Re: [PATCH 1/2] tpm2-sessions: Add full HMAC and encrypt/decrypt session handling

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Mon, 2018-03-05 at 13:35 +0200, Jarkko Sakkinen wrote:
> On Fri, Mar 02, 2018 at 10:06:15PM -0800, James Bottomley wrote:
> > 
> > diff --git a/drivers/char/tpm/tpm2b.h b/drivers/char/tpm/tpm2b.h
> > new file mode 100644
> > index 000000000000..c7726f2895aa
> > --- /dev/null
> > +++ b/drivers/char/tpm/tpm2b.h
> > @@ -0,0 +1,82 @@
> > +/* SPDX-License-Identifier: GPL-2.0 */
> > +#ifndef _TPM2_TPM2B_H
> > +#define _TPM2_TPM2B_H
> > +/*
> > + * Handing for tpm2b structures to facilitate the building of
> > commands
> > + */
> > +
> > +#include "tpm.h"
> > +
> > +#include <asm/unaligned.h>
> > +
> > +struct tpm2b {
> > +	struct tpm_buf buf;
> > +};
> > +
> > +/* opaque structure, holds auth session parameters like the
> > session key */
> > +struct tpm2_auth;
> > +
> > +static inline int tpm2b_init(struct tpm2b *buf)
> > +{
> > +	return tpm_buf_init(&buf->buf, 0, 0);
> > +}
> > +
> > +static inline void tpm2b_reset(struct tpm2b *buf)
> > +{
> > +	struct tpm_input_header *head;
> > +
> > +	head = (struct tpm_input_header *)buf->buf.data;
> > +	head->length = cpu_to_be32(sizeof(*head));
> > +}
> > +
> > +static inline void tpm2b_append(struct tpm2b *buf, const unsigned
> > char *data,
> > +				unsigned int len)
> > +{
> > +	tpm_buf_append(&buf->buf, data, len);
> > +}
> > +
> > +#define TPM2B_APPEND(type) \
> > +	static inline void tpm2b_append_##type(struct tpm2b *buf,
> > const type value) { tpm_buf_append_##type(&buf->buf, value); }
> > +
> > +TPM2B_APPEND(u8)
> > +TPM2B_APPEND(u16)
> > +TPM2B_APPEND(u32)
> > +
> > +static inline void *tpm2b_buffer(const struct tpm2b *buf)
> > +{
> > +	return buf->buf.data + sizeof(struct tpm_input_header);
> > +}
> > +
> > +static inline u16 tpm2b_len(struct tpm2b *buf)
> > +{
> > +	return tpm_buf_length(&buf->buf) - sizeof(struct
> > tpm_input_header);
> > +}
> > +
> > +static inline void tpm2b_destroy(struct tpm2b *buf)
> > +{
> > +	tpm_buf_destroy(&buf->buf);
> > +}
> > +
> > +static inline void tpm_buf_append_2b(struct tpm_buf *buf, struct
> > tpm2b *tpm2b)
> > +{
> > +	u16 len = tpm2b_len(tpm2b);
> > +
> > +	tpm_buf_append_u16(buf, len);
> > +	tpm_buf_append(buf, tpm2b_buffer(tpm2b), len);
> > +	/* clear the buf for reuse */
> > +	tpm2b_reset(tpm2b);
> > +}
> > +
> > +/* Macros for unmarshalling known size BE data */
> > +#define GET_INC(type)					\
> > +static inline u##type get_inc_##type(const u8 **ptr) {	\
> > +	u##type val;					\
> > +	val = get_unaligned_be##type(*ptr);		\
> > +	*ptr += sizeof(val);				\
> > +	return val;					\
> > +}
> > +
> > +GET_INC(16)
> > +GET_INC(32)
> > +
> > +#endif
> > -- 
> > 2.12.3
> > 
> 
> Some meta stuff:
> 
> * Add me to TO-field because I should probably review and eventually
>   test these, right?

Eventually; they're an RFC because we need to get the API right first
(I've already got a couple of fixes to it).

> * CC to linux-security-module

There's no change to anything in security module, so why?  All security
module people who care about the TPM should be on linux-integrity and
those who don't likely don't want to see the changes.  The reason
linux-crypto is on the cc is because I want them to make sure I'm using
their crypto system correctly.

> * Why there is no RFC tag given that these are so quite large
> changes?

There is an RFC tag on 0/2

> * Why in hell tpm2b.h?

Because all sized TPM structures are in 2B form and manipulating them
can be made a lot easier with helper routines.

James