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                 opth = (void *)skb->data;
  50                 proto = opth->nexthdr;
  51                 __skb_pull(skb, len);
  52         }
  53
  54         return proto;
  55 }
  56
  57 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
  58         netdev_features_t features)
  59 {
  60         struct sk_buff *segs = ERR_PTR(-EINVAL);
  61         struct ipv6hdr *ipv6h;
  62         const struct net_offload *ops;
  63         int proto;
  64         struct frag_hdr *fptr;
  65         unsigned int unfrag_ip6hlen;
  66         unsigned int payload_len;
  67         u8 *prevhdr;
  68         int offset = 0;
  69         bool encap, udpfrag;
  70         int nhoff;
  71
  72         if (unlikely(skb_shinfo(skb)->gso_type &
  73                      ~(SKB_GSO_TCPV4 |
  74                        SKB_GSO_UDP |
  75                        SKB_GSO_DODGY |
  76                        SKB_GSO_TCP_ECN |
  77                        SKB_GSO_TCP_FIXEDID |
  78                        SKB_GSO_TCPV6 |
  79                        SKB_GSO_GRE |
  80                        SKB_GSO_GRE_CSUM |
  81                        SKB_GSO_IPIP |
  82                        SKB_GSO_SIT |
  83                        SKB_GSO_UDP_TUNNEL |
  84                        SKB_GSO_UDP_TUNNEL_CSUM |
  85                        SKB_GSO_TUNNEL_REMCSUM |
  86                        SKB_GSO_PARTIAL |
  87                        0)))
  88                 goto out;
  89
  90         skb_reset_network_header(skb);
  91         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  92         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  93                 goto out;
  94
  95         encap = SKB_GSO_CB(skb)->encap_level > 0;
  96         if (encap)
  97                 features &= skb->dev->hw_enc_features;
  98         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  99
 100         ipv6h = ipv6_hdr(skb);
 101         __skb_pull(skb, sizeof(*ipv6h));
 102         segs = ERR_PTR(-EPROTONOSUPPORT);
 103
 104         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
 105
 106         if (skb->encapsulation &&
 107             skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
 108                 udpfrag = proto == IPPROTO_UDP && encap;
 109         else
 110                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
 111
 112         ops = rcu_dereference(inet6_offloads[proto]);
 113         if (likely(ops && ops->callbacks.gso_segment)) {
 114                 skb_reset_transport_header(skb);
 115                 segs = ops->callbacks.gso_segment(skb, features);
 116         }
 117
 118         if (IS_ERR(segs))
 119                 goto out;
 120
 121         for (skb = segs; skb; skb = skb->next) {
 122                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 123                 if (skb_is_gso(skb))
 124                         payload_len = skb_shinfo(skb)->gso_size +
 125                                       SKB_GSO_CB(skb)->data_offset +
 126                                       skb->head - (unsigned char *)(ipv6h + 1);
 127                 else
 128                         payload_len = skb->len - nhoff - sizeof(*ipv6h);
 129                 ipv6h->payload_len = htons(payload_len);
 130                 skb->network_header = (u8 *)ipv6h - skb->head;
 131
 132                 if (udpfrag) {
 133                         unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
 134                         fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
 135                         fptr->frag_off = htons(offset);
 136                         if (skb->next)
 137                                 fptr->frag_off |= htons(IP6_MF);
 138                         offset += (ntohs(ipv6h->payload_len) -
 139                                    sizeof(struct frag_hdr));
 140                 }
 141                 if (encap)
 142                         skb_reset_inner_headers(skb);
 143         }
 144
 145 out:
 146         return segs;
 147 }
 148
 149 /* Return the total length of all the extension hdrs, following the same
 150  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 151  */
 152 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 153                             const struct net_offload **opps)
 154 {
 155         struct ipv6_opt_hdr *opth = (void *)iph;
 156         int len = 0, proto, optlen = sizeof(*iph);
 157
 158         proto = iph->nexthdr;
 159         for (;;) {
 160                 if (proto != NEXTHDR_HOP) {
 161                         *opps = rcu_dereference(inet6_offloads[proto]);
 162                         if (unlikely(!(*opps)))
 163                                 break;
 164                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 165                                 break;
 166                 }
 167                 opth = (void *)opth + optlen;
 168                 optlen = ipv6_optlen(opth);
 169                 len += optlen;
 170                 proto = opth->nexthdr;
 171         }
 172         return len;
 173 }
 174
 175 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 176                                          struct sk_buff *skb)
 177 {
 178         const struct net_offload *ops;
 179         struct sk_buff **pp = NULL;
 180         struct sk_buff *p;
 181         struct ipv6hdr *iph;
 182         unsigned int nlen;
 183         unsigned int hlen;
 184         unsigned int off;
 185         u16 flush = 1;
 186         int proto;
 187
 188         off = skb_gro_offset(skb);
 189         hlen = off + sizeof(*iph);
 190         iph = skb_gro_header_fast(skb, off);
 191         if (skb_gro_header_hard(skb, hlen)) {
 192                 iph = skb_gro_header_slow(skb, hlen, off);
 193                 if (unlikely(!iph))
 194                         goto out;
 195         }
 196
 197         skb_set_network_header(skb, off);
 198         skb_gro_pull(skb, sizeof(*iph));
 199         skb_set_transport_header(skb, skb_gro_offset(skb));
 200
 201         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 202
 203         rcu_read_lock();
 204         proto = iph->nexthdr;
 205         ops = rcu_dereference(inet6_offloads[proto]);
 206         if (!ops || !ops->callbacks.gro_receive) {
 207                 __pskb_pull(skb, skb_gro_offset(skb));
 208                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 209                 skb_gro_pull(skb, -skb_transport_offset(skb));
 210                 skb_reset_transport_header(skb);
 211                 __skb_push(skb, skb_gro_offset(skb));
 212
 213                 ops = rcu_dereference(inet6_offloads[proto]);
 214                 if (!ops || !ops->callbacks.gro_receive)
 215                         goto out_unlock;
 216
 217                 iph = ipv6_hdr(skb);
 218         }
 219
 220         NAPI_GRO_CB(skb)->proto = proto;
 221
 222         flush--;
 223         nlen = skb_network_header_len(skb);
 224
 225         for (p = *head; p; p = p->next) {
 226                 const struct ipv6hdr *iph2;
 227                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 228
 229                 if (!NAPI_GRO_CB(p)->same_flow)
 230                         continue;
 231
 232                 iph2 = (struct ipv6hdr *)(p->data + off);
 233                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 234
 235                 /* All fields must match except length and Traffic Class.
 236                  * XXX skbs on the gro_list have all been parsed and pulled
 237                  * already so we don't need to compare nlen
 238                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 239                  * memcmp() alone below is suffcient, right?
 240                  */
 241                  if ((first_word & htonl(0xF00FFFFF)) ||
 242                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 243                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 244                         NAPI_GRO_CB(p)->same_flow = 0;
 245                         continue;
 246                 }
 247                 /* flush if Traffic Class fields are different */
 248                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 249                 NAPI_GRO_CB(p)->flush |= flush;
 250
 251                 /* If the previous IP ID value was based on an atomic
 252                  * datagram we can overwrite the value and ignore it.
 253                  */
 254                 if (NAPI_GRO_CB(skb)->is_atomic)
 255                         NAPI_GRO_CB(p)->flush_id = 0;
 256         }
 257
 258         NAPI_GRO_CB(skb)->is_atomic = true;
 259         NAPI_GRO_CB(skb)->flush |= flush;
 260
 261         skb_gro_postpull_rcsum(skb, iph, nlen);
 262
 263         pp = ops->callbacks.gro_receive(head, skb);
 264
 265 out_unlock:
 266         rcu_read_unlock();
 267
 268 out:
 269         NAPI_GRO_CB(skb)->flush |= flush;
 270
 271         return pp;
 272 }
 273
 274 static struct sk_buff **sit_gro_receive(struct sk_buff **head,
 275                                         struct sk_buff *skb)
 276 {
 277         if (NAPI_GRO_CB(skb)->encap_mark) {
 278                 NAPI_GRO_CB(skb)->flush = 1;
 279                 return NULL;
 280         }
 281
 282         NAPI_GRO_CB(skb)->encap_mark = 1;
 283
 284         return ipv6_gro_receive(head, skb);
 285 }
 286
 287 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 288 {
 289         const struct net_offload *ops;
 290         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 291         int err = -ENOSYS;
 292
 293         if (skb->encapsulation)
 294                 skb_set_inner_network_header(skb, nhoff);
 295
 296         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 297
 298         rcu_read_lock();
 299
 300         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 301         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 302                 goto out_unlock;
 303
 304         err = ops->callbacks.gro_complete(skb, nhoff);
 305
 306 out_unlock:
 307         rcu_read_unlock();
 308
 309         return err;
 310 }
 311
 312 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 313 {
 314         skb->encapsulation = 1;
 315         skb_shinfo(skb)->gso_type |= SKB_GSO_SIT;
 316         return ipv6_gro_complete(skb, nhoff);
 317 }
 318
 319 static struct packet_offload ipv6_packet_offload __read_mostly = {
 320         .type = cpu_to_be16(ETH_P_IPV6),
 321         .callbacks = {
 322                 .gso_segment = ipv6_gso_segment,
 323                 .gro_receive = ipv6_gro_receive,
 324                 .gro_complete = ipv6_gro_complete,
 325         },
 326 };
 327
 328 static const struct net_offload sit_offload = {
 329         .callbacks = {
 330                 .gso_segment    = ipv6_gso_segment,
 331                 .gro_receive    = sit_gro_receive,
 332                 .gro_complete   = sit_gro_complete,
 333         },
 334 };
 335
 336 static int __init ipv6_offload_init(void)
 337 {
 338
 339         if (tcpv6_offload_init() < 0)
 340                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 341         if (ipv6_exthdrs_offload_init() < 0)
 342                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 343
 344         dev_add_offload(&ipv6_packet_offload);
 345
 346         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 347
 348         return 0;
 349 }
 350
 351 fs_initcall(ipv6_offload_init);