2019-04-01 18:33:20

by Kees Cook

[permalink] [raw]
Subject: [PATCH] tpm: Actually fail on TPM errors during "get random"

A "get random" may fail with a TPM error, but those codes were returned
as-is to the caller, which assumed the result was the number of bytes
that had been written to the target buffer, which could lead to an kernel
heap memory exposure and over-read.

This fixes tpm1_get_random() to mask positive TPM errors into -EIO, as
before.

[ 18.092103] tpm tpm0: A TPM error (379) occurred attempting get random
[ 18.092106] usercopy: Kernel memory exposure attempt detected from SLUB object 'kmalloc-64' (offset 0, size 379)!

Link: https://bugzilla.redhat.com/show_bug.cgi?id=1650989
Reported-by: Phil Baker <[email protected]>
Reported-by: Craig Robson <[email protected]>
Fixes: 7aee9c52d7ac ("tpm: tpm1: rewrite tpm1_get_random() using tpm_buf structure")
Cc: Laura Abbott <[email protected]>
Cc: Tomas Winkler <[email protected]>
Cc: Jarkko Sakkinen <[email protected]>
Cc: [email protected]
Signed-off-by: Kees Cook <[email protected]>
---
drivers/char/tpm/tpm1-cmd.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/char/tpm/tpm1-cmd.c b/drivers/char/tpm/tpm1-cmd.c
index 85dcf2654d11..faeb78ecf960 100644
--- a/drivers/char/tpm/tpm1-cmd.c
+++ b/drivers/char/tpm/tpm1-cmd.c
@@ -510,7 +510,7 @@ struct tpm1_get_random_out {
*
* Return:
* * number of bytes read
- * * -errno or a TPM return code otherwise
+ * * -errno (positive TPM return codes are masked to -EIO)
*/
int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
{
@@ -524,7 +524,7 @@ int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)

rc = tpm_buf_init(&buf, TPM_TAG_RQU_COMMAND, TPM_ORD_GET_RANDOM);
if (rc)
- return rc;
+ goto fail;

do {
tpm_buf_append_u32(&buf, num_bytes);
@@ -559,7 +559,10 @@ int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
rc = total ? (int)total : -EIO;
out:
tpm_buf_destroy(&buf);
- return rc;
+fail:
+ if (rc < 0)
+ return rc;
+ return -EIO;
}

#define TPM_ORD_PCRREAD 21
--
2.17.1


--
Kees Cook


2019-04-01 18:41:07

by Jason Gunthorpe

[permalink] [raw]
Subject: Re: [PATCH] tpm: Actually fail on TPM errors during "get random"

On Mon, Apr 01, 2019 at 11:32:11AM -0700, Kees Cook wrote:
> A "get random" may fail with a TPM error, but those codes were returned
> as-is to the caller, which assumed the result was the number of bytes
> that had been written to the target buffer, which could lead to an kernel
> heap memory exposure and over-read.
>
> This fixes tpm1_get_random() to mask positive TPM errors into -EIO, as
> before.
>
> [ 18.092103] tpm tpm0: A TPM error (379) occurred attempting get random
> [ 18.092106] usercopy: Kernel memory exposure attempt detected from SLUB object 'kmalloc-64' (offset 0, size 379)!
>
> Link: https://bugzilla.redhat.com/show_bug.cgi?id=1650989
> Reported-by: Phil Baker <[email protected]>
> Reported-by: Craig Robson <[email protected]>
> Fixes: 7aee9c52d7ac ("tpm: tpm1: rewrite tpm1_get_random() using tpm_buf structure")
> Cc: Laura Abbott <[email protected]>
> Cc: Tomas Winkler <[email protected]>
> Cc: Jarkko Sakkinen <[email protected]>
> Cc: [email protected]
> Signed-off-by: Kees Cook <[email protected]>
> drivers/char/tpm/tpm1-cmd.c | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/char/tpm/tpm1-cmd.c b/drivers/char/tpm/tpm1-cmd.c
> index 85dcf2654d11..faeb78ecf960 100644
> +++ b/drivers/char/tpm/tpm1-cmd.c
> @@ -510,7 +510,7 @@ struct tpm1_get_random_out {
> *
> * Return:
> * * number of bytes read
> - * * -errno or a TPM return code otherwise
> + * * -errno (positive TPM return codes are masked to -EIO)
> */
> int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
> {
> @@ -524,7 +524,7 @@ int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
>
> rc = tpm_buf_init(&buf, TPM_TAG_RQU_COMMAND, TPM_ORD_GET_RANDOM);
> if (rc)
> - return rc;
> + goto fail;

?? Does tpm_buf_init() return positive?? Shouldn't..

I think you actually want this:

diff --git a/drivers/char/tpm/tpm1-cmd.c b/drivers/char/tpm/tpm1-cmd.c
index 85dcf2654d1102..a01e6fba1aacb6 100644
--- a/drivers/char/tpm/tpm1-cmd.c
+++ b/drivers/char/tpm/tpm1-cmd.c
@@ -510,7 +510,7 @@ struct tpm1_get_random_out {
*
* Return:
* * number of bytes read
- * * -errno or a TPM return code otherwise
+ * * -errno
*/
int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
{
@@ -531,8 +531,10 @@ int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)

rc = tpm_transmit_cmd(chip, &buf, sizeof(out->rng_data_len),
"attempting get random");
- if (rc)
+ if (rc) {
+ rc = -EIO
goto out;
+ }

out = (struct tpm1_get_random_out *)&buf.data[TPM_HEADER_SIZE];

diff --git a/drivers/char/tpm/tpm2-cmd.c b/drivers/char/tpm/tpm2-cmd.c
index e74c5b7b64bfbd..9fa498b4cf8816 100644
--- a/drivers/char/tpm/tpm2-cmd.c
+++ b/drivers/char/tpm/tpm2-cmd.c
@@ -328,8 +328,10 @@ int tpm2_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
offsetof(struct tpm2_get_random_out,
buffer),
"attempting get random");
- if (err)
+ if (err) {
+ rc = -EIO;
goto out;
+ }

out = (struct tpm2_get_random_out *)
&buf.data[TPM_HEADER_SIZE];

2019-04-01 18:44:53

by Kees Cook

[permalink] [raw]
Subject: Re: [PATCH] tpm: Actually fail on TPM errors during "get random"

On Mon, Apr 1, 2019 at 11:39 AM Jason Gunthorpe <[email protected]> wrote:
>
> On Mon, Apr 01, 2019 at 11:32:11AM -0700, Kees Cook wrote:
> > A "get random" may fail with a TPM error, but those codes were returned
> > as-is to the caller, which assumed the result was the number of bytes
> > that had been written to the target buffer, which could lead to an kernel
> > heap memory exposure and over-read.
> >
> > This fixes tpm1_get_random() to mask positive TPM errors into -EIO, as
> > before.
> >
> > [ 18.092103] tpm tpm0: A TPM error (379) occurred attempting get random
> > [ 18.092106] usercopy: Kernel memory exposure attempt detected from SLUB object 'kmalloc-64' (offset 0, size 379)!
> >
> > Link: https://bugzilla.redhat.com/show_bug.cgi?id=1650989
> > Reported-by: Phil Baker <[email protected]>
> > Reported-by: Craig Robson <[email protected]>
> > Fixes: 7aee9c52d7ac ("tpm: tpm1: rewrite tpm1_get_random() using tpm_buf structure")
> > Cc: Laura Abbott <[email protected]>
> > Cc: Tomas Winkler <[email protected]>
> > Cc: Jarkko Sakkinen <[email protected]>
> > Cc: [email protected]
> > Signed-off-by: Kees Cook <[email protected]>
> > drivers/char/tpm/tpm1-cmd.c | 9 ++++++---
> > 1 file changed, 6 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/char/tpm/tpm1-cmd.c b/drivers/char/tpm/tpm1-cmd.c
> > index 85dcf2654d11..faeb78ecf960 100644
> > +++ b/drivers/char/tpm/tpm1-cmd.c
> > @@ -510,7 +510,7 @@ struct tpm1_get_random_out {
> > *
> > * Return:
> > * * number of bytes read
> > - * * -errno or a TPM return code otherwise
> > + * * -errno (positive TPM return codes are masked to -EIO)
> > */
> > int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
> > {
> > @@ -524,7 +524,7 @@ int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
> >
> > rc = tpm_buf_init(&buf, TPM_TAG_RQU_COMMAND, TPM_ORD_GET_RANDOM);
> > if (rc)
> > - return rc;
> > + goto fail;
>
> ?? Does tpm_buf_init() return positive?? Shouldn't..
>
> I think you actually want this:

I didn't check that, but it was tpm_transmit_cmd() that was failing
(and would be covered by the error path change I sent).

>
> diff --git a/drivers/char/tpm/tpm1-cmd.c b/drivers/char/tpm/tpm1-cmd.c
> index 85dcf2654d1102..a01e6fba1aacb6 100644
> --- a/drivers/char/tpm/tpm1-cmd.c
> +++ b/drivers/char/tpm/tpm1-cmd.c
> @@ -510,7 +510,7 @@ struct tpm1_get_random_out {
> *
> * Return:
> * * number of bytes read
> - * * -errno or a TPM return code otherwise
> + * * -errno
> */
> int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
> {
> @@ -531,8 +531,10 @@ int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
>
> rc = tpm_transmit_cmd(chip, &buf, sizeof(out->rng_data_len),
> "attempting get random");
> - if (rc)
> + if (rc) {
> + rc = -EIO
> goto out;
> + }
>
> out = (struct tpm1_get_random_out *)&buf.data[TPM_HEADER_SIZE];
>
> diff --git a/drivers/char/tpm/tpm2-cmd.c b/drivers/char/tpm/tpm2-cmd.c
> index e74c5b7b64bfbd..9fa498b4cf8816 100644
> --- a/drivers/char/tpm/tpm2-cmd.c
> +++ b/drivers/char/tpm/tpm2-cmd.c
> @@ -328,8 +328,10 @@ int tpm2_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
> offsetof(struct tpm2_get_random_out,
> buffer),
> "attempting get random");
> - if (err)
> + if (err) {
> + rc = -EIO;
> goto out;
> + }
>
> out = (struct tpm2_get_random_out *)
> &buf.data[TPM_HEADER_SIZE];

Eeks, I missed tpm2. In both cases, though a negative error would be
masked. I was trying to leave those as-is and only mask positive
errors.


--
Kees Cook

2019-04-01 18:53:33

by James Bottomley

[permalink] [raw]
Subject: Re: [PATCH] tpm: Actually fail on TPM errors during "get random"

On Mon, 2019-04-01 at 11:32 -0700, Kees Cook wrote:
[...]
> --- a/drivers/char/tpm/tpm1-cmd.c
> +++ b/drivers/char/tpm/tpm1-cmd.c
> @@ -510,7 +510,7 @@ struct tpm1_get_random_out {
> *
> * Return:
> * * number of bytes read
> - * * -errno or a TPM return code otherwise
> + * * -errno (positive TPM return codes are masked to -EIO)
> */
> int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
> {
> @@ -524,7 +524,7 @@ int tpm1_get_random(struct tpm_chip *chip, u8
> *dest, size_t max)
>
> rc = tpm_buf_init(&buf, TPM_TAG_RQU_COMMAND,
> TPM_ORD_GET_RANDOM);
> if (rc)
> - return rc;
> + goto fail;
>
> do {
> tpm_buf_append_u32(&buf, num_bytes);
> @@ -559,7 +559,10 @@ int tpm1_get_random(struct tpm_chip *chip, u8
> *dest, size_t max)
> rc = total ? (int)total : -EIO;
> out:
> tpm_buf_destroy(&buf);
> - return rc;

You can't remove this otherwise the only return will ever be a failure.

I think what you're trying to catch is tpm_transmit_cmd returning a
positive failure, So you need to check the output of tpm_transmit_cmd
as well and goto failure leaving the above return in place.

James


> +fail:
> + if (rc < 0)
> + return rc;
> + return -EIO;
> }
>
> #define TPM_ORD_PCRREAD 21
> --
> 2.17.1
>
>

2019-04-01 18:56:10

by Kees Cook

[permalink] [raw]
Subject: Re: [PATCH] tpm: Actually fail on TPM errors during "get random"

On Mon, Apr 1, 2019 at 11:52 AM James Bottomley
<[email protected]> wrote:
>
> On Mon, 2019-04-01 at 11:32 -0700, Kees Cook wrote:
> [...]
> > --- a/drivers/char/tpm/tpm1-cmd.c
> > +++ b/drivers/char/tpm/tpm1-cmd.c
> > @@ -510,7 +510,7 @@ struct tpm1_get_random_out {
> > *
> > * Return:
> > * * number of bytes read
> > - * * -errno or a TPM return code otherwise
> > + * * -errno (positive TPM return codes are masked to -EIO)
> > */
> > int tpm1_get_random(struct tpm_chip *chip, u8 *dest, size_t max)
> > {
> > @@ -524,7 +524,7 @@ int tpm1_get_random(struct tpm_chip *chip, u8
> > *dest, size_t max)
> >
> > rc = tpm_buf_init(&buf, TPM_TAG_RQU_COMMAND,
> > TPM_ORD_GET_RANDOM);
> > if (rc)
> > - return rc;
> > + goto fail;
> >
> > do {
> > tpm_buf_append_u32(&buf, num_bytes);
> > @@ -559,7 +559,10 @@ int tpm1_get_random(struct tpm_chip *chip, u8
> > *dest, size_t max)
> > rc = total ? (int)total : -EIO;
> > out:
> > tpm_buf_destroy(&buf);
> > - return rc;
>
> You can't remove this otherwise the only return will ever be a failure.
>
> I think what you're trying to catch is tpm_transmit_cmd returning a
> positive failure, So you need to check the output of tpm_transmit_cmd
> as well and goto failure leaving the above return in place.

eek, yes. I double-checked this in tpm2, but tpm1 is different...

--
Kees Cook