Re: [PATCH ghak90 (was ghak32) V4 03/10] audit: log container info of syscalls

From: Paul Moore <paul@paul-moore.com>
To: Richard Guy Briggs <rgb@redhat.com>
Cc: <containers@lists.linux-foundation.org>,
	<linux-api@vger.kernel.org>, <linux-audit@redhat.com>,
	<linux-fsdevel@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
	<netdev@vger.kernel.org>, <netfilter-devel@vger.kernel.org>,
	<ebiederm@xmission.com>, <luto@kernel.org>, <carlos@redhat.com>,
	<dhowells@redhat.com>, <viro@zeniv.linux.org.uk>,
	<simo@redhat.com>, Eric Paris <eparis@parisplace.org>,
	Serge Hallyn <serge@hallyn.com>
Subject: Re: [PATCH ghak90 (was ghak32) V4 03/10] audit: log container info of syscalls
Date: Thu, 25 Oct 2018 07:13:07 +0100	[thread overview]
Message-ID: <166a9dae538.280e.85c95baa4474aabc7814e68940a78392@paul-moore.com> (raw)
In-Reply-To: <20181025004255.zl7p7j6gztouh2hh@madcap2.tricolour.ca>

On October 25, 2018 1:43:16 AM Richard Guy Briggs <rgb@redhat.com> wrote:
> On 2018-10-24 16:55, Paul Moore wrote:
>> On Wed, Oct 24, 2018 at 11:15 AM Richard Guy Briggs <rgb@redhat.com> wrote:
>>> On 2018-10-19 19:16, Paul Moore wrote:
>>>> On Sun, Aug 5, 2018 at 4:32 AM Richard Guy Briggs <rgb@redhat.com> wrote:
>

...

>
>>>> However, I do care about the "op" field in this record.  It just
>>>> doesn't make any sense; the way you are using it it is more of a
>>>> context field than an operations field, and even then why is the
>>>> context important from a logging and/or security perspective?  Drop it
>>>> please.
>>>
>>> I'll rename it to whatever you like.  I'd suggest "ref=".  The reason I
>>> think it is important is there are multiple sources that aren't always
>>> obvious from the other records to which it is associated.  In the case
>>> of ptrace and signals, there can be many target tasks listed (OBJ_PID)
>>> with no other way to distinguish the matching audit container identifier
>>> records all for one event.  This is in addition to the default syscall
>>> container identifier record.  I'm not currently happy with the text
>>> content to link the two, but that should be solvable (most obvious is
>>> taret PID).  Throwing away this information seems shortsighted.
>>
>> It would be helpful if you could generate real audit events
>> demonstrating the problems you are describing, as well as a more
>> standard syscall event, so we can discuss some possible solutions.
>
> If the auditted process is in a container and it ptraces or signals
> another process in a container, there will be two AUDIT_CONTAINER
> records for the same event that won't be identified as to which record
> belongs to which process or other record (SYSCALL vs 1+ OBJ_PID
> records).  There could be many signals recorded, each with their own
> OBJ_PID record.  The first is stored in the audit context and additional
> ones are stored in a chained struct that can accommodate 16 entries each.
>
> (See audit_signal_info(), __audit_ptrace().)
>
> (As a side note, on code inspection it appears that a signal target
> would get overwritten by a ptrace action if they were to happen in that
> order.)

As requested above, please respond with real audit events generated by this patchset so that we can discuss possible solutions.

--
paul moore
www.paul-moore.com