~hc/RK356X_SDK_RELEASE.git

..	..	@@ -10,6 +10,7 @@
10	10	#include <linux/pagemap.h>
11	11	#include <linux/blkdev.h>
12	12	#include <linux/uuid.h>
	13	+#include "misc.h"
13	14	#include "ctree.h"
14	15	#include "disk-io.h"
15	16	#include "transaction.h"
..	..	@@ -19,12 +20,84 @@
19	20	#include "volumes.h"
20	21	#include "dev-replace.h"
21	22	#include "qgroup.h"
	23	+#include "block-group.h"
	24	+#include "space-info.h"
22	25
23	26	#define BTRFS_ROOT_TRANS_TAG 0
24	27
	28	+/*
	29	+ * Transaction states and transitions
	30	+ *
	31	+ * No running transaction (fs tree blocks are not modified)
	32	+ * \|
	33	+ * \| To next stage:
	34	+ * \| Call start_transaction() variants. Except btrfs_join_transaction_nostart().
	35	+ * V
	36	+ * Transaction N [[TRANS_STATE_RUNNING]]
	37	+ * \|
	38	+ * \| New trans handles can be attached to transaction N by calling all
	39	+ * \| start_transaction() variants.
	40	+ * \|
	41	+ * \| To next stage:
	42	+ * \| Call btrfs_commit_transaction() on any trans handle attached to
	43	+ * \| transaction N
	44	+ * V
	45	+ * Transaction N [[TRANS_STATE_COMMIT_START]]
	46	+ * \|
	47	+ * \| Will wait for previous running transaction to completely finish if there
	48	+ * \| is one
	49	+ * \|
	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.
	55	+ * \|
	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
	59	+ * \| transaction N+1.
	60	+ * \|
	61	+ * \| To next stage:
	62	+ * \| Caller is chosen to commit transaction N, and all other trans handle
	63	+ * \| haven been released.
	64	+ * V
	65	+ * Transaction N [[TRANS_STATE_COMMIT_DOING]]
	66	+ * \|
	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
	71	+ * \| trees.
	72	+ * \|
	73	+ * \| At this stage, all start_transaction() calls will wait for this
	74	+ * \| transaction to finish and attach to transaction N+1.
	75	+ * \|
	76	+ * \| To next stage:
	77	+ * \| Until all supporting trees are updated.
	78	+ * V
	79	+ * Transaction N [[TRANS_STATE_UNBLOCKED]]
	80	+ * \| Transaction N+1
	81	+ * \| All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]]
	82	+ * \| need to write them back to disk and update \|
	83	+ * \| super blocks. \|
	84	+ * \| \|
	85	+ * \| At this stage, new transaction is allowed to \|
	86	+ * \| start. \|
	87	+ * \| All new start_transaction() calls will be \|
	88	+ * \| attached to transid N+1. \|
	89	+ * \| \|
	90	+ * \| To next stage: \|
	91	+ * \| Until all tree blocks are super blocks are \|
	92	+ * \| written to block devices \|
	93	+ * V \|
	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
	98	+ */
25	99	static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = {
26	100	[TRANS_STATE_RUNNING] = 0U,
27		- [TRANS_STATE_BLOCKED] = __TRANS_START,
28	101	[TRANS_STATE_COMMIT_START] = (__TRANS_START \| __TRANS_ATTACH),
29	102	[TRANS_STATE_COMMIT_DOING] = (__TRANS_START \|
30	103	__TRANS_ATTACH \|
..	..	@@ -47,19 +120,14 @@
47	120	WARN_ON(refcount_read(&transaction->use_count) == 0);
48	121	if (refcount_dec_and_test(&transaction->use_count)) {
49	122	BUG_ON(!list_empty(&transaction->list));
50		- WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root));
	123	+ WARN_ON(!RB_EMPTY_ROOT(
	124	+ &transaction->delayed_refs.href_root.rb_root));
	125	+ WARN_ON(!RB_EMPTY_ROOT(
	126	+ &transaction->delayed_refs.dirty_extent_root));
51	127	if (transaction->delayed_refs.pending_csums)
52	128	btrfs_err(transaction->fs_info,
53	129	"pending csums is %llu",
54	130	transaction->delayed_refs.pending_csums);
55		- while (!list_empty(&transaction->pending_chunks)) {
56		- struct extent_map *em;
57		-
58		- em = list_first_entry(&transaction->pending_chunks,
59		- struct extent_map, list);
60		- list_del_init(&em->list);
61		- free_extent_map(em);
62		- }
63	131	/*
64	132	* If any block groups are found in ->deleted_bgs then it's
65	133	* because the transaction was aborted and a commit did not
..	..	@@ -68,75 +136,90 @@
68	136	* discard the physical locations of the block groups.
69	137	*/
70	138	while (!list_empty(&transaction->deleted_bgs)) {
71		- struct btrfs_block_group_cache *cache;
	139	+ struct btrfs_block_group *cache;
72	140
73	141	cache = list_first_entry(&transaction->deleted_bgs,
74		- struct btrfs_block_group_cache,
	142	+ struct btrfs_block_group,
75	143	bg_list);
76	144	list_del_init(&cache->bg_list);
77		- btrfs_put_block_group_trimming(cache);
	145	+ btrfs_unfreeze_block_group(cache);
78	146	btrfs_put_block_group(cache);
79	147	}
	148	+ WARN_ON(!list_empty(&transaction->dev_update_list));
80	149	kfree(transaction);
81	150	}
82	151	}
83	152
84		-static void clear_btree_io_tree(struct extent_io_tree *tree)
	153	+static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
85	154	{
86		- spin_lock(&tree->lock);
87		- /*
88		- * Do a single barrier for the waitqueue_active check here, the state
89		- * of the waitqueue should not change once clear_btree_io_tree is
90		- * called.
91		- */
92		- smp_mb();
93		- while (!RB_EMPTY_ROOT(&tree->state)) {
94		- struct rb_node *node;
95		- struct extent_state *state;
96		-
97		- node = rb_first(&tree->state);
98		- state = rb_entry(node, struct extent_state, rb_node);
99		- rb_erase(&state->rb_node, &tree->state);
100		- RB_CLEAR_NODE(&state->rb_node);
101		- /*
102		- * btree io trees aren't supposed to have tasks waiting for
103		- * changes in the flags of extent states ever.
104		- */
105		- ASSERT(!waitqueue_active(&state->wq));
106		- free_extent_state(state);
107		-
108		- cond_resched_lock(&tree->lock);
109		- }
110		- spin_unlock(&tree->lock);
111		-}
112		-
113		-static noinline void switch_commit_roots(struct btrfs_transaction *trans)
114		-{
	155	+ struct btrfs_transaction *cur_trans = trans->transaction;
115	156	struct btrfs_fs_info *fs_info = trans->fs_info;
116	157	struct btrfs_root root, tmp;
	158	+ struct btrfs_caching_control caching_ctl, next;
117	159
118	160	down_write(&fs_info->commit_root_sem);
119		- list_for_each_entry_safe(root, tmp, &trans->switch_commits,
	161	+ list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
120	162	dirty_list) {
121	163	list_del_init(&root->dirty_list);
122	164	free_extent_buffer(root->commit_root);
123	165	root->commit_root = btrfs_root_node(root);
124		- if (is_fstree(root->objectid))
	166	+ if (is_fstree(root->root_key.objectid))
125	167	btrfs_unpin_free_ino(root);
126		- clear_btree_io_tree(&root->dirty_log_pages);
	168	+ extent_io_tree_release(&root->dirty_log_pages);
	169	+ btrfs_qgroup_clean_swapped_blocks(root);
127	170	}
128	171
129	172	/* We can free old roots now. */
130		- spin_lock(&trans->dropped_roots_lock);
131		- while (!list_empty(&trans->dropped_roots)) {
132		- root = list_first_entry(&trans->dropped_roots,
	173	+ spin_lock(&cur_trans->dropped_roots_lock);
	174	+ while (!list_empty(&cur_trans->dropped_roots)) {
	175	+ root = list_first_entry(&cur_trans->dropped_roots,
133	176	struct btrfs_root, root_list);
134	177	list_del_init(&root->root_list);
135		- spin_unlock(&trans->dropped_roots_lock);
	178	+ spin_unlock(&cur_trans->dropped_roots_lock);
	179	+ btrfs_free_log(trans, root);
136	180	btrfs_drop_and_free_fs_root(fs_info, root);
137		- spin_lock(&trans->dropped_roots_lock);
	181	+ spin_lock(&cur_trans->dropped_roots_lock);
138	182	}
139		- spin_unlock(&trans->dropped_roots_lock);
	183	+ spin_unlock(&cur_trans->dropped_roots_lock);
	184	+
	185	+ /*
	186	+ * We have to update the last_byte_to_unpin under the commit_root_sem,
	187	+ * at the same time we swap out the commit roots.
	188	+ *
	189	+ * This is because we must have a real view of the last spot the caching
	190	+ * kthreads were while caching. Consider the following views of the
	191	+ * extent tree for a block group
	192	+ *
	193	+ * commit root
	194	+ * +----+----+----+----+----+----+----+
	195	+ * \|\\\\\| \|\\\\\|\\\\\| \|\\\\\|\\\\\|
	196	+ * +----+----+----+----+----+----+----+
	197	+ * 0 1 2 3 4 5 6 7
	198	+ *
	199	+ * new commit root
	200	+ * +----+----+----+----+----+----+----+
	201	+ * \| \| \| \|\\\\\| \| \|\\\\\|
	202	+ * +----+----+----+----+----+----+----+
	203	+ * 0 1 2 3 4 5 6 7
	204	+ *
	205	+ * If the cache_ctl->progress was at 3, then we are only allowed to
	206	+ * unpin [0,1) and [2,3], because the caching thread has already
	207	+ * processed those extents. We are not allowed to unpin [5,6), because
	208	+ * the caching thread will re-start it's search from 3, and thus find
	209	+ * the hole from [4,6) to add to the free space cache.
	210	+ */
	211	+ list_for_each_entry_safe(caching_ctl, next,
	212	+ &fs_info->caching_block_groups, list) {
	213	+ struct btrfs_block_group *cache = caching_ctl->block_group;
	214	+
	215	+ if (btrfs_block_group_done(cache)) {
	216	+ cache->last_byte_to_unpin = (u64)-1;
	217	+ list_del_init(&caching_ctl->list);
	218	+ btrfs_put_caching_control(caching_ctl);
	219	+ } else {
	220	+ cache->last_byte_to_unpin = caching_ctl->progress;
	221	+ }
	222	+ }
140	223	up_write(&fs_info->commit_root_sem);
141	224	}
142	225
..	..	@@ -166,6 +249,24 @@
166	249	}
167	250
168	251	/*
	252	+ * To be called after all the new block groups attached to the transaction
	253	+ * handle have been created (btrfs_create_pending_block_groups()).
	254	+ */
	255	+void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
	256	+{
	257	+ struct btrfs_fs_info *fs_info = trans->fs_info;
	258	+
	259	+ if (!trans->chunk_bytes_reserved)
	260	+ return;
	261	+
	262	+ WARN_ON_ONCE(!list_empty(&trans->new_bgs));
	263	+
	264	+ btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv,
	265	+ trans->chunk_bytes_reserved, NULL);
	266	+ trans->chunk_bytes_reserved = 0;
	267	+}
	268	+
	269	+/*
169	270	* either allocate a new transaction or hop into the existing one
170	271	*/
171	272	static noinline int join_transaction(struct btrfs_fs_info *fs_info,
..	..	@@ -183,7 +284,7 @@
183	284
184	285	cur_trans = fs_info->running_transaction;
185	286	if (cur_trans) {
186		- if (cur_trans->aborted) {
	287	+ if (TRANS_ABORTED(cur_trans)) {
187	288	spin_unlock(&fs_info->trans_lock);
188	289	return cur_trans->aborted;
189	290	}
..	..	@@ -231,24 +332,24 @@
231	332	}
232	333
233	334	cur_trans->fs_info = fs_info;
	335	+ atomic_set(&cur_trans->pending_ordered, 0);
	336	+ init_waitqueue_head(&cur_trans->pending_wait);
234	337	atomic_set(&cur_trans->num_writers, 1);
235	338	extwriter_counter_init(cur_trans, type);
236	339	init_waitqueue_head(&cur_trans->writer_wait);
237	340	init_waitqueue_head(&cur_trans->commit_wait);
238		- init_waitqueue_head(&cur_trans->pending_wait);
239	341	cur_trans->state = TRANS_STATE_RUNNING;
240	342	/*
241	343	* One for this trans handle, one so it will live on until we
242	344	* commit the transaction.
243	345	*/
244	346	refcount_set(&cur_trans->use_count, 2);
245		- atomic_set(&cur_trans->pending_ordered, 0);
246	347	cur_trans->flags = 0;
247	348	cur_trans->start_time = ktime_get_seconds();
248	349
249	350	memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs));
250	351
251		- cur_trans->delayed_refs.href_root = RB_ROOT;
	352	+ cur_trans->delayed_refs.href_root = RB_ROOT_CACHED;
252	353	cur_trans->delayed_refs.dirty_extent_root = RB_ROOT;
253	354	atomic_set(&cur_trans->delayed_refs.num_entries, 0);
254	355
..	..	@@ -266,19 +367,20 @@
266	367	spin_lock_init(&cur_trans->delayed_refs.lock);
267	368
268	369	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
269		- INIT_LIST_HEAD(&cur_trans->pending_chunks);
	370	+ INIT_LIST_HEAD(&cur_trans->dev_update_list);
270	371	INIT_LIST_HEAD(&cur_trans->switch_commits);
271	372	INIT_LIST_HEAD(&cur_trans->dirty_bgs);
272	373	INIT_LIST_HEAD(&cur_trans->io_bgs);
273	374	INIT_LIST_HEAD(&cur_trans->dropped_roots);
274	375	mutex_init(&cur_trans->cache_write_mutex);
275		- cur_trans->num_dirty_bgs = 0;
276	376	spin_lock_init(&cur_trans->dirty_bgs_lock);
277	377	INIT_LIST_HEAD(&cur_trans->deleted_bgs);
278	378	spin_lock_init(&cur_trans->dropped_roots_lock);
279	379	list_add_tail(&cur_trans->list, &fs_info->trans_list);
280		- extent_io_tree_init(&cur_trans->dirty_pages,
281		- fs_info->btree_inode);
	380	+ extent_io_tree_init(fs_info, &cur_trans->dirty_pages,
	381	+ IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode);
	382	+ extent_io_tree_init(fs_info, &cur_trans->pinned_extents,
	383	+ IO_TREE_FS_PINNED_EXTENTS, NULL);
282	384	fs_info->generation++;
283	385	cur_trans->transid = fs_info->generation;
284	386	fs_info->running_transaction = cur_trans;
..	..	@@ -289,10 +391,10 @@
289	391	}
290	392
291	393	/*
292		- * this does all the record keeping required to make sure that a reference
293		- * counted root is properly recorded in a given transaction. This is required
294		- * to make sure the old root from before we joined the transaction is deleted
295		- * when the transaction commits
	394	+ * This does all the record keeping required to make sure that a shareable root
	395	+ * is properly recorded in a given transaction. This is required to make sure
	396	+ * the old root from before we joined the transaction is deleted when the
	397	+ * transaction commits.
296	398	*/
297	399	static int record_root_in_trans(struct btrfs_trans_handle *trans,
298	400	struct btrfs_root *root,
..	..	@@ -300,7 +402,7 @@
300	402	{
301	403	struct btrfs_fs_info *fs_info = root->fs_info;
302	404
303		- if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
	405	+ if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
304	406	root->last_trans < trans->transid) \|\| force) {
305	407	WARN_ON(root == fs_info->extent_root);
306	408	WARN_ON(!force && root->commit_root != root->node);
..	..	@@ -379,7 +481,7 @@
379	481	{
380	482	struct btrfs_fs_info *fs_info = root->fs_info;
381	483
382		- if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
	484	+ if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
383	485	return 0;
384	486
385	487	/*
..	..	@@ -400,9 +502,9 @@
400	502
401	503	static inline int is_transaction_blocked(struct btrfs_transaction *trans)
402	504	{
403		- return (trans->state >= TRANS_STATE_BLOCKED &&
	505	+ return (trans->state >= TRANS_STATE_COMMIT_START &&
404	506	trans->state < TRANS_STATE_UNBLOCKED &&
405		- !trans->aborted);
	507	+ !TRANS_ABORTED(trans));
406	508	}
407	509
408	510	/* wait for commit against the current transaction to become unblocked
..	..	@@ -421,7 +523,7 @@
421	523
422	524	wait_event(fs_info->transaction_wait,
423	525	cur_trans->state >= TRANS_STATE_UNBLOCKED \|\|
424		- cur_trans->aborted);
	526	+ TRANS_ABORTED(cur_trans));
425	527	btrfs_put_transaction(cur_trans);
426	528	} else {
427	529	spin_unlock(&fs_info->trans_lock);
..	..	@@ -444,7 +546,7 @@
444	546	struct btrfs_fs_info *fs_info = root->fs_info;
445	547
446	548	if (!fs_info->reloc_ctl \|\|
447		- !test_bit(BTRFS_ROOT_REF_COWS, &root->state) \|\|
	549	+ !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) \|\|
448	550	root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID \|\|
449	551	root->reloc_root)
450	552	return false;
..	..	@@ -458,12 +560,13 @@
458	560	bool enforce_qgroups)
459	561	{
460	562	struct btrfs_fs_info *fs_info = root->fs_info;
461		-
	563	+ struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
462	564	struct btrfs_trans_handle *h;
463	565	struct btrfs_transaction *cur_trans;
464	566	u64 num_bytes = 0;
465	567	u64 qgroup_reserved = 0;
466	568	bool reloc_reserved = false;
	569	+ bool do_chunk_alloc = false;
467	570	int ret;
468	571
469	572	/* Send isn't supposed to start transactions. */
..	..	@@ -487,13 +590,29 @@
487	590	* the appropriate flushing if need be.
488	591	*/
489	592	if (num_items && root != fs_info->chunk_root) {
	593	+ struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
	594	+ u64 delayed_refs_bytes = 0;
	595	+
490	596	qgroup_reserved = num_items * fs_info->nodesize;
491	597	ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
492	598	enforce_qgroups);
493	599	if (ret)
494	600	return ERR_PTR(ret);
495	601
496		- num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items);
	602	+ /*
	603	+ * We want to reserve all the bytes we may need all at once, so
	604	+ * we only do 1 enospc flushing cycle per transaction start. We
	605	+ * accomplish this by simply assuming we'll do 2 x num_items
	606	+ * worth of delayed refs updates in this trans handle, and
	607	+ * refill that amount for whatever is missing in the reserve.
	608	+ */
	609	+ num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items);
	610	+ if (flush == BTRFS_RESERVE_FLUSH_ALL &&
	611	+ delayed_refs_rsv->full == 0) {
	612	+ delayed_refs_bytes = num_bytes;
	613	+ num_bytes <<= 1;
	614	+ }
	615	+
497	616	/*
498	617	* Do the reservation for the relocation root creation
499	618	*/
..	..	@@ -502,8 +621,27 @@
502	621	reloc_reserved = true;
503	622	}
504	623
505		- ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
506		- num_bytes, flush);
	624	+ ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
	625	+ if (ret)
	626	+ goto reserve_fail;
	627	+ if (delayed_refs_bytes) {
	628	+ btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
	629	+ delayed_refs_bytes);
	630	+ num_bytes -= delayed_refs_bytes;
	631	+ }
	632	+
	633	+ if (rsv->space_info->force_alloc)
	634	+ do_chunk_alloc = true;
	635	+ } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
	636	+ !delayed_refs_rsv->full) {
	637	+ /*
	638	+ * Some people call with btrfs_start_transaction(root, 0)
	639	+ * because they can be throttled, but have some other mechanism
	640	+ * for reserving space. We still want these guys to refill the
	641	+ * delayed block_rsv so just add 1 items worth of reservation
	642	+ * here.
	643	+ */
	644	+ ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
507	645	if (ret)
508	646	goto reserve_fail;
509	647	}
..	..	@@ -556,7 +694,7 @@
556	694	INIT_LIST_HEAD(&h->new_bgs);
557	695
558	696	smp_mb();
559		- if (cur_trans->state >= TRANS_STATE_BLOCKED &&
	697	+ if (cur_trans->state >= TRANS_STATE_COMMIT_START &&
560	698	may_wait_transaction(fs_info, type)) {
561	699	current->journal_info = h;
562	700	btrfs_commit_transaction(h);
..	..	@@ -574,6 +712,19 @@
574	712	got_it:
575	713	if (!current->journal_info)
576	714	current->journal_info = h;
	715	+
	716	+ /*
	717	+ * If the space_info is marked ALLOC_FORCE then we'll get upgraded to
	718	+ * ALLOC_FORCE the first run through, and then we won't allocate for
	719	+ * anybody else who races in later. We don't care about the return
	720	+ * value here.
	721	+ */
	722	+ if (do_chunk_alloc && num_bytes) {
	723	+ u64 flags = h->block_rsv->space_info->flags;
	724	+
	725	+ btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags),
	726	+ CHUNK_ALLOC_NO_FORCE);
	727	+ }
577	728
578	729	/*
579	730	* btrfs_record_root_in_trans() needs to alloc new extents, and may
..	..	@@ -594,7 +745,7 @@
594	745	alloc_fail:
595	746	if (num_bytes)
596	747	btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv,
597		- num_bytes);
	748	+ num_bytes, NULL);
598	749	reserve_fail:
599	750	btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved);
600	751	return ERR_PTR(ret);
..	..	@@ -609,43 +760,10 @@
609	760
610	761	struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
611	762	struct btrfs_root *root,
612		- unsigned int num_items,
613		- int min_factor)
	763	+ unsigned int num_items)
614	764	{
615		- struct btrfs_fs_info *fs_info = root->fs_info;
616		- struct btrfs_trans_handle *trans;
617		- u64 num_bytes;
618		- int ret;
619		-
620		- /*
621		- * We have two callers: unlink and block group removal. The
622		- * former should succeed even if we will temporarily exceed
623		- * quota and the latter operates on the extent root so
624		- * qgroup enforcement is ignored anyway.
625		- */
626		- trans = start_transaction(root, num_items, TRANS_START,
627		- BTRFS_RESERVE_FLUSH_ALL, false);
628		- if (!IS_ERR(trans) \|\| PTR_ERR(trans) != -ENOSPC)
629		- return trans;
630		-
631		- trans = btrfs_start_transaction(root, 0);
632		- if (IS_ERR(trans))
633		- return trans;
634		-
635		- num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items);
636		- ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv,
637		- num_bytes, min_factor);
638		- if (ret) {
639		- btrfs_end_transaction(trans);
640		- return ERR_PTR(ret);
641		- }
642		-
643		- trans->block_rsv = &fs_info->trans_block_rsv;
644		- trans->bytes_reserved = num_bytes;
645		- trace_btrfs_space_reservation(fs_info, "transaction",
646		- trans->transid, num_bytes, 1);
647		-
648		- return trans;
	765	+ return start_transaction(root, num_items, TRANS_START,
	766	+ BTRFS_RESERVE_FLUSH_ALL_STEAL, false);
649	767	}
650	768
651	769	struct btrfs_trans_handle btrfs_join_transaction(struct btrfs_root root)
..	..	@@ -654,7 +772,7 @@
654	772	true);
655	773	}
656	774
657		-struct btrfs_trans_handle btrfs_join_transaction_nolock(struct btrfs_root root)
	775	+struct btrfs_trans_handle btrfs_join_transaction_spacecache(struct btrfs_root root)
658	776	{
659	777	return start_transaction(root, 0, TRANS_JOIN_NOLOCK,
660	778	BTRFS_RESERVE_NO_FLUSH, true);
..	..	@@ -692,7 +810,7 @@
692	810	/*
693	811	* btrfs_attach_transaction_barrier() - catch the running transaction
694	812	*
695		- * It is similar to the above function, the differentia is this one
	813	+ * It is similar to the above function, the difference is this one
696	814	* will wait for all the inactive transactions until they fully
697	815	* complete.
698	816	*/
..	..	@@ -782,7 +900,7 @@
782	900	{
783	901	struct btrfs_fs_info *fs_info = trans->fs_info;
784	902
785		- if (btrfs_check_space_for_delayed_refs(trans, fs_info))
	903	+ if (btrfs_check_space_for_delayed_refs(fs_info))
786	904	return 1;
787	905
788	906	return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
..	..	@@ -791,21 +909,11 @@
791	909	int btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
792	910	{
793	911	struct btrfs_transaction *cur_trans = trans->transaction;
794		- int updates;
795		- int err;
796	912
797	913	smp_mb();
798		- if (cur_trans->state >= TRANS_STATE_BLOCKED \|\|
	914	+ if (cur_trans->state >= TRANS_STATE_COMMIT_START \|\|
799	915	cur_trans->delayed_refs.flushing)
800	916	return 1;
801		-
802		- updates = trans->delayed_ref_updates;
803		- trans->delayed_ref_updates = 0;
804		- if (updates) {
805		- err = btrfs_run_delayed_refs(trans, updates * 2);
806		- if (err) /* Error code will also eval true */
807		- return err;
808		- }
809	917
810	918	return should_end_transaction(trans);
811	919	}
..	..	@@ -827,7 +935,7 @@
827	935	trace_btrfs_space_reservation(fs_info, "transaction",
828	936	trans->transid, trans->bytes_reserved, 0);
829	937	btrfs_block_rsv_release(fs_info, trans->block_rsv,
830		- trans->bytes_reserved);
	938	+ trans->bytes_reserved, NULL);
831	939	trans->bytes_reserved = 0;
832	940	}
833	941
..	..	@@ -836,11 +944,7 @@
836	944	{
837	945	struct btrfs_fs_info *info = trans->fs_info;
838	946	struct btrfs_transaction *cur_trans = trans->transaction;
839		- u64 transid = trans->transid;
840		- unsigned long cur = trans->delayed_ref_updates;
841		- int lock = (trans->type != TRANS_JOIN_NOLOCK);
842	947	int err = 0;
843		- int must_run_delayed_refs = 0;
844	948
845	949	if (refcount_read(&trans->use_count) > 1) {
846	950	refcount_dec(&trans->use_count);
..	..	@@ -851,46 +955,9 @@
851	955	btrfs_trans_release_metadata(trans);
852	956	trans->block_rsv = NULL;
853	957
854		- if (!list_empty(&trans->new_bgs))
855		- btrfs_create_pending_block_groups(trans);
856		-
857		- trans->delayed_ref_updates = 0;
858		- if (!trans->sync) {
859		- must_run_delayed_refs =
860		- btrfs_should_throttle_delayed_refs(trans, info);
861		- cur = max_t(unsigned long, cur, 32);
862		-
863		- /*
864		- * don't make the caller wait if they are from a NOLOCK
865		- * or ATTACH transaction, it will deadlock with commit
866		- */
867		- if (must_run_delayed_refs == 1 &&
868		- (trans->type & (__TRANS_JOIN_NOLOCK \| __TRANS_ATTACH)))
869		- must_run_delayed_refs = 2;
870		- }
871		-
872		- btrfs_trans_release_metadata(trans);
873		- trans->block_rsv = NULL;
874		-
875		- if (!list_empty(&trans->new_bgs))
876		- btrfs_create_pending_block_groups(trans);
	958	+ btrfs_create_pending_block_groups(trans);
877	959
878	960	btrfs_trans_release_chunk_metadata(trans);
879		-
880		- if (lock && should_end_transaction(trans) &&
881		- READ_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) {
882		- spin_lock(&info->trans_lock);
883		- if (cur_trans->state == TRANS_STATE_RUNNING)
884		- cur_trans->state = TRANS_STATE_BLOCKED;
885		- spin_unlock(&info->trans_lock);
886		- }
887		-
888		- if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) {
889		- if (throttle)
890		- return btrfs_commit_transaction(trans);
891		- else
892		- wake_up_process(info->transaction_kthread);
893		- }
894	961
895	962	if (trans->type & __TRANS_FREEZABLE)
896	963	sb_end_intwrite(info->sb);
..	..	@@ -909,17 +976,16 @@
909	976	if (throttle)
910	977	btrfs_run_delayed_iputs(info);
911	978
912		- if (trans->aborted \|\|
	979	+ if (TRANS_ABORTED(trans) \|\|
913	980	test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) {
914	981	wake_up_process(info->transaction_kthread);
915		- err = -EIO;
	982	+ if (TRANS_ABORTED(trans))
	983	+ err = trans->aborted;
	984	+ else
	985	+ err = -EROFS;
916	986	}
917	987
918	988	kmem_cache_free(btrfs_trans_handle_cachep, trans);
919		- if (must_run_delayed_refs) {
920		- btrfs_async_run_delayed_refs(info, cur, transid,
921		- must_run_delayed_refs == 1);
922		- }
923	989	return err;
924	990	}
925	991
..	..	@@ -967,7 +1033,7 @@
967	1033	* superblock that points to btree nodes/leafs for which
968	1034	* writeback hasn't finished yet (and without errors).
969	1035	* We cleanup any entries left in the io tree when committing
970		- * the transaction (through clear_btree_io_tree()).
	1036	+ * the transaction (through extent_io_tree_release()).
971	1037	*/
972	1038	if (err == -ENOMEM) {
973	1039	err = 0;
..	..	@@ -1012,7 +1078,7 @@
1012	1078	* left in the io tree. For a log commit, we don't remove them
1013	1079	* after committing the log because the tree can be accessed
1014	1080	* concurrently - we do it only at transaction commit time when
1015		- * it's safe to do it (through clear_btree_io_tree()).
	1081	+ * it's safe to do it (through extent_io_tree_release()).
1016	1082	*/
1017	1083	err = clear_extent_bit(dirty_pages, start, end,
1018	1084	EXTENT_NEED_WAIT, 0, 0, &cached_state);
..	..	@@ -1032,7 +1098,7 @@
1032	1098	return werr;
1033	1099	}
1034	1100
1035		-int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
	1101	+static int btrfs_wait_extents(struct btrfs_fs_info *fs_info,
1036	1102	struct extent_io_tree *dirty_pages)
1037	1103	{
1038	1104	bool errors = false;
..	..	@@ -1090,7 +1156,7 @@
1090	1156	blk_finish_plug(&plug);
1091	1157	ret2 = btrfs_wait_extents(fs_info, dirty_pages);
1092	1158
1093		- clear_btree_io_tree(&trans->transaction->dirty_pages);
	1159	+ extent_io_tree_release(&trans->transaction->dirty_pages);
1094	1160
1095	1161	if (ret)
1096	1162	return ret;
..	..	@@ -1158,7 +1224,7 @@
1158	1224
1159	1225	eb = btrfs_lock_root_node(fs_info->tree_root);
1160	1226	ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
1161		- 0, &eb);
	1227	+ 0, &eb, BTRFS_NESTING_COW);
1162	1228	btrfs_tree_unlock(eb);
1163	1229	free_extent_buffer(eb);
1164	1230
..	..	@@ -1169,17 +1235,17 @@
1169	1235	if (ret)
1170	1236	return ret;
1171	1237
1172		- ret = btrfs_run_dev_stats(trans, fs_info);
	1238	+ ret = btrfs_run_dev_stats(trans);
1173	1239	if (ret)
1174	1240	return ret;
1175		- ret = btrfs_run_dev_replace(trans, fs_info);
	1241	+ ret = btrfs_run_dev_replace(trans);
1176	1242	if (ret)
1177	1243	return ret;
1178	1244	ret = btrfs_run_qgroups(trans);
1179	1245	if (ret)
1180	1246	return ret;
1181	1247
1182		- ret = btrfs_setup_space_cache(trans, fs_info);
	1248	+ ret = btrfs_setup_space_cache(trans);
1183	1249	if (ret)
1184	1250	return ret;
1185	1251
..	..	@@ -1207,7 +1273,7 @@
1207	1273	}
1208	1274
1209	1275	while (!list_empty(dirty_bgs) \|\| !list_empty(io_bgs)) {
1210		- ret = btrfs_write_dirty_block_groups(trans, fs_info);
	1276	+ ret = btrfs_write_dirty_block_groups(trans);
1211	1277	if (ret)
1212	1278	return ret;
1213	1279	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
..	..	@@ -1220,7 +1286,10 @@
1220	1286
1221	1287	list_add_tail(&fs_info->extent_root->dirty_list,
1222	1288	&trans->transaction->switch_commits);
1223		- btrfs_after_dev_replace_commit(fs_info);
	1289	+
	1290	+ /* Update dev-replace pointer once everything is committed */
	1291	+ fs_info->dev_replace.committed_cursor_left =
	1292	+ fs_info->dev_replace.cursor_left_last_write_of_item;
1224	1293
1225	1294	return 0;
1226	1295	}
..	..	@@ -1235,8 +1304,10 @@
1235	1304	struct btrfs_fs_info *fs_info = root->fs_info;
1236	1305
1237	1306	spin_lock(&fs_info->trans_lock);
1238		- if (list_empty(&root->root_list))
	1307	+ if (list_empty(&root->root_list)) {
	1308	+ btrfs_grab_root(root);
1239	1309	list_add_tail(&root->root_list, &fs_info->dead_roots);
	1310	+ }
1240	1311	spin_unlock(&fs_info->trans_lock);
1241	1312	}
1242	1313
..	..	@@ -1360,7 +1431,7 @@
1360	1431	return 0;
1361	1432
1362	1433	/*
1363		- * Ensure dirty @src will be commited. Or, after comming
	1434	+ * Ensure dirty @src will be committed. Or, after coming
1364	1435	* commit_fs_roots() and switch_commit_roots(), any dirty but not
1365	1436	* recorded root will never be updated again, causing an outdated root
1366	1437	* item.
..	..	@@ -1401,7 +1472,7 @@
1401	1472	ret = commit_cowonly_roots(trans);
1402	1473	if (ret)
1403	1474	goto out;
1404		- switch_commit_roots(trans->transaction);
	1475	+ switch_commit_roots(trans);
1405	1476	ret = btrfs_write_and_wait_transaction(trans);
1406	1477	if (ret)
1407	1478	btrfs_handle_fs_error(fs_info, ret,
..	..	@@ -1453,7 +1524,6 @@
1453	1524	u64 index = 0;
1454	1525	u64 objectid;
1455	1526	u64 root_flags;
1456		- uuid_le new_uuid;
1457	1527
1458	1528	ASSERT(pending->path);
1459	1529	path = pending->path;
..	..	@@ -1546,8 +1616,7 @@
1546	1616
1547	1617	btrfs_set_root_generation_v2(new_root_item,
1548	1618	trans->transid);
1549		- uuid_le_gen(&new_uuid);
1550		- memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
	1619	+ generate_random_guid(new_root_item->uuid);
1551	1620	memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1552	1621	BTRFS_UUID_SIZE);
1553	1622	if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
..	..	@@ -1563,7 +1632,8 @@
1563	1632	btrfs_set_root_otransid(new_root_item, trans->transid);
1564	1633
1565	1634	old = btrfs_lock_root_node(root);
1566		- ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
	1635	+ ret = btrfs_cow_block(trans, root, old, NULL, 0, &old,
	1636	+ BTRFS_NESTING_COW);
1567	1637	if (ret) {
1568	1638	btrfs_tree_unlock(old);
1569	1639	free_extent_buffer(old);
..	..	@@ -1571,7 +1641,7 @@
1571	1641	goto fail;
1572	1642	}
1573	1643
1574		- btrfs_set_lock_blocking(old);
	1644	+ btrfs_set_lock_blocking_write(old);
1575	1645
1576	1646	ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1577	1647	/* clean up in any case */
..	..	@@ -1609,9 +1679,10 @@
1609	1679	}
1610	1680
1611	1681	key.offset = (u64)-1;
1612		- pending->snap = btrfs_read_fs_root_no_name(fs_info, &key);
	1682	+ pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev);
1613	1683	if (IS_ERR(pending->snap)) {
1614	1684	ret = PTR_ERR(pending->snap);
	1685	+ pending->snap = NULL;
1615	1686	btrfs_abort_transaction(trans, ret);
1616	1687	goto fail;
1617	1688	}
..	..	@@ -1639,10 +1710,9 @@
1639	1710	if (ret < 0)
1640	1711	goto fail;
1641	1712
1642		- ret = btrfs_insert_dir_item(trans, parent_root,
1643		- dentry->d_name.name, dentry->d_name.len,
1644		- BTRFS_I(parent_inode), &key,
1645		- BTRFS_FT_DIR, index);
	1713	+ ret = btrfs_insert_dir_item(trans, dentry->d_name.name,
	1714	+ dentry->d_name.len, BTRFS_I(parent_inode),
	1715	+ &key, BTRFS_FT_DIR, index);
1646	1716	/* We have check then name at the beginning, so it is impossible. */
1647	1717	BUG_ON(ret == -EEXIST \|\| ret == -EOVERFLOW);
1648	1718	if (ret) {
..	..	@@ -1659,7 +1729,8 @@
1659	1729	btrfs_abort_transaction(trans, ret);
1660	1730	goto fail;
1661	1731	}
1662		- ret = btrfs_uuid_tree_add(trans, new_uuid.b, BTRFS_UUID_KEY_SUBVOL,
	1732	+ ret = btrfs_uuid_tree_add(trans, new_root_item->uuid,
	1733	+ BTRFS_UUID_KEY_SUBVOL,
1663	1734	objectid);
1664	1735	if (ret) {
1665	1736	btrfs_abort_transaction(trans, ret);
..	..	@@ -1771,7 +1842,8 @@
1771	1842	struct btrfs_transaction *trans)
1772	1843	{
1773	1844	wait_event(fs_info->transaction_blocked_wait,
1774		- trans->state >= TRANS_STATE_COMMIT_START \|\| trans->aborted);
	1845	+ trans->state >= TRANS_STATE_COMMIT_START \|\|
	1846	+ TRANS_ABORTED(trans));
1775	1847	}
1776	1848
1777	1849	/*
..	..	@@ -1783,7 +1855,8 @@
1783	1855	struct btrfs_transaction *trans)
1784	1856	{
1785	1857	wait_event(fs_info->transaction_wait,
1786		- trans->state >= TRANS_STATE_UNBLOCKED \|\| trans->aborted);
	1858	+ trans->state >= TRANS_STATE_UNBLOCKED \|\|
	1859	+ TRANS_ABORTED(trans));
1787	1860	}
1788	1861
1789	1862	/*
..	..	@@ -1865,7 +1938,6 @@
1865	1938	{
1866	1939	struct btrfs_fs_info *fs_info = trans->fs_info;
1867	1940	struct btrfs_transaction *cur_trans = trans->transaction;
1868		- DEFINE_WAIT(wait);
1869	1941
1870	1942	WARN_ON(refcount_read(&trans->use_count) > 1);
1871	1943
..	..	@@ -1880,7 +1952,6 @@
1880	1952	*/
1881	1953	BUG_ON(list_empty(&cur_trans->list));
1882	1954
1883		- list_del_init(&cur_trans->list);
1884	1955	if (cur_trans == fs_info->running_transaction) {
1885	1956	cur_trans->state = TRANS_STATE_COMMIT_DOING;
1886	1957	spin_unlock(&fs_info->trans_lock);
..	..	@@ -1889,6 +1960,17 @@
1889	1960
1890	1961	spin_lock(&fs_info->trans_lock);
1891	1962	}
	1963	+
	1964	+ /*
	1965	+ * Now that we know no one else is still using the transaction we can
	1966	+ * remove the transaction from the list of transactions. This avoids
	1967	+ * the transaction kthread from cleaning up the transaction while some
	1968	+ * other task is still using it, which could result in a use-after-free
	1969	+ * on things like log trees, as it forces the transaction kthread to
	1970	+ * wait for this transaction to be cleaned up by us.
	1971	+ */
	1972	+ list_del_init(&cur_trans->list);
	1973	+
1892	1974	spin_unlock(&fs_info->trans_lock);
1893	1975
1894	1976	btrfs_cleanup_one_transaction(trans->transaction, fs_info);
..	..	@@ -1912,8 +1994,25 @@
1912	1994	kmem_cache_free(btrfs_trans_handle_cachep, trans);
1913	1995	}
1914	1996
1915		-static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info)
	1997	+/*
	1998	+ * Release reserved delayed ref space of all pending block groups of the
	1999	+ * transaction and remove them from the list
	2000	+ */
	2001	+static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans)
1916	2002	{
	2003	+ struct btrfs_fs_info *fs_info = trans->fs_info;
	2004	+ struct btrfs_block_group block_group, tmp;
	2005	+
	2006	+ list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
	2007	+ btrfs_delayed_refs_rsv_release(fs_info, 1);
	2008	+ list_del_init(&block_group->bg_list);
	2009	+ }
	2010	+}
	2011	+
	2012	+static inline int btrfs_start_delalloc_flush(struct btrfs_trans_handle *trans)
	2013	+{
	2014	+ struct btrfs_fs_info *fs_info = trans->fs_info;
	2015	+
1917	2016	/*
1918	2017	* We use writeback_inodes_sb here because if we used
1919	2018	* btrfs_start_delalloc_roots we would deadlock with fs freeze.
..	..	@@ -1923,22 +2022,50 @@
1923	2022	* from already being in a transaction and our join_transaction doesn't
1924	2023	* have to re-take the fs freeze lock.
1925	2024	*/
1926		- if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
	2025	+ if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) {
1927	2026	writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC);
	2027	+ } else {
	2028	+ struct btrfs_pending_snapshot *pending;
	2029	+ struct list_head *head = &trans->transaction->pending_snapshots;
	2030	+
	2031	+ /*
	2032	+ * Flush dellaloc for any root that is going to be snapshotted.
	2033	+ * This is done to avoid a corrupted version of files, in the
	2034	+ * snapshots, that had both buffered and direct IO writes (even
	2035	+ * if they were done sequentially) due to an unordered update of
	2036	+ * the inode's size on disk.
	2037	+ */
	2038	+ list_for_each_entry(pending, head, list) {
	2039	+ int ret;
	2040	+
	2041	+ ret = btrfs_start_delalloc_snapshot(pending->root);
	2042	+ if (ret)
	2043	+ return ret;
	2044	+ }
	2045	+ }
1928	2046	return 0;
1929	2047	}
1930	2048
1931		-static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info)
	2049	+static inline void btrfs_wait_delalloc_flush(struct btrfs_trans_handle *trans)
1932	2050	{
1933		- if (btrfs_test_opt(fs_info, FLUSHONCOMMIT))
1934		- btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1935		-}
	2051	+ struct btrfs_fs_info *fs_info = trans->fs_info;
1936	2052
1937		-static inline void
1938		-btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans)
1939		-{
1940		- wait_event(cur_trans->pending_wait,
1941		- atomic_read(&cur_trans->pending_ordered) == 0);
	2053	+ if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) {
	2054	+ btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
	2055	+ } else {
	2056	+ struct btrfs_pending_snapshot *pending;
	2057	+ struct list_head *head = &trans->transaction->pending_snapshots;
	2058	+
	2059	+ /*
	2060	+ * Wait for any dellaloc that we started previously for the roots
	2061	+ * that are going to be snapshotted. This is to avoid a corrupted
	2062	+ * version of files in the snapshots that had both buffered and
	2063	+ * direct IO writes (even if they were done sequentially).
	2064	+ */
	2065	+ list_for_each_entry(pending, head, list)
	2066	+ btrfs_wait_ordered_extents(pending->root,
	2067	+ U64_MAX, 0, U64_MAX);
	2068	+ }
1942	2069	}
1943	2070
1944	2071	int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
..	..	@@ -1947,6 +2074,8 @@
1947	2074	struct btrfs_transaction *cur_trans = trans->transaction;
1948	2075	struct btrfs_transaction *prev_trans = NULL;
1949	2076	int ret;
	2077	+
	2078	+ ASSERT(refcount_read(&trans->use_count) == 1);
1950	2079
1951	2080	/*
1952	2081	* Some places just start a transaction to commit it. We need to make
..	..	@@ -1957,7 +2086,7 @@
1957	2086	trans->dirty = true;
1958	2087
1959	2088	/* Stop the commit early if ->aborted is set */
1960		- if (unlikely(READ_ONCE(cur_trans->aborted))) {
	2089	+ if (TRANS_ABORTED(cur_trans)) {
1961	2090	ret = cur_trans->aborted;
1962	2091	btrfs_end_transaction(trans);
1963	2092	return ret;
..	..	@@ -1984,8 +2113,7 @@
1984	2113	cur_trans->delayed_refs.flushing = 1;
1985	2114	smp_wmb();
1986	2115
1987		- if (!list_empty(&trans->new_bgs))
1988		- btrfs_create_pending_block_groups(trans);
	2116	+ btrfs_create_pending_block_groups(trans);
1989	2117
1990	2118	ret = btrfs_run_delayed_refs(trans, 0);
1991	2119	if (ret) {
..	..	@@ -2032,7 +2160,7 @@
2032	2160
2033	2161	wait_for_commit(cur_trans);
2034	2162
2035		- if (unlikely(cur_trans->aborted))
	2163	+ if (TRANS_ABORTED(cur_trans))
2036	2164	ret = cur_trans->aborted;
2037	2165
2038	2166	btrfs_put_transaction(cur_trans);
..	..	@@ -2051,7 +2179,7 @@
2051	2179	spin_unlock(&fs_info->trans_lock);
2052	2180
2053	2181	wait_for_commit(prev_trans);
2054		- ret = prev_trans->aborted;
	2182	+ ret = READ_ONCE(prev_trans->aborted);
2055	2183
2056	2184	btrfs_put_transaction(prev_trans);
2057	2185	if (ret)
..	..	@@ -2075,7 +2203,7 @@
2075	2203
2076	2204	extwriter_counter_dec(cur_trans, trans->type);
2077	2205
2078		- ret = btrfs_start_delalloc_flush(fs_info);
	2206	+ ret = btrfs_start_delalloc_flush(trans);
2079	2207	if (ret)
2080	2208	goto cleanup_transaction;
2081	2209
..	..	@@ -2091,9 +2219,15 @@
2091	2219	if (ret)
2092	2220	goto cleanup_transaction;
2093	2221
2094		- btrfs_wait_delalloc_flush(fs_info);
	2222	+ btrfs_wait_delalloc_flush(trans);
2095	2223
2096		- btrfs_wait_pending_ordered(cur_trans);
	2224	+ /*
	2225	+ * Wait for all ordered extents started by a fast fsync that joined this
	2226	+ * transaction. Otherwise if this transaction commits before the ordered
	2227	+ * extents complete we lose logged data after a power failure.
	2228	+ */
	2229	+ wait_event(cur_trans->pending_wait,
	2230	+ atomic_read(&cur_trans->pending_ordered) == 0);
2097	2231
2098	2232	btrfs_scrub_pause(fs_info);
2099	2233	/*
..	..	@@ -2107,8 +2241,7 @@
2107	2241	wait_event(cur_trans->writer_wait,
2108	2242	atomic_read(&cur_trans->num_writers) == 1);
2109	2243
2110		- /* ->aborted might be set after the previous check, so check it */
2111		- if (unlikely(READ_ONCE(cur_trans->aborted))) {
	2244	+ if (TRANS_ABORTED(cur_trans)) {
2112	2245	ret = cur_trans->aborted;
2113	2246	goto scrub_continue;
2114	2247	}
..	..	@@ -2125,10 +2258,8 @@
2125	2258	* core function of the snapshot creation.
2126	2259	*/
2127	2260	ret = create_pending_snapshots(trans);
2128		- if (ret) {
2129		- mutex_unlock(&fs_info->reloc_mutex);
2130		- goto scrub_continue;
2131		- }
	2261	+ if (ret)
	2262	+ goto unlock_reloc;
2132	2263
2133	2264	/*
2134	2265	* We insert the dir indexes of the snapshots and update the inode
..	..	@@ -2141,16 +2272,12 @@
2141	2272	* the nodes and leaves.
2142	2273	*/
2143	2274	ret = btrfs_run_delayed_items(trans);
2144		- if (ret) {
2145		- mutex_unlock(&fs_info->reloc_mutex);
2146		- goto scrub_continue;
2147		- }
	2275	+ if (ret)
	2276	+ goto unlock_reloc;
2148	2277
2149	2278	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2150		- if (ret) {
2151		- mutex_unlock(&fs_info->reloc_mutex);
2152		- goto scrub_continue;
2153		- }
	2279	+ if (ret)
	2280	+ goto unlock_reloc;
2154	2281
2155	2282	/*
2156	2283	* make sure none of the code above managed to slip in a
..	..	@@ -2176,11 +2303,8 @@
2176	2303	mutex_lock(&fs_info->tree_log_mutex);
2177	2304
2178	2305	ret = commit_fs_roots(trans);
2179		- if (ret) {
2180		- mutex_unlock(&fs_info->tree_log_mutex);
2181		- mutex_unlock(&fs_info->reloc_mutex);
2182		- goto scrub_continue;
2183		- }
	2306	+ if (ret)
	2307	+ goto unlock_tree_log;
2184	2308
2185	2309	/*
2186	2310	* Since the transaction is done, we can apply the pending changes
..	..	@@ -2198,42 +2322,29 @@
2198	2322	* new delayed refs. Must handle them or qgroup can be wrong.
2199	2323	*/
2200	2324	ret = btrfs_run_delayed_refs(trans, (unsigned long)-1);
2201		- if (ret) {
2202		- mutex_unlock(&fs_info->tree_log_mutex);
2203		- mutex_unlock(&fs_info->reloc_mutex);
2204		- goto scrub_continue;
2205		- }
	2325	+ if (ret)
	2326	+ goto unlock_tree_log;
2206	2327
2207	2328	/*
2208	2329	* Since fs roots are all committed, we can get a quite accurate
2209	2330	* new_roots. So let's do quota accounting.
2210	2331	*/
2211	2332	ret = btrfs_qgroup_account_extents(trans);
2212		- if (ret < 0) {
2213		- mutex_unlock(&fs_info->tree_log_mutex);
2214		- mutex_unlock(&fs_info->reloc_mutex);
2215		- goto scrub_continue;
2216		- }
	2333	+ if (ret < 0)
	2334	+ goto unlock_tree_log;
2217	2335
2218	2336	ret = commit_cowonly_roots(trans);
2219		- if (ret) {
2220		- mutex_unlock(&fs_info->tree_log_mutex);
2221		- mutex_unlock(&fs_info->reloc_mutex);
2222		- goto scrub_continue;
2223		- }
	2337	+ if (ret)
	2338	+ goto unlock_tree_log;
2224	2339
2225	2340	/*
2226	2341	* The tasks which save the space cache and inode cache may also
2227	2342	* update ->aborted, check it.
2228	2343	*/
2229		- if (unlikely(READ_ONCE(cur_trans->aborted))) {
	2344	+ if (TRANS_ABORTED(cur_trans)) {
2230	2345	ret = cur_trans->aborted;
2231		- mutex_unlock(&fs_info->tree_log_mutex);
2232		- mutex_unlock(&fs_info->reloc_mutex);
2233		- goto scrub_continue;
	2346	+ goto unlock_tree_log;
2234	2347	}
2235		-
2236		- btrfs_prepare_extent_commit(fs_info);
2237	2348
2238	2349	cur_trans = fs_info->running_transaction;
2239	2350
..	..	@@ -2247,7 +2358,7 @@
2247	2358	list_add_tail(&fs_info->chunk_root->dirty_list,
2248	2359	&cur_trans->switch_commits);
2249	2360
2250		- switch_commit_roots(cur_trans);
	2361	+ switch_commit_roots(trans);
2251	2362
2252	2363	ASSERT(list_empty(&cur_trans->dirty_bgs));
2253	2364	ASSERT(list_empty(&cur_trans->io_bgs));
..	..	@@ -2258,8 +2369,7 @@
2258	2369	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2259	2370	sizeof(*fs_info->super_copy));
2260	2371
2261		- btrfs_update_commit_device_size(fs_info);
2262		- btrfs_update_commit_device_bytes_used(cur_trans);
	2372	+ btrfs_commit_device_sizes(cur_trans);
2263	2373
2264	2374	clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2265	2375	clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
..	..	@@ -2278,6 +2388,10 @@
2278	2388	if (ret) {
2279	2389	btrfs_handle_fs_error(fs_info, ret,
2280	2390	"Error while writing out transaction");
	2391	+ /*
	2392	+ * reloc_mutex has been unlocked, tree_log_mutex is still held
	2393	+ * but we can't jump to unlock_tree_log causing double unlock
	2394	+ */
2281	2395	mutex_unlock(&fs_info->tree_log_mutex);
2282	2396	goto scrub_continue;
2283	2397	}
..	..	@@ -2303,7 +2417,6 @@
2303	2417	*/
2304	2418	cur_trans->state = TRANS_STATE_COMPLETED;
2305	2419	wake_up(&cur_trans->commit_wait);
2306		- clear_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags);
2307	2420
2308	2421	spin_lock(&fs_info->trans_lock);
2309	2422	list_del_init(&cur_trans->list);
..	..	@@ -2326,10 +2439,15 @@
2326	2439
2327	2440	return ret;
2328	2441
	2442	+unlock_tree_log:
	2443	+ mutex_unlock(&fs_info->tree_log_mutex);
	2444	+unlock_reloc:
	2445	+ mutex_unlock(&fs_info->reloc_mutex);
2329	2446	scrub_continue:
2330	2447	btrfs_scrub_continue(fs_info);
2331	2448	cleanup_transaction:
2332	2449	btrfs_trans_release_metadata(trans);
	2450	+ btrfs_cleanup_pending_block_groups(trans);
2333	2451	btrfs_trans_release_chunk_metadata(trans);
2334	2452	trans->block_rsv = NULL;
2335	2453	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
..	..	@@ -2365,16 +2483,21 @@
2365	2483	list_del_init(&root->root_list);
2366	2484	spin_unlock(&fs_info->trans_lock);
2367	2485
2368		- btrfs_debug(fs_info, "cleaner removing %llu", root->objectid);
	2486	+ btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid);
2369	2487
2370	2488	btrfs_kill_all_delayed_nodes(root);
	2489	+ if (root->ino_cache_inode) {
	2490	+ iput(root->ino_cache_inode);
	2491	+ root->ino_cache_inode = NULL;
	2492	+ }
2371	2493
2372	2494	if (btrfs_header_backref_rev(root->node) <
2373	2495	BTRFS_MIXED_BACKREF_REV)
2374		- ret = btrfs_drop_snapshot(root, NULL, 0, 0);
	2496	+ ret = btrfs_drop_snapshot(root, 0, 0);
2375	2497	else
2376		- ret = btrfs_drop_snapshot(root, NULL, 1, 0);
	2498	+ ret = btrfs_drop_snapshot(root, 1, 0);
2377	2499
	2500	+ btrfs_put_root(root);
2378	2501	return (ret < 0) ? 0 : 1;
2379	2502	}
2380	2503