1 /* 2 * Linux NET3: GRE over IP protocol decoder. 3 * 4 * Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.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 #include <linux/capability.h> 14 #include <linux/module.h> 15 #include <linux/types.h> 16 #include <linux/sched.h> 17 #include <linux/kernel.h> 18 #include <asm/uaccess.h> 19 #include <linux/skbuff.h> 20 #include <linux/netdevice.h> 21 #include <linux/in.h> 22 #include <linux/tcp.h> 23 #include <linux/udp.h> 24 #include <linux/if_arp.h> 25 #include <linux/mroute.h> 26 #include <linux/init.h> 27 #include <linux/in6.h> 28 #include <linux/inetdevice.h> 29 #include <linux/igmp.h> 30 #include <linux/netfilter_ipv4.h> 31 #include <linux/if_ether.h> 32 33 #include <net/sock.h> 34 #include <net/ip.h> 35 #include <net/icmp.h> 36 #include <net/protocol.h> 37 #include <net/ipip.h> 38 #include <net/arp.h> 39 #include <net/checksum.h> 40 #include <net/dsfield.h> 41 #include <net/inet_ecn.h> 42 #include <net/xfrm.h> 43 44 #ifdef CONFIG_IPV6 45 #include <net/ipv6.h> 46 #include <net/ip6_fib.h> 47 #include <net/ip6_route.h> 48 #endif 49 50 /* 51 Problems & solutions 52 -------------------- 53 54 1. The most important issue is detecting local dead loops. 55 They would cause complete host lockup in transmit, which 56 would be "resolved" by stack overflow or, if queueing is enabled, 57 with infinite looping in net_bh. 58 59 We cannot track such dead loops during route installation, 60 it is infeasible task. The most general solutions would be 61 to keep skb->encapsulation counter (sort of local ttl), 62 and silently drop packet when it expires. It is the best 63 solution, but it supposes maintaing new variable in ALL 64 skb, even if no tunneling is used. 65 66 Current solution: t->recursion lock breaks dead loops. It looks 67 like dev->tbusy flag, but I preferred new variable, because 68 the semantics is different. One day, when hard_start_xmit 69 will be multithreaded we will have to use skb->encapsulation. 70 71 72 73 2. Networking dead loops would not kill routers, but would really 74 kill network. IP hop limit plays role of "t->recursion" in this case, 75 if we copy it from packet being encapsulated to upper header. 76 It is very good solution, but it introduces two problems: 77 78 - Routing protocols, using packets with ttl=1 (OSPF, RIP2), 79 do not work over tunnels. 80 - traceroute does not work. I planned to relay ICMP from tunnel, 81 so that this problem would be solved and traceroute output 82 would even more informative. This idea appeared to be wrong: 83 only Linux complies to rfc1812 now (yes, guys, Linux is the only 84 true router now :-)), all routers (at least, in neighbourhood of mine) 85 return only 8 bytes of payload. It is the end. 86 87 Hence, if we want that OSPF worked or traceroute said something reasonable, 88 we should search for another solution. 89 90 One of them is to parse packet trying to detect inner encapsulation 91 made by our node. It is difficult or even impossible, especially, 92 taking into account fragmentation. TO be short, tt is not solution at all. 93 94 Current solution: The solution was UNEXPECTEDLY SIMPLE. 95 We force DF flag on tunnels with preconfigured hop limit, 96 that is ALL. :-) Well, it does not remove the problem completely, 97 but exponential growth of network traffic is changed to linear 98 (branches, that exceed pmtu are pruned) and tunnel mtu 99 fastly degrades to value <68, where looping stops. 100 Yes, it is not good if there exists a router in the loop, 101 which does not force DF, even when encapsulating packets have DF set. 102 But it is not our problem! Nobody could accuse us, we made 103 all that we could make. Even if it is your gated who injected 104 fatal route to network, even if it were you who configured 105 fatal static route: you are innocent. :-) 106 107 108 109 3. Really, ipv4/ipip.c, ipv4/ip_gre.c and ipv6/sit.c contain 110 practically identical code. It would be good to glue them 111 together, but it is not very evident, how to make them modular. 112 sit is integral part of IPv6, ipip and gre are naturally modular. 113 We could extract common parts (hash table, ioctl etc) 114 to a separate module (ip_tunnel.c). 115 116 Alexey Kuznetsov. 117 */ 118 119 static int ipgre_tunnel_init(struct net_device *dev); 120 static void ipgre_tunnel_setup(struct net_device *dev); 121 122 /* Fallback tunnel: no source, no destination, no key, no options */ 123 124 static int ipgre_fb_tunnel_init(struct net_device *dev); 125 126 static struct net_device *ipgre_fb_tunnel_dev; 127 128 /* Tunnel hash table */ 129 130 /* 131 4 hash tables: 132 133 3: (remote,local) 134 2: (remote,*) 135 1: (*,local) 136 0: (*,*) 137 138 We require exact key match i.e. if a key is present in packet 139 it will match only tunnel with the same key; if it is not present, 140 it will match only keyless tunnel. 141 142 All keysless packets, if not matched configured keyless tunnels 143 will match fallback tunnel. 144 */ 145 146 #define HASH_SIZE 16 147 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF) 148 149 static struct ip_tunnel *tunnels[4][HASH_SIZE]; 150 151 #define tunnels_r_l (tunnels[3]) 152 #define tunnels_r (tunnels[2]) 153 #define tunnels_l (tunnels[1]) 154 #define tunnels_wc (tunnels[0]) 155 156 static DEFINE_RWLOCK(ipgre_lock); 157 158 /* Given src, dst and key, find appropriate for input tunnel. */ 159 160 static struct ip_tunnel * ipgre_tunnel_lookup(__be32 remote, __be32 local, __be32 key) 161 { 162 unsigned h0 = HASH(remote); 163 unsigned h1 = HASH(key); 164 struct ip_tunnel *t; 165 166 for (t = tunnels_r_l[h0^h1]; t; t = t->next) { 167 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr) { 168 if (t->parms.i_key == key && (t->dev->flags&IFF_UP)) 169 return t; 170 } 171 } 172 for (t = tunnels_r[h0^h1]; t; t = t->next) { 173 if (remote == t->parms.iph.daddr) { 174 if (t->parms.i_key == key && (t->dev->flags&IFF_UP)) 175 return t; 176 } 177 } 178 for (t = tunnels_l[h1]; t; t = t->next) { 179 if (local == t->parms.iph.saddr || 180 (local == t->parms.iph.daddr && MULTICAST(local))) { 181 if (t->parms.i_key == key && (t->dev->flags&IFF_UP)) 182 return t; 183 } 184 } 185 for (t = tunnels_wc[h1]; t; t = t->next) { 186 if (t->parms.i_key == key && (t->dev->flags&IFF_UP)) 187 return t; 188 } 189 190 if (ipgre_fb_tunnel_dev->flags&IFF_UP) 191 return netdev_priv(ipgre_fb_tunnel_dev); 192 return NULL; 193 } 194 195 static struct ip_tunnel **ipgre_bucket(struct ip_tunnel *t) 196 { 197 __be32 remote = t->parms.iph.daddr; 198 __be32 local = t->parms.iph.saddr; 199 __be32 key = t->parms.i_key; 200 unsigned h = HASH(key); 201 int prio = 0; 202 203 if (local) 204 prio |= 1; 205 if (remote && !MULTICAST(remote)) { 206 prio |= 2; 207 h ^= HASH(remote); 208 } 209 210 return &tunnels[prio][h]; 211 } 212 213 static void ipgre_tunnel_link(struct ip_tunnel *t) 214 { 215 struct ip_tunnel **tp = ipgre_bucket(t); 216 217 t->next = *tp; 218 write_lock_bh(&ipgre_lock); 219 *tp = t; 220 write_unlock_bh(&ipgre_lock); 221 } 222 223 static void ipgre_tunnel_unlink(struct ip_tunnel *t) 224 { 225 struct ip_tunnel **tp; 226 227 for (tp = ipgre_bucket(t); *tp; tp = &(*tp)->next) { 228 if (t == *tp) { 229 write_lock_bh(&ipgre_lock); 230 *tp = t->next; 231 write_unlock_bh(&ipgre_lock); 232 break; 233 } 234 } 235 } 236 237 static struct ip_tunnel * ipgre_tunnel_locate(struct ip_tunnel_parm *parms, int create) 238 { 239 __be32 remote = parms->iph.daddr; 240 __be32 local = parms->iph.saddr; 241 __be32 key = parms->i_key; 242 struct ip_tunnel *t, **tp, *nt; 243 struct net_device *dev; 244 unsigned h = HASH(key); 245 int prio = 0; 246 char name[IFNAMSIZ]; 247 248 if (local) 249 prio |= 1; 250 if (remote && !MULTICAST(remote)) { 251 prio |= 2; 252 h ^= HASH(remote); 253 } 254 for (tp = &tunnels[prio][h]; (t = *tp) != NULL; tp = &t->next) { 255 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr) { 256 if (key == t->parms.i_key) 257 return t; 258 } 259 } 260 if (!create) 261 return NULL; 262 263 if (parms->name[0]) 264 strlcpy(name, parms->name, IFNAMSIZ); 265 else { 266 int i; 267 for (i=1; i<100; i++) { 268 sprintf(name, "gre%d", i); 269 if (__dev_get_by_name(name) == NULL) 270 break; 271 } 272 if (i==100) 273 goto failed; 274 } 275 276 dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup); 277 if (!dev) 278 return NULL; 279 280 dev->init = ipgre_tunnel_init; 281 nt = netdev_priv(dev); 282 nt->parms = *parms; 283 284 if (register_netdevice(dev) < 0) { 285 free_netdev(dev); 286 goto failed; 287 } 288 289 dev_hold(dev); 290 ipgre_tunnel_link(nt); 291 return nt; 292 293 failed: 294 return NULL; 295 } 296 297 static void ipgre_tunnel_uninit(struct net_device *dev) 298 { 299 ipgre_tunnel_unlink(netdev_priv(dev)); 300 dev_put(dev); 301 } 302 303 304 static void ipgre_err(struct sk_buff *skb, u32 info) 305 { 306 #ifndef I_WISH_WORLD_WERE_PERFECT 307 308 /* It is not :-( All the routers (except for Linux) return only 309 8 bytes of packet payload. It means, that precise relaying of 310 ICMP in the real Internet is absolutely infeasible. 311 312 Moreover, Cisco "wise men" put GRE key to the third word 313 in GRE header. It makes impossible maintaining even soft state for keyed 314 GRE tunnels with enabled checksum. Tell them "thank you". 315 316 Well, I wonder, rfc1812 was written by Cisco employee, 317 what the hell these idiots break standrads established 318 by themself??? 319 */ 320 321 struct iphdr *iph = (struct iphdr*)skb->data; 322 __be16 *p = (__be16*)(skb->data+(iph->ihl<<2)); 323 int grehlen = (iph->ihl<<2) + 4; 324 int type = skb->h.icmph->type; 325 int code = skb->h.icmph->code; 326 struct ip_tunnel *t; 327 __be16 flags; 328 329 flags = p[0]; 330 if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) { 331 if (flags&(GRE_VERSION|GRE_ROUTING)) 332 return; 333 if (flags&GRE_KEY) { 334 grehlen += 4; 335 if (flags&GRE_CSUM) 336 grehlen += 4; 337 } 338 } 339 340 /* If only 8 bytes returned, keyed message will be dropped here */ 341 if (skb_headlen(skb) < grehlen) 342 return; 343 344 switch (type) { 345 default: 346 case ICMP_PARAMETERPROB: 347 return; 348 349 case ICMP_DEST_UNREACH: 350 switch (code) { 351 case ICMP_SR_FAILED: 352 case ICMP_PORT_UNREACH: 353 /* Impossible event. */ 354 return; 355 case ICMP_FRAG_NEEDED: 356 /* Soft state for pmtu is maintained by IP core. */ 357 return; 358 default: 359 /* All others are translated to HOST_UNREACH. 360 rfc2003 contains "deep thoughts" about NET_UNREACH, 361 I believe they are just ether pollution. --ANK 362 */ 363 break; 364 } 365 break; 366 case ICMP_TIME_EXCEEDED: 367 if (code != ICMP_EXC_TTL) 368 return; 369 break; 370 } 371 372 read_lock(&ipgre_lock); 373 t = ipgre_tunnel_lookup(iph->daddr, iph->saddr, (flags&GRE_KEY) ? *(((__be32*)p) + (grehlen>>2) - 1) : 0); 374 if (t == NULL || t->parms.iph.daddr == 0 || MULTICAST(t->parms.iph.daddr)) 375 goto out; 376 377 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED) 378 goto out; 379 380 if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO) 381 t->err_count++; 382 else 383 t->err_count = 1; 384 t->err_time = jiffies; 385 out: 386 read_unlock(&ipgre_lock); 387 return; 388 #else 389 struct iphdr *iph = (struct iphdr*)dp; 390 struct iphdr *eiph; 391 __be16 *p = (__be16*)(dp+(iph->ihl<<2)); 392 int type = skb->h.icmph->type; 393 int code = skb->h.icmph->code; 394 int rel_type = 0; 395 int rel_code = 0; 396 __be32 rel_info = 0; 397 __u32 n = 0; 398 __be16 flags; 399 int grehlen = (iph->ihl<<2) + 4; 400 struct sk_buff *skb2; 401 struct flowi fl; 402 struct rtable *rt; 403 404 if (p[1] != htons(ETH_P_IP)) 405 return; 406 407 flags = p[0]; 408 if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) { 409 if (flags&(GRE_VERSION|GRE_ROUTING)) 410 return; 411 if (flags&GRE_CSUM) 412 grehlen += 4; 413 if (flags&GRE_KEY) 414 grehlen += 4; 415 if (flags&GRE_SEQ) 416 grehlen += 4; 417 } 418 if (len < grehlen + sizeof(struct iphdr)) 419 return; 420 eiph = (struct iphdr*)(dp + grehlen); 421 422 switch (type) { 423 default: 424 return; 425 case ICMP_PARAMETERPROB: 426 n = ntohl(skb->h.icmph->un.gateway) >> 24; 427 if (n < (iph->ihl<<2)) 428 return; 429 430 /* So... This guy found something strange INSIDE encapsulated 431 packet. Well, he is fool, but what can we do ? 432 */ 433 rel_type = ICMP_PARAMETERPROB; 434 n -= grehlen; 435 rel_info = htonl(n << 24); 436 break; 437 438 case ICMP_DEST_UNREACH: 439 switch (code) { 440 case ICMP_SR_FAILED: 441 case ICMP_PORT_UNREACH: 442 /* Impossible event. */ 443 return; 444 case ICMP_FRAG_NEEDED: 445 /* And it is the only really necessary thing :-) */ 446 n = ntohs(skb->h.icmph->un.frag.mtu); 447 if (n < grehlen+68) 448 return; 449 n -= grehlen; 450 /* BSD 4.2 MORE DOES NOT EXIST IN NATURE. */ 451 if (n > ntohs(eiph->tot_len)) 452 return; 453 rel_info = htonl(n); 454 break; 455 default: 456 /* All others are translated to HOST_UNREACH. 457 rfc2003 contains "deep thoughts" about NET_UNREACH, 458 I believe, it is just ether pollution. --ANK 459 */ 460 rel_type = ICMP_DEST_UNREACH; 461 rel_code = ICMP_HOST_UNREACH; 462 break; 463 } 464 break; 465 case ICMP_TIME_EXCEEDED: 466 if (code != ICMP_EXC_TTL) 467 return; 468 break; 469 } 470 471 /* Prepare fake skb to feed it to icmp_send */ 472 skb2 = skb_clone(skb, GFP_ATOMIC); 473 if (skb2 == NULL) 474 return; 475 dst_release(skb2->dst); 476 skb2->dst = NULL; 477 skb_pull(skb2, skb->data - (u8*)eiph); 478 skb2->nh.raw = skb2->data; 479 480 /* Try to guess incoming interface */ 481 memset(&fl, 0, sizeof(fl)); 482 fl.fl4_dst = eiph->saddr; 483 fl.fl4_tos = RT_TOS(eiph->tos); 484 fl.proto = IPPROTO_GRE; 485 if (ip_route_output_key(&rt, &fl)) { 486 kfree_skb(skb2); 487 return; 488 } 489 skb2->dev = rt->u.dst.dev; 490 491 /* route "incoming" packet */ 492 if (rt->rt_flags&RTCF_LOCAL) { 493 ip_rt_put(rt); 494 rt = NULL; 495 fl.fl4_dst = eiph->daddr; 496 fl.fl4_src = eiph->saddr; 497 fl.fl4_tos = eiph->tos; 498 if (ip_route_output_key(&rt, &fl) || 499 rt->u.dst.dev->type != ARPHRD_IPGRE) { 500 ip_rt_put(rt); 501 kfree_skb(skb2); 502 return; 503 } 504 } else { 505 ip_rt_put(rt); 506 if (ip_route_input(skb2, eiph->daddr, eiph->saddr, eiph->tos, skb2->dev) || 507 skb2->dst->dev->type != ARPHRD_IPGRE) { 508 kfree_skb(skb2); 509 return; 510 } 511 } 512 513 /* change mtu on this route */ 514 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) { 515 if (n > dst_mtu(skb2->dst)) { 516 kfree_skb(skb2); 517 return; 518 } 519 skb2->dst->ops->update_pmtu(skb2->dst, n); 520 } else if (type == ICMP_TIME_EXCEEDED) { 521 struct ip_tunnel *t = netdev_priv(skb2->dev); 522 if (t->parms.iph.ttl) { 523 rel_type = ICMP_DEST_UNREACH; 524 rel_code = ICMP_HOST_UNREACH; 525 } 526 } 527 528 icmp_send(skb2, rel_type, rel_code, rel_info); 529 kfree_skb(skb2); 530 #endif 531 } 532 533 static inline void ipgre_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb) 534 { 535 if (INET_ECN_is_ce(iph->tos)) { 536 if (skb->protocol == htons(ETH_P_IP)) { 537 IP_ECN_set_ce(skb->nh.iph); 538 } else if (skb->protocol == htons(ETH_P_IPV6)) { 539 IP6_ECN_set_ce(skb->nh.ipv6h); 540 } 541 } 542 } 543 544 static inline u8 545 ipgre_ecn_encapsulate(u8 tos, struct iphdr *old_iph, struct sk_buff *skb) 546 { 547 u8 inner = 0; 548 if (skb->protocol == htons(ETH_P_IP)) 549 inner = old_iph->tos; 550 else if (skb->protocol == htons(ETH_P_IPV6)) 551 inner = ipv6_get_dsfield((struct ipv6hdr *)old_iph); 552 return INET_ECN_encapsulate(tos, inner); 553 } 554 555 static int ipgre_rcv(struct sk_buff *skb) 556 { 557 struct iphdr *iph; 558 u8 *h; 559 __be16 flags; 560 __sum16 csum = 0; 561 __be32 key = 0; 562 u32 seqno = 0; 563 struct ip_tunnel *tunnel; 564 int offset = 4; 565 566 if (!pskb_may_pull(skb, 16)) 567 goto drop_nolock; 568 569 iph = skb->nh.iph; 570 h = skb->data; 571 flags = *(__be16*)h; 572 573 if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) { 574 /* - Version must be 0. 575 - We do not support routing headers. 576 */ 577 if (flags&(GRE_VERSION|GRE_ROUTING)) 578 goto drop_nolock; 579 580 if (flags&GRE_CSUM) { 581 switch (skb->ip_summed) { 582 case CHECKSUM_COMPLETE: 583 csum = csum_fold(skb->csum); 584 if (!csum) 585 break; 586 /* fall through */ 587 case CHECKSUM_NONE: 588 skb->csum = 0; 589 csum = __skb_checksum_complete(skb); 590 skb->ip_summed = CHECKSUM_COMPLETE; 591 } 592 offset += 4; 593 } 594 if (flags&GRE_KEY) { 595 key = *(__be32*)(h + offset); 596 offset += 4; 597 } 598 if (flags&GRE_SEQ) { 599 seqno = ntohl(*(__be32*)(h + offset)); 600 offset += 4; 601 } 602 } 603 604 read_lock(&ipgre_lock); 605 if ((tunnel = ipgre_tunnel_lookup(iph->saddr, iph->daddr, key)) != NULL) { 606 secpath_reset(skb); 607 608 skb->protocol = *(__be16*)(h + 2); 609 /* WCCP version 1 and 2 protocol decoding. 610 * - Change protocol to IP 611 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header 612 */ 613 if (flags == 0 && 614 skb->protocol == htons(ETH_P_WCCP)) { 615 skb->protocol = htons(ETH_P_IP); 616 if ((*(h + offset) & 0xF0) != 0x40) 617 offset += 4; 618 } 619 620 skb->mac.raw = skb->nh.raw; 621 skb->nh.raw = __pskb_pull(skb, offset); 622 skb_postpull_rcsum(skb, skb->h.raw, offset); 623 skb->pkt_type = PACKET_HOST; 624 #ifdef CONFIG_NET_IPGRE_BROADCAST 625 if (MULTICAST(iph->daddr)) { 626 /* Looped back packet, drop it! */ 627 if (((struct rtable*)skb->dst)->fl.iif == 0) 628 goto drop; 629 tunnel->stat.multicast++; 630 skb->pkt_type = PACKET_BROADCAST; 631 } 632 #endif 633 634 if (((flags&GRE_CSUM) && csum) || 635 (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) { 636 tunnel->stat.rx_crc_errors++; 637 tunnel->stat.rx_errors++; 638 goto drop; 639 } 640 if (tunnel->parms.i_flags&GRE_SEQ) { 641 if (!(flags&GRE_SEQ) || 642 (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) { 643 tunnel->stat.rx_fifo_errors++; 644 tunnel->stat.rx_errors++; 645 goto drop; 646 } 647 tunnel->i_seqno = seqno + 1; 648 } 649 tunnel->stat.rx_packets++; 650 tunnel->stat.rx_bytes += skb->len; 651 skb->dev = tunnel->dev; 652 dst_release(skb->dst); 653 skb->dst = NULL; 654 nf_reset(skb); 655 ipgre_ecn_decapsulate(iph, skb); 656 netif_rx(skb); 657 read_unlock(&ipgre_lock); 658 return(0); 659 } 660 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 661 662 drop: 663 read_unlock(&ipgre_lock); 664 drop_nolock: 665 kfree_skb(skb); 666 return(0); 667 } 668 669 static int ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev) 670 { 671 struct ip_tunnel *tunnel = netdev_priv(dev); 672 struct net_device_stats *stats = &tunnel->stat; 673 struct iphdr *old_iph = skb->nh.iph; 674 struct iphdr *tiph; 675 u8 tos; 676 __be16 df; 677 struct rtable *rt; /* Route to the other host */ 678 struct net_device *tdev; /* Device to other host */ 679 struct iphdr *iph; /* Our new IP header */ 680 int max_headroom; /* The extra header space needed */ 681 int gre_hlen; 682 __be32 dst; 683 int mtu; 684 685 if (tunnel->recursion++) { 686 tunnel->stat.collisions++; 687 goto tx_error; 688 } 689 690 if (dev->hard_header) { 691 gre_hlen = 0; 692 tiph = (struct iphdr*)skb->data; 693 } else { 694 gre_hlen = tunnel->hlen; 695 tiph = &tunnel->parms.iph; 696 } 697 698 if ((dst = tiph->daddr) == 0) { 699 /* NBMA tunnel */ 700 701 if (skb->dst == NULL) { 702 tunnel->stat.tx_fifo_errors++; 703 goto tx_error; 704 } 705 706 if (skb->protocol == htons(ETH_P_IP)) { 707 rt = (struct rtable*)skb->dst; 708 if ((dst = rt->rt_gateway) == 0) 709 goto tx_error_icmp; 710 } 711 #ifdef CONFIG_IPV6 712 else if (skb->protocol == htons(ETH_P_IPV6)) { 713 struct in6_addr *addr6; 714 int addr_type; 715 struct neighbour *neigh = skb->dst->neighbour; 716 717 if (neigh == NULL) 718 goto tx_error; 719 720 addr6 = (struct in6_addr*)&neigh->primary_key; 721 addr_type = ipv6_addr_type(addr6); 722 723 if (addr_type == IPV6_ADDR_ANY) { 724 addr6 = &skb->nh.ipv6h->daddr; 725 addr_type = ipv6_addr_type(addr6); 726 } 727 728 if ((addr_type & IPV6_ADDR_COMPATv4) == 0) 729 goto tx_error_icmp; 730 731 dst = addr6->s6_addr32[3]; 732 } 733 #endif 734 else 735 goto tx_error; 736 } 737 738 tos = tiph->tos; 739 if (tos&1) { 740 if (skb->protocol == htons(ETH_P_IP)) 741 tos = old_iph->tos; 742 tos &= ~1; 743 } 744 745 { 746 struct flowi fl = { .oif = tunnel->parms.link, 747 .nl_u = { .ip4_u = 748 { .daddr = dst, 749 .saddr = tiph->saddr, 750 .tos = RT_TOS(tos) } }, 751 .proto = IPPROTO_GRE }; 752 if (ip_route_output_key(&rt, &fl)) { 753 tunnel->stat.tx_carrier_errors++; 754 goto tx_error; 755 } 756 } 757 tdev = rt->u.dst.dev; 758 759 if (tdev == dev) { 760 ip_rt_put(rt); 761 tunnel->stat.collisions++; 762 goto tx_error; 763 } 764 765 df = tiph->frag_off; 766 if (df) 767 mtu = dst_mtu(&rt->u.dst) - tunnel->hlen; 768 else 769 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu; 770 771 if (skb->dst) 772 skb->dst->ops->update_pmtu(skb->dst, mtu); 773 774 if (skb->protocol == htons(ETH_P_IP)) { 775 df |= (old_iph->frag_off&htons(IP_DF)); 776 777 if ((old_iph->frag_off&htons(IP_DF)) && 778 mtu < ntohs(old_iph->tot_len)) { 779 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); 780 ip_rt_put(rt); 781 goto tx_error; 782 } 783 } 784 #ifdef CONFIG_IPV6 785 else if (skb->protocol == htons(ETH_P_IPV6)) { 786 struct rt6_info *rt6 = (struct rt6_info*)skb->dst; 787 788 if (rt6 && mtu < dst_mtu(skb->dst) && mtu >= IPV6_MIN_MTU) { 789 if ((tunnel->parms.iph.daddr && !MULTICAST(tunnel->parms.iph.daddr)) || 790 rt6->rt6i_dst.plen == 128) { 791 rt6->rt6i_flags |= RTF_MODIFIED; 792 skb->dst->metrics[RTAX_MTU-1] = mtu; 793 } 794 } 795 796 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) { 797 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev); 798 ip_rt_put(rt); 799 goto tx_error; 800 } 801 } 802 #endif 803 804 if (tunnel->err_count > 0) { 805 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) { 806 tunnel->err_count--; 807 808 dst_link_failure(skb); 809 } else 810 tunnel->err_count = 0; 811 } 812 813 max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen; 814 815 if (skb_headroom(skb) < max_headroom || skb_cloned(skb) || skb_shared(skb)) { 816 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom); 817 if (!new_skb) { 818 ip_rt_put(rt); 819 stats->tx_dropped++; 820 dev_kfree_skb(skb); 821 tunnel->recursion--; 822 return 0; 823 } 824 if (skb->sk) 825 skb_set_owner_w(new_skb, skb->sk); 826 dev_kfree_skb(skb); 827 skb = new_skb; 828 old_iph = skb->nh.iph; 829 } 830 831 skb->h.raw = skb->nh.raw; 832 skb->nh.raw = skb_push(skb, gre_hlen); 833 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 834 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED | 835 IPSKB_REROUTED); 836 dst_release(skb->dst); 837 skb->dst = &rt->u.dst; 838 839 /* 840 * Push down and install the IPIP header. 841 */ 842 843 iph = skb->nh.iph; 844 iph->version = 4; 845 iph->ihl = sizeof(struct iphdr) >> 2; 846 iph->frag_off = df; 847 iph->protocol = IPPROTO_GRE; 848 iph->tos = ipgre_ecn_encapsulate(tos, old_iph, skb); 849 iph->daddr = rt->rt_dst; 850 iph->saddr = rt->rt_src; 851 852 if ((iph->ttl = tiph->ttl) == 0) { 853 if (skb->protocol == htons(ETH_P_IP)) 854 iph->ttl = old_iph->ttl; 855 #ifdef CONFIG_IPV6 856 else if (skb->protocol == htons(ETH_P_IPV6)) 857 iph->ttl = ((struct ipv6hdr*)old_iph)->hop_limit; 858 #endif 859 else 860 iph->ttl = dst_metric(&rt->u.dst, RTAX_HOPLIMIT); 861 } 862 863 ((__be16*)(iph+1))[0] = tunnel->parms.o_flags; 864 ((__be16*)(iph+1))[1] = skb->protocol; 865 866 if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) { 867 __be32 *ptr = (__be32*)(((u8*)iph) + tunnel->hlen - 4); 868 869 if (tunnel->parms.o_flags&GRE_SEQ) { 870 ++tunnel->o_seqno; 871 *ptr = htonl(tunnel->o_seqno); 872 ptr--; 873 } 874 if (tunnel->parms.o_flags&GRE_KEY) { 875 *ptr = tunnel->parms.o_key; 876 ptr--; 877 } 878 if (tunnel->parms.o_flags&GRE_CSUM) { 879 *ptr = 0; 880 *(__sum16*)ptr = ip_compute_csum((void*)(iph+1), skb->len - sizeof(struct iphdr)); 881 } 882 } 883 884 nf_reset(skb); 885 886 IPTUNNEL_XMIT(); 887 tunnel->recursion--; 888 return 0; 889 890 tx_error_icmp: 891 dst_link_failure(skb); 892 893 tx_error: 894 stats->tx_errors++; 895 dev_kfree_skb(skb); 896 tunnel->recursion--; 897 return 0; 898 } 899 900 static int 901 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd) 902 { 903 int err = 0; 904 struct ip_tunnel_parm p; 905 struct ip_tunnel *t; 906 907 switch (cmd) { 908 case SIOCGETTUNNEL: 909 t = NULL; 910 if (dev == ipgre_fb_tunnel_dev) { 911 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) { 912 err = -EFAULT; 913 break; 914 } 915 t = ipgre_tunnel_locate(&p, 0); 916 } 917 if (t == NULL) 918 t = netdev_priv(dev); 919 memcpy(&p, &t->parms, sizeof(p)); 920 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p))) 921 err = -EFAULT; 922 break; 923 924 case SIOCADDTUNNEL: 925 case SIOCCHGTUNNEL: 926 err = -EPERM; 927 if (!capable(CAP_NET_ADMIN)) 928 goto done; 929 930 err = -EFAULT; 931 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) 932 goto done; 933 934 err = -EINVAL; 935 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE || 936 p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) || 937 ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))) 938 goto done; 939 if (p.iph.ttl) 940 p.iph.frag_off |= htons(IP_DF); 941 942 if (!(p.i_flags&GRE_KEY)) 943 p.i_key = 0; 944 if (!(p.o_flags&GRE_KEY)) 945 p.o_key = 0; 946 947 t = ipgre_tunnel_locate(&p, cmd == SIOCADDTUNNEL); 948 949 if (dev != ipgre_fb_tunnel_dev && cmd == SIOCCHGTUNNEL) { 950 if (t != NULL) { 951 if (t->dev != dev) { 952 err = -EEXIST; 953 break; 954 } 955 } else { 956 unsigned nflags=0; 957 958 t = netdev_priv(dev); 959 960 if (MULTICAST(p.iph.daddr)) 961 nflags = IFF_BROADCAST; 962 else if (p.iph.daddr) 963 nflags = IFF_POINTOPOINT; 964 965 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) { 966 err = -EINVAL; 967 break; 968 } 969 ipgre_tunnel_unlink(t); 970 t->parms.iph.saddr = p.iph.saddr; 971 t->parms.iph.daddr = p.iph.daddr; 972 t->parms.i_key = p.i_key; 973 t->parms.o_key = p.o_key; 974 memcpy(dev->dev_addr, &p.iph.saddr, 4); 975 memcpy(dev->broadcast, &p.iph.daddr, 4); 976 ipgre_tunnel_link(t); 977 netdev_state_change(dev); 978 } 979 } 980 981 if (t) { 982 err = 0; 983 if (cmd == SIOCCHGTUNNEL) { 984 t->parms.iph.ttl = p.iph.ttl; 985 t->parms.iph.tos = p.iph.tos; 986 t->parms.iph.frag_off = p.iph.frag_off; 987 } 988 if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p))) 989 err = -EFAULT; 990 } else 991 err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT); 992 break; 993 994 case SIOCDELTUNNEL: 995 err = -EPERM; 996 if (!capable(CAP_NET_ADMIN)) 997 goto done; 998 999 if (dev == ipgre_fb_tunnel_dev) { 1000 err = -EFAULT; 1001 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) 1002 goto done; 1003 err = -ENOENT; 1004 if ((t = ipgre_tunnel_locate(&p, 0)) == NULL) 1005 goto done; 1006 err = -EPERM; 1007 if (t == netdev_priv(ipgre_fb_tunnel_dev)) 1008 goto done; 1009 dev = t->dev; 1010 } 1011 err = unregister_netdevice(dev); 1012 break; 1013 1014 default: 1015 err = -EINVAL; 1016 } 1017 1018 done: 1019 return err; 1020 } 1021 1022 static struct net_device_stats *ipgre_tunnel_get_stats(struct net_device *dev) 1023 { 1024 return &(((struct ip_tunnel*)netdev_priv(dev))->stat); 1025 } 1026 1027 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu) 1028 { 1029 struct ip_tunnel *tunnel = netdev_priv(dev); 1030 if (new_mtu < 68 || new_mtu > 0xFFF8 - tunnel->hlen) 1031 return -EINVAL; 1032 dev->mtu = new_mtu; 1033 return 0; 1034 } 1035 1036 #ifdef CONFIG_NET_IPGRE_BROADCAST 1037 /* Nice toy. Unfortunately, useless in real life :-) 1038 It allows to construct virtual multiprotocol broadcast "LAN" 1039 over the Internet, provided multicast routing is tuned. 1040 1041 1042 I have no idea was this bicycle invented before me, 1043 so that I had to set ARPHRD_IPGRE to a random value. 1044 I have an impression, that Cisco could make something similar, 1045 but this feature is apparently missing in IOS<=11.2(8). 1046 1047 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks 1048 with broadcast 224.66.66.66. If you have access to mbone, play with me :-) 1049 1050 ping -t 255 224.66.66.66 1051 1052 If nobody answers, mbone does not work. 1053 1054 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255 1055 ip addr add 10.66.66.<somewhat>/24 dev Universe 1056 ifconfig Universe up 1057 ifconfig Universe add fe80::<Your_real_addr>/10 1058 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96 1059 ftp 10.66.66.66 1060 ... 1061 ftp fec0:6666:6666::193.233.7.65 1062 ... 1063 1064 */ 1065 1066 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, unsigned short type, 1067 void *daddr, void *saddr, unsigned len) 1068 { 1069 struct ip_tunnel *t = netdev_priv(dev); 1070 struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen); 1071 __be16 *p = (__be16*)(iph+1); 1072 1073 memcpy(iph, &t->parms.iph, sizeof(struct iphdr)); 1074 p[0] = t->parms.o_flags; 1075 p[1] = htons(type); 1076 1077 /* 1078 * Set the source hardware address. 1079 */ 1080 1081 if (saddr) 1082 memcpy(&iph->saddr, saddr, 4); 1083 1084 if (daddr) { 1085 memcpy(&iph->daddr, daddr, 4); 1086 return t->hlen; 1087 } 1088 if (iph->daddr && !MULTICAST(iph->daddr)) 1089 return t->hlen; 1090 1091 return -t->hlen; 1092 } 1093 1094 static int ipgre_open(struct net_device *dev) 1095 { 1096 struct ip_tunnel *t = netdev_priv(dev); 1097 1098 if (MULTICAST(t->parms.iph.daddr)) { 1099 struct flowi fl = { .oif = t->parms.link, 1100 .nl_u = { .ip4_u = 1101 { .daddr = t->parms.iph.daddr, 1102 .saddr = t->parms.iph.saddr, 1103 .tos = RT_TOS(t->parms.iph.tos) } }, 1104 .proto = IPPROTO_GRE }; 1105 struct rtable *rt; 1106 if (ip_route_output_key(&rt, &fl)) 1107 return -EADDRNOTAVAIL; 1108 dev = rt->u.dst.dev; 1109 ip_rt_put(rt); 1110 if (__in_dev_get_rtnl(dev) == NULL) 1111 return -EADDRNOTAVAIL; 1112 t->mlink = dev->ifindex; 1113 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr); 1114 } 1115 return 0; 1116 } 1117 1118 static int ipgre_close(struct net_device *dev) 1119 { 1120 struct ip_tunnel *t = netdev_priv(dev); 1121 if (MULTICAST(t->parms.iph.daddr) && t->mlink) { 1122 struct in_device *in_dev = inetdev_by_index(t->mlink); 1123 if (in_dev) { 1124 ip_mc_dec_group(in_dev, t->parms.iph.daddr); 1125 in_dev_put(in_dev); 1126 } 1127 } 1128 return 0; 1129 } 1130 1131 #endif 1132 1133 static void ipgre_tunnel_setup(struct net_device *dev) 1134 { 1135 SET_MODULE_OWNER(dev); 1136 dev->uninit = ipgre_tunnel_uninit; 1137 dev->destructor = free_netdev; 1138 dev->hard_start_xmit = ipgre_tunnel_xmit; 1139 dev->get_stats = ipgre_tunnel_get_stats; 1140 dev->do_ioctl = ipgre_tunnel_ioctl; 1141 dev->change_mtu = ipgre_tunnel_change_mtu; 1142 1143 dev->type = ARPHRD_IPGRE; 1144 dev->hard_header_len = LL_MAX_HEADER + sizeof(struct iphdr) + 4; 1145 dev->mtu = ETH_DATA_LEN - sizeof(struct iphdr) - 4; 1146 dev->flags = IFF_NOARP; 1147 dev->iflink = 0; 1148 dev->addr_len = 4; 1149 } 1150 1151 static int ipgre_tunnel_init(struct net_device *dev) 1152 { 1153 struct net_device *tdev = NULL; 1154 struct ip_tunnel *tunnel; 1155 struct iphdr *iph; 1156 int hlen = LL_MAX_HEADER; 1157 int mtu = ETH_DATA_LEN; 1158 int addend = sizeof(struct iphdr) + 4; 1159 1160 tunnel = netdev_priv(dev); 1161 iph = &tunnel->parms.iph; 1162 1163 tunnel->dev = dev; 1164 strcpy(tunnel->parms.name, dev->name); 1165 1166 memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4); 1167 memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4); 1168 1169 /* Guess output device to choose reasonable mtu and hard_header_len */ 1170 1171 if (iph->daddr) { 1172 struct flowi fl = { .oif = tunnel->parms.link, 1173 .nl_u = { .ip4_u = 1174 { .daddr = iph->daddr, 1175 .saddr = iph->saddr, 1176 .tos = RT_TOS(iph->tos) } }, 1177 .proto = IPPROTO_GRE }; 1178 struct rtable *rt; 1179 if (!ip_route_output_key(&rt, &fl)) { 1180 tdev = rt->u.dst.dev; 1181 ip_rt_put(rt); 1182 } 1183 1184 dev->flags |= IFF_POINTOPOINT; 1185 1186 #ifdef CONFIG_NET_IPGRE_BROADCAST 1187 if (MULTICAST(iph->daddr)) { 1188 if (!iph->saddr) 1189 return -EINVAL; 1190 dev->flags = IFF_BROADCAST; 1191 dev->hard_header = ipgre_header; 1192 dev->open = ipgre_open; 1193 dev->stop = ipgre_close; 1194 } 1195 #endif 1196 } 1197 1198 if (!tdev && tunnel->parms.link) 1199 tdev = __dev_get_by_index(tunnel->parms.link); 1200 1201 if (tdev) { 1202 hlen = tdev->hard_header_len; 1203 mtu = tdev->mtu; 1204 } 1205 dev->iflink = tunnel->parms.link; 1206 1207 /* Precalculate GRE options length */ 1208 if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) { 1209 if (tunnel->parms.o_flags&GRE_CSUM) 1210 addend += 4; 1211 if (tunnel->parms.o_flags&GRE_KEY) 1212 addend += 4; 1213 if (tunnel->parms.o_flags&GRE_SEQ) 1214 addend += 4; 1215 } 1216 dev->hard_header_len = hlen + addend; 1217 dev->mtu = mtu - addend; 1218 tunnel->hlen = addend; 1219 return 0; 1220 } 1221 1222 static int __init ipgre_fb_tunnel_init(struct net_device *dev) 1223 { 1224 struct ip_tunnel *tunnel = netdev_priv(dev); 1225 struct iphdr *iph = &tunnel->parms.iph; 1226 1227 tunnel->dev = dev; 1228 strcpy(tunnel->parms.name, dev->name); 1229 1230 iph->version = 4; 1231 iph->protocol = IPPROTO_GRE; 1232 iph->ihl = 5; 1233 tunnel->hlen = sizeof(struct iphdr) + 4; 1234 1235 dev_hold(dev); 1236 tunnels_wc[0] = tunnel; 1237 return 0; 1238 } 1239 1240 1241 static struct net_protocol ipgre_protocol = { 1242 .handler = ipgre_rcv, 1243 .err_handler = ipgre_err, 1244 }; 1245 1246 1247 /* 1248 * And now the modules code and kernel interface. 1249 */ 1250 1251 static int __init ipgre_init(void) 1252 { 1253 int err; 1254 1255 printk(KERN_INFO "GRE over IPv4 tunneling driver\n"); 1256 1257 if (inet_add_protocol(&ipgre_protocol, IPPROTO_GRE) < 0) { 1258 printk(KERN_INFO "ipgre init: can't add protocol\n"); 1259 return -EAGAIN; 1260 } 1261 1262 ipgre_fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0", 1263 ipgre_tunnel_setup); 1264 if (!ipgre_fb_tunnel_dev) { 1265 err = -ENOMEM; 1266 goto err1; 1267 } 1268 1269 ipgre_fb_tunnel_dev->init = ipgre_fb_tunnel_init; 1270 1271 if ((err = register_netdev(ipgre_fb_tunnel_dev))) 1272 goto err2; 1273 out: 1274 return err; 1275 err2: 1276 free_netdev(ipgre_fb_tunnel_dev); 1277 err1: 1278 inet_del_protocol(&ipgre_protocol, IPPROTO_GRE); 1279 goto out; 1280 } 1281 1282 static void __exit ipgre_destroy_tunnels(void) 1283 { 1284 int prio; 1285 1286 for (prio = 0; prio < 4; prio++) { 1287 int h; 1288 for (h = 0; h < HASH_SIZE; h++) { 1289 struct ip_tunnel *t; 1290 while ((t = tunnels[prio][h]) != NULL) 1291 unregister_netdevice(t->dev); 1292 } 1293 } 1294 } 1295 1296 static void __exit ipgre_fini(void) 1297 { 1298 if (inet_del_protocol(&ipgre_protocol, IPPROTO_GRE) < 0) 1299 printk(KERN_INFO "ipgre close: can't remove protocol\n"); 1300 1301 rtnl_lock(); 1302 ipgre_destroy_tunnels(); 1303 rtnl_unlock(); 1304 } 1305 1306 module_init(ipgre_init); 1307 module_exit(ipgre_fini); 1308 MODULE_LICENSE("GPL"); 1309