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"
32 #define BTRFS_ROOT_TRANS_TAG 0
34 void put_transaction(struct btrfs_transaction *transaction)
36 WARN_ON(atomic_read(&transaction->use_count) == 0);
37 if (atomic_dec_and_test(&transaction->use_count)) {
38 BUG_ON(!list_empty(&transaction->list));
39 WARN_ON(transaction->delayed_refs.root.rb_node);
40 WARN_ON(!list_empty(&transaction->delayed_refs.seq_head));
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;
104 atomic_set(&cur_trans->num_writers, 1);
105 cur_trans->num_joined = 0;
106 init_waitqueue_head(&cur_trans->writer_wait);
107 init_waitqueue_head(&cur_trans->commit_wait);
108 cur_trans->in_commit = 0;
109 cur_trans->blocked = 0;
111 * One for this trans handle, one so it will live on until we
112 * commit the transaction.
114 atomic_set(&cur_trans->use_count, 2);
115 cur_trans->commit_done = 0;
116 cur_trans->start_time = get_seconds();
118 cur_trans->delayed_refs.root = RB_ROOT;
119 cur_trans->delayed_refs.num_entries = 0;
120 cur_trans->delayed_refs.num_heads_ready = 0;
121 cur_trans->delayed_refs.num_heads = 0;
122 cur_trans->delayed_refs.flushing = 0;
123 cur_trans->delayed_refs.run_delayed_start = 0;
124 cur_trans->delayed_refs.seq = 1;
125 init_waitqueue_head(&cur_trans->delayed_refs.seq_wait);
126 spin_lock_init(&cur_trans->commit_lock);
127 spin_lock_init(&cur_trans->delayed_refs.lock);
128 INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head);
130 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
131 list_add_tail(&cur_trans->list, &fs_info->trans_list);
132 extent_io_tree_init(&cur_trans->dirty_pages,
133 fs_info->btree_inode->i_mapping);
134 fs_info->generation++;
135 cur_trans->transid = fs_info->generation;
136 fs_info->running_transaction = cur_trans;
137 cur_trans->aborted = 0;
138 spin_unlock(&fs_info->trans_lock);
144 * this does all the record keeping required to make sure that a reference
145 * counted root is properly recorded in a given transaction. This is required
146 * to make sure the old root from before we joined the transaction is deleted
147 * when the transaction commits
149 static int record_root_in_trans(struct btrfs_trans_handle *trans,
150 struct btrfs_root *root)
152 if (root->ref_cows && root->last_trans < trans->transid) {
153 WARN_ON(root == root->fs_info->extent_root);
154 WARN_ON(root->commit_root != root->node);
157 * see below for in_trans_setup usage rules
158 * we have the reloc mutex held now, so there
159 * is only one writer in this function
161 root->in_trans_setup = 1;
163 /* make sure readers find in_trans_setup before
164 * they find our root->last_trans update
168 spin_lock(&root->fs_info->fs_roots_radix_lock);
169 if (root->last_trans == trans->transid) {
170 spin_unlock(&root->fs_info->fs_roots_radix_lock);
173 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
174 (unsigned long)root->root_key.objectid,
175 BTRFS_ROOT_TRANS_TAG);
176 spin_unlock(&root->fs_info->fs_roots_radix_lock);
177 root->last_trans = trans->transid;
179 /* this is pretty tricky. We don't want to
180 * take the relocation lock in btrfs_record_root_in_trans
181 * unless we're really doing the first setup for this root in
184 * Normally we'd use root->last_trans as a flag to decide
185 * if we want to take the expensive mutex.
187 * But, we have to set root->last_trans before we
188 * init the relocation root, otherwise, we trip over warnings
189 * in ctree.c. The solution used here is to flag ourselves
190 * with root->in_trans_setup. When this is 1, we're still
191 * fixing up the reloc trees and everyone must wait.
193 * When this is zero, they can trust root->last_trans and fly
194 * through btrfs_record_root_in_trans without having to take the
195 * lock. smp_wmb() makes sure that all the writes above are
196 * done before we pop in the zero below
198 btrfs_init_reloc_root(trans, root);
200 root->in_trans_setup = 0;
206 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
207 struct btrfs_root *root)
213 * see record_root_in_trans for comments about in_trans_setup usage
217 if (root->last_trans == trans->transid &&
218 !root->in_trans_setup)
221 mutex_lock(&root->fs_info->reloc_mutex);
222 record_root_in_trans(trans, root);
223 mutex_unlock(&root->fs_info->reloc_mutex);
228 /* wait for commit against the current transaction to become unblocked
229 * when this is done, it is safe to start a new transaction, but the current
230 * transaction might not be fully on disk.
232 static void wait_current_trans(struct btrfs_root *root)
234 struct btrfs_transaction *cur_trans;
236 spin_lock(&root->fs_info->trans_lock);
237 cur_trans = root->fs_info->running_transaction;
238 if (cur_trans && cur_trans->blocked) {
239 atomic_inc(&cur_trans->use_count);
240 spin_unlock(&root->fs_info->trans_lock);
242 wait_event(root->fs_info->transaction_wait,
243 !cur_trans->blocked);
244 put_transaction(cur_trans);
246 spin_unlock(&root->fs_info->trans_lock);
250 enum btrfs_trans_type {
257 static int may_wait_transaction(struct btrfs_root *root, int type)
259 if (root->fs_info->log_root_recovering)
262 if (type == TRANS_USERSPACE)
265 if (type == TRANS_START &&
266 !atomic_read(&root->fs_info->open_ioctl_trans))
272 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
273 u64 num_items, int type)
275 struct btrfs_trans_handle *h;
276 struct btrfs_transaction *cur_trans;
280 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
281 return ERR_PTR(-EROFS);
283 if (current->journal_info) {
284 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
285 h = current->journal_info;
287 h->orig_rsv = h->block_rsv;
293 * Do the reservation before we join the transaction so we can do all
294 * the appropriate flushing if need be.
296 if (num_items > 0 && root != root->fs_info->chunk_root) {
297 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
298 ret = btrfs_block_rsv_add(root,
299 &root->fs_info->trans_block_rsv,
305 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
307 return ERR_PTR(-ENOMEM);
309 if (may_wait_transaction(root, type))
310 wait_current_trans(root);
313 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
315 wait_current_trans(root);
316 } while (ret == -EBUSY);
319 kmem_cache_free(btrfs_trans_handle_cachep, h);
323 cur_trans = root->fs_info->running_transaction;
325 h->transid = cur_trans->transid;
326 h->transaction = cur_trans;
328 h->bytes_reserved = 0;
329 h->delayed_ref_updates = 0;
336 if (cur_trans->blocked && may_wait_transaction(root, type)) {
337 btrfs_commit_transaction(h, root);
342 trace_btrfs_space_reservation(root->fs_info, "transaction",
343 h->transid, num_bytes, 1);
344 h->block_rsv = &root->fs_info->trans_block_rsv;
345 h->bytes_reserved = num_bytes;
349 btrfs_record_root_in_trans(h, root);
351 if (!current->journal_info && type != TRANS_USERSPACE)
352 current->journal_info = h;
356 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
359 return start_transaction(root, num_items, TRANS_START);
361 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
363 return start_transaction(root, 0, TRANS_JOIN);
366 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
368 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
371 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
373 return start_transaction(root, 0, TRANS_USERSPACE);
376 /* wait for a transaction commit to be fully complete */
377 static noinline void wait_for_commit(struct btrfs_root *root,
378 struct btrfs_transaction *commit)
380 wait_event(commit->commit_wait, commit->commit_done);
383 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
385 struct btrfs_transaction *cur_trans = NULL, *t;
390 if (transid <= root->fs_info->last_trans_committed)
393 /* find specified transaction */
394 spin_lock(&root->fs_info->trans_lock);
395 list_for_each_entry(t, &root->fs_info->trans_list, list) {
396 if (t->transid == transid) {
398 atomic_inc(&cur_trans->use_count);
401 if (t->transid > transid)
404 spin_unlock(&root->fs_info->trans_lock);
407 goto out; /* bad transid */
409 /* find newest transaction that is committing | committed */
410 spin_lock(&root->fs_info->trans_lock);
411 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
417 atomic_inc(&cur_trans->use_count);
421 spin_unlock(&root->fs_info->trans_lock);
423 goto out; /* nothing committing|committed */
426 wait_for_commit(root, cur_trans);
428 put_transaction(cur_trans);
434 void btrfs_throttle(struct btrfs_root *root)
436 if (!atomic_read(&root->fs_info->open_ioctl_trans))
437 wait_current_trans(root);
440 static int should_end_transaction(struct btrfs_trans_handle *trans,
441 struct btrfs_root *root)
445 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
449 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
450 struct btrfs_root *root)
452 struct btrfs_transaction *cur_trans = trans->transaction;
453 struct btrfs_block_rsv *rsv = trans->block_rsv;
458 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
462 * We need to do this in case we're deleting csums so the global block
463 * rsv get's used instead of the csum block rsv.
465 trans->block_rsv = NULL;
467 updates = trans->delayed_ref_updates;
468 trans->delayed_ref_updates = 0;
470 err = btrfs_run_delayed_refs(trans, root, updates);
471 if (err) /* Error code will also eval true */
475 trans->block_rsv = rsv;
477 return should_end_transaction(trans, root);
480 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
481 struct btrfs_root *root, int throttle, int lock)
483 struct btrfs_transaction *cur_trans = trans->transaction;
484 struct btrfs_fs_info *info = root->fs_info;
488 if (--trans->use_count) {
489 trans->block_rsv = trans->orig_rsv;
493 btrfs_trans_release_metadata(trans, root);
494 trans->block_rsv = NULL;
496 unsigned long cur = trans->delayed_ref_updates;
497 trans->delayed_ref_updates = 0;
499 trans->transaction->delayed_refs.num_heads_ready > 64) {
500 trans->delayed_ref_updates = 0;
501 btrfs_run_delayed_refs(trans, root, cur);
508 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
509 should_end_transaction(trans, root)) {
510 trans->transaction->blocked = 1;
514 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
517 * We may race with somebody else here so end up having
518 * to call end_transaction on ourselves again, so inc
522 return btrfs_commit_transaction(trans, root);
524 wake_up_process(info->transaction_kthread);
528 WARN_ON(cur_trans != info->running_transaction);
529 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
530 atomic_dec(&cur_trans->num_writers);
533 if (waitqueue_active(&cur_trans->writer_wait))
534 wake_up(&cur_trans->writer_wait);
535 put_transaction(cur_trans);
537 if (current->journal_info == trans)
538 current->journal_info = NULL;
541 btrfs_run_delayed_iputs(root);
543 if (trans->aborted ||
544 root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
548 memset(trans, 0, sizeof(*trans));
549 kmem_cache_free(btrfs_trans_handle_cachep, trans);
553 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
554 struct btrfs_root *root)
558 ret = __btrfs_end_transaction(trans, root, 0, 1);
564 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
565 struct btrfs_root *root)
569 ret = __btrfs_end_transaction(trans, root, 1, 1);
575 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
576 struct btrfs_root *root)
580 ret = __btrfs_end_transaction(trans, root, 0, 0);
586 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
587 struct btrfs_root *root)
589 return __btrfs_end_transaction(trans, root, 1, 1);
593 * when btree blocks are allocated, they have some corresponding bits set for
594 * them in one of two extent_io trees. This is used to make sure all of
595 * those extents are sent to disk but does not wait on them
597 int btrfs_write_marked_extents(struct btrfs_root *root,
598 struct extent_io_tree *dirty_pages, int mark)
602 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
606 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
608 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
610 err = filemap_fdatawrite_range(mapping, start, end);
622 * when btree blocks are allocated, they have some corresponding bits set for
623 * them in one of two extent_io trees. This is used to make sure all of
624 * those extents are on disk for transaction or log commit. We wait
625 * on all the pages and clear them from the dirty pages state tree
627 int btrfs_wait_marked_extents(struct btrfs_root *root,
628 struct extent_io_tree *dirty_pages, int mark)
632 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
636 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
638 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
639 err = filemap_fdatawait_range(mapping, start, end);
651 * when btree blocks are allocated, they have some corresponding bits set for
652 * them in one of two extent_io trees. This is used to make sure all of
653 * those extents are on disk for transaction or log commit
655 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
656 struct extent_io_tree *dirty_pages, int mark)
661 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
662 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
671 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
672 struct btrfs_root *root)
674 if (!trans || !trans->transaction) {
675 struct inode *btree_inode;
676 btree_inode = root->fs_info->btree_inode;
677 return filemap_write_and_wait(btree_inode->i_mapping);
679 return btrfs_write_and_wait_marked_extents(root,
680 &trans->transaction->dirty_pages,
685 * this is used to update the root pointer in the tree of tree roots.
687 * But, in the case of the extent allocation tree, updating the root
688 * pointer may allocate blocks which may change the root of the extent
691 * So, this loops and repeats and makes sure the cowonly root didn't
692 * change while the root pointer was being updated in the metadata.
694 static int update_cowonly_root(struct btrfs_trans_handle *trans,
695 struct btrfs_root *root)
700 struct btrfs_root *tree_root = root->fs_info->tree_root;
702 old_root_used = btrfs_root_used(&root->root_item);
703 btrfs_write_dirty_block_groups(trans, root);
706 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
707 if (old_root_bytenr == root->node->start &&
708 old_root_used == btrfs_root_used(&root->root_item))
711 btrfs_set_root_node(&root->root_item, root->node);
712 ret = btrfs_update_root(trans, tree_root,
718 old_root_used = btrfs_root_used(&root->root_item);
719 ret = btrfs_write_dirty_block_groups(trans, root);
724 if (root != root->fs_info->extent_root)
725 switch_commit_root(root);
731 * update all the cowonly tree roots on disk
733 * The error handling in this function may not be obvious. Any of the
734 * failures will cause the file system to go offline. We still need
735 * to clean up the delayed refs.
737 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
738 struct btrfs_root *root)
740 struct btrfs_fs_info *fs_info = root->fs_info;
741 struct list_head *next;
742 struct extent_buffer *eb;
745 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
749 eb = btrfs_lock_root_node(fs_info->tree_root);
750 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
752 btrfs_tree_unlock(eb);
753 free_extent_buffer(eb);
758 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
762 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
763 next = fs_info->dirty_cowonly_roots.next;
765 root = list_entry(next, struct btrfs_root, dirty_list);
767 ret = update_cowonly_root(trans, root);
772 down_write(&fs_info->extent_commit_sem);
773 switch_commit_root(fs_info->extent_root);
774 up_write(&fs_info->extent_commit_sem);
780 * dead roots are old snapshots that need to be deleted. This allocates
781 * a dirty root struct and adds it into the list of dead roots that need to
784 int btrfs_add_dead_root(struct btrfs_root *root)
786 spin_lock(&root->fs_info->trans_lock);
787 list_add(&root->root_list, &root->fs_info->dead_roots);
788 spin_unlock(&root->fs_info->trans_lock);
793 * update all the cowonly tree roots on disk
795 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
796 struct btrfs_root *root)
798 struct btrfs_root *gang[8];
799 struct btrfs_fs_info *fs_info = root->fs_info;
804 spin_lock(&fs_info->fs_roots_radix_lock);
806 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
809 BTRFS_ROOT_TRANS_TAG);
812 for (i = 0; i < ret; i++) {
814 radix_tree_tag_clear(&fs_info->fs_roots_radix,
815 (unsigned long)root->root_key.objectid,
816 BTRFS_ROOT_TRANS_TAG);
817 spin_unlock(&fs_info->fs_roots_radix_lock);
819 btrfs_free_log(trans, root);
820 btrfs_update_reloc_root(trans, root);
821 btrfs_orphan_commit_root(trans, root);
823 btrfs_save_ino_cache(root, trans);
825 /* see comments in should_cow_block() */
829 if (root->commit_root != root->node) {
830 mutex_lock(&root->fs_commit_mutex);
831 switch_commit_root(root);
832 btrfs_unpin_free_ino(root);
833 mutex_unlock(&root->fs_commit_mutex);
835 btrfs_set_root_node(&root->root_item,
839 err = btrfs_update_root(trans, fs_info->tree_root,
842 spin_lock(&fs_info->fs_roots_radix_lock);
847 spin_unlock(&fs_info->fs_roots_radix_lock);
852 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
853 * otherwise every leaf in the btree is read and defragged.
855 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
857 struct btrfs_fs_info *info = root->fs_info;
858 struct btrfs_trans_handle *trans;
862 if (xchg(&root->defrag_running, 1))
866 trans = btrfs_start_transaction(root, 0);
868 return PTR_ERR(trans);
870 ret = btrfs_defrag_leaves(trans, root, cacheonly);
872 nr = trans->blocks_used;
873 btrfs_end_transaction(trans, root);
874 btrfs_btree_balance_dirty(info->tree_root, nr);
877 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
880 root->defrag_running = 0;
885 * new snapshots need to be created at a very specific time in the
886 * transaction commit. This does the actual creation
888 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
889 struct btrfs_fs_info *fs_info,
890 struct btrfs_pending_snapshot *pending)
892 struct btrfs_key key;
893 struct btrfs_root_item *new_root_item;
894 struct btrfs_root *tree_root = fs_info->tree_root;
895 struct btrfs_root *root = pending->root;
896 struct btrfs_root *parent_root;
897 struct btrfs_block_rsv *rsv;
898 struct inode *parent_inode;
899 struct dentry *parent;
900 struct dentry *dentry;
901 struct extent_buffer *tmp;
902 struct extent_buffer *old;
909 rsv = trans->block_rsv;
911 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
912 if (!new_root_item) {
913 ret = pending->error = -ENOMEM;
917 ret = btrfs_find_free_objectid(tree_root, &objectid);
919 pending->error = ret;
923 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
925 if (to_reserve > 0) {
926 ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv,
929 pending->error = ret;
934 key.objectid = objectid;
935 key.offset = (u64)-1;
936 key.type = BTRFS_ROOT_ITEM_KEY;
938 trans->block_rsv = &pending->block_rsv;
940 dentry = pending->dentry;
941 parent = dget_parent(dentry);
942 parent_inode = parent->d_inode;
943 parent_root = BTRFS_I(parent_inode)->root;
944 record_root_in_trans(trans, parent_root);
947 * insert the directory item
949 ret = btrfs_set_inode_index(parent_inode, &index);
950 BUG_ON(ret); /* -ENOMEM */
951 ret = btrfs_insert_dir_item(trans, parent_root,
952 dentry->d_name.name, dentry->d_name.len,
954 BTRFS_FT_DIR, index);
955 if (ret == -EEXIST) {
956 pending->error = -EEXIST;
960 goto abort_trans_dput;
963 btrfs_i_size_write(parent_inode, parent_inode->i_size +
964 dentry->d_name.len * 2);
965 ret = btrfs_update_inode(trans, parent_root, parent_inode);
967 goto abort_trans_dput;
970 * pull in the delayed directory update
971 * and the delayed inode item
972 * otherwise we corrupt the FS during
975 ret = btrfs_run_delayed_items(trans, root);
976 if (ret) { /* Transaction aborted */
981 record_root_in_trans(trans, root);
982 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
983 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
984 btrfs_check_and_init_root_item(new_root_item);
986 root_flags = btrfs_root_flags(new_root_item);
987 if (pending->readonly)
988 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
990 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
991 btrfs_set_root_flags(new_root_item, root_flags);
993 old = btrfs_lock_root_node(root);
994 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
996 btrfs_tree_unlock(old);
997 free_extent_buffer(old);
998 goto abort_trans_dput;
1001 btrfs_set_lock_blocking(old);
1003 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1004 /* clean up in any case */
1005 btrfs_tree_unlock(old);
1006 free_extent_buffer(old);
1008 goto abort_trans_dput;
1010 /* see comments in should_cow_block() */
1011 root->force_cow = 1;
1014 btrfs_set_root_node(new_root_item, tmp);
1015 /* record when the snapshot was created in key.offset */
1016 key.offset = trans->transid;
1017 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1018 btrfs_tree_unlock(tmp);
1019 free_extent_buffer(tmp);
1021 goto abort_trans_dput;
1024 * insert root back/forward references
1026 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1027 parent_root->root_key.objectid,
1028 btrfs_ino(parent_inode), index,
1029 dentry->d_name.name, dentry->d_name.len);
1034 key.offset = (u64)-1;
1035 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1036 if (IS_ERR(pending->snap)) {
1037 ret = PTR_ERR(pending->snap);
1041 ret = btrfs_reloc_post_snapshot(trans, pending);
1046 kfree(new_root_item);
1047 trans->block_rsv = rsv;
1048 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1054 btrfs_abort_transaction(trans, root, ret);
1059 * create all the snapshots we've scheduled for creation
1061 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1062 struct btrfs_fs_info *fs_info)
1064 struct btrfs_pending_snapshot *pending;
1065 struct list_head *head = &trans->transaction->pending_snapshots;
1067 list_for_each_entry(pending, head, list)
1068 create_pending_snapshot(trans, fs_info, pending);
1072 static void update_super_roots(struct btrfs_root *root)
1074 struct btrfs_root_item *root_item;
1075 struct btrfs_super_block *super;
1077 super = root->fs_info->super_copy;
1079 root_item = &root->fs_info->chunk_root->root_item;
1080 super->chunk_root = root_item->bytenr;
1081 super->chunk_root_generation = root_item->generation;
1082 super->chunk_root_level = root_item->level;
1084 root_item = &root->fs_info->tree_root->root_item;
1085 super->root = root_item->bytenr;
1086 super->generation = root_item->generation;
1087 super->root_level = root_item->level;
1088 if (btrfs_test_opt(root, SPACE_CACHE))
1089 super->cache_generation = root_item->generation;
1092 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1095 spin_lock(&info->trans_lock);
1096 if (info->running_transaction)
1097 ret = info->running_transaction->in_commit;
1098 spin_unlock(&info->trans_lock);
1102 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1105 spin_lock(&info->trans_lock);
1106 if (info->running_transaction)
1107 ret = info->running_transaction->blocked;
1108 spin_unlock(&info->trans_lock);
1113 * wait for the current transaction commit to start and block subsequent
1116 static void wait_current_trans_commit_start(struct btrfs_root *root,
1117 struct btrfs_transaction *trans)
1119 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1123 * wait for the current transaction to start and then become unblocked.
1126 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1127 struct btrfs_transaction *trans)
1129 wait_event(root->fs_info->transaction_wait,
1130 trans->commit_done || (trans->in_commit && !trans->blocked));
1134 * commit transactions asynchronously. once btrfs_commit_transaction_async
1135 * returns, any subsequent transaction will not be allowed to join.
1137 struct btrfs_async_commit {
1138 struct btrfs_trans_handle *newtrans;
1139 struct btrfs_root *root;
1140 struct delayed_work work;
1143 static void do_async_commit(struct work_struct *work)
1145 struct btrfs_async_commit *ac =
1146 container_of(work, struct btrfs_async_commit, work.work);
1148 btrfs_commit_transaction(ac->newtrans, ac->root);
1152 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1153 struct btrfs_root *root,
1154 int wait_for_unblock)
1156 struct btrfs_async_commit *ac;
1157 struct btrfs_transaction *cur_trans;
1159 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1163 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1165 ac->newtrans = btrfs_join_transaction(root);
1166 if (IS_ERR(ac->newtrans)) {
1167 int err = PTR_ERR(ac->newtrans);
1172 /* take transaction reference */
1173 cur_trans = trans->transaction;
1174 atomic_inc(&cur_trans->use_count);
1176 btrfs_end_transaction(trans, root);
1177 schedule_delayed_work(&ac->work, 0);
1179 /* wait for transaction to start and unblock */
1180 if (wait_for_unblock)
1181 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1183 wait_current_trans_commit_start(root, cur_trans);
1185 if (current->journal_info == trans)
1186 current->journal_info = NULL;
1188 put_transaction(cur_trans);
1193 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1194 struct btrfs_root *root)
1196 struct btrfs_transaction *cur_trans = trans->transaction;
1198 WARN_ON(trans->use_count > 1);
1200 spin_lock(&root->fs_info->trans_lock);
1201 list_del_init(&cur_trans->list);
1202 spin_unlock(&root->fs_info->trans_lock);
1204 btrfs_cleanup_one_transaction(trans->transaction, root);
1206 put_transaction(cur_trans);
1207 put_transaction(cur_trans);
1209 trace_btrfs_transaction_commit(root);
1211 btrfs_scrub_continue(root);
1213 if (current->journal_info == trans)
1214 current->journal_info = NULL;
1216 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1220 * btrfs_transaction state sequence:
1221 * in_commit = 0, blocked = 0 (initial)
1222 * in_commit = 1, blocked = 1
1226 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1227 struct btrfs_root *root)
1229 unsigned long joined = 0;
1230 struct btrfs_transaction *cur_trans = trans->transaction;
1231 struct btrfs_transaction *prev_trans = NULL;
1234 int should_grow = 0;
1235 unsigned long now = get_seconds();
1236 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1238 btrfs_run_ordered_operations(root, 0);
1240 btrfs_trans_release_metadata(trans, root);
1241 trans->block_rsv = NULL;
1243 if (cur_trans->aborted)
1244 goto cleanup_transaction;
1246 /* make a pass through all the delayed refs we have so far
1247 * any runnings procs may add more while we are here
1249 ret = btrfs_run_delayed_refs(trans, root, 0);
1251 goto cleanup_transaction;
1253 cur_trans = trans->transaction;
1256 * set the flushing flag so procs in this transaction have to
1257 * start sending their work down.
1259 cur_trans->delayed_refs.flushing = 1;
1261 ret = btrfs_run_delayed_refs(trans, root, 0);
1263 goto cleanup_transaction;
1265 spin_lock(&cur_trans->commit_lock);
1266 if (cur_trans->in_commit) {
1267 spin_unlock(&cur_trans->commit_lock);
1268 atomic_inc(&cur_trans->use_count);
1269 ret = btrfs_end_transaction(trans, root);
1271 wait_for_commit(root, cur_trans);
1273 put_transaction(cur_trans);
1278 trans->transaction->in_commit = 1;
1279 trans->transaction->blocked = 1;
1280 spin_unlock(&cur_trans->commit_lock);
1281 wake_up(&root->fs_info->transaction_blocked_wait);
1283 spin_lock(&root->fs_info->trans_lock);
1284 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1285 prev_trans = list_entry(cur_trans->list.prev,
1286 struct btrfs_transaction, list);
1287 if (!prev_trans->commit_done) {
1288 atomic_inc(&prev_trans->use_count);
1289 spin_unlock(&root->fs_info->trans_lock);
1291 wait_for_commit(root, prev_trans);
1293 put_transaction(prev_trans);
1295 spin_unlock(&root->fs_info->trans_lock);
1298 spin_unlock(&root->fs_info->trans_lock);
1301 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1305 int snap_pending = 0;
1307 joined = cur_trans->num_joined;
1308 if (!list_empty(&trans->transaction->pending_snapshots))
1311 WARN_ON(cur_trans != trans->transaction);
1313 if (flush_on_commit || snap_pending) {
1314 btrfs_start_delalloc_inodes(root, 1);
1315 btrfs_wait_ordered_extents(root, 0, 1);
1318 ret = btrfs_run_delayed_items(trans, root);
1320 goto cleanup_transaction;
1323 * rename don't use btrfs_join_transaction, so, once we
1324 * set the transaction to blocked above, we aren't going
1325 * to get any new ordered operations. We can safely run
1326 * it here and no for sure that nothing new will be added
1329 btrfs_run_ordered_operations(root, 1);
1331 prepare_to_wait(&cur_trans->writer_wait, &wait,
1332 TASK_UNINTERRUPTIBLE);
1334 if (atomic_read(&cur_trans->num_writers) > 1)
1335 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1336 else if (should_grow)
1337 schedule_timeout(1);
1339 finish_wait(&cur_trans->writer_wait, &wait);
1340 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1341 (should_grow && cur_trans->num_joined != joined));
1344 * Ok now we need to make sure to block out any other joins while we
1345 * commit the transaction. We could have started a join before setting
1346 * no_join so make sure to wait for num_writers to == 1 again.
1348 spin_lock(&root->fs_info->trans_lock);
1349 root->fs_info->trans_no_join = 1;
1350 spin_unlock(&root->fs_info->trans_lock);
1351 wait_event(cur_trans->writer_wait,
1352 atomic_read(&cur_trans->num_writers) == 1);
1355 * the reloc mutex makes sure that we stop
1356 * the balancing code from coming in and moving
1357 * extents around in the middle of the commit
1359 mutex_lock(&root->fs_info->reloc_mutex);
1361 ret = btrfs_run_delayed_items(trans, root);
1363 mutex_unlock(&root->fs_info->reloc_mutex);
1364 goto cleanup_transaction;
1367 ret = create_pending_snapshots(trans, root->fs_info);
1369 mutex_unlock(&root->fs_info->reloc_mutex);
1370 goto cleanup_transaction;
1373 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1375 mutex_unlock(&root->fs_info->reloc_mutex);
1376 goto cleanup_transaction;
1380 * make sure none of the code above managed to slip in a
1383 btrfs_assert_delayed_root_empty(root);
1385 WARN_ON(cur_trans != trans->transaction);
1387 btrfs_scrub_pause(root);
1388 /* btrfs_commit_tree_roots is responsible for getting the
1389 * various roots consistent with each other. Every pointer
1390 * in the tree of tree roots has to point to the most up to date
1391 * root for every subvolume and other tree. So, we have to keep
1392 * the tree logging code from jumping in and changing any
1395 * At this point in the commit, there can't be any tree-log
1396 * writers, but a little lower down we drop the trans mutex
1397 * and let new people in. By holding the tree_log_mutex
1398 * from now until after the super is written, we avoid races
1399 * with the tree-log code.
1401 mutex_lock(&root->fs_info->tree_log_mutex);
1403 ret = commit_fs_roots(trans, root);
1405 mutex_unlock(&root->fs_info->tree_log_mutex);
1406 mutex_unlock(&root->fs_info->reloc_mutex);
1407 goto cleanup_transaction;
1410 /* commit_fs_roots gets rid of all the tree log roots, it is now
1411 * safe to free the root of tree log roots
1413 btrfs_free_log_root_tree(trans, root->fs_info);
1415 ret = commit_cowonly_roots(trans, root);
1417 mutex_unlock(&root->fs_info->tree_log_mutex);
1418 mutex_unlock(&root->fs_info->reloc_mutex);
1419 goto cleanup_transaction;
1422 btrfs_prepare_extent_commit(trans, root);
1424 cur_trans = root->fs_info->running_transaction;
1426 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1427 root->fs_info->tree_root->node);
1428 switch_commit_root(root->fs_info->tree_root);
1430 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1431 root->fs_info->chunk_root->node);
1432 switch_commit_root(root->fs_info->chunk_root);
1434 update_super_roots(root);
1436 if (!root->fs_info->log_root_recovering) {
1437 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1438 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1441 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1442 sizeof(*root->fs_info->super_copy));
1444 trans->transaction->blocked = 0;
1445 spin_lock(&root->fs_info->trans_lock);
1446 root->fs_info->running_transaction = NULL;
1447 root->fs_info->trans_no_join = 0;
1448 spin_unlock(&root->fs_info->trans_lock);
1449 mutex_unlock(&root->fs_info->reloc_mutex);
1451 wake_up(&root->fs_info->transaction_wait);
1453 ret = btrfs_write_and_wait_transaction(trans, root);
1455 btrfs_error(root->fs_info, ret,
1456 "Error while writing out transaction.");
1457 mutex_unlock(&root->fs_info->tree_log_mutex);
1458 goto cleanup_transaction;
1461 ret = write_ctree_super(trans, root, 0);
1463 mutex_unlock(&root->fs_info->tree_log_mutex);
1464 goto cleanup_transaction;
1468 * the super is written, we can safely allow the tree-loggers
1469 * to go about their business
1471 mutex_unlock(&root->fs_info->tree_log_mutex);
1473 btrfs_finish_extent_commit(trans, root);
1475 cur_trans->commit_done = 1;
1477 root->fs_info->last_trans_committed = cur_trans->transid;
1479 wake_up(&cur_trans->commit_wait);
1481 spin_lock(&root->fs_info->trans_lock);
1482 list_del_init(&cur_trans->list);
1483 spin_unlock(&root->fs_info->trans_lock);
1485 put_transaction(cur_trans);
1486 put_transaction(cur_trans);
1488 trace_btrfs_transaction_commit(root);
1490 btrfs_scrub_continue(root);
1492 if (current->journal_info == trans)
1493 current->journal_info = NULL;
1495 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1497 if (current != root->fs_info->transaction_kthread)
1498 btrfs_run_delayed_iputs(root);
1502 cleanup_transaction:
1503 btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n");
1505 if (current->journal_info == trans)
1506 current->journal_info = NULL;
1507 cleanup_transaction(trans, root);
1513 * interface function to delete all the snapshots we have scheduled for deletion
1515 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1518 struct btrfs_fs_info *fs_info = root->fs_info;
1520 spin_lock(&fs_info->trans_lock);
1521 list_splice_init(&fs_info->dead_roots, &list);
1522 spin_unlock(&fs_info->trans_lock);
1524 while (!list_empty(&list)) {
1527 root = list_entry(list.next, struct btrfs_root, root_list);
1528 list_del(&root->root_list);
1530 btrfs_kill_all_delayed_nodes(root);
1532 if (btrfs_header_backref_rev(root->node) <
1533 BTRFS_MIXED_BACKREF_REV)
1534 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1536 ret =btrfs_drop_snapshot(root, NULL, 1, 0);