Re: [PATCH v3 2/3] common/rc: Add a new _require_scratch_extsize helper function

[Nirjhar Roy <nirjhar@linux.ibm.com>]

On 11/22/24 21:34, Darrick J. Wong wrote:
> On Fri, Nov 22, 2024 at 12:22:41AM +0530, Ritesh Harjani wrote:
>> Nirjhar Roy <nirjhar@linux.ibm.com> writes:
>>
>>> On 11/21/24 13:23, Ritesh Harjani (IBM) wrote:
>>>> Nirjhar Roy <nirjhar@linux.ibm.com> writes:
>>>>
>>>>> _require_scratch_extsize helper function will be used in the
>>>>> the next patch to make the test run only on filesystems with
>>>>> extsize support.
>>>>>
>>>>> Reviewed-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
>>>>> Signed-off-by: Nirjhar Roy <nirjhar@linux.ibm.com>
>>>>> ---
>>>>>    common/rc | 17 +++++++++++++++++
>>>>>    1 file changed, 17 insertions(+)
>>>>>
>>>>> diff --git a/common/rc b/common/rc
>>>>> index cccc98f5..995979e9 100644
>>>>> --- a/common/rc
>>>>> +++ b/common/rc
>>>>> @@ -48,6 +48,23 @@ _test_fsxattr_xflag()
>>>>>    	grep -q "fsxattr.xflags.*\[.*$2.*\]" <($XFS_IO_PROG -c "stat -v" "$1")
>>>>>    }
>>>>>    
>>>>> +# This test requires extsize support on the  filesystem
>>>>> +_require_scratch_extsize()
>>>>> +{
>>>>> +	_require_scratch
>>>> _require_xfs_io_command "extsize"
>>>>
>>>> ^^^ Don't we need this too?
>>> Yes, good point. I will add this in the next revision.
>>>>> +	_scratch_mkfs > /dev/null
>>>>> +	_scratch_mount
>>>>> +	local filename=$SCRATCH_MNT/$RANDOM
>>>>> +	local blksz=$(_get_block_size $SCRATCH_MNT)
>>>>> +	local extsz=$(( blksz*2 ))
>>>>> +	local res=$($XFS_IO_PROG -c "open -f $filename" -c "extsize $extsz" \
>>>>> +		-c "extsize")
>>>>> +	_scratch_unmount
>>>>> +	grep -q "\[$extsz\] $filename" <(echo $res) || \
>>>>> +		_notrun "this test requires extsize support on the filesystem"
>>>> Why grep when we can simply just check the return value of previous xfs_io command?
>>> No, I don't think we can rely on the return value of xfs_io. For ex,
>>> let's look at the following set of commands which are ran on an ext4 system:
>>>
>>> root@AMARPC: /mnt1/test$ xfs_io -V
>>> xfs_io version 5.13.0
>>> root@AMARPC: /mnt1/test$ touch new
>>> root@AMARPC: /mnt1/test$ xfs_io -c "extsize 8k"  new
>>> foreign file active, extsize command is for XFS filesystems only
>>> root@AMARPC: /mnt1/test$ echo "$?"
>>> 0
>>> This incorrect return value might have been fixed in some later versions
>>> of xfs_io but there are still versions where we can't solely rely on the
>>> return value.
>> Ok. That's bad, we then have to rely on grep.
>> Sure, thanks for checking and confirming that.
> You all should add CMD_FOREIGN_OK to the extsize command in xfs_io,
> assuming that you've not already done that in your dev workspace.
>
> --D

Yes, I have tested with that as well. I have applied the following patch 
to xfsprogs and tested with the modified xfs_io.

diff --git a/io/open.c b/io/open.c
index 15850b55..6407b7e8 100644
--- a/io/open.c
+++ b/io/open.c
@@ -980,7 +980,7 @@ open_init(void)
         extsize_cmd.args = _("[-D | -R] [extsize]");
         extsize_cmd.argmin = 0;
         extsize_cmd.argmax = -1;
-       extsize_cmd.flags = CMD_NOMAP_OK;
+       extsize_cmd.flags = CMD_NOMAP_OK | CMD_FOREIGN_OK;
         extsize_cmd.oneline =
                 _("get/set preferred extent size (in bytes) for the 
open file");
         extsize_cmd.help = extsize_help;

The return values are similar.

root@AMARPC: /mnt1/scratch$ touch new
root@AMARPC: /mnt1/scratch$ /home/ubuntu/xfsprogs-dev/io/xfs_io -c 
"extsize 8k" new
root@AMARPC: /mnt1/scratch$ echo "$?"
0
root@AMARPC: /mnt1/scratch$ /home/ubuntu/xfsprogs-dev/io/xfs_io -c 
"extsize" new
[0] new

This is the reason I am not relying on the return value, instead I am 
checking if only the extsize gets changed, we will assume that the 
extsize support is there, else the test will _notrun.

Also,

root@AMARPC: /mnt1/scratch$ strace -f /mnt1/scratch$ 
/home/ubuntu/xfsprogs-dev/io/xfs_io -c "extsize 16k" new

...

...

ioctl(3, FS_IOC_FSGETXATTR, {fsx_xflags=0, fsx_extsize=0, 
fsx_nextents=0, fsx_projid=0, fsx_cowextsize=0}) = 0
ioctl(3, FS_IOC_FSSETXATTR, {fsx_xflags=FS_XFLAG_EXTSIZE, 
fsx_extsize=16384, fsx_projid=0, fsx_cowextsize=0}) = 0
exit_group(0)

Looking at the existing code for ext4_fileattr_set(), We validate the 
flags but I think we silently don't validate(and ignore) the xflags. 
Like, we have

int err = -EOPNOTSUPP;
if (flags & ~EXT4_FL_USER_VISIBLE)
         goto out;

BUT we do NOT have something like

int err = -EOPNOTSUPP;
if (fa->fsx_flags & ~EXT4_VALID_XFLAGS) // where EXT4_VALID_XFLAGS 
should be an || of all the supported xflags in ext4.
         goto out;

I am not sure what other filesystems do, but if we check whether the 
extsize got changed, then we can correctly determine extsize support.

Does that make sense?

--NR



>
>> -ritesh
>>
-- 
---
Nirjhar Roy
Linux Kernel Developer
IBM, Bangalore