On Thu, 2019-07-25 at 13:03 -0700, Kees Cook wrote: > On Wed, Jul 24, 2019 at 10:08:57AM -0700, Linus Torvalds wrote: > > On Wed, Jul 24, 2019 at 6:09 AM Rasmus Villemoes > > <li...@rasmusvillemoes.dk> wrote: > > > The kernel's snprintf() does not behave in a non-standard way, at least > > > not with respect to its return value. > > > > Note that the kernels snprintf() *does* very much protect against the > > overflow case - not by changing the return value, but simply by having > > > > /* Reject out-of-range values early. Large positive sizes are > > used for unknown buffer sizes. */ > > if (WARN_ON_ONCE(size > INT_MAX)) > > return 0; > > > > at the very top. > > > > So you can't actually overflow in the kernel by using the repeated > > > > offset += vsnprintf( .. size - offset ..); > > > > model. > > > > Yes, it's the wrong thing to do, but it is still _safe_. > > Actually, perhaps we should add this test to strscpy() too?
Doesn't seem to have a reason not to be added but maybe it's better to add another WARN_ON_ONCE. > diff --git a/lib/string.c b/lib/string.c [] > @@ -182,7 +182,7 @@ ssize_t strscpy(char *dest, const char *src, size_t count) > size_t max = count; > long res = 0; > > - if (count == 0) > + if (count == 0 || count > INT_MAX) > return -E2BIG; > > #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS >
