[pve-devel] [PATCH storage] add workaround for zfs rollback bug

[pve-devel] [PATCH storage] add workaround for zfs rollback bug

[Dominik Csapak]

as described in the zfs bug https://github.com/openzfs/zfs/issues/10931
the kernel keeps around cached data from mmaps after a rollback, thus
having invalid data in files that were allegedly rolled back

to workaround this (until a real fix comes along), we unmount the subvol,
invalidating the kernel cache anyway

the dataset gets mounted on the next 'activate_volume' again

Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
---
 PVE/Storage/ZFSPoolPlugin.pm | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/PVE/Storage/ZFSPoolPlugin.pm b/PVE/Storage/ZFSPoolPlugin.pm
index 3054331..105d802 100644
--- a/PVE/Storage/ZFSPoolPlugin.pm
+++ b/PVE/Storage/ZFSPoolPlugin.pm
@@ -467,9 +467,18 @@ sub volume_snapshot_delete {
 sub volume_snapshot_rollback {
     my ($class, $scfg, $storeid, $volname, $snap) = @_;
 
-    my $vname = ($class->parse_volname($volname))[1];
+    my (undef, $vname, undef, undef, undef, undef, $format) = $class->parse_volname($volname);
+
+    my $msg = $class->zfs_request($scfg, undef, 'rollback', "$scfg->{pool}/$vname\@$snap");
+
+    # we have to unmount rollbacked subvols, to invalidate wrong kernel
+    # caches, they get mounted in activate volume again
+    # see zfs bug #10931 https://github.com/openzfs/zfs/issues/10931
+    if ($format eq 'subvol') {
+	$class->zfs_request($scfg, undef, 'unmount', "$scfg->{pool}/$vname");
+    }
 
-    $class->zfs_request($scfg, undef, 'rollback', "$scfg->{pool}/$vname\@$snap");
+    return $msg;
 }
 
 sub volume_rollback_is_possible {
-- 
2.20.1

[pve-devel] applied: [PATCH storage] add workaround for zfs rollback bug

[Thomas Lamprecht]

On 21.01.21 16:47, Dominik Csapak wrote:
> as described in the zfs bug https://github.com/openzfs/zfs/issues/10931
> the kernel keeps around cached data from mmaps after a rollback, thus
> having invalid data in files that were allegedly rolled back
> 
> to workaround this (until a real fix comes along), we unmount the subvol,
> invalidating the kernel cache anyway
> 
> the dataset gets mounted on the next 'activate_volume' again
> 
> Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
> ---
>  PVE/Storage/ZFSPoolPlugin.pm | 13 +++++++++++--
>  1 file changed, 11 insertions(+), 2 deletions(-)
> 
>

applied, thanks!