Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / net / ipv4 / gre_offload.c
diff --git a/kernel/net/ipv4/gre_offload.c b/kernel/net/ipv4/gre_offload.c
new file mode 100644 (file)
index 0000000..5aa46d4
--- /dev/null
@@ -0,0 +1,268 @@
+/*
+ *     IPV4 GSO/GRO offload support
+ *     Linux INET implementation
+ *
+ *     This program is free software; you can redistribute it and/or
+ *     modify it under the terms of the GNU General Public License
+ *     as published by the Free Software Foundation; either version
+ *     2 of the License, or (at your option) any later version.
+ *
+ *     GRE GSO support
+ */
+
+#include <linux/skbuff.h>
+#include <linux/init.h>
+#include <net/protocol.h>
+#include <net/gre.h>
+
+static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
+                                      netdev_features_t features)
+{
+       struct sk_buff *segs = ERR_PTR(-EINVAL);
+       netdev_features_t enc_features;
+       int ghl;
+       struct gre_base_hdr *greh;
+       u16 mac_offset = skb->mac_header;
+       int mac_len = skb->mac_len;
+       __be16 protocol = skb->protocol;
+       int tnl_hlen;
+       bool csum;
+
+       if (unlikely(skb_shinfo(skb)->gso_type &
+                               ~(SKB_GSO_TCPV4 |
+                                 SKB_GSO_TCPV6 |
+                                 SKB_GSO_UDP |
+                                 SKB_GSO_DODGY |
+                                 SKB_GSO_TCP_ECN |
+                                 SKB_GSO_GRE |
+                                 SKB_GSO_GRE_CSUM |
+                                 SKB_GSO_IPIP)))
+               goto out;
+
+       if (!skb->encapsulation)
+               goto out;
+
+       if (unlikely(!pskb_may_pull(skb, sizeof(*greh))))
+               goto out;
+
+       greh = (struct gre_base_hdr *)skb_transport_header(skb);
+
+       ghl = skb_inner_mac_header(skb) - skb_transport_header(skb);
+       if (unlikely(ghl < sizeof(*greh)))
+               goto out;
+
+       csum = !!(greh->flags & GRE_CSUM);
+       if (csum)
+               skb->encap_hdr_csum = 1;
+
+       /* setup inner skb. */
+       skb->protocol = greh->protocol;
+       skb->encapsulation = 0;
+
+       if (unlikely(!pskb_may_pull(skb, ghl)))
+               goto out;
+
+       __skb_pull(skb, ghl);
+       skb_reset_mac_header(skb);
+       skb_set_network_header(skb, skb_inner_network_offset(skb));
+       skb->mac_len = skb_inner_network_offset(skb);
+
+       /* segment inner packet. */
+       enc_features = skb->dev->hw_enc_features & features;
+       segs = skb_mac_gso_segment(skb, enc_features);
+       if (IS_ERR_OR_NULL(segs)) {
+               skb_gso_error_unwind(skb, protocol, ghl, mac_offset, mac_len);
+               goto out;
+       }
+
+       skb = segs;
+       tnl_hlen = skb_tnl_header_len(skb);
+       do {
+               __skb_push(skb, ghl);
+               if (csum) {
+                       __be32 *pcsum;
+
+                       if (skb_has_shared_frag(skb)) {
+                               int err;
+
+                               err = __skb_linearize(skb);
+                               if (err) {
+                                       kfree_skb_list(segs);
+                                       segs = ERR_PTR(err);
+                                       goto out;
+                               }
+                       }
+
+                       skb_reset_transport_header(skb);
+
+                       greh = (struct gre_base_hdr *)
+                           skb_transport_header(skb);
+                       pcsum = (__be32 *)(greh + 1);
+                       *pcsum = 0;
+                       *(__sum16 *)pcsum = gso_make_checksum(skb, 0);
+               }
+               __skb_push(skb, tnl_hlen - ghl);
+
+               skb_reset_inner_headers(skb);
+               skb->encapsulation = 1;
+
+               skb_reset_mac_header(skb);
+               skb_set_network_header(skb, mac_len);
+               skb->mac_len = mac_len;
+               skb->protocol = protocol;
+       } while ((skb = skb->next));
+out:
+       return segs;
+}
+
+static struct sk_buff **gre_gro_receive(struct sk_buff **head,
+                                       struct sk_buff *skb)
+{
+       struct sk_buff **pp = NULL;
+       struct sk_buff *p;
+       const struct gre_base_hdr *greh;
+       unsigned int hlen, grehlen;
+       unsigned int off;
+       int flush = 1;
+       struct packet_offload *ptype;
+       __be16 type;
+
+       off = skb_gro_offset(skb);
+       hlen = off + sizeof(*greh);
+       greh = skb_gro_header_fast(skb, off);
+       if (skb_gro_header_hard(skb, hlen)) {
+               greh = skb_gro_header_slow(skb, hlen, off);
+               if (unlikely(!greh))
+                       goto out;
+       }
+
+       /* Only support version 0 and K (key), C (csum) flags. Note that
+        * although the support for the S (seq#) flag can be added easily
+        * for GRO, this is problematic for GSO hence can not be enabled
+        * here because a GRO pkt may end up in the forwarding path, thus
+        * requiring GSO support to break it up correctly.
+        */
+       if ((greh->flags & ~(GRE_KEY|GRE_CSUM)) != 0)
+               goto out;
+
+       type = greh->protocol;
+
+       rcu_read_lock();
+       ptype = gro_find_receive_by_type(type);
+       if (!ptype)
+               goto out_unlock;
+
+       grehlen = GRE_HEADER_SECTION;
+
+       if (greh->flags & GRE_KEY)
+               grehlen += GRE_HEADER_SECTION;
+
+       if (greh->flags & GRE_CSUM)
+               grehlen += GRE_HEADER_SECTION;
+
+       hlen = off + grehlen;
+       if (skb_gro_header_hard(skb, hlen)) {
+               greh = skb_gro_header_slow(skb, hlen, off);
+               if (unlikely(!greh))
+                       goto out_unlock;
+       }
+
+       /* Don't bother verifying checksum if we're going to flush anyway. */
+       if ((greh->flags & GRE_CSUM) && !NAPI_GRO_CB(skb)->flush) {
+               if (skb_gro_checksum_simple_validate(skb))
+                       goto out_unlock;
+
+               skb_gro_checksum_try_convert(skb, IPPROTO_GRE, 0,
+                                            null_compute_pseudo);
+       }
+
+       flush = 0;
+
+       for (p = *head; p; p = p->next) {
+               const struct gre_base_hdr *greh2;
+
+               if (!NAPI_GRO_CB(p)->same_flow)
+                       continue;
+
+               /* The following checks are needed to ensure only pkts
+                * from the same tunnel are considered for aggregation.
+                * The criteria for "the same tunnel" includes:
+                * 1) same version (we only support version 0 here)
+                * 2) same protocol (we only support ETH_P_IP for now)
+                * 3) same set of flags
+                * 4) same key if the key field is present.
+                */
+               greh2 = (struct gre_base_hdr *)(p->data + off);
+
+               if (greh2->flags != greh->flags ||
+                   greh2->protocol != greh->protocol) {
+                       NAPI_GRO_CB(p)->same_flow = 0;
+                       continue;
+               }
+               if (greh->flags & GRE_KEY) {
+                       /* compare keys */
+                       if (*(__be32 *)(greh2+1) != *(__be32 *)(greh+1)) {
+                               NAPI_GRO_CB(p)->same_flow = 0;
+                               continue;
+                       }
+               }
+       }
+
+       skb_gro_pull(skb, grehlen);
+
+       /* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/
+       skb_gro_postpull_rcsum(skb, greh, grehlen);
+
+       pp = ptype->callbacks.gro_receive(head, skb);
+
+out_unlock:
+       rcu_read_unlock();
+out:
+       NAPI_GRO_CB(skb)->flush |= flush;
+
+       return pp;
+}
+
+static int gre_gro_complete(struct sk_buff *skb, int nhoff)
+{
+       struct gre_base_hdr *greh = (struct gre_base_hdr *)(skb->data + nhoff);
+       struct packet_offload *ptype;
+       unsigned int grehlen = sizeof(*greh);
+       int err = -ENOENT;
+       __be16 type;
+
+       skb->encapsulation = 1;
+       skb_shinfo(skb)->gso_type = SKB_GSO_GRE;
+
+       type = greh->protocol;
+       if (greh->flags & GRE_KEY)
+               grehlen += GRE_HEADER_SECTION;
+
+       if (greh->flags & GRE_CSUM)
+               grehlen += GRE_HEADER_SECTION;
+
+       rcu_read_lock();
+       ptype = gro_find_complete_by_type(type);
+       if (ptype)
+               err = ptype->callbacks.gro_complete(skb, nhoff + grehlen);
+
+       rcu_read_unlock();
+
+       skb_set_inner_mac_header(skb, nhoff + grehlen);
+
+       return err;
+}
+
+static const struct net_offload gre_offload = {
+       .callbacks = {
+               .gso_segment = gre_gso_segment,
+               .gro_receive = gre_gro_receive,
+               .gro_complete = gre_gro_complete,
+       },
+};
+
+static int __init gre_offload_init(void)
+{
+       return inet_add_offload(&gre_offload, IPPROTO_GRE);
+}
+device_initcall(gre_offload_init);