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 __sum16 check_checksum(struct sk_buff *skb)
  97 {
  98         __sum16 csum = 0;
  99
 100         switch (skb->ip_summed) {
 101         case CHECKSUM_COMPLETE:
 102                 csum = csum_fold(skb->csum);
 103
 104                 if (!csum)
 105                         break;
 106                 /* Fall through. */
 107
 108         case CHECKSUM_NONE:
 109                 skb->csum = 0;
 110                 csum = __skb_checksum_complete(skb);
 111                 skb->ip_summed = CHECKSUM_COMPLETE;
 112                 break;
 113         }
 114
 115         return csum;
 116 }
 117
 118 static int parse_gre_header(struct sk_buff *skb, struct tnl_ptk_info *tpi,
 119                             bool *csum_err)
 120 {
 121         unsigned int ip_hlen = ip_hdrlen(skb);
 122         const struct gre_base_hdr *greh;
 123         __be32 *options;
 124         int hdr_len;
 125
 126         if (unlikely(!pskb_may_pull(skb, sizeof(struct gre_base_hdr))))
 127                 return -EINVAL;
 128
 129         greh = (struct gre_base_hdr *)(skb_network_header(skb) + ip_hlen);
 130         if (unlikely(greh->flags & (GRE_VERSION | GRE_ROUTING)))
 131                 return -EINVAL;
 132
 133         tpi->flags = gre_flags_to_tnl_flags(greh->flags);
 134         hdr_len = ip_gre_calc_hlen(tpi->flags);
 135
 136         if (!pskb_may_pull(skb, hdr_len))
 137                 return -EINVAL;
 138
 139         greh = (struct gre_base_hdr *)(skb_network_header(skb) + ip_hlen);
 140         tpi->proto = greh->protocol;
 141
 142         options = (__be32 *)(greh + 1);
 143         if (greh->flags & GRE_CSUM) {
 144                 if (check_checksum(skb)) {
 145                         *csum_err = true;
 146                         return -EINVAL;
 147                 }
 148                 options++;
 149         }
 150
 151         if (greh->flags & GRE_KEY) {
 152                 tpi->key = *options;
 153                 options++;
 154         } else
 155                 tpi->key = 0;
 156
 157         if (unlikely(greh->flags & GRE_SEQ)) {
 158                 tpi->seq = *options;
 159                 options++;
 160         } else
 161                 tpi->seq = 0;
 162
 163         /* WCCP version 1 and 2 protocol decoding.
 164          * - Change protocol to IP
 165          * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
 166          */
 167         if (greh->flags == 0 && tpi->proto == htons(ETH_P_WCCP)) {
 168                 tpi->proto = htons(ETH_P_IP);
 169                 if ((*(u8 *)options & 0xF0) != 0x40) {
 170                         hdr_len += 4;
 171                         if (!pskb_may_pull(skb, hdr_len))
 172                                 return -EINVAL;
 173                 }
 174         }
 175
 176         return iptunnel_pull_header(skb, hdr_len, tpi->proto);
 177 }
 178
 179 static int gre_cisco_rcv(struct sk_buff *skb)
 180 {
 181         struct tnl_ptk_info tpi;
 182         int i;
 183         bool csum_err = false;
 184
 185         if (parse_gre_header(skb, &tpi, &csum_err) < 0)
 186                 goto drop;
 187
 188         rcu_read_lock();
 189         for (i = 0; i < GRE_IP_PROTO_MAX; i++) {
 190                 struct gre_cisco_protocol *proto;
 191                 int ret;
 192
 193                 proto = rcu_dereference(gre_cisco_proto_list[i]);
 194                 if (!proto)
 195                         continue;
 196                 ret = proto->handler(skb, &tpi);
 197                 if (ret == PACKET_RCVD) {
 198                         rcu_read_unlock();
 199                         return 0;
 200                 }
 201         }
 202         rcu_read_unlock();
 203
 204         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 205 drop:
 206         kfree_skb(skb);
 207         return 0;
 208 }
 209
 210 static void gre_cisco_err(struct sk_buff *skb, u32 info)
 211 {
 212         /* All the routers (except for Linux) return only
 213          * 8 bytes of packet payload. It means, that precise relaying of
 214          * ICMP in the real Internet is absolutely infeasible.
 215          *
 216          * Moreover, Cisco "wise men" put GRE key to the third word
 217          * in GRE header. It makes impossible maintaining even soft
 218          * state for keyed
 219          * GRE tunnels with enabled checksum. Tell them "thank you".
 220          *
 221          * Well, I wonder, rfc1812 was written by Cisco employee,
 222          * what the hell these idiots break standards established
 223          * by themselves???
 224          */
 225
 226         const int type = icmp_hdr(skb)->type;
 227         const int code = icmp_hdr(skb)->code;
 228         struct tnl_ptk_info tpi;
 229         bool csum_err = false;
 230         int i;
 231
 232         if (parse_gre_header(skb, &tpi, &csum_err)) {
 233                 if (!csum_err)          /* ignore csum errors. */
 234                         return;
 235         }
 236
 237         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
 238                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
 239                                 skb->dev->ifindex, 0, IPPROTO_GRE, 0);
 240                 return;
 241         }
 242         if (type == ICMP_REDIRECT) {
 243                 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
 244                                 IPPROTO_GRE, 0);
 245                 return;
 246         }
 247
 248         rcu_read_lock();
 249         for (i = 0; i < GRE_IP_PROTO_MAX; i++) {
 250                 struct gre_cisco_protocol *proto;
 251
 252                 proto = rcu_dereference(gre_cisco_proto_list[i]);
 253                 if (!proto)
 254                         continue;
 255
 256                 if (proto->err_handler(skb, info, &tpi) == PACKET_RCVD)
 257                         goto out;
 258
 259         }
 260 out:
 261         rcu_read_unlock();
 262 }
 263
 264 static int gre_rcv(struct sk_buff *skb)
 265 {
 266         const struct gre_protocol *proto;
 267         u8 ver;
 268         int ret;
 269
 270         if (!pskb_may_pull(skb, 12))
 271                 goto drop;
 272
 273         ver = skb->data[1]&0x7f;
 274         if (ver >= GREPROTO_MAX)
 275                 goto drop;
 276
 277         rcu_read_lock();
 278         proto = rcu_dereference(gre_proto[ver]);
 279         if (!proto || !proto->handler)
 280                 goto drop_unlock;
 281         ret = proto->handler(skb);
 282         rcu_read_unlock();
 283         return ret;
 284
 285 drop_unlock:
 286         rcu_read_unlock();
 287 drop:
 288         kfree_skb(skb);
 289         return NET_RX_DROP;
 290 }
 291
 292 static void gre_err(struct sk_buff *skb, u32 info)
 293 {
 294         const struct gre_protocol *proto;
 295         const struct iphdr *iph = (const struct iphdr *)skb->data;
 296         u8 ver = skb->data[(iph->ihl<<2) + 1]&0x7f;
 297
 298         if (ver >= GREPROTO_MAX)
 299                 return;
 300
 301         rcu_read_lock();
 302         proto = rcu_dereference(gre_proto[ver]);
 303         if (proto && proto->err_handler)
 304                 proto->err_handler(skb, info);
 305         rcu_read_unlock();
 306 }
 307
 308 static const struct net_protocol net_gre_protocol = {
 309         .handler     = gre_rcv,
 310         .err_handler = gre_err,
 311         .netns_ok    = 1,
 312 };
 313
 314 static const struct gre_protocol ipgre_protocol = {
 315         .handler     = gre_cisco_rcv,
 316         .err_handler = gre_cisco_err,
 317 };
 318
 319 int gre_cisco_register(struct gre_cisco_protocol *newp)
 320 {
 321         struct gre_cisco_protocol **proto = (struct gre_cisco_protocol **)
 322                                             &gre_cisco_proto_list[newp->priority];
 323
 324         return (cmpxchg(proto, NULL, newp) == NULL) ? 0 : -EBUSY;
 325 }
 326 EXPORT_SYMBOL_GPL(gre_cisco_register);
 327
 328 int gre_cisco_unregister(struct gre_cisco_protocol *del_proto)
 329 {
 330         struct gre_cisco_protocol **proto = (struct gre_cisco_protocol **)
 331                                             &gre_cisco_proto_list[del_proto->priority];
 332         int ret;
 333
 334         ret = (cmpxchg(proto, del_proto, NULL) == del_proto) ? 0 : -EINVAL;
 335
 336         if (ret)
 337                 return ret;
 338
 339         synchronize_net();
 340         return 0;
 341 }
 342 EXPORT_SYMBOL_GPL(gre_cisco_unregister);
 343
 344 static int __init gre_init(void)
 345 {
 346         pr_info("GRE over IPv4 demultiplexor driver\n");
 347
 348         if (inet_add_protocol(&net_gre_protocol, IPPROTO_GRE) < 0) {
 349                 pr_err("can't add protocol\n");
 350                 goto err;
 351         }
 352
 353         if (gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0) {
 354                 pr_info("%s: can't add ipgre handler\n", __func__);
 355                 goto err_gre;
 356         }
 357
 358         if (gre_offload_init()) {
 359                 pr_err("can't add protocol offload\n");
 360                 goto err_gso;
 361         }
 362
 363         return 0;
 364 err_gso:
 365         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
 366 err_gre:
 367         inet_del_protocol(&net_gre_protocol, IPPROTO_GRE);
 368 err:
 369         return -EAGAIN;
 370 }
 371
 372 static void __exit gre_exit(void)
 373 {
 374         gre_offload_exit();
 375
 376         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
 377         inet_del_protocol(&net_gre_protocol, IPPROTO_GRE);
 378 }
 379
 380 module_init(gre_init);
 381 module_exit(gre_exit);
 382
 383 MODULE_DESCRIPTION("GRE over IPv4 demultiplexer driver");
 384 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
 385 MODULE_LICENSE("GPL");