net/ipv6/ip6_offload.c

   1 /*
   2  *      IPV6 GSO/GRO offload support
   3  *      Linux INET6 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
  11 #include <linux/kernel.h>
  12 #include <linux/socket.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/printk.h>
  16
  17 #include <net/protocol.h>
  18 #include <net/ipv6.h>
  19
  20 #include "ip6_offload.h"
  21
  22 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
  23 {
  24         const struct net_offload *ops = NULL;
  25
  26         for (;;) {
  27                 struct ipv6_opt_hdr *opth;
  28                 int len;
  29
  30                 if (proto != NEXTHDR_HOP) {
  31                         ops = rcu_dereference(inet6_offloads[proto]);
  32
  33                         if (unlikely(!ops))
  34                                 break;
  35
  36                         if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
  37                                 break;
  38                 }
  39
  40                 if (unlikely(!pskb_may_pull(skb, 8)))
  41                         break;
  42
  43                 opth = (void *)skb->data;
  44                 len = ipv6_optlen(opth);
  45
  46                 if (unlikely(!pskb_may_pull(skb, len)))
  47                         break;
  48
  49                 proto = opth->nexthdr;
  50                 __skb_pull(skb, len);
  51         }
  52
  53         return proto;
  54 }
  55
  56 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
  57         netdev_features_t features)
  58 {
  59         struct sk_buff *segs = ERR_PTR(-EINVAL);
  60         struct ipv6hdr *ipv6h;
  61         const struct net_offload *ops;
  62         int proto;
  63         struct frag_hdr *fptr;
  64         unsigned int unfrag_ip6hlen;
  65         u8 *prevhdr;
  66         int offset = 0;
  67         bool encap, udpfrag;
  68         int nhoff;
  69
  70         if (unlikely(skb_shinfo(skb)->gso_type &
  71                      ~(SKB_GSO_UDP |
  72                        SKB_GSO_DODGY |
  73                        SKB_GSO_TCP_ECN |
  74                        SKB_GSO_GRE |
  75                        SKB_GSO_GRE_CSUM |
  76                        SKB_GSO_IPIP |
  77                        SKB_GSO_SIT |
  78                        SKB_GSO_UDP_TUNNEL |
  79                        SKB_GSO_UDP_TUNNEL_CSUM |
  80                        SKB_GSO_MPLS |
  81                        SKB_GSO_TCPV6 |
  82                        0)))
  83                 goto out;
  84
  85         skb_reset_network_header(skb);
  86         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  87         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  88                 goto out;
  89
  90         encap = SKB_GSO_CB(skb)->encap_level > 0;
  91         if (encap)
  92                 features = skb->dev->hw_enc_features & netif_skb_features(skb);
  93         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  94
  95         ipv6h = ipv6_hdr(skb);
  96         __skb_pull(skb, sizeof(*ipv6h));
  97         segs = ERR_PTR(-EPROTONOSUPPORT);
  98
  99         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
 100
 101         if (skb->encapsulation &&
 102             skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
 103                 udpfrag = proto == IPPROTO_UDP && encap;
 104         else
 105                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
 106
 107         ops = rcu_dereference(inet6_offloads[proto]);
 108         if (likely(ops && ops->callbacks.gso_segment)) {
 109                 skb_reset_transport_header(skb);
 110                 segs = ops->callbacks.gso_segment(skb, features);
 111         }
 112
 113         if (IS_ERR(segs))
 114                 goto out;
 115
 116         for (skb = segs; skb; skb = skb->next) {
 117                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 118                 ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h));
 119                 skb->network_header = (u8 *)ipv6h - skb->head;
 120
 121                 if (udpfrag) {
 122                         unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
 123                         fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
 124                         fptr->frag_off = htons(offset);
 125                         if (skb->next != NULL)
 126                                 fptr->frag_off |= htons(IP6_MF);
 127                         offset += (ntohs(ipv6h->payload_len) -
 128                                    sizeof(struct frag_hdr));
 129                 }
 130                 if (encap)
 131                         skb_reset_inner_headers(skb);
 132         }
 133
 134 out:
 135         return segs;
 136 }
 137
 138 /* Return the total length of all the extension hdrs, following the same
 139  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 140  */
 141 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 142                             const struct net_offload **opps)
 143 {
 144         struct ipv6_opt_hdr *opth = (void *)iph;
 145         int len = 0, proto, optlen = sizeof(*iph);
 146
 147         proto = iph->nexthdr;
 148         for (;;) {
 149                 if (proto != NEXTHDR_HOP) {
 150                         *opps = rcu_dereference(inet6_offloads[proto]);
 151                         if (unlikely(!(*opps)))
 152                                 break;
 153                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 154                                 break;
 155                 }
 156                 opth = (void *)opth + optlen;
 157                 optlen = ipv6_optlen(opth);
 158                 len += optlen;
 159                 proto = opth->nexthdr;
 160         }
 161         return len;
 162 }
 163
 164 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 165                                          struct sk_buff *skb)
 166 {
 167         const struct net_offload *ops;
 168         struct sk_buff **pp = NULL;
 169         struct sk_buff *p;
 170         struct ipv6hdr *iph;
 171         unsigned int nlen;
 172         unsigned int hlen;
 173         unsigned int off;
 174         u16 flush = 1;
 175         int proto;
 176
 177         off = skb_gro_offset(skb);
 178         hlen = off + sizeof(*iph);
 179         iph = skb_gro_header_fast(skb, off);
 180         if (skb_gro_header_hard(skb, hlen)) {
 181                 iph = skb_gro_header_slow(skb, hlen, off);
 182                 if (unlikely(!iph))
 183                         goto out;
 184         }
 185
 186         skb_set_network_header(skb, off);
 187         skb_gro_pull(skb, sizeof(*iph));
 188         skb_set_transport_header(skb, skb_gro_offset(skb));
 189
 190         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 191
 192         rcu_read_lock();
 193         proto = iph->nexthdr;
 194         ops = rcu_dereference(inet6_offloads[proto]);
 195         if (!ops || !ops->callbacks.gro_receive) {
 196                 __pskb_pull(skb, skb_gro_offset(skb));
 197                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 198                 skb_gro_pull(skb, -skb_transport_offset(skb));
 199                 skb_reset_transport_header(skb);
 200                 __skb_push(skb, skb_gro_offset(skb));
 201
 202                 ops = rcu_dereference(inet6_offloads[proto]);
 203                 if (!ops || !ops->callbacks.gro_receive)
 204                         goto out_unlock;
 205
 206                 iph = ipv6_hdr(skb);
 207         }
 208
 209         NAPI_GRO_CB(skb)->proto = proto;
 210
 211         flush--;
 212         nlen = skb_network_header_len(skb);
 213
 214         for (p = *head; p; p = p->next) {
 215                 const struct ipv6hdr *iph2;
 216                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 217
 218                 if (!NAPI_GRO_CB(p)->same_flow)
 219                         continue;
 220
 221                 iph2 = (struct ipv6hdr *)(p->data + off);
 222                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 223
 224                 /* All fields must match except length and Traffic Class.
 225                  * XXX skbs on the gro_list have all been parsed and pulled
 226                  * already so we don't need to compare nlen
 227                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 228                  * memcmp() alone below is suffcient, right?
 229                  */
 230                  if ((first_word & htonl(0xF00FFFFF)) ||
 231                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 232                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 233                         NAPI_GRO_CB(p)->same_flow = 0;
 234                         continue;
 235                 }
 236                 /* flush if Traffic Class fields are different */
 237                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 238                 NAPI_GRO_CB(p)->flush |= flush;
 239
 240                 /* Clear flush_id, there's really no concept of ID in IPv6. */
 241                 NAPI_GRO_CB(p)->flush_id = 0;
 242         }
 243
 244         NAPI_GRO_CB(skb)->flush |= flush;
 245
 246         skb_gro_postpull_rcsum(skb, iph, nlen);
 247
 248         pp = ops->callbacks.gro_receive(head, skb);
 249
 250 out_unlock:
 251         rcu_read_unlock();
 252
 253 out:
 254         NAPI_GRO_CB(skb)->flush |= flush;
 255
 256         return pp;
 257 }
 258
 259 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 260 {
 261         const struct net_offload *ops;
 262         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 263         int err = -ENOSYS;
 264
 265         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 266
 267         rcu_read_lock();
 268
 269         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 270         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 271                 goto out_unlock;
 272
 273         err = ops->callbacks.gro_complete(skb, nhoff);
 274
 275 out_unlock:
 276         rcu_read_unlock();
 277
 278         return err;
 279 }
 280
 281 static struct packet_offload ipv6_packet_offload __read_mostly = {
 282         .type = cpu_to_be16(ETH_P_IPV6),
 283         .callbacks = {
 284                 .gso_segment = ipv6_gso_segment,
 285                 .gro_receive = ipv6_gro_receive,
 286                 .gro_complete = ipv6_gro_complete,
 287         },
 288 };
 289
 290 static const struct net_offload sit_offload = {
 291         .callbacks = {
 292                 .gso_segment    = ipv6_gso_segment,
 293                 .gro_receive    = ipv6_gro_receive,
 294                 .gro_complete   = ipv6_gro_complete,
 295         },
 296 };
 297
 298 static int __init ipv6_offload_init(void)
 299 {
 300
 301         if (tcpv6_offload_init() < 0)
 302                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 303         if (udp_offload_init() < 0)
 304                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
 305         if (ipv6_exthdrs_offload_init() < 0)
 306                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 307
 308         dev_add_offload(&ipv6_packet_offload);
 309
 310         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 311
 312         return 0;
 313 }
 314
 315 fs_initcall(ipv6_offload_init);