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>
25 #include <linux/uuid.h>
28 #include "transaction.h"
31 #include "inode-map.h"
33 #include "dev-replace.h"
36 #define BTRFS_ROOT_TRANS_TAG 0
38 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
39 [TRANS_STATE_RUNNING] = 0U,
40 [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE |
42 [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE |
45 [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE |
49 [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE |
54 [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE |
61 void btrfs_put_transaction(struct btrfs_transaction *transaction)
63 WARN_ON(atomic_read(&transaction->use_count) == 0);
64 if (atomic_dec_and_test(&transaction->use_count)) {
65 BUG_ON(!list_empty(&transaction->list));
66 WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
67 if (transaction->delayed_refs.pending_csums)
68 printk(KERN_ERR "pending csums is %llu\n",
69 transaction->delayed_refs.pending_csums);
70 while (!list_empty(&transaction->pending_chunks)) {
71 struct extent_map *em;
73 em = list_first_entry(&transaction->pending_chunks,
74 struct extent_map, list);
75 list_del_init(&em->list);
79 * If any block groups are found in ->deleted_bgs then it's
80 * because the transaction was aborted and a commit did not
81 * happen (things failed before writing the new superblock
82 * and calling btrfs_finish_extent_commit()), so we can not
83 * discard the physical locations of the block groups.
85 while (!list_empty(&transaction->deleted_bgs)) {
86 struct btrfs_block_group_cache *cache;
88 cache = list_first_entry(&transaction->deleted_bgs,
89 struct btrfs_block_group_cache,
91 list_del_init(&cache->bg_list);
92 btrfs_put_block_group_trimming(cache);
93 btrfs_put_block_group(cache);
95 kmem_cache_free(btrfs_transaction_cachep, transaction);
99 static void clear_btree_io_tree(struct extent_io_tree *tree)
101 spin_lock(&tree->lock);
103 * Do a single barrier for the waitqueue_active check here, the state
104 * of the waitqueue should not change once clear_btree_io_tree is
108 while (!RB_EMPTY_ROOT(&tree->state)) {
109 struct rb_node *node;
110 struct extent_state *state;
112 node = rb_first(&tree->state);
113 state = rb_entry(node, struct extent_state, rb_node);
114 rb_erase(&state->rb_node, &tree->state);
115 RB_CLEAR_NODE(&state->rb_node);
117 * btree io trees aren't supposed to have tasks waiting for
118 * changes in the flags of extent states ever.
120 ASSERT(!waitqueue_active(&state->wq));
121 free_extent_state(state);
123 cond_resched_lock(&tree->lock);
125 spin_unlock(&tree->lock);
128 static noinline void switch_commit_roots(struct btrfs_transaction *trans,
129 struct btrfs_fs_info *fs_info)
131 struct btrfs_root *root, *tmp;
133 down_write(&fs_info->commit_root_sem);
134 list_for_each_entry_safe(root, tmp, &trans->switch_commits,
136 list_del_init(&root->dirty_list);
137 free_extent_buffer(root->commit_root);
138 root->commit_root = btrfs_root_node(root);
139 if (is_fstree(root->objectid))
140 btrfs_unpin_free_ino(root);
141 clear_btree_io_tree(&root->dirty_log_pages);
144 /* We can free old roots now. */
145 spin_lock(&trans->dropped_roots_lock);
146 while (!list_empty(&trans->dropped_roots)) {
147 root = list_first_entry(&trans->dropped_roots,
148 struct btrfs_root, root_list);
149 list_del_init(&root->root_list);
150 spin_unlock(&trans->dropped_roots_lock);
151 btrfs_drop_and_free_fs_root(fs_info, root);
152 spin_lock(&trans->dropped_roots_lock);
154 spin_unlock(&trans->dropped_roots_lock);
155 up_write(&fs_info->commit_root_sem);
158 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
161 if (type & TRANS_EXTWRITERS)
162 atomic_inc(&trans->num_extwriters);
165 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
168 if (type & TRANS_EXTWRITERS)
169 atomic_dec(&trans->num_extwriters);
172 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
175 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
178 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
180 return atomic_read(&trans->num_extwriters);
184 * either allocate a new transaction or hop into the existing one
186 static noinline int join_transaction(struct btrfs_root *root, unsigned int type)
188 struct btrfs_transaction *cur_trans;
189 struct btrfs_fs_info *fs_info = root->fs_info;
191 spin_lock(&fs_info->trans_lock);
193 /* The file system has been taken offline. No new transactions. */
194 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
195 spin_unlock(&fs_info->trans_lock);
199 cur_trans = fs_info->running_transaction;
201 if (cur_trans->aborted) {
202 spin_unlock(&fs_info->trans_lock);
203 return cur_trans->aborted;
205 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
206 spin_unlock(&fs_info->trans_lock);
209 atomic_inc(&cur_trans->use_count);
210 atomic_inc(&cur_trans->num_writers);
211 extwriter_counter_inc(cur_trans, type);
212 spin_unlock(&fs_info->trans_lock);
215 spin_unlock(&fs_info->trans_lock);
218 * If we are ATTACH, we just want to catch the current transaction,
219 * and commit it. If there is no transaction, just return ENOENT.
221 if (type == TRANS_ATTACH)
225 * JOIN_NOLOCK only happens during the transaction commit, so
226 * it is impossible that ->running_transaction is NULL
228 BUG_ON(type == TRANS_JOIN_NOLOCK);
230 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
234 spin_lock(&fs_info->trans_lock);
235 if (fs_info->running_transaction) {
237 * someone started a transaction after we unlocked. Make sure
238 * to redo the checks above
240 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
242 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
243 spin_unlock(&fs_info->trans_lock);
244 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
248 atomic_set(&cur_trans->num_writers, 1);
249 extwriter_counter_init(cur_trans, type);
250 init_waitqueue_head(&cur_trans->writer_wait);
251 init_waitqueue_head(&cur_trans->commit_wait);
252 init_waitqueue_head(&cur_trans->pending_wait);
253 cur_trans->state = TRANS_STATE_RUNNING;
255 * One for this trans handle, one so it will live on until we
256 * commit the transaction.
258 atomic_set(&cur_trans->use_count, 2);
259 atomic_set(&cur_trans->pending_ordered, 0);
260 cur_trans->flags = 0;
261 cur_trans->start_time = get_seconds();
263 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
265 cur_trans->delayed_refs.href_root = RB_ROOT;
266 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
267 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
270 * although the tree mod log is per file system and not per transaction,
271 * the log must never go across transaction boundaries.
274 if (!list_empty(&fs_info->tree_mod_seq_list))
275 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when "
276 "creating a fresh transaction\n");
277 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
278 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when "
279 "creating a fresh transaction\n");
280 atomic64_set(&fs_info->tree_mod_seq, 0);
282 spin_lock_init(&cur_trans->delayed_refs.lock);
284 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
285 INIT_LIST_HEAD(&cur_trans->pending_chunks);
286 INIT_LIST_HEAD(&cur_trans->switch_commits);
287 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
288 INIT_LIST_HEAD(&cur_trans->io_bgs);
289 INIT_LIST_HEAD(&cur_trans->dropped_roots);
290 mutex_init(&cur_trans->cache_write_mutex);
291 cur_trans->num_dirty_bgs = 0;
292 spin_lock_init(&cur_trans->dirty_bgs_lock);
293 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
294 spin_lock_init(&cur_trans->dropped_roots_lock);
295 list_add_tail(&cur_trans->list, &fs_info->trans_list);
296 extent_io_tree_init(&cur_trans->dirty_pages,
297 fs_info->btree_inode->i_mapping);
298 fs_info->generation++;
299 cur_trans->transid = fs_info->generation;
300 fs_info->running_transaction = cur_trans;
301 cur_trans->aborted = 0;
302 spin_unlock(&fs_info->trans_lock);
308 * this does all the record keeping required to make sure that a reference
309 * counted root is properly recorded in a given transaction. This is required
310 * to make sure the old root from before we joined the transaction is deleted
311 * when the transaction commits
313 static int record_root_in_trans(struct btrfs_trans_handle *trans,
314 struct btrfs_root *root,
317 if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
318 root->last_trans < trans->transid) || force) {
319 WARN_ON(root == root->fs_info->extent_root);
320 WARN_ON(root->commit_root != root->node);
323 * see below for IN_TRANS_SETUP usage rules
324 * we have the reloc mutex held now, so there
325 * is only one writer in this function
327 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
329 /* make sure readers find IN_TRANS_SETUP before
330 * they find our root->last_trans update
334 spin_lock(&root->fs_info->fs_roots_radix_lock);
335 if (root->last_trans == trans->transid && !force) {
336 spin_unlock(&root->fs_info->fs_roots_radix_lock);
339 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
340 (unsigned long)root->root_key.objectid,
341 BTRFS_ROOT_TRANS_TAG);
342 spin_unlock(&root->fs_info->fs_roots_radix_lock);
343 root->last_trans = trans->transid;
345 /* this is pretty tricky. We don't want to
346 * take the relocation lock in btrfs_record_root_in_trans
347 * unless we're really doing the first setup for this root in
350 * Normally we'd use root->last_trans as a flag to decide
351 * if we want to take the expensive mutex.
353 * But, we have to set root->last_trans before we
354 * init the relocation root, otherwise, we trip over warnings
355 * in ctree.c. The solution used here is to flag ourselves
356 * with root IN_TRANS_SETUP. When this is 1, we're still
357 * fixing up the reloc trees and everyone must wait.
359 * When this is zero, they can trust root->last_trans and fly
360 * through btrfs_record_root_in_trans without having to take the
361 * lock. smp_wmb() makes sure that all the writes above are
362 * done before we pop in the zero below
364 btrfs_init_reloc_root(trans, root);
365 smp_mb__before_atomic();
366 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
372 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
373 struct btrfs_root *root)
375 struct btrfs_transaction *cur_trans = trans->transaction;
377 /* Add ourselves to the transaction dropped list */
378 spin_lock(&cur_trans->dropped_roots_lock);
379 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
380 spin_unlock(&cur_trans->dropped_roots_lock);
382 /* Make sure we don't try to update the root at commit time */
383 spin_lock(&root->fs_info->fs_roots_radix_lock);
384 radix_tree_tag_clear(&root->fs_info->fs_roots_radix,
385 (unsigned long)root->root_key.objectid,
386 BTRFS_ROOT_TRANS_TAG);
387 spin_unlock(&root->fs_info->fs_roots_radix_lock);
390 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
391 struct btrfs_root *root)
393 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
397 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
401 if (root->last_trans == trans->transid &&
402 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
405 mutex_lock(&root->fs_info->reloc_mutex);
406 record_root_in_trans(trans, root, 0);
407 mutex_unlock(&root->fs_info->reloc_mutex);
412 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
414 return (trans->state >= TRANS_STATE_BLOCKED &&
415 trans->state < TRANS_STATE_UNBLOCKED &&
419 /* wait for commit against the current transaction to become unblocked
420 * when this is done, it is safe to start a new transaction, but the current
421 * transaction might not be fully on disk.
423 static void wait_current_trans(struct btrfs_root *root)
425 struct btrfs_transaction *cur_trans;
427 spin_lock(&root->fs_info->trans_lock);
428 cur_trans = root->fs_info->running_transaction;
429 if (cur_trans && is_transaction_blocked(cur_trans)) {
430 atomic_inc(&cur_trans->use_count);
431 spin_unlock(&root->fs_info->trans_lock);
433 wait_event(root->fs_info->transaction_wait,
434 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
436 btrfs_put_transaction(cur_trans);
438 spin_unlock(&root->fs_info->trans_lock);
442 static int may_wait_transaction(struct btrfs_root *root, int type)
444 if (test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
447 if (type == TRANS_USERSPACE)
450 if (type == TRANS_START &&
451 !atomic_read(&root->fs_info->open_ioctl_trans))
457 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
459 if (!root->fs_info->reloc_ctl ||
460 !test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
461 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
468 static struct btrfs_trans_handle *
469 start_transaction(struct btrfs_root *root, unsigned int num_items,
470 unsigned int type, enum btrfs_reserve_flush_enum flush)
472 struct btrfs_trans_handle *h;
473 struct btrfs_transaction *cur_trans;
475 u64 qgroup_reserved = 0;
476 bool reloc_reserved = false;
479 /* Send isn't supposed to start transactions. */
480 ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB);
482 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
483 return ERR_PTR(-EROFS);
485 if (current->journal_info) {
486 WARN_ON(type & TRANS_EXTWRITERS);
487 h = current->journal_info;
489 WARN_ON(h->use_count > 2);
490 h->orig_rsv = h->block_rsv;
496 * Do the reservation before we join the transaction so we can do all
497 * the appropriate flushing if need be.
499 if (num_items > 0 && root != root->fs_info->chunk_root) {
500 qgroup_reserved = num_items * root->nodesize;
501 ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved);
505 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
507 * Do the reservation for the relocation root creation
509 if (need_reserve_reloc_root(root)) {
510 num_bytes += root->nodesize;
511 reloc_reserved = true;
514 ret = btrfs_block_rsv_add(root,
515 &root->fs_info->trans_block_rsv,
521 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
528 * If we are JOIN_NOLOCK we're already committing a transaction and
529 * waiting on this guy, so we don't need to do the sb_start_intwrite
530 * because we're already holding a ref. We need this because we could
531 * have raced in and did an fsync() on a file which can kick a commit
532 * and then we deadlock with somebody doing a freeze.
534 * If we are ATTACH, it means we just want to catch the current
535 * transaction and commit it, so we needn't do sb_start_intwrite().
537 if (type & __TRANS_FREEZABLE)
538 sb_start_intwrite(root->fs_info->sb);
540 if (may_wait_transaction(root, type))
541 wait_current_trans(root);
544 ret = join_transaction(root, type);
546 wait_current_trans(root);
547 if (unlikely(type == TRANS_ATTACH))
550 } while (ret == -EBUSY);
553 /* We must get the transaction if we are JOIN_NOLOCK. */
554 BUG_ON(type == TRANS_JOIN_NOLOCK);
558 cur_trans = root->fs_info->running_transaction;
560 h->transid = cur_trans->transid;
561 h->transaction = cur_trans;
564 h->fs_info = root->fs_info;
567 h->can_flush_pending_bgs = true;
568 INIT_LIST_HEAD(&h->qgroup_ref_list);
569 INIT_LIST_HEAD(&h->new_bgs);
572 if (cur_trans->state >= TRANS_STATE_BLOCKED &&
573 may_wait_transaction(root, type)) {
574 current->journal_info = h;
575 btrfs_commit_transaction(h, root);
580 trace_btrfs_space_reservation(root->fs_info, "transaction",
581 h->transid, num_bytes, 1);
582 h->block_rsv = &root->fs_info->trans_block_rsv;
583 h->bytes_reserved = num_bytes;
584 h->reloc_reserved = reloc_reserved;
588 btrfs_record_root_in_trans(h, root);
590 if (!current->journal_info && type != TRANS_USERSPACE)
591 current->journal_info = h;
595 if (type & __TRANS_FREEZABLE)
596 sb_end_intwrite(root->fs_info->sb);
597 kmem_cache_free(btrfs_trans_handle_cachep, h);
600 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
603 btrfs_qgroup_free_meta(root, qgroup_reserved);
607 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
608 unsigned int num_items)
610 return start_transaction(root, num_items, TRANS_START,
611 BTRFS_RESERVE_FLUSH_ALL);
613 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
614 struct btrfs_root *root,
615 unsigned int num_items,
618 struct btrfs_trans_handle *trans;
622 trans = btrfs_start_transaction(root, num_items);
623 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
626 trans = btrfs_start_transaction(root, 0);
630 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
631 ret = btrfs_cond_migrate_bytes(root->fs_info,
632 &root->fs_info->trans_block_rsv,
636 btrfs_end_transaction(trans, root);
640 trans->block_rsv = &root->fs_info->trans_block_rsv;
641 trans->bytes_reserved = num_bytes;
642 trace_btrfs_space_reservation(root->fs_info, "transaction",
643 trans->transid, num_bytes, 1);
648 struct btrfs_trans_handle *btrfs_start_transaction_lflush(
649 struct btrfs_root *root,
650 unsigned int num_items)
652 return start_transaction(root, num_items, TRANS_START,
653 BTRFS_RESERVE_FLUSH_LIMIT);
656 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
658 return start_transaction(root, 0, TRANS_JOIN,
659 BTRFS_RESERVE_NO_FLUSH);
662 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
664 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
665 BTRFS_RESERVE_NO_FLUSH);
668 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
670 return start_transaction(root, 0, TRANS_USERSPACE,
671 BTRFS_RESERVE_NO_FLUSH);
675 * btrfs_attach_transaction() - catch the running transaction
677 * It is used when we want to commit the current the transaction, but
678 * don't want to start a new one.
680 * Note: If this function return -ENOENT, it just means there is no
681 * running transaction. But it is possible that the inactive transaction
682 * is still in the memory, not fully on disk. If you hope there is no
683 * inactive transaction in the fs when -ENOENT is returned, you should
685 * btrfs_attach_transaction_barrier()
687 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
689 return start_transaction(root, 0, TRANS_ATTACH,
690 BTRFS_RESERVE_NO_FLUSH);
694 * btrfs_attach_transaction_barrier() - catch the running transaction
696 * It is similar to the above function, the differentia is this one
697 * will wait for all the inactive transactions until they fully
700 struct btrfs_trans_handle *
701 btrfs_attach_transaction_barrier(struct btrfs_root *root)
703 struct btrfs_trans_handle *trans;
705 trans = start_transaction(root, 0, TRANS_ATTACH,
706 BTRFS_RESERVE_NO_FLUSH);
707 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
708 btrfs_wait_for_commit(root, 0);
713 /* wait for a transaction commit to be fully complete */
714 static noinline void wait_for_commit(struct btrfs_root *root,
715 struct btrfs_transaction *commit)
717 wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED);
720 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
722 struct btrfs_transaction *cur_trans = NULL, *t;
726 if (transid <= root->fs_info->last_trans_committed)
729 /* find specified transaction */
730 spin_lock(&root->fs_info->trans_lock);
731 list_for_each_entry(t, &root->fs_info->trans_list, list) {
732 if (t->transid == transid) {
734 atomic_inc(&cur_trans->use_count);
738 if (t->transid > transid) {
743 spin_unlock(&root->fs_info->trans_lock);
746 * The specified transaction doesn't exist, or we
747 * raced with btrfs_commit_transaction
750 if (transid > root->fs_info->last_trans_committed)
755 /* find newest transaction that is committing | committed */
756 spin_lock(&root->fs_info->trans_lock);
757 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
759 if (t->state >= TRANS_STATE_COMMIT_START) {
760 if (t->state == TRANS_STATE_COMPLETED)
763 atomic_inc(&cur_trans->use_count);
767 spin_unlock(&root->fs_info->trans_lock);
769 goto out; /* nothing committing|committed */
772 wait_for_commit(root, cur_trans);
773 btrfs_put_transaction(cur_trans);
778 void btrfs_throttle(struct btrfs_root *root)
780 if (!atomic_read(&root->fs_info->open_ioctl_trans))
781 wait_current_trans(root);
784 static int should_end_transaction(struct btrfs_trans_handle *trans,
785 struct btrfs_root *root)
787 if (root->fs_info->global_block_rsv.space_info->full &&
788 btrfs_check_space_for_delayed_refs(trans, root))
791 return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
794 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
795 struct btrfs_root *root)
797 struct btrfs_transaction *cur_trans = trans->transaction;
802 if (cur_trans->state >= TRANS_STATE_BLOCKED ||
803 cur_trans->delayed_refs.flushing)
806 updates = trans->delayed_ref_updates;
807 trans->delayed_ref_updates = 0;
809 err = btrfs_run_delayed_refs(trans, root, updates * 2);
810 if (err) /* Error code will also eval true */
814 return should_end_transaction(trans, root);
817 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
818 struct btrfs_root *root, int throttle)
820 struct btrfs_transaction *cur_trans = trans->transaction;
821 struct btrfs_fs_info *info = root->fs_info;
822 u64 transid = trans->transid;
823 unsigned long cur = trans->delayed_ref_updates;
824 int lock = (trans->type != TRANS_JOIN_NOLOCK);
826 int must_run_delayed_refs = 0;
828 if (trans->use_count > 1) {
830 trans->block_rsv = trans->orig_rsv;
834 btrfs_trans_release_metadata(trans, root);
835 trans->block_rsv = NULL;
837 if (!list_empty(&trans->new_bgs))
838 btrfs_create_pending_block_groups(trans, root);
840 trans->delayed_ref_updates = 0;
842 must_run_delayed_refs =
843 btrfs_should_throttle_delayed_refs(trans, root);
844 cur = max_t(unsigned long, cur, 32);
847 * don't make the caller wait if they are from a NOLOCK
848 * or ATTACH transaction, it will deadlock with commit
850 if (must_run_delayed_refs == 1 &&
851 (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
852 must_run_delayed_refs = 2;
855 btrfs_trans_release_metadata(trans, root);
856 trans->block_rsv = NULL;
858 if (!list_empty(&trans->new_bgs))
859 btrfs_create_pending_block_groups(trans, root);
861 btrfs_trans_release_chunk_metadata(trans);
863 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
864 should_end_transaction(trans, root) &&
865 ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
866 spin_lock(&info->trans_lock);
867 if (cur_trans->state == TRANS_STATE_RUNNING)
868 cur_trans->state = TRANS_STATE_BLOCKED;
869 spin_unlock(&info->trans_lock);
872 if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
874 return btrfs_commit_transaction(trans, root);
876 wake_up_process(info->transaction_kthread);
879 if (trans->type & __TRANS_FREEZABLE)
880 sb_end_intwrite(root->fs_info->sb);
882 WARN_ON(cur_trans != info->running_transaction);
883 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
884 atomic_dec(&cur_trans->num_writers);
885 extwriter_counter_dec(cur_trans, trans->type);
888 * Make sure counter is updated before we wake up waiters.
891 if (waitqueue_active(&cur_trans->writer_wait))
892 wake_up(&cur_trans->writer_wait);
893 btrfs_put_transaction(cur_trans);
895 if (current->journal_info == trans)
896 current->journal_info = NULL;
899 btrfs_run_delayed_iputs(root);
901 if (trans->aborted ||
902 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
903 wake_up_process(info->transaction_kthread);
906 assert_qgroups_uptodate(trans);
908 kmem_cache_free(btrfs_trans_handle_cachep, trans);
909 if (must_run_delayed_refs) {
910 btrfs_async_run_delayed_refs(root, cur, transid,
911 must_run_delayed_refs == 1);
916 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
917 struct btrfs_root *root)
919 return __btrfs_end_transaction(trans, root, 0);
922 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
923 struct btrfs_root *root)
925 return __btrfs_end_transaction(trans, root, 1);
929 * when btree blocks are allocated, they have some corresponding bits set for
930 * them in one of two extent_io trees. This is used to make sure all of
931 * those extents are sent to disk but does not wait on them
933 int btrfs_write_marked_extents(struct btrfs_root *root,
934 struct extent_io_tree *dirty_pages, int mark)
938 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
939 struct extent_state *cached_state = NULL;
943 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
944 mark, &cached_state)) {
945 bool wait_writeback = false;
947 err = convert_extent_bit(dirty_pages, start, end,
949 mark, &cached_state);
951 * convert_extent_bit can return -ENOMEM, which is most of the
952 * time a temporary error. So when it happens, ignore the error
953 * and wait for writeback of this range to finish - because we
954 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
955 * to btrfs_wait_marked_extents() would not know that writeback
956 * for this range started and therefore wouldn't wait for it to
957 * finish - we don't want to commit a superblock that points to
958 * btree nodes/leafs for which writeback hasn't finished yet
959 * (and without errors).
960 * We cleanup any entries left in the io tree when committing
961 * the transaction (through clear_btree_io_tree()).
963 if (err == -ENOMEM) {
965 wait_writeback = true;
968 err = filemap_fdatawrite_range(mapping, start, end);
971 else if (wait_writeback)
972 werr = filemap_fdatawait_range(mapping, start, end);
973 free_extent_state(cached_state);
982 * when btree blocks are allocated, they have some corresponding bits set for
983 * them in one of two extent_io trees. This is used to make sure all of
984 * those extents are on disk for transaction or log commit. We wait
985 * on all the pages and clear them from the dirty pages state tree
987 int btrfs_wait_marked_extents(struct btrfs_root *root,
988 struct extent_io_tree *dirty_pages, int mark)
992 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
993 struct extent_state *cached_state = NULL;
998 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
999 EXTENT_NEED_WAIT, &cached_state)) {
1001 * Ignore -ENOMEM errors returned by clear_extent_bit().
1002 * When committing the transaction, we'll remove any entries
1003 * left in the io tree. For a log commit, we don't remove them
1004 * after committing the log because the tree can be accessed
1005 * concurrently - we do it only at transaction commit time when
1006 * it's safe to do it (through clear_btree_io_tree()).
1008 err = clear_extent_bit(dirty_pages, start, end,
1010 0, 0, &cached_state, GFP_NOFS);
1014 err = filemap_fdatawait_range(mapping, start, end);
1017 free_extent_state(cached_state);
1018 cached_state = NULL;
1025 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1026 if ((mark & EXTENT_DIRTY) &&
1027 test_and_clear_bit(BTRFS_FS_LOG1_ERR,
1028 &root->fs_info->flags))
1031 if ((mark & EXTENT_NEW) &&
1032 test_and_clear_bit(BTRFS_FS_LOG2_ERR,
1033 &root->fs_info->flags))
1036 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR,
1037 &root->fs_info->flags))
1041 if (errors && !werr)
1048 * when btree blocks are allocated, they have some corresponding bits set for
1049 * them in one of two extent_io trees. This is used to make sure all of
1050 * those extents are on disk for transaction or log commit
1052 static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
1053 struct extent_io_tree *dirty_pages, int mark)
1057 struct blk_plug plug;
1059 blk_start_plug(&plug);
1060 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
1061 blk_finish_plug(&plug);
1062 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
1071 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
1072 struct btrfs_root *root)
1076 ret = btrfs_write_and_wait_marked_extents(root,
1077 &trans->transaction->dirty_pages,
1079 clear_btree_io_tree(&trans->transaction->dirty_pages);
1085 * this is used to update the root pointer in the tree of tree roots.
1087 * But, in the case of the extent allocation tree, updating the root
1088 * pointer may allocate blocks which may change the root of the extent
1091 * So, this loops and repeats and makes sure the cowonly root didn't
1092 * change while the root pointer was being updated in the metadata.
1094 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1095 struct btrfs_root *root)
1098 u64 old_root_bytenr;
1100 struct btrfs_root *tree_root = root->fs_info->tree_root;
1102 old_root_used = btrfs_root_used(&root->root_item);
1105 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1106 if (old_root_bytenr == root->node->start &&
1107 old_root_used == btrfs_root_used(&root->root_item))
1110 btrfs_set_root_node(&root->root_item, root->node);
1111 ret = btrfs_update_root(trans, tree_root,
1117 old_root_used = btrfs_root_used(&root->root_item);
1124 * update all the cowonly tree roots on disk
1126 * The error handling in this function may not be obvious. Any of the
1127 * failures will cause the file system to go offline. We still need
1128 * to clean up the delayed refs.
1130 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
1131 struct btrfs_root *root)
1133 struct btrfs_fs_info *fs_info = root->fs_info;
1134 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1135 struct list_head *io_bgs = &trans->transaction->io_bgs;
1136 struct list_head *next;
1137 struct extent_buffer *eb;
1140 eb = btrfs_lock_root_node(fs_info->tree_root);
1141 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1143 btrfs_tree_unlock(eb);
1144 free_extent_buffer(eb);
1149 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1153 ret = btrfs_run_dev_stats(trans, root->fs_info);
1156 ret = btrfs_run_dev_replace(trans, root->fs_info);
1159 ret = btrfs_run_qgroups(trans, root->fs_info);
1163 ret = btrfs_setup_space_cache(trans, root);
1167 /* run_qgroups might have added some more refs */
1168 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1172 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1173 next = fs_info->dirty_cowonly_roots.next;
1174 list_del_init(next);
1175 root = list_entry(next, struct btrfs_root, dirty_list);
1176 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1178 if (root != fs_info->extent_root)
1179 list_add_tail(&root->dirty_list,
1180 &trans->transaction->switch_commits);
1181 ret = update_cowonly_root(trans, root);
1184 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1189 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1190 ret = btrfs_write_dirty_block_groups(trans, root);
1193 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1198 if (!list_empty(&fs_info->dirty_cowonly_roots))
1201 list_add_tail(&fs_info->extent_root->dirty_list,
1202 &trans->transaction->switch_commits);
1203 btrfs_after_dev_replace_commit(fs_info);
1209 * dead roots are old snapshots that need to be deleted. This allocates
1210 * a dirty root struct and adds it into the list of dead roots that need to
1213 void btrfs_add_dead_root(struct btrfs_root *root)
1215 spin_lock(&root->fs_info->trans_lock);
1216 if (list_empty(&root->root_list))
1217 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
1218 spin_unlock(&root->fs_info->trans_lock);
1222 * update all the cowonly tree roots on disk
1224 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
1225 struct btrfs_root *root)
1227 struct btrfs_root *gang[8];
1228 struct btrfs_fs_info *fs_info = root->fs_info;
1233 spin_lock(&fs_info->fs_roots_radix_lock);
1235 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1238 BTRFS_ROOT_TRANS_TAG);
1241 for (i = 0; i < ret; i++) {
1243 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1244 (unsigned long)root->root_key.objectid,
1245 BTRFS_ROOT_TRANS_TAG);
1246 spin_unlock(&fs_info->fs_roots_radix_lock);
1248 btrfs_free_log(trans, root);
1249 btrfs_update_reloc_root(trans, root);
1250 btrfs_orphan_commit_root(trans, root);
1252 btrfs_save_ino_cache(root, trans);
1254 /* see comments in should_cow_block() */
1255 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1256 smp_mb__after_atomic();
1258 if (root->commit_root != root->node) {
1259 list_add_tail(&root->dirty_list,
1260 &trans->transaction->switch_commits);
1261 btrfs_set_root_node(&root->root_item,
1265 err = btrfs_update_root(trans, fs_info->tree_root,
1268 spin_lock(&fs_info->fs_roots_radix_lock);
1271 btrfs_qgroup_free_meta_all(root);
1274 spin_unlock(&fs_info->fs_roots_radix_lock);
1279 * defrag a given btree.
1280 * Every leaf in the btree is read and defragged.
1282 int btrfs_defrag_root(struct btrfs_root *root)
1284 struct btrfs_fs_info *info = root->fs_info;
1285 struct btrfs_trans_handle *trans;
1288 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1292 trans = btrfs_start_transaction(root, 0);
1294 return PTR_ERR(trans);
1296 ret = btrfs_defrag_leaves(trans, root);
1298 btrfs_end_transaction(trans, root);
1299 btrfs_btree_balance_dirty(info->tree_root);
1302 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1305 if (btrfs_defrag_cancelled(root->fs_info)) {
1306 pr_debug("BTRFS: defrag_root cancelled\n");
1311 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1316 * Do all special snapshot related qgroup dirty hack.
1318 * Will do all needed qgroup inherit and dirty hack like switch commit
1319 * roots inside one transaction and write all btree into disk, to make
1322 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1323 struct btrfs_root *src,
1324 struct btrfs_root *parent,
1325 struct btrfs_qgroup_inherit *inherit,
1328 struct btrfs_fs_info *fs_info = src->fs_info;
1332 * Save some performance in the case that qgroups are not
1333 * enabled. If this check races with the ioctl, rescan will
1336 mutex_lock(&fs_info->qgroup_ioctl_lock);
1337 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
1338 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1341 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1344 * We are going to commit transaction, see btrfs_commit_transaction()
1345 * comment for reason locking tree_log_mutex
1347 mutex_lock(&fs_info->tree_log_mutex);
1349 ret = commit_fs_roots(trans, src);
1352 ret = btrfs_qgroup_prepare_account_extents(trans, fs_info);
1355 ret = btrfs_qgroup_account_extents(trans, fs_info);
1359 /* Now qgroup are all updated, we can inherit it to new qgroups */
1360 ret = btrfs_qgroup_inherit(trans, fs_info,
1361 src->root_key.objectid, dst_objectid,
1367 * Now we do a simplified commit transaction, which will:
1368 * 1) commit all subvolume and extent tree
1369 * To ensure all subvolume and extent tree have a valid
1370 * commit_root to accounting later insert_dir_item()
1371 * 2) write all btree blocks onto disk
1372 * This is to make sure later btree modification will be cowed
1373 * Or commit_root can be populated and cause wrong qgroup numbers
1374 * In this simplified commit, we don't really care about other trees
1375 * like chunk and root tree, as they won't affect qgroup.
1376 * And we don't write super to avoid half committed status.
1378 ret = commit_cowonly_roots(trans, src);
1381 switch_commit_roots(trans->transaction, fs_info);
1382 ret = btrfs_write_and_wait_transaction(trans, src);
1384 btrfs_handle_fs_error(fs_info, ret,
1385 "Error while writing out transaction for qgroup");
1388 mutex_unlock(&fs_info->tree_log_mutex);
1391 * Force parent root to be updated, as we recorded it before so its
1392 * last_trans == cur_transid.
1393 * Or it won't be committed again onto disk after later
1397 record_root_in_trans(trans, parent, 1);
1402 * new snapshots need to be created at a very specific time in the
1403 * transaction commit. This does the actual creation.
1406 * If the error which may affect the commitment of the current transaction
1407 * happens, we should return the error number. If the error which just affect
1408 * the creation of the pending snapshots, just return 0.
1410 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1411 struct btrfs_fs_info *fs_info,
1412 struct btrfs_pending_snapshot *pending)
1414 struct btrfs_key key;
1415 struct btrfs_root_item *new_root_item;
1416 struct btrfs_root *tree_root = fs_info->tree_root;
1417 struct btrfs_root *root = pending->root;
1418 struct btrfs_root *parent_root;
1419 struct btrfs_block_rsv *rsv;
1420 struct inode *parent_inode;
1421 struct btrfs_path *path;
1422 struct btrfs_dir_item *dir_item;
1423 struct dentry *dentry;
1424 struct extent_buffer *tmp;
1425 struct extent_buffer *old;
1426 struct timespec cur_time;
1434 ASSERT(pending->path);
1435 path = pending->path;
1437 ASSERT(pending->root_item);
1438 new_root_item = pending->root_item;
1440 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1442 goto no_free_objectid;
1445 * Make qgroup to skip current new snapshot's qgroupid, as it is
1446 * accounted by later btrfs_qgroup_inherit().
1448 btrfs_set_skip_qgroup(trans, objectid);
1450 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1452 if (to_reserve > 0) {
1453 pending->error = btrfs_block_rsv_add(root,
1454 &pending->block_rsv,
1456 BTRFS_RESERVE_NO_FLUSH);
1458 goto clear_skip_qgroup;
1461 key.objectid = objectid;
1462 key.offset = (u64)-1;
1463 key.type = BTRFS_ROOT_ITEM_KEY;
1465 rsv = trans->block_rsv;
1466 trans->block_rsv = &pending->block_rsv;
1467 trans->bytes_reserved = trans->block_rsv->reserved;
1468 trace_btrfs_space_reservation(root->fs_info, "transaction",
1470 trans->bytes_reserved, 1);
1471 dentry = pending->dentry;
1472 parent_inode = pending->dir;
1473 parent_root = BTRFS_I(parent_inode)->root;
1474 record_root_in_trans(trans, parent_root, 0);
1476 cur_time = current_fs_time(parent_inode->i_sb);
1479 * insert the directory item
1481 ret = btrfs_set_inode_index(parent_inode, &index);
1482 BUG_ON(ret); /* -ENOMEM */
1484 /* check if there is a file/dir which has the same name. */
1485 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1486 btrfs_ino(parent_inode),
1487 dentry->d_name.name,
1488 dentry->d_name.len, 0);
1489 if (dir_item != NULL && !IS_ERR(dir_item)) {
1490 pending->error = -EEXIST;
1491 goto dir_item_existed;
1492 } else if (IS_ERR(dir_item)) {
1493 ret = PTR_ERR(dir_item);
1494 btrfs_abort_transaction(trans, ret);
1497 btrfs_release_path(path);
1500 * pull in the delayed directory update
1501 * and the delayed inode item
1502 * otherwise we corrupt the FS during
1505 ret = btrfs_run_delayed_items(trans, root);
1506 if (ret) { /* Transaction aborted */
1507 btrfs_abort_transaction(trans, ret);
1511 record_root_in_trans(trans, root, 0);
1512 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1513 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1514 btrfs_check_and_init_root_item(new_root_item);
1516 root_flags = btrfs_root_flags(new_root_item);
1517 if (pending->readonly)
1518 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1520 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1521 btrfs_set_root_flags(new_root_item, root_flags);
1523 btrfs_set_root_generation_v2(new_root_item,
1525 uuid_le_gen(&new_uuid);
1526 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1527 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1529 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1530 memset(new_root_item->received_uuid, 0,
1531 sizeof(new_root_item->received_uuid));
1532 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1533 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1534 btrfs_set_root_stransid(new_root_item, 0);
1535 btrfs_set_root_rtransid(new_root_item, 0);
1537 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1538 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1539 btrfs_set_root_otransid(new_root_item, trans->transid);
1541 old = btrfs_lock_root_node(root);
1542 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1544 btrfs_tree_unlock(old);
1545 free_extent_buffer(old);
1546 btrfs_abort_transaction(trans, ret);
1550 btrfs_set_lock_blocking(old);
1552 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1553 /* clean up in any case */
1554 btrfs_tree_unlock(old);
1555 free_extent_buffer(old);
1557 btrfs_abort_transaction(trans, ret);
1560 /* see comments in should_cow_block() */
1561 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1564 btrfs_set_root_node(new_root_item, tmp);
1565 /* record when the snapshot was created in key.offset */
1566 key.offset = trans->transid;
1567 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1568 btrfs_tree_unlock(tmp);
1569 free_extent_buffer(tmp);
1571 btrfs_abort_transaction(trans, ret);
1576 * insert root back/forward references
1578 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1579 parent_root->root_key.objectid,
1580 btrfs_ino(parent_inode), index,
1581 dentry->d_name.name, dentry->d_name.len);
1583 btrfs_abort_transaction(trans, ret);
1587 key.offset = (u64)-1;
1588 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1589 if (IS_ERR(pending->snap)) {
1590 ret = PTR_ERR(pending->snap);
1591 btrfs_abort_transaction(trans, ret);
1595 ret = btrfs_reloc_post_snapshot(trans, pending);
1597 btrfs_abort_transaction(trans, ret);
1601 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1603 btrfs_abort_transaction(trans, ret);
1608 * Do special qgroup accounting for snapshot, as we do some qgroup
1609 * snapshot hack to do fast snapshot.
1610 * To co-operate with that hack, we do hack again.
1611 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1613 ret = qgroup_account_snapshot(trans, root, parent_root,
1614 pending->inherit, objectid);
1618 ret = btrfs_insert_dir_item(trans, parent_root,
1619 dentry->d_name.name, dentry->d_name.len,
1621 BTRFS_FT_DIR, index);
1622 /* We have check then name at the beginning, so it is impossible. */
1623 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1625 btrfs_abort_transaction(trans, ret);
1629 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1630 dentry->d_name.len * 2);
1631 parent_inode->i_mtime = parent_inode->i_ctime =
1632 current_fs_time(parent_inode->i_sb);
1633 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1635 btrfs_abort_transaction(trans, ret);
1638 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b,
1639 BTRFS_UUID_KEY_SUBVOL, objectid);
1641 btrfs_abort_transaction(trans, ret);
1644 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1645 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
1646 new_root_item->received_uuid,
1647 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1649 if (ret && ret != -EEXIST) {
1650 btrfs_abort_transaction(trans, ret);
1655 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1657 btrfs_abort_transaction(trans, ret);
1662 pending->error = ret;
1664 trans->block_rsv = rsv;
1665 trans->bytes_reserved = 0;
1667 btrfs_clear_skip_qgroup(trans);
1669 kfree(new_root_item);
1670 pending->root_item = NULL;
1671 btrfs_free_path(path);
1672 pending->path = NULL;
1678 * create all the snapshots we've scheduled for creation
1680 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1681 struct btrfs_fs_info *fs_info)
1683 struct btrfs_pending_snapshot *pending, *next;
1684 struct list_head *head = &trans->transaction->pending_snapshots;
1687 list_for_each_entry_safe(pending, next, head, list) {
1688 list_del(&pending->list);
1689 ret = create_pending_snapshot(trans, fs_info, pending);
1696 static void update_super_roots(struct btrfs_root *root)
1698 struct btrfs_root_item *root_item;
1699 struct btrfs_super_block *super;
1701 super = root->fs_info->super_copy;
1703 root_item = &root->fs_info->chunk_root->root_item;
1704 super->chunk_root = root_item->bytenr;
1705 super->chunk_root_generation = root_item->generation;
1706 super->chunk_root_level = root_item->level;
1708 root_item = &root->fs_info->tree_root->root_item;
1709 super->root = root_item->bytenr;
1710 super->generation = root_item->generation;
1711 super->root_level = root_item->level;
1712 if (btrfs_test_opt(root->fs_info, SPACE_CACHE))
1713 super->cache_generation = root_item->generation;
1714 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &root->fs_info->flags))
1715 super->uuid_tree_generation = root_item->generation;
1718 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1720 struct btrfs_transaction *trans;
1723 spin_lock(&info->trans_lock);
1724 trans = info->running_transaction;
1726 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1727 spin_unlock(&info->trans_lock);
1731 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1733 struct btrfs_transaction *trans;
1736 spin_lock(&info->trans_lock);
1737 trans = info->running_transaction;
1739 ret = is_transaction_blocked(trans);
1740 spin_unlock(&info->trans_lock);
1745 * wait for the current transaction commit to start and block subsequent
1748 static void wait_current_trans_commit_start(struct btrfs_root *root,
1749 struct btrfs_transaction *trans)
1751 wait_event(root->fs_info->transaction_blocked_wait,
1752 trans->state >= TRANS_STATE_COMMIT_START ||
1757 * wait for the current transaction to start and then become unblocked.
1760 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1761 struct btrfs_transaction *trans)
1763 wait_event(root->fs_info->transaction_wait,
1764 trans->state >= TRANS_STATE_UNBLOCKED ||
1769 * commit transactions asynchronously. once btrfs_commit_transaction_async
1770 * returns, any subsequent transaction will not be allowed to join.
1772 struct btrfs_async_commit {
1773 struct btrfs_trans_handle *newtrans;
1774 struct btrfs_root *root;
1775 struct work_struct work;
1778 static void do_async_commit(struct work_struct *work)
1780 struct btrfs_async_commit *ac =
1781 container_of(work, struct btrfs_async_commit, work);
1784 * We've got freeze protection passed with the transaction.
1785 * Tell lockdep about it.
1787 if (ac->newtrans->type & __TRANS_FREEZABLE)
1788 __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS);
1790 current->journal_info = ac->newtrans;
1792 btrfs_commit_transaction(ac->newtrans, ac->root);
1796 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1797 struct btrfs_root *root,
1798 int wait_for_unblock)
1800 struct btrfs_async_commit *ac;
1801 struct btrfs_transaction *cur_trans;
1803 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1807 INIT_WORK(&ac->work, do_async_commit);
1809 ac->newtrans = btrfs_join_transaction(root);
1810 if (IS_ERR(ac->newtrans)) {
1811 int err = PTR_ERR(ac->newtrans);
1816 /* take transaction reference */
1817 cur_trans = trans->transaction;
1818 atomic_inc(&cur_trans->use_count);
1820 btrfs_end_transaction(trans, root);
1823 * Tell lockdep we've released the freeze rwsem, since the
1824 * async commit thread will be the one to unlock it.
1826 if (ac->newtrans->type & __TRANS_FREEZABLE)
1827 __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS);
1829 schedule_work(&ac->work);
1831 /* wait for transaction to start and unblock */
1832 if (wait_for_unblock)
1833 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1835 wait_current_trans_commit_start(root, cur_trans);
1837 if (current->journal_info == trans)
1838 current->journal_info = NULL;
1840 btrfs_put_transaction(cur_trans);
1845 static void cleanup_transaction(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root, int err)
1848 struct btrfs_transaction *cur_trans = trans->transaction;
1851 WARN_ON(trans->use_count > 1);
1853 btrfs_abort_transaction(trans, err);
1855 spin_lock(&root->fs_info->trans_lock);
1858 * If the transaction is removed from the list, it means this
1859 * transaction has been committed successfully, so it is impossible
1860 * to call the cleanup function.
1862 BUG_ON(list_empty(&cur_trans->list));
1864 list_del_init(&cur_trans->list);
1865 if (cur_trans == root->fs_info->running_transaction) {
1866 cur_trans->state = TRANS_STATE_COMMIT_DOING;
1867 spin_unlock(&root->fs_info->trans_lock);
1868 wait_event(cur_trans->writer_wait,
1869 atomic_read(&cur_trans->num_writers) == 1);
1871 spin_lock(&root->fs_info->trans_lock);
1873 spin_unlock(&root->fs_info->trans_lock);
1875 btrfs_cleanup_one_transaction(trans->transaction, root);
1877 spin_lock(&root->fs_info->trans_lock);
1878 if (cur_trans == root->fs_info->running_transaction)
1879 root->fs_info->running_transaction = NULL;
1880 spin_unlock(&root->fs_info->trans_lock);
1882 if (trans->type & __TRANS_FREEZABLE)
1883 sb_end_intwrite(root->fs_info->sb);
1884 btrfs_put_transaction(cur_trans);
1885 btrfs_put_transaction(cur_trans);
1887 trace_btrfs_transaction_commit(root);
1889 if (current->journal_info == trans)
1890 current->journal_info = NULL;
1891 btrfs_scrub_cancel(root->fs_info);
1893 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1896 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
1898 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
1899 return btrfs_start_delalloc_roots(fs_info, 1, -1);
1903 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
1905 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
1906 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
1910 btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans)
1912 wait_event(cur_trans->pending_wait,
1913 atomic_read(&cur_trans->pending_ordered) == 0);
1916 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1917 struct btrfs_root *root)
1919 struct btrfs_transaction *cur_trans = trans->transaction;
1920 struct btrfs_transaction *prev_trans = NULL;
1923 /* Stop the commit early if ->aborted is set */
1924 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1925 ret = cur_trans->aborted;
1926 btrfs_end_transaction(trans, root);
1930 /* make a pass through all the delayed refs we have so far
1931 * any runnings procs may add more while we are here
1933 ret = btrfs_run_delayed_refs(trans, root, 0);
1935 btrfs_end_transaction(trans, root);
1939 btrfs_trans_release_metadata(trans, root);
1940 trans->block_rsv = NULL;
1942 cur_trans = trans->transaction;
1945 * set the flushing flag so procs in this transaction have to
1946 * start sending their work down.
1948 cur_trans->delayed_refs.flushing = 1;
1951 if (!list_empty(&trans->new_bgs))
1952 btrfs_create_pending_block_groups(trans, root);
1954 ret = btrfs_run_delayed_refs(trans, root, 0);
1956 btrfs_end_transaction(trans, root);
1960 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
1963 /* this mutex is also taken before trying to set
1964 * block groups readonly. We need to make sure
1965 * that nobody has set a block group readonly
1966 * after a extents from that block group have been
1967 * allocated for cache files. btrfs_set_block_group_ro
1968 * will wait for the transaction to commit if it
1969 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
1971 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
1972 * only one process starts all the block group IO. It wouldn't
1973 * hurt to have more than one go through, but there's no
1974 * real advantage to it either.
1976 mutex_lock(&root->fs_info->ro_block_group_mutex);
1977 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
1980 mutex_unlock(&root->fs_info->ro_block_group_mutex);
1983 ret = btrfs_start_dirty_block_groups(trans, root);
1986 btrfs_end_transaction(trans, root);
1990 spin_lock(&root->fs_info->trans_lock);
1991 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
1992 spin_unlock(&root->fs_info->trans_lock);
1993 atomic_inc(&cur_trans->use_count);
1994 ret = btrfs_end_transaction(trans, root);
1996 wait_for_commit(root, cur_trans);
1998 if (unlikely(cur_trans->aborted))
1999 ret = cur_trans->aborted;
2001 btrfs_put_transaction(cur_trans);
2006 cur_trans->state = TRANS_STATE_COMMIT_START;
2007 wake_up(&root->fs_info->transaction_blocked_wait);
2009 if (cur_trans->list.prev != &root->fs_info->trans_list) {
2010 prev_trans = list_entry(cur_trans->list.prev,
2011 struct btrfs_transaction, list);
2012 if (prev_trans->state != TRANS_STATE_COMPLETED) {
2013 atomic_inc(&prev_trans->use_count);
2014 spin_unlock(&root->fs_info->trans_lock);
2016 wait_for_commit(root, prev_trans);
2017 ret = prev_trans->aborted;
2019 btrfs_put_transaction(prev_trans);
2021 goto cleanup_transaction;
2023 spin_unlock(&root->fs_info->trans_lock);
2026 spin_unlock(&root->fs_info->trans_lock);
2029 extwriter_counter_dec(cur_trans, trans->type);
2031 ret = btrfs_start_delalloc_flush(root->fs_info);
2033 goto cleanup_transaction;
2035 ret = btrfs_run_delayed_items(trans, root);
2037 goto cleanup_transaction;
2039 wait_event(cur_trans->writer_wait,
2040 extwriter_counter_read(cur_trans) == 0);
2042 /* some pending stuffs might be added after the previous flush. */
2043 ret = btrfs_run_delayed_items(trans, root);
2045 goto cleanup_transaction;
2047 btrfs_wait_delalloc_flush(root->fs_info);
2049 btrfs_wait_pending_ordered(cur_trans);
2051 btrfs_scrub_pause(root);
2053 * Ok now we need to make sure to block out any other joins while we
2054 * commit the transaction. We could have started a join before setting
2055 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2057 spin_lock(&root->fs_info->trans_lock);
2058 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2059 spin_unlock(&root->fs_info->trans_lock);
2060 wait_event(cur_trans->writer_wait,
2061 atomic_read(&cur_trans->num_writers) == 1);
2063 /* ->aborted might be set after the previous check, so check it */
2064 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2065 ret = cur_trans->aborted;
2066 goto scrub_continue;
2069 * the reloc mutex makes sure that we stop
2070 * the balancing code from coming in and moving
2071 * extents around in the middle of the commit
2073 mutex_lock(&root->fs_info->reloc_mutex);
2076 * We needn't worry about the delayed items because we will
2077 * deal with them in create_pending_snapshot(), which is the
2078 * core function of the snapshot creation.
2080 ret = create_pending_snapshots(trans, root->fs_info);
2082 mutex_unlock(&root->fs_info->reloc_mutex);
2083 goto scrub_continue;
2087 * We insert the dir indexes of the snapshots and update the inode
2088 * of the snapshots' parents after the snapshot creation, so there
2089 * are some delayed items which are not dealt with. Now deal with
2092 * We needn't worry that this operation will corrupt the snapshots,
2093 * because all the tree which are snapshoted will be forced to COW
2094 * the nodes and leaves.
2096 ret = btrfs_run_delayed_items(trans, root);
2098 mutex_unlock(&root->fs_info->reloc_mutex);
2099 goto scrub_continue;
2102 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
2104 mutex_unlock(&root->fs_info->reloc_mutex);
2105 goto scrub_continue;
2108 /* Reocrd old roots for later qgroup accounting */
2109 ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info);
2111 mutex_unlock(&root->fs_info->reloc_mutex);
2112 goto scrub_continue;
2116 * make sure none of the code above managed to slip in a
2119 btrfs_assert_delayed_root_empty(root);
2121 WARN_ON(cur_trans != trans->transaction);
2123 /* btrfs_commit_tree_roots is responsible for getting the
2124 * various roots consistent with each other. Every pointer
2125 * in the tree of tree roots has to point to the most up to date
2126 * root for every subvolume and other tree. So, we have to keep
2127 * the tree logging code from jumping in and changing any
2130 * At this point in the commit, there can't be any tree-log
2131 * writers, but a little lower down we drop the trans mutex
2132 * and let new people in. By holding the tree_log_mutex
2133 * from now until after the super is written, we avoid races
2134 * with the tree-log code.
2136 mutex_lock(&root->fs_info->tree_log_mutex);
2138 ret = commit_fs_roots(trans, root);
2140 mutex_unlock(&root->fs_info->tree_log_mutex);
2141 mutex_unlock(&root->fs_info->reloc_mutex);
2142 goto scrub_continue;
2146 * Since the transaction is done, we can apply the pending changes
2147 * before the next transaction.
2149 btrfs_apply_pending_changes(root->fs_info);
2151 /* commit_fs_roots gets rid of all the tree log roots, it is now
2152 * safe to free the root of tree log roots
2154 btrfs_free_log_root_tree(trans, root->fs_info);
2157 * Since fs roots are all committed, we can get a quite accurate
2158 * new_roots. So let's do quota accounting.
2160 ret = btrfs_qgroup_account_extents(trans, root->fs_info);
2162 mutex_unlock(&root->fs_info->tree_log_mutex);
2163 mutex_unlock(&root->fs_info->reloc_mutex);
2164 goto scrub_continue;
2167 ret = commit_cowonly_roots(trans, root);
2169 mutex_unlock(&root->fs_info->tree_log_mutex);
2170 mutex_unlock(&root->fs_info->reloc_mutex);
2171 goto scrub_continue;
2175 * The tasks which save the space cache and inode cache may also
2176 * update ->aborted, check it.
2178 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
2179 ret = cur_trans->aborted;
2180 mutex_unlock(&root->fs_info->tree_log_mutex);
2181 mutex_unlock(&root->fs_info->reloc_mutex);
2182 goto scrub_continue;
2185 btrfs_prepare_extent_commit(trans, root);
2187 cur_trans = root->fs_info->running_transaction;
2189 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
2190 root->fs_info->tree_root->node);
2191 list_add_tail(&root->fs_info->tree_root->dirty_list,
2192 &cur_trans->switch_commits);
2194 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
2195 root->fs_info->chunk_root->node);
2196 list_add_tail(&root->fs_info->chunk_root->dirty_list,
2197 &cur_trans->switch_commits);
2199 switch_commit_roots(cur_trans, root->fs_info);
2201 assert_qgroups_uptodate(trans);
2202 ASSERT(list_empty(&cur_trans->dirty_bgs));
2203 ASSERT(list_empty(&cur_trans->io_bgs));
2204 update_super_roots(root);
2206 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
2207 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
2208 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
2209 sizeof(*root->fs_info->super_copy));
2211 btrfs_update_commit_device_size(root->fs_info);
2212 btrfs_update_commit_device_bytes_used(root, cur_trans);
2214 clear_bit(BTRFS_FS_LOG1_ERR, &root->fs_info->flags);
2215 clear_bit(BTRFS_FS_LOG2_ERR, &root->fs_info->flags);
2217 btrfs_trans_release_chunk_metadata(trans);
2219 spin_lock(&root->fs_info->trans_lock);
2220 cur_trans->state = TRANS_STATE_UNBLOCKED;
2221 root->fs_info->running_transaction = NULL;
2222 spin_unlock(&root->fs_info->trans_lock);
2223 mutex_unlock(&root->fs_info->reloc_mutex);
2225 wake_up(&root->fs_info->transaction_wait);
2227 ret = btrfs_write_and_wait_transaction(trans, root);
2229 btrfs_handle_fs_error(root->fs_info, ret,
2230 "Error while writing out transaction");
2231 mutex_unlock(&root->fs_info->tree_log_mutex);
2232 goto scrub_continue;
2235 ret = write_ctree_super(trans, root, 0);
2237 mutex_unlock(&root->fs_info->tree_log_mutex);
2238 goto scrub_continue;
2242 * the super is written, we can safely allow the tree-loggers
2243 * to go about their business
2245 mutex_unlock(&root->fs_info->tree_log_mutex);
2247 btrfs_finish_extent_commit(trans, root);
2249 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2250 btrfs_clear_space_info_full(root->fs_info);
2252 root->fs_info->last_trans_committed = cur_trans->transid;
2254 * We needn't acquire the lock here because there is no other task
2255 * which can change it.
2257 cur_trans->state = TRANS_STATE_COMPLETED;
2258 wake_up(&cur_trans->commit_wait);
2260 spin_lock(&root->fs_info->trans_lock);
2261 list_del_init(&cur_trans->list);
2262 spin_unlock(&root->fs_info->trans_lock);
2264 btrfs_put_transaction(cur_trans);
2265 btrfs_put_transaction(cur_trans);
2267 if (trans->type & __TRANS_FREEZABLE)
2268 sb_end_intwrite(root->fs_info->sb);
2270 trace_btrfs_transaction_commit(root);
2272 btrfs_scrub_continue(root);
2274 if (current->journal_info == trans)
2275 current->journal_info = NULL;
2277 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2280 * If fs has been frozen, we can not handle delayed iputs, otherwise
2281 * it'll result in deadlock about SB_FREEZE_FS.
2283 if (current != root->fs_info->transaction_kthread &&
2284 current != root->fs_info->cleaner_kthread &&
2285 !root->fs_info->fs_frozen)
2286 btrfs_run_delayed_iputs(root);
2291 btrfs_scrub_continue(root);
2292 cleanup_transaction:
2293 btrfs_trans_release_metadata(trans, root);
2294 btrfs_trans_release_chunk_metadata(trans);
2295 trans->block_rsv = NULL;
2296 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
2297 if (current->journal_info == trans)
2298 current->journal_info = NULL;
2299 cleanup_transaction(trans, root, ret);
2305 * return < 0 if error
2306 * 0 if there are no more dead_roots at the time of call
2307 * 1 there are more to be processed, call me again
2309 * The return value indicates there are certainly more snapshots to delete, but
2310 * if there comes a new one during processing, it may return 0. We don't mind,
2311 * because btrfs_commit_super will poke cleaner thread and it will process it a
2312 * few seconds later.
2314 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2317 struct btrfs_fs_info *fs_info = root->fs_info;
2319 spin_lock(&fs_info->trans_lock);
2320 if (list_empty(&fs_info->dead_roots)) {
2321 spin_unlock(&fs_info->trans_lock);
2324 root = list_first_entry(&fs_info->dead_roots,
2325 struct btrfs_root, root_list);
2326 list_del_init(&root->root_list);
2327 spin_unlock(&fs_info->trans_lock);
2329 pr_debug("BTRFS: cleaner removing %llu\n", root->objectid);
2331 btrfs_kill_all_delayed_nodes(root);
2333 if (btrfs_header_backref_rev(root->node) <
2334 BTRFS_MIXED_BACKREF_REV)
2335 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
2337 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
2339 return (ret < 0) ? 0 : 1;
2342 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2347 prev = xchg(&fs_info->pending_changes, 0);
2351 bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE;
2353 btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2356 bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE;
2358 btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE);
2361 bit = 1 << BTRFS_PENDING_COMMIT;
2363 btrfs_debug(fs_info, "pending commit done");
2368 "unknown pending changes left 0x%lx, ignoring", prev);