Re: [PATCH bpf-next v2 04/11] bpf: Hold RCU read lock in bpf_prog_ksym_find

[Eduard Zingerman <eddyz87@gmail.com>]

Kumar Kartikeya Dwivedi <memxor@gmail.com> writes:

[...]

>> --- a/kernel/bpf/core.c
>> +++ b/kernel/bpf/core.c
>> @@ -782,7 +782,11 @@ bool is_bpf_text_address(unsigned long addr)
>>
>>  struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
>>  {
>> -       struct bpf_ksym *ksym = bpf_ksym_find(addr);
>> +       struct bpf_ksym *ksym;
>> +
>> +       rcu_read_lock();
>> +       ksym = bpf_ksym_find(addr);
>> +       rcu_read_unlock();
>>
>>         return ksym && ksym->prog ?
>>                container_of(ksym, struct bpf_prog_aux, ksym)->prog :
>
> This isn't right, we need to have the read section open around ksym
> access as well.
> We can end the section and return the prog pointer.
> The caller is responsible to ensure prog outlives RCU protection, or
> otherwise hold it if necessary for prog's lifetime.
>
> We're using this to pick programs who have an active stack frame, so
> they aren't going away.
> But the ksym access itself needs to happen under correct protection.
>
> I can fix it in a respin, whatever is best.

Are rcu_read_{lock,unlock} necessary in core.c:search_bpf_extables()
after this change?
Also, helpers.c:bpf_stack_walker.c does not have lock/unlock in it,
this patch needs a fixes tag for commit f18b03fabaa9 ("bpf: Implement BPF exceptions")?