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