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         u8 *prevhdr;
  67         int offset = 0;
  68         bool encap, udpfrag;
  69         int nhoff;
  70
  71         if (unlikely(skb_shinfo(skb)->gso_type &
  72                      ~(SKB_GSO_TCPV4 |
  73                        SKB_GSO_UDP |
  74                        SKB_GSO_DODGY |
  75                        SKB_GSO_TCP_ECN |
  76                        SKB_GSO_GRE |
  77                        SKB_GSO_GRE_CSUM |
  78                        SKB_GSO_IPIP |
  79                        SKB_GSO_SIT |
  80                        SKB_GSO_UDP_TUNNEL |
  81                        SKB_GSO_UDP_TUNNEL_CSUM |
  82                        SKB_GSO_TUNNEL_REMCSUM |
  83                        SKB_GSO_TCPV6 |
  84                        0)))
  85                 goto out;
  86
  87         skb_reset_network_header(skb);
  88         nhoff = skb_network_header(skb) - skb_mac_header(skb);
  89         if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
  90                 goto out;
  91
  92         encap = SKB_GSO_CB(skb)->encap_level > 0;
  93         if (encap)
  94                 features &= skb->dev->hw_enc_features;
  95         SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
  96
  97         ipv6h = ipv6_hdr(skb);
  98         __skb_pull(skb, sizeof(*ipv6h));
  99         segs = ERR_PTR(-EPROTONOSUPPORT);
 100
 101         proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
 102
 103         if (skb->encapsulation &&
 104             skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
 105                 udpfrag = proto == IPPROTO_UDP && encap;
 106         else
 107                 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
 108
 109         ops = rcu_dereference(inet6_offloads[proto]);
 110         if (likely(ops && ops->callbacks.gso_segment)) {
 111                 skb_reset_transport_header(skb);
 112                 segs = ops->callbacks.gso_segment(skb, features);
 113         }
 114
 115         if (IS_ERR(segs))
 116                 goto out;
 117
 118         for (skb = segs; skb; skb = skb->next) {
 119                 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
 120                 ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h));
 121                 skb->network_header = (u8 *)ipv6h - skb->head;
 122
 123                 if (udpfrag) {
 124                         unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
 125                         fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
 126                         fptr->frag_off = htons(offset);
 127                         if (skb->next)
 128                                 fptr->frag_off |= htons(IP6_MF);
 129                         offset += (ntohs(ipv6h->payload_len) -
 130                                    sizeof(struct frag_hdr));
 131                 }
 132                 if (encap)
 133                         skb_reset_inner_headers(skb);
 134         }
 135
 136 out:
 137         return segs;
 138 }
 139
 140 /* Return the total length of all the extension hdrs, following the same
 141  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
 142  */
 143 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 144                             const struct net_offload **opps)
 145 {
 146         struct ipv6_opt_hdr *opth = (void *)iph;
 147         int len = 0, proto, optlen = sizeof(*iph);
 148
 149         proto = iph->nexthdr;
 150         for (;;) {
 151                 if (proto != NEXTHDR_HOP) {
 152                         *opps = rcu_dereference(inet6_offloads[proto]);
 153                         if (unlikely(!(*opps)))
 154                                 break;
 155                         if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 156                                 break;
 157                 }
 158                 opth = (void *)opth + optlen;
 159                 optlen = ipv6_optlen(opth);
 160                 len += optlen;
 161                 proto = opth->nexthdr;
 162         }
 163         return len;
 164 }
 165
 166 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 167                                          struct sk_buff *skb)
 168 {
 169         const struct net_offload *ops;
 170         struct sk_buff **pp = NULL;
 171         struct sk_buff *p;
 172         struct ipv6hdr *iph;
 173         unsigned int nlen;
 174         unsigned int hlen;
 175         unsigned int off;
 176         u16 flush = 1;
 177         int proto;
 178
 179         off = skb_gro_offset(skb);
 180         hlen = off + sizeof(*iph);
 181         iph = skb_gro_header_fast(skb, off);
 182         if (skb_gro_header_hard(skb, hlen)) {
 183                 iph = skb_gro_header_slow(skb, hlen, off);
 184                 if (unlikely(!iph))
 185                         goto out;
 186         }
 187
 188         skb_set_network_header(skb, off);
 189         skb_gro_pull(skb, sizeof(*iph));
 190         skb_set_transport_header(skb, skb_gro_offset(skb));
 191
 192         flush += ntohs(iph->payload_len) != skb_gro_len(skb);
 193
 194         rcu_read_lock();
 195         proto = iph->nexthdr;
 196         ops = rcu_dereference(inet6_offloads[proto]);
 197         if (!ops || !ops->callbacks.gro_receive) {
 198                 __pskb_pull(skb, skb_gro_offset(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                 /* Clear flush_id, there's really no concept of ID in IPv6. */
 243                 NAPI_GRO_CB(p)->flush_id = 0;
 244         }
 245
 246         NAPI_GRO_CB(skb)->flush |= flush;
 247
 248         skb_gro_postpull_rcsum(skb, iph, nlen);
 249
 250         pp = 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 int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
 262 {
 263         const struct net_offload *ops;
 264         struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
 265         int err = -ENOSYS;
 266
 267         iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
 268
 269         rcu_read_lock();
 270
 271         nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
 272         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 273                 goto out_unlock;
 274
 275         err = ops->callbacks.gro_complete(skb, nhoff);
 276
 277 out_unlock:
 278         rcu_read_unlock();
 279
 280         return err;
 281 }
 282
 283 static struct packet_offload ipv6_packet_offload __read_mostly = {
 284         .type = cpu_to_be16(ETH_P_IPV6),
 285         .callbacks = {
 286                 .gso_segment = ipv6_gso_segment,
 287                 .gro_receive = ipv6_gro_receive,
 288                 .gro_complete = ipv6_gro_complete,
 289         },
 290 };
 291
 292 static const struct net_offload sit_offload = {
 293         .callbacks = {
 294                 .gso_segment    = ipv6_gso_segment,
 295         },
 296 };
 297
 298 static int __init ipv6_offload_init(void)
 299 {
 300
 301         if (tcpv6_offload_init() < 0)
 302                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
 303         if (udp_offload_init() < 0)
 304                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
 305         if (ipv6_exthdrs_offload_init() < 0)
 306                 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 307
 308         dev_add_offload(&ipv6_packet_offload);
 309
 310         inet_add_offload(&sit_offload, IPPROTO_IPV6);
 311
 312         return 0;
 313 }
 314
 315 fs_initcall(ipv6_offload_init);