linux-kernel - Re: [PATCH bpf-next v4 0/8] MAC and Audit policy using eBPF (KRSI)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Sat, 22 Feb 2020 00:09:33 +0100
From:   KP Singh <kpsingh@...omium.org>
To:     Casey Schaufler <casey@...aufler-ca.com>
Cc:     Linux Security Module list 
        <linux-security-module@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>, bpf <bpf@...r.kernel.org>,
        James Morris <jmorris@...ei.org>,
        Kees Cook <keescook@...omium.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>
Subject: Re: [PATCH bpf-next v4 0/8] MAC and Audit policy using eBPF (KRSI)

Thanks Casey,

I appreciate your quick responses!

On 21-Feb 14:31, Casey Schaufler wrote:
> On 2/21/2020 11:41 AM, KP Singh wrote:
> > On 21-Feb 11:19, Casey Schaufler wrote:
> >> On 2/20/2020 9:52 AM, KP Singh wrote:
> >>> From: KP Singh <kpsingh@...gle.com>
> >> Again, apologies for the CC list trimming.
> >>
> >>> # v3 -> v4
> >>>
> >>>   https://lkml.org/lkml/2020/1/23/515
> >>>
> >>> * Moved away from allocating a separate security_hook_heads and adding a
> >>>   new special case for arch_prepare_bpf_trampoline to using BPF fexit
> >>>   trampolines called from the right place in the LSM hook and toggled by
> >>>   static keys based on the discussion in:
> >>>
> >>>     https://lore.kernel.org/bpf/CAG48ez25mW+_oCxgCtbiGMX07g_ph79UOJa07h=o_6B6+Q-u5g@mail.gmail.com/
> >>>
> >>> * Since the code does not deal with security_hook_heads anymore, it goes
> >>>   from "being a BPF LSM" to "BPF program attachment to LSM hooks".

[...]

> >> likely harmful.
> > We will be happy to document each of the macros in detail. Do note a
> > few things here:
> >
> > * There is really nothing magical about them though,
> 
> 
> +#define LSM_HOOK_void(NAME, ...) \
> +	noinline void bpf_lsm_##NAME(__VA_ARGS__) {}
> +
> +#include <linux/lsm_hook_names.h>
> +#undef LSM_HOOK
> 
> I haven't seen anything this ... novel ... in a very long time.

This is not "novel", it's a fairly common pattern followed in tracing:

For example, the TRACE_INCLUDE macro which is used for tracepoints:

  include/trace/define_trace.h

and used in:

  * include/trace/bpf_probe.h

    https://github.com/torvalds/linux/blob/master/include/trace/bpf_probe.h#L110

  * include/trace/perf.h

    https://github.com/torvalds/linux/blob/master/include/trace/perf.h#L90

  * include/trace/trace_events.h

    https://github.com/torvalds/linux/blob/master/include/trace/trace_events.h#L402

> I see why you want to do this, but you're tying the two sets
> of code together unnaturally. When (not if) the two sets diverge
> you're going to be introducing another clever way to deal with

I don't fully understand what "two sets diverge means" here. All the
BPF headers need is the name, return type and the args. This is the
same information which is needed by the call_{int, void}_hooks and the
LSM declarataions (i.e. security_hook_heads and
security_list_options).

> the special case.
> 
> It's not that I don't understand what you're doing. It's that
> I don't like what you're doing. Explanation doesn't make me like
> it better.

As I have previously said, we will be happy to (and have already)
updated our approach based on the consensus we arrive at here. The
best outcome would be to not sacrifice performance as the LSM hooks
are called from various performance critical code-paths.

It would be great to know the maintainers' (BPF and Security)
perspective on this as well.

- KP

> 
> >  the LSM hooks are
> >   collectively declared in lsm_hook_names.h and are used to delcare
> >   the security_list_options and security_hook_heads for the LSM
> >   framework (this was previously maitained in two different places):
> >
> >   For BPF, they declare:
> >
> >     * bpf_lsm_<name> attachment points and their prototypes.
> >     * A static key (bpf_lsm_key_<name>) to enable and disable these
> >        hooks with a function to set its value i.e.
> >        (bpf_lsm_<name>_set_enabled).
> >
> > * We have kept the BPF related macros out of security/.
> > * All the BPF calls in the LSM infrastructure are guarded by
> >   CONFIG_BPF_LSM (there are only two main calls though, i.e.
> >   call_int_hook, call_void_hook).
> >
> > Honestly, the macros aren't any more complicated than
> > call_int_progs/call_void_progs.
> >
> > - KP
> >
> >> Would you please drop the excessive optimization? I understand
> >> that there's been a lot of discussion and debate about it,
> >> but this implementation is out of control, disruptive, and
> >> dangerous to the code around it.
> >>
> >>
>