linux-kernel - Re: [PATCH bpf-next 3/5] libbpf: add low level TC-BPF API

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]

Message-ID: <5d59b5ee-a21e-1860-e2e5-d03f89306fd8@iogearbox.net>
Date:   Fri, 2 Apr 2021 02:19:29 +0200
From:   Daniel Borkmann <daniel@...earbox.net>
To:     Kumar Kartikeya Dwivedi <memxor@...il.com>
Cc:     Andrii Nakryiko <andrii.nakryiko@...il.com>,
        bpf <bpf@...r.kernel.org>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        Toke Høiland-Jørgensen <toke@...hat.com>,
        Alexei Starovoitov <ast@...nel.org>,
        Andrii Nakryiko <andrii@...nel.org>,
        Martin KaFai Lau <kafai@...com>,
        Song Liu <songliubraving@...com>, Yonghong Song <yhs@...com>,
        John Fastabend <john.fastabend@...il.com>,
        KP Singh <kpsingh@...nel.org>, Shuah Khan <shuah@...nel.org>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Jesper Dangaard Brouer <hawk@...nel.org>,
        Peter Zijlstra <peterz@...radead.org>,
        open list <linux-kernel@...r.kernel.org>,
        Networking <netdev@...r.kernel.org>,
        "open list:KERNEL SELFTEST FRAMEWORK" 
        <linux-kselftest@...r.kernel.org>
Subject: Re: [PATCH bpf-next 3/5] libbpf: add low level TC-BPF API

On 3/31/21 11:44 AM, Kumar Kartikeya Dwivedi wrote:
> On Wed, Mar 31, 2021 at 02:55:47AM IST, Daniel Borkmann wrote:
>> Do we even need the _block variant? I would rather prefer to take the chance
>> and make it as simple as possible, and only iff really needed extend with
>> other APIs, for example:
> 
> The block variant can be dropped, I'll use the TC_BLOCK/TC_DEV alternative which
> sets parent_id/ifindex properly.
> 
>>    bpf_tc_attach(prog_fd, ifindex, {INGRESS,EGRESS});
>>
>> Internally, this will create the sch_clsact qdisc & cls_bpf filter instance
>> iff not present yet, and attach to a default prio 1 handle 1, and _always_ in
>> direct-action mode. This is /as simple as it gets/ and we don't need to bother
>> users with more complex tc/cls_bpf internals unless desired. For example,
>> extended APIs could add prio/parent so that multi-prog can be attached to a
>> single cls_bpf instance, but even that could be a second step, imho.
> 
> I am not opposed to clsact qdisc setup if INGRESS/EGRESS is supplied (not sure
> how others feel about it).

What speaks against it? It would be 100% clear from API side where the prog is
being attached. Same as with tc cmdline where you specify 'ingress'/'egress'.

> We could make direct_action mode default, and similarly choose prio

To be honest, I wouldn't even support a mode from the lib/API side where direct_action
is not set. It should always be forced to true. Everything else is rather broken
setup-wise, imho, since it won't scale. We added direct_action a bit later to the
kernel than original cls_bpf, but if I would do it again today, I'd make it the
only available option. I don't see a reasonable use-case where you have it to false.

> as 1 by default instead of letting the kernel do it. Then you can just pass in
> NULL for bpf_tc_cls_opts and be close to what you're proposing. For protocol we
> can choose ETH_P_ALL by default too if the user doesn't set it.

Same here with ETH_P_ALL, I'm not sure anyone uses anything other than ETH_P_ALL,
so yes, that should be default.

> With these modifications, the equivalent would look like
> 	bpf_tc_cls_attach(prog_fd, TC_DEV(ifindex, INGRESS), NULL, &id);

Few things compared to bpf_tc_attach(prog_fd, ifindex, {INGRESS,EGRESS}):

1) nit, but why even 'cls' in the name. I think we shouldn't expose such old-days
    tc semantics to a user. Just bpf_tc_attach() is cleaner/simpler to understand.
2) What's the 'TC_DEV(ifindex, INGRESS)' macro doing exactly? Looks unnecessary,
    why not regular args to the API?
3) Exposed bpf_tc_attach() API could internally call a bpf_tc_attach_opts() API
    with preset defaults, and the latter could have all the custom bits if the user
    needs to go beyond the simple API, so from your bpf_tc_cls_attach() I'd also
    drop the NULL.
4) For the simple API I'd likely also drop the id (you could have a query API if
    needed).

> So as long as the user doesn't care about other details, they can just pass opts
> as NULL.