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