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         bool gso_partial;
  73
  74         skb_reset_network_header(skb);
  75         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  76         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  77                 goto out;
  78
  79         encap = SKB_GSO_CB(skb)->encap_level > 0;
  80         if (encap)
  81                 features &= skb->dev->hw_enc_features;
  82         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  83
  84         ipv6h = ipv6_hdr(skb);
  85         __skb_pull(skb, sizeof(*ipv6h));
  86         segs = ERR_PTR(-EPROTONOSUPPORT);
  87
  88         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
  89
  90         if (skb->encapsulation &&
  91             skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
  92                 udpfrag = proto == IPPROTO_UDP && encap;
  93         else
  94                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
  95
  96         ops = rcu_dereference(inet6_offloads[proto]);
  97         if (likely(ops && ops->callbacks.gso_segment)) {
  98                 skb_reset_transport_header(skb);
  99                 segs = ops->callbacks.gso_segment(skb, features);
 100         }
 101
 102         if (IS_ERR_OR_NULL(segs))
 103                 goto out;
 104
 105         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
 106
 107         for (skb = segs; skb; skb = skb->next) {
 108                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 109                 if (gso_partial)
 110                         payload_len = skb_shinfo(skb)->gso_size +
 111                                       SKB_GSO_CB(skb)->data_offset +
 112                                       skb->head - (unsigned char *)(ipv6h + 1);
 113                 else
 114                         payload_len = skb->len - nhoff - sizeof(*ipv6h);
 115                 ipv6h->payload_len = htons(payload_len);
 116                 skb->network_header = (u8 *)ipv6h - skb->head;
 117
 118                 if (udpfrag) {
 119                         unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
 120                         fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
 121                         fptr->frag_off = htons(offset);
 122                         if (skb->next)
 123                                 fptr->frag_off |= htons(IP6_MF);
 124                         offset += (ntohs(ipv6h->payload_len) -
 125                                    sizeof(struct frag_hdr));
 126                 }
 127                 if (encap)
 128                         skb_reset_inner_headers(skb);
 129         }
 130
 131 out:
 132         return segs;
 133 }
 134
 135 /* Return the total length of all the extension hdrs, following the same
 136  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 137  */
 138 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 139                             const struct net_offload **opps)
 140 {
 141         struct ipv6_opt_hdr *opth = (void *)iph;
 142         int len = 0, proto, optlen = sizeof(*iph);
 143
 144         proto = iph->nexthdr;
 145         for (;;) {
 146                 if (proto != NEXTHDR_HOP) {
 147                         *opps = rcu_dereference(inet6_offloads[proto]);
 148                         if (unlikely(!(*opps)))
 149                                 break;
 150                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 151                                 break;
 152                 }
 153                 opth = (void *)opth + optlen;
 154                 optlen = ipv6_optlen(opth);
 155                 len += optlen;
 156                 proto = opth->nexthdr;
 157         }
 158         return len;
 159 }
 160
 161 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 162                                          struct sk_buff *skb)
 163 {
 164         const struct net_offload *ops;
 165         struct sk_buff **pp = NULL;
 166         struct sk_buff *p;
 167         struct ipv6hdr *iph;
 168         unsigned int nlen;
 169         unsigned int hlen;
 170         unsigned int off;
 171         u16 flush = 1;
 172         int proto;
 173
 174         off = skb_gro_offset(skb);
 175         hlen = off + sizeof(*iph);
 176         iph = skb_gro_header_fast(skb, off);
 177         if (skb_gro_header_hard(skb, hlen)) {
 178                 iph = skb_gro_header_slow(skb, hlen, off);
 179                 if (unlikely(!iph))
 180                         goto out;
 181         }
 182
 183         skb_set_network_header(skb, off);
 184         skb_gro_pull(skb, sizeof(*iph));
 185         skb_set_transport_header(skb, skb_gro_offset(skb));
 186
 187         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 188
 189         rcu_read_lock();
 190         proto = iph->nexthdr;
 191         ops = rcu_dereference(inet6_offloads[proto]);
 192         if (!ops || !ops->callbacks.gro_receive) {
 193                 __pskb_pull(skb, skb_gro_offset(skb));
 194                 skb_gro_frag0_invalidate(skb);
 195                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 196                 skb_gro_pull(skb, -skb_transport_offset(skb));
 197                 skb_reset_transport_header(skb);
 198                 __skb_push(skb, skb_gro_offset(skb));
 199
 200                 ops = rcu_dereference(inet6_offloads[proto]);
 201                 if (!ops || !ops->callbacks.gro_receive)
 202                         goto out_unlock;
 203
 204                 iph = ipv6_hdr(skb);
 205         }
 206
 207         NAPI_GRO_CB(skb)->proto = proto;
 208
 209         flush--;
 210         nlen = skb_network_header_len(skb);
 211
 212         for (p = *head; p; p = p->next) {
 213                 const struct ipv6hdr *iph2;
 214                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 215
 216                 if (!NAPI_GRO_CB(p)->same_flow)
 217                         continue;
 218
 219                 iph2 = (struct ipv6hdr *)(p->data + off);
 220                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 221
 222                 /* All fields must match except length and Traffic Class.
 223                  * XXX skbs on the gro_list have all been parsed and pulled
 224                  * already so we don't need to compare nlen
 225                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 226                  * memcmp() alone below is suffcient, right?
 227                  */
 228                  if ((first_word & htonl(0xF00FFFFF)) ||
 229                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 230                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 231                         NAPI_GRO_CB(p)->same_flow = 0;
 232                         continue;
 233                 }
 234                 /* flush if Traffic Class fields are different */
 235                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 236                 NAPI_GRO_CB(p)->flush |= flush;
 237
 238                 /* If the previous IP ID value was based on an atomic
 239                  * datagram we can overwrite the value and ignore it.
 240                  */
 241                 if (NAPI_GRO_CB(skb)->is_atomic)
 242                         NAPI_GRO_CB(p)->flush_id = 0;
 243         }
 244
 245         NAPI_GRO_CB(skb)->is_atomic = true;
 246         NAPI_GRO_CB(skb)->flush |= flush;
 247
 248         skb_gro_postpull_rcsum(skb, iph, nlen);
 249
 250         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 251
 252 out_unlock:
 253         rcu_read_unlock();
 254
 255 out:
 256         NAPI_GRO_CB(skb)->flush |= flush;
 257
 258         return pp;
 259 }
 260
 261 static struct sk_buff **sit_ip6ip6_gro_receive(struct sk_buff **head,
 262                                                struct sk_buff *skb)
 263 {
 264         /* Common GRO receive for SIT and IP6IP6 */
 265
 266         if (NAPI_GRO_CB(skb)->encap_mark) {
 267                 NAPI_GRO_CB(skb)->flush = 1;
 268                 return NULL;
 269         }
 270
 271         NAPI_GRO_CB(skb)->encap_mark = 1;
 272
 273         return ipv6_gro_receive(head, skb);
 274 }
 275
 276 static struct sk_buff **ip4ip6_gro_receive(struct sk_buff **head,
 277                                            struct sk_buff *skb)
 278 {
 279         /* Common GRO receive for SIT and IP6IP6 */
 280
 281         if (NAPI_GRO_CB(skb)->encap_mark) {
 282                 NAPI_GRO_CB(skb)->flush = 1;
 283                 return NULL;
 284         }
 285
 286         NAPI_GRO_CB(skb)->encap_mark = 1;
 287
 288         return inet_gro_receive(head, skb);
 289 }
 290
 291 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 292 {
 293         const struct net_offload *ops;
 294         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 295         int err = -ENOSYS;
 296
 297         if (skb->encapsulation)
 298                 skb_set_inner_network_header(skb, nhoff);
 299
 300         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 301
 302         rcu_read_lock();
 303
 304         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 305         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 306                 goto out_unlock;
 307
 308         err = ops->callbacks.gro_complete(skb, nhoff);
 309
 310 out_unlock:
 311         rcu_read_unlock();
 312
 313         return err;
 314 }
 315
 316 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 317 {
 318         skb->encapsulation = 1;
 319         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
 320         return ipv6_gro_complete(skb, nhoff);
 321 }
 322
 323 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
 324 {
 325         skb->encapsulation = 1;
 326         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 327         return ipv6_gro_complete(skb, nhoff);
 328 }
 329
 330 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
 331 {
 332         skb->encapsulation = 1;
 333         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
 334         return inet_gro_complete(skb, nhoff);
 335 }
 336
 337 static struct packet_offload ipv6_packet_offload __read_mostly = {
 338         .type = cpu_to_be16(ETH_P_IPV6),
 339         .callbacks = {
 340                 .gso_segment = ipv6_gso_segment,
 341                 .gro_receive = ipv6_gro_receive,
 342                 .gro_complete = ipv6_gro_complete,
 343         },
 344 };
 345
 346 static const struct net_offload sit_offload = {
 347         .callbacks = {
 348                 .gso_segment    = ipv6_gso_segment,
 349                 .gro_receive    = sit_ip6ip6_gro_receive,
 350                 .gro_complete   = sit_gro_complete,
 351         },
 352 };
 353
 354 static const struct net_offload ip4ip6_offload = {
 355         .callbacks = {
 356                 .gso_segment    = inet_gso_segment,
 357                 .gro_receive    = ip4ip6_gro_receive,
 358                 .gro_complete   = ip4ip6_gro_complete,
 359         },
 360 };
 361
 362 static const struct net_offload ip6ip6_offload = {
 363         .callbacks = {
 364                 .gso_segment    = ipv6_gso_segment,
 365                 .gro_receive    = sit_ip6ip6_gro_receive,
 366                 .gro_complete   = ip6ip6_gro_complete,
 367         },
 368 };
 369 static int __init ipv6_offload_init(void)
 370 {
 371
 372         if (tcpv6_offload_init() < 0)
 373                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 374         if (ipv6_exthdrs_offload_init() < 0)
 375                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 376
 377         dev_add_offload(&ipv6_packet_offload);
 378
 379         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 380         inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
 381         inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
 382
 383         return 0;
 384 }
 385
 386 fs_initcall(ipv6_offload_init);