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 int ipv6_gso_send_check(struct sk_buff *skb)
  57 {
  58         const struct ipv6hdr *ipv6h;
  59         const struct net_offload *ops;
  60         int err = -EINVAL;
  61
  62         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  63                 goto out;
  64
  65         ipv6h = ipv6_hdr(skb);
  66         __skb_pull(skb, sizeof(*ipv6h));
  67         err = -EPROTONOSUPPORT;
  68
  69         ops = rcu_dereference(inet6_offloads[
  70                 ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr)]);
  71
  72         if (likely(ops && ops->callbacks.gso_send_check)) {
  73                 skb_reset_transport_header(skb);
  74                 err = ops->callbacks.gso_send_check(skb);
  75         }
  76
  77 out:
  78         return err;
  79 }
  80
  81 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
  82         netdev_features_t features)
  83 {
  84         struct sk_buff *segs = ERR_PTR(-EINVAL);
  85         struct ipv6hdr *ipv6h;
  86         const struct net_offload *ops;
  87         int proto;
  88         struct frag_hdr *fptr;
  89         unsigned int unfrag_ip6hlen;
  90         u8 *prevhdr;
  91         int offset = 0;
  92         bool tunnel;
  93         int nhoff;
  94
  95         if (unlikely(skb_shinfo(skb)->gso_type &
  96                      ~(SKB_GSO_UDP |
  97                        SKB_GSO_DODGY |
  98                        SKB_GSO_TCP_ECN |
  99                        SKB_GSO_GRE |
 100                        SKB_GSO_IPIP |
 101                        SKB_GSO_SIT |
 102                        SKB_GSO_UDP_TUNNEL |
 103                        SKB_GSO_MPLS |
 104                        SKB_GSO_TCPV6 |
 105                        0)))
 106                 goto out;
 107
 108         skb_reset_network_header(skb);
 109         nhoff = skb_network_header(skb) - skb_mac_header(skb);
 110         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
 111                 goto out;
 112
 113         tunnel = SKB_GSO_CB(skb)->encap_level > 0;
 114         if (tunnel)
 115                 features = skb->dev->hw_enc_features & netif_skb_features(skb);
 116         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
 117
 118         ipv6h = ipv6_hdr(skb);
 119         __skb_pull(skb, sizeof(*ipv6h));
 120         segs = ERR_PTR(-EPROTONOSUPPORT);
 121
 122         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
 123
 124         ops = rcu_dereference(inet6_offloads[proto]);
 125         if (likely(ops && ops->callbacks.gso_segment)) {
 126                 skb_reset_transport_header(skb);
 127                 segs = ops->callbacks.gso_segment(skb, features);
 128         }
 129
 130         if (IS_ERR(segs))
 131                 goto out;
 132
 133         for (skb = segs; skb; skb = skb->next) {
 134                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 135                 ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h));
 136                 if (tunnel) {
 137                         skb_reset_inner_headers(skb);
 138                         skb->encapsulation = 1;
 139                 }
 140                 skb->network_header = (u8 *)ipv6h - skb->head;
 141
 142                 if (!tunnel && proto == IPPROTO_UDP) {
 143                         unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
 144                         fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
 145                         fptr->frag_off = htons(offset);
 146                         if (skb->next != NULL)
 147                                 fptr->frag_off |= htons(IP6_MF);
 148                         offset += (ntohs(ipv6h->payload_len) -
 149                                    sizeof(struct frag_hdr));
 150                 }
 151         }
 152
 153 out:
 154         return segs;
 155 }
 156
 157 /* Return the total length of all the extension hdrs, following the same
 158  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 159  */
 160 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 161                             const struct net_offload **opps)
 162 {
 163         struct ipv6_opt_hdr *opth = (void *)iph;
 164         int len = 0, proto, optlen = sizeof(*iph);
 165
 166         proto = iph->nexthdr;
 167         for (;;) {
 168                 if (proto != NEXTHDR_HOP) {
 169                         *opps = rcu_dereference(inet6_offloads[proto]);
 170                         if (unlikely(!(*opps)))
 171                                 break;
 172                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 173                                 break;
 174                 }
 175                 opth = (void *)opth + optlen;
 176                 optlen = ipv6_optlen(opth);
 177                 len += optlen;
 178                 proto = opth->nexthdr;
 179         }
 180         return len;
 181 }
 182
 183 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 184                                          struct sk_buff *skb)
 185 {
 186         const struct net_offload *ops;
 187         struct sk_buff **pp = NULL;
 188         struct sk_buff *p;
 189         struct ipv6hdr *iph;
 190         unsigned int nlen;
 191         unsigned int hlen;
 192         unsigned int off;
 193         u16 flush = 1;
 194         int proto;
 195         __wsum csum;
 196
 197         off = skb_gro_offset(skb);
 198         hlen = off + sizeof(*iph);
 199         iph = skb_gro_header_fast(skb, off);
 200         if (skb_gro_header_hard(skb, hlen)) {
 201                 iph = skb_gro_header_slow(skb, hlen, off);
 202                 if (unlikely(!iph))
 203                         goto out;
 204         }
 205
 206         skb_set_network_header(skb, off);
 207         skb_gro_pull(skb, sizeof(*iph));
 208         skb_set_transport_header(skb, skb_gro_offset(skb));
 209
 210         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 211
 212         rcu_read_lock();
 213         proto = iph->nexthdr;
 214         ops = rcu_dereference(inet6_offloads[proto]);
 215         if (!ops || !ops->callbacks.gro_receive) {
 216                 __pskb_pull(skb, skb_gro_offset(skb));
 217                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 218                 skb_gro_pull(skb, -skb_transport_offset(skb));
 219                 skb_reset_transport_header(skb);
 220                 __skb_push(skb, skb_gro_offset(skb));
 221
 222                 ops = rcu_dereference(inet6_offloads[proto]);
 223                 if (!ops || !ops->callbacks.gro_receive)
 224                         goto out_unlock;
 225
 226                 iph = ipv6_hdr(skb);
 227         }
 228
 229         NAPI_GRO_CB(skb)->proto = proto;
 230
 231         flush--;
 232         nlen = skb_network_header_len(skb);
 233
 234         for (p = *head; p; p = p->next) {
 235                 const struct ipv6hdr *iph2;
 236                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 237
 238                 if (!NAPI_GRO_CB(p)->same_flow)
 239                         continue;
 240
 241                 iph2 = (struct ipv6hdr *)(p->data + off);
 242                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2 ;
 243
 244                 /* All fields must match except length and Traffic Class.
 245                  * XXX skbs on the gro_list have all been parsed and pulled
 246                  * already so we don't need to compare nlen
 247                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 248                  * memcmp() alone below is suffcient, right?
 249                  */
 250                  if ((first_word & htonl(0xF00FFFFF)) ||
 251                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 252                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 253                         NAPI_GRO_CB(p)->same_flow = 0;
 254                         continue;
 255                 }
 256                 /* flush if Traffic Class fields are different */
 257                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 258                 NAPI_GRO_CB(p)->flush |= flush;
 259         }
 260
 261         NAPI_GRO_CB(skb)->flush |= flush;
 262
 263         csum = skb->csum;
 264         skb_postpull_rcsum(skb, iph, skb_network_header_len(skb));
 265
 266         pp = ops->callbacks.gro_receive(head, skb);
 267
 268         skb->csum = csum;
 269
 270 out_unlock:
 271         rcu_read_unlock();
 272
 273 out:
 274         NAPI_GRO_CB(skb)->flush |= flush;
 275
 276         return pp;
 277 }
 278
 279 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 280 {
 281         const struct net_offload *ops;
 282         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 283         int err = -ENOSYS;
 284
 285         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 286
 287         rcu_read_lock();
 288
 289         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 290         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 291                 goto out_unlock;
 292
 293         err = ops->callbacks.gro_complete(skb, nhoff);
 294
 295 out_unlock:
 296         rcu_read_unlock();
 297
 298         return err;
 299 }
 300
 301 static struct packet_offload ipv6_packet_offload __read_mostly = {
 302         .type = cpu_to_be16(ETH_P_IPV6),
 303         .callbacks = {
 304                 .gso_send_check = ipv6_gso_send_check,
 305                 .gso_segment = ipv6_gso_segment,
 306                 .gro_receive = ipv6_gro_receive,
 307                 .gro_complete = ipv6_gro_complete,
 308         },
 309 };
 310
 311 static const struct net_offload sit_offload = {
 312         .callbacks = {
 313                 .gso_send_check = ipv6_gso_send_check,
 314                 .gso_segment    = ipv6_gso_segment,
 315         },
 316 };
 317
 318 static int __init ipv6_offload_init(void)
 319 {
 320
 321         if (tcpv6_offload_init() < 0)
 322                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 323         if (udp_offload_init() < 0)
 324                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
 325         if (ipv6_exthdrs_offload_init() < 0)
 326                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 327
 328         dev_add_offload(&ipv6_packet_offload);
 329
 330         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 331
 332         return 0;
 333 }
 334
 335 fs_initcall(ipv6_offload_init);