net/ipv4/gre_demux.c

   1 /*
   2  *      GRE over IPv4 demultiplexer driver
   3  *
   4  *      Authors: Dmitry Kozlov (xeb@mail.ru)
   5  *
   6  *      This program is free software; you can redistribute it and/or
   7  *      modify it under the terms of the GNU General Public License
   8  *      as published by the Free Software Foundation; either version
   9  *      2 of the License, or (at your option) any later version.
  10  *
  11  */
  12
  13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  15 #include <linux/module.h>
  16 #include <linux/if.h>
  17 #include <linux/icmp.h>
  18 #include <linux/kernel.h>
  19 #include <linux/kmod.h>
  20 #include <linux/skbuff.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/if_tunnel.h>
  25 #include <linux/spinlock.h>
  26 #include <net/protocol.h>
  27 #include <net/gre.h>
  28
  29 #include <net/icmp.h>
  30 #include <net/route.h>
  31 #include <net/xfrm.h>
  32
  33 static const struct gre_protocol __rcu *gre_proto[GREPROTO_MAX] __read_mostly;
  34 static struct gre_cisco_protocol __rcu *gre_cisco_proto_list[GRE_IP_PROTO_MAX];
  35
  36 int gre_add_protocol(const struct gre_protocol *proto, u8 version)
  37 {
  38         if (version >= GREPROTO_MAX)
  39                 return -EINVAL;
  40
  41         return (cmpxchg((const struct gre_protocol **)&gre_proto[version], NULL, proto) == NULL) ?
  42                 0 : -EBUSY;
  43 }
  44 EXPORT_SYMBOL_GPL(gre_add_protocol);
  45
  46 int gre_del_protocol(const struct gre_protocol *proto, u8 version)
  47 {
  48         int ret;
  49
  50         if (version >= GREPROTO_MAX)
  51                 return -EINVAL;
  52
  53         ret = (cmpxchg((const struct gre_protocol **)&gre_proto[version], proto, NULL) == proto) ?
  54                 0 : -EBUSY;
  55
  56         if (ret)
  57                 return ret;
  58
  59         synchronize_rcu();
  60         return 0;
  61 }
  62 EXPORT_SYMBOL_GPL(gre_del_protocol);
  63
  64 void gre_build_header(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
  65                       int hdr_len)
  66 {
  67         struct gre_base_hdr *greh;
  68
  69         skb_push(skb, hdr_len);
  70
  71         greh = (struct gre_base_hdr *)skb->data;
  72         greh->flags = tnl_flags_to_gre_flags(tpi->flags);
  73         greh->protocol = tpi->proto;
  74
  75         if (tpi->flags&(TUNNEL_KEY|TUNNEL_CSUM|TUNNEL_SEQ)) {
  76                 __be32 *ptr = (__be32 *)(((u8 *)greh) + hdr_len - 4);
  77
  78                 if (tpi->flags&TUNNEL_SEQ) {
  79                         *ptr = tpi->seq;
  80                         ptr--;
  81                 }
  82                 if (tpi->flags&TUNNEL_KEY) {
  83                         *ptr = tpi->key;
  84                         ptr--;
  85                 }
  86                 if (tpi->flags&TUNNEL_CSUM &&
  87                     !(skb_shinfo(skb)->gso_type & SKB_GSO_GRE)) {
  88                         *ptr = 0;
  89                         *(__sum16 *)ptr = csum_fold(skb_checksum(skb, 0,
  90                                                                  skb->len, 0));
  91                 }
  92         }
  93 }
  94 EXPORT_SYMBOL_GPL(gre_build_header);
  95
  96 static int parse_gre_header(struct sk_buff *skb, struct tnl_ptk_info *tpi,
  97                             bool *csum_err)
  98 {
  99         unsigned int ip_hlen = ip_hdrlen(skb);
 100         const struct gre_base_hdr *greh;
 101         __be32 *options;
 102         int hdr_len;
 103
 104         if (unlikely(!pskb_may_pull(skb, sizeof(struct gre_base_hdr))))
 105                 return -EINVAL;
 106
 107         greh = (struct gre_base_hdr *)(skb_network_header(skb) + ip_hlen);
 108         if (unlikely(greh->flags & (GRE_VERSION | GRE_ROUTING)))
 109                 return -EINVAL;
 110
 111         tpi->flags = gre_flags_to_tnl_flags(greh->flags);
 112         hdr_len = ip_gre_calc_hlen(tpi->flags);
 113
 114         if (!pskb_may_pull(skb, hdr_len))
 115                 return -EINVAL;
 116
 117         greh = (struct gre_base_hdr *)(skb_network_header(skb) + ip_hlen);
 118         tpi->proto = greh->protocol;
 119
 120         options = (__be32 *)(greh + 1);
 121         if (greh->flags & GRE_CSUM) {
 122                 if (skb_checksum_simple_validate(skb)) {
 123                         *csum_err = true;
 124                         return -EINVAL;
 125                 }
 126                 options++;
 127         }
 128
 129         if (greh->flags & GRE_KEY) {
 130                 tpi->key = *options;
 131                 options++;
 132         } else
 133                 tpi->key = 0;
 134
 135         if (unlikely(greh->flags & GRE_SEQ)) {
 136                 tpi->seq = *options;
 137                 options++;
 138         } else
 139                 tpi->seq = 0;
 140
 141         /* WCCP version 1 and 2 protocol decoding.
 142          * - Change protocol to IP
 143          * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
 144          */
 145         if (greh->flags == 0 && tpi->proto == htons(ETH_P_WCCP)) {
 146                 tpi->proto = htons(ETH_P_IP);
 147                 if ((*(u8 *)options & 0xF0) != 0x40) {
 148                         hdr_len += 4;
 149                         if (!pskb_may_pull(skb, hdr_len))
 150                                 return -EINVAL;
 151                 }
 152         }
 153
 154         return iptunnel_pull_header(skb, hdr_len, tpi->proto);
 155 }
 156
 157 static int gre_cisco_rcv(struct sk_buff *skb)
 158 {
 159         struct tnl_ptk_info tpi;
 160         int i;
 161         bool csum_err = false;
 162
 163 #ifdef CONFIG_NET_IPGRE_BROADCAST
 164         if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
 165                 /* Looped back packet, drop it! */
 166                 if (rt_is_output_route(skb_rtable(skb)))
 167                         goto drop;
 168         }
 169 #endif
 170
 171         if (parse_gre_header(skb, &tpi, &csum_err) < 0)
 172                 goto drop;
 173
 174         rcu_read_lock();
 175         for (i = 0; i < GRE_IP_PROTO_MAX; i++) {
 176                 struct gre_cisco_protocol *proto;
 177                 int ret;
 178
 179                 proto = rcu_dereference(gre_cisco_proto_list[i]);
 180                 if (!proto)
 181                         continue;
 182                 ret = proto->handler(skb, &tpi);
 183                 if (ret == PACKET_RCVD) {
 184                         rcu_read_unlock();
 185                         return 0;
 186                 }
 187         }
 188         rcu_read_unlock();
 189
 190         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 191 drop:
 192         kfree_skb(skb);
 193         return 0;
 194 }
 195
 196 static void gre_cisco_err(struct sk_buff *skb, u32 info)
 197 {
 198         /* All the routers (except for Linux) return only
 199          * 8 bytes of packet payload. It means, that precise relaying of
 200          * ICMP in the real Internet is absolutely infeasible.
 201          *
 202          * Moreover, Cisco "wise men" put GRE key to the third word
 203          * in GRE header. It makes impossible maintaining even soft
 204          * state for keyed
 205          * GRE tunnels with enabled checksum. Tell them "thank you".
 206          *
 207          * Well, I wonder, rfc1812 was written by Cisco employee,
 208          * what the hell these idiots break standards established
 209          * by themselves???
 210          */
 211
 212         const int type = icmp_hdr(skb)->type;
 213         const int code = icmp_hdr(skb)->code;
 214         struct tnl_ptk_info tpi;
 215         bool csum_err = false;
 216         int i;
 217
 218         if (parse_gre_header(skb, &tpi, &csum_err)) {
 219                 if (!csum_err)          /* ignore csum errors. */
 220                         return;
 221         }
 222
 223         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
 224                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
 225                                 skb->dev->ifindex, 0, IPPROTO_GRE, 0);
 226                 return;
 227         }
 228         if (type == ICMP_REDIRECT) {
 229                 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
 230                                 IPPROTO_GRE, 0);
 231                 return;
 232         }
 233
 234         rcu_read_lock();
 235         for (i = 0; i < GRE_IP_PROTO_MAX; i++) {
 236                 struct gre_cisco_protocol *proto;
 237
 238                 proto = rcu_dereference(gre_cisco_proto_list[i]);
 239                 if (!proto)
 240                         continue;
 241
 242                 if (proto->err_handler(skb, info, &tpi) == PACKET_RCVD)
 243                         goto out;
 244
 245         }
 246 out:
 247         rcu_read_unlock();
 248 }
 249
 250 static int gre_rcv(struct sk_buff *skb)
 251 {
 252         const struct gre_protocol *proto;
 253         u8 ver;
 254         int ret;
 255
 256         if (!pskb_may_pull(skb, 12))
 257                 goto drop;
 258
 259         ver = skb->data[1]&0x7f;
 260         if (ver >= GREPROTO_MAX)
 261                 goto drop;
 262
 263         rcu_read_lock();
 264         proto = rcu_dereference(gre_proto[ver]);
 265         if (!proto || !proto->handler)
 266                 goto drop_unlock;
 267         ret = proto->handler(skb);
 268         rcu_read_unlock();
 269         return ret;
 270
 271 drop_unlock:
 272         rcu_read_unlock();
 273 drop:
 274         kfree_skb(skb);
 275         return NET_RX_DROP;
 276 }
 277
 278 static void gre_err(struct sk_buff *skb, u32 info)
 279 {
 280         const struct gre_protocol *proto;
 281         const struct iphdr *iph = (const struct iphdr *)skb->data;
 282         u8 ver = skb->data[(iph->ihl<<2) + 1]&0x7f;
 283
 284         if (ver >= GREPROTO_MAX)
 285                 return;
 286
 287         rcu_read_lock();
 288         proto = rcu_dereference(gre_proto[ver]);
 289         if (proto && proto->err_handler)
 290                 proto->err_handler(skb, info);
 291         rcu_read_unlock();
 292 }
 293
 294 static const struct net_protocol net_gre_protocol = {
 295         .handler     = gre_rcv,
 296         .err_handler = gre_err,
 297         .netns_ok    = 1,
 298 };
 299
 300 static const struct gre_protocol ipgre_protocol = {
 301         .handler     = gre_cisco_rcv,
 302         .err_handler = gre_cisco_err,
 303 };
 304
 305 int gre_cisco_register(struct gre_cisco_protocol *newp)
 306 {
 307         struct gre_cisco_protocol **proto = (struct gre_cisco_protocol **)
 308                                             &gre_cisco_proto_list[newp->priority];
 309
 310         return (cmpxchg(proto, NULL, newp) == NULL) ? 0 : -EBUSY;
 311 }
 312 EXPORT_SYMBOL_GPL(gre_cisco_register);
 313
 314 int gre_cisco_unregister(struct gre_cisco_protocol *del_proto)
 315 {
 316         struct gre_cisco_protocol **proto = (struct gre_cisco_protocol **)
 317                                             &gre_cisco_proto_list[del_proto->priority];
 318         int ret;
 319
 320         ret = (cmpxchg(proto, del_proto, NULL) == del_proto) ? 0 : -EINVAL;
 321
 322         if (ret)
 323                 return ret;
 324
 325         synchronize_net();
 326         return 0;
 327 }
 328 EXPORT_SYMBOL_GPL(gre_cisco_unregister);
 329
 330 static int __init gre_init(void)
 331 {
 332         pr_info("GRE over IPv4 demultiplexor driver\n");
 333
 334         if (inet_add_protocol(&net_gre_protocol, IPPROTO_GRE) < 0) {
 335                 pr_err("can't add protocol\n");
 336                 goto err;
 337         }
 338
 339         if (gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0) {
 340                 pr_info("%s: can't add ipgre handler\n", __func__);
 341                 goto err_gre;
 342         }
 343
 344         return 0;
 345 err_gre:
 346         inet_del_protocol(&net_gre_protocol, IPPROTO_GRE);
 347 err:
 348         return -EAGAIN;
 349 }
 350
 351 static void __exit gre_exit(void)
 352 {
 353         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
 354         inet_del_protocol(&net_gre_protocol, IPPROTO_GRE);
 355 }
 356
 357 module_init(gre_init);
 358 module_exit(gre_exit);
 359
 360 MODULE_DESCRIPTION("GRE over IPv4 demultiplexer driver");
 361 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
 362 MODULE_LICENSE("GPL");