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 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
229 cur_trans->delayed_refs.num_heads_ready = 0;
230 cur_trans->delayed_refs.pending_csums = 0;
231 cur_trans->delayed_refs.num_heads = 0;
232 cur_trans->delayed_refs.flushing = 0;
233 cur_trans->delayed_refs.run_delayed_start = 0;
236 * although the tree mod log is per file system and not per transaction,
237 * the log must never go across transaction boundaries.
240 if (!list_empty(&fs_info->tree_mod_seq_list))
241 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
242 "creating a fresh transaction\n");
243 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
244 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
245 "creating a fresh transaction\n");
246 atomic64_set(&fs_info->tree_mod_seq, 0);
248 spin_lock_init(&cur_trans->delayed_refs.lock);
250 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
251 INIT_LIST_HEAD(&cur_trans->pending_chunks);
252 INIT_LIST_HEAD(&cur_trans->switch_commits);
253 INIT_LIST_HEAD(&cur_trans->pending_ordered);
254 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
255 INIT_LIST_HEAD(&cur_trans->io_bgs);
256 mutex_init(&cur_trans->cache_write_mutex);
257 cur_trans->num_dirty_bgs = 0;
258 spin_lock_init(&cur_trans->dirty_bgs_lock);
259 list_add_tail(&cur_trans->list, &fs_info->trans_list);
260 extent_io_tree_init(&cur_trans->dirty_pages,
261 fs_info->btree_inode->i_mapping);
262 fs_info->generation++;
263 cur_trans->transid = fs_info->generation;
264 fs_info->running_transaction = cur_trans;
265 cur_trans->aborted = 0;
266 spin_unlock(&fs_info->trans_lock);
272 * this does all the record keeping required to make sure that a reference
273 * counted root is properly recorded in a given transaction. This is required
274 * to make sure the old root from before we joined the transaction is deleted
275 * when the transaction commits
277 static int record_root_in_trans(struct btrfs_trans_handle *trans,
278 struct btrfs_root *root)
280 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
281 root->last_trans < trans->transid) {
282 WARN_ON(root == root->fs_info->extent_root);
283 WARN_ON(root->commit_root != root->node);
286 * see below for IN_TRANS_SETUP usage rules
287 * we have the reloc mutex held now, so there
288 * is only one writer in this function
290 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
292 /* make sure readers find IN_TRANS_SETUP before
293 * they find our root->last_trans update
297 spin_lock(&root->fs_info->fs_roots_radix_lock);
298 if (root->last_trans == trans->transid) {
299 spin_unlock(&root->fs_info->fs_roots_radix_lock);
302 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
303 (unsigned long)root->root_key.objectid,
304 BTRFS_ROOT_TRANS_TAG);
305 spin_unlock(&root->fs_info->fs_roots_radix_lock);
306 root->last_trans = trans->transid;
308 /* this is pretty tricky. We don't want to
309 * take the relocation lock in btrfs_record_root_in_trans
310 * unless we're really doing the first setup for this root in
313 * Normally we'd use root->last_trans as a flag to decide
314 * if we want to take the expensive mutex.
316 * But, we have to set root->last_trans before we
317 * init the relocation root, otherwise, we trip over warnings
318 * in ctree.c. The solution used here is to flag ourselves
319 * with root IN_TRANS_SETUP. When this is 1, we're still
320 * fixing up the reloc trees and everyone must wait.
322 * When this is zero, they can trust root->last_trans and fly
323 * through btrfs_record_root_in_trans without having to take the
324 * lock. smp_wmb() makes sure that all the writes above are
325 * done before we pop in the zero below
327 btrfs_init_reloc_root(trans, root);
328 smp_mb__before_atomic();
329 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
335 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
336 struct btrfs_root *root)
338 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
342 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
346 if (root->last_trans == trans->transid &&
347 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
350 mutex_lock(&root->fs_info->reloc_mutex);
351 record_root_in_trans(trans, root);
352 mutex_unlock(&root->fs_info->reloc_mutex);
357 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
359 return (trans->state >= TRANS_STATE_BLOCKED &&
360 trans->state < TRANS_STATE_UNBLOCKED &&
364 /* wait for commit against the current transaction to become unblocked
365 * when this is done, it is safe to start a new transaction, but the current
366 * transaction might not be fully on disk.
368 static void wait_current_trans(struct btrfs_root *root)
370 struct btrfs_transaction *cur_trans;
372 spin_lock(&root->fs_info->trans_lock);
373 cur_trans = root->fs_info->running_transaction;
374 if (cur_trans && is_transaction_blocked(cur_trans)) {
375 atomic_inc(&cur_trans->use_count);
376 spin_unlock(&root->fs_info->trans_lock);
378 wait_event(root->fs_info->transaction_wait,
379 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
381 btrfs_put_transaction(cur_trans);
383 spin_unlock(&root->fs_info->trans_lock);
387 static int may_wait_transaction(struct btrfs_root *root, int type)
389 if (root->fs_info->log_root_recovering)
392 if (type == TRANS_USERSPACE)
395 if (type == TRANS_START &&
396 !atomic_read(&root->fs_info->open_ioctl_trans))
402 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
404 if (!root->fs_info->reloc_ctl ||
405 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
406 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
413 static struct btrfs_trans_handle *
414 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
415 enum btrfs_reserve_flush_enum flush)
417 struct btrfs_trans_handle *h;
418 struct btrfs_transaction *cur_trans;
420 u64 qgroup_reserved = 0;
421 bool reloc_reserved = false;
424 /* Send isn't supposed to start transactions. */
425 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
427 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
428 return ERR_PTR(-EROFS);
430 if (current->journal_info) {
431 WARN_ON(type & TRANS_EXTWRITERS);
432 h = current->journal_info;
434 WARN_ON(h->use_count > 2);
435 h->orig_rsv = h->block_rsv;
441 * Do the reservation before we join the transaction so we can do all
442 * the appropriate flushing if need be.
444 if (num_items > 0 && root != root->fs_info->chunk_root) {
445 if (root->fs_info->quota_enabled &&
446 is_fstree(root->root_key.objectid)) {
447 qgroup_reserved = num_items * root->nodesize;
448 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
453 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
455 * Do the reservation for the relocation root creation
457 if (need_reserve_reloc_root(root)) {
458 num_bytes += root->nodesize;
459 reloc_reserved = true;
462 ret = btrfs_block_rsv_add(root,
463 &root->fs_info->trans_block_rsv,
469 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
476 * If we are JOIN_NOLOCK we're already committing a transaction and
477 * waiting on this guy, so we don't need to do the sb_start_intwrite
478 * because we're already holding a ref. We need this because we could
479 * have raced in and did an fsync() on a file which can kick a commit
480 * and then we deadlock with somebody doing a freeze.
482 * If we are ATTACH, it means we just want to catch the current
483 * transaction and commit it, so we needn't do sb_start_intwrite().
485 if (type & __TRANS_FREEZABLE)
486 sb_start_intwrite(root->fs_info->sb);
488 if (may_wait_transaction(root, type))
489 wait_current_trans(root);
492 ret = join_transaction(root, type);
494 wait_current_trans(root);
495 if (unlikely(type == TRANS_ATTACH))
498 } while (ret == -EBUSY);
501 /* We must get the transaction if we are JOIN_NOLOCK. */
502 BUG_ON(type == TRANS_JOIN_NOLOCK);
506 cur_trans = root->fs_info->running_transaction;
508 h->transid = cur_trans->transid;
509 h->transaction = cur_trans;
511 h->bytes_reserved = 0;
513 h->delayed_ref_updates = 0;
519 h->qgroup_reserved = 0;
520 h->delayed_ref_elem.seq = 0;
522 h->allocating_chunk = false;
523 h->reloc_reserved = false;
525 INIT_LIST_HEAD(&h->qgroup_ref_list);
526 INIT_LIST_HEAD(&h->new_bgs);
527 INIT_LIST_HEAD(&h->ordered);
530 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
531 may_wait_transaction(root, type)) {
532 current->journal_info = h;
533 btrfs_commit_transaction(h, root);
538 trace_btrfs_space_reservation(root->fs_info, "transaction",
539 h->transid, num_bytes, 1);
540 h->block_rsv = &root->fs_info->trans_block_rsv;
541 h->bytes_reserved = num_bytes;
542 h->reloc_reserved = reloc_reserved;
544 h->qgroup_reserved = qgroup_reserved;
547 btrfs_record_root_in_trans(h, root);
549 if (!current->journal_info && type != TRANS_USERSPACE)
550 current->journal_info = h;
554 if (type & __TRANS_FREEZABLE)
555 sb_end_intwrite(root->fs_info->sb);
556 kmem_cache_free(btrfs_trans_handle_cachep, h);
559 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
563 btrfs_qgroup_free(root, qgroup_reserved);
567 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
570 return start_transaction(root, num_items, TRANS_START,
571 BTRFS_RESERVE_FLUSH_ALL);
574 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
575 struct btrfs_root *root, int num_items)
577 return start_transaction(root, num_items, TRANS_START,
578 BTRFS_RESERVE_FLUSH_LIMIT);
581 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
583 return start_transaction(root, 0, TRANS_JOIN, 0);
586 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
588 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
591 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
593 return start_transaction(root, 0, TRANS_USERSPACE, 0);
597 * btrfs_attach_transaction() - catch the running transaction
599 * It is used when we want to commit the current the transaction, but
600 * don't want to start a new one.
602 * Note: If this function return -ENOENT, it just means there is no
603 * running transaction. But it is possible that the inactive transaction
604 * is still in the memory, not fully on disk. If you hope there is no
605 * inactive transaction in the fs when -ENOENT is returned, you should
607 * btrfs_attach_transaction_barrier()
609 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
611 return start_transaction(root, 0, TRANS_ATTACH, 0);
615 * btrfs_attach_transaction_barrier() - catch the running transaction
617 * It is similar to the above function, the differentia is this one
618 * will wait for all the inactive transactions until they fully
621 struct btrfs_trans_handle *
622 btrfs_attach_transaction_barrier(struct btrfs_root *root)
624 struct btrfs_trans_handle *trans;
626 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
627 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
628 btrfs_wait_for_commit(root, 0);
633 /* wait for a transaction commit to be fully complete */
634 static noinline void wait_for_commit(struct btrfs_root *root,
635 struct btrfs_transaction *commit)
637 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
640 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
642 struct btrfs_transaction *cur_trans = NULL, *t;
646 if (transid <= root->fs_info->last_trans_committed)
649 /* find specified transaction */
650 spin_lock(&root->fs_info->trans_lock);
651 list_for_each_entry(t, &root->fs_info->trans_list, list) {
652 if (t->transid == transid) {
654 atomic_inc(&cur_trans->use_count);
658 if (t->transid > transid) {
663 spin_unlock(&root->fs_info->trans_lock);
666 * The specified transaction doesn't exist, or we
667 * raced with btrfs_commit_transaction
670 if (transid > root->fs_info->last_trans_committed)
675 /* find newest transaction that is committing | committed */
676 spin_lock(&root->fs_info->trans_lock);
677 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
679 if (t->state >= TRANS_STATE_COMMIT_START) {
680 if (t->state == TRANS_STATE_COMPLETED)
683 atomic_inc(&cur_trans->use_count);
687 spin_unlock(&root->fs_info->trans_lock);
689 goto out; /* nothing committing|committed */
692 wait_for_commit(root, cur_trans);
693 btrfs_put_transaction(cur_trans);
698 void btrfs_throttle(struct btrfs_root *root)
700 if (!atomic_read(&root->fs_info->open_ioctl_trans))
701 wait_current_trans(root);
704 static int should_end_transaction(struct btrfs_trans_handle *trans,
705 struct btrfs_root *root)
707 if (root->fs_info->global_block_rsv.space_info->full &&
708 btrfs_check_space_for_delayed_refs(trans, root))
711 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
714 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
715 struct btrfs_root *root)
717 struct btrfs_transaction *cur_trans = trans->transaction;
722 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
723 cur_trans->delayed_refs.flushing)
726 updates = trans->delayed_ref_updates;
727 trans->delayed_ref_updates = 0;
729 err = btrfs_run_delayed_refs(trans, root, updates * 2);
730 if (err) /* Error code will also eval true */
734 return should_end_transaction(trans, root);
737 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
738 struct btrfs_root *root, int throttle)
740 struct btrfs_transaction *cur_trans = trans->transaction;
741 struct btrfs_fs_info *info = root->fs_info;
742 unsigned long cur = trans->delayed_ref_updates;
743 int lock = (trans->type != TRANS_JOIN_NOLOCK);
745 int must_run_delayed_refs = 0;
747 if (trans->use_count > 1) {
749 trans->block_rsv = trans->orig_rsv;
753 btrfs_trans_release_metadata(trans, root);
754 trans->block_rsv = NULL;
756 if (!list_empty(&trans->new_bgs))
757 btrfs_create_pending_block_groups(trans, root);
759 if (!list_empty(&trans->ordered)) {
760 spin_lock(&info->trans_lock);
761 list_splice(&trans->ordered, &cur_trans->pending_ordered);
762 spin_unlock(&info->trans_lock);
765 trans->delayed_ref_updates = 0;
767 must_run_delayed_refs =
768 btrfs_should_throttle_delayed_refs(trans, root);
769 cur = max_t(unsigned long, cur, 32);
772 * don't make the caller wait if they are from a NOLOCK
773 * or ATTACH transaction, it will deadlock with commit
775 if (must_run_delayed_refs == 1 &&
776 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
777 must_run_delayed_refs = 2;
780 if (trans->qgroup_reserved) {
782 * the same root has to be passed here between start_transaction
783 * and end_transaction. Subvolume quota depends on this.
785 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
786 trans->qgroup_reserved = 0;
789 btrfs_trans_release_metadata(trans, root);
790 trans->block_rsv = NULL;
792 if (!list_empty(&trans->new_bgs))
793 btrfs_create_pending_block_groups(trans, root);
795 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
796 should_end_transaction(trans, root) &&
797 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
798 spin_lock(&info->trans_lock);
799 if (cur_trans->state == TRANS_STATE_RUNNING)
800 cur_trans->state = TRANS_STATE_BLOCKED;
801 spin_unlock(&info->trans_lock);
804 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
806 return btrfs_commit_transaction(trans, root);
808 wake_up_process(info->transaction_kthread);
811 if (trans->type & __TRANS_FREEZABLE)
812 sb_end_intwrite(root->fs_info->sb);
814 WARN_ON(cur_trans != info->running_transaction);
815 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
816 atomic_dec(&cur_trans->num_writers);
817 extwriter_counter_dec(cur_trans, trans->type);
820 if (waitqueue_active(&cur_trans->writer_wait))
821 wake_up(&cur_trans->writer_wait);
822 btrfs_put_transaction(cur_trans);
824 if (current->journal_info == trans)
825 current->journal_info = NULL;
828 btrfs_run_delayed_iputs(root);
830 if (trans->aborted ||
831 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
832 wake_up_process(info->transaction_kthread);
835 assert_qgroups_uptodate(trans);
837 kmem_cache_free(btrfs_trans_handle_cachep, trans);
838 if (must_run_delayed_refs) {
839 btrfs_async_run_delayed_refs(root, cur,
840 must_run_delayed_refs == 1);
845 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
846 struct btrfs_root *root)
848 return __btrfs_end_transaction(trans, root, 0);
851 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
852 struct btrfs_root *root)
854 return __btrfs_end_transaction(trans, root, 1);
858 * when btree blocks are allocated, they have some corresponding bits set for
859 * them in one of two extent_io trees. This is used to make sure all of
860 * those extents are sent to disk but does not wait on them
862 int btrfs_write_marked_extents(struct btrfs_root *root,
863 struct extent_io_tree *dirty_pages, int mark)
867 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
868 struct extent_state *cached_state = NULL;
872 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
873 mark, &cached_state)) {
874 bool wait_writeback = false;
876 err = convert_extent_bit(dirty_pages, start, end,
878 mark, &cached_state, GFP_NOFS);
880 * convert_extent_bit can return -ENOMEM, which is most of the
881 * time a temporary error. So when it happens, ignore the error
882 * and wait for writeback of this range to finish - because we
883 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
884 * to btrfs_wait_marked_extents() would not know that writeback
885 * for this range started and therefore wouldn't wait for it to
886 * finish - we don't want to commit a superblock that points to
887 * btree nodes/leafs for which writeback hasn't finished yet
888 * (and without errors).
889 * We cleanup any entries left in the io tree when committing
890 * the transaction (through clear_btree_io_tree()).
892 if (err == -ENOMEM) {
894 wait_writeback = true;
897 err = filemap_fdatawrite_range(mapping, start, end);
900 else if (wait_writeback)
901 werr = filemap_fdatawait_range(mapping, start, end);
902 free_extent_state(cached_state);
911 * when btree blocks are allocated, they have some corresponding bits set for
912 * them in one of two extent_io trees. This is used to make sure all of
913 * those extents are on disk for transaction or log commit. We wait
914 * on all the pages and clear them from the dirty pages state tree
916 int btrfs_wait_marked_extents(struct btrfs_root *root,
917 struct extent_io_tree *dirty_pages, int mark)
921 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
922 struct extent_state *cached_state = NULL;
925 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
928 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
929 EXTENT_NEED_WAIT, &cached_state)) {
931 * Ignore -ENOMEM errors returned by clear_extent_bit().
932 * When committing the transaction, we'll remove any entries
933 * left in the io tree. For a log commit, we don't remove them
934 * after committing the log because the tree can be accessed
935 * concurrently - we do it only at transaction commit time when
936 * it's safe to do it (through clear_btree_io_tree()).
938 err = clear_extent_bit(dirty_pages, start, end,
940 0, 0, &cached_state, GFP_NOFS);
944 err = filemap_fdatawait_range(mapping, start, end);
947 free_extent_state(cached_state);
955 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
956 if ((mark & EXTENT_DIRTY) &&
957 test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR,
958 &btree_ino->runtime_flags))
961 if ((mark & EXTENT_NEW) &&
962 test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR,
963 &btree_ino->runtime_flags))
966 if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR,
967 &btree_ino->runtime_flags))
978 * when btree blocks are allocated, they have some corresponding bits set for
979 * them in one of two extent_io trees. This is used to make sure all of
980 * those extents are on disk for transaction or log commit
982 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
983 struct extent_io_tree *dirty_pages, int mark)
987 struct blk_plug plug;
989 blk_start_plug(&plug);
990 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
991 blk_finish_plug(&plug);
992 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
1001 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
1002 struct btrfs_root *root)
1006 ret = btrfs_write_and_wait_marked_extents(root,
1007 &trans->transaction->dirty_pages,
1009 clear_btree_io_tree(&trans->transaction->dirty_pages);
1015 * this is used to update the root pointer in the tree of tree roots.
1017 * But, in the case of the extent allocation tree, updating the root
1018 * pointer may allocate blocks which may change the root of the extent
1021 * So, this loops and repeats and makes sure the cowonly root didn't
1022 * change while the root pointer was being updated in the metadata.
1024 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1025 struct btrfs_root *root)
1028 u64 old_root_bytenr;
1030 struct btrfs_root *tree_root = root->fs_info->tree_root;
1032 old_root_used = btrfs_root_used(&root->root_item);
1035 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1036 if (old_root_bytenr == root->node->start &&
1037 old_root_used == btrfs_root_used(&root->root_item))
1040 btrfs_set_root_node(&root->root_item, root->node);
1041 ret = btrfs_update_root(trans, tree_root,
1047 old_root_used = btrfs_root_used(&root->root_item);
1054 * update all the cowonly tree roots on disk
1056 * The error handling in this function may not be obvious. Any of the
1057 * failures will cause the file system to go offline. We still need
1058 * to clean up the delayed refs.
1060 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1061 struct btrfs_root *root)
1063 struct btrfs_fs_info *fs_info = root->fs_info;
1064 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1065 struct list_head *io_bgs = &trans->transaction->io_bgs;
1066 struct list_head *next;
1067 struct extent_buffer *eb;
1070 eb = btrfs_lock_root_node(fs_info->tree_root);
1071 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1073 btrfs_tree_unlock(eb);
1074 free_extent_buffer(eb);
1079 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1083 ret = btrfs_run_dev_stats(trans, root->fs_info);
1086 ret = btrfs_run_dev_replace(trans, root->fs_info);
1089 ret = btrfs_run_qgroups(trans, root->fs_info);
1093 ret = btrfs_setup_space_cache(trans, root);
1097 /* run_qgroups might have added some more refs */
1098 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1102 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1103 next = fs_info->dirty_cowonly_roots.next;
1104 list_del_init(next);
1105 root = list_entry(next, struct btrfs_root, dirty_list);
1106 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1108 if (root != fs_info->extent_root)
1109 list_add_tail(&root->dirty_list,
1110 &trans->transaction->switch_commits);
1111 ret = update_cowonly_root(trans, root);
1114 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1119 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1120 ret = btrfs_write_dirty_block_groups(trans, root);
1123 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1128 if (!list_empty(&fs_info->dirty_cowonly_roots))
1131 list_add_tail(&fs_info->extent_root->dirty_list,
1132 &trans->transaction->switch_commits);
1133 btrfs_after_dev_replace_commit(fs_info);
1139 * dead roots are old snapshots that need to be deleted. This allocates
1140 * a dirty root struct and adds it into the list of dead roots that need to
1143 void btrfs_add_dead_root(struct btrfs_root *root)
1145 spin_lock(&root->fs_info->trans_lock);
1146 if (list_empty(&root->root_list))
1147 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1148 spin_unlock(&root->fs_info->trans_lock);
1152 * update all the cowonly tree roots on disk
1154 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1155 struct btrfs_root *root)
1157 struct btrfs_root *gang[8];
1158 struct btrfs_fs_info *fs_info = root->fs_info;
1163 spin_lock(&fs_info->fs_roots_radix_lock);
1165 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1168 BTRFS_ROOT_TRANS_TAG);
1171 for (i = 0; i < ret; i++) {
1173 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1174 (unsigned long)root->root_key.objectid,
1175 BTRFS_ROOT_TRANS_TAG);
1176 spin_unlock(&fs_info->fs_roots_radix_lock);
1178 btrfs_free_log(trans, root);
1179 btrfs_update_reloc_root(trans, root);
1180 btrfs_orphan_commit_root(trans, root);
1182 btrfs_save_ino_cache(root, trans);
1184 /* see comments in should_cow_block() */
1185 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1186 smp_mb__after_atomic();
1188 if (root->commit_root != root->node) {
1189 list_add_tail(&root->dirty_list,
1190 &trans->transaction->switch_commits);
1191 btrfs_set_root_node(&root->root_item,
1195 err = btrfs_update_root(trans, fs_info->tree_root,
1198 spin_lock(&fs_info->fs_roots_radix_lock);
1203 spin_unlock(&fs_info->fs_roots_radix_lock);
1208 * defrag a given btree.
1209 * Every leaf in the btree is read and defragged.
1211 int btrfs_defrag_root(struct btrfs_root *root)
1213 struct btrfs_fs_info *info = root->fs_info;
1214 struct btrfs_trans_handle *trans;
1217 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1221 trans = btrfs_start_transaction(root, 0);
1223 return PTR_ERR(trans);
1225 ret = btrfs_defrag_leaves(trans, root);
1227 btrfs_end_transaction(trans, root);
1228 btrfs_btree_balance_dirty(info->tree_root);
1231 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1234 if (btrfs_defrag_cancelled(root->fs_info)) {
1235 pr_debug("BTRFS: defrag_root cancelled\n");
1240 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1245 * new snapshots need to be created at a very specific time in the
1246 * transaction commit. This does the actual creation.
1249 * If the error which may affect the commitment of the current transaction
1250 * happens, we should return the error number. If the error which just affect
1251 * the creation of the pending snapshots, just return 0.
1253 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1254 struct btrfs_fs_info *fs_info,
1255 struct btrfs_pending_snapshot *pending)
1257 struct btrfs_key key;
1258 struct btrfs_root_item *new_root_item;
1259 struct btrfs_root *tree_root = fs_info->tree_root;
1260 struct btrfs_root *root = pending->root;
1261 struct btrfs_root *parent_root;
1262 struct btrfs_block_rsv *rsv;
1263 struct inode *parent_inode;
1264 struct btrfs_path *path;
1265 struct btrfs_dir_item *dir_item;
1266 struct dentry *dentry;
1267 struct extent_buffer *tmp;
1268 struct extent_buffer *old;
1269 struct timespec cur_time = CURRENT_TIME;
1277 path = btrfs_alloc_path();
1279 pending->error = -ENOMEM;
1283 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1284 if (!new_root_item) {
1285 pending->error = -ENOMEM;
1286 goto root_item_alloc_fail;
1289 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1291 goto no_free_objectid;
1293 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1295 if (to_reserve > 0) {
1296 pending->error = btrfs_block_rsv_add(root,
1297 &pending->block_rsv,
1299 BTRFS_RESERVE_NO_FLUSH);
1301 goto no_free_objectid;
1304 key.objectid = objectid;
1305 key.offset = (u64)-1;
1306 key.type = BTRFS_ROOT_ITEM_KEY;
1308 rsv = trans->block_rsv;
1309 trans->block_rsv = &pending->block_rsv;
1310 trans->bytes_reserved = trans->block_rsv->reserved;
1312 dentry = pending->dentry;
1313 parent_inode = pending->dir;
1314 parent_root = BTRFS_I(parent_inode)->root;
1315 record_root_in_trans(trans, parent_root);
1318 * insert the directory item
1320 ret = btrfs_set_inode_index(parent_inode, &index);
1321 BUG_ON(ret); /* -ENOMEM */
1323 /* check if there is a file/dir which has the same name. */
1324 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1325 btrfs_ino(parent_inode),
1326 dentry->d_name.name,
1327 dentry->d_name.len, 0);
1328 if (dir_item != NULL && !IS_ERR(dir_item)) {
1329 pending->error = -EEXIST;
1330 goto dir_item_existed;
1331 } else if (IS_ERR(dir_item)) {
1332 ret = PTR_ERR(dir_item);
1333 btrfs_abort_transaction(trans, root, ret);
1336 btrfs_release_path(path);
1339 * pull in the delayed directory update
1340 * and the delayed inode item
1341 * otherwise we corrupt the FS during
1344 ret = btrfs_run_delayed_items(trans, root);
1345 if (ret) { /* Transaction aborted */
1346 btrfs_abort_transaction(trans, root, ret);
1350 record_root_in_trans(trans, root);
1351 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1352 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1353 btrfs_check_and_init_root_item(new_root_item);
1355 root_flags = btrfs_root_flags(new_root_item);
1356 if (pending->readonly)
1357 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1359 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1360 btrfs_set_root_flags(new_root_item, root_flags);
1362 btrfs_set_root_generation_v2(new_root_item,
1364 uuid_le_gen(&new_uuid);
1365 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1366 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1368 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1369 memset(new_root_item->received_uuid, 0,
1370 sizeof(new_root_item->received_uuid));
1371 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1372 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1373 btrfs_set_root_stransid(new_root_item, 0);
1374 btrfs_set_root_rtransid(new_root_item, 0);
1376 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1377 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1378 btrfs_set_root_otransid(new_root_item, trans->transid);
1380 old = btrfs_lock_root_node(root);
1381 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1383 btrfs_tree_unlock(old);
1384 free_extent_buffer(old);
1385 btrfs_abort_transaction(trans, root, ret);
1389 btrfs_set_lock_blocking(old);
1391 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1392 /* clean up in any case */
1393 btrfs_tree_unlock(old);
1394 free_extent_buffer(old);
1396 btrfs_abort_transaction(trans, root, ret);
1401 * We need to flush delayed refs in order to make sure all of our quota
1402 * operations have been done before we call btrfs_qgroup_inherit.
1404 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1406 btrfs_abort_transaction(trans, root, ret);
1410 ret = btrfs_qgroup_inherit(trans, fs_info,
1411 root->root_key.objectid,
1412 objectid, pending->inherit);
1414 btrfs_abort_transaction(trans, root, ret);
1418 /* see comments in should_cow_block() */
1419 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1422 btrfs_set_root_node(new_root_item, tmp);
1423 /* record when the snapshot was created in key.offset */
1424 key.offset = trans->transid;
1425 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1426 btrfs_tree_unlock(tmp);
1427 free_extent_buffer(tmp);
1429 btrfs_abort_transaction(trans, root, ret);
1434 * insert root back/forward references
1436 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1437 parent_root->root_key.objectid,
1438 btrfs_ino(parent_inode), index,
1439 dentry->d_name.name, dentry->d_name.len);
1441 btrfs_abort_transaction(trans, root, ret);
1445 key.offset = (u64)-1;
1446 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1447 if (IS_ERR(pending->snap)) {
1448 ret = PTR_ERR(pending->snap);
1449 btrfs_abort_transaction(trans, root, ret);
1453 ret = btrfs_reloc_post_snapshot(trans, pending);
1455 btrfs_abort_transaction(trans, root, ret);
1459 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1461 btrfs_abort_transaction(trans, root, ret);
1465 ret = btrfs_insert_dir_item(trans, parent_root,
1466 dentry->d_name.name, dentry->d_name.len,
1468 BTRFS_FT_DIR, index);
1469 /* We have check then name at the beginning, so it is impossible. */
1470 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1472 btrfs_abort_transaction(trans, root, ret);
1476 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1477 dentry->d_name.len * 2);
1478 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1479 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1481 btrfs_abort_transaction(trans, root, ret);
1484 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1485 BTRFS_UUID_KEY_SUBVOL, objectid);
1487 btrfs_abort_transaction(trans, root, ret);
1490 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1491 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1492 new_root_item->received_uuid,
1493 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1495 if (ret && ret != -EEXIST) {
1496 btrfs_abort_transaction(trans, root, ret);
1501 pending->error = ret;
1503 trans->block_rsv = rsv;
1504 trans->bytes_reserved = 0;
1506 kfree(new_root_item);
1507 root_item_alloc_fail:
1508 btrfs_free_path(path);
1513 * create all the snapshots we've scheduled for creation
1515 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1516 struct btrfs_fs_info *fs_info)
1518 struct btrfs_pending_snapshot *pending, *next;
1519 struct list_head *head = &trans->transaction->pending_snapshots;
1522 list_for_each_entry_safe(pending, next, head, list) {
1523 list_del(&pending->list);
1524 ret = create_pending_snapshot(trans, fs_info, pending);
1531 static void update_super_roots(struct btrfs_root *root)
1533 struct btrfs_root_item *root_item;
1534 struct btrfs_super_block *super;
1536 super = root->fs_info->super_copy;
1538 root_item = &root->fs_info->chunk_root->root_item;
1539 super->chunk_root = root_item->bytenr;
1540 super->chunk_root_generation = root_item->generation;
1541 super->chunk_root_level = root_item->level;
1543 root_item = &root->fs_info->tree_root->root_item;
1544 super->root = root_item->bytenr;
1545 super->generation = root_item->generation;
1546 super->root_level = root_item->level;
1547 if (btrfs_test_opt(root, SPACE_CACHE))
1548 super->cache_generation = root_item->generation;
1549 if (root->fs_info->update_uuid_tree_gen)
1550 super->uuid_tree_generation = root_item->generation;
1553 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1555 struct btrfs_transaction *trans;
1558 spin_lock(&info->trans_lock);
1559 trans = info->running_transaction;
1561 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1562 spin_unlock(&info->trans_lock);
1566 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1568 struct btrfs_transaction *trans;
1571 spin_lock(&info->trans_lock);
1572 trans = info->running_transaction;
1574 ret = is_transaction_blocked(trans);
1575 spin_unlock(&info->trans_lock);
1580 * wait for the current transaction commit to start and block subsequent
1583 static void wait_current_trans_commit_start(struct btrfs_root *root,
1584 struct btrfs_transaction *trans)
1586 wait_event(root->fs_info->transaction_blocked_wait,
1587 trans->state >= TRANS_STATE_COMMIT_START ||
1592 * wait for the current transaction to start and then become unblocked.
1595 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1596 struct btrfs_transaction *trans)
1598 wait_event(root->fs_info->transaction_wait,
1599 trans->state >= TRANS_STATE_UNBLOCKED ||
1604 * commit transactions asynchronously. once btrfs_commit_transaction_async
1605 * returns, any subsequent transaction will not be allowed to join.
1607 struct btrfs_async_commit {
1608 struct btrfs_trans_handle *newtrans;
1609 struct btrfs_root *root;
1610 struct work_struct work;
1613 static void do_async_commit(struct work_struct *work)
1615 struct btrfs_async_commit *ac =
1616 container_of(work, struct btrfs_async_commit, work);
1619 * We've got freeze protection passed with the transaction.
1620 * Tell lockdep about it.
1622 if (ac->newtrans->type & __TRANS_FREEZABLE)
1624 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1627 current->journal_info = ac->newtrans;
1629 btrfs_commit_transaction(ac->newtrans, ac->root);
1633 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1634 struct btrfs_root *root,
1635 int wait_for_unblock)
1637 struct btrfs_async_commit *ac;
1638 struct btrfs_transaction *cur_trans;
1640 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1644 INIT_WORK(&ac->work, do_async_commit);
1646 ac->newtrans = btrfs_join_transaction(root);
1647 if (IS_ERR(ac->newtrans)) {
1648 int err = PTR_ERR(ac->newtrans);
1653 /* take transaction reference */
1654 cur_trans = trans->transaction;
1655 atomic_inc(&cur_trans->use_count);
1657 btrfs_end_transaction(trans, root);
1660 * Tell lockdep we've released the freeze rwsem, since the
1661 * async commit thread will be the one to unlock it.
1663 if (ac->newtrans->type & __TRANS_FREEZABLE)
1665 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1668 schedule_work(&ac->work);
1670 /* wait for transaction to start and unblock */
1671 if (wait_for_unblock)
1672 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1674 wait_current_trans_commit_start(root, cur_trans);
1676 if (current->journal_info == trans)
1677 current->journal_info = NULL;
1679 btrfs_put_transaction(cur_trans);
1684 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1685 struct btrfs_root *root, int err)
1687 struct btrfs_transaction *cur_trans = trans->transaction;
1690 WARN_ON(trans->use_count > 1);
1692 btrfs_abort_transaction(trans, root, err);
1694 spin_lock(&root->fs_info->trans_lock);
1697 * If the transaction is removed from the list, it means this
1698 * transaction has been committed successfully, so it is impossible
1699 * to call the cleanup function.
1701 BUG_ON(list_empty(&cur_trans->list));
1703 list_del_init(&cur_trans->list);
1704 if (cur_trans == root->fs_info->running_transaction) {
1705 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1706 spin_unlock(&root->fs_info->trans_lock);
1707 wait_event(cur_trans->writer_wait,
1708 atomic_read(&cur_trans->num_writers) == 1);
1710 spin_lock(&root->fs_info->trans_lock);
1712 spin_unlock(&root->fs_info->trans_lock);
1714 btrfs_cleanup_one_transaction(trans->transaction, root);
1716 spin_lock(&root->fs_info->trans_lock);
1717 if (cur_trans == root->fs_info->running_transaction)
1718 root->fs_info->running_transaction = NULL;
1719 spin_unlock(&root->fs_info->trans_lock);
1721 if (trans->type & __TRANS_FREEZABLE)
1722 sb_end_intwrite(root->fs_info->sb);
1723 btrfs_put_transaction(cur_trans);
1724 btrfs_put_transaction(cur_trans);
1726 trace_btrfs_transaction_commit(root);
1728 if (current->journal_info == trans)
1729 current->journal_info = NULL;
1730 btrfs_scrub_cancel(root->fs_info);
1732 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1735 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1737 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1738 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1742 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1744 if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT))
1745 btrfs_wait_ordered_roots(fs_info, -1);
1749 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans,
1750 struct btrfs_fs_info *fs_info)
1752 struct btrfs_ordered_extent *ordered;
1754 spin_lock(&fs_info->trans_lock);
1755 while (!list_empty(&cur_trans->pending_ordered)) {
1756 ordered = list_first_entry(&cur_trans->pending_ordered,
1757 struct btrfs_ordered_extent,
1759 list_del_init(&ordered->trans_list);
1760 spin_unlock(&fs_info->trans_lock);
1762 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE,
1764 btrfs_put_ordered_extent(ordered);
1765 spin_lock(&fs_info->trans_lock);
1767 spin_unlock(&fs_info->trans_lock);
1770 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1771 struct btrfs_root *root)
1773 struct btrfs_transaction *cur_trans = trans->transaction;
1774 struct btrfs_transaction *prev_trans = NULL;
1775 struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode);
1778 /* Stop the commit early if ->aborted is set */
1779 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1780 ret = cur_trans->aborted;
1781 btrfs_end_transaction(trans, root);
1785 /* make a pass through all the delayed refs we have so far
1786 * any runnings procs may add more while we are here
1788 ret = btrfs_run_delayed_refs(trans, root, 0);
1790 btrfs_end_transaction(trans, root);
1794 btrfs_trans_release_metadata(trans, root);
1795 trans->block_rsv = NULL;
1796 if (trans->qgroup_reserved) {
1797 btrfs_qgroup_free(root, trans->qgroup_reserved);
1798 trans->qgroup_reserved = 0;
1801 cur_trans = trans->transaction;
1804 * set the flushing flag so procs in this transaction have to
1805 * start sending their work down.
1807 cur_trans->delayed_refs.flushing = 1;
1810 if (!list_empty(&trans->new_bgs))
1811 btrfs_create_pending_block_groups(trans, root);
1813 ret = btrfs_run_delayed_refs(trans, root, 0);
1815 btrfs_end_transaction(trans, root);
1819 if (!cur_trans->dirty_bg_run) {
1822 /* this mutex is also taken before trying to set
1823 * block groups readonly. We need to make sure
1824 * that nobody has set a block group readonly
1825 * after a extents from that block group have been
1826 * allocated for cache files. btrfs_set_block_group_ro
1827 * will wait for the transaction to commit if it
1828 * finds dirty_bg_run = 1
1830 * The dirty_bg_run flag is also used to make sure only
1831 * one process starts all the block group IO. It wouldn't
1832 * hurt to have more than one go through, but there's no
1833 * real advantage to it either.
1835 mutex_lock(&root->fs_info->ro_block_group_mutex);
1836 if (!cur_trans->dirty_bg_run) {
1838 cur_trans->dirty_bg_run = 1;
1840 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1843 ret = btrfs_start_dirty_block_groups(trans, root);
1846 btrfs_end_transaction(trans, root);
1850 spin_lock(&root->fs_info->trans_lock);
1851 list_splice(&trans->ordered, &cur_trans->pending_ordered);
1852 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1853 spin_unlock(&root->fs_info->trans_lock);
1854 atomic_inc(&cur_trans->use_count);
1855 ret = btrfs_end_transaction(trans, root);
1857 wait_for_commit(root, cur_trans);
1859 if (unlikely(cur_trans->aborted))
1860 ret = cur_trans->aborted;
1862 btrfs_put_transaction(cur_trans);
1867 cur_trans->state = TRANS_STATE_COMMIT_START;
1868 wake_up(&root->fs_info->transaction_blocked_wait);
1870 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1871 prev_trans = list_entry(cur_trans->list.prev,
1872 struct btrfs_transaction, list);
1873 if (prev_trans->state != TRANS_STATE_COMPLETED) {
1874 atomic_inc(&prev_trans->use_count);
1875 spin_unlock(&root->fs_info->trans_lock);
1877 wait_for_commit(root, prev_trans);
1879 btrfs_put_transaction(prev_trans);
1881 spin_unlock(&root->fs_info->trans_lock);
1884 spin_unlock(&root->fs_info->trans_lock);
1887 extwriter_counter_dec(cur_trans, trans->type);
1889 ret = btrfs_start_delalloc_flush(root->fs_info);
1891 goto cleanup_transaction;
1893 ret = btrfs_run_delayed_items(trans, root);
1895 goto cleanup_transaction;
1897 wait_event(cur_trans->writer_wait,
1898 extwriter_counter_read(cur_trans) == 0);
1900 /* some pending stuffs might be added after the previous flush. */
1901 ret = btrfs_run_delayed_items(trans, root);
1903 goto cleanup_transaction;
1905 btrfs_wait_delalloc_flush(root->fs_info);
1907 btrfs_wait_pending_ordered(cur_trans, root->fs_info);
1909 btrfs_scrub_pause(root);
1911 * Ok now we need to make sure to block out any other joins while we
1912 * commit the transaction. We could have started a join before setting
1913 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
1915 spin_lock(&root->fs_info->trans_lock);
1916 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1917 spin_unlock(&root->fs_info->trans_lock);
1918 wait_event(cur_trans->writer_wait,
1919 atomic_read(&cur_trans->num_writers) == 1);
1921 /* ->aborted might be set after the previous check, so check it */
1922 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1923 ret = cur_trans->aborted;
1924 goto scrub_continue;
1927 * the reloc mutex makes sure that we stop
1928 * the balancing code from coming in and moving
1929 * extents around in the middle of the commit
1931 mutex_lock(&root->fs_info->reloc_mutex);
1934 * We needn't worry about the delayed items because we will
1935 * deal with them in create_pending_snapshot(), which is the
1936 * core function of the snapshot creation.
1938 ret = create_pending_snapshots(trans, root->fs_info);
1940 mutex_unlock(&root->fs_info->reloc_mutex);
1941 goto scrub_continue;
1945 * We insert the dir indexes of the snapshots and update the inode
1946 * of the snapshots' parents after the snapshot creation, so there
1947 * are some delayed items which are not dealt with. Now deal with
1950 * We needn't worry that this operation will corrupt the snapshots,
1951 * because all the tree which are snapshoted will be forced to COW
1952 * the nodes and leaves.
1954 ret = btrfs_run_delayed_items(trans, root);
1956 mutex_unlock(&root->fs_info->reloc_mutex);
1957 goto scrub_continue;
1960 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1962 mutex_unlock(&root->fs_info->reloc_mutex);
1963 goto scrub_continue;
1967 * make sure none of the code above managed to slip in a
1970 btrfs_assert_delayed_root_empty(root);
1972 WARN_ON(cur_trans != trans->transaction);
1974 /* btrfs_commit_tree_roots is responsible for getting the
1975 * various roots consistent with each other. Every pointer
1976 * in the tree of tree roots has to point to the most up to date
1977 * root for every subvolume and other tree. So, we have to keep
1978 * the tree logging code from jumping in and changing any
1981 * At this point in the commit, there can't be any tree-log
1982 * writers, but a little lower down we drop the trans mutex
1983 * and let new people in. By holding the tree_log_mutex
1984 * from now until after the super is written, we avoid races
1985 * with the tree-log code.
1987 mutex_lock(&root->fs_info->tree_log_mutex);
1989 ret = commit_fs_roots(trans, root);
1991 mutex_unlock(&root->fs_info->tree_log_mutex);
1992 mutex_unlock(&root->fs_info->reloc_mutex);
1993 goto scrub_continue;
1997 * Since the transaction is done, we can apply the pending changes
1998 * before the next transaction.
2000 btrfs_apply_pending_changes(root->fs_info);
2002 /* commit_fs_roots gets rid of all the tree log roots, it is now
2003 * safe to free the root of tree log roots
2005 btrfs_free_log_root_tree(trans, root->fs_info);
2007 ret = commit_cowonly_roots(trans, root);
2009 mutex_unlock(&root->fs_info->tree_log_mutex);
2010 mutex_unlock(&root->fs_info->reloc_mutex);
2011 goto scrub_continue;
2015 * The tasks which save the space cache and inode cache may also
2016 * update ->aborted, check it.
2018 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2019 ret = cur_trans->aborted;
2020 mutex_unlock(&root->fs_info->tree_log_mutex);
2021 mutex_unlock(&root->fs_info->reloc_mutex);
2022 goto scrub_continue;
2025 btrfs_prepare_extent_commit(trans, root);
2027 cur_trans = root->fs_info->running_transaction;
2029 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2030 root->fs_info->tree_root->node);
2031 list_add_tail(&root->fs_info->tree_root->dirty_list,
2032 &cur_trans->switch_commits);
2034 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2035 root->fs_info->chunk_root->node);
2036 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2037 &cur_trans->switch_commits);
2039 switch_commit_roots(cur_trans, root->fs_info);
2041 assert_qgroups_uptodate(trans);
2042 ASSERT(list_empty(&cur_trans->dirty_bgs));
2043 ASSERT(list_empty(&cur_trans->io_bgs));
2044 update_super_roots(root);
2046 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2047 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
2048 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2049 sizeof(*root->fs_info->super_copy));
2051 btrfs_update_commit_device_size(root->fs_info);
2052 btrfs_update_commit_device_bytes_used(root, cur_trans);
2054 clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
2055 clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
2057 spin_lock(&root->fs_info->trans_lock);
2058 cur_trans->state = TRANS_STATE_UNBLOCKED;
2059 root->fs_info->running_transaction = NULL;
2060 spin_unlock(&root->fs_info->trans_lock);
2061 mutex_unlock(&root->fs_info->reloc_mutex);
2063 wake_up(&root->fs_info->transaction_wait);
2065 ret = btrfs_write_and_wait_transaction(trans, root);
2067 btrfs_error(root->fs_info, ret,
2068 "Error while writing out transaction");
2069 mutex_unlock(&root->fs_info->tree_log_mutex);
2070 goto scrub_continue;
2073 ret = write_ctree_super(trans, root, 0);
2075 mutex_unlock(&root->fs_info->tree_log_mutex);
2076 goto scrub_continue;
2080 * the super is written, we can safely allow the tree-loggers
2081 * to go about their business
2083 mutex_unlock(&root->fs_info->tree_log_mutex);
2085 btrfs_finish_extent_commit(trans, root);
2087 if (cur_trans->have_free_bgs)
2088 btrfs_clear_space_info_full(root->fs_info);
2090 root->fs_info->last_trans_committed = cur_trans->transid;
2092 * We needn't acquire the lock here because there is no other task
2093 * which can change it.
2095 cur_trans->state = TRANS_STATE_COMPLETED;
2096 wake_up(&cur_trans->commit_wait);
2098 spin_lock(&root->fs_info->trans_lock);
2099 list_del_init(&cur_trans->list);
2100 spin_unlock(&root->fs_info->trans_lock);
2102 btrfs_put_transaction(cur_trans);
2103 btrfs_put_transaction(cur_trans);
2105 if (trans->type & __TRANS_FREEZABLE)
2106 sb_end_intwrite(root->fs_info->sb);
2108 trace_btrfs_transaction_commit(root);
2110 btrfs_scrub_continue(root);
2112 if (current->journal_info == trans)
2113 current->journal_info = NULL;
2115 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2117 if (current != root->fs_info->transaction_kthread)
2118 btrfs_run_delayed_iputs(root);
2123 btrfs_scrub_continue(root);
2124 cleanup_transaction:
2125 btrfs_trans_release_metadata(trans, root);
2126 trans->block_rsv = NULL;
2127 if (trans->qgroup_reserved) {
2128 btrfs_qgroup_free(root, trans->qgroup_reserved);
2129 trans->qgroup_reserved = 0;
2131 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
2132 if (current->journal_info == trans)
2133 current->journal_info = NULL;
2134 cleanup_transaction(trans, root, ret);
2140 * return < 0 if error
2141 * 0 if there are no more dead_roots at the time of call
2142 * 1 there are more to be processed, call me again
2144 * The return value indicates there are certainly more snapshots to delete, but
2145 * if there comes a new one during processing, it may return 0. We don't mind,
2146 * because btrfs_commit_super will poke cleaner thread and it will process it a
2147 * few seconds later.
2149 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2152 struct btrfs_fs_info *fs_info = root->fs_info;
2154 spin_lock(&fs_info->trans_lock);
2155 if (list_empty(&fs_info->dead_roots)) {
2156 spin_unlock(&fs_info->trans_lock);
2159 root = list_first_entry(&fs_info->dead_roots,
2160 struct btrfs_root, root_list);
2161 list_del_init(&root->root_list);
2162 spin_unlock(&fs_info->trans_lock);
2164 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2166 btrfs_kill_all_delayed_nodes(root);
2168 if (btrfs_header_backref_rev(root->node) <
2169 BTRFS_MIXED_BACKREF_REV)
2170 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2172 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2174 return (ret < 0) ? 0 : 1;
2177 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2182 prev = xchg(&fs_info->pending_changes, 0);
2186 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2188 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2191 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2193 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2196 bit = 1 << BTRFS_PENDING_COMMIT;
2198 btrfs_debug(fs_info, "pending commit done");
2203 "unknown pending changes left 0x%lx, ignoring", prev);