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 #include <net/inet_common.h>
  20
  21 #include "ip6_offload.h"
  22
  23 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
  24 {
  25         const struct net_offload *ops = NULL;
  26
  27         for (;;) {
  28                 struct ipv6_opt_hdr *opth;
  29                 int len;
  30
  31                 if (proto != NEXTHDR_HOP) {
  32                         ops = rcu_dereference(inet6_offloads[proto]);
  33
  34                         if (unlikely(!ops))
  35                                 break;
  36
  37                         if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
  38                                 break;
  39                 }
  40
  41                 if (unlikely(!pskb_may_pull(skb, 8)))
  42                         break;
  43
  44                 opth = (void *)skb->data;
  45                 len = ipv6_optlen(opth);
  46
  47                 if (unlikely(!pskb_may_pull(skb, len)))
  48                         break;
  49
  50                 opth = (void *)skb->data;
  51                 proto = opth->nexthdr;
  52                 __skb_pull(skb, len);
  53         }
  54
  55         return proto;
  56 }
  57
  58 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
  59         netdev_features_t features)
  60 {
  61         struct sk_buff *segs = ERR_PTR(-EINVAL);
  62         struct ipv6hdr *ipv6h;
  63         const struct net_offload *ops;
  64         int proto;
  65         struct frag_hdr *fptr;
  66         unsigned int payload_len;
  67         u8 *prevhdr;
  68         int offset = 0;
  69         bool encap, udpfrag;
  70         int nhoff;
  71         bool gso_partial;
  72
  73         skb_reset_network_header(skb);
  74         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  75         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  76                 goto out;
  77
  78         encap = SKB_GSO_CB(skb)->encap_level > 0;
  79         if (encap)
  80                 features &= skb->dev->hw_enc_features;
  81         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  82
  83         ipv6h = ipv6_hdr(skb);
  84         __skb_pull(skb, sizeof(*ipv6h));
  85         segs = ERR_PTR(-EPROTONOSUPPORT);
  86
  87         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
  88
  89         if (skb->encapsulation &&
  90             skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
  91                 udpfrag = proto == IPPROTO_UDP && encap;
  92         else
  93                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
  94
  95         ops = rcu_dereference(inet6_offloads[proto]);
  96         if (likely(ops && ops->callbacks.gso_segment)) {
  97                 skb_reset_transport_header(skb);
  98                 segs = ops->callbacks.gso_segment(skb, features);
  99         }
 100
 101         if (IS_ERR_OR_NULL(segs))
 102                 goto out;
 103
 104         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
 105
 106         for (skb = segs; skb; skb = skb->next) {
 107                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 108                 if (gso_partial)
 109                         payload_len = skb_shinfo(skb)->gso_size +
 110                                       SKB_GSO_CB(skb)->data_offset +
 111                                       skb->head - (unsigned char *)(ipv6h + 1);
 112                 else
 113                         payload_len = skb->len - nhoff - sizeof(*ipv6h);
 114                 ipv6h->payload_len = htons(payload_len);
 115                 skb->network_header = (u8 *)ipv6h - skb->head;
 116
 117                 if (udpfrag) {
 118                         int err = ip6_find_1stfragopt(skb, &prevhdr);
 119                         if (err < 0) {
 120                                 kfree_skb_list(segs);
 121                                 return ERR_PTR(err);
 122                         }
 123                         fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
 124                         fptr->frag_off = htons(offset);
 125                         if (skb->next)
 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                 skb_gro_frag0_invalidate(skb);
 198                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 199                 skb_gro_pull(skb, -skb_transport_offset(skb));
 200                 skb_reset_transport_header(skb);
 201                 __skb_push(skb, skb_gro_offset(skb));
 202
 203                 ops = rcu_dereference(inet6_offloads[proto]);
 204                 if (!ops || !ops->callbacks.gro_receive)
 205                         goto out_unlock;
 206
 207                 iph = ipv6_hdr(skb);
 208         }
 209
 210         NAPI_GRO_CB(skb)->proto = proto;
 211
 212         flush--;
 213         nlen = skb_network_header_len(skb);
 214
 215         for (p = *head; p; p = p->next) {
 216                 const struct ipv6hdr *iph2;
 217                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 218
 219                 if (!NAPI_GRO_CB(p)->same_flow)
 220                         continue;
 221
 222                 iph2 = (struct ipv6hdr *)(p->data + off);
 223                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 224
 225                 /* All fields must match except length and Traffic Class.
 226                  * XXX skbs on the gro_list have all been parsed and pulled
 227                  * already so we don't need to compare nlen
 228                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 229                  * memcmp() alone below is suffcient, right?
 230                  */
 231                  if ((first_word & htonl(0xF00FFFFF)) ||
 232                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 233                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 234                         NAPI_GRO_CB(p)->same_flow = 0;
 235                         continue;
 236                 }
 237                 /* flush if Traffic Class fields are different */
 238                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 239                 NAPI_GRO_CB(p)->flush |= flush;
 240
 241                 /* If the previous IP ID value was based on an atomic
 242                  * datagram we can overwrite the value and ignore it.
 243                  */
 244                 if (NAPI_GRO_CB(skb)->is_atomic)
 245                         NAPI_GRO_CB(p)->flush_id = 0;
 246         }
 247
 248         NAPI_GRO_CB(skb)->is_atomic = true;
 249         NAPI_GRO_CB(skb)->flush |= flush;
 250
 251         skb_gro_postpull_rcsum(skb, iph, nlen);
 252
 253         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 254
 255 out_unlock:
 256         rcu_read_unlock();
 257
 258 out:
 259         skb_gro_flush_final(skb, pp, flush);
 260
 261         return pp;
 262 }
 263
 264 static struct sk_buff **sit_ip6ip6_gro_receive(struct sk_buff **head,
 265                                                struct sk_buff *skb)
 266 {
 267         /* Common GRO receive for SIT and IP6IP6 */
 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 struct sk_buff **ip4ip6_gro_receive(struct sk_buff **head,
 280                                            struct sk_buff *skb)
 281 {
 282         /* Common GRO receive for SIT and IP6IP6 */
 283
 284         if (NAPI_GRO_CB(skb)->encap_mark) {
 285                 NAPI_GRO_CB(skb)->flush = 1;
 286                 return NULL;
 287         }
 288
 289         NAPI_GRO_CB(skb)->encap_mark = 1;
 290
 291         return inet_gro_receive(head, skb);
 292 }
 293
 294 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 295 {
 296         const struct net_offload *ops;
 297         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 298         int err = -ENOSYS;
 299
 300         if (skb->encapsulation) {
 301                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
 302                 skb_set_inner_network_header(skb, nhoff);
 303         }
 304
 305         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 306
 307         rcu_read_lock();
 308
 309         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 310         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 311                 goto out_unlock;
 312
 313         err = ops->callbacks.gro_complete(skb, nhoff);
 314
 315 out_unlock:
 316         rcu_read_unlock();
 317
 318         return err;
 319 }
 320
 321 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 322 {
 323         skb->encapsulation = 1;
 324         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
 325         return ipv6_gro_complete(skb, nhoff);
 326 }
 327
 328 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
 329 {
 330         skb->encapsulation = 1;
 331         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 332         return ipv6_gro_complete(skb, nhoff);
 333 }
 334
 335 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 336 {
 337         skb->encapsulation = 1;
 338         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 339         return inet_gro_complete(skb, nhoff);
 340 }
 341
 342 static struct packet_offload ipv6_packet_offload __read_mostly = {
 343         .type = cpu_to_be16(ETH_P_IPV6),
 344         .callbacks = {
 345                 .gso_segment = ipv6_gso_segment,
 346                 .gro_receive = ipv6_gro_receive,
 347                 .gro_complete = ipv6_gro_complete,
 348         },
 349 };
 350
 351 static const struct net_offload sit_offload = {
 352         .callbacks = {
 353                 .gso_segment    = ipv6_gso_segment,
 354                 .gro_receive    = sit_ip6ip6_gro_receive,
 355                 .gro_complete   = sit_gro_complete,
 356         },
 357 };
 358
 359 static const struct net_offload ip4ip6_offload = {
 360         .callbacks = {
 361                 .gso_segment    = inet_gso_segment,
 362                 .gro_receive    = ip4ip6_gro_receive,
 363                 .gro_complete   = ip4ip6_gro_complete,
 364         },
 365 };
 366
 367 static const struct net_offload ip6ip6_offload = {
 368         .callbacks = {
 369                 .gso_segment    = ipv6_gso_segment,
 370                 .gro_receive    = sit_ip6ip6_gro_receive,
 371                 .gro_complete   = ip6ip6_gro_complete,
 372         },
 373 };
 374 static int __init ipv6_offload_init(void)
 375 {
 376
 377         if (tcpv6_offload_init() < 0)
 378                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 379         if (ipv6_exthdrs_offload_init() < 0)
 380                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 381
 382         dev_add_offload(&ipv6_packet_offload);
 383
 384         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 385         inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
 386         inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
 387
 388         return 0;
 389 }
 390
 391 fs_initcall(ipv6_offload_init);