Re: [PATCH v2 net-next 1/2] cls_bpf: introduce integrated actions

[Alexei Starovoitov <ast@plumgrid.com>]

On 9/17/15 6:13 AM, Daniel Borkmann wrote:
> Hi Jamal,
>
> On 09/17/2015 02:37 PM, Jamal Hadi Salim wrote:
>> On 09/16/15 02:05, Alexei Starovoitov wrote:
>>> From: Daniel Borkmann <daniel@iogearbox.net>
>>>
>>> Often cls_bpf classifier is used with single action drop attached.
>>> Optimize this use case and let cls_bpf return both classid and action.
>>> For backwards compatibility reasons enable this feature under
>>> TCA_BPF_FLAG_ACT_DIRECT flag.
>>>
>>
>> This is going off in a different direction really.
>> You are replicating the infrastructure inside bpf.
>
> Hmm, I don't really agree. With cls_bpf you have non-linear
> classifications as opposed to walking a chain of classifiers:
> worst case, I have to walk through N classifiers just to find
> out that the last one matches that I need to drop - this doesn't
> scale at all. Given that we can make this decision right here,
> we can use this fact and have simple return codes provided as
> well. It only supplements non-linear classification that was
> from the very beginning of cls_bpf a core part of it.

I don't see the replication either. May be the commit log was
misread as bpf program now executes the actions and bypasses
tcf_exts_exec() ? Well, that may be interesting idea for
the future, but that's not what the patch is doing.
With this patch cls_bpf can return single integer like
TC_ACT_SHOT/TC_ACT_OK that gact/act_bpf can already do as
an _optimization_ to avoid extra hops. To do full-fledged
action chaining the tcf_exts_exec() is used.