net: qualcomm: rmnet: Add support for RX checksum offload
authorSubash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Sun, 7 Jan 2018 18:36:37 +0000 (11:36 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 8 Jan 2018 18:58:49 +0000 (13:58 -0500)
When using the MAPv4 packet format, receive checksum offload can be
enabled in hardware. The checksum computation over pseudo header is
not offloaded but the rest of the checksum computation over
the payload is offloaded. This applies only for TCP / UDP packets
which are not fragmented.

rmnet validates the TCP/UDP checksum for the packet using the checksum
from the checksum trailer added to the packet by hardware. The
validation performed is as following -

1. Perform 1's complement over the checksum value from the trailer
2. Compute 1's complement checksum over IPv4 / IPv6 header and
   subtracts it from the value from step 1
3. Computes 1's complement checksum over IPv4 / IPv6 pseudo header and
   adds it to the value from step 2
4. Subtracts the checksum value from the TCP / UDP header from the
   value from step 3.
5. Compares the value from step 4 to the checksum value from the
   TCP / UDP header.
6. If the comparison in step 5 succeeds, CHECKSUM_UNNECESSARY is set
   and the packet is passed on to network stack. If there is a
   failure, then the packet is passed on as such without modifying
   the ip_summed field.

The checksum field is also checked for UDP checksum 0 as per RFC 768
and for unexpected TCP checksum of 0.

If checksum offload is disabled when using MAPv4 packet format in
receive path, the packet is queued as is to network stack without
the validations above.

Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/qualcomm/rmnet/rmnet_handlers.c
drivers/net/ethernet/qualcomm/rmnet/rmnet_map.h
drivers/net/ethernet/qualcomm/rmnet/rmnet_map_data.c
drivers/net/ethernet/qualcomm/rmnet/rmnet_vnd.c

index 8f8c4f205166f33e1bf6e3ecc37785290fb3ae28..340945857d23d5cf8879023556eaf362c74343eb 100644 (file)
@@ -66,8 +66,8 @@ __rmnet_map_ingress_handler(struct sk_buff *skb,
                            struct rmnet_port *port)
 {
        struct rmnet_endpoint *ep;
+       u16 len, pad;
        u8 mux_id;
-       u16 len;
 
        if (RMNET_MAP_GET_CD_BIT(skb)) {
                if (port->data_format & RMNET_INGRESS_FORMAT_MAP_COMMANDS)
@@ -77,7 +77,8 @@ __rmnet_map_ingress_handler(struct sk_buff *skb,
        }
 
        mux_id = RMNET_MAP_GET_MUX_ID(skb);
-       len = RMNET_MAP_GET_LENGTH(skb) - RMNET_MAP_GET_PAD(skb);
+       pad = RMNET_MAP_GET_PAD(skb);
+       len = RMNET_MAP_GET_LENGTH(skb) - pad;
 
        if (mux_id >= RMNET_MAX_LOGICAL_EP)
                goto free_skb;
@@ -90,8 +91,14 @@ __rmnet_map_ingress_handler(struct sk_buff *skb,
 
        /* Subtract MAP header */
        skb_pull(skb, sizeof(struct rmnet_map_header));
-       skb_trim(skb, len);
        rmnet_set_skb_proto(skb);
+
+       if (port->data_format & RMNET_INGRESS_FORMAT_MAP_CKSUMV4) {
+               if (!rmnet_map_checksum_downlink_packet(skb, len + pad))
+                       skb->ip_summed = CHECKSUM_UNNECESSARY;
+       }
+
+       skb_trim(skb, len);
        rmnet_deliver_skb(skb);
        return;
 
@@ -115,7 +122,7 @@ rmnet_map_ingress_handler(struct sk_buff *skb,
        }
 
        if (port->data_format & RMNET_INGRESS_FORMAT_DEAGGREGATION) {
-               while ((skbn = rmnet_map_deaggregate(skb)) != NULL)
+               while ((skbn = rmnet_map_deaggregate(skb, port)) != NULL)
                        __rmnet_map_ingress_handler(skbn, port);
 
                consume_skb(skb);
index 50c50cdc3e22f27ddab7330d45f69444e78843e9..ca9f4731f53ca066dec09fa68f06d0e964c5635d 100644 (file)
@@ -83,9 +83,11 @@ struct rmnet_map_ul_csum_header {
 #define RMNET_MAP_NO_PAD_BYTES        0
 #define RMNET_MAP_ADD_PAD_BYTES       1
 
-struct sk_buff *rmnet_map_deaggregate(struct sk_buff *skb);
+struct sk_buff *rmnet_map_deaggregate(struct sk_buff *skb,
+                                     struct rmnet_port *port);
 struct rmnet_map_header *rmnet_map_add_map_header(struct sk_buff *skb,
                                                  int hdrlen, int pad);
 void rmnet_map_command(struct sk_buff *skb, struct rmnet_port *port);
+int rmnet_map_checksum_downlink_packet(struct sk_buff *skb, u16 len);
 
 #endif /* _RMNET_MAP_H_ */
index 978ce2679b235f1c417ce3ad7d1f7efc3dc6b93b..881c1dc9d7aea6dfd180364e2a255bec14758ab8 100644 (file)
@@ -14,6 +14,9 @@
  */
 
 #include <linux/netdevice.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <net/ip6_checksum.h>
 #include "rmnet_config.h"
 #include "rmnet_map.h"
 #include "rmnet_private.h"
 #define RMNET_MAP_DEAGGR_SPACING  64
 #define RMNET_MAP_DEAGGR_HEADROOM (RMNET_MAP_DEAGGR_SPACING / 2)
 
+static __sum16 *rmnet_map_get_csum_field(unsigned char protocol,
+                                        const void *txporthdr)
+{
+       __sum16 *check = NULL;
+
+       switch (protocol) {
+       case IPPROTO_TCP:
+               check = &(((struct tcphdr *)txporthdr)->check);
+               break;
+
+       case IPPROTO_UDP:
+               check = &(((struct udphdr *)txporthdr)->check);
+               break;
+
+       default:
+               check = NULL;
+               break;
+       }
+
+       return check;
+}
+
+static int
+rmnet_map_ipv4_dl_csum_trailer(struct sk_buff *skb,
+                              struct rmnet_map_dl_csum_trailer *csum_trailer)
+{
+       __sum16 *csum_field, csum_temp, pseudo_csum, hdr_csum, ip_payload_csum;
+       u16 csum_value, csum_value_final;
+       struct iphdr *ip4h;
+       void *txporthdr;
+       __be16 addend;
+
+       ip4h = (struct iphdr *)(skb->data);
+       if ((ntohs(ip4h->frag_off) & IP_MF) ||
+           ((ntohs(ip4h->frag_off) & IP_OFFSET) > 0))
+               return -EOPNOTSUPP;
+
+       txporthdr = skb->data + ip4h->ihl * 4;
+
+       csum_field = rmnet_map_get_csum_field(ip4h->protocol, txporthdr);
+
+       if (!csum_field)
+               return -EPROTONOSUPPORT;
+
+       /* RFC 768 - Skip IPv4 UDP packets where sender checksum field is 0 */
+       if (*csum_field == 0 && ip4h->protocol == IPPROTO_UDP)
+               return 0;
+
+       csum_value = ~ntohs(csum_trailer->csum_value);
+       hdr_csum = ~ip_fast_csum(ip4h, (int)ip4h->ihl);
+       ip_payload_csum = csum16_sub((__force __sum16)csum_value,
+                                    (__force __be16)hdr_csum);
+
+       pseudo_csum = ~csum_tcpudp_magic(ip4h->saddr, ip4h->daddr,
+                                        ntohs(ip4h->tot_len) - ip4h->ihl * 4,
+                                        ip4h->protocol, 0);
+       addend = (__force __be16)ntohs((__force __be16)pseudo_csum);
+       pseudo_csum = csum16_add(ip_payload_csum, addend);
+
+       addend = (__force __be16)ntohs((__force __be16)*csum_field);
+       csum_temp = ~csum16_sub(pseudo_csum, addend);
+       csum_value_final = (__force u16)csum_temp;
+
+       if (unlikely(csum_value_final == 0)) {
+               switch (ip4h->protocol) {
+               case IPPROTO_UDP:
+                       /* RFC 768 - DL4 1's complement rule for UDP csum 0 */
+                       csum_value_final = ~csum_value_final;
+                       break;
+
+               case IPPROTO_TCP:
+                       /* DL4 Non-RFC compliant TCP checksum found */
+                       if (*csum_field == (__force __sum16)0xFFFF)
+                               csum_value_final = ~csum_value_final;
+                       break;
+               }
+       }
+
+       if (csum_value_final == ntohs((__force __be16)*csum_field))
+               return 0;
+       else
+               return -EINVAL;
+}
+
+#if IS_ENABLED(CONFIG_IPV6)
+static int
+rmnet_map_ipv6_dl_csum_trailer(struct sk_buff *skb,
+                              struct rmnet_map_dl_csum_trailer *csum_trailer)
+{
+       __sum16 *csum_field, ip6_payload_csum, pseudo_csum, csum_temp;
+       u16 csum_value, csum_value_final;
+       __be16 ip6_hdr_csum, addend;
+       struct ipv6hdr *ip6h;
+       void *txporthdr;
+       u32 length;
+
+       ip6h = (struct ipv6hdr *)(skb->data);
+
+       txporthdr = skb->data + sizeof(struct ipv6hdr);
+       csum_field = rmnet_map_get_csum_field(ip6h->nexthdr, txporthdr);
+
+       if (!csum_field)
+               return -EPROTONOSUPPORT;
+
+       csum_value = ~ntohs(csum_trailer->csum_value);
+       ip6_hdr_csum = (__force __be16)
+                       ~ntohs((__force __be16)ip_compute_csum(ip6h,
+                              (int)(txporthdr - (void *)(skb->data))));
+       ip6_payload_csum = csum16_sub((__force __sum16)csum_value,
+                                     ip6_hdr_csum);
+
+       length = (ip6h->nexthdr == IPPROTO_UDP) ?
+                ntohs(((struct udphdr *)txporthdr)->len) :
+                ntohs(ip6h->payload_len);
+       pseudo_csum = ~(csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
+                            length, ip6h->nexthdr, 0));
+       addend = (__force __be16)ntohs((__force __be16)pseudo_csum);
+       pseudo_csum = csum16_add(ip6_payload_csum, addend);
+
+       addend = (__force __be16)ntohs((__force __be16)*csum_field);
+       csum_temp = ~csum16_sub(pseudo_csum, addend);
+       csum_value_final = (__force u16)csum_temp;
+
+       if (unlikely(csum_value_final == 0)) {
+               switch (ip6h->nexthdr) {
+               case IPPROTO_UDP:
+                       /* RFC 2460 section 8.1
+                        * DL6 One's complement rule for UDP checksum 0
+                        */
+                       csum_value_final = ~csum_value_final;
+                       break;
+
+               case IPPROTO_TCP:
+                       /* DL6 Non-RFC compliant TCP checksum found */
+                       if (*csum_field == (__force __sum16)0xFFFF)
+                               csum_value_final = ~csum_value_final;
+                       break;
+               }
+       }
+
+       if (csum_value_final == ntohs((__force __be16)*csum_field))
+               return 0;
+       else
+               return -EINVAL;
+}
+#endif
+
 /* Adds MAP header to front of skb->data
  * Padding is calculated and set appropriately in MAP header. Mux ID is
  * initialized to 0.
@@ -66,7 +216,8 @@ done:
  * returned, indicating that there are no more packets to deaggregate. Caller
  * is responsible for freeing the original skb.
  */
-struct sk_buff *rmnet_map_deaggregate(struct sk_buff *skb)
+struct sk_buff *rmnet_map_deaggregate(struct sk_buff *skb,
+                                     struct rmnet_port *port)
 {
        struct rmnet_map_header *maph;
        struct sk_buff *skbn;
@@ -78,6 +229,9 @@ struct sk_buff *rmnet_map_deaggregate(struct sk_buff *skb)
        maph = (struct rmnet_map_header *)skb->data;
        packet_len = ntohs(maph->pkt_len) + sizeof(struct rmnet_map_header);
 
+       if (port->data_format & RMNET_INGRESS_FORMAT_MAP_CKSUMV4)
+               packet_len += sizeof(struct rmnet_map_dl_csum_trailer);
+
        if (((int)skb->len - (int)packet_len) < 0)
                return NULL;
 
@@ -97,3 +251,33 @@ struct sk_buff *rmnet_map_deaggregate(struct sk_buff *skb)
 
        return skbn;
 }
+
+/* Validates packet checksums. Function takes a pointer to
+ * the beginning of a buffer which contains the IP payload +
+ * padding + checksum trailer.
+ * Only IPv4 and IPv6 are supported along with TCP & UDP.
+ * Fragmented or tunneled packets are not supported.
+ */
+int rmnet_map_checksum_downlink_packet(struct sk_buff *skb, u16 len)
+{
+       struct rmnet_map_dl_csum_trailer *csum_trailer;
+
+       if (unlikely(!(skb->dev->features & NETIF_F_RXCSUM)))
+               return -EOPNOTSUPP;
+
+       csum_trailer = (struct rmnet_map_dl_csum_trailer *)(skb->data + len);
+
+       if (!csum_trailer->valid)
+               return -EINVAL;
+
+       if (skb->protocol == htons(ETH_P_IP))
+               return rmnet_map_ipv4_dl_csum_trailer(skb, csum_trailer);
+       else if (skb->protocol == htons(ETH_P_IPV6))
+#if IS_ENABLED(CONFIG_IPV6)
+               return rmnet_map_ipv6_dl_csum_trailer(skb, csum_trailer);
+#else
+               return -EPROTONOSUPPORT;
+#endif
+
+       return 0;
+}
index 5bb29f44d114a53adca1b6b04084560e7928d6b0..879a2e01acdbf9b4c328ce1680ab0aceddc5fb12 100644 (file)
@@ -188,6 +188,8 @@ int rmnet_vnd_newlink(u8 id, struct net_device *rmnet_dev,
        if (rmnet_get_endpoint(port, id))
                return -EBUSY;
 
+       rmnet_dev->hw_features = NETIF_F_RXCSUM;
+
        rc = register_netdevice(rmnet_dev);
        if (!rc) {
                ep->egress_dev = rmnet_dev;