Re: [PATCH bpf] bpf: fix potential race in tail call compatibility check

["Toke Høiland-Jørgensen" <toke@redhat.com>]

Alexei Starovoitov <alexei.starovoitov@gmail.com> writes:

> On Thu, Oct 21, 2021 at 11:40 AM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>> +       map_type = READ_ONCE(array->aux->type);
>> +       if (!map_type) {
>> +               /* There's no owner yet where we could check for compatibility.
>> +                * Do an atomic swap to prevent racing with another invocation
>> +                * of this branch (via simultaneous map_update syscalls).
>>                  */
>> -               array->aux->type  = fp->type;
>> -               array->aux->jited = fp->jited;
>> +               if (cmpxchg(&array->aux->type, 0, prog_type))
>> +                       return false;
>
> Other fields might be used in the compatibility check in the future.
> This hack is too fragile.
> Just use a spin_lock.

Well, yeah, we're adding another field for xdp_mb. I was just going to
eat more bits of the 'type' field, but OK, can switch to a spinlock
instead :)

-Toke