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"
33 #define BTRFS_ROOT_TRANS_TAG 0
35 void put_transaction(struct btrfs_transaction *transaction)
37 WARN_ON(atomic_read(&transaction->use_count) == 0);
38 if (atomic_dec_and_test(&transaction->use_count)) {
39 BUG_ON(!list_empty(&transaction->list));
40 WARN_ON(transaction->delayed_refs.root.rb_node);
41 WARN_ON(!list_empty(&transaction->delayed_refs.seq_head));
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 (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
104 spin_unlock(&root->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;
129 cur_trans->delayed_refs.seq = 1;
132 * although the tree mod log is per file system and not per transaction,
133 * the log must never go across transaction boundaries.
136 if (!list_empty(&fs_info->tree_mod_seq_list)) {
137 printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when "
138 "creating a fresh transaction\n");
141 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) {
142 printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
143 "creating a fresh transaction\n");
146 atomic_set(&fs_info->tree_mod_seq, 0);
148 init_waitqueue_head(&cur_trans->delayed_refs.seq_wait);
149 spin_lock_init(&cur_trans->commit_lock);
150 spin_lock_init(&cur_trans->delayed_refs.lock);
151 INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head);
153 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
154 list_add_tail(&cur_trans->list, &fs_info->trans_list);
155 extent_io_tree_init(&cur_trans->dirty_pages,
156 fs_info->btree_inode->i_mapping);
157 fs_info->generation++;
158 cur_trans->transid = fs_info->generation;
159 fs_info->running_transaction = cur_trans;
160 cur_trans->aborted = 0;
161 spin_unlock(&fs_info->trans_lock);
167 * this does all the record keeping required to make sure that a reference
168 * counted root is properly recorded in a given transaction. This is required
169 * to make sure the old root from before we joined the transaction is deleted
170 * when the transaction commits
172 static int record_root_in_trans(struct btrfs_trans_handle *trans,
173 struct btrfs_root *root)
175 if (root->ref_cows && root->last_trans < trans->transid) {
176 WARN_ON(root == root->fs_info->extent_root);
177 WARN_ON(root->commit_root != root->node);
180 * see below for in_trans_setup usage rules
181 * we have the reloc mutex held now, so there
182 * is only one writer in this function
184 root->in_trans_setup = 1;
186 /* make sure readers find in_trans_setup before
187 * they find our root->last_trans update
191 spin_lock(&root->fs_info->fs_roots_radix_lock);
192 if (root->last_trans == trans->transid) {
193 spin_unlock(&root->fs_info->fs_roots_radix_lock);
196 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
197 (unsigned long)root->root_key.objectid,
198 BTRFS_ROOT_TRANS_TAG);
199 spin_unlock(&root->fs_info->fs_roots_radix_lock);
200 root->last_trans = trans->transid;
202 /* this is pretty tricky. We don't want to
203 * take the relocation lock in btrfs_record_root_in_trans
204 * unless we're really doing the first setup for this root in
207 * Normally we'd use root->last_trans as a flag to decide
208 * if we want to take the expensive mutex.
210 * But, we have to set root->last_trans before we
211 * init the relocation root, otherwise, we trip over warnings
212 * in ctree.c. The solution used here is to flag ourselves
213 * with root->in_trans_setup. When this is 1, we're still
214 * fixing up the reloc trees and everyone must wait.
216 * When this is zero, they can trust root->last_trans and fly
217 * through btrfs_record_root_in_trans without having to take the
218 * lock. smp_wmb() makes sure that all the writes above are
219 * done before we pop in the zero below
221 btrfs_init_reloc_root(trans, root);
223 root->in_trans_setup = 0;
229 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
230 struct btrfs_root *root)
236 * see record_root_in_trans for comments about in_trans_setup usage
240 if (root->last_trans == trans->transid &&
241 !root->in_trans_setup)
244 mutex_lock(&root->fs_info->reloc_mutex);
245 record_root_in_trans(trans, root);
246 mutex_unlock(&root->fs_info->reloc_mutex);
251 /* wait for commit against the current transaction to become unblocked
252 * when this is done, it is safe to start a new transaction, but the current
253 * transaction might not be fully on disk.
255 static void wait_current_trans(struct btrfs_root *root)
257 struct btrfs_transaction *cur_trans;
259 spin_lock(&root->fs_info->trans_lock);
260 cur_trans = root->fs_info->running_transaction;
261 if (cur_trans && cur_trans->blocked) {
262 atomic_inc(&cur_trans->use_count);
263 spin_unlock(&root->fs_info->trans_lock);
265 wait_event(root->fs_info->transaction_wait,
266 !cur_trans->blocked);
267 put_transaction(cur_trans);
269 spin_unlock(&root->fs_info->trans_lock);
273 enum btrfs_trans_type {
280 static int may_wait_transaction(struct btrfs_root *root, int type)
282 if (root->fs_info->log_root_recovering)
285 if (type == TRANS_USERSPACE)
288 if (type == TRANS_START &&
289 !atomic_read(&root->fs_info->open_ioctl_trans))
295 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
296 u64 num_items, int type)
298 struct btrfs_trans_handle *h;
299 struct btrfs_transaction *cur_trans;
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 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
321 ret = btrfs_block_rsv_add(root,
322 &root->fs_info->trans_block_rsv,
328 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
330 return ERR_PTR(-ENOMEM);
332 sb_start_intwrite(root->fs_info->sb);
334 if (may_wait_transaction(root, type))
335 wait_current_trans(root);
338 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
340 wait_current_trans(root);
341 } while (ret == -EBUSY);
344 sb_end_intwrite(root->fs_info->sb);
345 kmem_cache_free(btrfs_trans_handle_cachep, h);
349 cur_trans = root->fs_info->running_transaction;
351 h->transid = cur_trans->transid;
352 h->transaction = cur_trans;
354 h->bytes_reserved = 0;
355 h->delayed_ref_updates = 0;
362 if (cur_trans->blocked && may_wait_transaction(root, type)) {
363 btrfs_commit_transaction(h, root);
368 trace_btrfs_space_reservation(root->fs_info, "transaction",
369 h->transid, num_bytes, 1);
370 h->block_rsv = &root->fs_info->trans_block_rsv;
371 h->bytes_reserved = num_bytes;
375 btrfs_record_root_in_trans(h, root);
377 if (!current->journal_info && type != TRANS_USERSPACE)
378 current->journal_info = h;
382 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
385 return start_transaction(root, num_items, TRANS_START);
387 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
389 return start_transaction(root, 0, TRANS_JOIN);
392 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
394 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
397 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
399 return start_transaction(root, 0, TRANS_USERSPACE);
402 /* wait for a transaction commit to be fully complete */
403 static noinline void wait_for_commit(struct btrfs_root *root,
404 struct btrfs_transaction *commit)
406 wait_event(commit->commit_wait, commit->commit_done);
409 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
411 struct btrfs_transaction *cur_trans = NULL, *t;
416 if (transid <= root->fs_info->last_trans_committed)
419 /* find specified transaction */
420 spin_lock(&root->fs_info->trans_lock);
421 list_for_each_entry(t, &root->fs_info->trans_list, list) {
422 if (t->transid == transid) {
424 atomic_inc(&cur_trans->use_count);
427 if (t->transid > transid)
430 spin_unlock(&root->fs_info->trans_lock);
433 goto out; /* bad transid */
435 /* find newest transaction that is committing | committed */
436 spin_lock(&root->fs_info->trans_lock);
437 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
443 atomic_inc(&cur_trans->use_count);
447 spin_unlock(&root->fs_info->trans_lock);
449 goto out; /* nothing committing|committed */
452 wait_for_commit(root, cur_trans);
454 put_transaction(cur_trans);
460 void btrfs_throttle(struct btrfs_root *root)
462 if (!atomic_read(&root->fs_info->open_ioctl_trans))
463 wait_current_trans(root);
466 static int should_end_transaction(struct btrfs_trans_handle *trans,
467 struct btrfs_root *root)
471 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
475 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
476 struct btrfs_root *root)
478 struct btrfs_transaction *cur_trans = trans->transaction;
479 struct btrfs_block_rsv *rsv = trans->block_rsv;
484 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
488 * We need to do this in case we're deleting csums so the global block
489 * rsv get's used instead of the csum block rsv.
491 trans->block_rsv = NULL;
493 updates = trans->delayed_ref_updates;
494 trans->delayed_ref_updates = 0;
496 err = btrfs_run_delayed_refs(trans, root, updates);
497 if (err) /* Error code will also eval true */
501 trans->block_rsv = rsv;
503 return should_end_transaction(trans, root);
506 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
507 struct btrfs_root *root, int throttle, int lock)
509 struct btrfs_transaction *cur_trans = trans->transaction;
510 struct btrfs_fs_info *info = root->fs_info;
514 if (--trans->use_count) {
515 trans->block_rsv = trans->orig_rsv;
519 btrfs_trans_release_metadata(trans, root);
520 trans->block_rsv = NULL;
522 unsigned long cur = trans->delayed_ref_updates;
523 trans->delayed_ref_updates = 0;
525 trans->transaction->delayed_refs.num_heads_ready > 64) {
526 trans->delayed_ref_updates = 0;
527 btrfs_run_delayed_refs(trans, root, cur);
534 sb_end_intwrite(root->fs_info->sb);
536 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
537 should_end_transaction(trans, root)) {
538 trans->transaction->blocked = 1;
542 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
545 * We may race with somebody else here so end up having
546 * to call end_transaction on ourselves again, so inc
550 return btrfs_commit_transaction(trans, root);
552 wake_up_process(info->transaction_kthread);
556 WARN_ON(cur_trans != info->running_transaction);
557 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
558 atomic_dec(&cur_trans->num_writers);
561 if (waitqueue_active(&cur_trans->writer_wait))
562 wake_up(&cur_trans->writer_wait);
563 put_transaction(cur_trans);
565 if (current->journal_info == trans)
566 current->journal_info = NULL;
569 btrfs_run_delayed_iputs(root);
571 if (trans->aborted ||
572 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
576 memset(trans, 0, sizeof(*trans));
577 kmem_cache_free(btrfs_trans_handle_cachep, trans);
581 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
582 struct btrfs_root *root)
586 ret = __btrfs_end_transaction(trans, root, 0, 1);
592 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
593 struct btrfs_root *root)
597 ret = __btrfs_end_transaction(trans, root, 1, 1);
603 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
604 struct btrfs_root *root)
608 ret = __btrfs_end_transaction(trans, root, 0, 0);
614 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
615 struct btrfs_root *root)
617 return __btrfs_end_transaction(trans, root, 1, 1);
621 * when btree blocks are allocated, they have some corresponding bits set for
622 * them in one of two extent_io trees. This is used to make sure all of
623 * those extents are sent to disk but does not wait on them
625 int btrfs_write_marked_extents(struct btrfs_root *root,
626 struct extent_io_tree *dirty_pages, int mark)
630 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
634 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
636 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
638 err = filemap_fdatawrite_range(mapping, start, end);
650 * when btree blocks are allocated, they have some corresponding bits set for
651 * them in one of two extent_io trees. This is used to make sure all of
652 * those extents are on disk for transaction or log commit. We wait
653 * on all the pages and clear them from the dirty pages state tree
655 int btrfs_wait_marked_extents(struct btrfs_root *root,
656 struct extent_io_tree *dirty_pages, int mark)
660 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
664 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
666 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
667 err = filemap_fdatawait_range(mapping, start, end);
679 * when btree blocks are allocated, they have some corresponding bits set for
680 * them in one of two extent_io trees. This is used to make sure all of
681 * those extents are on disk for transaction or log commit
683 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
684 struct extent_io_tree *dirty_pages, int mark)
689 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
690 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
699 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
700 struct btrfs_root *root)
702 if (!trans || !trans->transaction) {
703 struct inode *btree_inode;
704 btree_inode = root->fs_info->btree_inode;
705 return filemap_write_and_wait(btree_inode->i_mapping);
707 return btrfs_write_and_wait_marked_extents(root,
708 &trans->transaction->dirty_pages,
713 * this is used to update the root pointer in the tree of tree roots.
715 * But, in the case of the extent allocation tree, updating the root
716 * pointer may allocate blocks which may change the root of the extent
719 * So, this loops and repeats and makes sure the cowonly root didn't
720 * change while the root pointer was being updated in the metadata.
722 static int update_cowonly_root(struct btrfs_trans_handle *trans,
723 struct btrfs_root *root)
728 struct btrfs_root *tree_root = root->fs_info->tree_root;
730 old_root_used = btrfs_root_used(&root->root_item);
731 btrfs_write_dirty_block_groups(trans, root);
734 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
735 if (old_root_bytenr == root->node->start &&
736 old_root_used == btrfs_root_used(&root->root_item))
739 btrfs_set_root_node(&root->root_item, root->node);
740 ret = btrfs_update_root(trans, tree_root,
746 old_root_used = btrfs_root_used(&root->root_item);
747 ret = btrfs_write_dirty_block_groups(trans, root);
752 if (root != root->fs_info->extent_root)
753 switch_commit_root(root);
759 * update all the cowonly tree roots on disk
761 * The error handling in this function may not be obvious. Any of the
762 * failures will cause the file system to go offline. We still need
763 * to clean up the delayed refs.
765 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
766 struct btrfs_root *root)
768 struct btrfs_fs_info *fs_info = root->fs_info;
769 struct list_head *next;
770 struct extent_buffer *eb;
773 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
777 eb = btrfs_lock_root_node(fs_info->tree_root);
778 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
780 btrfs_tree_unlock(eb);
781 free_extent_buffer(eb);
786 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
790 ret = btrfs_run_dev_stats(trans, root->fs_info);
793 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
794 next = fs_info->dirty_cowonly_roots.next;
796 root = list_entry(next, struct btrfs_root, dirty_list);
798 ret = update_cowonly_root(trans, root);
803 down_write(&fs_info->extent_commit_sem);
804 switch_commit_root(fs_info->extent_root);
805 up_write(&fs_info->extent_commit_sem);
811 * dead roots are old snapshots that need to be deleted. This allocates
812 * a dirty root struct and adds it into the list of dead roots that need to
815 int btrfs_add_dead_root(struct btrfs_root *root)
817 spin_lock(&root->fs_info->trans_lock);
818 list_add(&root->root_list, &root->fs_info->dead_roots);
819 spin_unlock(&root->fs_info->trans_lock);
824 * update all the cowonly tree roots on disk
826 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
827 struct btrfs_root *root)
829 struct btrfs_root *gang[8];
830 struct btrfs_fs_info *fs_info = root->fs_info;
835 spin_lock(&fs_info->fs_roots_radix_lock);
837 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
840 BTRFS_ROOT_TRANS_TAG);
843 for (i = 0; i < ret; i++) {
845 radix_tree_tag_clear(&fs_info->fs_roots_radix,
846 (unsigned long)root->root_key.objectid,
847 BTRFS_ROOT_TRANS_TAG);
848 spin_unlock(&fs_info->fs_roots_radix_lock);
850 btrfs_free_log(trans, root);
851 btrfs_update_reloc_root(trans, root);
852 btrfs_orphan_commit_root(trans, root);
854 btrfs_save_ino_cache(root, trans);
856 /* see comments in should_cow_block() */
860 if (root->commit_root != root->node) {
861 mutex_lock(&root->fs_commit_mutex);
862 switch_commit_root(root);
863 btrfs_unpin_free_ino(root);
864 mutex_unlock(&root->fs_commit_mutex);
866 btrfs_set_root_node(&root->root_item,
870 err = btrfs_update_root(trans, fs_info->tree_root,
873 spin_lock(&fs_info->fs_roots_radix_lock);
878 spin_unlock(&fs_info->fs_roots_radix_lock);
883 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
884 * otherwise every leaf in the btree is read and defragged.
886 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
888 struct btrfs_fs_info *info = root->fs_info;
889 struct btrfs_trans_handle *trans;
893 if (xchg(&root->defrag_running, 1))
897 trans = btrfs_start_transaction(root, 0);
899 return PTR_ERR(trans);
901 ret = btrfs_defrag_leaves(trans, root, cacheonly);
903 nr = trans->blocks_used;
904 btrfs_end_transaction(trans, root);
905 btrfs_btree_balance_dirty(info->tree_root, nr);
908 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
911 root->defrag_running = 0;
916 * new snapshots need to be created at a very specific time in the
917 * transaction commit. This does the actual creation
919 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
920 struct btrfs_fs_info *fs_info,
921 struct btrfs_pending_snapshot *pending)
923 struct btrfs_key key;
924 struct btrfs_root_item *new_root_item;
925 struct btrfs_root *tree_root = fs_info->tree_root;
926 struct btrfs_root *root = pending->root;
927 struct btrfs_root *parent_root;
928 struct btrfs_block_rsv *rsv;
929 struct inode *parent_inode;
930 struct dentry *parent;
931 struct dentry *dentry;
932 struct extent_buffer *tmp;
933 struct extent_buffer *old;
940 rsv = trans->block_rsv;
942 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
943 if (!new_root_item) {
944 ret = pending->error = -ENOMEM;
948 ret = btrfs_find_free_objectid(tree_root, &objectid);
950 pending->error = ret;
954 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
956 if (to_reserve > 0) {
957 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
960 pending->error = ret;
965 key.objectid = objectid;
966 key.offset = (u64)-1;
967 key.type = BTRFS_ROOT_ITEM_KEY;
969 trans->block_rsv = &pending->block_rsv;
971 dentry = pending->dentry;
972 parent = dget_parent(dentry);
973 parent_inode = parent->d_inode;
974 parent_root = BTRFS_I(parent_inode)->root;
975 record_root_in_trans(trans, parent_root);
978 * insert the directory item
980 ret = btrfs_set_inode_index(parent_inode, &index);
981 BUG_ON(ret); /* -ENOMEM */
982 ret = btrfs_insert_dir_item(trans, parent_root,
983 dentry->d_name.name, dentry->d_name.len,
985 BTRFS_FT_DIR, index);
986 if (ret == -EEXIST) {
987 pending->error = -EEXIST;
991 goto abort_trans_dput;
994 btrfs_i_size_write(parent_inode, parent_inode->i_size +
995 dentry->d_name.len * 2);
996 ret = btrfs_update_inode(trans, parent_root, parent_inode);
998 goto abort_trans_dput;
1001 * pull in the delayed directory update
1002 * and the delayed inode item
1003 * otherwise we corrupt the FS during
1006 ret = btrfs_run_delayed_items(trans, root);
1007 if (ret) { /* Transaction aborted */
1012 record_root_in_trans(trans, root);
1013 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1014 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1015 btrfs_check_and_init_root_item(new_root_item);
1017 root_flags = btrfs_root_flags(new_root_item);
1018 if (pending->readonly)
1019 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1021 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1022 btrfs_set_root_flags(new_root_item, root_flags);
1024 old = btrfs_lock_root_node(root);
1025 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1027 btrfs_tree_unlock(old);
1028 free_extent_buffer(old);
1029 goto abort_trans_dput;
1032 btrfs_set_lock_blocking(old);
1034 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1035 /* clean up in any case */
1036 btrfs_tree_unlock(old);
1037 free_extent_buffer(old);
1039 goto abort_trans_dput;
1041 /* see comments in should_cow_block() */
1042 root->force_cow = 1;
1045 btrfs_set_root_node(new_root_item, tmp);
1046 /* record when the snapshot was created in key.offset */
1047 key.offset = trans->transid;
1048 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1049 btrfs_tree_unlock(tmp);
1050 free_extent_buffer(tmp);
1052 goto abort_trans_dput;
1055 * insert root back/forward references
1057 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1058 parent_root->root_key.objectid,
1059 btrfs_ino(parent_inode), index,
1060 dentry->d_name.name, dentry->d_name.len);
1065 key.offset = (u64)-1;
1066 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1067 if (IS_ERR(pending->snap)) {
1068 ret = PTR_ERR(pending->snap);
1072 ret = btrfs_reloc_post_snapshot(trans, pending);
1077 kfree(new_root_item);
1078 trans->block_rsv = rsv;
1079 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1085 btrfs_abort_transaction(trans, root, ret);
1090 * create all the snapshots we've scheduled for creation
1092 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1093 struct btrfs_fs_info *fs_info)
1095 struct btrfs_pending_snapshot *pending;
1096 struct list_head *head = &trans->transaction->pending_snapshots;
1098 list_for_each_entry(pending, head, list)
1099 create_pending_snapshot(trans, fs_info, pending);
1103 static void update_super_roots(struct btrfs_root *root)
1105 struct btrfs_root_item *root_item;
1106 struct btrfs_super_block *super;
1108 super = root->fs_info->super_copy;
1110 root_item = &root->fs_info->chunk_root->root_item;
1111 super->chunk_root = root_item->bytenr;
1112 super->chunk_root_generation = root_item->generation;
1113 super->chunk_root_level = root_item->level;
1115 root_item = &root->fs_info->tree_root->root_item;
1116 super->root = root_item->bytenr;
1117 super->generation = root_item->generation;
1118 super->root_level = root_item->level;
1119 if (btrfs_test_opt(root, SPACE_CACHE))
1120 super->cache_generation = root_item->generation;
1123 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1126 spin_lock(&info->trans_lock);
1127 if (info->running_transaction)
1128 ret = info->running_transaction->in_commit;
1129 spin_unlock(&info->trans_lock);
1133 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1136 spin_lock(&info->trans_lock);
1137 if (info->running_transaction)
1138 ret = info->running_transaction->blocked;
1139 spin_unlock(&info->trans_lock);
1144 * wait for the current transaction commit to start and block subsequent
1147 static void wait_current_trans_commit_start(struct btrfs_root *root,
1148 struct btrfs_transaction *trans)
1150 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1154 * wait for the current transaction to start and then become unblocked.
1157 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1158 struct btrfs_transaction *trans)
1160 wait_event(root->fs_info->transaction_wait,
1161 trans->commit_done || (trans->in_commit && !trans->blocked));
1165 * commit transactions asynchronously. once btrfs_commit_transaction_async
1166 * returns, any subsequent transaction will not be allowed to join.
1168 struct btrfs_async_commit {
1169 struct btrfs_trans_handle *newtrans;
1170 struct btrfs_root *root;
1171 struct delayed_work work;
1174 static void do_async_commit(struct work_struct *work)
1176 struct btrfs_async_commit *ac =
1177 container_of(work, struct btrfs_async_commit, work.work);
1179 btrfs_commit_transaction(ac->newtrans, ac->root);
1183 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1184 struct btrfs_root *root,
1185 int wait_for_unblock)
1187 struct btrfs_async_commit *ac;
1188 struct btrfs_transaction *cur_trans;
1190 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1194 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1196 ac->newtrans = btrfs_join_transaction(root);
1197 if (IS_ERR(ac->newtrans)) {
1198 int err = PTR_ERR(ac->newtrans);
1203 /* take transaction reference */
1204 cur_trans = trans->transaction;
1205 atomic_inc(&cur_trans->use_count);
1207 btrfs_end_transaction(trans, root);
1208 schedule_delayed_work(&ac->work, 0);
1210 /* wait for transaction to start and unblock */
1211 if (wait_for_unblock)
1212 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1214 wait_current_trans_commit_start(root, cur_trans);
1216 if (current->journal_info == trans)
1217 current->journal_info = NULL;
1219 put_transaction(cur_trans);
1224 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1225 struct btrfs_root *root, int err)
1227 struct btrfs_transaction *cur_trans = trans->transaction;
1229 WARN_ON(trans->use_count > 1);
1231 btrfs_abort_transaction(trans, root, err);
1233 spin_lock(&root->fs_info->trans_lock);
1234 list_del_init(&cur_trans->list);
1235 if (cur_trans == root->fs_info->running_transaction) {
1236 root->fs_info->running_transaction = NULL;
1237 root->fs_info->trans_no_join = 0;
1239 spin_unlock(&root->fs_info->trans_lock);
1241 btrfs_cleanup_one_transaction(trans->transaction, root);
1243 put_transaction(cur_trans);
1244 put_transaction(cur_trans);
1246 trace_btrfs_transaction_commit(root);
1248 btrfs_scrub_continue(root);
1250 if (current->journal_info == trans)
1251 current->journal_info = NULL;
1253 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1257 * btrfs_transaction state sequence:
1258 * in_commit = 0, blocked = 0 (initial)
1259 * in_commit = 1, blocked = 1
1263 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1264 struct btrfs_root *root)
1266 unsigned long joined = 0;
1267 struct btrfs_transaction *cur_trans = trans->transaction;
1268 struct btrfs_transaction *prev_trans = NULL;
1271 int should_grow = 0;
1272 unsigned long now = get_seconds();
1273 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1275 btrfs_run_ordered_operations(root, 0);
1277 btrfs_trans_release_metadata(trans, root);
1278 trans->block_rsv = NULL;
1280 if (cur_trans->aborted)
1281 goto cleanup_transaction;
1283 /* make a pass through all the delayed refs we have so far
1284 * any runnings procs may add more while we are here
1286 ret = btrfs_run_delayed_refs(trans, root, 0);
1288 goto cleanup_transaction;
1290 cur_trans = trans->transaction;
1293 * set the flushing flag so procs in this transaction have to
1294 * start sending their work down.
1296 cur_trans->delayed_refs.flushing = 1;
1298 ret = btrfs_run_delayed_refs(trans, root, 0);
1300 goto cleanup_transaction;
1302 spin_lock(&cur_trans->commit_lock);
1303 if (cur_trans->in_commit) {
1304 spin_unlock(&cur_trans->commit_lock);
1305 atomic_inc(&cur_trans->use_count);
1306 ret = btrfs_end_transaction(trans, root);
1308 wait_for_commit(root, cur_trans);
1310 put_transaction(cur_trans);
1315 trans->transaction->in_commit = 1;
1316 trans->transaction->blocked = 1;
1317 spin_unlock(&cur_trans->commit_lock);
1318 wake_up(&root->fs_info->transaction_blocked_wait);
1320 spin_lock(&root->fs_info->trans_lock);
1321 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1322 prev_trans = list_entry(cur_trans->list.prev,
1323 struct btrfs_transaction, list);
1324 if (!prev_trans->commit_done) {
1325 atomic_inc(&prev_trans->use_count);
1326 spin_unlock(&root->fs_info->trans_lock);
1328 wait_for_commit(root, prev_trans);
1330 put_transaction(prev_trans);
1332 spin_unlock(&root->fs_info->trans_lock);
1335 spin_unlock(&root->fs_info->trans_lock);
1338 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1342 int snap_pending = 0;
1344 joined = cur_trans->num_joined;
1345 if (!list_empty(&trans->transaction->pending_snapshots))
1348 WARN_ON(cur_trans != trans->transaction);
1350 if (flush_on_commit || snap_pending) {
1351 btrfs_start_delalloc_inodes(root, 1);
1352 btrfs_wait_ordered_extents(root, 0, 1);
1355 ret = btrfs_run_delayed_items(trans, root);
1357 goto cleanup_transaction;
1360 * rename don't use btrfs_join_transaction, so, once we
1361 * set the transaction to blocked above, we aren't going
1362 * to get any new ordered operations. We can safely run
1363 * it here and no for sure that nothing new will be added
1366 btrfs_run_ordered_operations(root, 1);
1368 prepare_to_wait(&cur_trans->writer_wait, &wait,
1369 TASK_UNINTERRUPTIBLE);
1371 if (atomic_read(&cur_trans->num_writers) > 1)
1372 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1373 else if (should_grow)
1374 schedule_timeout(1);
1376 finish_wait(&cur_trans->writer_wait, &wait);
1377 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1378 (should_grow && cur_trans->num_joined != joined));
1381 * Ok now we need to make sure to block out any other joins while we
1382 * commit the transaction. We could have started a join before setting
1383 * no_join so make sure to wait for num_writers to == 1 again.
1385 spin_lock(&root->fs_info->trans_lock);
1386 root->fs_info->trans_no_join = 1;
1387 spin_unlock(&root->fs_info->trans_lock);
1388 wait_event(cur_trans->writer_wait,
1389 atomic_read(&cur_trans->num_writers) == 1);
1392 * the reloc mutex makes sure that we stop
1393 * the balancing code from coming in and moving
1394 * extents around in the middle of the commit
1396 mutex_lock(&root->fs_info->reloc_mutex);
1398 ret = btrfs_run_delayed_items(trans, root);
1400 mutex_unlock(&root->fs_info->reloc_mutex);
1401 goto cleanup_transaction;
1404 ret = create_pending_snapshots(trans, root->fs_info);
1406 mutex_unlock(&root->fs_info->reloc_mutex);
1407 goto cleanup_transaction;
1410 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1412 mutex_unlock(&root->fs_info->reloc_mutex);
1413 goto cleanup_transaction;
1417 * make sure none of the code above managed to slip in a
1420 btrfs_assert_delayed_root_empty(root);
1422 WARN_ON(cur_trans != trans->transaction);
1424 btrfs_scrub_pause(root);
1425 /* btrfs_commit_tree_roots is responsible for getting the
1426 * various roots consistent with each other. Every pointer
1427 * in the tree of tree roots has to point to the most up to date
1428 * root for every subvolume and other tree. So, we have to keep
1429 * the tree logging code from jumping in and changing any
1432 * At this point in the commit, there can't be any tree-log
1433 * writers, but a little lower down we drop the trans mutex
1434 * and let new people in. By holding the tree_log_mutex
1435 * from now until after the super is written, we avoid races
1436 * with the tree-log code.
1438 mutex_lock(&root->fs_info->tree_log_mutex);
1440 ret = commit_fs_roots(trans, root);
1442 mutex_unlock(&root->fs_info->tree_log_mutex);
1443 mutex_unlock(&root->fs_info->reloc_mutex);
1444 goto cleanup_transaction;
1447 /* commit_fs_roots gets rid of all the tree log roots, it is now
1448 * safe to free the root of tree log roots
1450 btrfs_free_log_root_tree(trans, root->fs_info);
1452 ret = commit_cowonly_roots(trans, root);
1454 mutex_unlock(&root->fs_info->tree_log_mutex);
1455 mutex_unlock(&root->fs_info->reloc_mutex);
1456 goto cleanup_transaction;
1459 btrfs_prepare_extent_commit(trans, root);
1461 cur_trans = root->fs_info->running_transaction;
1463 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1464 root->fs_info->tree_root->node);
1465 switch_commit_root(root->fs_info->tree_root);
1467 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1468 root->fs_info->chunk_root->node);
1469 switch_commit_root(root->fs_info->chunk_root);
1471 update_super_roots(root);
1473 if (!root->fs_info->log_root_recovering) {
1474 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1475 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1478 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1479 sizeof(*root->fs_info->super_copy));
1481 trans->transaction->blocked = 0;
1482 spin_lock(&root->fs_info->trans_lock);
1483 root->fs_info->running_transaction = NULL;
1484 root->fs_info->trans_no_join = 0;
1485 spin_unlock(&root->fs_info->trans_lock);
1486 mutex_unlock(&root->fs_info->reloc_mutex);
1488 wake_up(&root->fs_info->transaction_wait);
1490 ret = btrfs_write_and_wait_transaction(trans, root);
1492 btrfs_error(root->fs_info, ret,
1493 "Error while writing out transaction.");
1494 mutex_unlock(&root->fs_info->tree_log_mutex);
1495 goto cleanup_transaction;
1498 ret = write_ctree_super(trans, root, 0);
1500 mutex_unlock(&root->fs_info->tree_log_mutex);
1501 goto cleanup_transaction;
1505 * the super is written, we can safely allow the tree-loggers
1506 * to go about their business
1508 mutex_unlock(&root->fs_info->tree_log_mutex);
1510 btrfs_finish_extent_commit(trans, root);
1512 cur_trans->commit_done = 1;
1514 root->fs_info->last_trans_committed = cur_trans->transid;
1516 wake_up(&cur_trans->commit_wait);
1518 spin_lock(&root->fs_info->trans_lock);
1519 list_del_init(&cur_trans->list);
1520 spin_unlock(&root->fs_info->trans_lock);
1522 put_transaction(cur_trans);
1523 put_transaction(cur_trans);
1525 sb_end_intwrite(root->fs_info->sb);
1527 trace_btrfs_transaction_commit(root);
1529 btrfs_scrub_continue(root);
1531 if (current->journal_info == trans)
1532 current->journal_info = NULL;
1534 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1536 if (current != root->fs_info->transaction_kthread)
1537 btrfs_run_delayed_iputs(root);
1541 cleanup_transaction:
1542 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1544 if (current->journal_info == trans)
1545 current->journal_info = NULL;
1546 cleanup_transaction(trans, root, ret);
1552 * interface function to delete all the snapshots we have scheduled for deletion
1554 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1557 struct btrfs_fs_info *fs_info = root->fs_info;
1559 spin_lock(&fs_info->trans_lock);
1560 list_splice_init(&fs_info->dead_roots, &list);
1561 spin_unlock(&fs_info->trans_lock);
1563 while (!list_empty(&list)) {
1566 root = list_entry(list.next, struct btrfs_root, root_list);
1567 list_del(&root->root_list);
1569 btrfs_kill_all_delayed_nodes(root);
1571 if (btrfs_header_backref_rev(root->node) <
1572 BTRFS_MIXED_BACKREF_REV)
1573 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1575 ret =btrfs_drop_snapshot(root, NULL, 1, 0);