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                           (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
  93         else
  94                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
  95                           (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
  96
  97         ops = rcu_dereference(inet6_offloads[proto]);
  98         if (likely(ops && ops->callbacks.gso_segment)) {
  99                 skb_reset_transport_header(skb);
 100                 segs = ops->callbacks.gso_segment(skb, features);
 101                 if (!segs)
 102                         skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
 103         }
 104
 105         if (IS_ERR_OR_NULL(segs))
 106                 goto out;
 107
 108         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
 109
 110         for (skb = segs; skb; skb = skb->next) {
 111                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 112                 if (gso_partial && skb_is_gso(skb))
 113                         payload_len = skb_shinfo(skb)->gso_size +
 114                                       SKB_GSO_CB(skb)->data_offset +
 115                                       skb->head - (unsigned char *)(ipv6h + 1);
 116                 else
 117                         payload_len = skb->len - nhoff - sizeof(*ipv6h);
 118                 ipv6h->payload_len = htons(payload_len);
 119                 skb->network_header = (u8 *)ipv6h - skb->head;
 120                 skb_reset_mac_len(skb);
 121
 122                 if (udpfrag) {
 123                         int err = ip6_find_1stfragopt(skb, &prevhdr);
 124                         if (err < 0) {
 125                                 kfree_skb_list(segs);
 126                                 return ERR_PTR(err);
 127                         }
 128                         fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
 129                         fptr->frag_off = htons(offset);
 130                         if (skb->next)
 131                                 fptr->frag_off |= htons(IP6_MF);
 132                         offset += (ntohs(ipv6h->payload_len) -
 133                                    sizeof(struct frag_hdr));
 134                 }
 135                 if (encap)
 136                         skb_reset_inner_headers(skb);
 137         }
 138
 139 out:
 140         return segs;
 141 }
 142
 143 /* Return the total length of all the extension hdrs, following the same
 144  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 145  */
 146 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 147                             const struct net_offload **opps)
 148 {
 149         struct ipv6_opt_hdr *opth = (void *)iph;
 150         int len = 0, proto, optlen = sizeof(*iph);
 151
 152         proto = iph->nexthdr;
 153         for (;;) {
 154                 if (proto != NEXTHDR_HOP) {
 155                         *opps = rcu_dereference(inet6_offloads[proto]);
 156                         if (unlikely(!(*opps)))
 157                                 break;
 158                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 159                                 break;
 160                 }
 161                 opth = (void *)opth + optlen;
 162                 optlen = ipv6_optlen(opth);
 163                 len += optlen;
 164                 proto = opth->nexthdr;
 165         }
 166         return len;
 167 }
 168
 169 static struct sk_buff *ipv6_gro_receive(struct list_head *head,
 170                                         struct sk_buff *skb)
 171 {
 172         const struct net_offload *ops;
 173         struct sk_buff *pp = NULL;
 174         struct sk_buff *p;
 175         struct ipv6hdr *iph;
 176         unsigned int nlen;
 177         unsigned int hlen;
 178         unsigned int off;
 179         u16 flush = 1;
 180         int proto;
 181
 182         off = skb_gro_offset(skb);
 183         hlen = off + sizeof(*iph);
 184         iph = skb_gro_header_fast(skb, off);
 185         if (skb_gro_header_hard(skb, hlen)) {
 186                 iph = skb_gro_header_slow(skb, hlen, off);
 187                 if (unlikely(!iph))
 188                         goto out;
 189         }
 190
 191         skb_set_network_header(skb, off);
 192         skb_gro_pull(skb, sizeof(*iph));
 193         skb_set_transport_header(skb, skb_gro_offset(skb));
 194
 195         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 196
 197         rcu_read_lock();
 198         proto = iph->nexthdr;
 199         ops = rcu_dereference(inet6_offloads[proto]);
 200         if (!ops || !ops->callbacks.gro_receive) {
 201                 __pskb_pull(skb, skb_gro_offset(skb));
 202                 skb_gro_frag0_invalidate(skb);
 203                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 204                 skb_gro_pull(skb, -skb_transport_offset(skb));
 205                 skb_reset_transport_header(skb);
 206                 __skb_push(skb, skb_gro_offset(skb));
 207
 208                 ops = rcu_dereference(inet6_offloads[proto]);
 209                 if (!ops || !ops->callbacks.gro_receive)
 210                         goto out_unlock;
 211
 212                 iph = ipv6_hdr(skb);
 213         }
 214
 215         NAPI_GRO_CB(skb)->proto = proto;
 216
 217         flush--;
 218         nlen = skb_network_header_len(skb);
 219
 220         list_for_each_entry(p, head, list) {
 221                 const struct ipv6hdr *iph2;
 222                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 223
 224                 if (!NAPI_GRO_CB(p)->same_flow)
 225                         continue;
 226
 227                 iph2 = (struct ipv6hdr *)(p->data + off);
 228                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 229
 230                 /* All fields must match except length and Traffic Class.
 231                  * XXX skbs on the gro_list have all been parsed and pulled
 232                  * already so we don't need to compare nlen
 233                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 234                  * memcmp() alone below is suffcient, right?
 235                  */
 236                  if ((first_word & htonl(0xF00FFFFF)) ||
 237                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 238                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 239                         NAPI_GRO_CB(p)->same_flow = 0;
 240                         continue;
 241                 }
 242                 /* flush if Traffic Class fields are different */
 243                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 244                 NAPI_GRO_CB(p)->flush |= flush;
 245
 246                 /* If the previous IP ID value was based on an atomic
 247                  * datagram we can overwrite the value and ignore it.
 248                  */
 249                 if (NAPI_GRO_CB(skb)->is_atomic)
 250                         NAPI_GRO_CB(p)->flush_id = 0;
 251         }
 252
 253         NAPI_GRO_CB(skb)->is_atomic = true;
 254         NAPI_GRO_CB(skb)->flush |= flush;
 255
 256         skb_gro_postpull_rcsum(skb, iph, nlen);
 257
 258         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 259
 260 out_unlock:
 261         rcu_read_unlock();
 262
 263 out:
 264         skb_gro_flush_final(skb, pp, flush);
 265
 266         return pp;
 267 }
 268
 269 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
 270                                               struct sk_buff *skb)
 271 {
 272         /* Common GRO receive for SIT and IP6IP6 */
 273
 274         if (NAPI_GRO_CB(skb)->encap_mark) {
 275                 NAPI_GRO_CB(skb)->flush = 1;
 276                 return NULL;
 277         }
 278
 279         NAPI_GRO_CB(skb)->encap_mark = 1;
 280
 281         return ipv6_gro_receive(head, skb);
 282 }
 283
 284 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
 285                                           struct sk_buff *skb)
 286 {
 287         /* Common GRO receive for SIT and IP6IP6 */
 288
 289         if (NAPI_GRO_CB(skb)->encap_mark) {
 290                 NAPI_GRO_CB(skb)->flush = 1;
 291                 return NULL;
 292         }
 293
 294         NAPI_GRO_CB(skb)->encap_mark = 1;
 295
 296         return inet_gro_receive(head, skb);
 297 }
 298
 299 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 300 {
 301         const struct net_offload *ops;
 302         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 303         int err = -ENOSYS;
 304
 305         if (skb->encapsulation) {
 306                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
 307                 skb_set_inner_network_header(skb, nhoff);
 308         }
 309
 310         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 311
 312         rcu_read_lock();
 313
 314         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 315         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 316                 goto out_unlock;
 317
 318         err = ops->callbacks.gro_complete(skb, nhoff);
 319
 320 out_unlock:
 321         rcu_read_unlock();
 322
 323         return err;
 324 }
 325
 326 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 327 {
 328         skb->encapsulation = 1;
 329         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
 330         return ipv6_gro_complete(skb, nhoff);
 331 }
 332
 333 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
 334 {
 335         skb->encapsulation = 1;
 336         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 337         return ipv6_gro_complete(skb, nhoff);
 338 }
 339
 340 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 341 {
 342         skb->encapsulation = 1;
 343         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 344         return inet_gro_complete(skb, nhoff);
 345 }
 346
 347 static struct packet_offload ipv6_packet_offload __read_mostly = {
 348         .type = cpu_to_be16(ETH_P_IPV6),
 349         .callbacks = {
 350                 .gso_segment = ipv6_gso_segment,
 351                 .gro_receive = ipv6_gro_receive,
 352                 .gro_complete = ipv6_gro_complete,
 353         },
 354 };
 355
 356 static const struct net_offload sit_offload = {
 357         .callbacks = {
 358                 .gso_segment    = ipv6_gso_segment,
 359                 .gro_receive    = sit_ip6ip6_gro_receive,
 360                 .gro_complete   = sit_gro_complete,
 361         },
 362 };
 363
 364 static const struct net_offload ip4ip6_offload = {
 365         .callbacks = {
 366                 .gso_segment    = inet_gso_segment,
 367                 .gro_receive    = ip4ip6_gro_receive,
 368                 .gro_complete   = ip4ip6_gro_complete,
 369         },
 370 };
 371
 372 static const struct net_offload ip6ip6_offload = {
 373         .callbacks = {
 374                 .gso_segment    = ipv6_gso_segment,
 375                 .gro_receive    = sit_ip6ip6_gro_receive,
 376                 .gro_complete   = ip6ip6_gro_complete,
 377         },
 378 };
 379 static int __init ipv6_offload_init(void)
 380 {
 381
 382         if (tcpv6_offload_init() < 0)
 383                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 384         if (ipv6_exthdrs_offload_init() < 0)
 385                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 386
 387         dev_add_offload(&ipv6_packet_offload);
 388
 389         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 390         inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
 391         inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
 392
 393         return 0;
 394 }
 395
 396 fs_initcall(ipv6_offload_init);