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

[Daniel Borkmann <daniel@iogearbox.net>]

On 4/16/21 12:22 AM, Andrii Nakryiko wrote:
> On Thu, Apr 15, 2021 at 3:10 PM Daniel Borkmann <daniel@iogearbox.net> wrote:
>> On 4/15/21 1:58 AM, Andrii Nakryiko wrote:
>>> On Wed, Apr 14, 2021 at 4:32 PM Daniel Borkmann <daniel@iogearbox.net> wrote:
>>>> On 4/15/21 1:19 AM, Andrii Nakryiko wrote:
>>>>> On Wed, Apr 14, 2021 at 3:51 PM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>>>>>> Andrii Nakryiko <andrii.nakryiko@gmail.com> writes:
>>>>>>> On Wed, Apr 14, 2021 at 3:58 AM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>>>>>>>> Andrii Nakryiko <andrii.nakryiko@gmail.com> writes:
>>>>>>>>> On Tue, Apr 6, 2021 at 3:06 AM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>>>>>>>>>> Andrii Nakryiko <andrii.nakryiko@gmail.com> writes:
>>>>>>>>>>> On Sat, Apr 3, 2021 at 10:47 AM Alexei Starovoitov
>>>>>>>>>>> <alexei.starovoitov@gmail.com> wrote:
>>>>>>>>>>>> On Sat, Apr 03, 2021 at 12:38:06AM +0530, Kumar Kartikeya Dwivedi wrote:
>>>>>>>>>>>>> On Sat, Apr 03, 2021 at 12:02:14AM IST, Alexei Starovoitov wrote:
>>>>>>>>>>>>>> On Fri, Apr 2, 2021 at 8:27 AM Kumar Kartikeya Dwivedi <memxor@gmail.com> wrote:
>>>>>>>>>>>>>>> [...]
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> All of these things are messy because of tc legacy. bpf tried to follow tc style
>>>>>>>>>>>>>> with cls and act distinction and it didn't quite work. cls with
>>>>>>>>>>>>>> direct-action is the only
>>>>>>>>>>>>>> thing that became mainstream while tc style attach wasn't really addressed.
>>>>>>>>>>>>>> There were several incidents where tc had tens of thousands of progs attached
>>>>>>>>>>>>>> because of this attach/query/index weirdness described above.
>>>>>>>>>>>>>> I think the only way to address this properly is to introduce bpf_link style of
>>>>>>>>>>>>>> attaching to tc. Such bpf_link would support ingress/egress only.
>>>>>>>>>>>>>> direction-action will be implied. There won't be any index and query
>>>>>>>>>>>>>> will be obvious.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Note that we already have bpf_link support working (without support for pinning
>>>>>>>>>>>>> ofcourse) in a limited way. The ifindex, protocol, parent_id, priority, handle,
>>>>>>>>>>>>> chain_index tuple uniquely identifies a filter, so we stash this in the bpf_link
>>>>>>>>>>>>> and are able to operate on the exact filter during release.
>>>>>>>>>>>>
>>>>>>>>>>>> Except they're not unique. The library can stash them, but something else
>>>>>>>>>>>> doing detach via iproute2 or their own netlink calls will detach the prog.
>>>>>>>>>>>> This other app can attach to the same spot a different prog and now
>>>>>>>>>>>> bpf_link__destroy will be detaching somebody else prog.
>>>>>>>>>>>>
>>>>>>>>>>>>>> So I would like to propose to take this patch set a step further from
>>>>>>>>>>>>>> what Daniel said:
>>>>>>>>>>>>>> int bpf_tc_attach(prog_fd, ifindex, {INGRESS,EGRESS}):
>>>>>>>>>>>>>> and make this proposed api to return FD.
>>>>>>>>>>>>>> To detach from tc ingress/egress just close(fd).
>>>>>>>>>>>>>
>>>>>>>>>>>>> You mean adding an fd-based TC API to the kernel?
>>>>>>>>>>>>
>>>>>>>>>>>> yes.
>>>>>>>>>>>
>>>>>>>>>>> I'm totally for bpf_link-based TC attachment.
>>>>>>>>>>>
>>>>>>>>>>> But I think *also* having "legacy" netlink-based APIs will allow
>>>>>>>>>>> applications to handle older kernels in a much nicer way without extra
>>>>>>>>>>> dependency on iproute2. We have a similar situation with kprobe, where
>>>>>>>>>>> currently libbpf only supports "modern" fd-based attachment, but users
>>>>>>>>>>> periodically ask questions and struggle to figure out issues on older
>>>>>>>>>>> kernels that don't support new APIs.
>>>>>>>>>>
>>>>>>>>>> +1; I am OK with adding a new bpf_link-based way to attach TC programs,
>>>>>>>>>> but we still need to support the netlink API in libbpf.
>>>>>>>>>>
>>>>>>>>>>> So I think we'd have to support legacy TC APIs, but I agree with
>>>>>>>>>>> Alexei and Daniel that we should keep it to the simplest and most
>>>>>>>>>>> straightforward API of supporting direction-action attachments and
>>>>>>>>>>> setting up qdisc transparently (if I'm getting all the terminology
>>>>>>>>>>> right, after reading Quentin's blog post). That coincidentally should
>>>>>>>>>>> probably match how bpf_link-based TC API will look like, so all that
>>>>>>>>>>> can be abstracted behind a single bpf_link__attach_tc() API as well,
>>>>>>>>>>> right? That's the plan for dealing with kprobe right now, btw. Libbpf
>>>>>>>>>>> will detect the best available API and transparently fall back (maybe
>>>>>>>>>>> with some warning for awareness, due to inherent downsides of legacy
>>>>>>>>>>> APIs: no auto-cleanup being the most prominent one).
>>>>>>>>>>
>>>>>>>>>> Yup, SGTM: Expose both in the low-level API (in bpf.c), and make the
>>>>>>>>>> high-level API auto-detect. That way users can also still use the
>>>>>>>>>> netlink attach function if they don't want the fd-based auto-close
>>>>>>>>>> behaviour of bpf_link.
>>>>>>>>>
>>>>>>>>> So I thought a bit more about this, and it feels like the right move
>>>>>>>>> would be to expose only higher-level TC BPF API behind bpf_link. It
>>>>>>>>> will keep the API complexity and amount of APIs that libbpf will have
>>>>>>>>> to support to the minimum, and will keep the API itself simple:
>>>>>>>>> direct-attach with the minimum amount of input arguments. By not
>>>>>>>>> exposing low-level APIs we also table the whole bpf_tc_cls_attach_id
>>>>>>>>> design discussion, as we now can keep as much info as needed inside
>>>>>>>>> bpf_link_tc (which will embed bpf_link internally as well) to support
>>>>>>>>> detachment and possibly some additional querying, if needed.
>>>>>>>>
>>>>>>>> But then there would be no way for the caller to explicitly select a
>>>>>>>> mechanism? I.e., if I write a BPF program using this mechanism targeting
>>>>>>>> a 5.12 kernel, I'll get netlink attachment, which can stick around when
>>>>>>>> I do bpf_link__disconnect(). But then if the kernel gets upgraded to
>>>>>>>> support bpf_link for TC programs I'll suddenly transparently get
>>>>>>>> bpf_link and the attachments will go away unless I pin them. This
>>>>>>>> seems... less than ideal?
>>>>>>>
>>>>>>> That's what we are doing with bpf_program__attach_kprobe(), though.
>>>>>>> And so far I've only seen people (privately) saying how good it would
>>>>>>> be to have bpf_link-based TC APIs, doesn't seem like anyone with a
>>>>>>> realistic use case prefers the current APIs. So I suspect it's not
>>>>>>> going to be a problem in practice. But at least I'd start there and
>>>>>>> see how people are using it and if they need anything else.
>>>>>>
>>>>>> *sigh* - I really wish you would stop arbitrarily declaring your own use
>>>>>> cases "realistic" and mine (implied) "unrealistic". Makes it really hard
>>>>>> to have a productive discussion...
>>>>>
>>>>> Well (sigh?..), this wasn't my intention, sorry you read it this way.
>>>>> But we had similar discussions when I was adding bpf_link-based XDP
>>>>> attach APIs. And guess what, now I see that samples/bpf/whatever_xdp
>>>>> is switched to bpf_link-based XDP, because that makes everything
>>>>> simpler and more reliable. What I also know is that in production we
>>>>> ran into multiple issues with anything that doesn't auto-detach on
>>>>> process exit/crash (unless pinned explicitly, of course). And that
>>>>> people that are trying to use TC right now are saying how having
>>>>> bpf_link-based TC APIs would make everything *simpler* and *safer*. So
>>>>> I don't know... I understand it might be convenient in some cases to
>>>>> not care about a lifetime of BPF programs you are attaching, but then
>>>>> there are usually explicit and intentional ways to achieve at least
>>>>> similar behavior with safety by default.
>>>>
>>>> [...]
>>>>
>>>>    >>> There are many ways to skin this cat. I'd prioritize bpf_link-based TC
>>>>    >>> APIs to be added with legacy TC API as a fallback.
>>>>
>>>> I think the problem here is though that this would need to be deterministic
>>>> when upgrading from one kernel version to another where we don't use the
>>>> fallback anymore, e.g. in case of Cilium we always want to keep the progs
>>>> attached to allow headless updates on the agent, meaning, traffic keeps
>>>> flowing through the BPF datapath while in user space, our agent restarts
>>>> after upgrade, and atomically replaces the BPF progs once up and running
>>>> (we're doing this for the whole range of 4.9 to 5.x kernels that we support).
>>>> While we use the 'simple' api that is discussed here internally in Cilium,
>>>> this attach behavior would have to be consistent, so transparent fallback
>>>> inside libbpf on link vs non-link availability won't work (at least in our
>>>> case).
>>>
>>> What about pinning? It's not exactly the same, but bpf_link could
>>> actually pin a BPF program, if using legacy TC, and pin bpf_link, if
>>> using bpf_link-based APIs. Of course before switching from iproute2 to
>>> libbpf APIs you'd need to design your applications to use pinning
>>> instead of relying implicitly on permanently attached BPF program.
>>
>> All the progs we load from Cilium in a K8s setting w/ Pods, we could have easily
>> over 100 loaded at the same time on a node, and we template the per Pod ones, so
>> the complexity of managing those pinned lifecycles from the agent and dealing with
>> the semantic/fallback differences between kernels feels probably not worth the
>> gain. So if there would be a libbpf tc simplified attach API, I'd for the time
>> being stick to the existing aka legacy means.
> 
> Sure. Then what do you think about keeping only low-level TC APIs, and
> in the future add bpf_program__attach_tc(), which will use
> bpf_link-based one. It seems like it's not worth it to pretend we have
> bpf_link-based semantics with "legacy" current TC APIs. Similarly how
> we have a low-level XDP attach API, and bpf_link-based (only)
> bpf_program__attach_xdp().

I think that's okay. I guess question is what do we define as initial scope for
the low-level TC API. cls_bpf w/ fixed direct-action mode + fixed eth_p_all,
allowing to flexibly specify handle / priority or a block_index feels reasonable.