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 memset(transaction, 0, sizeof(*transaction));
42 kmem_cache_free(btrfs_transaction_cachep, transaction);
46 static noinline void switch_commit_root(struct btrfs_root *root)
48 free_extent_buffer(root->commit_root);
49 root->commit_root = btrfs_root_node(root);
53 * either allocate a new transaction or hop into the existing one
55 static noinline int join_transaction(struct btrfs_root *root, int nofail)
57 struct btrfs_transaction *cur_trans;
58 struct btrfs_fs_info *fs_info = root->fs_info;
60 spin_lock(&fs_info->trans_lock);
62 /* The file system has been taken offline. No new transactions. */
63 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
64 spin_unlock(&fs_info->trans_lock);
68 if (fs_info->trans_no_join) {
70 spin_unlock(&fs_info->trans_lock);
75 cur_trans = fs_info->running_transaction;
77 if (cur_trans->aborted) {
78 spin_unlock(&fs_info->trans_lock);
79 return cur_trans->aborted;
81 atomic_inc(&cur_trans->use_count);
82 atomic_inc(&cur_trans->num_writers);
83 cur_trans->num_joined++;
84 spin_unlock(&fs_info->trans_lock);
87 spin_unlock(&fs_info->trans_lock);
89 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
93 spin_lock(&fs_info->trans_lock);
94 if (fs_info->running_transaction) {
96 * someone started a transaction after we unlocked. Make sure
97 * to redo the trans_no_join checks above
99 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
100 cur_trans = fs_info->running_transaction;
102 } else if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
103 spin_unlock(&root->fs_info->trans_lock);
104 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
108 atomic_set(&cur_trans->num_writers, 1);
109 cur_trans->num_joined = 0;
110 init_waitqueue_head(&cur_trans->writer_wait);
111 init_waitqueue_head(&cur_trans->commit_wait);
112 cur_trans->in_commit = 0;
113 cur_trans->blocked = 0;
115 * One for this trans handle, one so it will live on until we
116 * commit the transaction.
118 atomic_set(&cur_trans->use_count, 2);
119 cur_trans->commit_done = 0;
120 cur_trans->start_time = get_seconds();
122 cur_trans->delayed_refs.root = RB_ROOT;
123 cur_trans->delayed_refs.num_entries = 0;
124 cur_trans->delayed_refs.num_heads_ready = 0;
125 cur_trans->delayed_refs.num_heads = 0;
126 cur_trans->delayed_refs.flushing = 0;
127 cur_trans->delayed_refs.run_delayed_start = 0;
130 * although the tree mod log is per file system and not per transaction,
131 * the log must never go across transaction boundaries.
134 if (!list_empty(&fs_info->tree_mod_seq_list)) {
135 printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when "
136 "creating a fresh transaction\n");
139 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) {
140 printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
141 "creating a fresh transaction\n");
144 atomic_set(&fs_info->tree_mod_seq, 0);
146 spin_lock_init(&cur_trans->commit_lock);
147 spin_lock_init(&cur_trans->delayed_refs.lock);
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;
298 u64 qgroup_reserved = 0;
300 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
301 return ERR_PTR(-EROFS);
303 if (current->journal_info) {
304 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
305 h = current->journal_info;
307 h->orig_rsv = h->block_rsv;
313 * Do the reservation before we join the transaction so we can do all
314 * the appropriate flushing if need be.
316 if (num_items > 0 && root != root->fs_info->chunk_root) {
317 if (root->fs_info->quota_enabled &&
318 is_fstree(root->root_key.objectid)) {
319 qgroup_reserved = num_items * root->leafsize;
320 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
325 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
326 ret = btrfs_block_rsv_add(root,
327 &root->fs_info->trans_block_rsv,
333 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
335 return ERR_PTR(-ENOMEM);
337 if (may_wait_transaction(root, type))
338 wait_current_trans(root);
341 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
343 wait_current_trans(root);
344 } while (ret == -EBUSY);
347 kmem_cache_free(btrfs_trans_handle_cachep, h);
351 cur_trans = root->fs_info->running_transaction;
353 h->transid = cur_trans->transid;
354 h->transaction = cur_trans;
356 h->bytes_reserved = 0;
358 h->delayed_ref_updates = 0;
363 h->qgroup_reserved = qgroup_reserved;
364 h->delayed_ref_elem.seq = 0;
365 INIT_LIST_HEAD(&h->qgroup_ref_list);
368 if (cur_trans->blocked && may_wait_transaction(root, type)) {
369 btrfs_commit_transaction(h, root);
374 trace_btrfs_space_reservation(root->fs_info, "transaction",
375 h->transid, num_bytes, 1);
376 h->block_rsv = &root->fs_info->trans_block_rsv;
377 h->bytes_reserved = num_bytes;
381 btrfs_record_root_in_trans(h, root);
383 if (!current->journal_info && type != TRANS_USERSPACE)
384 current->journal_info = h;
388 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
391 return start_transaction(root, num_items, TRANS_START);
393 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
395 return start_transaction(root, 0, TRANS_JOIN);
398 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
400 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
403 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
405 return start_transaction(root, 0, TRANS_USERSPACE);
408 /* wait for a transaction commit to be fully complete */
409 static noinline void wait_for_commit(struct btrfs_root *root,
410 struct btrfs_transaction *commit)
412 wait_event(commit->commit_wait, commit->commit_done);
415 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
417 struct btrfs_transaction *cur_trans = NULL, *t;
422 if (transid <= root->fs_info->last_trans_committed)
425 /* find specified transaction */
426 spin_lock(&root->fs_info->trans_lock);
427 list_for_each_entry(t, &root->fs_info->trans_list, list) {
428 if (t->transid == transid) {
430 atomic_inc(&cur_trans->use_count);
433 if (t->transid > transid)
436 spin_unlock(&root->fs_info->trans_lock);
439 goto out; /* bad transid */
441 /* find newest transaction that is committing | committed */
442 spin_lock(&root->fs_info->trans_lock);
443 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
449 atomic_inc(&cur_trans->use_count);
453 spin_unlock(&root->fs_info->trans_lock);
455 goto out; /* nothing committing|committed */
458 wait_for_commit(root, cur_trans);
460 put_transaction(cur_trans);
466 void btrfs_throttle(struct btrfs_root *root)
468 if (!atomic_read(&root->fs_info->open_ioctl_trans))
469 wait_current_trans(root);
472 static int should_end_transaction(struct btrfs_trans_handle *trans,
473 struct btrfs_root *root)
477 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
481 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
482 struct btrfs_root *root)
484 struct btrfs_transaction *cur_trans = trans->transaction;
485 struct btrfs_block_rsv *rsv = trans->block_rsv;
490 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
494 * We need to do this in case we're deleting csums so the global block
495 * rsv get's used instead of the csum block rsv.
497 trans->block_rsv = NULL;
499 updates = trans->delayed_ref_updates;
500 trans->delayed_ref_updates = 0;
502 err = btrfs_run_delayed_refs(trans, root, updates);
503 if (err) /* Error code will also eval true */
507 trans->block_rsv = rsv;
509 return should_end_transaction(trans, root);
512 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
513 struct btrfs_root *root, int throttle, int lock)
515 struct btrfs_transaction *cur_trans = trans->transaction;
516 struct btrfs_fs_info *info = root->fs_info;
520 if (--trans->use_count) {
521 trans->block_rsv = trans->orig_rsv;
526 * do the qgroup accounting as early as possible
528 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
530 btrfs_trans_release_metadata(trans, root);
531 trans->block_rsv = NULL;
533 * the same root has to be passed to start_transaction and
534 * end_transaction. Subvolume quota depends on this.
536 WARN_ON(trans->root != root);
538 if (trans->qgroup_reserved) {
539 btrfs_qgroup_free(root, trans->qgroup_reserved);
540 trans->qgroup_reserved = 0;
544 unsigned long cur = trans->delayed_ref_updates;
545 trans->delayed_ref_updates = 0;
547 trans->transaction->delayed_refs.num_heads_ready > 64) {
548 trans->delayed_ref_updates = 0;
549 btrfs_run_delayed_refs(trans, root, cur);
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;
968 rsv = trans->block_rsv;
970 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
971 if (!new_root_item) {
972 ret = pending->error = -ENOMEM;
976 ret = btrfs_find_free_objectid(tree_root, &objectid);
978 pending->error = ret;
982 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
984 if (to_reserve > 0) {
985 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
988 pending->error = ret;
993 ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
994 objectid, pending->inherit);
995 kfree(pending->inherit);
997 pending->error = ret;
1001 key.objectid = objectid;
1002 key.offset = (u64)-1;
1003 key.type = BTRFS_ROOT_ITEM_KEY;
1005 trans->block_rsv = &pending->block_rsv;
1007 dentry = pending->dentry;
1008 parent = dget_parent(dentry);
1009 parent_inode = parent->d_inode;
1010 parent_root = BTRFS_I(parent_inode)->root;
1011 record_root_in_trans(trans, parent_root);
1014 * insert the directory item
1016 ret = btrfs_set_inode_index(parent_inode, &index);
1017 BUG_ON(ret); /* -ENOMEM */
1018 ret = btrfs_insert_dir_item(trans, parent_root,
1019 dentry->d_name.name, dentry->d_name.len,
1021 BTRFS_FT_DIR, index);
1022 if (ret == -EEXIST) {
1023 pending->error = -EEXIST;
1027 goto abort_trans_dput;
1030 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1031 dentry->d_name.len * 2);
1032 ret = btrfs_update_inode(trans, parent_root, parent_inode);
1034 goto abort_trans_dput;
1037 * pull in the delayed directory update
1038 * and the delayed inode item
1039 * otherwise we corrupt the FS during
1042 ret = btrfs_run_delayed_items(trans, root);
1043 if (ret) { /* Transaction aborted */
1048 record_root_in_trans(trans, root);
1049 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1050 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1051 btrfs_check_and_init_root_item(new_root_item);
1053 root_flags = btrfs_root_flags(new_root_item);
1054 if (pending->readonly)
1055 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1057 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1058 btrfs_set_root_flags(new_root_item, root_flags);
1060 old = btrfs_lock_root_node(root);
1061 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1063 btrfs_tree_unlock(old);
1064 free_extent_buffer(old);
1065 goto abort_trans_dput;
1068 btrfs_set_lock_blocking(old);
1070 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1071 /* clean up in any case */
1072 btrfs_tree_unlock(old);
1073 free_extent_buffer(old);
1075 goto abort_trans_dput;
1077 /* see comments in should_cow_block() */
1078 root->force_cow = 1;
1081 btrfs_set_root_node(new_root_item, tmp);
1082 /* record when the snapshot was created in key.offset */
1083 key.offset = trans->transid;
1084 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1085 btrfs_tree_unlock(tmp);
1086 free_extent_buffer(tmp);
1088 goto abort_trans_dput;
1091 * insert root back/forward references
1093 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1094 parent_root->root_key.objectid,
1095 btrfs_ino(parent_inode), index,
1096 dentry->d_name.name, dentry->d_name.len);
1101 key.offset = (u64)-1;
1102 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1103 if (IS_ERR(pending->snap)) {
1104 ret = PTR_ERR(pending->snap);
1108 ret = btrfs_reloc_post_snapshot(trans, pending);
1113 kfree(new_root_item);
1114 trans->block_rsv = rsv;
1115 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1121 btrfs_abort_transaction(trans, root, ret);
1126 * create all the snapshots we've scheduled for creation
1128 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1129 struct btrfs_fs_info *fs_info)
1131 struct btrfs_pending_snapshot *pending;
1132 struct list_head *head = &trans->transaction->pending_snapshots;
1134 list_for_each_entry(pending, head, list)
1135 create_pending_snapshot(trans, fs_info, pending);
1139 static void update_super_roots(struct btrfs_root *root)
1141 struct btrfs_root_item *root_item;
1142 struct btrfs_super_block *super;
1144 super = root->fs_info->super_copy;
1146 root_item = &root->fs_info->chunk_root->root_item;
1147 super->chunk_root = root_item->bytenr;
1148 super->chunk_root_generation = root_item->generation;
1149 super->chunk_root_level = root_item->level;
1151 root_item = &root->fs_info->tree_root->root_item;
1152 super->root = root_item->bytenr;
1153 super->generation = root_item->generation;
1154 super->root_level = root_item->level;
1155 if (btrfs_test_opt(root, SPACE_CACHE))
1156 super->cache_generation = root_item->generation;
1159 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1162 spin_lock(&info->trans_lock);
1163 if (info->running_transaction)
1164 ret = info->running_transaction->in_commit;
1165 spin_unlock(&info->trans_lock);
1169 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1172 spin_lock(&info->trans_lock);
1173 if (info->running_transaction)
1174 ret = info->running_transaction->blocked;
1175 spin_unlock(&info->trans_lock);
1180 * wait for the current transaction commit to start and block subsequent
1183 static void wait_current_trans_commit_start(struct btrfs_root *root,
1184 struct btrfs_transaction *trans)
1186 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1190 * wait for the current transaction to start and then become unblocked.
1193 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1194 struct btrfs_transaction *trans)
1196 wait_event(root->fs_info->transaction_wait,
1197 trans->commit_done || (trans->in_commit && !trans->blocked));
1201 * commit transactions asynchronously. once btrfs_commit_transaction_async
1202 * returns, any subsequent transaction will not be allowed to join.
1204 struct btrfs_async_commit {
1205 struct btrfs_trans_handle *newtrans;
1206 struct btrfs_root *root;
1207 struct delayed_work work;
1210 static void do_async_commit(struct work_struct *work)
1212 struct btrfs_async_commit *ac =
1213 container_of(work, struct btrfs_async_commit, work.work);
1215 btrfs_commit_transaction(ac->newtrans, ac->root);
1219 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1220 struct btrfs_root *root,
1221 int wait_for_unblock)
1223 struct btrfs_async_commit *ac;
1224 struct btrfs_transaction *cur_trans;
1226 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1230 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1232 ac->newtrans = btrfs_join_transaction(root);
1233 if (IS_ERR(ac->newtrans)) {
1234 int err = PTR_ERR(ac->newtrans);
1239 /* take transaction reference */
1240 cur_trans = trans->transaction;
1241 atomic_inc(&cur_trans->use_count);
1243 btrfs_end_transaction(trans, root);
1244 schedule_delayed_work(&ac->work, 0);
1246 /* wait for transaction to start and unblock */
1247 if (wait_for_unblock)
1248 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1250 wait_current_trans_commit_start(root, cur_trans);
1252 if (current->journal_info == trans)
1253 current->journal_info = NULL;
1255 put_transaction(cur_trans);
1260 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1261 struct btrfs_root *root, int err)
1263 struct btrfs_transaction *cur_trans = trans->transaction;
1265 WARN_ON(trans->use_count > 1);
1267 btrfs_abort_transaction(trans, root, err);
1269 spin_lock(&root->fs_info->trans_lock);
1270 list_del_init(&cur_trans->list);
1271 if (cur_trans == root->fs_info->running_transaction) {
1272 root->fs_info->running_transaction = NULL;
1273 root->fs_info->trans_no_join = 0;
1275 spin_unlock(&root->fs_info->trans_lock);
1277 btrfs_cleanup_one_transaction(trans->transaction, root);
1279 put_transaction(cur_trans);
1280 put_transaction(cur_trans);
1282 trace_btrfs_transaction_commit(root);
1284 btrfs_scrub_continue(root);
1286 if (current->journal_info == trans)
1287 current->journal_info = NULL;
1289 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1293 * btrfs_transaction state sequence:
1294 * in_commit = 0, blocked = 0 (initial)
1295 * in_commit = 1, blocked = 1
1299 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1300 struct btrfs_root *root)
1302 unsigned long joined = 0;
1303 struct btrfs_transaction *cur_trans = trans->transaction;
1304 struct btrfs_transaction *prev_trans = NULL;
1307 int should_grow = 0;
1308 unsigned long now = get_seconds();
1309 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1311 btrfs_run_ordered_operations(root, 0);
1313 btrfs_trans_release_metadata(trans, root);
1314 trans->block_rsv = NULL;
1316 if (cur_trans->aborted)
1317 goto cleanup_transaction;
1319 /* make a pass through all the delayed refs we have so far
1320 * any runnings procs may add more while we are here
1322 ret = btrfs_run_delayed_refs(trans, root, 0);
1324 goto cleanup_transaction;
1326 cur_trans = trans->transaction;
1329 * set the flushing flag so procs in this transaction have to
1330 * start sending their work down.
1332 cur_trans->delayed_refs.flushing = 1;
1334 ret = btrfs_run_delayed_refs(trans, root, 0);
1336 goto cleanup_transaction;
1338 spin_lock(&cur_trans->commit_lock);
1339 if (cur_trans->in_commit) {
1340 spin_unlock(&cur_trans->commit_lock);
1341 atomic_inc(&cur_trans->use_count);
1342 ret = btrfs_end_transaction(trans, root);
1344 wait_for_commit(root, cur_trans);
1346 put_transaction(cur_trans);
1351 trans->transaction->in_commit = 1;
1352 trans->transaction->blocked = 1;
1353 spin_unlock(&cur_trans->commit_lock);
1354 wake_up(&root->fs_info->transaction_blocked_wait);
1356 spin_lock(&root->fs_info->trans_lock);
1357 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1358 prev_trans = list_entry(cur_trans->list.prev,
1359 struct btrfs_transaction, list);
1360 if (!prev_trans->commit_done) {
1361 atomic_inc(&prev_trans->use_count);
1362 spin_unlock(&root->fs_info->trans_lock);
1364 wait_for_commit(root, prev_trans);
1366 put_transaction(prev_trans);
1368 spin_unlock(&root->fs_info->trans_lock);
1371 spin_unlock(&root->fs_info->trans_lock);
1374 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1378 int snap_pending = 0;
1380 joined = cur_trans->num_joined;
1381 if (!list_empty(&trans->transaction->pending_snapshots))
1384 WARN_ON(cur_trans != trans->transaction);
1386 if (flush_on_commit || snap_pending) {
1387 btrfs_start_delalloc_inodes(root, 1);
1388 btrfs_wait_ordered_extents(root, 0, 1);
1391 ret = btrfs_run_delayed_items(trans, root);
1393 goto cleanup_transaction;
1396 * running the delayed items may have added new refs. account
1397 * them now so that they hinder processing of more delayed refs
1398 * as little as possible.
1400 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1403 * rename don't use btrfs_join_transaction, so, once we
1404 * set the transaction to blocked above, we aren't going
1405 * to get any new ordered operations. We can safely run
1406 * it here and no for sure that nothing new will be added
1409 btrfs_run_ordered_operations(root, 1);
1411 prepare_to_wait(&cur_trans->writer_wait, &wait,
1412 TASK_UNINTERRUPTIBLE);
1414 if (atomic_read(&cur_trans->num_writers) > 1)
1415 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1416 else if (should_grow)
1417 schedule_timeout(1);
1419 finish_wait(&cur_trans->writer_wait, &wait);
1420 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1421 (should_grow && cur_trans->num_joined != joined));
1424 * Ok now we need to make sure to block out any other joins while we
1425 * commit the transaction. We could have started a join before setting
1426 * no_join so make sure to wait for num_writers to == 1 again.
1428 spin_lock(&root->fs_info->trans_lock);
1429 root->fs_info->trans_no_join = 1;
1430 spin_unlock(&root->fs_info->trans_lock);
1431 wait_event(cur_trans->writer_wait,
1432 atomic_read(&cur_trans->num_writers) == 1);
1435 * the reloc mutex makes sure that we stop
1436 * the balancing code from coming in and moving
1437 * extents around in the middle of the commit
1439 mutex_lock(&root->fs_info->reloc_mutex);
1441 ret = btrfs_run_delayed_items(trans, root);
1443 mutex_unlock(&root->fs_info->reloc_mutex);
1444 goto cleanup_transaction;
1447 ret = create_pending_snapshots(trans, root->fs_info);
1449 mutex_unlock(&root->fs_info->reloc_mutex);
1450 goto cleanup_transaction;
1453 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1455 mutex_unlock(&root->fs_info->reloc_mutex);
1456 goto cleanup_transaction;
1460 * make sure none of the code above managed to slip in a
1463 btrfs_assert_delayed_root_empty(root);
1465 WARN_ON(cur_trans != trans->transaction);
1467 btrfs_scrub_pause(root);
1468 /* btrfs_commit_tree_roots is responsible for getting the
1469 * various roots consistent with each other. Every pointer
1470 * in the tree of tree roots has to point to the most up to date
1471 * root for every subvolume and other tree. So, we have to keep
1472 * the tree logging code from jumping in and changing any
1475 * At this point in the commit, there can't be any tree-log
1476 * writers, but a little lower down we drop the trans mutex
1477 * and let new people in. By holding the tree_log_mutex
1478 * from now until after the super is written, we avoid races
1479 * with the tree-log code.
1481 mutex_lock(&root->fs_info->tree_log_mutex);
1483 ret = commit_fs_roots(trans, root);
1485 mutex_unlock(&root->fs_info->tree_log_mutex);
1486 mutex_unlock(&root->fs_info->reloc_mutex);
1487 goto cleanup_transaction;
1490 /* commit_fs_roots gets rid of all the tree log roots, it is now
1491 * safe to free the root of tree log roots
1493 btrfs_free_log_root_tree(trans, root->fs_info);
1495 ret = commit_cowonly_roots(trans, root);
1497 mutex_unlock(&root->fs_info->tree_log_mutex);
1498 mutex_unlock(&root->fs_info->reloc_mutex);
1499 goto cleanup_transaction;
1502 btrfs_prepare_extent_commit(trans, root);
1504 cur_trans = root->fs_info->running_transaction;
1506 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1507 root->fs_info->tree_root->node);
1508 switch_commit_root(root->fs_info->tree_root);
1510 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1511 root->fs_info->chunk_root->node);
1512 switch_commit_root(root->fs_info->chunk_root);
1514 assert_qgroups_uptodate(trans);
1515 update_super_roots(root);
1517 if (!root->fs_info->log_root_recovering) {
1518 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1519 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1522 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1523 sizeof(*root->fs_info->super_copy));
1525 trans->transaction->blocked = 0;
1526 spin_lock(&root->fs_info->trans_lock);
1527 root->fs_info->running_transaction = NULL;
1528 root->fs_info->trans_no_join = 0;
1529 spin_unlock(&root->fs_info->trans_lock);
1530 mutex_unlock(&root->fs_info->reloc_mutex);
1532 wake_up(&root->fs_info->transaction_wait);
1534 ret = btrfs_write_and_wait_transaction(trans, root);
1536 btrfs_error(root->fs_info, ret,
1537 "Error while writing out transaction.");
1538 mutex_unlock(&root->fs_info->tree_log_mutex);
1539 goto cleanup_transaction;
1542 ret = write_ctree_super(trans, root, 0);
1544 mutex_unlock(&root->fs_info->tree_log_mutex);
1545 goto cleanup_transaction;
1549 * the super is written, we can safely allow the tree-loggers
1550 * to go about their business
1552 mutex_unlock(&root->fs_info->tree_log_mutex);
1554 btrfs_finish_extent_commit(trans, root);
1556 cur_trans->commit_done = 1;
1558 root->fs_info->last_trans_committed = cur_trans->transid;
1560 wake_up(&cur_trans->commit_wait);
1562 spin_lock(&root->fs_info->trans_lock);
1563 list_del_init(&cur_trans->list);
1564 spin_unlock(&root->fs_info->trans_lock);
1566 put_transaction(cur_trans);
1567 put_transaction(cur_trans);
1569 trace_btrfs_transaction_commit(root);
1571 btrfs_scrub_continue(root);
1573 if (current->journal_info == trans)
1574 current->journal_info = NULL;
1576 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1578 if (current != root->fs_info->transaction_kthread)
1579 btrfs_run_delayed_iputs(root);
1583 cleanup_transaction:
1584 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1586 if (current->journal_info == trans)
1587 current->journal_info = NULL;
1588 cleanup_transaction(trans, root, ret);
1594 * interface function to delete all the snapshots we have scheduled for deletion
1596 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1599 struct btrfs_fs_info *fs_info = root->fs_info;
1601 spin_lock(&fs_info->trans_lock);
1602 list_splice_init(&fs_info->dead_roots, &list);
1603 spin_unlock(&fs_info->trans_lock);
1605 while (!list_empty(&list)) {
1608 root = list_entry(list.next, struct btrfs_root, root_list);
1609 list_del(&root->root_list);
1611 btrfs_kill_all_delayed_nodes(root);
1613 if (btrfs_header_backref_rev(root->node) <
1614 BTRFS_MIXED_BACKREF_REV)
1615 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1617 ret =btrfs_drop_snapshot(root, NULL, 1, 0);