[PATCH 00/42][v5] My current patch queue

[PATCH 00/42][v5] My current patch queue

[Josef Bacik]

v3->v4:
- added stacktraces to all the changelogs
- added the various reviewed-by's.
- fixed the loop in inode_rsv_refill to not use goto again;

v2->v3:
- reworked the truncate/evict throttling, we were still occasionally hitting
  enospc aborts in production in these paths because we were too aggressive with
  space usage.
- reworked the delayed iput stuff to be a little less racey and less deadlocky.
- Addressed the comments from Dave and Omar.
- A lot of production testing.

v1->v2:
- addressed all of the issues brought up.
- added more comments.
- split up some patches.

original message:

This is the current queue of things that I've been working on.  The main thing
these patches are doing is separating out the delayed refs reservations from the
global reserve into their own block rsv.  We have been consistently hitting
issues in production where we abort a transaction because we run out of the
global reserve either while running delayed refs or while updating dirty block
groups.  This is because the math around global reserves is made up bullshit
magic that has been tweaked more and more throughout the years.  The result is
something that is inconsistent across the board and sometimes wrong.  So instead
we need a way to know exactly how much space we need to keep around in order to
satisfy our outstanding delayed refs and our dirty block groups.

Since we don't know how many delayed refs we need at the start of any
modification we simply use the nr_items passed into btrfs_start_transaction() as
a guess for what we may need.  This has the side effect of putting more pressure
on the ENOSPC system, but it's pressure we can deal with more intelligently
because we always know how much space we have outstanding, instead of guessing
with weird global reserve math.

This works similar to every other reservation we have, we reserve the worst case
up front, and then at transaction end time we free up any space we didn't
actually use for delayed refs.

My performance tests show that we are bit faster now since we can do more
intelligent flushing and don't have to fall back on simply committing the
transaction in hopes that we have enough space for everything we need to do.

That leads me to the 2nd part of this pull, there's a bunch of fixes around
ENOSPC.  Because we are a bit faster now there were a bunch of things uncovered
in testing, but they seem to be all resolved now.

The final chunk of fixes are around transaction aborts.  There were a lot of
accounting bugs I was running into while running generic/435, so I fixed a bunch
of those up so now it runs cleanly.

I have been running these patches through xfstests on multiple machines for a
while, they are pretty solid and ready for wider testing and review.  Thanks,

Josef

[PATCH 01/42] btrfs: add btrfs_delete_ref_head helper

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

We do this dance in cleanup_ref_head and check_ref_cleanup, unify it
into a helper and cleanup the calling functions.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
---
 fs/btrfs/delayed-ref.c | 14 ++++++++++++++
 fs/btrfs/delayed-ref.h |  3 ++-
 fs/btrfs/extent-tree.c | 22 +++-------------------
 3 files changed, 19 insertions(+), 20 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index 62ff545ba1f7..3a9e4ac21794 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -393,6 +393,20 @@ btrfs_select_ref_head(struct btrfs_trans_handle *trans)
 	return head;
 }
 
+void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
+			   struct btrfs_delayed_ref_head *head)
+{
+	lockdep_assert_held(&delayed_refs->lock);
+	lockdep_assert_held(&head->lock);
+
+	rb_erase(&head->href_node, &delayed_refs->href_root);
+	RB_CLEAR_NODE(&head->href_node);
+	atomic_dec(&delayed_refs->num_entries);
+	delayed_refs->num_heads--;
+	if (head->processing == 0)
+		delayed_refs->num_heads_ready--;
+}
+
 /*
  * Helper to insert the ref_node to the tail or merge with tail.
  *
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
index d9f2a4ebd5db..7769177b489e 100644
--- a/fs/btrfs/delayed-ref.h
+++ b/fs/btrfs/delayed-ref.h
@@ -261,7 +261,8 @@ static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
 {
 	mutex_unlock(&head->mutex);
 }
-
+void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
+			   struct btrfs_delayed_ref_head *head);
 
 struct btrfs_delayed_ref_head *
 btrfs_select_ref_head(struct btrfs_trans_handle *trans);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index f77226d8020a..d24a0de4a2e7 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2492,12 +2492,9 @@ static int cleanup_ref_head(struct btrfs_trans_handle *trans,
 		spin_unlock(&delayed_refs->lock);
 		return 1;
 	}
-	delayed_refs->num_heads--;
-	rb_erase(&head->href_node, &delayed_refs->href_root);
-	RB_CLEAR_NODE(&head->href_node);
+	btrfs_delete_ref_head(delayed_refs, head);
 	spin_unlock(&head->lock);
 	spin_unlock(&delayed_refs->lock);
-	atomic_dec(&delayed_refs->num_entries);
 
 	trace_run_delayed_ref_head(fs_info, head, 0);
 
@@ -6984,22 +6981,9 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	if (!mutex_trylock(&head->mutex))
 		goto out;
 
-	/*
-	 * at this point we have a head with no other entries.  Go
-	 * ahead and process it.
-	 */
-	rb_erase(&head->href_node, &delayed_refs->href_root);
-	RB_CLEAR_NODE(&head->href_node);
-	atomic_dec(&delayed_refs->num_entries);
-
-	/*
-	 * we don't take a ref on the node because we're removing it from the
-	 * tree, so we just steal the ref the tree was holding.
-	 */
-	delayed_refs->num_heads--;
-	if (head->processing == 0)
-		delayed_refs->num_heads_ready--;
+	btrfs_delete_ref_head(delayed_refs, head);
 	head->processing = 0;
+
 	spin_unlock(&head->lock);
 	spin_unlock(&delayed_refs->lock);
 
-- 
2.14.3

[PATCH 02/42] btrfs: add cleanup_ref_head_accounting helper

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

We were missing some quota cleanups in check_ref_cleanup, so break the
ref head accounting cleanup into a helper and call that from both
check_ref_cleanup and cleanup_ref_head.  This will hopefully ensure that
we don't screw up accounting in the future for other things that we add.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
---
 fs/btrfs/extent-tree.c | 67 +++++++++++++++++++++++++++++---------------------
 1 file changed, 39 insertions(+), 28 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index d24a0de4a2e7..a44d55e36e11 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2461,6 +2461,41 @@ static int cleanup_extent_op(struct btrfs_trans_handle *trans,
 	return ret ? ret : 1;
 }
 
+static void cleanup_ref_head_accounting(struct btrfs_trans_handle *trans,
+					struct btrfs_delayed_ref_head *head)
+{
+	struct btrfs_fs_info *fs_info = trans->fs_info;
+	struct btrfs_delayed_ref_root *delayed_refs =
+		&trans->transaction->delayed_refs;
+
+	if (head->total_ref_mod < 0) {
+		struct btrfs_space_info *space_info;
+		u64 flags;
+
+		if (head->is_data)
+			flags = BTRFS_BLOCK_GROUP_DATA;
+		else if (head->is_system)
+			flags = BTRFS_BLOCK_GROUP_SYSTEM;
+		else
+			flags = BTRFS_BLOCK_GROUP_METADATA;
+		space_info = __find_space_info(fs_info, flags);
+		ASSERT(space_info);
+		percpu_counter_add_batch(&space_info->total_bytes_pinned,
+				   -head->num_bytes,
+				   BTRFS_TOTAL_BYTES_PINNED_BATCH);
+
+		if (head->is_data) {
+			spin_lock(&delayed_refs->lock);
+			delayed_refs->pending_csums -= head->num_bytes;
+			spin_unlock(&delayed_refs->lock);
+		}
+	}
+
+	/* Also free its reserved qgroup space */
+	btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
+				      head->qgroup_reserved);
+}
+
 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
 			    struct btrfs_delayed_ref_head *head)
 {
@@ -2496,31 +2531,6 @@ static int cleanup_ref_head(struct btrfs_trans_handle *trans,
 	spin_unlock(&head->lock);
 	spin_unlock(&delayed_refs->lock);
 
-	trace_run_delayed_ref_head(fs_info, head, 0);
-
-	if (head->total_ref_mod < 0) {
-		struct btrfs_space_info *space_info;
-		u64 flags;
-
-		if (head->is_data)
-			flags = BTRFS_BLOCK_GROUP_DATA;
-		else if (head->is_system)
-			flags = BTRFS_BLOCK_GROUP_SYSTEM;
-		else
-			flags = BTRFS_BLOCK_GROUP_METADATA;
-		space_info = __find_space_info(fs_info, flags);
-		ASSERT(space_info);
-		percpu_counter_add_batch(&space_info->total_bytes_pinned,
-				   -head->num_bytes,
-				   BTRFS_TOTAL_BYTES_PINNED_BATCH);
-
-		if (head->is_data) {
-			spin_lock(&delayed_refs->lock);
-			delayed_refs->pending_csums -= head->num_bytes;
-			spin_unlock(&delayed_refs->lock);
-		}
-	}
-
 	if (head->must_insert_reserved) {
 		btrfs_pin_extent(fs_info, head->bytenr,
 				 head->num_bytes, 1);
@@ -2530,9 +2540,9 @@ static int cleanup_ref_head(struct btrfs_trans_handle *trans,
 		}
 	}
 
-	/* Also free its reserved qgroup space */
-	btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
-				      head->qgroup_reserved);
+	cleanup_ref_head_accounting(trans, head);
+
+	trace_run_delayed_ref_head(fs_info, head, 0);
 	btrfs_delayed_ref_unlock(head);
 	btrfs_put_delayed_ref_head(head);
 	return 0;
@@ -6991,6 +7001,7 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	if (head->must_insert_reserved)
 		ret = 1;
 
+	cleanup_ref_head_accounting(trans, head);
 	mutex_unlock(&head->mutex);
 	btrfs_put_delayed_ref_head(head);
 	return ret;
-- 
2.14.3

[PATCH 03/42] btrfs: cleanup extent_op handling

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

The cleanup_extent_op function actually would run the extent_op if it
needed running, which made the name sort of a misnomer.  Change it to
run_and_cleanup_extent_op, and move the actual cleanup work to
cleanup_extent_op so it can be used by check_ref_cleanup() in order to
unify the extent op handling.

Signed-off-by: Josef Bacik <jbacik@fb.com>
---
 fs/btrfs/extent-tree.c | 36 +++++++++++++++++++++++-------------
 1 file changed, 23 insertions(+), 13 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index a44d55e36e11..98f36dfeccb0 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2442,19 +2442,33 @@ static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_ref
 	btrfs_delayed_ref_unlock(head);
 }
 
-static int cleanup_extent_op(struct btrfs_trans_handle *trans,
-			     struct btrfs_delayed_ref_head *head)
+static struct btrfs_delayed_extent_op *
+cleanup_extent_op(struct btrfs_trans_handle *trans,
+		  struct btrfs_delayed_ref_head *head)
 {
 	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
-	int ret;
 
 	if (!extent_op)
-		return 0;
-	head->extent_op = NULL;
+		return NULL;
+
 	if (head->must_insert_reserved) {
+		head->extent_op = NULL;
 		btrfs_free_delayed_extent_op(extent_op);
-		return 0;
+		return NULL;
 	}
+	return extent_op;
+}
+
+static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
+				     struct btrfs_delayed_ref_head *head)
+{
+	struct btrfs_delayed_extent_op *extent_op =
+		cleanup_extent_op(trans, head);
+	int ret;
+
+	if (!extent_op)
+		return 0;
+	head->extent_op = NULL;
 	spin_unlock(&head->lock);
 	ret = run_delayed_extent_op(trans, head, extent_op);
 	btrfs_free_delayed_extent_op(extent_op);
@@ -2506,7 +2520,7 @@ static int cleanup_ref_head(struct btrfs_trans_handle *trans,
 
 	delayed_refs = &trans->transaction->delayed_refs;
 
-	ret = cleanup_extent_op(trans, head);
+	ret = run_and_cleanup_extent_op(trans, head);
 	if (ret < 0) {
 		unselect_delayed_ref_head(delayed_refs, head);
 		btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
@@ -6977,12 +6991,8 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	if (!RB_EMPTY_ROOT(&head->ref_tree))
 		goto out;
 
-	if (head->extent_op) {
-		if (!head->must_insert_reserved)
-			goto out;
-		btrfs_free_delayed_extent_op(head->extent_op);
-		head->extent_op = NULL;
-	}
+	if (cleanup_extent_op(trans, head) != NULL)
+		goto out;
 
 	/*
 	 * waiting for the lock here would deadlock.  If someone else has it
-- 
2.14.3

[PATCH 04/42] btrfs: only track ref_heads in delayed_ref_updates

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

We use this number to figure out how many delayed refs to run, but
__btrfs_run_delayed_refs really only checks every time we need a new
delayed ref head, so we always run at least one ref head completely no
matter what the number of items on it.  Fix the accounting to only be
adjusted when we add/remove a ref head.

Signed-off-by: Josef Bacik <jbacik@fb.com>
---
 fs/btrfs/delayed-ref.c | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index 3a9e4ac21794..27f7dd4e3d52 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -234,8 +234,6 @@ static inline void drop_delayed_ref(struct btrfs_trans_handle *trans,
 	ref->in_tree = 0;
 	btrfs_put_delayed_ref(ref);
 	atomic_dec(&delayed_refs->num_entries);
-	if (trans->delayed_ref_updates)
-		trans->delayed_ref_updates--;
 }
 
 static bool merge_ref(struct btrfs_trans_handle *trans,
@@ -460,7 +458,6 @@ static int insert_delayed_ref(struct btrfs_trans_handle *trans,
 	if (ref->action == BTRFS_ADD_DELAYED_REF)
 		list_add_tail(&ref->add_list, &href->ref_add_list);
 	atomic_inc(&root->num_entries);
-	trans->delayed_ref_updates++;
 	spin_unlock(&href->lock);
 	return ret;
 }
-- 
2.14.3

[PATCH 05/42] btrfs: only count ref heads run in __btrfs_run_delayed_refs

[Josef Bacik]

We pick the number of ref's to run based on the number of ref heads, and
only make the decision to stop once we've processed entire ref heads, so
only count the ref heads we've run and bail once we've hit the number of
ref heads we wanted to process.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 98f36dfeccb0..b32bd38390dd 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2592,6 +2592,7 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 				spin_unlock(&delayed_refs->lock);
 				break;
 			}
+			count++;
 
 			/* grab the lock that says we are going to process
 			 * all the refs for this head */
@@ -2605,7 +2606,6 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 			 */
 			if (ret == -EAGAIN) {
 				locked_ref = NULL;
-				count++;
 				continue;
 			}
 		}
@@ -2633,7 +2633,6 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 			unselect_delayed_ref_head(delayed_refs, locked_ref);
 			locked_ref = NULL;
 			cond_resched();
-			count++;
 			continue;
 		}
 
@@ -2651,7 +2650,6 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 				return ret;
 			}
 			locked_ref = NULL;
-			count++;
 			continue;
 		}
 
@@ -2702,7 +2700,6 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 		}
 
 		btrfs_put_delayed_ref(ref);
-		count++;
 		cond_resched();
 	}
 
-- 
2.14.3

[PATCH 06/42] btrfs: introduce delayed_refs_rsv

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

Traditionally we've had voodoo in btrfs to account for the space that
delayed refs may take up by having a global_block_rsv.  This works most
of the time, except when it doesn't.  We've had issues reported and seen
in production where sometimes the global reserve is exhausted during
transaction commit before we can run all of our delayed refs, resulting
in an aborted transaction.  Because of this voodoo we have equally
dubious flushing semantics around throttling delayed refs which we often
get wrong.

So instead give them their own block_rsv.  This way we can always know
exactly how much outstanding space we need for delayed refs.  This
allows us to make sure we are constantly filling that reservation up
with space, and allows us to put more precise pressure on the enospc
system.  Instead of doing math to see if its a good time to throttle,
the normal enospc code will be invoked if we have a lot of delayed refs
pending, and they will be run via the normal flushing mechanism.

For now the delayed_refs_rsv will hold the reservations for the delayed
refs, the block group updates, and deleting csums.  We could have a
separate rsv for the block group updates, but the csum deletion stuff is
still handled via the delayed_refs so that will stay there.

Signed-off-by: Josef Bacik <jbacik@fb.com>
---
 fs/btrfs/ctree.h             |  27 +++--
 fs/btrfs/delayed-ref.c       |  28 ++++-
 fs/btrfs/disk-io.c           |   4 +
 fs/btrfs/extent-tree.c       | 279 +++++++++++++++++++++++++++++++++++--------
 fs/btrfs/inode.c             |   2 +-
 fs/btrfs/transaction.c       |  77 ++++++------
 include/trace/events/btrfs.h |   2 +
 7 files changed, 312 insertions(+), 107 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 66f1d3895bca..1a2c3b629af2 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -452,8 +452,9 @@ struct btrfs_space_info {
 #define	BTRFS_BLOCK_RSV_TRANS		3
 #define	BTRFS_BLOCK_RSV_CHUNK		4
 #define	BTRFS_BLOCK_RSV_DELOPS		5
-#define	BTRFS_BLOCK_RSV_EMPTY		6
-#define	BTRFS_BLOCK_RSV_TEMP		7
+#define BTRFS_BLOCK_RSV_DELREFS		6
+#define	BTRFS_BLOCK_RSV_EMPTY		7
+#define	BTRFS_BLOCK_RSV_TEMP		8
 
 struct btrfs_block_rsv {
 	u64 size;
@@ -794,6 +795,8 @@ struct btrfs_fs_info {
 	struct btrfs_block_rsv chunk_block_rsv;
 	/* block reservation for delayed operations */
 	struct btrfs_block_rsv delayed_block_rsv;
+	/* block reservation for delayed refs */
+	struct btrfs_block_rsv delayed_refs_rsv;
 
 	struct btrfs_block_rsv empty_block_rsv;
 
@@ -2608,8 +2611,7 @@ static inline u64 btrfs_calc_trunc_metadata_size(struct btrfs_fs_info *fs_info,
 
 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
 				       struct btrfs_fs_info *fs_info);
-int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
-				       struct btrfs_fs_info *fs_info);
+bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
 					 const u64 start);
 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg);
@@ -2723,10 +2725,12 @@ enum btrfs_reserve_flush_enum {
 enum btrfs_flush_state {
 	FLUSH_DELAYED_ITEMS_NR	=	1,
 	FLUSH_DELAYED_ITEMS	=	2,
-	FLUSH_DELALLOC		=	3,
-	FLUSH_DELALLOC_WAIT	=	4,
-	ALLOC_CHUNK		=	5,
-	COMMIT_TRANS		=	6,
+	FLUSH_DELAYED_REFS_NR	=	3,
+	FLUSH_DELAYED_REFS	=	4,
+	FLUSH_DELALLOC		=	5,
+	FLUSH_DELALLOC_WAIT	=	6,
+	ALLOC_CHUNK		=	7,
+	COMMIT_TRANS		=	8,
 };
 
 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes);
@@ -2777,6 +2781,13 @@ int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
 			     struct btrfs_block_rsv *block_rsv,
 			     u64 num_bytes);
+void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr);
+void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
+int btrfs_throttle_delayed_refs(struct btrfs_fs_info *fs_info,
+				enum btrfs_reserve_flush_enum flush);
+void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
+				       struct btrfs_block_rsv *src,
+				       u64 num_bytes);
 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache);
 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache);
 void btrfs_put_block_group_cache(struct btrfs_fs_info *info);
diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index 27f7dd4e3d52..96ce087747b2 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -467,11 +467,14 @@ static int insert_delayed_ref(struct btrfs_trans_handle *trans,
  * existing and update must have the same bytenr
  */
 static noinline void
-update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs,
+update_existing_head_ref(struct btrfs_trans_handle *trans,
 			 struct btrfs_delayed_ref_head *existing,
 			 struct btrfs_delayed_ref_head *update,
 			 int *old_ref_mod_ret)
 {
+	struct btrfs_delayed_ref_root *delayed_refs =
+		&trans->transaction->delayed_refs;
+	struct btrfs_fs_info *fs_info = trans->fs_info;
 	int old_ref_mod;
 
 	BUG_ON(existing->is_data != update->is_data);
@@ -529,10 +532,18 @@ update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs,
 	 * versa we need to make sure to adjust pending_csums accordingly.
 	 */
 	if (existing->is_data) {
-		if (existing->total_ref_mod >= 0 && old_ref_mod < 0)
+		u64 csum_items =
+			btrfs_csum_bytes_to_leaves(fs_info,
+						   existing->num_bytes);
+
+		if (existing->total_ref_mod >= 0 && old_ref_mod < 0) {
 			delayed_refs->pending_csums -= existing->num_bytes;
-		if (existing->total_ref_mod < 0 && old_ref_mod >= 0)
+			btrfs_delayed_refs_rsv_release(fs_info, csum_items);
+		}
+		if (existing->total_ref_mod < 0 && old_ref_mod >= 0) {
 			delayed_refs->pending_csums += existing->num_bytes;
+			trans->delayed_ref_updates += csum_items;
+		}
 	}
 	spin_unlock(&existing->lock);
 }
@@ -638,7 +649,7 @@ add_delayed_ref_head(struct btrfs_trans_handle *trans,
 			&& head_ref->qgroup_reserved
 			&& existing->qgroup_ref_root
 			&& existing->qgroup_reserved);
-		update_existing_head_ref(delayed_refs, existing, head_ref,
+		update_existing_head_ref(trans, existing, head_ref,
 					 old_ref_mod);
 		/*
 		 * we've updated the existing ref, free the newly
@@ -649,8 +660,12 @@ add_delayed_ref_head(struct btrfs_trans_handle *trans,
 	} else {
 		if (old_ref_mod)
 			*old_ref_mod = 0;
-		if (head_ref->is_data && head_ref->ref_mod < 0)
+		if (head_ref->is_data && head_ref->ref_mod < 0) {
 			delayed_refs->pending_csums += head_ref->num_bytes;
+			trans->delayed_ref_updates +=
+				btrfs_csum_bytes_to_leaves(trans->fs_info,
+							   head_ref->num_bytes);
+		}
 		delayed_refs->num_heads++;
 		delayed_refs->num_heads_ready++;
 		atomic_inc(&delayed_refs->num_entries);
@@ -785,6 +800,7 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 
 	ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node);
 	spin_unlock(&delayed_refs->lock);
+	btrfs_update_delayed_refs_rsv(trans);
 
 	trace_add_delayed_tree_ref(fs_info, &ref->node, ref,
 				   action == BTRFS_ADD_DELAYED_EXTENT ?
@@ -866,6 +882,7 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 
 	ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node);
 	spin_unlock(&delayed_refs->lock);
+	btrfs_update_delayed_refs_rsv(trans);
 
 	trace_add_delayed_data_ref(trans->fs_info, &ref->node, ref,
 				   action == BTRFS_ADD_DELAYED_EXTENT ?
@@ -903,6 +920,7 @@ int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
 			     NULL, NULL, NULL);
 
 	spin_unlock(&delayed_refs->lock);
+	btrfs_update_delayed_refs_rsv(trans);
 	return 0;
 }
 
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 5124c15705ce..377ad9c1cb17 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -2692,6 +2692,9 @@ int open_ctree(struct super_block *sb,
 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
 			     BTRFS_BLOCK_RSV_DELOPS);
+	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
+			     BTRFS_BLOCK_RSV_DELREFS);
+
 	atomic_set(&fs_info->async_delalloc_pages, 0);
 	atomic_set(&fs_info->defrag_running, 0);
 	atomic_set(&fs_info->qgroup_op_seq, 0);
@@ -4419,6 +4422,7 @@ void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
 
 		spin_unlock(&cur_trans->dirty_bgs_lock);
 		btrfs_put_block_group(cache);
+		btrfs_delayed_refs_rsv_release(fs_info, 1);
 		spin_lock(&cur_trans->dirty_bgs_lock);
 	}
 	spin_unlock(&cur_trans->dirty_bgs_lock);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index b32bd38390dd..1213f573eea2 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2481,6 +2481,7 @@ static void cleanup_ref_head_accounting(struct btrfs_trans_handle *trans,
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_delayed_ref_root *delayed_refs =
 		&trans->transaction->delayed_refs;
+	int nr_items = 1;
 
 	if (head->total_ref_mod < 0) {
 		struct btrfs_space_info *space_info;
@@ -2502,12 +2503,15 @@ static void cleanup_ref_head_accounting(struct btrfs_trans_handle *trans,
 			spin_lock(&delayed_refs->lock);
 			delayed_refs->pending_csums -= head->num_bytes;
 			spin_unlock(&delayed_refs->lock);
+			nr_items += btrfs_csum_bytes_to_leaves(fs_info,
+				head->num_bytes);
 		}
 	}
 
 	/* Also free its reserved qgroup space */
 	btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
 				      head->qgroup_reserved);
+	btrfs_delayed_refs_rsv_release(fs_info, nr_items);
 }
 
 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
@@ -2802,40 +2806,22 @@ u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
 	return num_csums;
 }
 
-int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
-				       struct btrfs_fs_info *fs_info)
+bool btrfs_check_space_for_delayed_refs( struct btrfs_fs_info *fs_info)
 {
-	struct btrfs_block_rsv *global_rsv;
-	u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
-	u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
-	unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
-	u64 num_bytes, num_dirty_bgs_bytes;
-	int ret = 0;
-
-	num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
-	num_heads = heads_to_leaves(fs_info, num_heads);
-	if (num_heads > 1)
-		num_bytes += (num_heads - 1) * fs_info->nodesize;
-	num_bytes <<= 1;
-	num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
-							fs_info->nodesize;
-	num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
-							     num_dirty_bgs);
-	global_rsv = &fs_info->global_block_rsv;
-
-	/*
-	 * If we can't allocate any more chunks lets make sure we have _lots_ of
-	 * wiggle room since running delayed refs can create more delayed refs.
-	 */
-	if (global_rsv->space_info->full) {
-		num_dirty_bgs_bytes <<= 1;
-		num_bytes <<= 1;
-	}
+	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
+	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
+	u64 reserved;
+	bool ret = false;
 
 	spin_lock(&global_rsv->lock);
-	if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
-		ret = 1;
+	reserved = global_rsv->reserved;
 	spin_unlock(&global_rsv->lock);
+
+	spin_lock(&delayed_refs_rsv->lock);
+	reserved += delayed_refs_rsv->reserved;
+	if (delayed_refs_rsv->size >= reserved)
+		ret = true;
+	spin_unlock(&delayed_refs_rsv->lock);
 	return ret;
 }
 
@@ -2855,7 +2841,7 @@ int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
 	if (val >= NSEC_PER_SEC / 2)
 		return 2;
 
-	return btrfs_check_space_for_delayed_refs(trans, fs_info);
+	return btrfs_check_space_for_delayed_refs(fs_info) ? 1 : 0;
 }
 
 struct async_delayed_refs {
@@ -3610,6 +3596,8 @@ int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
 	 */
 	mutex_lock(&trans->transaction->cache_write_mutex);
 	while (!list_empty(&dirty)) {
+		bool drop_reserve = true;
+
 		cache = list_first_entry(&dirty,
 					 struct btrfs_block_group_cache,
 					 dirty_list);
@@ -3682,6 +3670,7 @@ int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
 					list_add_tail(&cache->dirty_list,
 						      &cur_trans->dirty_bgs);
 					btrfs_get_block_group(cache);
+					drop_reserve = false;
 				}
 				spin_unlock(&cur_trans->dirty_bgs_lock);
 			} else if (ret) {
@@ -3692,6 +3681,8 @@ int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
 		/* if its not on the io list, we need to put the block group */
 		if (should_put)
 			btrfs_put_block_group(cache);
+		if (drop_reserve)
+			btrfs_delayed_refs_rsv_release(fs_info, 1);
 
 		if (ret)
 			break;
@@ -3840,6 +3831,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		/* if its not on the io list, we need to put the block group */
 		if (should_put)
 			btrfs_put_block_group(cache);
+		btrfs_delayed_refs_rsv_release(fs_info, 1);
 		spin_lock(&cur_trans->dirty_bgs_lock);
 	}
 	spin_unlock(&cur_trans->dirty_bgs_lock);
@@ -4816,8 +4808,10 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 {
 	struct reserve_ticket *ticket = NULL;
 	struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
+	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
 	struct btrfs_trans_handle *trans;
 	u64 bytes;
+	u64 reclaim_bytes = 0;
 
 	trans = (struct btrfs_trans_handle *)current->journal_info;
 	if (trans)
@@ -4850,12 +4844,16 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 		return -ENOSPC;
 
 	spin_lock(&delayed_rsv->lock);
-	if (delayed_rsv->size > bytes)
-		bytes = 0;
-	else
-		bytes -= delayed_rsv->size;
+	reclaim_bytes += delayed_rsv->reserved;
 	spin_unlock(&delayed_rsv->lock);
 
+	spin_lock(&delayed_refs_rsv->lock);
+	reclaim_bytes += delayed_refs_rsv->reserved;
+	spin_unlock(&delayed_refs_rsv->lock);
+	if (reclaim_bytes >= bytes)
+		goto commit;
+	bytes -= reclaim_bytes;
+
 	if (__percpu_counter_compare(&space_info->total_bytes_pinned,
 				   bytes,
 				   BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) {
@@ -4905,6 +4903,20 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 		shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
 				state == FLUSH_DELALLOC_WAIT);
 		break;
+	case FLUSH_DELAYED_REFS_NR:
+	case FLUSH_DELAYED_REFS:
+		trans = btrfs_join_transaction(root);
+		if (IS_ERR(trans)) {
+			ret = PTR_ERR(trans);
+			break;
+		}
+		if (state == FLUSH_DELAYED_REFS_NR)
+			nr = calc_reclaim_items_nr(fs_info, num_bytes);
+		else
+			nr = 0;
+		btrfs_run_delayed_refs(trans, nr);
+		btrfs_end_transaction(trans);
+		break;
 	case ALLOC_CHUNK:
 		trans = btrfs_join_transaction(root);
 		if (IS_ERR(trans)) {
@@ -5377,6 +5389,91 @@ int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
 	return 0;
 }
 
+/**
+ * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv.
+ * @fs_info - the fs info for our fs.
+ * @src - the source block rsv to transfer from.
+ * @num_bytes - the number of bytes to transfer.
+ *
+ * This transfers up to the num_bytes amount from the src rsv to the
+ * delayed_refs_rsv.  Any extra bytes are returned to the space info.
+ */
+void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
+				       struct btrfs_block_rsv *src,
+				       u64 num_bytes)
+{
+	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
+	u64 to_free = 0;
+
+	spin_lock(&src->lock);
+	src->reserved -= num_bytes;
+	src->size -= num_bytes;
+	spin_unlock(&src->lock);
+
+	spin_lock(&delayed_refs_rsv->lock);
+	if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) {
+		u64 delta = delayed_refs_rsv->size -
+			delayed_refs_rsv->reserved;
+		if (num_bytes > delta) {
+			to_free = num_bytes - delta;
+			num_bytes = delta;
+		}
+	} else {
+		to_free = num_bytes;
+		num_bytes = 0;
+	}
+
+	if (num_bytes)
+		delayed_refs_rsv->reserved += num_bytes;
+	if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size)
+		delayed_refs_rsv->full = 1;
+	spin_unlock(&delayed_refs_rsv->lock);
+
+	if (num_bytes)
+		trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
+					      0, num_bytes, 1);
+	if (to_free)
+		space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info,
+					 to_free);
+}
+
+/**
+ * btrfs_throttle_delayed_refs - throttle based on our delayed refs usage.
+ * @fs_info - the fs_info for our fs.
+ * @flush - control how we can flush for this reservation.
+ *
+ * This will refill the delayed block_rsv up to 1 items size worth of space and
+ * will return -ENOSPC if we can't make the reservation.
+ */
+int btrfs_throttle_delayed_refs(struct btrfs_fs_info *fs_info,
+				enum btrfs_reserve_flush_enum flush)
+{
+	struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
+	u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1);
+	u64 num_bytes = 0;
+	int ret = -ENOSPC;
+
+	spin_lock(&block_rsv->lock);
+	if (block_rsv->reserved < block_rsv->size) {
+		num_bytes = block_rsv->size - block_rsv->reserved;
+		num_bytes = min(num_bytes, limit);
+	}
+	spin_unlock(&block_rsv->lock);
+
+	if (!num_bytes)
+		return 0;
+
+	ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv,
+				     num_bytes, flush);
+	if (ret)
+		return ret;
+	block_rsv_add_bytes(block_rsv, num_bytes, 0);
+	trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
+				      0, num_bytes, 1);
+	return 0;
+}
+
+
 /*
  * This is for space we already have accounted in space_info->bytes_may_use, so
  * basically when we're returning space from block_rsv's.
@@ -5699,6 +5796,31 @@ static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
 	return ret;
 }
 
+static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
+				     struct btrfs_block_rsv *block_rsv,
+				     u64 num_bytes, u64 *qgroup_to_release)
+{
+	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
+	struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
+	struct btrfs_block_rsv *target = delayed_rsv;
+
+	if (target->full || target == block_rsv)
+		target = global_rsv;
+
+	if (block_rsv->space_info != target->space_info)
+		target = NULL;
+
+	return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
+				       qgroup_to_release);
+}
+
+void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
+			     struct btrfs_block_rsv *block_rsv,
+			     u64 num_bytes)
+{
+	__btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL);
+}
+
 /**
  * btrfs_inode_rsv_release - release any excessive reservation.
  * @inode - the inode we need to release from.
@@ -5713,7 +5835,6 @@ static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
 {
 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
-	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
 	struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
 	u64 released = 0;
 	u64 qgroup_to_release = 0;
@@ -5723,8 +5844,8 @@ static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
 	 * are releasing 0 bytes, and then we'll just get the reservation over
 	 * the size free'd.
 	 */
-	released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
-					   &qgroup_to_release);
+	released = __btrfs_block_rsv_release(fs_info, block_rsv, 0,
+					     &qgroup_to_release);
 	if (released > 0)
 		trace_btrfs_space_reservation(fs_info, "delalloc",
 					      btrfs_ino(inode), released, 0);
@@ -5735,16 +5856,26 @@ static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
 						   qgroup_to_release);
 }
 
-void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
-			     struct btrfs_block_rsv *block_rsv,
-			     u64 num_bytes)
+/**
+ * btrfs_delayed_refs_rsv_release - release a ref head's reservation.
+ * @fs_info - the fs_info for our fs.
+ * @nr - the number of items to drop.
+ *
+ * This drops the delayed ref head's count from the delayed refs rsv and free's
+ * any excess reservation we had.
+ */
+void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr)
 {
+	struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
+	u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr);
+	u64 released = 0;
 
-	if (global_rsv == block_rsv ||
-	    block_rsv->space_info != global_rsv->space_info)
-		global_rsv = NULL;
-	block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
+	released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv,
+					   num_bytes, NULL);
+	if (released)
+		trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
+					      0, released, 0);
 }
 
 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
@@ -5809,9 +5940,10 @@ static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
 	fs_info->trans_block_rsv.space_info = space_info;
 	fs_info->empty_block_rsv.space_info = space_info;
 	fs_info->delayed_block_rsv.space_info = space_info;
+	fs_info->delayed_refs_rsv.space_info = space_info;
 
-	fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
-	fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
+	fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
+	fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
 	fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
 	fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
 	if (fs_info->quota_root)
@@ -5831,8 +5963,34 @@ static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
 	WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
 	WARN_ON(fs_info->delayed_block_rsv.size > 0);
 	WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
+	WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
+	WARN_ON(fs_info->delayed_refs_rsv.size > 0);
 }
 
+/*
+ * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv
+ * @trans - the trans that may have generated delayed refs
+ *
+ * This is to be called anytime we may have adjusted trans->delayed_ref_updates,
+ * it'll calculate the additional size and add it to the delayed_refs_rsv.
+ */
+void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
+{
+	struct btrfs_fs_info *fs_info = trans->fs_info;
+	struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
+	u64 num_bytes;
+
+	if (!trans->delayed_ref_updates)
+		return;
+
+	num_bytes = btrfs_calc_trans_metadata_size(fs_info,
+						   trans->delayed_ref_updates);
+	spin_lock(&delayed_rsv->lock);
+	delayed_rsv->size += num_bytes;
+	delayed_rsv->full = 0;
+	spin_unlock(&delayed_rsv->lock);
+	trans->delayed_ref_updates = 0;
+}
 
 /*
  * To be called after all the new block groups attached to the transaction
@@ -6126,6 +6284,7 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 	u64 old_val;
 	u64 byte_in_group;
 	int factor;
+	int ret = 0;
 
 	/* block accounting for super block */
 	spin_lock(&info->delalloc_root_lock);
@@ -6139,8 +6298,10 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 
 	while (total) {
 		cache = btrfs_lookup_block_group(info, bytenr);
-		if (!cache)
-			return -ENOENT;
+		if (!cache) {
+			ret = -ENOENT;
+			break;
+		}
 		factor = btrfs_bg_type_to_factor(cache->flags);
 
 		/*
@@ -6199,6 +6360,7 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 			list_add_tail(&cache->dirty_list,
 				      &trans->transaction->dirty_bgs);
 			trans->transaction->num_dirty_bgs++;
+			trans->delayed_ref_updates++;
 			btrfs_get_block_group(cache);
 		}
 		spin_unlock(&trans->transaction->dirty_bgs_lock);
@@ -6216,7 +6378,10 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 		total -= num_bytes;
 		bytenr += num_bytes;
 	}
-	return 0;
+
+	/* Modified block groups are accounted for in the delayed_refs_rsv. */
+	btrfs_update_delayed_refs_rsv(trans);
+	return ret;
 }
 
 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
@@ -8230,7 +8395,12 @@ use_block_rsv(struct btrfs_trans_handle *trans,
 		goto again;
 	}
 
-	if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
+	/*
+	 * The global reserve still exists to save us from ourselves, so don't
+	 * warn_on if we are short on our delayed refs reserve.
+	 */
+	if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
+	    btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
 		static DEFINE_RATELIMIT_STATE(_rs,
 				DEFAULT_RATELIMIT_INTERVAL * 10,
 				/*DEFAULT_RATELIMIT_BURST*/ 1);
@@ -10146,6 +10316,7 @@ void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
 		add_block_group_free_space(trans, block_group);
 		/* already aborted the transaction if it failed. */
 next:
+		btrfs_delayed_refs_rsv_release(fs_info, 1);
 		list_del_init(&block_group->bg_list);
 	}
 	trans->can_flush_pending_bgs = can_flush_pending_bgs;
@@ -10223,6 +10394,8 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
 	link_block_group(cache);
 
 	list_add_tail(&cache->bg_list, &trans->new_bgs);
+	trans->delayed_ref_updates++;
+	btrfs_update_delayed_refs_rsv(trans);
 
 	set_avail_alloc_bits(fs_info, type);
 	return 0;
@@ -10260,6 +10433,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
 	int factor;
 	struct btrfs_caching_control *caching_ctl = NULL;
 	bool remove_em;
+	bool remove_rsv = false;
 
 	block_group = btrfs_lookup_block_group(fs_info, group_start);
 	BUG_ON(!block_group);
@@ -10324,6 +10498,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
 
 	if (!list_empty(&block_group->dirty_list)) {
 		list_del_init(&block_group->dirty_list);
+		remove_rsv = true;
 		btrfs_put_block_group(block_group);
 	}
 	spin_unlock(&trans->transaction->dirty_bgs_lock);
@@ -10533,6 +10708,8 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
 
 	ret = btrfs_del_item(trans, root, path);
 out:
+	if (remove_rsv)
+		btrfs_delayed_refs_rsv_release(fs_info, 1);
 	btrfs_free_path(path);
 	return ret;
 }
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 212fa71317d6..cd00ec869c96 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -5382,7 +5382,7 @@ static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
 		 * Try to steal from the global reserve if there is space for
 		 * it.
 		 */
-		if (!btrfs_check_space_for_delayed_refs(trans, fs_info) &&
+		if (!btrfs_check_space_for_delayed_refs(fs_info) &&
 		    !btrfs_block_rsv_migrate(global_rsv, rsv, min_size, 0))
 			return trans;
 
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 3b84f5015029..117e0c4a914a 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -455,7 +455,7 @@ start_transaction(struct btrfs_root *root, unsigned int num_items,
 		  bool enforce_qgroups)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
-
+	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
 	struct btrfs_trans_handle *h;
 	struct btrfs_transaction *cur_trans;
 	u64 num_bytes = 0;
@@ -484,13 +484,28 @@ start_transaction(struct btrfs_root *root, unsigned int num_items,
 	 * the appropriate flushing if need be.
 	 */
 	if (num_items && root != fs_info->chunk_root) {
+		struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
+		u64 delayed_refs_bytes = 0;
+
 		qgroup_reserved = num_items * fs_info->nodesize;
 		ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
 				enforce_qgroups);
 		if (ret)
 			return ERR_PTR(ret);
 
+		/*
+		 * We want to reserve all the bytes we may need all at once, so
+		 * we only do 1 enospc flushing cycle per transaction start.  We
+		 * accomplish this by simply assuming we'll do 2 x num_items
+		 * worth of delayed refs updates in this trans handle, and
+		 * refill that amount for whatever is missing in the reserve.
+		 */
 		num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items);
+		if (delayed_refs_rsv->full == 0) {
+			delayed_refs_bytes = num_bytes;
+			num_bytes <<= 1;
+		}
+
 		/*
 		 * Do the reservation for the relocation root creation
 		 */
@@ -499,8 +514,24 @@ start_transaction(struct btrfs_root *root, unsigned int num_items,
 			reloc_reserved = true;
 		}
 
-		ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
-					  num_bytes, flush);
+		ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
+		if (ret)
+			goto reserve_fail;
+		if (delayed_refs_bytes) {
+			btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
+							  delayed_refs_bytes);
+			num_bytes -= delayed_refs_bytes;
+		}
+	} else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
+		   !delayed_refs_rsv->full) {
+		/*
+		 * Some people call with btrfs_start_transaction(root, 0)
+		 * because they can be throttled, but have some other mechanism
+		 * for reserving space.  We still want these guys to refill the
+		 * delayed block_rsv so just add 1 items worth of reservation
+		 * here.
+		 */
+		ret = btrfs_throttle_delayed_refs(fs_info, flush);
 		if (ret)
 			goto reserve_fail;
 	}
@@ -759,7 +790,7 @@ static int should_end_transaction(struct btrfs_trans_handle *trans)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 
-	if (btrfs_check_space_for_delayed_refs(trans, fs_info))
+	if (btrfs_check_space_for_delayed_refs(fs_info))
 		return 1;
 
 	return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5);
@@ -768,22 +799,12 @@ static int should_end_transaction(struct btrfs_trans_handle *trans)
 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans)
 {
 	struct btrfs_transaction *cur_trans = trans->transaction;
-	int updates;
-	int err;
 
 	smp_mb();
 	if (cur_trans->state >= TRANS_STATE_BLOCKED ||
 	    cur_trans->delayed_refs.flushing)
 		return 1;
 
-	updates = trans->delayed_ref_updates;
-	trans->delayed_ref_updates = 0;
-	if (updates) {
-		err = btrfs_run_delayed_refs(trans, updates * 2);
-		if (err) /* Error code will also eval true */
-			return err;
-	}
-
 	return should_end_transaction(trans);
 }
 
@@ -813,11 +834,8 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 {
 	struct btrfs_fs_info *info = trans->fs_info;
 	struct btrfs_transaction *cur_trans = trans->transaction;
-	u64 transid = trans->transid;
-	unsigned long cur = trans->delayed_ref_updates;
 	int lock = (trans->type != TRANS_JOIN_NOLOCK);
 	int err = 0;
-	int must_run_delayed_refs = 0;
 
 	if (refcount_read(&trans->use_count) > 1) {
 		refcount_dec(&trans->use_count);
@@ -828,27 +846,6 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 	btrfs_trans_release_metadata(trans);
 	trans->block_rsv = NULL;
 
-	if (!list_empty(&trans->new_bgs))
-		btrfs_create_pending_block_groups(trans);
-
-	trans->delayed_ref_updates = 0;
-	if (!trans->sync) {
-		must_run_delayed_refs =
-			btrfs_should_throttle_delayed_refs(trans, info);
-		cur = max_t(unsigned long, cur, 32);
-
-		/*
-		 * don't make the caller wait if they are from a NOLOCK
-		 * or ATTACH transaction, it will deadlock with commit
-		 */
-		if (must_run_delayed_refs == 1 &&
-		    (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH)))
-			must_run_delayed_refs = 2;
-	}
-
-	btrfs_trans_release_metadata(trans);
-	trans->block_rsv = NULL;
-
 	if (!list_empty(&trans->new_bgs))
 		btrfs_create_pending_block_groups(trans);
 
@@ -893,10 +890,6 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 	}
 
 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
-	if (must_run_delayed_refs) {
-		btrfs_async_run_delayed_refs(info, cur, transid,
-					     must_run_delayed_refs == 1);
-	}
 	return err;
 }
 
diff --git a/include/trace/events/btrfs.h b/include/trace/events/btrfs.h
index b401c4e36394..7d205e50b09c 100644
--- a/include/trace/events/btrfs.h
+++ b/include/trace/events/btrfs.h
@@ -1048,6 +1048,8 @@ TRACE_EVENT(btrfs_trigger_flush,
 		{ FLUSH_DELAYED_ITEMS,		"FLUSH_DELAYED_ITEMS"},		\
 		{ FLUSH_DELALLOC,		"FLUSH_DELALLOC"},		\
 		{ FLUSH_DELALLOC_WAIT,		"FLUSH_DELALLOC_WAIT"},		\
+		{ FLUSH_DELAYED_REFS_NR,	"FLUSH_DELAYED_REFS_NR"},	\
+		{ FLUSH_DELAYED_REFS,		"FLUSH_ELAYED_REFS"},		\
 		{ ALLOC_CHUNK,			"ALLOC_CHUNK"},			\
 		{ COMMIT_TRANS,			"COMMIT_TRANS"})
 
-- 
2.14.3

[PATCH 07/42] btrfs: check if free bgs for commit

[Josef Bacik]

may_commit_transaction will skip committing the transaction if we don't
have enough pinned space or if we're trying to find space for a SYSTEM
chunk.  However if we have pending free block groups in this transaction
we still want to commit as we may be able to allocate a chunk to make
our reservation.  So instead of just returning ENOSPC, check if we have
free block groups pending, and if so commit the transaction to allow us
to use that free space.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
---
 fs/btrfs/extent-tree.c | 33 +++++++++++++++++++--------------
 1 file changed, 19 insertions(+), 14 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 1213f573eea2..4aab49debfc3 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -4830,10 +4830,18 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	if (!bytes)
 		return 0;
 
-	/* See if there is enough pinned space to make this reservation */
-	if (__percpu_counter_compare(&space_info->total_bytes_pinned,
-				   bytes,
-				   BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
+	trans = btrfs_join_transaction(fs_info->extent_root);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+
+	/*
+	 * See if there is enough pinned space to make this reservation, or if
+	 * we have bg's that are going to be freed, allowing us to possibly do a
+	 * chunk allocation the next loop through.
+	 */
+	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
+	    __percpu_counter_compare(&space_info->total_bytes_pinned, bytes,
+				     BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
 		goto commit;
 
 	/*
@@ -4841,7 +4849,7 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	 * this reservation.
 	 */
 	if (space_info != delayed_rsv->space_info)
-		return -ENOSPC;
+		goto enospc;
 
 	spin_lock(&delayed_rsv->lock);
 	reclaim_bytes += delayed_rsv->reserved;
@@ -4855,17 +4863,14 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	bytes -= reclaim_bytes;
 
 	if (__percpu_counter_compare(&space_info->total_bytes_pinned,
-				   bytes,
-				   BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) {
-		return -ENOSPC;
-	}
-
+				     bytes,
+				     BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0)
+		goto enospc;
 commit:
-	trans = btrfs_join_transaction(fs_info->extent_root);
-	if (IS_ERR(trans))
-		return -ENOSPC;
-
 	return btrfs_commit_transaction(trans);
+enospc:
+	btrfs_end_transaction(trans);
+	return -ENOSPC;
 }
 
 /*
-- 
2.14.3

[PATCH 08/42] btrfs: dump block_rsv whe dumping space info

[Josef Bacik]

For enospc_debug having the block rsvs is super helpful to see if we've
done something wrong.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
---
 fs/btrfs/extent-tree.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 4aab49debfc3..8b00c658deb3 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -7918,6 +7918,15 @@ static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
 	return ret;
 }
 
+#define DUMP_BLOCK_RSV(fs_info, rsv_name)				\
+do {									\
+	struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name;		\
+	spin_lock(&__rsv->lock);					\
+	btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu",	\
+		   __rsv->size, __rsv->reserved);			\
+	spin_unlock(&__rsv->lock);					\
+} while (0)
+
 static void dump_space_info(struct btrfs_fs_info *fs_info,
 			    struct btrfs_space_info *info, u64 bytes,
 			    int dump_block_groups)
@@ -7937,6 +7946,12 @@ static void dump_space_info(struct btrfs_fs_info *fs_info,
 		info->bytes_readonly);
 	spin_unlock(&info->lock);
 
+	DUMP_BLOCK_RSV(fs_info, global_block_rsv);
+	DUMP_BLOCK_RSV(fs_info, trans_block_rsv);
+	DUMP_BLOCK_RSV(fs_info, chunk_block_rsv);
+	DUMP_BLOCK_RSV(fs_info, delayed_block_rsv);
+	DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv);
+
 	if (!dump_block_groups)
 		return;
 
-- 
2.14.3

[PATCH 09/42] btrfs: release metadata before running delayed refs

[Josef Bacik]

We want to release the unused reservation we have since it refills the
delayed refs reserve, which will make everything go smoother when
running the delayed refs if we're short on our reservation.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/transaction.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 117e0c4a914a..a0f19ca0bd6c 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -1922,6 +1922,9 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 		return ret;
 	}
 
+	btrfs_trans_release_metadata(trans);
+	trans->block_rsv = NULL;
+
 	/* make a pass through all the delayed refs we have so far
 	 * any runnings procs may add more while we are here
 	 */
@@ -1931,9 +1934,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 		return ret;
 	}
 
-	btrfs_trans_release_metadata(trans);
-	trans->block_rsv = NULL;
-
 	cur_trans = trans->transaction;
 
 	/*
-- 
2.14.3

[PATCH 10/42] btrfs: protect space cache inode alloc with nofs

[Josef Bacik]

If we're allocating a new space cache inode it's likely going to be
under a transaction handle, so we need to use memalloc_nofs_save() in
order to avoid deadlocks, and more importantly lockdep messages that
make xfstests fail.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
---
 fs/btrfs/free-space-cache.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index c3888c113d81..e077ad3b4549 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -10,6 +10,7 @@
 #include <linux/math64.h>
 #include <linux/ratelimit.h>
 #include <linux/error-injection.h>
+#include <linux/sched/mm.h>
 #include "ctree.h"
 #include "free-space-cache.h"
 #include "transaction.h"
@@ -47,6 +48,7 @@ static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
 	struct btrfs_free_space_header *header;
 	struct extent_buffer *leaf;
 	struct inode *inode = NULL;
+	unsigned nofs_flag;
 	int ret;
 
 	key.objectid = BTRFS_FREE_SPACE_OBJECTID;
@@ -68,7 +70,13 @@ static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
 	btrfs_disk_key_to_cpu(&location, &disk_key);
 	btrfs_release_path(path);
 
+	/*
+	 * We are often under a trans handle at this point, so we need to make
+	 * sure NOFS is set to keep us from deadlocking.
+	 */
+	nofs_flag = memalloc_nofs_save();
 	inode = btrfs_iget(fs_info->sb, &location, root, NULL);
+	memalloc_nofs_restore(nofs_flag);
 	if (IS_ERR(inode))
 		return inode;
 
-- 
2.14.3

[PATCH 11/42] btrfs: fix truncate throttling

[Josef Bacik]

We have a bunch of magic to make sure we're throttling delayed refs when
truncating a file.  Now that we have a delayed refs rsv and a mechanism
for refilling that reserve simply use that instead of all of this magic.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/inode.c | 79 ++++++++++++--------------------------------------------
 1 file changed, 17 insertions(+), 62 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index cd00ec869c96..5a91055a13b2 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -4493,31 +4493,6 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 	return err;
 }
 
-static int truncate_space_check(struct btrfs_trans_handle *trans,
-				struct btrfs_root *root,
-				u64 bytes_deleted)
-{
-	struct btrfs_fs_info *fs_info = root->fs_info;
-	int ret;
-
-	/*
-	 * This is only used to apply pressure to the enospc system, we don't
-	 * intend to use this reservation at all.
-	 */
-	bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
-	bytes_deleted *= fs_info->nodesize;
-	ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
-				  bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
-	if (!ret) {
-		trace_btrfs_space_reservation(fs_info, "transaction",
-					      trans->transid,
-					      bytes_deleted, 1);
-		trans->bytes_reserved += bytes_deleted;
-	}
-	return ret;
-
-}
-
 /*
  * Return this if we need to call truncate_block for the last bit of the
  * truncate.
@@ -4562,7 +4537,6 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 	u64 bytes_deleted = 0;
 	bool be_nice = false;
 	bool should_throttle = false;
-	bool should_end = false;
 
 	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
 
@@ -4775,15 +4749,7 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 				btrfs_abort_transaction(trans, ret);
 				break;
 			}
-			if (btrfs_should_throttle_delayed_refs(trans, fs_info))
-				btrfs_async_run_delayed_refs(fs_info,
-					trans->delayed_ref_updates * 2,
-					trans->transid, 0);
 			if (be_nice) {
-				if (truncate_space_check(trans, root,
-							 extent_num_bytes)) {
-					should_end = true;
-				}
 				if (btrfs_should_throttle_delayed_refs(trans,
 								       fs_info))
 					should_throttle = true;
@@ -4795,7 +4761,7 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 
 		if (path->slots[0] == 0 ||
 		    path->slots[0] != pending_del_slot ||
-		    should_throttle || should_end) {
+		    should_throttle) {
 			if (pending_del_nr) {
 				ret = btrfs_del_items(trans, root, path,
 						pending_del_slot,
@@ -4807,23 +4773,24 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 				pending_del_nr = 0;
 			}
 			btrfs_release_path(path);
-			if (should_throttle) {
-				unsigned long updates = trans->delayed_ref_updates;
-				if (updates) {
-					trans->delayed_ref_updates = 0;
-					ret = btrfs_run_delayed_refs(trans,
-								   updates * 2);
-					if (ret)
-						break;
-				}
-			}
+
 			/*
-			 * if we failed to refill our space rsv, bail out
-			 * and let the transaction restart
+			 * We can generate a lot of delayed refs, so we need to
+			 * throttle every once and a while and make sure we're
+			 * adding enough space to keep up with the work we are
+			 * generating.  Since we hold a transaction here we
+			 * can't flush, and we don't want to FLUSH_LIMIT because
+			 * we could have generated too many delayed refs to
+			 * actually allocate, so just bail if we're short and
+			 * let the normal reservation dance happen higher up.
 			 */
-			if (should_end) {
-				ret = -EAGAIN;
-				break;
+			if (should_throttle) {
+				ret = btrfs_throttle_delayed_refs(fs_info,
+							BTRFS_RESERVE_NO_FLUSH);
+				if (ret) {
+					ret = -EAGAIN;
+					break;
+				}
 			}
 			goto search_again;
 		} else {
@@ -4849,18 +4816,6 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 	}
 
 	btrfs_free_path(path);
-
-	if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
-		unsigned long updates = trans->delayed_ref_updates;
-		int err;
-
-		if (updates) {
-			trans->delayed_ref_updates = 0;
-			err = btrfs_run_delayed_refs(trans, updates * 2);
-			if (err)
-				ret = err;
-		}
-	}
 	return ret;
 }
 
-- 
2.14.3

[PATCH 12/42] btrfs: don't use global rsv for chunk allocation

[Josef Bacik]

We've done this forever because of the voodoo around knowing how much
space we have.  However we have better ways of doing this now, and on
normal file systems we'll easily have a global reserve of 512MiB, and
since metadata chunks are usually 1GiB that means we'll allocate
metadata chunks more readily.  Instead use the actual used amount when
determining if we need to allocate a chunk or not.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 8b00c658deb3..828b82db439c 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -4374,21 +4374,12 @@ static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
 			      struct btrfs_space_info *sinfo, int force)
 {
-	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
 	u64 bytes_used = btrfs_space_info_used(sinfo, false);
 	u64 thresh;
 
 	if (force == CHUNK_ALLOC_FORCE)
 		return 1;
 
-	/*
-	 * We need to take into account the global rsv because for all intents
-	 * and purposes it's used space.  Don't worry about locking the
-	 * global_rsv, it doesn't change except when the transaction commits.
-	 */
-	if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
-		bytes_used += calc_global_rsv_need_space(global_rsv);
-
 	/*
 	 * in limited mode, we want to have some free space up to
 	 * about 1% of the FS size.
-- 
2.14.3

[PATCH 13/42] btrfs: add ALLOC_CHUNK_FORCE to the flushing code

[Josef Bacik]

With my change to no longer take into account the global reserve for
metadata allocation chunks we have this side-effect for mixed block
group fs'es where we are no longer allocating enough chunks for the
data/metadata requirements.  To deal with this add a ALLOC_CHUNK_FORCE
step to the flushing state machine.  This will only get used if we've
already made a full loop through the flushing machinery and tried
committing the transaction.  If we have then we can try and force a
chunk allocation since we likely need it to make progress.  This
resolves the issues I was seeing with the mixed bg tests in xfstests
with my previous patch.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/ctree.h             |  3 ++-
 fs/btrfs/extent-tree.c       | 18 +++++++++++++++++-
 include/trace/events/btrfs.h |  1 +
 3 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 1a2c3b629af2..29db902511c1 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -2730,7 +2730,8 @@ enum btrfs_flush_state {
 	FLUSH_DELALLOC		=	5,
 	FLUSH_DELALLOC_WAIT	=	6,
 	ALLOC_CHUNK		=	7,
-	COMMIT_TRANS		=	8,
+	ALLOC_CHUNK_FORCE	=	8,
+	COMMIT_TRANS		=	9,
 };
 
 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 828b82db439c..640ec640b21c 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -4914,6 +4914,7 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 		btrfs_end_transaction(trans);
 		break;
 	case ALLOC_CHUNK:
+	case ALLOC_CHUNK_FORCE:
 		trans = btrfs_join_transaction(root);
 		if (IS_ERR(trans)) {
 			ret = PTR_ERR(trans);
@@ -4921,7 +4922,9 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 		}
 		ret = do_chunk_alloc(trans,
 				     btrfs_metadata_alloc_profile(fs_info),
-				     CHUNK_ALLOC_NO_FORCE);
+				     (state == ALLOC_CHUNK) ?
+				     CHUNK_ALLOC_NO_FORCE :
+				     CHUNK_ALLOC_FORCE);
 		btrfs_end_transaction(trans);
 		if (ret > 0 || ret == -ENOSPC)
 			ret = 0;
@@ -5057,6 +5060,19 @@ static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
 				commit_cycles--;
 		}
 
+		/*
+		 * We don't want to force a chunk allocation until we've tried
+		 * pretty hard to reclaim space.  Think of the case where we
+		 * free'd up a bunch of space and so have a lot of pinned space
+		 * to reclaim.  We would rather use that than possibly create a
+		 * underutilized metadata chunk.  So if this is our first run
+		 * through the flushing state machine skip ALLOC_CHUNK_FORCE and
+		 * commit the transaction.  If nothing has changed the next go
+		 * around then we can force a chunk allocation.
+		 */
+		if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles)
+			flush_state++;
+
 		if (flush_state > COMMIT_TRANS) {
 			commit_cycles++;
 			if (commit_cycles > 2) {
diff --git a/include/trace/events/btrfs.h b/include/trace/events/btrfs.h
index 7d205e50b09c..fdb23181b5b7 100644
--- a/include/trace/events/btrfs.h
+++ b/include/trace/events/btrfs.h
@@ -1051,6 +1051,7 @@ TRACE_EVENT(btrfs_trigger_flush,
 		{ FLUSH_DELAYED_REFS_NR,	"FLUSH_DELAYED_REFS_NR"},	\
 		{ FLUSH_DELAYED_REFS,		"FLUSH_ELAYED_REFS"},		\
 		{ ALLOC_CHUNK,			"ALLOC_CHUNK"},			\
+		{ ALLOC_CHUNK_FORCE,		"ALLOC_CHUNK_FORCE"},		\
 		{ COMMIT_TRANS,			"COMMIT_TRANS"})
 
 TRACE_EVENT(btrfs_flush_space,
-- 
2.14.3

[PATCH 14/42] btrfs: reset max_extent_size properly

[Josef Bacik]

If we use up our block group before allocating a new one we'll easily
get a max_extent_size that's set really really low, which will result in
a lot of fragmentation.  We need to make sure we're resetting the
max_extent_size when we add a new chunk or add new space.

Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 640ec640b21c..2b1704331d21 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -4573,6 +4573,7 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
 			goto out;
 	} else {
 		ret = 1;
+		space_info->max_extent_size = 0;
 	}
 
 	space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
@@ -6671,6 +6672,7 @@ static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
 		space_info->bytes_readonly += num_bytes;
 	cache->reserved -= num_bytes;
 	space_info->bytes_reserved -= num_bytes;
+	space_info->max_extent_size = 0;
 
 	if (delalloc)
 		cache->delalloc_bytes -= num_bytes;
-- 
2.14.3

[PATCH 15/42] btrfs: don't enospc all tickets on flush failure

[Josef Bacik]

With the introduction of the per-inode block_rsv it became possible to
have really really large reservation requests made because of data
fragmentation.  Since the ticket stuff assumed that we'd always have
relatively small reservation requests it just killed all tickets if we
were unable to satisfy the current request.  However this is generally
not the case anymore.  So fix this logic to instead see if we had a
ticket that we were able to give some reservation to, and if we were
continue the flushing loop again.  Likewise we make the tickets use the
space_info_add_old_bytes() method of returning what reservation they did
receive in hopes that it could satisfy reservations down the line.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 45 +++++++++++++++++++++++++--------------------
 1 file changed, 25 insertions(+), 20 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 2b1704331d21..19449ee93693 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -4779,6 +4779,7 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
 }
 
 struct reserve_ticket {
+	u64 orig_bytes;
 	u64 bytes;
 	int error;
 	struct list_head list;
@@ -5000,7 +5001,7 @@ static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
 		!test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
 }
 
-static void wake_all_tickets(struct list_head *head)
+static bool wake_all_tickets(struct list_head *head)
 {
 	struct reserve_ticket *ticket;
 
@@ -5009,7 +5010,10 @@ static void wake_all_tickets(struct list_head *head)
 		list_del_init(&ticket->list);
 		ticket->error = -ENOSPC;
 		wake_up(&ticket->wait);
+		if (ticket->bytes != ticket->orig_bytes)
+			return true;
 	}
+	return false;
 }
 
 /*
@@ -5077,8 +5081,12 @@ static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
 		if (flush_state > COMMIT_TRANS) {
 			commit_cycles++;
 			if (commit_cycles > 2) {
-				wake_all_tickets(&space_info->tickets);
-				space_info->flush = 0;
+				if (wake_all_tickets(&space_info->tickets)) {
+					flush_state = FLUSH_DELAYED_ITEMS_NR;
+					commit_cycles--;
+				} else {
+					space_info->flush = 0;
+				}
 			} else {
 				flush_state = FLUSH_DELAYED_ITEMS_NR;
 			}
@@ -5130,10 +5138,11 @@ static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
 
 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
 			       struct btrfs_space_info *space_info,
-			       struct reserve_ticket *ticket, u64 orig_bytes)
+			       struct reserve_ticket *ticket)
 
 {
 	DEFINE_WAIT(wait);
+	u64 reclaim_bytes = 0;
 	int ret = 0;
 
 	spin_lock(&space_info->lock);
@@ -5154,14 +5163,12 @@ static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
 		ret = ticket->error;
 	if (!list_empty(&ticket->list))
 		list_del_init(&ticket->list);
-	if (ticket->bytes && ticket->bytes < orig_bytes) {
-		u64 num_bytes = orig_bytes - ticket->bytes;
-		space_info->bytes_may_use -= num_bytes;
-		trace_btrfs_space_reservation(fs_info, "space_info",
-					      space_info->flags, num_bytes, 0);
-	}
+	if (ticket->bytes && ticket->bytes < ticket->orig_bytes)
+		reclaim_bytes = ticket->orig_bytes - ticket->bytes;
 	spin_unlock(&space_info->lock);
 
+	if (reclaim_bytes)
+		space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
 	return ret;
 }
 
@@ -5187,6 +5194,7 @@ static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
 {
 	struct reserve_ticket ticket;
 	u64 used;
+	u64 reclaim_bytes = 0;
 	int ret = 0;
 
 	ASSERT(orig_bytes);
@@ -5222,6 +5230,7 @@ static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
 	 * the list and we will do our own flushing further down.
 	 */
 	if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
+		ticket.orig_bytes = orig_bytes;
 		ticket.bytes = orig_bytes;
 		ticket.error = 0;
 		init_waitqueue_head(&ticket.wait);
@@ -5262,25 +5271,21 @@ static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
 		return ret;
 
 	if (flush == BTRFS_RESERVE_FLUSH_ALL)
-		return wait_reserve_ticket(fs_info, space_info, &ticket,
-					   orig_bytes);
+		return wait_reserve_ticket(fs_info, space_info, &ticket);
 
 	ret = 0;
 	priority_reclaim_metadata_space(fs_info, space_info, &ticket);
 	spin_lock(&space_info->lock);
 	if (ticket.bytes) {
-		if (ticket.bytes < orig_bytes) {
-			u64 num_bytes = orig_bytes - ticket.bytes;
-			space_info->bytes_may_use -= num_bytes;
-			trace_btrfs_space_reservation(fs_info, "space_info",
-						      space_info->flags,
-						      num_bytes, 0);
-
-		}
+		if (ticket.bytes < orig_bytes)
+			reclaim_bytes = orig_bytes - ticket.bytes;
 		list_del_init(&ticket.list);
 		ret = -ENOSPC;
 	}
 	spin_unlock(&space_info->lock);
+
+	if (reclaim_bytes)
+		space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
 	ASSERT(list_empty(&ticket.list));
 	return ret;
 }
-- 
2.14.3

[PATCH 16/42] btrfs: loop in inode_rsv_refill

[Josef Bacik]

With severe fragmentation we can end up with our inode rsv size being
huge during writeout, which would cause us to need to make very large
metadata reservations.  However we may not actually need that much once
writeout is complete.  So instead try to make our reservation, and if we
couldn't make it re-calculate our new reservation size and try again.
If our reservation size doesn't change between tries then we know we are
actually out of space and can error out.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 56 ++++++++++++++++++++++++++++++++++++--------------
 1 file changed, 41 insertions(+), 15 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 19449ee93693..3757ca42b8e0 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -5766,6 +5766,21 @@ int btrfs_block_rsv_refill(struct btrfs_root *root,
 	return ret;
 }
 
+static inline void __get_refill_bytes(struct btrfs_block_rsv *block_rsv,
+				      u64 *metadata_bytes, u64 *qgroup_bytes)
+{
+	*metadata_bytes = 0;
+	*qgroup_bytes = 0;
+
+	spin_lock(&block_rsv->lock);
+	if (block_rsv->reserved < block_rsv->size)
+		*metadata_bytes = block_rsv->size - block_rsv->reserved;
+	if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
+		*qgroup_bytes = block_rsv->qgroup_rsv_size -
+			block_rsv->qgroup_rsv_reserved;
+	spin_unlock(&block_rsv->lock);
+}
+
 /**
  * btrfs_inode_rsv_refill - refill the inode block rsv.
  * @inode - the inode we are refilling.
@@ -5781,25 +5796,37 @@ static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
 {
 	struct btrfs_root *root = inode->root;
 	struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
-	u64 num_bytes = 0;
+	u64 num_bytes = 0, last = 0;
 	u64 qgroup_num_bytes = 0;
 	int ret = -ENOSPC;
 
-	spin_lock(&block_rsv->lock);
-	if (block_rsv->reserved < block_rsv->size)
-		num_bytes = block_rsv->size - block_rsv->reserved;
-	if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
-		qgroup_num_bytes = block_rsv->qgroup_rsv_size -
-				   block_rsv->qgroup_rsv_reserved;
-	spin_unlock(&block_rsv->lock);
-
+	__get_refill_bytes(block_rsv, &num_bytes, &qgroup_num_bytes);
 	if (num_bytes == 0)
 		return 0;
 
-	ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
-	if (ret)
-		return ret;
-	ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
+	do {
+		ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
+		if (ret)
+			return ret;
+		ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
+		if (ret) {
+			btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
+			last = num_bytes;
+			/*
+			 * If we are fragmented we can end up with a lot of
+			 * outstanding extents which will make our size be much
+			 * larger than our reserved amount.  If we happen to
+			 * try to do a reservation here that may result in us
+			 * trying to do a pretty hefty reservation, which we may
+			 * not need once delalloc flushing happens.  If this is
+			 * the case try and do the reserve again.
+			 */
+			if (flush == BTRFS_RESERVE_FLUSH_ALL)
+				__get_refill_bytes(block_rsv, &num_bytes,
+						   &qgroup_num_bytes);
+		}
+	} while (ret && last != num_bytes);
+
 	if (!ret) {
 		block_rsv_add_bytes(block_rsv, num_bytes, 0);
 		trace_btrfs_space_reservation(root->fs_info, "delalloc",
@@ -5809,8 +5836,7 @@ static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
 		spin_lock(&block_rsv->lock);
 		block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
 		spin_unlock(&block_rsv->lock);
-	} else
-		btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
+	}
 	return ret;
 }
 
-- 
2.14.3

[PATCH 17/42] btrfs: run delayed iputs before committing

[Josef Bacik]

Delayed iputs means we can have final iputs of deleted inodes in the
queue, which could potentially generate a lot of pinned space that could
be free'd.  So before we decide to commit the transaction for ENOPSC
reasons, run the delayed iputs so that any potential space is free'd up.
If there is and we freed enough we can then commit the transaction and
potentially be able to make our reservation.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
---
 fs/btrfs/extent-tree.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 3757ca42b8e0..ef852d1911d1 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -4823,6 +4823,15 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	if (!bytes)
 		return 0;
 
+	/*
+	 * If we have pending delayed iputs then we could free up a bunch of
+	 * pinned space, so make sure we run the iputs before we do our pinned
+	 * bytes check below.
+	 */
+	mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
+	btrfs_run_delayed_iputs(fs_info);
+	mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
+
 	trans = btrfs_join_transaction(fs_info->extent_root);
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
-- 
2.14.3

[PATCH 18/42] btrfs: move the dio_sem higher up the callchain

[Josef Bacik]

We're getting a lockdep splat because we take the dio_sem under the
log_mutex.  What we really need is to protect fsync() from logging an
extent map for an extent we never waited on higher up, so just guard the
whole thing with dio_sem.

======================================================
WARNING: possible circular locking dependency detected
4.18.0-rc4-xfstests-00025-g5de5edbaf1d4 #411 Not tainted
------------------------------------------------------
aio-dio-invalid/30928 is trying to acquire lock:
0000000092621cfd (&mm->mmap_sem){++++}, at: get_user_pages_unlocked+0x5a/0x1e0

but task is already holding lock:
00000000cefe6b35 (&ei->dio_sem){++++}, at: btrfs_direct_IO+0x3be/0x400

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #5 (&ei->dio_sem){++++}:
       lock_acquire+0xbd/0x220
       down_write+0x51/0xb0
       btrfs_log_changed_extents+0x80/0xa40
       btrfs_log_inode+0xbaf/0x1000
       btrfs_log_inode_parent+0x26f/0xa80
       btrfs_log_dentry_safe+0x50/0x70
       btrfs_sync_file+0x357/0x540
       do_fsync+0x38/0x60
       __ia32_sys_fdatasync+0x12/0x20
       do_fast_syscall_32+0x9a/0x2f0
       entry_SYSENTER_compat+0x84/0x96

-> #4 (&ei->log_mutex){+.+.}:
       lock_acquire+0xbd/0x220
       __mutex_lock+0x86/0xa10
       btrfs_record_unlink_dir+0x2a/0xa0
       btrfs_unlink+0x5a/0xc0
       vfs_unlink+0xb1/0x1a0
       do_unlinkat+0x264/0x2b0
       do_fast_syscall_32+0x9a/0x2f0
       entry_SYSENTER_compat+0x84/0x96

-> #3 (sb_internal#2){.+.+}:
       lock_acquire+0xbd/0x220
       __sb_start_write+0x14d/0x230
       start_transaction+0x3e6/0x590
       btrfs_evict_inode+0x475/0x640
       evict+0xbf/0x1b0
       btrfs_run_delayed_iputs+0x6c/0x90
       cleaner_kthread+0x124/0x1a0
       kthread+0x106/0x140
       ret_from_fork+0x3a/0x50

-> #2 (&fs_info->cleaner_delayed_iput_mutex){+.+.}:
       lock_acquire+0xbd/0x220
       __mutex_lock+0x86/0xa10
       btrfs_alloc_data_chunk_ondemand+0x197/0x530
       btrfs_check_data_free_space+0x4c/0x90
       btrfs_delalloc_reserve_space+0x20/0x60
       btrfs_page_mkwrite+0x87/0x520
       do_page_mkwrite+0x31/0xa0
       __handle_mm_fault+0x799/0xb00
       handle_mm_fault+0x7c/0xe0
       __do_page_fault+0x1d3/0x4a0
       async_page_fault+0x1e/0x30

-> #1 (sb_pagefaults){.+.+}:
       lock_acquire+0xbd/0x220
       __sb_start_write+0x14d/0x230
       btrfs_page_mkwrite+0x6a/0x520
       do_page_mkwrite+0x31/0xa0
       __handle_mm_fault+0x799/0xb00
       handle_mm_fault+0x7c/0xe0
       __do_page_fault+0x1d3/0x4a0
       async_page_fault+0x1e/0x30

-> #0 (&mm->mmap_sem){++++}:
       __lock_acquire+0x42e/0x7a0
       lock_acquire+0xbd/0x220
       down_read+0x48/0xb0
       get_user_pages_unlocked+0x5a/0x1e0
       get_user_pages_fast+0xa4/0x150
       iov_iter_get_pages+0xc3/0x340
       do_direct_IO+0xf93/0x1d70
       __blockdev_direct_IO+0x32d/0x1c20
       btrfs_direct_IO+0x227/0x400
       generic_file_direct_write+0xcf/0x180
       btrfs_file_write_iter+0x308/0x58c
       aio_write+0xf8/0x1d0
       io_submit_one+0x3a9/0x620
       __ia32_compat_sys_io_submit+0xb2/0x270
       do_int80_syscall_32+0x5b/0x1a0
       entry_INT80_compat+0x88/0xa0

other info that might help us debug this:

Chain exists of:
  &mm->mmap_sem --> &ei->log_mutex --> &ei->dio_sem

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&ei->dio_sem);
                               lock(&ei->log_mutex);
                               lock(&ei->dio_sem);
  lock(&mm->mmap_sem);

 *** DEADLOCK ***

1 lock held by aio-dio-invalid/30928:
 #0: 00000000cefe6b35 (&ei->dio_sem){++++}, at: btrfs_direct_IO+0x3be/0x400

stack backtrace:
CPU: 0 PID: 30928 Comm: aio-dio-invalid Not tainted 4.18.0-rc4-xfstests-00025-g5de5edbaf1d4 #411
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-2.el7 04/01/2014
Call Trace:
 dump_stack+0x7c/0xbb
 print_circular_bug.isra.37+0x297/0x2a4
 check_prev_add.constprop.45+0x781/0x7a0
 ? __lock_acquire+0x42e/0x7a0
 validate_chain.isra.41+0x7f0/0xb00
 __lock_acquire+0x42e/0x7a0
 lock_acquire+0xbd/0x220
 ? get_user_pages_unlocked+0x5a/0x1e0
 down_read+0x48/0xb0
 ? get_user_pages_unlocked+0x5a/0x1e0
 get_user_pages_unlocked+0x5a/0x1e0
 get_user_pages_fast+0xa4/0x150
 iov_iter_get_pages+0xc3/0x340
 do_direct_IO+0xf93/0x1d70
 ? __alloc_workqueue_key+0x358/0x490
 ? __blockdev_direct_IO+0x14b/0x1c20
 __blockdev_direct_IO+0x32d/0x1c20
 ? btrfs_run_delalloc_work+0x40/0x40
 ? can_nocow_extent+0x490/0x490
 ? kvm_clock_read+0x1f/0x30
 ? can_nocow_extent+0x490/0x490
 ? btrfs_run_delalloc_work+0x40/0x40
 btrfs_direct_IO+0x227/0x400
 ? btrfs_run_delalloc_work+0x40/0x40
 generic_file_direct_write+0xcf/0x180
 btrfs_file_write_iter+0x308/0x58c
 aio_write+0xf8/0x1d0
 ? kvm_clock_read+0x1f/0x30
 ? __might_fault+0x3e/0x90
 io_submit_one+0x3a9/0x620
 ? io_submit_one+0xe5/0x620
 __ia32_compat_sys_io_submit+0xb2/0x270
 do_int80_syscall_32+0x5b/0x1a0
 entry_INT80_compat+0x88/0xa0

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
---
 fs/btrfs/file.c     | 12 ++++++++++++
 fs/btrfs/tree-log.c |  2 --
 2 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 095f0bb86bb7..c07110edb9de 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -2079,6 +2079,14 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 		goto out;
 
 	inode_lock(inode);
+
+	/*
+	 * We take the dio_sem here because the tree log stuff can race with
+	 * lockless dio writes and get an extent map logged for an extent we
+	 * never waited on.  We need it this high up for lockdep reasons.
+	 */
+	down_write(&BTRFS_I(inode)->dio_sem);
+
 	atomic_inc(&root->log_batch);
 
 	/*
@@ -2087,6 +2095,7 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 	 */
 	ret = btrfs_wait_ordered_range(inode, start, len);
 	if (ret) {
+		up_write(&BTRFS_I(inode)->dio_sem);
 		inode_unlock(inode);
 		goto out;
 	}
@@ -2110,6 +2119,7 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 		 * checked called fsync.
 		 */
 		ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
+		up_write(&BTRFS_I(inode)->dio_sem);
 		inode_unlock(inode);
 		goto out;
 	}
@@ -2128,6 +2138,7 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 	trans = btrfs_start_transaction(root, 0);
 	if (IS_ERR(trans)) {
 		ret = PTR_ERR(trans);
+		up_write(&BTRFS_I(inode)->dio_sem);
 		inode_unlock(inode);
 		goto out;
 	}
@@ -2149,6 +2160,7 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 	 * file again, but that will end up using the synchronization
 	 * inside btrfs_sync_log to keep things safe.
 	 */
+	up_write(&BTRFS_I(inode)->dio_sem);
 	inode_unlock(inode);
 
 	/*
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index 1650dc44a5e3..66b7e059b765 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -4374,7 +4374,6 @@ static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
 
 	INIT_LIST_HEAD(&extents);
 
-	down_write(&inode->dio_sem);
 	write_lock(&tree->lock);
 	test_gen = root->fs_info->last_trans_committed;
 	logged_start = start;
@@ -4440,7 +4439,6 @@ static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
 	}
 	WARN_ON(!list_empty(&extents));
 	write_unlock(&tree->lock);
-	up_write(&inode->dio_sem);
 
 	btrfs_release_path(path);
 	if (!ret)
-- 
2.14.3

Re: [PATCH 18/42] btrfs: move the dio_sem higher up the callchain

[David Sterba]

On Fri, Oct 12, 2018 at 03:32:32PM -0400, Josef Bacik wrote:
> --- a/fs/btrfs/tree-log.c
> +++ b/fs/btrfs/tree-log.c
> @@ -4374,7 +4374,6 @@ static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
>  
>  	INIT_LIST_HEAD(&extents);
>  
> -	down_write(&inode->dio_sem);

I'll add lockdep_assert_held(&inode->dio_sem) so it's documented and
runtime checked.

>  	write_lock(&tree->lock);
>  	test_gen = root->fs_info->last_trans_committed;
>  	logged_start = start;
> @@ -4440,7 +4439,6 @@ static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
>  	}
>  	WARN_ON(!list_empty(&extents));
>  	write_unlock(&tree->lock);
> -	up_write(&inode->dio_sem);
>  
>  	btrfs_release_path(path);
>  	if (!ret)
> -- 
> 2.14.3

[PATCH 19/42] btrfs: set max_extent_size properly

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

We can't use entry->bytes if our entry is a bitmap entry, we need to use
entry->max_extent_size in that case.  Fix up all the logic to make this
consistent.

Signed-off-by: Josef Bacik <jbacik@fb.com>
---
 fs/btrfs/free-space-cache.c | 30 ++++++++++++++++++++----------
 1 file changed, 20 insertions(+), 10 deletions(-)

diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index e077ad3b4549..2dd773e96530 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -1770,6 +1770,13 @@ static int search_bitmap(struct btrfs_free_space_ctl *ctl,
 	return -1;
 }
 
+static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
+{
+	if (entry->bitmap)
+		return entry->max_extent_size;
+	return entry->bytes;
+}
+
 /* Cache the size of the max extent in bytes */
 static struct btrfs_free_space *
 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
@@ -1791,8 +1798,8 @@ find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
 	for (node = &entry->offset_index; node; node = rb_next(node)) {
 		entry = rb_entry(node, struct btrfs_free_space, offset_index);
 		if (entry->bytes < *bytes) {
-			if (entry->bytes > *max_extent_size)
-				*max_extent_size = entry->bytes;
+			*max_extent_size = max(get_max_extent_size(entry),
+					       *max_extent_size);
 			continue;
 		}
 
@@ -1810,8 +1817,8 @@ find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
 		}
 
 		if (entry->bytes < *bytes + align_off) {
-			if (entry->bytes > *max_extent_size)
-				*max_extent_size = entry->bytes;
+			*max_extent_size = max(get_max_extent_size(entry),
+					       *max_extent_size);
 			continue;
 		}
 
@@ -1823,8 +1830,10 @@ find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
 				*offset = tmp;
 				*bytes = size;
 				return entry;
-			} else if (size > *max_extent_size) {
-				*max_extent_size = size;
+			} else {
+				*max_extent_size =
+					max(get_max_extent_size(entry),
+					    *max_extent_size);
 			}
 			continue;
 		}
@@ -2684,8 +2693,8 @@ static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
 
 	err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
 	if (err) {
-		if (search_bytes > *max_extent_size)
-			*max_extent_size = search_bytes;
+		*max_extent_size = max(get_max_extent_size(entry),
+				       *max_extent_size);
 		return 0;
 	}
 
@@ -2722,8 +2731,9 @@ u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
 
 	entry = rb_entry(node, struct btrfs_free_space, offset_index);
 	while (1) {
-		if (entry->bytes < bytes && entry->bytes > *max_extent_size)
-			*max_extent_size = entry->bytes;
+		if (entry->bytes < bytes)
+			*max_extent_size = max(get_max_extent_size(entry),
+					       *max_extent_size);
 
 		if (entry->bytes < bytes ||
 		    (!entry->bitmap && entry->offset < min_start)) {
-- 
2.14.3

Re: [PATCH 19/42] btrfs: set max_extent_size properly

[David Sterba]

On Fri, Oct 12, 2018 at 03:32:33PM -0400, Josef Bacik wrote:
> From: Josef Bacik <jbacik@fb.com>
> 
> We can't use entry->bytes if our entry is a bitmap entry, we need to use
> entry->max_extent_size in that case.  Fix up all the logic to make this
> consistent.
> 
> Signed-off-by: Josef Bacik <jbacik@fb.com>
> ---
>  fs/btrfs/free-space-cache.c | 30 ++++++++++++++++++++----------
>  1 file changed, 20 insertions(+), 10 deletions(-)
> 
> diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
> index e077ad3b4549..2dd773e96530 100644
> --- a/fs/btrfs/free-space-cache.c
> +++ b/fs/btrfs/free-space-cache.c
> @@ -1770,6 +1770,13 @@ static int search_bitmap(struct btrfs_free_space_ctl *ctl,
>  	return -1;
>  }
>  
> +static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
> +{
> +	if (entry->bitmap)
> +		return entry->max_extent_size;
> +	return entry->bytes;
> +}

> +			*max_extent_size = max(get_max_extent_size(entry),
> +					       *max_extent_size);

Looks reasonable.

Reviewed-by: David Sterba <dsterba@suse.com>

[PATCH 20/42] btrfs: don't use ctl->free_space for max_extent_size

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

max_extent_size is supposed to be the largest contiguous range for the
space info, and ctl->free_space is the total free space in the block
group.  We need to keep track of these separately and _only_ use the
max_free_space if we don't have a max_extent_size, as that means our
original request was too large to search any of the block groups for and
therefore wouldn't have a max_extent_size set.

Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
---
 fs/btrfs/extent-tree.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index ef852d1911d1..6721698ab8aa 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -7496,6 +7496,7 @@ static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
 	struct btrfs_block_group_cache *block_group = NULL;
 	u64 search_start = 0;
 	u64 max_extent_size = 0;
+	u64 max_free_space = 0;
 	u64 empty_cluster = 0;
 	struct btrfs_space_info *space_info;
 	int loop = 0;
@@ -7791,8 +7792,8 @@ static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
 			spin_lock(&ctl->tree_lock);
 			if (ctl->free_space <
 			    num_bytes + empty_cluster + empty_size) {
-				if (ctl->free_space > max_extent_size)
-					max_extent_size = ctl->free_space;
+				max_free_space = max(max_free_space,
+						     ctl->free_space);
 				spin_unlock(&ctl->tree_lock);
 				goto loop;
 			}
@@ -7959,6 +7960,8 @@ static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
 	}
 out:
 	if (ret == -ENOSPC) {
+		if (!max_extent_size)
+			max_extent_size = max_free_space;
 		spin_lock(&space_info->lock);
 		space_info->max_extent_size = max_extent_size;
 		spin_unlock(&space_info->lock);
-- 
2.14.3

[PATCH 21/42] btrfs: reset max_extent_size on clear in a bitmap

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

We need to clear the max_extent_size when we clear bits from a bitmap
since it could have been from the range that contains the
max_extent_size.

Reviewed-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
---
 fs/btrfs/free-space-cache.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 2dd773e96530..411acffbd45d 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -1687,6 +1687,8 @@ static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
 	bitmap_clear(info->bitmap, start, count);
 
 	info->bytes -= bytes;
+	if (info->max_extent_size > ctl->unit)
+		info->max_extent_size = 0;
 }
 
 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
-- 
2.14.3

[PATCH 22/42] btrfs: only run delayed refs if we're committing

[Josef Bacik]

I noticed in a giant dbench run that we spent a lot of time on lock
contention while running transaction commit.  This is because dbench
results in a lot of fsync()'s that do a btrfs_transaction_commit(), and
they all run the delayed refs first thing, so they all contend with
each other.  This leads to seconds of 0 throughput.  Change this to only
run the delayed refs if we're the ones committing the transaction.  This
makes the latency go away and we get no more lock contention.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/transaction.c | 24 +++++++++---------------
 1 file changed, 9 insertions(+), 15 deletions(-)

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index a0f19ca0bd6c..39a2bddb0b29 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -1925,15 +1925,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	btrfs_trans_release_metadata(trans);
 	trans->block_rsv = NULL;
 
-	/* make a pass through all the delayed refs we have so far
-	 * any runnings procs may add more while we are here
-	 */
-	ret = btrfs_run_delayed_refs(trans, 0);
-	if (ret) {
-		btrfs_end_transaction(trans);
-		return ret;
-	}
-
 	cur_trans = trans->transaction;
 
 	/*
@@ -1946,12 +1937,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	if (!list_empty(&trans->new_bgs))
 		btrfs_create_pending_block_groups(trans);
 
-	ret = btrfs_run_delayed_refs(trans, 0);
-	if (ret) {
-		btrfs_end_transaction(trans);
-		return ret;
-	}
-
 	if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
 		int run_it = 0;
 
@@ -2022,6 +2007,15 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 		spin_unlock(&fs_info->trans_lock);
 	}
 
+	/*
+	 * We are now the only one in the commit area, we can run delayed refs
+	 * without hitting a bunch of lock contention from a lot of people
+	 * trying to commit the transaction at once.
+	 */
+	ret = btrfs_run_delayed_refs(trans, 0);
+	if (ret)
+		goto cleanup_transaction;
+
 	extwriter_counter_dec(cur_trans, trans->type);
 
 	ret = btrfs_start_delalloc_flush(fs_info);
-- 
2.14.3

[PATCH 23/42] btrfs: make sure we create all new bgs

[Josef Bacik]

Allocating new chunks modifies both the extent and chunk tree, which can
trigger new chunk allocations.  So instead of doing list_for_each_safe,
just do while (!list_empty()) so we make sure we don't exit with other
pending bg's still on our list.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 6721698ab8aa..0d7fc07c34ab 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -10361,7 +10361,7 @@ int btrfs_read_block_groups(struct btrfs_fs_info *info)
 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
-	struct btrfs_block_group_cache *block_group, *tmp;
+	struct btrfs_block_group_cache *block_group;
 	struct btrfs_root *extent_root = fs_info->extent_root;
 	struct btrfs_block_group_item item;
 	struct btrfs_key key;
@@ -10369,7 +10369,10 @@ void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
 	bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
 
 	trans->can_flush_pending_bgs = false;
-	list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
+	while (!list_empty(&trans->new_bgs)) {
+		block_group = list_first_entry(&trans->new_bgs,
+					       struct btrfs_block_group_cache,
+					       bg_list);
 		if (ret)
 			goto next;
 
-- 
2.14.3

[PATCH 24/42] btrfs: assert on non-empty delayed iputs

[Josef Bacik]

I ran into an issue where there was some reference being held on an
inode that I couldn't track.  This assert wasn't triggered, but it at
least rules out we're doing something stupid.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/disk-io.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 377ad9c1cb17..39bd158466cd 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -3979,6 +3979,7 @@ void close_ctree(struct btrfs_fs_info *fs_info)
 	kthread_stop(fs_info->transaction_kthread);
 	kthread_stop(fs_info->cleaner_kthread);
 
+	ASSERT(list_empty(&fs_info->delayed_iputs));
 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
 
 	btrfs_free_qgroup_config(fs_info);
-- 
2.14.3

[PATCH 25/42] btrfs: pass delayed_refs_root to btrfs_delayed_ref_lock

[Josef Bacik]

We don't need the trans except to get the delayed_refs_root, so just
pass the delayed_refs_root into btrfs_delayed_ref_lock and call it a
day.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/delayed-ref.c | 5 +----
 fs/btrfs/delayed-ref.h | 2 +-
 fs/btrfs/extent-tree.c | 2 +-
 3 files changed, 3 insertions(+), 6 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index 96ce087747b2..87778645bf4a 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -197,12 +197,9 @@ find_ref_head(struct rb_root *root, u64 bytenr,
 	return NULL;
 }
 
-int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
+int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
 			   struct btrfs_delayed_ref_head *head)
 {
-	struct btrfs_delayed_ref_root *delayed_refs;
-
-	delayed_refs = &trans->transaction->delayed_refs;
 	lockdep_assert_held(&delayed_refs->lock);
 	if (mutex_trylock(&head->mutex))
 		return 0;
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
index 7769177b489e..ee636d7a710a 100644
--- a/fs/btrfs/delayed-ref.h
+++ b/fs/btrfs/delayed-ref.h
@@ -255,7 +255,7 @@ void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
 struct btrfs_delayed_ref_head *
 btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
 			    u64 bytenr);
-int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
+int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
 			   struct btrfs_delayed_ref_head *head);
 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
 {
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 0d7fc07c34ab..529c2aefbbf4 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2600,7 +2600,7 @@ static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 
 			/* grab the lock that says we are going to process
 			 * all the refs for this head */
-			ret = btrfs_delayed_ref_lock(trans, locked_ref);
+			ret = btrfs_delayed_ref_lock(delayed_refs, locked_ref);
 			spin_unlock(&delayed_refs->lock);
 			/*
 			 * we may have dropped the spin lock to get the head
-- 
2.14.3

[PATCH 26/42] btrfs: make btrfs_destroy_delayed_refs use btrfs_delayed_ref_lock

[Josef Bacik]

We have this open coded in btrfs_destroy_delayed_refs, use the helper
instead.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/disk-io.c | 11 ++---------
 1 file changed, 2 insertions(+), 9 deletions(-)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 39bd158466cd..121ab180a78a 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4214,16 +4214,9 @@ static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
 
 		head = rb_entry(node, struct btrfs_delayed_ref_head,
 				href_node);
-		if (!mutex_trylock(&head->mutex)) {
-			refcount_inc(&head->refs);
-			spin_unlock(&delayed_refs->lock);
-
-			mutex_lock(&head->mutex);
-			mutex_unlock(&head->mutex);
-			btrfs_put_delayed_ref_head(head);
-			spin_lock(&delayed_refs->lock);
+		if (btrfs_delayed_ref_lock(delayed_refs, head))
 			continue;
-		}
+
 		spin_lock(&head->lock);
 		while ((n = rb_first(&head->ref_tree)) != NULL) {
 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
-- 
2.14.3

[PATCH 27/42] btrfs: make btrfs_destroy_delayed_refs use btrfs_delete_ref_head

[Josef Bacik]

Instead of open coding this stuff use the helper instead.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/disk-io.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 121ab180a78a..fe1f229320ef 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4232,12 +4232,7 @@ static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
 		if (head->must_insert_reserved)
 			pin_bytes = true;
 		btrfs_free_delayed_extent_op(head->extent_op);
-		delayed_refs->num_heads--;
-		if (head->processing == 0)
-			delayed_refs->num_heads_ready--;
-		atomic_dec(&delayed_refs->num_entries);
-		rb_erase(&head->href_node, &delayed_refs->href_root);
-		RB_CLEAR_NODE(&head->href_node);
+		btrfs_delete_ref_head(delayed_refs, head);
 		spin_unlock(&head->lock);
 		spin_unlock(&delayed_refs->lock);
 		mutex_unlock(&head->mutex);
-- 
2.14.3

[PATCH 28/42] btrfs: handle delayed ref head accounting cleanup in abort

[Josef Bacik]

We weren't doing any of the accounting cleanup when we aborted
transactions.  Fix this by making cleanup_ref_head_accounting global and
calling it from the abort code, this fixes the issue where our
accounting was all wrong after the fs aborts.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/ctree.h       |  5 +++++
 fs/btrfs/disk-io.c     |  1 +
 fs/btrfs/extent-tree.c | 13 ++++++-------
 3 files changed, 12 insertions(+), 7 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 29db902511c1..e40356ca0295 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -35,6 +35,7 @@
 struct btrfs_trans_handle;
 struct btrfs_transaction;
 struct btrfs_pending_snapshot;
+struct btrfs_delayed_ref_root;
 extern struct kmem_cache *btrfs_trans_handle_cachep;
 extern struct kmem_cache *btrfs_bit_radix_cachep;
 extern struct kmem_cache *btrfs_path_cachep;
@@ -2623,6 +2624,10 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 			   unsigned long count);
 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
 				 unsigned long count, u64 transid, int wait);
+void
+btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
+				  struct btrfs_delayed_ref_root *delayed_refs,
+				  struct btrfs_delayed_ref_head *head);
 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len);
 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 			     struct btrfs_fs_info *fs_info, u64 bytenr,
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index fe1f229320ef..54fbdc944a3f 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4240,6 +4240,7 @@ static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
 		if (pin_bytes)
 			btrfs_pin_extent(fs_info, head->bytenr,
 					 head->num_bytes, 1);
+		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
 		btrfs_put_delayed_ref_head(head);
 		cond_resched();
 		spin_lock(&delayed_refs->lock);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 529c2aefbbf4..5b5a28b69fa9 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2475,12 +2475,11 @@ static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
 	return ret ? ret : 1;
 }
 
-static void cleanup_ref_head_accounting(struct btrfs_trans_handle *trans,
-					struct btrfs_delayed_ref_head *head)
+void
+btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
+				  struct btrfs_delayed_ref_root *delayed_refs,
+				  struct btrfs_delayed_ref_head *head)
 {
-	struct btrfs_fs_info *fs_info = trans->fs_info;
-	struct btrfs_delayed_ref_root *delayed_refs =
-		&trans->transaction->delayed_refs;
 	int nr_items = 1;
 
 	if (head->total_ref_mod < 0) {
@@ -2558,7 +2557,7 @@ static int cleanup_ref_head(struct btrfs_trans_handle *trans,
 		}
 	}
 
-	cleanup_ref_head_accounting(trans, head);
+	btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
 
 	trace_run_delayed_ref_head(fs_info, head, 0);
 	btrfs_delayed_ref_unlock(head);
@@ -7227,7 +7226,7 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	if (head->must_insert_reserved)
 		ret = 1;
 
-	cleanup_ref_head_accounting(trans, head);
+	btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
 	mutex_unlock(&head->mutex);
 	btrfs_put_delayed_ref_head(head);
 	return ret;
-- 
2.14.3

[PATCH 29/42] btrfs: call btrfs_create_pending_block_groups unconditionally

[Josef Bacik]

The first thing we do is loop through the list, this

if (!list_empty())
	btrfs_create_pending_block_groups();

thing is just wasted space.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 3 +--
 fs/btrfs/transaction.c | 6 ++----
 2 files changed, 3 insertions(+), 6 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 5b5a28b69fa9..14495c030301 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2978,8 +2978,7 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 	}
 
 	if (run_all) {
-		if (!list_empty(&trans->new_bgs))
-			btrfs_create_pending_block_groups(trans);
+		btrfs_create_pending_block_groups(trans);
 
 		spin_lock(&delayed_refs->lock);
 		node = rb_first(&delayed_refs->href_root);
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 39a2bddb0b29..46ca775a709e 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -846,8 +846,7 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
 	btrfs_trans_release_metadata(trans);
 	trans->block_rsv = NULL;
 
-	if (!list_empty(&trans->new_bgs))
-		btrfs_create_pending_block_groups(trans);
+	btrfs_create_pending_block_groups(trans);
 
 	btrfs_trans_release_chunk_metadata(trans);
 
@@ -1934,8 +1933,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	cur_trans->delayed_refs.flushing = 1;
 	smp_wmb();
 
-	if (!list_empty(&trans->new_bgs))
-		btrfs_create_pending_block_groups(trans);
+	btrfs_create_pending_block_groups(trans);
 
 	if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) {
 		int run_it = 0;
-- 
2.14.3

[PATCH 30/42] btrfs: just delete pending bgs if we are aborted

[Josef Bacik]

We still need to do all of the accounting cleanup for pending block
groups if we abort.  So set the ret to trans->aborted so if we aborted
the cleanup happens and everybody is happy.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 14495c030301..933cba06c9fb 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -10363,9 +10363,15 @@ void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
 	struct btrfs_root *extent_root = fs_info->extent_root;
 	struct btrfs_block_group_item item;
 	struct btrfs_key key;
-	int ret = 0;
+	int ret;
 	bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
 
+	/*
+	 * If we aborted the transaction with pending bg's we need to just
+	 * cleanup the list and carry on.
+	 */
+	ret = trans->aborted;
+
 	trans->can_flush_pending_bgs = false;
 	while (!list_empty(&trans->new_bgs)) {
 		block_group = list_first_entry(&trans->new_bgs,
-- 
2.14.3

[PATCH 31/42] btrfs: cleanup pending bgs on transaction abort

[Josef Bacik]

We may abort the transaction during a commit and not have a chance to
run the pending bgs stuff, which will leave block groups on our list and
cause us accounting issues and leaked memory.  Fix this by running the
pending bgs when we cleanup a transaction.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/transaction.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 46ca775a709e..9168efaca37e 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -2280,6 +2280,10 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	btrfs_scrub_continue(fs_info);
 cleanup_transaction:
 	btrfs_trans_release_metadata(trans);
+	/* This cleans up the pending block groups list properly. */
+	if (!trans->aborted)
+		trans->aborted = ret;
+	btrfs_create_pending_block_groups(trans);
 	btrfs_trans_release_chunk_metadata(trans);
 	trans->block_rsv = NULL;
 	btrfs_warn(fs_info, "Skipping commit of aborted transaction.");
-- 
2.14.3

[PATCH 32/42] btrfs: only free reserved extent if we didn't insert it

[Josef Bacik]

When we insert the file extent once the ordered extent completes we free
the reserved extent reservation as it'll have been migrated to the
bytes_used counter.  However if we error out after this step we'll still
clear the reserved extent reservation, resulting in a negative
accounting of the reserved bytes for the block group and space info.
Fix this by only doing the free if we didn't successfully insert a file
extent for this extent.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
---
 fs/btrfs/inode.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 5a91055a13b2..2b257d14bd3d 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2992,6 +2992,7 @@ static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
 	bool truncated = false;
 	bool range_locked = false;
 	bool clear_new_delalloc_bytes = false;
+	bool clear_reserved_extent = true;
 
 	if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
 	    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
@@ -3095,10 +3096,12 @@ static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
 						logical_len, logical_len,
 						compress_type, 0, 0,
 						BTRFS_FILE_EXTENT_REG);
-		if (!ret)
+		if (!ret) {
+			clear_reserved_extent = false;
 			btrfs_release_delalloc_bytes(fs_info,
 						     ordered_extent->start,
 						     ordered_extent->disk_len);
+		}
 	}
 	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
 			   ordered_extent->file_offset, ordered_extent->len,
@@ -3159,8 +3162,13 @@ static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
 		 * wrong we need to return the space for this ordered extent
 		 * back to the allocator.  We only free the extent in the
 		 * truncated case if we didn't write out the extent at all.
+		 *
+		 * If we made it past insert_reserved_file_extent before we
+		 * errored out then we don't need to do this as the accounting
+		 * has already been done.
 		 */
 		if ((ret || !logical_len) &&
+		    clear_reserved_extent &&
 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
 		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
 			btrfs_free_reserved_extent(fs_info,
-- 
2.14.3

[PATCH 33/42] btrfs: fix insert_reserved error handling

[Josef Bacik]

We were not handling the reserved byte accounting properly for data
references.  Metadata was fine, if it errored out the error paths would
free the bytes_reserved count and pin the extent, but it even missed one
of the error cases.  So instead move this handling up into
run_one_delayed_ref so we are sure that both cases are properly cleaned
up in case of a transaction abort.

Reviewed-by: David Sterba <dsterba@suse.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 12 ++++--------
 1 file changed, 4 insertions(+), 8 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 933cba06c9fb..9b5366932298 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2405,6 +2405,9 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 					   insert_reserved);
 	else
 		BUG();
+	if (ret && insert_reserved)
+		btrfs_pin_extent(trans->fs_info, node->bytenr,
+				 node->num_bytes, 1);
 	return ret;
 }
 
@@ -8257,21 +8260,14 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 	}
 
 	path = btrfs_alloc_path();
-	if (!path) {
-		btrfs_free_and_pin_reserved_extent(fs_info,
-						   extent_key.objectid,
-						   fs_info->nodesize);
+	if (!path)
 		return -ENOMEM;
-	}
 
 	path->leave_spinning = 1;
 	ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
 				      &extent_key, size);
 	if (ret) {
 		btrfs_free_path(path);
-		btrfs_free_and_pin_reserved_extent(fs_info,
-						   extent_key.objectid,
-						   fs_info->nodesize);
 		return ret;
 	}
 
-- 
2.14.3

[PATCH 34/42] btrfs: wait on ordered extents on abort cleanup

[Josef Bacik]

If we flip read-only before we initiate writeback on all dirty pages for
ordered extents we've created then we'll have ordered extents left over
on umount, which results in all sorts of bad things happening.  Fix this
by making sure we wait on ordered extents if we have to do the aborted
transaction cleanup stuff.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/disk-io.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 54fbdc944a3f..51b2a5bf25e5 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4188,6 +4188,14 @@ static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
 		spin_lock(&fs_info->ordered_root_lock);
 	}
 	spin_unlock(&fs_info->ordered_root_lock);
+
+	/*
+	 * We need this here because if we've been flipped read-only we won't
+	 * get sync() from the umount, so we need to make sure any ordered
+	 * extents that haven't had their dirty pages IO start writeout yet
+	 * actually get run and error out properly.
+	 */
+	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 }
 
 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
-- 
2.14.3

[PATCH 35/42] MAINTAINERS: update my email address for btrfs

[Josef Bacik]

My work email is completely useless, switch it to my personal address so
I get emails on a account I actually pay attention to.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 MAINTAINERS | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/MAINTAINERS b/MAINTAINERS
index 32fbc6f732d4..7723dc958e99 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -3095,7 +3095,7 @@ F:	drivers/gpio/gpio-bt8xx.c
 
 BTRFS FILE SYSTEM
 M:	Chris Mason <clm@fb.com>
-M:	Josef Bacik <jbacik@fb.com>
+M:	Josef Bacik <josef@toxicpanda.com>
 M:	David Sterba <dsterba@suse.com>
 L:	linux-btrfs@vger.kernel.org
 W:	http://btrfs.wiki.kernel.org/
-- 
2.14.3

[PATCH 36/42] btrfs: wait on caching when putting the bg cache

[Josef Bacik]

While testing my backport I noticed there was a panic if I ran
generic/416 generic/417 generic/418 all in a row.  This just happened to
uncover a race where we had outstanding IO after we destroy all of our
workqueues, and then we'd go to queue the endio work on those free'd
workqueues.  This is because we aren't waiting for the caching threads
to be done before freeing everything up, so to fix this make sure we
wait on any outstanding caching that's being done before we free up the
block group, so we're sure to be done with all IO by the time we get to
btrfs_stop_all_workers().  This fixes the panic I was seeing
consistently in testing.

------------[ cut here ]------------
kernel BUG at fs/btrfs/volumes.c:6112!
SMP PTI
Modules linked in:
CPU: 1 PID: 27165 Comm: kworker/u4:7 Not tainted 4.16.0-02155-g3553e54a578d-dirty #875
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-2.el7 04/01/2014
Workqueue: btrfs-cache btrfs_cache_helper
RIP: 0010:btrfs_map_bio+0x346/0x370
RSP: 0000:ffffc900061e79d0 EFLAGS: 00010202
RAX: 0000000000000000 RBX: ffff880071542e00 RCX: 0000000000533000
RDX: ffff88006bb74380 RSI: 0000000000000008 RDI: ffff880078160000
RBP: 0000000000000001 R08: ffff8800781cd200 R09: 0000000000503000
R10: ffff88006cd21200 R11: 0000000000000000 R12: 0000000000000000
R13: 0000000000000000 R14: ffff8800781cd200 R15: ffff880071542e00
FS:  0000000000000000(0000) GS:ffff88007fd00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000000000817ffc4 CR3: 0000000078314000 CR4: 00000000000006e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 btree_submit_bio_hook+0x8a/0xd0
 submit_one_bio+0x5d/0x80
 read_extent_buffer_pages+0x18a/0x320
 btree_read_extent_buffer_pages+0xbc/0x200
 ? alloc_extent_buffer+0x359/0x3e0
 read_tree_block+0x3d/0x60
 read_block_for_search.isra.30+0x1a5/0x360
 btrfs_search_slot+0x41b/0xa10
 btrfs_next_old_leaf+0x212/0x470
 caching_thread+0x323/0x490
 normal_work_helper+0xc5/0x310
 process_one_work+0x141/0x340
 worker_thread+0x44/0x3c0
 kthread+0xf8/0x130
 ? process_one_work+0x340/0x340
 ? kthread_bind+0x10/0x10
 ret_from_fork+0x35/0x40
Code: ff ff 48 8b 4c 24 28 48 89 de 48 8b 7c 24 08 e8 d1 e5 04 00 89 c3 e9 08 ff ff ff 4d 89 c6 49 89 df e9 27 fe ff ff e8 5a 3a bb ff <0f> 0b 0f 0b e9 57 ff ff ff 48 8b 7c 24 08 4c 89 f9 4c 89 ea 48
RIP: btrfs_map_bio+0x346/0x370 RSP: ffffc900061e79d0
---[ end trace 827eb13e50846033 ]---
Kernel panic - not syncing: Fatal exception
Kernel Offset: disabled
---[ end Kernel panic - not syncing: Fatal exception

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
---
 fs/btrfs/extent-tree.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 9b5366932298..4bacb5042d08 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -9894,6 +9894,7 @@ void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
 
 		block_group = btrfs_lookup_first_block_group(info, last);
 		while (block_group) {
+			wait_block_group_cache_done(block_group);
 			spin_lock(&block_group->lock);
 			if (block_group->iref)
 				break;
-- 
2.14.3

[PATCH 37/42] btrfs: wakeup cleaner thread when adding delayed iput

[Josef Bacik]

The cleaner thread usually takes care of delayed iputs, with the
exception of the btrfs_end_transaction_throttle path.  The cleaner
thread only gets woken up every 30 seconds, so instead wake it up to do
it's work so that we can free up that space as quickly as possible.

Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/inode.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 2b257d14bd3d..0a1671fb03bf 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3323,6 +3323,7 @@ void btrfs_add_delayed_iput(struct inode *inode)
 	ASSERT(list_empty(&binode->delayed_iput));
 	list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
 	spin_unlock(&fs_info->delayed_iput_lock);
+	wake_up_process(fs_info->cleaner_kthread);
 }
 
 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
-- 
2.14.3

[PATCH 38/42] btrfs: be more explicit about allowed flush states

[Josef Bacik]

For FLUSH_LIMIT flushers we really can only allocate chunks and flush
delayed inode items, everything else is problematic.  I added a bunch of
new states and it lead to weirdness in the FLUSH_LIMIT case because I
forgot about how it worked.  So instead explicitly declare the states
that are ok for flushing with FLUSH_LIMIT and use that for our state
machine.  Then as we add new things that are safe we can just add them
to this list.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent-tree.c | 21 ++++++++++-----------
 1 file changed, 10 insertions(+), 11 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 4bacb5042d08..2056d5d37dfe 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -5110,12 +5110,18 @@ void btrfs_init_async_reclaim_work(struct work_struct *work)
 	INIT_WORK(work, btrfs_async_reclaim_metadata_space);
 }
 
+static const enum btrfs_flush_state priority_flush_states[] = {
+	FLUSH_DELAYED_ITEMS_NR,
+	FLUSH_DELAYED_ITEMS,
+	ALLOC_CHUNK,
+};
+
 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
 					    struct btrfs_space_info *space_info,
 					    struct reserve_ticket *ticket)
 {
 	u64 to_reclaim;
-	int flush_state = FLUSH_DELAYED_ITEMS_NR;
+	int flush_state = 0;
 
 	spin_lock(&space_info->lock);
 	to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
@@ -5127,7 +5133,8 @@ static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
 	spin_unlock(&space_info->lock);
 
 	do {
-		flush_space(fs_info, space_info, to_reclaim, flush_state);
+		flush_space(fs_info, space_info, to_reclaim,
+			    priority_flush_states[flush_state]);
 		flush_state++;
 		spin_lock(&space_info->lock);
 		if (ticket->bytes == 0) {
@@ -5135,15 +5142,7 @@ static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
 			return;
 		}
 		spin_unlock(&space_info->lock);
-
-		/*
-		 * Priority flushers can't wait on delalloc without
-		 * deadlocking.
-		 */
-		if (flush_state == FLUSH_DELALLOC ||
-		    flush_state == FLUSH_DELALLOC_WAIT)
-			flush_state = ALLOC_CHUNK;
-	} while (flush_state < COMMIT_TRANS);
+	} while (flush_state < ARRAY_SIZE(priority_flush_states));
 }
 
 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
-- 
2.14.3

[PATCH 39/42] btrfs: replace cleaner_delayed_iput_mutex with a waitqueue

[Josef Bacik]

The throttle path doesn't take cleaner_delayed_iput_mutex, which means
we could think we're done flushing iputs in the data space reservation
path when we could have a throttler doing an iput.  There's no real
reason to serialize the delayed iput flushing, so instead of taking the
cleaner_delayed_iput_mutex whenever we flush the delayed iputs just
replace it with an atomic counter and a waitqueue.  This removes the
short (or long depending on how big the inode is) window where we think
there are no more pending iputs when there really are some.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/ctree.h       |  4 +++-
 fs/btrfs/disk-io.c     |  5 ++---
 fs/btrfs/extent-tree.c |  9 +++++----
 fs/btrfs/inode.c       | 21 +++++++++++++++++++++
 4 files changed, 31 insertions(+), 8 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index e40356ca0295..1ef0b1649cad 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -894,7 +894,8 @@ struct btrfs_fs_info {
 
 	spinlock_t delayed_iput_lock;
 	struct list_head delayed_iputs;
-	struct mutex cleaner_delayed_iput_mutex;
+	atomic_t nr_delayed_iputs;
+	wait_queue_head_t delayed_iputs_wait;
 
 	/* this protects tree_mod_seq_list */
 	spinlock_t tree_mod_seq_lock;
@@ -3212,6 +3213,7 @@ int btrfs_orphan_cleanup(struct btrfs_root *root);
 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size);
 void btrfs_add_delayed_iput(struct inode *inode);
 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info);
+int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info);
 int btrfs_prealloc_file_range(struct inode *inode, int mode,
 			      u64 start, u64 num_bytes, u64 min_size,
 			      loff_t actual_len, u64 *alloc_hint);
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 51b2a5bf25e5..3dce9ff72e41 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1692,9 +1692,7 @@ static int cleaner_kthread(void *arg)
 			goto sleep;
 		}
 
-		mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
 		btrfs_run_delayed_iputs(fs_info);
-		mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
 
 		again = btrfs_clean_one_deleted_snapshot(root);
 		mutex_unlock(&fs_info->cleaner_mutex);
@@ -2677,7 +2675,6 @@ int open_ctree(struct super_block *sb,
 	mutex_init(&fs_info->delete_unused_bgs_mutex);
 	mutex_init(&fs_info->reloc_mutex);
 	mutex_init(&fs_info->delalloc_root_mutex);
-	mutex_init(&fs_info->cleaner_delayed_iput_mutex);
 	seqlock_init(&fs_info->profiles_lock);
 
 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
@@ -2699,6 +2696,7 @@ int open_ctree(struct super_block *sb,
 	atomic_set(&fs_info->defrag_running, 0);
 	atomic_set(&fs_info->qgroup_op_seq, 0);
 	atomic_set(&fs_info->reada_works_cnt, 0);
+	atomic_set(&fs_info->nr_delayed_iputs, 0);
 	atomic64_set(&fs_info->tree_mod_seq, 0);
 	fs_info->sb = sb;
 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
@@ -2776,6 +2774,7 @@ int open_ctree(struct super_block *sb,
 	init_waitqueue_head(&fs_info->transaction_wait);
 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
 	init_waitqueue_head(&fs_info->async_submit_wait);
+	init_waitqueue_head(&fs_info->delayed_iputs_wait);
 
 	INIT_LIST_HEAD(&fs_info->pinned_chunks);
 
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 2056d5d37dfe..d0a7b0589826 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -4258,8 +4258,9 @@ int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
 				 * operations. Wait for it to finish so that
 				 * more space is released.
 				 */
-				mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
-				mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
+				ret = btrfs_wait_on_delayed_iputs(fs_info);
+				if (ret)
+					return ret;
 				goto again;
 			} else {
 				btrfs_end_transaction(trans);
@@ -4829,9 +4830,9 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	 * pinned space, so make sure we run the iputs before we do our pinned
 	 * bytes check below.
 	 */
-	mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
 	btrfs_run_delayed_iputs(fs_info);
-	mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
+	wait_event(fs_info->delayed_iputs_wait,
+		   atomic_read(&fs_info->nr_delayed_iputs) == 0);
 
 	trans = btrfs_join_transaction(fs_info->extent_root);
 	if (IS_ERR(trans))
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 0a1671fb03bf..ab8242b10601 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3319,6 +3319,7 @@ void btrfs_add_delayed_iput(struct inode *inode)
 	if (atomic_add_unless(&inode->i_count, -1, 1))
 		return;
 
+	atomic_inc(&fs_info->nr_delayed_iputs);
 	spin_lock(&fs_info->delayed_iput_lock);
 	ASSERT(list_empty(&binode->delayed_iput));
 	list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
@@ -3338,11 +3339,31 @@ void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
 		list_del_init(&inode->delayed_iput);
 		spin_unlock(&fs_info->delayed_iput_lock);
 		iput(&inode->vfs_inode);
+		if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
+			wake_up(&fs_info->delayed_iputs_wait);
 		spin_lock(&fs_info->delayed_iput_lock);
 	}
 	spin_unlock(&fs_info->delayed_iput_lock);
 }
 
+/**
+ * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
+ * @fs_info - the fs_info for this fs
+ * @return - EINTR if we were killed, 0 if nothing's pending
+ *
+ * This will wait on any delayed iputs that are currently running with KILLABLE
+ * set.  Once they are all done running we will return, unless we are killed in
+ * which case we return EINTR.
+ */
+int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
+{
+	int ret = wait_event_killable(fs_info->delayed_iputs_wait,
+			atomic_read(&fs_info->nr_delayed_iputs) == 0);
+	if (ret)
+		return -EINTR;
+	return 0;
+}
+
 /*
  * This creates an orphan entry for the given inode in case something goes wrong
  * in the middle of an unlink.
-- 
2.14.3

[PATCH 40/42] btrfs: drop min_size from evict_refill_and_join

[Josef Bacik]

We don't need it, rsv->size is set once and never changes throughout
its lifetime, so just use that for the reserve size.

Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/inode.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index ab8242b10601..dbcca915e681 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -5339,8 +5339,7 @@ static void evict_inode_truncate_pages(struct inode *inode)
 }
 
 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
-							struct btrfs_block_rsv *rsv,
-							u64 min_size)
+							struct btrfs_block_rsv *rsv)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
@@ -5350,7 +5349,7 @@ static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
 		struct btrfs_trans_handle *trans;
 		int ret;
 
-		ret = btrfs_block_rsv_refill(root, rsv, min_size,
+		ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
 					     BTRFS_RESERVE_FLUSH_LIMIT);
 
 		if (ret && ++failures > 2) {
@@ -5368,7 +5367,7 @@ static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
 		 * it.
 		 */
 		if (!btrfs_check_space_for_delayed_refs(fs_info) &&
-		    !btrfs_block_rsv_migrate(global_rsv, rsv, min_size, 0))
+		    !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0))
 			return trans;
 
 		/* If not, commit and try again. */
@@ -5384,7 +5383,6 @@ void btrfs_evict_inode(struct inode *inode)
 	struct btrfs_trans_handle *trans;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_block_rsv *rsv;
-	u64 min_size;
 	int ret;
 
 	trace_btrfs_inode_evict(inode);
@@ -5394,8 +5392,6 @@ void btrfs_evict_inode(struct inode *inode)
 		return;
 	}
 
-	min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
-
 	evict_inode_truncate_pages(inode);
 
 	if (inode->i_nlink &&
@@ -5428,13 +5424,13 @@ void btrfs_evict_inode(struct inode *inode)
 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
 	if (!rsv)
 		goto no_delete;
-	rsv->size = min_size;
+	rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
 	rsv->failfast = 1;
 
 	btrfs_i_size_write(BTRFS_I(inode), 0);
 
 	while (1) {
-		trans = evict_refill_and_join(root, rsv, min_size);
+		trans = evict_refill_and_join(root, rsv);
 		if (IS_ERR(trans))
 			goto free_rsv;
 
@@ -5459,7 +5455,7 @@ void btrfs_evict_inode(struct inode *inode)
 	 * If it turns out that we are dropping too many of these, we might want
 	 * to add a mechanism for retrying these after a commit.
 	 */
-	trans = evict_refill_and_join(root, rsv, min_size);
+	trans = evict_refill_and_join(root, rsv);
 	if (!IS_ERR(trans)) {
 		trans->block_rsv = rsv;
 		btrfs_orphan_del(trans, BTRFS_I(inode));
-- 
2.14.3

[PATCH 41/42] btrfs: reserve extra space during evict()

[Josef Bacik]

We could generate a lot of delayed refs in evict but never have any left
over space from our block rsv to make up for that fact.  So reserve some
extra space and give it to the transaction so it can be used to refill
the delayed refs rsv every loop through the truncate path.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/inode.c | 25 +++++++++++++++++++++++--
 1 file changed, 23 insertions(+), 2 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index dbcca915e681..9f7da5e3c741 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -5343,13 +5343,15 @@ static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
+	u64 delayed_refs_extra = btrfs_calc_trans_metadata_size(fs_info, 1);
 	int failures = 0;
 
 	for (;;) {
 		struct btrfs_trans_handle *trans;
 		int ret;
 
-		ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
+		ret = btrfs_block_rsv_refill(root, rsv,
+					     rsv->size + delayed_refs_extra,
 					     BTRFS_RESERVE_FLUSH_LIMIT);
 
 		if (ret && ++failures > 2) {
@@ -5358,9 +5360,28 @@ static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
 			return ERR_PTR(-ENOSPC);
 		}
 
+		/*
+		 * Evict can generate a large amount of delayed refs without
+		 * having a way to add space back since we exhaust our temporary
+		 * block rsv.  We aren't allowed to do FLUSH_ALL in this case
+		 * because we could deadlock with so many things in the flushing
+		 * code, so we have to try and hold some extra space to
+		 * compensate for our delayed ref generation.  If we can't get
+		 * that space then we need see if we can steal our minimum from
+		 * the global reserve.  We will be ratelimited by the amount of
+		 * space we have for the delayed refs rsv, so we'll end up
+		 * committing and trying again.
+		 */
 		trans = btrfs_join_transaction(root);
-		if (IS_ERR(trans) || !ret)
+		if (IS_ERR(trans) || !ret) {
+			if (!IS_ERR(trans)) {
+				trans->block_rsv = &fs_info->trans_block_rsv;
+				trans->bytes_reserved = delayed_refs_extra;
+				btrfs_block_rsv_migrate(rsv, trans->block_rsv,
+							delayed_refs_extra, 1);
+			}
 			return trans;
+		}
 
 		/*
 		 * Try to steal from the global reserve if there is space for
-- 
2.14.3

[PATCH 42/42] btrfs: don't run delayed_iputs in commit

[Josef Bacik]

This could result in a really bad case where we do something like

evict
  evict_refill_and_join
    btrfs_commit_transaction
      btrfs_run_delayed_iputs
        evict
          evict_refill_and_join
            btrfs_commit_transaction
... forever

We have plenty of other places where we run delayed iputs that are much
safer, let those do the work.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/transaction.c | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 9168efaca37e..c91dc36fccae 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -2265,15 +2265,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 
 	kmem_cache_free(btrfs_trans_handle_cachep, trans);
 
-	/*
-	 * If fs has been frozen, we can not handle delayed iputs, otherwise
-	 * it'll result in deadlock about SB_FREEZE_FS.
-	 */
-	if (current != fs_info->transaction_kthread &&
-	    current != fs_info->cleaner_kthread &&
-	    !test_bit(BTRFS_FS_FROZEN, &fs_info->flags))
-		btrfs_run_delayed_iputs(fs_info);
-
 	return ret;
 
 scrub_continue:
-- 
2.14.3

Re: [PATCH 42/42] btrfs: don't run delayed_iputs in commit

[Filipe Manana]

On Fri, Oct 12, 2018 at 8:35 PM Josef Bacik <josef@toxicpanda.com> wrote:
>
> This could result in a really bad case where we do something like
>
> evict
>   evict_refill_and_join
>     btrfs_commit_transaction
>       btrfs_run_delayed_iputs
>         evict
>           evict_refill_and_join
>             btrfs_commit_transaction
> ... forever
>
> We have plenty of other places where we run delayed iputs that are much
> safer, let those do the work.
>
> Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>

> ---
>  fs/btrfs/transaction.c | 9 ---------
>  1 file changed, 9 deletions(-)
>
> diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
> index 9168efaca37e..c91dc36fccae 100644
> --- a/fs/btrfs/transaction.c
> +++ b/fs/btrfs/transaction.c
> @@ -2265,15 +2265,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
>
>         kmem_cache_free(btrfs_trans_handle_cachep, trans);
>
> -       /*
> -        * If fs has been frozen, we can not handle delayed iputs, otherwise
> -        * it'll result in deadlock about SB_FREEZE_FS.
> -        */
> -       if (current != fs_info->transaction_kthread &&
> -           current != fs_info->cleaner_kthread &&
> -           !test_bit(BTRFS_FS_FROZEN, &fs_info->flags))
> -               btrfs_run_delayed_iputs(fs_info);
> -
>         return ret;
>
>  scrub_continue:
> --
> 2.14.3
>


-- 
Filipe David Manana,

“Whether you think you can, or you think you can't — you're right.”

Re: [PATCH 42/42] btrfs: don't run delayed_iputs in commit

[David Sterba]

On Fri, Oct 12, 2018 at 03:32:56PM -0400, Josef Bacik wrote:
> This could result in a really bad case where we do something like
> 
> evict
>   evict_refill_and_join
>     btrfs_commit_transaction
>       btrfs_run_delayed_iputs
>         evict
>           evict_refill_and_join
>             btrfs_commit_transaction
> ... forever
> 
> We have plenty of other places where we run delayed iputs that are much
> safer, let those do the work.
> 
> Signed-off-by: Josef Bacik <josef@toxicpanda.com>

Reviewed-by: David Sterba <dsterba@suse.com>

[PATCH 00/42][v4] My current patch queue

[Josef Bacik]

v3->v4:
- added stacktraces to all the changelogs
- added the various reviewed-by's.
- fixed the loop in inode_rsv_refill to not use goto again;

v2->v3:
- reworked the truncate/evict throttling, we were still occasionally hitting
  enospc aborts in production in these paths because we were too aggressive with
  space usage.
- reworked the delayed iput stuff to be a little less racey and less deadlocky.
- Addressed the comments from Dave and Omar.
- A lot of production testing.

v1->v2:
- addressed all of the issues brought up.
- added more comments.
- split up some patches.

original message:

This is the current queue of things that I've been working on.  The main thing
these patches are doing is separating out the delayed refs reservations from the
global reserve into their own block rsv.  We have been consistently hitting
issues in production where we abort a transaction because we run out of the
global reserve either while running delayed refs or while updating dirty block
groups.  This is because the math around global reserves is made up bullshit
magic that has been tweaked more and more throughout the years.  The result is
something that is inconsistent across the board and sometimes wrong.  So instead
we need a way to know exactly how much space we need to keep around in order to
satisfy our outstanding delayed refs and our dirty block groups.

Since we don't know how many delayed refs we need at the start of any
modification we simply use the nr_items passed into btrfs_start_transaction() as
a guess for what we may need.  This has the side effect of putting more pressure
on the ENOSPC system, but it's pressure we can deal with more intelligently
because we always know how much space we have outstanding, instead of guessing
with weird global reserve math.

This works similar to every other reservation we have, we reserve the worst case
up front, and then at transaction end time we free up any space we didn't
actually use for delayed refs.

My performance tests show that we are bit faster now since we can do more
intelligent flushing and don't have to fall back on simply committing the
transaction in hopes that we have enough space for everything we need to do.

That leads me to the 2nd part of this pull, there's a bunch of fixes around
ENOSPC.  Because we are a bit faster now there were a bunch of things uncovered
in testing, but they seem to be all resolved now.

The final chunk of fixes are around transaction aborts.  There were a lot of
accounting bugs I was running into while running generic/435, so I fixed a bunch
of those up so now it runs cleanly.

I have been running these patches through xfstests on multiple machines for a
while, they are pretty solid and ready for wider testing and review.  Thanks,

Josef

[PATCH 01/42] btrfs: add btrfs_delete_ref_head helper

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

We do this dance in cleanup_ref_head and check_ref_cleanup, unify it
into a helper and cleanup the calling functions.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
---
 fs/btrfs/delayed-ref.c | 14 ++++++++++++++
 fs/btrfs/delayed-ref.h |  3 ++-
 fs/btrfs/extent-tree.c | 22 +++-------------------
 3 files changed, 19 insertions(+), 20 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index 62ff545ba1f7..3a9e4ac21794 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -393,6 +393,20 @@ btrfs_select_ref_head(struct btrfs_trans_handle *trans)
 	return head;
 }
 
+void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
+			   struct btrfs_delayed_ref_head *head)
+{
+	lockdep_assert_held(&delayed_refs->lock);
+	lockdep_assert_held(&head->lock);
+
+	rb_erase(&head->href_node, &delayed_refs->href_root);
+	RB_CLEAR_NODE(&head->href_node);
+	atomic_dec(&delayed_refs->num_entries);
+	delayed_refs->num_heads--;
+	if (head->processing == 0)
+		delayed_refs->num_heads_ready--;
+}
+
 /*
  * Helper to insert the ref_node to the tail or merge with tail.
  *
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
index d9f2a4ebd5db..7769177b489e 100644
--- a/fs/btrfs/delayed-ref.h
+++ b/fs/btrfs/delayed-ref.h
@@ -261,7 +261,8 @@ static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
 {
 	mutex_unlock(&head->mutex);
 }
-
+void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
+			   struct btrfs_delayed_ref_head *head);
 
 struct btrfs_delayed_ref_head *
 btrfs_select_ref_head(struct btrfs_trans_handle *trans);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index f77226d8020a..d24a0de4a2e7 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2492,12 +2492,9 @@ static int cleanup_ref_head(struct btrfs_trans_handle *trans,
 		spin_unlock(&delayed_refs->lock);
 		return 1;
 	}
-	delayed_refs->num_heads--;
-	rb_erase(&head->href_node, &delayed_refs->href_root);
-	RB_CLEAR_NODE(&head->href_node);
+	btrfs_delete_ref_head(delayed_refs, head);
 	spin_unlock(&head->lock);
 	spin_unlock(&delayed_refs->lock);
-	atomic_dec(&delayed_refs->num_entries);
 
 	trace_run_delayed_ref_head(fs_info, head, 0);
 
@@ -6984,22 +6981,9 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	if (!mutex_trylock(&head->mutex))
 		goto out;
 
-	/*
-	 * at this point we have a head with no other entries.  Go
-	 * ahead and process it.
-	 */
-	rb_erase(&head->href_node, &delayed_refs->href_root);
-	RB_CLEAR_NODE(&head->href_node);
-	atomic_dec(&delayed_refs->num_entries);
-
-	/*
-	 * we don't take a ref on the node because we're removing it from the
-	 * tree, so we just steal the ref the tree was holding.
-	 */
-	delayed_refs->num_heads--;
-	if (head->processing == 0)
-		delayed_refs->num_heads_ready--;
+	btrfs_delete_ref_head(delayed_refs, head);
 	head->processing = 0;
+
 	spin_unlock(&head->lock);
 	spin_unlock(&delayed_refs->lock);
 
-- 
2.14.3

[PATCH 00/42][v3] My current patch queue

[Josef Bacik]

v2->v3:
- reworked the truncate/evict throttling, we were still occasionally hitting
  enospc aborts in production in these paths because we were too aggressive with
  space usage.
- reworked the delayed iput stuff to be a little less racey and less deadlocky.
- Addressed the comments from Dave and Omar.
- A lot of production testing.

v1->v2:
- addressed all of the issues brought up.
- added more comments.
- split up some patches.

original message:

This is the current queue of things that I've been working on.  The main thing
these patches are doing is separating out the delayed refs reservations from the
global reserve into their own block rsv.  We have been consistently hitting
issues in production where we abort a transaction because we run out of the
global reserve either while running delayed refs or while updating dirty block
groups.  This is because the math around global reserves is made up bullshit
magic that has been tweaked more and more throughout the years.  The result is
something that is inconsistent across the board and sometimes wrong.  So instead
we need a way to know exactly how much space we need to keep around in order to
satisfy our outstanding delayed refs and our dirty block groups.

Since we don't know how many delayed refs we need at the start of any
modification we simply use the nr_items passed into btrfs_start_transaction() as
a guess for what we may need.  This has the side effect of putting more pressure
on the ENOSPC system, but it's pressure we can deal with more intelligently
because we always know how much space we have outstanding, instead of guessing
with weird global reserve math.

This works similar to every other reservation we have, we reserve the worst case
up front, and then at transaction end time we free up any space we didn't
actually use for delayed refs.

My performance tests show that we are bit faster now since we can do more
intelligent flushing and don't have to fall back on simply committing the
transaction in hopes that we have enough space for everything we need to do.

That leads me to the 2nd part of this pull, there's a bunch of fixes around
ENOSPC.  Because we are a bit faster now there were a bunch of things uncovered
in testing, but they seem to be all resolved now.

The final chunk of fixes are around transaction aborts.  There were a lot of
accounting bugs I was running into while running generic/435, so I fixed a bunch
of those up so now it runs cleanly.

I have been running these patches through xfstests on multiple machines for a
while, they are pretty solid and ready for wider testing and review.  Thanks,

Josef

[PATCH 01/42] btrfs: add btrfs_delete_ref_head helper

[Josef Bacik]

From: Josef Bacik <jbacik@fb.com>

We do this dance in cleanup_ref_head and check_ref_cleanup, unify it
into a helper and cleanup the calling functions.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
---
 fs/btrfs/delayed-ref.c | 14 ++++++++++++++
 fs/btrfs/delayed-ref.h |  3 ++-
 fs/btrfs/extent-tree.c | 22 +++-------------------
 3 files changed, 19 insertions(+), 20 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index 62ff545ba1f7..3a9e4ac21794 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -393,6 +393,20 @@ btrfs_select_ref_head(struct btrfs_trans_handle *trans)
 	return head;
 }
 
+void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
+			   struct btrfs_delayed_ref_head *head)
+{
+	lockdep_assert_held(&delayed_refs->lock);
+	lockdep_assert_held(&head->lock);
+
+	rb_erase(&head->href_node, &delayed_refs->href_root);
+	RB_CLEAR_NODE(&head->href_node);
+	atomic_dec(&delayed_refs->num_entries);
+	delayed_refs->num_heads--;
+	if (head->processing == 0)
+		delayed_refs->num_heads_ready--;
+}
+
 /*
  * Helper to insert the ref_node to the tail or merge with tail.
  *
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
index d9f2a4ebd5db..7769177b489e 100644
--- a/fs/btrfs/delayed-ref.h
+++ b/fs/btrfs/delayed-ref.h
@@ -261,7 +261,8 @@ static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
 {
 	mutex_unlock(&head->mutex);
 }
-
+void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
+			   struct btrfs_delayed_ref_head *head);
 
 struct btrfs_delayed_ref_head *
 btrfs_select_ref_head(struct btrfs_trans_handle *trans);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index f77226d8020a..d24a0de4a2e7 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2492,12 +2492,9 @@ static int cleanup_ref_head(struct btrfs_trans_handle *trans,
 		spin_unlock(&delayed_refs->lock);
 		return 1;
 	}
-	delayed_refs->num_heads--;
-	rb_erase(&head->href_node, &delayed_refs->href_root);
-	RB_CLEAR_NODE(&head->href_node);
+	btrfs_delete_ref_head(delayed_refs, head);
 	spin_unlock(&head->lock);
 	spin_unlock(&delayed_refs->lock);
-	atomic_dec(&delayed_refs->num_entries);
 
 	trace_run_delayed_ref_head(fs_info, head, 0);
 
@@ -6984,22 +6981,9 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	if (!mutex_trylock(&head->mutex))
 		goto out;
 
-	/*
-	 * at this point we have a head with no other entries.  Go
-	 * ahead and process it.
-	 */
-	rb_erase(&head->href_node, &delayed_refs->href_root);
-	RB_CLEAR_NODE(&head->href_node);
-	atomic_dec(&delayed_refs->num_entries);
-
-	/*
-	 * we don't take a ref on the node because we're removing it from the
-	 * tree, so we just steal the ref the tree was holding.
-	 */
-	delayed_refs->num_heads--;
-	if (head->processing == 0)
-		delayed_refs->num_heads_ready--;
+	btrfs_delete_ref_head(delayed_refs, head);
 	head->processing = 0;
+
 	spin_unlock(&head->lock);
 	spin_unlock(&delayed_refs->lock);
 
-- 
2.14.3