Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Steve Grubb <sgrubb@redhat.com>
To:     Richard Guy Briggs <rgb@redhat.com>
Cc:     Paul Moore <paul@paul-moore.com>,
        Linux-Audit Mailing List <linux-audit@redhat.com>,
        LKML <linux-kernel@vger.kernel.org>,
        linux-fsdevel@vger.kernel.org, Eric Paris <eparis@parisplace.org>,
        Jan Kara <jack@suse.cz>, Amir Goldstein <amir73il@gmail.com>
Subject: Re: [PATCH v4 3/4] fanotify,audit: Allow audit to use the full permission event response
Date:   Wed, 31 Aug 2022 19:55:38 -0400
Message-ID: <5600292.DvuYhMxLoT@x2>
Organization: Red Hat
In-Reply-To: <Yw/efLafvmimtCDq@madcap2.tricolour.ca>
References: <cover.1659996830.git.rgb@redhat.com> <12063373.O9o76ZdvQC@x2> <Yw/efLafvmimtCDq@madcap2.tricolour.ca>
MIME-Version: 1.0
Content-Transfer-Encoding: 7Bit
Content-Type: text/plain; charset="us-ascii"
Precedence: bulk

On Wednesday, August 31, 2022 6:19:40 PM EDT Richard Guy Briggs wrote:
> On 2022-08-31 17:25, Steve Grubb wrote:
> > On Wednesday, August 31, 2022 5:07:25 PM EDT Richard Guy Briggs wrote:
> > > > > diff --git a/kernel/auditsc.c b/kernel/auditsc.c
> > > > > index 433418d73584..f000fec52360 100644
> > > > > --- a/kernel/auditsc.c
> > > > > +++ b/kernel/auditsc.c
> > > > > @@ -64,6 +64,7 @@
> > > > > #include <uapi/linux/limits.h>
> > > > > #include <uapi/linux/netfilter/nf_tables.h>
> > > > > #include <uapi/linux/openat2.h> // struct open_how
> > > > > +#include <uapi/linux/fanotify.h>
> > > > > 
> > > > > #include "audit.h"
> > > > > 
> > > > > @@ -2899,10 +2900,34 @@ void __audit_log_kern_module(char *name)
> > > > > context->type = AUDIT_KERN_MODULE;
> > > > > }
> > > > > 
> > > > > -void __audit_fanotify(u32 response)
> > > > > +void __audit_fanotify(u32 response, size_t len, char *buf)
> > > > > {
> > > > > -       audit_log(audit_context(), GFP_KERNEL,
> > > > > -               AUDIT_FANOTIFY, "resp=%u", response);
> > > > > +       struct fanotify_response_info_audit_rule *friar;
> > > > > +       size_t c = len;
> > > > > +       char *ib = buf;
> > > > > +
> > > > > +       if (!(len && buf)) {
> > > > > +               audit_log(audit_context(), GFP_KERNEL,
> > > > > AUDIT_FANOTIFY,
> > > > > +                         "resp=%u fan_type=0 fan_info=?",
> > > > > response);
> > > > > +               return;
> > > > > +       }
> > > > > +       while (c >= sizeof(struct fanotify_response_info_header)) {
> > > > > +               friar = (struct fanotify_response_info_audit_rule
> > > > > *)buf;
> > > > 
> > > > Since the only use of this at the moment is the
> > > > fanotify_response_info_rule, why not pass the
> > > > fanotify_response_info_rule struct directly into this function?  We
> > > > can always change it if we need to in the future without affecting
> > > > userspace, and it would simplify the code.
> > > 
> > > Steve, would it make any sense for there to be more than one
> > > FAN_RESPONSE_INFO_AUDIT_RULE header in a message?  Could there be more
> > > than one rule that contributes to a notify reason?  If not, would it be
> > > reasonable to return -EINVAL if there is more than one?
> > 
> > I don't see a reason for sending more than one header. What is more
> > probable is the need to send additional data in that header. I was
> > thinking of maybe bit mapping it in the rule number. But I'd suggest
> > padding the struct just in case it needs expanding some day.
> 
> This doesn't exactly answer my question about multiple rules
> contributing to one decision.

I don't forsee that.
 
> The need for more as yet undefined information sounds like a good reason
> to define a new header if that happens.

It's much better to pad the struct so that the size doesn't change.

> At this point, is it reasonable to throw an error if more than one RULE
> header appears in a message?

It is a write syscall. I'd silently discard everything else and document that 
in the man pages. But the fanotify maintainers should really weigh in on 
this.

> The way I had coded this last patchset was to allow for more than one RULE
> header and each one would get its own record in the event.

I do not forsee a need for this.

> How many rules total are likely to exist?

Could be a thousand. But I already know some missing information we'd like to 
return to user space in an audit event, so the bit mapping on the rule number 
might happen. I'd suggest padding one u32 for future use.

-Steve