[RFC][PATCH] selinux: Report result in avc messages

[William Roberts]

Yes Stephen is correct. I was suggesting a change to the variable not to
the key in the output KVP.

However...Stephens patch would grow in size and he would have to type that
much more :-)
On Apr 30, 2014 9:08 AM, "Stephen Smalley" <stephen.smalley at gmail.com>
wrote:

> The revised patch switched from result=allowed|denied to
> permissive=0|1 in the avc message.  I think Bill's point was with
> respect to the code, which still internally is passing around the
> result of the decision and inferring the permissive state from it,
> rather than the output string itself.
>
> On Wed, Apr 30, 2014 at 9:01 AM, Steve Grubb <sgrubb at redhat.com> wrote:
> > On Wednesday, April 30, 2014 08:48:31 AM William Roberts wrote:
> >> My only nit would be the variable name result....would it be better
> named
> >> is_permissive or something?
> >
> > That adds more bytes. My personal taste would be to abbreviate it to save
> > bytes. They add up when you have 100's of thousands of events per day.
> >
> > -Steve
> >
> >
> >> Otherwise LGTM. From the Android camp, this will be very helpful.
> >> On Apr 30, 2014 8:43 AM, "Stephen Smalley" <stephen.smalley at gmail.com>
> >>
> >> wrote:
> >> > Attached patch switches to reporting permissive=0|1 and only does it
> >> > for avc: denied messages.
> >> >
> >> > On Wed, Apr 30, 2014 at 8:18 AM, Stephen Smalley
> >> >
> >> > <stephen.smalley at gmail.com> wrote:
> >> > > I could make it permissive=0 or permissive=1 if that is less
> >> > > confusing.  It doesn't necessarily correspond to the result of the
> >> > > system call, just the avc_has_perm call, as e.g. the kernel checks
> >> > > CAP_DAC_OVERRIDE and falls back to CAP_DAC_READ_SEARCH if only
> >> > > read/search access was requested, and there are other cases where a
> >> > > permission denial has a side effect rather than preventing the
> system
> >> > > call (e.g. CAP_FSETID).
> >> > >
> >> > > On Wed, Apr 30, 2014 at 6:34 AM, Daniel J Walsh <dwalsh at redhat.com>
> >> >
> >> > wrote:
> >> > >> On 04/30/2014 09:29 AM, Steve Grubb wrote:
> >> > >>> On Wednesday, April 30, 2014 08:59:50 AM Daniel J Walsh wrote:
> >> > >>>> How about permitted rather then allowed.
> >> > >>>
> >> > >>> I think permitted is already in an AVC.
> >> > >>
> >> > >> Not sure where.
> >> > >>
> >> > >>>> On 04/29/2014 10:59 PM, Eric Paris wrote:
> >> > >>>>> On Tue, 2014-04-29 at 16:54 -0700, Stephen Smalley wrote:
> >> > >>>>>> Requested for Android in order to distinguish denials that are
> not
> >> >
> >> > in
> >> >
> >> > >>>>>> fact breaking anything yet due to permissive domains versus
> denials
> >> > >>>>>> that are being enforced, but seems generally useful.  result
> field
> >> >
> >> > was
> >> >
> >> > >>>>>> already in the selinux audit data structure and was being
> passed to
> >> > >>>>>> avc_audit() but wasn't being used.  Seems to cause no harm to
> >> >
> >> > ausearch
> >> >
> >> > >>>>>> or audit2allow to add it as a field.  Comments?
> >> > >>>>>
> >> > >>>>> I think it's a great idea, but I'm worried that Steve is going
> to
> >> > >>>>> get
> >> > >>>>> grumpy because an AVC record is going to have a result= field
> which
> >> >
> >> > is
> >> >
> >> > >>>>> similar, but not necessarily related to the res= field of a
> SYSCALL
> >> > >>>>> record.
> >> > >>>
> >> > >>> I think that I'll have to parse this field no matter what. Its
> >> >
> >> > probably that
> >> >
> >> > >>> important. In the syscall, we use success= to be the final
> >> >
> >> > determination.
> >> >
> >> > >>>>> Seems easily confused (although probably 9999 times out of
> >> > >>>>> 10000 they will be the same)
> >> > >>>
> >> > >>> Why would this ever not be correct? Are there times when we get
> an AVC
> >> >
> >> > with a
> >> >
> >> > >>> denial _and_ the syscall completes successfully?
> >> > >>>
> >> > >>> I'd suggest using res= since its in the audit dictionary and means
> >> >
> >> > exactly
> >> >
> >> > >>> what you are wanting to use it for. In it, 1 is success, 0 is
> failure.
> >> > >>
> >> > >> I have seen AVC's where the success=yes in enforcing mode.
>  Basically
> >> > >> the kernel takes a different code path and the syscall succeeds.
>  Most
> >> > >> of these end up as dontaudits.
> >> > >>
> >> > >>>>> So while I wholeheartedly think we should take the idea, I
> wonder if
> >> > >>>>> someone can dream up a name that isn't confusingly similar...
> >> > >>>>>
> >> > >>>>> I can't think of anything...
> >> > >>>
> >> > >>> There is thesaurus.com. :-)
> >> > >>>
> >> > >>> consequence, outcome, effect, reaction,  conclusion, verdict,
> >> > >>> decision,
> >> > >>> judgement, finding, ruling, answer, solution, recommendation,
> order,
> >> >
> >> >  ...
> >> >
> >> > >>> -Steve
> >> >
> >> > --
> >> > Linux-audit mailing list
> >> > Linux-audit at redhat.com
> >> > https://www.redhat.com/mailman/listinfo/linux-audit
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/linux-audit/attachments/20140430/2c317573/attachment.htm>