Devuan bug report logs - #498
libc6: Permission denied, intermittent in execve

Reply or subscribe to this bug.

Toggle useless messages

Message #5 received at submit@bugs.devuan.org (full text, mbox, reply):

Package: libc6
Version: GNU C Library (Debian GLIBC 2.28-10) stable release version 2.28.
Severity: normal


-------- Forwarded Message --------
Subject: libc6: Permission denied, intermittent in execve
Date: Mon, 27 Jul 2020 10:25:27 +0200
From: Alessandro Vesely <vesely@tana.it>
To: Devuan Bug Tracking System <submit@bugs.devuan.org>


Dear Maintainer,

in certain situations, execve fails setting errno to EACCESS.  The same
program, launched by the same user in different ways, succeeds or fails
according to preceding actions.

None of the failure conditions for EACCESS is met.

The case at hand happens with an old version of Thunderbird and a LibreOffice
attachment.  After saving the attachment, Thunderbird execs gio-launch-desktop.
The latter tries to exec libreoffice6.4 and fails.

I strace'd the full arguments used in the failed execve(), and copied them to a
simple C program which runs just that execve() call.  When called from the
command line, the program succeeds.  Then I replaced the gio-launch-desktop
executable with my straw men.  When called from Thunderbird, the program fails.

See also:
https://unix.stackexchange.com/questions/600174/permission-denied-intermittent-in-execve


-- System Information:
Distributor ID: Debian
Description:    Devuan GNU/Linux 3 (beowulf)
Release:        3
Codename:       beowulf
Architecture: x86_64

Kernel: Linux 4.19.0-9-amd64 (SMP w/8 CPU cores)
Locale: LANG=en_IE.UTF-8, LC_CTYPE=en_IE.UTF-8 (charmap=UTF-8),
LANGUAGE=en_IE.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash

Message #8 received at 498@bugs.devuan.org (full text, mbox, reply):

Control: forcemerge 498 497
Control: tags -1 debian
Control: forwarded -1 https://bugs.debian.org/966343

Alessandro

On Mon, Jul 27, 2020 at 10:32:15AM +0200, Alessandro Vesely wrote:
> Package: libc6
> Version: GNU C Library (Debian GLIBC 2.28-10) stable release version 2.28.
> Severity: normal
> 
> 
> -------- Forwarded Message --------
> Subject: libc6: Permission denied, intermittent in execve
> Date: Mon, 27 Jul 2020 10:25:27 +0200
> From: Alessandro Vesely <vesely@tana.it>
> To: Devuan Bug Tracking System <submit@bugs.devuan.org>
> 
> 
> Dear Maintainer,
> 
> in certain situations, execve fails setting errno to EACCESS.  The same
> program, launched by the same user in different ways, succeeds or fails
> according to preceding actions.

Thanks for this. As you have realised, libc6 is a Debian package that Devuan
uses directly without recompilation so this issue is correctly dealt with in
Debian's BTS.

However, one thought that occurs to me is whether apparmor is causing this? Does
disabling it[1] restore predictable behaviour?

Thanks.

Mark


[1]  https://wiki.debian.org/AppArmor/HowToUse#Diagnose_if_a_bug_might_have_been_caused_by_AppArmor

Message #21 received at 498@bugs.devuan.org (full text, mbox, reply):

Hi Mark,

On Mon 27/Jul/2020 11:14:01 +0200 Mark Hindley wrote:
> On Mon, Jul 27, 2020 at 10:32:15AM +0200, Alessandro Vesely wrote:
>> Package: libc6
>> Version: GNU C Library (Debian GLIBC 2.28-10) stable release version 2.28.
>> Severity: normal
>> 
>> in certain situations, execve fails setting errno to EACCESS.  The same
>> program, launched by the same user in different ways, succeeds or fails
>> according to preceding actions.
> 
> Thanks for this. As you have realised, libc6 is a Debian package that Devuan
> uses directly without recompilation so this issue is correctly dealt with in
> Debian's BTS.
> 
> However, one thought that occurs to me is whether apparmor is causing this? Does
> disabling it[1] restore predictable behaviour?


Bingo!

Jul 27 09:47:25 pcale kernel: [ 1569.887279] audit: type=1400 audit(1595836045.642:33): apparmor="DENIED" operation="exec" profile="thunderbird" name="/opt/lib
reoffice6.4/program/soffice" pid=5402 comm="gio-launch-desk" requested_mask="x" denied_mask="x" fsuid=1000 ouid=0

I dunno how come apparmor got installed.  Probably it happened when I upgraded to Beowulf.

After aa-teardown and purging apparmor, execve works as expected.

So this turns out to be a documentation bug.  The execve man page should mention that EACCESS can result as an (unforeseen) apparmor impediment.


Thank you so much
Ale

Message #26 received at 498@bugs.devuan.org (full text, mbox, reply):

On Mon, Jul 27, 2020 at 11:47:34AM +0200, Alessandro Vesely wrote:
> > However, one thought that occurs to me is whether apparmor is causing this? Does
> > disabling it[1] restore predictable behaviour?
> 
> Bingo!
> 
> Jul 27 09:47:25 pcale kernel: [ 1569.887279] audit: type=1400 audit(1595836045.642:33): apparmor="DENIED" operation="exec" profile="thunderbird" name="/opt/lib
> reoffice6.4/program/soffice" pid=5402 comm="gio-launch-desk" requested_mask="x" denied_mask="x" fsuid=1000 ouid=0
> 
> I dunno how come apparmor got installed.  Probably it happened when I upgraded
> to Beowulf.

Yes, it is now the default in Debian buster and Devuan beowulf has inherited
that.

Mark

Message #31 received at 498@bugs.devuan.org (full text, mbox, reply):

Alessandro Vesely, le lun. 27 juil. 2020 11:47:34 +0200, a ecrit:
> So this turns out to be a documentation bug.  The execve man page should mention that EACCESS can result as an (unforeseen) apparmor impediment.

Well, basically all system calls would then need this...

Samuel

Message #36 received at 498@bugs.devuan.org (full text, mbox, reply):

On Mon, 27 Jul 2020 12:13:44 +0200 Samuel Thibault <sthibault@debian.org> wrote:
> Alessandro Vesely, le lun. 27 juil. 2020 11:47:34 +0200, a ecrit:
> > So this turns out to be a documentation bug.  The execve man page should mention that EACCESS can result as an (unforeseen) apparmor impediment.
> 
> Well, basically all system calls would then need this...


Yeah, likely.  How many man pages have snippets like "[...] denied for one of the directories in the path [...]"?

Yet, considering the following examples, they seem to have been written manually rather than resorting to some sort of script:


       EACCES The requested access to the file is not allowed, or search permission is denied for one of the directories in the  path
              prefix  of  pathname, or the file did not exist yet and write access to the parent directory is not allowed.  (See also
              path_resolution(7).)

       EACCES Search permission is denied on a component of the path prefix of filename or the name of a  script  interpreter.   (See
              also path_resolution(7).)

       EACCES Write access to the directory containing newpath is denied, or search permission is denied for one of  the  directories
              in the path prefix of oldpath or newpath.  (See also path_resolution(7).)

       EACCES Search permission is denied for a component of the path prefix, or the named file is not writable by  the  user.
              (See also path_resolution(7).)

       EACCES Search permission is denied on a component of the path prefix.  (See also path_resolution(7).)


Philip Couling commented that the man page /could/ mention security extensions since they are prevelent. See:
https://unix.stackexchange.com/questions/600174/identical-execve-causes-permission-denied-for-one-program-but-not-another/600529#comment1121270_600529

For execve, for example, one could add that permissions are not derived from file flags only.  For example:

OLD:

       EACCES Execute permission is denied for the file or a script or ELF interpreter.

NEW:

       EACCES Execute permission for the file or a script or ELF interpreter is denied either by flags or by security modules.


Would that be correct?  Do all "DENIED" operations result in EACCES?  And what do other security modules do?  Hmm...  Starting to document that mess from the point of view of programs getting such failure codes would allow better logging and better troubleshooting.


Best
Ale

Message #41 received at 498@bugs.devuan.org (full text, mbox, reply):

Hi,

On 2020-07-27 19:01, Alessandro Vesely wrote:
> On Mon, 27 Jul 2020 12:13:44 +0200 Samuel Thibault <sthibault@debian.org> wrote:
> > Alessandro Vesely, le lun. 27 juil. 2020 11:47:34 +0200, a ecrit:
> > > So this turns out to be a documentation bug.  The execve man page should mention that EACCESS can result as an (unforeseen) apparmor impediment.
> > 
> > Well, basically all system calls would then need this...
> 
> 
> Yeah, likely.  How many man pages have snippets like "[...] denied for one of the directories in the path [...]"?
>
> Yet, considering the following examples, they seem to have been written manually rather than resorting to some sort of script:

There are many syscalls that might end up returning EACCESS. Mapping
them to the corresponding libc functions might not be always fully
linear.

[snip]

> Philip Couling commented that the man page /could/ mention security extensions since they are prevelent. See:
> https://unix.stackexchange.com/questions/600174/identical-execve-causes-permission-denied-for-one-program-but-not-another/600529#comment1121270_600529
> 
> For execve, for example, one could add that permissions are not derived from file flags only.  For example:
> 
> OLD:
> 
>        EACCES Execute permission is denied for the file or a script or ELF interpreter.
> 
> NEW:
> 
>        EACCES Execute permission for the file or a script or ELF interpreter is denied either by flags or by security modules.
> 
> 
> Would that be correct?  Do all "DENIED" operations result in EACCES?  And what do other security modules do?  Hmm...  Starting to document that mess from the point of view of programs getting such failure codes would allow better logging and better troubleshooting.

That seems correct for apparmor (although it can also return ENOENT in
case of race condition), not sure about other security modules. seccomp
just kills the process, not sure you want to mention that in every
possible manpages.

In any case this is now out of scope for the glibc package, so I am
closing the bug. If you feel that the doc needs to be updated, please
open a bug against the corresponding package.

Regards,
Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Send a report that this bug log contains spam.