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                 skb_reset_mac_len(skb);
 119
 120                 if (udpfrag) {
 121                         int err = ip6_find_1stfragopt(skb, &prevhdr);
 122                         if (err < 0) {
 123                                 kfree_skb_list(segs);
 124                                 return ERR_PTR(err);
 125                         }
 126                         fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
 127                         fptr->frag_off = htons(offset);
 128                         if (skb->next)
 129                                 fptr->frag_off |= htons(IP6_MF);
 130                         offset += (ntohs(ipv6h->payload_len) -
 131                                    sizeof(struct frag_hdr));
 132                 }
 133                 if (encap)
 134                         skb_reset_inner_headers(skb);
 135         }
 136
 137 out:
 138         return segs;
 139 }
 140
 141 /* Return the total length of all the extension hdrs, following the same
 142  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 143  */
 144 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 145                             const struct net_offload **opps)
 146 {
 147         struct ipv6_opt_hdr *opth = (void *)iph;
 148         int len = 0, proto, optlen = sizeof(*iph);
 149
 150         proto = iph->nexthdr;
 151         for (;;) {
 152                 if (proto != NEXTHDR_HOP) {
 153                         *opps = rcu_dereference(inet6_offloads[proto]);
 154                         if (unlikely(!(*opps)))
 155                                 break;
 156                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 157                                 break;
 158                 }
 159                 opth = (void *)opth + optlen;
 160                 optlen = ipv6_optlen(opth);
 161                 len += optlen;
 162                 proto = opth->nexthdr;
 163         }
 164         return len;
 165 }
 166
 167 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
 168                                                          struct sk_buff *skb)
 169 {
 170         const struct net_offload *ops;
 171         struct sk_buff *pp = NULL;
 172         struct sk_buff *p;
 173         struct ipv6hdr *iph;
 174         unsigned int nlen;
 175         unsigned int hlen;
 176         unsigned int off;
 177         u16 flush = 1;
 178         int proto;
 179
 180         off = skb_gro_offset(skb);
 181         hlen = off + sizeof(*iph);
 182         iph = skb_gro_header_fast(skb, off);
 183         if (skb_gro_header_hard(skb, hlen)) {
 184                 iph = skb_gro_header_slow(skb, hlen, off);
 185                 if (unlikely(!iph))
 186                         goto out;
 187         }
 188
 189         skb_set_network_header(skb, off);
 190         skb_gro_pull(skb, sizeof(*iph));
 191         skb_set_transport_header(skb, skb_gro_offset(skb));
 192
 193         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 194
 195         rcu_read_lock();
 196         proto = iph->nexthdr;
 197         ops = rcu_dereference(inet6_offloads[proto]);
 198         if (!ops || !ops->callbacks.gro_receive) {
 199                 __pskb_pull(skb, skb_gro_offset(skb));
 200                 skb_gro_frag0_invalidate(skb);
 201                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 202                 skb_gro_pull(skb, -skb_transport_offset(skb));
 203                 skb_reset_transport_header(skb);
 204                 __skb_push(skb, skb_gro_offset(skb));
 205
 206                 ops = rcu_dereference(inet6_offloads[proto]);
 207                 if (!ops || !ops->callbacks.gro_receive)
 208                         goto out_unlock;
 209
 210                 iph = ipv6_hdr(skb);
 211         }
 212
 213         NAPI_GRO_CB(skb)->proto = proto;
 214
 215         flush--;
 216         nlen = skb_network_header_len(skb);
 217
 218         list_for_each_entry(p, head, list) {
 219                 const struct ipv6hdr *iph2;
 220                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 221
 222                 if (!NAPI_GRO_CB(p)->same_flow)
 223                         continue;
 224
 225                 iph2 = (struct ipv6hdr *)(p->data + off);
 226                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 227
 228                 /* All fields must match except length and Traffic Class.
 229                  * XXX skbs on the gro_list have all been parsed and pulled
 230                  * already so we don't need to compare nlen
 231                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 232                  * memcmp() alone below is sufficient, right?
 233                  */
 234                  if ((first_word & htonl(0xF00FFFFF)) ||
 235                     !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
 236                     !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
 237                     *(u16 *)&iph->nexthdr != *(u16 *)&iph2->nexthdr) {
 238 not_same_flow:
 239                         NAPI_GRO_CB(p)->same_flow = 0;
 240                         continue;
 241                 }
 242                 if (unlikely(nlen > sizeof(struct ipv6hdr))) {
 243                         if (memcmp(iph + 1, iph2 + 1,
 244                                    nlen - sizeof(struct ipv6hdr)))
 245                                 goto not_same_flow;
 246                 }
 247                 /* flush if Traffic Class fields are different */
 248                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 249                 NAPI_GRO_CB(p)->flush |= flush;
 250
 251                 /* If the previous IP ID value was based on an atomic
 252                  * datagram we can overwrite the value and ignore it.
 253                  */
 254                 if (NAPI_GRO_CB(skb)->is_atomic)
 255                         NAPI_GRO_CB(p)->flush_id = 0;
 256         }
 257
 258         NAPI_GRO_CB(skb)->is_atomic = true;
 259         NAPI_GRO_CB(skb)->flush |= flush;
 260
 261         skb_gro_postpull_rcsum(skb, iph, nlen);
 262
 263         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 264
 265 out_unlock:
 266         rcu_read_unlock();
 267
 268 out:
 269         skb_gro_flush_final(skb, pp, flush);
 270
 271         return pp;
 272 }
 273
 274 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
 275                                               struct sk_buff *skb)
 276 {
 277         /* Common GRO receive for SIT and IP6IP6 */
 278
 279         if (NAPI_GRO_CB(skb)->encap_mark) {
 280                 NAPI_GRO_CB(skb)->flush = 1;
 281                 return NULL;
 282         }
 283
 284         NAPI_GRO_CB(skb)->encap_mark = 1;
 285
 286         return ipv6_gro_receive(head, skb);
 287 }
 288
 289 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
 290                                           struct sk_buff *skb)
 291 {
 292         /* Common GRO receive for SIT and IP6IP6 */
 293
 294         if (NAPI_GRO_CB(skb)->encap_mark) {
 295                 NAPI_GRO_CB(skb)->flush = 1;
 296                 return NULL;
 297         }
 298
 299         NAPI_GRO_CB(skb)->encap_mark = 1;
 300
 301         return inet_gro_receive(head, skb);
 302 }
 303
 304 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 305 {
 306         const struct net_offload *ops;
 307         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 308         int err = -ENOSYS;
 309
 310         if (skb->encapsulation) {
 311                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
 312                 skb_set_inner_network_header(skb, nhoff);
 313         }
 314
 315         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 316
 317         rcu_read_lock();
 318
 319         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 320         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 321                 goto out_unlock;
 322
 323         err = ops->callbacks.gro_complete(skb, nhoff);
 324
 325 out_unlock:
 326         rcu_read_unlock();
 327
 328         return err;
 329 }
 330
 331 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 332 {
 333         skb->encapsulation = 1;
 334         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
 335         return ipv6_gro_complete(skb, nhoff);
 336 }
 337
 338 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
 339 {
 340         skb->encapsulation = 1;
 341         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 342         return ipv6_gro_complete(skb, nhoff);
 343 }
 344
 345 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 346 {
 347         skb->encapsulation = 1;
 348         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 349         return inet_gro_complete(skb, nhoff);
 350 }
 351
 352 static struct packet_offload ipv6_packet_offload __read_mostly = {
 353         .type = cpu_to_be16(ETH_P_IPV6),
 354         .callbacks = {
 355                 .gso_segment = ipv6_gso_segment,
 356                 .gro_receive = ipv6_gro_receive,
 357                 .gro_complete = ipv6_gro_complete,
 358         },
 359 };
 360
 361 static const struct net_offload sit_offload = {
 362         .callbacks = {
 363                 .gso_segment    = ipv6_gso_segment,
 364                 .gro_receive    = sit_ip6ip6_gro_receive,
 365                 .gro_complete   = sit_gro_complete,
 366         },
 367 };
 368
 369 static const struct net_offload ip4ip6_offload = {
 370         .callbacks = {
 371                 .gso_segment    = inet_gso_segment,
 372                 .gro_receive    = ip4ip6_gro_receive,
 373                 .gro_complete   = ip4ip6_gro_complete,
 374         },
 375 };
 376
 377 static const struct net_offload ip6ip6_offload = {
 378         .callbacks = {
 379                 .gso_segment    = ipv6_gso_segment,
 380                 .gro_receive    = sit_ip6ip6_gro_receive,
 381                 .gro_complete   = ip6ip6_gro_complete,
 382         },
 383 };
 384 static int __init ipv6_offload_init(void)
 385 {
 386
 387         if (tcpv6_offload_init() < 0)
 388                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 389         if (ipv6_exthdrs_offload_init() < 0)
 390                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 391
 392         dev_add_offload(&ipv6_packet_offload);
 393
 394         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 395         inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
 396         inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
 397
 398         return 0;
 399 }
 400
 401 fs_initcall(ipv6_offload_init);