execve()ing non-existent command returns EPERM instead of ENOENT in Koji

[Petr Pisar]

Reviewer of my package found the package does pass tests in Koji
(dist-f16). I found the reason:

A test executes non-existent command and expects ENOENT. This is how it
works even in my local Rawhide. However Koji glibc returns EPERM.

You can see it in build.log of scratch build
<http://koji.fedoraproject.org/koji/taskinfo?taskID=3092204>;:

  + /usr/bin/perl -le 'system qq{x-foo}; printf qq{%d %s\n}, $!, $!'
  13 Permission denied

and than failing tests #3 and #10.

The Perl on-liner executes `x-foo' commands and print errno and
strerror(errno).

Of course I can adjust tests to accept EPERM too. But, preferably, I'd
like to know if the spotted difference is intended or it's just a bug in
Koji setup.

-- Petr

[Petr Pisar]

On 2011-05-25, Petr Pisar <ppisar(a)redhat.com&gt; wrote:
>  A test executes non-existent command and expects ENOENT. This is how it
>  works even in my local Rawhide. However Koji glibc returns EPERM.
>  
Correction: error 13 is EACCES.

-- Petr

[Andreas Schwab]

Petr Pisar <ppisar(a)redhat.com&gt; writes:

>  Reviewer of my package found the package does pass tests in Koji
>  (dist-f16). I found the reason:
>
>  A test executes non-existent command and expects ENOENT. 
That's a bug in the test.  EACCES is a valid error if a PATH element is
inaccessible.

Andreas.

-- 
Andreas Schwab, schwab(a)redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."

[Petr Pisar]

On 2011-05-26, Andreas Schwab <schwab(a)redhat.com&gt; wrote:
>  Petr Pisar <ppisar(a)redhat.com&gt; writes:
>
> > Reviewer of my package found the package does pass tests in Koji
> > (dist-f16). I found the reason:
> >
> > A test executes non-existent command and expects ENOENT.
>
>  That's a bug in the test.  EACCES is a valid error if a PATH element is
>  inaccessible.
>  
You are right. Returning EACCES if command does not exist in any
accesible directory and other directories have insufficient permissions
makes sense.

-- Petr

[Kevin Kofler]

Petr Pisar wrote:
>  A test executes non-existent command and expects ENOENT. This is how
> it
>  works even in my local Rawhide. However Koji glibc returns [EACCESS]. 
I guess the difference is due to SELinux.

        Kevin Kofler

[Daniel J Walsh]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 05/26/2011 06:57 AM, Kevin Kofler wrote:
>  Petr Pisar wrote:
> > A test executes non-existent command and expects ENOENT. This is how it
> > works even in my local Rawhide. However Koji glibc returns [EACCESS].
>  
>  I guess the difference is due to SELinux.
>  
>          Kevin Kofler
>   
Is there a chance SELinux is preventing the search of the path, before
the kernel realizes there is noent?
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.11 (GNU/Linux)
> Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/
>
> iEYEARECAAYFAk3eTvwACgkQrlYvE4MpobMbkACfRpT2zJegSe2P7HD2g3sJrLkE
> 27cAmwTaks33L+29IL8TbBKnjUResAfA
> =70yt
> -----END PGP SIGNATURE-----