Skip to contentSkip to conversation

btrfs: remove unnecessary EXTENT_UPTODATE state in buffered I/O path

PATCH In review ethanlien 3 participants 3 messages Aug 19, 2022 – Aug 22, 2022 Btrfs file system

lore
Share & tools
lore

ethanlien

author
#0 lore raw

PATCH v2 1/1 btrfs: remove unnecessary EXTENT_UPTODATE state in buffered I/O path

From: Ethan Lien <[email protected]>

After we copied data to page cache in buffered I/O, we
1. Insert a EXTENT_UPTODATE state into inode's io_tree, by endio_readpage_release_extent(), set_extent_delalloc() or set_extent_defrag().
2. Set page uptodate before we unlock the page.

But the only place we check io_tree's EXTENT_UPTODATE state is in btrfs_do_readpage(). We know we enter btrfs_do_readpage() only when we have a non-uptodate page, so it is unnecessary to set EXTENT_UPTODATE.

For example, when performing a buffered random read:

	fio --rw=randread --ioengine=libaio --direct=0 --numjobs=4 \
		--filesize=32G --size=4G --bs=4k --name=job \
		--filename=/mnt/file --name=job

Then check how many extent_state in io_tree:

cat /proc/slabinfo | grep btrfs_extent_state | awk '{print $2}'

w/o this patch, we got 640567 btrfs_extent_state.
w/ this patch, we got 204 btrfs_extent_state.

Maintaining such a big tree brings overhead since every I/O needs to insert EXTENT_LOCKED, insert EXTENT_UPTODATE, then remove EXTENT_LOCKED. And in every insert or remove, we need to lock io_tree, do tree search, alloc or dealloc extent states. By removing unnecessary EXTENT_UPTODATE, we keep io_tree in a minimal size and reduce overhead when performing buffered I/O.

---

V2: Remove set_extent_uptodate() from btrfs_get_extent(), and when we found
a inline extent, set page uptodate in btrfs_do_readpage().

 fs/btrfs/extent-io-tree.h |  4 ++--
 fs/btrfs/extent_io.c      | 16 +---------------
 fs/btrfs/inode.c          |  2 --
 3 files changed, 3 insertions(+), 19 deletions(-)
fs/btrfs/extent-io-tree.h +2 -2
index c3eb52dbe61c..53ae849d0248 100644
@@ -211,7 +211,7 @@ static inline int set_extent_delalloc(struct extent_io_tree *tree, u64 start,
 				      struct extent_state **cached_state)
 {
 	return set_extent_bit(tree, start, end,
-			      EXTENT_DELALLOC | EXTENT_UPTODATE | extra_bits,
+			      EXTENT_DELALLOC | extra_bits,
 			      0, NULL, cached_state, GFP_NOFS, NULL);
 }
 
@@ -219,7 +219,7 @@ static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start,
 		u64 end, struct extent_state **cached_state)
 {
 	return set_extent_bit(tree, start, end,
-			      EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
+			      EXTENT_DELALLOC | EXTENT_DEFRAG,
 			      0, NULL, cached_state, GFP_NOFS, NULL);
 }
 
fs/btrfs/extent_io.c +1 -15
index bfae67c593c5..7e082770a088 100644
@@ -2924,9 +2924,6 @@ static void endio_readpage_release_extent(struct processed_extent *processed,
 	 * Now we don't have range contiguous to the processed range, release
 	 * the processed range now.
 	 */
-	if (processed->uptodate && tree->track_uptodate)
-		set_extent_uptodate(tree, processed->start, processed->end,
-				    &cached, GFP_ATOMIC);
 	unlock_extent_cached_atomic(tree, processed->start, processed->end,
 				    &cached);
 
@@ -3718,20 +3715,9 @@ static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
 			continue;
 		}
 		/* the get_extent function already copied into the page */
-		if (test_range_bit(tree, cur, cur_end,
-				   EXTENT_UPTODATE, 1, NULL)) {
-			unlock_extent(tree, cur, cur + iosize - 1);
-			end_page_read(page, true, cur, iosize);
-			cur = cur + iosize;
-			pg_offset += iosize;
-			continue;
-		}
-		/* we have an inline extent but it didn't get marked up
-		 * to date.  Error out
-		 */
 		if (block_start == EXTENT_MAP_INLINE) {
 			unlock_extent(tree, cur, cur + iosize - 1);
-			end_page_read(page, false, cur, iosize);
+			end_page_read(page, true, cur, iosize);
 			cur = cur + iosize;
 			pg_offset += iosize;
 			continue;
fs/btrfs/inode.c +0 -2
index f0c97d25b4a0..639edbcce9fe 100644
@@ -7031,8 +7031,6 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
 			}
 			flush_dcache_page(page);
 		}
-		set_extent_uptodate(io_tree, em->start,
-				    extent_map_end(em) - 1, NULL, GFP_NOFS);
 		goto insert;
 	}
 not_found:
Signature · 1 line
2.17.1
Reviewed-by on v2 1/1
#1 lore raw
Quoted 31 lines · ethanlien

From: Ethan Lien <[email protected]>

After we copied data to page cache in buffered I/O, we
1. Insert a EXTENT_UPTODATE state into inode's io_tree, by endio_readpage_release_extent(), set_extent_delalloc() or set_extent_defrag().
2. Set page uptodate before we unlock the page.

But the only place we check io_tree's EXTENT_UPTODATE state is in btrfs_do_readpage(). We know we enter btrfs_do_readpage() only when we have a non-uptodate page, so it is unnecessary to set EXTENT_UPTODATE.

For example, when performing a buffered random read:

	fio --rw=randread --ioengine=libaio --direct=0 --numjobs=4 \
		--filesize=32G --size=4G --bs=4k --name=job \
		--filename=/mnt/file --name=job

Then check how many extent_state in io_tree:

cat /proc/slabinfo | grep btrfs_extent_state | awk '{print $2}'

w/o this patch, we got 640567 btrfs_extent_state.
w/ this patch, we got 204 btrfs_extent_state.

Maintaining such a big tree brings overhead since every I/O needs to insert EXTENT_LOCKED, insert EXTENT_UPTODATE, then remove EXTENT_LOCKED. And in every insert or remove, we need to lock io_tree, do tree search, alloc or dealloc extent states. By removing unnecessary EXTENT_UPTODATE, we keep io_tree in a minimal size and reduce overhead when performing buffered I/O.

Quoted 6 lines

---

V2: Remove set_extent_uptodate() from btrfs_get_extent(), and when we found
a inline extent, set page uptodate in btrfs_do_readpage().

Yep, that was what I had in mind.
Thanks for trying it out. I've given it some testing, including a full run of fstests just to confirm I hadn't missed something.

Looks good, so:

Thanks.

Quoted 80 lines
 fs/btrfs/extent-io-tree.h |  4 ++--
 fs/btrfs/extent_io.c      | 16 +---------------
 fs/btrfs/inode.c          |  2 --
 3 files changed, 3 insertions(+), 19 deletions(-)
fs/btrfs/extent-io-tree.h +2 -2
index c3eb52dbe61c..53ae849d0248 100644
@@ -211,7 +211,7 @@ static inline int set_extent_delalloc(struct extent_io_tree *tree, u64 start,
 				      struct extent_state **cached_state)
 {
 	return set_extent_bit(tree, start, end,
-			      EXTENT_DELALLOC | EXTENT_UPTODATE | extra_bits,
+			      EXTENT_DELALLOC | extra_bits,
 			      0, NULL, cached_state, GFP_NOFS, NULL);
 }
 
@@ -219,7 +219,7 @@ static inline int set_extent_defrag(struct extent_io_tree *tree, u64 start,
 		u64 end, struct extent_state **cached_state)
 {
 	return set_extent_bit(tree, start, end,
-			      EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
+			      EXTENT_DELALLOC | EXTENT_DEFRAG,
 			      0, NULL, cached_state, GFP_NOFS, NULL);
 }
 
fs/btrfs/extent_io.c +1 -15
index bfae67c593c5..7e082770a088 100644
@@ -2924,9 +2924,6 @@ static void endio_readpage_release_extent(struct processed_extent *processed,
 	 * Now we don't have range contiguous to the processed range, release
 	 * the processed range now.
 	 */
-	if (processed->uptodate && tree->track_uptodate)
-		set_extent_uptodate(tree, processed->start, processed->end,
-				    &cached, GFP_ATOMIC);
 	unlock_extent_cached_atomic(tree, processed->start, processed->end,
 				    &cached);
 
@@ -3718,20 +3715,9 @@ static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
 			continue;
 		}
 		/* the get_extent function already copied into the page */
-		if (test_range_bit(tree, cur, cur_end,
-				   EXTENT_UPTODATE, 1, NULL)) {
-			unlock_extent(tree, cur, cur + iosize - 1);
-			end_page_read(page, true, cur, iosize);
-			cur = cur + iosize;
-			pg_offset += iosize;
-			continue;
-		}
-		/* we have an inline extent but it didn't get marked up
-		 * to date.  Error out
-		 */
 		if (block_start == EXTENT_MAP_INLINE) {
 			unlock_extent(tree, cur, cur + iosize - 1);
-			end_page_read(page, false, cur, iosize);
+			end_page_read(page, true, cur, iosize);
 			cur = cur + iosize;
 			pg_offset += iosize;
 			continue;
fs/btrfs/inode.c +0 -2
index f0c97d25b4a0..639edbcce9fe 100644
@@ -7031,8 +7031,6 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
 			}
 			flush_dcache_page(page);
 		}
-		set_extent_uptodate(io_tree, em->start,
-				    extent_map_end(em) - 1, NULL, GFP_NOFS);
 		goto insert;
 	}
 not_found:
Signature · 1 line
2.17.1
#2 lore raw
In reply to ethanlien
Quoted 31 lines · ethanlien

From: Ethan Lien <[email protected]>

After we copied data to page cache in buffered I/O, we
1. Insert a EXTENT_UPTODATE state into inode's io_tree, by endio_readpage_release_extent(), set_extent_delalloc() or set_extent_defrag().
2. Set page uptodate before we unlock the page.

But the only place we check io_tree's EXTENT_UPTODATE state is in btrfs_do_readpage(). We know we enter btrfs_do_readpage() only when we have a non-uptodate page, so it is unnecessary to set EXTENT_UPTODATE.

For example, when performing a buffered random read:

	fio --rw=randread --ioengine=libaio --direct=0 --numjobs=4 \
		--filesize=32G --size=4G --bs=4k --name=job \
		--filename=/mnt/file --name=job

Then check how many extent_state in io_tree:

cat /proc/slabinfo | grep btrfs_extent_state | awk '{print $2}'

w/o this patch, we got 640567 btrfs_extent_state.
w/ this patch, we got 204 btrfs_extent_state.

Maintaining such a big tree brings overhead since every I/O needs to insert EXTENT_LOCKED, insert EXTENT_UPTODATE, then remove EXTENT_LOCKED. And in every insert or remove, we need to lock io_tree, do tree search, alloc or dealloc extent states. By removing unnecessary EXTENT_UPTODATE, we keep io_tree in a minimal size and reduce overhead when performing buffered I/O.

Quoted 6 lines

---

V2: Remove set_extent_uptodate() from btrfs_get_extent(), and when we found
a inline extent, set page uptodate in btrfs_do_readpage().

Added to misc-next, thanks.