Abnormal Apache behavior.

[Stefan Held]

Hi guys. 

Dunno if this is not new to you, but i am experiencing a strange
behavior of the apache in FC4 with selinux enabled.

Ok. What have i done? 

First i wrote some php stuff and was wondering why the Server did not
allow to get some files in /css and does not allow to connect via an
network socket to the postgresql server.

Then i restarted the Server with apachectl stop and apachectl start.
From now on everything worked fine and like expected.

Then i did an Kernel update rebooted the machine and my Site was not
reachable again. So i did some investigation and saw in the audit.log
that selinux is disabling some stuff. 

Then i restarted again with apachectl stop and start. And like expected
the httpd started working again. 

Is this an issue? I think this behavior is not normal :-)



-- 
 Stefan Held                    VI has only 2 Modes:
 obi(a)unixkiste.org              The first one is for beeping all the time, 
 IRCNet: Obi_Wan                the second destroys the text.   
---------------------------------------------------------------------------
perl -e'map{print pack c,($|++?1:13)+ord,select$,,$,,$,,$|}split//,ESEL.$/'
---------------------------------------------------------------------------
    GPG-Keyprint = EAF2 6A65 D102 F2DB 4970  2A67 455B 98F2 572C 3FA9

[Colin Walters]

On Wed, 2005-07-06 at 02:43 +0200, Stefan Held wrote:
>  Hi guys. 
>  
>  Dunno if this is not new to you, but i am experiencing a strange
>  behavior of the apache in FC4 with selinux enabled.
>  
>  Ok. What have i done? 
>  
>  First i wrote some php stuff and was wondering why the Server did not
>  allow to get some files in /css and does not allow to connect via an
>  network socket to the postgresql server. 
Did you have a look at this guide?

http://fedora.redhat.com/docs/selinux-apache-fc3/

It needs to be updated for FC4, but should be helpful nonetheless.

>  Then i restarted the Server with apachectl stop and apachectl start.
>  From now on everything worked fine and like expected. 
The reason I believe is because apachectl restarts the Apache httpd
daemon on its own.  The way the Fedora targeted policy works for daemons
is that they are only confined when executed via the /etc/init.d/*
scripts, so when apachectl executes httpd it stays in unconfined_t.

This is to prevent issues such as the system administrator executing
"httpd -t" and causing a domain transition to httpd_t which isn't
allowed access to the administrator's terminal.

Probably we shouldn't ship the apachectl command in Fedora, instead
requiring using "service httpd restart".

[Joe Orton]

On Tue, Jul 05, 2005 at 10:47:15PM -0400, Colin Walters wrote:
>  On Wed, 2005-07-06 at 02:43 +0200, Stefan Held wrote:
>  > Then i restarted the Server with apachectl stop and apachectl start.
>  > From now on everything worked fine and like expected.
>  
>  The reason I believe is because apachectl restarts the Apache httpd
>  daemon on its own.  The way the Fedora targeted policy works for daemons
>  is that they are only confined when executed via the /etc/init.d/*
>  scripts, so when apachectl executes httpd it stays in unconfined_t. 
Eh?  I thought the transition happens upon exec of httpd regardless of 
who performs the exec.  Empirical evidence suggests that's the case 
anyway...

[root@tango ~]# service httpd stop
Stopping httpd:                                            [  OK  ]
[root@tango ~]# apachectl start
[root@tango ~]# ps axZ | grep httpd
root:system_r:httpd_t           30536 ?        Ss     0:00 /usr/sbin/httpd -k start

joe

[Stephen Smalley]

On Fri, 2005-07-08 at 14:15 +0100, Joe Orton wrote:
>  Eh?  I thought the transition happens upon exec of httpd regardless
> of 
>  who performs the exec.  Empirical evidence suggests that's the case 
>  anyway...
>  
>  [root@tango ~]# service httpd stop
>  Stopping httpd:                                            [  OK  ]
>  [root@tango ~]# apachectl start
>  [root@tango ~]# ps axZ | grep httpd
>  root:system_r:httpd_t           30536 ?        Ss     0:00 /usr/sbin/httpd -k start
On FC4, apachectl start leaves it running in unconfined_t.  In FC3,
since the system starts in unconfined_t (so both rc scripts and user
shells are in the same domain), there is no distinction, so you wouldn't
see a difference there.

-- 
Stephen Smalley
National Security Agency

[Joe Orton]

On Fri, Jul 08, 2005 at 09:43:30AM -0400, Stephen Smalley wrote:
>  On Fri, 2005-07-08 at 14:15 +0100, Joe Orton wrote:
>  > Eh?  I thought the transition happens upon exec of httpd regardless of 
>  > who performs the exec.  Empirical evidence suggests that's the case 
>  > anyway...
>  > 
>  > [root@tango ~]# service httpd stop
>  > Stopping httpd:                                            [  OK  ]
>  > [root@tango ~]# apachectl start
>  > [root@tango ~]# ps axZ | grep httpd
>  > root:system_r:httpd_t           30536 ?        Ss     0:00 /usr/sbin/httpd -k start
>  
>  On FC4, apachectl start leaves it running in unconfined_t.  In FC3,
>  since the system starts in unconfined_t (so both rc scripts and user
>  shells are in the same domain), there is no distinction, so you wouldn't
>  see a difference there. 
OK - can that be changed?  I'd really much rather that apachectl, the 
init script, and direct invocation of /usr/sbin/httpd all had the same 
behaviour, as has been (mostly) the case forever.

joe

[Daniel J Walsh]

Joe Orton wrote:

> On Fri, Jul 08, 2005 at 09:43:30AM -0400, Stephen Smalley wrote:
>   
>
> >On Fri, 2005-07-08 at 14:15 +0100, Joe Orton wrote:
> >    
> >
> >>Eh?  I thought the transition happens upon exec of httpd regardless of 
> >>who performs the exec.  Empirical evidence suggests that's the case 
> >>anyway...
> >>
> >>[root@tango ~]# service httpd stop
> >>Stopping httpd:                                            [  OK  ]
> >>[root@tango ~]# apachectl start
> >>[root@tango ~]# ps axZ | grep httpd
> >>root:system_r:httpd_t           30536 ?        Ss     0:00 /usr/sbin/httpd -k
> start
> >>      
> >>
> >On FC4, apachectl start leaves it running in unconfined_t.  In FC3,
> >since the system starts in unconfined_t (so both rc scripts and user
> >shells are in the same domain), there is no distinction, so you wouldn't
> >see a difference there.
> >    
> >
>
> OK - can that be changed?  I'd really much rather that apachectl, the 
> init script, and direct invocation of /usr/sbin/httpd all had the same 
> behaviour, as has been (mostly) the case forever.
>
> joe
>
>
>
>   
>  
It already has been.  apachectl is set to initrc_exec_t whith will start 
httpd in the correct context.  Install
the latest policy for FC4 and run restorecon on apachectl if it is not 
set to initrc_exec_t.

Dan

> --
> fedora-selinux-list mailing list
> fedora-selinux-list(a)redhat.com
> http://www.redhat.com/mailman/listinfo/fedora-selinux-list
>   
>  

--

[Joe Orton]

On Fri, Jul 29, 2005 at 09:15:18AM -0400, Daniel J Walsh wrote:
>  It already has been.  apachectl is set to initrc_exec_t whith will 
>  start httpd in the correct context.  Install the latest policy for FC4 
>  and run restorecon on apachectl if it is not set to initrc_exec_t. 
Ah, great, yes, I see it on my current system.  Thanks Dan.

joe

[Colin Walters]

On Fri, 2005-07-29 at 11:39 +0100, Joe Orton wrote:
>  On Fri, Jul 08, 2005 at 09:43:30AM -0400, Stephen Smalley wrote:
>  > On Fri, 2005-07-08 at 14:15 +0100, Joe Orton wrote:
>  > > Eh?  I thought the transition happens upon exec of httpd regardless of 
>  > > who performs the exec.  Empirical evidence suggests that's the case 
>  > > anyway...
>  > > 
>  > > [root@tango ~]# service httpd stop
>  > > Stopping httpd:                                            [  OK  ]
>  > > [root@tango ~]# apachectl start
>  > > [root@tango ~]# ps axZ | grep httpd
>  > > root:system_r:httpd_t           30536 ?        Ss     0:00 /usr/sbin/httpd -k
> start
>  > 
>  > On FC4, apachectl start leaves it running in unconfined_t.  In FC3,
>  > since the system starts in unconfined_t (so both rc scripts and user
>  > shells are in the same domain), there is no distinction, so you wouldn't
>  > see a difference there.
>  
>  OK - can that be changed?  I'd really much rather that apachectl, the 
>  init script, and direct invocation of /usr/sbin/httpd all had the same 
>  behaviour, as has been (mostly) the case forever. 
For direct invocation of /usr/sbin/httpd; we can't have it both ways.
It has to either be confined or not confined.  People seem to want it
unconfined so e.g. httpd -t can still print to the terminal.

[Stefan Held]

Am Freitag, den 08.07.2005, 14:15 +0100 schrieb Joe Orton:
[root@extranet ~]# service httpd restart
httpd beenden:                                             [  OK  ]
httpd starten:                                             [  OK  ]

[root@extranet ~]# ps auxZ | grep httpd
root:system_r:httpd_t           root     25934  2.9  0.4  19080  9252 ?
Ss   01:40   0:00 /usr/sbin/httpd
root:system_r:httpd_t           apache   25936  0.0  0.4  19212  9272 ?
S    01:40   0:00 /usr/sbin/httpd
root:system_r:httpd_t           apache   25937  0.0  0.4  19212  9272 ?
S    01:40   0:00 /usr/sbin/httpd
root:system_r:httpd_t           apache   25938  0.0  0.4  19212  9272 ?
S    01:40   0:00 /usr/sbin/httpd
root:system_r:httpd_t           apache   25939  0.0  0.4  19212  9272 ?
S    01:40   0:00 /usr/sbin/httpd

[root@extranet ~]# apachectl stop
[root@extranet ~]# apachectl start
[root@extranet ~]# ps auxZ | grep httpd
root:system_r:unconfined_t      root     25956 14.3  0.4  21160  9244 ?
Ss   01:40   0:00 /usr/sbin/httpd -k start
root:system_r:unconfined_t      apache   25957  0.0  0.4  21292  9264 ?
S    01:40   0:00 /usr/sbin/httpd -k start
root:system_r:unconfined_t      apache   25958  0.0  0.4  21292  9264 ?
S    01:40   0:00 /usr/sbin/httpd -k start
root:system_r:unconfined_t      apache   25959  0.0  0.4  21292  9264 ?
S    01:40   0:00 /usr/sbin/httpd -k start




-- 
 Stefan Held                    VI has only 2 Modes:
 obi(a)unixkiste.org              The first one is for beeping all the time, 
 IRCNet: Obi_Wan                the second destroys the text.   
---------------------------------------------------------------------------
perl -e'map{print pack c,($|++?1:13)+ord,select$,,$,,$,,$|}split//,ESEL.$/'
---------------------------------------------------------------------------
    GPG-Keyprint = EAF2 6A65 D102 F2DB 4970  2A67 455B 98F2 572C 3FA9