1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/writeback.h>
10 #include <linux/pagemap.h>
11 #include <linux/blkdev.h>
12 #include <linux/uuid.h>
16 #include "transaction.h"
20 #include "dev-replace.h"
22 #include "block-group.h"
23 #include "space-info.h"
26 #define BTRFS_ROOT_TRANS_TAG 0
29 * Transaction states and transitions
31 * No running transaction (fs tree blocks are not modified)
34 * | Call start_transaction() variants. Except btrfs_join_transaction_nostart().
36 * Transaction N [[TRANS_STATE_RUNNING]]
38 * | New trans handles can be attached to transaction N by calling all
39 * | start_transaction() variants.
42 * | Call btrfs_commit_transaction() on any trans handle attached to
45 * Transaction N [[TRANS_STATE_COMMIT_START]]
47 * | Will wait for previous running transaction to completely finish if there
50 * | Then one of the following happes:
51 * | - Wait for all other trans handle holders to release.
52 * | The btrfs_commit_transaction() caller will do the commit work.
53 * | - Wait for current transaction to be committed by others.
54 * | Other btrfs_commit_transaction() caller will do the commit work.
56 * | At this stage, only btrfs_join_transaction*() variants can attach
57 * | to this running transaction.
58 * | All other variants will wait for current one to finish and attach to
62 * | Caller is chosen to commit transaction N, and all other trans handle
63 * | haven been released.
65 * Transaction N [[TRANS_STATE_COMMIT_DOING]]
67 * | The heavy lifting transaction work is started.
68 * | From running delayed refs (modifying extent tree) to creating pending
69 * | snapshots, running qgroups.
70 * | In short, modify supporting trees to reflect modifications of subvolume
73 * | At this stage, all start_transaction() calls will wait for this
74 * | transaction to finish and attach to transaction N+1.
77 * | Until all supporting trees are updated.
79 * Transaction N [[TRANS_STATE_UNBLOCKED]]
81 * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
82 * | need to write them back to disk and update |
85 * | At this stage, new transaction is allowed to |
87 * | All new start_transaction() calls will be |
88 * | attached to transid N+1. |
91 * | Until all tree blocks are super blocks are |
92 * | written to block devices |
94 * Transaction N [[TRANS_STATE_COMPLETED]] V
95 * All tree blocks and super blocks are written. Transaction N+1
96 * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]]
97 * data structures will be cleaned up. | Life goes on
99 static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
100 [TRANS_STATE_RUNNING] = 0U,
101 [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH),
102 [TRANS_STATE_COMMIT_DOING] = (__TRANS_START |
105 __TRANS_JOIN_NOSTART),
106 [TRANS_STATE_UNBLOCKED] = (__TRANS_START |
109 __TRANS_JOIN_NOLOCK |
110 __TRANS_JOIN_NOSTART),
111 [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START |
114 __TRANS_JOIN_NOLOCK |
115 __TRANS_JOIN_NOSTART),
116 [TRANS_STATE_COMPLETED] = (__TRANS_START |
119 __TRANS_JOIN_NOLOCK |
120 __TRANS_JOIN_NOSTART),
123 void btrfs_put_transaction(struct btrfs_transaction *transaction)
125 WARN_ON(refcount_read(&transaction->use_count) == 0);
126 if (refcount_dec_and_test(&transaction->use_count)) {
127 BUG_ON(!list_empty(&transaction->list));
128 WARN_ON(!RB_EMPTY_ROOT(
129 &transaction->delayed_refs.href_root.rb_root));
130 WARN_ON(!RB_EMPTY_ROOT(
131 &transaction->delayed_refs.dirty_extent_root));
132 if (transaction->delayed_refs.pending_csums)
133 btrfs_err(transaction->fs_info,
134 "pending csums is %llu",
135 transaction->delayed_refs.pending_csums);
137 * If any block groups are found in ->deleted_bgs then it's
138 * because the transaction was aborted and a commit did not
139 * happen (things failed before writing the new superblock
140 * and calling btrfs_finish_extent_commit()), so we can not
141 * discard the physical locations of the block groups.
143 while (!list_empty(&transaction->deleted_bgs)) {
144 struct btrfs_block_group *cache;
146 cache = list_first_entry(&transaction->deleted_bgs,
147 struct btrfs_block_group,
149 list_del_init(&cache->bg_list);
150 btrfs_unfreeze_block_group(cache);
151 btrfs_put_block_group(cache);
153 WARN_ON(!list_empty(&transaction->dev_update_list));
158 static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
160 struct btrfs_transaction *cur_trans = trans->transaction;
161 struct btrfs_fs_info *fs_info = trans->fs_info;
162 struct btrfs_root *root, *tmp;
163 struct btrfs_caching_control *caching_ctl, *next;
165 down_write(&fs_info->commit_root_sem);
167 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
168 fs_info->last_reloc_trans = trans->transid;
170 list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
172 list_del_init(&root->dirty_list);
173 free_extent_buffer(root->commit_root);
174 root->commit_root = btrfs_root_node(root);
175 extent_io_tree_release(&root->dirty_log_pages);
176 btrfs_qgroup_clean_swapped_blocks(root);
179 /* We can free old roots now. */
180 spin_lock(&cur_trans->dropped_roots_lock);
181 while (!list_empty(&cur_trans->dropped_roots)) {
182 root = list_first_entry(&cur_trans->dropped_roots,
183 struct btrfs_root, root_list);
184 list_del_init(&root->root_list);
185 spin_unlock(&cur_trans->dropped_roots_lock);
186 btrfs_free_log(trans, root);
187 btrfs_drop_and_free_fs_root(fs_info, root);
188 spin_lock(&cur_trans->dropped_roots_lock);
190 spin_unlock(&cur_trans->dropped_roots_lock);
193 * We have to update the last_byte_to_unpin under the commit_root_sem,
194 * at the same time we swap out the commit roots.
196 * This is because we must have a real view of the last spot the caching
197 * kthreads were while caching. Consider the following views of the
198 * extent tree for a block group
201 * +----+----+----+----+----+----+----+
202 * |\\\\| |\\\\|\\\\| |\\\\|\\\\|
203 * +----+----+----+----+----+----+----+
207 * +----+----+----+----+----+----+----+
208 * | | | |\\\\| | |\\\\|
209 * +----+----+----+----+----+----+----+
212 * If the cache_ctl->progress was at 3, then we are only allowed to
213 * unpin [0,1) and [2,3], because the caching thread has already
214 * processed those extents. We are not allowed to unpin [5,6), because
215 * the caching thread will re-start it's search from 3, and thus find
216 * the hole from [4,6) to add to the free space cache.
218 spin_lock(&fs_info->block_group_cache_lock);
219 list_for_each_entry_safe(caching_ctl, next,
220 &fs_info->caching_block_groups, list) {
221 struct btrfs_block_group *cache = caching_ctl->block_group;
223 if (btrfs_block_group_done(cache)) {
224 cache->last_byte_to_unpin = (u64)-1;
225 list_del_init(&caching_ctl->list);
226 btrfs_put_caching_control(caching_ctl);
228 cache->last_byte_to_unpin = caching_ctl->progress;
231 spin_unlock(&fs_info->block_group_cache_lock);
232 up_write(&fs_info->commit_root_sem);
235 static inline void extwriter_counter_inc(struct btrfs_transaction *trans,
238 if (type & TRANS_EXTWRITERS)
239 atomic_inc(&trans->num_extwriters);
242 static inline void extwriter_counter_dec(struct btrfs_transaction *trans,
245 if (type & TRANS_EXTWRITERS)
246 atomic_dec(&trans->num_extwriters);
249 static inline void extwriter_counter_init(struct btrfs_transaction *trans,
252 atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0));
255 static inline int extwriter_counter_read(struct btrfs_transaction *trans)
257 return atomic_read(&trans->num_extwriters);
261 * To be called after doing the chunk btree updates right after allocating a new
262 * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a
263 * chunk after all chunk btree updates and after finishing the second phase of
264 * chunk allocation (btrfs_create_pending_block_groups()) in case some block
265 * group had its chunk item insertion delayed to the second phase.
267 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
269 struct btrfs_fs_info *fs_info = trans->fs_info;
271 if (!trans->chunk_bytes_reserved)
274 btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
275 trans->chunk_bytes_reserved, NULL);
276 trans->chunk_bytes_reserved = 0;
280 * either allocate a new transaction or hop into the existing one
282 static noinline int join_transaction(struct btrfs_fs_info *fs_info,
285 struct btrfs_transaction *cur_trans;
287 spin_lock(&fs_info->trans_lock);
289 /* The file system has been taken offline. No new transactions. */
290 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
291 spin_unlock(&fs_info->trans_lock);
295 cur_trans = fs_info->running_transaction;
297 if (TRANS_ABORTED(cur_trans)) {
298 spin_unlock(&fs_info->trans_lock);
299 return cur_trans->aborted;
301 if (btrfs_blocked_trans_types[cur_trans->state] & type) {
302 spin_unlock(&fs_info->trans_lock);
305 refcount_inc(&cur_trans->use_count);
306 atomic_inc(&cur_trans->num_writers);
307 extwriter_counter_inc(cur_trans, type);
308 spin_unlock(&fs_info->trans_lock);
311 spin_unlock(&fs_info->trans_lock);
314 * If we are ATTACH, we just want to catch the current transaction,
315 * and commit it. If there is no transaction, just return ENOENT.
317 if (type == TRANS_ATTACH)
321 * JOIN_NOLOCK only happens during the transaction commit, so
322 * it is impossible that ->running_transaction is NULL
324 BUG_ON(type == TRANS_JOIN_NOLOCK);
326 cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS);
330 spin_lock(&fs_info->trans_lock);
331 if (fs_info->running_transaction) {
333 * someone started a transaction after we unlocked. Make sure
334 * to redo the checks above
338 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
339 spin_unlock(&fs_info->trans_lock);
344 cur_trans->fs_info = fs_info;
345 atomic_set(&cur_trans->pending_ordered, 0);
346 init_waitqueue_head(&cur_trans->pending_wait);
347 atomic_set(&cur_trans->num_writers, 1);
348 extwriter_counter_init(cur_trans, type);
349 init_waitqueue_head(&cur_trans->writer_wait);
350 init_waitqueue_head(&cur_trans->commit_wait);
351 cur_trans->state = TRANS_STATE_RUNNING;
353 * One for this trans handle, one so it will live on until we
354 * commit the transaction.
356 refcount_set(&cur_trans->use_count, 2);
357 cur_trans->flags = 0;
358 cur_trans->start_time = ktime_get_seconds();
360 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
362 cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
363 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
364 atomic_set(&cur_trans->delayed_refs.num_entries, 0);
367 * although the tree mod log is per file system and not per transaction,
368 * the log must never go across transaction boundaries.
371 if (!list_empty(&fs_info->tree_mod_seq_list))
372 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n");
373 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
374 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n");
375 atomic64_set(&fs_info->tree_mod_seq, 0);
377 spin_lock_init(&cur_trans->delayed_refs.lock);
379 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
380 INIT_LIST_HEAD(&cur_trans->dev_update_list);
381 INIT_LIST_HEAD(&cur_trans->switch_commits);
382 INIT_LIST_HEAD(&cur_trans->dirty_bgs);
383 INIT_LIST_HEAD(&cur_trans->io_bgs);
384 INIT_LIST_HEAD(&cur_trans->dropped_roots);
385 mutex_init(&cur_trans->cache_write_mutex);
386 spin_lock_init(&cur_trans->dirty_bgs_lock);
387 INIT_LIST_HEAD(&cur_trans->deleted_bgs);
388 spin_lock_init(&cur_trans->dropped_roots_lock);
389 INIT_LIST_HEAD(&cur_trans->releasing_ebs);
390 spin_lock_init(&cur_trans->releasing_ebs_lock);
391 list_add_tail(&cur_trans->list, &fs_info->trans_list);
392 extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
393 IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode);
394 extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
395 IO_TREE_FS_PINNED_EXTENTS, NULL);
396 fs_info->generation++;
397 cur_trans->transid = fs_info->generation;
398 fs_info->running_transaction = cur_trans;
399 cur_trans->aborted = 0;
400 spin_unlock(&fs_info->trans_lock);
406 * This does all the record keeping required to make sure that a shareable root
407 * is properly recorded in a given transaction. This is required to make sure
408 * the old root from before we joined the transaction is deleted when the
409 * transaction commits.
411 static int record_root_in_trans(struct btrfs_trans_handle *trans,
412 struct btrfs_root *root,
415 struct btrfs_fs_info *fs_info = root->fs_info;
418 if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
419 root->last_trans < trans->transid) || force) {
420 WARN_ON(root == fs_info->extent_root);
421 WARN_ON(!force && root->commit_root != root->node);
424 * see below for IN_TRANS_SETUP usage rules
425 * we have the reloc mutex held now, so there
426 * is only one writer in this function
428 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
430 /* make sure readers find IN_TRANS_SETUP before
431 * they find our root->last_trans update
435 spin_lock(&fs_info->fs_roots_radix_lock);
436 if (root->last_trans == trans->transid && !force) {
437 spin_unlock(&fs_info->fs_roots_radix_lock);
440 radix_tree_tag_set(&fs_info->fs_roots_radix,
441 (unsigned long)root->root_key.objectid,
442 BTRFS_ROOT_TRANS_TAG);
443 spin_unlock(&fs_info->fs_roots_radix_lock);
444 root->last_trans = trans->transid;
446 /* this is pretty tricky. We don't want to
447 * take the relocation lock in btrfs_record_root_in_trans
448 * unless we're really doing the first setup for this root in
451 * Normally we'd use root->last_trans as a flag to decide
452 * if we want to take the expensive mutex.
454 * But, we have to set root->last_trans before we
455 * init the relocation root, otherwise, we trip over warnings
456 * in ctree.c. The solution used here is to flag ourselves
457 * with root IN_TRANS_SETUP. When this is 1, we're still
458 * fixing up the reloc trees and everyone must wait.
460 * When this is zero, they can trust root->last_trans and fly
461 * through btrfs_record_root_in_trans without having to take the
462 * lock. smp_wmb() makes sure that all the writes above are
463 * done before we pop in the zero below
465 ret = btrfs_init_reloc_root(trans, root);
466 smp_mb__before_atomic();
467 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state);
473 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
474 struct btrfs_root *root)
476 struct btrfs_fs_info *fs_info = root->fs_info;
477 struct btrfs_transaction *cur_trans = trans->transaction;
479 /* Add ourselves to the transaction dropped list */
480 spin_lock(&cur_trans->dropped_roots_lock);
481 list_add_tail(&root->root_list, &cur_trans->dropped_roots);
482 spin_unlock(&cur_trans->dropped_roots_lock);
484 /* Make sure we don't try to update the root at commit time */
485 spin_lock(&fs_info->fs_roots_radix_lock);
486 radix_tree_tag_clear(&fs_info->fs_roots_radix,
487 (unsigned long)root->root_key.objectid,
488 BTRFS_ROOT_TRANS_TAG);
489 spin_unlock(&fs_info->fs_roots_radix_lock);
492 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
493 struct btrfs_root *root)
495 struct btrfs_fs_info *fs_info = root->fs_info;
498 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
502 * see record_root_in_trans for comments about IN_TRANS_SETUP usage
506 if (root->last_trans == trans->transid &&
507 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state))
510 mutex_lock(&fs_info->reloc_mutex);
511 ret = record_root_in_trans(trans, root, 0);
512 mutex_unlock(&fs_info->reloc_mutex);
517 static inline int is_transaction_blocked(struct btrfs_transaction *trans)
519 return (trans->state >= TRANS_STATE_COMMIT_START &&
520 trans->state < TRANS_STATE_UNBLOCKED &&
521 !TRANS_ABORTED(trans));
524 /* wait for commit against the current transaction to become unblocked
525 * when this is done, it is safe to start a new transaction, but the current
526 * transaction might not be fully on disk.
528 static void wait_current_trans(struct btrfs_fs_info *fs_info)
530 struct btrfs_transaction *cur_trans;
532 spin_lock(&fs_info->trans_lock);
533 cur_trans = fs_info->running_transaction;
534 if (cur_trans && is_transaction_blocked(cur_trans)) {
535 refcount_inc(&cur_trans->use_count);
536 spin_unlock(&fs_info->trans_lock);
538 wait_event(fs_info->transaction_wait,
539 cur_trans->state >= TRANS_STATE_UNBLOCKED ||
540 TRANS_ABORTED(cur_trans));
541 btrfs_put_transaction(cur_trans);
543 spin_unlock(&fs_info->trans_lock);
547 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type)
549 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
552 if (type == TRANS_START)
558 static inline bool need_reserve_reloc_root(struct btrfs_root *root)
560 struct btrfs_fs_info *fs_info = root->fs_info;
562 if (!fs_info->reloc_ctl ||
563 !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
564 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
571 static struct btrfs_trans_handle *
572 start_transaction(struct btrfs_root *root, unsigned int num_items,
573 unsigned int type, enum btrfs_reserve_flush_enum flush,
574 bool enforce_qgroups)
576 struct btrfs_fs_info *fs_info = root->fs_info;
577 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
578 struct btrfs_trans_handle *h;
579 struct btrfs_transaction *cur_trans;
581 u64 qgroup_reserved = 0;
582 bool reloc_reserved = false;
583 bool do_chunk_alloc = false;
586 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
587 return ERR_PTR(-EROFS);
589 if (current->journal_info) {
590 WARN_ON(type & TRANS_EXTWRITERS);
591 h = current->journal_info;
592 refcount_inc(&h->use_count);
593 WARN_ON(refcount_read(&h->use_count) > 2);
594 h->orig_rsv = h->block_rsv;
600 * Do the reservation before we join the transaction so we can do all
601 * the appropriate flushing if need be.
603 if (num_items && root != fs_info->chunk_root) {
604 struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
605 u64 delayed_refs_bytes = 0;
607 qgroup_reserved = num_items * fs_info->nodesize;
608 ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
614 * We want to reserve all the bytes we may need all at once, so
615 * we only do 1 enospc flushing cycle per transaction start. We
616 * accomplish this by simply assuming we'll do 2 x num_items
617 * worth of delayed refs updates in this trans handle, and
618 * refill that amount for whatever is missing in the reserve.
620 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
621 if (flush == BTRFS_RESERVE_FLUSH_ALL &&
622 delayed_refs_rsv->full == 0) {
623 delayed_refs_bytes = num_bytes;
628 * Do the reservation for the relocation root creation
630 if (need_reserve_reloc_root(root)) {
631 num_bytes += fs_info->nodesize;
632 reloc_reserved = true;
635 ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
638 if (delayed_refs_bytes) {
639 btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
641 num_bytes -= delayed_refs_bytes;
644 if (rsv->space_info->force_alloc)
645 do_chunk_alloc = true;
646 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
647 !delayed_refs_rsv->full) {
649 * Some people call with btrfs_start_transaction(root, 0)
650 * because they can be throttled, but have some other mechanism
651 * for reserving space. We still want these guys to refill the
652 * delayed block_rsv so just add 1 items worth of reservation
655 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
660 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS);
667 * If we are JOIN_NOLOCK we're already committing a transaction and
668 * waiting on this guy, so we don't need to do the sb_start_intwrite
669 * because we're already holding a ref. We need this because we could
670 * have raced in and did an fsync() on a file which can kick a commit
671 * and then we deadlock with somebody doing a freeze.
673 * If we are ATTACH, it means we just want to catch the current
674 * transaction and commit it, so we needn't do sb_start_intwrite().
676 if (type & __TRANS_FREEZABLE)
677 sb_start_intwrite(fs_info->sb);
679 if (may_wait_transaction(fs_info, type))
680 wait_current_trans(fs_info);
683 ret = join_transaction(fs_info, type);
685 wait_current_trans(fs_info);
686 if (unlikely(type == TRANS_ATTACH ||
687 type == TRANS_JOIN_NOSTART))
690 } while (ret == -EBUSY);
695 cur_trans = fs_info->running_transaction;
697 h->transid = cur_trans->transid;
698 h->transaction = cur_trans;
700 refcount_set(&h->use_count, 1);
701 h->fs_info = root->fs_info;
704 INIT_LIST_HEAD(&h->new_bgs);
707 if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
708 may_wait_transaction(fs_info, type)) {
709 current->journal_info = h;
710 btrfs_commit_transaction(h);
715 trace_btrfs_space_reservation(fs_info, "transaction",
716 h->transid, num_bytes, 1);
717 h->block_rsv = &fs_info->trans_block_rsv;
718 h->bytes_reserved = num_bytes;
719 h->reloc_reserved = reloc_reserved;
723 if (!current->journal_info)
724 current->journal_info = h;
727 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
728 * ALLOC_FORCE the first run through, and then we won't allocate for
729 * anybody else who races in later. We don't care about the return
732 if (do_chunk_alloc && num_bytes) {
733 u64 flags = h->block_rsv->space_info->flags;
735 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
736 CHUNK_ALLOC_NO_FORCE);
740 * btrfs_record_root_in_trans() needs to alloc new extents, and may
741 * call btrfs_join_transaction() while we're also starting a
744 * Thus it need to be called after current->journal_info initialized,
745 * or we can deadlock.
747 ret = btrfs_record_root_in_trans(h, root);
750 * The transaction handle is fully initialized and linked with
751 * other structures so it needs to be ended in case of errors,
754 btrfs_end_transaction(h);
761 if (type & __TRANS_FREEZABLE)
762 sb_end_intwrite(fs_info->sb);
763 kmem_cache_free(btrfs_trans_handle_cachep, h);
766 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
769 btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
773 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
774 unsigned int num_items)
776 return start_transaction(root, num_items, TRANS_START,
777 BTRFS_RESERVE_FLUSH_ALL, true);
780 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
781 struct btrfs_root *root,
782 unsigned int num_items)
784 return start_transaction(root, num_items, TRANS_START,
785 BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
788 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
790 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH,
794 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root)
796 return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
797 BTRFS_RESERVE_NO_FLUSH, true);
801 * Similar to regular join but it never starts a transaction when none is
802 * running or after waiting for the current one to finish.
804 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root)
806 return start_transaction(root, 0, TRANS_JOIN_NOSTART,
807 BTRFS_RESERVE_NO_FLUSH, true);
811 * btrfs_attach_transaction() - catch the running transaction
813 * It is used when we want to commit the current the transaction, but
814 * don't want to start a new one.
816 * Note: If this function return -ENOENT, it just means there is no
817 * running transaction. But it is possible that the inactive transaction
818 * is still in the memory, not fully on disk. If you hope there is no
819 * inactive transaction in the fs when -ENOENT is returned, you should
821 * btrfs_attach_transaction_barrier()
823 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
825 return start_transaction(root, 0, TRANS_ATTACH,
826 BTRFS_RESERVE_NO_FLUSH, true);
830 * btrfs_attach_transaction_barrier() - catch the running transaction
832 * It is similar to the above function, the difference is this one
833 * will wait for all the inactive transactions until they fully
836 struct btrfs_trans_handle *
837 btrfs_attach_transaction_barrier(struct btrfs_root *root)
839 struct btrfs_trans_handle *trans;
841 trans = start_transaction(root, 0, TRANS_ATTACH,
842 BTRFS_RESERVE_NO_FLUSH, true);
843 if (trans == ERR_PTR(-ENOENT))
844 btrfs_wait_for_commit(root->fs_info, 0);
849 /* Wait for a transaction commit to reach at least the given state. */
850 static noinline void wait_for_commit(struct btrfs_transaction *commit,
851 const enum btrfs_trans_state min_state)
853 struct btrfs_fs_info *fs_info = commit->fs_info;
854 u64 transid = commit->transid;
858 wait_event(commit->commit_wait, commit->state >= min_state);
860 btrfs_put_transaction(commit);
862 if (min_state < TRANS_STATE_COMPLETED)
866 * A transaction isn't really completed until all of the
867 * previous transactions are completed, but with fsync we can
868 * end up with SUPER_COMMITTED transactions before a COMPLETED
869 * transaction. Wait for those.
872 spin_lock(&fs_info->trans_lock);
873 commit = list_first_entry_or_null(&fs_info->trans_list,
874 struct btrfs_transaction,
876 if (!commit || commit->transid > transid) {
877 spin_unlock(&fs_info->trans_lock);
880 refcount_inc(&commit->use_count);
882 spin_unlock(&fs_info->trans_lock);
886 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid)
888 struct btrfs_transaction *cur_trans = NULL, *t;
892 if (transid <= fs_info->last_trans_committed)
895 /* find specified transaction */
896 spin_lock(&fs_info->trans_lock);
897 list_for_each_entry(t, &fs_info->trans_list, list) {
898 if (t->transid == transid) {
900 refcount_inc(&cur_trans->use_count);
904 if (t->transid > transid) {
909 spin_unlock(&fs_info->trans_lock);
912 * The specified transaction doesn't exist, or we
913 * raced with btrfs_commit_transaction
916 if (transid > fs_info->last_trans_committed)
921 /* find newest transaction that is committing | committed */
922 spin_lock(&fs_info->trans_lock);
923 list_for_each_entry_reverse(t, &fs_info->trans_list,
925 if (t->state >= TRANS_STATE_COMMIT_START) {
926 if (t->state == TRANS_STATE_COMPLETED)
929 refcount_inc(&cur_trans->use_count);
933 spin_unlock(&fs_info->trans_lock);
935 goto out; /* nothing committing|committed */
938 wait_for_commit(cur_trans, TRANS_STATE_COMPLETED);
939 btrfs_put_transaction(cur_trans);
944 void btrfs_throttle(struct btrfs_fs_info *fs_info)
946 wait_current_trans(fs_info);
949 static bool should_end_transaction(struct btrfs_trans_handle *trans)
951 struct btrfs_fs_info *fs_info = trans->fs_info;
953 if (btrfs_check_space_for_delayed_refs(fs_info))
956 return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
959 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
961 struct btrfs_transaction *cur_trans = trans->transaction;
963 if (cur_trans->state >= TRANS_STATE_COMMIT_START ||
964 test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags))
967 return should_end_transaction(trans);
970 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans)
973 struct btrfs_fs_info *fs_info = trans->fs_info;
975 if (!trans->block_rsv) {
976 ASSERT(!trans->bytes_reserved);
980 if (!trans->bytes_reserved)
983 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv);
984 trace_btrfs_space_reservation(fs_info, "transaction",
985 trans->transid, trans->bytes_reserved, 0);
986 btrfs_block_rsv_release(fs_info, trans->block_rsv,
987 trans->bytes_reserved, NULL);
988 trans->bytes_reserved = 0;
991 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
994 struct btrfs_fs_info *info = trans->fs_info;
995 struct btrfs_transaction *cur_trans = trans->transaction;
998 if (refcount_read(&trans->use_count) > 1) {
999 refcount_dec(&trans->use_count);
1000 trans->block_rsv = trans->orig_rsv;
1004 btrfs_trans_release_metadata(trans);
1005 trans->block_rsv = NULL;
1007 btrfs_create_pending_block_groups(trans);
1009 btrfs_trans_release_chunk_metadata(trans);
1011 if (trans->type & __TRANS_FREEZABLE)
1012 sb_end_intwrite(info->sb);
1014 WARN_ON(cur_trans != info->running_transaction);
1015 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
1016 atomic_dec(&cur_trans->num_writers);
1017 extwriter_counter_dec(cur_trans, trans->type);
1019 cond_wake_up(&cur_trans->writer_wait);
1020 btrfs_put_transaction(cur_trans);
1022 if (current->journal_info == trans)
1023 current->journal_info = NULL;
1026 btrfs_run_delayed_iputs(info);
1028 if (TRANS_ABORTED(trans) ||
1029 test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
1030 wake_up_process(info->transaction_kthread);
1031 if (TRANS_ABORTED(trans))
1032 err = trans->aborted;
1037 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1041 int btrfs_end_transaction(struct btrfs_trans_handle *trans)
1043 return __btrfs_end_transaction(trans, 0);
1046 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans)
1048 return __btrfs_end_transaction(trans, 1);
1052 * when btree blocks are allocated, they have some corresponding bits set for
1053 * them in one of two extent_io trees. This is used to make sure all of
1054 * those extents are sent to disk but does not wait on them
1056 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
1057 struct extent_io_tree *dirty_pages, int mark)
1061 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1062 struct extent_state *cached_state = NULL;
1066 atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1067 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1068 mark, &cached_state)) {
1069 bool wait_writeback = false;
1071 err = convert_extent_bit(dirty_pages, start, end,
1073 mark, &cached_state);
1075 * convert_extent_bit can return -ENOMEM, which is most of the
1076 * time a temporary error. So when it happens, ignore the error
1077 * and wait for writeback of this range to finish - because we
1078 * failed to set the bit EXTENT_NEED_WAIT for the range, a call
1079 * to __btrfs_wait_marked_extents() would not know that
1080 * writeback for this range started and therefore wouldn't
1081 * wait for it to finish - we don't want to commit a
1082 * superblock that points to btree nodes/leafs for which
1083 * writeback hasn't finished yet (and without errors).
1084 * We cleanup any entries left in the io tree when committing
1085 * the transaction (through extent_io_tree_release()).
1087 if (err == -ENOMEM) {
1089 wait_writeback = true;
1092 err = filemap_fdatawrite_range(mapping, start, end);
1095 else if (wait_writeback)
1096 werr = filemap_fdatawait_range(mapping, start, end);
1097 free_extent_state(cached_state);
1098 cached_state = NULL;
1102 atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers);
1107 * when btree blocks are allocated, they have some corresponding bits set for
1108 * them in one of two extent_io trees. This is used to make sure all of
1109 * those extents are on disk for transaction or log commit. We wait
1110 * on all the pages and clear them from the dirty pages state tree
1112 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info,
1113 struct extent_io_tree *dirty_pages)
1117 struct address_space *mapping = fs_info->btree_inode->i_mapping;
1118 struct extent_state *cached_state = NULL;
1122 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
1123 EXTENT_NEED_WAIT, &cached_state)) {
1125 * Ignore -ENOMEM errors returned by clear_extent_bit().
1126 * When committing the transaction, we'll remove any entries
1127 * left in the io tree. For a log commit, we don't remove them
1128 * after committing the log because the tree can be accessed
1129 * concurrently - we do it only at transaction commit time when
1130 * it's safe to do it (through extent_io_tree_release()).
1132 err = clear_extent_bit(dirty_pages, start, end,
1133 EXTENT_NEED_WAIT, 0, 0, &cached_state);
1137 err = filemap_fdatawait_range(mapping, start, end);
1140 free_extent_state(cached_state);
1141 cached_state = NULL;
1150 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1151 struct extent_io_tree *dirty_pages)
1153 bool errors = false;
1156 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1157 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags))
1165 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark)
1167 struct btrfs_fs_info *fs_info = log_root->fs_info;
1168 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages;
1169 bool errors = false;
1172 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
1174 err = __btrfs_wait_marked_extents(fs_info, dirty_pages);
1175 if ((mark & EXTENT_DIRTY) &&
1176 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags))
1179 if ((mark & EXTENT_NEW) &&
1180 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags))
1189 * When btree blocks are allocated the corresponding extents are marked dirty.
1190 * This function ensures such extents are persisted on disk for transaction or
1193 * @trans: transaction whose dirty pages we'd like to write
1195 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans)
1199 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages;
1200 struct btrfs_fs_info *fs_info = trans->fs_info;
1201 struct blk_plug plug;
1203 blk_start_plug(&plug);
1204 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY);
1205 blk_finish_plug(&plug);
1206 ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1208 extent_io_tree_release(&trans->transaction->dirty_pages);
1219 * this is used to update the root pointer in the tree of tree roots.
1221 * But, in the case of the extent allocation tree, updating the root
1222 * pointer may allocate blocks which may change the root of the extent
1225 * So, this loops and repeats and makes sure the cowonly root didn't
1226 * change while the root pointer was being updated in the metadata.
1228 static int update_cowonly_root(struct btrfs_trans_handle *trans,
1229 struct btrfs_root *root)
1232 u64 old_root_bytenr;
1234 struct btrfs_fs_info *fs_info = root->fs_info;
1235 struct btrfs_root *tree_root = fs_info->tree_root;
1237 old_root_used = btrfs_root_used(&root->root_item);
1240 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
1241 if (old_root_bytenr == root->node->start &&
1242 old_root_used == btrfs_root_used(&root->root_item))
1245 btrfs_set_root_node(&root->root_item, root->node);
1246 ret = btrfs_update_root(trans, tree_root,
1252 old_root_used = btrfs_root_used(&root->root_item);
1259 * update all the cowonly tree roots on disk
1261 * The error handling in this function may not be obvious. Any of the
1262 * failures will cause the file system to go offline. We still need
1263 * to clean up the delayed refs.
1265 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
1267 struct btrfs_fs_info *fs_info = trans->fs_info;
1268 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs;
1269 struct list_head *io_bgs = &trans->transaction->io_bgs;
1270 struct list_head *next;
1271 struct extent_buffer *eb;
1274 eb = btrfs_lock_root_node(fs_info->tree_root);
1275 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1276 0, &eb, BTRFS_NESTING_COW);
1277 btrfs_tree_unlock(eb);
1278 free_extent_buffer(eb);
1283 ret = btrfs_run_dev_stats(trans);
1286 ret = btrfs_run_dev_replace(trans);
1289 ret = btrfs_run_qgroups(trans);
1293 ret = btrfs_setup_space_cache(trans);
1298 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
1299 struct btrfs_root *root;
1300 next = fs_info->dirty_cowonly_roots.next;
1301 list_del_init(next);
1302 root = list_entry(next, struct btrfs_root, dirty_list);
1303 clear_bit(BTRFS_ROOT_DIRTY, &root->state);
1305 if (root != fs_info->extent_root)
1306 list_add_tail(&root->dirty_list,
1307 &trans->transaction->switch_commits);
1308 ret = update_cowonly_root(trans, root);
1313 /* Now flush any delayed refs generated by updating all of the roots */
1314 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1318 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) {
1319 ret = btrfs_write_dirty_block_groups(trans);
1324 * We're writing the dirty block groups, which could generate
1325 * delayed refs, which could generate more dirty block groups,
1326 * so we want to keep this flushing in this loop to make sure
1327 * everything gets run.
1329 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1334 if (!list_empty(&fs_info->dirty_cowonly_roots))
1337 list_add_tail(&fs_info->extent_root->dirty_list,
1338 &trans->transaction->switch_commits);
1340 /* Update dev-replace pointer once everything is committed */
1341 fs_info->dev_replace.committed_cursor_left =
1342 fs_info->dev_replace.cursor_left_last_write_of_item;
1348 * If we had a pending drop we need to see if there are any others left in our
1349 * dead roots list, and if not clear our bit and wake any waiters.
1351 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
1354 * We put the drop in progress roots at the front of the list, so if the
1355 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
1358 spin_lock(&fs_info->trans_lock);
1359 if (!list_empty(&fs_info->dead_roots)) {
1360 struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
1363 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
1364 spin_unlock(&fs_info->trans_lock);
1368 spin_unlock(&fs_info->trans_lock);
1370 btrfs_wake_unfinished_drop(fs_info);
1374 * dead roots are old snapshots that need to be deleted. This allocates
1375 * a dirty root struct and adds it into the list of dead roots that need to
1378 void btrfs_add_dead_root(struct btrfs_root *root)
1380 struct btrfs_fs_info *fs_info = root->fs_info;
1382 spin_lock(&fs_info->trans_lock);
1383 if (list_empty(&root->root_list)) {
1384 btrfs_grab_root(root);
1386 /* We want to process the partially complete drops first. */
1387 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
1388 list_add(&root->root_list, &fs_info->dead_roots);
1390 list_add_tail(&root->root_list, &fs_info->dead_roots);
1392 spin_unlock(&fs_info->trans_lock);
1396 * update all the cowonly tree roots on disk
1398 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans)
1400 struct btrfs_fs_info *fs_info = trans->fs_info;
1401 struct btrfs_root *gang[8];
1405 spin_lock(&fs_info->fs_roots_radix_lock);
1407 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
1410 BTRFS_ROOT_TRANS_TAG);
1413 for (i = 0; i < ret; i++) {
1414 struct btrfs_root *root = gang[i];
1417 radix_tree_tag_clear(&fs_info->fs_roots_radix,
1418 (unsigned long)root->root_key.objectid,
1419 BTRFS_ROOT_TRANS_TAG);
1420 spin_unlock(&fs_info->fs_roots_radix_lock);
1422 btrfs_free_log(trans, root);
1423 ret2 = btrfs_update_reloc_root(trans, root);
1427 /* see comments in should_cow_block() */
1428 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1429 smp_mb__after_atomic();
1431 if (root->commit_root != root->node) {
1432 list_add_tail(&root->dirty_list,
1433 &trans->transaction->switch_commits);
1434 btrfs_set_root_node(&root->root_item,
1438 ret2 = btrfs_update_root(trans, fs_info->tree_root,
1443 spin_lock(&fs_info->fs_roots_radix_lock);
1444 btrfs_qgroup_free_meta_all_pertrans(root);
1447 spin_unlock(&fs_info->fs_roots_radix_lock);
1452 * defrag a given btree.
1453 * Every leaf in the btree is read and defragged.
1455 int btrfs_defrag_root(struct btrfs_root *root)
1457 struct btrfs_fs_info *info = root->fs_info;
1458 struct btrfs_trans_handle *trans;
1461 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state))
1465 trans = btrfs_start_transaction(root, 0);
1466 if (IS_ERR(trans)) {
1467 ret = PTR_ERR(trans);
1471 ret = btrfs_defrag_leaves(trans, root);
1473 btrfs_end_transaction(trans);
1474 btrfs_btree_balance_dirty(info);
1477 if (btrfs_fs_closing(info) || ret != -EAGAIN)
1480 if (btrfs_defrag_cancelled(info)) {
1481 btrfs_debug(info, "defrag_root cancelled");
1486 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state);
1491 * Do all special snapshot related qgroup dirty hack.
1493 * Will do all needed qgroup inherit and dirty hack like switch commit
1494 * roots inside one transaction and write all btree into disk, to make
1497 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans,
1498 struct btrfs_root *src,
1499 struct btrfs_root *parent,
1500 struct btrfs_qgroup_inherit *inherit,
1503 struct btrfs_fs_info *fs_info = src->fs_info;
1507 * Save some performance in the case that qgroups are not
1508 * enabled. If this check races with the ioctl, rescan will
1511 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1515 * Ensure dirty @src will be committed. Or, after coming
1516 * commit_fs_roots() and switch_commit_roots(), any dirty but not
1517 * recorded root will never be updated again, causing an outdated root
1520 ret = record_root_in_trans(trans, src, 1);
1525 * btrfs_qgroup_inherit relies on a consistent view of the usage for the
1526 * src root, so we must run the delayed refs here.
1528 * However this isn't particularly fool proof, because there's no
1529 * synchronization keeping us from changing the tree after this point
1530 * before we do the qgroup_inherit, or even from making changes while
1531 * we're doing the qgroup_inherit. But that's a problem for the future,
1532 * for now flush the delayed refs to narrow the race window where the
1533 * qgroup counters could end up wrong.
1535 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
1537 btrfs_abort_transaction(trans, ret);
1542 * We are going to commit transaction, see btrfs_commit_transaction()
1543 * comment for reason locking tree_log_mutex
1545 mutex_lock(&fs_info->tree_log_mutex);
1547 ret = commit_fs_roots(trans);
1550 ret = btrfs_qgroup_account_extents(trans);
1554 /* Now qgroup are all updated, we can inherit it to new qgroups */
1555 ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid,
1561 * Now we do a simplified commit transaction, which will:
1562 * 1) commit all subvolume and extent tree
1563 * To ensure all subvolume and extent tree have a valid
1564 * commit_root to accounting later insert_dir_item()
1565 * 2) write all btree blocks onto disk
1566 * This is to make sure later btree modification will be cowed
1567 * Or commit_root can be populated and cause wrong qgroup numbers
1568 * In this simplified commit, we don't really care about other trees
1569 * like chunk and root tree, as they won't affect qgroup.
1570 * And we don't write super to avoid half committed status.
1572 ret = commit_cowonly_roots(trans);
1575 switch_commit_roots(trans);
1576 ret = btrfs_write_and_wait_transaction(trans);
1578 btrfs_handle_fs_error(fs_info, ret,
1579 "Error while writing out transaction for qgroup");
1582 mutex_unlock(&fs_info->tree_log_mutex);
1585 * Force parent root to be updated, as we recorded it before so its
1586 * last_trans == cur_transid.
1587 * Or it won't be committed again onto disk after later
1591 ret = record_root_in_trans(trans, parent, 1);
1596 * new snapshots need to be created at a very specific time in the
1597 * transaction commit. This does the actual creation.
1600 * If the error which may affect the commitment of the current transaction
1601 * happens, we should return the error number. If the error which just affect
1602 * the creation of the pending snapshots, just return 0.
1604 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1605 struct btrfs_pending_snapshot *pending)
1608 struct btrfs_fs_info *fs_info = trans->fs_info;
1609 struct btrfs_key key;
1610 struct btrfs_root_item *new_root_item;
1611 struct btrfs_root *tree_root = fs_info->tree_root;
1612 struct btrfs_root *root = pending->root;
1613 struct btrfs_root *parent_root;
1614 struct btrfs_block_rsv *rsv;
1615 struct inode *parent_inode;
1616 struct btrfs_path *path;
1617 struct btrfs_dir_item *dir_item;
1618 struct dentry *dentry;
1619 struct extent_buffer *tmp;
1620 struct extent_buffer *old;
1621 struct timespec64 cur_time;
1628 ASSERT(pending->path);
1629 path = pending->path;
1631 ASSERT(pending->root_item);
1632 new_root_item = pending->root_item;
1634 pending->error = btrfs_get_free_objectid(tree_root, &objectid);
1636 goto no_free_objectid;
1639 * Make qgroup to skip current new snapshot's qgroupid, as it is
1640 * accounted by later btrfs_qgroup_inherit().
1642 btrfs_set_skip_qgroup(trans, objectid);
1644 btrfs_reloc_pre_snapshot(pending, &to_reserve);
1646 if (to_reserve > 0) {
1647 pending->error = btrfs_block_rsv_add(root,
1648 &pending->block_rsv,
1650 BTRFS_RESERVE_NO_FLUSH);
1652 goto clear_skip_qgroup;
1655 key.objectid = objectid;
1656 key.offset = (u64)-1;
1657 key.type = BTRFS_ROOT_ITEM_KEY;
1659 rsv = trans->block_rsv;
1660 trans->block_rsv = &pending->block_rsv;
1661 trans->bytes_reserved = trans->block_rsv->reserved;
1662 trace_btrfs_space_reservation(fs_info, "transaction",
1664 trans->bytes_reserved, 1);
1665 dentry = pending->dentry;
1666 parent_inode = pending->dir;
1667 parent_root = BTRFS_I(parent_inode)->root;
1668 ret = record_root_in_trans(trans, parent_root, 0);
1671 cur_time = current_time(parent_inode);
1674 * insert the directory item
1676 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index);
1677 BUG_ON(ret); /* -ENOMEM */
1679 /* check if there is a file/dir which has the same name. */
1680 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1681 btrfs_ino(BTRFS_I(parent_inode)),
1682 dentry->d_name.name,
1683 dentry->d_name.len, 0);
1684 if (dir_item != NULL && !IS_ERR(dir_item)) {
1685 pending->error = -EEXIST;
1686 goto dir_item_existed;
1687 } else if (IS_ERR(dir_item)) {
1688 ret = PTR_ERR(dir_item);
1689 btrfs_abort_transaction(trans, ret);
1692 btrfs_release_path(path);
1695 * pull in the delayed directory update
1696 * and the delayed inode item
1697 * otherwise we corrupt the FS during
1700 ret = btrfs_run_delayed_items(trans);
1701 if (ret) { /* Transaction aborted */
1702 btrfs_abort_transaction(trans, ret);
1706 ret = record_root_in_trans(trans, root, 0);
1708 btrfs_abort_transaction(trans, ret);
1711 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1712 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1713 btrfs_check_and_init_root_item(new_root_item);
1715 root_flags = btrfs_root_flags(new_root_item);
1716 if (pending->readonly)
1717 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1719 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1720 btrfs_set_root_flags(new_root_item, root_flags);
1722 btrfs_set_root_generation_v2(new_root_item,
1724 generate_random_guid(new_root_item->uuid);
1725 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1727 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1728 memset(new_root_item->received_uuid, 0,
1729 sizeof(new_root_item->received_uuid));
1730 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1731 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1732 btrfs_set_root_stransid(new_root_item, 0);
1733 btrfs_set_root_rtransid(new_root_item, 0);
1735 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec);
1736 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec);
1737 btrfs_set_root_otransid(new_root_item, trans->transid);
1739 old = btrfs_lock_root_node(root);
1740 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
1743 btrfs_tree_unlock(old);
1744 free_extent_buffer(old);
1745 btrfs_abort_transaction(trans, ret);
1749 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1750 /* clean up in any case */
1751 btrfs_tree_unlock(old);
1752 free_extent_buffer(old);
1754 btrfs_abort_transaction(trans, ret);
1757 /* see comments in should_cow_block() */
1758 set_bit(BTRFS_ROOT_FORCE_COW, &root->state);
1761 btrfs_set_root_node(new_root_item, tmp);
1762 /* record when the snapshot was created in key.offset */
1763 key.offset = trans->transid;
1764 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1765 btrfs_tree_unlock(tmp);
1766 free_extent_buffer(tmp);
1768 btrfs_abort_transaction(trans, ret);
1773 * insert root back/forward references
1775 ret = btrfs_add_root_ref(trans, objectid,
1776 parent_root->root_key.objectid,
1777 btrfs_ino(BTRFS_I(parent_inode)), index,
1778 dentry->d_name.name, dentry->d_name.len);
1780 btrfs_abort_transaction(trans, ret);
1784 key.offset = (u64)-1;
1785 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1786 if (IS_ERR(pending->snap)) {
1787 ret = PTR_ERR(pending->snap);
1788 pending->snap = NULL;
1789 btrfs_abort_transaction(trans, ret);
1793 ret = btrfs_reloc_post_snapshot(trans, pending);
1795 btrfs_abort_transaction(trans, ret);
1800 * Do special qgroup accounting for snapshot, as we do some qgroup
1801 * snapshot hack to do fast snapshot.
1802 * To co-operate with that hack, we do hack again.
1803 * Or snapshot will be greatly slowed down by a subtree qgroup rescan
1805 ret = qgroup_account_snapshot(trans, root, parent_root,
1806 pending->inherit, objectid);
1810 ret = btrfs_insert_dir_item(trans, dentry->d_name.name,
1811 dentry->d_name.len, BTRFS_I(parent_inode),
1812 &key, BTRFS_FT_DIR, index);
1813 /* We have check then name at the beginning, so it is impossible. */
1814 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1816 btrfs_abort_transaction(trans, ret);
1820 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size +
1821 dentry->d_name.len * 2);
1822 parent_inode->i_mtime = parent_inode->i_ctime =
1823 current_time(parent_inode);
1824 ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode));
1826 btrfs_abort_transaction(trans, ret);
1829 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
1830 BTRFS_UUID_KEY_SUBVOL,
1833 btrfs_abort_transaction(trans, ret);
1836 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) {
1837 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid,
1838 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
1840 if (ret && ret != -EEXIST) {
1841 btrfs_abort_transaction(trans, ret);
1847 pending->error = ret;
1849 trans->block_rsv = rsv;
1850 trans->bytes_reserved = 0;
1852 btrfs_clear_skip_qgroup(trans);
1854 kfree(new_root_item);
1855 pending->root_item = NULL;
1856 btrfs_free_path(path);
1857 pending->path = NULL;
1863 * create all the snapshots we've scheduled for creation
1865 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans)
1867 struct btrfs_pending_snapshot *pending, *next;
1868 struct list_head *head = &trans->transaction->pending_snapshots;
1871 list_for_each_entry_safe(pending, next, head, list) {
1872 list_del(&pending->list);
1873 ret = create_pending_snapshot(trans, pending);
1880 static void update_super_roots(struct btrfs_fs_info *fs_info)
1882 struct btrfs_root_item *root_item;
1883 struct btrfs_super_block *super;
1885 super = fs_info->super_copy;
1887 root_item = &fs_info->chunk_root->root_item;
1888 super->chunk_root = root_item->bytenr;
1889 super->chunk_root_generation = root_item->generation;
1890 super->chunk_root_level = root_item->level;
1892 root_item = &fs_info->tree_root->root_item;
1893 super->root = root_item->bytenr;
1894 super->generation = root_item->generation;
1895 super->root_level = root_item->level;
1896 if (btrfs_test_opt(fs_info, SPACE_CACHE))
1897 super->cache_generation = root_item->generation;
1898 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags))
1899 super->cache_generation = 0;
1900 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags))
1901 super->uuid_tree_generation = root_item->generation;
1904 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1906 struct btrfs_transaction *trans;
1909 spin_lock(&info->trans_lock);
1910 trans = info->running_transaction;
1912 ret = (trans->state >= TRANS_STATE_COMMIT_START);
1913 spin_unlock(&info->trans_lock);
1917 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1919 struct btrfs_transaction *trans;
1922 spin_lock(&info->trans_lock);
1923 trans = info->running_transaction;
1925 ret = is_transaction_blocked(trans);
1926 spin_unlock(&info->trans_lock);
1931 * commit transactions asynchronously. once btrfs_commit_transaction_async
1932 * returns, any subsequent transaction will not be allowed to join.
1934 struct btrfs_async_commit {
1935 struct btrfs_trans_handle *newtrans;
1936 struct work_struct work;
1939 static void do_async_commit(struct work_struct *work)
1941 struct btrfs_async_commit *ac =
1942 container_of(work, struct btrfs_async_commit, work);
1945 * We've got freeze protection passed with the transaction.
1946 * Tell lockdep about it.
1948 if (ac->newtrans->type & __TRANS_FREEZABLE)
1949 __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS);
1951 current->journal_info = ac->newtrans;
1953 btrfs_commit_transaction(ac->newtrans);
1957 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans)
1959 struct btrfs_fs_info *fs_info = trans->fs_info;
1960 struct btrfs_async_commit *ac;
1961 struct btrfs_transaction *cur_trans;
1963 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1967 INIT_WORK(&ac->work, do_async_commit);
1968 ac->newtrans = btrfs_join_transaction(trans->root);
1969 if (IS_ERR(ac->newtrans)) {
1970 int err = PTR_ERR(ac->newtrans);
1975 /* take transaction reference */
1976 cur_trans = trans->transaction;
1977 refcount_inc(&cur_trans->use_count);
1979 btrfs_end_transaction(trans);
1982 * Tell lockdep we've released the freeze rwsem, since the
1983 * async commit thread will be the one to unlock it.
1985 if (ac->newtrans->type & __TRANS_FREEZABLE)
1986 __sb_writers_release(fs_info->sb, SB_FREEZE_FS);
1988 schedule_work(&ac->work);
1990 * Wait for the current transaction commit to start and block
1991 * subsequent transaction joins
1993 wait_event(fs_info->transaction_blocked_wait,
1994 cur_trans->state >= TRANS_STATE_COMMIT_START ||
1995 TRANS_ABORTED(cur_trans));
1996 if (current->journal_info == trans)
1997 current->journal_info = NULL;
1999 btrfs_put_transaction(cur_trans);
2004 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err)
2006 struct btrfs_fs_info *fs_info = trans->fs_info;
2007 struct btrfs_transaction *cur_trans = trans->transaction;
2009 WARN_ON(refcount_read(&trans->use_count) > 1);
2011 btrfs_abort_transaction(trans, err);
2013 spin_lock(&fs_info->trans_lock);
2016 * If the transaction is removed from the list, it means this
2017 * transaction has been committed successfully, so it is impossible
2018 * to call the cleanup function.
2020 BUG_ON(list_empty(&cur_trans->list));
2022 if (cur_trans == fs_info->running_transaction) {
2023 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2024 spin_unlock(&fs_info->trans_lock);
2025 wait_event(cur_trans->writer_wait,
2026 atomic_read(&cur_trans->num_writers) == 1);
2028 spin_lock(&fs_info->trans_lock);
2032 * Now that we know no one else is still using the transaction we can
2033 * remove the transaction from the list of transactions. This avoids
2034 * the transaction kthread from cleaning up the transaction while some
2035 * other task is still using it, which could result in a use-after-free
2036 * on things like log trees, as it forces the transaction kthread to
2037 * wait for this transaction to be cleaned up by us.
2039 list_del_init(&cur_trans->list);
2041 spin_unlock(&fs_info->trans_lock);
2043 btrfs_cleanup_one_transaction(trans->transaction, fs_info);
2045 spin_lock(&fs_info->trans_lock);
2046 if (cur_trans == fs_info->running_transaction)
2047 fs_info->running_transaction = NULL;
2048 spin_unlock(&fs_info->trans_lock);
2050 if (trans->type & __TRANS_FREEZABLE)
2051 sb_end_intwrite(fs_info->sb);
2052 btrfs_put_transaction(cur_trans);
2053 btrfs_put_transaction(cur_trans);
2055 trace_btrfs_transaction_commit(trans->root);
2057 if (current->journal_info == trans)
2058 current->journal_info = NULL;
2059 btrfs_scrub_cancel(fs_info);
2061 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2065 * Release reserved delayed ref space of all pending block groups of the
2066 * transaction and remove them from the list
2068 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
2070 struct btrfs_fs_info *fs_info = trans->fs_info;
2071 struct btrfs_block_group *block_group, *tmp;
2073 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
2074 btrfs_delayed_refs_rsv_release(fs_info, 1);
2075 list_del_init(&block_group->bg_list);
2079 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
2082 * We use try_to_writeback_inodes_sb() here because if we used
2083 * btrfs_start_delalloc_roots we would deadlock with fs freeze.
2084 * Currently are holding the fs freeze lock, if we do an async flush
2085 * we'll do btrfs_join_transaction() and deadlock because we need to
2086 * wait for the fs freeze lock. Using the direct flushing we benefit
2087 * from already being in a transaction and our join_transaction doesn't
2088 * have to re-take the fs freeze lock.
2090 * Note that try_to_writeback_inodes_sb() will only trigger writeback
2091 * if it can read lock sb->s_umount. It will always be able to lock it,
2092 * except when the filesystem is being unmounted or being frozen, but in
2093 * those cases sync_filesystem() is called, which results in calling
2094 * writeback_inodes_sb() while holding a write lock on sb->s_umount.
2095 * Note that we don't call writeback_inodes_sb() directly, because it
2096 * will emit a warning if sb->s_umount is not locked.
2098 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2099 try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
2103 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
2105 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
2106 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
2110 * Add a pending snapshot associated with the given transaction handle to the
2111 * respective handle. This must be called after the transaction commit started
2112 * and while holding fs_info->trans_lock.
2113 * This serves to guarantee a caller of btrfs_commit_transaction() that it can
2114 * safely free the pending snapshot pointer in case btrfs_commit_transaction()
2117 static void add_pending_snapshot(struct btrfs_trans_handle *trans)
2119 struct btrfs_transaction *cur_trans = trans->transaction;
2121 if (!trans->pending_snapshot)
2124 lockdep_assert_held(&trans->fs_info->trans_lock);
2125 ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START);
2127 list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots);
2130 int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
2132 struct btrfs_fs_info *fs_info = trans->fs_info;
2133 struct btrfs_transaction *cur_trans = trans->transaction;
2134 struct btrfs_transaction *prev_trans = NULL;
2137 ASSERT(refcount_read(&trans->use_count) == 1);
2139 /* Stop the commit early if ->aborted is set */
2140 if (TRANS_ABORTED(cur_trans)) {
2141 ret = cur_trans->aborted;
2142 btrfs_end_transaction(trans);
2146 btrfs_trans_release_metadata(trans);
2147 trans->block_rsv = NULL;
2150 * We only want one transaction commit doing the flushing so we do not
2151 * waste a bunch of time on lock contention on the extent root node.
2153 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING,
2154 &cur_trans->delayed_refs.flags)) {
2156 * Make a pass through all the delayed refs we have so far.
2157 * Any running threads may add more while we are here.
2159 ret = btrfs_run_delayed_refs(trans, 0);
2161 btrfs_end_transaction(trans);
2166 btrfs_create_pending_block_groups(trans);
2168 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
2171 /* this mutex is also taken before trying to set
2172 * block groups readonly. We need to make sure
2173 * that nobody has set a block group readonly
2174 * after a extents from that block group have been
2175 * allocated for cache files. btrfs_set_block_group_ro
2176 * will wait for the transaction to commit if it
2177 * finds BTRFS_TRANS_DIRTY_BG_RUN set.
2179 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure
2180 * only one process starts all the block group IO. It wouldn't
2181 * hurt to have more than one go through, but there's no
2182 * real advantage to it either.
2184 mutex_lock(&fs_info->ro_block_group_mutex);
2185 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN,
2188 mutex_unlock(&fs_info->ro_block_group_mutex);
2191 ret = btrfs_start_dirty_block_groups(trans);
2193 btrfs_end_transaction(trans);
2199 spin_lock(&fs_info->trans_lock);
2200 if (cur_trans->state >= TRANS_STATE_COMMIT_START) {
2201 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2203 add_pending_snapshot(trans);
2205 spin_unlock(&fs_info->trans_lock);
2206 refcount_inc(&cur_trans->use_count);
2208 if (trans->in_fsync)
2209 want_state = TRANS_STATE_SUPER_COMMITTED;
2210 ret = btrfs_end_transaction(trans);
2211 wait_for_commit(cur_trans, want_state);
2213 if (TRANS_ABORTED(cur_trans))
2214 ret = cur_trans->aborted;
2216 btrfs_put_transaction(cur_trans);
2221 cur_trans->state = TRANS_STATE_COMMIT_START;
2222 wake_up(&fs_info->transaction_blocked_wait);
2224 if (cur_trans->list.prev != &fs_info->trans_list) {
2225 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED;
2227 if (trans->in_fsync)
2228 want_state = TRANS_STATE_SUPER_COMMITTED;
2230 prev_trans = list_entry(cur_trans->list.prev,
2231 struct btrfs_transaction, list);
2232 if (prev_trans->state < want_state) {
2233 refcount_inc(&prev_trans->use_count);
2234 spin_unlock(&fs_info->trans_lock);
2236 wait_for_commit(prev_trans, want_state);
2238 ret = READ_ONCE(prev_trans->aborted);
2240 btrfs_put_transaction(prev_trans);
2242 goto cleanup_transaction;
2244 spin_unlock(&fs_info->trans_lock);
2247 spin_unlock(&fs_info->trans_lock);
2249 * The previous transaction was aborted and was already removed
2250 * from the list of transactions at fs_info->trans_list. So we
2251 * abort to prevent writing a new superblock that reflects a
2252 * corrupt state (pointing to trees with unwritten nodes/leafs).
2254 if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) {
2256 goto cleanup_transaction;
2260 extwriter_counter_dec(cur_trans, trans->type);
2262 ret = btrfs_start_delalloc_flush(fs_info);
2264 goto cleanup_transaction;
2266 ret = btrfs_run_delayed_items(trans);
2268 goto cleanup_transaction;
2270 wait_event(cur_trans->writer_wait,
2271 extwriter_counter_read(cur_trans) == 0);
2273 /* some pending stuffs might be added after the previous flush. */
2274 ret = btrfs_run_delayed_items(trans);
2276 goto cleanup_transaction;
2278 btrfs_wait_delalloc_flush(fs_info);
2281 * Wait for all ordered extents started by a fast fsync that joined this
2282 * transaction. Otherwise if this transaction commits before the ordered
2283 * extents complete we lose logged data after a power failure.
2285 wait_event(cur_trans->pending_wait,
2286 atomic_read(&cur_trans->pending_ordered) == 0);
2288 btrfs_scrub_pause(fs_info);
2290 * Ok now we need to make sure to block out any other joins while we
2291 * commit the transaction. We could have started a join before setting
2292 * COMMIT_DOING so make sure to wait for num_writers to == 1 again.
2294 spin_lock(&fs_info->trans_lock);
2295 add_pending_snapshot(trans);
2296 cur_trans->state = TRANS_STATE_COMMIT_DOING;
2297 spin_unlock(&fs_info->trans_lock);
2298 wait_event(cur_trans->writer_wait,
2299 atomic_read(&cur_trans->num_writers) == 1);
2301 if (TRANS_ABORTED(cur_trans)) {
2302 ret = cur_trans->aborted;
2303 goto scrub_continue;
2306 * the reloc mutex makes sure that we stop
2307 * the balancing code from coming in and moving
2308 * extents around in the middle of the commit
2310 mutex_lock(&fs_info->reloc_mutex);
2313 * We needn't worry about the delayed items because we will
2314 * deal with them in create_pending_snapshot(), which is the
2315 * core function of the snapshot creation.
2317 ret = create_pending_snapshots(trans);
2322 * We insert the dir indexes of the snapshots and update the inode
2323 * of the snapshots' parents after the snapshot creation, so there
2324 * are some delayed items which are not dealt with. Now deal with
2327 * We needn't worry that this operation will corrupt the snapshots,
2328 * because all the tree which are snapshoted will be forced to COW
2329 * the nodes and leaves.
2331 ret = btrfs_run_delayed_items(trans);
2335 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2340 * make sure none of the code above managed to slip in a
2343 btrfs_assert_delayed_root_empty(fs_info);
2345 WARN_ON(cur_trans != trans->transaction);
2347 /* btrfs_commit_tree_roots is responsible for getting the
2348 * various roots consistent with each other. Every pointer
2349 * in the tree of tree roots has to point to the most up to date
2350 * root for every subvolume and other tree. So, we have to keep
2351 * the tree logging code from jumping in and changing any
2354 * At this point in the commit, there can't be any tree-log
2355 * writers, but a little lower down we drop the trans mutex
2356 * and let new people in. By holding the tree_log_mutex
2357 * from now until after the super is written, we avoid races
2358 * with the tree-log code.
2360 mutex_lock(&fs_info->tree_log_mutex);
2362 ret = commit_fs_roots(trans);
2364 goto unlock_tree_log;
2367 * Since the transaction is done, we can apply the pending changes
2368 * before the next transaction.
2370 btrfs_apply_pending_changes(fs_info);
2372 /* commit_fs_roots gets rid of all the tree log roots, it is now
2373 * safe to free the root of tree log roots
2375 btrfs_free_log_root_tree(trans, fs_info);
2378 * Since fs roots are all committed, we can get a quite accurate
2379 * new_roots. So let's do quota accounting.
2381 ret = btrfs_qgroup_account_extents(trans);
2383 goto unlock_tree_log;
2385 ret = commit_cowonly_roots(trans);
2387 goto unlock_tree_log;
2390 * The tasks which save the space cache and inode cache may also
2391 * update ->aborted, check it.
2393 if (TRANS_ABORTED(cur_trans)) {
2394 ret = cur_trans->aborted;
2395 goto unlock_tree_log;
2398 cur_trans = fs_info->running_transaction;
2400 btrfs_set_root_node(&fs_info->tree_root->root_item,
2401 fs_info->tree_root->node);
2402 list_add_tail(&fs_info->tree_root->dirty_list,
2403 &cur_trans->switch_commits);
2405 btrfs_set_root_node(&fs_info->chunk_root->root_item,
2406 fs_info->chunk_root->node);
2407 list_add_tail(&fs_info->chunk_root->dirty_list,
2408 &cur_trans->switch_commits);
2410 switch_commit_roots(trans);
2412 ASSERT(list_empty(&cur_trans->dirty_bgs));
2413 ASSERT(list_empty(&cur_trans->io_bgs));
2414 update_super_roots(fs_info);
2416 btrfs_set_super_log_root(fs_info->super_copy, 0);
2417 btrfs_set_super_log_root_level(fs_info->super_copy, 0);
2418 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2419 sizeof(*fs_info->super_copy));
2421 btrfs_commit_device_sizes(cur_trans);
2423 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2424 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2426 btrfs_trans_release_chunk_metadata(trans);
2428 spin_lock(&fs_info->trans_lock);
2429 cur_trans->state = TRANS_STATE_UNBLOCKED;
2430 fs_info->running_transaction = NULL;
2431 spin_unlock(&fs_info->trans_lock);
2432 mutex_unlock(&fs_info->reloc_mutex);
2434 wake_up(&fs_info->transaction_wait);
2436 ret = btrfs_write_and_wait_transaction(trans);
2438 btrfs_handle_fs_error(fs_info, ret,
2439 "Error while writing out transaction");
2441 * reloc_mutex has been unlocked, tree_log_mutex is still held
2442 * but we can't jump to unlock_tree_log causing double unlock
2444 mutex_unlock(&fs_info->tree_log_mutex);
2445 goto scrub_continue;
2449 * At this point, we should have written all the tree blocks allocated
2450 * in this transaction. So it's now safe to free the redirtyied extent
2453 btrfs_free_redirty_list(cur_trans);
2455 ret = write_all_supers(fs_info, 0);
2457 * the super is written, we can safely allow the tree-loggers
2458 * to go about their business
2460 mutex_unlock(&fs_info->tree_log_mutex);
2462 goto scrub_continue;
2465 * We needn't acquire the lock here because there is no other task
2466 * which can change it.
2468 cur_trans->state = TRANS_STATE_SUPER_COMMITTED;
2469 wake_up(&cur_trans->commit_wait);
2471 btrfs_finish_extent_commit(trans);
2473 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags))
2474 btrfs_clear_space_info_full(fs_info);
2476 fs_info->last_trans_committed = cur_trans->transid;
2478 * We needn't acquire the lock here because there is no other task
2479 * which can change it.
2481 cur_trans->state = TRANS_STATE_COMPLETED;
2482 wake_up(&cur_trans->commit_wait);
2484 spin_lock(&fs_info->trans_lock);
2485 list_del_init(&cur_trans->list);
2486 spin_unlock(&fs_info->trans_lock);
2488 btrfs_put_transaction(cur_trans);
2489 btrfs_put_transaction(cur_trans);
2491 if (trans->type & __TRANS_FREEZABLE)
2492 sb_end_intwrite(fs_info->sb);
2494 trace_btrfs_transaction_commit(trans->root);
2496 btrfs_scrub_continue(fs_info);
2498 if (current->journal_info == trans)
2499 current->journal_info = NULL;
2501 kmem_cache_free(btrfs_trans_handle_cachep, trans);
2506 mutex_unlock(&fs_info->tree_log_mutex);
2508 mutex_unlock(&fs_info->reloc_mutex);
2510 btrfs_scrub_continue(fs_info);
2511 cleanup_transaction:
2512 btrfs_trans_release_metadata(trans);
2513 btrfs_cleanup_pending_block_groups(trans);
2514 btrfs_trans_release_chunk_metadata(trans);
2515 trans->block_rsv = NULL;
2516 btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
2517 if (current->journal_info == trans)
2518 current->journal_info = NULL;
2519 cleanup_transaction(trans, ret);
2525 * return < 0 if error
2526 * 0 if there are no more dead_roots at the time of call
2527 * 1 there are more to be processed, call me again
2529 * The return value indicates there are certainly more snapshots to delete, but
2530 * if there comes a new one during processing, it may return 0. We don't mind,
2531 * because btrfs_commit_super will poke cleaner thread and it will process it a
2532 * few seconds later.
2534 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
2537 struct btrfs_fs_info *fs_info = root->fs_info;
2539 spin_lock(&fs_info->trans_lock);
2540 if (list_empty(&fs_info->dead_roots)) {
2541 spin_unlock(&fs_info->trans_lock);
2544 root = list_first_entry(&fs_info->dead_roots,
2545 struct btrfs_root, root_list);
2546 list_del_init(&root->root_list);
2547 spin_unlock(&fs_info->trans_lock);
2549 btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2551 btrfs_kill_all_delayed_nodes(root);
2553 if (btrfs_header_backref_rev(root->node) <
2554 BTRFS_MIXED_BACKREF_REV)
2555 ret = btrfs_drop_snapshot(root, 0, 0);
2557 ret = btrfs_drop_snapshot(root, 1, 0);
2559 btrfs_put_root(root);
2560 return (ret < 0) ? 0 : 1;
2563 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info)
2568 prev = xchg(&fs_info->pending_changes, 0);
2572 bit = 1 << BTRFS_PENDING_COMMIT;
2574 btrfs_debug(fs_info, "pending commit done");
2579 "unknown pending changes left 0x%lx, ignoring", prev);