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 encap, udpfrag;
  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         encap = SKB_GSO_CB(skb)->encap_level > 0;
 114         if (encap)
 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         if (skb->encapsulation &&
 125             skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
 126                 udpfrag = proto == IPPROTO_UDP && encap;
 127         else
 128                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
 129
 130         ops = rcu_dereference(inet6_offloads[proto]);
 131         if (likely(ops && ops->callbacks.gso_segment)) {
 132                 skb_reset_transport_header(skb);
 133                 segs = ops->callbacks.gso_segment(skb, features);
 134         }
 135
 136         if (IS_ERR(segs))
 137                 goto out;
 138
 139         for (skb = segs; skb; skb = skb->next) {
 140                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 141                 ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h));
 142                 skb->network_header = (u8 *)ipv6h - skb->head;
 143
 144                 if (udpfrag) {
 145                         unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
 146                         fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
 147                         fptr->frag_off = htons(offset);
 148                         if (skb->next != NULL)
 149                                 fptr->frag_off |= htons(IP6_MF);
 150                         offset += (ntohs(ipv6h->payload_len) -
 151                                    sizeof(struct frag_hdr));
 152                 }
 153                 if (encap)
 154                         skb_reset_inner_headers(skb);
 155         }
 156
 157 out:
 158         return segs;
 159 }
 160
 161 /* Return the total length of all the extension hdrs, following the same
 162  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 163  */
 164 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 165                             const struct net_offload **opps)
 166 {
 167         struct ipv6_opt_hdr *opth = (void *)iph;
 168         int len = 0, proto, optlen = sizeof(*iph);
 169
 170         proto = iph->nexthdr;
 171         for (;;) {
 172                 if (proto != NEXTHDR_HOP) {
 173                         *opps = rcu_dereference(inet6_offloads[proto]);
 174                         if (unlikely(!(*opps)))
 175                                 break;
 176                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 177                                 break;
 178                 }
 179                 opth = (void *)opth + optlen;
 180                 optlen = ipv6_optlen(opth);
 181                 len += optlen;
 182                 proto = opth->nexthdr;
 183         }
 184         return len;
 185 }
 186
 187 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 188                                          struct sk_buff *skb)
 189 {
 190         const struct net_offload *ops;
 191         struct sk_buff **pp = NULL;
 192         struct sk_buff *p;
 193         struct ipv6hdr *iph;
 194         unsigned int nlen;
 195         unsigned int hlen;
 196         unsigned int off;
 197         u16 flush = 1;
 198         int proto;
 199
 200         off = skb_gro_offset(skb);
 201         hlen = off + sizeof(*iph);
 202         iph = skb_gro_header_fast(skb, off);
 203         if (skb_gro_header_hard(skb, hlen)) {
 204                 iph = skb_gro_header_slow(skb, hlen, off);
 205                 if (unlikely(!iph))
 206                         goto out;
 207         }
 208
 209         skb_set_network_header(skb, off);
 210         skb_gro_pull(skb, sizeof(*iph));
 211         skb_set_transport_header(skb, skb_gro_offset(skb));
 212
 213         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 214
 215         rcu_read_lock();
 216         proto = iph->nexthdr;
 217         ops = rcu_dereference(inet6_offloads[proto]);
 218         if (!ops || !ops->callbacks.gro_receive) {
 219                 __pskb_pull(skb, skb_gro_offset(skb));
 220                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 221                 skb_gro_pull(skb, -skb_transport_offset(skb));
 222                 skb_reset_transport_header(skb);
 223                 __skb_push(skb, skb_gro_offset(skb));
 224
 225                 ops = rcu_dereference(inet6_offloads[proto]);
 226                 if (!ops || !ops->callbacks.gro_receive)
 227                         goto out_unlock;
 228
 229                 iph = ipv6_hdr(skb);
 230         }
 231
 232         NAPI_GRO_CB(skb)->proto = proto;
 233
 234         flush--;
 235         nlen = skb_network_header_len(skb);
 236
 237         for (p = *head; p; p = p->next) {
 238                 const struct ipv6hdr *iph2;
 239                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 240
 241                 if (!NAPI_GRO_CB(p)->same_flow)
 242                         continue;
 243
 244                 iph2 = (struct ipv6hdr *)(p->data + off);
 245                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2 ;
 246
 247                 /* All fields must match except length and Traffic Class.
 248                  * XXX skbs on the gro_list have all been parsed and pulled
 249                  * already so we don't need to compare nlen
 250                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 251                  * memcmp() alone below is suffcient, right?
 252                  */
 253                  if ((first_word & htonl(0xF00FFFFF)) ||
 254                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 255                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 256                         NAPI_GRO_CB(p)->same_flow = 0;
 257                         continue;
 258                 }
 259                 /* flush if Traffic Class fields are different */
 260                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 261                 NAPI_GRO_CB(p)->flush |= flush;
 262         }
 263
 264         NAPI_GRO_CB(skb)->flush |= flush;
 265
 266         skb_gro_postpull_rcsum(skb, iph, nlen);
 267
 268         pp = ops->callbacks.gro_receive(head, skb);
 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);