On Tue, Jan 31, 2017 at 10:50:06PM +0200, Jarkko Sakkinen wrote:
> On Wed, Feb 01, 2017 at 12:14:12AM +0530, Nayna wrote:
> > > I already sent my pull request to 4.11 and even today I found something
> > > fishy. You declared a function local array by using a variable in "tpm:
> > > enhance TPM 2.0 PCR extend to support multiple banks" (max_active_banks
> > > or something). And the event log patches have just passed the review.
> >
> > Yes. I have checked using clang and it has passed the clang.. and I also
> > verified there were no complains during build.
>
> What we can deduce from that is that they didn't expose the issue in
> question.
>
> I found this by running sparse with make C=2 M=drives/char/tpm
>
> > What type of problem do you see ?
>
> It is disallowed to do stack allocation in the kernel code even if C
> standard would allow it. Stack is scarce resource so you need to know
> its usage at compile time.
>
> In this case you actually know the allocation because the value is not
> changed during the course of the function but it is still bad. Probably
> compiler will optimize it out. Still it is not a good practice.
>
> > Also, to understand, this is related to multi-bank patchset. I mean how does
> > it affect for event log patchset ?
>
> Well in both cases these have landed fairly late but I asked from James
> whether I'll have to postpone these to 4.12.
>
> Usually when I've sent my release pull request I do not want to make any
> radical changes to the codebase because they always require extra QA and
> thus take extra time.
rc = tpm_transmit_cmd(chip, buf.data, tpm_buf_length(&buf), 0, 0,
"attempting extend a PCR value");
This should be
rc = tpm_transmit_cmd(chip, buf.data, PAGE_SIZE, 0, 0,
"attempting extend a PCR value");
The second parameter is the size of the buffer, not length of the input
data.
I'll update this...
/Jarkko
