dh key: fix rounding up KDF output length

Commit Message

Eric Biggers June 7, 2018, 7:12 p.m. UTC

From: Eric Biggers <[e-mail]>

Commit 383203eff718 ("dh key: get rid of stack allocated array") changed
kdf_ctr() to assume that the length of key material to derive is a
multiple of the digest size.  The length was supposed to be rounded up
accordingly.  However, the round_up() macro was used which only gives
the correct result on power-of-2 arguments, whereas not all hash
algorithms have power-of-2 digest sizes.  In some cases this resulted in
a write past the end of the 'outbuf' buffer.

Fix it by switching to roundup(), which works for non-power-of-2 inputs.

Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Reported-by: [e-mail]
Fixes: 383203eff718 ("dh key: get rid of stack allocated array")
Signed-off-by: Eric Biggers <[e-mail]>
---
 security/keys/dh.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

Comments

Kees Cook June 7, 2018, 7:16 p.m. UTC | #1

On Thu, Jun 7, 2018 at 12:12 PM, Eric Biggers <[e-mail]> wrote:
> From: Eric Biggers <[e-mail]>
>
> Commit 383203eff718 ("dh key: get rid of stack allocated array") changed
> kdf_ctr() to assume that the length of key material to derive is a
> multiple of the digest size.  The length was supposed to be rounded up
> accordingly.  However, the round_up() macro was used which only gives
> the correct result on power-of-2 arguments, whereas not all hash
> algorithms have power-of-2 digest sizes.  In some cases this resulted in
> a write past the end of the 'outbuf' buffer.
>
> Fix it by switching to roundup(), which works for non-power-of-2 inputs.

round_up() vs roundup(). Wow, that's not confusing. :( I wonder if we
should rename the former to roundup_pow2() or something?

> Reported-by: [e-mail]
> Reported-by: [e-mail]
> Reported-by: [e-mail]
> Reported-by: [e-mail]
> Fixes: 383203eff718 ("dh key: get rid of stack allocated array")
> Signed-off-by: Eric Biggers <[e-mail]>

Regardless:

Acked-by: Kees Cook <[e-mail]>

-Kees

> ---
>  security/keys/dh.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/security/keys/dh.c b/security/keys/dh.c
> index f7403821db7f0..b203f7758f976 100644
> --- a/security/keys/dh.c
> +++ b/security/keys/dh.c
> @@ -142,6 +142,8 @@ static void kdf_dealloc(struct kdf_sdesc *sdesc)
>   * The src pointer is defined as Z || other info where Z is the shared secret
>   * from DH and other info is an arbitrary string (see SP800-56A section
>   * 5.8.1.2).
> + *
> + * 'dlen' must be a multiple of the digest size.
>   */
>  static int kdf_ctr(struct kdf_sdesc *sdesc, const u8 *src, unsigned int slen,
>                    u8 *dst, unsigned int dlen, unsigned int zlen)
> @@ -205,8 +207,8 @@ static int keyctl_dh_compute_kdf(struct kdf_sdesc *sdesc,
>  {
>         uint8_t *outbuf = NULL;
>         int ret;
> -       size_t outbuf_len = round_up(buflen,
> -                                    crypto_shash_digestsize(sdesc->shash.tfm));
> +       size_t outbuf_len = roundup(buflen,
> +                                   crypto_shash_digestsize(sdesc->shash.tfm));
>
>         outbuf = kmalloc(outbuf_len, GFP_KERNEL);
>         if (!outbuf) {
> --
> 2.17.1.1185.g55be947832-goog
>

James Morris June 7, 2018, 7:28 p.m. UTC | #2

On Thu, 7 Jun 2018, Kees Cook wrote:

> On Thu, Jun 7, 2018 at 12:12 PM, Eric Biggers <[e-mail]> wrote:
> > From: Eric Biggers <[e-mail]>
> >
> > Commit 383203eff718 ("dh key: get rid of stack allocated array") changed
> > kdf_ctr() to assume that the length of key material to derive is a
> > multiple of the digest size.  The length was supposed to be rounded up
> > accordingly.  However, the round_up() macro was used which only gives
> > the correct result on power-of-2 arguments, whereas not all hash
> > algorithms have power-of-2 digest sizes.  In some cases this resulted in
> > a write past the end of the 'outbuf' buffer.
> >
> > Fix it by switching to roundup(), which works for non-power-of-2 inputs.
> 
> round_up() vs roundup(). Wow, that's not confusing. :( I wonder if we
> should rename the former to roundup_pow2() or something?

Good idea, in a separate patch(set).

Eric Biggers June 7, 2018, 7:28 p.m. UTC | #3

On Thu, Jun 07, 2018 at 12:16:16PM -0700, Kees Cook wrote:
> On Thu, Jun 7, 2018 at 12:12 PM, Eric Biggers <[e-mail]> wrote:
> > From: Eric Biggers <[e-mail]>
> >
> > Commit 383203eff718 ("dh key: get rid of stack allocated array") changed
> > kdf_ctr() to assume that the length of key material to derive is a
> > multiple of the digest size.  The length was supposed to be rounded up
> > accordingly.  However, the round_up() macro was used which only gives
> > the correct result on power-of-2 arguments, whereas not all hash
> > algorithms have power-of-2 digest sizes.  In some cases this resulted in
> > a write past the end of the 'outbuf' buffer.
> >
> > Fix it by switching to roundup(), which works for non-power-of-2 inputs.
> 
> round_up() vs roundup(). Wow, that's not confusing. :( I wonder if we
> should rename the former to roundup_pow2() or something?

Yes, it's very confusing, and I wish the names were clearer, or that there was
one macro that just did the right thing (but then the power-of-2 optimization
could only be done for constants, where it might not be necessary anyway).
roundup_pow2() would still be confused with roundup_pow_of_two(), unfortunately.

Eric

Tycho Andersen June 7, 2018, 8:28 p.m. UTC | #4

On Thu, Jun 07, 2018 at 12:12:01PM -0700, Eric Biggers wrote:
> From: Eric Biggers <[e-mail]>
> 
> Commit 383203eff718 ("dh key: get rid of stack allocated array") changed
> kdf_ctr() to assume that the length of key material to derive is a
> multiple of the digest size.  The length was supposed to be rounded up
> accordingly.  However, the round_up() macro was used which only gives
> the correct result on power-of-2 arguments, whereas not all hash
> algorithms have power-of-2 digest sizes.  In some cases this resulted in
> a write past the end of the 'outbuf' buffer.
> 
> Fix it by switching to roundup(), which works for non-power-of-2 inputs.
> 
> Reported-by: [e-mail]
> Reported-by: [e-mail]
> Reported-by: [e-mail]
> Reported-by: [e-mail]
> Fixes: 383203eff718 ("dh key: get rid of stack allocated array")
> Signed-off-by: Eric Biggers <[e-mail]>

Arg, thanks.

Acked-by: Tycho Andersen <[e-mail]>

David Howells June 8, 2018, 3:37 p.m. UTC | #5

Eric Biggers <[e-mail]> wrote:

> Commit 383203eff718 ("dh key: get rid of stack allocated array") changed
> kdf_ctr() to assume that the length of key material to derive is a
> multiple of the digest size.  The length was supposed to be rounded up
> accordingly.  However, the round_up() macro was used which only gives
> the correct result on power-of-2 arguments, whereas not all hash
> algorithms have power-of-2 digest sizes.  In some cases this resulted in
> a write past the end of the 'outbuf' buffer.
> 
> Fix it by switching to roundup(), which works for non-power-of-2 inputs.

Applied.

Eric Biggers June 25, 2018, 5:14 p.m. UTC | #6

Hi David,

On Fri, Jun 08, 2018 at 04:37:58PM +0100, David Howells wrote:
> Eric Biggers <[e-mail]> wrote:
> 
> > Commit 383203eff718 ("dh key: get rid of stack allocated array") changed
> > kdf_ctr() to assume that the length of key material to derive is a
> > multiple of the digest size.  The length was supposed to be rounded up
> > accordingly.  However, the round_up() macro was used which only gives
> > the correct result on power-of-2 arguments, whereas not all hash
> > algorithms have power-of-2 digest sizes.  In some cases this resulted in
> > a write past the end of the 'outbuf' buffer.
> > 
> > Fix it by switching to roundup(), which works for non-power-of-2 inputs.
> 
> Applied.

Applied to where?  When are you planning to send this to Linus?  Note that this
was a regression from v4.17 to v4.18-rc1.

Thanks,

- Eric

Patch

diff --git a/security/keys/dh.c b/security/keys/dh.c
index f7403821db7f0..b203f7758f976 100644
--- a/security/keys/dh.c
+++ b/security/keys/dh.c
@@ -142,6 +142,8 @@  static void kdf_dealloc(struct kdf_sdesc *sdesc)
  * The src pointer is defined as Z || other info where Z is the shared secret
  * from DH and other info is an arbitrary string (see SP800-56A section
  * 5.8.1.2).
+ *
+ * 'dlen' must be a multiple of the digest size.
  */
 static int kdf_ctr(struct kdf_sdesc *sdesc, const u8 *src, unsigned int slen,
 		   u8 *dst, unsigned int dlen, unsigned int zlen)
@@ -205,8 +207,8 @@  static int keyctl_dh_compute_kdf(struct kdf_sdesc *sdesc,
 {
 	uint8_t *outbuf = NULL;
 	int ret;
-	size_t outbuf_len = round_up(buflen,
-			             crypto_shash_digestsize(sdesc->shash.tfm));
+	size_t outbuf_len = roundup(buflen,
+				    crypto_shash_digestsize(sdesc->shash.tfm));
 
 	outbuf = kmalloc(outbuf_len, GFP_KERNEL);
 	if (!outbuf) {