An obscure problem with v2.37 make check

An obscure problem with v2.37 make check

[Bruce Dubbs]

When I built version v2.37, make check hung on me.  The problem was in 
my kernel configuration, but I thought I would pass this on.

First of all, my biggest problem was finding out what had failed.  A 
simple 'make check' ran the tests and when it got to the end, hung.

I could not find any documentation about running the tests, but diving 
into the Makefile led me to tests/run.sh.  By default this runs tests 
using all cores, but it is unclear if this is meant to be run directly.

In any case running run.sh --parallel=1 (instead of the default 24) 
allowed me to find the problem test, tests/ts/lsns/ioctl_ns.  In this 
test, it is doing:

my_userns=$(stat -c %i -L /proc/self/ns/user)

My problem was that /proc/self/ns/user did not exist on my system.  This 
was due to a missing CONFIG option in my kernel configuration.  Adding 
that and rebuilding the kernel allowed all tests to pass.

One solution to this problem may be to test for the existence of the 
file before running 'stat' similar to the tests for programs like 
'touch' and 'uniq'.  Since this is such an unusual situation, I can 
understand if you just ignore the issue, however I thought you should 
know about it.

   -- Bruce

Re: An obscure problem with v2.37 make check

[Karel Zak]

On Wed, Jun 16, 2021 at 03:42:04PM -0500, Bruce Dubbs wrote:
> When I built version v2.37, make check hung on me.  The problem was in my
> kernel configuration, but I thought I would pass this on.
> 
> First of all, my biggest problem was finding out what had failed.  A simple
> 'make check' ran the tests and when it got to the end, hung.
> 
> I could not find any documentation about running the tests, but diving into
> the Makefile led me to tests/run.sh.  By default this runs tests using all
> cores, but it is unclear if this is meant to be run directly.
 
The tests are designed for smart people, like you who, are able to debug it ;-)

> In any case running run.sh --parallel=1 (instead of the default 24) allowed
> me to find the problem test, tests/ts/lsns/ioctl_ns.  In this test, it is
> doing:

The --parallel=1 is the default when you execute ./run.sh from command
line. This is the way I usually use the tests. The "make check"
executes it in parallel to make it faster for automated execution
(github actions, travis, random QA, etc.)

> 
> my_userns=$(stat -c %i -L /proc/self/ns/user)
> 
> My problem was that /proc/self/ns/user did not exist on my system.  This was
> due to a missing CONFIG option in my kernel configuration.  Adding that and
> rebuilding the kernel allowed all tests to pass.
> 
> One solution to this problem may be to test for the existence of the file
> before running 'stat' similar to the tests for programs like 'touch' and

Sounds good. Please, send patch.

For 3rd party programs you need "ts_check_prog <progname>" at the
beginning of the test.

> 'uniq'.  Since this is such an unusual situation, I can understand if you
> just ignore the issue, however I thought you should know about it.
 
It's definitely important to run the tests in more environments to
make it stable and I'm happy that we have contributors from
non-mainstream distributions.

So, thanks for your feedback.

  Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: An obscure problem with v2.37 make check

[Anatoly Pugachev]

On Thu, Jun 17, 2021 at 12:00 PM Karel Zak <kzak@redhat.com> wrote:
> > One solution to this problem may be to test for the existence of the file
> > before running 'stat' similar to the tests for programs like 'touch' and
>
> Sounds good. Please, send patch.

Bruce, Karel,

something like this?

util-linux$ git diff
diff --git a/tests/ts/lsns/ioctl_ns b/tests/ts/lsns/ioctl_ns
index e91f6743f..b8f35e2e5 100755
--- a/tests/ts/lsns/ioctl_ns
+++ b/tests/ts/lsns/ioctl_ns
@@ -24,6 +24,9 @@ ts_init "$*"
 # ts_skip_nonroot
 grep -q '#define HAVE_LINUX_NSFS_H' ${top_builddir}/config.h ||
ts_skip "no ioctl_ns support"

+[ -a /proc/self/ns/user ] || ts_skip "no USER namespace kernel support"
+[ -a /proc/self/ns/pid ] || ts_skip "no PID namespace kernel support"
+
 ts_check_test_command "$TS_CMD_LSNS"
 ts_check_test_command "$TS_CMD_UNSHARE"
 ts_check_prog "stat"

Re: An obscure problem with v2.37 make check

[Bruce Dubbs]

On 6/17/21 6:37 AM, Anatoly Pugachev wrote:
> On Thu, Jun 17, 2021 at 12:00 PM Karel Zak <kzak@redhat.com> wrote:
>>> One solution to this problem may be to test for the existence of the file
>>> before running 'stat' similar to the tests for programs like 'touch' and
>>
>> Sounds good. Please, send patch.
> 
> Bruce, Karel,
> 
> something like this?
> 
> util-linux$ git diff
> diff --git a/tests/ts/lsns/ioctl_ns b/tests/ts/lsns/ioctl_ns
> index e91f6743f..b8f35e2e5 100755
> --- a/tests/ts/lsns/ioctl_ns
> +++ b/tests/ts/lsns/ioctl_ns
> @@ -24,6 +24,9 @@ ts_init "$*"
>   # ts_skip_nonroot
>   grep -q '#define HAVE_LINUX_NSFS_H' ${top_builddir}/config.h ||
> ts_skip "no ioctl_ns support"
> 
> +[ -a /proc/self/ns/user ] || ts_skip "no USER namespace kernel support"
> +[ -a /proc/self/ns/pid ] || ts_skip "no PID namespace kernel support"
> +
>   ts_check_test_command "$TS_CMD_LSNS"
>   ts_check_test_command "$TS_CMD_UNSHARE"
>   ts_check_prog "stat"
> 

Looks pretty good to me, but I would use -r instead of -a.

   -- Bruce

Re: An obscure problem with v2.37 make check

[Karel Zak]

On Thu, Jun 17, 2021 at 11:50:42AM -0500, Bruce Dubbs wrote:
> On 6/17/21 6:37 AM, Anatoly Pugachev wrote:
> > util-linux$ git diff
> > diff --git a/tests/ts/lsns/ioctl_ns b/tests/ts/lsns/ioctl_ns
> > index e91f6743f..b8f35e2e5 100755
> > --- a/tests/ts/lsns/ioctl_ns
> > +++ b/tests/ts/lsns/ioctl_ns
> > @@ -24,6 +24,9 @@ ts_init "$*"
> >   # ts_skip_nonroot
> >   grep -q '#define HAVE_LINUX_NSFS_H' ${top_builddir}/config.h ||
> > ts_skip "no ioctl_ns support"
> > 
> > +[ -a /proc/self/ns/user ] || ts_skip "no USER namespace kernel support"
> > +[ -a /proc/self/ns/pid ] || ts_skip "no PID namespace kernel support"
> > +
> >   ts_check_test_command "$TS_CMD_LSNS"
> >   ts_check_test_command "$TS_CMD_UNSHARE"
> >   ts_check_prog "stat"
> > 
> 
> Looks pretty good to me, but I would use -r instead of -a.

Applied with -r, thanks guys!

  Karel


-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com