Re: [RFC PATCH] tst: Add test for gai_suspend

[Adhemerval Zanella <adhemerval.zanella@linaro.org>]

On 12/03/2021 13:09, Lukasz Majewski wrote:
> Dear Community,
> 
>> ---
>>   resolv/Makefile          |  2 ++
>>   resolv/tst-gai_suspend.c | 53
>> ++++++++++++++++++++++++++++++++++++++++ 2 files changed, 55
>> insertions(+) create mode 100644 resolv/tst-gai_suspend.c
>>
>> diff --git a/resolv/Makefile b/resolv/Makefile
>> index 1047bb6ae5..e7e66fe903 100644
>> --- a/resolv/Makefile
>> +++ b/resolv/Makefile
>> @@ -60,6 +60,7 @@ tests += \
>>     tst-resolv-res_init-multi \
>>     tst-resolv-search \
>>     tst-resolv-trailing \
>> +  tst-gai_suspend \
>>   
>>   # This test calls __res_context_send directly, which is not exported
>>   # from libresolv.
>> @@ -209,3 +210,4 @@ $(objpfx)tst-ns_name_compress:
>> $(objpfx)libresolv.so $(objpfx)tst-ns_name_pton: $(objpfx)libresolv.so
>>   $(objpfx)tst-res_hnok: $(objpfx)libresolv.so
>>   $(objpfx)tst-p_secstodate: $(objpfx)libresolv.so
>> +$(objpfx)tst-gai_suspend: $(objpfx)libanl.so
>> diff --git a/resolv/tst-gai_suspend.c b/resolv/tst-gai_suspend.c
>> new file mode 100644
>> index 0000000000..bbafd13fc4
>> --- /dev/null
>> +++ b/resolv/tst-gai_suspend.c
>> @@ -0,0 +1,53 @@
>> +/* Test for gai_suspend timeout
>> +   Copyright (C) 2021 Free Software Foundation, Inc.
>> +   This file is part of the GNU C Library.
>> +
>> +   The GNU C Library is free software; you can redistribute it and/or
>> +   modify it under the terms of the GNU Lesser General Public
>> +   License as published by the Free Software Foundation; either
>> +   version 2.1 of the License, or (at your option) any later version.
>> +
>> +   The GNU C Library is distributed in the hope that it will be
>> useful,
>> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
>> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
>> +   Lesser General Public License for more details.
>> +
>> +   You should have received a copy of the GNU Lesser General Public
>> +   License along with the GNU C Library; if not, see
>> +   <https://www.gnu.org/licenses/>.  */
>> +
>> +#include <time.h>
>> +#include <netdb.h>
>> +#include <errno.h>
>> +#include <string.h>
>> +#include <support/check.h>
>> +#include <support/xtime.h>
>> +#include <support/timespec.h>
>> +
>> +static int
>> +do_test (void)
>> +{
>> +  struct timespec ts;
>> +  xclock_gettime (CLOCK_REALTIME, &ts);
>> +  struct timespec timeout = make_timespec (1, 0);
>> +  ts = timespec_add (ts, timeout);
>> +
>> +  /* Ignore gaicb content - just wait for timeout.  */
>> +  struct addrinfo result;
>> +
>> +  struct gaicb gai = { "foo.baz", "0", NULL, &result };
>> +  const struct gaicb * const gai_s[] = { &gai };
>> +  struct gaicb *gai_p = &gai;
>> +
>> +  int ret = getaddrinfo_a (GAI_NOWAIT, &gai_p, 1, NULL);
>> +  if (ret != 0)
>> +    FAIL_EXIT1 ("getaddrinfo_a failed: %m\n");
>> +
>> +  ret = gai_suspend (gai_s, 1, &timeout);
>> +  TEST_COMPARE (ret, EAI_AGAIN);
>> +  TEST_TIMESPEC_NOW_OR_AFTER (CLOCK_REALTIME, ts);
>> +
>> +  return 0;
>> +}
>> +
>> +#include <support/test-driver.c>
> 
> I'm a bit puzzled with the above gai_suspend() test.
> 
> 
> The idea for Y2038 test is to check if one can timeout when passed
> crafted struct gaicb causes the gai_suspend() to exit with timeout.
> This approach is similar to one from tst-ppoll.c
> 
> 
> 1. When I do run it by hand on the remove machine (from glibc test
> suite):
> --------------------------------------------------
> 
> rm -rf ./resolv/*
> make PARALLELMFLAGS="-j8"
> test-wrapper='/home/lukma/work/yocto/y2038/build/workspace/sources/y2038-glibc/scripts/cross-test-ssh.sh
> --allow-time-setting root@192.168.7.2' subdirs=resolv check
> 
> I do see the following error:
> -----------------------------
> 
> cat ./resolv/tst-gai_suspend.out
> Didn't expect signal from child: got `Segmentation fault'
> 
> This is strange, as I've passed NULL to getaddrinfo_a as the last
> parameter, which causes passing no signals to program running it.
> 
> What am I missing here? Has the test setup any strict policy to handle
> LWPs ?
> 
> 
> 2. On the VM (armv7):
> 
> gai_suspend.c test program:
> ---------------------------
> 
> #define _GNU_SOURCE
> 
> #include <stdio.h>
> #include <time.h>
> #include <netdb.h>
> 
> int main (int argc, char **argv)
> {
>          struct timespec ts, timeout, now;
> 
>          clock_gettime (CLOCK_REALTIME, &ts);
> 
>          timeout.tv_sec = ts.tv_sec + 3;
>          timeout.tv_nsec = ts.tv_nsec;
> 
>          struct addrinfo result;
>          struct gaicb gai = { "foo.baz", "0", NULL, &result };
>          const struct gaicb * const gai_s[] = { &gai };
>          struct gaicb *gai_p = &gai;
> 
>          int ret = getaddrinfo_a (GAI_NOWAIT, &gai_p, 1, NULL);
>          if (ret != 0)
>            printf("getaddrinfo_a error");
> 
>          ret = gai_suspend (gai_s, 1, &timeout);
>          // According to man getaddrinfo_s I shall receive ret == -3
>          // when conversion from name to IP credentials is performed
>          if (ret != 0)
>            printf("gai_suspend error");
> 
>          clock_gettime (CLOCK_REALTIME, &now);
> 
>          printf("gai_suspend seconds diff: %llu\n", now.tv_sec -
>          ts.tv_sec);
> 
>          return 0;
> }
> 
> 
> gcc -Wall -ggdb -D_TIME_BITS=64 -D_FILE_OFFSET_BITS=64 -I/opt/include \
>      -I/opt/usr/include -L/opt/usr/lib -Wl,-rpath=/opt/lib \
>      -Wl,--dynamic-linker=/opt/lib/ld-linux.so.2  gai_suspend.c -o gs
> -lanl \ -lpthread -static
> 
> The above line has some manual adjustments for linker, but this works
> for other syscalls for some long time now - more info [1].
> 
> 
> In this case the program quits without any issues with threads, but the
> output and behavior of gai_suspend is not as expected (according at
> least to the manual [2]). The return code from gai_suspend shall be
> EAI_AGAIN and the result struct has some garbage in it.
> 
> 
> The question(s):
> ================
> 
> 1. Why I do experience thread error in glibc test suite?
> 
> 2. Is the code for gai_suspend testing wrong (not much reference in
> glibc), or do we have a bug in the the implementation?
> 
> 
> Thanks in advance for any help/hints.
> 
> 
> 
> Links:
> [1] - https://github.com/lmajewski/meta-y2038/
> [2] - https://linux.die.net/man/3/gai_suspend
> 
> Best regards,
> 
> Lukasz Majewski

I am not sure from where the failure come from, on my environment I am
not seeing the SEGFAULT failure on neither x86_64 nor on i686.

But I am seeing:

$ ./testrun.sh resolv/tst-gai_suspend
tst-gai_suspend.c:47: numeric comparison failure
    left: 0 (0x0); from: ret
   right: -3 (0xfffffffd); from: EAI_AGAIN
tst-gai_suspend.c:48: 1616504556.995728406s not after 
1616504557.971222237s (difference 0.975493831s)
error: 2 test failures

The issue is tests is using the system NSS and possible DNS cache so the
expected name resolution timeout does not happen reliable.  I am not
sure how we can accomplish it, maybe through running it on a container
with a custom NSS module to actually emulate the delay.