Fw: How to define rule for SERVICE_START/STOP?

Fw: How to define rule for SERVICE_START/STOP?

[Gisela Cheng]

[-- Attachment #1.1: Type: text/plain, Size: 2848 bytes --]

So I understand the SERVICE_START/STOP are sent by systemd.  We do have a 
unit file for systemd as following:

 [Unit]
Description=......

[Service]
Type=forking
WorkingDirectory=/var/lib/....

PIDFile=/var/run/xxxx.pid

ExecStart=/usr/sbin/xxxx

Restart=always

# Wait a bit before restarting xxx
RestartSec=5s

# xxxx kicks the watchdog 
WatchdogSec=60

# If xxxx doesn't signal that it's ready, consider it failed
TimeoutStartSec=300

# Time between SIGTERM and SIGKILL on shutdown
TimeoutStopSec=20

# Allows calls to sd_notify from main process
NotifyAccess=main

StandardInput=null
StandardOutput=journal
StandardError=journal
 
[Install]
WantedBy=multi-user.target


Would this trigger the creation of the audit record?  Our application is 
expected to be up and running all the time.  But we can shut it down, or
it could be down due to error.  We want audit record when it is down. 
Would TimeoutStopSec=  tag (don't know if this is the proper name) 
does it, or I need other special tag? 

Thanks.


 Gisela Cheng 
 giselac@us.ibm.com

----- Forwarded by Gisela Cheng/Poughkeepsie/IBM on 08/05/2014 09:24 AM 
-----

From:   Steve Grubb <sgrubb@redhat.com>
To:     linux-audit@redhat.com
Cc:     Gisela Cheng/Poughkeepsie/IBM@IBMUS
Date:   08/04/2014 04:34 PM
Subject:        Re: How to define rule for SERVICE_START/STOP?



On Monday, August 04, 2014 03:16:18 PM Gisela Cheng wrote:
>  We want to use Linux audit type SERVICE_START/STOP for our application
> running as service.
>  But I am not able to find example on how to use auditctl to define the
> rule.  It seems to me that all the examples are of rules defined for
> system_calls.

There are 2 kinds of events. Some are hardwired into the applications (or 
kernel) and they send them if auditing is enabled. The other kind are 
discretionary in that the admin defines what to audit. The problem with 
the 
discretionary rules are that it is from the kernel's point of view. 
Meaning 
that you only get events as the process transitions through something the 
kernel controls like files or syscalls.


> Questions:
>  1.  Can I use audit type SERVICE_START/STOP for my application runs as
> service?

The SERVICE_START/STOP command are intended to be sent by the init daemon. 

Systemd already sends these. Upstart could be patched, but if that is done 
it 
would need to match the layout, order, and formatting of the systemd 
events.

> or would it be considered as type USR_CMD?

USER_CMD is for something like sudo to record what command the user will 
be 
running.

>  2.  How do I use auditctl to define rule for SERVICE_START/STOP?  Can 
you
> direct/point me to URL/documentation where it is documented?

It can't define these events because the kernel only sees a process start 
or 
stop. It has no idea that its a service. Only init can tell the 
difference.

-Steve


[-- Attachment #1.2: Type: text/html, Size: 5819 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: Fw: How to define rule for SERVICE_START/STOP?

[Steve Grubb]

On Tuesday, August 05, 2014 10:36:17 AM Gisela Cheng wrote:
> Would this trigger the creation of the audit record? 

I don't think anything in the config file would affect this. The requirements are 
to record all starts and stops. I don't think it can record a segfault kind of 
stop. That would, however, generate an ANOM_ABEND event.


> Our application is expected to be up and running all the time.  But we can
> shut it down, or it could be down due to error.  We want audit record when
> it is down. Would TimeoutStopSec=  tag (don't know if this is the proper
> name)  does it, or I need other special tag? 

No tag should affect this.

-Steve

Re: Fw: How to define rule for SERVICE_START/STOP?

[Eric Paris]

On Tue, 2014-08-05 at 12:48 -0400, Steve Grubb wrote:
> On Tuesday, August 05, 2014 10:36:17 AM Gisela Cheng wrote:
> > Would this trigger the creation of the audit record? 
> 
> I don't think anything in the config file would affect this. The requirements are 
> to record all starts and stops. I don't think it can record a segfault kind of 
> stop. That would, however, generate an ANOM_ABEND event.

All stops of a non-forking daemon will be properly tracked with a stop
event (including segfault).  systemd doesn't just start processes and
forget about them.  It isn't sysv.  We don't live in the stone ages any
more   :)

For forking services (like yours) it gets it right most of the time,
especially since you told it the pid file, but I understand it isn't
perfect.  sendmail is a forking service if you kill -SIGV $SENDMAIL_PID
you will get something like:

type=ANOM_ABEND msg=audit(1407329684.562:419): auid=4294967295 uid=0 gid=0 ses=4294967295 subj=system_u:system_r:sendmail_t:s0 pid=1114 comm="sendmail" exe="/usr/sbin/sendmail.sendmail" sig=11
type=SERVICE_STOP msg=audit(1407329684.570:420): pid=1 uid=0 auid=4294967295 ses=4294967295 subj=system_u:system_r:init_t:s0 msg=' comm="sendmail" exe="/usr/lib/systemd/systemd" hostname=? addr=? terminal=? res=failed'

Aka, Steve is right and you are right.