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 struct sk_buff *gre_handle_offloads(struct sk_buff *skb, bool gre_csum)
  97 {
  98         int err;
  99
 100         if (likely(!skb->encapsulation)) {
 101                 skb_reset_inner_headers(skb);
 102                 skb->encapsulation = 1;
 103         }
 104
 105         if (skb_is_gso(skb)) {
 106                 err = skb_unclone(skb, GFP_ATOMIC);
 107                 if (unlikely(err))
 108                         goto error;
 109                 skb_shinfo(skb)->gso_type |= SKB_GSO_GRE;
 110                 return skb;
 111         } else if (skb->ip_summed == CHECKSUM_PARTIAL && gre_csum) {
 112                 err = skb_checksum_help(skb);
 113                 if (unlikely(err))
 114                         goto error;
 115         } else if (skb->ip_summed != CHECKSUM_PARTIAL)
 116                 skb->ip_summed = CHECKSUM_NONE;
 117
 118         return skb;
 119 error:
 120         kfree_skb(skb);
 121         return ERR_PTR(err);
 122 }
 123 EXPORT_SYMBOL_GPL(gre_handle_offloads);
 124
 125 static __sum16 check_checksum(struct sk_buff *skb)
 126 {
 127         __sum16 csum = 0;
 128
 129         switch (skb->ip_summed) {
 130         case CHECKSUM_COMPLETE:
 131                 csum = csum_fold(skb->csum);
 132
 133                 if (!csum)
 134                         break;
 135                 /* Fall through. */
 136
 137         case CHECKSUM_NONE:
 138                 skb->csum = 0;
 139                 csum = __skb_checksum_complete(skb);
 140                 skb->ip_summed = CHECKSUM_COMPLETE;
 141                 break;
 142         }
 143
 144         return csum;
 145 }
 146
 147 static int parse_gre_header(struct sk_buff *skb, struct tnl_ptk_info *tpi,
 148                             bool *csum_err)
 149 {
 150         unsigned int ip_hlen = ip_hdrlen(skb);
 151         const struct gre_base_hdr *greh;
 152         __be32 *options;
 153         int hdr_len;
 154
 155         if (unlikely(!pskb_may_pull(skb, sizeof(struct gre_base_hdr))))
 156                 return -EINVAL;
 157
 158         greh = (struct gre_base_hdr *)(skb_network_header(skb) + ip_hlen);
 159         if (unlikely(greh->flags & (GRE_VERSION | GRE_ROUTING)))
 160                 return -EINVAL;
 161
 162         tpi->flags = gre_flags_to_tnl_flags(greh->flags);
 163         hdr_len = ip_gre_calc_hlen(tpi->flags);
 164
 165         if (!pskb_may_pull(skb, hdr_len))
 166                 return -EINVAL;
 167
 168         greh = (struct gre_base_hdr *)(skb_network_header(skb) + ip_hlen);
 169         tpi->proto = greh->protocol;
 170
 171         options = (__be32 *)(greh + 1);
 172         if (greh->flags & GRE_CSUM) {
 173                 if (check_checksum(skb)) {
 174                         *csum_err = true;
 175                         return -EINVAL;
 176                 }
 177                 options++;
 178         }
 179
 180         if (greh->flags & GRE_KEY) {
 181                 tpi->key = *options;
 182                 options++;
 183         } else
 184                 tpi->key = 0;
 185
 186         if (unlikely(greh->flags & GRE_SEQ)) {
 187                 tpi->seq = *options;
 188                 options++;
 189         } else
 190                 tpi->seq = 0;
 191
 192         /* WCCP version 1 and 2 protocol decoding.
 193          * - Change protocol to IP
 194          * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
 195          */
 196         if (greh->flags == 0 && tpi->proto == htons(ETH_P_WCCP)) {
 197                 tpi->proto = htons(ETH_P_IP);
 198                 if ((*(u8 *)options & 0xF0) != 0x40) {
 199                         hdr_len += 4;
 200                         if (!pskb_may_pull(skb, hdr_len))
 201                                 return -EINVAL;
 202                 }
 203         }
 204
 205         return iptunnel_pull_header(skb, hdr_len, tpi->proto);
 206 }
 207
 208 static int gre_cisco_rcv(struct sk_buff *skb)
 209 {
 210         struct tnl_ptk_info tpi;
 211         int i;
 212         bool csum_err = false;
 213
 214         if (parse_gre_header(skb, &tpi, &csum_err) < 0)
 215                 goto drop;
 216
 217         rcu_read_lock();
 218         for (i = 0; i < GRE_IP_PROTO_MAX; i++) {
 219                 struct gre_cisco_protocol *proto;
 220                 int ret;
 221
 222                 proto = rcu_dereference(gre_cisco_proto_list[i]);
 223                 if (!proto)
 224                         continue;
 225                 ret = proto->handler(skb, &tpi);
 226                 if (ret == PACKET_RCVD) {
 227                         rcu_read_unlock();
 228                         return 0;
 229                 }
 230         }
 231         rcu_read_unlock();
 232
 233         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 234 drop:
 235         kfree_skb(skb);
 236         return 0;
 237 }
 238
 239 static void gre_cisco_err(struct sk_buff *skb, u32 info)
 240 {
 241         /* All the routers (except for Linux) return only
 242          * 8 bytes of packet payload. It means, that precise relaying of
 243          * ICMP in the real Internet is absolutely infeasible.
 244          *
 245          * Moreover, Cisco "wise men" put GRE key to the third word
 246          * in GRE header. It makes impossible maintaining even soft
 247          * state for keyed
 248          * GRE tunnels with enabled checksum. Tell them "thank you".
 249          *
 250          * Well, I wonder, rfc1812 was written by Cisco employee,
 251          * what the hell these idiots break standards established
 252          * by themselves???
 253          */
 254
 255         const int type = icmp_hdr(skb)->type;
 256         const int code = icmp_hdr(skb)->code;
 257         struct tnl_ptk_info tpi;
 258         bool csum_err = false;
 259         int i;
 260
 261         if (parse_gre_header(skb, &tpi, &csum_err)) {
 262                 if (!csum_err)          /* ignore csum errors. */
 263                         return;
 264         }
 265
 266         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
 267                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
 268                                 skb->dev->ifindex, 0, IPPROTO_GRE, 0);
 269                 return;
 270         }
 271         if (type == ICMP_REDIRECT) {
 272                 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
 273                                 IPPROTO_GRE, 0);
 274                 return;
 275         }
 276
 277         rcu_read_lock();
 278         for (i = 0; i < GRE_IP_PROTO_MAX; i++) {
 279                 struct gre_cisco_protocol *proto;
 280
 281                 proto = rcu_dereference(gre_cisco_proto_list[i]);
 282                 if (!proto)
 283                         continue;
 284
 285                 if (proto->err_handler(skb, info, &tpi) == PACKET_RCVD)
 286                         goto out;
 287
 288         }
 289 out:
 290         rcu_read_unlock();
 291 }
 292
 293 static int gre_rcv(struct sk_buff *skb)
 294 {
 295         const struct gre_protocol *proto;
 296         u8 ver;
 297         int ret;
 298
 299         if (!pskb_may_pull(skb, 12))
 300                 goto drop;
 301
 302         ver = skb->data[1]&0x7f;
 303         if (ver >= GREPROTO_MAX)
 304                 goto drop;
 305
 306         rcu_read_lock();
 307         proto = rcu_dereference(gre_proto[ver]);
 308         if (!proto || !proto->handler)
 309                 goto drop_unlock;
 310         ret = proto->handler(skb);
 311         rcu_read_unlock();
 312         return ret;
 313
 314 drop_unlock:
 315         rcu_read_unlock();
 316 drop:
 317         kfree_skb(skb);
 318         return NET_RX_DROP;
 319 }
 320
 321 static void gre_err(struct sk_buff *skb, u32 info)
 322 {
 323         const struct gre_protocol *proto;
 324         const struct iphdr *iph = (const struct iphdr *)skb->data;
 325         u8 ver = skb->data[(iph->ihl<<2) + 1]&0x7f;
 326
 327         if (ver >= GREPROTO_MAX)
 328                 return;
 329
 330         rcu_read_lock();
 331         proto = rcu_dereference(gre_proto[ver]);
 332         if (proto && proto->err_handler)
 333                 proto->err_handler(skb, info);
 334         rcu_read_unlock();
 335 }
 336
 337 static const struct net_protocol net_gre_protocol = {
 338         .handler     = gre_rcv,
 339         .err_handler = gre_err,
 340         .netns_ok    = 1,
 341 };
 342
 343 static const struct gre_protocol ipgre_protocol = {
 344         .handler     = gre_cisco_rcv,
 345         .err_handler = gre_cisco_err,
 346 };
 347
 348 int gre_cisco_register(struct gre_cisco_protocol *newp)
 349 {
 350         struct gre_cisco_protocol **proto = (struct gre_cisco_protocol **)
 351                                             &gre_cisco_proto_list[newp->priority];
 352
 353         return (cmpxchg(proto, NULL, newp) == NULL) ? 0 : -EBUSY;
 354 }
 355 EXPORT_SYMBOL_GPL(gre_cisco_register);
 356
 357 int gre_cisco_unregister(struct gre_cisco_protocol *del_proto)
 358 {
 359         struct gre_cisco_protocol **proto = (struct gre_cisco_protocol **)
 360                                             &gre_cisco_proto_list[del_proto->priority];
 361         int ret;
 362
 363         ret = (cmpxchg(proto, del_proto, NULL) == del_proto) ? 0 : -EINVAL;
 364
 365         if (ret)
 366                 return ret;
 367
 368         synchronize_net();
 369         return 0;
 370 }
 371 EXPORT_SYMBOL_GPL(gre_cisco_unregister);
 372
 373 static int __init gre_init(void)
 374 {
 375         pr_info("GRE over IPv4 demultiplexor driver\n");
 376
 377         if (inet_add_protocol(&net_gre_protocol, IPPROTO_GRE) < 0) {
 378                 pr_err("can't add protocol\n");
 379                 goto err;
 380         }
 381
 382         if (gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0) {
 383                 pr_info("%s: can't add ipgre handler\n", __func__);
 384                 goto err_gre;
 385         }
 386
 387         if (gre_offload_init()) {
 388                 pr_err("can't add protocol offload\n");
 389                 goto err_gso;
 390         }
 391
 392         return 0;
 393 err_gso:
 394         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
 395 err_gre:
 396         inet_del_protocol(&net_gre_protocol, IPPROTO_GRE);
 397 err:
 398         return -EAGAIN;
 399 }
 400
 401 static void __exit gre_exit(void)
 402 {
 403         gre_offload_exit();
 404
 405         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
 406         inet_del_protocol(&net_gre_protocol, IPPROTO_GRE);
 407 }
 408
 409 module_init(gre_init);
 410 module_exit(gre_exit);
 411
 412 MODULE_DESCRIPTION("GRE over IPv4 demultiplexer driver");
 413 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
 414 MODULE_LICENSE("GPL");