On Tue, Sep 09, 2025 at 11:25 PM +02, Michal Luczaj wrote:
> On 9/9/25 12:15, Jakub Sitnicki wrote:
>> On Tue, Sep 09, 2025 at 11:51 AM +02, Jakub Sitnicki wrote:
>>> On Fri, Sep 05, 2025 at 01:11 PM +02, Michal Luczaj wrote:
>>>> try_recv() was meant to support both @expect_success cases, but all the
>>>> callers use @expect_success=false anyway. Drop the unused logic and fold in
>>>> MSG_DONTWAIT. Adapt callers.
>>>>
>>>> Subtle change here: recv() return value of 0 will also be considered (an
>>>> unexpected) success.
>>>>
>>>> Signed-off-by: Michal Luczaj <mhal@xxxxxxx>
>>>> ---
>>>> .../selftests/bpf/prog_tests/sockmap_redir.c | 25 +++++++++-------------
>>>> 1 file changed, 10 insertions(+), 15 deletions(-)
>>>>
>>>> diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c b/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
>>>> index 9c461d93113db20de65ac353f92dfdbe32ffbd3b..c1bf1076e8152b7d83c3e07e2dce746b5a39cf7e 100644
>>>> --- a/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
>>>> +++ b/tools/testing/selftests/bpf/prog_tests/sockmap_redir.c
>>>> @@ -144,17 +144,14 @@ static void get_redir_params(struct redir_spec *redir,
>>>> *redirect_flags = 0;
>>>> }
>>>>
>>>> -static void try_recv(const char *prefix, int fd, int flags, bool expect_success)
>>>> +static void fail_recv(const char *prefix, int fd, int more_flags)
>>>> {
>>>> ssize_t n;
>>>> char buf;
>>>>
>>>> - errno = 0;
>>>> - n = recv(fd, &buf, 1, flags);
>>>> - if (n < 0 && expect_success)
>>>> - FAIL_ERRNO("%s: unexpected failure: retval=%zd", prefix, n);
>>>> - if (!n && !expect_success)
>>>> - FAIL("%s: expected failure: retval=%zd", prefix, n);
>>>> + n = recv(fd, &buf, 1, MSG_DONTWAIT | more_flags);
>>>> + if (n >= 0)
>>>> + FAIL("%s: unexpected success: retval=%zd", prefix, n);
>>>> }
>>>
>>> This bit, which you highlighted in the description, I don't get.
>>>
>>> If we're expecting to receive exactly one byte, why treat a short read
>>> as a succcess? Why not make it a strict "n != 1" check?
>>>
>>> [...]
>>
>> Nevermind. It makes sense now. We do want to report a failure for 0-len
>> msg recv as well. You're effectively checking if the rcv queue is empty.
>>
>> I'd add MSG_PEEK, to signal that we're _just checking_ if the socket is
>> readable, and turn the check into the below to succeed only when
>> queue is empty:
>>
>> (n != -1 || (errno != EAGAIN && errno != EWOULDBLOCK))
>
> Well, looks like adding MSG_PEEK exposed a bug in the test. I'll fix that.
The gift that keeps on giving xD
Other alternatives that should also work, but who knows:
- select/poll/epoll readability check
- ioctl(SIOCINQ) but no way to tell if 0-len msg is pending
