[PATCH 1/2] Btrfs: do not try and lookup the file extent when finishing ordered io

[PATCH 1/2] Btrfs: do not try and lookup the file extent when finishing ordered io

[Josef Bacik]

When running the following fio job

[torrent]
filename=torrent-test
rw=randwrite
size=4g
filesize=4g
bs=4k
ioengine=sync

you would see long stalls where no work was being done.  That is because we were
doing all this extra work to read in the file extent outside of the transaction,
however in the random io case this ends up hurting us because the file extents
are not there to begin with.  So axe this logic, since we end up reading in the
file extent when we go to update it anyway.  This took the fio job from 11 mb/s
with several ~10 second stalls to 24 mb/s to a couple of 1-2 second stalls.

Signed-off-by: Josef Bacik <josef@redhat.com>
---
 fs/btrfs/inode.c |   46 ++--------------------------------------------
 1 files changed, 2 insertions(+), 44 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index e8be185..9a1c746 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -1679,24 +1679,6 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
  * before we start the transaction.  It limits the amount of btree
  * reads required while inside the transaction.
  */
-static noinline void reada_csum(struct btrfs_root *root,
-				struct btrfs_path *path,
-				struct btrfs_ordered_extent *ordered_extent)
-{
-	struct btrfs_ordered_sum *sum;
-	u64 bytenr;
-
-	sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
-			 list);
-	bytenr = sum->sums[0].bytenr;
-
-	/*
-	 * we don't care about the results, the point of this search is
-	 * just to get the btree leaves into ram
-	 */
-	btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
-}
-
 /* as ordered data IO finishes, this gets called so we can finish
  * an ordered extent if the range of bytes in the file it covers are
  * fully written.
@@ -1707,7 +1689,6 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 	struct btrfs_trans_handle *trans;
 	struct btrfs_ordered_extent *ordered_extent = NULL;
 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
-	struct btrfs_path *path;
 	int compressed = 0;
 	int ret;
 
@@ -1715,32 +1696,9 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 	if (!ret)
 		return 0;
 
-	/*
-	 * before we join the transaction, try to do some of our IO.
-	 * This will limit the amount of IO that we have to do with
-	 * the transaction running.  We're unlikely to need to do any
-	 * IO if the file extents are new, the disk_i_size checks
-	 * covers the most common case.
-	 */
-	if (start < BTRFS_I(inode)->disk_i_size) {
-		path = btrfs_alloc_path();
-		if (path) {
-			ret = btrfs_lookup_file_extent(NULL, root, path,
-						       inode->i_ino,
-						       start, 0);
-			ordered_extent = btrfs_lookup_ordered_extent(inode,
-								     start);
-			if (!list_empty(&ordered_extent->list)) {
-				btrfs_release_path(root, path);
-				reada_csum(root, path, ordered_extent);
-			}
-			btrfs_free_path(path);
-		}
-	}
-
-	if (!ordered_extent)
-		ordered_extent = btrfs_lookup_ordered_extent(inode, start);
+	ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 	BUG_ON(!ordered_extent);
+
 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
 		BUG_ON(!list_empty(&ordered_extent->list));
 		ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
-- 
1.6.6

[PATCH 2/2] Btrfs: cache ordered extent when completing io

[Josef Bacik]

When finishing io we run btrfs_dec_test_ordered_pending, and then immediately
run btrfs_lookup_ordered_extent, but btrfs_dec_test_ordered_pending does that
already, so we're searching twice when we don't have to.  This patch lets us
pass a btrfs_ordered_extent in to btrfs_dec_test_ordered_pending so if we do
complete io on that ordered extent we can just use the one we found then instead
of having to do another btrfs_lookup_ordered_extent.  This made my fio job with
the other patch go from 24 mb/s to 29 mb/s.

Signed-off-by: Josef Bacik <josef@redhat.com>
---
 fs/btrfs/inode.c        |    5 ++---
 fs/btrfs/ordered-data.c |    7 ++++++-
 fs/btrfs/ordered-data.h |    3 ++-
 3 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 9a1c746..26574d8 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -1692,11 +1692,10 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 	int compressed = 0;
 	int ret;
 
-	ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
+	ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
+					     end - start + 1);
 	if (!ret)
 		return 0;
-
-	ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 	BUG_ON(!ordered_extent);
 
 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 5c2a9e7..c44054b 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -232,11 +232,12 @@ int btrfs_add_ordered_sum(struct inode *inode,
  * to make sure this function only returns 1 once for a given ordered extent.
  */
 int btrfs_dec_test_ordered_pending(struct inode *inode,
+				   struct btrfs_ordered_extent **cached,
 				   u64 file_offset, u64 io_size)
 {
 	struct btrfs_ordered_inode_tree *tree;
 	struct rb_node *node;
-	struct btrfs_ordered_extent *entry;
+	struct btrfs_ordered_extent *entry = NULL;
 	int ret;
 
 	tree = &BTRFS_I(inode)->ordered_tree;
@@ -264,6 +265,10 @@ int btrfs_dec_test_ordered_pending(struct inode *inode,
 	else
 		ret = 1;
 out:
+	if (!ret && cached && entry) {
+		*cached = entry;
+		atomic_inc(&entry->refs);
+	}
 	mutex_unlock(&tree->mutex);
 	return ret == 0;
 }
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index 1fe1282..398bfc6 100644
--- a/fs/btrfs/ordered-data.h
+++ b/fs/btrfs/ordered-data.h
@@ -137,7 +137,8 @@ int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry);
 int btrfs_remove_ordered_extent(struct inode *inode,
 				struct btrfs_ordered_extent *entry);
 int btrfs_dec_test_ordered_pending(struct inode *inode,
-				       u64 file_offset, u64 io_size);
+				   struct btrfs_ordered_extent **cached,
+				   u64 file_offset, u64 io_size);
 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 			     u64 start, u64 len, u64 disk_len, int tyep);
 int btrfs_add_ordered_sum(struct inode *inode,
-- 
1.6.6