Re: [PATCH] kobject: Replace strlcpy with strscpy

[Kees Cook]

On Tue, Jul 11, 2023 at 08:14:30AM +0000, David Laight wrote:
> From: Azeem Shaikh
> > Sent: 10 July 2023 19:07
> > 
> > On Mon, Jul 10, 2023 at 9:13 AM David Laight <David.Laight@xxxxxxxxxx> wrote:
> > >
> > > >       int len;
> > > >
> > > > -     len = strlcpy(&env->buf[env->buflen], subsystem, buffer_size);
> > > > -     if (len >= buffer_size) {
> > > > +     len = strscpy(&env->buf[env->buflen], subsystem, buffer_size);
> > > > +     if (len < 0) {
> > > >               pr_warn("init_uevent_argv: buffer size of %d too small, needed %d\n",
> > > >                       buffer_size, len);
> > > >               return -ENOMEM;
> > >
> > > The size in the error message is now wrong.
> > 
> > Thanks for catching this.
> > 
> > > It has to be said that mostly all the strings that get copied
> > > in the kernel are '\0' terminated - so maybe it is all moot.
> > > OTOH printing (at least some of) the string that didn't fit
> > > is a lot more useful than its length.
> > 
> > How about printing out strlen(subsystem) along with the entire value
> > of @subsystem? So that the warn reads:
> > 
> > pr_warn("init_uevent_argv: buffer size of %d too small for %s, needed
> > %d\n", buffer_size, subsystem, strlen(subsystem));
> > 
> > Does that seem better?

Yeah, that'll retain the intention of the warning. It shouldn't really
even be possible to hit that warning, so I don't think we need to worry
about the "extra" call to strlen().

> Not with the justification for not using strlcpy() :-)

What?

-Kees

-- 
Kees Cook