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)
295 struct btrfs_trans_handle *h;
296 struct btrfs_transaction *cur_trans;
299 u64 qgroup_reserved = 0;
301 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
302 return ERR_PTR(-EROFS);
304 if (current->journal_info) {
305 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
306 h = current->journal_info;
308 h->orig_rsv = h->block_rsv;
314 * Do the reservation before we join the transaction so we can do all
315 * the appropriate flushing if need be.
317 if (num_items > 0 && root != root->fs_info->chunk_root) {
318 if (root->fs_info->quota_enabled &&
319 is_fstree(root->root_key.objectid)) {
320 qgroup_reserved = num_items * root->leafsize;
321 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
326 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
327 ret = btrfs_block_rsv_add(root,
328 &root->fs_info->trans_block_rsv,
334 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
336 return ERR_PTR(-ENOMEM);
338 if (may_wait_transaction(root, type))
339 wait_current_trans(root);
342 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
344 wait_current_trans(root);
345 } while (ret == -EBUSY);
348 kmem_cache_free(btrfs_trans_handle_cachep, h);
352 cur_trans = root->fs_info->running_transaction;
354 h->transid = cur_trans->transid;
355 h->transaction = cur_trans;
357 h->bytes_reserved = 0;
359 h->delayed_ref_updates = 0;
365 h->qgroup_reserved = qgroup_reserved;
366 h->delayed_ref_elem.seq = 0;
367 INIT_LIST_HEAD(&h->qgroup_ref_list);
370 if (cur_trans->blocked && may_wait_transaction(root, type)) {
371 btrfs_commit_transaction(h, root);
376 trace_btrfs_space_reservation(root->fs_info, "transaction",
377 h->transid, num_bytes, 1);
378 h->block_rsv = &root->fs_info->trans_block_rsv;
379 h->bytes_reserved = num_bytes;
383 btrfs_record_root_in_trans(h, root);
385 if (!current->journal_info && type != TRANS_USERSPACE)
386 current->journal_info = h;
390 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
393 return start_transaction(root, num_items, TRANS_START);
395 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
397 return start_transaction(root, 0, TRANS_JOIN);
400 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
402 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
405 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
407 return start_transaction(root, 0, TRANS_USERSPACE);
410 /* wait for a transaction commit to be fully complete */
411 static noinline void wait_for_commit(struct btrfs_root *root,
412 struct btrfs_transaction *commit)
414 wait_event(commit->commit_wait, commit->commit_done);
417 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
419 struct btrfs_transaction *cur_trans = NULL, *t;
424 if (transid <= root->fs_info->last_trans_committed)
427 /* find specified transaction */
428 spin_lock(&root->fs_info->trans_lock);
429 list_for_each_entry(t, &root->fs_info->trans_list, list) {
430 if (t->transid == transid) {
432 atomic_inc(&cur_trans->use_count);
435 if (t->transid > transid)
438 spin_unlock(&root->fs_info->trans_lock);
441 goto out; /* bad transid */
443 /* find newest transaction that is committing | committed */
444 spin_lock(&root->fs_info->trans_lock);
445 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
451 atomic_inc(&cur_trans->use_count);
455 spin_unlock(&root->fs_info->trans_lock);
457 goto out; /* nothing committing|committed */
460 wait_for_commit(root, cur_trans);
462 put_transaction(cur_trans);
468 void btrfs_throttle(struct btrfs_root *root)
470 if (!atomic_read(&root->fs_info->open_ioctl_trans))
471 wait_current_trans(root);
474 static int should_end_transaction(struct btrfs_trans_handle *trans,
475 struct btrfs_root *root)
479 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
483 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
484 struct btrfs_root *root)
486 struct btrfs_transaction *cur_trans = trans->transaction;
491 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
494 updates = trans->delayed_ref_updates;
495 trans->delayed_ref_updates = 0;
497 err = btrfs_run_delayed_refs(trans, root, updates);
498 if (err) /* Error code will also eval true */
502 return should_end_transaction(trans, root);
505 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
506 struct btrfs_root *root, int throttle, int lock)
508 struct btrfs_transaction *cur_trans = trans->transaction;
509 struct btrfs_fs_info *info = root->fs_info;
513 if (--trans->use_count) {
514 trans->block_rsv = trans->orig_rsv;
519 * do the qgroup accounting as early as possible
521 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
523 btrfs_trans_release_metadata(trans, root);
524 trans->block_rsv = NULL;
526 * the same root has to be passed to start_transaction and
527 * end_transaction. Subvolume quota depends on this.
529 WARN_ON(trans->root != root);
531 if (trans->qgroup_reserved) {
532 btrfs_qgroup_free(root, trans->qgroup_reserved);
533 trans->qgroup_reserved = 0;
537 unsigned long cur = trans->delayed_ref_updates;
538 trans->delayed_ref_updates = 0;
540 trans->transaction->delayed_refs.num_heads_ready > 64) {
541 trans->delayed_ref_updates = 0;
542 btrfs_run_delayed_refs(trans, root, cur);
548 btrfs_trans_release_metadata(trans, root);
549 trans->block_rsv = NULL;
551 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
552 should_end_transaction(trans, root)) {
553 trans->transaction->blocked = 1;
557 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
560 * We may race with somebody else here so end up having
561 * to call end_transaction on ourselves again, so inc
565 return btrfs_commit_transaction(trans, root);
567 wake_up_process(info->transaction_kthread);
571 WARN_ON(cur_trans != info->running_transaction);
572 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
573 atomic_dec(&cur_trans->num_writers);
576 if (waitqueue_active(&cur_trans->writer_wait))
577 wake_up(&cur_trans->writer_wait);
578 put_transaction(cur_trans);
580 if (current->journal_info == trans)
581 current->journal_info = NULL;
584 btrfs_run_delayed_iputs(root);
586 if (trans->aborted ||
587 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
590 assert_qgroups_uptodate(trans);
592 memset(trans, 0, sizeof(*trans));
593 kmem_cache_free(btrfs_trans_handle_cachep, trans);
597 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
598 struct btrfs_root *root)
602 ret = __btrfs_end_transaction(trans, root, 0, 1);
608 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
609 struct btrfs_root *root)
613 ret = __btrfs_end_transaction(trans, root, 1, 1);
619 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
620 struct btrfs_root *root)
624 ret = __btrfs_end_transaction(trans, root, 0, 0);
630 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
631 struct btrfs_root *root)
633 return __btrfs_end_transaction(trans, root, 1, 1);
637 * when btree blocks are allocated, they have some corresponding bits set for
638 * them in one of two extent_io trees. This is used to make sure all of
639 * those extents are sent to disk but does not wait on them
641 int btrfs_write_marked_extents(struct btrfs_root *root,
642 struct extent_io_tree *dirty_pages, int mark)
646 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
650 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
652 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
654 err = filemap_fdatawrite_range(mapping, start, end);
666 * when btree blocks are allocated, they have some corresponding bits set for
667 * them in one of two extent_io trees. This is used to make sure all of
668 * those extents are on disk for transaction or log commit. We wait
669 * on all the pages and clear them from the dirty pages state tree
671 int btrfs_wait_marked_extents(struct btrfs_root *root,
672 struct extent_io_tree *dirty_pages, int mark)
676 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
680 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
682 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
683 err = filemap_fdatawait_range(mapping, start, end);
695 * when btree blocks are allocated, they have some corresponding bits set for
696 * them in one of two extent_io trees. This is used to make sure all of
697 * those extents are on disk for transaction or log commit
699 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
700 struct extent_io_tree *dirty_pages, int mark)
705 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
706 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
715 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
716 struct btrfs_root *root)
718 if (!trans || !trans->transaction) {
719 struct inode *btree_inode;
720 btree_inode = root->fs_info->btree_inode;
721 return filemap_write_and_wait(btree_inode->i_mapping);
723 return btrfs_write_and_wait_marked_extents(root,
724 &trans->transaction->dirty_pages,
729 * this is used to update the root pointer in the tree of tree roots.
731 * But, in the case of the extent allocation tree, updating the root
732 * pointer may allocate blocks which may change the root of the extent
735 * So, this loops and repeats and makes sure the cowonly root didn't
736 * change while the root pointer was being updated in the metadata.
738 static int update_cowonly_root(struct btrfs_trans_handle *trans,
739 struct btrfs_root *root)
744 struct btrfs_root *tree_root = root->fs_info->tree_root;
746 old_root_used = btrfs_root_used(&root->root_item);
747 btrfs_write_dirty_block_groups(trans, root);
750 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
751 if (old_root_bytenr == root->node->start &&
752 old_root_used == btrfs_root_used(&root->root_item))
755 btrfs_set_root_node(&root->root_item, root->node);
756 ret = btrfs_update_root(trans, tree_root,
762 old_root_used = btrfs_root_used(&root->root_item);
763 ret = btrfs_write_dirty_block_groups(trans, root);
768 if (root != root->fs_info->extent_root)
769 switch_commit_root(root);
775 * update all the cowonly tree roots on disk
777 * The error handling in this function may not be obvious. Any of the
778 * failures will cause the file system to go offline. We still need
779 * to clean up the delayed refs.
781 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
782 struct btrfs_root *root)
784 struct btrfs_fs_info *fs_info = root->fs_info;
785 struct list_head *next;
786 struct extent_buffer *eb;
789 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
793 eb = btrfs_lock_root_node(fs_info->tree_root);
794 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
796 btrfs_tree_unlock(eb);
797 free_extent_buffer(eb);
802 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
806 ret = btrfs_run_dev_stats(trans, root->fs_info);
809 ret = btrfs_run_qgroups(trans, root->fs_info);
812 /* run_qgroups might have added some more refs */
813 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
816 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
817 next = fs_info->dirty_cowonly_roots.next;
819 root = list_entry(next, struct btrfs_root, dirty_list);
821 ret = update_cowonly_root(trans, root);
826 down_write(&fs_info->extent_commit_sem);
827 switch_commit_root(fs_info->extent_root);
828 up_write(&fs_info->extent_commit_sem);
834 * dead roots are old snapshots that need to be deleted. This allocates
835 * a dirty root struct and adds it into the list of dead roots that need to
838 int btrfs_add_dead_root(struct btrfs_root *root)
840 spin_lock(&root->fs_info->trans_lock);
841 list_add(&root->root_list, &root->fs_info->dead_roots);
842 spin_unlock(&root->fs_info->trans_lock);
847 * update all the cowonly tree roots on disk
849 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
850 struct btrfs_root *root)
852 struct btrfs_root *gang[8];
853 struct btrfs_fs_info *fs_info = root->fs_info;
858 spin_lock(&fs_info->fs_roots_radix_lock);
860 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
863 BTRFS_ROOT_TRANS_TAG);
866 for (i = 0; i < ret; i++) {
868 radix_tree_tag_clear(&fs_info->fs_roots_radix,
869 (unsigned long)root->root_key.objectid,
870 BTRFS_ROOT_TRANS_TAG);
871 spin_unlock(&fs_info->fs_roots_radix_lock);
873 btrfs_free_log(trans, root);
874 btrfs_update_reloc_root(trans, root);
875 btrfs_orphan_commit_root(trans, root);
877 btrfs_save_ino_cache(root, trans);
879 /* see comments in should_cow_block() */
883 if (root->commit_root != root->node) {
884 mutex_lock(&root->fs_commit_mutex);
885 switch_commit_root(root);
886 btrfs_unpin_free_ino(root);
887 mutex_unlock(&root->fs_commit_mutex);
889 btrfs_set_root_node(&root->root_item,
893 err = btrfs_update_root(trans, fs_info->tree_root,
896 spin_lock(&fs_info->fs_roots_radix_lock);
901 spin_unlock(&fs_info->fs_roots_radix_lock);
906 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
907 * otherwise every leaf in the btree is read and defragged.
909 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
911 struct btrfs_fs_info *info = root->fs_info;
912 struct btrfs_trans_handle *trans;
916 if (xchg(&root->defrag_running, 1))
920 trans = btrfs_start_transaction(root, 0);
922 return PTR_ERR(trans);
924 ret = btrfs_defrag_leaves(trans, root, cacheonly);
926 nr = trans->blocks_used;
927 btrfs_end_transaction(trans, root);
928 btrfs_btree_balance_dirty(info->tree_root, nr);
931 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
934 root->defrag_running = 0;
939 * new snapshots need to be created at a very specific time in the
940 * transaction commit. This does the actual creation
942 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
943 struct btrfs_fs_info *fs_info,
944 struct btrfs_pending_snapshot *pending)
946 struct btrfs_key key;
947 struct btrfs_root_item *new_root_item;
948 struct btrfs_root *tree_root = fs_info->tree_root;
949 struct btrfs_root *root = pending->root;
950 struct btrfs_root *parent_root;
951 struct btrfs_block_rsv *rsv;
952 struct inode *parent_inode;
953 struct dentry *parent;
954 struct dentry *dentry;
955 struct extent_buffer *tmp;
956 struct extent_buffer *old;
957 struct timespec cur_time = CURRENT_TIME;
965 rsv = trans->block_rsv;
967 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
968 if (!new_root_item) {
969 ret = pending->error = -ENOMEM;
973 ret = btrfs_find_free_objectid(tree_root, &objectid);
975 pending->error = ret;
979 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
981 if (to_reserve > 0) {
982 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
985 pending->error = ret;
990 ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
991 objectid, pending->inherit);
992 kfree(pending->inherit);
994 pending->error = ret;
998 key.objectid = objectid;
999 key.offset = (u64)-1;
1000 key.type = BTRFS_ROOT_ITEM_KEY;
1002 trans->block_rsv = &pending->block_rsv;
1004 dentry = pending->dentry;
1005 parent = dget_parent(dentry);
1006 parent_inode = parent->d_inode;
1007 parent_root = BTRFS_I(parent_inode)->root;
1008 record_root_in_trans(trans, parent_root);
1011 * insert the directory item
1013 ret = btrfs_set_inode_index(parent_inode, &index);
1014 BUG_ON(ret); /* -ENOMEM */
1015 ret = btrfs_insert_dir_item(trans, parent_root,
1016 dentry->d_name.name, dentry->d_name.len,
1018 BTRFS_FT_DIR, index);
1019 if (ret == -EEXIST) {
1020 pending->error = -EEXIST;
1024 goto abort_trans_dput;
1027 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1028 dentry->d_name.len * 2);
1029 ret = btrfs_update_inode(trans, parent_root, parent_inode);
1031 goto abort_trans_dput;
1034 * pull in the delayed directory update
1035 * and the delayed inode item
1036 * otherwise we corrupt the FS during
1039 ret = btrfs_run_delayed_items(trans, root);
1040 if (ret) { /* Transaction aborted */
1045 record_root_in_trans(trans, root);
1046 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1047 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1048 btrfs_check_and_init_root_item(new_root_item);
1050 root_flags = btrfs_root_flags(new_root_item);
1051 if (pending->readonly)
1052 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1054 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1055 btrfs_set_root_flags(new_root_item, root_flags);
1057 btrfs_set_root_generation_v2(new_root_item,
1059 uuid_le_gen(&new_uuid);
1060 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1061 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1063 new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1064 new_root_item->otime.nsec = cpu_to_le64(cur_time.tv_nsec);
1065 btrfs_set_root_otransid(new_root_item, trans->transid);
1066 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1067 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1068 btrfs_set_root_stransid(new_root_item, 0);
1069 btrfs_set_root_rtransid(new_root_item, 0);
1071 old = btrfs_lock_root_node(root);
1072 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1074 btrfs_tree_unlock(old);
1075 free_extent_buffer(old);
1076 goto abort_trans_dput;
1079 btrfs_set_lock_blocking(old);
1081 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1082 /* clean up in any case */
1083 btrfs_tree_unlock(old);
1084 free_extent_buffer(old);
1086 goto abort_trans_dput;
1088 /* see comments in should_cow_block() */
1089 root->force_cow = 1;
1092 btrfs_set_root_node(new_root_item, tmp);
1093 /* record when the snapshot was created in key.offset */
1094 key.offset = trans->transid;
1095 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1096 btrfs_tree_unlock(tmp);
1097 free_extent_buffer(tmp);
1099 goto abort_trans_dput;
1102 * insert root back/forward references
1104 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1105 parent_root->root_key.objectid,
1106 btrfs_ino(parent_inode), index,
1107 dentry->d_name.name, dentry->d_name.len);
1112 key.offset = (u64)-1;
1113 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1114 if (IS_ERR(pending->snap)) {
1115 ret = PTR_ERR(pending->snap);
1119 ret = btrfs_reloc_post_snapshot(trans, pending);
1124 kfree(new_root_item);
1125 trans->block_rsv = rsv;
1126 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1132 btrfs_abort_transaction(trans, root, ret);
1137 * create all the snapshots we've scheduled for creation
1139 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1140 struct btrfs_fs_info *fs_info)
1142 struct btrfs_pending_snapshot *pending;
1143 struct list_head *head = &trans->transaction->pending_snapshots;
1145 list_for_each_entry(pending, head, list)
1146 create_pending_snapshot(trans, fs_info, pending);
1150 static void update_super_roots(struct btrfs_root *root)
1152 struct btrfs_root_item *root_item;
1153 struct btrfs_super_block *super;
1155 super = root->fs_info->super_copy;
1157 root_item = &root->fs_info->chunk_root->root_item;
1158 super->chunk_root = root_item->bytenr;
1159 super->chunk_root_generation = root_item->generation;
1160 super->chunk_root_level = root_item->level;
1162 root_item = &root->fs_info->tree_root->root_item;
1163 super->root = root_item->bytenr;
1164 super->generation = root_item->generation;
1165 super->root_level = root_item->level;
1166 if (btrfs_test_opt(root, SPACE_CACHE))
1167 super->cache_generation = root_item->generation;
1170 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1173 spin_lock(&info->trans_lock);
1174 if (info->running_transaction)
1175 ret = info->running_transaction->in_commit;
1176 spin_unlock(&info->trans_lock);
1180 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1183 spin_lock(&info->trans_lock);
1184 if (info->running_transaction)
1185 ret = info->running_transaction->blocked;
1186 spin_unlock(&info->trans_lock);
1191 * wait for the current transaction commit to start and block subsequent
1194 static void wait_current_trans_commit_start(struct btrfs_root *root,
1195 struct btrfs_transaction *trans)
1197 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1201 * wait for the current transaction to start and then become unblocked.
1204 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1205 struct btrfs_transaction *trans)
1207 wait_event(root->fs_info->transaction_wait,
1208 trans->commit_done || (trans->in_commit && !trans->blocked));
1212 * commit transactions asynchronously. once btrfs_commit_transaction_async
1213 * returns, any subsequent transaction will not be allowed to join.
1215 struct btrfs_async_commit {
1216 struct btrfs_trans_handle *newtrans;
1217 struct btrfs_root *root;
1218 struct delayed_work work;
1221 static void do_async_commit(struct work_struct *work)
1223 struct btrfs_async_commit *ac =
1224 container_of(work, struct btrfs_async_commit, work.work);
1226 btrfs_commit_transaction(ac->newtrans, ac->root);
1230 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1231 struct btrfs_root *root,
1232 int wait_for_unblock)
1234 struct btrfs_async_commit *ac;
1235 struct btrfs_transaction *cur_trans;
1237 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1241 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1243 ac->newtrans = btrfs_join_transaction(root);
1244 if (IS_ERR(ac->newtrans)) {
1245 int err = PTR_ERR(ac->newtrans);
1250 /* take transaction reference */
1251 cur_trans = trans->transaction;
1252 atomic_inc(&cur_trans->use_count);
1254 btrfs_end_transaction(trans, root);
1255 schedule_delayed_work(&ac->work, 0);
1257 /* wait for transaction to start and unblock */
1258 if (wait_for_unblock)
1259 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1261 wait_current_trans_commit_start(root, cur_trans);
1263 if (current->journal_info == trans)
1264 current->journal_info = NULL;
1266 put_transaction(cur_trans);
1271 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1272 struct btrfs_root *root, int err)
1274 struct btrfs_transaction *cur_trans = trans->transaction;
1276 WARN_ON(trans->use_count > 1);
1278 btrfs_abort_transaction(trans, root, err);
1280 spin_lock(&root->fs_info->trans_lock);
1281 list_del_init(&cur_trans->list);
1282 if (cur_trans == root->fs_info->running_transaction) {
1283 root->fs_info->running_transaction = NULL;
1284 root->fs_info->trans_no_join = 0;
1286 spin_unlock(&root->fs_info->trans_lock);
1288 btrfs_cleanup_one_transaction(trans->transaction, root);
1290 put_transaction(cur_trans);
1291 put_transaction(cur_trans);
1293 trace_btrfs_transaction_commit(root);
1295 btrfs_scrub_continue(root);
1297 if (current->journal_info == trans)
1298 current->journal_info = NULL;
1300 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1304 * btrfs_transaction state sequence:
1305 * in_commit = 0, blocked = 0 (initial)
1306 * in_commit = 1, blocked = 1
1310 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1311 struct btrfs_root *root)
1313 unsigned long joined = 0;
1314 struct btrfs_transaction *cur_trans = trans->transaction;
1315 struct btrfs_transaction *prev_trans = NULL;
1318 int should_grow = 0;
1319 unsigned long now = get_seconds();
1320 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1322 btrfs_run_ordered_operations(root, 0);
1324 if (cur_trans->aborted)
1325 goto cleanup_transaction;
1327 /* make a pass through all the delayed refs we have so far
1328 * any runnings procs may add more while we are here
1330 ret = btrfs_run_delayed_refs(trans, root, 0);
1332 goto cleanup_transaction;
1334 btrfs_trans_release_metadata(trans, root);
1335 trans->block_rsv = NULL;
1337 cur_trans = trans->transaction;
1340 * set the flushing flag so procs in this transaction have to
1341 * start sending their work down.
1343 cur_trans->delayed_refs.flushing = 1;
1345 ret = btrfs_run_delayed_refs(trans, root, 0);
1347 goto cleanup_transaction;
1349 spin_lock(&cur_trans->commit_lock);
1350 if (cur_trans->in_commit) {
1351 spin_unlock(&cur_trans->commit_lock);
1352 atomic_inc(&cur_trans->use_count);
1353 ret = btrfs_end_transaction(trans, root);
1355 wait_for_commit(root, cur_trans);
1357 put_transaction(cur_trans);
1362 trans->transaction->in_commit = 1;
1363 trans->transaction->blocked = 1;
1364 spin_unlock(&cur_trans->commit_lock);
1365 wake_up(&root->fs_info->transaction_blocked_wait);
1367 spin_lock(&root->fs_info->trans_lock);
1368 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1369 prev_trans = list_entry(cur_trans->list.prev,
1370 struct btrfs_transaction, list);
1371 if (!prev_trans->commit_done) {
1372 atomic_inc(&prev_trans->use_count);
1373 spin_unlock(&root->fs_info->trans_lock);
1375 wait_for_commit(root, prev_trans);
1377 put_transaction(prev_trans);
1379 spin_unlock(&root->fs_info->trans_lock);
1382 spin_unlock(&root->fs_info->trans_lock);
1385 if (!btrfs_test_opt(root, SSD) &&
1386 (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1390 int snap_pending = 0;
1392 joined = cur_trans->num_joined;
1393 if (!list_empty(&trans->transaction->pending_snapshots))
1396 WARN_ON(cur_trans != trans->transaction);
1398 if (flush_on_commit || snap_pending) {
1399 btrfs_start_delalloc_inodes(root, 1);
1400 btrfs_wait_ordered_extents(root, 0, 1);
1403 ret = btrfs_run_delayed_items(trans, root);
1405 goto cleanup_transaction;
1408 * running the delayed items may have added new refs. account
1409 * them now so that they hinder processing of more delayed refs
1410 * as little as possible.
1412 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1415 * rename don't use btrfs_join_transaction, so, once we
1416 * set the transaction to blocked above, we aren't going
1417 * to get any new ordered operations. We can safely run
1418 * it here and no for sure that nothing new will be added
1421 btrfs_run_ordered_operations(root, 1);
1423 prepare_to_wait(&cur_trans->writer_wait, &wait,
1424 TASK_UNINTERRUPTIBLE);
1426 if (atomic_read(&cur_trans->num_writers) > 1)
1427 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1428 else if (should_grow)
1429 schedule_timeout(1);
1431 finish_wait(&cur_trans->writer_wait, &wait);
1432 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1433 (should_grow && cur_trans->num_joined != joined));
1436 * Ok now we need to make sure to block out any other joins while we
1437 * commit the transaction. We could have started a join before setting
1438 * no_join so make sure to wait for num_writers to == 1 again.
1440 spin_lock(&root->fs_info->trans_lock);
1441 root->fs_info->trans_no_join = 1;
1442 spin_unlock(&root->fs_info->trans_lock);
1443 wait_event(cur_trans->writer_wait,
1444 atomic_read(&cur_trans->num_writers) == 1);
1447 * the reloc mutex makes sure that we stop
1448 * the balancing code from coming in and moving
1449 * extents around in the middle of the commit
1451 mutex_lock(&root->fs_info->reloc_mutex);
1453 ret = btrfs_run_delayed_items(trans, root);
1455 mutex_unlock(&root->fs_info->reloc_mutex);
1456 goto cleanup_transaction;
1459 ret = create_pending_snapshots(trans, root->fs_info);
1461 mutex_unlock(&root->fs_info->reloc_mutex);
1462 goto cleanup_transaction;
1465 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1467 mutex_unlock(&root->fs_info->reloc_mutex);
1468 goto cleanup_transaction;
1472 * make sure none of the code above managed to slip in a
1475 btrfs_assert_delayed_root_empty(root);
1477 WARN_ON(cur_trans != trans->transaction);
1479 btrfs_scrub_pause(root);
1480 /* btrfs_commit_tree_roots is responsible for getting the
1481 * various roots consistent with each other. Every pointer
1482 * in the tree of tree roots has to point to the most up to date
1483 * root for every subvolume and other tree. So, we have to keep
1484 * the tree logging code from jumping in and changing any
1487 * At this point in the commit, there can't be any tree-log
1488 * writers, but a little lower down we drop the trans mutex
1489 * and let new people in. By holding the tree_log_mutex
1490 * from now until after the super is written, we avoid races
1491 * with the tree-log code.
1493 mutex_lock(&root->fs_info->tree_log_mutex);
1495 ret = commit_fs_roots(trans, root);
1497 mutex_unlock(&root->fs_info->tree_log_mutex);
1498 mutex_unlock(&root->fs_info->reloc_mutex);
1499 goto cleanup_transaction;
1502 /* commit_fs_roots gets rid of all the tree log roots, it is now
1503 * safe to free the root of tree log roots
1505 btrfs_free_log_root_tree(trans, root->fs_info);
1507 ret = commit_cowonly_roots(trans, root);
1509 mutex_unlock(&root->fs_info->tree_log_mutex);
1510 mutex_unlock(&root->fs_info->reloc_mutex);
1511 goto cleanup_transaction;
1514 btrfs_prepare_extent_commit(trans, root);
1516 cur_trans = root->fs_info->running_transaction;
1518 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1519 root->fs_info->tree_root->node);
1520 switch_commit_root(root->fs_info->tree_root);
1522 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1523 root->fs_info->chunk_root->node);
1524 switch_commit_root(root->fs_info->chunk_root);
1526 assert_qgroups_uptodate(trans);
1527 update_super_roots(root);
1529 if (!root->fs_info->log_root_recovering) {
1530 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1531 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1534 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1535 sizeof(*root->fs_info->super_copy));
1537 trans->transaction->blocked = 0;
1538 spin_lock(&root->fs_info->trans_lock);
1539 root->fs_info->running_transaction = NULL;
1540 root->fs_info->trans_no_join = 0;
1541 spin_unlock(&root->fs_info->trans_lock);
1542 mutex_unlock(&root->fs_info->reloc_mutex);
1544 wake_up(&root->fs_info->transaction_wait);
1546 ret = btrfs_write_and_wait_transaction(trans, root);
1548 btrfs_error(root->fs_info, ret,
1549 "Error while writing out transaction.");
1550 mutex_unlock(&root->fs_info->tree_log_mutex);
1551 goto cleanup_transaction;
1554 ret = write_ctree_super(trans, root, 0);
1556 mutex_unlock(&root->fs_info->tree_log_mutex);
1557 goto cleanup_transaction;
1561 * the super is written, we can safely allow the tree-loggers
1562 * to go about their business
1564 mutex_unlock(&root->fs_info->tree_log_mutex);
1566 btrfs_finish_extent_commit(trans, root);
1568 cur_trans->commit_done = 1;
1570 root->fs_info->last_trans_committed = cur_trans->transid;
1572 wake_up(&cur_trans->commit_wait);
1574 spin_lock(&root->fs_info->trans_lock);
1575 list_del_init(&cur_trans->list);
1576 spin_unlock(&root->fs_info->trans_lock);
1578 put_transaction(cur_trans);
1579 put_transaction(cur_trans);
1581 trace_btrfs_transaction_commit(root);
1583 btrfs_scrub_continue(root);
1585 if (current->journal_info == trans)
1586 current->journal_info = NULL;
1588 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1590 if (current != root->fs_info->transaction_kthread)
1591 btrfs_run_delayed_iputs(root);
1595 cleanup_transaction:
1596 btrfs_trans_release_metadata(trans, root);
1597 trans->block_rsv = NULL;
1598 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1600 if (current->journal_info == trans)
1601 current->journal_info = NULL;
1602 cleanup_transaction(trans, root, ret);
1608 * interface function to delete all the snapshots we have scheduled for deletion
1610 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1613 struct btrfs_fs_info *fs_info = root->fs_info;
1615 spin_lock(&fs_info->trans_lock);
1616 list_splice_init(&fs_info->dead_roots, &list);
1617 spin_unlock(&fs_info->trans_lock);
1619 while (!list_empty(&list)) {
1622 root = list_entry(list.next, struct btrfs_root, root_list);
1623 list_del(&root->root_list);
1625 btrfs_kill_all_delayed_nodes(root);
1627 if (btrfs_header_backref_rev(root->node) <
1628 BTRFS_MIXED_BACKREF_REV)
1629 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1631 ret =btrfs_drop_snapshot(root, NULL, 1, 0);