Re: [PATCH bpf-next 1/4] bpf, x64: Fix tailcall infinite loop caused by freplace

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> writes:

> On Tue, Aug 27, 2024 at 5:48 AM Leon Hwang <leon.hwang@xxxxxxxxx> wrote:
>>
>> > I wonder if disallowing to freplace programs when
>> > replacement.tail_call_reachable != replaced.tail_call_reachable
>> > would be a better option?
>> >
>>
>> This idea is wonderful.
>>
>> We can disallow attaching tail_call_reachable freplace prog to
>> not-tail_call_reachable bpf prog. So, the following 3 cases are allowed.
>>
>> 1. attach tail_call_reachable freplace prog to tail_call_reachable bpf prog.
>> 2. attach not-tail_call_reachable freplace prog to tail_call_reachable
>> bpf prog.
>> 3. attach not-tail_call_reachable freplace prog to
>> not-tail_call_reachable bpf prog.
>
> I think it's fine to disable freplace and tail_call combination
> altogether.

In the libxdp dispatcher we rely on the fact that an freplace program is
equivalent to a directly attached XDP program. And we've definitely seen
people using tail calls along with the libxdp dispatcher (e.g.,
https://github.com/xdp-project/xdp-tools/issues/377), so I don't think
it's a good idea to disable it entirely.

I think restricting the combinations should be fine, though - the libxdp
dispatcher will not end up in a tail call map unless someone is going
out of their way to do weird things :)

-Toke






[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux