nfs41: implement DESTROY_CLIENTID operation
[linux-btrfs-devel.git] / fs / btrfs / transaction.c
blob7dc36fab4afc5b9df1c4655e455c8df8bd3ee39b
1 /*
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.
19 #include <linux/fs.h>
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>
25 #include "ctree.h"
26 #include "disk-io.h"
27 #include "transaction.h"
28 #include "locking.h"
29 #include "tree-log.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) {
59 if (!nofail) {
60 spin_unlock(&root->fs_info->trans_lock);
61 return -EBUSY;
65 cur_trans = root->fs_info->running_transaction;
66 if (cur_trans) {
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);
71 return 0;
73 spin_unlock(&root->fs_info->trans_lock);
75 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
76 if (!cur_trans)
77 return -ENOMEM;
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);
86 return 0;
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);
120 return 0;
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
146 smp_wmb();
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);
151 return 0;
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
162 * this transaction.
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);
179 smp_wmb();
180 root->in_trans_setup = 0;
182 return 0;
186 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
187 struct btrfs_root *root)
189 if (!root->ref_cows)
190 return 0;
193 * see record_root_in_trans for comments about in_trans_setup usage
194 * and barriers
196 smp_rmb();
197 if (root->last_trans == trans->transid &&
198 !root->in_trans_setup)
199 return 0;
201 mutex_lock(&root->fs_info->reloc_mutex);
202 record_root_in_trans(trans, root);
203 mutex_unlock(&root->fs_info->reloc_mutex);
205 return 0;
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);
225 } else {
226 spin_unlock(&root->fs_info->trans_lock);
230 enum btrfs_trans_type {
231 TRANS_START,
232 TRANS_JOIN,
233 TRANS_USERSPACE,
234 TRANS_JOIN_NOLOCK,
237 static int may_wait_transaction(struct btrfs_root *root, int type)
239 if (root->fs_info->log_root_recovering)
240 return 0;
242 if (type == TRANS_USERSPACE)
243 return 1;
245 if (type == TRANS_START &&
246 !atomic_read(&root->fs_info->open_ioctl_trans))
247 return 1;
249 return 0;
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;
257 u64 num_bytes = 0;
258 int ret;
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;
266 h->use_count++;
267 h->orig_rsv = h->block_rsv;
268 h->block_rsv = NULL;
269 goto got_it;
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,
280 num_bytes);
281 if (ret)
282 return ERR_PTR(ret);
284 again:
285 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
286 if (!h)
287 return ERR_PTR(-ENOMEM);
289 if (may_wait_transaction(root, type))
290 wait_current_trans(root);
292 do {
293 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
294 if (ret == -EBUSY)
295 wait_current_trans(root);
296 } while (ret == -EBUSY);
298 if (ret < 0) {
299 kmem_cache_free(btrfs_trans_handle_cachep, h);
300 return ERR_PTR(ret);
303 cur_trans = root->fs_info->running_transaction;
305 h->transid = cur_trans->transid;
306 h->transaction = cur_trans;
307 h->blocks_used = 0;
308 h->bytes_reserved = 0;
309 h->delayed_ref_updates = 0;
310 h->use_count = 1;
311 h->block_rsv = NULL;
312 h->orig_rsv = NULL;
314 smp_mb();
315 if (cur_trans->blocked && may_wait_transaction(root, type)) {
316 btrfs_commit_transaction(h, root);
317 goto again;
320 if (num_bytes) {
321 h->block_rsv = &root->fs_info->trans_block_rsv;
322 h->bytes_reserved = num_bytes;
325 got_it:
326 btrfs_record_root_in_trans(h, root);
328 if (!current->journal_info && type != TRANS_USERSPACE)
329 current->journal_info = h;
330 return h;
333 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
334 int num_items)
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;
363 int ret;
365 ret = 0;
366 if (transid) {
367 if (transid <= root->fs_info->last_trans_committed)
368 goto out;
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) {
374 cur_trans = t;
375 atomic_inc(&cur_trans->use_count);
376 break;
378 if (t->transid > transid)
379 break;
381 spin_unlock(&root->fs_info->trans_lock);
382 ret = -EINVAL;
383 if (!cur_trans)
384 goto out; /* bad transid */
385 } else {
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,
389 list) {
390 if (t->in_commit) {
391 if (t->commit_done)
392 break;
393 cur_trans = t;
394 atomic_inc(&cur_trans->use_count);
395 break;
398 spin_unlock(&root->fs_info->trans_lock);
399 if (!cur_trans)
400 goto out; /* nothing committing|committed */
403 wait_for_commit(root, cur_trans);
405 put_transaction(cur_trans);
406 ret = 0;
407 out:
408 return ret;
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)
420 int ret;
421 ret = btrfs_block_rsv_check(trans, root,
422 &root->fs_info->global_block_rsv, 0, 5);
423 return ret ? 1 : 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;
430 int updates;
432 smp_mb();
433 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
434 return 1;
436 updates = trans->delayed_ref_updates;
437 trans->delayed_ref_updates = 0;
438 if (updates)
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;
449 int count = 0;
451 if (--trans->use_count) {
452 trans->block_rsv = trans->orig_rsv;
453 return 0;
456 while (count < 4) {
457 unsigned long cur = trans->delayed_ref_updates;
458 trans->delayed_ref_updates = 0;
459 if (cur &&
460 trans->transaction->delayed_refs.num_heads_ready > 64) {
461 trans->delayed_ref_updates = 0;
464 * do a full flush if the transaction is trying
465 * to close
467 if (trans->transaction->delayed_refs.flushing)
468 cur = 0;
469 btrfs_run_delayed_refs(trans, root, cur);
470 } else {
471 break;
473 count++;
476 btrfs_trans_release_metadata(trans, root);
478 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
479 should_end_transaction(trans, root)) {
480 trans->transaction->blocked = 1;
481 smp_wmb();
484 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
485 if (throttle) {
487 * We may race with somebody else here so end up having
488 * to call end_transaction on ourselves again, so inc
489 * our use_count.
491 trans->use_count++;
492 return btrfs_commit_transaction(trans, root);
493 } else {
494 wake_up_process(info->transaction_kthread);
498 WARN_ON(cur_trans != info->running_transaction);
499 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
500 atomic_dec(&cur_trans->num_writers);
502 smp_mb();
503 if (waitqueue_active(&cur_trans->writer_wait))
504 wake_up(&cur_trans->writer_wait);
505 put_transaction(cur_trans);
507 if (current->journal_info == trans)
508 current->journal_info = NULL;
509 memset(trans, 0, sizeof(*trans));
510 kmem_cache_free(btrfs_trans_handle_cachep, trans);
512 if (throttle)
513 btrfs_run_delayed_iputs(root);
515 return 0;
518 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
519 struct btrfs_root *root)
521 int ret;
523 ret = __btrfs_end_transaction(trans, root, 0, 1);
524 if (ret)
525 return ret;
526 return 0;
529 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
530 struct btrfs_root *root)
532 int ret;
534 ret = __btrfs_end_transaction(trans, root, 1, 1);
535 if (ret)
536 return ret;
537 return 0;
540 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
541 struct btrfs_root *root)
543 int ret;
545 ret = __btrfs_end_transaction(trans, root, 0, 0);
546 if (ret)
547 return ret;
548 return 0;
551 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
552 struct btrfs_root *root)
554 return __btrfs_end_transaction(trans, root, 1, 1);
558 * when btree blocks are allocated, they have some corresponding bits set for
559 * them in one of two extent_io trees. This is used to make sure all of
560 * those extents are sent to disk but does not wait on them
562 int btrfs_write_marked_extents(struct btrfs_root *root,
563 struct extent_io_tree *dirty_pages, int mark)
565 int ret;
566 int err = 0;
567 int werr = 0;
568 struct page *page;
569 struct inode *btree_inode = root->fs_info->btree_inode;
570 u64 start = 0;
571 u64 end;
572 unsigned long index;
574 while (1) {
575 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
576 mark);
577 if (ret)
578 break;
579 while (start <= end) {
580 cond_resched();
582 index = start >> PAGE_CACHE_SHIFT;
583 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
584 page = find_get_page(btree_inode->i_mapping, index);
585 if (!page)
586 continue;
588 btree_lock_page_hook(page);
589 if (!page->mapping) {
590 unlock_page(page);
591 page_cache_release(page);
592 continue;
595 if (PageWriteback(page)) {
596 if (PageDirty(page))
597 wait_on_page_writeback(page);
598 else {
599 unlock_page(page);
600 page_cache_release(page);
601 continue;
604 err = write_one_page(page, 0);
605 if (err)
606 werr = err;
607 page_cache_release(page);
610 if (err)
611 werr = err;
612 return werr;
616 * when btree blocks are allocated, they have some corresponding bits set for
617 * them in one of two extent_io trees. This is used to make sure all of
618 * those extents are on disk for transaction or log commit. We wait
619 * on all the pages and clear them from the dirty pages state tree
621 int btrfs_wait_marked_extents(struct btrfs_root *root,
622 struct extent_io_tree *dirty_pages, int mark)
624 int ret;
625 int err = 0;
626 int werr = 0;
627 struct page *page;
628 struct inode *btree_inode = root->fs_info->btree_inode;
629 u64 start = 0;
630 u64 end;
631 unsigned long index;
633 while (1) {
634 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
635 mark);
636 if (ret)
637 break;
639 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
640 while (start <= end) {
641 index = start >> PAGE_CACHE_SHIFT;
642 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
643 page = find_get_page(btree_inode->i_mapping, index);
644 if (!page)
645 continue;
646 if (PageDirty(page)) {
647 btree_lock_page_hook(page);
648 wait_on_page_writeback(page);
649 err = write_one_page(page, 0);
650 if (err)
651 werr = err;
653 wait_on_page_writeback(page);
654 page_cache_release(page);
655 cond_resched();
658 if (err)
659 werr = err;
660 return werr;
664 * when btree blocks are allocated, they have some corresponding bits set for
665 * them in one of two extent_io trees. This is used to make sure all of
666 * those extents are on disk for transaction or log commit
668 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
669 struct extent_io_tree *dirty_pages, int mark)
671 int ret;
672 int ret2;
674 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
675 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
676 return ret || ret2;
679 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
680 struct btrfs_root *root)
682 if (!trans || !trans->transaction) {
683 struct inode *btree_inode;
684 btree_inode = root->fs_info->btree_inode;
685 return filemap_write_and_wait(btree_inode->i_mapping);
687 return btrfs_write_and_wait_marked_extents(root,
688 &trans->transaction->dirty_pages,
689 EXTENT_DIRTY);
693 * this is used to update the root pointer in the tree of tree roots.
695 * But, in the case of the extent allocation tree, updating the root
696 * pointer may allocate blocks which may change the root of the extent
697 * allocation tree.
699 * So, this loops and repeats and makes sure the cowonly root didn't
700 * change while the root pointer was being updated in the metadata.
702 static int update_cowonly_root(struct btrfs_trans_handle *trans,
703 struct btrfs_root *root)
705 int ret;
706 u64 old_root_bytenr;
707 u64 old_root_used;
708 struct btrfs_root *tree_root = root->fs_info->tree_root;
710 old_root_used = btrfs_root_used(&root->root_item);
711 btrfs_write_dirty_block_groups(trans, root);
713 while (1) {
714 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
715 if (old_root_bytenr == root->node->start &&
716 old_root_used == btrfs_root_used(&root->root_item))
717 break;
719 btrfs_set_root_node(&root->root_item, root->node);
720 ret = btrfs_update_root(trans, tree_root,
721 &root->root_key,
722 &root->root_item);
723 BUG_ON(ret);
725 old_root_used = btrfs_root_used(&root->root_item);
726 ret = btrfs_write_dirty_block_groups(trans, root);
727 BUG_ON(ret);
730 if (root != root->fs_info->extent_root)
731 switch_commit_root(root);
733 return 0;
737 * update all the cowonly tree roots on disk
739 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
740 struct btrfs_root *root)
742 struct btrfs_fs_info *fs_info = root->fs_info;
743 struct list_head *next;
744 struct extent_buffer *eb;
745 int ret;
747 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
748 BUG_ON(ret);
750 eb = btrfs_lock_root_node(fs_info->tree_root);
751 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
752 btrfs_tree_unlock(eb);
753 free_extent_buffer(eb);
755 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
756 BUG_ON(ret);
758 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
759 next = fs_info->dirty_cowonly_roots.next;
760 list_del_init(next);
761 root = list_entry(next, struct btrfs_root, dirty_list);
763 update_cowonly_root(trans, root);
766 down_write(&fs_info->extent_commit_sem);
767 switch_commit_root(fs_info->extent_root);
768 up_write(&fs_info->extent_commit_sem);
770 return 0;
774 * dead roots are old snapshots that need to be deleted. This allocates
775 * a dirty root struct and adds it into the list of dead roots that need to
776 * be deleted
778 int btrfs_add_dead_root(struct btrfs_root *root)
780 spin_lock(&root->fs_info->trans_lock);
781 list_add(&root->root_list, &root->fs_info->dead_roots);
782 spin_unlock(&root->fs_info->trans_lock);
783 return 0;
787 * update all the cowonly tree roots on disk
789 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
790 struct btrfs_root *root)
792 struct btrfs_root *gang[8];
793 struct btrfs_fs_info *fs_info = root->fs_info;
794 int i;
795 int ret;
796 int err = 0;
798 spin_lock(&fs_info->fs_roots_radix_lock);
799 while (1) {
800 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
801 (void **)gang, 0,
802 ARRAY_SIZE(gang),
803 BTRFS_ROOT_TRANS_TAG);
804 if (ret == 0)
805 break;
806 for (i = 0; i < ret; i++) {
807 root = gang[i];
808 radix_tree_tag_clear(&fs_info->fs_roots_radix,
809 (unsigned long)root->root_key.objectid,
810 BTRFS_ROOT_TRANS_TAG);
811 spin_unlock(&fs_info->fs_roots_radix_lock);
813 btrfs_free_log(trans, root);
814 btrfs_update_reloc_root(trans, root);
815 btrfs_orphan_commit_root(trans, root);
817 btrfs_save_ino_cache(root, trans);
819 if (root->commit_root != root->node) {
820 mutex_lock(&root->fs_commit_mutex);
821 switch_commit_root(root);
822 btrfs_unpin_free_ino(root);
823 mutex_unlock(&root->fs_commit_mutex);
825 btrfs_set_root_node(&root->root_item,
826 root->node);
829 err = btrfs_update_root(trans, fs_info->tree_root,
830 &root->root_key,
831 &root->root_item);
832 spin_lock(&fs_info->fs_roots_radix_lock);
833 if (err)
834 break;
837 spin_unlock(&fs_info->fs_roots_radix_lock);
838 return err;
842 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
843 * otherwise every leaf in the btree is read and defragged.
845 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
847 struct btrfs_fs_info *info = root->fs_info;
848 struct btrfs_trans_handle *trans;
849 int ret;
850 unsigned long nr;
852 if (xchg(&root->defrag_running, 1))
853 return 0;
855 while (1) {
856 trans = btrfs_start_transaction(root, 0);
857 if (IS_ERR(trans))
858 return PTR_ERR(trans);
860 ret = btrfs_defrag_leaves(trans, root, cacheonly);
862 nr = trans->blocks_used;
863 btrfs_end_transaction(trans, root);
864 btrfs_btree_balance_dirty(info->tree_root, nr);
865 cond_resched();
867 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
868 break;
870 root->defrag_running = 0;
871 return ret;
875 * new snapshots need to be created at a very specific time in the
876 * transaction commit. This does the actual creation
878 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
879 struct btrfs_fs_info *fs_info,
880 struct btrfs_pending_snapshot *pending)
882 struct btrfs_key key;
883 struct btrfs_root_item *new_root_item;
884 struct btrfs_root *tree_root = fs_info->tree_root;
885 struct btrfs_root *root = pending->root;
886 struct btrfs_root *parent_root;
887 struct inode *parent_inode;
888 struct dentry *parent;
889 struct dentry *dentry;
890 struct extent_buffer *tmp;
891 struct extent_buffer *old;
892 int ret;
893 u64 to_reserve = 0;
894 u64 index = 0;
895 u64 objectid;
896 u64 root_flags;
898 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
899 if (!new_root_item) {
900 pending->error = -ENOMEM;
901 goto fail;
904 ret = btrfs_find_free_objectid(tree_root, &objectid);
905 if (ret) {
906 pending->error = ret;
907 goto fail;
910 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
911 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
913 if (to_reserve > 0) {
914 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
915 to_reserve);
916 if (ret) {
917 pending->error = ret;
918 goto fail;
922 key.objectid = objectid;
923 key.offset = (u64)-1;
924 key.type = BTRFS_ROOT_ITEM_KEY;
926 trans->block_rsv = &pending->block_rsv;
928 dentry = pending->dentry;
929 parent = dget_parent(dentry);
930 parent_inode = parent->d_inode;
931 parent_root = BTRFS_I(parent_inode)->root;
932 record_root_in_trans(trans, parent_root);
935 * insert the directory item
937 ret = btrfs_set_inode_index(parent_inode, &index);
938 BUG_ON(ret);
939 ret = btrfs_insert_dir_item(trans, parent_root,
940 dentry->d_name.name, dentry->d_name.len,
941 parent_inode, &key,
942 BTRFS_FT_DIR, index);
943 BUG_ON(ret);
945 btrfs_i_size_write(parent_inode, parent_inode->i_size +
946 dentry->d_name.len * 2);
947 ret = btrfs_update_inode(trans, parent_root, parent_inode);
948 BUG_ON(ret);
951 * pull in the delayed directory update
952 * and the delayed inode item
953 * otherwise we corrupt the FS during
954 * snapshot
956 ret = btrfs_run_delayed_items(trans, root);
957 BUG_ON(ret);
959 record_root_in_trans(trans, root);
960 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
961 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
962 btrfs_check_and_init_root_item(new_root_item);
964 root_flags = btrfs_root_flags(new_root_item);
965 if (pending->readonly)
966 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
967 else
968 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
969 btrfs_set_root_flags(new_root_item, root_flags);
971 old = btrfs_lock_root_node(root);
972 btrfs_cow_block(trans, root, old, NULL, 0, &old);
973 btrfs_set_lock_blocking(old);
975 btrfs_copy_root(trans, root, old, &tmp, objectid);
976 btrfs_tree_unlock(old);
977 free_extent_buffer(old);
979 btrfs_set_root_node(new_root_item, tmp);
980 /* record when the snapshot was created in key.offset */
981 key.offset = trans->transid;
982 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
983 btrfs_tree_unlock(tmp);
984 free_extent_buffer(tmp);
985 BUG_ON(ret);
988 * insert root back/forward references
990 ret = btrfs_add_root_ref(trans, tree_root, objectid,
991 parent_root->root_key.objectid,
992 btrfs_ino(parent_inode), index,
993 dentry->d_name.name, dentry->d_name.len);
994 BUG_ON(ret);
995 dput(parent);
997 key.offset = (u64)-1;
998 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
999 BUG_ON(IS_ERR(pending->snap));
1001 btrfs_reloc_post_snapshot(trans, pending);
1002 btrfs_orphan_post_snapshot(trans, pending);
1003 fail:
1004 kfree(new_root_item);
1005 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1006 return 0;
1010 * create all the snapshots we've scheduled for creation
1012 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1013 struct btrfs_fs_info *fs_info)
1015 struct btrfs_pending_snapshot *pending;
1016 struct list_head *head = &trans->transaction->pending_snapshots;
1017 int ret;
1019 list_for_each_entry(pending, head, list) {
1020 ret = create_pending_snapshot(trans, fs_info, pending);
1021 BUG_ON(ret);
1023 return 0;
1026 static void update_super_roots(struct btrfs_root *root)
1028 struct btrfs_root_item *root_item;
1029 struct btrfs_super_block *super;
1031 super = &root->fs_info->super_copy;
1033 root_item = &root->fs_info->chunk_root->root_item;
1034 super->chunk_root = root_item->bytenr;
1035 super->chunk_root_generation = root_item->generation;
1036 super->chunk_root_level = root_item->level;
1038 root_item = &root->fs_info->tree_root->root_item;
1039 super->root = root_item->bytenr;
1040 super->generation = root_item->generation;
1041 super->root_level = root_item->level;
1042 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1043 super->cache_generation = root_item->generation;
1046 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1048 int ret = 0;
1049 spin_lock(&info->trans_lock);
1050 if (info->running_transaction)
1051 ret = info->running_transaction->in_commit;
1052 spin_unlock(&info->trans_lock);
1053 return ret;
1056 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1058 int ret = 0;
1059 spin_lock(&info->trans_lock);
1060 if (info->running_transaction)
1061 ret = info->running_transaction->blocked;
1062 spin_unlock(&info->trans_lock);
1063 return ret;
1067 * wait for the current transaction commit to start and block subsequent
1068 * transaction joins
1070 static void wait_current_trans_commit_start(struct btrfs_root *root,
1071 struct btrfs_transaction *trans)
1073 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1077 * wait for the current transaction to start and then become unblocked.
1078 * caller holds ref.
1080 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1081 struct btrfs_transaction *trans)
1083 wait_event(root->fs_info->transaction_wait,
1084 trans->commit_done || (trans->in_commit && !trans->blocked));
1088 * commit transactions asynchronously. once btrfs_commit_transaction_async
1089 * returns, any subsequent transaction will not be allowed to join.
1091 struct btrfs_async_commit {
1092 struct btrfs_trans_handle *newtrans;
1093 struct btrfs_root *root;
1094 struct delayed_work work;
1097 static void do_async_commit(struct work_struct *work)
1099 struct btrfs_async_commit *ac =
1100 container_of(work, struct btrfs_async_commit, work.work);
1102 btrfs_commit_transaction(ac->newtrans, ac->root);
1103 kfree(ac);
1106 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1107 struct btrfs_root *root,
1108 int wait_for_unblock)
1110 struct btrfs_async_commit *ac;
1111 struct btrfs_transaction *cur_trans;
1113 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1114 if (!ac)
1115 return -ENOMEM;
1117 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1118 ac->root = root;
1119 ac->newtrans = btrfs_join_transaction(root);
1120 if (IS_ERR(ac->newtrans)) {
1121 int err = PTR_ERR(ac->newtrans);
1122 kfree(ac);
1123 return err;
1126 /* take transaction reference */
1127 cur_trans = trans->transaction;
1128 atomic_inc(&cur_trans->use_count);
1130 btrfs_end_transaction(trans, root);
1131 schedule_delayed_work(&ac->work, 0);
1133 /* wait for transaction to start and unblock */
1134 if (wait_for_unblock)
1135 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1136 else
1137 wait_current_trans_commit_start(root, cur_trans);
1139 if (current->journal_info == trans)
1140 current->journal_info = NULL;
1142 put_transaction(cur_trans);
1143 return 0;
1147 * btrfs_transaction state sequence:
1148 * in_commit = 0, blocked = 0 (initial)
1149 * in_commit = 1, blocked = 1
1150 * blocked = 0
1151 * commit_done = 1
1153 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1154 struct btrfs_root *root)
1156 unsigned long joined = 0;
1157 struct btrfs_transaction *cur_trans;
1158 struct btrfs_transaction *prev_trans = NULL;
1159 DEFINE_WAIT(wait);
1160 int ret;
1161 int should_grow = 0;
1162 unsigned long now = get_seconds();
1163 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1165 btrfs_run_ordered_operations(root, 0);
1167 /* make a pass through all the delayed refs we have so far
1168 * any runnings procs may add more while we are here
1170 ret = btrfs_run_delayed_refs(trans, root, 0);
1171 BUG_ON(ret);
1173 btrfs_trans_release_metadata(trans, root);
1175 cur_trans = trans->transaction;
1177 * set the flushing flag so procs in this transaction have to
1178 * start sending their work down.
1180 cur_trans->delayed_refs.flushing = 1;
1182 ret = btrfs_run_delayed_refs(trans, root, 0);
1183 BUG_ON(ret);
1185 spin_lock(&cur_trans->commit_lock);
1186 if (cur_trans->in_commit) {
1187 spin_unlock(&cur_trans->commit_lock);
1188 atomic_inc(&cur_trans->use_count);
1189 btrfs_end_transaction(trans, root);
1191 wait_for_commit(root, cur_trans);
1193 put_transaction(cur_trans);
1195 return 0;
1198 trans->transaction->in_commit = 1;
1199 trans->transaction->blocked = 1;
1200 spin_unlock(&cur_trans->commit_lock);
1201 wake_up(&root->fs_info->transaction_blocked_wait);
1203 spin_lock(&root->fs_info->trans_lock);
1204 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1205 prev_trans = list_entry(cur_trans->list.prev,
1206 struct btrfs_transaction, list);
1207 if (!prev_trans->commit_done) {
1208 atomic_inc(&prev_trans->use_count);
1209 spin_unlock(&root->fs_info->trans_lock);
1211 wait_for_commit(root, prev_trans);
1213 put_transaction(prev_trans);
1214 } else {
1215 spin_unlock(&root->fs_info->trans_lock);
1217 } else {
1218 spin_unlock(&root->fs_info->trans_lock);
1221 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1222 should_grow = 1;
1224 do {
1225 int snap_pending = 0;
1227 joined = cur_trans->num_joined;
1228 if (!list_empty(&trans->transaction->pending_snapshots))
1229 snap_pending = 1;
1231 WARN_ON(cur_trans != trans->transaction);
1233 if (flush_on_commit || snap_pending) {
1234 btrfs_start_delalloc_inodes(root, 1);
1235 ret = btrfs_wait_ordered_extents(root, 0, 1);
1236 BUG_ON(ret);
1239 ret = btrfs_run_delayed_items(trans, root);
1240 BUG_ON(ret);
1243 * rename don't use btrfs_join_transaction, so, once we
1244 * set the transaction to blocked above, we aren't going
1245 * to get any new ordered operations. We can safely run
1246 * it here and no for sure that nothing new will be added
1247 * to the list
1249 btrfs_run_ordered_operations(root, 1);
1251 prepare_to_wait(&cur_trans->writer_wait, &wait,
1252 TASK_UNINTERRUPTIBLE);
1254 if (atomic_read(&cur_trans->num_writers) > 1)
1255 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1256 else if (should_grow)
1257 schedule_timeout(1);
1259 finish_wait(&cur_trans->writer_wait, &wait);
1260 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1261 (should_grow && cur_trans->num_joined != joined));
1264 * Ok now we need to make sure to block out any other joins while we
1265 * commit the transaction. We could have started a join before setting
1266 * no_join so make sure to wait for num_writers to == 1 again.
1268 spin_lock(&root->fs_info->trans_lock);
1269 root->fs_info->trans_no_join = 1;
1270 spin_unlock(&root->fs_info->trans_lock);
1271 wait_event(cur_trans->writer_wait,
1272 atomic_read(&cur_trans->num_writers) == 1);
1275 * the reloc mutex makes sure that we stop
1276 * the balancing code from coming in and moving
1277 * extents around in the middle of the commit
1279 mutex_lock(&root->fs_info->reloc_mutex);
1281 ret = btrfs_run_delayed_items(trans, root);
1282 BUG_ON(ret);
1284 ret = create_pending_snapshots(trans, root->fs_info);
1285 BUG_ON(ret);
1287 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1288 BUG_ON(ret);
1291 * make sure none of the code above managed to slip in a
1292 * delayed item
1294 btrfs_assert_delayed_root_empty(root);
1296 WARN_ON(cur_trans != trans->transaction);
1298 btrfs_scrub_pause(root);
1299 /* btrfs_commit_tree_roots is responsible for getting the
1300 * various roots consistent with each other. Every pointer
1301 * in the tree of tree roots has to point to the most up to date
1302 * root for every subvolume and other tree. So, we have to keep
1303 * the tree logging code from jumping in and changing any
1304 * of the trees.
1306 * At this point in the commit, there can't be any tree-log
1307 * writers, but a little lower down we drop the trans mutex
1308 * and let new people in. By holding the tree_log_mutex
1309 * from now until after the super is written, we avoid races
1310 * with the tree-log code.
1312 mutex_lock(&root->fs_info->tree_log_mutex);
1314 ret = commit_fs_roots(trans, root);
1315 BUG_ON(ret);
1317 /* commit_fs_roots gets rid of all the tree log roots, it is now
1318 * safe to free the root of tree log roots
1320 btrfs_free_log_root_tree(trans, root->fs_info);
1322 ret = commit_cowonly_roots(trans, root);
1323 BUG_ON(ret);
1325 btrfs_prepare_extent_commit(trans, root);
1327 cur_trans = root->fs_info->running_transaction;
1329 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1330 root->fs_info->tree_root->node);
1331 switch_commit_root(root->fs_info->tree_root);
1333 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1334 root->fs_info->chunk_root->node);
1335 switch_commit_root(root->fs_info->chunk_root);
1337 update_super_roots(root);
1339 if (!root->fs_info->log_root_recovering) {
1340 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1341 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1344 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1345 sizeof(root->fs_info->super_copy));
1347 trans->transaction->blocked = 0;
1348 spin_lock(&root->fs_info->trans_lock);
1349 root->fs_info->running_transaction = NULL;
1350 root->fs_info->trans_no_join = 0;
1351 spin_unlock(&root->fs_info->trans_lock);
1352 mutex_unlock(&root->fs_info->reloc_mutex);
1354 wake_up(&root->fs_info->transaction_wait);
1356 ret = btrfs_write_and_wait_transaction(trans, root);
1357 BUG_ON(ret);
1358 write_ctree_super(trans, root, 0);
1361 * the super is written, we can safely allow the tree-loggers
1362 * to go about their business
1364 mutex_unlock(&root->fs_info->tree_log_mutex);
1366 btrfs_finish_extent_commit(trans, root);
1368 cur_trans->commit_done = 1;
1370 root->fs_info->last_trans_committed = cur_trans->transid;
1372 wake_up(&cur_trans->commit_wait);
1374 spin_lock(&root->fs_info->trans_lock);
1375 list_del_init(&cur_trans->list);
1376 spin_unlock(&root->fs_info->trans_lock);
1378 put_transaction(cur_trans);
1379 put_transaction(cur_trans);
1381 trace_btrfs_transaction_commit(root);
1383 btrfs_scrub_continue(root);
1385 if (current->journal_info == trans)
1386 current->journal_info = NULL;
1388 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1390 if (current != root->fs_info->transaction_kthread)
1391 btrfs_run_delayed_iputs(root);
1393 return ret;
1397 * interface function to delete all the snapshots we have scheduled for deletion
1399 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1401 LIST_HEAD(list);
1402 struct btrfs_fs_info *fs_info = root->fs_info;
1404 spin_lock(&fs_info->trans_lock);
1405 list_splice_init(&fs_info->dead_roots, &list);
1406 spin_unlock(&fs_info->trans_lock);
1408 while (!list_empty(&list)) {
1409 root = list_entry(list.next, struct btrfs_root, root_list);
1410 list_del(&root->root_list);
1412 btrfs_kill_all_delayed_nodes(root);
1414 if (btrfs_header_backref_rev(root->node) <
1415 BTRFS_MIXED_BACKREF_REV)
1416 btrfs_drop_snapshot(root, NULL, 0);
1417 else
1418 btrfs_drop_snapshot(root, NULL, 1);
1420 return 0;