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