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 if (may_wait_transaction(root, type))
333 wait_current_trans(root);
336 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
338 wait_current_trans(root);
339 } while (ret == -EBUSY);
342 kmem_cache_free(btrfs_trans_handle_cachep, h);
346 cur_trans = root->fs_info->running_transaction;
348 h->transid = cur_trans->transid;
349 h->transaction = cur_trans;
351 h->bytes_reserved = 0;
352 h->delayed_ref_updates = 0;
359 if (cur_trans->blocked && may_wait_transaction(root, type)) {
360 btrfs_commit_transaction(h, root);
365 trace_btrfs_space_reservation(root->fs_info, "transaction",
366 h->transid, num_bytes, 1);
367 h->block_rsv = &root->fs_info->trans_block_rsv;
368 h->bytes_reserved = num_bytes;
372 btrfs_record_root_in_trans(h, root);
374 if (!current->journal_info && type != TRANS_USERSPACE)
375 current->journal_info = h;
379 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
382 return start_transaction(root, num_items, TRANS_START);
384 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
386 return start_transaction(root, 0, TRANS_JOIN);
389 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
391 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
394 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
396 return start_transaction(root, 0, TRANS_USERSPACE);
399 /* wait for a transaction commit to be fully complete */
400 static noinline void wait_for_commit(struct btrfs_root *root,
401 struct btrfs_transaction *commit)
403 wait_event(commit->commit_wait, commit->commit_done);
406 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
408 struct btrfs_transaction *cur_trans = NULL, *t;
413 if (transid <= root->fs_info->last_trans_committed)
416 /* find specified transaction */
417 spin_lock(&root->fs_info->trans_lock);
418 list_for_each_entry(t, &root->fs_info->trans_list, list) {
419 if (t->transid == transid) {
421 atomic_inc(&cur_trans->use_count);
424 if (t->transid > transid)
427 spin_unlock(&root->fs_info->trans_lock);
430 goto out; /* bad transid */
432 /* find newest transaction that is committing | committed */
433 spin_lock(&root->fs_info->trans_lock);
434 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
440 atomic_inc(&cur_trans->use_count);
444 spin_unlock(&root->fs_info->trans_lock);
446 goto out; /* nothing committing|committed */
449 wait_for_commit(root, cur_trans);
451 put_transaction(cur_trans);
457 void btrfs_throttle(struct btrfs_root *root)
459 if (!atomic_read(&root->fs_info->open_ioctl_trans))
460 wait_current_trans(root);
463 static int should_end_transaction(struct btrfs_trans_handle *trans,
464 struct btrfs_root *root)
468 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
472 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
473 struct btrfs_root *root)
475 struct btrfs_transaction *cur_trans = trans->transaction;
476 struct btrfs_block_rsv *rsv = trans->block_rsv;
481 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
485 * We need to do this in case we're deleting csums so the global block
486 * rsv get's used instead of the csum block rsv.
488 trans->block_rsv = NULL;
490 updates = trans->delayed_ref_updates;
491 trans->delayed_ref_updates = 0;
493 err = btrfs_run_delayed_refs(trans, root, updates);
494 if (err) /* Error code will also eval true */
498 trans->block_rsv = rsv;
500 return should_end_transaction(trans, root);
503 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
504 struct btrfs_root *root, int throttle, int lock)
506 struct btrfs_transaction *cur_trans = trans->transaction;
507 struct btrfs_fs_info *info = root->fs_info;
511 if (--trans->use_count) {
512 trans->block_rsv = trans->orig_rsv;
516 btrfs_trans_release_metadata(trans, root);
517 trans->block_rsv = NULL;
519 unsigned long cur = trans->delayed_ref_updates;
520 trans->delayed_ref_updates = 0;
522 trans->transaction->delayed_refs.num_heads_ready > 64) {
523 trans->delayed_ref_updates = 0;
524 btrfs_run_delayed_refs(trans, root, cur);
531 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
532 should_end_transaction(trans, root)) {
533 trans->transaction->blocked = 1;
537 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
540 * We may race with somebody else here so end up having
541 * to call end_transaction on ourselves again, so inc
545 return btrfs_commit_transaction(trans, root);
547 wake_up_process(info->transaction_kthread);
551 WARN_ON(cur_trans != info->running_transaction);
552 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
553 atomic_dec(&cur_trans->num_writers);
556 if (waitqueue_active(&cur_trans->writer_wait))
557 wake_up(&cur_trans->writer_wait);
558 put_transaction(cur_trans);
560 if (current->journal_info == trans)
561 current->journal_info = NULL;
564 btrfs_run_delayed_iputs(root);
566 if (trans->aborted ||
567 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
571 memset(trans, 0, sizeof(*trans));
572 kmem_cache_free(btrfs_trans_handle_cachep, trans);
576 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
577 struct btrfs_root *root)
581 ret = __btrfs_end_transaction(trans, root, 0, 1);
587 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
588 struct btrfs_root *root)
592 ret = __btrfs_end_transaction(trans, root, 1, 1);
598 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
599 struct btrfs_root *root)
603 ret = __btrfs_end_transaction(trans, root, 0, 0);
609 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
610 struct btrfs_root *root)
612 return __btrfs_end_transaction(trans, root, 1, 1);
616 * when btree blocks are allocated, they have some corresponding bits set for
617 * them in one of two extent_io trees. This is used to make sure all of
618 * those extents are sent to disk but does not wait on them
620 int btrfs_write_marked_extents(struct btrfs_root *root,
621 struct extent_io_tree *dirty_pages, int mark)
625 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
629 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
631 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
633 err = filemap_fdatawrite_range(mapping, start, end);
645 * when btree blocks are allocated, they have some corresponding bits set for
646 * them in one of two extent_io trees. This is used to make sure all of
647 * those extents are on disk for transaction or log commit. We wait
648 * on all the pages and clear them from the dirty pages state tree
650 int btrfs_wait_marked_extents(struct btrfs_root *root,
651 struct extent_io_tree *dirty_pages, int mark)
655 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
659 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
661 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
662 err = filemap_fdatawait_range(mapping, start, end);
674 * when btree blocks are allocated, they have some corresponding bits set for
675 * them in one of two extent_io trees. This is used to make sure all of
676 * those extents are on disk for transaction or log commit
678 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
679 struct extent_io_tree *dirty_pages, int mark)
684 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
685 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
694 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
695 struct btrfs_root *root)
697 if (!trans || !trans->transaction) {
698 struct inode *btree_inode;
699 btree_inode = root->fs_info->btree_inode;
700 return filemap_write_and_wait(btree_inode->i_mapping);
702 return btrfs_write_and_wait_marked_extents(root,
703 &trans->transaction->dirty_pages,
708 * this is used to update the root pointer in the tree of tree roots.
710 * But, in the case of the extent allocation tree, updating the root
711 * pointer may allocate blocks which may change the root of the extent
714 * So, this loops and repeats and makes sure the cowonly root didn't
715 * change while the root pointer was being updated in the metadata.
717 static int update_cowonly_root(struct btrfs_trans_handle *trans,
718 struct btrfs_root *root)
723 struct btrfs_root *tree_root = root->fs_info->tree_root;
725 old_root_used = btrfs_root_used(&root->root_item);
726 btrfs_write_dirty_block_groups(trans, root);
729 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
730 if (old_root_bytenr == root->node->start &&
731 old_root_used == btrfs_root_used(&root->root_item))
734 btrfs_set_root_node(&root->root_item, root->node);
735 ret = btrfs_update_root(trans, tree_root,
741 old_root_used = btrfs_root_used(&root->root_item);
742 ret = btrfs_write_dirty_block_groups(trans, root);
747 if (root != root->fs_info->extent_root)
748 switch_commit_root(root);
754 * update all the cowonly tree roots on disk
756 * The error handling in this function may not be obvious. Any of the
757 * failures will cause the file system to go offline. We still need
758 * to clean up the delayed refs.
760 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
761 struct btrfs_root *root)
763 struct btrfs_fs_info *fs_info = root->fs_info;
764 struct list_head *next;
765 struct extent_buffer *eb;
768 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
772 eb = btrfs_lock_root_node(fs_info->tree_root);
773 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
775 btrfs_tree_unlock(eb);
776 free_extent_buffer(eb);
781 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
785 ret = btrfs_run_dev_stats(trans, root->fs_info);
788 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
789 next = fs_info->dirty_cowonly_roots.next;
791 root = list_entry(next, struct btrfs_root, dirty_list);
793 ret = update_cowonly_root(trans, root);
798 down_write(&fs_info->extent_commit_sem);
799 switch_commit_root(fs_info->extent_root);
800 up_write(&fs_info->extent_commit_sem);
806 * dead roots are old snapshots that need to be deleted. This allocates
807 * a dirty root struct and adds it into the list of dead roots that need to
810 int btrfs_add_dead_root(struct btrfs_root *root)
812 spin_lock(&root->fs_info->trans_lock);
813 list_add(&root->root_list, &root->fs_info->dead_roots);
814 spin_unlock(&root->fs_info->trans_lock);
819 * update all the cowonly tree roots on disk
821 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
822 struct btrfs_root *root)
824 struct btrfs_root *gang[8];
825 struct btrfs_fs_info *fs_info = root->fs_info;
830 spin_lock(&fs_info->fs_roots_radix_lock);
832 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
835 BTRFS_ROOT_TRANS_TAG);
838 for (i = 0; i < ret; i++) {
840 radix_tree_tag_clear(&fs_info->fs_roots_radix,
841 (unsigned long)root->root_key.objectid,
842 BTRFS_ROOT_TRANS_TAG);
843 spin_unlock(&fs_info->fs_roots_radix_lock);
845 btrfs_free_log(trans, root);
846 btrfs_update_reloc_root(trans, root);
847 btrfs_orphan_commit_root(trans, root);
849 btrfs_save_ino_cache(root, trans);
851 /* see comments in should_cow_block() */
855 if (root->commit_root != root->node) {
856 mutex_lock(&root->fs_commit_mutex);
857 switch_commit_root(root);
858 btrfs_unpin_free_ino(root);
859 mutex_unlock(&root->fs_commit_mutex);
861 btrfs_set_root_node(&root->root_item,
865 err = btrfs_update_root(trans, fs_info->tree_root,
868 spin_lock(&fs_info->fs_roots_radix_lock);
873 spin_unlock(&fs_info->fs_roots_radix_lock);
878 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
879 * otherwise every leaf in the btree is read and defragged.
881 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
883 struct btrfs_fs_info *info = root->fs_info;
884 struct btrfs_trans_handle *trans;
888 if (xchg(&root->defrag_running, 1))
892 trans = btrfs_start_transaction(root, 0);
894 return PTR_ERR(trans);
896 ret = btrfs_defrag_leaves(trans, root, cacheonly);
898 nr = trans->blocks_used;
899 btrfs_end_transaction(trans, root);
900 btrfs_btree_balance_dirty(info->tree_root, nr);
903 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
906 root->defrag_running = 0;
911 * new snapshots need to be created at a very specific time in the
912 * transaction commit. This does the actual creation
914 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
915 struct btrfs_fs_info *fs_info,
916 struct btrfs_pending_snapshot *pending)
918 struct btrfs_key key;
919 struct btrfs_root_item *new_root_item;
920 struct btrfs_root *tree_root = fs_info->tree_root;
921 struct btrfs_root *root = pending->root;
922 struct btrfs_root *parent_root;
923 struct btrfs_block_rsv *rsv;
924 struct inode *parent_inode;
925 struct dentry *parent;
926 struct dentry *dentry;
927 struct extent_buffer *tmp;
928 struct extent_buffer *old;
935 rsv = trans->block_rsv;
937 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
938 if (!new_root_item) {
939 ret = pending->error = -ENOMEM;
943 ret = btrfs_find_free_objectid(tree_root, &objectid);
945 pending->error = ret;
949 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
951 if (to_reserve > 0) {
952 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
955 pending->error = ret;
960 key.objectid = objectid;
961 key.offset = (u64)-1;
962 key.type = BTRFS_ROOT_ITEM_KEY;
964 trans->block_rsv = &pending->block_rsv;
966 dentry = pending->dentry;
967 parent = dget_parent(dentry);
968 parent_inode = parent->d_inode;
969 parent_root = BTRFS_I(parent_inode)->root;
970 record_root_in_trans(trans, parent_root);
973 * insert the directory item
975 ret = btrfs_set_inode_index(parent_inode, &index);
976 BUG_ON(ret); /* -ENOMEM */
977 ret = btrfs_insert_dir_item(trans, parent_root,
978 dentry->d_name.name, dentry->d_name.len,
980 BTRFS_FT_DIR, index);
981 if (ret == -EEXIST) {
982 pending->error = -EEXIST;
986 goto abort_trans_dput;
989 btrfs_i_size_write(parent_inode, parent_inode->i_size +
990 dentry->d_name.len * 2);
991 ret = btrfs_update_inode(trans, parent_root, parent_inode);
993 goto abort_trans_dput;
996 * pull in the delayed directory update
997 * and the delayed inode item
998 * otherwise we corrupt the FS during
1001 ret = btrfs_run_delayed_items(trans, root);
1002 if (ret) { /* Transaction aborted */
1007 record_root_in_trans(trans, root);
1008 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1009 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1010 btrfs_check_and_init_root_item(new_root_item);
1012 root_flags = btrfs_root_flags(new_root_item);
1013 if (pending->readonly)
1014 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1016 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1017 btrfs_set_root_flags(new_root_item, root_flags);
1019 old = btrfs_lock_root_node(root);
1020 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1022 btrfs_tree_unlock(old);
1023 free_extent_buffer(old);
1024 goto abort_trans_dput;
1027 btrfs_set_lock_blocking(old);
1029 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1030 /* clean up in any case */
1031 btrfs_tree_unlock(old);
1032 free_extent_buffer(old);
1034 goto abort_trans_dput;
1036 /* see comments in should_cow_block() */
1037 root->force_cow = 1;
1040 btrfs_set_root_node(new_root_item, tmp);
1041 /* record when the snapshot was created in key.offset */
1042 key.offset = trans->transid;
1043 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1044 btrfs_tree_unlock(tmp);
1045 free_extent_buffer(tmp);
1047 goto abort_trans_dput;
1050 * insert root back/forward references
1052 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1053 parent_root->root_key.objectid,
1054 btrfs_ino(parent_inode), index,
1055 dentry->d_name.name, dentry->d_name.len);
1060 key.offset = (u64)-1;
1061 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1062 if (IS_ERR(pending->snap)) {
1063 ret = PTR_ERR(pending->snap);
1067 ret = btrfs_reloc_post_snapshot(trans, pending);
1072 kfree(new_root_item);
1073 trans->block_rsv = rsv;
1074 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1080 btrfs_abort_transaction(trans, root, ret);
1085 * create all the snapshots we've scheduled for creation
1087 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1088 struct btrfs_fs_info *fs_info)
1090 struct btrfs_pending_snapshot *pending;
1091 struct list_head *head = &trans->transaction->pending_snapshots;
1093 list_for_each_entry(pending, head, list)
1094 create_pending_snapshot(trans, fs_info, pending);
1098 static void update_super_roots(struct btrfs_root *root)
1100 struct btrfs_root_item *root_item;
1101 struct btrfs_super_block *super;
1103 super = root->fs_info->super_copy;
1105 root_item = &root->fs_info->chunk_root->root_item;
1106 super->chunk_root = root_item->bytenr;
1107 super->chunk_root_generation = root_item->generation;
1108 super->chunk_root_level = root_item->level;
1110 root_item = &root->fs_info->tree_root->root_item;
1111 super->root = root_item->bytenr;
1112 super->generation = root_item->generation;
1113 super->root_level = root_item->level;
1114 if (btrfs_test_opt(root, SPACE_CACHE))
1115 super->cache_generation = root_item->generation;
1118 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1121 spin_lock(&info->trans_lock);
1122 if (info->running_transaction)
1123 ret = info->running_transaction->in_commit;
1124 spin_unlock(&info->trans_lock);
1128 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1131 spin_lock(&info->trans_lock);
1132 if (info->running_transaction)
1133 ret = info->running_transaction->blocked;
1134 spin_unlock(&info->trans_lock);
1139 * wait for the current transaction commit to start and block subsequent
1142 static void wait_current_trans_commit_start(struct btrfs_root *root,
1143 struct btrfs_transaction *trans)
1145 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1149 * wait for the current transaction to start and then become unblocked.
1152 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1153 struct btrfs_transaction *trans)
1155 wait_event(root->fs_info->transaction_wait,
1156 trans->commit_done || (trans->in_commit && !trans->blocked));
1160 * commit transactions asynchronously. once btrfs_commit_transaction_async
1161 * returns, any subsequent transaction will not be allowed to join.
1163 struct btrfs_async_commit {
1164 struct btrfs_trans_handle *newtrans;
1165 struct btrfs_root *root;
1166 struct delayed_work work;
1169 static void do_async_commit(struct work_struct *work)
1171 struct btrfs_async_commit *ac =
1172 container_of(work, struct btrfs_async_commit, work.work);
1174 btrfs_commit_transaction(ac->newtrans, ac->root);
1178 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1179 struct btrfs_root *root,
1180 int wait_for_unblock)
1182 struct btrfs_async_commit *ac;
1183 struct btrfs_transaction *cur_trans;
1185 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1189 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1191 ac->newtrans = btrfs_join_transaction(root);
1192 if (IS_ERR(ac->newtrans)) {
1193 int err = PTR_ERR(ac->newtrans);
1198 /* take transaction reference */
1199 cur_trans = trans->transaction;
1200 atomic_inc(&cur_trans->use_count);
1202 btrfs_end_transaction(trans, root);
1203 schedule_delayed_work(&ac->work, 0);
1205 /* wait for transaction to start and unblock */
1206 if (wait_for_unblock)
1207 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1209 wait_current_trans_commit_start(root, cur_trans);
1211 if (current->journal_info == trans)
1212 current->journal_info = NULL;
1214 put_transaction(cur_trans);
1219 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1220 struct btrfs_root *root, int err)
1222 struct btrfs_transaction *cur_trans = trans->transaction;
1224 WARN_ON(trans->use_count > 1);
1226 btrfs_abort_transaction(trans, root, err);
1228 spin_lock(&root->fs_info->trans_lock);
1229 list_del_init(&cur_trans->list);
1230 if (cur_trans == root->fs_info->running_transaction) {
1231 root->fs_info->running_transaction = NULL;
1232 root->fs_info->trans_no_join = 0;
1234 spin_unlock(&root->fs_info->trans_lock);
1236 btrfs_cleanup_one_transaction(trans->transaction, root);
1238 put_transaction(cur_trans);
1239 put_transaction(cur_trans);
1241 trace_btrfs_transaction_commit(root);
1243 btrfs_scrub_continue(root);
1245 if (current->journal_info == trans)
1246 current->journal_info = NULL;
1248 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1252 * btrfs_transaction state sequence:
1253 * in_commit = 0, blocked = 0 (initial)
1254 * in_commit = 1, blocked = 1
1258 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1259 struct btrfs_root *root)
1261 unsigned long joined = 0;
1262 struct btrfs_transaction *cur_trans = trans->transaction;
1263 struct btrfs_transaction *prev_trans = NULL;
1266 int should_grow = 0;
1267 unsigned long now = get_seconds();
1268 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1270 btrfs_run_ordered_operations(root, 0);
1272 btrfs_trans_release_metadata(trans, root);
1273 trans->block_rsv = NULL;
1275 if (cur_trans->aborted)
1276 goto cleanup_transaction;
1278 /* make a pass through all the delayed refs we have so far
1279 * any runnings procs may add more while we are here
1281 ret = btrfs_run_delayed_refs(trans, root, 0);
1283 goto cleanup_transaction;
1285 cur_trans = trans->transaction;
1288 * set the flushing flag so procs in this transaction have to
1289 * start sending their work down.
1291 cur_trans->delayed_refs.flushing = 1;
1293 ret = btrfs_run_delayed_refs(trans, root, 0);
1295 goto cleanup_transaction;
1297 spin_lock(&cur_trans->commit_lock);
1298 if (cur_trans->in_commit) {
1299 spin_unlock(&cur_trans->commit_lock);
1300 atomic_inc(&cur_trans->use_count);
1301 ret = btrfs_end_transaction(trans, root);
1303 wait_for_commit(root, cur_trans);
1305 put_transaction(cur_trans);
1310 trans->transaction->in_commit = 1;
1311 trans->transaction->blocked = 1;
1312 spin_unlock(&cur_trans->commit_lock);
1313 wake_up(&root->fs_info->transaction_blocked_wait);
1315 spin_lock(&root->fs_info->trans_lock);
1316 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1317 prev_trans = list_entry(cur_trans->list.prev,
1318 struct btrfs_transaction, list);
1319 if (!prev_trans->commit_done) {
1320 atomic_inc(&prev_trans->use_count);
1321 spin_unlock(&root->fs_info->trans_lock);
1323 wait_for_commit(root, prev_trans);
1325 put_transaction(prev_trans);
1327 spin_unlock(&root->fs_info->trans_lock);
1330 spin_unlock(&root->fs_info->trans_lock);
1333 if (!btrfs_test_opt(root, SSD) &&
1334 (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1338 int snap_pending = 0;
1340 joined = cur_trans->num_joined;
1341 if (!list_empty(&trans->transaction->pending_snapshots))
1344 WARN_ON(cur_trans != trans->transaction);
1346 if (flush_on_commit || snap_pending) {
1347 btrfs_start_delalloc_inodes(root, 1);
1348 btrfs_wait_ordered_extents(root, 0, 1);
1351 ret = btrfs_run_delayed_items(trans, root);
1353 goto cleanup_transaction;
1356 * rename don't use btrfs_join_transaction, so, once we
1357 * set the transaction to blocked above, we aren't going
1358 * to get any new ordered operations. We can safely run
1359 * it here and no for sure that nothing new will be added
1362 btrfs_run_ordered_operations(root, 1);
1364 prepare_to_wait(&cur_trans->writer_wait, &wait,
1365 TASK_UNINTERRUPTIBLE);
1367 if (atomic_read(&cur_trans->num_writers) > 1)
1368 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1369 else if (should_grow)
1370 schedule_timeout(1);
1372 finish_wait(&cur_trans->writer_wait, &wait);
1373 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1374 (should_grow && cur_trans->num_joined != joined));
1377 * Ok now we need to make sure to block out any other joins while we
1378 * commit the transaction. We could have started a join before setting
1379 * no_join so make sure to wait for num_writers to == 1 again.
1381 spin_lock(&root->fs_info->trans_lock);
1382 root->fs_info->trans_no_join = 1;
1383 spin_unlock(&root->fs_info->trans_lock);
1384 wait_event(cur_trans->writer_wait,
1385 atomic_read(&cur_trans->num_writers) == 1);
1388 * the reloc mutex makes sure that we stop
1389 * the balancing code from coming in and moving
1390 * extents around in the middle of the commit
1392 mutex_lock(&root->fs_info->reloc_mutex);
1394 ret = btrfs_run_delayed_items(trans, root);
1396 mutex_unlock(&root->fs_info->reloc_mutex);
1397 goto cleanup_transaction;
1400 ret = create_pending_snapshots(trans, root->fs_info);
1402 mutex_unlock(&root->fs_info->reloc_mutex);
1403 goto cleanup_transaction;
1406 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1408 mutex_unlock(&root->fs_info->reloc_mutex);
1409 goto cleanup_transaction;
1413 * make sure none of the code above managed to slip in a
1416 btrfs_assert_delayed_root_empty(root);
1418 WARN_ON(cur_trans != trans->transaction);
1420 btrfs_scrub_pause(root);
1421 /* btrfs_commit_tree_roots is responsible for getting the
1422 * various roots consistent with each other. Every pointer
1423 * in the tree of tree roots has to point to the most up to date
1424 * root for every subvolume and other tree. So, we have to keep
1425 * the tree logging code from jumping in and changing any
1428 * At this point in the commit, there can't be any tree-log
1429 * writers, but a little lower down we drop the trans mutex
1430 * and let new people in. By holding the tree_log_mutex
1431 * from now until after the super is written, we avoid races
1432 * with the tree-log code.
1434 mutex_lock(&root->fs_info->tree_log_mutex);
1436 ret = commit_fs_roots(trans, root);
1438 mutex_unlock(&root->fs_info->tree_log_mutex);
1439 mutex_unlock(&root->fs_info->reloc_mutex);
1440 goto cleanup_transaction;
1443 /* commit_fs_roots gets rid of all the tree log roots, it is now
1444 * safe to free the root of tree log roots
1446 btrfs_free_log_root_tree(trans, root->fs_info);
1448 ret = commit_cowonly_roots(trans, root);
1450 mutex_unlock(&root->fs_info->tree_log_mutex);
1451 mutex_unlock(&root->fs_info->reloc_mutex);
1452 goto cleanup_transaction;
1455 btrfs_prepare_extent_commit(trans, root);
1457 cur_trans = root->fs_info->running_transaction;
1459 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1460 root->fs_info->tree_root->node);
1461 switch_commit_root(root->fs_info->tree_root);
1463 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1464 root->fs_info->chunk_root->node);
1465 switch_commit_root(root->fs_info->chunk_root);
1467 update_super_roots(root);
1469 if (!root->fs_info->log_root_recovering) {
1470 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1471 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1474 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1475 sizeof(*root->fs_info->super_copy));
1477 trans->transaction->blocked = 0;
1478 spin_lock(&root->fs_info->trans_lock);
1479 root->fs_info->running_transaction = NULL;
1480 root->fs_info->trans_no_join = 0;
1481 spin_unlock(&root->fs_info->trans_lock);
1482 mutex_unlock(&root->fs_info->reloc_mutex);
1484 wake_up(&root->fs_info->transaction_wait);
1486 ret = btrfs_write_and_wait_transaction(trans, root);
1488 btrfs_error(root->fs_info, ret,
1489 "Error while writing out transaction.");
1490 mutex_unlock(&root->fs_info->tree_log_mutex);
1491 goto cleanup_transaction;
1494 ret = write_ctree_super(trans, root, 0);
1496 mutex_unlock(&root->fs_info->tree_log_mutex);
1497 goto cleanup_transaction;
1501 * the super is written, we can safely allow the tree-loggers
1502 * to go about their business
1504 mutex_unlock(&root->fs_info->tree_log_mutex);
1506 btrfs_finish_extent_commit(trans, root);
1508 cur_trans->commit_done = 1;
1510 root->fs_info->last_trans_committed = cur_trans->transid;
1512 wake_up(&cur_trans->commit_wait);
1514 spin_lock(&root->fs_info->trans_lock);
1515 list_del_init(&cur_trans->list);
1516 spin_unlock(&root->fs_info->trans_lock);
1518 put_transaction(cur_trans);
1519 put_transaction(cur_trans);
1521 trace_btrfs_transaction_commit(root);
1523 btrfs_scrub_continue(root);
1525 if (current->journal_info == trans)
1526 current->journal_info = NULL;
1528 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1530 if (current != root->fs_info->transaction_kthread)
1531 btrfs_run_delayed_iputs(root);
1535 cleanup_transaction:
1536 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1538 if (current->journal_info == trans)
1539 current->journal_info = NULL;
1540 cleanup_transaction(trans, root, ret);
1546 * interface function to delete all the snapshots we have scheduled for deletion
1548 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1551 struct btrfs_fs_info *fs_info = root->fs_info;
1553 spin_lock(&fs_info->trans_lock);
1554 list_splice_init(&fs_info->dead_roots, &list);
1555 spin_unlock(&fs_info->trans_lock);
1557 while (!list_empty(&list)) {
1560 root = list_entry(list.next, struct btrfs_root, root_list);
1561 list_del(&root->root_list);
1563 btrfs_kill_all_delayed_nodes(root);
1565 if (btrfs_header_backref_rev(root->node) <
1566 BTRFS_MIXED_BACKREF_REV)
1567 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1569 ret =btrfs_drop_snapshot(root, NULL, 1, 0);