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         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_TCPV4 |
  72                        SKB_GSO_UDP |
  73                        SKB_GSO_DODGY |
  74                        SKB_GSO_TCP_ECN |
  75                        SKB_GSO_GRE |
  76                        SKB_GSO_GRE_CSUM |
  77                        SKB_GSO_IPIP |
  78                        SKB_GSO_SIT |
  79                        SKB_GSO_UDP_TUNNEL |
  80                        SKB_GSO_UDP_TUNNEL_CSUM |
  81                        SKB_GSO_TUNNEL_REMCSUM |
  82                        SKB_GSO_TCPV6 |
  83                        0)))
  84                 goto out;
  85
  86         skb_reset_network_header(skb);
  87         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  88         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  89                 goto out;
  90
  91         encap = SKB_GSO_CB(skb)->encap_level > 0;
  92         if (encap)
  93                 features &= skb->dev->hw_enc_features;
  94         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  95
  96         ipv6h = ipv6_hdr(skb);
  97         __skb_pull(skb, sizeof(*ipv6h));
  98         segs = ERR_PTR(-EPROTONOSUPPORT);
  99
 100         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
 101
 102         if (skb->encapsulation &&
 103             skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
 104                 udpfrag = proto == IPPROTO_UDP && encap;
 105         else
 106                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
 107
 108         ops = rcu_dereference(inet6_offloads[proto]);
 109         if (likely(ops && ops->callbacks.gso_segment)) {
 110                 skb_reset_transport_header(skb);
 111                 segs = ops->callbacks.gso_segment(skb, features);
 112         }
 113
 114         if (IS_ERR(segs))
 115                 goto out;
 116
 117         for (skb = segs; skb; skb = skb->next) {
 118                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 119                 ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h));
 120                 skb->network_header = (u8 *)ipv6h - skb->head;
 121                 skb_reset_mac_len(skb);
 122
 123                 if (udpfrag) {
 124                         int err = ip6_find_1stfragopt(skb, &prevhdr);
 125                         if (err < 0) {
 126                                 kfree_skb_list(segs);
 127                                 return ERR_PTR(err);
 128                         }
 129                         fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
 130                         fptr->frag_off = htons(offset);
 131                         if (skb->next)
 132                                 fptr->frag_off |= htons(IP6_MF);
 133                         offset += (ntohs(ipv6h->payload_len) -
 134                                    sizeof(struct frag_hdr));
 135                 }
 136                 if (encap)
 137                         skb_reset_inner_headers(skb);
 138         }
 139
 140 out:
 141         return segs;
 142 }
 143
 144 /* Return the total length of all the extension hdrs, following the same
 145  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 146  */
 147 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 148                             const struct net_offload **opps)
 149 {
 150         struct ipv6_opt_hdr *opth = (void *)iph;
 151         int len = 0, proto, optlen = sizeof(*iph);
 152
 153         proto = iph->nexthdr;
 154         for (;;) {
 155                 if (proto != NEXTHDR_HOP) {
 156                         *opps = rcu_dereference(inet6_offloads[proto]);
 157                         if (unlikely(!(*opps)))
 158                                 break;
 159                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 160                                 break;
 161                 }
 162                 opth = (void *)opth + optlen;
 163                 optlen = ipv6_optlen(opth);
 164                 len += optlen;
 165                 proto = opth->nexthdr;
 166         }
 167         return len;
 168 }
 169
 170 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 171                                          struct sk_buff *skb)
 172 {
 173         const struct net_offload *ops;
 174         struct sk_buff **pp = NULL;
 175         struct sk_buff *p;
 176         struct ipv6hdr *iph;
 177         unsigned int nlen;
 178         unsigned int hlen;
 179         unsigned int off;
 180         u16 flush = 1;
 181         int proto;
 182
 183         off = skb_gro_offset(skb);
 184         hlen = off + sizeof(*iph);
 185         iph = skb_gro_header_fast(skb, off);
 186         if (skb_gro_header_hard(skb, hlen)) {
 187                 iph = skb_gro_header_slow(skb, hlen, off);
 188                 if (unlikely(!iph))
 189                         goto out;
 190         }
 191
 192         skb_set_network_header(skb, off);
 193         skb_gro_pull(skb, sizeof(*iph));
 194         skb_set_transport_header(skb, skb_gro_offset(skb));
 195
 196         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 197
 198         rcu_read_lock();
 199         proto = iph->nexthdr;
 200         ops = rcu_dereference(inet6_offloads[proto]);
 201         if (!ops || !ops->callbacks.gro_receive) {
 202                 __pskb_pull(skb, skb_gro_offset(skb));
 203                 skb_gro_frag0_invalidate(skb);
 204                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 205                 skb_gro_pull(skb, -skb_transport_offset(skb));
 206                 skb_reset_transport_header(skb);
 207                 __skb_push(skb, skb_gro_offset(skb));
 208
 209                 ops = rcu_dereference(inet6_offloads[proto]);
 210                 if (!ops || !ops->callbacks.gro_receive)
 211                         goto out_unlock;
 212
 213                 iph = ipv6_hdr(skb);
 214         }
 215
 216         NAPI_GRO_CB(skb)->proto = proto;
 217
 218         flush--;
 219         nlen = skb_network_header_len(skb);
 220
 221         for (p = *head; p; p = p->next) {
 222                 const struct ipv6hdr *iph2;
 223                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 224
 225                 if (!NAPI_GRO_CB(p)->same_flow)
 226                         continue;
 227
 228                 iph2 = (struct ipv6hdr *)(p->data + off);
 229                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 230
 231                 /* All fields must match except length and Traffic Class.
 232                  * XXX skbs on the gro_list have all been parsed and pulled
 233                  * already so we don't need to compare nlen
 234                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 235                  * memcmp() alone below is suffcient, right?
 236                  */
 237                  if ((first_word & htonl(0xF00FFFFF)) ||
 238                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 239                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 240                         NAPI_GRO_CB(p)->same_flow = 0;
 241                         continue;
 242                 }
 243                 /* flush if Traffic Class fields are different */
 244                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 245                 NAPI_GRO_CB(p)->flush |= flush;
 246
 247                 /* Clear flush_id, there's really no concept of ID in IPv6. */
 248                 NAPI_GRO_CB(p)->flush_id = 0;
 249         }
 250
 251         NAPI_GRO_CB(skb)->flush |= flush;
 252
 253         skb_gro_postpull_rcsum(skb, iph, nlen);
 254
 255         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 256
 257 out_unlock:
 258         rcu_read_unlock();
 259
 260 out:
 261         NAPI_GRO_CB(skb)->flush |= flush;
 262
 263         return pp;
 264 }
 265
 266 static struct sk_buff **sit_gro_receive(struct sk_buff **head,
 267                                         struct sk_buff *skb)
 268 {
 269         if (NAPI_GRO_CB(skb)->encap_mark) {
 270                 NAPI_GRO_CB(skb)->flush = 1;
 271                 return NULL;
 272         }
 273
 274         NAPI_GRO_CB(skb)->encap_mark = 1;
 275
 276         return ipv6_gro_receive(head, skb);
 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         if (skb->encapsulation)
 286                 skb_set_inner_network_header(skb, nhoff);
 287
 288         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 289
 290         rcu_read_lock();
 291
 292         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 293         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 294                 goto out_unlock;
 295
 296         err = ops->callbacks.gro_complete(skb, nhoff);
 297
 298 out_unlock:
 299         rcu_read_unlock();
 300
 301         return err;
 302 }
 303
 304 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 305 {
 306         skb->encapsulation = 1;
 307         skb_shinfo(skb)->gso_type |= SKB_GSO_SIT;
 308         return ipv6_gro_complete(skb, nhoff);
 309 }
 310
 311 static struct packet_offload ipv6_packet_offload __read_mostly = {
 312         .type = cpu_to_be16(ETH_P_IPV6),
 313         .callbacks = {
 314                 .gso_segment = ipv6_gso_segment,
 315                 .gro_receive = ipv6_gro_receive,
 316                 .gro_complete = ipv6_gro_complete,
 317         },
 318 };
 319
 320 static const struct net_offload sit_offload = {
 321         .callbacks = {
 322                 .gso_segment    = ipv6_gso_segment,
 323                 .gro_receive    = sit_gro_receive,
 324                 .gro_complete   = sit_gro_complete,
 325         },
 326 };
 327
 328 static int __init ipv6_offload_init(void)
 329 {
 330
 331         if (tcpv6_offload_init() < 0)
 332                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 333         if (udp_offload_init() < 0)
 334                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
 335         if (ipv6_exthdrs_offload_init() < 0)
 336                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 337
 338         dev_add_offload(&ipv6_packet_offload);
 339
 340         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 341
 342         return 0;
 343 }
 344
 345 fs_initcall(ipv6_offload_init);