OSDN Git Service

btrfs: lift some btrfs_cross_ref_exist checks in nocow path
authorEthan Lien <ethanlien@synology.com>
Thu, 17 May 2018 06:58:29 +0000 (14:58 +0800)
committerDavid Sterba <dsterba@suse.com>
Wed, 30 May 2018 14:46:53 +0000 (16:46 +0200)
In nocow path, we check if the extent is snapshotted in
btrfs_cross_ref_exist(). We can do the similar check earlier and avoid
unnecessary search into extent tree.

A fio test on a Intel D-1531, 16GB RAM, SSD RAID-5 machine as follows:

[global]
group_reporting
time_based
thread=1
ioengine=libaio
bs=4k
iodepth=32
size=64G
runtime=180
numjobs=8
rw=randwrite

[file1]
filename=/mnt/nocow/testfile

IOPS result:   unpatched     patched

1 fio round:     46670        46958
snapshot
2 fio round:     51826        54498
3 fio round:     59767        61289

After snapshot, the first fio get about 5% performance gain. As we
continually write to the same file, all writes will resume to nocow mode
and eventually we have no performance gain.

Signed-off-by: Ethan Lien <ethanlien@synology.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update comments ]
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/inode.c

index 5d370bb..27293ff 100644 (file)
@@ -1370,6 +1370,13 @@ next_slot:
                            btrfs_file_extent_encryption(leaf, fi) ||
                            btrfs_file_extent_other_encoding(leaf, fi))
                                goto out_check;
+                       /*
+                        * Do the same check as in btrfs_cross_ref_exist but
+                        * without the unnecessary search.
+                        */
+                       if (btrfs_file_extent_generation(leaf, fi) <=
+                           btrfs_root_last_snapshot(&root->root_item))
+                               goto out_check;
                        if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
                                goto out_check;
                        if (btrfs_extent_readonly(fs_info, disk_bytenr))
@@ -7322,6 +7329,14 @@ noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
            btrfs_file_extent_other_encoding(leaf, fi))
                goto out;
 
+       /*
+        * Do the same check as in btrfs_cross_ref_exist but without the
+        * unnecessary search.
+        */
+       if (btrfs_file_extent_generation(leaf, fi) <=
+           btrfs_root_last_snapshot(&root->root_item))
+               goto out;
+
        backref_offset = btrfs_file_extent_offset(leaf, fi);
 
        if (orig_start) {