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;
105 atomic_set(&cur_trans->num_writers, 1);
106 cur_trans->num_joined = 0;
107 init_waitqueue_head(&cur_trans->writer_wait);
108 init_waitqueue_head(&cur_trans->commit_wait);
109 cur_trans->in_commit = 0;
110 cur_trans->blocked = 0;
112 * One for this trans handle, one so it will live on until we
113 * commit the transaction.
115 atomic_set(&cur_trans->use_count, 2);
116 cur_trans->commit_done = 0;
117 cur_trans->start_time = get_seconds();
119 cur_trans->delayed_refs.root = RB_ROOT;
120 cur_trans->delayed_refs.num_entries = 0;
121 cur_trans->delayed_refs.num_heads_ready = 0;
122 cur_trans->delayed_refs.num_heads = 0;
123 cur_trans->delayed_refs.flushing = 0;
124 cur_trans->delayed_refs.run_delayed_start = 0;
125 cur_trans->delayed_refs.seq = 1;
128 * although the tree mod log is per file system and not per transaction,
129 * the log must never go across transaction boundaries.
132 if (!list_empty(&fs_info->tree_mod_seq_list)) {
133 printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when "
134 "creating a fresh transaction\n");
137 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) {
138 printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
139 "creating a fresh transaction\n");
142 atomic_set(&fs_info->tree_mod_seq, 0);
144 init_waitqueue_head(&cur_trans->delayed_refs.seq_wait);
145 spin_lock_init(&cur_trans->commit_lock);
146 spin_lock_init(&cur_trans->delayed_refs.lock);
147 INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head);
149 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
150 list_add_tail(&cur_trans->list, &fs_info->trans_list);
151 extent_io_tree_init(&cur_trans->dirty_pages,
152 fs_info->btree_inode->i_mapping);
153 fs_info->generation++;
154 cur_trans->transid = fs_info->generation;
155 fs_info->running_transaction = cur_trans;
156 cur_trans->aborted = 0;
157 spin_unlock(&fs_info->trans_lock);
163 * this does all the record keeping required to make sure that a reference
164 * counted root is properly recorded in a given transaction. This is required
165 * to make sure the old root from before we joined the transaction is deleted
166 * when the transaction commits
168 static int record_root_in_trans(struct btrfs_trans_handle *trans,
169 struct btrfs_root *root)
171 if (root->ref_cows && root->last_trans < trans->transid) {
172 WARN_ON(root == root->fs_info->extent_root);
173 WARN_ON(root->commit_root != root->node);
176 * see below for in_trans_setup usage rules
177 * we have the reloc mutex held now, so there
178 * is only one writer in this function
180 root->in_trans_setup = 1;
182 /* make sure readers find in_trans_setup before
183 * they find our root->last_trans update
187 spin_lock(&root->fs_info->fs_roots_radix_lock);
188 if (root->last_trans == trans->transid) {
189 spin_unlock(&root->fs_info->fs_roots_radix_lock);
192 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
193 (unsigned long)root->root_key.objectid,
194 BTRFS_ROOT_TRANS_TAG);
195 spin_unlock(&root->fs_info->fs_roots_radix_lock);
196 root->last_trans = trans->transid;
198 /* this is pretty tricky. We don't want to
199 * take the relocation lock in btrfs_record_root_in_trans
200 * unless we're really doing the first setup for this root in
203 * Normally we'd use root->last_trans as a flag to decide
204 * if we want to take the expensive mutex.
206 * But, we have to set root->last_trans before we
207 * init the relocation root, otherwise, we trip over warnings
208 * in ctree.c. The solution used here is to flag ourselves
209 * with root->in_trans_setup. When this is 1, we're still
210 * fixing up the reloc trees and everyone must wait.
212 * When this is zero, they can trust root->last_trans and fly
213 * through btrfs_record_root_in_trans without having to take the
214 * lock. smp_wmb() makes sure that all the writes above are
215 * done before we pop in the zero below
217 btrfs_init_reloc_root(trans, root);
219 root->in_trans_setup = 0;
225 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
226 struct btrfs_root *root)
232 * see record_root_in_trans for comments about in_trans_setup usage
236 if (root->last_trans == trans->transid &&
237 !root->in_trans_setup)
240 mutex_lock(&root->fs_info->reloc_mutex);
241 record_root_in_trans(trans, root);
242 mutex_unlock(&root->fs_info->reloc_mutex);
247 /* wait for commit against the current transaction to become unblocked
248 * when this is done, it is safe to start a new transaction, but the current
249 * transaction might not be fully on disk.
251 static void wait_current_trans(struct btrfs_root *root)
253 struct btrfs_transaction *cur_trans;
255 spin_lock(&root->fs_info->trans_lock);
256 cur_trans = root->fs_info->running_transaction;
257 if (cur_trans && cur_trans->blocked) {
258 atomic_inc(&cur_trans->use_count);
259 spin_unlock(&root->fs_info->trans_lock);
261 wait_event(root->fs_info->transaction_wait,
262 !cur_trans->blocked);
263 put_transaction(cur_trans);
265 spin_unlock(&root->fs_info->trans_lock);
269 enum btrfs_trans_type {
276 static int may_wait_transaction(struct btrfs_root *root, int type)
278 if (root->fs_info->log_root_recovering)
281 if (type == TRANS_USERSPACE)
284 if (type == TRANS_START &&
285 !atomic_read(&root->fs_info->open_ioctl_trans))
291 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
292 u64 num_items, int type)
294 struct btrfs_trans_handle *h;
295 struct btrfs_transaction *cur_trans;
299 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
300 return ERR_PTR(-EROFS);
302 if (current->journal_info) {
303 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
304 h = current->journal_info;
306 h->orig_rsv = h->block_rsv;
312 * Do the reservation before we join the transaction so we can do all
313 * the appropriate flushing if need be.
315 if (num_items > 0 && root != root->fs_info->chunk_root) {
316 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
317 ret = btrfs_block_rsv_add(root,
318 &root->fs_info->trans_block_rsv,
324 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
326 return ERR_PTR(-ENOMEM);
328 if (may_wait_transaction(root, type))
329 wait_current_trans(root);
332 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
334 wait_current_trans(root);
335 } while (ret == -EBUSY);
338 kmem_cache_free(btrfs_trans_handle_cachep, h);
342 cur_trans = root->fs_info->running_transaction;
344 h->transid = cur_trans->transid;
345 h->transaction = cur_trans;
347 h->bytes_reserved = 0;
348 h->delayed_ref_updates = 0;
355 if (cur_trans->blocked && may_wait_transaction(root, type)) {
356 btrfs_commit_transaction(h, root);
361 trace_btrfs_space_reservation(root->fs_info, "transaction",
362 h->transid, num_bytes, 1);
363 h->block_rsv = &root->fs_info->trans_block_rsv;
364 h->bytes_reserved = num_bytes;
368 btrfs_record_root_in_trans(h, root);
370 if (!current->journal_info && type != TRANS_USERSPACE)
371 current->journal_info = h;
375 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
378 return start_transaction(root, num_items, TRANS_START);
380 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
382 return start_transaction(root, 0, TRANS_JOIN);
385 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
387 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
390 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
392 return start_transaction(root, 0, TRANS_USERSPACE);
395 /* wait for a transaction commit to be fully complete */
396 static noinline void wait_for_commit(struct btrfs_root *root,
397 struct btrfs_transaction *commit)
399 wait_event(commit->commit_wait, commit->commit_done);
402 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
404 struct btrfs_transaction *cur_trans = NULL, *t;
409 if (transid <= root->fs_info->last_trans_committed)
412 /* find specified transaction */
413 spin_lock(&root->fs_info->trans_lock);
414 list_for_each_entry(t, &root->fs_info->trans_list, list) {
415 if (t->transid == transid) {
417 atomic_inc(&cur_trans->use_count);
420 if (t->transid > transid)
423 spin_unlock(&root->fs_info->trans_lock);
426 goto out; /* bad transid */
428 /* find newest transaction that is committing | committed */
429 spin_lock(&root->fs_info->trans_lock);
430 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
436 atomic_inc(&cur_trans->use_count);
440 spin_unlock(&root->fs_info->trans_lock);
442 goto out; /* nothing committing|committed */
445 wait_for_commit(root, cur_trans);
447 put_transaction(cur_trans);
453 void btrfs_throttle(struct btrfs_root *root)
455 if (!atomic_read(&root->fs_info->open_ioctl_trans))
456 wait_current_trans(root);
459 static int should_end_transaction(struct btrfs_trans_handle *trans,
460 struct btrfs_root *root)
464 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
468 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
469 struct btrfs_root *root)
471 struct btrfs_transaction *cur_trans = trans->transaction;
472 struct btrfs_block_rsv *rsv = trans->block_rsv;
477 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
481 * We need to do this in case we're deleting csums so the global block
482 * rsv get's used instead of the csum block rsv.
484 trans->block_rsv = NULL;
486 updates = trans->delayed_ref_updates;
487 trans->delayed_ref_updates = 0;
489 err = btrfs_run_delayed_refs(trans, root, updates);
490 if (err) /* Error code will also eval true */
494 trans->block_rsv = rsv;
496 return should_end_transaction(trans, root);
499 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
500 struct btrfs_root *root, int throttle, int lock)
502 struct btrfs_transaction *cur_trans = trans->transaction;
503 struct btrfs_fs_info *info = root->fs_info;
507 if (--trans->use_count) {
508 trans->block_rsv = trans->orig_rsv;
512 btrfs_trans_release_metadata(trans, root);
513 trans->block_rsv = NULL;
515 unsigned long cur = trans->delayed_ref_updates;
516 trans->delayed_ref_updates = 0;
518 trans->transaction->delayed_refs.num_heads_ready > 64) {
519 trans->delayed_ref_updates = 0;
520 btrfs_run_delayed_refs(trans, root, cur);
527 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
528 should_end_transaction(trans, root)) {
529 trans->transaction->blocked = 1;
533 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
536 * We may race with somebody else here so end up having
537 * to call end_transaction on ourselves again, so inc
541 return btrfs_commit_transaction(trans, root);
543 wake_up_process(info->transaction_kthread);
547 WARN_ON(cur_trans != info->running_transaction);
548 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
549 atomic_dec(&cur_trans->num_writers);
552 if (waitqueue_active(&cur_trans->writer_wait))
553 wake_up(&cur_trans->writer_wait);
554 put_transaction(cur_trans);
556 if (current->journal_info == trans)
557 current->journal_info = NULL;
560 btrfs_run_delayed_iputs(root);
562 if (trans->aborted ||
563 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
567 memset(trans, 0, sizeof(*trans));
568 kmem_cache_free(btrfs_trans_handle_cachep, trans);
572 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
573 struct btrfs_root *root)
577 ret = __btrfs_end_transaction(trans, root, 0, 1);
583 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
584 struct btrfs_root *root)
588 ret = __btrfs_end_transaction(trans, root, 1, 1);
594 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
595 struct btrfs_root *root)
599 ret = __btrfs_end_transaction(trans, root, 0, 0);
605 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
606 struct btrfs_root *root)
608 return __btrfs_end_transaction(trans, root, 1, 1);
612 * when btree blocks are allocated, they have some corresponding bits set for
613 * them in one of two extent_io trees. This is used to make sure all of
614 * those extents are sent to disk but does not wait on them
616 int btrfs_write_marked_extents(struct btrfs_root *root,
617 struct extent_io_tree *dirty_pages, int mark)
621 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
625 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
627 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
629 err = filemap_fdatawrite_range(mapping, start, end);
641 * when btree blocks are allocated, they have some corresponding bits set for
642 * them in one of two extent_io trees. This is used to make sure all of
643 * those extents are on disk for transaction or log commit. We wait
644 * on all the pages and clear them from the dirty pages state tree
646 int btrfs_wait_marked_extents(struct btrfs_root *root,
647 struct extent_io_tree *dirty_pages, int mark)
651 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
655 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
657 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
658 err = filemap_fdatawait_range(mapping, start, end);
670 * when btree blocks are allocated, they have some corresponding bits set for
671 * them in one of two extent_io trees. This is used to make sure all of
672 * those extents are on disk for transaction or log commit
674 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
675 struct extent_io_tree *dirty_pages, int mark)
680 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
681 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
690 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
691 struct btrfs_root *root)
693 if (!trans || !trans->transaction) {
694 struct inode *btree_inode;
695 btree_inode = root->fs_info->btree_inode;
696 return filemap_write_and_wait(btree_inode->i_mapping);
698 return btrfs_write_and_wait_marked_extents(root,
699 &trans->transaction->dirty_pages,
704 * this is used to update the root pointer in the tree of tree roots.
706 * But, in the case of the extent allocation tree, updating the root
707 * pointer may allocate blocks which may change the root of the extent
710 * So, this loops and repeats and makes sure the cowonly root didn't
711 * change while the root pointer was being updated in the metadata.
713 static int update_cowonly_root(struct btrfs_trans_handle *trans,
714 struct btrfs_root *root)
719 struct btrfs_root *tree_root = root->fs_info->tree_root;
721 old_root_used = btrfs_root_used(&root->root_item);
722 btrfs_write_dirty_block_groups(trans, root);
725 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
726 if (old_root_bytenr == root->node->start &&
727 old_root_used == btrfs_root_used(&root->root_item))
730 btrfs_set_root_node(&root->root_item, root->node);
731 ret = btrfs_update_root(trans, tree_root,
737 old_root_used = btrfs_root_used(&root->root_item);
738 ret = btrfs_write_dirty_block_groups(trans, root);
743 if (root != root->fs_info->extent_root)
744 switch_commit_root(root);
750 * update all the cowonly tree roots on disk
752 * The error handling in this function may not be obvious. Any of the
753 * failures will cause the file system to go offline. We still need
754 * to clean up the delayed refs.
756 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
757 struct btrfs_root *root)
759 struct btrfs_fs_info *fs_info = root->fs_info;
760 struct list_head *next;
761 struct extent_buffer *eb;
764 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
768 eb = btrfs_lock_root_node(fs_info->tree_root);
769 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
771 btrfs_tree_unlock(eb);
772 free_extent_buffer(eb);
777 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
781 ret = btrfs_run_dev_stats(trans, root->fs_info);
784 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
785 next = fs_info->dirty_cowonly_roots.next;
787 root = list_entry(next, struct btrfs_root, dirty_list);
789 ret = update_cowonly_root(trans, root);
794 down_write(&fs_info->extent_commit_sem);
795 switch_commit_root(fs_info->extent_root);
796 up_write(&fs_info->extent_commit_sem);
802 * dead roots are old snapshots that need to be deleted. This allocates
803 * a dirty root struct and adds it into the list of dead roots that need to
806 int btrfs_add_dead_root(struct btrfs_root *root)
808 spin_lock(&root->fs_info->trans_lock);
809 list_add(&root->root_list, &root->fs_info->dead_roots);
810 spin_unlock(&root->fs_info->trans_lock);
815 * update all the cowonly tree roots on disk
817 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
818 struct btrfs_root *root)
820 struct btrfs_root *gang[8];
821 struct btrfs_fs_info *fs_info = root->fs_info;
826 spin_lock(&fs_info->fs_roots_radix_lock);
828 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
831 BTRFS_ROOT_TRANS_TAG);
834 for (i = 0; i < ret; i++) {
836 radix_tree_tag_clear(&fs_info->fs_roots_radix,
837 (unsigned long)root->root_key.objectid,
838 BTRFS_ROOT_TRANS_TAG);
839 spin_unlock(&fs_info->fs_roots_radix_lock);
841 btrfs_free_log(trans, root);
842 btrfs_update_reloc_root(trans, root);
843 btrfs_orphan_commit_root(trans, root);
845 btrfs_save_ino_cache(root, trans);
847 /* see comments in should_cow_block() */
851 if (root->commit_root != root->node) {
852 mutex_lock(&root->fs_commit_mutex);
853 switch_commit_root(root);
854 btrfs_unpin_free_ino(root);
855 mutex_unlock(&root->fs_commit_mutex);
857 btrfs_set_root_node(&root->root_item,
861 err = btrfs_update_root(trans, fs_info->tree_root,
864 spin_lock(&fs_info->fs_roots_radix_lock);
869 spin_unlock(&fs_info->fs_roots_radix_lock);
874 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
875 * otherwise every leaf in the btree is read and defragged.
877 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
879 struct btrfs_fs_info *info = root->fs_info;
880 struct btrfs_trans_handle *trans;
884 if (xchg(&root->defrag_running, 1))
888 trans = btrfs_start_transaction(root, 0);
890 return PTR_ERR(trans);
892 ret = btrfs_defrag_leaves(trans, root, cacheonly);
894 nr = trans->blocks_used;
895 btrfs_end_transaction(trans, root);
896 btrfs_btree_balance_dirty(info->tree_root, nr);
899 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
902 root->defrag_running = 0;
907 * new snapshots need to be created at a very specific time in the
908 * transaction commit. This does the actual creation
910 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
911 struct btrfs_fs_info *fs_info,
912 struct btrfs_pending_snapshot *pending)
914 struct btrfs_key key;
915 struct btrfs_root_item *new_root_item;
916 struct btrfs_root *tree_root = fs_info->tree_root;
917 struct btrfs_root *root = pending->root;
918 struct btrfs_root *parent_root;
919 struct btrfs_block_rsv *rsv;
920 struct inode *parent_inode;
921 struct dentry *parent;
922 struct dentry *dentry;
923 struct extent_buffer *tmp;
924 struct extent_buffer *old;
931 rsv = trans->block_rsv;
933 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
934 if (!new_root_item) {
935 ret = pending->error = -ENOMEM;
939 ret = btrfs_find_free_objectid(tree_root, &objectid);
941 pending->error = ret;
945 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
947 if (to_reserve > 0) {
948 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
951 pending->error = ret;
956 key.objectid = objectid;
957 key.offset = (u64)-1;
958 key.type = BTRFS_ROOT_ITEM_KEY;
960 trans->block_rsv = &pending->block_rsv;
962 dentry = pending->dentry;
963 parent = dget_parent(dentry);
964 parent_inode = parent->d_inode;
965 parent_root = BTRFS_I(parent_inode)->root;
966 record_root_in_trans(trans, parent_root);
969 * insert the directory item
971 ret = btrfs_set_inode_index(parent_inode, &index);
972 BUG_ON(ret); /* -ENOMEM */
973 ret = btrfs_insert_dir_item(trans, parent_root,
974 dentry->d_name.name, dentry->d_name.len,
976 BTRFS_FT_DIR, index);
977 if (ret == -EEXIST) {
978 pending->error = -EEXIST;
982 goto abort_trans_dput;
985 btrfs_i_size_write(parent_inode, parent_inode->i_size +
986 dentry->d_name.len * 2);
987 ret = btrfs_update_inode(trans, parent_root, parent_inode);
989 goto abort_trans_dput;
992 * pull in the delayed directory update
993 * and the delayed inode item
994 * otherwise we corrupt the FS during
997 ret = btrfs_run_delayed_items(trans, root);
998 if (ret) { /* Transaction aborted */
1003 record_root_in_trans(trans, root);
1004 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1005 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1006 btrfs_check_and_init_root_item(new_root_item);
1008 root_flags = btrfs_root_flags(new_root_item);
1009 if (pending->readonly)
1010 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1012 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1013 btrfs_set_root_flags(new_root_item, root_flags);
1015 old = btrfs_lock_root_node(root);
1016 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1018 btrfs_tree_unlock(old);
1019 free_extent_buffer(old);
1020 goto abort_trans_dput;
1023 btrfs_set_lock_blocking(old);
1025 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1026 /* clean up in any case */
1027 btrfs_tree_unlock(old);
1028 free_extent_buffer(old);
1030 goto abort_trans_dput;
1032 /* see comments in should_cow_block() */
1033 root->force_cow = 1;
1036 btrfs_set_root_node(new_root_item, tmp);
1037 /* record when the snapshot was created in key.offset */
1038 key.offset = trans->transid;
1039 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1040 btrfs_tree_unlock(tmp);
1041 free_extent_buffer(tmp);
1043 goto abort_trans_dput;
1046 * insert root back/forward references
1048 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1049 parent_root->root_key.objectid,
1050 btrfs_ino(parent_inode), index,
1051 dentry->d_name.name, dentry->d_name.len);
1056 key.offset = (u64)-1;
1057 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1058 if (IS_ERR(pending->snap)) {
1059 ret = PTR_ERR(pending->snap);
1063 ret = btrfs_reloc_post_snapshot(trans, pending);
1068 kfree(new_root_item);
1069 trans->block_rsv = rsv;
1070 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1076 btrfs_abort_transaction(trans, root, ret);
1081 * create all the snapshots we've scheduled for creation
1083 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1084 struct btrfs_fs_info *fs_info)
1086 struct btrfs_pending_snapshot *pending;
1087 struct list_head *head = &trans->transaction->pending_snapshots;
1089 list_for_each_entry(pending, head, list)
1090 create_pending_snapshot(trans, fs_info, pending);
1094 static void update_super_roots(struct btrfs_root *root)
1096 struct btrfs_root_item *root_item;
1097 struct btrfs_super_block *super;
1099 super = root->fs_info->super_copy;
1101 root_item = &root->fs_info->chunk_root->root_item;
1102 super->chunk_root = root_item->bytenr;
1103 super->chunk_root_generation = root_item->generation;
1104 super->chunk_root_level = root_item->level;
1106 root_item = &root->fs_info->tree_root->root_item;
1107 super->root = root_item->bytenr;
1108 super->generation = root_item->generation;
1109 super->root_level = root_item->level;
1110 if (btrfs_test_opt(root, SPACE_CACHE))
1111 super->cache_generation = root_item->generation;
1114 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1117 spin_lock(&info->trans_lock);
1118 if (info->running_transaction)
1119 ret = info->running_transaction->in_commit;
1120 spin_unlock(&info->trans_lock);
1124 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1127 spin_lock(&info->trans_lock);
1128 if (info->running_transaction)
1129 ret = info->running_transaction->blocked;
1130 spin_unlock(&info->trans_lock);
1135 * wait for the current transaction commit to start and block subsequent
1138 static void wait_current_trans_commit_start(struct btrfs_root *root,
1139 struct btrfs_transaction *trans)
1141 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1145 * wait for the current transaction to start and then become unblocked.
1148 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1149 struct btrfs_transaction *trans)
1151 wait_event(root->fs_info->transaction_wait,
1152 trans->commit_done || (trans->in_commit && !trans->blocked));
1156 * commit transactions asynchronously. once btrfs_commit_transaction_async
1157 * returns, any subsequent transaction will not be allowed to join.
1159 struct btrfs_async_commit {
1160 struct btrfs_trans_handle *newtrans;
1161 struct btrfs_root *root;
1162 struct delayed_work work;
1165 static void do_async_commit(struct work_struct *work)
1167 struct btrfs_async_commit *ac =
1168 container_of(work, struct btrfs_async_commit, work.work);
1170 btrfs_commit_transaction(ac->newtrans, ac->root);
1174 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1175 struct btrfs_root *root,
1176 int wait_for_unblock)
1178 struct btrfs_async_commit *ac;
1179 struct btrfs_transaction *cur_trans;
1181 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1185 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1187 ac->newtrans = btrfs_join_transaction(root);
1188 if (IS_ERR(ac->newtrans)) {
1189 int err = PTR_ERR(ac->newtrans);
1194 /* take transaction reference */
1195 cur_trans = trans->transaction;
1196 atomic_inc(&cur_trans->use_count);
1198 btrfs_end_transaction(trans, root);
1199 schedule_delayed_work(&ac->work, 0);
1201 /* wait for transaction to start and unblock */
1202 if (wait_for_unblock)
1203 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1205 wait_current_trans_commit_start(root, cur_trans);
1207 if (current->journal_info == trans)
1208 current->journal_info = NULL;
1210 put_transaction(cur_trans);
1215 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1216 struct btrfs_root *root)
1218 struct btrfs_transaction *cur_trans = trans->transaction;
1220 WARN_ON(trans->use_count > 1);
1222 spin_lock(&root->fs_info->trans_lock);
1223 list_del_init(&cur_trans->list);
1224 spin_unlock(&root->fs_info->trans_lock);
1226 btrfs_cleanup_one_transaction(trans->transaction, root);
1228 put_transaction(cur_trans);
1229 put_transaction(cur_trans);
1231 trace_btrfs_transaction_commit(root);
1233 btrfs_scrub_continue(root);
1235 if (current->journal_info == trans)
1236 current->journal_info = NULL;
1238 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1242 * btrfs_transaction state sequence:
1243 * in_commit = 0, blocked = 0 (initial)
1244 * in_commit = 1, blocked = 1
1248 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1249 struct btrfs_root *root)
1251 unsigned long joined = 0;
1252 struct btrfs_transaction *cur_trans = trans->transaction;
1253 struct btrfs_transaction *prev_trans = NULL;
1256 int should_grow = 0;
1257 unsigned long now = get_seconds();
1258 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1260 btrfs_run_ordered_operations(root, 0);
1262 btrfs_trans_release_metadata(trans, root);
1263 trans->block_rsv = NULL;
1265 if (cur_trans->aborted)
1266 goto cleanup_transaction;
1268 /* make a pass through all the delayed refs we have so far
1269 * any runnings procs may add more while we are here
1271 ret = btrfs_run_delayed_refs(trans, root, 0);
1273 goto cleanup_transaction;
1275 cur_trans = trans->transaction;
1278 * set the flushing flag so procs in this transaction have to
1279 * start sending their work down.
1281 cur_trans->delayed_refs.flushing = 1;
1283 ret = btrfs_run_delayed_refs(trans, root, 0);
1285 goto cleanup_transaction;
1287 spin_lock(&cur_trans->commit_lock);
1288 if (cur_trans->in_commit) {
1289 spin_unlock(&cur_trans->commit_lock);
1290 atomic_inc(&cur_trans->use_count);
1291 ret = btrfs_end_transaction(trans, root);
1293 wait_for_commit(root, cur_trans);
1295 put_transaction(cur_trans);
1300 trans->transaction->in_commit = 1;
1301 trans->transaction->blocked = 1;
1302 spin_unlock(&cur_trans->commit_lock);
1303 wake_up(&root->fs_info->transaction_blocked_wait);
1305 spin_lock(&root->fs_info->trans_lock);
1306 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1307 prev_trans = list_entry(cur_trans->list.prev,
1308 struct btrfs_transaction, list);
1309 if (!prev_trans->commit_done) {
1310 atomic_inc(&prev_trans->use_count);
1311 spin_unlock(&root->fs_info->trans_lock);
1313 wait_for_commit(root, prev_trans);
1315 put_transaction(prev_trans);
1317 spin_unlock(&root->fs_info->trans_lock);
1320 spin_unlock(&root->fs_info->trans_lock);
1323 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1327 int snap_pending = 0;
1329 joined = cur_trans->num_joined;
1330 if (!list_empty(&trans->transaction->pending_snapshots))
1333 WARN_ON(cur_trans != trans->transaction);
1335 if (flush_on_commit || snap_pending) {
1336 btrfs_start_delalloc_inodes(root, 1);
1337 btrfs_wait_ordered_extents(root, 0, 1);
1340 ret = btrfs_run_delayed_items(trans, root);
1342 goto cleanup_transaction;
1345 * rename don't use btrfs_join_transaction, so, once we
1346 * set the transaction to blocked above, we aren't going
1347 * to get any new ordered operations. We can safely run
1348 * it here and no for sure that nothing new will be added
1351 btrfs_run_ordered_operations(root, 1);
1353 prepare_to_wait(&cur_trans->writer_wait, &wait,
1354 TASK_UNINTERRUPTIBLE);
1356 if (atomic_read(&cur_trans->num_writers) > 1)
1357 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1358 else if (should_grow)
1359 schedule_timeout(1);
1361 finish_wait(&cur_trans->writer_wait, &wait);
1362 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1363 (should_grow && cur_trans->num_joined != joined));
1366 * Ok now we need to make sure to block out any other joins while we
1367 * commit the transaction. We could have started a join before setting
1368 * no_join so make sure to wait for num_writers to == 1 again.
1370 spin_lock(&root->fs_info->trans_lock);
1371 root->fs_info->trans_no_join = 1;
1372 spin_unlock(&root->fs_info->trans_lock);
1373 wait_event(cur_trans->writer_wait,
1374 atomic_read(&cur_trans->num_writers) == 1);
1377 * the reloc mutex makes sure that we stop
1378 * the balancing code from coming in and moving
1379 * extents around in the middle of the commit
1381 mutex_lock(&root->fs_info->reloc_mutex);
1383 ret = btrfs_run_delayed_items(trans, root);
1385 mutex_unlock(&root->fs_info->reloc_mutex);
1386 goto cleanup_transaction;
1389 ret = create_pending_snapshots(trans, root->fs_info);
1391 mutex_unlock(&root->fs_info->reloc_mutex);
1392 goto cleanup_transaction;
1395 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1397 mutex_unlock(&root->fs_info->reloc_mutex);
1398 goto cleanup_transaction;
1402 * make sure none of the code above managed to slip in a
1405 btrfs_assert_delayed_root_empty(root);
1407 WARN_ON(cur_trans != trans->transaction);
1409 btrfs_scrub_pause(root);
1410 /* btrfs_commit_tree_roots is responsible for getting the
1411 * various roots consistent with each other. Every pointer
1412 * in the tree of tree roots has to point to the most up to date
1413 * root for every subvolume and other tree. So, we have to keep
1414 * the tree logging code from jumping in and changing any
1417 * At this point in the commit, there can't be any tree-log
1418 * writers, but a little lower down we drop the trans mutex
1419 * and let new people in. By holding the tree_log_mutex
1420 * from now until after the super is written, we avoid races
1421 * with the tree-log code.
1423 mutex_lock(&root->fs_info->tree_log_mutex);
1425 ret = commit_fs_roots(trans, root);
1427 mutex_unlock(&root->fs_info->tree_log_mutex);
1428 mutex_unlock(&root->fs_info->reloc_mutex);
1429 goto cleanup_transaction;
1432 /* commit_fs_roots gets rid of all the tree log roots, it is now
1433 * safe to free the root of tree log roots
1435 btrfs_free_log_root_tree(trans, root->fs_info);
1437 ret = commit_cowonly_roots(trans, root);
1439 mutex_unlock(&root->fs_info->tree_log_mutex);
1440 mutex_unlock(&root->fs_info->reloc_mutex);
1441 goto cleanup_transaction;
1444 btrfs_prepare_extent_commit(trans, root);
1446 cur_trans = root->fs_info->running_transaction;
1448 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1449 root->fs_info->tree_root->node);
1450 switch_commit_root(root->fs_info->tree_root);
1452 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1453 root->fs_info->chunk_root->node);
1454 switch_commit_root(root->fs_info->chunk_root);
1456 update_super_roots(root);
1458 if (!root->fs_info->log_root_recovering) {
1459 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1460 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1463 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1464 sizeof(*root->fs_info->super_copy));
1466 trans->transaction->blocked = 0;
1467 spin_lock(&root->fs_info->trans_lock);
1468 root->fs_info->running_transaction = NULL;
1469 root->fs_info->trans_no_join = 0;
1470 spin_unlock(&root->fs_info->trans_lock);
1471 mutex_unlock(&root->fs_info->reloc_mutex);
1473 wake_up(&root->fs_info->transaction_wait);
1475 ret = btrfs_write_and_wait_transaction(trans, root);
1477 btrfs_error(root->fs_info, ret,
1478 "Error while writing out transaction.");
1479 mutex_unlock(&root->fs_info->tree_log_mutex);
1480 goto cleanup_transaction;
1483 ret = write_ctree_super(trans, root, 0);
1485 mutex_unlock(&root->fs_info->tree_log_mutex);
1486 goto cleanup_transaction;
1490 * the super is written, we can safely allow the tree-loggers
1491 * to go about their business
1493 mutex_unlock(&root->fs_info->tree_log_mutex);
1495 btrfs_finish_extent_commit(trans, root);
1497 cur_trans->commit_done = 1;
1499 root->fs_info->last_trans_committed = cur_trans->transid;
1501 wake_up(&cur_trans->commit_wait);
1503 spin_lock(&root->fs_info->trans_lock);
1504 list_del_init(&cur_trans->list);
1505 spin_unlock(&root->fs_info->trans_lock);
1507 put_transaction(cur_trans);
1508 put_transaction(cur_trans);
1510 trace_btrfs_transaction_commit(root);
1512 btrfs_scrub_continue(root);
1514 if (current->journal_info == trans)
1515 current->journal_info = NULL;
1517 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1519 if (current != root->fs_info->transaction_kthread)
1520 btrfs_run_delayed_iputs(root);
1524 cleanup_transaction:
1525 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1527 if (current->journal_info == trans)
1528 current->journal_info = NULL;
1529 cleanup_transaction(trans, root);
1535 * interface function to delete all the snapshots we have scheduled for deletion
1537 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1540 struct btrfs_fs_info *fs_info = root->fs_info;
1542 spin_lock(&fs_info->trans_lock);
1543 list_splice_init(&fs_info->dead_roots, &list);
1544 spin_unlock(&fs_info->trans_lock);
1546 while (!list_empty(&list)) {
1549 root = list_entry(list.next, struct btrfs_root, root_list);
1550 list_del(&root->root_list);
1552 btrfs_kill_all_delayed_nodes(root);
1554 if (btrfs_header_backref_rev(root->node) <
1555 BTRFS_MIXED_BACKREF_REV)
1556 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1558 ret =btrfs_drop_snapshot(root, NULL, 1, 0);