On Thu, 2023-01-26 at 11:03 -0800, Andrii Nakryiko wrote:
> On Thu, Jan 26, 2023 at 3:41 AM Ilya Leoshkevich <iii@xxxxxxxxxxxxx>
> wrote:
> >
> > On Wed, 2023-01-25 at 16:26 -0800, Andrii Nakryiko wrote:
> > > On Wed, Jan 25, 2023 at 1:39 PM Ilya Leoshkevich
> > > <iii@xxxxxxxxxxxxx>
> > > wrote:
> > > >
> > > > Loading programs that use bpf_usdt_arg() on s390x fails with:
> > > >
> > > > ; switch (arg_spec->arg_type) {
> > > > 139: (61) r1 = *(u32 *)(r2 +8)
> > > > R2 unbounded memory access, make sure to bounds check any
> > > > such
> > > > access
> > >
> > > can you show a bit longer log? we shouldn't just use
> > > bpf_probe_read_kernel for this. I suspect strategically placed
> > > barrier_var() calls will solve this. This is usually an issue
> > > with
> > > compiler reordering operations and doing actual check after it
> > > already
> > > speculatively adjusted pointer (which is technically safe and ok
> > > if
> > > we
> > > never deref that pointer, but verifier doesn't recognize such
> > > pattern)
> >
> > The full log is here:
> >
> > https://gist.github.com/iii-i/b6149ee99b37078ec920ab1d3bb45134
[...]
> > --- a/tools/lib/bpf/usdt.bpf.h
> > +++ b/tools/lib/bpf/usdt.bpf.h
> > @@ -130,7 +130,10 @@ int bpf_usdt_arg(struct pt_regs *ctx, __u64
> > arg_num, long *res)
> > if (!spec)
> > return -ESRCH;
> >
> > - if (arg_num >= BPF_USDT_MAX_ARG_CNT || arg_num >= spec-
> > > arg_cnt)
> > + if (arg_num >= BPF_USDT_MAX_ARG_CNT)
> > + return -ENOENT;
> > + barrier_var(arg_num);
> > + if (arg_num >= spec->arg_cnt)
> > return -ENOENT;
> >
> > arg_spec = &spec->args[arg_num];
> >
> > I can use this in v2 if it looks good.
>
> arg_num -> spec->arg_cnt is "real" check, arg_num >=
> BPF_USDT_MAX_ARG_CNT is more to satisfy verifier (we know that
> spec->arg_cnt won't be >= BPF_USDT_MAX_ARG_CNT). Let's swap two
> checks
> in order and keep BPF_USDT_MAX_ARG_CNT close to spec->args[arg_num]
> use? And if barrier_var() is necessary, then so be it.
Unfortunately just swapping did not help, so let's use the barrier.
> > Btw, I looked at the barrier_var() definition:
> >
> > #define barrier_var(var) asm volatile("" : "=r"(var) : "0"(var))
> >
> > and I'm curious why it's not defined like this:
> >
> > #define barrier_var(var) asm volatile("" : "+r"(var))
> >
> > which is a bit simpler?
> > >
>
> no reason, just unfamiliarity with embedded asm back then, we can
> update it we they are equivalent
Thanks! I will add a simplification in v2 then.
