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"
31 #define BTRFS_ROOT_TRANS_TAG 0
33 static noinline void put_transaction(struct btrfs_transaction *transaction)
35 WARN_ON(atomic_read(&transaction->use_count) == 0);
36 if (atomic_dec_and_test(&transaction->use_count)) {
37 memset(transaction, 0, sizeof(*transaction));
38 kmem_cache_free(btrfs_transaction_cachep, transaction);
42 static noinline void switch_commit_root(struct btrfs_root *root)
44 free_extent_buffer(root->commit_root);
45 root->commit_root = btrfs_root_node(root);
49 * either allocate a new transaction or hop into the existing one
51 static noinline int join_transaction(struct btrfs_root *root)
53 struct btrfs_transaction *cur_trans;
54 cur_trans = root->fs_info->running_transaction;
56 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
60 root->fs_info->generation++;
61 atomic_set(&cur_trans->num_writers, 1);
62 cur_trans->num_joined = 0;
63 cur_trans->transid = root->fs_info->generation;
64 init_waitqueue_head(&cur_trans->writer_wait);
65 init_waitqueue_head(&cur_trans->commit_wait);
66 cur_trans->in_commit = 0;
67 cur_trans->blocked = 0;
68 atomic_set(&cur_trans->use_count, 1);
69 cur_trans->commit_done = 0;
70 cur_trans->start_time = get_seconds();
72 cur_trans->delayed_refs.root = RB_ROOT;
73 cur_trans->delayed_refs.num_entries = 0;
74 cur_trans->delayed_refs.num_heads_ready = 0;
75 cur_trans->delayed_refs.num_heads = 0;
76 cur_trans->delayed_refs.flushing = 0;
77 cur_trans->delayed_refs.run_delayed_start = 0;
78 spin_lock_init(&cur_trans->delayed_refs.lock);
80 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
81 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
82 extent_io_tree_init(&cur_trans->dirty_pages,
83 root->fs_info->btree_inode->i_mapping,
85 spin_lock(&root->fs_info->new_trans_lock);
86 root->fs_info->running_transaction = cur_trans;
87 spin_unlock(&root->fs_info->new_trans_lock);
89 atomic_inc(&cur_trans->num_writers);
90 cur_trans->num_joined++;
97 * this does all the record keeping required to make sure that a reference
98 * counted root is properly recorded in a given transaction. This is required
99 * to make sure the old root from before we joined the transaction is deleted
100 * when the transaction commits
102 static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root)
105 if (root->ref_cows && root->last_trans < trans->transid) {
106 WARN_ON(root == root->fs_info->extent_root);
107 WARN_ON(root->commit_root != root->node);
109 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
110 (unsigned long)root->root_key.objectid,
111 BTRFS_ROOT_TRANS_TAG);
112 root->last_trans = trans->transid;
113 btrfs_init_reloc_root(trans, root);
118 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
119 struct btrfs_root *root)
124 mutex_lock(&root->fs_info->trans_mutex);
125 if (root->last_trans == trans->transid) {
126 mutex_unlock(&root->fs_info->trans_mutex);
130 record_root_in_trans(trans, root);
131 mutex_unlock(&root->fs_info->trans_mutex);
135 /* wait for commit against the current transaction to become unblocked
136 * when this is done, it is safe to start a new transaction, but the current
137 * transaction might not be fully on disk.
139 static void wait_current_trans(struct btrfs_root *root)
141 struct btrfs_transaction *cur_trans;
143 cur_trans = root->fs_info->running_transaction;
144 if (cur_trans && cur_trans->blocked) {
146 atomic_inc(&cur_trans->use_count);
148 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
149 TASK_UNINTERRUPTIBLE);
150 if (!cur_trans->blocked)
152 mutex_unlock(&root->fs_info->trans_mutex);
154 mutex_lock(&root->fs_info->trans_mutex);
156 finish_wait(&root->fs_info->transaction_wait, &wait);
157 put_transaction(cur_trans);
161 enum btrfs_trans_type {
168 static int may_wait_transaction(struct btrfs_root *root, int type)
170 if (!root->fs_info->log_root_recovering &&
171 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
172 type == TRANS_USERSPACE))
177 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
178 u64 num_items, int type)
180 struct btrfs_trans_handle *h;
181 struct btrfs_transaction *cur_trans;
185 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
186 return ERR_PTR(-EROFS);
188 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
190 return ERR_PTR(-ENOMEM);
192 if (type != TRANS_JOIN_NOLOCK)
193 mutex_lock(&root->fs_info->trans_mutex);
194 if (may_wait_transaction(root, type))
195 wait_current_trans(root);
197 ret = join_transaction(root);
199 kmem_cache_free(btrfs_trans_handle_cachep, h);
200 if (type != TRANS_JOIN_NOLOCK)
201 mutex_unlock(&root->fs_info->trans_mutex);
205 cur_trans = root->fs_info->running_transaction;
206 atomic_inc(&cur_trans->use_count);
207 if (type != TRANS_JOIN_NOLOCK)
208 mutex_unlock(&root->fs_info->trans_mutex);
210 h->transid = cur_trans->transid;
211 h->transaction = cur_trans;
214 h->bytes_reserved = 0;
215 h->delayed_ref_updates = 0;
219 if (cur_trans->blocked && may_wait_transaction(root, type)) {
220 btrfs_commit_transaction(h, root);
225 ret = btrfs_trans_reserve_metadata(h, root, num_items);
226 if (ret == -EAGAIN && !retries) {
228 btrfs_commit_transaction(h, root);
230 } else if (ret == -EAGAIN) {
232 * We have already retried and got EAGAIN, so really we
233 * don't have space, so set ret to -ENOSPC.
239 btrfs_end_transaction(h, root);
244 if (type != TRANS_JOIN_NOLOCK)
245 mutex_lock(&root->fs_info->trans_mutex);
246 record_root_in_trans(h, root);
247 if (type != TRANS_JOIN_NOLOCK)
248 mutex_unlock(&root->fs_info->trans_mutex);
250 if (!current->journal_info && type != TRANS_USERSPACE)
251 current->journal_info = h;
255 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
258 return start_transaction(root, num_items, TRANS_START);
260 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
263 return start_transaction(root, 0, TRANS_JOIN);
266 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root,
269 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
272 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
275 return start_transaction(r, 0, TRANS_USERSPACE);
278 /* wait for a transaction commit to be fully complete */
279 static noinline int wait_for_commit(struct btrfs_root *root,
280 struct btrfs_transaction *commit)
283 mutex_lock(&root->fs_info->trans_mutex);
284 while (!commit->commit_done) {
285 prepare_to_wait(&commit->commit_wait, &wait,
286 TASK_UNINTERRUPTIBLE);
287 if (commit->commit_done)
289 mutex_unlock(&root->fs_info->trans_mutex);
291 mutex_lock(&root->fs_info->trans_mutex);
293 mutex_unlock(&root->fs_info->trans_mutex);
294 finish_wait(&commit->commit_wait, &wait);
298 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
300 struct btrfs_transaction *cur_trans = NULL, *t;
303 mutex_lock(&root->fs_info->trans_mutex);
307 if (transid <= root->fs_info->last_trans_committed)
310 /* find specified transaction */
311 list_for_each_entry(t, &root->fs_info->trans_list, list) {
312 if (t->transid == transid) {
316 if (t->transid > transid)
321 goto out_unlock; /* bad transid */
323 /* find newest transaction that is committing | committed */
324 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
334 goto out_unlock; /* nothing committing|committed */
337 atomic_inc(&cur_trans->use_count);
338 mutex_unlock(&root->fs_info->trans_mutex);
340 wait_for_commit(root, cur_trans);
342 mutex_lock(&root->fs_info->trans_mutex);
343 put_transaction(cur_trans);
346 mutex_unlock(&root->fs_info->trans_mutex);
352 * rate limit against the drop_snapshot code. This helps to slow down new
353 * operations if the drop_snapshot code isn't able to keep up.
355 static void throttle_on_drops(struct btrfs_root *root)
357 struct btrfs_fs_info *info = root->fs_info;
358 int harder_count = 0;
361 if (atomic_read(&info->throttles)) {
364 thr = atomic_read(&info->throttle_gen);
367 prepare_to_wait(&info->transaction_throttle,
368 &wait, TASK_UNINTERRUPTIBLE);
369 if (!atomic_read(&info->throttles)) {
370 finish_wait(&info->transaction_throttle, &wait);
374 finish_wait(&info->transaction_throttle, &wait);
375 } while (thr == atomic_read(&info->throttle_gen));
378 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
382 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
386 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
393 void btrfs_throttle(struct btrfs_root *root)
395 mutex_lock(&root->fs_info->trans_mutex);
396 if (!root->fs_info->open_ioctl_trans)
397 wait_current_trans(root);
398 mutex_unlock(&root->fs_info->trans_mutex);
401 static int should_end_transaction(struct btrfs_trans_handle *trans,
402 struct btrfs_root *root)
405 ret = btrfs_block_rsv_check(trans, root,
406 &root->fs_info->global_block_rsv, 0, 5);
410 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
411 struct btrfs_root *root)
413 struct btrfs_transaction *cur_trans = trans->transaction;
416 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
419 updates = trans->delayed_ref_updates;
420 trans->delayed_ref_updates = 0;
422 btrfs_run_delayed_refs(trans, root, updates);
424 return should_end_transaction(trans, root);
427 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
428 struct btrfs_root *root, int throttle, int lock)
430 struct btrfs_transaction *cur_trans = trans->transaction;
431 struct btrfs_fs_info *info = root->fs_info;
435 unsigned long cur = trans->delayed_ref_updates;
436 trans->delayed_ref_updates = 0;
438 trans->transaction->delayed_refs.num_heads_ready > 64) {
439 trans->delayed_ref_updates = 0;
442 * do a full flush if the transaction is trying
445 if (trans->transaction->delayed_refs.flushing)
447 btrfs_run_delayed_refs(trans, root, cur);
454 btrfs_trans_release_metadata(trans, root);
456 if (lock && !root->fs_info->open_ioctl_trans &&
457 should_end_transaction(trans, root))
458 trans->transaction->blocked = 1;
460 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
462 return btrfs_commit_transaction(trans, root);
464 wake_up_process(info->transaction_kthread);
467 WARN_ON(cur_trans != info->running_transaction);
468 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
469 atomic_dec(&cur_trans->num_writers);
472 if (waitqueue_active(&cur_trans->writer_wait))
473 wake_up(&cur_trans->writer_wait);
474 put_transaction(cur_trans);
476 if (current->journal_info == trans)
477 current->journal_info = NULL;
478 memset(trans, 0, sizeof(*trans));
479 kmem_cache_free(btrfs_trans_handle_cachep, trans);
482 btrfs_run_delayed_iputs(root);
487 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
488 struct btrfs_root *root)
490 return __btrfs_end_transaction(trans, root, 0, 1);
493 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
494 struct btrfs_root *root)
496 return __btrfs_end_transaction(trans, root, 1, 1);
499 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
500 struct btrfs_root *root)
502 return __btrfs_end_transaction(trans, root, 0, 0);
506 * when btree blocks are allocated, they have some corresponding bits set for
507 * them in one of two extent_io trees. This is used to make sure all of
508 * those extents are sent to disk but does not wait on them
510 int btrfs_write_marked_extents(struct btrfs_root *root,
511 struct extent_io_tree *dirty_pages, int mark)
517 struct inode *btree_inode = root->fs_info->btree_inode;
523 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
527 while (start <= end) {
530 index = start >> PAGE_CACHE_SHIFT;
531 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
532 page = find_get_page(btree_inode->i_mapping, index);
536 btree_lock_page_hook(page);
537 if (!page->mapping) {
539 page_cache_release(page);
543 if (PageWriteback(page)) {
545 wait_on_page_writeback(page);
548 page_cache_release(page);
552 err = write_one_page(page, 0);
555 page_cache_release(page);
564 * when btree blocks are allocated, they have some corresponding bits set for
565 * them in one of two extent_io trees. This is used to make sure all of
566 * those extents are on disk for transaction or log commit. We wait
567 * on all the pages and clear them from the dirty pages state tree
569 int btrfs_wait_marked_extents(struct btrfs_root *root,
570 struct extent_io_tree *dirty_pages, int mark)
576 struct inode *btree_inode = root->fs_info->btree_inode;
582 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
587 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
588 while (start <= end) {
589 index = start >> PAGE_CACHE_SHIFT;
590 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
591 page = find_get_page(btree_inode->i_mapping, index);
594 if (PageDirty(page)) {
595 btree_lock_page_hook(page);
596 wait_on_page_writeback(page);
597 err = write_one_page(page, 0);
601 wait_on_page_writeback(page);
602 page_cache_release(page);
612 * when btree blocks are allocated, they have some corresponding bits set for
613 * them in one of two extent_io trees. This is used to make sure all of
614 * those extents are on disk for transaction or log commit
616 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
617 struct extent_io_tree *dirty_pages, int mark)
622 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
623 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
627 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
628 struct btrfs_root *root)
630 if (!trans || !trans->transaction) {
631 struct inode *btree_inode;
632 btree_inode = root->fs_info->btree_inode;
633 return filemap_write_and_wait(btree_inode->i_mapping);
635 return btrfs_write_and_wait_marked_extents(root,
636 &trans->transaction->dirty_pages,
641 * this is used to update the root pointer in the tree of tree roots.
643 * But, in the case of the extent allocation tree, updating the root
644 * pointer may allocate blocks which may change the root of the extent
647 * So, this loops and repeats and makes sure the cowonly root didn't
648 * change while the root pointer was being updated in the metadata.
650 static int update_cowonly_root(struct btrfs_trans_handle *trans,
651 struct btrfs_root *root)
656 struct btrfs_root *tree_root = root->fs_info->tree_root;
658 old_root_used = btrfs_root_used(&root->root_item);
659 btrfs_write_dirty_block_groups(trans, root);
662 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
663 if (old_root_bytenr == root->node->start &&
664 old_root_used == btrfs_root_used(&root->root_item))
667 btrfs_set_root_node(&root->root_item, root->node);
668 ret = btrfs_update_root(trans, tree_root,
673 old_root_used = btrfs_root_used(&root->root_item);
674 ret = btrfs_write_dirty_block_groups(trans, root);
678 if (root != root->fs_info->extent_root)
679 switch_commit_root(root);
685 * update all the cowonly tree roots on disk
687 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
688 struct btrfs_root *root)
690 struct btrfs_fs_info *fs_info = root->fs_info;
691 struct list_head *next;
692 struct extent_buffer *eb;
695 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
698 eb = btrfs_lock_root_node(fs_info->tree_root);
699 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
700 btrfs_tree_unlock(eb);
701 free_extent_buffer(eb);
703 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
706 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
707 next = fs_info->dirty_cowonly_roots.next;
709 root = list_entry(next, struct btrfs_root, dirty_list);
711 update_cowonly_root(trans, root);
714 down_write(&fs_info->extent_commit_sem);
715 switch_commit_root(fs_info->extent_root);
716 up_write(&fs_info->extent_commit_sem);
722 * dead roots are old snapshots that need to be deleted. This allocates
723 * a dirty root struct and adds it into the list of dead roots that need to
726 int btrfs_add_dead_root(struct btrfs_root *root)
728 mutex_lock(&root->fs_info->trans_mutex);
729 list_add(&root->root_list, &root->fs_info->dead_roots);
730 mutex_unlock(&root->fs_info->trans_mutex);
735 * update all the cowonly tree roots on disk
737 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
738 struct btrfs_root *root)
740 struct btrfs_root *gang[8];
741 struct btrfs_fs_info *fs_info = root->fs_info;
747 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
750 BTRFS_ROOT_TRANS_TAG);
753 for (i = 0; i < ret; i++) {
755 radix_tree_tag_clear(&fs_info->fs_roots_radix,
756 (unsigned long)root->root_key.objectid,
757 BTRFS_ROOT_TRANS_TAG);
759 btrfs_free_log(trans, root);
760 btrfs_update_reloc_root(trans, root);
761 btrfs_orphan_commit_root(trans, root);
763 if (root->commit_root != root->node) {
764 switch_commit_root(root);
765 btrfs_set_root_node(&root->root_item,
769 err = btrfs_update_root(trans, fs_info->tree_root,
780 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
781 * otherwise every leaf in the btree is read and defragged.
783 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
785 struct btrfs_fs_info *info = root->fs_info;
786 struct btrfs_trans_handle *trans;
790 if (xchg(&root->defrag_running, 1))
794 trans = btrfs_start_transaction(root, 0);
796 return PTR_ERR(trans);
798 ret = btrfs_defrag_leaves(trans, root, cacheonly);
800 nr = trans->blocks_used;
801 btrfs_end_transaction(trans, root);
802 btrfs_btree_balance_dirty(info->tree_root, nr);
805 if (root->fs_info->closing || ret != -EAGAIN)
808 root->defrag_running = 0;
814 * when dropping snapshots, we generate a ton of delayed refs, and it makes
815 * sense not to join the transaction while it is trying to flush the current
816 * queue of delayed refs out.
818 * This is used by the drop snapshot code only
820 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
824 mutex_lock(&info->trans_mutex);
825 while (info->running_transaction &&
826 info->running_transaction->delayed_refs.flushing) {
827 prepare_to_wait(&info->transaction_wait, &wait,
828 TASK_UNINTERRUPTIBLE);
829 mutex_unlock(&info->trans_mutex);
833 mutex_lock(&info->trans_mutex);
834 finish_wait(&info->transaction_wait, &wait);
836 mutex_unlock(&info->trans_mutex);
841 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
844 int btrfs_drop_dead_root(struct btrfs_root *root)
846 struct btrfs_trans_handle *trans;
847 struct btrfs_root *tree_root = root->fs_info->tree_root;
853 * we don't want to jump in and create a bunch of
854 * delayed refs if the transaction is starting to close
856 wait_transaction_pre_flush(tree_root->fs_info);
857 trans = btrfs_start_transaction(tree_root, 1);
860 * we've joined a transaction, make sure it isn't
863 if (trans->transaction->delayed_refs.flushing) {
864 btrfs_end_transaction(trans, tree_root);
868 ret = btrfs_drop_snapshot(trans, root);
872 ret = btrfs_update_root(trans, tree_root,
878 nr = trans->blocks_used;
879 ret = btrfs_end_transaction(trans, tree_root);
882 btrfs_btree_balance_dirty(tree_root, nr);
887 ret = btrfs_del_root(trans, tree_root, &root->root_key);
890 nr = trans->blocks_used;
891 ret = btrfs_end_transaction(trans, tree_root);
894 free_extent_buffer(root->node);
895 free_extent_buffer(root->commit_root);
898 btrfs_btree_balance_dirty(tree_root, nr);
904 * new snapshots need to be created at a very specific time in the
905 * transaction commit. This does the actual creation
907 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
908 struct btrfs_fs_info *fs_info,
909 struct btrfs_pending_snapshot *pending)
911 struct btrfs_key key;
912 struct btrfs_root_item *new_root_item;
913 struct btrfs_root *tree_root = fs_info->tree_root;
914 struct btrfs_root *root = pending->root;
915 struct btrfs_root *parent_root;
916 struct inode *parent_inode;
917 struct dentry *parent;
918 struct dentry *dentry;
919 struct extent_buffer *tmp;
920 struct extent_buffer *old;
927 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
928 if (!new_root_item) {
929 pending->error = -ENOMEM;
933 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
935 pending->error = ret;
939 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
940 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
942 if (to_reserve > 0) {
943 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
946 pending->error = ret;
951 key.objectid = objectid;
952 key.offset = (u64)-1;
953 key.type = BTRFS_ROOT_ITEM_KEY;
955 trans->block_rsv = &pending->block_rsv;
957 dentry = pending->dentry;
958 parent = dget_parent(dentry);
959 parent_inode = parent->d_inode;
960 parent_root = BTRFS_I(parent_inode)->root;
961 record_root_in_trans(trans, parent_root);
964 * insert the directory item
966 ret = btrfs_set_inode_index(parent_inode, &index);
968 ret = btrfs_insert_dir_item(trans, parent_root,
969 dentry->d_name.name, dentry->d_name.len,
970 parent_inode->i_ino, &key,
971 BTRFS_FT_DIR, index);
974 btrfs_i_size_write(parent_inode, parent_inode->i_size +
975 dentry->d_name.len * 2);
976 ret = btrfs_update_inode(trans, parent_root, parent_inode);
979 record_root_in_trans(trans, root);
980 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
981 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
982 btrfs_check_and_init_root_item(new_root_item);
984 root_flags = btrfs_root_flags(new_root_item);
985 if (pending->readonly)
986 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
988 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
989 btrfs_set_root_flags(new_root_item, root_flags);
991 old = btrfs_lock_root_node(root);
992 btrfs_cow_block(trans, root, old, NULL, 0, &old);
993 btrfs_set_lock_blocking(old);
995 btrfs_copy_root(trans, root, old, &tmp, objectid);
996 btrfs_tree_unlock(old);
997 free_extent_buffer(old);
999 btrfs_set_root_node(new_root_item, tmp);
1000 /* record when the snapshot was created in key.offset */
1001 key.offset = trans->transid;
1002 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1003 btrfs_tree_unlock(tmp);
1004 free_extent_buffer(tmp);
1008 * insert root back/forward references
1010 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1011 parent_root->root_key.objectid,
1012 parent_inode->i_ino, index,
1013 dentry->d_name.name, dentry->d_name.len);
1017 key.offset = (u64)-1;
1018 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1019 BUG_ON(IS_ERR(pending->snap));
1021 btrfs_reloc_post_snapshot(trans, pending);
1022 btrfs_orphan_post_snapshot(trans, pending);
1024 kfree(new_root_item);
1025 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1030 * create all the snapshots we've scheduled for creation
1032 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1033 struct btrfs_fs_info *fs_info)
1035 struct btrfs_pending_snapshot *pending;
1036 struct list_head *head = &trans->transaction->pending_snapshots;
1039 list_for_each_entry(pending, head, list) {
1040 ret = create_pending_snapshot(trans, fs_info, pending);
1046 static void update_super_roots(struct btrfs_root *root)
1048 struct btrfs_root_item *root_item;
1049 struct btrfs_super_block *super;
1051 super = &root->fs_info->super_copy;
1053 root_item = &root->fs_info->chunk_root->root_item;
1054 super->chunk_root = root_item->bytenr;
1055 super->chunk_root_generation = root_item->generation;
1056 super->chunk_root_level = root_item->level;
1058 root_item = &root->fs_info->tree_root->root_item;
1059 super->root = root_item->bytenr;
1060 super->generation = root_item->generation;
1061 super->root_level = root_item->level;
1062 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1063 super->cache_generation = root_item->generation;
1066 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1069 spin_lock(&info->new_trans_lock);
1070 if (info->running_transaction)
1071 ret = info->running_transaction->in_commit;
1072 spin_unlock(&info->new_trans_lock);
1076 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1079 spin_lock(&info->new_trans_lock);
1080 if (info->running_transaction)
1081 ret = info->running_transaction->blocked;
1082 spin_unlock(&info->new_trans_lock);
1087 * wait for the current transaction commit to start and block subsequent
1090 static void wait_current_trans_commit_start(struct btrfs_root *root,
1091 struct btrfs_transaction *trans)
1095 if (trans->in_commit)
1099 prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
1100 TASK_UNINTERRUPTIBLE);
1101 if (trans->in_commit) {
1102 finish_wait(&root->fs_info->transaction_blocked_wait,
1106 mutex_unlock(&root->fs_info->trans_mutex);
1108 mutex_lock(&root->fs_info->trans_mutex);
1109 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1114 * wait for the current transaction to start and then become unblocked.
1117 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1118 struct btrfs_transaction *trans)
1122 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1126 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1127 TASK_UNINTERRUPTIBLE);
1128 if (trans->commit_done ||
1129 (trans->in_commit && !trans->blocked)) {
1130 finish_wait(&root->fs_info->transaction_wait,
1134 mutex_unlock(&root->fs_info->trans_mutex);
1136 mutex_lock(&root->fs_info->trans_mutex);
1137 finish_wait(&root->fs_info->transaction_wait,
1143 * commit transactions asynchronously. once btrfs_commit_transaction_async
1144 * returns, any subsequent transaction will not be allowed to join.
1146 struct btrfs_async_commit {
1147 struct btrfs_trans_handle *newtrans;
1148 struct btrfs_root *root;
1149 struct delayed_work work;
1152 static void do_async_commit(struct work_struct *work)
1154 struct btrfs_async_commit *ac =
1155 container_of(work, struct btrfs_async_commit, work.work);
1157 btrfs_commit_transaction(ac->newtrans, ac->root);
1161 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1162 struct btrfs_root *root,
1163 int wait_for_unblock)
1165 struct btrfs_async_commit *ac;
1166 struct btrfs_transaction *cur_trans;
1168 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1172 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1174 ac->newtrans = btrfs_join_transaction(root, 0);
1175 if (IS_ERR(ac->newtrans)) {
1176 int err = PTR_ERR(ac->newtrans);
1181 /* take transaction reference */
1182 mutex_lock(&root->fs_info->trans_mutex);
1183 cur_trans = trans->transaction;
1184 atomic_inc(&cur_trans->use_count);
1185 mutex_unlock(&root->fs_info->trans_mutex);
1187 btrfs_end_transaction(trans, root);
1188 schedule_delayed_work(&ac->work, 0);
1190 /* wait for transaction to start and unblock */
1191 mutex_lock(&root->fs_info->trans_mutex);
1192 if (wait_for_unblock)
1193 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1195 wait_current_trans_commit_start(root, cur_trans);
1196 put_transaction(cur_trans);
1197 mutex_unlock(&root->fs_info->trans_mutex);
1203 * btrfs_transaction state sequence:
1204 * in_commit = 0, blocked = 0 (initial)
1205 * in_commit = 1, blocked = 1
1209 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1210 struct btrfs_root *root)
1212 unsigned long joined = 0;
1213 struct btrfs_transaction *cur_trans;
1214 struct btrfs_transaction *prev_trans = NULL;
1217 int should_grow = 0;
1218 unsigned long now = get_seconds();
1219 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1221 btrfs_run_ordered_operations(root, 0);
1223 /* make a pass through all the delayed refs we have so far
1224 * any runnings procs may add more while we are here
1226 ret = btrfs_run_delayed_refs(trans, root, 0);
1229 btrfs_trans_release_metadata(trans, root);
1231 cur_trans = trans->transaction;
1233 * set the flushing flag so procs in this transaction have to
1234 * start sending their work down.
1236 cur_trans->delayed_refs.flushing = 1;
1238 ret = btrfs_run_delayed_refs(trans, root, 0);
1241 mutex_lock(&root->fs_info->trans_mutex);
1242 if (cur_trans->in_commit) {
1243 atomic_inc(&cur_trans->use_count);
1244 mutex_unlock(&root->fs_info->trans_mutex);
1245 btrfs_end_transaction(trans, root);
1247 ret = wait_for_commit(root, cur_trans);
1250 mutex_lock(&root->fs_info->trans_mutex);
1251 put_transaction(cur_trans);
1252 mutex_unlock(&root->fs_info->trans_mutex);
1257 trans->transaction->in_commit = 1;
1258 trans->transaction->blocked = 1;
1259 wake_up(&root->fs_info->transaction_blocked_wait);
1261 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1262 prev_trans = list_entry(cur_trans->list.prev,
1263 struct btrfs_transaction, list);
1264 if (!prev_trans->commit_done) {
1265 atomic_inc(&prev_trans->use_count);
1266 mutex_unlock(&root->fs_info->trans_mutex);
1268 wait_for_commit(root, prev_trans);
1270 mutex_lock(&root->fs_info->trans_mutex);
1271 put_transaction(prev_trans);
1275 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1279 int snap_pending = 0;
1280 joined = cur_trans->num_joined;
1281 if (!list_empty(&trans->transaction->pending_snapshots))
1284 WARN_ON(cur_trans != trans->transaction);
1285 mutex_unlock(&root->fs_info->trans_mutex);
1287 if (flush_on_commit || snap_pending) {
1288 btrfs_start_delalloc_inodes(root, 1);
1289 ret = btrfs_wait_ordered_extents(root, 0, 1);
1294 * rename don't use btrfs_join_transaction, so, once we
1295 * set the transaction to blocked above, we aren't going
1296 * to get any new ordered operations. We can safely run
1297 * it here and no for sure that nothing new will be added
1300 btrfs_run_ordered_operations(root, 1);
1302 prepare_to_wait(&cur_trans->writer_wait, &wait,
1303 TASK_UNINTERRUPTIBLE);
1306 if (atomic_read(&cur_trans->num_writers) > 1)
1307 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1308 else if (should_grow)
1309 schedule_timeout(1);
1311 mutex_lock(&root->fs_info->trans_mutex);
1312 finish_wait(&cur_trans->writer_wait, &wait);
1313 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1314 (should_grow && cur_trans->num_joined != joined));
1316 ret = create_pending_snapshots(trans, root->fs_info);
1319 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1322 WARN_ON(cur_trans != trans->transaction);
1324 /* btrfs_commit_tree_roots is responsible for getting the
1325 * various roots consistent with each other. Every pointer
1326 * in the tree of tree roots has to point to the most up to date
1327 * root for every subvolume and other tree. So, we have to keep
1328 * the tree logging code from jumping in and changing any
1331 * At this point in the commit, there can't be any tree-log
1332 * writers, but a little lower down we drop the trans mutex
1333 * and let new people in. By holding the tree_log_mutex
1334 * from now until after the super is written, we avoid races
1335 * with the tree-log code.
1337 mutex_lock(&root->fs_info->tree_log_mutex);
1339 ret = commit_fs_roots(trans, root);
1342 /* commit_fs_roots gets rid of all the tree log roots, it is now
1343 * safe to free the root of tree log roots
1345 btrfs_free_log_root_tree(trans, root->fs_info);
1347 ret = commit_cowonly_roots(trans, root);
1350 btrfs_prepare_extent_commit(trans, root);
1352 cur_trans = root->fs_info->running_transaction;
1353 spin_lock(&root->fs_info->new_trans_lock);
1354 root->fs_info->running_transaction = NULL;
1355 spin_unlock(&root->fs_info->new_trans_lock);
1357 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1358 root->fs_info->tree_root->node);
1359 switch_commit_root(root->fs_info->tree_root);
1361 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1362 root->fs_info->chunk_root->node);
1363 switch_commit_root(root->fs_info->chunk_root);
1365 update_super_roots(root);
1367 if (!root->fs_info->log_root_recovering) {
1368 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1369 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1372 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1373 sizeof(root->fs_info->super_copy));
1375 trans->transaction->blocked = 0;
1377 wake_up(&root->fs_info->transaction_wait);
1379 mutex_unlock(&root->fs_info->trans_mutex);
1380 ret = btrfs_write_and_wait_transaction(trans, root);
1382 write_ctree_super(trans, root, 0);
1385 * the super is written, we can safely allow the tree-loggers
1386 * to go about their business
1388 mutex_unlock(&root->fs_info->tree_log_mutex);
1390 btrfs_finish_extent_commit(trans, root);
1392 mutex_lock(&root->fs_info->trans_mutex);
1394 cur_trans->commit_done = 1;
1396 root->fs_info->last_trans_committed = cur_trans->transid;
1398 wake_up(&cur_trans->commit_wait);
1400 list_del_init(&cur_trans->list);
1401 put_transaction(cur_trans);
1402 put_transaction(cur_trans);
1404 trace_btrfs_transaction_commit(root);
1406 mutex_unlock(&root->fs_info->trans_mutex);
1408 if (current->journal_info == trans)
1409 current->journal_info = NULL;
1411 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1413 if (current != root->fs_info->transaction_kthread)
1414 btrfs_run_delayed_iputs(root);
1420 * interface function to delete all the snapshots we have scheduled for deletion
1422 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1425 struct btrfs_fs_info *fs_info = root->fs_info;
1427 mutex_lock(&fs_info->trans_mutex);
1428 list_splice_init(&fs_info->dead_roots, &list);
1429 mutex_unlock(&fs_info->trans_mutex);
1431 while (!list_empty(&list)) {
1432 root = list_entry(list.next, struct btrfs_root, root_list);
1433 list_del(&root->root_list);
1435 if (btrfs_header_backref_rev(root->node) <
1436 BTRFS_MIXED_BACKREF_REV)
1437 btrfs_drop_snapshot(root, NULL, 0);
1439 btrfs_drop_snapshot(root, NULL, 1);