[PATCH] can: test size of struct sockaddr

[PATCH] can: test size of struct sockaddr

[Kurt Van Dijck]

I think this patch makes the CAN socket code comform to the
manpages of sendmsg & recvmsg.

Signed-off-by: Kurt Van Dijck <kurt.van.dijck-/BeEPy95v10@public.gmane.org>
---
 net/can/bcm.c |    6 +++++-
 net/can/raw.c |    6 +++++-
 2 files changed, 10 insertions(+), 2 deletions(-)

diff --git a/net/can/bcm.c b/net/can/bcm.c
index 6faa825..dc0d5d6 100644
--- a/net/can/bcm.c
+++ b/net/can/bcm.c
@@ -1256,6 +1256,9 @@ static int bcm_sendmsg(struct kiocb *iocb, struct socket *sock,
 		struct sockaddr_can *addr =
 			(struct sockaddr_can *)msg->msg_name;
 
+		if (msg->msg_namelen < sizeof(*addr))
+			return -EINVAL;
+
 		if (addr->can_family != AF_CAN)
 			return -EINVAL;
 
@@ -1557,8 +1560,9 @@ static int bcm_recvmsg(struct kiocb *iocb, struct socket *sock,
 	sock_recv_ts_and_drops(msg, sk, skb);
 
 	if (msg->msg_name) {
+		memcpy(msg->msg_name, skb->cb, MIN(msg->msg_namelen,
+					sizeof(struct sockaddr_can)));
 		msg->msg_namelen = sizeof(struct sockaddr_can);
-		memcpy(msg->msg_name, skb->cb, msg->msg_namelen);
 	}
 
 	skb_free_datagram(sk, skb);
diff --git a/net/can/raw.c b/net/can/raw.c
index e88f610..e68a6d3 100644
--- a/net/can/raw.c
+++ b/net/can/raw.c
@@ -649,6 +649,9 @@ static int raw_sendmsg(struct kiocb *iocb, struct socket *sock,
 		struct sockaddr_can *addr =
 			(struct sockaddr_can *)msg->msg_name;
 
+		if (msg->msg_namelen < sizeof(*addr))
+			return -EINVAL;
+
 		if (addr->can_family != AF_CAN)
 			return -EINVAL;
 
@@ -727,8 +730,9 @@ static int raw_recvmsg(struct kiocb *iocb, struct socket *sock,
 	sock_recv_ts_and_drops(msg, sk, skb);
 
 	if (msg->msg_name) {
+		memcpy(msg->msg_name, skb->cb, MIN(msg->msg_namelen,
+					sizeof(struct sockaddr_can)));
 		msg->msg_namelen = sizeof(struct sockaddr_can);
-		memcpy(msg->msg_name, skb->cb, msg->msg_namelen);
 	}
 
 	/* assign the flags that have been recorded in raw_rcv() */

Re: [PATCH] can: test size of struct sockaddr

[Oliver Hartkopp]

On 14.01.2011 18:23, Kurt Van Dijck wrote:
> I think this patch makes the CAN socket code comform to the
> manpages of sendmsg & recvmsg.

Hello Kurt,

if you check similar code sniplets in the kernel you would see, that in
recvmsg()-cases the given namelen is not used from userspace.

If msg->msg_name is not NULL the msg_namelen is just set by the kernel. E.g.
see af_packet.c, af_econet.c, etc.

So the code in candump.c setting the msg_namelen before recvmsg() is obviously
obsolete ...

Btw. your two patches below for bcm_sendmsg() and raw_sendmesg() look good.

If you would like to resubmit these two patches, you may add my Acked-by

Thanks,
Oliver



> diff --git a/net/can/bcm.c b/net/can/bcm.c
> index 6faa825..dc0d5d6 100644
> --- a/net/can/bcm.c
> +++ b/net/can/bcm.c
> @@ -1256,6 +1256,9 @@ static int bcm_sendmsg(struct kiocb *iocb, struct socket *sock,
>  		struct sockaddr_can *addr =
>  			(struct sockaddr_can *)msg->msg_name;
>  
> +		if (msg->msg_namelen < sizeof(*addr))
> +			return -EINVAL;
> +
>  		if (addr->can_family != AF_CAN)
>  			return -EINVAL;
>  


> diff --git a/net/can/raw.c b/net/can/raw.c
> index e88f610..e68a6d3 100644
> --- a/net/can/raw.c
> +++ b/net/can/raw.c
> @@ -649,6 +649,9 @@ static int raw_sendmsg(struct kiocb *iocb, struct socket *sock,
>  		struct sockaddr_can *addr =
>  			(struct sockaddr_can *)msg->msg_name;
>  
> +		if (msg->msg_namelen < sizeof(*addr))
> +			return -EINVAL;
> +
>  		if (addr->can_family != AF_CAN)
>  			return -EINVAL;
>  

Re: [PATCH] can: test size of struct sockaddr

[Kurt Van Dijck]

On Sat, Jan 15, 2011 at 06:08:54PM +0100, Oliver Hartkopp wrote:
> On 14.01.2011 18:23, Kurt Van Dijck wrote:
> > I think this patch makes the CAN socket code comform to the
> > manpages of sendmsg & recvmsg.
> 
> Hello Kurt,
> 
> if you check similar code sniplets in the kernel you would see, that in
> recvmsg()-cases the given namelen is not used from userspace.
I went into udp code, but my mind got troubled there.
> 
> If msg->msg_name is not NULL the msg_namelen is just set by the kernel. E.g.
> see af_packet.c, af_econet.c, etc.
I should have looked further ...
Given this, I went up to net/socket.c:__sys_recvmsg.
Within kernel space, a struct sockaddr_storage is used in fact ....
This solves indeed the problem I was trying to address.
> 
> So the code in candump.c setting the msg_namelen before recvmsg() is obviously
> obsolete ...
I think no, since after the socket's recvmsg() has been called, 
since move_addr_to_user() will do checks similar as I tried to reimplement.
> 
> Btw. your two patches below for bcm_sendmsg() and raw_sendmesg() look good.
> 
> If you would like to resubmit these two patches, you may add my Acked-by
Thanks, I will do so.

Kurt