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

[Lukasz Majewski <lukma@denx.de>]

[-- Attachment #1: Type: text/plain, Size: 8871 bytes --]

Hi Adhemerval,

> 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.

Even more strange (tested on x86_64):

1. For "sourceware.org" + "http"
struct gaicb gai = { "sourceware.org", "http", NULL, &result };

getaddrinfo_a (GAI_NOWAIT, &gai_p, 1, NULL);
gai_suspend (gai_s, 1, &timeout);

ret = gai_error(gai_p);

Here the ret is 0 as expected.

2. For "foo.baz" + "0" (the gai_suspend and getaddrinfo_a are called in
the same way as above)

struct gaicb gai = { "foo.baz", "0", NULL, &result };

Here the ret is -2 as expected (EAI_NONAME - NAME or SERVICE unknown) -
the "0" here is wrong.


3. For "foo.baz" + "http" (the gai_suspend and getaddrinfo_a are called
in the same way as above)

struct gaicb gai = { "foo.baz", "http", NULL, &result };

Here the ret is also -2, which I don't understand. I would expect that
the foo.baz name is going to be resolved as it "looks" correctly and
hence the timeout will occur.

Any hints on how timeout can be forced?

>  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.


Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]