[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5BC1D7AD-32C1-4CDC-BA99-F4DABE61EEA3@fb.com>
Date: Wed, 5 Aug 2020 03:59:13 +0000
From: Song Liu <songliubraving@...com>
To: Andrii Nakryiko <andrii.nakryiko@...il.com>
CC: open list <linux-kernel@...r.kernel.org>,
bpf <bpf@...r.kernel.org>, Networking <netdev@...r.kernel.org>,
Alexei Starovoitov <ast@...nel.org>,
Daniel Borkmann <daniel@...earbox.net>,
Kernel Team <Kernel-team@...com>,
john fastabend <john.fastabend@...il.com>,
KP Singh <kpsingh@...omium.org>,
Jesper Dangaard Brouer <brouer@...hat.com>,
Daniel Xu <dlxu@...com>
Subject: Re: [PATCH bpf-next 2/5] libbpf: support BPF_PROG_TYPE_USER programs
> On Aug 4, 2020, at 6:38 PM, Andrii Nakryiko <andrii.nakryiko@...il.com> wrote:
>
> On Mon, Aug 3, 2020 at 6:18 PM Song Liu <songliubraving@...com> wrote:
>>
>>
>>
>>> On Aug 2, 2020, at 6:40 PM, Andrii Nakryiko <andrii.nakryiko@...il.com> wrote:
>>>
>>> On Sat, Aug 1, 2020 at 1:50 AM Song Liu <songliubraving@...com> wrote:
>>>>
>>
>> [...]
>>
>>>
>>>> };
>>>>
>>>> LIBBPF_API int bpf_prog_test_run_xattr(struct bpf_prog_test_run_attr *test_attr);
>>>> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
>>>> index b9f11f854985b..9ce175a486214 100644
>>>> --- a/tools/lib/bpf/libbpf.c
>>>> +++ b/tools/lib/bpf/libbpf.c
>>>> @@ -6922,6 +6922,7 @@ static const struct bpf_sec_def section_defs[] = {
>>>> BPF_PROG_SEC("lwt_out", BPF_PROG_TYPE_LWT_OUT),
>>>> BPF_PROG_SEC("lwt_xmit", BPF_PROG_TYPE_LWT_XMIT),
>>>> BPF_PROG_SEC("lwt_seg6local", BPF_PROG_TYPE_LWT_SEG6LOCAL),
>>>> + BPF_PROG_SEC("user", BPF_PROG_TYPE_USER),
>>>
>>> let's do "user/" for consistency with most other prog types (and nice
>>> separation between prog type and custom user name)
>>
>> About "user" vs. "user/", I still think "user" is better.
>>
>> Unlike kprobe and tracepoint, user prog doesn't use the part after "/".
>> This is similar to "perf_event" for BPF_PROG_TYPE_PERF_EVENT, "xdl" for
>> BPF_PROG_TYPE_XDP, etc. If we specify "user" here, "user/" and "user/xxx"
>> would also work. However, if we specify "user/" here, programs that used
>> "user" by accident will fail to load, with a message like:
>>
>> libbpf: failed to load program 'user'
>>
>> which is confusing.
>
> xdp, perf_event and a bunch of others don't enforce it, that's true,
> they are a bit of a legacy,
I don't see w/o "/" is a legacy thing. BPF_PROG_TYPE_STRUCT_OPS just uses
"struct_ops".
> unfortunately. But all the recent ones do,
> and we explicitly did that for xdp_dev/xdp_cpu, for instance.
> Specifying just "user" in the spec would allow something nonsensical
> like "userargh", for instance, due to this being treated as a prefix.
> There is no harm to require users to do "user/my_prog", though.
I don't see why allowing "userargh" is a problem. Failing "user" is
more confusing. We can probably improve that by a hint like:
libbpf: failed to load program 'user', do you mean "user/"?
But it is pretty silly. "user/something_never_used" also looks weird.
> Alternatively, we could introduce a new convention in the spec,
> something like "user?", which would accept either "user" or
> "user/something", but not "user/" nor "userblah". We can try that as
> well.
Again, I don't really understand why allowing "userblah" is a problem.
We already have "xdp", "xdp_devmap/", and "xdp_cpumap/", they all work
fine so far.
Thanks,
Song
Powered by blists - more mailing lists