Re: [PATCH 6.12.y-5.10.y] selftests: mptcp: connect modes: re-add exec mode

[Matthieu Baerts <matttbe@kernel.org>]

Hi Greg,

On 04/11/2025 00:54, Greg KH wrote:
> On Mon, Nov 03, 2025 at 05:54:34PM +0100, Matthieu Baerts (NGI0) wrote:
>> It looks like the execution permissions (+x) got lost during the
>> backports of these new files.
>>
>> The issue is that some CIs don't execute these tests without that.
>>
>> Fixes: 37848a456fc3 ("selftests: mptcp: connect: also cover alt modes")
>> Fixes: fdf0f60a2bb0 ("selftests: mptcp: connect: also cover checksum")
>> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
>> ---
>> I'm not sure why they got lost, maybe Quilt doesn't support that? But
>> then, can this patch still be applied?
>> The same patch can be applied up to v5.10. In v5.10, only
>> mptcp_connect_mmap.sh file is present, but I can send a dedicated patch
>> for v5.10.
>> ---
>>  tools/testing/selftests/net/mptcp/mptcp_connect_checksum.sh | 0
>>  tools/testing/selftests/net/mptcp/mptcp_connect_mmap.sh     | 0
>>  tools/testing/selftests/net/mptcp/mptcp_connect_sendfile.sh | 0
>>  3 files changed, 0 insertions(+), 0 deletions(-)
>>  mode change 100644 => 100755 tools/testing/selftests/net/mptcp/mptcp_connect_checksum.sh
>>  mode change 100644 => 100755 tools/testing/selftests/net/mptcp/mptcp_connect_mmap.sh
>>  mode change 100644 => 100755 tools/testing/selftests/net/mptcp/mptcp_connect_sendfile.sh
>>
>> diff --git a/tools/testing/selftests/net/mptcp/mptcp_connect_checksum.sh b/tools/testing/selftests/net/mptcp/mptcp_connect_checksum.sh
>> old mode 100644
>> new mode 100755
>> diff --git a/tools/testing/selftests/net/mptcp/mptcp_connect_mmap.sh b/tools/testing/selftests/net/mptcp/mptcp_connect_mmap.sh
>> old mode 100644
>> new mode 100755
>> diff --git a/tools/testing/selftests/net/mptcp/mptcp_connect_sendfile.sh b/tools/testing/selftests/net/mptcp/mptcp_connect_sendfile.sh
>> old mode 100644
>> new mode 100755
>> -- 
>> 2.51.0
>>
>>
> 
> This is going to be a pain to apply, given that we use quilt, and that
> does not handle modes well, if at all.
> 
> So yes, that is why these files are not marked executable, but I thought
> we were moving away from that anyway, most scripts should not be marked
> that way.

I see. I should be able to find a solution:

- some CIs use kselftest's runner.sh script, and kernels >= 5.10 have a
fix thanks to 303f8e2d0200 ("selftests/kselftest/runner/run_one(): allow
running non-executable files"): it prints a warning, so I thought I had
to fix it, but fine to keep the warning and drop the fix.

- Others try to execute the scripts directly, but I should be able to
fix those.

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.