Re: [PATCH] common/btrfs: handle dmdust as mounted device in _btrfs_buffered_read_on_mirror()

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

On 2023/6/28 19:34, Zorro Lang wrote:
> On Tue, Jun 27, 2023 at 05:23:31AM +0800, Qu Wenruo wrote:
>>
>>
>> On 2023/6/27 01:32, Zorro Lang wrote:
>>> On Mon, Jun 26, 2023 at 02:00:52PM +0800, Qu Wenruo wrote:
>>>> [BUG]
>>>> After commit ab41f0bddb73 ("common/btrfs: use _scratch_cycle_mount to
>>>> ensure all page caches are dropped"), the test case btrfs/143 can fail
>>>> like below:
>>>>
>>>>    btrfs/143 6s ... [failed, exit status 1]- output mismatch (see ~/xfstests/results//btrfs/143.out.bad)
>>>>       --- tests/btrfs/143.out 2020-06-10 19:29:03.818519162 +0100
>>>>       +++ ~/xfstests/results//btrfs/143.out.bad 2023-06-19 17:04:00.575033899 +0100
>>>>       @@ -1,37 +1,6 @@
>>>>        QA output created by 143
>>>>        wrote 131072/131072 bytes
>>>>        XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
>>>>       -XXXXXXXX:  aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa
>>>> ................
>>>>       -XXXXXXXX:  aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa
>>>> ................
>>>>       -XXXXXXXX:  aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa
>>>> ................
>>>>       -XXXXXXXX:  aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa
>>>> ................
>>>>
>>>> [CAUSE]
>>>> Test case btrfs/143 uses dm-dust device to emulate read errors, this
>>>> means we can not use _scratch_cycle_mount to cycle mount $SCRATCH_MNT.
>>>>
>>>> As it would go mount $SCRATCH_DEV, not the dm-dust device to
>>>> $SCRATCH_MNT.
>>>> This prevents us to trigger read-repair (since no error would be hit)
>>>> thus fail the test.
>>>>
>>>> [FIX]
>>>> Since we can mount whatever device at $SCRATCH_MNT, we can not use
>>>> _scratch_cycle_mount in this case.
>>>>
>>>> Instead implement a small helper to grab the mounted device and its
>>>> mount options, and use the same device and mount options to cycle
>>>> $SCRATCH_MNT mount.
>>>>
>>>> This would fix btrfs/143 and hopefully future test cases which use dm
>>>> devices.
>>>>
>>>> Reported-by: Filipe Manana <fdmanana@suse.com>
>>>> Signed-off-by: Qu Wenruo <wqu@suse.com>
>>>> ---
>>>>    common/btrfs | 14 ++++++++++++--
>>>>    1 file changed, 12 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/common/btrfs b/common/btrfs
>>>> index 175b33ae..4a02b2cc 100644
>>>> --- a/common/btrfs
>>>> +++ b/common/btrfs
>>>> @@ -601,8 +601,18 @@ _btrfs_buffered_read_on_mirror()
>>>>    	# The drop_caches doesn't seem to drop every pages on aarch64 with
>>>>    	# 64K page size.
>>>>    	# So here as another workaround, cycle mount the SCRATCH_MNT to ensure
>>>> -	# the cache are dropped.
>>>> -	_scratch_cycle_mount
>>>> +	# the cache are dropped, but we can not use _scratch_cycle_mount, as
>>>> +	# we may mount whatever dm device at SCRATCH_MNT.
>>>> +	# So here we grab the mounted block device and its mount options, then
>>>> +	# unmount and re-mount with the same device and options.
>>>> +	local mount_info=$(_mount | grep "$SCRATCH_MNT")
>>>> +	if [ -z "$mount_info" ]; then
>>>> +		_fail "failed to grab mount info of $SCRATCH_MNT"
>>>> +	fi
>>>> +	local dev=$(echo $mount_info | $AWK_PROG '{print $1}')
>>>> +	local opts=$(echo $mount_info | $AWK_PROG '{print $6}' | sed 's/[()]//g')
>>>
>>> The `findmnt` can help to get $dev and $opts:
>>>
>>>     local dev=$(findmnt -n -T $SCRATCH_MNT -o SOURCE)
>>>     local opts=$(findmnt -n -T $SCRATCH_MNT -o OPTIONS)
>>>
>>> If you hope to check you can keep:
>>>
>>>     if [ -z "$dev" -o -z "$opts" ];then
>>>             _fail "failed to grab mount info of $SCRATCH_MNT"
>>>     fi
>>
>> That's really helpful!
>>
>>>
>>>> +	_scratch_unmount
>>>> +	_mount $dev -o $opts $SCRATCH_MNT
>>>
>>> I'm wondering can this help that, after you get the "real" device name:
>>>
>>>     SCRATCH_DEV=$dev _scratch_cycle_mount
>>
>> AFAIK we still need to specify the mount option.
>>
>> As it's possible previous mount is specifying certain mount option
>> that's not in MOUNT_OPTIONS environment variables.
>>
>> E.g. mounting a specific subvolume or a temporary mount option.
>>
>> Thus I believe we may still need to specific the mount options.
>
> Hmm... if the _scratch_cycle_mount doesn't support dmdust, others dmxxxx
> (e.g. dmdelay, dmthin, dmerror, dmflaky) have similar problem, right?

Yes, but my point here is, although "SCRATCH_DEV=$dev
_scratch_cycle_mount" can work for most cases, it can still miss the
specific mount option of the current mount.

Thus we still need to go "_mount $dev -o $opts $SCRATCH_MNT", just for
the extra mount options.

Thanks,
Qu
>
> Thanks,
> Zorro
>
>>
>> Thanks,
>> Qu
>>
>>>
>>> Thanks,
>>> Zorro
>>>
>>>>    	while [[ -z $( (( BASHPID % nr_mirrors == mirror )) &&
>>>>    		exec $XFS_IO_PROG \
>>>>    			-c "pread -b $size $offset $size" $file) ]]; do
>>>> --
>>>> 2.39.0
>>>>
>>>
>>
>