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 static noinline 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 memset(transaction, 0, sizeof(*transaction));
40 kmem_cache_free(btrfs_transaction_cachep, transaction);
44 static noinline void switch_commit_root(struct btrfs_root *root)
46 free_extent_buffer(root->commit_root);
47 root->commit_root = btrfs_root_node(root);
51 * either allocate a new transaction or hop into the existing one
53 static noinline int join_transaction(struct btrfs_root *root, int nofail)
55 struct btrfs_transaction *cur_trans;
57 spin_lock(&root->fs_info->trans_lock);
58 if (root->fs_info->trans_no_join) {
60 spin_unlock(&root->fs_info->trans_lock);
65 cur_trans = root->fs_info->running_transaction;
67 atomic_inc(&cur_trans->use_count);
68 atomic_inc(&cur_trans->num_writers);
69 cur_trans->num_joined++;
70 spin_unlock(&root->fs_info->trans_lock);
73 spin_unlock(&root->fs_info->trans_lock);
75 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
78 spin_lock(&root->fs_info->trans_lock);
79 if (root->fs_info->running_transaction) {
80 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
81 cur_trans = root->fs_info->running_transaction;
82 atomic_inc(&cur_trans->use_count);
83 atomic_inc(&cur_trans->num_writers);
84 cur_trans->num_joined++;
85 spin_unlock(&root->fs_info->trans_lock);
88 atomic_set(&cur_trans->num_writers, 1);
89 cur_trans->num_joined = 0;
90 init_waitqueue_head(&cur_trans->writer_wait);
91 init_waitqueue_head(&cur_trans->commit_wait);
92 cur_trans->in_commit = 0;
93 cur_trans->blocked = 0;
95 * One for this trans handle, one so it will live on until we
96 * commit the transaction.
98 atomic_set(&cur_trans->use_count, 2);
99 cur_trans->commit_done = 0;
100 cur_trans->start_time = get_seconds();
102 cur_trans->delayed_refs.root = RB_ROOT;
103 cur_trans->delayed_refs.num_entries = 0;
104 cur_trans->delayed_refs.num_heads_ready = 0;
105 cur_trans->delayed_refs.num_heads = 0;
106 cur_trans->delayed_refs.flushing = 0;
107 cur_trans->delayed_refs.run_delayed_start = 0;
108 spin_lock_init(&cur_trans->commit_lock);
109 spin_lock_init(&cur_trans->delayed_refs.lock);
111 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
112 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
113 extent_io_tree_init(&cur_trans->dirty_pages,
114 root->fs_info->btree_inode->i_mapping);
115 root->fs_info->generation++;
116 cur_trans->transid = root->fs_info->generation;
117 root->fs_info->running_transaction = cur_trans;
118 spin_unlock(&root->fs_info->trans_lock);
124 * this does all the record keeping required to make sure that a reference
125 * counted root is properly recorded in a given transaction. This is required
126 * to make sure the old root from before we joined the transaction is deleted
127 * when the transaction commits
129 static int record_root_in_trans(struct btrfs_trans_handle *trans,
130 struct btrfs_root *root)
132 if (root->ref_cows && root->last_trans < trans->transid) {
133 WARN_ON(root == root->fs_info->extent_root);
134 WARN_ON(root->commit_root != root->node);
137 * see below for in_trans_setup usage rules
138 * we have the reloc mutex held now, so there
139 * is only one writer in this function
141 root->in_trans_setup = 1;
143 /* make sure readers find in_trans_setup before
144 * they find our root->last_trans update
148 spin_lock(&root->fs_info->fs_roots_radix_lock);
149 if (root->last_trans == trans->transid) {
150 spin_unlock(&root->fs_info->fs_roots_radix_lock);
153 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
154 (unsigned long)root->root_key.objectid,
155 BTRFS_ROOT_TRANS_TAG);
156 spin_unlock(&root->fs_info->fs_roots_radix_lock);
157 root->last_trans = trans->transid;
159 /* this is pretty tricky. We don't want to
160 * take the relocation lock in btrfs_record_root_in_trans
161 * unless we're really doing the first setup for this root in
164 * Normally we'd use root->last_trans as a flag to decide
165 * if we want to take the expensive mutex.
167 * But, we have to set root->last_trans before we
168 * init the relocation root, otherwise, we trip over warnings
169 * in ctree.c. The solution used here is to flag ourselves
170 * with root->in_trans_setup. When this is 1, we're still
171 * fixing up the reloc trees and everyone must wait.
173 * When this is zero, they can trust root->last_trans and fly
174 * through btrfs_record_root_in_trans without having to take the
175 * lock. smp_wmb() makes sure that all the writes above are
176 * done before we pop in the zero below
178 btrfs_init_reloc_root(trans, root);
180 root->in_trans_setup = 0;
186 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
187 struct btrfs_root *root)
193 * see record_root_in_trans for comments about in_trans_setup usage
197 if (root->last_trans == trans->transid &&
198 !root->in_trans_setup)
201 mutex_lock(&root->fs_info->reloc_mutex);
202 record_root_in_trans(trans, root);
203 mutex_unlock(&root->fs_info->reloc_mutex);
208 /* wait for commit against the current transaction to become unblocked
209 * when this is done, it is safe to start a new transaction, but the current
210 * transaction might not be fully on disk.
212 static void wait_current_trans(struct btrfs_root *root)
214 struct btrfs_transaction *cur_trans;
216 spin_lock(&root->fs_info->trans_lock);
217 cur_trans = root->fs_info->running_transaction;
218 if (cur_trans && cur_trans->blocked) {
219 atomic_inc(&cur_trans->use_count);
220 spin_unlock(&root->fs_info->trans_lock);
222 wait_event(root->fs_info->transaction_wait,
223 !cur_trans->blocked);
224 put_transaction(cur_trans);
226 spin_unlock(&root->fs_info->trans_lock);
230 enum btrfs_trans_type {
237 static int may_wait_transaction(struct btrfs_root *root, int type)
239 if (root->fs_info->log_root_recovering)
242 if (type == TRANS_USERSPACE)
245 if (type == TRANS_START &&
246 !atomic_read(&root->fs_info->open_ioctl_trans))
252 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
253 u64 num_items, int type)
255 struct btrfs_trans_handle *h;
256 struct btrfs_transaction *cur_trans;
260 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
261 return ERR_PTR(-EROFS);
263 if (current->journal_info) {
264 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
265 h = current->journal_info;
267 h->orig_rsv = h->block_rsv;
273 * Do the reservation before we join the transaction so we can do all
274 * the appropriate flushing if need be.
276 if (num_items > 0 && root != root->fs_info->chunk_root) {
277 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
278 ret = btrfs_block_rsv_add(NULL, root,
279 &root->fs_info->trans_block_rsv,
285 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
287 return ERR_PTR(-ENOMEM);
289 if (may_wait_transaction(root, type))
290 wait_current_trans(root);
293 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
295 wait_current_trans(root);
296 } while (ret == -EBUSY);
299 kmem_cache_free(btrfs_trans_handle_cachep, h);
303 cur_trans = root->fs_info->running_transaction;
305 h->transid = cur_trans->transid;
306 h->transaction = cur_trans;
308 h->bytes_reserved = 0;
309 h->delayed_ref_updates = 0;
315 if (cur_trans->blocked && may_wait_transaction(root, type)) {
316 btrfs_commit_transaction(h, root);
321 h->block_rsv = &root->fs_info->trans_block_rsv;
322 h->bytes_reserved = num_bytes;
326 btrfs_record_root_in_trans(h, root);
328 if (!current->journal_info && type != TRANS_USERSPACE)
329 current->journal_info = h;
333 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
336 return start_transaction(root, num_items, TRANS_START);
338 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
340 return start_transaction(root, 0, TRANS_JOIN);
343 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
345 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
348 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
350 return start_transaction(root, 0, TRANS_USERSPACE);
353 /* wait for a transaction commit to be fully complete */
354 static noinline void wait_for_commit(struct btrfs_root *root,
355 struct btrfs_transaction *commit)
357 wait_event(commit->commit_wait, commit->commit_done);
360 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
362 struct btrfs_transaction *cur_trans = NULL, *t;
367 if (transid <= root->fs_info->last_trans_committed)
370 /* find specified transaction */
371 spin_lock(&root->fs_info->trans_lock);
372 list_for_each_entry(t, &root->fs_info->trans_list, list) {
373 if (t->transid == transid) {
375 atomic_inc(&cur_trans->use_count);
378 if (t->transid > transid)
381 spin_unlock(&root->fs_info->trans_lock);
384 goto out; /* bad transid */
386 /* find newest transaction that is committing | committed */
387 spin_lock(&root->fs_info->trans_lock);
388 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
394 atomic_inc(&cur_trans->use_count);
398 spin_unlock(&root->fs_info->trans_lock);
400 goto out; /* nothing committing|committed */
403 wait_for_commit(root, cur_trans);
405 put_transaction(cur_trans);
411 void btrfs_throttle(struct btrfs_root *root)
413 if (!atomic_read(&root->fs_info->open_ioctl_trans))
414 wait_current_trans(root);
417 static int should_end_transaction(struct btrfs_trans_handle *trans,
418 struct btrfs_root *root)
421 ret = btrfs_block_rsv_check(trans, root,
422 &root->fs_info->global_block_rsv, 0, 5, 0);
426 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
427 struct btrfs_root *root)
429 struct btrfs_transaction *cur_trans = trans->transaction;
433 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
436 updates = trans->delayed_ref_updates;
437 trans->delayed_ref_updates = 0;
439 btrfs_run_delayed_refs(trans, root, updates);
441 return should_end_transaction(trans, root);
444 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
445 struct btrfs_root *root, int throttle, int lock)
447 struct btrfs_transaction *cur_trans = trans->transaction;
448 struct btrfs_fs_info *info = root->fs_info;
451 if (--trans->use_count) {
452 trans->block_rsv = trans->orig_rsv;
456 trans->block_rsv = NULL;
458 unsigned long cur = trans->delayed_ref_updates;
459 trans->delayed_ref_updates = 0;
461 trans->transaction->delayed_refs.num_heads_ready > 64) {
462 trans->delayed_ref_updates = 0;
465 * do a full flush if the transaction is trying
468 if (trans->transaction->delayed_refs.flushing)
470 btrfs_run_delayed_refs(trans, root, cur);
477 btrfs_trans_release_metadata(trans, root);
479 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
480 should_end_transaction(trans, root)) {
481 trans->transaction->blocked = 1;
485 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
488 * We may race with somebody else here so end up having
489 * to call end_transaction on ourselves again, so inc
493 return btrfs_commit_transaction(trans, root);
495 wake_up_process(info->transaction_kthread);
499 WARN_ON(cur_trans != info->running_transaction);
500 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
501 atomic_dec(&cur_trans->num_writers);
504 if (waitqueue_active(&cur_trans->writer_wait))
505 wake_up(&cur_trans->writer_wait);
506 put_transaction(cur_trans);
508 if (current->journal_info == trans)
509 current->journal_info = NULL;
510 memset(trans, 0, sizeof(*trans));
511 kmem_cache_free(btrfs_trans_handle_cachep, trans);
514 btrfs_run_delayed_iputs(root);
519 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
520 struct btrfs_root *root)
524 ret = __btrfs_end_transaction(trans, root, 0, 1);
530 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
531 struct btrfs_root *root)
535 ret = __btrfs_end_transaction(trans, root, 1, 1);
541 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
542 struct btrfs_root *root)
546 ret = __btrfs_end_transaction(trans, root, 0, 0);
552 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
553 struct btrfs_root *root)
555 return __btrfs_end_transaction(trans, root, 1, 1);
559 * when btree blocks are allocated, they have some corresponding bits set for
560 * them in one of two extent_io trees. This is used to make sure all of
561 * those extents are sent to disk but does not wait on them
563 int btrfs_write_marked_extents(struct btrfs_root *root,
564 struct extent_io_tree *dirty_pages, int mark)
570 struct inode *btree_inode = root->fs_info->btree_inode;
576 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
580 while (start <= end) {
583 index = start >> PAGE_CACHE_SHIFT;
584 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
585 page = find_get_page(btree_inode->i_mapping, index);
589 btree_lock_page_hook(page);
590 if (!page->mapping) {
592 page_cache_release(page);
596 if (PageWriteback(page)) {
598 wait_on_page_writeback(page);
601 page_cache_release(page);
605 err = write_one_page(page, 0);
608 page_cache_release(page);
617 * when btree blocks are allocated, they have some corresponding bits set for
618 * them in one of two extent_io trees. This is used to make sure all of
619 * those extents are on disk for transaction or log commit. We wait
620 * on all the pages and clear them from the dirty pages state tree
622 int btrfs_wait_marked_extents(struct btrfs_root *root,
623 struct extent_io_tree *dirty_pages, int mark)
629 struct inode *btree_inode = root->fs_info->btree_inode;
635 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
640 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
641 while (start <= end) {
642 index = start >> PAGE_CACHE_SHIFT;
643 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
644 page = find_get_page(btree_inode->i_mapping, index);
647 if (PageDirty(page)) {
648 btree_lock_page_hook(page);
649 wait_on_page_writeback(page);
650 err = write_one_page(page, 0);
654 wait_on_page_writeback(page);
655 page_cache_release(page);
665 * when btree blocks are allocated, they have some corresponding bits set for
666 * them in one of two extent_io trees. This is used to make sure all of
667 * those extents are on disk for transaction or log commit
669 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
670 struct extent_io_tree *dirty_pages, int mark)
675 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
676 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
680 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
681 struct btrfs_root *root)
683 if (!trans || !trans->transaction) {
684 struct inode *btree_inode;
685 btree_inode = root->fs_info->btree_inode;
686 return filemap_write_and_wait(btree_inode->i_mapping);
688 return btrfs_write_and_wait_marked_extents(root,
689 &trans->transaction->dirty_pages,
694 * this is used to update the root pointer in the tree of tree roots.
696 * But, in the case of the extent allocation tree, updating the root
697 * pointer may allocate blocks which may change the root of the extent
700 * So, this loops and repeats and makes sure the cowonly root didn't
701 * change while the root pointer was being updated in the metadata.
703 static int update_cowonly_root(struct btrfs_trans_handle *trans,
704 struct btrfs_root *root)
709 struct btrfs_root *tree_root = root->fs_info->tree_root;
711 old_root_used = btrfs_root_used(&root->root_item);
712 btrfs_write_dirty_block_groups(trans, root);
715 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
716 if (old_root_bytenr == root->node->start &&
717 old_root_used == btrfs_root_used(&root->root_item))
720 btrfs_set_root_node(&root->root_item, root->node);
721 ret = btrfs_update_root(trans, tree_root,
726 old_root_used = btrfs_root_used(&root->root_item);
727 ret = btrfs_write_dirty_block_groups(trans, root);
731 if (root != root->fs_info->extent_root)
732 switch_commit_root(root);
738 * update all the cowonly tree roots on disk
740 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
741 struct btrfs_root *root)
743 struct btrfs_fs_info *fs_info = root->fs_info;
744 struct list_head *next;
745 struct extent_buffer *eb;
748 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
751 eb = btrfs_lock_root_node(fs_info->tree_root);
752 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
753 btrfs_tree_unlock(eb);
754 free_extent_buffer(eb);
756 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
759 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
760 next = fs_info->dirty_cowonly_roots.next;
762 root = list_entry(next, struct btrfs_root, dirty_list);
764 update_cowonly_root(trans, root);
767 down_write(&fs_info->extent_commit_sem);
768 switch_commit_root(fs_info->extent_root);
769 up_write(&fs_info->extent_commit_sem);
775 * dead roots are old snapshots that need to be deleted. This allocates
776 * a dirty root struct and adds it into the list of dead roots that need to
779 int btrfs_add_dead_root(struct btrfs_root *root)
781 spin_lock(&root->fs_info->trans_lock);
782 list_add(&root->root_list, &root->fs_info->dead_roots);
783 spin_unlock(&root->fs_info->trans_lock);
788 * update all the cowonly tree roots on disk
790 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
791 struct btrfs_root *root)
793 struct btrfs_root *gang[8];
794 struct btrfs_fs_info *fs_info = root->fs_info;
799 spin_lock(&fs_info->fs_roots_radix_lock);
801 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
804 BTRFS_ROOT_TRANS_TAG);
807 for (i = 0; i < ret; i++) {
809 radix_tree_tag_clear(&fs_info->fs_roots_radix,
810 (unsigned long)root->root_key.objectid,
811 BTRFS_ROOT_TRANS_TAG);
812 spin_unlock(&fs_info->fs_roots_radix_lock);
814 btrfs_free_log(trans, root);
815 btrfs_update_reloc_root(trans, root);
816 btrfs_orphan_commit_root(trans, root);
818 btrfs_save_ino_cache(root, trans);
820 if (root->commit_root != root->node) {
821 mutex_lock(&root->fs_commit_mutex);
822 switch_commit_root(root);
823 btrfs_unpin_free_ino(root);
824 mutex_unlock(&root->fs_commit_mutex);
826 btrfs_set_root_node(&root->root_item,
830 err = btrfs_update_root(trans, fs_info->tree_root,
833 spin_lock(&fs_info->fs_roots_radix_lock);
838 spin_unlock(&fs_info->fs_roots_radix_lock);
843 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
844 * otherwise every leaf in the btree is read and defragged.
846 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
848 struct btrfs_fs_info *info = root->fs_info;
849 struct btrfs_trans_handle *trans;
853 if (xchg(&root->defrag_running, 1))
857 trans = btrfs_start_transaction(root, 0);
859 return PTR_ERR(trans);
861 ret = btrfs_defrag_leaves(trans, root, cacheonly);
863 nr = trans->blocks_used;
864 btrfs_end_transaction(trans, root);
865 btrfs_btree_balance_dirty(info->tree_root, nr);
868 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
871 root->defrag_running = 0;
876 * new snapshots need to be created at a very specific time in the
877 * transaction commit. This does the actual creation
879 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
880 struct btrfs_fs_info *fs_info,
881 struct btrfs_pending_snapshot *pending)
883 struct btrfs_key key;
884 struct btrfs_root_item *new_root_item;
885 struct btrfs_root *tree_root = fs_info->tree_root;
886 struct btrfs_root *root = pending->root;
887 struct btrfs_root *parent_root;
888 struct btrfs_block_rsv *rsv;
889 struct inode *parent_inode;
890 struct dentry *parent;
891 struct dentry *dentry;
892 struct extent_buffer *tmp;
893 struct extent_buffer *old;
900 rsv = trans->block_rsv;
902 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
903 if (!new_root_item) {
904 pending->error = -ENOMEM;
908 ret = btrfs_find_free_objectid(tree_root, &objectid);
910 pending->error = ret;
914 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
916 if (to_reserve > 0) {
917 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
920 pending->error = ret;
925 key.objectid = objectid;
926 key.offset = (u64)-1;
927 key.type = BTRFS_ROOT_ITEM_KEY;
929 trans->block_rsv = &pending->block_rsv;
931 dentry = pending->dentry;
932 parent = dget_parent(dentry);
933 parent_inode = parent->d_inode;
934 parent_root = BTRFS_I(parent_inode)->root;
935 record_root_in_trans(trans, parent_root);
938 * insert the directory item
940 ret = btrfs_set_inode_index(parent_inode, &index);
942 ret = btrfs_insert_dir_item(trans, parent_root,
943 dentry->d_name.name, dentry->d_name.len,
945 BTRFS_FT_DIR, index);
948 btrfs_i_size_write(parent_inode, parent_inode->i_size +
949 dentry->d_name.len * 2);
950 ret = btrfs_update_inode(trans, parent_root, parent_inode);
954 * pull in the delayed directory update
955 * and the delayed inode item
956 * otherwise we corrupt the FS during
959 ret = btrfs_run_delayed_items(trans, root);
962 record_root_in_trans(trans, root);
963 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
964 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
965 btrfs_check_and_init_root_item(new_root_item);
967 root_flags = btrfs_root_flags(new_root_item);
968 if (pending->readonly)
969 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
971 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
972 btrfs_set_root_flags(new_root_item, root_flags);
974 old = btrfs_lock_root_node(root);
975 btrfs_cow_block(trans, root, old, NULL, 0, &old);
976 btrfs_set_lock_blocking(old);
978 btrfs_copy_root(trans, root, old, &tmp, objectid);
979 btrfs_tree_unlock(old);
980 free_extent_buffer(old);
982 btrfs_set_root_node(new_root_item, tmp);
983 /* record when the snapshot was created in key.offset */
984 key.offset = trans->transid;
985 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
986 btrfs_tree_unlock(tmp);
987 free_extent_buffer(tmp);
991 * insert root back/forward references
993 ret = btrfs_add_root_ref(trans, tree_root, objectid,
994 parent_root->root_key.objectid,
995 btrfs_ino(parent_inode), index,
996 dentry->d_name.name, dentry->d_name.len);
1000 key.offset = (u64)-1;
1001 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1002 BUG_ON(IS_ERR(pending->snap));
1004 btrfs_reloc_post_snapshot(trans, pending);
1006 kfree(new_root_item);
1007 trans->block_rsv = rsv;
1008 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1013 * create all the snapshots we've scheduled for creation
1015 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1016 struct btrfs_fs_info *fs_info)
1018 struct btrfs_pending_snapshot *pending;
1019 struct list_head *head = &trans->transaction->pending_snapshots;
1022 list_for_each_entry(pending, head, list) {
1023 ret = create_pending_snapshot(trans, fs_info, pending);
1029 static void update_super_roots(struct btrfs_root *root)
1031 struct btrfs_root_item *root_item;
1032 struct btrfs_super_block *super;
1034 super = &root->fs_info->super_copy;
1036 root_item = &root->fs_info->chunk_root->root_item;
1037 super->chunk_root = root_item->bytenr;
1038 super->chunk_root_generation = root_item->generation;
1039 super->chunk_root_level = root_item->level;
1041 root_item = &root->fs_info->tree_root->root_item;
1042 super->root = root_item->bytenr;
1043 super->generation = root_item->generation;
1044 super->root_level = root_item->level;
1045 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1046 super->cache_generation = root_item->generation;
1049 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1052 spin_lock(&info->trans_lock);
1053 if (info->running_transaction)
1054 ret = info->running_transaction->in_commit;
1055 spin_unlock(&info->trans_lock);
1059 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1062 spin_lock(&info->trans_lock);
1063 if (info->running_transaction)
1064 ret = info->running_transaction->blocked;
1065 spin_unlock(&info->trans_lock);
1070 * wait for the current transaction commit to start and block subsequent
1073 static void wait_current_trans_commit_start(struct btrfs_root *root,
1074 struct btrfs_transaction *trans)
1076 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1080 * wait for the current transaction to start and then become unblocked.
1083 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1084 struct btrfs_transaction *trans)
1086 wait_event(root->fs_info->transaction_wait,
1087 trans->commit_done || (trans->in_commit && !trans->blocked));
1091 * commit transactions asynchronously. once btrfs_commit_transaction_async
1092 * returns, any subsequent transaction will not be allowed to join.
1094 struct btrfs_async_commit {
1095 struct btrfs_trans_handle *newtrans;
1096 struct btrfs_root *root;
1097 struct delayed_work work;
1100 static void do_async_commit(struct work_struct *work)
1102 struct btrfs_async_commit *ac =
1103 container_of(work, struct btrfs_async_commit, work.work);
1105 btrfs_commit_transaction(ac->newtrans, ac->root);
1109 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1110 struct btrfs_root *root,
1111 int wait_for_unblock)
1113 struct btrfs_async_commit *ac;
1114 struct btrfs_transaction *cur_trans;
1116 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1120 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1122 ac->newtrans = btrfs_join_transaction(root);
1123 if (IS_ERR(ac->newtrans)) {
1124 int err = PTR_ERR(ac->newtrans);
1129 /* take transaction reference */
1130 cur_trans = trans->transaction;
1131 atomic_inc(&cur_trans->use_count);
1133 btrfs_end_transaction(trans, root);
1134 schedule_delayed_work(&ac->work, 0);
1136 /* wait for transaction to start and unblock */
1137 if (wait_for_unblock)
1138 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1140 wait_current_trans_commit_start(root, cur_trans);
1142 if (current->journal_info == trans)
1143 current->journal_info = NULL;
1145 put_transaction(cur_trans);
1150 * btrfs_transaction state sequence:
1151 * in_commit = 0, blocked = 0 (initial)
1152 * in_commit = 1, blocked = 1
1156 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1157 struct btrfs_root *root)
1159 unsigned long joined = 0;
1160 struct btrfs_transaction *cur_trans;
1161 struct btrfs_transaction *prev_trans = NULL;
1164 int should_grow = 0;
1165 unsigned long now = get_seconds();
1166 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1168 btrfs_run_ordered_operations(root, 0);
1170 /* make a pass through all the delayed refs we have so far
1171 * any runnings procs may add more while we are here
1173 ret = btrfs_run_delayed_refs(trans, root, 0);
1176 btrfs_trans_release_metadata(trans, root);
1178 cur_trans = trans->transaction;
1180 * set the flushing flag so procs in this transaction have to
1181 * start sending their work down.
1183 cur_trans->delayed_refs.flushing = 1;
1185 ret = btrfs_run_delayed_refs(trans, root, 0);
1188 spin_lock(&cur_trans->commit_lock);
1189 if (cur_trans->in_commit) {
1190 spin_unlock(&cur_trans->commit_lock);
1191 atomic_inc(&cur_trans->use_count);
1192 btrfs_end_transaction(trans, root);
1194 wait_for_commit(root, cur_trans);
1196 put_transaction(cur_trans);
1201 trans->transaction->in_commit = 1;
1202 trans->transaction->blocked = 1;
1203 spin_unlock(&cur_trans->commit_lock);
1204 wake_up(&root->fs_info->transaction_blocked_wait);
1206 spin_lock(&root->fs_info->trans_lock);
1207 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1208 prev_trans = list_entry(cur_trans->list.prev,
1209 struct btrfs_transaction, list);
1210 if (!prev_trans->commit_done) {
1211 atomic_inc(&prev_trans->use_count);
1212 spin_unlock(&root->fs_info->trans_lock);
1214 wait_for_commit(root, prev_trans);
1216 put_transaction(prev_trans);
1218 spin_unlock(&root->fs_info->trans_lock);
1221 spin_unlock(&root->fs_info->trans_lock);
1224 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1228 int snap_pending = 0;
1230 joined = cur_trans->num_joined;
1231 if (!list_empty(&trans->transaction->pending_snapshots))
1234 WARN_ON(cur_trans != trans->transaction);
1236 if (flush_on_commit || snap_pending) {
1237 btrfs_start_delalloc_inodes(root, 1);
1238 ret = btrfs_wait_ordered_extents(root, 0, 1);
1242 ret = btrfs_run_delayed_items(trans, root);
1246 * rename don't use btrfs_join_transaction, so, once we
1247 * set the transaction to blocked above, we aren't going
1248 * to get any new ordered operations. We can safely run
1249 * it here and no for sure that nothing new will be added
1252 btrfs_run_ordered_operations(root, 1);
1254 prepare_to_wait(&cur_trans->writer_wait, &wait,
1255 TASK_UNINTERRUPTIBLE);
1257 if (atomic_read(&cur_trans->num_writers) > 1)
1258 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1259 else if (should_grow)
1260 schedule_timeout(1);
1262 finish_wait(&cur_trans->writer_wait, &wait);
1263 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1264 (should_grow && cur_trans->num_joined != joined));
1267 * Ok now we need to make sure to block out any other joins while we
1268 * commit the transaction. We could have started a join before setting
1269 * no_join so make sure to wait for num_writers to == 1 again.
1271 spin_lock(&root->fs_info->trans_lock);
1272 root->fs_info->trans_no_join = 1;
1273 spin_unlock(&root->fs_info->trans_lock);
1274 wait_event(cur_trans->writer_wait,
1275 atomic_read(&cur_trans->num_writers) == 1);
1278 * the reloc mutex makes sure that we stop
1279 * the balancing code from coming in and moving
1280 * extents around in the middle of the commit
1282 mutex_lock(&root->fs_info->reloc_mutex);
1284 ret = btrfs_run_delayed_items(trans, root);
1287 ret = create_pending_snapshots(trans, root->fs_info);
1290 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1294 * make sure none of the code above managed to slip in a
1297 btrfs_assert_delayed_root_empty(root);
1299 WARN_ON(cur_trans != trans->transaction);
1301 btrfs_scrub_pause(root);
1302 /* btrfs_commit_tree_roots is responsible for getting the
1303 * various roots consistent with each other. Every pointer
1304 * in the tree of tree roots has to point to the most up to date
1305 * root for every subvolume and other tree. So, we have to keep
1306 * the tree logging code from jumping in and changing any
1309 * At this point in the commit, there can't be any tree-log
1310 * writers, but a little lower down we drop the trans mutex
1311 * and let new people in. By holding the tree_log_mutex
1312 * from now until after the super is written, we avoid races
1313 * with the tree-log code.
1315 mutex_lock(&root->fs_info->tree_log_mutex);
1317 ret = commit_fs_roots(trans, root);
1320 /* commit_fs_roots gets rid of all the tree log roots, it is now
1321 * safe to free the root of tree log roots
1323 btrfs_free_log_root_tree(trans, root->fs_info);
1325 ret = commit_cowonly_roots(trans, root);
1328 btrfs_prepare_extent_commit(trans, root);
1330 cur_trans = root->fs_info->running_transaction;
1332 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1333 root->fs_info->tree_root->node);
1334 switch_commit_root(root->fs_info->tree_root);
1336 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1337 root->fs_info->chunk_root->node);
1338 switch_commit_root(root->fs_info->chunk_root);
1340 update_super_roots(root);
1342 if (!root->fs_info->log_root_recovering) {
1343 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1344 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1347 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1348 sizeof(root->fs_info->super_copy));
1350 trans->transaction->blocked = 0;
1351 spin_lock(&root->fs_info->trans_lock);
1352 root->fs_info->running_transaction = NULL;
1353 root->fs_info->trans_no_join = 0;
1354 spin_unlock(&root->fs_info->trans_lock);
1355 mutex_unlock(&root->fs_info->reloc_mutex);
1357 wake_up(&root->fs_info->transaction_wait);
1359 ret = btrfs_write_and_wait_transaction(trans, root);
1361 write_ctree_super(trans, root, 0);
1364 * the super is written, we can safely allow the tree-loggers
1365 * to go about their business
1367 mutex_unlock(&root->fs_info->tree_log_mutex);
1369 btrfs_finish_extent_commit(trans, root);
1371 cur_trans->commit_done = 1;
1373 root->fs_info->last_trans_committed = cur_trans->transid;
1375 wake_up(&cur_trans->commit_wait);
1377 spin_lock(&root->fs_info->trans_lock);
1378 list_del_init(&cur_trans->list);
1379 spin_unlock(&root->fs_info->trans_lock);
1381 put_transaction(cur_trans);
1382 put_transaction(cur_trans);
1384 trace_btrfs_transaction_commit(root);
1386 btrfs_scrub_continue(root);
1388 if (current->journal_info == trans)
1389 current->journal_info = NULL;
1391 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1393 if (current != root->fs_info->transaction_kthread)
1394 btrfs_run_delayed_iputs(root);
1400 * interface function to delete all the snapshots we have scheduled for deletion
1402 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1405 struct btrfs_fs_info *fs_info = root->fs_info;
1407 spin_lock(&fs_info->trans_lock);
1408 list_splice_init(&fs_info->dead_roots, &list);
1409 spin_unlock(&fs_info->trans_lock);
1411 while (!list_empty(&list)) {
1412 root = list_entry(list.next, struct btrfs_root, root_list);
1413 list_del(&root->root_list);
1415 btrfs_kill_all_delayed_nodes(root);
1417 if (btrfs_header_backref_rev(root->node) <
1418 BTRFS_MIXED_BACKREF_REV)
1419 btrfs_drop_snapshot(root, NULL, 0);
1421 btrfs_drop_snapshot(root, NULL, 1);