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
  20 #include "ip6_offload.h"
  21
  22 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
  23 {
  24         const struct net_offload *ops = NULL;
  25
  26         for (;;) {
  27                 struct ipv6_opt_hdr *opth;
  28                 int len;
  29
  30                 if (proto != NEXTHDR_HOP) {
  31                         ops = rcu_dereference(inet6_offloads[proto]);
  32
  33                         if (unlikely(!ops))
  34                                 break;
  35
  36                         if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
  37                                 break;
  38                 }
  39
  40                 if (unlikely(!pskb_may_pull(skb, 8)))
  41                         break;
  42
  43                 opth = (void *)skb->data;
  44                 len = ipv6_optlen(opth);
  45
  46                 if (unlikely(!pskb_may_pull(skb, len)))
  47                         break;
  48
  49                 opth = (void *)skb->data;
  50                 proto = opth->nexthdr;
  51                 __skb_pull(skb, len);
  52         }
  53
  54         return proto;
  55 }
  56
  57 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
  58         netdev_features_t features)
  59 {
  60         struct sk_buff *segs = ERR_PTR(-EINVAL);
  61         struct ipv6hdr *ipv6h;
  62         const struct net_offload *ops;
  63         int proto;
  64         struct frag_hdr *fptr;
  65         unsigned int unfrag_ip6hlen;
  66         unsigned int payload_len;
  67         u8 *prevhdr;
  68         int offset = 0;
  69         bool encap, udpfrag;
  70         int nhoff;
  71
  72         skb_reset_network_header(skb);
  73         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  74         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  75                 goto out;
  76
  77         encap = SKB_GSO_CB(skb)->encap_level > 0;
  78         if (encap)
  79                 features &= skb->dev->hw_enc_features;
  80         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  81
  82         ipv6h = ipv6_hdr(skb);
  83         __skb_pull(skb, sizeof(*ipv6h));
  84         segs = ERR_PTR(-EPROTONOSUPPORT);
  85
  86         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
  87
  88         if (skb->encapsulation &&
  89             skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
  90                 udpfrag = proto == IPPROTO_UDP && encap;
  91         else
  92                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
  93
  94         ops = rcu_dereference(inet6_offloads[proto]);
  95         if (likely(ops && ops->callbacks.gso_segment)) {
  96                 skb_reset_transport_header(skb);
  97                 segs = ops->callbacks.gso_segment(skb, features);
  98         }
  99
 100         if (IS_ERR(segs))
 101                 goto out;
 102
 103         for (skb = segs; skb; skb = skb->next) {
 104                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 105                 if (skb_is_gso(skb))
 106                         payload_len = skb_shinfo(skb)->gso_size +
 107                                       SKB_GSO_CB(skb)->data_offset +
 108                                       skb->head - (unsigned char *)(ipv6h + 1);
 109                 else
 110                         payload_len = skb->len - nhoff - sizeof(*ipv6h);
 111                 ipv6h->payload_len = htons(payload_len);
 112                 skb->network_header = (u8 *)ipv6h - skb->head;
 113
 114                 if (udpfrag) {
 115                         unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
 116                         fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
 117                         fptr->frag_off = htons(offset);
 118                         if (skb->next)
 119                                 fptr->frag_off |= htons(IP6_MF);
 120                         offset += (ntohs(ipv6h->payload_len) -
 121                                    sizeof(struct frag_hdr));
 122                 }
 123                 if (encap)
 124                         skb_reset_inner_headers(skb);
 125         }
 126
 127 out:
 128         return segs;
 129 }
 130
 131 /* Return the total length of all the extension hdrs, following the same
 132  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 133  */
 134 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 135                             const struct net_offload **opps)
 136 {
 137         struct ipv6_opt_hdr *opth = (void *)iph;
 138         int len = 0, proto, optlen = sizeof(*iph);
 139
 140         proto = iph->nexthdr;
 141         for (;;) {
 142                 if (proto != NEXTHDR_HOP) {
 143                         *opps = rcu_dereference(inet6_offloads[proto]);
 144                         if (unlikely(!(*opps)))
 145                                 break;
 146                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 147                                 break;
 148                 }
 149                 opth = (void *)opth + optlen;
 150                 optlen = ipv6_optlen(opth);
 151                 len += optlen;
 152                 proto = opth->nexthdr;
 153         }
 154         return len;
 155 }
 156
 157 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 158                                          struct sk_buff *skb)
 159 {
 160         const struct net_offload *ops;
 161         struct sk_buff **pp = NULL;
 162         struct sk_buff *p;
 163         struct ipv6hdr *iph;
 164         unsigned int nlen;
 165         unsigned int hlen;
 166         unsigned int off;
 167         u16 flush = 1;
 168         int proto;
 169
 170         off = skb_gro_offset(skb);
 171         hlen = off + sizeof(*iph);
 172         iph = skb_gro_header_fast(skb, off);
 173         if (skb_gro_header_hard(skb, hlen)) {
 174                 iph = skb_gro_header_slow(skb, hlen, off);
 175                 if (unlikely(!iph))
 176                         goto out;
 177         }
 178
 179         skb_set_network_header(skb, off);
 180         skb_gro_pull(skb, sizeof(*iph));
 181         skb_set_transport_header(skb, skb_gro_offset(skb));
 182
 183         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 184
 185         rcu_read_lock();
 186         proto = iph->nexthdr;
 187         ops = rcu_dereference(inet6_offloads[proto]);
 188         if (!ops || !ops->callbacks.gro_receive) {
 189                 __pskb_pull(skb, skb_gro_offset(skb));
 190                 proto = ipv6_gso_pull_exthdrs(skb, proto);
 191                 skb_gro_pull(skb, -skb_transport_offset(skb));
 192                 skb_reset_transport_header(skb);
 193                 __skb_push(skb, skb_gro_offset(skb));
 194
 195                 ops = rcu_dereference(inet6_offloads[proto]);
 196                 if (!ops || !ops->callbacks.gro_receive)
 197                         goto out_unlock;
 198
 199                 iph = ipv6_hdr(skb);
 200         }
 201
 202         NAPI_GRO_CB(skb)->proto = proto;
 203
 204         flush--;
 205         nlen = skb_network_header_len(skb);
 206
 207         for (p = *head; p; p = p->next) {
 208                 const struct ipv6hdr *iph2;
 209                 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
 210
 211                 if (!NAPI_GRO_CB(p)->same_flow)
 212                         continue;
 213
 214                 iph2 = (struct ipv6hdr *)(p->data + off);
 215                 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
 216
 217                 /* All fields must match except length and Traffic Class.
 218                  * XXX skbs on the gro_list have all been parsed and pulled
 219                  * already so we don't need to compare nlen
 220                  * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
 221                  * memcmp() alone below is suffcient, right?
 222                  */
 223                  if ((first_word & htonl(0xF00FFFFF)) ||
 224                     memcmp(&iph->nexthdr, &iph2->nexthdr,
 225                            nlen - offsetof(struct ipv6hdr, nexthdr))) {
 226                         NAPI_GRO_CB(p)->same_flow = 0;
 227                         continue;
 228                 }
 229                 /* flush if Traffic Class fields are different */
 230                 NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
 231                 NAPI_GRO_CB(p)->flush |= flush;
 232
 233                 /* If the previous IP ID value was based on an atomic
 234                  * datagram we can overwrite the value and ignore it.
 235                  */
 236                 if (NAPI_GRO_CB(skb)->is_atomic)
 237                         NAPI_GRO_CB(p)->flush_id = 0;
 238         }
 239
 240         NAPI_GRO_CB(skb)->is_atomic = true;
 241         NAPI_GRO_CB(skb)->flush |= flush;
 242
 243         skb_gro_postpull_rcsum(skb, iph, nlen);
 244
 245         pp = ops->callbacks.gro_receive(head, skb);
 246
 247 out_unlock:
 248         rcu_read_unlock();
 249
 250 out:
 251         NAPI_GRO_CB(skb)->flush |= flush;
 252
 253         return pp;
 254 }
 255
 256 static struct sk_buff **sit_gro_receive(struct sk_buff **head,
 257                                         struct sk_buff *skb)
 258 {
 259         if (NAPI_GRO_CB(skb)->encap_mark) {
 260                 NAPI_GRO_CB(skb)->flush = 1;
 261                 return NULL;
 262         }
 263
 264         NAPI_GRO_CB(skb)->encap_mark = 1;
 265
 266         return ipv6_gro_receive(head, skb);
 267 }
 268
 269 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 270 {
 271         const struct net_offload *ops;
 272         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 273         int err = -ENOSYS;
 274
 275         if (skb->encapsulation)
 276                 skb_set_inner_network_header(skb, nhoff);
 277
 278         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 279
 280         rcu_read_lock();
 281
 282         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 283         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 284                 goto out_unlock;
 285
 286         err = ops->callbacks.gro_complete(skb, nhoff);
 287
 288 out_unlock:
 289         rcu_read_unlock();
 290
 291         return err;
 292 }
 293
 294 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
 295 {
 296         skb->encapsulation = 1;
 297         skb_shinfo(skb)->gso_type |= SKB_GSO_SIT;
 298         return ipv6_gro_complete(skb, nhoff);
 299 }
 300
 301 static struct packet_offload ipv6_packet_offload __read_mostly = {
 302         .type = cpu_to_be16(ETH_P_IPV6),
 303         .callbacks = {
 304                 .gso_segment = ipv6_gso_segment,
 305                 .gro_receive = ipv6_gro_receive,
 306                 .gro_complete = ipv6_gro_complete,
 307         },
 308 };
 309
 310 static const struct net_offload sit_offload = {
 311         .callbacks = {
 312                 .gso_segment    = ipv6_gso_segment,
 313                 .gro_receive    = sit_gro_receive,
 314                 .gro_complete   = sit_gro_complete,
 315         },
 316 };
 317
 318 static int __init ipv6_offload_init(void)
 319 {
 320
 321         if (tcpv6_offload_init() < 0)
 322                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 323         if (ipv6_exthdrs_offload_init() < 0)
 324                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 325
 326         dev_add_offload(&ipv6_packet_offload);
 327
 328         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 329
 330         return 0;
 331 }
 332
 333 fs_initcall(ipv6_offload_init);