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