Re: [PATCH] arm64/mm: Reject invalid NUMA option

From: Steven Rostedt
Date: Mon Apr 27 2020 - 23:12:59 EST


On Mon, 27 Apr 2020 22:59:44 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Mon, 27 Apr 2020 22:54:06 -0400
> Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
>
> > On Tue, 28 Apr 2020 10:59:14 +1000
> > Gavin Shan <gshan@xxxxxxxxxx> wrote:
> >
> > > Hi Mark,
> > >
> > > On 4/24/20 8:11 PM, Mark Rutland wrote:
> > > > [Adding Steve, who added str_has_prefix()]
> > > >
> > > > On Fri, Apr 24, 2020 at 02:53:14PM +1000, Gavin Shan wrote:
> > > >> The NUMA option is parsed by str_has_prefix() and the invalid option
> > > >> like "numa=o" can be regarded as "numa=off" wrongly.
> > > >
> > > > Are you certain that can pass? If that can happen, str_has_prefix() is
> > > > misnamed and does not seem to do what its kerneldoc says it does, as
> > > > "off" is not a prefix of "o".
> > > >
> > >
> > > Yes, It's possible. str_has_prefix() depends on strncmp(). In this particular
> > > case, it's equal to the snippet of code as below: strncmp() returns zero.
> > > str_has_prefix() returns 3.
> >
> > Wait! strncmp("o", "off", 3) returns zero?
> >
> > That to me looks like a bug!
> >
> > This means str_has_prefix() is broken in other areas as well.
> >
> >
> > >
> > > int strncmp(const char *cs, const char *ct, size_t count)
> > > {
> > > unsigned char c1, c2;
> > >
> > > while (count) {
> > > c1 = *cs++;
> > > c2 = *ct++;
> > > if (c1 != c2)
> > > return c1 < c2 ? -1 : 1;
> > > if (!c1) /* break after first character is compared */
> >
> > Crap! That is totally wrong!
>
> Looking at this again, it's not wrong. But how did we get here if c2 isn't
> zero as well?
>

Could this be a bug in the implementation of strncmp() in
arch/arm64/lib/strncmp.S. As I don't know arm64 assembly, I have no idea
what it is trying to do.

But strncmp("o","off",3) returning zero *is* a bug.

-- Steve