Date:   Mon, 19 Apr 2021 23:57:23 +0200
From:   Daniel Borkmann <daniel@...earbox.net>
To:     Toke Høiland-Jørgensen <toke@...hat.com>,
        Kumar Kartikeya Dwivedi <memxor@...il.com>,
        bpf@...r.kernel.org
Cc:     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>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Jesper Dangaard Brouer <hawk@...nel.org>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        linux-kernel@...r.kernel.org, netdev@...r.kernel.org
Subject: Re: [PATCH bpf-next v2 3/4] libbpf: add low level TC-BPF API

On 4/19/21 11:43 PM, Toke Høiland-Jørgensen wrote:
> Daniel Borkmann <daniel@...earbox.net> writes:
>> On 4/19/21 2:18 PM, Kumar Kartikeya Dwivedi wrote:
>>> This adds functions that wrap the netlink API used for adding,
>>> manipulating, and removing traffic control filters. These functions
>>> operate directly on the loaded prog's fd, and return a handle to the
>>> filter using an out parameter named id.
>>>
>>> The basic featureset is covered to allow for attaching, manipulation of
>>> properties, and removal of filters. Some additional features like
>>> TCA_BPF_POLICE and TCA_RATE for tc_cls have been omitted. These can
>>> added on top later by extending the bpf_tc_cls_opts struct.
>>>
>>> Support for binding actions directly to a classifier by passing them in
>>> during filter creation has also been omitted for now. These actions have
>>> an auto clean up property because their lifetime is bound to the filter
>>> they are attached to. This can be added later, but was omitted for now
>>> as direct action mode is a better alternative to it, which is enabled by
>>> default.
>>>
>>> An API summary:
>>>
>>> bpf_tc_act_{attach, change, replace} may be used to attach, change, and
>>
>> typo on bpf_tc_act_{...} ?
>>                  ^^^
>>> replace SCHED_CLS bpf classifier. The protocol field can be set as 0, in
>>> which case it is subsitituted as ETH_P_ALL by default.
>>
>> Do you have an actual user that needs anything other than ETH_P_ALL? Why is it
>> even needed? Why not stick to just ETH_P_ALL?
>>
>>> The behavior of the three functions is as follows:
>>>
>>> attach = create filter if it does not exist, fail otherwise
>>> change = change properties of the classifier of existing filter
>>> replace = create filter, and replace any existing filter
>>
>> This touches on tc oddities quite a bit. Why do we need to expose them? Can't we
>> simplify/abstract this e.g. i) create or update instance, ii) delete instance,
>> iii) get instance ? What concrete use case do you have that you need those three
>> above?
>>
>>> bpf_tc_cls_detach may be used to detach existing SCHED_CLS
>>> filter. The bpf_tc_cls_attach_id object filled in during attach,
>>> change, or replace must be passed in to the detach functions for them to
>>> remove the filter and its attached classififer correctly.
>>>
>>> bpf_tc_cls_get_info is a helper that can be used to obtain attributes
>>> for the filter and classififer. The opts structure may be used to
>>> choose the granularity of search, such that info for a specific filter
>>> corresponding to the same loaded bpf program can be obtained. By
>>> default, the first match is returned to the user.
>>>
>>> Examples:
>>>
>>> 	struct bpf_tc_cls_attach_id id = {};
>>> 	struct bpf_object *obj;
>>> 	struct bpf_program *p;
>>> 	int fd, r;
>>>
>>> 	obj = bpf_object_open("foo.o");
>>> 	if (IS_ERR_OR_NULL(obj))
>>> 		return PTR_ERR(obj);
>>>
>>> 	p = bpf_object__find_program_by_title(obj, "classifier");
>>> 	if (IS_ERR_OR_NULL(p))
>>> 		return PTR_ERR(p);
>>>
>>> 	if (bpf_object__load(obj) < 0)
>>> 		return -1;
>>>
>>> 	fd = bpf_program__fd(p);
>>>
>>> 	r = bpf_tc_cls_attach(fd, if_nametoindex("lo"),
>>> 			      BPF_TC_CLSACT_INGRESS,
>>> 			      NULL, &id);
>>> 	if (r < 0)
>>> 		return r;
>>>
>>> ... which is roughly equivalent to (after clsact qdisc setup):
>>>     # tc filter add dev lo ingress bpf obj foo.o sec classifier da
>>>
>>> ... as direct action mode is always enabled.
>>>
>>> If a user wishes to modify existing options on an attached classifier,
>>> bpf_tc_cls_change API may be used.
>>>
>>> Only parameters class_id can be modified, the rest are filled in to
>>> identify the correct filter. protocol can be left out if it was not
>>> chosen explicitly (defaulting to ETH_P_ALL).
>>>
>>> Example:
>>>
>>> 	/* Optional parameters necessary to select the right filter */
>>> 	DECLARE_LIBBPF_OPTS(bpf_tc_cls_opts, opts,
>>> 			    .handle = id.handle,
>>> 			    .priority = id.priority,
>>> 			    .chain_index = id.chain_index)
>>
>> Why do we need chain_index as part of the basic API?
>>
>>> 	opts.class_id = TC_H_MAKE(1UL << 16, 12);
>>> 	r = bpf_tc_cls_change(fd, if_nametoindex("lo"),
>>> 			      BPF_TC_CLSACT_INGRESS,
>>> 			      &opts, &id);
>>
>> Also, I'm not sure whether the prefix should even be named  bpf_tc_cls_*() tbh,
>> yes, despite being "low level" api. I think in the context of bpf we should stop
>> regarding this as 'classifier' and 'action' objects since it's really just a
>> single entity and not separate ones. It's weird enough to explain this concept
>> to new users and if a libbpf based api could cleanly abstract it, I would be all
>> for it. I don't think we need to map 1:1 the old tc legacy even in the low level
>> api, tbh, as it feels backwards. I think the 'handle' & 'priority' bits are okay,
>> but I would remove the others.
> 
> Hmm, I'm OK with dropping the TC oddities (including the cls_ in the
> name), but I think we should be documenting it so that users that do
> come from TC will not be completely lost :)

Yeah, that sounds good to me. :) All I'm trying to say is that /we/ are used to the
terminology and quirks that come with it, but I'm hoping that such API will be used
by *new* folks who have zero context on the underlying details, and they also really
shouldn't have to care. Even if on the lower level we expose handle/priority at least,
the API should be designed with a mindset of no prior tc experience required. Simple
and easy to use. I think the handle/priority concept can be explained/documented fairly
straight forward. There is a chance to hide all the nasty historic/legacy details in
something clean, easy to use and scaleable (implied by the da mode), we should use this
opportunity. ;)

Thanks,
Daniel

Powered by blists - more mailing lists