[PATCH net-next] net/mlx4_en: Fix IPv6 csum calculation

[PATCH net-next] net/mlx4_en: Fix IPv6 csum calculation

[clsoto]

From: Carol L Soto <clsoto@linux.vnet.ibm.com>

Seeing this message with mlx4_eni with IPv6: hw csum failure

Changing IPv6 csum calculation to be based on OFED 2.4 code.
When calculate IPv6 csum based also on source and destination
addresses.

Signed-off-by: Muhammad Mahajna <muhammadm@mellanox.com>
Signed-off-by: Carol L Soto <clsoto@linux.vnet.ibm.com>
---
 drivers/net/ethernet/mellanox/mlx4/en_rx.c | 46 +++++++++++++++++++++---------
 1 file changed, 33 insertions(+), 13 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
index 4402a1e..d43c512 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
@@ -693,26 +693,46 @@ static void get_fixed_ipv4_csum(__wsum hw_checksum, struct sk_buff *skb,
 }
 
 #if IS_ENABLED(CONFIG_IPV6)
-/* In IPv6 packets, besides subtracting the pseudo header checksum,
- * we also compute/add the IP header checksum which
- * is not added by the HW.
- */
+static __wsum csum_ipv6_magic_nofold(const struct in6_addr *saddr,
+				     const struct in6_addr *daddr,
+				     __u32 len, unsigned short proto)
+{
+	__wsum res = 0;
+
+	res = csum_add(res, saddr->in6_u.u6_addr32[0]);
+	res = csum_add(res, saddr->in6_u.u6_addr32[1]);
+	res = csum_add(res, saddr->in6_u.u6_addr32[2]);
+	res = csum_add(res, saddr->in6_u.u6_addr32[3]);
+	res = csum_add(res, daddr->in6_u.u6_addr32[0]);
+	res = csum_add(res, daddr->in6_u.u6_addr32[1]);
+	res = csum_add(res, daddr->in6_u.u6_addr32[2]);
+	res = csum_add(res, daddr->in6_u.u6_addr32[3]);
+	res = csum_add(res, len);
+	res = csum_add(res, htonl(proto));
+
+	return res;
+}
+
 static int get_fixed_ipv6_csum(__wsum hw_checksum, struct sk_buff *skb,
 			       struct ipv6hdr *ipv6h)
 {
-	__wsum csum_pseudo_hdr = 0;
+	__wsum csum_pseudo_header = 0;
 
-	if (ipv6h->nexthdr == IPPROTO_FRAGMENT || ipv6h->nexthdr == IPPROTO_HOPOPTS)
+	if (ipv6h->nexthdr == IPPROTO_FRAGMENT ||
+	    ipv6h->nexthdr == IPPROTO_HOPOPTS)
 		return -1;
-	hw_checksum = csum_add(hw_checksum, (__force __wsum)(ipv6h->nexthdr << 8));
 
-	csum_pseudo_hdr = csum_partial(&ipv6h->saddr,
-				       sizeof(ipv6h->saddr) + sizeof(ipv6h->daddr), 0);
-	csum_pseudo_hdr = csum_add(csum_pseudo_hdr, (__force __wsum)ipv6h->payload_len);
-	csum_pseudo_hdr = csum_add(csum_pseudo_hdr, (__force __wsum)ntohs(ipv6h->nexthdr));
+	hw_checksum = csum_add(hw_checksum, cpu_to_be16(ipv6h->nexthdr));
+	csum_pseudo_header = csum_ipv6_magic_nofold(&ipv6h->saddr,
+						    &ipv6h->daddr,
+						    ipv6h->payload_len,
+						    ipv6h->nexthdr);
+	hw_checksum = csum_sub(hw_checksum, csum_pseudo_header);
+	hw_checksum = csum_partial(ipv6h, sizeof(struct ipv6hdr), hw_checksum);
+	if (!hw_checksum)
+		return -1;
 
-	skb->csum = csum_sub(hw_checksum, csum_pseudo_hdr);
-	skb->csum = csum_add(skb->csum, csum_partial(ipv6h, sizeof(struct ipv6hdr), 0));
+	skb->csum = hw_checksum;
 	return 0;
 }
 #endif
-- 
1.8.3.1

Re: [PATCH net-next] net/mlx4_en: Fix IPv6 csum calculation

[Joe Perches]

On Fri, 2015-08-28 at 14:28 -0500, clsoto@linux.vnet.ibm.com wrote:
> From: Carol L Soto <clsoto@linux.vnet.ibm.com>
> 
> Seeing this message with mlx4_eni with IPv6: hw csum failure
> 
> Changing IPv6 csum calculation to be based on OFED 2.4 code.
> When calculate IPv6 csum based also on source and destination
> addresses.
[]
> diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
[]
> +static __wsum csum_ipv6_magic_nofold(const struct in6_addr *saddr,
> +				     const struct in6_addr *daddr,
> +				     __u32 len, unsigned short proto)
> +{
> +	__wsum res = 0;
> +
> +	res = csum_add(res, saddr->in6_u.u6_addr32[0]);
> +	res = csum_add(res, saddr->in6_u.u6_addr32[1]);
> +	res = csum_add(res, saddr->in6_u.u6_addr32[2]);
> +	res = csum_add(res, saddr->in6_u.u6_addr32[3]);
> +	res = csum_add(res, daddr->in6_u.u6_addr32[0]);
> +	res = csum_add(res, daddr->in6_u.u6_addr32[1]);
> +	res = csum_add(res, daddr->in6_u.u6_addr32[2]);
> +	res = csum_add(res, daddr->in6_u.u6_addr32[3]);
> +	res = csum_add(res, len);
> +	res = csum_add(res, htonl(proto));

You should try running sparse on this code.

Re: [PATCH net-next] net/mlx4_en: Fix IPv6 csum calculation

[Or Gerlitz]

On Fri, Aug 28, 2015 at 12:46 PM, Joe Perches <joe@perches.com> wrote:
> On Fri, 2015-08-28 at 14:28 -0500, clsoto@linux.vnet.ibm.com wrote:
>> From: Carol L Soto <clsoto@linux.vnet.ibm.com>
>>
>> Seeing this message with mlx4_eni with IPv6: hw csum failure
>>
>> Changing IPv6 csum calculation to be based on OFED 2.4 code.
>> When calculate IPv6 csum based also on source and destination
>> addresses.
> []
>> diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
> []
>> +static __wsum csum_ipv6_magic_nofold(const struct in6_addr *saddr,
>> +                                  const struct in6_addr *daddr,
>> +                                  __u32 len, unsigned short proto)
>> +{
>> +     __wsum res = 0;
>> +
>> +     res = csum_add(res, saddr->in6_u.u6_addr32[0]);
>> +     res = csum_add(res, saddr->in6_u.u6_addr32[1]);
>> +     res = csum_add(res, saddr->in6_u.u6_addr32[2]);
>> +     res = csum_add(res, saddr->in6_u.u6_addr32[3]);
>> +     res = csum_add(res, daddr->in6_u.u6_addr32[0]);
>> +     res = csum_add(res, daddr->in6_u.u6_addr32[1]);
>> +     res = csum_add(res, daddr->in6_u.u6_addr32[2]);
>> +     res = csum_add(res, daddr->in6_u.u6_addr32[3]);
>> +     res = csum_add(res, len);
>> +     res = csum_add(res, htonl(proto));
>
> You should try running sparse on this code.


Yep, if I remember correct, the patch is still under internal review
with making sparse/smatch happy on it being a pending comment.


Or.