Re: [PATCH/RFC bpf-next 03/16] bpf: split read liveness into REG_LIVE_READ64 and REG_LIVE_READ32

[Jiong Wang <jiong.wang@netronome.com>]

Jann Horn writes:

> On Tue, Mar 26, 2019 at 7:06 PM Jiong Wang <jiong.wang@netronome.com> wrote:
>>
>> In previous patch, we have split register arg type for sub-register read,
>> but haven't touch read liveness.
>>
>> This patch further split read liveness into REG_LIVE_READ64 and
>> REG_LIVE_READ32. Liveness propagation code are updated accordingly.
>>
>> After this split, customized actions could be defined when propagating full
>> register read (REG_LIVE_READ64) or sub-register read (REG_LIVE_READ32).
>>
>> Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
> [...]
>> @@ -1374,7 +1374,8 @@ static int check_stack_read(struct bpf_verifier_env *env,
>>                         return -EACCES;
>>                 }
>>                 mark_reg_read(env, &reg_state->stack[spi].spilled_ptr,
>> -                             reg_state->stack[spi].spilled_ptr.parent);
>> +                             reg_state->stack[spi].spilled_ptr.parent,
>> +                             size == BPF_REG_SIZE);
>
> Isn't it possible to use a 4-byte read on the upper half of an 8-byte
> stack slot?

I think that's fine, and is irrelevant with zero-extension on register.

If it is a 8-byte stack slot comes from spill of register, then the
definition of the register should have been marked as needing
zero-extension if that register was generated by sub-register write.

Regards,
Jiong

>
>>                 if (value_regno >= 0) {
>>                         if (zeros == size) {
>>                                 /* any size read into register is zero extended,
>> @@ -2220,7 +2221,8 @@ static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
>>                  * the whole slot to be marked as 'read'
>>                  */
>>                 mark_reg_read(env, &state->stack[spi].spilled_ptr,
>> -                             state->stack[spi].spilled_ptr.parent);
>> +                             state->stack[spi].spilled_ptr.parent,
>> +                             access_size == BPF_REG_SIZE);
>
> Same thing as above.
>
>>         }
>>         return update_stack_depth(env, state, off);
>>  }
> [...]