1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Linux NET3: GRE over IP protocol decoder. 4 * 5 * Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru) 6 */ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/capability.h> 11 #include <linux/module.h> 12 #include <linux/types.h> 13 #include <linux/kernel.h> 14 #include <linux/slab.h> 15 #include <linux/uaccess.h> 16 #include <linux/skbuff.h> 17 #include <linux/netdevice.h> 18 #include <linux/in.h> 19 #include <linux/tcp.h> 20 #include <linux/udp.h> 21 #include <linux/if_arp.h> 22 #include <linux/if_vlan.h> 23 #include <linux/init.h> 24 #include <linux/in6.h> 25 #include <linux/inetdevice.h> 26 #include <linux/igmp.h> 27 #include <linux/netfilter_ipv4.h> 28 #include <linux/etherdevice.h> 29 #include <linux/if_ether.h> 30 31 #include <net/sock.h> 32 #include <net/ip.h> 33 #include <net/icmp.h> 34 #include <net/protocol.h> 35 #include <net/ip_tunnels.h> 36 #include <net/arp.h> 37 #include <net/checksum.h> 38 #include <net/dsfield.h> 39 #include <net/inet_ecn.h> 40 #include <net/xfrm.h> 41 #include <net/net_namespace.h> 42 #include <net/netns/generic.h> 43 #include <net/rtnetlink.h> 44 #include <net/gre.h> 45 #include <net/dst_metadata.h> 46 #include <net/erspan.h> 47 48 /* 49 Problems & solutions 50 -------------------- 51 52 1. The most important issue is detecting local dead loops. 53 They would cause complete host lockup in transmit, which 54 would be "resolved" by stack overflow or, if queueing is enabled, 55 with infinite looping in net_bh. 56 57 We cannot track such dead loops during route installation, 58 it is infeasible task. The most general solutions would be 59 to keep skb->encapsulation counter (sort of local ttl), 60 and silently drop packet when it expires. It is a good 61 solution, but it supposes maintaining new variable in ALL 62 skb, even if no tunneling is used. 63 64 Current solution: xmit_recursion breaks dead loops. This is a percpu 65 counter, since when we enter the first ndo_xmit(), cpu migration is 66 forbidden. We force an exit if this counter reaches RECURSION_LIMIT 67 68 2. Networking dead loops would not kill routers, but would really 69 kill network. IP hop limit plays role of "t->recursion" in this case, 70 if we copy it from packet being encapsulated to upper header. 71 It is very good solution, but it introduces two problems: 72 73 - Routing protocols, using packets with ttl=1 (OSPF, RIP2), 74 do not work over tunnels. 75 - traceroute does not work. I planned to relay ICMP from tunnel, 76 so that this problem would be solved and traceroute output 77 would even more informative. This idea appeared to be wrong: 78 only Linux complies to rfc1812 now (yes, guys, Linux is the only 79 true router now :-)), all routers (at least, in neighbourhood of mine) 80 return only 8 bytes of payload. It is the end. 81 82 Hence, if we want that OSPF worked or traceroute said something reasonable, 83 we should search for another solution. 84 85 One of them is to parse packet trying to detect inner encapsulation 86 made by our node. It is difficult or even impossible, especially, 87 taking into account fragmentation. TO be short, ttl is not solution at all. 88 89 Current solution: The solution was UNEXPECTEDLY SIMPLE. 90 We force DF flag on tunnels with preconfigured hop limit, 91 that is ALL. :-) Well, it does not remove the problem completely, 92 but exponential growth of network traffic is changed to linear 93 (branches, that exceed pmtu are pruned) and tunnel mtu 94 rapidly degrades to value <68, where looping stops. 95 Yes, it is not good if there exists a router in the loop, 96 which does not force DF, even when encapsulating packets have DF set. 97 But it is not our problem! Nobody could accuse us, we made 98 all that we could make. Even if it is your gated who injected 99 fatal route to network, even if it were you who configured 100 fatal static route: you are innocent. :-) 101 102 Alexey Kuznetsov. 103 */ 104 105 static bool log_ecn_error = true; 106 module_param(log_ecn_error, bool, 0644); 107 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); 108 109 static struct rtnl_link_ops ipgre_link_ops __read_mostly; 110 static const struct header_ops ipgre_header_ops; 111 112 static int ipgre_tunnel_init(struct net_device *dev); 113 static void erspan_build_header(struct sk_buff *skb, 114 u32 id, u32 index, 115 bool truncate, bool is_ipv4); 116 117 static unsigned int ipgre_net_id __read_mostly; 118 static unsigned int gre_tap_net_id __read_mostly; 119 static unsigned int erspan_net_id __read_mostly; 120 121 static int ipgre_err(struct sk_buff *skb, u32 info, 122 const struct tnl_ptk_info *tpi) 123 { 124 125 /* All the routers (except for Linux) return only 126 8 bytes of packet payload. It means, that precise relaying of 127 ICMP in the real Internet is absolutely infeasible. 128 129 Moreover, Cisco "wise men" put GRE key to the third word 130 in GRE header. It makes impossible maintaining even soft 131 state for keyed GRE tunnels with enabled checksum. Tell 132 them "thank you". 133 134 Well, I wonder, rfc1812 was written by Cisco employee, 135 what the hell these idiots break standards established 136 by themselves??? 137 */ 138 struct net *net = dev_net(skb->dev); 139 struct ip_tunnel_net *itn; 140 const struct iphdr *iph; 141 const int type = icmp_hdr(skb)->type; 142 const int code = icmp_hdr(skb)->code; 143 unsigned int data_len = 0; 144 struct ip_tunnel *t; 145 146 if (tpi->proto == htons(ETH_P_TEB)) 147 itn = net_generic(net, gre_tap_net_id); 148 else if (tpi->proto == htons(ETH_P_ERSPAN) || 149 tpi->proto == htons(ETH_P_ERSPAN2)) 150 itn = net_generic(net, erspan_net_id); 151 else 152 itn = net_generic(net, ipgre_net_id); 153 154 iph = (const struct iphdr *)(icmp_hdr(skb) + 1); 155 t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags, 156 iph->daddr, iph->saddr, tpi->key); 157 158 if (!t) 159 return -ENOENT; 160 161 switch (type) { 162 default: 163 case ICMP_PARAMETERPROB: 164 return 0; 165 166 case ICMP_DEST_UNREACH: 167 switch (code) { 168 case ICMP_SR_FAILED: 169 case ICMP_PORT_UNREACH: 170 /* Impossible event. */ 171 return 0; 172 default: 173 /* All others are translated to HOST_UNREACH. 174 rfc2003 contains "deep thoughts" about NET_UNREACH, 175 I believe they are just ether pollution. --ANK 176 */ 177 break; 178 } 179 break; 180 181 case ICMP_TIME_EXCEEDED: 182 if (code != ICMP_EXC_TTL) 183 return 0; 184 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */ 185 break; 186 187 case ICMP_REDIRECT: 188 break; 189 } 190 191 #if IS_ENABLED(CONFIG_IPV6) 192 if (tpi->proto == htons(ETH_P_IPV6) && 193 !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len, 194 type, data_len)) 195 return 0; 196 #endif 197 198 if (t->parms.iph.daddr == 0 || 199 ipv4_is_multicast(t->parms.iph.daddr)) 200 return 0; 201 202 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED) 203 return 0; 204 205 if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO)) 206 t->err_count++; 207 else 208 t->err_count = 1; 209 t->err_time = jiffies; 210 211 return 0; 212 } 213 214 static void gre_err(struct sk_buff *skb, u32 info) 215 { 216 /* All the routers (except for Linux) return only 217 * 8 bytes of packet payload. It means, that precise relaying of 218 * ICMP in the real Internet is absolutely infeasible. 219 * 220 * Moreover, Cisco "wise men" put GRE key to the third word 221 * in GRE header. It makes impossible maintaining even soft 222 * state for keyed 223 * GRE tunnels with enabled checksum. Tell them "thank you". 224 * 225 * Well, I wonder, rfc1812 was written by Cisco employee, 226 * what the hell these idiots break standards established 227 * by themselves??? 228 */ 229 230 const struct iphdr *iph = (struct iphdr *)skb->data; 231 const int type = icmp_hdr(skb)->type; 232 const int code = icmp_hdr(skb)->code; 233 struct tnl_ptk_info tpi; 234 235 if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP), 236 iph->ihl * 4) < 0) 237 return; 238 239 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) { 240 ipv4_update_pmtu(skb, dev_net(skb->dev), info, 241 skb->dev->ifindex, IPPROTO_GRE); 242 return; 243 } 244 if (type == ICMP_REDIRECT) { 245 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 246 IPPROTO_GRE); 247 return; 248 } 249 250 ipgre_err(skb, info, &tpi); 251 } 252 253 static bool is_erspan_type1(int gre_hdr_len) 254 { 255 /* Both ERSPAN type I (version 0) and type II (version 1) use 256 * protocol 0x88BE, but the type I has only 4-byte GRE header, 257 * while type II has 8-byte. 258 */ 259 return gre_hdr_len == 4; 260 } 261 262 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi, 263 int gre_hdr_len) 264 { 265 struct net *net = dev_net(skb->dev); 266 struct metadata_dst *tun_dst = NULL; 267 struct erspan_base_hdr *ershdr; 268 struct ip_tunnel_net *itn; 269 struct ip_tunnel *tunnel; 270 const struct iphdr *iph; 271 struct erspan_md2 *md2; 272 int ver; 273 int len; 274 275 itn = net_generic(net, erspan_net_id); 276 iph = ip_hdr(skb); 277 if (is_erspan_type1(gre_hdr_len)) { 278 ver = 0; 279 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, 280 tpi->flags | TUNNEL_NO_KEY, 281 iph->saddr, iph->daddr, 0); 282 } else { 283 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len); 284 ver = ershdr->ver; 285 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, 286 tpi->flags | TUNNEL_KEY, 287 iph->saddr, iph->daddr, tpi->key); 288 } 289 290 if (tunnel) { 291 if (is_erspan_type1(gre_hdr_len)) 292 len = gre_hdr_len; 293 else 294 len = gre_hdr_len + erspan_hdr_len(ver); 295 296 if (unlikely(!pskb_may_pull(skb, len))) 297 return PACKET_REJECT; 298 299 if (__iptunnel_pull_header(skb, 300 len, 301 htons(ETH_P_TEB), 302 false, false) < 0) 303 goto drop; 304 305 if (tunnel->collect_md) { 306 struct erspan_metadata *pkt_md, *md; 307 struct ip_tunnel_info *info; 308 unsigned char *gh; 309 __be64 tun_id; 310 __be16 flags; 311 312 tpi->flags |= TUNNEL_KEY; 313 flags = tpi->flags; 314 tun_id = key32_to_tunnel_id(tpi->key); 315 316 tun_dst = ip_tun_rx_dst(skb, flags, 317 tun_id, sizeof(*md)); 318 if (!tun_dst) 319 return PACKET_REJECT; 320 321 /* skb can be uncloned in __iptunnel_pull_header, so 322 * old pkt_md is no longer valid and we need to reset 323 * it 324 */ 325 gh = skb_network_header(skb) + 326 skb_network_header_len(skb); 327 pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len + 328 sizeof(*ershdr)); 329 md = ip_tunnel_info_opts(&tun_dst->u.tun_info); 330 md->version = ver; 331 md2 = &md->u.md2; 332 memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE : 333 ERSPAN_V2_MDSIZE); 334 335 info = &tun_dst->u.tun_info; 336 info->key.tun_flags |= TUNNEL_ERSPAN_OPT; 337 info->options_len = sizeof(*md); 338 } 339 340 skb_reset_mac_header(skb); 341 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error); 342 return PACKET_RCVD; 343 } 344 return PACKET_REJECT; 345 346 drop: 347 kfree_skb(skb); 348 return PACKET_RCVD; 349 } 350 351 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi, 352 struct ip_tunnel_net *itn, int hdr_len, bool raw_proto) 353 { 354 struct metadata_dst *tun_dst = NULL; 355 const struct iphdr *iph; 356 struct ip_tunnel *tunnel; 357 358 iph = ip_hdr(skb); 359 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags, 360 iph->saddr, iph->daddr, tpi->key); 361 362 if (tunnel) { 363 const struct iphdr *tnl_params; 364 365 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto, 366 raw_proto, false) < 0) 367 goto drop; 368 369 /* Special case for ipgre_header_parse(), which expects the 370 * mac_header to point to the outer IP header. 371 */ 372 if (tunnel->dev->header_ops == &ipgre_header_ops) 373 skb_pop_mac_header(skb); 374 else 375 skb_reset_mac_header(skb); 376 377 tnl_params = &tunnel->parms.iph; 378 if (tunnel->collect_md || tnl_params->daddr == 0) { 379 __be16 flags; 380 __be64 tun_id; 381 382 flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY); 383 tun_id = key32_to_tunnel_id(tpi->key); 384 tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0); 385 if (!tun_dst) 386 return PACKET_REJECT; 387 } 388 389 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error); 390 return PACKET_RCVD; 391 } 392 return PACKET_NEXT; 393 394 drop: 395 kfree_skb(skb); 396 return PACKET_RCVD; 397 } 398 399 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi, 400 int hdr_len) 401 { 402 struct net *net = dev_net(skb->dev); 403 struct ip_tunnel_net *itn; 404 int res; 405 406 if (tpi->proto == htons(ETH_P_TEB)) 407 itn = net_generic(net, gre_tap_net_id); 408 else 409 itn = net_generic(net, ipgre_net_id); 410 411 res = __ipgre_rcv(skb, tpi, itn, hdr_len, false); 412 if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) { 413 /* ipgre tunnels in collect metadata mode should receive 414 * also ETH_P_TEB traffic. 415 */ 416 itn = net_generic(net, ipgre_net_id); 417 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true); 418 } 419 return res; 420 } 421 422 static int gre_rcv(struct sk_buff *skb) 423 { 424 struct tnl_ptk_info tpi; 425 bool csum_err = false; 426 int hdr_len; 427 428 #ifdef CONFIG_NET_IPGRE_BROADCAST 429 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) { 430 /* Looped back packet, drop it! */ 431 if (rt_is_output_route(skb_rtable(skb))) 432 goto drop; 433 } 434 #endif 435 436 hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0); 437 if (hdr_len < 0) 438 goto drop; 439 440 if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) || 441 tpi.proto == htons(ETH_P_ERSPAN2))) { 442 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD) 443 return 0; 444 goto out; 445 } 446 447 if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD) 448 return 0; 449 450 out: 451 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 452 drop: 453 kfree_skb(skb); 454 return 0; 455 } 456 457 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev, 458 const struct iphdr *tnl_params, 459 __be16 proto) 460 { 461 struct ip_tunnel *tunnel = netdev_priv(dev); 462 __be16 flags = tunnel->parms.o_flags; 463 464 /* Push GRE header. */ 465 gre_build_header(skb, tunnel->tun_hlen, 466 flags, proto, tunnel->parms.o_key, 467 (flags & TUNNEL_SEQ) ? htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0); 468 469 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol); 470 } 471 472 static int gre_handle_offloads(struct sk_buff *skb, bool csum) 473 { 474 return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE); 475 } 476 477 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev, 478 __be16 proto) 479 { 480 struct ip_tunnel *tunnel = netdev_priv(dev); 481 struct ip_tunnel_info *tun_info; 482 const struct ip_tunnel_key *key; 483 int tunnel_hlen; 484 __be16 flags; 485 486 tun_info = skb_tunnel_info(skb); 487 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 488 ip_tunnel_info_af(tun_info) != AF_INET)) 489 goto err_free_skb; 490 491 key = &tun_info->key; 492 tunnel_hlen = gre_calc_hlen(key->tun_flags); 493 494 if (skb_cow_head(skb, dev->needed_headroom)) 495 goto err_free_skb; 496 497 /* Push Tunnel header. */ 498 if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM))) 499 goto err_free_skb; 500 501 flags = tun_info->key.tun_flags & 502 (TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ); 503 gre_build_header(skb, tunnel_hlen, flags, proto, 504 tunnel_id_to_key32(tun_info->key.tun_id), 505 (flags & TUNNEL_SEQ) ? htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0); 506 507 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen); 508 509 return; 510 511 err_free_skb: 512 kfree_skb(skb); 513 DEV_STATS_INC(dev, tx_dropped); 514 } 515 516 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev) 517 { 518 struct ip_tunnel *tunnel = netdev_priv(dev); 519 struct ip_tunnel_info *tun_info; 520 const struct ip_tunnel_key *key; 521 struct erspan_metadata *md; 522 bool truncate = false; 523 __be16 proto; 524 int tunnel_hlen; 525 int version; 526 int nhoff; 527 528 tun_info = skb_tunnel_info(skb); 529 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 530 ip_tunnel_info_af(tun_info) != AF_INET)) 531 goto err_free_skb; 532 533 key = &tun_info->key; 534 if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT)) 535 goto err_free_skb; 536 if (tun_info->options_len < sizeof(*md)) 537 goto err_free_skb; 538 md = ip_tunnel_info_opts(tun_info); 539 540 /* ERSPAN has fixed 8 byte GRE header */ 541 version = md->version; 542 tunnel_hlen = 8 + erspan_hdr_len(version); 543 544 if (skb_cow_head(skb, dev->needed_headroom)) 545 goto err_free_skb; 546 547 if (gre_handle_offloads(skb, false)) 548 goto err_free_skb; 549 550 if (skb->len > dev->mtu + dev->hard_header_len) { 551 if (pskb_trim(skb, dev->mtu + dev->hard_header_len)) 552 goto err_free_skb; 553 truncate = true; 554 } 555 556 nhoff = skb_network_offset(skb); 557 if (skb->protocol == htons(ETH_P_IP) && 558 (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff)) 559 truncate = true; 560 561 if (skb->protocol == htons(ETH_P_IPV6)) { 562 int thoff; 563 564 if (skb_transport_header_was_set(skb)) 565 thoff = skb_transport_offset(skb); 566 else 567 thoff = nhoff + sizeof(struct ipv6hdr); 568 if (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff) 569 truncate = true; 570 } 571 572 if (version == 1) { 573 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)), 574 ntohl(md->u.index), truncate, true); 575 proto = htons(ETH_P_ERSPAN); 576 } else if (version == 2) { 577 erspan_build_header_v2(skb, 578 ntohl(tunnel_id_to_key32(key->tun_id)), 579 md->u.md2.dir, 580 get_hwid(&md->u.md2), 581 truncate, true); 582 proto = htons(ETH_P_ERSPAN2); 583 } else { 584 goto err_free_skb; 585 } 586 587 gre_build_header(skb, 8, TUNNEL_SEQ, 588 proto, 0, htonl(atomic_fetch_inc(&tunnel->o_seqno))); 589 590 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen); 591 592 return; 593 594 err_free_skb: 595 kfree_skb(skb); 596 DEV_STATS_INC(dev, tx_dropped); 597 } 598 599 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 600 { 601 struct ip_tunnel_info *info = skb_tunnel_info(skb); 602 const struct ip_tunnel_key *key; 603 struct rtable *rt; 604 struct flowi4 fl4; 605 606 if (ip_tunnel_info_af(info) != AF_INET) 607 return -EINVAL; 608 609 key = &info->key; 610 ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src, 611 tunnel_id_to_key32(key->tun_id), 612 key->tos & ~INET_ECN_MASK, dev_net(dev), 0, 613 skb->mark, skb_get_hash(skb), key->flow_flags); 614 rt = ip_route_output_key(dev_net(dev), &fl4); 615 if (IS_ERR(rt)) 616 return PTR_ERR(rt); 617 618 ip_rt_put(rt); 619 info->key.u.ipv4.src = fl4.saddr; 620 return 0; 621 } 622 623 static netdev_tx_t ipgre_xmit(struct sk_buff *skb, 624 struct net_device *dev) 625 { 626 struct ip_tunnel *tunnel = netdev_priv(dev); 627 const struct iphdr *tnl_params; 628 629 if (!pskb_inet_may_pull(skb)) 630 goto free_skb; 631 632 if (tunnel->collect_md) { 633 gre_fb_xmit(skb, dev, skb->protocol); 634 return NETDEV_TX_OK; 635 } 636 637 if (dev->header_ops) { 638 int pull_len = tunnel->hlen + sizeof(struct iphdr); 639 640 if (skb_cow_head(skb, 0)) 641 goto free_skb; 642 643 tnl_params = (const struct iphdr *)skb->data; 644 645 if (!pskb_network_may_pull(skb, pull_len)) 646 goto free_skb; 647 648 /* ip_tunnel_xmit() needs skb->data pointing to gre header. */ 649 skb_pull(skb, pull_len); 650 skb_reset_mac_header(skb); 651 652 if (skb->ip_summed == CHECKSUM_PARTIAL && 653 skb_checksum_start(skb) < skb->data) 654 goto free_skb; 655 } else { 656 if (skb_cow_head(skb, dev->needed_headroom)) 657 goto free_skb; 658 659 tnl_params = &tunnel->parms.iph; 660 } 661 662 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 663 goto free_skb; 664 665 __gre_xmit(skb, dev, tnl_params, skb->protocol); 666 return NETDEV_TX_OK; 667 668 free_skb: 669 kfree_skb(skb); 670 DEV_STATS_INC(dev, tx_dropped); 671 return NETDEV_TX_OK; 672 } 673 674 static netdev_tx_t erspan_xmit(struct sk_buff *skb, 675 struct net_device *dev) 676 { 677 struct ip_tunnel *tunnel = netdev_priv(dev); 678 bool truncate = false; 679 __be16 proto; 680 681 if (!pskb_inet_may_pull(skb)) 682 goto free_skb; 683 684 if (tunnel->collect_md) { 685 erspan_fb_xmit(skb, dev); 686 return NETDEV_TX_OK; 687 } 688 689 if (gre_handle_offloads(skb, false)) 690 goto free_skb; 691 692 if (skb_cow_head(skb, dev->needed_headroom)) 693 goto free_skb; 694 695 if (skb->len > dev->mtu + dev->hard_header_len) { 696 if (pskb_trim(skb, dev->mtu + dev->hard_header_len)) 697 goto free_skb; 698 truncate = true; 699 } 700 701 /* Push ERSPAN header */ 702 if (tunnel->erspan_ver == 0) { 703 proto = htons(ETH_P_ERSPAN); 704 tunnel->parms.o_flags &= ~TUNNEL_SEQ; 705 } else if (tunnel->erspan_ver == 1) { 706 erspan_build_header(skb, ntohl(tunnel->parms.o_key), 707 tunnel->index, 708 truncate, true); 709 proto = htons(ETH_P_ERSPAN); 710 } else if (tunnel->erspan_ver == 2) { 711 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key), 712 tunnel->dir, tunnel->hwid, 713 truncate, true); 714 proto = htons(ETH_P_ERSPAN2); 715 } else { 716 goto free_skb; 717 } 718 719 tunnel->parms.o_flags &= ~TUNNEL_KEY; 720 __gre_xmit(skb, dev, &tunnel->parms.iph, proto); 721 return NETDEV_TX_OK; 722 723 free_skb: 724 kfree_skb(skb); 725 DEV_STATS_INC(dev, tx_dropped); 726 return NETDEV_TX_OK; 727 } 728 729 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb, 730 struct net_device *dev) 731 { 732 struct ip_tunnel *tunnel = netdev_priv(dev); 733 734 if (!pskb_inet_may_pull(skb)) 735 goto free_skb; 736 737 if (tunnel->collect_md) { 738 gre_fb_xmit(skb, dev, htons(ETH_P_TEB)); 739 return NETDEV_TX_OK; 740 } 741 742 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 743 goto free_skb; 744 745 if (skb_cow_head(skb, dev->needed_headroom)) 746 goto free_skb; 747 748 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB)); 749 return NETDEV_TX_OK; 750 751 free_skb: 752 kfree_skb(skb); 753 DEV_STATS_INC(dev, tx_dropped); 754 return NETDEV_TX_OK; 755 } 756 757 static void ipgre_link_update(struct net_device *dev, bool set_mtu) 758 { 759 struct ip_tunnel *tunnel = netdev_priv(dev); 760 __be16 flags; 761 int len; 762 763 len = tunnel->tun_hlen; 764 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 765 len = tunnel->tun_hlen - len; 766 tunnel->hlen = tunnel->hlen + len; 767 768 if (dev->header_ops) 769 dev->hard_header_len += len; 770 else 771 dev->needed_headroom += len; 772 773 if (set_mtu) 774 dev->mtu = max_t(int, dev->mtu - len, 68); 775 776 flags = tunnel->parms.o_flags; 777 778 if (flags & TUNNEL_SEQ || 779 (flags & TUNNEL_CSUM && tunnel->encap.type != TUNNEL_ENCAP_NONE)) { 780 dev->features &= ~NETIF_F_GSO_SOFTWARE; 781 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 782 } else { 783 dev->features |= NETIF_F_GSO_SOFTWARE; 784 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 785 } 786 } 787 788 static int ipgre_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, 789 int cmd) 790 { 791 int err; 792 793 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) { 794 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE || 795 p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) || 796 ((p->i_flags | p->o_flags) & (GRE_VERSION | GRE_ROUTING))) 797 return -EINVAL; 798 } 799 800 p->i_flags = gre_flags_to_tnl_flags(p->i_flags); 801 p->o_flags = gre_flags_to_tnl_flags(p->o_flags); 802 803 err = ip_tunnel_ctl(dev, p, cmd); 804 if (err) 805 return err; 806 807 if (cmd == SIOCCHGTUNNEL) { 808 struct ip_tunnel *t = netdev_priv(dev); 809 810 t->parms.i_flags = p->i_flags; 811 t->parms.o_flags = p->o_flags; 812 813 if (strcmp(dev->rtnl_link_ops->kind, "erspan")) 814 ipgre_link_update(dev, true); 815 } 816 817 p->i_flags = gre_tnl_flags_to_gre_flags(p->i_flags); 818 p->o_flags = gre_tnl_flags_to_gre_flags(p->o_flags); 819 return 0; 820 } 821 822 /* Nice toy. Unfortunately, useless in real life :-) 823 It allows to construct virtual multiprotocol broadcast "LAN" 824 over the Internet, provided multicast routing is tuned. 825 826 827 I have no idea was this bicycle invented before me, 828 so that I had to set ARPHRD_IPGRE to a random value. 829 I have an impression, that Cisco could make something similar, 830 but this feature is apparently missing in IOS<=11.2(8). 831 832 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks 833 with broadcast 224.66.66.66. If you have access to mbone, play with me :-) 834 835 ping -t 255 224.66.66.66 836 837 If nobody answers, mbone does not work. 838 839 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255 840 ip addr add 10.66.66.<somewhat>/24 dev Universe 841 ifconfig Universe up 842 ifconfig Universe add fe80::<Your_real_addr>/10 843 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96 844 ftp 10.66.66.66 845 ... 846 ftp fec0:6666:6666::193.233.7.65 847 ... 848 */ 849 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, 850 unsigned short type, 851 const void *daddr, const void *saddr, unsigned int len) 852 { 853 struct ip_tunnel *t = netdev_priv(dev); 854 struct iphdr *iph; 855 struct gre_base_hdr *greh; 856 857 iph = skb_push(skb, t->hlen + sizeof(*iph)); 858 greh = (struct gre_base_hdr *)(iph+1); 859 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags); 860 greh->protocol = htons(type); 861 862 memcpy(iph, &t->parms.iph, sizeof(struct iphdr)); 863 864 /* Set the source hardware address. */ 865 if (saddr) 866 memcpy(&iph->saddr, saddr, 4); 867 if (daddr) 868 memcpy(&iph->daddr, daddr, 4); 869 if (iph->daddr) 870 return t->hlen + sizeof(*iph); 871 872 return -(t->hlen + sizeof(*iph)); 873 } 874 875 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr) 876 { 877 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb); 878 memcpy(haddr, &iph->saddr, 4); 879 return 4; 880 } 881 882 static const struct header_ops ipgre_header_ops = { 883 .create = ipgre_header, 884 .parse = ipgre_header_parse, 885 }; 886 887 #ifdef CONFIG_NET_IPGRE_BROADCAST 888 static int ipgre_open(struct net_device *dev) 889 { 890 struct ip_tunnel *t = netdev_priv(dev); 891 892 if (ipv4_is_multicast(t->parms.iph.daddr)) { 893 struct flowi4 fl4; 894 struct rtable *rt; 895 896 rt = ip_route_output_gre(t->net, &fl4, 897 t->parms.iph.daddr, 898 t->parms.iph.saddr, 899 t->parms.o_key, 900 RT_TOS(t->parms.iph.tos), 901 t->parms.link); 902 if (IS_ERR(rt)) 903 return -EADDRNOTAVAIL; 904 dev = rt->dst.dev; 905 ip_rt_put(rt); 906 if (!__in_dev_get_rtnl(dev)) 907 return -EADDRNOTAVAIL; 908 t->mlink = dev->ifindex; 909 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr); 910 } 911 return 0; 912 } 913 914 static int ipgre_close(struct net_device *dev) 915 { 916 struct ip_tunnel *t = netdev_priv(dev); 917 918 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) { 919 struct in_device *in_dev; 920 in_dev = inetdev_by_index(t->net, t->mlink); 921 if (in_dev) 922 ip_mc_dec_group(in_dev, t->parms.iph.daddr); 923 } 924 return 0; 925 } 926 #endif 927 928 static const struct net_device_ops ipgre_netdev_ops = { 929 .ndo_init = ipgre_tunnel_init, 930 .ndo_uninit = ip_tunnel_uninit, 931 #ifdef CONFIG_NET_IPGRE_BROADCAST 932 .ndo_open = ipgre_open, 933 .ndo_stop = ipgre_close, 934 #endif 935 .ndo_start_xmit = ipgre_xmit, 936 .ndo_siocdevprivate = ip_tunnel_siocdevprivate, 937 .ndo_change_mtu = ip_tunnel_change_mtu, 938 .ndo_get_stats64 = dev_get_tstats64, 939 .ndo_get_iflink = ip_tunnel_get_iflink, 940 .ndo_tunnel_ctl = ipgre_tunnel_ctl, 941 }; 942 943 #define GRE_FEATURES (NETIF_F_SG | \ 944 NETIF_F_FRAGLIST | \ 945 NETIF_F_HIGHDMA | \ 946 NETIF_F_HW_CSUM) 947 948 static void ipgre_tunnel_setup(struct net_device *dev) 949 { 950 dev->netdev_ops = &ipgre_netdev_ops; 951 dev->type = ARPHRD_IPGRE; 952 ip_tunnel_setup(dev, ipgre_net_id); 953 } 954 955 static void __gre_tunnel_init(struct net_device *dev) 956 { 957 struct ip_tunnel *tunnel; 958 __be16 flags; 959 960 tunnel = netdev_priv(dev); 961 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 962 tunnel->parms.iph.protocol = IPPROTO_GRE; 963 964 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen; 965 dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph); 966 967 dev->features |= GRE_FEATURES | NETIF_F_LLTX; 968 dev->hw_features |= GRE_FEATURES; 969 970 flags = tunnel->parms.o_flags; 971 972 /* TCP offload with GRE SEQ is not supported, nor can we support 2 973 * levels of outer headers requiring an update. 974 */ 975 if (flags & TUNNEL_SEQ) 976 return; 977 if (flags & TUNNEL_CSUM && tunnel->encap.type != TUNNEL_ENCAP_NONE) 978 return; 979 980 dev->features |= NETIF_F_GSO_SOFTWARE; 981 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 982 } 983 984 static int ipgre_tunnel_init(struct net_device *dev) 985 { 986 struct ip_tunnel *tunnel = netdev_priv(dev); 987 struct iphdr *iph = &tunnel->parms.iph; 988 989 __gre_tunnel_init(dev); 990 991 __dev_addr_set(dev, &iph->saddr, 4); 992 memcpy(dev->broadcast, &iph->daddr, 4); 993 994 dev->flags = IFF_NOARP; 995 netif_keep_dst(dev); 996 dev->addr_len = 4; 997 998 if (iph->daddr && !tunnel->collect_md) { 999 #ifdef CONFIG_NET_IPGRE_BROADCAST 1000 if (ipv4_is_multicast(iph->daddr)) { 1001 if (!iph->saddr) 1002 return -EINVAL; 1003 dev->flags = IFF_BROADCAST; 1004 dev->header_ops = &ipgre_header_ops; 1005 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 1006 dev->needed_headroom = 0; 1007 } 1008 #endif 1009 } else if (!tunnel->collect_md) { 1010 dev->header_ops = &ipgre_header_ops; 1011 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 1012 dev->needed_headroom = 0; 1013 } 1014 1015 return ip_tunnel_init(dev); 1016 } 1017 1018 static const struct gre_protocol ipgre_protocol = { 1019 .handler = gre_rcv, 1020 .err_handler = gre_err, 1021 }; 1022 1023 static int __net_init ipgre_init_net(struct net *net) 1024 { 1025 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL); 1026 } 1027 1028 static void __net_exit ipgre_exit_batch_rtnl(struct list_head *list_net, 1029 struct list_head *dev_to_kill) 1030 { 1031 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops, 1032 dev_to_kill); 1033 } 1034 1035 static struct pernet_operations ipgre_net_ops = { 1036 .init = ipgre_init_net, 1037 .exit_batch_rtnl = ipgre_exit_batch_rtnl, 1038 .id = &ipgre_net_id, 1039 .size = sizeof(struct ip_tunnel_net), 1040 }; 1041 1042 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[], 1043 struct netlink_ext_ack *extack) 1044 { 1045 __be16 flags; 1046 1047 if (!data) 1048 return 0; 1049 1050 flags = 0; 1051 if (data[IFLA_GRE_IFLAGS]) 1052 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1053 if (data[IFLA_GRE_OFLAGS]) 1054 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1055 if (flags & (GRE_VERSION|GRE_ROUTING)) 1056 return -EINVAL; 1057 1058 if (data[IFLA_GRE_COLLECT_METADATA] && 1059 data[IFLA_GRE_ENCAP_TYPE] && 1060 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE) 1061 return -EINVAL; 1062 1063 return 0; 1064 } 1065 1066 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[], 1067 struct netlink_ext_ack *extack) 1068 { 1069 __be32 daddr; 1070 1071 if (tb[IFLA_ADDRESS]) { 1072 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1073 return -EINVAL; 1074 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1075 return -EADDRNOTAVAIL; 1076 } 1077 1078 if (!data) 1079 goto out; 1080 1081 if (data[IFLA_GRE_REMOTE]) { 1082 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4); 1083 if (!daddr) 1084 return -EINVAL; 1085 } 1086 1087 out: 1088 return ipgre_tunnel_validate(tb, data, extack); 1089 } 1090 1091 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[], 1092 struct netlink_ext_ack *extack) 1093 { 1094 __be16 flags = 0; 1095 int ret; 1096 1097 if (!data) 1098 return 0; 1099 1100 ret = ipgre_tap_validate(tb, data, extack); 1101 if (ret) 1102 return ret; 1103 1104 if (data[IFLA_GRE_ERSPAN_VER] && 1105 nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0) 1106 return 0; 1107 1108 /* ERSPAN type II/III should only have GRE sequence and key flag */ 1109 if (data[IFLA_GRE_OFLAGS]) 1110 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1111 if (data[IFLA_GRE_IFLAGS]) 1112 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1113 if (!data[IFLA_GRE_COLLECT_METADATA] && 1114 flags != (GRE_SEQ | GRE_KEY)) 1115 return -EINVAL; 1116 1117 /* ERSPAN Session ID only has 10-bit. Since we reuse 1118 * 32-bit key field as ID, check it's range. 1119 */ 1120 if (data[IFLA_GRE_IKEY] && 1121 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK)) 1122 return -EINVAL; 1123 1124 if (data[IFLA_GRE_OKEY] && 1125 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK)) 1126 return -EINVAL; 1127 1128 return 0; 1129 } 1130 1131 static int ipgre_netlink_parms(struct net_device *dev, 1132 struct nlattr *data[], 1133 struct nlattr *tb[], 1134 struct ip_tunnel_parm *parms, 1135 __u32 *fwmark) 1136 { 1137 struct ip_tunnel *t = netdev_priv(dev); 1138 1139 memset(parms, 0, sizeof(*parms)); 1140 1141 parms->iph.protocol = IPPROTO_GRE; 1142 1143 if (!data) 1144 return 0; 1145 1146 if (data[IFLA_GRE_LINK]) 1147 parms->link = nla_get_u32(data[IFLA_GRE_LINK]); 1148 1149 if (data[IFLA_GRE_IFLAGS]) 1150 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS])); 1151 1152 if (data[IFLA_GRE_OFLAGS]) 1153 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS])); 1154 1155 if (data[IFLA_GRE_IKEY]) 1156 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]); 1157 1158 if (data[IFLA_GRE_OKEY]) 1159 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]); 1160 1161 if (data[IFLA_GRE_LOCAL]) 1162 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]); 1163 1164 if (data[IFLA_GRE_REMOTE]) 1165 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]); 1166 1167 if (data[IFLA_GRE_TTL]) 1168 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]); 1169 1170 if (data[IFLA_GRE_TOS]) 1171 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]); 1172 1173 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) { 1174 if (t->ignore_df) 1175 return -EINVAL; 1176 parms->iph.frag_off = htons(IP_DF); 1177 } 1178 1179 if (data[IFLA_GRE_COLLECT_METADATA]) { 1180 t->collect_md = true; 1181 if (dev->type == ARPHRD_IPGRE) 1182 dev->type = ARPHRD_NONE; 1183 } 1184 1185 if (data[IFLA_GRE_IGNORE_DF]) { 1186 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF]) 1187 && (parms->iph.frag_off & htons(IP_DF))) 1188 return -EINVAL; 1189 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]); 1190 } 1191 1192 if (data[IFLA_GRE_FWMARK]) 1193 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]); 1194 1195 return 0; 1196 } 1197 1198 static int erspan_netlink_parms(struct net_device *dev, 1199 struct nlattr *data[], 1200 struct nlattr *tb[], 1201 struct ip_tunnel_parm *parms, 1202 __u32 *fwmark) 1203 { 1204 struct ip_tunnel *t = netdev_priv(dev); 1205 int err; 1206 1207 err = ipgre_netlink_parms(dev, data, tb, parms, fwmark); 1208 if (err) 1209 return err; 1210 if (!data) 1211 return 0; 1212 1213 if (data[IFLA_GRE_ERSPAN_VER]) { 1214 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]); 1215 1216 if (t->erspan_ver > 2) 1217 return -EINVAL; 1218 } 1219 1220 if (t->erspan_ver == 1) { 1221 if (data[IFLA_GRE_ERSPAN_INDEX]) { 1222 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]); 1223 if (t->index & ~INDEX_MASK) 1224 return -EINVAL; 1225 } 1226 } else if (t->erspan_ver == 2) { 1227 if (data[IFLA_GRE_ERSPAN_DIR]) { 1228 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]); 1229 if (t->dir & ~(DIR_MASK >> DIR_OFFSET)) 1230 return -EINVAL; 1231 } 1232 if (data[IFLA_GRE_ERSPAN_HWID]) { 1233 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]); 1234 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET)) 1235 return -EINVAL; 1236 } 1237 } 1238 1239 return 0; 1240 } 1241 1242 /* This function returns true when ENCAP attributes are present in the nl msg */ 1243 static bool ipgre_netlink_encap_parms(struct nlattr *data[], 1244 struct ip_tunnel_encap *ipencap) 1245 { 1246 bool ret = false; 1247 1248 memset(ipencap, 0, sizeof(*ipencap)); 1249 1250 if (!data) 1251 return ret; 1252 1253 if (data[IFLA_GRE_ENCAP_TYPE]) { 1254 ret = true; 1255 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]); 1256 } 1257 1258 if (data[IFLA_GRE_ENCAP_FLAGS]) { 1259 ret = true; 1260 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]); 1261 } 1262 1263 if (data[IFLA_GRE_ENCAP_SPORT]) { 1264 ret = true; 1265 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]); 1266 } 1267 1268 if (data[IFLA_GRE_ENCAP_DPORT]) { 1269 ret = true; 1270 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]); 1271 } 1272 1273 return ret; 1274 } 1275 1276 static int gre_tap_init(struct net_device *dev) 1277 { 1278 __gre_tunnel_init(dev); 1279 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1280 netif_keep_dst(dev); 1281 1282 return ip_tunnel_init(dev); 1283 } 1284 1285 static const struct net_device_ops gre_tap_netdev_ops = { 1286 .ndo_init = gre_tap_init, 1287 .ndo_uninit = ip_tunnel_uninit, 1288 .ndo_start_xmit = gre_tap_xmit, 1289 .ndo_set_mac_address = eth_mac_addr, 1290 .ndo_validate_addr = eth_validate_addr, 1291 .ndo_change_mtu = ip_tunnel_change_mtu, 1292 .ndo_get_stats64 = dev_get_tstats64, 1293 .ndo_get_iflink = ip_tunnel_get_iflink, 1294 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1295 }; 1296 1297 static int erspan_tunnel_init(struct net_device *dev) 1298 { 1299 struct ip_tunnel *tunnel = netdev_priv(dev); 1300 1301 if (tunnel->erspan_ver == 0) 1302 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */ 1303 else 1304 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */ 1305 1306 tunnel->parms.iph.protocol = IPPROTO_GRE; 1307 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen + 1308 erspan_hdr_len(tunnel->erspan_ver); 1309 1310 dev->features |= GRE_FEATURES; 1311 dev->hw_features |= GRE_FEATURES; 1312 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1313 netif_keep_dst(dev); 1314 1315 return ip_tunnel_init(dev); 1316 } 1317 1318 static const struct net_device_ops erspan_netdev_ops = { 1319 .ndo_init = erspan_tunnel_init, 1320 .ndo_uninit = ip_tunnel_uninit, 1321 .ndo_start_xmit = erspan_xmit, 1322 .ndo_set_mac_address = eth_mac_addr, 1323 .ndo_validate_addr = eth_validate_addr, 1324 .ndo_change_mtu = ip_tunnel_change_mtu, 1325 .ndo_get_stats64 = dev_get_tstats64, 1326 .ndo_get_iflink = ip_tunnel_get_iflink, 1327 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1328 }; 1329 1330 static void ipgre_tap_setup(struct net_device *dev) 1331 { 1332 ether_setup(dev); 1333 dev->max_mtu = 0; 1334 dev->netdev_ops = &gre_tap_netdev_ops; 1335 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1336 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1337 ip_tunnel_setup(dev, gre_tap_net_id); 1338 } 1339 1340 static int 1341 ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[]) 1342 { 1343 struct ip_tunnel_encap ipencap; 1344 1345 if (ipgre_netlink_encap_parms(data, &ipencap)) { 1346 struct ip_tunnel *t = netdev_priv(dev); 1347 int err = ip_tunnel_encap_setup(t, &ipencap); 1348 1349 if (err < 0) 1350 return err; 1351 } 1352 1353 return 0; 1354 } 1355 1356 static int ipgre_newlink(struct net *src_net, struct net_device *dev, 1357 struct nlattr *tb[], struct nlattr *data[], 1358 struct netlink_ext_ack *extack) 1359 { 1360 struct ip_tunnel_parm p; 1361 __u32 fwmark = 0; 1362 int err; 1363 1364 err = ipgre_newlink_encap_setup(dev, data); 1365 if (err) 1366 return err; 1367 1368 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1369 if (err < 0) 1370 return err; 1371 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1372 } 1373 1374 static int erspan_newlink(struct net *src_net, struct net_device *dev, 1375 struct nlattr *tb[], struct nlattr *data[], 1376 struct netlink_ext_ack *extack) 1377 { 1378 struct ip_tunnel_parm p; 1379 __u32 fwmark = 0; 1380 int err; 1381 1382 err = ipgre_newlink_encap_setup(dev, data); 1383 if (err) 1384 return err; 1385 1386 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1387 if (err) 1388 return err; 1389 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1390 } 1391 1392 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[], 1393 struct nlattr *data[], 1394 struct netlink_ext_ack *extack) 1395 { 1396 struct ip_tunnel *t = netdev_priv(dev); 1397 __u32 fwmark = t->fwmark; 1398 struct ip_tunnel_parm p; 1399 int err; 1400 1401 err = ipgre_newlink_encap_setup(dev, data); 1402 if (err) 1403 return err; 1404 1405 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1406 if (err < 0) 1407 return err; 1408 1409 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1410 if (err < 0) 1411 return err; 1412 1413 t->parms.i_flags = p.i_flags; 1414 t->parms.o_flags = p.o_flags; 1415 1416 ipgre_link_update(dev, !tb[IFLA_MTU]); 1417 1418 return 0; 1419 } 1420 1421 static int erspan_changelink(struct net_device *dev, struct nlattr *tb[], 1422 struct nlattr *data[], 1423 struct netlink_ext_ack *extack) 1424 { 1425 struct ip_tunnel *t = netdev_priv(dev); 1426 __u32 fwmark = t->fwmark; 1427 struct ip_tunnel_parm p; 1428 int err; 1429 1430 err = ipgre_newlink_encap_setup(dev, data); 1431 if (err) 1432 return err; 1433 1434 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1435 if (err < 0) 1436 return err; 1437 1438 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1439 if (err < 0) 1440 return err; 1441 1442 t->parms.i_flags = p.i_flags; 1443 t->parms.o_flags = p.o_flags; 1444 1445 return 0; 1446 } 1447 1448 static size_t ipgre_get_size(const struct net_device *dev) 1449 { 1450 return 1451 /* IFLA_GRE_LINK */ 1452 nla_total_size(4) + 1453 /* IFLA_GRE_IFLAGS */ 1454 nla_total_size(2) + 1455 /* IFLA_GRE_OFLAGS */ 1456 nla_total_size(2) + 1457 /* IFLA_GRE_IKEY */ 1458 nla_total_size(4) + 1459 /* IFLA_GRE_OKEY */ 1460 nla_total_size(4) + 1461 /* IFLA_GRE_LOCAL */ 1462 nla_total_size(4) + 1463 /* IFLA_GRE_REMOTE */ 1464 nla_total_size(4) + 1465 /* IFLA_GRE_TTL */ 1466 nla_total_size(1) + 1467 /* IFLA_GRE_TOS */ 1468 nla_total_size(1) + 1469 /* IFLA_GRE_PMTUDISC */ 1470 nla_total_size(1) + 1471 /* IFLA_GRE_ENCAP_TYPE */ 1472 nla_total_size(2) + 1473 /* IFLA_GRE_ENCAP_FLAGS */ 1474 nla_total_size(2) + 1475 /* IFLA_GRE_ENCAP_SPORT */ 1476 nla_total_size(2) + 1477 /* IFLA_GRE_ENCAP_DPORT */ 1478 nla_total_size(2) + 1479 /* IFLA_GRE_COLLECT_METADATA */ 1480 nla_total_size(0) + 1481 /* IFLA_GRE_IGNORE_DF */ 1482 nla_total_size(1) + 1483 /* IFLA_GRE_FWMARK */ 1484 nla_total_size(4) + 1485 /* IFLA_GRE_ERSPAN_INDEX */ 1486 nla_total_size(4) + 1487 /* IFLA_GRE_ERSPAN_VER */ 1488 nla_total_size(1) + 1489 /* IFLA_GRE_ERSPAN_DIR */ 1490 nla_total_size(1) + 1491 /* IFLA_GRE_ERSPAN_HWID */ 1492 nla_total_size(2) + 1493 0; 1494 } 1495 1496 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev) 1497 { 1498 struct ip_tunnel *t = netdev_priv(dev); 1499 struct ip_tunnel_parm *p = &t->parms; 1500 __be16 o_flags = p->o_flags; 1501 1502 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) || 1503 nla_put_be16(skb, IFLA_GRE_IFLAGS, 1504 gre_tnl_flags_to_gre_flags(p->i_flags)) || 1505 nla_put_be16(skb, IFLA_GRE_OFLAGS, 1506 gre_tnl_flags_to_gre_flags(o_flags)) || 1507 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) || 1508 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) || 1509 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) || 1510 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) || 1511 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) || 1512 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) || 1513 nla_put_u8(skb, IFLA_GRE_PMTUDISC, 1514 !!(p->iph.frag_off & htons(IP_DF))) || 1515 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark)) 1516 goto nla_put_failure; 1517 1518 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE, 1519 t->encap.type) || 1520 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT, 1521 t->encap.sport) || 1522 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT, 1523 t->encap.dport) || 1524 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS, 1525 t->encap.flags)) 1526 goto nla_put_failure; 1527 1528 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df)) 1529 goto nla_put_failure; 1530 1531 if (t->collect_md) { 1532 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA)) 1533 goto nla_put_failure; 1534 } 1535 1536 return 0; 1537 1538 nla_put_failure: 1539 return -EMSGSIZE; 1540 } 1541 1542 static int erspan_fill_info(struct sk_buff *skb, const struct net_device *dev) 1543 { 1544 struct ip_tunnel *t = netdev_priv(dev); 1545 1546 if (t->erspan_ver <= 2) { 1547 if (t->erspan_ver != 0 && !t->collect_md) 1548 t->parms.o_flags |= TUNNEL_KEY; 1549 1550 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver)) 1551 goto nla_put_failure; 1552 1553 if (t->erspan_ver == 1) { 1554 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index)) 1555 goto nla_put_failure; 1556 } else if (t->erspan_ver == 2) { 1557 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir)) 1558 goto nla_put_failure; 1559 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid)) 1560 goto nla_put_failure; 1561 } 1562 } 1563 1564 return ipgre_fill_info(skb, dev); 1565 1566 nla_put_failure: 1567 return -EMSGSIZE; 1568 } 1569 1570 static void erspan_setup(struct net_device *dev) 1571 { 1572 struct ip_tunnel *t = netdev_priv(dev); 1573 1574 ether_setup(dev); 1575 dev->max_mtu = 0; 1576 dev->netdev_ops = &erspan_netdev_ops; 1577 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1578 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1579 ip_tunnel_setup(dev, erspan_net_id); 1580 t->erspan_ver = 1; 1581 } 1582 1583 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = { 1584 [IFLA_GRE_LINK] = { .type = NLA_U32 }, 1585 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 }, 1586 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 }, 1587 [IFLA_GRE_IKEY] = { .type = NLA_U32 }, 1588 [IFLA_GRE_OKEY] = { .type = NLA_U32 }, 1589 [IFLA_GRE_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 1590 [IFLA_GRE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1591 [IFLA_GRE_TTL] = { .type = NLA_U8 }, 1592 [IFLA_GRE_TOS] = { .type = NLA_U8 }, 1593 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 }, 1594 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 }, 1595 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 }, 1596 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 }, 1597 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 }, 1598 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1599 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 }, 1600 [IFLA_GRE_FWMARK] = { .type = NLA_U32 }, 1601 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 }, 1602 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 }, 1603 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 }, 1604 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 }, 1605 }; 1606 1607 static struct rtnl_link_ops ipgre_link_ops __read_mostly = { 1608 .kind = "gre", 1609 .maxtype = IFLA_GRE_MAX, 1610 .policy = ipgre_policy, 1611 .priv_size = sizeof(struct ip_tunnel), 1612 .setup = ipgre_tunnel_setup, 1613 .validate = ipgre_tunnel_validate, 1614 .newlink = ipgre_newlink, 1615 .changelink = ipgre_changelink, 1616 .dellink = ip_tunnel_dellink, 1617 .get_size = ipgre_get_size, 1618 .fill_info = ipgre_fill_info, 1619 .get_link_net = ip_tunnel_get_link_net, 1620 }; 1621 1622 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = { 1623 .kind = "gretap", 1624 .maxtype = IFLA_GRE_MAX, 1625 .policy = ipgre_policy, 1626 .priv_size = sizeof(struct ip_tunnel), 1627 .setup = ipgre_tap_setup, 1628 .validate = ipgre_tap_validate, 1629 .newlink = ipgre_newlink, 1630 .changelink = ipgre_changelink, 1631 .dellink = ip_tunnel_dellink, 1632 .get_size = ipgre_get_size, 1633 .fill_info = ipgre_fill_info, 1634 .get_link_net = ip_tunnel_get_link_net, 1635 }; 1636 1637 static struct rtnl_link_ops erspan_link_ops __read_mostly = { 1638 .kind = "erspan", 1639 .maxtype = IFLA_GRE_MAX, 1640 .policy = ipgre_policy, 1641 .priv_size = sizeof(struct ip_tunnel), 1642 .setup = erspan_setup, 1643 .validate = erspan_validate, 1644 .newlink = erspan_newlink, 1645 .changelink = erspan_changelink, 1646 .dellink = ip_tunnel_dellink, 1647 .get_size = ipgre_get_size, 1648 .fill_info = erspan_fill_info, 1649 .get_link_net = ip_tunnel_get_link_net, 1650 }; 1651 1652 struct net_device *gretap_fb_dev_create(struct net *net, const char *name, 1653 u8 name_assign_type) 1654 { 1655 struct nlattr *tb[IFLA_MAX + 1]; 1656 struct net_device *dev; 1657 LIST_HEAD(list_kill); 1658 struct ip_tunnel *t; 1659 int err; 1660 1661 memset(&tb, 0, sizeof(tb)); 1662 1663 dev = rtnl_create_link(net, name, name_assign_type, 1664 &ipgre_tap_ops, tb, NULL); 1665 if (IS_ERR(dev)) 1666 return dev; 1667 1668 /* Configure flow based GRE device. */ 1669 t = netdev_priv(dev); 1670 t->collect_md = true; 1671 1672 err = ipgre_newlink(net, dev, tb, NULL, NULL); 1673 if (err < 0) { 1674 free_netdev(dev); 1675 return ERR_PTR(err); 1676 } 1677 1678 /* openvswitch users expect packet sizes to be unrestricted, 1679 * so set the largest MTU we can. 1680 */ 1681 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false); 1682 if (err) 1683 goto out; 1684 1685 err = rtnl_configure_link(dev, NULL, 0, NULL); 1686 if (err < 0) 1687 goto out; 1688 1689 return dev; 1690 out: 1691 ip_tunnel_dellink(dev, &list_kill); 1692 unregister_netdevice_many(&list_kill); 1693 return ERR_PTR(err); 1694 } 1695 EXPORT_SYMBOL_GPL(gretap_fb_dev_create); 1696 1697 static int __net_init ipgre_tap_init_net(struct net *net) 1698 { 1699 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0"); 1700 } 1701 1702 static void __net_exit ipgre_tap_exit_batch_rtnl(struct list_head *list_net, 1703 struct list_head *dev_to_kill) 1704 { 1705 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops, 1706 dev_to_kill); 1707 } 1708 1709 static struct pernet_operations ipgre_tap_net_ops = { 1710 .init = ipgre_tap_init_net, 1711 .exit_batch_rtnl = ipgre_tap_exit_batch_rtnl, 1712 .id = &gre_tap_net_id, 1713 .size = sizeof(struct ip_tunnel_net), 1714 }; 1715 1716 static int __net_init erspan_init_net(struct net *net) 1717 { 1718 return ip_tunnel_init_net(net, erspan_net_id, 1719 &erspan_link_ops, "erspan0"); 1720 } 1721 1722 static void __net_exit erspan_exit_batch_rtnl(struct list_head *net_list, 1723 struct list_head *dev_to_kill) 1724 { 1725 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops, 1726 dev_to_kill); 1727 } 1728 1729 static struct pernet_operations erspan_net_ops = { 1730 .init = erspan_init_net, 1731 .exit_batch_rtnl = erspan_exit_batch_rtnl, 1732 .id = &erspan_net_id, 1733 .size = sizeof(struct ip_tunnel_net), 1734 }; 1735 1736 static int __init ipgre_init(void) 1737 { 1738 int err; 1739 1740 pr_info("GRE over IPv4 tunneling driver\n"); 1741 1742 err = register_pernet_device(&ipgre_net_ops); 1743 if (err < 0) 1744 return err; 1745 1746 err = register_pernet_device(&ipgre_tap_net_ops); 1747 if (err < 0) 1748 goto pnet_tap_failed; 1749 1750 err = register_pernet_device(&erspan_net_ops); 1751 if (err < 0) 1752 goto pnet_erspan_failed; 1753 1754 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO); 1755 if (err < 0) { 1756 pr_info("%s: can't add protocol\n", __func__); 1757 goto add_proto_failed; 1758 } 1759 1760 err = rtnl_link_register(&ipgre_link_ops); 1761 if (err < 0) 1762 goto rtnl_link_failed; 1763 1764 err = rtnl_link_register(&ipgre_tap_ops); 1765 if (err < 0) 1766 goto tap_ops_failed; 1767 1768 err = rtnl_link_register(&erspan_link_ops); 1769 if (err < 0) 1770 goto erspan_link_failed; 1771 1772 return 0; 1773 1774 erspan_link_failed: 1775 rtnl_link_unregister(&ipgre_tap_ops); 1776 tap_ops_failed: 1777 rtnl_link_unregister(&ipgre_link_ops); 1778 rtnl_link_failed: 1779 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1780 add_proto_failed: 1781 unregister_pernet_device(&erspan_net_ops); 1782 pnet_erspan_failed: 1783 unregister_pernet_device(&ipgre_tap_net_ops); 1784 pnet_tap_failed: 1785 unregister_pernet_device(&ipgre_net_ops); 1786 return err; 1787 } 1788 1789 static void __exit ipgre_fini(void) 1790 { 1791 rtnl_link_unregister(&ipgre_tap_ops); 1792 rtnl_link_unregister(&ipgre_link_ops); 1793 rtnl_link_unregister(&erspan_link_ops); 1794 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1795 unregister_pernet_device(&ipgre_tap_net_ops); 1796 unregister_pernet_device(&ipgre_net_ops); 1797 unregister_pernet_device(&erspan_net_ops); 1798 } 1799 1800 module_init(ipgre_init); 1801 module_exit(ipgre_fini); 1802 MODULE_DESCRIPTION("IPv4 GRE tunnels over IP library"); 1803 MODULE_LICENSE("GPL"); 1804 MODULE_ALIAS_RTNL_LINK("gre"); 1805 MODULE_ALIAS_RTNL_LINK("gretap"); 1806 MODULE_ALIAS_RTNL_LINK("erspan"); 1807 MODULE_ALIAS_NETDEV("gre0"); 1808 MODULE_ALIAS_NETDEV("gretap0"); 1809 MODULE_ALIAS_NETDEV("erspan0"); 1810