Re: [PATCH RFC] tpm: migrate pubek_show to struct tpm_buf

[Jarkko Sakkinen]

On Thu, May 25, 2017 at 03:16:13PM -0600, Jason Gunthorpe wrote:
> On Thu, May 25, 2017 at 02:11:04PM -0700, Jarkko Sakkinen wrote:
> >  	struct tpm_chip *chip = to_tpm_chip(dev);
> > +	char anti_replay[20];
> >  
> > -	tpm_cmd.header.in = tpm_readpubek_header;
> > -	err = tpm_transmit_cmd(chip, NULL, &tpm_cmd, READ_PUBEK_RESULT_SIZE,
> > +	rc = tpm_buf_init(&tpm_buf, TPM_TAG_RQU_COMMAND, TPM_ORD_READPUBEK);
> > +	if (rc)
> > +		return rc;
> > +
> > +	/* The checksum is ignored so it doesn't matter what the contents are.
> > +	 */
> > +	tpm_buf_append(&tpm_buf, anti_replay, sizeof(anti_replay));
> 
> It does matter, we do not want to leak random kernel memory incase it
> has something sensitive. Zero anti_replay.

If there was a leak it has existed before this change as tpm_cmd was
also allocated from stack. And there is not leak because the checksum is
not printed.

> > +
> > -	/*
> > -	   ignore header 10 bytes
> > -	   algorithm 32 bits (1 == RSA )
> > -	   encscheme 16 bits
> > -	   sigscheme 16 bits
> > -	   parameters (RSA 12->bytes: keybit, #primes, expbit)
> > -	   keylenbytes 32 bits
> > -	   256 byte modulus
> > -	   ignore checksum 20 bytes
> > -	 */
> 
> Not sure we should delete the comment, tpm buf does not make the parse
> any clearer.

I think better idea would be to move struct tpm_readpubek_params_out
declaration here and use it to refer different fields. Previously this
has been a complete mess. The structure has been declared but it has not
been used for anything. I wonder what is the history here...

/Jarkko