Skip to contentSkip to conversation

generic/604: Fix for overlayfs

PATCH In review Amir Goldstein 2 participants 3 messages Jun 18, 2023 – Jun 20, 2023

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 generic/604: Fix for overlayfs

Since v6.3, I noticed that generic/604 does not run on overlayfs because:

generic/604 -- upper fs needs to support d_type

This is odd because the base fs I am using (xfs) does support d_type.

The reason is that for overlayfs, this sequence run by this test:

_scratch_unmount &
_scratch_mount

Translates to:

umount $OVL_MNT; umount $BASE_MNT &
mount $BASE_MNT ...; mount $OVL_MNT ...

Which can end up reordred as:

umount $OVL_MNT;
mount $BASE_MNT ...
umount $BASE_MNT &
mount $OVL_MNT ...

and overlayfs is trying to use a non-existing upper fs.

Use UMOUNT_PROG directly instead of the _scratch_unmount helper, to avoid unmounting the base fs.

Incidently, the only thing that has changed in overlayfs in v6.3 is idmapped mounts support and the test in question was run without idmapped mounts enabled, so the cahnge in behavior must be related to some subtle timing change.

---
 tests/generic/604 | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
tests/generic/604 +3 -1
index 9c53fd57..cc6a4b21 100755
@@ -24,7 +24,9 @@ _scratch_mount
 for i in $(seq 0 500); do
 	$XFS_IO_PROG -f -c "pwrite 0 4K" $SCRATCH_MNT/$i >/dev/null
 done
-_scratch_unmount &
+# For overlayfs, avoid unmouting the base fs after _scratch_mount
+# tries to mount the base fs
+$UMOUNT_PROG $SCRATCH_MNT &
 _scratch_mount
 wait
 
Signature · 1 line
2.34.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 26 lines · Amir Goldstein

Since v6.3, I noticed that generic/604 does not run on overlayfs because:

generic/604 -- upper fs needs to support d_type

This is odd because the base fs I am using (xfs) does support d_type.

The reason is that for overlayfs, this sequence run by this test:

_scratch_unmount &
_scratch_mount

Translates to:

umount $OVL_MNT; umount $BASE_MNT &
mount $BASE_MNT ...; mount $OVL_MNT ...

Which can end up reordred as:

umount $OVL_MNT;
mount $BASE_MNT ...
umount $BASE_MNT &
mount $OVL_MNT ...

and overlayfs is trying to use a non-existing upper fs.

Quoted 7 lines

Use UMOUNT_PROG directly instead of the _scratch_unmount helper, to avoid unmounting the base fs.

Incidently, the only thing that has changed in overlayfs in v6.3 is idmapped mounts support and the test in question was run without idmapped mounts enabled, so the cahnge in behavior must be related to some subtle timing change.

I implemented testing overlayfs on top of idmapped mounts in xfstests and changed a lot of the overlayfs test infrastructure. So I wouldn't be surprised if that's the reason.

Quoted 3 lines

---

Looks good to me,

#2 lore raw
Quoted 31 lines · Christian Brauner
Quoted 29 lines · Amir Goldstein

Since v6.3, I noticed that generic/604 does not run on overlayfs because:

generic/604 -- upper fs needs to support d_type

This is odd because the base fs I am using (xfs) does support d_type.

The reason is that for overlayfs, this sequence run by this test:

_scratch_unmount &
_scratch_mount

Translates to:

umount $OVL_MNT; umount $BASE_MNT &
mount $BASE_MNT ...; mount $OVL_MNT ...

Which can end up reordred as:

umount $OVL_MNT;
mount $BASE_MNT ...
umount $BASE_MNT &
mount $OVL_MNT ...

and overlayfs is trying to use a non-existing upper fs.

Use UMOUNT_PROG directly instead of the _scratch_unmount helper, to avoid unmounting the base fs.

Quoted 8 lines
Quoted 4 lines

Incidently, the only thing that has changed in overlayfs in v6.3 is idmapped mounts support and the test in question was run without idmapped mounts enabled, so the cahnge in behavior must be related to some subtle timing change.

I implemented testing overlayfs on top of idmapped mounts in xfstests and changed a lot of the overlayfs test infrastructure. So I wouldn't be surprised if that's the reason.

Could be, but to clarify, I was comparing kernel v6.2 to v6.3 with same xfstests and in both cases I did not set IDMAPPED_MOUNTS, so _idmapped_mount() should have been a noop.

Thanks,
Amir.