Re: [PATCH bpf-next] bpf: Fix a verifier message for alloc size helper arg

[Brendan Jackman]

Sorry, duplicate - seems I had my mail client in HTML mode the first
time around.

On Tue, 12 Jan 2021 at 14:14, KP Singh <kpsingh@xxxxxxxxxx> wrote:
>
> On Tue, Jan 12, 2021 at 1:39 PM Brendan Jackman <jackmanb@xxxxxxxxxx> wrote:
> >
> > The error message here is misleading, the argument will be rejected
> > unless it is a known constant.
> >
> > Signed-off-by: Brendan Jackman <jackmanb@xxxxxxxxxx>
> > ---
> >  kernel/bpf/verifier.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 17270b8404f1..5534e667bdb1 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -4319,7 +4319,7 @@ static int check_func_arg(struct bpf_verifier_env *env, u32 arg,
> >                         err = mark_chain_precision(env, regno);
> >         } else if (arg_type_is_alloc_size(arg_type)) {
> >                 if (!tnum_is_const(reg->var_off)) {
> > -                       verbose(env, "R%d unbounded size, use 'var &= const' or 'if (var < const)'\n",
>
> Can you check if:
>
> int var = 1000;
> var += 1;
>
> if (var < 2000)
>    // call helper
>
> and then using var in the argument works? If so, the existing error
> message would be correct.

I think that would work because var is already a known constant before
the conditional.. but the error message is still wrong, the `if (var <
2000)` is irrelevant. If var was not already a known constant (e.g.
came from the return value of a bpf_probe_read_kernel_str) it would
fail verification.