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