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