2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/slab.h>
21 #include <linux/sched.h>
22 #include <linux/writeback.h>
23 #include <linux/pagemap.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
28 #include "transaction.h"
31 #include "inode-map.h"
34 #define BTRFS_ROOT_TRANS_TAG 0
36 void put_transaction(struct btrfs_transaction *transaction)
38 WARN_ON(atomic_read(&transaction->use_count) == 0);
39 if (atomic_dec_and_test(&transaction->use_count)) {
40 BUG_ON(!list_empty(&transaction->list));
41 WARN_ON(transaction->delayed_refs.root.rb_node);
42 memset(transaction, 0, sizeof(*transaction));
43 kmem_cache_free(btrfs_transaction_cachep, transaction);
47 static noinline void switch_commit_root(struct btrfs_root *root)
49 free_extent_buffer(root->commit_root);
50 root->commit_root = btrfs_root_node(root);
54 * either allocate a new transaction or hop into the existing one
56 static noinline int join_transaction(struct btrfs_root *root, int nofail)
58 struct btrfs_transaction *cur_trans;
59 struct btrfs_fs_info *fs_info = root->fs_info;
61 spin_lock(&fs_info->trans_lock);
63 /* The file system has been taken offline. No new transactions. */
64 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
65 spin_unlock(&fs_info->trans_lock);
69 if (fs_info->trans_no_join) {
71 spin_unlock(&fs_info->trans_lock);
76 cur_trans = fs_info->running_transaction;
78 if (cur_trans->aborted) {
79 spin_unlock(&fs_info->trans_lock);
80 return cur_trans->aborted;
82 atomic_inc(&cur_trans->use_count);
83 atomic_inc(&cur_trans->num_writers);
84 cur_trans->num_joined++;
85 spin_unlock(&fs_info->trans_lock);
88 spin_unlock(&fs_info->trans_lock);
90 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
94 spin_lock(&fs_info->trans_lock);
95 if (fs_info->running_transaction) {
97 * someone started a transaction after we unlocked. Make sure
98 * to redo the trans_no_join checks above
100 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
101 cur_trans = fs_info->running_transaction;
103 } else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
104 spin_unlock(&fs_info->trans_lock);
105 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
109 atomic_set(&cur_trans->num_writers, 1);
110 cur_trans->num_joined = 0;
111 init_waitqueue_head(&cur_trans->writer_wait);
112 init_waitqueue_head(&cur_trans->commit_wait);
113 cur_trans->in_commit = 0;
114 cur_trans->blocked = 0;
116 * One for this trans handle, one so it will live on until we
117 * commit the transaction.
119 atomic_set(&cur_trans->use_count, 2);
120 cur_trans->commit_done = 0;
121 cur_trans->start_time = get_seconds();
123 cur_trans->delayed_refs.root = RB_ROOT;
124 cur_trans->delayed_refs.num_entries = 0;
125 cur_trans->delayed_refs.num_heads_ready = 0;
126 cur_trans->delayed_refs.num_heads = 0;
127 cur_trans->delayed_refs.flushing = 0;
128 cur_trans->delayed_refs.run_delayed_start = 0;
131 * although the tree mod log is per file system and not per transaction,
132 * the log must never go across transaction boundaries.
135 if (!list_empty(&fs_info->tree_mod_seq_list)) {
136 printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when "
137 "creating a fresh transaction\n");
140 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) {
141 printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
142 "creating a fresh transaction\n");
145 atomic_set(&fs_info->tree_mod_seq, 0);
147 spin_lock_init(&cur_trans->commit_lock);
148 spin_lock_init(&cur_trans->delayed_refs.lock);
150 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
151 list_add_tail(&cur_trans->list, &fs_info->trans_list);
152 extent_io_tree_init(&cur_trans->dirty_pages,
153 fs_info->btree_inode->i_mapping);
154 fs_info->generation++;
155 cur_trans->transid = fs_info->generation;
156 fs_info->running_transaction = cur_trans;
157 cur_trans->aborted = 0;
158 spin_unlock(&fs_info->trans_lock);
164 * this does all the record keeping required to make sure that a reference
165 * counted root is properly recorded in a given transaction. This is required
166 * to make sure the old root from before we joined the transaction is deleted
167 * when the transaction commits
169 static int record_root_in_trans(struct btrfs_trans_handle *trans,
170 struct btrfs_root *root)
172 if (root->ref_cows && root->last_trans < trans->transid) {
173 WARN_ON(root == root->fs_info->extent_root);
174 WARN_ON(root->commit_root != root->node);
177 * see below for in_trans_setup usage rules
178 * we have the reloc mutex held now, so there
179 * is only one writer in this function
181 root->in_trans_setup = 1;
183 /* make sure readers find in_trans_setup before
184 * they find our root->last_trans update
188 spin_lock(&root->fs_info->fs_roots_radix_lock);
189 if (root->last_trans == trans->transid) {
190 spin_unlock(&root->fs_info->fs_roots_radix_lock);
193 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
194 (unsigned long)root->root_key.objectid,
195 BTRFS_ROOT_TRANS_TAG);
196 spin_unlock(&root->fs_info->fs_roots_radix_lock);
197 root->last_trans = trans->transid;
199 /* this is pretty tricky. We don't want to
200 * take the relocation lock in btrfs_record_root_in_trans
201 * unless we're really doing the first setup for this root in
204 * Normally we'd use root->last_trans as a flag to decide
205 * if we want to take the expensive mutex.
207 * But, we have to set root->last_trans before we
208 * init the relocation root, otherwise, we trip over warnings
209 * in ctree.c. The solution used here is to flag ourselves
210 * with root->in_trans_setup. When this is 1, we're still
211 * fixing up the reloc trees and everyone must wait.
213 * When this is zero, they can trust root->last_trans and fly
214 * through btrfs_record_root_in_trans without having to take the
215 * lock. smp_wmb() makes sure that all the writes above are
216 * done before we pop in the zero below
218 btrfs_init_reloc_root(trans, root);
220 root->in_trans_setup = 0;
226 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
227 struct btrfs_root *root)
233 * see record_root_in_trans for comments about in_trans_setup usage
237 if (root->last_trans == trans->transid &&
238 !root->in_trans_setup)
241 mutex_lock(&root->fs_info->reloc_mutex);
242 record_root_in_trans(trans, root);
243 mutex_unlock(&root->fs_info->reloc_mutex);
248 /* wait for commit against the current transaction to become unblocked
249 * when this is done, it is safe to start a new transaction, but the current
250 * transaction might not be fully on disk.
252 static void wait_current_trans(struct btrfs_root *root)
254 struct btrfs_transaction *cur_trans;
256 spin_lock(&root->fs_info->trans_lock);
257 cur_trans = root->fs_info->running_transaction;
258 if (cur_trans && cur_trans->blocked) {
259 atomic_inc(&cur_trans->use_count);
260 spin_unlock(&root->fs_info->trans_lock);
262 wait_event(root->fs_info->transaction_wait,
263 !cur_trans->blocked);
264 put_transaction(cur_trans);
266 spin_unlock(&root->fs_info->trans_lock);
270 enum btrfs_trans_type {
277 static int may_wait_transaction(struct btrfs_root *root, int type)
279 if (root->fs_info->log_root_recovering)
282 if (type == TRANS_USERSPACE)
285 if (type == TRANS_START &&
286 !atomic_read(&root->fs_info->open_ioctl_trans))
292 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
293 u64 num_items, int type)
295 struct btrfs_trans_handle *h;
296 struct btrfs_transaction *cur_trans;
299 u64 qgroup_reserved = 0;
301 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
302 return ERR_PTR(-EROFS);
304 if (current->journal_info) {
305 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
306 h = current->journal_info;
308 h->orig_rsv = h->block_rsv;
314 * Do the reservation before we join the transaction so we can do all
315 * the appropriate flushing if need be.
317 if (num_items > 0 && root != root->fs_info->chunk_root) {
318 if (root->fs_info->quota_enabled &&
319 is_fstree(root->root_key.objectid)) {
320 qgroup_reserved = num_items * root->leafsize;
321 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
326 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
327 ret = btrfs_block_rsv_add(root,
328 &root->fs_info->trans_block_rsv,
334 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
336 return ERR_PTR(-ENOMEM);
338 sb_start_intwrite(root->fs_info->sb);
340 if (may_wait_transaction(root, type))
341 wait_current_trans(root);
344 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
346 wait_current_trans(root);
347 } while (ret == -EBUSY);
350 sb_end_intwrite(root->fs_info->sb);
351 kmem_cache_free(btrfs_trans_handle_cachep, h);
355 cur_trans = root->fs_info->running_transaction;
357 h->transid = cur_trans->transid;
358 h->transaction = cur_trans;
360 h->bytes_reserved = 0;
362 h->delayed_ref_updates = 0;
368 h->qgroup_reserved = qgroup_reserved;
369 h->delayed_ref_elem.seq = 0;
370 INIT_LIST_HEAD(&h->qgroup_ref_list);
373 if (cur_trans->blocked && may_wait_transaction(root, type)) {
374 btrfs_commit_transaction(h, root);
379 trace_btrfs_space_reservation(root->fs_info, "transaction",
380 h->transid, num_bytes, 1);
381 h->block_rsv = &root->fs_info->trans_block_rsv;
382 h->bytes_reserved = num_bytes;
386 btrfs_record_root_in_trans(h, root);
388 if (!current->journal_info && type != TRANS_USERSPACE)
389 current->journal_info = h;
393 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
396 return start_transaction(root, num_items, TRANS_START);
398 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
400 return start_transaction(root, 0, TRANS_JOIN);
403 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
405 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
408 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
410 return start_transaction(root, 0, TRANS_USERSPACE);
413 /* wait for a transaction commit to be fully complete */
414 static noinline void wait_for_commit(struct btrfs_root *root,
415 struct btrfs_transaction *commit)
417 wait_event(commit->commit_wait, commit->commit_done);
420 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
422 struct btrfs_transaction *cur_trans = NULL, *t;
427 if (transid <= root->fs_info->last_trans_committed)
430 /* find specified transaction */
431 spin_lock(&root->fs_info->trans_lock);
432 list_for_each_entry(t, &root->fs_info->trans_list, list) {
433 if (t->transid == transid) {
435 atomic_inc(&cur_trans->use_count);
438 if (t->transid > transid)
441 spin_unlock(&root->fs_info->trans_lock);
444 goto out; /* bad transid */
446 /* find newest transaction that is committing | committed */
447 spin_lock(&root->fs_info->trans_lock);
448 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
454 atomic_inc(&cur_trans->use_count);
458 spin_unlock(&root->fs_info->trans_lock);
460 goto out; /* nothing committing|committed */
463 wait_for_commit(root, cur_trans);
465 put_transaction(cur_trans);
471 void btrfs_throttle(struct btrfs_root *root)
473 if (!atomic_read(&root->fs_info->open_ioctl_trans))
474 wait_current_trans(root);
477 static int should_end_transaction(struct btrfs_trans_handle *trans,
478 struct btrfs_root *root)
482 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
486 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
487 struct btrfs_root *root)
489 struct btrfs_transaction *cur_trans = trans->transaction;
494 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
497 updates = trans->delayed_ref_updates;
498 trans->delayed_ref_updates = 0;
500 err = btrfs_run_delayed_refs(trans, root, updates);
501 if (err) /* Error code will also eval true */
505 return should_end_transaction(trans, root);
508 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
509 struct btrfs_root *root, int throttle, int lock)
511 struct btrfs_transaction *cur_trans = trans->transaction;
512 struct btrfs_fs_info *info = root->fs_info;
516 if (--trans->use_count) {
517 trans->block_rsv = trans->orig_rsv;
522 * do the qgroup accounting as early as possible
524 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
526 btrfs_trans_release_metadata(trans, root);
527 trans->block_rsv = NULL;
529 * the same root has to be passed to start_transaction and
530 * end_transaction. Subvolume quota depends on this.
532 WARN_ON(trans->root != root);
534 if (trans->qgroup_reserved) {
535 btrfs_qgroup_free(root, trans->qgroup_reserved);
536 trans->qgroup_reserved = 0;
540 unsigned long cur = trans->delayed_ref_updates;
541 trans->delayed_ref_updates = 0;
543 trans->transaction->delayed_refs.num_heads_ready > 64) {
544 trans->delayed_ref_updates = 0;
545 btrfs_run_delayed_refs(trans, root, cur);
551 btrfs_trans_release_metadata(trans, root);
552 trans->block_rsv = NULL;
554 sb_end_intwrite(root->fs_info->sb);
556 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
557 should_end_transaction(trans, root)) {
558 trans->transaction->blocked = 1;
562 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
565 * We may race with somebody else here so end up having
566 * to call end_transaction on ourselves again, so inc
570 return btrfs_commit_transaction(trans, root);
572 wake_up_process(info->transaction_kthread);
576 WARN_ON(cur_trans != info->running_transaction);
577 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
578 atomic_dec(&cur_trans->num_writers);
581 if (waitqueue_active(&cur_trans->writer_wait))
582 wake_up(&cur_trans->writer_wait);
583 put_transaction(cur_trans);
585 if (current->journal_info == trans)
586 current->journal_info = NULL;
589 btrfs_run_delayed_iputs(root);
591 if (trans->aborted ||
592 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
595 assert_qgroups_uptodate(trans);
597 memset(trans, 0, sizeof(*trans));
598 kmem_cache_free(btrfs_trans_handle_cachep, trans);
602 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
603 struct btrfs_root *root)
607 ret = __btrfs_end_transaction(trans, root, 0, 1);
613 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
614 struct btrfs_root *root)
618 ret = __btrfs_end_transaction(trans, root, 1, 1);
624 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
625 struct btrfs_root *root)
629 ret = __btrfs_end_transaction(trans, root, 0, 0);
635 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
636 struct btrfs_root *root)
638 return __btrfs_end_transaction(trans, root, 1, 1);
642 * when btree blocks are allocated, they have some corresponding bits set for
643 * them in one of two extent_io trees. This is used to make sure all of
644 * those extents are sent to disk but does not wait on them
646 int btrfs_write_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 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
659 err = filemap_fdatawrite_range(mapping, start, end);
671 * when btree blocks are allocated, they have some corresponding bits set for
672 * them in one of two extent_io trees. This is used to make sure all of
673 * those extents are on disk for transaction or log commit. We wait
674 * on all the pages and clear them from the dirty pages state tree
676 int btrfs_wait_marked_extents(struct btrfs_root *root,
677 struct extent_io_tree *dirty_pages, int mark)
681 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
685 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
687 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
688 err = filemap_fdatawait_range(mapping, start, end);
700 * when btree blocks are allocated, they have some corresponding bits set for
701 * them in one of two extent_io trees. This is used to make sure all of
702 * those extents are on disk for transaction or log commit
704 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
705 struct extent_io_tree *dirty_pages, int mark)
710 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
711 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
720 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
721 struct btrfs_root *root)
723 if (!trans || !trans->transaction) {
724 struct inode *btree_inode;
725 btree_inode = root->fs_info->btree_inode;
726 return filemap_write_and_wait(btree_inode->i_mapping);
728 return btrfs_write_and_wait_marked_extents(root,
729 &trans->transaction->dirty_pages,
734 * this is used to update the root pointer in the tree of tree roots.
736 * But, in the case of the extent allocation tree, updating the root
737 * pointer may allocate blocks which may change the root of the extent
740 * So, this loops and repeats and makes sure the cowonly root didn't
741 * change while the root pointer was being updated in the metadata.
743 static int update_cowonly_root(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root)
749 struct btrfs_root *tree_root = root->fs_info->tree_root;
751 old_root_used = btrfs_root_used(&root->root_item);
752 btrfs_write_dirty_block_groups(trans, root);
755 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
756 if (old_root_bytenr == root->node->start &&
757 old_root_used == btrfs_root_used(&root->root_item))
760 btrfs_set_root_node(&root->root_item, root->node);
761 ret = btrfs_update_root(trans, tree_root,
767 old_root_used = btrfs_root_used(&root->root_item);
768 ret = btrfs_write_dirty_block_groups(trans, root);
773 if (root != root->fs_info->extent_root)
774 switch_commit_root(root);
780 * update all the cowonly tree roots on disk
782 * The error handling in this function may not be obvious. Any of the
783 * failures will cause the file system to go offline. We still need
784 * to clean up the delayed refs.
786 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
787 struct btrfs_root *root)
789 struct btrfs_fs_info *fs_info = root->fs_info;
790 struct list_head *next;
791 struct extent_buffer *eb;
794 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
798 eb = btrfs_lock_root_node(fs_info->tree_root);
799 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
801 btrfs_tree_unlock(eb);
802 free_extent_buffer(eb);
807 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
811 ret = btrfs_run_dev_stats(trans, root->fs_info);
814 ret = btrfs_run_qgroups(trans, root->fs_info);
817 /* run_qgroups might have added some more refs */
818 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
821 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
822 next = fs_info->dirty_cowonly_roots.next;
824 root = list_entry(next, struct btrfs_root, dirty_list);
826 ret = update_cowonly_root(trans, root);
831 down_write(&fs_info->extent_commit_sem);
832 switch_commit_root(fs_info->extent_root);
833 up_write(&fs_info->extent_commit_sem);
839 * dead roots are old snapshots that need to be deleted. This allocates
840 * a dirty root struct and adds it into the list of dead roots that need to
843 int btrfs_add_dead_root(struct btrfs_root *root)
845 spin_lock(&root->fs_info->trans_lock);
846 list_add(&root->root_list, &root->fs_info->dead_roots);
847 spin_unlock(&root->fs_info->trans_lock);
852 * update all the cowonly tree roots on disk
854 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
855 struct btrfs_root *root)
857 struct btrfs_root *gang[8];
858 struct btrfs_fs_info *fs_info = root->fs_info;
863 spin_lock(&fs_info->fs_roots_radix_lock);
865 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
868 BTRFS_ROOT_TRANS_TAG);
871 for (i = 0; i < ret; i++) {
873 radix_tree_tag_clear(&fs_info->fs_roots_radix,
874 (unsigned long)root->root_key.objectid,
875 BTRFS_ROOT_TRANS_TAG);
876 spin_unlock(&fs_info->fs_roots_radix_lock);
878 btrfs_free_log(trans, root);
879 btrfs_update_reloc_root(trans, root);
880 btrfs_orphan_commit_root(trans, root);
882 btrfs_save_ino_cache(root, trans);
884 /* see comments in should_cow_block() */
888 if (root->commit_root != root->node) {
889 mutex_lock(&root->fs_commit_mutex);
890 switch_commit_root(root);
891 btrfs_unpin_free_ino(root);
892 mutex_unlock(&root->fs_commit_mutex);
894 btrfs_set_root_node(&root->root_item,
898 err = btrfs_update_root(trans, fs_info->tree_root,
901 spin_lock(&fs_info->fs_roots_radix_lock);
906 spin_unlock(&fs_info->fs_roots_radix_lock);
911 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
912 * otherwise every leaf in the btree is read and defragged.
914 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
916 struct btrfs_fs_info *info = root->fs_info;
917 struct btrfs_trans_handle *trans;
921 if (xchg(&root->defrag_running, 1))
925 trans = btrfs_start_transaction(root, 0);
927 return PTR_ERR(trans);
929 ret = btrfs_defrag_leaves(trans, root, cacheonly);
931 nr = trans->blocks_used;
932 btrfs_end_transaction(trans, root);
933 btrfs_btree_balance_dirty(info->tree_root, nr);
936 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
939 root->defrag_running = 0;
944 * new snapshots need to be created at a very specific time in the
945 * transaction commit. This does the actual creation
947 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
948 struct btrfs_fs_info *fs_info,
949 struct btrfs_pending_snapshot *pending)
951 struct btrfs_key key;
952 struct btrfs_root_item *new_root_item;
953 struct btrfs_root *tree_root = fs_info->tree_root;
954 struct btrfs_root *root = pending->root;
955 struct btrfs_root *parent_root;
956 struct btrfs_block_rsv *rsv;
957 struct inode *parent_inode;
958 struct dentry *parent;
959 struct dentry *dentry;
960 struct extent_buffer *tmp;
961 struct extent_buffer *old;
962 struct timespec cur_time = CURRENT_TIME;
970 rsv = trans->block_rsv;
972 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
973 if (!new_root_item) {
974 ret = pending->error = -ENOMEM;
978 ret = btrfs_find_free_objectid(tree_root, &objectid);
980 pending->error = ret;
984 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
986 if (to_reserve > 0) {
987 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
990 pending->error = ret;
995 ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
996 objectid, pending->inherit);
997 kfree(pending->inherit);
999 pending->error = ret;
1003 key.objectid = objectid;
1004 key.offset = (u64)-1;
1005 key.type = BTRFS_ROOT_ITEM_KEY;
1007 trans->block_rsv = &pending->block_rsv;
1009 dentry = pending->dentry;
1010 parent = dget_parent(dentry);
1011 parent_inode = parent->d_inode;
1012 parent_root = BTRFS_I(parent_inode)->root;
1013 record_root_in_trans(trans, parent_root);
1016 * insert the directory item
1018 ret = btrfs_set_inode_index(parent_inode, &index);
1019 BUG_ON(ret); /* -ENOMEM */
1020 ret = btrfs_insert_dir_item(trans, parent_root,
1021 dentry->d_name.name, dentry->d_name.len,
1023 BTRFS_FT_DIR, index);
1024 if (ret == -EEXIST) {
1025 pending->error = -EEXIST;
1029 goto abort_trans_dput;
1032 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1033 dentry->d_name.len * 2);
1034 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1035 ret = btrfs_update_inode(trans, parent_root, parent_inode);
1037 goto abort_trans_dput;
1040 * pull in the delayed directory update
1041 * and the delayed inode item
1042 * otherwise we corrupt the FS during
1045 ret = btrfs_run_delayed_items(trans, root);
1046 if (ret) { /* Transaction aborted */
1051 record_root_in_trans(trans, root);
1052 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1053 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1054 btrfs_check_and_init_root_item(new_root_item);
1056 root_flags = btrfs_root_flags(new_root_item);
1057 if (pending->readonly)
1058 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1060 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1061 btrfs_set_root_flags(new_root_item, root_flags);
1063 btrfs_set_root_generation_v2(new_root_item,
1065 uuid_le_gen(&new_uuid);
1066 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1067 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1069 new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1070 new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
1071 btrfs_set_root_otransid(new_root_item, trans->transid);
1072 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1073 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1074 btrfs_set_root_stransid(new_root_item, 0);
1075 btrfs_set_root_rtransid(new_root_item, 0);
1077 old = btrfs_lock_root_node(root);
1078 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1080 btrfs_tree_unlock(old);
1081 free_extent_buffer(old);
1082 goto abort_trans_dput;
1085 btrfs_set_lock_blocking(old);
1087 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1088 /* clean up in any case */
1089 btrfs_tree_unlock(old);
1090 free_extent_buffer(old);
1092 goto abort_trans_dput;
1094 /* see comments in should_cow_block() */
1095 root->force_cow = 1;
1098 btrfs_set_root_node(new_root_item, tmp);
1099 /* record when the snapshot was created in key.offset */
1100 key.offset = trans->transid;
1101 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1102 btrfs_tree_unlock(tmp);
1103 free_extent_buffer(tmp);
1105 goto abort_trans_dput;
1108 * insert root back/forward references
1110 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1111 parent_root->root_key.objectid,
1112 btrfs_ino(parent_inode), index,
1113 dentry->d_name.name, dentry->d_name.len);
1118 key.offset = (u64)-1;
1119 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1120 if (IS_ERR(pending->snap)) {
1121 ret = PTR_ERR(pending->snap);
1125 ret = btrfs_reloc_post_snapshot(trans, pending);
1130 kfree(new_root_item);
1131 trans->block_rsv = rsv;
1132 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1138 btrfs_abort_transaction(trans, root, ret);
1143 * create all the snapshots we've scheduled for creation
1145 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1146 struct btrfs_fs_info *fs_info)
1148 struct btrfs_pending_snapshot *pending;
1149 struct list_head *head = &trans->transaction->pending_snapshots;
1151 list_for_each_entry(pending, head, list)
1152 create_pending_snapshot(trans, fs_info, pending);
1156 static void update_super_roots(struct btrfs_root *root)
1158 struct btrfs_root_item *root_item;
1159 struct btrfs_super_block *super;
1161 super = root->fs_info->super_copy;
1163 root_item = &root->fs_info->chunk_root->root_item;
1164 super->chunk_root = root_item->bytenr;
1165 super->chunk_root_generation = root_item->generation;
1166 super->chunk_root_level = root_item->level;
1168 root_item = &root->fs_info->tree_root->root_item;
1169 super->root = root_item->bytenr;
1170 super->generation = root_item->generation;
1171 super->root_level = root_item->level;
1172 if (btrfs_test_opt(root, SPACE_CACHE))
1173 super->cache_generation = root_item->generation;
1176 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1179 spin_lock(&info->trans_lock);
1180 if (info->running_transaction)
1181 ret = info->running_transaction->in_commit;
1182 spin_unlock(&info->trans_lock);
1186 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1189 spin_lock(&info->trans_lock);
1190 if (info->running_transaction)
1191 ret = info->running_transaction->blocked;
1192 spin_unlock(&info->trans_lock);
1197 * wait for the current transaction commit to start and block subsequent
1200 static void wait_current_trans_commit_start(struct btrfs_root *root,
1201 struct btrfs_transaction *trans)
1203 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1207 * wait for the current transaction to start and then become unblocked.
1210 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1211 struct btrfs_transaction *trans)
1213 wait_event(root->fs_info->transaction_wait,
1214 trans->commit_done || (trans->in_commit && !trans->blocked));
1218 * commit transactions asynchronously. once btrfs_commit_transaction_async
1219 * returns, any subsequent transaction will not be allowed to join.
1221 struct btrfs_async_commit {
1222 struct btrfs_trans_handle *newtrans;
1223 struct btrfs_root *root;
1224 struct delayed_work work;
1227 static void do_async_commit(struct work_struct *work)
1229 struct btrfs_async_commit *ac =
1230 container_of(work, struct btrfs_async_commit, work.work);
1232 btrfs_commit_transaction(ac->newtrans, ac->root);
1236 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1237 struct btrfs_root *root,
1238 int wait_for_unblock)
1240 struct btrfs_async_commit *ac;
1241 struct btrfs_transaction *cur_trans;
1243 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1247 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1249 ac->newtrans = btrfs_join_transaction(root);
1250 if (IS_ERR(ac->newtrans)) {
1251 int err = PTR_ERR(ac->newtrans);
1256 /* take transaction reference */
1257 cur_trans = trans->transaction;
1258 atomic_inc(&cur_trans->use_count);
1260 btrfs_end_transaction(trans, root);
1261 schedule_delayed_work(&ac->work, 0);
1263 /* wait for transaction to start and unblock */
1264 if (wait_for_unblock)
1265 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1267 wait_current_trans_commit_start(root, cur_trans);
1269 if (current->journal_info == trans)
1270 current->journal_info = NULL;
1272 put_transaction(cur_trans);
1277 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1278 struct btrfs_root *root, int err)
1280 struct btrfs_transaction *cur_trans = trans->transaction;
1282 WARN_ON(trans->use_count > 1);
1284 btrfs_abort_transaction(trans, root, err);
1286 spin_lock(&root->fs_info->trans_lock);
1287 list_del_init(&cur_trans->list);
1288 if (cur_trans == root->fs_info->running_transaction) {
1289 root->fs_info->running_transaction = NULL;
1290 root->fs_info->trans_no_join = 0;
1292 spin_unlock(&root->fs_info->trans_lock);
1294 btrfs_cleanup_one_transaction(trans->transaction, root);
1296 put_transaction(cur_trans);
1297 put_transaction(cur_trans);
1299 trace_btrfs_transaction_commit(root);
1301 btrfs_scrub_continue(root);
1303 if (current->journal_info == trans)
1304 current->journal_info = NULL;
1306 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1310 * btrfs_transaction state sequence:
1311 * in_commit = 0, blocked = 0 (initial)
1312 * in_commit = 1, blocked = 1
1316 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1317 struct btrfs_root *root)
1319 unsigned long joined = 0;
1320 struct btrfs_transaction *cur_trans = trans->transaction;
1321 struct btrfs_transaction *prev_trans = NULL;
1324 int should_grow = 0;
1325 unsigned long now = get_seconds();
1326 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1328 btrfs_run_ordered_operations(root, 0);
1330 if (cur_trans->aborted)
1331 goto cleanup_transaction;
1333 /* make a pass through all the delayed refs we have so far
1334 * any runnings procs may add more while we are here
1336 ret = btrfs_run_delayed_refs(trans, root, 0);
1338 goto cleanup_transaction;
1340 btrfs_trans_release_metadata(trans, root);
1341 trans->block_rsv = NULL;
1343 cur_trans = trans->transaction;
1346 * set the flushing flag so procs in this transaction have to
1347 * start sending their work down.
1349 cur_trans->delayed_refs.flushing = 1;
1351 ret = btrfs_run_delayed_refs(trans, root, 0);
1353 goto cleanup_transaction;
1355 spin_lock(&cur_trans->commit_lock);
1356 if (cur_trans->in_commit) {
1357 spin_unlock(&cur_trans->commit_lock);
1358 atomic_inc(&cur_trans->use_count);
1359 ret = btrfs_end_transaction(trans, root);
1361 wait_for_commit(root, cur_trans);
1363 put_transaction(cur_trans);
1368 trans->transaction->in_commit = 1;
1369 trans->transaction->blocked = 1;
1370 spin_unlock(&cur_trans->commit_lock);
1371 wake_up(&root->fs_info->transaction_blocked_wait);
1373 spin_lock(&root->fs_info->trans_lock);
1374 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1375 prev_trans = list_entry(cur_trans->list.prev,
1376 struct btrfs_transaction, list);
1377 if (!prev_trans->commit_done) {
1378 atomic_inc(&prev_trans->use_count);
1379 spin_unlock(&root->fs_info->trans_lock);
1381 wait_for_commit(root, prev_trans);
1383 put_transaction(prev_trans);
1385 spin_unlock(&root->fs_info->trans_lock);
1388 spin_unlock(&root->fs_info->trans_lock);
1391 if (!btrfs_test_opt(root, SSD) &&
1392 (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1396 int snap_pending = 0;
1398 joined = cur_trans->num_joined;
1399 if (!list_empty(&trans->transaction->pending_snapshots))
1402 WARN_ON(cur_trans != trans->transaction);
1404 if (flush_on_commit || snap_pending) {
1405 btrfs_start_delalloc_inodes(root, 1);
1406 btrfs_wait_ordered_extents(root, 0, 1);
1409 ret = btrfs_run_delayed_items(trans, root);
1411 goto cleanup_transaction;
1414 * running the delayed items may have added new refs. account
1415 * them now so that they hinder processing of more delayed refs
1416 * as little as possible.
1418 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1421 * rename don't use btrfs_join_transaction, so, once we
1422 * set the transaction to blocked above, we aren't going
1423 * to get any new ordered operations. We can safely run
1424 * it here and no for sure that nothing new will be added
1427 btrfs_run_ordered_operations(root, 1);
1429 prepare_to_wait(&cur_trans->writer_wait, &wait,
1430 TASK_UNINTERRUPTIBLE);
1432 if (atomic_read(&cur_trans->num_writers) > 1)
1433 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1434 else if (should_grow)
1435 schedule_timeout(1);
1437 finish_wait(&cur_trans->writer_wait, &wait);
1438 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1439 (should_grow && cur_trans->num_joined != joined));
1442 * Ok now we need to make sure to block out any other joins while we
1443 * commit the transaction. We could have started a join before setting
1444 * no_join so make sure to wait for num_writers to == 1 again.
1446 spin_lock(&root->fs_info->trans_lock);
1447 root->fs_info->trans_no_join = 1;
1448 spin_unlock(&root->fs_info->trans_lock);
1449 wait_event(cur_trans->writer_wait,
1450 atomic_read(&cur_trans->num_writers) == 1);
1453 * the reloc mutex makes sure that we stop
1454 * the balancing code from coming in and moving
1455 * extents around in the middle of the commit
1457 mutex_lock(&root->fs_info->reloc_mutex);
1459 ret = btrfs_run_delayed_items(trans, root);
1461 mutex_unlock(&root->fs_info->reloc_mutex);
1462 goto cleanup_transaction;
1465 ret = create_pending_snapshots(trans, root->fs_info);
1467 mutex_unlock(&root->fs_info->reloc_mutex);
1468 goto cleanup_transaction;
1471 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1473 mutex_unlock(&root->fs_info->reloc_mutex);
1474 goto cleanup_transaction;
1478 * make sure none of the code above managed to slip in a
1481 btrfs_assert_delayed_root_empty(root);
1483 WARN_ON(cur_trans != trans->transaction);
1485 btrfs_scrub_pause(root);
1486 /* btrfs_commit_tree_roots is responsible for getting the
1487 * various roots consistent with each other. Every pointer
1488 * in the tree of tree roots has to point to the most up to date
1489 * root for every subvolume and other tree. So, we have to keep
1490 * the tree logging code from jumping in and changing any
1493 * At this point in the commit, there can't be any tree-log
1494 * writers, but a little lower down we drop the trans mutex
1495 * and let new people in. By holding the tree_log_mutex
1496 * from now until after the super is written, we avoid races
1497 * with the tree-log code.
1499 mutex_lock(&root->fs_info->tree_log_mutex);
1501 ret = commit_fs_roots(trans, root);
1503 mutex_unlock(&root->fs_info->tree_log_mutex);
1504 mutex_unlock(&root->fs_info->reloc_mutex);
1505 goto cleanup_transaction;
1508 /* commit_fs_roots gets rid of all the tree log roots, it is now
1509 * safe to free the root of tree log roots
1511 btrfs_free_log_root_tree(trans, root->fs_info);
1513 ret = commit_cowonly_roots(trans, root);
1515 mutex_unlock(&root->fs_info->tree_log_mutex);
1516 mutex_unlock(&root->fs_info->reloc_mutex);
1517 goto cleanup_transaction;
1520 btrfs_prepare_extent_commit(trans, root);
1522 cur_trans = root->fs_info->running_transaction;
1524 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1525 root->fs_info->tree_root->node);
1526 switch_commit_root(root->fs_info->tree_root);
1528 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1529 root->fs_info->chunk_root->node);
1530 switch_commit_root(root->fs_info->chunk_root);
1532 assert_qgroups_uptodate(trans);
1533 update_super_roots(root);
1535 if (!root->fs_info->log_root_recovering) {
1536 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1537 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1540 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1541 sizeof(*root->fs_info->super_copy));
1543 trans->transaction->blocked = 0;
1544 spin_lock(&root->fs_info->trans_lock);
1545 root->fs_info->running_transaction = NULL;
1546 root->fs_info->trans_no_join = 0;
1547 spin_unlock(&root->fs_info->trans_lock);
1548 mutex_unlock(&root->fs_info->reloc_mutex);
1550 wake_up(&root->fs_info->transaction_wait);
1552 ret = btrfs_write_and_wait_transaction(trans, root);
1554 btrfs_error(root->fs_info, ret,
1555 "Error while writing out transaction.");
1556 mutex_unlock(&root->fs_info->tree_log_mutex);
1557 goto cleanup_transaction;
1560 ret = write_ctree_super(trans, root, 0);
1562 mutex_unlock(&root->fs_info->tree_log_mutex);
1563 goto cleanup_transaction;
1567 * the super is written, we can safely allow the tree-loggers
1568 * to go about their business
1570 mutex_unlock(&root->fs_info->tree_log_mutex);
1572 btrfs_finish_extent_commit(trans, root);
1574 cur_trans->commit_done = 1;
1576 root->fs_info->last_trans_committed = cur_trans->transid;
1578 wake_up(&cur_trans->commit_wait);
1580 spin_lock(&root->fs_info->trans_lock);
1581 list_del_init(&cur_trans->list);
1582 spin_unlock(&root->fs_info->trans_lock);
1584 put_transaction(cur_trans);
1585 put_transaction(cur_trans);
1587 sb_end_intwrite(root->fs_info->sb);
1589 trace_btrfs_transaction_commit(root);
1591 btrfs_scrub_continue(root);
1593 if (current->journal_info == trans)
1594 current->journal_info = NULL;
1596 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1598 if (current != root->fs_info->transaction_kthread)
1599 btrfs_run_delayed_iputs(root);
1603 cleanup_transaction:
1604 btrfs_trans_release_metadata(trans, root);
1605 trans->block_rsv = NULL;
1606 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1608 if (current->journal_info == trans)
1609 current->journal_info = NULL;
1610 cleanup_transaction(trans, root, ret);
1616 * interface function to delete all the snapshots we have scheduled for deletion
1618 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1621 struct btrfs_fs_info *fs_info = root->fs_info;
1623 spin_lock(&fs_info->trans_lock);
1624 list_splice_init(&fs_info->dead_roots, &list);
1625 spin_unlock(&fs_info->trans_lock);
1627 while (!list_empty(&list)) {
1630 root = list_entry(list.next, struct btrfs_root, root_list);
1631 list_del(&root->root_list);
1633 btrfs_kill_all_delayed_nodes(root);
1635 if (btrfs_header_backref_rev(root->node) <
1636 BTRFS_MIXED_BACKREF_REV)
1637 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1639 ret =btrfs_drop_snapshot(root, NULL, 1, 0);