net/ipv4/gre_offload.c

   1 /*
   2  *      IPV4 GSO/GRO offload support
   3  *      Linux INET implementation
   4  *
   5  *      This program is free software; you can redistribute it and/or
   6  *      modify it under the terms of the GNU General Public License
   7  *      as published by the Free Software Foundation; either version
   8  *      2 of the License, or (at your option) any later version.
   9  *
  10  *      GRE GSO support
  11  */
  12
  13 #include <linux/skbuff.h>
  14 #include <linux/init.h>
  15 #include <net/protocol.h>
  16 #include <net/gre.h>
  17
  18 static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
  19                                        netdev_features_t features)
  20 {
  21         struct sk_buff *segs = ERR_PTR(-EINVAL);
  22         netdev_features_t enc_features;
  23         int ghl;
  24         struct gre_base_hdr *greh;
  25         u16 mac_offset = skb->mac_header;
  26         int mac_len = skb->mac_len;
  27         __be16 protocol = skb->protocol;
  28         int tnl_hlen;
  29         bool csum;
  30
  31         if (unlikely(skb_shinfo(skb)->gso_type &
  32                                 ~(SKB_GSO_TCPV4 |
  33                                   SKB_GSO_TCPV6 |
  34                                   SKB_GSO_UDP |
  35                                   SKB_GSO_DODGY |
  36                                   SKB_GSO_TCP_ECN |
  37                                   SKB_GSO_GRE |
  38                                   SKB_GSO_GRE_CSUM |
  39                                   SKB_GSO_IPIP |
  40                                   SKB_GSO_SIT)))
  41                 goto out;
  42
  43         if (!skb->encapsulation)
  44                 goto out;
  45
  46         if (unlikely(!pskb_may_pull(skb, sizeof(*greh))))
  47                 goto out;
  48
  49         greh = (struct gre_base_hdr *)skb_transport_header(skb);
  50
  51         ghl = skb_inner_mac_header(skb) - skb_transport_header(skb);
  52         if (unlikely(ghl < sizeof(*greh)))
  53                 goto out;
  54
  55         csum = !!(greh->flags & GRE_CSUM);
  56         if (csum)
  57                 skb->encap_hdr_csum = 1;
  58
  59         /* setup inner skb. */
  60         skb->protocol = greh->protocol;
  61         skb->encapsulation = 0;
  62
  63         if (unlikely(!pskb_may_pull(skb, ghl)))
  64                 goto out;
  65
  66         __skb_pull(skb, ghl);
  67         skb_reset_mac_header(skb);
  68         skb_set_network_header(skb, skb_inner_network_offset(skb));
  69         skb->mac_len = skb_inner_network_offset(skb);
  70
  71         /* segment inner packet. */
  72         enc_features = skb->dev->hw_enc_features & features;
  73         segs = skb_mac_gso_segment(skb, enc_features);
  74         if (IS_ERR_OR_NULL(segs)) {
  75                 skb_gso_error_unwind(skb, protocol, ghl, mac_offset, mac_len);
  76                 goto out;
  77         }
  78
  79         skb = segs;
  80         tnl_hlen = skb_tnl_header_len(skb);
  81         do {
  82                 __skb_push(skb, ghl);
  83                 if (csum) {
  84                         __be32 *pcsum;
  85
  86                         if (skb_has_shared_frag(skb)) {
  87                                 int err;
  88
  89                                 err = __skb_linearize(skb);
  90                                 if (err) {
  91                                         kfree_skb_list(segs);
  92                                         segs = ERR_PTR(err);
  93                                         goto out;
  94                                 }
  95                         }
  96
  97                         skb_reset_transport_header(skb);
  98
  99                         greh = (struct gre_base_hdr *)
 100                             skb_transport_header(skb);
 101                         pcsum = (__be32 *)(greh + 1);
 102                         *pcsum = 0;
 103                         *(__sum16 *)pcsum = gso_make_checksum(skb, 0);
 104                 }
 105                 __skb_push(skb, tnl_hlen - ghl);
 106
 107                 skb_reset_inner_headers(skb);
 108                 skb->encapsulation = 1;
 109
 110                 skb_reset_mac_header(skb);
 111                 skb_set_network_header(skb, mac_len);
 112                 skb->mac_len = mac_len;
 113                 skb->protocol = protocol;
 114         } while ((skb = skb->next));
 115 out:
 116         return segs;
 117 }
 118
 119 static struct sk_buff **gre_gro_receive(struct sk_buff **head,
 120                                         struct sk_buff *skb)
 121 {
 122         struct sk_buff **pp = NULL;
 123         struct sk_buff *p;
 124         const struct gre_base_hdr *greh;
 125         unsigned int hlen, grehlen;
 126         unsigned int off;
 127         int flush = 1;
 128         struct packet_offload *ptype;
 129         __be16 type;
 130
 131         if (NAPI_GRO_CB(skb)->encap_mark)
 132                 goto out;
 133
 134         NAPI_GRO_CB(skb)->encap_mark = 1;
 135
 136         off = skb_gro_offset(skb);
 137         hlen = off + sizeof(*greh);
 138         greh = skb_gro_header_fast(skb, off);
 139         if (skb_gro_header_hard(skb, hlen)) {
 140                 greh = skb_gro_header_slow(skb, hlen, off);
 141                 if (unlikely(!greh))
 142                         goto out;
 143         }
 144
 145         /* Only support version 0 and K (key), C (csum) flags. Note that
 146          * although the support for the S (seq#) flag can be added easily
 147          * for GRO, this is problematic for GSO hence can not be enabled
 148          * here because a GRO pkt may end up in the forwarding path, thus
 149          * requiring GSO support to break it up correctly.
 150          */
 151         if ((greh->flags & ~(GRE_KEY|GRE_CSUM)) != 0)
 152                 goto out;
 153
 154         type = greh->protocol;
 155
 156         rcu_read_lock();
 157         ptype = gro_find_receive_by_type(type);
 158         if (!ptype)
 159                 goto out_unlock;
 160
 161         grehlen = GRE_HEADER_SECTION;
 162
 163         if (greh->flags & GRE_KEY)
 164                 grehlen += GRE_HEADER_SECTION;
 165
 166         if (greh->flags & GRE_CSUM)
 167                 grehlen += GRE_HEADER_SECTION;
 168
 169         hlen = off + grehlen;
 170         if (skb_gro_header_hard(skb, hlen)) {
 171                 greh = skb_gro_header_slow(skb, hlen, off);
 172                 if (unlikely(!greh))
 173                         goto out_unlock;
 174         }
 175
 176         /* Don't bother verifying checksum if we're going to flush anyway. */
 177         if ((greh->flags & GRE_CSUM) && !NAPI_GRO_CB(skb)->flush) {
 178                 if (skb_gro_checksum_simple_validate(skb))
 179                         goto out_unlock;
 180
 181                 skb_gro_checksum_try_convert(skb, IPPROTO_GRE, 0,
 182                                              null_compute_pseudo);
 183         }
 184
 185         flush = 0;
 186
 187         for (p = *head; p; p = p->next) {
 188                 const struct gre_base_hdr *greh2;
 189
 190                 if (!NAPI_GRO_CB(p)->same_flow)
 191                         continue;
 192
 193                 /* The following checks are needed to ensure only pkts
 194                  * from the same tunnel are considered for aggregation.
 195                  * The criteria for "the same tunnel" includes:
 196                  * 1) same version (we only support version 0 here)
 197                  * 2) same protocol (we only support ETH_P_IP for now)
 198                  * 3) same set of flags
 199                  * 4) same key if the key field is present.
 200                  */
 201                 greh2 = (struct gre_base_hdr *)(p->data + off);
 202
 203                 if (greh2->flags != greh->flags ||
 204                     greh2->protocol != greh->protocol) {
 205                         NAPI_GRO_CB(p)->same_flow = 0;
 206                         continue;
 207                 }
 208                 if (greh->flags & GRE_KEY) {
 209                         /* compare keys */
 210                         if (*(__be32 *)(greh2+1) != *(__be32 *)(greh+1)) {
 211                                 NAPI_GRO_CB(p)->same_flow = 0;
 212                                 continue;
 213                         }
 214                 }
 215         }
 216
 217         skb_gro_pull(skb, grehlen);
 218
 219         /* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/
 220         skb_gro_postpull_rcsum(skb, greh, grehlen);
 221
 222         pp = ptype->callbacks.gro_receive(head, skb);
 223
 224 out_unlock:
 225         rcu_read_unlock();
 226 out:
 227         NAPI_GRO_CB(skb)->flush |= flush;
 228
 229         return pp;
 230 }
 231
 232 static int gre_gro_complete(struct sk_buff *skb, int nhoff)
 233 {
 234         struct gre_base_hdr *greh = (struct gre_base_hdr *)(skb->data + nhoff);
 235         struct packet_offload *ptype;
 236         unsigned int grehlen = sizeof(*greh);
 237         int err = -ENOENT;
 238         __be16 type;
 239
 240         skb->encapsulation = 1;
 241         skb_shinfo(skb)->gso_type = SKB_GSO_GRE;
 242
 243         type = greh->protocol;
 244         if (greh->flags & GRE_KEY)
 245                 grehlen += GRE_HEADER_SECTION;
 246
 247         if (greh->flags & GRE_CSUM)
 248                 grehlen += GRE_HEADER_SECTION;
 249
 250         rcu_read_lock();
 251         ptype = gro_find_complete_by_type(type);
 252         if (ptype)
 253                 err = ptype->callbacks.gro_complete(skb, nhoff + grehlen);
 254
 255         rcu_read_unlock();
 256
 257         skb_set_inner_mac_header(skb, nhoff + grehlen);
 258
 259         return err;
 260 }
 261
 262 static const struct net_offload gre_offload = {
 263         .callbacks = {
 264                 .gso_segment = gre_gso_segment,
 265                 .gro_receive = gre_gro_receive,
 266                 .gro_complete = gre_gro_complete,
 267         },
 268 };
 269
 270 static int __init gre_offload_init(void)
 271 {
 272         return inet_add_offload(&gre_offload, IPPROTO_GRE);
 273 }
 274 device_initcall(gre_offload_init);