffer *old;
1274 struct timespec cur_time = CURRENT_TIME;
1282 path = btrfs_alloc_path();
1284 pending->error = -ENOMEM;
1288 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1289 if (!new_root_item) {
1290 pending->error = -ENOMEM;
1291 goto root_item_alloc_fail;
1294 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1296 goto no_free_objectid;
1298 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1300 if (to_reserve > 0) {
1301 pending->error = btrfs_block_rsv_add(root,
1302 &pending->block_rsv,
1304 BTRFS_RESERVE_NO_FLUSH);
1306 goto no_free_objectid;
1309 key.objectid = objectid;
1310 key.offset = (u64)-1;
1311 key.type = BTRFS_ROOT_ITEM_KEY;
1313 rsv = trans->block_rsv;
1314 trans->block_rsv = &pending->block_rsv;
1315 trans->bytes_reserved = trans->block_rsv->reserved;
1317 dentry = pending->dentry;
1318 parent_inode = pending->dir;
1319 parent_root = BTRFS_I(parent_inode)->root;
1320 record_root_in_trans(trans, parent_root);
1323 * insert the directory item
1325 ret = btrfs_set_inode_index(parent_inode, &index);
1326 BUG_ON(ret); /* -ENOMEM */
1328 /* check if there is a file/dir which has the same name. */
1329 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1330 btrfs_ino(parent_inode),
1331 dentry->d_name.name,
1332 dentry->d_name.len, 0);
1333 if (dir_item != NULL && !IS_ERR(dir_item)) {
1334 pending->error = -EEXIST;
1335 goto dir_item_existed;
1336 } else if (IS_ERR(dir_item)) {
1337 ret = PTR_ERR(dir_item);
1338 btrfs_abort_transaction(trans, root, ret);
1341 btrfs_release_path(path);
1344 * pull in the delayed directory update
1345 * and the delayed inode item
1346 * otherwise we corrupt the FS during
1349 ret = btrfs_run_delayed_items(trans, root);
1350 if (ret) { /* Transaction aborted */
1351 btrfs_abort_transaction(trans, root, ret);
1355 record_root_in_trans(trans, root);
1356 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1357 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1358 btrfs_check_and_init_root_item(new_root_item);
1360 root_flags = btrfs_root_flags(new_root_item);
1361 if (pending->readonly)
1362 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1364 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1365 btrfs_set_root_flags(new_root_item, root_flags);
1367 btrfs_set_root_generation_v2(new_root_item,
1369 uuid_le_gen(&new_uuid);
1370 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1371 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1373 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1374 memset(new_root_item->received_uuid, 0,
1375 sizeof(new_root_item->received_uuid));
1376 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1377 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1378 btrfs_set_root_stransid(new_root_item, 0);
1379 btrfs_set_root_rtransid(new_root_item, 0);
1381 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1382 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1383 btrfs_set_root_otransid(new_root_item, trans->transid);
1385 old = btrfs_lock_root_node(root);
1386 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1388 btrfs_tree_unlock(old);
1389 free_extent_buffer(old);
1390 btrfs_abort_transaction(trans, root, ret);
1394 btrfs_set_lock_blocking(old);
1396 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1397 /* clean up in any case */
1398 btrfs_tree_unlock(old);
1399 free_extent_buffer(old);
1401 btrfs_abort_transaction(trans, root, ret);
1406 * We need to flush delayed refs in order to make sure all of our quota
1407 * operations have been done before we call btrfs_qgroup_inherit.
1409 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1411 btrfs_abort_transaction(trans, root, ret);
1415 ret = btrfs_qgroup_inherit(trans, fs_info,
1416 root->root_key.objectid,
1417 objectid, pending->inherit);
1419 btrfs_abort_transaction(trans, root, ret);
1423 /* see comments in should_cow_block() */
1424 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1427 btrfs_set_root_node(new_root_item, tmp);
1428 /* record when the snapshot was created in key.offset */
1429 key.offset = trans->transid;
1430 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1431 btrfs_tree_unlock(tmp);
1432 free_extent_buffer(tmp);
1434 btrfs_abort_transaction(trans, root, ret);
1439 * insert root back/forward references
1441 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1442 parent_root->root_key.objectid,
1443 btrfs_ino(parent_inode), index,
1444 dentry->d_name.name, dentry->d_name.len);
1446 btrfs_abort_transaction(trans, root, ret);
1450 key.offset = (u64)-1;
1451 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1452 if (IS_ERR(pending->snap)) {
1453 ret = PTR_ERR(pending->snap);
1454 btrfs_abort_transaction(trans, root, ret);
1458 ret = btrfs_reloc_post_snapshot(trans, pending);
1460 btrfs_abort_transaction(trans, root, ret);
1464 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1466 btrfs_abort_transaction(trans, root, ret);
1470 ret = btrfs_insert_dir_item(trans, parent_root,
1471 dentry->d_name.name, dentry->d_name.len,
1473 BTRFS_FT_DIR, index);
1474 /* We have check then name at the beginning, so it is impossible. */
1475 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1477 btrfs_abort_transaction(trans, root, ret);
1481 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1482 dentry->d_name.len * 2);
1483 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1484 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1486 btrfs_abort_transaction(trans, root, ret);
1489 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1490 BTRFS_UUID_KEY_SUBVOL, objectid);
1492 btrfs_abort_transaction(trans, root, ret);
1495 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1496 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1497 new_root_item->received_uuid,
1498 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1500 if (ret && ret != -EEXIST) {
1501 btrfs_abort_transaction(trans, root, ret);
1506 pending->error = ret;
1508 trans->block_rsv = rsv;
1509 trans->bytes_reserved = 0;
1511 kfree(new_root_item);
1512 root_item_alloc_fail:
1513 btrfs_free_path(path);
1518 * create all the snapshots we've scheduled for creation
1520 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1521 struct btrfs_fs_info *fs_info)
1523 struct btrfs_pending_snapshot *pending, *next;
1524 struct list_head *head = &trans->transaction->pending_snapshots;
1527 list_for_each_entry_safe(pending, next, head, list) {
1528 list_del(&pending->list);
1529 ret = create_pending_snapshot(trans, fs_info, pending);
1536 static void update_super_roots(struct btrfs_root *root)
1538 struct btrfs_root_item *root_item;
1539 struct btrfs_super_block *super;
1541 super = root->fs_info->super_copy;
1543 root_item = &root->fs_info->chunk_root->root_item;
1544 super->chunk_root = root_item->bytenr;
1545 super->chunk_root_generation = root_item->generation;
1546 super->chunk_root_level = root_item->level;
1548 root_item = &root->fs_info->tree_root->root_item;
1549 super->root = root_item->bytenr;
1550 super->generation = root_item->generation;
1551 super->root_level = root_item->level;
1552 if (btrfs_test_opt(root, SPACE_CACHE))
1553 super->cache_generation = root_item->generation;
1554 if (root->fs_info->update_uuid_tree_gen)
1555 super->uuid_tree_generation = root_item->generation;
1558 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1560 struct btrfs_transaction *trans;
1563 spin_lock(&info->trans_lock);
1564 trans = info->running_transaction;
1566 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1567 spin_unlock(&info->trans_lock);
1571 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1573 struct btrfs_transaction *trans;
1576 spin_lock(&info->trans_lock);
1577 trans = info->running_transaction;
1579 ret = is_transaction_blocked(trans);
1580 spin_unlock(&info->trans_lock);
1585 * wait for the current transaction commit to start and block subsequent
1588 static void wait_current_trans_commit_start(struct btrfs_root *root,
1589 struct btrfs_transaction *trans)
1591 wait_event(root->fs_info->transaction_blocked_wait,
1592 trans->state >= TRANS_STATE_COMMIT_START ||
1597 * wait for the current transaction to start and then become unblocked.
1600 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1601 struct btrfs_transaction *trans)
1603 wait_event(root->fs_info->transaction_wait,
1604 trans->state >= TRANS_STATE_UNBLOCKED ||
1609 * commit transactions asynchronously. once btrfs_commit_transaction_async
1610 * returns, any subsequent transaction will not be allowed to join.
1612 struct btrfs_async_commit {
1613 struct btrfs_trans_handle *newtrans;
1614 struct btrfs_root *root;
1615 struct work_struct work;
1618 static void do_async_commit(struct work_struct *work)
1620 struct btrfs_async_commit *ac =
1621 container_of(work, struct btrfs_async_commit, work);
1624 * We've got freeze protection passed with the transaction.
1625 * Tell lockdep about it.
1627 if (ac->newtrans->type & __TRANS_FREEZABLE)
1629 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1632 current->journal_info = ac->newtrans;
1634 btrfs_commit_transaction(ac->newtrans, ac->root);
1638 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1639 struct btrfs_root *root,
1640 int wait_for_unblock)
1642 struct btrfs_async_commit *ac;
1643 struct btrfs_transaction *cur_trans;
1645 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1649 INIT_WORK(&ac->work, do_async_commit);
1651 ac->newtrans = btrfs_join_transaction(root);
1652 if (IS_ERR(ac->newtrans)) {
1653 int err = PTR_ERR(ac->newtrans);
1658 /* take transaction reference */
1659 cur_trans = trans->transaction;
1660 atomic_inc(&cur_trans->use_count);
1662 btrfs_end_transaction(trans, root);
1665 * Tell lockdep we've released the freeze rwsem, since the
1666 * async commit thread will be the one to unlock it.
1668 if (ac->newtrans->type & __TRANS_FREEZABLE)
1670 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1673 schedule_work(&ac->work);
1675 /* wait for transaction to start and unblock */
1676 if (wait_for_unblock)
1677 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1679 wait_current_trans_commit_start(root, cur_trans);
1681 if (current->journal_info == trans)
1682 current->journal_info = NULL;
1684 btrfs_put_transaction(cur_trans);
1689 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1690 struct btrfs_root *root, int err)
1692 struct btrfs_transaction *cur_trans = trans->transaction;
1695 WARN_ON(trans->use_count > 1);
1697 btrfs_abort_transaction(trans, root, err);
1699 spin_lock(&root->fs_info->trans_lock);
1702 * If the transaction is removed from the list, it means this
1703 * transaction has been committed successfully, so it is impossible
1704 * to call the cleanup function.
1706 BUG_ON(list_empty(&cur_trans->list));
1708 list_del_init(&cur_trans->list);
1709 if (cur_trans == root->fs_info->running_transaction) {
1710 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1711 spin_unlock(&root->fs_info->trans_lock);
1712 wait_event(cur_trans->writer_wait,
1713 atomic_read(&cur_trans->num_writers) == 1);
1715 spin_lock(&root->fs_info->trans_lock);
1717 spin_unlock(&root->fs_info->trans_lock);
1719 btrfs_cleanup_one_transaction(trans->transaction, root);
1721 spin_lock(&root->fs_info->trans_lock);
1722 if (cur_trans == root->fs_info->running_transaction)
1723 root->fs_info->running_transaction = NULL;
1724 spin_unlock(&root->fs_info->trans_lock);
1726 if (trans->type & __TRANS_FREEZABLE)
1727 sb_end_intwrite(root->fs_info->sb);
1728 btrfs_put_transaction(cur_trans);
1729 btrfs_put_transaction(cur_trans);
1731 trace_btrfs_transaction_commit(root);
1733 if (current->journal_info == trans)
1734 current->journal_info = NULL;
1735 btrfs_scrub_cancel(root->fs_info);
1737 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1740 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1742 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1743 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1747 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1749 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1750 btrfs_wait_ordered_roots(fs_info, -1);
1754 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans,
1755 struct btrfs_fs_info *fs_info)
1757 struct btrfs_ordered_extent *ordered;
1759 spin_lock(&fs_info->trans_lock);
1760 while (!list_empty(&cur_trans->pending_ordered)) {
1761 ordered = list_first_entry(&cur_trans->pending_ordered,
1762 struct btrfs_ordered_extent,
1764 list_del_init(&ordered->trans_list);
1765 spin_unlock(&fs_info->trans_lock);
1767 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE,
1769 btrfs_put_ordered_extent(ordered);
1770 spin_lock(&fs_info->trans_lock);
1772 spin_unlock(&fs_info->trans_lock);
1775 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1776 struct btrfs_root *root)
1778 struct btrfs_transaction *cur_trans = trans->transaction;
1779 struct btrfs_transaction *prev_trans = NULL;
1780 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
1783 /* Stop the commit early if ->aborted is set */
1784 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1785 ret = cur_trans->aborted;
1786 btrfs_end_transaction(trans, root);
1790 /* make a pass through all the delayed refs we have so far
1791 * any runnings procs may add more while we are here
1793 ret = btrfs_run_delayed_refs(trans, root, 0);
1795 btrfs_end_transaction(trans, root);
1799 btrfs_trans_release_metadata(trans, root);
1800 trans->block_rsv = NULL;
1801 if (trans->qgroup_reserved) {
1802 btrfs_qgroup_free(root, trans->qgroup_reserved);
1803 trans->qgroup_reserved = 0;
1806 cur_trans = trans->transaction;
1809 * set the flushing flag so procs in this transaction have to
1810 * start sending their work down.
1812 cur_trans->delayed_refs.flushing = 1;
1815 if (!list_empty(&trans->new_bgs))
1816 btrfs_create_pending_block_groups(trans, root);
1818 ret = btrfs_run_delayed_refs(trans, root, 0);
1820 btrfs_end_transaction(trans, root);
1824 if (!cur_trans->dirty_bg_run) {
1827 /* this mutex is also taken before trying to set
1828 * block groups readonly. We need to make sure
1829 * that nobody has set a block group readonly
1830 * after a extents from that block group have been
1831 * allocated for cache files. btrfs_set_block_group_ro
1832 * will wait for the transaction to commit if it
1833 * finds dirty_bg_run = 1
1835 * The dirty_bg_run flag is also used to make sure only
1836 * one process starts all the block group IO. It wouldn't
1837 * hurt to have more than one go through, but there's no
1838 * real advantage to it either.
1840 mutex_lock(&root->fs_info->ro_block_group_mutex);
1841 if (!cur_trans->dirty_bg_run) {
1843 cur_trans->dirty_bg_run = 1;
1845 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1848 ret = btrfs_start_dirty_block_groups(trans, root);
1851 btrfs_end_transaction(trans, root);
1855 spin_lock(&root->fs_info->trans_lock);
1856 list_splice(&trans->ordered, &cur_trans->pending_ordered);
1857 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1858 spin_unlock(&root->fs_info->trans_lock);
1859 atomic_inc(&cur_trans->use_count);
1860 ret = btrfs_end_transaction(trans, root);
1862 wait_for_commit(root, cur_trans);
1864 if (unlikely(cur_trans->aborted))
1865 ret = cur_trans->aborted;
1867 btrfs_put_transaction(cur_trans);
1872 cur_trans->state = TRANS_STATE_COMMIT_START;
1873 wake_up(&root->fs_info->transaction_blocked_wait);
1875 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1876 prev_trans = list_entry(cur_trans->list.prev,
1877 struct btrfs_transaction, list);
1878 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1879 atomic_inc(&prev_trans->use_count);
1880 spin_unlock(&root->fs_info->trans_lock);
1882 wait_for_commit(root, prev_trans);
1884 btrfs_put_transaction(prev_trans);
1886 spin_unlock(&root->fs_info->trans_lock);
1889 spin_unlock(&root->fs_info->trans_lock);
1892 extwriter_counter_dec(cur_trans, trans->type);
1894 ret = btrfs_start_delalloc_flush(root->fs_info);
1896 goto cleanup_transaction;
1898 ret = btrfs_run_delayed_items(trans, root);
1900 goto cleanup_transaction;
1902 wait_event(cur_trans->writer_wait,
1903 extwriter_counter_read(cur_trans) == 0);
1905 /* some pending stuffs might be added after the previous flush. */
1906 ret = btrfs_run_delayed_items(trans, root);
1908 goto cleanup_transaction;
1910 btrfs_wait_delalloc_flush(root->fs_info);
1912 btrfs_wait_pending_ordered(cur_trans, root->fs_info);
1914 btrfs_scrub_pause(root);
1916 * Ok now we need to make sure to block out any other joins while we
1917 * commit the transaction. We could have started a join before setting
1918 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1920 spin_lock(&root->fs_info->trans_lock);
1921 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1922 spin_unlock(&root->fs_info->trans_lock);
1923 wait_event(cur_trans->writer_wait,
1924 atomic_read(&cur_trans->num_writers) == 1);
1926 /* ->aborted might be set after the previous check, so check it */
1927 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1928 ret = cur_trans->aborted;
1929 goto scrub_continue;
1932 * the reloc mutex makes sure that we stop
1933 * the balancing code from coming in and moving
1934 * extents around in the middle of the commit
1936 mutex_lock(&root->fs_info->reloc_mutex);
1939 * We needn't worry about the delayed items because we will
1940 * deal with them in create_pending_snapshot(), which is the
1941 * core function of the snapshot creation.
1943 ret = create_pending_snapshots(trans, root->fs_info);
1945 mutex_unlock(&root->fs_info->reloc_mutex);
1946 goto scrub_continue;
1950 * We insert the dir indexes of the snapshots and update the inode
1951 * of the snapshots' parents after the snapshot creation, so there
1952 * are some delayed items which are not dealt with. Now deal with
1955 * We needn't worry that this operation will corrupt the snapshots,
1956 * because all the tree which are snapshoted will be forced to COW
1957 * the nodes and leaves.
1959 ret = btrfs_run_delayed_items(trans, root);
1961 mutex_unlock(&root->fs_info->reloc_mutex);
1962 goto scrub_continue;
1965 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1967 mutex_unlock(&root->fs_info->reloc_mutex);
1968 goto scrub_continue;
1971 /* Reocrd old roots for later qgroup accounting */
1972 ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info);
1974 mutex_unlock(&root->fs_info->reloc_mutex);
1975 goto scrub_continue;
1979 * make sure none of the code above managed to slip in a
1982 btrfs_assert_delayed_root_empty(root);
1984 WARN_ON(cur_trans != trans->transaction);
1986 /* btrfs_commit_tree_roots is responsible for getting the
1987 * various roots consistent with each other. Every pointer
1988 * in the tree of tree roots has to point to the most up to date
1989 * root for every subvolume and other tree. So, we have to keep
1990 * the tree logging code from jumping in and changing any
1993 * At this point in the commit, there can't be any tree-log
1994 * writers, but a little lower down we drop the trans mutex
1995 * and let new people in. By holding the tree_log_mutex
1996 * from now until after the super is written, we avoid races
1997 * with the tree-log code.
1999 mutex_lock(&root->fs_info->tree_log_mutex);
2001 ret = commit_fs_roots(trans, root);
2003 mutex_unlock(&root->fs_info->tree_log_mutex);
2004 mutex_unlock(&root->fs_info->reloc_mutex);
2005 goto scrub_continue;
2009 * Since the transaction is done, we can apply the pending changes
2010 * before the next transaction.
2012 btrfs_apply_pending_changes(root->fs_info);
2014 /* commit_fs_roots gets rid of all the tree log roots, it is now
2015 * safe to free the root of tree log roots
2017 btrfs_free_log_root_tree(trans, root->fs_info);
2020 * Since fs roots are all committed, we can get a quite accurate
2021 * new_roots. So let's do quota accounting.
2023 ret = btrfs_qgroup_account_extents(trans, root->fs_info);
2025 mutex_unlock(&root->fs_info->tree_log_mutex);
2026 mutex_unlock(&root->fs_info->reloc_mutex);
2027 goto scrub_continue;
2030 ret = commit_cowonly_roots(trans, root);
2032 mutex_unlock(&root->fs_info->tree_log_mutex);
2033 mutex_unlock(&root->fs_info->reloc_mutex);
2034 goto scrub_continue;
2038 * The tasks which save the space cache and inode cache may also
2039 * update ->aborted, check it.
2041 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2042 ret = cur_trans->aborted;
2043 mutex_unlock(&root->fs_info->tree_log_mutex);
2044 mutex_unlock(&root->fs_info->reloc_mutex);
2045 goto scrub_continue;
2048 btrfs_prepare_extent_commit(trans, root);
2050 cur_trans = root->fs_info->running_transaction;
2052 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2053 root->fs_info->tree_root->node);
2054 list_add_tail(&root->fs_info->tree_root->dirty_list,
2055 &cur_trans->switch_commits);
2057 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2058 root->fs_info->chunk_root->node);
2059 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2060 &cur_trans->switch_commits);
2062 switch_commit_roots(cur_trans, root->fs_info);
2064 assert_qgroups_uptodate(trans);
2065 ASSERT(list_empty(&cur_trans->dirty_bgs));
2066 ASSERT(list_empty(&cur_trans->io_bgs));
2067 update_super_roots(root);
2069 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2070 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
2071 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2072 sizeof(*root->fs_info->super_copy));
2074 btrfs_update_commit_device_size(root->fs_info);
2075 btrfs_update_commit_device_bytes_used(root, cur_trans);
2077 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
2078 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
2080 btrfs_trans_release_chunk_metadata(trans);
2082 spin_lock(&root->fs_info->trans_lock);
2083 cur_trans->state = TRANS_STATE_UNBLOCKED;
2084 root->fs_info->running_transaction = NULL;
2085 spin_unlock(&root->fs_info->trans_lock);
2086 mutex_unlock(&root->fs_info->reloc_mutex);
2088 wake_up(&root->fs_info->transaction_wait);
2090 ret = btrfs_write_and_wait_transaction(trans, root);
2092 btrfs_error(root->fs_info, ret,
2093 "Error while writing out transaction");
2094 mutex_unlock(&root->fs_info->tree_log_mutex);
2095 goto scrub_continue;
2098 ret = write_ctree_super(trans, root, 0);
2100 mutex_unlock(&root->fs_info->tree_log_mutex);
2101 goto scrub_continue;
2105 * the super is written, we can safely allow the tree-loggers
2106 * to go about their business
2108 mutex_unlock(&root->fs_info->tree_log_mutex);
2110 btrfs_finish_extent_commit(trans, root);
2112 if (cur_trans->have_free_bgs)
2113 btrfs_clear_space_info_full(root->fs_info);
2115 root->fs_info->last_trans_committed = cur_trans->transid;
2117 * We needn't acquire the lock here because there is no other task
2118 * which can change it.
2120 cur_trans->state = TRANS_STATE_COMPLETED;
2121 wake_up(&cur_trans->commit_wait);
2123 spin_lock(&root->fs_info->trans_lock);
2124 list_del_init(&cur_trans->list);
2125 spin_unlock(&root->fs_info->trans_lock);
2127 btrfs_put_transaction(cur_trans);
2128 btrfs_put_transaction(cur_trans);
2130 if (trans->type & __TRANS_FREEZABLE)
2131 sb_end_intwrite(root->fs_info->sb);
2133 trace_btrfs_transaction_commit(root);
2135 btrfs_scrub_continue(root);
2137 if (current->journal_info == trans)
2138 current->journal_info = NULL;
2140 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2142 if (current != root->fs_info->transaction_kthread)
2143 btrfs_run_delayed_iputs(root);
2148 btrfs_scrub_continue(root);
2149 cleanup_transaction:
2150 btrfs_trans_release_metadata(trans, root);
2151 btrfs_trans_release_chunk_metadata(trans);
2152 trans->block_rsv = NULL;
2153 if (trans->qgroup_reserved) {
2154 btrfs_qgroup_free(root, trans->qgroup_reserved);
2155 trans->qgroup_reserved = 0;
2157 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
2158 if (current->journal_info == trans)
2159 current->journal_info = NULL;
2160 cleanup_transaction(trans, root, ret);
2166 * return < 0 if error
2167 * 0 if there are no more dead_roots at the time of call
2168 * 1 there are more to be processed, call me again
2170 * The return value indicates there are certainly more snapshots to delete, but
2171 * if there comes a new one during processing, it may return 0. We don't mind,
2172 * because btrfs_commit_super will poke cleaner thread and it will process it a
2173 * few seconds later.
2175 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2178 struct btrfs_fs_info *fs_info = root->fs_info;
2180 spin_lock(&fs_info->trans_lock);
2181 if (list_empty(&fs_info->dead_roots)) {
2182 spin_unlock(&fs_info->trans_lock);
2185 root = list_first_entry(&fs_info->dead_roots,
2186 struct btrfs_root, root_list);
2187 list_del_init(&root->root_list);
2188 spin_unlock(&fs_info->trans_lock);
2190 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2192 btrfs_kill_all_delayed_nodes(root);
2194 if (btrfs_header_backref_rev(root->node) <
2195 BTRFS_MIXED_BACKREF_REV)
2196 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2198 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2200 return (ret < 0) ? 0 : 1;
2203 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2208 prev = xchg(&fs_info->pending_changes, 0);
2212 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2214 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2217 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2219 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2222 bit = 1 << BTRFS_PENDING_COMMIT;
2224 btrfs_debug(fs_info, "pending commit done");
2229 "unknown pending changes left 0x%lx, ignoring", prev);