Viewing: ext4-optimize-find_delayed_extent.patch
commit 3dd73b5c5d61a219c702873711055cb1cc80394a
Author: Andrew Perepechko <andrew.perepechko@hpe.com>
AuthorDate: Tue Nov 1 19:26:54 2022 +0300
LU-16286 ldiskfs: reimplement nodelalloc optimization
fiemap calls perform costly delayed extent search affecting
BRW performance, however, in Lustre we don't use delayed
allocation at all. Let's skip this search completely as we did
in RHEL7.
Signed-off-by: Andrew Perepechko <andrew.perepechko@hpe.com>
HPE-bug-id: LUS-11161
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Alexander Boyko <alexander.boyko@hpe.com>
Change-Id: I2c3562cf5cbdf3c5532e4b79b28a040a995322b7
Reviewed-on: https://review.whamcloud.com/49007
--- linux-stage.orig/fs/ext4/extents.c 2022-09-05 09:04:31.628122705 -0600
+++ linux-stage/fs/ext4/extents.c 2022-09-06 05:32:57.083369853 -0600
@@ -5064,6 +5064,13 @@ static int ext4_find_delayed_extent(s
struct extent_status es;
ext4_lblk_t block, next_del;
+ if (!test_opt(inode->i_sb, DELALLOC)) {
+ if (newes->es_pblk == 0)
+ return 0;
+
+ return EXT_MAX_BLOCKS;
+ }
+
if (newes->es_pblk == 0) {
ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
newes->es_lblk,
