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 int ipgre_tunnel_init(struct net_device *dev); 111 static void erspan_build_header(struct sk_buff *skb, 112 u32 id, u32 index, 113 bool truncate, bool is_ipv4); 114 115 static unsigned int ipgre_net_id __read_mostly; 116 static unsigned int gre_tap_net_id __read_mostly; 117 static unsigned int erspan_net_id __read_mostly; 118 119 static int ipgre_err(struct sk_buff *skb, u32 info, 120 const struct tnl_ptk_info *tpi) 121 { 122 123 /* All the routers (except for Linux) return only 124 8 bytes of packet payload. It means, that precise relaying of 125 ICMP in the real Internet is absolutely infeasible. 126 127 Moreover, Cisco "wise men" put GRE key to the third word 128 in GRE header. It makes impossible maintaining even soft 129 state for keyed GRE tunnels with enabled checksum. Tell 130 them "thank you". 131 132 Well, I wonder, rfc1812 was written by Cisco employee, 133 what the hell these idiots break standards established 134 by themselves??? 135 */ 136 struct net *net = dev_net(skb->dev); 137 struct ip_tunnel_net *itn; 138 const struct iphdr *iph; 139 const int type = icmp_hdr(skb)->type; 140 const int code = icmp_hdr(skb)->code; 141 unsigned int data_len = 0; 142 struct ip_tunnel *t; 143 144 if (tpi->proto == htons(ETH_P_TEB)) 145 itn = net_generic(net, gre_tap_net_id); 146 else if (tpi->proto == htons(ETH_P_ERSPAN) || 147 tpi->proto == htons(ETH_P_ERSPAN2)) 148 itn = net_generic(net, erspan_net_id); 149 else 150 itn = net_generic(net, ipgre_net_id); 151 152 iph = (const struct iphdr *)(icmp_hdr(skb) + 1); 153 t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags, 154 iph->daddr, iph->saddr, tpi->key); 155 156 if (!t) 157 return -ENOENT; 158 159 switch (type) { 160 default: 161 case ICMP_PARAMETERPROB: 162 return 0; 163 164 case ICMP_DEST_UNREACH: 165 switch (code) { 166 case ICMP_SR_FAILED: 167 case ICMP_PORT_UNREACH: 168 /* Impossible event. */ 169 return 0; 170 default: 171 /* All others are translated to HOST_UNREACH. 172 rfc2003 contains "deep thoughts" about NET_UNREACH, 173 I believe they are just ether pollution. --ANK 174 */ 175 break; 176 } 177 break; 178 179 case ICMP_TIME_EXCEEDED: 180 if (code != ICMP_EXC_TTL) 181 return 0; 182 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */ 183 break; 184 185 case ICMP_REDIRECT: 186 break; 187 } 188 189 #if IS_ENABLED(CONFIG_IPV6) 190 if (tpi->proto == htons(ETH_P_IPV6) && 191 !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len, 192 type, data_len)) 193 return 0; 194 #endif 195 196 if (t->parms.iph.daddr == 0 || 197 ipv4_is_multicast(t->parms.iph.daddr)) 198 return 0; 199 200 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED) 201 return 0; 202 203 if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO)) 204 t->err_count++; 205 else 206 t->err_count = 1; 207 t->err_time = jiffies; 208 209 return 0; 210 } 211 212 static void gre_err(struct sk_buff *skb, u32 info) 213 { 214 /* All the routers (except for Linux) return only 215 * 8 bytes of packet payload. It means, that precise relaying of 216 * ICMP in the real Internet is absolutely infeasible. 217 * 218 * Moreover, Cisco "wise men" put GRE key to the third word 219 * in GRE header. It makes impossible maintaining even soft 220 * state for keyed 221 * GRE tunnels with enabled checksum. Tell them "thank you". 222 * 223 * Well, I wonder, rfc1812 was written by Cisco employee, 224 * what the hell these idiots break standards established 225 * by themselves??? 226 */ 227 228 const struct iphdr *iph = (struct iphdr *)skb->data; 229 const int type = icmp_hdr(skb)->type; 230 const int code = icmp_hdr(skb)->code; 231 struct tnl_ptk_info tpi; 232 233 if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP), 234 iph->ihl * 4) < 0) 235 return; 236 237 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) { 238 ipv4_update_pmtu(skb, dev_net(skb->dev), info, 239 skb->dev->ifindex, IPPROTO_GRE); 240 return; 241 } 242 if (type == ICMP_REDIRECT) { 243 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 244 IPPROTO_GRE); 245 return; 246 } 247 248 ipgre_err(skb, info, &tpi); 249 } 250 251 static bool is_erspan_type1(int gre_hdr_len) 252 { 253 /* Both ERSPAN type I (version 0) and type II (version 1) use 254 * protocol 0x88BE, but the type I has only 4-byte GRE header, 255 * while type II has 8-byte. 256 */ 257 return gre_hdr_len == 4; 258 } 259 260 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi, 261 int gre_hdr_len) 262 { 263 struct net *net = dev_net(skb->dev); 264 struct metadata_dst *tun_dst = NULL; 265 struct erspan_base_hdr *ershdr; 266 struct ip_tunnel_net *itn; 267 struct ip_tunnel *tunnel; 268 const struct iphdr *iph; 269 struct erspan_md2 *md2; 270 int ver; 271 int len; 272 273 itn = net_generic(net, erspan_net_id); 274 iph = ip_hdr(skb); 275 if (is_erspan_type1(gre_hdr_len)) { 276 ver = 0; 277 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, 278 tpi->flags | TUNNEL_NO_KEY, 279 iph->saddr, iph->daddr, 0); 280 } else { 281 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len); 282 ver = ershdr->ver; 283 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, 284 tpi->flags | TUNNEL_KEY, 285 iph->saddr, iph->daddr, tpi->key); 286 } 287 288 if (tunnel) { 289 if (is_erspan_type1(gre_hdr_len)) 290 len = gre_hdr_len; 291 else 292 len = gre_hdr_len + erspan_hdr_len(ver); 293 294 if (unlikely(!pskb_may_pull(skb, len))) 295 return PACKET_REJECT; 296 297 if (__iptunnel_pull_header(skb, 298 len, 299 htons(ETH_P_TEB), 300 false, false) < 0) 301 goto drop; 302 303 if (tunnel->collect_md) { 304 struct erspan_metadata *pkt_md, *md; 305 struct ip_tunnel_info *info; 306 unsigned char *gh; 307 __be64 tun_id; 308 __be16 flags; 309 310 tpi->flags |= TUNNEL_KEY; 311 flags = tpi->flags; 312 tun_id = key32_to_tunnel_id(tpi->key); 313 314 tun_dst = ip_tun_rx_dst(skb, flags, 315 tun_id, sizeof(*md)); 316 if (!tun_dst) 317 return PACKET_REJECT; 318 319 /* skb can be uncloned in __iptunnel_pull_header, so 320 * old pkt_md is no longer valid and we need to reset 321 * it 322 */ 323 gh = skb_network_header(skb) + 324 skb_network_header_len(skb); 325 pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len + 326 sizeof(*ershdr)); 327 md = ip_tunnel_info_opts(&tun_dst->u.tun_info); 328 md->version = ver; 329 md2 = &md->u.md2; 330 memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE : 331 ERSPAN_V2_MDSIZE); 332 333 info = &tun_dst->u.tun_info; 334 info->key.tun_flags |= TUNNEL_ERSPAN_OPT; 335 info->options_len = sizeof(*md); 336 } 337 338 skb_reset_mac_header(skb); 339 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error); 340 return PACKET_RCVD; 341 } 342 return PACKET_REJECT; 343 344 drop: 345 kfree_skb(skb); 346 return PACKET_RCVD; 347 } 348 349 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi, 350 struct ip_tunnel_net *itn, int hdr_len, bool raw_proto) 351 { 352 struct metadata_dst *tun_dst = NULL; 353 const struct iphdr *iph; 354 struct ip_tunnel *tunnel; 355 356 iph = ip_hdr(skb); 357 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags, 358 iph->saddr, iph->daddr, tpi->key); 359 360 if (tunnel) { 361 const struct iphdr *tnl_params; 362 363 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto, 364 raw_proto, false) < 0) 365 goto drop; 366 367 if (tunnel->dev->type != ARPHRD_NONE) 368 skb_pop_mac_header(skb); 369 else 370 skb_reset_mac_header(skb); 371 372 tnl_params = &tunnel->parms.iph; 373 if (tunnel->collect_md || tnl_params->daddr == 0) { 374 __be16 flags; 375 __be64 tun_id; 376 377 flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY); 378 tun_id = key32_to_tunnel_id(tpi->key); 379 tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0); 380 if (!tun_dst) 381 return PACKET_REJECT; 382 } 383 384 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error); 385 return PACKET_RCVD; 386 } 387 return PACKET_NEXT; 388 389 drop: 390 kfree_skb(skb); 391 return PACKET_RCVD; 392 } 393 394 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi, 395 int hdr_len) 396 { 397 struct net *net = dev_net(skb->dev); 398 struct ip_tunnel_net *itn; 399 int res; 400 401 if (tpi->proto == htons(ETH_P_TEB)) 402 itn = net_generic(net, gre_tap_net_id); 403 else 404 itn = net_generic(net, ipgre_net_id); 405 406 res = __ipgre_rcv(skb, tpi, itn, hdr_len, false); 407 if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) { 408 /* ipgre tunnels in collect metadata mode should receive 409 * also ETH_P_TEB traffic. 410 */ 411 itn = net_generic(net, ipgre_net_id); 412 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true); 413 } 414 return res; 415 } 416 417 static int gre_rcv(struct sk_buff *skb) 418 { 419 struct tnl_ptk_info tpi; 420 bool csum_err = false; 421 int hdr_len; 422 423 #ifdef CONFIG_NET_IPGRE_BROADCAST 424 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) { 425 /* Looped back packet, drop it! */ 426 if (rt_is_output_route(skb_rtable(skb))) 427 goto drop; 428 } 429 #endif 430 431 hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0); 432 if (hdr_len < 0) 433 goto drop; 434 435 if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) || 436 tpi.proto == htons(ETH_P_ERSPAN2))) { 437 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD) 438 return 0; 439 goto out; 440 } 441 442 if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD) 443 return 0; 444 445 out: 446 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 447 drop: 448 kfree_skb(skb); 449 return 0; 450 } 451 452 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev, 453 const struct iphdr *tnl_params, 454 __be16 proto) 455 { 456 struct ip_tunnel *tunnel = netdev_priv(dev); 457 458 if (tunnel->parms.o_flags & TUNNEL_SEQ) 459 tunnel->o_seqno++; 460 461 /* Push GRE header. */ 462 gre_build_header(skb, tunnel->tun_hlen, 463 tunnel->parms.o_flags, proto, tunnel->parms.o_key, 464 htonl(tunnel->o_seqno)); 465 466 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol); 467 } 468 469 static int gre_handle_offloads(struct sk_buff *skb, bool csum) 470 { 471 return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE); 472 } 473 474 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev, 475 __be16 proto) 476 { 477 struct ip_tunnel *tunnel = netdev_priv(dev); 478 struct ip_tunnel_info *tun_info; 479 const struct ip_tunnel_key *key; 480 int tunnel_hlen; 481 __be16 flags; 482 483 tun_info = skb_tunnel_info(skb); 484 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 485 ip_tunnel_info_af(tun_info) != AF_INET)) 486 goto err_free_skb; 487 488 key = &tun_info->key; 489 tunnel_hlen = gre_calc_hlen(key->tun_flags); 490 491 if (skb_cow_head(skb, dev->needed_headroom)) 492 goto err_free_skb; 493 494 /* Push Tunnel header. */ 495 if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM))) 496 goto err_free_skb; 497 498 flags = tun_info->key.tun_flags & 499 (TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ); 500 gre_build_header(skb, tunnel_hlen, flags, proto, 501 tunnel_id_to_key32(tun_info->key.tun_id), 502 (flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++) : 0); 503 504 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen); 505 506 return; 507 508 err_free_skb: 509 kfree_skb(skb); 510 dev->stats.tx_dropped++; 511 } 512 513 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev) 514 { 515 struct ip_tunnel *tunnel = netdev_priv(dev); 516 struct ip_tunnel_info *tun_info; 517 const struct ip_tunnel_key *key; 518 struct erspan_metadata *md; 519 bool truncate = false; 520 __be16 proto; 521 int tunnel_hlen; 522 int version; 523 int nhoff; 524 int thoff; 525 526 tun_info = skb_tunnel_info(skb); 527 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 528 ip_tunnel_info_af(tun_info) != AF_INET)) 529 goto err_free_skb; 530 531 key = &tun_info->key; 532 if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT)) 533 goto err_free_skb; 534 if (tun_info->options_len < sizeof(*md)) 535 goto err_free_skb; 536 md = ip_tunnel_info_opts(tun_info); 537 538 /* ERSPAN has fixed 8 byte GRE header */ 539 version = md->version; 540 tunnel_hlen = 8 + erspan_hdr_len(version); 541 542 if (skb_cow_head(skb, dev->needed_headroom)) 543 goto err_free_skb; 544 545 if (gre_handle_offloads(skb, false)) 546 goto err_free_skb; 547 548 if (skb->len > dev->mtu + dev->hard_header_len) { 549 pskb_trim(skb, dev->mtu + dev->hard_header_len); 550 truncate = true; 551 } 552 553 nhoff = skb_network_header(skb) - skb_mac_header(skb); 554 if (skb->protocol == htons(ETH_P_IP) && 555 (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff)) 556 truncate = true; 557 558 thoff = skb_transport_header(skb) - skb_mac_header(skb); 559 if (skb->protocol == htons(ETH_P_IPV6) && 560 (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff)) 561 truncate = true; 562 563 if (version == 1) { 564 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)), 565 ntohl(md->u.index), truncate, true); 566 proto = htons(ETH_P_ERSPAN); 567 } else if (version == 2) { 568 erspan_build_header_v2(skb, 569 ntohl(tunnel_id_to_key32(key->tun_id)), 570 md->u.md2.dir, 571 get_hwid(&md->u.md2), 572 truncate, true); 573 proto = htons(ETH_P_ERSPAN2); 574 } else { 575 goto err_free_skb; 576 } 577 578 gre_build_header(skb, 8, TUNNEL_SEQ, 579 proto, 0, htonl(tunnel->o_seqno++)); 580 581 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen); 582 583 return; 584 585 err_free_skb: 586 kfree_skb(skb); 587 dev->stats.tx_dropped++; 588 } 589 590 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 591 { 592 struct ip_tunnel_info *info = skb_tunnel_info(skb); 593 const struct ip_tunnel_key *key; 594 struct rtable *rt; 595 struct flowi4 fl4; 596 597 if (ip_tunnel_info_af(info) != AF_INET) 598 return -EINVAL; 599 600 key = &info->key; 601 ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src, 602 tunnel_id_to_key32(key->tun_id), key->tos, 0, 603 skb->mark, skb_get_hash(skb)); 604 rt = ip_route_output_key(dev_net(dev), &fl4); 605 if (IS_ERR(rt)) 606 return PTR_ERR(rt); 607 608 ip_rt_put(rt); 609 info->key.u.ipv4.src = fl4.saddr; 610 return 0; 611 } 612 613 static netdev_tx_t ipgre_xmit(struct sk_buff *skb, 614 struct net_device *dev) 615 { 616 struct ip_tunnel *tunnel = netdev_priv(dev); 617 const struct iphdr *tnl_params; 618 619 if (!pskb_inet_may_pull(skb)) 620 goto free_skb; 621 622 if (tunnel->collect_md) { 623 gre_fb_xmit(skb, dev, skb->protocol); 624 return NETDEV_TX_OK; 625 } 626 627 if (dev->header_ops) { 628 /* Need space for new headers */ 629 if (skb_cow_head(skb, dev->needed_headroom - 630 (tunnel->hlen + sizeof(struct iphdr)))) 631 goto free_skb; 632 633 tnl_params = (const struct iphdr *)skb->data; 634 635 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing 636 * to gre header. 637 */ 638 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr)); 639 skb_reset_mac_header(skb); 640 } else { 641 if (skb_cow_head(skb, dev->needed_headroom)) 642 goto free_skb; 643 644 tnl_params = &tunnel->parms.iph; 645 } 646 647 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 648 goto free_skb; 649 650 __gre_xmit(skb, dev, tnl_params, skb->protocol); 651 return NETDEV_TX_OK; 652 653 free_skb: 654 kfree_skb(skb); 655 dev->stats.tx_dropped++; 656 return NETDEV_TX_OK; 657 } 658 659 static netdev_tx_t erspan_xmit(struct sk_buff *skb, 660 struct net_device *dev) 661 { 662 struct ip_tunnel *tunnel = netdev_priv(dev); 663 bool truncate = false; 664 __be16 proto; 665 666 if (!pskb_inet_may_pull(skb)) 667 goto free_skb; 668 669 if (tunnel->collect_md) { 670 erspan_fb_xmit(skb, dev); 671 return NETDEV_TX_OK; 672 } 673 674 if (gre_handle_offloads(skb, false)) 675 goto free_skb; 676 677 if (skb_cow_head(skb, dev->needed_headroom)) 678 goto free_skb; 679 680 if (skb->len > dev->mtu + dev->hard_header_len) { 681 pskb_trim(skb, dev->mtu + dev->hard_header_len); 682 truncate = true; 683 } 684 685 /* Push ERSPAN header */ 686 if (tunnel->erspan_ver == 0) { 687 proto = htons(ETH_P_ERSPAN); 688 tunnel->parms.o_flags &= ~TUNNEL_SEQ; 689 } else if (tunnel->erspan_ver == 1) { 690 erspan_build_header(skb, ntohl(tunnel->parms.o_key), 691 tunnel->index, 692 truncate, true); 693 proto = htons(ETH_P_ERSPAN); 694 } else if (tunnel->erspan_ver == 2) { 695 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key), 696 tunnel->dir, tunnel->hwid, 697 truncate, true); 698 proto = htons(ETH_P_ERSPAN2); 699 } else { 700 goto free_skb; 701 } 702 703 tunnel->parms.o_flags &= ~TUNNEL_KEY; 704 __gre_xmit(skb, dev, &tunnel->parms.iph, proto); 705 return NETDEV_TX_OK; 706 707 free_skb: 708 kfree_skb(skb); 709 dev->stats.tx_dropped++; 710 return NETDEV_TX_OK; 711 } 712 713 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb, 714 struct net_device *dev) 715 { 716 struct ip_tunnel *tunnel = netdev_priv(dev); 717 718 if (!pskb_inet_may_pull(skb)) 719 goto free_skb; 720 721 if (tunnel->collect_md) { 722 gre_fb_xmit(skb, dev, htons(ETH_P_TEB)); 723 return NETDEV_TX_OK; 724 } 725 726 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 727 goto free_skb; 728 729 if (skb_cow_head(skb, dev->needed_headroom)) 730 goto free_skb; 731 732 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB)); 733 return NETDEV_TX_OK; 734 735 free_skb: 736 kfree_skb(skb); 737 dev->stats.tx_dropped++; 738 return NETDEV_TX_OK; 739 } 740 741 static void ipgre_link_update(struct net_device *dev, bool set_mtu) 742 { 743 struct ip_tunnel *tunnel = netdev_priv(dev); 744 int len; 745 746 len = tunnel->tun_hlen; 747 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 748 len = tunnel->tun_hlen - len; 749 tunnel->hlen = tunnel->hlen + len; 750 751 dev->needed_headroom = dev->needed_headroom + len; 752 if (set_mtu) 753 dev->mtu = max_t(int, dev->mtu - len, 68); 754 755 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) { 756 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) || 757 tunnel->encap.type == TUNNEL_ENCAP_NONE) { 758 dev->features |= NETIF_F_GSO_SOFTWARE; 759 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 760 } else { 761 dev->features &= ~NETIF_F_GSO_SOFTWARE; 762 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 763 } 764 dev->features |= NETIF_F_LLTX; 765 } else { 766 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 767 dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE); 768 } 769 } 770 771 static int ipgre_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, 772 int cmd) 773 { 774 int err; 775 776 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) { 777 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE || 778 p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) || 779 ((p->i_flags | p->o_flags) & (GRE_VERSION | GRE_ROUTING))) 780 return -EINVAL; 781 } 782 783 p->i_flags = gre_flags_to_tnl_flags(p->i_flags); 784 p->o_flags = gre_flags_to_tnl_flags(p->o_flags); 785 786 err = ip_tunnel_ctl(dev, p, cmd); 787 if (err) 788 return err; 789 790 if (cmd == SIOCCHGTUNNEL) { 791 struct ip_tunnel *t = netdev_priv(dev); 792 793 t->parms.i_flags = p->i_flags; 794 t->parms.o_flags = p->o_flags; 795 796 if (strcmp(dev->rtnl_link_ops->kind, "erspan")) 797 ipgre_link_update(dev, true); 798 } 799 800 p->i_flags = gre_tnl_flags_to_gre_flags(p->i_flags); 801 p->o_flags = gre_tnl_flags_to_gre_flags(p->o_flags); 802 return 0; 803 } 804 805 /* Nice toy. Unfortunately, useless in real life :-) 806 It allows to construct virtual multiprotocol broadcast "LAN" 807 over the Internet, provided multicast routing is tuned. 808 809 810 I have no idea was this bicycle invented before me, 811 so that I had to set ARPHRD_IPGRE to a random value. 812 I have an impression, that Cisco could make something similar, 813 but this feature is apparently missing in IOS<=11.2(8). 814 815 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks 816 with broadcast 224.66.66.66. If you have access to mbone, play with me :-) 817 818 ping -t 255 224.66.66.66 819 820 If nobody answers, mbone does not work. 821 822 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255 823 ip addr add 10.66.66.<somewhat>/24 dev Universe 824 ifconfig Universe up 825 ifconfig Universe add fe80::<Your_real_addr>/10 826 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96 827 ftp 10.66.66.66 828 ... 829 ftp fec0:6666:6666::193.233.7.65 830 ... 831 */ 832 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, 833 unsigned short type, 834 const void *daddr, const void *saddr, unsigned int len) 835 { 836 struct ip_tunnel *t = netdev_priv(dev); 837 struct iphdr *iph; 838 struct gre_base_hdr *greh; 839 840 iph = skb_push(skb, t->hlen + sizeof(*iph)); 841 greh = (struct gre_base_hdr *)(iph+1); 842 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags); 843 greh->protocol = htons(type); 844 845 memcpy(iph, &t->parms.iph, sizeof(struct iphdr)); 846 847 /* Set the source hardware address. */ 848 if (saddr) 849 memcpy(&iph->saddr, saddr, 4); 850 if (daddr) 851 memcpy(&iph->daddr, daddr, 4); 852 if (iph->daddr) 853 return t->hlen + sizeof(*iph); 854 855 return -(t->hlen + sizeof(*iph)); 856 } 857 858 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr) 859 { 860 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb); 861 memcpy(haddr, &iph->saddr, 4); 862 return 4; 863 } 864 865 static const struct header_ops ipgre_header_ops = { 866 .create = ipgre_header, 867 .parse = ipgre_header_parse, 868 }; 869 870 #ifdef CONFIG_NET_IPGRE_BROADCAST 871 static int ipgre_open(struct net_device *dev) 872 { 873 struct ip_tunnel *t = netdev_priv(dev); 874 875 if (ipv4_is_multicast(t->parms.iph.daddr)) { 876 struct flowi4 fl4; 877 struct rtable *rt; 878 879 rt = ip_route_output_gre(t->net, &fl4, 880 t->parms.iph.daddr, 881 t->parms.iph.saddr, 882 t->parms.o_key, 883 RT_TOS(t->parms.iph.tos), 884 t->parms.link); 885 if (IS_ERR(rt)) 886 return -EADDRNOTAVAIL; 887 dev = rt->dst.dev; 888 ip_rt_put(rt); 889 if (!__in_dev_get_rtnl(dev)) 890 return -EADDRNOTAVAIL; 891 t->mlink = dev->ifindex; 892 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr); 893 } 894 return 0; 895 } 896 897 static int ipgre_close(struct net_device *dev) 898 { 899 struct ip_tunnel *t = netdev_priv(dev); 900 901 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) { 902 struct in_device *in_dev; 903 in_dev = inetdev_by_index(t->net, t->mlink); 904 if (in_dev) 905 ip_mc_dec_group(in_dev, t->parms.iph.daddr); 906 } 907 return 0; 908 } 909 #endif 910 911 static const struct net_device_ops ipgre_netdev_ops = { 912 .ndo_init = ipgre_tunnel_init, 913 .ndo_uninit = ip_tunnel_uninit, 914 #ifdef CONFIG_NET_IPGRE_BROADCAST 915 .ndo_open = ipgre_open, 916 .ndo_stop = ipgre_close, 917 #endif 918 .ndo_start_xmit = ipgre_xmit, 919 .ndo_do_ioctl = ip_tunnel_ioctl, 920 .ndo_change_mtu = ip_tunnel_change_mtu, 921 .ndo_get_stats64 = ip_tunnel_get_stats64, 922 .ndo_get_iflink = ip_tunnel_get_iflink, 923 .ndo_tunnel_ctl = ipgre_tunnel_ctl, 924 }; 925 926 #define GRE_FEATURES (NETIF_F_SG | \ 927 NETIF_F_FRAGLIST | \ 928 NETIF_F_HIGHDMA | \ 929 NETIF_F_HW_CSUM) 930 931 static void ipgre_tunnel_setup(struct net_device *dev) 932 { 933 dev->netdev_ops = &ipgre_netdev_ops; 934 dev->type = ARPHRD_IPGRE; 935 ip_tunnel_setup(dev, ipgre_net_id); 936 } 937 938 static void __gre_tunnel_init(struct net_device *dev) 939 { 940 struct ip_tunnel *tunnel; 941 942 tunnel = netdev_priv(dev); 943 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 944 tunnel->parms.iph.protocol = IPPROTO_GRE; 945 946 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen; 947 948 dev->features |= GRE_FEATURES; 949 dev->hw_features |= GRE_FEATURES; 950 951 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) { 952 /* TCP offload with GRE SEQ is not supported, nor 953 * can we support 2 levels of outer headers requiring 954 * an update. 955 */ 956 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) || 957 (tunnel->encap.type == TUNNEL_ENCAP_NONE)) { 958 dev->features |= NETIF_F_GSO_SOFTWARE; 959 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 960 } 961 962 /* Can use a lockless transmit, unless we generate 963 * output sequences 964 */ 965 dev->features |= NETIF_F_LLTX; 966 } 967 } 968 969 static int ipgre_tunnel_init(struct net_device *dev) 970 { 971 struct ip_tunnel *tunnel = netdev_priv(dev); 972 struct iphdr *iph = &tunnel->parms.iph; 973 974 __gre_tunnel_init(dev); 975 976 memcpy(dev->dev_addr, &iph->saddr, 4); 977 memcpy(dev->broadcast, &iph->daddr, 4); 978 979 dev->flags = IFF_NOARP; 980 netif_keep_dst(dev); 981 dev->addr_len = 4; 982 983 if (iph->daddr && !tunnel->collect_md) { 984 #ifdef CONFIG_NET_IPGRE_BROADCAST 985 if (ipv4_is_multicast(iph->daddr)) { 986 if (!iph->saddr) 987 return -EINVAL; 988 dev->flags = IFF_BROADCAST; 989 dev->header_ops = &ipgre_header_ops; 990 } 991 #endif 992 } else if (!tunnel->collect_md) { 993 dev->header_ops = &ipgre_header_ops; 994 } 995 996 return ip_tunnel_init(dev); 997 } 998 999 static const struct gre_protocol ipgre_protocol = { 1000 .handler = gre_rcv, 1001 .err_handler = gre_err, 1002 }; 1003 1004 static int __net_init ipgre_init_net(struct net *net) 1005 { 1006 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL); 1007 } 1008 1009 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net) 1010 { 1011 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops); 1012 } 1013 1014 static struct pernet_operations ipgre_net_ops = { 1015 .init = ipgre_init_net, 1016 .exit_batch = ipgre_exit_batch_net, 1017 .id = &ipgre_net_id, 1018 .size = sizeof(struct ip_tunnel_net), 1019 }; 1020 1021 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[], 1022 struct netlink_ext_ack *extack) 1023 { 1024 __be16 flags; 1025 1026 if (!data) 1027 return 0; 1028 1029 flags = 0; 1030 if (data[IFLA_GRE_IFLAGS]) 1031 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1032 if (data[IFLA_GRE_OFLAGS]) 1033 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1034 if (flags & (GRE_VERSION|GRE_ROUTING)) 1035 return -EINVAL; 1036 1037 if (data[IFLA_GRE_COLLECT_METADATA] && 1038 data[IFLA_GRE_ENCAP_TYPE] && 1039 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE) 1040 return -EINVAL; 1041 1042 return 0; 1043 } 1044 1045 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[], 1046 struct netlink_ext_ack *extack) 1047 { 1048 __be32 daddr; 1049 1050 if (tb[IFLA_ADDRESS]) { 1051 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1052 return -EINVAL; 1053 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1054 return -EADDRNOTAVAIL; 1055 } 1056 1057 if (!data) 1058 goto out; 1059 1060 if (data[IFLA_GRE_REMOTE]) { 1061 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4); 1062 if (!daddr) 1063 return -EINVAL; 1064 } 1065 1066 out: 1067 return ipgre_tunnel_validate(tb, data, extack); 1068 } 1069 1070 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[], 1071 struct netlink_ext_ack *extack) 1072 { 1073 __be16 flags = 0; 1074 int ret; 1075 1076 if (!data) 1077 return 0; 1078 1079 ret = ipgre_tap_validate(tb, data, extack); 1080 if (ret) 1081 return ret; 1082 1083 if (data[IFLA_GRE_ERSPAN_VER] && 1084 nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0) 1085 return 0; 1086 1087 /* ERSPAN type II/III should only have GRE sequence and key flag */ 1088 if (data[IFLA_GRE_OFLAGS]) 1089 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1090 if (data[IFLA_GRE_IFLAGS]) 1091 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1092 if (!data[IFLA_GRE_COLLECT_METADATA] && 1093 flags != (GRE_SEQ | GRE_KEY)) 1094 return -EINVAL; 1095 1096 /* ERSPAN Session ID only has 10-bit. Since we reuse 1097 * 32-bit key field as ID, check it's range. 1098 */ 1099 if (data[IFLA_GRE_IKEY] && 1100 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK)) 1101 return -EINVAL; 1102 1103 if (data[IFLA_GRE_OKEY] && 1104 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK)) 1105 return -EINVAL; 1106 1107 return 0; 1108 } 1109 1110 static int ipgre_netlink_parms(struct net_device *dev, 1111 struct nlattr *data[], 1112 struct nlattr *tb[], 1113 struct ip_tunnel_parm *parms, 1114 __u32 *fwmark) 1115 { 1116 struct ip_tunnel *t = netdev_priv(dev); 1117 1118 memset(parms, 0, sizeof(*parms)); 1119 1120 parms->iph.protocol = IPPROTO_GRE; 1121 1122 if (!data) 1123 return 0; 1124 1125 if (data[IFLA_GRE_LINK]) 1126 parms->link = nla_get_u32(data[IFLA_GRE_LINK]); 1127 1128 if (data[IFLA_GRE_IFLAGS]) 1129 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS])); 1130 1131 if (data[IFLA_GRE_OFLAGS]) 1132 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS])); 1133 1134 if (data[IFLA_GRE_IKEY]) 1135 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]); 1136 1137 if (data[IFLA_GRE_OKEY]) 1138 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]); 1139 1140 if (data[IFLA_GRE_LOCAL]) 1141 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]); 1142 1143 if (data[IFLA_GRE_REMOTE]) 1144 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]); 1145 1146 if (data[IFLA_GRE_TTL]) 1147 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]); 1148 1149 if (data[IFLA_GRE_TOS]) 1150 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]); 1151 1152 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) { 1153 if (t->ignore_df) 1154 return -EINVAL; 1155 parms->iph.frag_off = htons(IP_DF); 1156 } 1157 1158 if (data[IFLA_GRE_COLLECT_METADATA]) { 1159 t->collect_md = true; 1160 if (dev->type == ARPHRD_IPGRE) 1161 dev->type = ARPHRD_NONE; 1162 } 1163 1164 if (data[IFLA_GRE_IGNORE_DF]) { 1165 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF]) 1166 && (parms->iph.frag_off & htons(IP_DF))) 1167 return -EINVAL; 1168 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]); 1169 } 1170 1171 if (data[IFLA_GRE_FWMARK]) 1172 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]); 1173 1174 return 0; 1175 } 1176 1177 static int erspan_netlink_parms(struct net_device *dev, 1178 struct nlattr *data[], 1179 struct nlattr *tb[], 1180 struct ip_tunnel_parm *parms, 1181 __u32 *fwmark) 1182 { 1183 struct ip_tunnel *t = netdev_priv(dev); 1184 int err; 1185 1186 err = ipgre_netlink_parms(dev, data, tb, parms, fwmark); 1187 if (err) 1188 return err; 1189 if (!data) 1190 return 0; 1191 1192 if (data[IFLA_GRE_ERSPAN_VER]) { 1193 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]); 1194 1195 if (t->erspan_ver > 2) 1196 return -EINVAL; 1197 } 1198 1199 if (t->erspan_ver == 1) { 1200 if (data[IFLA_GRE_ERSPAN_INDEX]) { 1201 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]); 1202 if (t->index & ~INDEX_MASK) 1203 return -EINVAL; 1204 } 1205 } else if (t->erspan_ver == 2) { 1206 if (data[IFLA_GRE_ERSPAN_DIR]) { 1207 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]); 1208 if (t->dir & ~(DIR_MASK >> DIR_OFFSET)) 1209 return -EINVAL; 1210 } 1211 if (data[IFLA_GRE_ERSPAN_HWID]) { 1212 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]); 1213 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET)) 1214 return -EINVAL; 1215 } 1216 } 1217 1218 return 0; 1219 } 1220 1221 /* This function returns true when ENCAP attributes are present in the nl msg */ 1222 static bool ipgre_netlink_encap_parms(struct nlattr *data[], 1223 struct ip_tunnel_encap *ipencap) 1224 { 1225 bool ret = false; 1226 1227 memset(ipencap, 0, sizeof(*ipencap)); 1228 1229 if (!data) 1230 return ret; 1231 1232 if (data[IFLA_GRE_ENCAP_TYPE]) { 1233 ret = true; 1234 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]); 1235 } 1236 1237 if (data[IFLA_GRE_ENCAP_FLAGS]) { 1238 ret = true; 1239 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]); 1240 } 1241 1242 if (data[IFLA_GRE_ENCAP_SPORT]) { 1243 ret = true; 1244 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]); 1245 } 1246 1247 if (data[IFLA_GRE_ENCAP_DPORT]) { 1248 ret = true; 1249 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]); 1250 } 1251 1252 return ret; 1253 } 1254 1255 static int gre_tap_init(struct net_device *dev) 1256 { 1257 __gre_tunnel_init(dev); 1258 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1259 netif_keep_dst(dev); 1260 1261 return ip_tunnel_init(dev); 1262 } 1263 1264 static const struct net_device_ops gre_tap_netdev_ops = { 1265 .ndo_init = gre_tap_init, 1266 .ndo_uninit = ip_tunnel_uninit, 1267 .ndo_start_xmit = gre_tap_xmit, 1268 .ndo_set_mac_address = eth_mac_addr, 1269 .ndo_validate_addr = eth_validate_addr, 1270 .ndo_change_mtu = ip_tunnel_change_mtu, 1271 .ndo_get_stats64 = ip_tunnel_get_stats64, 1272 .ndo_get_iflink = ip_tunnel_get_iflink, 1273 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1274 }; 1275 1276 static int erspan_tunnel_init(struct net_device *dev) 1277 { 1278 struct ip_tunnel *tunnel = netdev_priv(dev); 1279 1280 if (tunnel->erspan_ver == 0) 1281 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */ 1282 else 1283 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */ 1284 1285 tunnel->parms.iph.protocol = IPPROTO_GRE; 1286 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen + 1287 erspan_hdr_len(tunnel->erspan_ver); 1288 1289 dev->features |= GRE_FEATURES; 1290 dev->hw_features |= GRE_FEATURES; 1291 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1292 netif_keep_dst(dev); 1293 1294 return ip_tunnel_init(dev); 1295 } 1296 1297 static const struct net_device_ops erspan_netdev_ops = { 1298 .ndo_init = erspan_tunnel_init, 1299 .ndo_uninit = ip_tunnel_uninit, 1300 .ndo_start_xmit = erspan_xmit, 1301 .ndo_set_mac_address = eth_mac_addr, 1302 .ndo_validate_addr = eth_validate_addr, 1303 .ndo_change_mtu = ip_tunnel_change_mtu, 1304 .ndo_get_stats64 = ip_tunnel_get_stats64, 1305 .ndo_get_iflink = ip_tunnel_get_iflink, 1306 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1307 }; 1308 1309 static void ipgre_tap_setup(struct net_device *dev) 1310 { 1311 ether_setup(dev); 1312 dev->max_mtu = 0; 1313 dev->netdev_ops = &gre_tap_netdev_ops; 1314 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1315 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1316 ip_tunnel_setup(dev, gre_tap_net_id); 1317 } 1318 1319 static int 1320 ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[]) 1321 { 1322 struct ip_tunnel_encap ipencap; 1323 1324 if (ipgre_netlink_encap_parms(data, &ipencap)) { 1325 struct ip_tunnel *t = netdev_priv(dev); 1326 int err = ip_tunnel_encap_setup(t, &ipencap); 1327 1328 if (err < 0) 1329 return err; 1330 } 1331 1332 return 0; 1333 } 1334 1335 static int ipgre_newlink(struct net *src_net, struct net_device *dev, 1336 struct nlattr *tb[], struct nlattr *data[], 1337 struct netlink_ext_ack *extack) 1338 { 1339 struct ip_tunnel_parm p; 1340 __u32 fwmark = 0; 1341 int err; 1342 1343 err = ipgre_newlink_encap_setup(dev, data); 1344 if (err) 1345 return err; 1346 1347 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1348 if (err < 0) 1349 return err; 1350 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1351 } 1352 1353 static int erspan_newlink(struct net *src_net, struct net_device *dev, 1354 struct nlattr *tb[], struct nlattr *data[], 1355 struct netlink_ext_ack *extack) 1356 { 1357 struct ip_tunnel_parm p; 1358 __u32 fwmark = 0; 1359 int err; 1360 1361 err = ipgre_newlink_encap_setup(dev, data); 1362 if (err) 1363 return err; 1364 1365 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1366 if (err) 1367 return err; 1368 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1369 } 1370 1371 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[], 1372 struct nlattr *data[], 1373 struct netlink_ext_ack *extack) 1374 { 1375 struct ip_tunnel *t = netdev_priv(dev); 1376 __u32 fwmark = t->fwmark; 1377 struct ip_tunnel_parm p; 1378 int err; 1379 1380 err = ipgre_newlink_encap_setup(dev, data); 1381 if (err) 1382 return err; 1383 1384 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1385 if (err < 0) 1386 return err; 1387 1388 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1389 if (err < 0) 1390 return err; 1391 1392 t->parms.i_flags = p.i_flags; 1393 t->parms.o_flags = p.o_flags; 1394 1395 ipgre_link_update(dev, !tb[IFLA_MTU]); 1396 1397 return 0; 1398 } 1399 1400 static int erspan_changelink(struct net_device *dev, struct nlattr *tb[], 1401 struct nlattr *data[], 1402 struct netlink_ext_ack *extack) 1403 { 1404 struct ip_tunnel *t = netdev_priv(dev); 1405 __u32 fwmark = t->fwmark; 1406 struct ip_tunnel_parm p; 1407 int err; 1408 1409 err = ipgre_newlink_encap_setup(dev, data); 1410 if (err) 1411 return err; 1412 1413 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1414 if (err < 0) 1415 return err; 1416 1417 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1418 if (err < 0) 1419 return err; 1420 1421 t->parms.i_flags = p.i_flags; 1422 t->parms.o_flags = p.o_flags; 1423 1424 return 0; 1425 } 1426 1427 static size_t ipgre_get_size(const struct net_device *dev) 1428 { 1429 return 1430 /* IFLA_GRE_LINK */ 1431 nla_total_size(4) + 1432 /* IFLA_GRE_IFLAGS */ 1433 nla_total_size(2) + 1434 /* IFLA_GRE_OFLAGS */ 1435 nla_total_size(2) + 1436 /* IFLA_GRE_IKEY */ 1437 nla_total_size(4) + 1438 /* IFLA_GRE_OKEY */ 1439 nla_total_size(4) + 1440 /* IFLA_GRE_LOCAL */ 1441 nla_total_size(4) + 1442 /* IFLA_GRE_REMOTE */ 1443 nla_total_size(4) + 1444 /* IFLA_GRE_TTL */ 1445 nla_total_size(1) + 1446 /* IFLA_GRE_TOS */ 1447 nla_total_size(1) + 1448 /* IFLA_GRE_PMTUDISC */ 1449 nla_total_size(1) + 1450 /* IFLA_GRE_ENCAP_TYPE */ 1451 nla_total_size(2) + 1452 /* IFLA_GRE_ENCAP_FLAGS */ 1453 nla_total_size(2) + 1454 /* IFLA_GRE_ENCAP_SPORT */ 1455 nla_total_size(2) + 1456 /* IFLA_GRE_ENCAP_DPORT */ 1457 nla_total_size(2) + 1458 /* IFLA_GRE_COLLECT_METADATA */ 1459 nla_total_size(0) + 1460 /* IFLA_GRE_IGNORE_DF */ 1461 nla_total_size(1) + 1462 /* IFLA_GRE_FWMARK */ 1463 nla_total_size(4) + 1464 /* IFLA_GRE_ERSPAN_INDEX */ 1465 nla_total_size(4) + 1466 /* IFLA_GRE_ERSPAN_VER */ 1467 nla_total_size(1) + 1468 /* IFLA_GRE_ERSPAN_DIR */ 1469 nla_total_size(1) + 1470 /* IFLA_GRE_ERSPAN_HWID */ 1471 nla_total_size(2) + 1472 0; 1473 } 1474 1475 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev) 1476 { 1477 struct ip_tunnel *t = netdev_priv(dev); 1478 struct ip_tunnel_parm *p = &t->parms; 1479 __be16 o_flags = p->o_flags; 1480 1481 if (t->erspan_ver <= 2) { 1482 if (t->erspan_ver != 0 && !t->collect_md) 1483 o_flags |= TUNNEL_KEY; 1484 1485 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver)) 1486 goto nla_put_failure; 1487 1488 if (t->erspan_ver == 1) { 1489 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index)) 1490 goto nla_put_failure; 1491 } else if (t->erspan_ver == 2) { 1492 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir)) 1493 goto nla_put_failure; 1494 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid)) 1495 goto nla_put_failure; 1496 } 1497 } 1498 1499 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) || 1500 nla_put_be16(skb, IFLA_GRE_IFLAGS, 1501 gre_tnl_flags_to_gre_flags(p->i_flags)) || 1502 nla_put_be16(skb, IFLA_GRE_OFLAGS, 1503 gre_tnl_flags_to_gre_flags(o_flags)) || 1504 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) || 1505 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) || 1506 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) || 1507 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) || 1508 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) || 1509 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) || 1510 nla_put_u8(skb, IFLA_GRE_PMTUDISC, 1511 !!(p->iph.frag_off & htons(IP_DF))) || 1512 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark)) 1513 goto nla_put_failure; 1514 1515 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE, 1516 t->encap.type) || 1517 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT, 1518 t->encap.sport) || 1519 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT, 1520 t->encap.dport) || 1521 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS, 1522 t->encap.flags)) 1523 goto nla_put_failure; 1524 1525 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df)) 1526 goto nla_put_failure; 1527 1528 if (t->collect_md) { 1529 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA)) 1530 goto nla_put_failure; 1531 } 1532 1533 return 0; 1534 1535 nla_put_failure: 1536 return -EMSGSIZE; 1537 } 1538 1539 static void erspan_setup(struct net_device *dev) 1540 { 1541 struct ip_tunnel *t = netdev_priv(dev); 1542 1543 ether_setup(dev); 1544 dev->max_mtu = 0; 1545 dev->netdev_ops = &erspan_netdev_ops; 1546 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1547 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1548 ip_tunnel_setup(dev, erspan_net_id); 1549 t->erspan_ver = 1; 1550 } 1551 1552 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = { 1553 [IFLA_GRE_LINK] = { .type = NLA_U32 }, 1554 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 }, 1555 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 }, 1556 [IFLA_GRE_IKEY] = { .type = NLA_U32 }, 1557 [IFLA_GRE_OKEY] = { .type = NLA_U32 }, 1558 [IFLA_GRE_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 1559 [IFLA_GRE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1560 [IFLA_GRE_TTL] = { .type = NLA_U8 }, 1561 [IFLA_GRE_TOS] = { .type = NLA_U8 }, 1562 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 }, 1563 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 }, 1564 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 }, 1565 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 }, 1566 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 }, 1567 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1568 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 }, 1569 [IFLA_GRE_FWMARK] = { .type = NLA_U32 }, 1570 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 }, 1571 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 }, 1572 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 }, 1573 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 }, 1574 }; 1575 1576 static struct rtnl_link_ops ipgre_link_ops __read_mostly = { 1577 .kind = "gre", 1578 .maxtype = IFLA_GRE_MAX, 1579 .policy = ipgre_policy, 1580 .priv_size = sizeof(struct ip_tunnel), 1581 .setup = ipgre_tunnel_setup, 1582 .validate = ipgre_tunnel_validate, 1583 .newlink = ipgre_newlink, 1584 .changelink = ipgre_changelink, 1585 .dellink = ip_tunnel_dellink, 1586 .get_size = ipgre_get_size, 1587 .fill_info = ipgre_fill_info, 1588 .get_link_net = ip_tunnel_get_link_net, 1589 }; 1590 1591 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = { 1592 .kind = "gretap", 1593 .maxtype = IFLA_GRE_MAX, 1594 .policy = ipgre_policy, 1595 .priv_size = sizeof(struct ip_tunnel), 1596 .setup = ipgre_tap_setup, 1597 .validate = ipgre_tap_validate, 1598 .newlink = ipgre_newlink, 1599 .changelink = ipgre_changelink, 1600 .dellink = ip_tunnel_dellink, 1601 .get_size = ipgre_get_size, 1602 .fill_info = ipgre_fill_info, 1603 .get_link_net = ip_tunnel_get_link_net, 1604 }; 1605 1606 static struct rtnl_link_ops erspan_link_ops __read_mostly = { 1607 .kind = "erspan", 1608 .maxtype = IFLA_GRE_MAX, 1609 .policy = ipgre_policy, 1610 .priv_size = sizeof(struct ip_tunnel), 1611 .setup = erspan_setup, 1612 .validate = erspan_validate, 1613 .newlink = erspan_newlink, 1614 .changelink = erspan_changelink, 1615 .dellink = ip_tunnel_dellink, 1616 .get_size = ipgre_get_size, 1617 .fill_info = ipgre_fill_info, 1618 .get_link_net = ip_tunnel_get_link_net, 1619 }; 1620 1621 struct net_device *gretap_fb_dev_create(struct net *net, const char *name, 1622 u8 name_assign_type) 1623 { 1624 struct nlattr *tb[IFLA_MAX + 1]; 1625 struct net_device *dev; 1626 LIST_HEAD(list_kill); 1627 struct ip_tunnel *t; 1628 int err; 1629 1630 memset(&tb, 0, sizeof(tb)); 1631 1632 dev = rtnl_create_link(net, name, name_assign_type, 1633 &ipgre_tap_ops, tb, NULL); 1634 if (IS_ERR(dev)) 1635 return dev; 1636 1637 /* Configure flow based GRE device. */ 1638 t = netdev_priv(dev); 1639 t->collect_md = true; 1640 1641 err = ipgre_newlink(net, dev, tb, NULL, NULL); 1642 if (err < 0) { 1643 free_netdev(dev); 1644 return ERR_PTR(err); 1645 } 1646 1647 /* openvswitch users expect packet sizes to be unrestricted, 1648 * so set the largest MTU we can. 1649 */ 1650 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false); 1651 if (err) 1652 goto out; 1653 1654 err = rtnl_configure_link(dev, NULL); 1655 if (err < 0) 1656 goto out; 1657 1658 return dev; 1659 out: 1660 ip_tunnel_dellink(dev, &list_kill); 1661 unregister_netdevice_many(&list_kill); 1662 return ERR_PTR(err); 1663 } 1664 EXPORT_SYMBOL_GPL(gretap_fb_dev_create); 1665 1666 static int __net_init ipgre_tap_init_net(struct net *net) 1667 { 1668 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0"); 1669 } 1670 1671 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net) 1672 { 1673 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops); 1674 } 1675 1676 static struct pernet_operations ipgre_tap_net_ops = { 1677 .init = ipgre_tap_init_net, 1678 .exit_batch = ipgre_tap_exit_batch_net, 1679 .id = &gre_tap_net_id, 1680 .size = sizeof(struct ip_tunnel_net), 1681 }; 1682 1683 static int __net_init erspan_init_net(struct net *net) 1684 { 1685 return ip_tunnel_init_net(net, erspan_net_id, 1686 &erspan_link_ops, "erspan0"); 1687 } 1688 1689 static void __net_exit erspan_exit_batch_net(struct list_head *net_list) 1690 { 1691 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops); 1692 } 1693 1694 static struct pernet_operations erspan_net_ops = { 1695 .init = erspan_init_net, 1696 .exit_batch = erspan_exit_batch_net, 1697 .id = &erspan_net_id, 1698 .size = sizeof(struct ip_tunnel_net), 1699 }; 1700 1701 static int __init ipgre_init(void) 1702 { 1703 int err; 1704 1705 pr_info("GRE over IPv4 tunneling driver\n"); 1706 1707 err = register_pernet_device(&ipgre_net_ops); 1708 if (err < 0) 1709 return err; 1710 1711 err = register_pernet_device(&ipgre_tap_net_ops); 1712 if (err < 0) 1713 goto pnet_tap_failed; 1714 1715 err = register_pernet_device(&erspan_net_ops); 1716 if (err < 0) 1717 goto pnet_erspan_failed; 1718 1719 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO); 1720 if (err < 0) { 1721 pr_info("%s: can't add protocol\n", __func__); 1722 goto add_proto_failed; 1723 } 1724 1725 err = rtnl_link_register(&ipgre_link_ops); 1726 if (err < 0) 1727 goto rtnl_link_failed; 1728 1729 err = rtnl_link_register(&ipgre_tap_ops); 1730 if (err < 0) 1731 goto tap_ops_failed; 1732 1733 err = rtnl_link_register(&erspan_link_ops); 1734 if (err < 0) 1735 goto erspan_link_failed; 1736 1737 return 0; 1738 1739 erspan_link_failed: 1740 rtnl_link_unregister(&ipgre_tap_ops); 1741 tap_ops_failed: 1742 rtnl_link_unregister(&ipgre_link_ops); 1743 rtnl_link_failed: 1744 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1745 add_proto_failed: 1746 unregister_pernet_device(&erspan_net_ops); 1747 pnet_erspan_failed: 1748 unregister_pernet_device(&ipgre_tap_net_ops); 1749 pnet_tap_failed: 1750 unregister_pernet_device(&ipgre_net_ops); 1751 return err; 1752 } 1753 1754 static void __exit ipgre_fini(void) 1755 { 1756 rtnl_link_unregister(&ipgre_tap_ops); 1757 rtnl_link_unregister(&ipgre_link_ops); 1758 rtnl_link_unregister(&erspan_link_ops); 1759 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1760 unregister_pernet_device(&ipgre_tap_net_ops); 1761 unregister_pernet_device(&ipgre_net_ops); 1762 unregister_pernet_device(&erspan_net_ops); 1763 } 1764 1765 module_init(ipgre_init); 1766 module_exit(ipgre_fini); 1767 MODULE_LICENSE("GPL"); 1768 MODULE_ALIAS_RTNL_LINK("gre"); 1769 MODULE_ALIAS_RTNL_LINK("gretap"); 1770 MODULE_ALIAS_RTNL_LINK("erspan"); 1771 MODULE_ALIAS_NETDEV("gre0"); 1772 MODULE_ALIAS_NETDEV("gretap0"); 1773 MODULE_ALIAS_NETDEV("erspan0"); 1774