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/sched.h>
21 #include <linux/writeback.h>
22 #include <linux/pagemap.h>
25 #include "transaction.h"
27 #include "ref-cache.h"
29 static int total_trans = 0;
30 extern struct kmem_cache *btrfs_trans_handle_cachep;
31 extern struct kmem_cache *btrfs_transaction_cachep;
33 #define BTRFS_ROOT_TRANS_TAG 0
35 static noinline void put_transaction(struct btrfs_transaction *transaction)
37 WARN_ON(transaction->use_count == 0);
38 transaction->use_count--;
39 if (transaction->use_count == 0) {
40 WARN_ON(total_trans == 0);
42 list_del_init(&transaction->list);
43 memset(transaction, 0, sizeof(*transaction));
44 kmem_cache_free(btrfs_transaction_cachep, transaction);
48 static noinline int join_transaction(struct btrfs_root *root)
50 struct btrfs_transaction *cur_trans;
51 cur_trans = root->fs_info->running_transaction;
53 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
57 root->fs_info->generation++;
58 root->fs_info->last_alloc = 0;
59 root->fs_info->last_data_alloc = 0;
60 cur_trans->num_writers = 1;
61 cur_trans->num_joined = 0;
62 cur_trans->transid = root->fs_info->generation;
63 init_waitqueue_head(&cur_trans->writer_wait);
64 init_waitqueue_head(&cur_trans->commit_wait);
65 cur_trans->in_commit = 0;
66 cur_trans->blocked = 0;
67 cur_trans->use_count = 1;
68 cur_trans->commit_done = 0;
69 cur_trans->start_time = get_seconds();
70 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
71 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
72 extent_io_tree_init(&cur_trans->dirty_pages,
73 root->fs_info->btree_inode->i_mapping,
75 spin_lock(&root->fs_info->new_trans_lock);
76 root->fs_info->running_transaction = cur_trans;
77 spin_unlock(&root->fs_info->new_trans_lock);
79 cur_trans->num_writers++;
80 cur_trans->num_joined++;
86 static noinline int record_root_in_trans(struct btrfs_root *root)
88 struct btrfs_dirty_root *dirty;
89 u64 running_trans_id = root->fs_info->running_transaction->transid;
90 if (root->ref_cows && root->last_trans < running_trans_id) {
91 WARN_ON(root == root->fs_info->extent_root);
92 if (root->root_item.refs != 0) {
93 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
94 (unsigned long)root->root_key.objectid,
95 BTRFS_ROOT_TRANS_TAG);
97 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
99 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
100 BUG_ON(!dirty->root);
101 dirty->latest_root = root;
102 INIT_LIST_HEAD(&dirty->list);
104 root->commit_root = btrfs_root_node(root);
106 memcpy(dirty->root, root, sizeof(*root));
107 spin_lock_init(&dirty->root->node_lock);
108 spin_lock_init(&dirty->root->list_lock);
109 mutex_init(&dirty->root->objectid_mutex);
110 INIT_LIST_HEAD(&dirty->root->dead_list);
111 dirty->root->node = root->commit_root;
112 dirty->root->commit_root = NULL;
114 spin_lock(&root->list_lock);
115 list_add(&dirty->root->dead_list, &root->dead_list);
116 spin_unlock(&root->list_lock);
118 root->dirty_root = dirty;
122 root->last_trans = running_trans_id;
127 static void wait_current_trans(struct btrfs_root *root)
129 struct btrfs_transaction *cur_trans;
131 cur_trans = root->fs_info->running_transaction;
132 if (cur_trans && cur_trans->blocked) {
134 cur_trans->use_count++;
136 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
137 TASK_UNINTERRUPTIBLE);
138 if (cur_trans->blocked) {
139 mutex_unlock(&root->fs_info->trans_mutex);
141 mutex_lock(&root->fs_info->trans_mutex);
142 finish_wait(&root->fs_info->transaction_wait,
145 finish_wait(&root->fs_info->transaction_wait,
150 put_transaction(cur_trans);
154 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
155 int num_blocks, int join)
157 struct btrfs_trans_handle *h =
158 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
161 mutex_lock(&root->fs_info->trans_mutex);
163 wait_current_trans(root);
164 ret = join_transaction(root);
167 record_root_in_trans(root);
168 h->transid = root->fs_info->running_transaction->transid;
169 h->transaction = root->fs_info->running_transaction;
170 h->blocks_reserved = num_blocks;
172 h->block_group = NULL;
173 h->alloc_exclude_nr = 0;
174 h->alloc_exclude_start = 0;
175 root->fs_info->running_transaction->use_count++;
176 mutex_unlock(&root->fs_info->trans_mutex);
180 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
183 return start_transaction(root, num_blocks, 0);
185 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
188 return start_transaction(root, num_blocks, 1);
191 static noinline int wait_for_commit(struct btrfs_root *root,
192 struct btrfs_transaction *commit)
195 mutex_lock(&root->fs_info->trans_mutex);
196 while(!commit->commit_done) {
197 prepare_to_wait(&commit->commit_wait, &wait,
198 TASK_UNINTERRUPTIBLE);
199 if (commit->commit_done)
201 mutex_unlock(&root->fs_info->trans_mutex);
203 mutex_lock(&root->fs_info->trans_mutex);
205 mutex_unlock(&root->fs_info->trans_mutex);
206 finish_wait(&commit->commit_wait, &wait);
210 static void throttle_on_drops(struct btrfs_root *root)
212 struct btrfs_fs_info *info = root->fs_info;
215 if (atomic_read(&info->throttles)) {
218 int harder_count = 0;
219 thr = atomic_read(&info->throttle_gen);
222 prepare_to_wait(&info->transaction_throttle,
223 &wait, TASK_UNINTERRUPTIBLE);
224 if (!atomic_read(&info->throttles)) {
225 finish_wait(&info->transaction_throttle, &wait);
229 finish_wait(&info->transaction_throttle, &wait);
230 } while (thr == atomic_read(&info->throttle_gen));
232 if (harder_count < 5 &&
233 info->total_ref_cache_size > 1 * 1024 * 1024) {
238 if (harder_count < 10 &&
239 info->total_ref_cache_size > 5 * 1024 * 1024) {
246 void btrfs_throttle(struct btrfs_root *root)
248 mutex_lock(&root->fs_info->trans_mutex);
249 wait_current_trans(root);
250 mutex_unlock(&root->fs_info->trans_mutex);
252 throttle_on_drops(root);
255 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
256 struct btrfs_root *root, int throttle)
258 struct btrfs_transaction *cur_trans;
259 struct btrfs_fs_info *info = root->fs_info;
261 mutex_lock(&info->trans_mutex);
262 cur_trans = info->running_transaction;
263 WARN_ON(cur_trans != trans->transaction);
264 WARN_ON(cur_trans->num_writers < 1);
265 cur_trans->num_writers--;
267 if (waitqueue_active(&cur_trans->writer_wait))
268 wake_up(&cur_trans->writer_wait);
269 put_transaction(cur_trans);
270 mutex_unlock(&info->trans_mutex);
271 memset(trans, 0, sizeof(*trans));
272 kmem_cache_free(btrfs_trans_handle_cachep, trans);
275 throttle_on_drops(root);
280 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
281 struct btrfs_root *root)
283 return __btrfs_end_transaction(trans, root, 0);
286 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
287 struct btrfs_root *root)
289 return __btrfs_end_transaction(trans, root, 1);
293 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
294 struct btrfs_root *root)
299 struct extent_io_tree *dirty_pages;
301 struct inode *btree_inode = root->fs_info->btree_inode;
306 if (!trans || !trans->transaction) {
307 return filemap_write_and_wait(btree_inode->i_mapping);
309 dirty_pages = &trans->transaction->dirty_pages;
311 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
315 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
316 while(start <= end) {
317 index = start >> PAGE_CACHE_SHIFT;
318 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
319 page = find_lock_page(btree_inode->i_mapping, index);
322 if (PageWriteback(page)) {
324 wait_on_page_writeback(page);
327 page_cache_release(page);
331 err = write_one_page(page, 0);
334 page_cache_release(page);
337 err = filemap_fdatawait(btree_inode->i_mapping);
343 static int update_cowonly_root(struct btrfs_trans_handle *trans,
344 struct btrfs_root *root)
348 struct btrfs_root *tree_root = root->fs_info->tree_root;
350 btrfs_write_dirty_block_groups(trans, root);
352 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
353 if (old_root_bytenr == root->node->start)
355 btrfs_set_root_bytenr(&root->root_item,
357 btrfs_set_root_level(&root->root_item,
358 btrfs_header_level(root->node));
359 ret = btrfs_update_root(trans, tree_root,
363 btrfs_write_dirty_block_groups(trans, root);
368 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
369 struct btrfs_root *root)
371 struct btrfs_fs_info *fs_info = root->fs_info;
372 struct list_head *next;
374 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
375 next = fs_info->dirty_cowonly_roots.next;
377 root = list_entry(next, struct btrfs_root, dirty_list);
378 update_cowonly_root(trans, root);
383 int btrfs_add_dead_root(struct btrfs_root *root,
384 struct btrfs_root *latest,
385 struct list_head *dead_list)
387 struct btrfs_dirty_root *dirty;
389 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
393 dirty->latest_root = latest;
394 list_add(&dirty->list, dead_list);
398 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
399 struct radix_tree_root *radix,
400 struct list_head *list)
402 struct btrfs_dirty_root *dirty;
403 struct btrfs_root *gang[8];
404 struct btrfs_root *root;
411 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
413 BTRFS_ROOT_TRANS_TAG);
416 for (i = 0; i < ret; i++) {
418 radix_tree_tag_clear(radix,
419 (unsigned long)root->root_key.objectid,
420 BTRFS_ROOT_TRANS_TAG);
422 BUG_ON(!root->ref_tree);
423 dirty = root->dirty_root;
425 if (root->commit_root == root->node) {
426 WARN_ON(root->node->start !=
427 btrfs_root_bytenr(&root->root_item));
429 free_extent_buffer(root->commit_root);
430 root->commit_root = NULL;
432 spin_lock(&root->list_lock);
433 list_del_init(&dirty->root->dead_list);
434 spin_unlock(&root->list_lock);
439 /* make sure to update the root on disk
440 * so we get any updates to the block used
443 err = btrfs_update_root(trans,
444 root->fs_info->tree_root,
450 memset(&root->root_item.drop_progress, 0,
451 sizeof(struct btrfs_disk_key));
452 root->root_item.drop_level = 0;
453 root->commit_root = NULL;
454 root->root_key.offset = root->fs_info->generation;
455 btrfs_set_root_bytenr(&root->root_item,
457 btrfs_set_root_level(&root->root_item,
458 btrfs_header_level(root->node));
459 err = btrfs_insert_root(trans, root->fs_info->tree_root,
465 refs = btrfs_root_refs(&dirty->root->root_item);
466 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
467 err = btrfs_update_root(trans, root->fs_info->tree_root,
468 &dirty->root->root_key,
469 &dirty->root->root_item);
473 list_add(&dirty->list, list);
476 free_extent_buffer(dirty->root->node);
485 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
487 struct btrfs_fs_info *info = root->fs_info;
489 struct btrfs_trans_handle *trans;
493 if (root->defrag_running)
495 trans = btrfs_start_transaction(root, 1);
497 root->defrag_running = 1;
498 ret = btrfs_defrag_leaves(trans, root, cacheonly);
499 nr = trans->blocks_used;
500 btrfs_end_transaction(trans, root);
501 btrfs_btree_balance_dirty(info->tree_root, nr);
504 trans = btrfs_start_transaction(root, 1);
505 if (root->fs_info->closing || ret != -EAGAIN)
508 root->defrag_running = 0;
510 btrfs_end_transaction(trans, root);
514 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
515 struct list_head *list)
517 struct btrfs_dirty_root *dirty;
518 struct btrfs_trans_handle *trans;
526 while(!list_empty(list)) {
527 struct btrfs_root *root;
529 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
530 list_del_init(&dirty->list);
532 num_bytes = btrfs_root_used(&dirty->root->root_item);
533 root = dirty->latest_root;
534 atomic_inc(&root->fs_info->throttles);
536 mutex_lock(&root->fs_info->drop_mutex);
538 trans = btrfs_start_transaction(tree_root, 1);
539 ret = btrfs_drop_snapshot(trans, dirty->root);
540 if (ret != -EAGAIN) {
544 err = btrfs_update_root(trans,
546 &dirty->root->root_key,
547 &dirty->root->root_item);
550 nr = trans->blocks_used;
551 ret = btrfs_end_transaction(trans, tree_root);
554 mutex_unlock(&root->fs_info->drop_mutex);
555 btrfs_btree_balance_dirty(tree_root, nr);
557 mutex_lock(&root->fs_info->drop_mutex);
560 atomic_dec(&root->fs_info->throttles);
561 wake_up(&root->fs_info->transaction_throttle);
563 mutex_lock(&root->fs_info->alloc_mutex);
564 num_bytes -= btrfs_root_used(&dirty->root->root_item);
565 bytes_used = btrfs_root_used(&root->root_item);
567 record_root_in_trans(root);
568 btrfs_set_root_used(&root->root_item,
569 bytes_used - num_bytes);
571 mutex_unlock(&root->fs_info->alloc_mutex);
573 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
578 mutex_unlock(&root->fs_info->drop_mutex);
580 spin_lock(&root->list_lock);
581 list_del_init(&dirty->root->dead_list);
582 if (!list_empty(&root->dead_list)) {
583 struct btrfs_root *oldest;
584 oldest = list_entry(root->dead_list.prev,
585 struct btrfs_root, dead_list);
586 max_useless = oldest->root_key.offset - 1;
588 max_useless = root->root_key.offset - 1;
590 spin_unlock(&root->list_lock);
592 nr = trans->blocks_used;
593 ret = btrfs_end_transaction(trans, tree_root);
596 ret = btrfs_remove_leaf_refs(root, max_useless);
599 free_extent_buffer(dirty->root->node);
603 btrfs_btree_balance_dirty(tree_root, nr);
609 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
610 struct btrfs_fs_info *fs_info,
611 struct btrfs_pending_snapshot *pending)
613 struct btrfs_key key;
614 struct btrfs_root_item *new_root_item;
615 struct btrfs_root *tree_root = fs_info->tree_root;
616 struct btrfs_root *root = pending->root;
617 struct extent_buffer *tmp;
618 struct extent_buffer *old;
623 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
624 if (!new_root_item) {
628 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
632 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
634 key.objectid = objectid;
636 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
638 old = btrfs_lock_root_node(root);
639 btrfs_cow_block(trans, root, old, NULL, 0, &old);
641 btrfs_copy_root(trans, root, old, &tmp, objectid);
642 btrfs_tree_unlock(old);
643 free_extent_buffer(old);
645 btrfs_set_root_bytenr(new_root_item, tmp->start);
646 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
647 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
649 btrfs_tree_unlock(tmp);
650 free_extent_buffer(tmp);
655 * insert the directory item
657 key.offset = (u64)-1;
658 namelen = strlen(pending->name);
659 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
660 pending->name, namelen,
661 root->fs_info->sb->s_root->d_inode->i_ino,
662 &key, BTRFS_FT_DIR, 0);
667 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
668 pending->name, strlen(pending->name), objectid,
669 root->fs_info->sb->s_root->d_inode->i_ino, 0);
671 /* Invalidate existing dcache entry for new snapshot. */
672 btrfs_invalidate_dcache_root(root, pending->name, namelen);
675 kfree(new_root_item);
679 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
680 struct btrfs_fs_info *fs_info)
682 struct btrfs_pending_snapshot *pending;
683 struct list_head *head = &trans->transaction->pending_snapshots;
686 while(!list_empty(head)) {
687 pending = list_entry(head->next,
688 struct btrfs_pending_snapshot, list);
689 ret = create_pending_snapshot(trans, fs_info, pending);
691 list_del(&pending->list);
692 kfree(pending->name);
698 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
699 struct btrfs_root *root)
701 unsigned long joined = 0;
702 unsigned long timeout = 1;
703 struct btrfs_transaction *cur_trans;
704 struct btrfs_transaction *prev_trans = NULL;
705 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
706 struct list_head dirty_fs_roots;
707 struct extent_io_tree *pinned_copy;
711 INIT_LIST_HEAD(&dirty_fs_roots);
713 mutex_lock(&root->fs_info->trans_mutex);
714 if (trans->transaction->in_commit) {
715 cur_trans = trans->transaction;
716 trans->transaction->use_count++;
717 mutex_unlock(&root->fs_info->trans_mutex);
718 btrfs_end_transaction(trans, root);
720 ret = wait_for_commit(root, cur_trans);
723 mutex_lock(&root->fs_info->trans_mutex);
724 put_transaction(cur_trans);
725 mutex_unlock(&root->fs_info->trans_mutex);
730 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
734 extent_io_tree_init(pinned_copy,
735 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
737 trans->transaction->in_commit = 1;
738 trans->transaction->blocked = 1;
739 cur_trans = trans->transaction;
740 if (cur_trans->list.prev != &root->fs_info->trans_list) {
741 prev_trans = list_entry(cur_trans->list.prev,
742 struct btrfs_transaction, list);
743 if (!prev_trans->commit_done) {
744 prev_trans->use_count++;
745 mutex_unlock(&root->fs_info->trans_mutex);
747 wait_for_commit(root, prev_trans);
749 mutex_lock(&root->fs_info->trans_mutex);
750 put_transaction(prev_trans);
755 joined = cur_trans->num_joined;
756 WARN_ON(cur_trans != trans->transaction);
757 prepare_to_wait(&cur_trans->writer_wait, &wait,
758 TASK_UNINTERRUPTIBLE);
760 if (cur_trans->num_writers > 1)
761 timeout = MAX_SCHEDULE_TIMEOUT;
765 mutex_unlock(&root->fs_info->trans_mutex);
767 schedule_timeout(timeout);
769 mutex_lock(&root->fs_info->trans_mutex);
770 finish_wait(&cur_trans->writer_wait, &wait);
771 } while (cur_trans->num_writers > 1 ||
772 (cur_trans->num_joined != joined));
774 ret = create_pending_snapshots(trans, root->fs_info);
777 WARN_ON(cur_trans != trans->transaction);
779 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
783 ret = btrfs_commit_tree_roots(trans, root);
786 cur_trans = root->fs_info->running_transaction;
787 spin_lock(&root->fs_info->new_trans_lock);
788 root->fs_info->running_transaction = NULL;
789 spin_unlock(&root->fs_info->new_trans_lock);
790 btrfs_set_super_generation(&root->fs_info->super_copy,
792 btrfs_set_super_root(&root->fs_info->super_copy,
793 root->fs_info->tree_root->node->start);
794 btrfs_set_super_root_level(&root->fs_info->super_copy,
795 btrfs_header_level(root->fs_info->tree_root->node));
797 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
798 chunk_root->node->start);
799 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
800 btrfs_header_level(chunk_root->node));
801 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
802 sizeof(root->fs_info->super_copy));
804 btrfs_copy_pinned(root, pinned_copy);
806 trans->transaction->blocked = 0;
807 wake_up(&root->fs_info->transaction_throttle);
808 wake_up(&root->fs_info->transaction_wait);
810 mutex_unlock(&root->fs_info->trans_mutex);
811 ret = btrfs_write_and_wait_transaction(trans, root);
813 write_ctree_super(trans, root);
815 btrfs_finish_extent_commit(trans, root, pinned_copy);
816 mutex_lock(&root->fs_info->trans_mutex);
820 cur_trans->commit_done = 1;
821 root->fs_info->last_trans_committed = cur_trans->transid;
822 wake_up(&cur_trans->commit_wait);
823 put_transaction(cur_trans);
824 put_transaction(cur_trans);
826 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
827 if (root->fs_info->closing)
828 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
830 mutex_unlock(&root->fs_info->trans_mutex);
831 kmem_cache_free(btrfs_trans_handle_cachep, trans);
833 if (root->fs_info->closing) {
834 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
839 int btrfs_clean_old_snapshots(struct btrfs_root *root)
841 struct list_head dirty_roots;
842 INIT_LIST_HEAD(&dirty_roots);
844 mutex_lock(&root->fs_info->trans_mutex);
845 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
846 mutex_unlock(&root->fs_info->trans_mutex);
848 if (!list_empty(&dirty_roots)) {
849 drop_dirty_roots(root, &dirty_roots);