2019-07-02 00:03:09

by Cfir Cohen

Subject: [PATCH] crypto: ccp/gcm - use const time tag comparison.

Avoid leaking GCM tag through timing side channel.

Signed-off-by: Cfir Cohen <[email protected]>
---
drivers/crypto/ccp/ccp-ops.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/crypto/ccp/ccp-ops.c b/drivers/crypto/ccp/ccp-ops.c
index db8de89d990f..633670220f6c 100644
--- a/drivers/crypto/ccp/ccp-ops.c
+++ b/drivers/crypto/ccp/ccp-ops.c
@@ -840,7 +840,8 @@ static int ccp_run_aes_gcm_cmd(struct ccp_cmd_queue *cmd_q,
if (ret)
goto e_tag;

- ret = memcmp(tag.address, final_wa.address, AES_BLOCK_SIZE);
+ ret = crypto_memneq(tag.address, final_wa.address,
+ AES_BLOCK_SIZE) ? -EBADMSG : 0;
ccp_dm_free(&tag);
}

--
2.22.0.410.gd8fdbe21b5-goog

2019-07-02 15:55:01

by Eric Biggers

Subject: Re: [PATCH] crypto: ccp/gcm - use const time tag comparison.

On Tue, Jul 02, 2019 at 03:41:23PM +0000, Gary R Hook wrote:
> On 7/1/19 7:25 PM, Eric Biggers wrote:
> > On Mon, Jul 01, 2019 at 05:01:32PM -0700, Cfir Cohen wrote:
> >> Avoid leaking GCM tag through timing side channel.
> >>
> >> Signed-off-by: Cfir Cohen <[email protected]>
> >> ---
> >> drivers/crypto/ccp/ccp-ops.c | 3 ++-
> >> 1 file changed, 2 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/crypto/ccp/ccp-ops.c b/drivers/crypto/ccp/ccp-ops.c
> >> index db8de89d990f..633670220f6c 100644
> >> --- a/drivers/crypto/ccp/ccp-ops.c
> >> +++ b/drivers/crypto/ccp/ccp-ops.c
> >> @@ -840,7 +840,8 @@ static int ccp_run_aes_gcm_cmd(struct ccp_cmd_queue *cmd_q,
> >> if (ret)
> >> goto e_tag;
> >>
> >> - ret = memcmp(tag.address, final_wa.address, AES_BLOCK_SIZE);
> >> + ret = crypto_memneq(tag.address, final_wa.address,
> >> + AES_BLOCK_SIZE) ? -EBADMSG : 0;
> >> ccp_dm_free(&tag);
> >> }
> >>
> >> --
> >> 2.22.0.410.gd8fdbe21b5-goog
> >>
> >
> > Looks like this needs:
> >
> > Fixes: 36cf515b9bbe ("crypto: ccp - Enable support for AES GCM on v5 CCPs")
> > Cc: <[email protected]> # v4.12+
>
>
> Yes, it does. For clarity, does that mean you've taken care of this?
>

Herbert is the person who will apply this, so he'd need to do it. But it might
be better just to resend.

- Eric