Re: [PATCH bpf-next v10 2/6] bpf: Add bpf_prog_run_array_sleepable()

["Alexei Starovoitov" <alexei.starovoitov@gmail.com>]

On Fri Apr 17, 2026 at 10:22 AM PDT, Mykyta Yatsenko wrote:
>
>
> On 4/16/26 3:40 PM, Alexei Starovoitov wrote:
>> On Wed, Apr 15, 2026 at 6:41 PM Alexei Starovoitov
>> <alexei.starovoitov@gmail.com> wrote:
>> One more thing..
>> Please go back to my 'do your homework' request from v9
>> and answer the question why it's not safe to do bpf_prog_run
>> on dummy prog like this patch does.
>> I missed it yesterday and only complained about the waste of cpu cycles.
>> It's more than that.
>> 
>> Do not send v11.
>
> I see that very similar bug was fixed in 7dc211c1159d ("bpf: Fix invalid 
> prog->stats access when update_effective_progs fails"), the problem is 
> the same - NULL field access on the dummy prog. So the approach to just 
> allow dummy to run further is fragile, because other NULL fields may be 
> accessed.

exactly. so far we've hit 'active == NULL' and 'stats == NULL',
and 'stats == NULL' had to be workaround-ed like that in that commit.
There is also 'aux'. I hope it's not used at run-time, but let's not take chances.
Filter dummy prog as immediate skip.