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"
34 #define BTRFS_ROOT_TRANS_TAG 0
36 void put_transaction(struct btrfs_transaction *transaction)
38 WARN_ON(atomic_read(&transaction->use_count) == 0);
39 if (atomic_dec_and_test(&transaction->use_count)) {
40 BUG_ON(!list_empty(&transaction->list));
41 WARN_ON(transaction->delayed_refs.root.rb_node);
42 memset(transaction, 0, sizeof(*transaction));
43 kmem_cache_free(btrfs_transaction_cachep, transaction);
47 static noinline void switch_commit_root(struct btrfs_root *root)
49 free_extent_buffer(root->commit_root);
50 root->commit_root = btrfs_root_node(root);
54 * either allocate a new transaction or hop into the existing one
56 static noinline int join_transaction(struct btrfs_root *root, int nofail)
58 struct btrfs_transaction *cur_trans;
59 struct btrfs_fs_info *fs_info = root->fs_info;
61 spin_lock(&fs_info->trans_lock);
63 /* The file system has been taken offline. No new transactions. */
64 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
65 spin_unlock(&fs_info->trans_lock);
69 if (fs_info->trans_no_join) {
71 spin_unlock(&fs_info->trans_lock);
76 cur_trans = fs_info->running_transaction;
78 if (cur_trans->aborted) {
79 spin_unlock(&fs_info->trans_lock);
80 return cur_trans->aborted;
82 atomic_inc(&cur_trans->use_count);
83 atomic_inc(&cur_trans->num_writers);
84 cur_trans->num_joined++;
85 spin_unlock(&fs_info->trans_lock);
88 spin_unlock(&fs_info->trans_lock);
90 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
94 spin_lock(&fs_info->trans_lock);
95 if (fs_info->running_transaction) {
97 * someone started a transaction after we unlocked. Make sure
98 * to redo the trans_no_join checks above
100 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
101 cur_trans = fs_info->running_transaction;
103 } else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
104 spin_unlock(&fs_info->trans_lock);
105 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
109 atomic_set(&cur_trans->num_writers, 1);
110 cur_trans->num_joined = 0;
111 init_waitqueue_head(&cur_trans->writer_wait);
112 init_waitqueue_head(&cur_trans->commit_wait);
113 cur_trans->in_commit = 0;
114 cur_trans->blocked = 0;
116 * One for this trans handle, one so it will live on until we
117 * commit the transaction.
119 atomic_set(&cur_trans->use_count, 2);
120 cur_trans->commit_done = 0;
121 cur_trans->start_time = get_seconds();
123 cur_trans->delayed_refs.root = RB_ROOT;
124 cur_trans->delayed_refs.num_entries = 0;
125 cur_trans->delayed_refs.num_heads_ready = 0;
126 cur_trans->delayed_refs.num_heads = 0;
127 cur_trans->delayed_refs.flushing = 0;
128 cur_trans->delayed_refs.run_delayed_start = 0;
131 * although the tree mod log is per file system and not per transaction,
132 * the log must never go across transaction boundaries.
135 if (!list_empty(&fs_info->tree_mod_seq_list)) {
136 printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when "
137 "creating a fresh transaction\n");
140 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) {
141 printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
142 "creating a fresh transaction\n");
145 atomic_set(&fs_info->tree_mod_seq, 0);
147 spin_lock_init(&cur_trans->commit_lock);
148 spin_lock_init(&cur_trans->delayed_refs.lock);
150 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
151 list_add_tail(&cur_trans->list, &fs_info->trans_list);
152 extent_io_tree_init(&cur_trans->dirty_pages,
153 fs_info->btree_inode->i_mapping);
154 fs_info->generation++;
155 cur_trans->transid = fs_info->generation;
156 fs_info->running_transaction = cur_trans;
157 cur_trans->aborted = 0;
158 spin_unlock(&fs_info->trans_lock);
164 * this does all the record keeping required to make sure that a reference
165 * counted root is properly recorded in a given transaction. This is required
166 * to make sure the old root from before we joined the transaction is deleted
167 * when the transaction commits
169 static int record_root_in_trans(struct btrfs_trans_handle *trans,
170 struct btrfs_root *root)
172 if (root->ref_cows && root->last_trans < trans->transid) {
173 WARN_ON(root == root->fs_info->extent_root);
174 WARN_ON(root->commit_root != root->node);
177 * see below for in_trans_setup usage rules
178 * we have the reloc mutex held now, so there
179 * is only one writer in this function
181 root->in_trans_setup = 1;
183 /* make sure readers find in_trans_setup before
184 * they find our root->last_trans update
188 spin_lock(&root->fs_info->fs_roots_radix_lock);
189 if (root->last_trans == trans->transid) {
190 spin_unlock(&root->fs_info->fs_roots_radix_lock);
193 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
194 (unsigned long)root->root_key.objectid,
195 BTRFS_ROOT_TRANS_TAG);
196 spin_unlock(&root->fs_info->fs_roots_radix_lock);
197 root->last_trans = trans->transid;
199 /* this is pretty tricky. We don't want to
200 * take the relocation lock in btrfs_record_root_in_trans
201 * unless we're really doing the first setup for this root in
204 * Normally we'd use root->last_trans as a flag to decide
205 * if we want to take the expensive mutex.
207 * But, we have to set root->last_trans before we
208 * init the relocation root, otherwise, we trip over warnings
209 * in ctree.c. The solution used here is to flag ourselves
210 * with root->in_trans_setup. When this is 1, we're still
211 * fixing up the reloc trees and everyone must wait.
213 * When this is zero, they can trust root->last_trans and fly
214 * through btrfs_record_root_in_trans without having to take the
215 * lock. smp_wmb() makes sure that all the writes above are
216 * done before we pop in the zero below
218 btrfs_init_reloc_root(trans, root);
220 root->in_trans_setup = 0;
226 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
227 struct btrfs_root *root)
233 * see record_root_in_trans for comments about in_trans_setup usage
237 if (root->last_trans == trans->transid &&
238 !root->in_trans_setup)
241 mutex_lock(&root->fs_info->reloc_mutex);
242 record_root_in_trans(trans, root);
243 mutex_unlock(&root->fs_info->reloc_mutex);
248 /* wait for commit against the current transaction to become unblocked
249 * when this is done, it is safe to start a new transaction, but the current
250 * transaction might not be fully on disk.
252 static void wait_current_trans(struct btrfs_root *root)
254 struct btrfs_transaction *cur_trans;
256 spin_lock(&root->fs_info->trans_lock);
257 cur_trans = root->fs_info->running_transaction;
258 if (cur_trans && cur_trans->blocked) {
259 atomic_inc(&cur_trans->use_count);
260 spin_unlock(&root->fs_info->trans_lock);
262 wait_event(root->fs_info->transaction_wait,
263 !cur_trans->blocked);
264 put_transaction(cur_trans);
266 spin_unlock(&root->fs_info->trans_lock);
270 enum btrfs_trans_type {
277 static int may_wait_transaction(struct btrfs_root *root, int type)
279 if (root->fs_info->log_root_recovering)
282 if (type == TRANS_USERSPACE)
285 if (type == TRANS_START &&
286 !atomic_read(&root->fs_info->open_ioctl_trans))
292 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
293 u64 num_items, int type,
296 struct btrfs_trans_handle *h;
297 struct btrfs_transaction *cur_trans;
300 u64 qgroup_reserved = 0;
302 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
303 return ERR_PTR(-EROFS);
305 if (current->journal_info) {
306 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
307 h = current->journal_info;
309 h->orig_rsv = h->block_rsv;
315 * Do the reservation before we join the transaction so we can do all
316 * the appropriate flushing if need be.
318 if (num_items > 0 && root != root->fs_info->chunk_root) {
319 if (root->fs_info->quota_enabled &&
320 is_fstree(root->root_key.objectid)) {
321 qgroup_reserved = num_items * root->leafsize;
322 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
327 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
329 ret = btrfs_block_rsv_add_noflush(root,
330 &root->fs_info->trans_block_rsv,
333 ret = btrfs_block_rsv_add(root,
334 &root->fs_info->trans_block_rsv,
340 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
342 return ERR_PTR(-ENOMEM);
344 sb_start_intwrite(root->fs_info->sb);
346 if (may_wait_transaction(root, type))
347 wait_current_trans(root);
350 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
352 wait_current_trans(root);
353 } while (ret == -EBUSY);
356 sb_end_intwrite(root->fs_info->sb);
357 kmem_cache_free(btrfs_trans_handle_cachep, h);
361 cur_trans = root->fs_info->running_transaction;
363 h->transid = cur_trans->transid;
364 h->transaction = cur_trans;
366 h->bytes_reserved = 0;
368 h->delayed_ref_updates = 0;
374 h->qgroup_reserved = qgroup_reserved;
375 h->delayed_ref_elem.seq = 0;
376 INIT_LIST_HEAD(&h->qgroup_ref_list);
379 if (cur_trans->blocked && may_wait_transaction(root, type)) {
380 btrfs_commit_transaction(h, root);
385 trace_btrfs_space_reservation(root->fs_info, "transaction",
386 h->transid, num_bytes, 1);
387 h->block_rsv = &root->fs_info->trans_block_rsv;
388 h->bytes_reserved = num_bytes;
392 btrfs_record_root_in_trans(h, root);
394 if (!current->journal_info && type != TRANS_USERSPACE)
395 current->journal_info = h;
399 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
402 return start_transaction(root, num_items, TRANS_START, 0);
405 struct btrfs_trans_handle *btrfs_start_transaction_noflush(
406 struct btrfs_root *root, int num_items)
408 return start_transaction(root, num_items, TRANS_START, 1);
411 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
413 return start_transaction(root, 0, TRANS_JOIN, 0);
416 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
418 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
421 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
423 return start_transaction(root, 0, TRANS_USERSPACE, 0);
426 /* wait for a transaction commit to be fully complete */
427 static noinline void wait_for_commit(struct btrfs_root *root,
428 struct btrfs_transaction *commit)
430 wait_event(commit->commit_wait, commit->commit_done);
433 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
435 struct btrfs_transaction *cur_trans = NULL, *t;
440 if (transid <= root->fs_info->last_trans_committed)
443 /* find specified transaction */
444 spin_lock(&root->fs_info->trans_lock);
445 list_for_each_entry(t, &root->fs_info->trans_list, list) {
446 if (t->transid == transid) {
448 atomic_inc(&cur_trans->use_count);
451 if (t->transid > transid)
454 spin_unlock(&root->fs_info->trans_lock);
457 goto out; /* bad transid */
459 /* find newest transaction that is committing | committed */
460 spin_lock(&root->fs_info->trans_lock);
461 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
467 atomic_inc(&cur_trans->use_count);
471 spin_unlock(&root->fs_info->trans_lock);
473 goto out; /* nothing committing|committed */
476 wait_for_commit(root, cur_trans);
478 put_transaction(cur_trans);
484 void btrfs_throttle(struct btrfs_root *root)
486 if (!atomic_read(&root->fs_info->open_ioctl_trans))
487 wait_current_trans(root);
490 static int should_end_transaction(struct btrfs_trans_handle *trans,
491 struct btrfs_root *root)
495 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
499 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
500 struct btrfs_root *root)
502 struct btrfs_transaction *cur_trans = trans->transaction;
507 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
510 updates = trans->delayed_ref_updates;
511 trans->delayed_ref_updates = 0;
513 err = btrfs_run_delayed_refs(trans, root, updates);
514 if (err) /* Error code will also eval true */
518 return should_end_transaction(trans, root);
521 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
522 struct btrfs_root *root, int throttle, int lock)
524 struct btrfs_transaction *cur_trans = trans->transaction;
525 struct btrfs_fs_info *info = root->fs_info;
529 if (--trans->use_count) {
530 trans->block_rsv = trans->orig_rsv;
535 * do the qgroup accounting as early as possible
537 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
539 btrfs_trans_release_metadata(trans, root);
540 trans->block_rsv = NULL;
542 * the same root has to be passed to start_transaction and
543 * end_transaction. Subvolume quota depends on this.
545 WARN_ON(trans->root != root);
547 if (trans->qgroup_reserved) {
548 btrfs_qgroup_free(root, trans->qgroup_reserved);
549 trans->qgroup_reserved = 0;
553 unsigned long cur = trans->delayed_ref_updates;
554 trans->delayed_ref_updates = 0;
556 trans->transaction->delayed_refs.num_heads_ready > 64) {
557 trans->delayed_ref_updates = 0;
558 btrfs_run_delayed_refs(trans, root, cur);
564 btrfs_trans_release_metadata(trans, root);
565 trans->block_rsv = NULL;
567 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
568 should_end_transaction(trans, root)) {
569 trans->transaction->blocked = 1;
573 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
576 * We may race with somebody else here so end up having
577 * to call end_transaction on ourselves again, so inc
581 return btrfs_commit_transaction(trans, root);
583 wake_up_process(info->transaction_kthread);
587 sb_end_intwrite(root->fs_info->sb);
589 WARN_ON(cur_trans != info->running_transaction);
590 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
591 atomic_dec(&cur_trans->num_writers);
594 if (waitqueue_active(&cur_trans->writer_wait))
595 wake_up(&cur_trans->writer_wait);
596 put_transaction(cur_trans);
598 if (current->journal_info == trans)
599 current->journal_info = NULL;
602 btrfs_run_delayed_iputs(root);
604 if (trans->aborted ||
605 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
608 assert_qgroups_uptodate(trans);
610 memset(trans, 0, sizeof(*trans));
611 kmem_cache_free(btrfs_trans_handle_cachep, trans);
615 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
616 struct btrfs_root *root)
620 ret = __btrfs_end_transaction(trans, root, 0, 1);
626 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
627 struct btrfs_root *root)
631 ret = __btrfs_end_transaction(trans, root, 1, 1);
637 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
638 struct btrfs_root *root)
642 ret = __btrfs_end_transaction(trans, root, 0, 0);
648 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
649 struct btrfs_root *root)
651 return __btrfs_end_transaction(trans, root, 1, 1);
655 * when btree blocks are allocated, they have some corresponding bits set for
656 * them in one of two extent_io trees. This is used to make sure all of
657 * those extents are sent to disk but does not wait on them
659 int btrfs_write_marked_extents(struct btrfs_root *root,
660 struct extent_io_tree *dirty_pages, int mark)
664 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
668 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
670 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
672 err = filemap_fdatawrite_range(mapping, start, end);
684 * when btree blocks are allocated, they have some corresponding bits set for
685 * them in one of two extent_io trees. This is used to make sure all of
686 * those extents are on disk for transaction or log commit. We wait
687 * on all the pages and clear them from the dirty pages state tree
689 int btrfs_wait_marked_extents(struct btrfs_root *root,
690 struct extent_io_tree *dirty_pages, int mark)
694 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
698 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
700 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
701 err = filemap_fdatawait_range(mapping, start, end);
713 * when btree blocks are allocated, they have some corresponding bits set for
714 * them in one of two extent_io trees. This is used to make sure all of
715 * those extents are on disk for transaction or log commit
717 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
718 struct extent_io_tree *dirty_pages, int mark)
723 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
724 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
733 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
734 struct btrfs_root *root)
736 if (!trans || !trans->transaction) {
737 struct inode *btree_inode;
738 btree_inode = root->fs_info->btree_inode;
739 return filemap_write_and_wait(btree_inode->i_mapping);
741 return btrfs_write_and_wait_marked_extents(root,
742 &trans->transaction->dirty_pages,
747 * this is used to update the root pointer in the tree of tree roots.
749 * But, in the case of the extent allocation tree, updating the root
750 * pointer may allocate blocks which may change the root of the extent
753 * So, this loops and repeats and makes sure the cowonly root didn't
754 * change while the root pointer was being updated in the metadata.
756 static int update_cowonly_root(struct btrfs_trans_handle *trans,
757 struct btrfs_root *root)
762 struct btrfs_root *tree_root = root->fs_info->tree_root;
764 old_root_used = btrfs_root_used(&root->root_item);
765 btrfs_write_dirty_block_groups(trans, root);
768 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
769 if (old_root_bytenr == root->node->start &&
770 old_root_used == btrfs_root_used(&root->root_item))
773 btrfs_set_root_node(&root->root_item, root->node);
774 ret = btrfs_update_root(trans, tree_root,
780 old_root_used = btrfs_root_used(&root->root_item);
781 ret = btrfs_write_dirty_block_groups(trans, root);
786 if (root != root->fs_info->extent_root)
787 switch_commit_root(root);
793 * update all the cowonly tree roots on disk
795 * The error handling in this function may not be obvious. Any of the
796 * failures will cause the file system to go offline. We still need
797 * to clean up the delayed refs.
799 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
800 struct btrfs_root *root)
802 struct btrfs_fs_info *fs_info = root->fs_info;
803 struct list_head *next;
804 struct extent_buffer *eb;
807 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
811 eb = btrfs_lock_root_node(fs_info->tree_root);
812 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
814 btrfs_tree_unlock(eb);
815 free_extent_buffer(eb);
820 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
824 ret = btrfs_run_dev_stats(trans, root->fs_info);
827 ret = btrfs_run_qgroups(trans, root->fs_info);
830 /* run_qgroups might have added some more refs */
831 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
834 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
835 next = fs_info->dirty_cowonly_roots.next;
837 root = list_entry(next, struct btrfs_root, dirty_list);
839 ret = update_cowonly_root(trans, root);
844 down_write(&fs_info->extent_commit_sem);
845 switch_commit_root(fs_info->extent_root);
846 up_write(&fs_info->extent_commit_sem);
852 * dead roots are old snapshots that need to be deleted. This allocates
853 * a dirty root struct and adds it into the list of dead roots that need to
856 int btrfs_add_dead_root(struct btrfs_root *root)
858 spin_lock(&root->fs_info->trans_lock);
859 list_add(&root->root_list, &root->fs_info->dead_roots);
860 spin_unlock(&root->fs_info->trans_lock);
865 * update all the cowonly tree roots on disk
867 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
868 struct btrfs_root *root)
870 struct btrfs_root *gang[8];
871 struct btrfs_fs_info *fs_info = root->fs_info;
876 spin_lock(&fs_info->fs_roots_radix_lock);
878 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
881 BTRFS_ROOT_TRANS_TAG);
884 for (i = 0; i < ret; i++) {
886 radix_tree_tag_clear(&fs_info->fs_roots_radix,
887 (unsigned long)root->root_key.objectid,
888 BTRFS_ROOT_TRANS_TAG);
889 spin_unlock(&fs_info->fs_roots_radix_lock);
891 btrfs_free_log(trans, root);
892 btrfs_update_reloc_root(trans, root);
893 btrfs_orphan_commit_root(trans, root);
895 btrfs_save_ino_cache(root, trans);
897 /* see comments in should_cow_block() */
901 if (root->commit_root != root->node) {
902 mutex_lock(&root->fs_commit_mutex);
903 switch_commit_root(root);
904 btrfs_unpin_free_ino(root);
905 mutex_unlock(&root->fs_commit_mutex);
907 btrfs_set_root_node(&root->root_item,
911 err = btrfs_update_root(trans, fs_info->tree_root,
914 spin_lock(&fs_info->fs_roots_radix_lock);
919 spin_unlock(&fs_info->fs_roots_radix_lock);
924 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
925 * otherwise every leaf in the btree is read and defragged.
927 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
929 struct btrfs_fs_info *info = root->fs_info;
930 struct btrfs_trans_handle *trans;
934 if (xchg(&root->defrag_running, 1))
938 trans = btrfs_start_transaction(root, 0);
940 return PTR_ERR(trans);
942 ret = btrfs_defrag_leaves(trans, root, cacheonly);
944 nr = trans->blocks_used;
945 btrfs_end_transaction(trans, root);
946 btrfs_btree_balance_dirty(info->tree_root, nr);
949 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
952 root->defrag_running = 0;
957 * new snapshots need to be created at a very specific time in the
958 * transaction commit. This does the actual creation
960 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
961 struct btrfs_fs_info *fs_info,
962 struct btrfs_pending_snapshot *pending)
964 struct btrfs_key key;
965 struct btrfs_root_item *new_root_item;
966 struct btrfs_root *tree_root = fs_info->tree_root;
967 struct btrfs_root *root = pending->root;
968 struct btrfs_root *parent_root;
969 struct btrfs_block_rsv *rsv;
970 struct inode *parent_inode;
971 struct btrfs_path *path;
972 struct btrfs_dir_item *dir_item;
973 struct dentry *parent;
974 struct dentry *dentry;
975 struct extent_buffer *tmp;
976 struct extent_buffer *old;
977 struct timespec cur_time = CURRENT_TIME;
985 path = btrfs_alloc_path();
987 ret = pending->error = -ENOMEM;
988 goto path_alloc_fail;
991 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
992 if (!new_root_item) {
993 ret = pending->error = -ENOMEM;
994 goto root_item_alloc_fail;
997 ret = btrfs_find_free_objectid(tree_root, &objectid);
999 pending->error = ret;
1000 goto no_free_objectid;
1003 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1005 if (to_reserve > 0) {
1006 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
1009 pending->error = ret;
1010 goto no_free_objectid;
1014 ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
1015 objectid, pending->inherit);
1017 pending->error = ret;
1018 goto no_free_objectid;
1021 key.objectid = objectid;
1022 key.offset = (u64)-1;
1023 key.type = BTRFS_ROOT_ITEM_KEY;
1025 rsv = trans->block_rsv;
1026 trans->block_rsv = &pending->block_rsv;
1028 dentry = pending->dentry;
1029 parent = dget_parent(dentry);
1030 parent_inode = parent->d_inode;
1031 parent_root = BTRFS_I(parent_inode)->root;
1032 record_root_in_trans(trans, parent_root);
1035 * insert the directory item
1037 ret = btrfs_set_inode_index(parent_inode, &index);
1038 BUG_ON(ret); /* -ENOMEM */
1040 /* check if there is a file/dir which has the same name. */
1041 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1042 btrfs_ino(parent_inode),
1043 dentry->d_name.name,
1044 dentry->d_name.len, 0);
1045 if (dir_item != NULL && !IS_ERR(dir_item)) {
1046 pending->error = -EEXIST;
1048 } else if (IS_ERR(dir_item)) {
1049 ret = PTR_ERR(dir_item);
1052 btrfs_release_path(path);
1055 * pull in the delayed directory update
1056 * and the delayed inode item
1057 * otherwise we corrupt the FS during
1060 ret = btrfs_run_delayed_items(trans, root);
1061 if (ret) /* Transaction aborted */
1064 record_root_in_trans(trans, root);
1065 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1066 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1067 btrfs_check_and_init_root_item(new_root_item);
1069 root_flags = btrfs_root_flags(new_root_item);
1070 if (pending->readonly)
1071 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1073 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1074 btrfs_set_root_flags(new_root_item, root_flags);
1076 btrfs_set_root_generation_v2(new_root_item,
1078 uuid_le_gen(&new_uuid);
1079 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1080 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1082 new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1083 new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
1084 btrfs_set_root_otransid(new_root_item, trans->transid);
1085 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1086 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1087 btrfs_set_root_stransid(new_root_item, 0);
1088 btrfs_set_root_rtransid(new_root_item, 0);
1090 old = btrfs_lock_root_node(root);
1091 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1093 btrfs_tree_unlock(old);
1094 free_extent_buffer(old);
1098 btrfs_set_lock_blocking(old);
1100 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1101 /* clean up in any case */
1102 btrfs_tree_unlock(old);
1103 free_extent_buffer(old);
1107 /* see comments in should_cow_block() */
1108 root->force_cow = 1;
1111 btrfs_set_root_node(new_root_item, tmp);
1112 /* record when the snapshot was created in key.offset */
1113 key.offset = trans->transid;
1114 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1115 btrfs_tree_unlock(tmp);
1116 free_extent_buffer(tmp);
1121 * insert root back/forward references
1123 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1124 parent_root->root_key.objectid,
1125 btrfs_ino(parent_inode), index,
1126 dentry->d_name.name, dentry->d_name.len);
1130 key.offset = (u64)-1;
1131 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1132 if (IS_ERR(pending->snap)) {
1133 ret = PTR_ERR(pending->snap);
1137 ret = btrfs_reloc_post_snapshot(trans, pending);
1141 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1145 ret = btrfs_insert_dir_item(trans, parent_root,
1146 dentry->d_name.name, dentry->d_name.len,
1148 BTRFS_FT_DIR, index);
1149 /* We have check then name at the beginning, so it is impossible. */
1150 BUG_ON(ret == -EEXIST);
1154 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1155 dentry->d_name.len * 2);
1156 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1157 ret = btrfs_update_inode(trans, parent_root, parent_inode);
1162 trans->block_rsv = rsv;
1164 kfree(new_root_item);
1165 root_item_alloc_fail:
1166 btrfs_free_path(path);
1168 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1172 btrfs_abort_transaction(trans, root, ret);
1177 * create all the snapshots we've scheduled for creation
1179 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1180 struct btrfs_fs_info *fs_info)
1182 struct btrfs_pending_snapshot *pending;
1183 struct list_head *head = &trans->transaction->pending_snapshots;
1185 list_for_each_entry(pending, head, list)
1186 create_pending_snapshot(trans, fs_info, pending);
1190 static void update_super_roots(struct btrfs_root *root)
1192 struct btrfs_root_item *root_item;
1193 struct btrfs_super_block *super;
1195 super = root->fs_info->super_copy;
1197 root_item = &root->fs_info->chunk_root->root_item;
1198 super->chunk_root = root_item->bytenr;
1199 super->chunk_root_generation = root_item->generation;
1200 super->chunk_root_level = root_item->level;
1202 root_item = &root->fs_info->tree_root->root_item;
1203 super->root = root_item->bytenr;
1204 super->generation = root_item->generation;
1205 super->root_level = root_item->level;
1206 if (btrfs_test_opt(root, SPACE_CACHE))
1207 super->cache_generation = root_item->generation;
1210 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1213 spin_lock(&info->trans_lock);
1214 if (info->running_transaction)
1215 ret = info->running_transaction->in_commit;
1216 spin_unlock(&info->trans_lock);
1220 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1223 spin_lock(&info->trans_lock);
1224 if (info->running_transaction)
1225 ret = info->running_transaction->blocked;
1226 spin_unlock(&info->trans_lock);
1231 * wait for the current transaction commit to start and block subsequent
1234 static void wait_current_trans_commit_start(struct btrfs_root *root,
1235 struct btrfs_transaction *trans)
1237 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1241 * wait for the current transaction to start and then become unblocked.
1244 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1245 struct btrfs_transaction *trans)
1247 wait_event(root->fs_info->transaction_wait,
1248 trans->commit_done || (trans->in_commit && !trans->blocked));
1252 * commit transactions asynchronously. once btrfs_commit_transaction_async
1253 * returns, any subsequent transaction will not be allowed to join.
1255 struct btrfs_async_commit {
1256 struct btrfs_trans_handle *newtrans;
1257 struct btrfs_root *root;
1258 struct delayed_work work;
1261 static void do_async_commit(struct work_struct *work)
1263 struct btrfs_async_commit *ac =
1264 container_of(work, struct btrfs_async_commit, work.work);
1267 * We've got freeze protection passed with the transaction.
1268 * Tell lockdep about it.
1271 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1274 current->journal_info = ac->newtrans;
1276 btrfs_commit_transaction(ac->newtrans, ac->root);
1280 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1281 struct btrfs_root *root,
1282 int wait_for_unblock)
1284 struct btrfs_async_commit *ac;
1285 struct btrfs_transaction *cur_trans;
1287 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1291 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1293 ac->newtrans = btrfs_join_transaction(root);
1294 if (IS_ERR(ac->newtrans)) {
1295 int err = PTR_ERR(ac->newtrans);
1300 /* take transaction reference */
1301 cur_trans = trans->transaction;
1302 atomic_inc(&cur_trans->use_count);
1304 btrfs_end_transaction(trans, root);
1307 * Tell lockdep we've released the freeze rwsem, since the
1308 * async commit thread will be the one to unlock it.
1310 rwsem_release(&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1313 schedule_delayed_work(&ac->work, 0);
1315 /* wait for transaction to start and unblock */
1316 if (wait_for_unblock)
1317 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1319 wait_current_trans_commit_start(root, cur_trans);
1321 if (current->journal_info == trans)
1322 current->journal_info = NULL;
1324 put_transaction(cur_trans);
1329 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1330 struct btrfs_root *root, int err)
1332 struct btrfs_transaction *cur_trans = trans->transaction;
1334 WARN_ON(trans->use_count > 1);
1336 btrfs_abort_transaction(trans, root, err);
1338 spin_lock(&root->fs_info->trans_lock);
1339 list_del_init(&cur_trans->list);
1340 if (cur_trans == root->fs_info->running_transaction) {
1341 root->fs_info->running_transaction = NULL;
1342 root->fs_info->trans_no_join = 0;
1344 spin_unlock(&root->fs_info->trans_lock);
1346 btrfs_cleanup_one_transaction(trans->transaction, root);
1348 put_transaction(cur_trans);
1349 put_transaction(cur_trans);
1351 trace_btrfs_transaction_commit(root);
1353 btrfs_scrub_continue(root);
1355 if (current->journal_info == trans)
1356 current->journal_info = NULL;
1358 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1362 * btrfs_transaction state sequence:
1363 * in_commit = 0, blocked = 0 (initial)
1364 * in_commit = 1, blocked = 1
1368 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1369 struct btrfs_root *root)
1371 unsigned long joined = 0;
1372 struct btrfs_transaction *cur_trans = trans->transaction;
1373 struct btrfs_transaction *prev_trans = NULL;
1376 int should_grow = 0;
1377 unsigned long now = get_seconds();
1378 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1380 btrfs_run_ordered_operations(root, 0);
1382 if (cur_trans->aborted)
1383 goto cleanup_transaction;
1385 /* make a pass through all the delayed refs we have so far
1386 * any runnings procs may add more while we are here
1388 ret = btrfs_run_delayed_refs(trans, root, 0);
1390 goto cleanup_transaction;
1392 btrfs_trans_release_metadata(trans, root);
1393 trans->block_rsv = NULL;
1395 cur_trans = trans->transaction;
1398 * set the flushing flag so procs in this transaction have to
1399 * start sending their work down.
1401 cur_trans->delayed_refs.flushing = 1;
1403 ret = btrfs_run_delayed_refs(trans, root, 0);
1405 goto cleanup_transaction;
1407 spin_lock(&cur_trans->commit_lock);
1408 if (cur_trans->in_commit) {
1409 spin_unlock(&cur_trans->commit_lock);
1410 atomic_inc(&cur_trans->use_count);
1411 ret = btrfs_end_transaction(trans, root);
1413 wait_for_commit(root, cur_trans);
1415 put_transaction(cur_trans);
1420 trans->transaction->in_commit = 1;
1421 trans->transaction->blocked = 1;
1422 spin_unlock(&cur_trans->commit_lock);
1423 wake_up(&root->fs_info->transaction_blocked_wait);
1425 spin_lock(&root->fs_info->trans_lock);
1426 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1427 prev_trans = list_entry(cur_trans->list.prev,
1428 struct btrfs_transaction, list);
1429 if (!prev_trans->commit_done) {
1430 atomic_inc(&prev_trans->use_count);
1431 spin_unlock(&root->fs_info->trans_lock);
1433 wait_for_commit(root, prev_trans);
1435 put_transaction(prev_trans);
1437 spin_unlock(&root->fs_info->trans_lock);
1440 spin_unlock(&root->fs_info->trans_lock);
1443 if (!btrfs_test_opt(root, SSD) &&
1444 (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1448 int snap_pending = 0;
1450 joined = cur_trans->num_joined;
1451 if (!list_empty(&trans->transaction->pending_snapshots))
1454 WARN_ON(cur_trans != trans->transaction);
1456 if (flush_on_commit || snap_pending) {
1457 btrfs_start_delalloc_inodes(root, 1);
1458 btrfs_wait_ordered_extents(root, 0, 1);
1461 ret = btrfs_run_delayed_items(trans, root);
1463 goto cleanup_transaction;
1466 * running the delayed items may have added new refs. account
1467 * them now so that they hinder processing of more delayed refs
1468 * as little as possible.
1470 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1473 * rename don't use btrfs_join_transaction, so, once we
1474 * set the transaction to blocked above, we aren't going
1475 * to get any new ordered operations. We can safely run
1476 * it here and no for sure that nothing new will be added
1479 btrfs_run_ordered_operations(root, 1);
1481 prepare_to_wait(&cur_trans->writer_wait, &wait,
1482 TASK_UNINTERRUPTIBLE);
1484 if (atomic_read(&cur_trans->num_writers) > 1)
1485 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1486 else if (should_grow)
1487 schedule_timeout(1);
1489 finish_wait(&cur_trans->writer_wait, &wait);
1490 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1491 (should_grow && cur_trans->num_joined != joined));
1494 * Ok now we need to make sure to block out any other joins while we
1495 * commit the transaction. We could have started a join before setting
1496 * no_join so make sure to wait for num_writers to == 1 again.
1498 spin_lock(&root->fs_info->trans_lock);
1499 root->fs_info->trans_no_join = 1;
1500 spin_unlock(&root->fs_info->trans_lock);
1501 wait_event(cur_trans->writer_wait,
1502 atomic_read(&cur_trans->num_writers) == 1);
1505 * the reloc mutex makes sure that we stop
1506 * the balancing code from coming in and moving
1507 * extents around in the middle of the commit
1509 mutex_lock(&root->fs_info->reloc_mutex);
1512 * We needn't worry about the delayed items because we will
1513 * deal with them in create_pending_snapshot(), which is the
1514 * core function of the snapshot creation.
1516 ret = create_pending_snapshots(trans, root->fs_info);
1518 mutex_unlock(&root->fs_info->reloc_mutex);
1519 goto cleanup_transaction;
1523 * We insert the dir indexes of the snapshots and update the inode
1524 * of the snapshots' parents after the snapshot creation, so there
1525 * are some delayed items which are not dealt with. Now deal with
1528 * We needn't worry that this operation will corrupt the snapshots,
1529 * because all the tree which are snapshoted will be forced to COW
1530 * the nodes and leaves.
1532 ret = btrfs_run_delayed_items(trans, root);
1534 mutex_unlock(&root->fs_info->reloc_mutex);
1535 goto cleanup_transaction;
1538 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1540 mutex_unlock(&root->fs_info->reloc_mutex);
1541 goto cleanup_transaction;
1545 * make sure none of the code above managed to slip in a
1548 btrfs_assert_delayed_root_empty(root);
1550 WARN_ON(cur_trans != trans->transaction);
1552 btrfs_scrub_pause(root);
1553 /* btrfs_commit_tree_roots is responsible for getting the
1554 * various roots consistent with each other. Every pointer
1555 * in the tree of tree roots has to point to the most up to date
1556 * root for every subvolume and other tree. So, we have to keep
1557 * the tree logging code from jumping in and changing any
1560 * At this point in the commit, there can't be any tree-log
1561 * writers, but a little lower down we drop the trans mutex
1562 * and let new people in. By holding the tree_log_mutex
1563 * from now until after the super is written, we avoid races
1564 * with the tree-log code.
1566 mutex_lock(&root->fs_info->tree_log_mutex);
1568 ret = commit_fs_roots(trans, root);
1570 mutex_unlock(&root->fs_info->tree_log_mutex);
1571 mutex_unlock(&root->fs_info->reloc_mutex);
1572 goto cleanup_transaction;
1575 /* commit_fs_roots gets rid of all the tree log roots, it is now
1576 * safe to free the root of tree log roots
1578 btrfs_free_log_root_tree(trans, root->fs_info);
1580 ret = commit_cowonly_roots(trans, root);
1582 mutex_unlock(&root->fs_info->tree_log_mutex);
1583 mutex_unlock(&root->fs_info->reloc_mutex);
1584 goto cleanup_transaction;
1587 btrfs_prepare_extent_commit(trans, root);
1589 cur_trans = root->fs_info->running_transaction;
1591 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1592 root->fs_info->tree_root->node);
1593 switch_commit_root(root->fs_info->tree_root);
1595 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1596 root->fs_info->chunk_root->node);
1597 switch_commit_root(root->fs_info->chunk_root);
1599 assert_qgroups_uptodate(trans);
1600 update_super_roots(root);
1602 if (!root->fs_info->log_root_recovering) {
1603 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1604 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1607 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1608 sizeof(*root->fs_info->super_copy));
1610 trans->transaction->blocked = 0;
1611 spin_lock(&root->fs_info->trans_lock);
1612 root->fs_info->running_transaction = NULL;
1613 root->fs_info->trans_no_join = 0;
1614 spin_unlock(&root->fs_info->trans_lock);
1615 mutex_unlock(&root->fs_info->reloc_mutex);
1617 wake_up(&root->fs_info->transaction_wait);
1619 ret = btrfs_write_and_wait_transaction(trans, root);
1621 btrfs_error(root->fs_info, ret,
1622 "Error while writing out transaction.");
1623 mutex_unlock(&root->fs_info->tree_log_mutex);
1624 goto cleanup_transaction;
1627 ret = write_ctree_super(trans, root, 0);
1629 mutex_unlock(&root->fs_info->tree_log_mutex);
1630 goto cleanup_transaction;
1634 * the super is written, we can safely allow the tree-loggers
1635 * to go about their business
1637 mutex_unlock(&root->fs_info->tree_log_mutex);
1639 btrfs_finish_extent_commit(trans, root);
1641 cur_trans->commit_done = 1;
1643 root->fs_info->last_trans_committed = cur_trans->transid;
1645 wake_up(&cur_trans->commit_wait);
1647 spin_lock(&root->fs_info->trans_lock);
1648 list_del_init(&cur_trans->list);
1649 spin_unlock(&root->fs_info->trans_lock);
1651 put_transaction(cur_trans);
1652 put_transaction(cur_trans);
1654 sb_end_intwrite(root->fs_info->sb);
1656 trace_btrfs_transaction_commit(root);
1658 btrfs_scrub_continue(root);
1660 if (current->journal_info == trans)
1661 current->journal_info = NULL;
1663 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1665 if (current != root->fs_info->transaction_kthread)
1666 btrfs_run_delayed_iputs(root);
1670 cleanup_transaction:
1671 btrfs_trans_release_metadata(trans, root);
1672 trans->block_rsv = NULL;
1673 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1675 if (current->journal_info == trans)
1676 current->journal_info = NULL;
1677 cleanup_transaction(trans, root, ret);
1683 * interface function to delete all the snapshots we have scheduled for deletion
1685 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1688 struct btrfs_fs_info *fs_info = root->fs_info;
1690 spin_lock(&fs_info->trans_lock);
1691 list_splice_init(&fs_info->dead_roots, &list);
1692 spin_unlock(&fs_info->trans_lock);
1694 while (!list_empty(&list)) {
1697 root = list_entry(list.next, struct btrfs_root, root_list);
1698 list_del(&root->root_list);
1700 btrfs_kill_all_delayed_nodes(root);
1702 if (btrfs_header_backref_rev(root->node) <
1703 BTRFS_MIXED_BACKREF_REV)
1704 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1706 ret =btrfs_drop_snapshot(root, NULL, 1, 0);