[PATCH v2] crypto: ccm - avoid negative wrapping of integers

[PATCH v2] crypto: ccm - avoid negative wrapping of integers

[Jordy Zomer]

Set csize to unsigned int to avoid it from wrapping as a negative number (since format input sends an unsigned integer to this function). This would also result in undefined behavior in the left shift when msg len is checked, potentially resulting in a buffer overflow in the memcpy call.

Signed-off-by: Jordy Zomer <jordy@pwning.systems>
---
To address was corrected, and ccm was added to the topic to indicate that this is just for ccm.

 crypto/ccm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/crypto/ccm.c b/crypto/ccm.c
index 6b815ece51c6..e14201edf9db 100644
--- a/crypto/ccm.c
+++ b/crypto/ccm.c
@@ -66,7 +66,7 @@ static inline struct crypto_ccm_req_priv_ctx *crypto_ccm_reqctx(
 	return (void *)PTR_ALIGN((u8 *)aead_request_ctx(req), align + 1);
 }
 
-static int set_msg_len(u8 *block, unsigned int msglen, int csize)
+static int set_msg_len(u8 *block, unsigned int msglen, unsigned int csize)
 {
 	__be32 data;
 
-- 
2.27.0

Re: [PATCH v2] crypto: ccm - avoid negative wrapping of integers

[Eric Biggers]

On Mon, Jul 26, 2021 at 07:01:20PM +0200, Jordy Zomer wrote:
> Set csize to unsigned int to avoid it from wrapping as a negative number (since format input sends an unsigned integer to this function). This would also result in undefined behavior in the left shift when msg len is checked, potentially resulting in a buffer overflow in the memcpy call.
> 
> Signed-off-by: Jordy Zomer <jordy@pwning.systems>
> ---
> To address was corrected, and ccm was added to the topic to indicate that this is just for ccm.
> 
>  crypto/ccm.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/crypto/ccm.c b/crypto/ccm.c
> index 6b815ece51c6..e14201edf9db 100644
> --- a/crypto/ccm.c
> +++ b/crypto/ccm.c
> @@ -66,7 +66,7 @@ static inline struct crypto_ccm_req_priv_ctx *crypto_ccm_reqctx(
>  	return (void *)PTR_ALIGN((u8 *)aead_request_ctx(req), align + 1);
>  }
>  
> -static int set_msg_len(u8 *block, unsigned int msglen, int csize)
> +static int set_msg_len(u8 *block, unsigned int msglen, unsigned int csize)
>  {
>  	__be32 data;

This isn't necessarily a bad change, but the value of csize is clearly in
[1, 256] if you read format_input(), and in fact is in [2, 8] if you read the
whole file, so please don't claim this is actually fixing anything, as it's not.
Also please line wrap your commit message.

- Eric

Re: [PATCH v2] crypto: ccm - avoid negative wrapping of integers

[Greg Kroah-Hartman]

On Mon, Jul 26, 2021 at 10:18:47AM -0700, Eric Biggers wrote:
> On Mon, Jul 26, 2021 at 07:01:20PM +0200, Jordy Zomer wrote:
> > Set csize to unsigned int to avoid it from wrapping as a negative number (since format input sends an unsigned integer to this function). This would also result in undefined behavior in the left shift when msg len is checked, potentially resulting in a buffer overflow in the memcpy call.
> > 
> > Signed-off-by: Jordy Zomer <jordy@pwning.systems>
> > ---
> > To address was corrected, and ccm was added to the topic to indicate that this is just for ccm.
> > 
> >  crypto/ccm.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/crypto/ccm.c b/crypto/ccm.c
> > index 6b815ece51c6..e14201edf9db 100644
> > --- a/crypto/ccm.c
> > +++ b/crypto/ccm.c
> > @@ -66,7 +66,7 @@ static inline struct crypto_ccm_req_priv_ctx *crypto_ccm_reqctx(
> >  	return (void *)PTR_ALIGN((u8 *)aead_request_ctx(req), align + 1);
> >  }
> >  
> > -static int set_msg_len(u8 *block, unsigned int msglen, int csize)
> > +static int set_msg_len(u8 *block, unsigned int msglen, unsigned int csize)
> >  {
> >  	__be32 data;
> 
> This isn't necessarily a bad change, but the value of csize is clearly in
> [1, 256] if you read format_input(), and in fact is in [2, 8] if you read the
> whole file, so please don't claim this is actually fixing anything, as it's not.

Oh that was not obvious at all, I looked at that for a long time and
missed the place where this was checked earlier.  Perhaps just make
csize here a u8 and that would take away all question about what is
happening?

thanks,

greg k-h

Re: [PATCH v2] crypto: ccm - avoid negative wrapping of integers

[Greg Kroah-Hartman]

On Tue, Jul 27, 2021 at 07:36:56AM +0200, Greg Kroah-Hartman wrote:
> On Mon, Jul 26, 2021 at 10:18:47AM -0700, Eric Biggers wrote:
> > On Mon, Jul 26, 2021 at 07:01:20PM +0200, Jordy Zomer wrote:
> > > Set csize to unsigned int to avoid it from wrapping as a negative number (since format input sends an unsigned integer to this function). This would also result in undefined behavior in the left shift when msg len is checked, potentially resulting in a buffer overflow in the memcpy call.
> > > 
> > > Signed-off-by: Jordy Zomer <jordy@pwning.systems>
> > > ---
> > > To address was corrected, and ccm was added to the topic to indicate that this is just for ccm.
> > > 
> > >  crypto/ccm.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/crypto/ccm.c b/crypto/ccm.c
> > > index 6b815ece51c6..e14201edf9db 100644
> > > --- a/crypto/ccm.c
> > > +++ b/crypto/ccm.c
> > > @@ -66,7 +66,7 @@ static inline struct crypto_ccm_req_priv_ctx *crypto_ccm_reqctx(
> > >  	return (void *)PTR_ALIGN((u8 *)aead_request_ctx(req), align + 1);
> > >  }
> > >  
> > > -static int set_msg_len(u8 *block, unsigned int msglen, int csize)
> > > +static int set_msg_len(u8 *block, unsigned int msglen, unsigned int csize)
> > >  {
> > >  	__be32 data;
> > 
> > This isn't necessarily a bad change, but the value of csize is clearly in
> > [1, 256] if you read format_input(), and in fact is in [2, 8] if you read the
> > whole file, so please don't claim this is actually fixing anything, as it's not.
> 
> Oh that was not obvious at all, I looked at that for a long time and
> missed the place where this was checked earlier.  Perhaps just make
> csize here a u8 and that would take away all question about what is
> happening?

And part of this effort is to make it obvious that there is no overflow
happening, to allow tools to check this type of thing.  When you have to
work backwards long long ways like this, it makes automatic auditing
almost impossible, along with manual auditing like I tried to do :)

thanks,

greg k-h