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 {
278 static int may_wait_transaction(struct btrfs_root *root, int type)
280 if (root->fs_info->log_root_recovering)
283 if (type == TRANS_USERSPACE)
286 if (type == TRANS_START &&
287 !atomic_read(&root->fs_info->open_ioctl_trans))
293 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
294 u64 num_items, int type,
297 struct btrfs_trans_handle *h;
298 struct btrfs_transaction *cur_trans;
301 u64 qgroup_reserved = 0;
303 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
304 return ERR_PTR(-EROFS);
306 if (current->journal_info) {
307 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
308 h = current->journal_info;
310 h->orig_rsv = h->block_rsv;
316 * Do the reservation before we join the transaction so we can do all
317 * the appropriate flushing if need be.
319 if (num_items > 0 && root != root->fs_info->chunk_root) {
320 if (root->fs_info->quota_enabled &&
321 is_fstree(root->root_key.objectid)) {
322 qgroup_reserved = num_items * root->leafsize;
323 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
328 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
330 ret = btrfs_block_rsv_add_noflush(root,
331 &root->fs_info->trans_block_rsv,
334 ret = btrfs_block_rsv_add(root,
335 &root->fs_info->trans_block_rsv,
341 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
343 return ERR_PTR(-ENOMEM);
345 if (!__sb_start_write(root->fs_info->sb, SB_FREEZE_FS, false)) {
346 if (type == TRANS_JOIN_FREEZE)
347 return ERR_PTR(-EPERM);
348 sb_start_intwrite(root->fs_info->sb);
351 if (may_wait_transaction(root, type))
352 wait_current_trans(root);
355 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
357 wait_current_trans(root);
358 } while (ret == -EBUSY);
361 sb_end_intwrite(root->fs_info->sb);
362 kmem_cache_free(btrfs_trans_handle_cachep, h);
366 cur_trans = root->fs_info->running_transaction;
368 h->transid = cur_trans->transid;
369 h->transaction = cur_trans;
371 h->bytes_reserved = 0;
373 h->delayed_ref_updates = 0;
379 h->qgroup_reserved = qgroup_reserved;
380 h->delayed_ref_elem.seq = 0;
381 INIT_LIST_HEAD(&h->qgroup_ref_list);
382 INIT_LIST_HEAD(&h->new_bgs);
385 if (cur_trans->blocked && may_wait_transaction(root, type)) {
386 btrfs_commit_transaction(h, root);
391 trace_btrfs_space_reservation(root->fs_info, "transaction",
392 h->transid, num_bytes, 1);
393 h->block_rsv = &root->fs_info->trans_block_rsv;
394 h->bytes_reserved = num_bytes;
398 btrfs_record_root_in_trans(h, root);
400 if (!current->journal_info && type != TRANS_USERSPACE)
401 current->journal_info = h;
405 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
408 return start_transaction(root, num_items, TRANS_START, 0);
411 struct btrfs_trans_handle *btrfs_start_transaction_noflush(
412 struct btrfs_root *root, int num_items)
414 return start_transaction(root, num_items, TRANS_START, 1);
417 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
419 return start_transaction(root, 0, TRANS_JOIN, 0);
422 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
424 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
427 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
429 return start_transaction(root, 0, TRANS_USERSPACE, 0);
432 struct btrfs_trans_handle *btrfs_join_transaction_freeze(struct btrfs_root *root)
434 return start_transaction(root, 0, TRANS_JOIN_FREEZE, 0);
437 /* wait for a transaction commit to be fully complete */
438 static noinline void wait_for_commit(struct btrfs_root *root,
439 struct btrfs_transaction *commit)
441 wait_event(commit->commit_wait, commit->commit_done);
444 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
446 struct btrfs_transaction *cur_trans = NULL, *t;
451 if (transid <= root->fs_info->last_trans_committed)
454 /* find specified transaction */
455 spin_lock(&root->fs_info->trans_lock);
456 list_for_each_entry(t, &root->fs_info->trans_list, list) {
457 if (t->transid == transid) {
459 atomic_inc(&cur_trans->use_count);
462 if (t->transid > transid)
465 spin_unlock(&root->fs_info->trans_lock);
468 goto out; /* bad transid */
470 /* find newest transaction that is committing | committed */
471 spin_lock(&root->fs_info->trans_lock);
472 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
478 atomic_inc(&cur_trans->use_count);
482 spin_unlock(&root->fs_info->trans_lock);
484 goto out; /* nothing committing|committed */
487 wait_for_commit(root, cur_trans);
489 put_transaction(cur_trans);
495 void btrfs_throttle(struct btrfs_root *root)
497 if (!atomic_read(&root->fs_info->open_ioctl_trans))
498 wait_current_trans(root);
501 static int should_end_transaction(struct btrfs_trans_handle *trans,
502 struct btrfs_root *root)
506 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
510 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
511 struct btrfs_root *root)
513 struct btrfs_transaction *cur_trans = trans->transaction;
518 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
521 updates = trans->delayed_ref_updates;
522 trans->delayed_ref_updates = 0;
524 err = btrfs_run_delayed_refs(trans, root, updates);
525 if (err) /* Error code will also eval true */
529 return should_end_transaction(trans, root);
532 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
533 struct btrfs_root *root, int throttle, int lock)
535 struct btrfs_transaction *cur_trans = trans->transaction;
536 struct btrfs_fs_info *info = root->fs_info;
540 if (--trans->use_count) {
541 trans->block_rsv = trans->orig_rsv;
546 * do the qgroup accounting as early as possible
548 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
550 btrfs_trans_release_metadata(trans, root);
551 trans->block_rsv = NULL;
553 * the same root has to be passed to start_transaction and
554 * end_transaction. Subvolume quota depends on this.
556 WARN_ON(trans->root != root);
558 if (trans->qgroup_reserved) {
559 btrfs_qgroup_free(root, trans->qgroup_reserved);
560 trans->qgroup_reserved = 0;
563 if (!list_empty(&trans->new_bgs))
564 btrfs_create_pending_block_groups(trans, root);
567 unsigned long cur = trans->delayed_ref_updates;
568 trans->delayed_ref_updates = 0;
570 trans->transaction->delayed_refs.num_heads_ready > 64) {
571 trans->delayed_ref_updates = 0;
572 btrfs_run_delayed_refs(trans, root, cur);
578 btrfs_trans_release_metadata(trans, root);
579 trans->block_rsv = NULL;
581 if (!list_empty(&trans->new_bgs))
582 btrfs_create_pending_block_groups(trans, root);
584 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
585 should_end_transaction(trans, root)) {
586 trans->transaction->blocked = 1;
590 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
593 * We may race with somebody else here so end up having
594 * to call end_transaction on ourselves again, so inc
598 return btrfs_commit_transaction(trans, root);
600 wake_up_process(info->transaction_kthread);
604 sb_end_intwrite(root->fs_info->sb);
606 WARN_ON(cur_trans != info->running_transaction);
607 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
608 atomic_dec(&cur_trans->num_writers);
611 if (waitqueue_active(&cur_trans->writer_wait))
612 wake_up(&cur_trans->writer_wait);
613 put_transaction(cur_trans);
615 if (current->journal_info == trans)
616 current->journal_info = NULL;
619 btrfs_run_delayed_iputs(root);
621 if (trans->aborted ||
622 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
625 assert_qgroups_uptodate(trans);
627 memset(trans, 0, sizeof(*trans));
628 kmem_cache_free(btrfs_trans_handle_cachep, trans);
632 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
633 struct btrfs_root *root)
637 ret = __btrfs_end_transaction(trans, root, 0, 1);
643 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
644 struct btrfs_root *root)
648 ret = __btrfs_end_transaction(trans, root, 1, 1);
654 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
655 struct btrfs_root *root)
659 ret = __btrfs_end_transaction(trans, root, 0, 0);
665 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
666 struct btrfs_root *root)
668 return __btrfs_end_transaction(trans, root, 1, 1);
672 * when btree blocks are allocated, they have some corresponding bits set for
673 * them in one of two extent_io trees. This is used to make sure all of
674 * those extents are sent to disk but does not wait on them
676 int btrfs_write_marked_extents(struct btrfs_root *root,
677 struct extent_io_tree *dirty_pages, int mark)
681 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
685 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
687 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
689 err = filemap_fdatawrite_range(mapping, start, end);
701 * when btree blocks are allocated, they have some corresponding bits set for
702 * them in one of two extent_io trees. This is used to make sure all of
703 * those extents are on disk for transaction or log commit. We wait
704 * on all the pages and clear them from the dirty pages state tree
706 int btrfs_wait_marked_extents(struct btrfs_root *root,
707 struct extent_io_tree *dirty_pages, int mark)
711 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
715 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
717 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
718 err = filemap_fdatawait_range(mapping, start, end);
730 * when btree blocks are allocated, they have some corresponding bits set for
731 * them in one of two extent_io trees. This is used to make sure all of
732 * those extents are on disk for transaction or log commit
734 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
735 struct extent_io_tree *dirty_pages, int mark)
740 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
741 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
750 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
751 struct btrfs_root *root)
753 if (!trans || !trans->transaction) {
754 struct inode *btree_inode;
755 btree_inode = root->fs_info->btree_inode;
756 return filemap_write_and_wait(btree_inode->i_mapping);
758 return btrfs_write_and_wait_marked_extents(root,
759 &trans->transaction->dirty_pages,
764 * this is used to update the root pointer in the tree of tree roots.
766 * But, in the case of the extent allocation tree, updating the root
767 * pointer may allocate blocks which may change the root of the extent
770 * So, this loops and repeats and makes sure the cowonly root didn't
771 * change while the root pointer was being updated in the metadata.
773 static int update_cowonly_root(struct btrfs_trans_handle *trans,
774 struct btrfs_root *root)
779 struct btrfs_root *tree_root = root->fs_info->tree_root;
781 old_root_used = btrfs_root_used(&root->root_item);
782 btrfs_write_dirty_block_groups(trans, root);
785 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
786 if (old_root_bytenr == root->node->start &&
787 old_root_used == btrfs_root_used(&root->root_item))
790 btrfs_set_root_node(&root->root_item, root->node);
791 ret = btrfs_update_root(trans, tree_root,
797 old_root_used = btrfs_root_used(&root->root_item);
798 ret = btrfs_write_dirty_block_groups(trans, root);
803 if (root != root->fs_info->extent_root)
804 switch_commit_root(root);
810 * update all the cowonly tree roots on disk
812 * The error handling in this function may not be obvious. Any of the
813 * failures will cause the file system to go offline. We still need
814 * to clean up the delayed refs.
816 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
817 struct btrfs_root *root)
819 struct btrfs_fs_info *fs_info = root->fs_info;
820 struct list_head *next;
821 struct extent_buffer *eb;
824 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
828 eb = btrfs_lock_root_node(fs_info->tree_root);
829 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
831 btrfs_tree_unlock(eb);
832 free_extent_buffer(eb);
837 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
841 ret = btrfs_run_dev_stats(trans, root->fs_info);
844 ret = btrfs_run_qgroups(trans, root->fs_info);
847 /* run_qgroups might have added some more refs */
848 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
851 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
852 next = fs_info->dirty_cowonly_roots.next;
854 root = list_entry(next, struct btrfs_root, dirty_list);
856 ret = update_cowonly_root(trans, root);
861 down_write(&fs_info->extent_commit_sem);
862 switch_commit_root(fs_info->extent_root);
863 up_write(&fs_info->extent_commit_sem);
869 * dead roots are old snapshots that need to be deleted. This allocates
870 * a dirty root struct and adds it into the list of dead roots that need to
873 int btrfs_add_dead_root(struct btrfs_root *root)
875 spin_lock(&root->fs_info->trans_lock);
876 list_add(&root->root_list, &root->fs_info->dead_roots);
877 spin_unlock(&root->fs_info->trans_lock);
882 * update all the cowonly tree roots on disk
884 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
885 struct btrfs_root *root)
887 struct btrfs_root *gang[8];
888 struct btrfs_fs_info *fs_info = root->fs_info;
893 spin_lock(&fs_info->fs_roots_radix_lock);
895 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
898 BTRFS_ROOT_TRANS_TAG);
901 for (i = 0; i < ret; i++) {
903 radix_tree_tag_clear(&fs_info->fs_roots_radix,
904 (unsigned long)root->root_key.objectid,
905 BTRFS_ROOT_TRANS_TAG);
906 spin_unlock(&fs_info->fs_roots_radix_lock);
908 btrfs_free_log(trans, root);
909 btrfs_update_reloc_root(trans, root);
910 btrfs_orphan_commit_root(trans, root);
912 btrfs_save_ino_cache(root, trans);
914 /* see comments in should_cow_block() */
918 if (root->commit_root != root->node) {
919 mutex_lock(&root->fs_commit_mutex);
920 switch_commit_root(root);
921 btrfs_unpin_free_ino(root);
922 mutex_unlock(&root->fs_commit_mutex);
924 btrfs_set_root_node(&root->root_item,
928 err = btrfs_update_root(trans, fs_info->tree_root,
931 spin_lock(&fs_info->fs_roots_radix_lock);
936 spin_unlock(&fs_info->fs_roots_radix_lock);
941 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
942 * otherwise every leaf in the btree is read and defragged.
944 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
946 struct btrfs_fs_info *info = root->fs_info;
947 struct btrfs_trans_handle *trans;
951 if (xchg(&root->defrag_running, 1))
955 trans = btrfs_start_transaction(root, 0);
957 return PTR_ERR(trans);
959 ret = btrfs_defrag_leaves(trans, root, cacheonly);
961 nr = trans->blocks_used;
962 btrfs_end_transaction(trans, root);
963 btrfs_btree_balance_dirty(info->tree_root, nr);
966 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
969 root->defrag_running = 0;
974 * new snapshots need to be created at a very specific time in the
975 * transaction commit. This does the actual creation
977 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
978 struct btrfs_fs_info *fs_info,
979 struct btrfs_pending_snapshot *pending)
981 struct btrfs_key key;
982 struct btrfs_root_item *new_root_item;
983 struct btrfs_root *tree_root = fs_info->tree_root;
984 struct btrfs_root *root = pending->root;
985 struct btrfs_root *parent_root;
986 struct btrfs_block_rsv *rsv;
987 struct inode *parent_inode;
988 struct btrfs_path *path;
989 struct btrfs_dir_item *dir_item;
990 struct dentry *parent;
991 struct dentry *dentry;
992 struct extent_buffer *tmp;
993 struct extent_buffer *old;
994 struct timespec cur_time = CURRENT_TIME;
1002 path = btrfs_alloc_path();
1004 ret = pending->error = -ENOMEM;
1005 goto path_alloc_fail;
1008 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1009 if (!new_root_item) {
1010 ret = pending->error = -ENOMEM;
1011 goto root_item_alloc_fail;
1014 ret = btrfs_find_free_objectid(tree_root, &objectid);
1016 pending->error = ret;
1017 goto no_free_objectid;
1020 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1022 if (to_reserve > 0) {
1023 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
1026 pending->error = ret;
1027 goto no_free_objectid;
1031 ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
1032 objectid, pending->inherit);
1034 pending->error = ret;
1035 goto no_free_objectid;
1038 key.objectid = objectid;
1039 key.offset = (u64)-1;
1040 key.type = BTRFS_ROOT_ITEM_KEY;
1042 rsv = trans->block_rsv;
1043 trans->block_rsv = &pending->block_rsv;
1045 dentry = pending->dentry;
1046 parent = dget_parent(dentry);
1047 parent_inode = parent->d_inode;
1048 parent_root = BTRFS_I(parent_inode)->root;
1049 record_root_in_trans(trans, parent_root);
1052 * insert the directory item
1054 ret = btrfs_set_inode_index(parent_inode, &index);
1055 BUG_ON(ret); /* -ENOMEM */
1057 /* check if there is a file/dir which has the same name. */
1058 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1059 btrfs_ino(parent_inode),
1060 dentry->d_name.name,
1061 dentry->d_name.len, 0);
1062 if (dir_item != NULL && !IS_ERR(dir_item)) {
1063 pending->error = -EEXIST;
1065 } else if (IS_ERR(dir_item)) {
1066 ret = PTR_ERR(dir_item);
1069 btrfs_release_path(path);
1072 * pull in the delayed directory update
1073 * and the delayed inode item
1074 * otherwise we corrupt the FS during
1077 ret = btrfs_run_delayed_items(trans, root);
1078 if (ret) /* Transaction aborted */
1081 record_root_in_trans(trans, root);
1082 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1083 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1084 btrfs_check_and_init_root_item(new_root_item);
1086 root_flags = btrfs_root_flags(new_root_item);
1087 if (pending->readonly)
1088 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1090 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1091 btrfs_set_root_flags(new_root_item, root_flags);
1093 btrfs_set_root_generation_v2(new_root_item,
1095 uuid_le_gen(&new_uuid);
1096 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1097 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1099 new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1100 new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
1101 btrfs_set_root_otransid(new_root_item, trans->transid);
1102 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1103 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1104 btrfs_set_root_stransid(new_root_item, 0);
1105 btrfs_set_root_rtransid(new_root_item, 0);
1107 old = btrfs_lock_root_node(root);
1108 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1110 btrfs_tree_unlock(old);
1111 free_extent_buffer(old);
1115 btrfs_set_lock_blocking(old);
1117 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1118 /* clean up in any case */
1119 btrfs_tree_unlock(old);
1120 free_extent_buffer(old);
1124 /* see comments in should_cow_block() */
1125 root->force_cow = 1;
1128 btrfs_set_root_node(new_root_item, tmp);
1129 /* record when the snapshot was created in key.offset */
1130 key.offset = trans->transid;
1131 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1132 btrfs_tree_unlock(tmp);
1133 free_extent_buffer(tmp);
1138 * insert root back/forward references
1140 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1141 parent_root->root_key.objectid,
1142 btrfs_ino(parent_inode), index,
1143 dentry->d_name.name, dentry->d_name.len);
1147 key.offset = (u64)-1;
1148 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1149 if (IS_ERR(pending->snap)) {
1150 ret = PTR_ERR(pending->snap);
1154 ret = btrfs_reloc_post_snapshot(trans, pending);
1158 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1162 ret = btrfs_insert_dir_item(trans, parent_root,
1163 dentry->d_name.name, dentry->d_name.len,
1165 BTRFS_FT_DIR, index);
1166 /* We have check then name at the beginning, so it is impossible. */
1167 BUG_ON(ret == -EEXIST);
1171 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1172 dentry->d_name.len * 2);
1173 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1174 ret = btrfs_update_inode(trans, parent_root, parent_inode);
1179 trans->block_rsv = rsv;
1181 kfree(new_root_item);
1182 root_item_alloc_fail:
1183 btrfs_free_path(path);
1185 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1189 btrfs_abort_transaction(trans, root, ret);
1194 * create all the snapshots we've scheduled for creation
1196 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1197 struct btrfs_fs_info *fs_info)
1199 struct btrfs_pending_snapshot *pending;
1200 struct list_head *head = &trans->transaction->pending_snapshots;
1202 list_for_each_entry(pending, head, list)
1203 create_pending_snapshot(trans, fs_info, pending);
1207 static void update_super_roots(struct btrfs_root *root)
1209 struct btrfs_root_item *root_item;
1210 struct btrfs_super_block *super;
1212 super = root->fs_info->super_copy;
1214 root_item = &root->fs_info->chunk_root->root_item;
1215 super->chunk_root = root_item->bytenr;
1216 super->chunk_root_generation = root_item->generation;
1217 super->chunk_root_level = root_item->level;
1219 root_item = &root->fs_info->tree_root->root_item;
1220 super->root = root_item->bytenr;
1221 super->generation = root_item->generation;
1222 super->root_level = root_item->level;
1223 if (btrfs_test_opt(root, SPACE_CACHE))
1224 super->cache_generation = root_item->generation;
1227 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1230 spin_lock(&info->trans_lock);
1231 if (info->running_transaction)
1232 ret = info->running_transaction->in_commit;
1233 spin_unlock(&info->trans_lock);
1237 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1240 spin_lock(&info->trans_lock);
1241 if (info->running_transaction)
1242 ret = info->running_transaction->blocked;
1243 spin_unlock(&info->trans_lock);
1248 * wait for the current transaction commit to start and block subsequent
1251 static void wait_current_trans_commit_start(struct btrfs_root *root,
1252 struct btrfs_transaction *trans)
1254 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1258 * wait for the current transaction to start and then become unblocked.
1261 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1262 struct btrfs_transaction *trans)
1264 wait_event(root->fs_info->transaction_wait,
1265 trans->commit_done || (trans->in_commit && !trans->blocked));
1269 * commit transactions asynchronously. once btrfs_commit_transaction_async
1270 * returns, any subsequent transaction will not be allowed to join.
1272 struct btrfs_async_commit {
1273 struct btrfs_trans_handle *newtrans;
1274 struct btrfs_root *root;
1275 struct delayed_work work;
1278 static void do_async_commit(struct work_struct *work)
1280 struct btrfs_async_commit *ac =
1281 container_of(work, struct btrfs_async_commit, work.work);
1284 * We've got freeze protection passed with the transaction.
1285 * Tell lockdep about it.
1288 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1291 current->journal_info = ac->newtrans;
1293 btrfs_commit_transaction(ac->newtrans, ac->root);
1297 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1298 struct btrfs_root *root,
1299 int wait_for_unblock)
1301 struct btrfs_async_commit *ac;
1302 struct btrfs_transaction *cur_trans;
1304 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1308 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1310 ac->newtrans = btrfs_join_transaction(root);
1311 if (IS_ERR(ac->newtrans)) {
1312 int err = PTR_ERR(ac->newtrans);
1317 /* take transaction reference */
1318 cur_trans = trans->transaction;
1319 atomic_inc(&cur_trans->use_count);
1321 btrfs_end_transaction(trans, root);
1324 * Tell lockdep we've released the freeze rwsem, since the
1325 * async commit thread will be the one to unlock it.
1327 rwsem_release(&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1330 schedule_delayed_work(&ac->work, 0);
1332 /* wait for transaction to start and unblock */
1333 if (wait_for_unblock)
1334 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1336 wait_current_trans_commit_start(root, cur_trans);
1338 if (current->journal_info == trans)
1339 current->journal_info = NULL;
1341 put_transaction(cur_trans);
1346 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1347 struct btrfs_root *root, int err)
1349 struct btrfs_transaction *cur_trans = trans->transaction;
1351 WARN_ON(trans->use_count > 1);
1353 btrfs_abort_transaction(trans, root, err);
1355 spin_lock(&root->fs_info->trans_lock);
1356 list_del_init(&cur_trans->list);
1357 if (cur_trans == root->fs_info->running_transaction) {
1358 root->fs_info->running_transaction = NULL;
1359 root->fs_info->trans_no_join = 0;
1361 spin_unlock(&root->fs_info->trans_lock);
1363 btrfs_cleanup_one_transaction(trans->transaction, root);
1365 put_transaction(cur_trans);
1366 put_transaction(cur_trans);
1368 trace_btrfs_transaction_commit(root);
1370 btrfs_scrub_continue(root);
1372 if (current->journal_info == trans)
1373 current->journal_info = NULL;
1375 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1379 * btrfs_transaction state sequence:
1380 * in_commit = 0, blocked = 0 (initial)
1381 * in_commit = 1, blocked = 1
1385 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1386 struct btrfs_root *root)
1388 unsigned long joined = 0;
1389 struct btrfs_transaction *cur_trans = trans->transaction;
1390 struct btrfs_transaction *prev_trans = NULL;
1393 int should_grow = 0;
1394 unsigned long now = get_seconds();
1395 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1397 btrfs_run_ordered_operations(root, 0);
1399 if (cur_trans->aborted)
1400 goto cleanup_transaction;
1402 /* make a pass through all the delayed refs we have so far
1403 * any runnings procs may add more while we are here
1405 ret = btrfs_run_delayed_refs(trans, root, 0);
1407 goto cleanup_transaction;
1409 btrfs_trans_release_metadata(trans, root);
1410 trans->block_rsv = NULL;
1412 cur_trans = trans->transaction;
1415 * set the flushing flag so procs in this transaction have to
1416 * start sending their work down.
1418 cur_trans->delayed_refs.flushing = 1;
1420 if (!list_empty(&trans->new_bgs))
1421 btrfs_create_pending_block_groups(trans, root);
1423 ret = btrfs_run_delayed_refs(trans, root, 0);
1425 goto cleanup_transaction;
1427 spin_lock(&cur_trans->commit_lock);
1428 if (cur_trans->in_commit) {
1429 spin_unlock(&cur_trans->commit_lock);
1430 atomic_inc(&cur_trans->use_count);
1431 ret = btrfs_end_transaction(trans, root);
1433 wait_for_commit(root, cur_trans);
1435 put_transaction(cur_trans);
1440 trans->transaction->in_commit = 1;
1441 trans->transaction->blocked = 1;
1442 spin_unlock(&cur_trans->commit_lock);
1443 wake_up(&root->fs_info->transaction_blocked_wait);
1445 spin_lock(&root->fs_info->trans_lock);
1446 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1447 prev_trans = list_entry(cur_trans->list.prev,
1448 struct btrfs_transaction, list);
1449 if (!prev_trans->commit_done) {
1450 atomic_inc(&prev_trans->use_count);
1451 spin_unlock(&root->fs_info->trans_lock);
1453 wait_for_commit(root, prev_trans);
1455 put_transaction(prev_trans);
1457 spin_unlock(&root->fs_info->trans_lock);
1460 spin_unlock(&root->fs_info->trans_lock);
1463 if (!btrfs_test_opt(root, SSD) &&
1464 (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1468 int snap_pending = 0;
1470 joined = cur_trans->num_joined;
1471 if (!list_empty(&trans->transaction->pending_snapshots))
1474 WARN_ON(cur_trans != trans->transaction);
1476 if (flush_on_commit || snap_pending) {
1477 btrfs_start_delalloc_inodes(root, 1);
1478 btrfs_wait_ordered_extents(root, 0, 1);
1481 ret = btrfs_run_delayed_items(trans, root);
1483 goto cleanup_transaction;
1486 * running the delayed items may have added new refs. account
1487 * them now so that they hinder processing of more delayed refs
1488 * as little as possible.
1490 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1493 * rename don't use btrfs_join_transaction, so, once we
1494 * set the transaction to blocked above, we aren't going
1495 * to get any new ordered operations. We can safely run
1496 * it here and no for sure that nothing new will be added
1499 btrfs_run_ordered_operations(root, 1);
1501 prepare_to_wait(&cur_trans->writer_wait, &wait,
1502 TASK_UNINTERRUPTIBLE);
1504 if (atomic_read(&cur_trans->num_writers) > 1)
1505 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1506 else if (should_grow)
1507 schedule_timeout(1);
1509 finish_wait(&cur_trans->writer_wait, &wait);
1510 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1511 (should_grow && cur_trans->num_joined != joined));
1514 * Ok now we need to make sure to block out any other joins while we
1515 * commit the transaction. We could have started a join before setting
1516 * no_join so make sure to wait for num_writers to == 1 again.
1518 spin_lock(&root->fs_info->trans_lock);
1519 root->fs_info->trans_no_join = 1;
1520 spin_unlock(&root->fs_info->trans_lock);
1521 wait_event(cur_trans->writer_wait,
1522 atomic_read(&cur_trans->num_writers) == 1);
1525 * the reloc mutex makes sure that we stop
1526 * the balancing code from coming in and moving
1527 * extents around in the middle of the commit
1529 mutex_lock(&root->fs_info->reloc_mutex);
1532 * We needn't worry about the delayed items because we will
1533 * deal with them in create_pending_snapshot(), which is the
1534 * core function of the snapshot creation.
1536 ret = create_pending_snapshots(trans, root->fs_info);
1538 mutex_unlock(&root->fs_info->reloc_mutex);
1539 goto cleanup_transaction;
1543 * We insert the dir indexes of the snapshots and update the inode
1544 * of the snapshots' parents after the snapshot creation, so there
1545 * are some delayed items which are not dealt with. Now deal with
1548 * We needn't worry that this operation will corrupt the snapshots,
1549 * because all the tree which are snapshoted will be forced to COW
1550 * the nodes and leaves.
1552 ret = btrfs_run_delayed_items(trans, root);
1554 mutex_unlock(&root->fs_info->reloc_mutex);
1555 goto cleanup_transaction;
1558 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1560 mutex_unlock(&root->fs_info->reloc_mutex);
1561 goto cleanup_transaction;
1565 * make sure none of the code above managed to slip in a
1568 btrfs_assert_delayed_root_empty(root);
1570 WARN_ON(cur_trans != trans->transaction);
1572 btrfs_scrub_pause(root);
1573 /* btrfs_commit_tree_roots is responsible for getting the
1574 * various roots consistent with each other. Every pointer
1575 * in the tree of tree roots has to point to the most up to date
1576 * root for every subvolume and other tree. So, we have to keep
1577 * the tree logging code from jumping in and changing any
1580 * At this point in the commit, there can't be any tree-log
1581 * writers, but a little lower down we drop the trans mutex
1582 * and let new people in. By holding the tree_log_mutex
1583 * from now until after the super is written, we avoid races
1584 * with the tree-log code.
1586 mutex_lock(&root->fs_info->tree_log_mutex);
1588 ret = commit_fs_roots(trans, root);
1590 mutex_unlock(&root->fs_info->tree_log_mutex);
1591 mutex_unlock(&root->fs_info->reloc_mutex);
1592 goto cleanup_transaction;
1595 /* commit_fs_roots gets rid of all the tree log roots, it is now
1596 * safe to free the root of tree log roots
1598 btrfs_free_log_root_tree(trans, root->fs_info);
1600 ret = commit_cowonly_roots(trans, root);
1602 mutex_unlock(&root->fs_info->tree_log_mutex);
1603 mutex_unlock(&root->fs_info->reloc_mutex);
1604 goto cleanup_transaction;
1607 btrfs_prepare_extent_commit(trans, root);
1609 cur_trans = root->fs_info->running_transaction;
1611 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1612 root->fs_info->tree_root->node);
1613 switch_commit_root(root->fs_info->tree_root);
1615 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1616 root->fs_info->chunk_root->node);
1617 switch_commit_root(root->fs_info->chunk_root);
1619 assert_qgroups_uptodate(trans);
1620 update_super_roots(root);
1622 if (!root->fs_info->log_root_recovering) {
1623 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1624 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1627 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1628 sizeof(*root->fs_info->super_copy));
1630 trans->transaction->blocked = 0;
1631 spin_lock(&root->fs_info->trans_lock);
1632 root->fs_info->running_transaction = NULL;
1633 root->fs_info->trans_no_join = 0;
1634 spin_unlock(&root->fs_info->trans_lock);
1635 mutex_unlock(&root->fs_info->reloc_mutex);
1637 wake_up(&root->fs_info->transaction_wait);
1639 ret = btrfs_write_and_wait_transaction(trans, root);
1641 btrfs_error(root->fs_info, ret,
1642 "Error while writing out transaction.");
1643 mutex_unlock(&root->fs_info->tree_log_mutex);
1644 goto cleanup_transaction;
1647 ret = write_ctree_super(trans, root, 0);
1649 mutex_unlock(&root->fs_info->tree_log_mutex);
1650 goto cleanup_transaction;
1654 * the super is written, we can safely allow the tree-loggers
1655 * to go about their business
1657 mutex_unlock(&root->fs_info->tree_log_mutex);
1659 btrfs_finish_extent_commit(trans, root);
1661 cur_trans->commit_done = 1;
1663 root->fs_info->last_trans_committed = cur_trans->transid;
1665 wake_up(&cur_trans->commit_wait);
1667 spin_lock(&root->fs_info->trans_lock);
1668 list_del_init(&cur_trans->list);
1669 spin_unlock(&root->fs_info->trans_lock);
1671 put_transaction(cur_trans);
1672 put_transaction(cur_trans);
1674 sb_end_intwrite(root->fs_info->sb);
1676 trace_btrfs_transaction_commit(root);
1678 btrfs_scrub_continue(root);
1680 if (current->journal_info == trans)
1681 current->journal_info = NULL;
1683 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1685 if (current != root->fs_info->transaction_kthread)
1686 btrfs_run_delayed_iputs(root);
1690 cleanup_transaction:
1691 btrfs_trans_release_metadata(trans, root);
1692 trans->block_rsv = NULL;
1693 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1695 if (current->journal_info == trans)
1696 current->journal_info = NULL;
1697 cleanup_transaction(trans, root, ret);
1703 * interface function to delete all the snapshots we have scheduled for deletion
1705 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1708 struct btrfs_fs_info *fs_info = root->fs_info;
1710 spin_lock(&fs_info->trans_lock);
1711 list_splice_init(&fs_info->dead_roots, &list);
1712 spin_unlock(&fs_info->trans_lock);
1714 while (!list_empty(&list)) {
1717 root = list_entry(list.next, struct btrfs_root, root_list);
1718 list_del(&root->root_list);
1720 btrfs_kill_all_delayed_nodes(root);
1722 if (btrfs_header_backref_rev(root->node) <
1723 BTRFS_MIXED_BACKREF_REV)
1724 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1726 ret =btrfs_drop_snapshot(root, NULL, 1, 0);