On Thu Sep 12, 2024 at 8:36 PM EEST, Sergey Shtylyov wrote:
> On 9/12/24 5:27 PM, Jarkko Sakkinen wrote:
> [...]
>
> >>>>>> In find_asymmetric_key(), if all NULLs are passed in id_{0,1,2} parameters
> >>>>>> the kernel will first emit WARN and then have an oops because id_2 gets
> >>>>>> dereferenced anyway.
> >>>>>>
> >>>>>> Found by Linux Verification Center (linuxtesting.org) with Svace static
> >>>>>> analysis tool.
> >>>>>
> >>>>> Weird, I recall that I've either sent a patch to address the same site
> >>>>> OR have commented a patch with similar reasoning. Well, it does not
> >>>>> matter, I think it this makes sense to me.
> >>>>>
> >>>>> You could further add to the motivation that given the panic_on_warn
> >>>>> kernel command-line parameter, it is for the best limit the scope and
> >>>>> use of the WARN-macro.
> >>>>
> >>>> I don't understand what you mean -- this version of the patch keeps
> >>>> the WARN_ON() call, it just moves that call, so that the duplicate id_{0,1,2}
> >>>> checks are avoided...
> >>>
> >>> I overlooked the code change (my bad sorry). Here's a better version of
> >>> the first paragraph:
> >>>
> >>> "find_asymmetric_keys() has nullity checks of id_0 and id_1 but ignores
> >>> validation for id_2. Check nullity also for id_2."
> >>
> >> Hm, what about WARN_ON(!id_0 && !id_1 && !id_2) -- it used to check all
> >> the pointers, right? I think our variant was closer to reality... :-)
> >
> > Right (lazy validation, first null ignores rest)
>
> No, contrariwise: since we use && and !, first non-NULL would ignore the rest.
Oops correct :-/
BR, Jarkko
