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