netdev - Re: [PATCH bpf v3 3/5] selftests/bpf: test

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <df1d9ec2-783a-09f4-29d7-544d20d74465@gmail.com>
Date:   Wed, 30 May 2018 12:59:03 -0700
From:   John Fastabend <john.fastabend@...il.com>
To:     Alexei Starovoitov <alexei.starovoitov@...il.com>,
        Prashant Bhole <bhole_prashant_q7@....ntt.co.jp>
Cc:     Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        "David S . Miller" <davem@...emloft.net>,
        Shuah Khan <shuah@...nel.org>, netdev@...r.kernel.org,
        linux-kselftest@...r.kernel.org
Subject: Re: [PATCH bpf v3 3/5] selftests/bpf: test_sockmap, fix test timeout

On 05/30/2018 12:29 PM, Alexei Starovoitov wrote:
> On Wed, May 30, 2018 at 02:56:09PM +0900, Prashant Bhole wrote:
>> In order to reduce runtime of tests, recently timout for select() call
>> was reduced from 1sec to 10usec. This was causing many tests failures.
>> It was caught with failure handling commits in this series.
>>
>> Restoring the timeout from 10usec to 1sec
>>
>> Fixes: a18fda1a62c3 ("bpf: reduce runtime of test_sockmap tests")
>> Signed-off-by: Prashant Bhole <bhole_prashant_q7@....ntt.co.jp>
>> ---
>>  tools/testing/selftests/bpf/test_sockmap.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/tools/testing/selftests/bpf/test_sockmap.c b/tools/testing/selftests/bpf/test_sockmap.c
>> index 64f9e25c451f..9d01f5c2abe2 100644
>> --- a/tools/testing/selftests/bpf/test_sockmap.c
>> +++ b/tools/testing/selftests/bpf/test_sockmap.c
>> @@ -345,8 +345,8 @@ static int msg_loop(int fd, int iov_count, int iov_length, int cnt,
>>  		if (err < 0)
>>  			perror("recv start time: ");
>>  		while (s->bytes_recvd < total_bytes) {
>> -			timeout.tv_sec = 0;
>> -			timeout.tv_usec = 10;
>> +			timeout.tv_sec = 1;
>> +			timeout.tv_usec = 0;
> 
> I've applied the set, but had to revert it, since it takes too long.
> 
> real	1m40.124s
> user	0m0.375s
> sys	0m14.521s
> 

Dang, I thought it would be a bit longer but not minutes.

> Myself and Daniel run the test semi-manually when we apply patches.> Adding 2 extra minutes of wait time is unnecessary.

Yep.

> Especially since most of it is idle time.
> Please find a way to fix tests differently.
> btw I don't see any failures today. Not sure what is being fixed
> by incresing a timeout.
> 

Calling these fixes is a bit much, they are primarily improvements.

The background is, when I originally wrote the tests my goal was to
exercise the kernel code paths. Because of this I didn't really care if
the tests actually sent/recv all bytes in the test. (I have long
running tests using netperf/wrk/apached/etc. for that) But, the manual
tests do have an option to verify the data if specified. The 'verify'
option is a bit fragile in that with the right tests (e.g. drop)
or the certain options (e.g. cork) it can fail which is expected.

What Prashant added was support to actually verify the data correctly.
And also fix a few cgroup handling and some pretty printing as well.
He noticed the low timeout causing issue in these cases though so
increased it.

@Prashant, how about increasing this less dramatically because now
all cork tests are going to stall for 1s unless perfectly aligned.
How about 100us? Or even better we can conditionally set it based
on if tx_cork is set. If tx_cork is set use 1us otherwise use 200us
or something. (1s is really to high in any cases for lo)

Also capturing some of the above in the cover letter would help
folks understand the context a bit better.

Thanks!
John