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>
27 #include "transaction.h"
30 #include "inode-map.h"
32 #define BTRFS_ROOT_TRANS_TAG 0
34 void put_transaction(struct btrfs_transaction *transaction)
36 WARN_ON(atomic_read(&transaction->use_count) == 0);
37 if (atomic_dec_and_test(&transaction->use_count)) {
38 BUG_ON(!list_empty(&transaction->list));
39 WARN_ON(transaction->delayed_refs.root.rb_node);
40 WARN_ON(!list_empty(&transaction->delayed_refs.seq_head));
41 memset(transaction, 0, sizeof(*transaction));
42 kmem_cache_free(btrfs_transaction_cachep, transaction);
46 static noinline void switch_commit_root(struct btrfs_root *root)
48 free_extent_buffer(root->commit_root);
49 root->commit_root = btrfs_root_node(root);
53 * either allocate a new transaction or hop into the existing one
55 static noinline int join_transaction(struct btrfs_root *root, int nofail)
57 struct btrfs_transaction *cur_trans;
59 spin_lock(&root->fs_info->trans_lock);
61 /* The file system has been taken offline. No new transactions. */
62 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
63 spin_unlock(&root->fs_info->trans_lock);
67 if (root->fs_info->trans_no_join) {
69 spin_unlock(&root->fs_info->trans_lock);
74 cur_trans = root->fs_info->running_transaction;
76 if (cur_trans->aborted)
77 return cur_trans->aborted;
78 atomic_inc(&cur_trans->use_count);
79 atomic_inc(&cur_trans->num_writers);
80 cur_trans->num_joined++;
81 spin_unlock(&root->fs_info->trans_lock);
84 spin_unlock(&root->fs_info->trans_lock);
86 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
90 spin_lock(&root->fs_info->trans_lock);
91 if (root->fs_info->running_transaction) {
93 * someone started a transaction after we unlocked. Make sure
94 * to redo the trans_no_join checks above
96 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
97 cur_trans = root->fs_info->running_transaction;
101 atomic_set(&cur_trans->num_writers, 1);
102 cur_trans->num_joined = 0;
103 init_waitqueue_head(&cur_trans->writer_wait);
104 init_waitqueue_head(&cur_trans->commit_wait);
105 cur_trans->in_commit = 0;
106 cur_trans->blocked = 0;
108 * One for this trans handle, one so it will live on until we
109 * commit the transaction.
111 atomic_set(&cur_trans->use_count, 2);
112 cur_trans->commit_done = 0;
113 cur_trans->start_time = get_seconds();
115 cur_trans->delayed_refs.root = RB_ROOT;
116 cur_trans->delayed_refs.num_entries = 0;
117 cur_trans->delayed_refs.num_heads_ready = 0;
118 cur_trans->delayed_refs.num_heads = 0;
119 cur_trans->delayed_refs.flushing = 0;
120 cur_trans->delayed_refs.run_delayed_start = 0;
121 cur_trans->delayed_refs.seq = 1;
122 init_waitqueue_head(&cur_trans->delayed_refs.seq_wait);
123 spin_lock_init(&cur_trans->commit_lock);
124 spin_lock_init(&cur_trans->delayed_refs.lock);
125 INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head);
127 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
128 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
129 extent_io_tree_init(&cur_trans->dirty_pages,
130 root->fs_info->btree_inode->i_mapping);
131 root->fs_info->generation++;
132 cur_trans->transid = root->fs_info->generation;
133 root->fs_info->running_transaction = cur_trans;
134 cur_trans->aborted = 0;
135 spin_unlock(&root->fs_info->trans_lock);
141 * this does all the record keeping required to make sure that a reference
142 * counted root is properly recorded in a given transaction. This is required
143 * to make sure the old root from before we joined the transaction is deleted
144 * when the transaction commits
146 static int record_root_in_trans(struct btrfs_trans_handle *trans,
147 struct btrfs_root *root)
149 if (root->ref_cows && root->last_trans < trans->transid) {
150 WARN_ON(root == root->fs_info->extent_root);
151 WARN_ON(root->commit_root != root->node);
154 * see below for in_trans_setup usage rules
155 * we have the reloc mutex held now, so there
156 * is only one writer in this function
158 root->in_trans_setup = 1;
160 /* make sure readers find in_trans_setup before
161 * they find our root->last_trans update
165 spin_lock(&root->fs_info->fs_roots_radix_lock);
166 if (root->last_trans == trans->transid) {
167 spin_unlock(&root->fs_info->fs_roots_radix_lock);
170 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
171 (unsigned long)root->root_key.objectid,
172 BTRFS_ROOT_TRANS_TAG);
173 spin_unlock(&root->fs_info->fs_roots_radix_lock);
174 root->last_trans = trans->transid;
176 /* this is pretty tricky. We don't want to
177 * take the relocation lock in btrfs_record_root_in_trans
178 * unless we're really doing the first setup for this root in
181 * Normally we'd use root->last_trans as a flag to decide
182 * if we want to take the expensive mutex.
184 * But, we have to set root->last_trans before we
185 * init the relocation root, otherwise, we trip over warnings
186 * in ctree.c. The solution used here is to flag ourselves
187 * with root->in_trans_setup. When this is 1, we're still
188 * fixing up the reloc trees and everyone must wait.
190 * When this is zero, they can trust root->last_trans and fly
191 * through btrfs_record_root_in_trans without having to take the
192 * lock. smp_wmb() makes sure that all the writes above are
193 * done before we pop in the zero below
195 btrfs_init_reloc_root(trans, root);
197 root->in_trans_setup = 0;
203 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
204 struct btrfs_root *root)
210 * see record_root_in_trans for comments about in_trans_setup usage
214 if (root->last_trans == trans->transid &&
215 !root->in_trans_setup)
218 mutex_lock(&root->fs_info->reloc_mutex);
219 record_root_in_trans(trans, root);
220 mutex_unlock(&root->fs_info->reloc_mutex);
225 /* wait for commit against the current transaction to become unblocked
226 * when this is done, it is safe to start a new transaction, but the current
227 * transaction might not be fully on disk.
229 static void wait_current_trans(struct btrfs_root *root)
231 struct btrfs_transaction *cur_trans;
233 spin_lock(&root->fs_info->trans_lock);
234 cur_trans = root->fs_info->running_transaction;
235 if (cur_trans && cur_trans->blocked) {
236 atomic_inc(&cur_trans->use_count);
237 spin_unlock(&root->fs_info->trans_lock);
239 wait_event(root->fs_info->transaction_wait,
240 !cur_trans->blocked);
241 put_transaction(cur_trans);
243 spin_unlock(&root->fs_info->trans_lock);
247 enum btrfs_trans_type {
254 static int may_wait_transaction(struct btrfs_root *root, int type)
256 if (root->fs_info->log_root_recovering)
259 if (type == TRANS_USERSPACE)
262 if (type == TRANS_START &&
263 !atomic_read(&root->fs_info->open_ioctl_trans))
269 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
270 u64 num_items, int type)
272 struct btrfs_trans_handle *h;
273 struct btrfs_transaction *cur_trans;
277 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
278 return ERR_PTR(-EROFS);
280 if (current->journal_info) {
281 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
282 h = current->journal_info;
284 h->orig_rsv = h->block_rsv;
290 * Do the reservation before we join the transaction so we can do all
291 * the appropriate flushing if need be.
293 if (num_items > 0 && root != root->fs_info->chunk_root) {
294 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
295 ret = btrfs_block_rsv_add(root,
296 &root->fs_info->trans_block_rsv,
302 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
304 return ERR_PTR(-ENOMEM);
306 if (may_wait_transaction(root, type))
307 wait_current_trans(root);
310 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
312 wait_current_trans(root);
313 } while (ret == -EBUSY);
316 kmem_cache_free(btrfs_trans_handle_cachep, h);
320 cur_trans = root->fs_info->running_transaction;
322 h->transid = cur_trans->transid;
323 h->transaction = cur_trans;
325 h->bytes_reserved = 0;
326 h->delayed_ref_updates = 0;
333 if (cur_trans->blocked && may_wait_transaction(root, type)) {
334 btrfs_commit_transaction(h, root);
339 trace_btrfs_space_reservation(root->fs_info, "transaction",
340 h->transid, num_bytes, 1);
341 h->block_rsv = &root->fs_info->trans_block_rsv;
342 h->bytes_reserved = num_bytes;
346 btrfs_record_root_in_trans(h, root);
348 if (!current->journal_info && type != TRANS_USERSPACE)
349 current->journal_info = h;
353 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
356 return start_transaction(root, num_items, TRANS_START);
358 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
360 return start_transaction(root, 0, TRANS_JOIN);
363 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
365 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
368 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
370 return start_transaction(root, 0, TRANS_USERSPACE);
373 /* wait for a transaction commit to be fully complete */
374 static noinline void wait_for_commit(struct btrfs_root *root,
375 struct btrfs_transaction *commit)
377 wait_event(commit->commit_wait, commit->commit_done);
380 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
382 struct btrfs_transaction *cur_trans = NULL, *t;
387 if (transid <= root->fs_info->last_trans_committed)
390 /* find specified transaction */
391 spin_lock(&root->fs_info->trans_lock);
392 list_for_each_entry(t, &root->fs_info->trans_list, list) {
393 if (t->transid == transid) {
395 atomic_inc(&cur_trans->use_count);
398 if (t->transid > transid)
401 spin_unlock(&root->fs_info->trans_lock);
404 goto out; /* bad transid */
406 /* find newest transaction that is committing | committed */
407 spin_lock(&root->fs_info->trans_lock);
408 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
414 atomic_inc(&cur_trans->use_count);
418 spin_unlock(&root->fs_info->trans_lock);
420 goto out; /* nothing committing|committed */
423 wait_for_commit(root, cur_trans);
425 put_transaction(cur_trans);
431 void btrfs_throttle(struct btrfs_root *root)
433 if (!atomic_read(&root->fs_info->open_ioctl_trans))
434 wait_current_trans(root);
437 static int should_end_transaction(struct btrfs_trans_handle *trans,
438 struct btrfs_root *root)
442 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
446 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
447 struct btrfs_root *root)
449 struct btrfs_transaction *cur_trans = trans->transaction;
450 struct btrfs_block_rsv *rsv = trans->block_rsv;
455 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
459 * We need to do this in case we're deleting csums so the global block
460 * rsv get's used instead of the csum block rsv.
462 trans->block_rsv = NULL;
464 updates = trans->delayed_ref_updates;
465 trans->delayed_ref_updates = 0;
467 err = btrfs_run_delayed_refs(trans, root, updates);
468 if (err) /* Error code will also eval true */
472 trans->block_rsv = rsv;
474 return should_end_transaction(trans, root);
477 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
478 struct btrfs_root *root, int throttle, int lock)
480 struct btrfs_transaction *cur_trans = trans->transaction;
481 struct btrfs_fs_info *info = root->fs_info;
485 if (--trans->use_count) {
486 trans->block_rsv = trans->orig_rsv;
490 btrfs_trans_release_metadata(trans, root);
491 trans->block_rsv = NULL;
493 unsigned long cur = trans->delayed_ref_updates;
494 trans->delayed_ref_updates = 0;
496 trans->transaction->delayed_refs.num_heads_ready > 64) {
497 trans->delayed_ref_updates = 0;
498 btrfs_run_delayed_refs(trans, root, cur);
505 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
506 should_end_transaction(trans, root)) {
507 trans->transaction->blocked = 1;
511 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
514 * We may race with somebody else here so end up having
515 * to call end_transaction on ourselves again, so inc
519 return btrfs_commit_transaction(trans, root);
521 wake_up_process(info->transaction_kthread);
525 WARN_ON(cur_trans != info->running_transaction);
526 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
527 atomic_dec(&cur_trans->num_writers);
530 if (waitqueue_active(&cur_trans->writer_wait))
531 wake_up(&cur_trans->writer_wait);
532 put_transaction(cur_trans);
534 if (current->journal_info == trans)
535 current->journal_info = NULL;
538 btrfs_run_delayed_iputs(root);
540 if (trans->aborted ||
541 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
545 memset(trans, 0, sizeof(*trans));
546 kmem_cache_free(btrfs_trans_handle_cachep, trans);
550 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
551 struct btrfs_root *root)
555 ret = __btrfs_end_transaction(trans, root, 0, 1);
561 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
562 struct btrfs_root *root)
566 ret = __btrfs_end_transaction(trans, root, 1, 1);
572 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
573 struct btrfs_root *root)
577 ret = __btrfs_end_transaction(trans, root, 0, 0);
583 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
584 struct btrfs_root *root)
586 return __btrfs_end_transaction(trans, root, 1, 1);
590 * when btree blocks are allocated, they have some corresponding bits set for
591 * them in one of two extent_io trees. This is used to make sure all of
592 * those extents are sent to disk but does not wait on them
594 int btrfs_write_marked_extents(struct btrfs_root *root,
595 struct extent_io_tree *dirty_pages, int mark)
599 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
603 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
605 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
607 err = filemap_fdatawrite_range(mapping, start, end);
619 * when btree blocks are allocated, they have some corresponding bits set for
620 * them in one of two extent_io trees. This is used to make sure all of
621 * those extents are on disk for transaction or log commit. We wait
622 * on all the pages and clear them from the dirty pages state tree
624 int btrfs_wait_marked_extents(struct btrfs_root *root,
625 struct extent_io_tree *dirty_pages, int mark)
629 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
633 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
635 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
636 err = filemap_fdatawait_range(mapping, start, end);
648 * when btree blocks are allocated, they have some corresponding bits set for
649 * them in one of two extent_io trees. This is used to make sure all of
650 * those extents are on disk for transaction or log commit
652 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
653 struct extent_io_tree *dirty_pages, int mark)
658 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
659 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
668 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
669 struct btrfs_root *root)
671 if (!trans || !trans->transaction) {
672 struct inode *btree_inode;
673 btree_inode = root->fs_info->btree_inode;
674 return filemap_write_and_wait(btree_inode->i_mapping);
676 return btrfs_write_and_wait_marked_extents(root,
677 &trans->transaction->dirty_pages,
682 * this is used to update the root pointer in the tree of tree roots.
684 * But, in the case of the extent allocation tree, updating the root
685 * pointer may allocate blocks which may change the root of the extent
688 * So, this loops and repeats and makes sure the cowonly root didn't
689 * change while the root pointer was being updated in the metadata.
691 static int update_cowonly_root(struct btrfs_trans_handle *trans,
692 struct btrfs_root *root)
697 struct btrfs_root *tree_root = root->fs_info->tree_root;
699 old_root_used = btrfs_root_used(&root->root_item);
700 btrfs_write_dirty_block_groups(trans, root);
703 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
704 if (old_root_bytenr == root->node->start &&
705 old_root_used == btrfs_root_used(&root->root_item))
708 btrfs_set_root_node(&root->root_item, root->node);
709 ret = btrfs_update_root(trans, tree_root,
715 old_root_used = btrfs_root_used(&root->root_item);
716 ret = btrfs_write_dirty_block_groups(trans, root);
721 if (root != root->fs_info->extent_root)
722 switch_commit_root(root);
728 * update all the cowonly tree roots on disk
730 * The error handling in this function may not be obvious. Any of the
731 * failures will cause the file system to go offline. We still need
732 * to clean up the delayed refs.
734 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root)
737 struct btrfs_fs_info *fs_info = root->fs_info;
738 struct list_head *next;
739 struct extent_buffer *eb;
742 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
746 eb = btrfs_lock_root_node(fs_info->tree_root);
747 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
749 btrfs_tree_unlock(eb);
750 free_extent_buffer(eb);
755 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
759 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
760 next = fs_info->dirty_cowonly_roots.next;
762 root = list_entry(next, struct btrfs_root, dirty_list);
764 ret = update_cowonly_root(trans, root);
769 down_write(&fs_info->extent_commit_sem);
770 switch_commit_root(fs_info->extent_root);
771 up_write(&fs_info->extent_commit_sem);
777 * dead roots are old snapshots that need to be deleted. This allocates
778 * a dirty root struct and adds it into the list of dead roots that need to
781 int btrfs_add_dead_root(struct btrfs_root *root)
783 spin_lock(&root->fs_info->trans_lock);
784 list_add(&root->root_list, &root->fs_info->dead_roots);
785 spin_unlock(&root->fs_info->trans_lock);
790 * update all the cowonly tree roots on disk
792 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
793 struct btrfs_root *root)
795 struct btrfs_root *gang[8];
796 struct btrfs_fs_info *fs_info = root->fs_info;
801 spin_lock(&fs_info->fs_roots_radix_lock);
803 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
806 BTRFS_ROOT_TRANS_TAG);
809 for (i = 0; i < ret; i++) {
811 radix_tree_tag_clear(&fs_info->fs_roots_radix,
812 (unsigned long)root->root_key.objectid,
813 BTRFS_ROOT_TRANS_TAG);
814 spin_unlock(&fs_info->fs_roots_radix_lock);
816 btrfs_free_log(trans, root);
817 btrfs_update_reloc_root(trans, root);
818 btrfs_orphan_commit_root(trans, root);
820 btrfs_save_ino_cache(root, trans);
822 /* see comments in should_cow_block() */
826 if (root->commit_root != root->node) {
827 mutex_lock(&root->fs_commit_mutex);
828 switch_commit_root(root);
829 btrfs_unpin_free_ino(root);
830 mutex_unlock(&root->fs_commit_mutex);
832 btrfs_set_root_node(&root->root_item,
836 err = btrfs_update_root(trans, fs_info->tree_root,
839 spin_lock(&fs_info->fs_roots_radix_lock);
844 spin_unlock(&fs_info->fs_roots_radix_lock);
849 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
850 * otherwise every leaf in the btree is read and defragged.
852 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
854 struct btrfs_fs_info *info = root->fs_info;
855 struct btrfs_trans_handle *trans;
859 if (xchg(&root->defrag_running, 1))
863 trans = btrfs_start_transaction(root, 0);
865 return PTR_ERR(trans);
867 ret = btrfs_defrag_leaves(trans, root, cacheonly);
869 nr = trans->blocks_used;
870 btrfs_end_transaction(trans, root);
871 btrfs_btree_balance_dirty(info->tree_root, nr);
874 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
877 root->defrag_running = 0;
882 * new snapshots need to be created at a very specific time in the
883 * transaction commit. This does the actual creation
885 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
886 struct btrfs_fs_info *fs_info,
887 struct btrfs_pending_snapshot *pending)
889 struct btrfs_key key;
890 struct btrfs_root_item *new_root_item;
891 struct btrfs_root *tree_root = fs_info->tree_root;
892 struct btrfs_root *root = pending->root;
893 struct btrfs_root *parent_root;
894 struct btrfs_block_rsv *rsv;
895 struct inode *parent_inode;
896 struct dentry *parent;
897 struct dentry *dentry;
898 struct extent_buffer *tmp;
899 struct extent_buffer *old;
906 rsv = trans->block_rsv;
908 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
909 if (!new_root_item) {
910 ret = pending->error = -ENOMEM;
914 ret = btrfs_find_free_objectid(tree_root, &objectid);
916 pending->error = ret;
920 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
922 if (to_reserve > 0) {
923 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
926 pending->error = ret;
931 key.objectid = objectid;
932 key.offset = (u64)-1;
933 key.type = BTRFS_ROOT_ITEM_KEY;
935 trans->block_rsv = &pending->block_rsv;
937 dentry = pending->dentry;
938 parent = dget_parent(dentry);
939 parent_inode = parent->d_inode;
940 parent_root = BTRFS_I(parent_inode)->root;
941 record_root_in_trans(trans, parent_root);
944 * insert the directory item
946 ret = btrfs_set_inode_index(parent_inode, &index);
947 BUG_ON(ret); /* -ENOMEM */
948 ret = btrfs_insert_dir_item(trans, parent_root,
949 dentry->d_name.name, dentry->d_name.len,
951 BTRFS_FT_DIR, index);
952 if (ret == -EEXIST) {
953 pending->error = -EEXIST;
957 goto abort_trans_dput;
960 btrfs_i_size_write(parent_inode, parent_inode->i_size +
961 dentry->d_name.len * 2);
962 ret = btrfs_update_inode(trans, parent_root, parent_inode);
964 goto abort_trans_dput;
967 * pull in the delayed directory update
968 * and the delayed inode item
969 * otherwise we corrupt the FS during
972 ret = btrfs_run_delayed_items(trans, root);
973 if (ret) { /* Transaction aborted */
978 record_root_in_trans(trans, root);
979 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
980 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
981 btrfs_check_and_init_root_item(new_root_item);
983 root_flags = btrfs_root_flags(new_root_item);
984 if (pending->readonly)
985 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
987 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
988 btrfs_set_root_flags(new_root_item, root_flags);
990 old = btrfs_lock_root_node(root);
991 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
993 btrfs_tree_unlock(old);
994 free_extent_buffer(old);
995 goto abort_trans_dput;
998 btrfs_set_lock_blocking(old);
1000 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1001 /* clean up in any case */
1002 btrfs_tree_unlock(old);
1003 free_extent_buffer(old);
1005 goto abort_trans_dput;
1007 /* see comments in should_cow_block() */
1008 root->force_cow = 1;
1011 btrfs_set_root_node(new_root_item, tmp);
1012 /* record when the snapshot was created in key.offset */
1013 key.offset = trans->transid;
1014 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1015 btrfs_tree_unlock(tmp);
1016 free_extent_buffer(tmp);
1018 goto abort_trans_dput;
1021 * insert root back/forward references
1023 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1024 parent_root->root_key.objectid,
1025 btrfs_ino(parent_inode), index,
1026 dentry->d_name.name, dentry->d_name.len);
1031 key.offset = (u64)-1;
1032 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1033 if (IS_ERR(pending->snap)) {
1034 ret = PTR_ERR(pending->snap);
1038 ret = btrfs_reloc_post_snapshot(trans, pending);
1043 kfree(new_root_item);
1044 trans->block_rsv = rsv;
1045 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1051 btrfs_abort_transaction(trans, root, ret);
1056 * create all the snapshots we've scheduled for creation
1058 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1059 struct btrfs_fs_info *fs_info)
1061 struct btrfs_pending_snapshot *pending;
1062 struct list_head *head = &trans->transaction->pending_snapshots;
1064 list_for_each_entry(pending, head, list)
1065 create_pending_snapshot(trans, fs_info, pending);
1069 static void update_super_roots(struct btrfs_root *root)
1071 struct btrfs_root_item *root_item;
1072 struct btrfs_super_block *super;
1074 super = root->fs_info->super_copy;
1076 root_item = &root->fs_info->chunk_root->root_item;
1077 super->chunk_root = root_item->bytenr;
1078 super->chunk_root_generation = root_item->generation;
1079 super->chunk_root_level = root_item->level;
1081 root_item = &root->fs_info->tree_root->root_item;
1082 super->root = root_item->bytenr;
1083 super->generation = root_item->generation;
1084 super->root_level = root_item->level;
1085 if (btrfs_test_opt(root, SPACE_CACHE))
1086 super->cache_generation = root_item->generation;
1089 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1092 spin_lock(&info->trans_lock);
1093 if (info->running_transaction)
1094 ret = info->running_transaction->in_commit;
1095 spin_unlock(&info->trans_lock);
1099 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1102 spin_lock(&info->trans_lock);
1103 if (info->running_transaction)
1104 ret = info->running_transaction->blocked;
1105 spin_unlock(&info->trans_lock);
1110 * wait for the current transaction commit to start and block subsequent
1113 static void wait_current_trans_commit_start(struct btrfs_root *root,
1114 struct btrfs_transaction *trans)
1116 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1120 * wait for the current transaction to start and then become unblocked.
1123 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1124 struct btrfs_transaction *trans)
1126 wait_event(root->fs_info->transaction_wait,
1127 trans->commit_done || (trans->in_commit && !trans->blocked));
1131 * commit transactions asynchronously. once btrfs_commit_transaction_async
1132 * returns, any subsequent transaction will not be allowed to join.
1134 struct btrfs_async_commit {
1135 struct btrfs_trans_handle *newtrans;
1136 struct btrfs_root *root;
1137 struct delayed_work work;
1140 static void do_async_commit(struct work_struct *work)
1142 struct btrfs_async_commit *ac =
1143 container_of(work, struct btrfs_async_commit, work.work);
1145 btrfs_commit_transaction(ac->newtrans, ac->root);
1149 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1150 struct btrfs_root *root,
1151 int wait_for_unblock)
1153 struct btrfs_async_commit *ac;
1154 struct btrfs_transaction *cur_trans;
1156 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1160 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1162 ac->newtrans = btrfs_join_transaction(root);
1163 if (IS_ERR(ac->newtrans)) {
1164 int err = PTR_ERR(ac->newtrans);
1169 /* take transaction reference */
1170 cur_trans = trans->transaction;
1171 atomic_inc(&cur_trans->use_count);
1173 btrfs_end_transaction(trans, root);
1174 schedule_delayed_work(&ac->work, 0);
1176 /* wait for transaction to start and unblock */
1177 if (wait_for_unblock)
1178 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1180 wait_current_trans_commit_start(root, cur_trans);
1182 if (current->journal_info == trans)
1183 current->journal_info = NULL;
1185 put_transaction(cur_trans);
1190 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1191 struct btrfs_root *root)
1193 struct btrfs_transaction *cur_trans = trans->transaction;
1195 WARN_ON(trans->use_count > 1);
1197 spin_lock(&root->fs_info->trans_lock);
1198 list_del_init(&cur_trans->list);
1199 spin_unlock(&root->fs_info->trans_lock);
1201 btrfs_cleanup_one_transaction(trans->transaction, root);
1203 put_transaction(cur_trans);
1204 put_transaction(cur_trans);
1206 trace_btrfs_transaction_commit(root);
1208 btrfs_scrub_continue(root);
1210 if (current->journal_info == trans)
1211 current->journal_info = NULL;
1213 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1217 * btrfs_transaction state sequence:
1218 * in_commit = 0, blocked = 0 (initial)
1219 * in_commit = 1, blocked = 1
1223 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1224 struct btrfs_root *root)
1226 unsigned long joined = 0;
1227 struct btrfs_transaction *cur_trans = trans->transaction;
1228 struct btrfs_transaction *prev_trans = NULL;
1231 int should_grow = 0;
1232 unsigned long now = get_seconds();
1233 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1235 btrfs_run_ordered_operations(root, 0);
1237 btrfs_trans_release_metadata(trans, root);
1238 trans->block_rsv = NULL;
1240 if (cur_trans->aborted)
1241 goto cleanup_transaction;
1243 /* make a pass through all the delayed refs we have so far
1244 * any runnings procs may add more while we are here
1246 ret = btrfs_run_delayed_refs(trans, root, 0);
1248 goto cleanup_transaction;
1250 cur_trans = trans->transaction;
1253 * set the flushing flag so procs in this transaction have to
1254 * start sending their work down.
1256 cur_trans->delayed_refs.flushing = 1;
1258 ret = btrfs_run_delayed_refs(trans, root, 0);
1260 goto cleanup_transaction;
1262 spin_lock(&cur_trans->commit_lock);
1263 if (cur_trans->in_commit) {
1264 spin_unlock(&cur_trans->commit_lock);
1265 atomic_inc(&cur_trans->use_count);
1266 ret = btrfs_end_transaction(trans, root);
1268 wait_for_commit(root, cur_trans);
1270 put_transaction(cur_trans);
1275 trans->transaction->in_commit = 1;
1276 trans->transaction->blocked = 1;
1277 spin_unlock(&cur_trans->commit_lock);
1278 wake_up(&root->fs_info->transaction_blocked_wait);
1280 spin_lock(&root->fs_info->trans_lock);
1281 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1282 prev_trans = list_entry(cur_trans->list.prev,
1283 struct btrfs_transaction, list);
1284 if (!prev_trans->commit_done) {
1285 atomic_inc(&prev_trans->use_count);
1286 spin_unlock(&root->fs_info->trans_lock);
1288 wait_for_commit(root, prev_trans);
1290 put_transaction(prev_trans);
1292 spin_unlock(&root->fs_info->trans_lock);
1295 spin_unlock(&root->fs_info->trans_lock);
1298 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1302 int snap_pending = 0;
1304 joined = cur_trans->num_joined;
1305 if (!list_empty(&trans->transaction->pending_snapshots))
1308 WARN_ON(cur_trans != trans->transaction);
1310 if (flush_on_commit || snap_pending) {
1311 btrfs_start_delalloc_inodes(root, 1);
1312 btrfs_wait_ordered_extents(root, 0, 1);
1315 ret = btrfs_run_delayed_items(trans, root);
1317 goto cleanup_transaction;
1320 * rename don't use btrfs_join_transaction, so, once we
1321 * set the transaction to blocked above, we aren't going
1322 * to get any new ordered operations. We can safely run
1323 * it here and no for sure that nothing new will be added
1326 btrfs_run_ordered_operations(root, 1);
1328 prepare_to_wait(&cur_trans->writer_wait, &wait,
1329 TASK_UNINTERRUPTIBLE);
1331 if (atomic_read(&cur_trans->num_writers) > 1)
1332 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1333 else if (should_grow)
1334 schedule_timeout(1);
1336 finish_wait(&cur_trans->writer_wait, &wait);
1337 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1338 (should_grow && cur_trans->num_joined != joined));
1341 * Ok now we need to make sure to block out any other joins while we
1342 * commit the transaction. We could have started a join before setting
1343 * no_join so make sure to wait for num_writers to == 1 again.
1345 spin_lock(&root->fs_info->trans_lock);
1346 root->fs_info->trans_no_join = 1;
1347 spin_unlock(&root->fs_info->trans_lock);
1348 wait_event(cur_trans->writer_wait,
1349 atomic_read(&cur_trans->num_writers) == 1);
1352 * the reloc mutex makes sure that we stop
1353 * the balancing code from coming in and moving
1354 * extents around in the middle of the commit
1356 mutex_lock(&root->fs_info->reloc_mutex);
1358 ret = btrfs_run_delayed_items(trans, root);
1360 mutex_unlock(&root->fs_info->reloc_mutex);
1361 goto cleanup_transaction;
1364 ret = create_pending_snapshots(trans, root->fs_info);
1366 mutex_unlock(&root->fs_info->reloc_mutex);
1367 goto cleanup_transaction;
1370 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1372 mutex_unlock(&root->fs_info->reloc_mutex);
1373 goto cleanup_transaction;
1377 * make sure none of the code above managed to slip in a
1380 btrfs_assert_delayed_root_empty(root);
1382 WARN_ON(cur_trans != trans->transaction);
1384 btrfs_scrub_pause(root);
1385 /* btrfs_commit_tree_roots is responsible for getting the
1386 * various roots consistent with each other. Every pointer
1387 * in the tree of tree roots has to point to the most up to date
1388 * root for every subvolume and other tree. So, we have to keep
1389 * the tree logging code from jumping in and changing any
1392 * At this point in the commit, there can't be any tree-log
1393 * writers, but a little lower down we drop the trans mutex
1394 * and let new people in. By holding the tree_log_mutex
1395 * from now until after the super is written, we avoid races
1396 * with the tree-log code.
1398 mutex_lock(&root->fs_info->tree_log_mutex);
1400 ret = commit_fs_roots(trans, root);
1402 mutex_unlock(&root->fs_info->tree_log_mutex);
1403 goto cleanup_transaction;
1406 /* commit_fs_roots gets rid of all the tree log roots, it is now
1407 * safe to free the root of tree log roots
1409 btrfs_free_log_root_tree(trans, root->fs_info);
1411 ret = commit_cowonly_roots(trans, root);
1413 mutex_unlock(&root->fs_info->tree_log_mutex);
1414 goto cleanup_transaction;
1417 btrfs_prepare_extent_commit(trans, root);
1419 cur_trans = root->fs_info->running_transaction;
1421 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1422 root->fs_info->tree_root->node);
1423 switch_commit_root(root->fs_info->tree_root);
1425 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1426 root->fs_info->chunk_root->node);
1427 switch_commit_root(root->fs_info->chunk_root);
1429 update_super_roots(root);
1431 if (!root->fs_info->log_root_recovering) {
1432 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1433 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1436 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1437 sizeof(*root->fs_info->super_copy));
1439 trans->transaction->blocked = 0;
1440 spin_lock(&root->fs_info->trans_lock);
1441 root->fs_info->running_transaction = NULL;
1442 root->fs_info->trans_no_join = 0;
1443 spin_unlock(&root->fs_info->trans_lock);
1444 mutex_unlock(&root->fs_info->reloc_mutex);
1446 wake_up(&root->fs_info->transaction_wait);
1448 ret = btrfs_write_and_wait_transaction(trans, root);
1450 btrfs_error(root->fs_info, ret,
1451 "Error while writing out transaction.");
1452 mutex_unlock(&root->fs_info->tree_log_mutex);
1453 goto cleanup_transaction;
1456 ret = write_ctree_super(trans, root, 0);
1458 mutex_unlock(&root->fs_info->tree_log_mutex);
1459 goto cleanup_transaction;
1463 * the super is written, we can safely allow the tree-loggers
1464 * to go about their business
1466 mutex_unlock(&root->fs_info->tree_log_mutex);
1468 btrfs_finish_extent_commit(trans, root);
1470 cur_trans->commit_done = 1;
1472 root->fs_info->last_trans_committed = cur_trans->transid;
1474 wake_up(&cur_trans->commit_wait);
1476 spin_lock(&root->fs_info->trans_lock);
1477 list_del_init(&cur_trans->list);
1478 spin_unlock(&root->fs_info->trans_lock);
1480 put_transaction(cur_trans);
1481 put_transaction(cur_trans);
1483 trace_btrfs_transaction_commit(root);
1485 btrfs_scrub_continue(root);
1487 if (current->journal_info == trans)
1488 current->journal_info = NULL;
1490 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1492 if (current != root->fs_info->transaction_kthread)
1493 btrfs_run_delayed_iputs(root);
1497 cleanup_transaction:
1498 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1500 if (current->journal_info == trans)
1501 current->journal_info = NULL;
1502 cleanup_transaction(trans, root);
1508 * interface function to delete all the snapshots we have scheduled for deletion
1510 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1513 struct btrfs_fs_info *fs_info = root->fs_info;
1515 spin_lock(&fs_info->trans_lock);
1516 list_splice_init(&fs_info->dead_roots, &list);
1517 spin_unlock(&fs_info->trans_lock);
1519 while (!list_empty(&list)) {
1522 root = list_entry(list.next, struct btrfs_root, root_list);
1523 list_del(&root->root_list);
1525 btrfs_kill_all_delayed_nodes(root);
1527 if (btrfs_header_backref_rev(root->node) <
1528 BTRFS_MIXED_BACKREF_REV)
1529 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1531 ret =btrfs_drop_snapshot(root, NULL, 1, 0);