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 unfrag_ip6hlen;
  67         unsigned int payload_len;
  68         u8 *prevhdr;
  69         int offset = 0;
  70         bool encap, udpfrag;
  71         int nhoff;
  72         bool gso_partial;
  73
  74         skb_reset_network_header(skb);
  75         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  76         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  77                 goto out;
  78
  79         encap = SKB_GSO_CB(skb)->encap_level > 0;
  80         if (encap)
  81                 features &= skb->dev->hw_enc_features;
  82         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  83
  84         ipv6h = ipv6_hdr(skb);
  85         __skb_pull(skb, sizeof(*ipv6h));
  86         segs = ERR_PTR(-EPROTONOSUPPORT);
  87
  88         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
  89
  90         if (skb->encapsulation &&
  91             skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
  92                 udpfrag = proto == IPPROTO_UDP && encap;
  93         else
  94                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
  95
  96         ops = rcu_dereference(inet6_offloads[proto]);
  97         if (likely(ops && ops->callbacks.gso_segment)) {
  98                 skb_reset_transport_header(skb);
  99                 segs = ops->callbacks.gso_segment(skb, features);
 100         }
 101
 102         if (IS_ERR_OR_NULL(segs))
 103                 goto out;
 104
 105         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
 106
 107         for (skb = segs; skb; skb = skb->next) {
 108                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 109                 if (gso_partial)
 110                         payload_len = skb_shinfo(skb)->gso_size +
 111                                       SKB_GSO_CB(skb)->data_offset +
 112                                       skb->head - (unsigned char *)(ipv6h + 1);
 113                 else
 114                         payload_len = skb->len - nhoff - sizeof(*ipv6h);
 115                 ipv6h->payload_len = htons(payload_len);
 116                 skb->network_header = (u8 *)ipv6h - skb->head;
 117
 118                 if (udpfrag) {
 119                         unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
 120                         if (unfrag_ip6hlen < 0)
 121                                 return ERR_PTR(unfrag_ip6hlen);
 122                         fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
 123                         fptr->frag_off = htons(offset);
 124                         if (skb->next)
 125                                 fptr->frag_off |= htons(IP6_MF);
 126                         offset += (ntohs(ipv6h->payload_len) -
 127                                    sizeof(struct frag_hdr));
 128                 }
 129                 if (encap)
 130                         skb_reset_inner_headers(skb);
 131         }
 132
 133 out:
 134         return segs;
 135 }
 136
 137 /* Return the total length of all the extension hdrs, following the same
 138  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 139  */
 140 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 141                             const struct net_offload **opps)
 142 {
 143         struct ipv6_opt_hdr *opth = (void *)iph;
 144         int len = 0, proto, optlen = sizeof(*iph);
 145
 146         proto = iph->nexthdr;
 147         for (;;) {
 148                 if (proto != NEXTHDR_HOP) {
 149                         *opps = rcu_dereference(inet6_offloads[proto]);
 150                         if (unlikely(!(*opps)))
 151                                 break;
 152                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 153                                 break;
 154                 }
 155                 opth = (void *)opth + optlen;
 156                 optlen = ipv6_optlen(opth);
 157                 len += optlen;
 158                 proto = opth->nexthdr;
 159         }
 160         return len;
 161 }
 162
 163 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 164                                          struct sk_buff *skb)
 165 {
 166         const struct net_offload *ops;
 167         struct sk_buff **pp = NULL;
 168         struct sk_buff *p;
 169         struct ipv6hdr *iph;
 170         unsigned int nlen;
 171         unsigned int hlen;
 172         unsigned int off;
 173         u16 flush = 1;
 174         int proto;
 175
 176         off = skb_gro_offset(skb);
 177         hlen = off + sizeof(*iph);
 178         iph = skb_gro_header_fast(skb, off);
 179         if (skb_gro_header_hard(skb, hlen)) {
 180                 iph = skb_gro_header_slow(skb, hlen, off);
 181                 if (unlikely(!iph))
 182                         goto out;
 183         }
 184
 185         skb_set_network_header(skb, off);
 186         skb_gro_pull(skb, sizeof(*iph));
 187         skb_set_transport_header(skb, skb_gro_offset(skb));
 188
 189         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 190
 191         rcu_read_lock();
 192         proto = iph->nexthdr;
 193         ops = rcu_dereference(inet6_offloads[proto]);
 194         if (!ops || !ops->callbacks.gro_receive) {
 195                 __pskb_pull(skb, skb_gro_offset(skb));
 196                 skb_gro_frag0_invalidate(skb);
 197                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 198                 skb_gro_pull(skb, -skb_transport_offset(skb));
 199                 skb_reset_transport_header(skb);
 200                 __skb_push(skb, skb_gro_offset(skb));
 201
 202                 ops = rcu_dereference(inet6_offloads[proto]);
 203                 if (!ops || !ops->callbacks.gro_receive)
 204                         goto out_unlock;
 205
 206                 iph = ipv6_hdr(skb);
 207         }
 208
 209         NAPI_GRO_CB(skb)->proto = proto;
 210
 211         flush--;
 212         nlen = skb_network_header_len(skb);
 213
 214         for (p = *head; p; p = p->next) {
 215                 const struct ipv6hdr *iph2;
 216                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 217
 218                 if (!NAPI_GRO_CB(p)->same_flow)
 219                         continue;
 220
 221                 iph2 = (struct ipv6hdr *)(p->data + off);
 222                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 223
 224                 /* All fields must match except length and Traffic Class.
 225                  * XXX skbs on the gro_list have all been parsed and pulled
 226                  * already so we don't need to compare nlen
 227                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 228                  * memcmp() alone below is suffcient, right?
 229                  */
 230                  if ((first_word & htonl(0xF00FFFFF)) ||
 231                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 232                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 233                         NAPI_GRO_CB(p)->same_flow = 0;
 234                         continue;
 235                 }
 236                 /* flush if Traffic Class fields are different */
 237                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 238                 NAPI_GRO_CB(p)->flush |= flush;
 239
 240                 /* If the previous IP ID value was based on an atomic
 241                  * datagram we can overwrite the value and ignore it.
 242                  */
 243                 if (NAPI_GRO_CB(skb)->is_atomic)
 244                         NAPI_GRO_CB(p)->flush_id = 0;
 245         }
 246
 247         NAPI_GRO_CB(skb)->is_atomic = true;
 248         NAPI_GRO_CB(skb)->flush |= flush;
 249
 250         skb_gro_postpull_rcsum(skb, iph, nlen);
 251
 252         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 253
 254 out_unlock:
 255         rcu_read_unlock();
 256
 257 out:
 258         skb_gro_flush_final(skb, pp, flush);
 259
 260         return pp;
 261 }
 262
 263 static struct sk_buff **sit_ip6ip6_gro_receive(struct sk_buff **head,
 264                                                struct sk_buff *skb)
 265 {
 266         /* Common GRO receive for SIT and IP6IP6 */
 267
 268         if (NAPI_GRO_CB(skb)->encap_mark) {
 269                 NAPI_GRO_CB(skb)->flush = 1;
 270                 return NULL;
 271         }
 272
 273         NAPI_GRO_CB(skb)->encap_mark = 1;
 274
 275         return ipv6_gro_receive(head, skb);
 276 }
 277
 278 static struct sk_buff **ip4ip6_gro_receive(struct sk_buff **head,
 279                                            struct sk_buff *skb)
 280 {
 281         /* Common GRO receive for SIT and IP6IP6 */
 282
 283         if (NAPI_GRO_CB(skb)->encap_mark) {
 284                 NAPI_GRO_CB(skb)->flush = 1;
 285                 return NULL;
 286         }
 287
 288         NAPI_GRO_CB(skb)->encap_mark = 1;
 289
 290         return inet_gro_receive(head, skb);
 291 }
 292
 293 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 294 {
 295         const struct net_offload *ops;
 296         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 297         int err = -ENOSYS;
 298
 299         if (skb->encapsulation) {
 300                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
 301                 skb_set_inner_network_header(skb, nhoff);
 302         }
 303
 304         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 305
 306         rcu_read_lock();
 307
 308         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 309         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 310                 goto out_unlock;
 311
 312         err = ops->callbacks.gro_complete(skb, nhoff);
 313
 314 out_unlock:
 315         rcu_read_unlock();
 316
 317         return err;
 318 }
 319
 320 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 321 {
 322         skb->encapsulation = 1;
 323         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
 324         return ipv6_gro_complete(skb, nhoff);
 325 }
 326
 327 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
 328 {
 329         skb->encapsulation = 1;
 330         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 331         return ipv6_gro_complete(skb, nhoff);
 332 }
 333
 334 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 335 {
 336         skb->encapsulation = 1;
 337         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 338         return inet_gro_complete(skb, nhoff);
 339 }
 340
 341 static struct packet_offload ipv6_packet_offload __read_mostly = {
 342         .type = cpu_to_be16(ETH_P_IPV6),
 343         .callbacks = {
 344                 .gso_segment = ipv6_gso_segment,
 345                 .gro_receive = ipv6_gro_receive,
 346                 .gro_complete = ipv6_gro_complete,
 347         },
 348 };
 349
 350 static const struct net_offload sit_offload = {
 351         .callbacks = {
 352                 .gso_segment    = ipv6_gso_segment,
 353                 .gro_receive    = sit_ip6ip6_gro_receive,
 354                 .gro_complete   = sit_gro_complete,
 355         },
 356 };
 357
 358 static const struct net_offload ip4ip6_offload = {
 359         .callbacks = {
 360                 .gso_segment    = inet_gso_segment,
 361                 .gro_receive    = ip4ip6_gro_receive,
 362                 .gro_complete   = ip4ip6_gro_complete,
 363         },
 364 };
 365
 366 static const struct net_offload ip6ip6_offload = {
 367         .callbacks = {
 368                 .gso_segment    = ipv6_gso_segment,
 369                 .gro_receive    = sit_ip6ip6_gro_receive,
 370                 .gro_complete   = ip6ip6_gro_complete,
 371         },
 372 };
 373 static int __init ipv6_offload_init(void)
 374 {
 375
 376         if (tcpv6_offload_init() < 0)
 377                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 378         if (ipv6_exthdrs_offload_init() < 0)
 379                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 380
 381         dev_add_offload(&ipv6_packet_offload);
 382
 383         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 384         inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
 385         inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
 386
 387         return 0;
 388 }
 389
 390 fs_initcall(ipv6_offload_init);