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.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 int thoff; 528 529 tun_info = skb_tunnel_info(skb); 530 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 531 ip_tunnel_info_af(tun_info) != AF_INET)) 532 goto err_free_skb; 533 534 key = &tun_info->key; 535 if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT)) 536 goto err_free_skb; 537 if (tun_info->options_len < sizeof(*md)) 538 goto err_free_skb; 539 md = ip_tunnel_info_opts(tun_info); 540 541 /* ERSPAN has fixed 8 byte GRE header */ 542 version = md->version; 543 tunnel_hlen = 8 + erspan_hdr_len(version); 544 545 if (skb_cow_head(skb, dev->needed_headroom)) 546 goto err_free_skb; 547 548 if (gre_handle_offloads(skb, false)) 549 goto err_free_skb; 550 551 if (skb->len > dev->mtu + dev->hard_header_len) { 552 pskb_trim(skb, dev->mtu + dev->hard_header_len); 553 truncate = true; 554 } 555 556 nhoff = skb_network_header(skb) - skb_mac_header(skb); 557 if (skb->protocol == htons(ETH_P_IP) && 558 (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff)) 559 truncate = true; 560 561 thoff = skb_transport_header(skb) - skb_mac_header(skb); 562 if (skb->protocol == htons(ETH_P_IPV6) && 563 (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff)) 564 truncate = true; 565 566 if (version == 1) { 567 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)), 568 ntohl(md->u.index), truncate, true); 569 proto = htons(ETH_P_ERSPAN); 570 } else if (version == 2) { 571 erspan_build_header_v2(skb, 572 ntohl(tunnel_id_to_key32(key->tun_id)), 573 md->u.md2.dir, 574 get_hwid(&md->u.md2), 575 truncate, true); 576 proto = htons(ETH_P_ERSPAN2); 577 } else { 578 goto err_free_skb; 579 } 580 581 gre_build_header(skb, 8, TUNNEL_SEQ, 582 proto, 0, htonl(atomic_fetch_inc(&tunnel->o_seqno))); 583 584 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen); 585 586 return; 587 588 err_free_skb: 589 kfree_skb(skb); 590 dev->stats.tx_dropped++; 591 } 592 593 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 594 { 595 struct ip_tunnel_info *info = skb_tunnel_info(skb); 596 const struct ip_tunnel_key *key; 597 struct rtable *rt; 598 struct flowi4 fl4; 599 600 if (ip_tunnel_info_af(info) != AF_INET) 601 return -EINVAL; 602 603 key = &info->key; 604 ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src, 605 tunnel_id_to_key32(key->tun_id), 606 key->tos & ~INET_ECN_MASK, dev_net(dev), 0, 607 skb->mark, skb_get_hash(skb)); 608 rt = ip_route_output_key(dev_net(dev), &fl4); 609 if (IS_ERR(rt)) 610 return PTR_ERR(rt); 611 612 ip_rt_put(rt); 613 info->key.u.ipv4.src = fl4.saddr; 614 return 0; 615 } 616 617 static netdev_tx_t ipgre_xmit(struct sk_buff *skb, 618 struct net_device *dev) 619 { 620 struct ip_tunnel *tunnel = netdev_priv(dev); 621 const struct iphdr *tnl_params; 622 623 if (!pskb_inet_may_pull(skb)) 624 goto free_skb; 625 626 if (tunnel->collect_md) { 627 gre_fb_xmit(skb, dev, skb->protocol); 628 return NETDEV_TX_OK; 629 } 630 631 if (dev->header_ops) { 632 if (skb_cow_head(skb, 0)) 633 goto free_skb; 634 635 tnl_params = (const struct iphdr *)skb->data; 636 637 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing 638 * to gre header. 639 */ 640 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr)); 641 skb_reset_mac_header(skb); 642 643 if (skb->ip_summed == CHECKSUM_PARTIAL && 644 skb_checksum_start(skb) < skb->data) 645 goto free_skb; 646 } else { 647 if (skb_cow_head(skb, dev->needed_headroom)) 648 goto free_skb; 649 650 tnl_params = &tunnel->parms.iph; 651 } 652 653 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 654 goto free_skb; 655 656 __gre_xmit(skb, dev, tnl_params, skb->protocol); 657 return NETDEV_TX_OK; 658 659 free_skb: 660 kfree_skb(skb); 661 dev->stats.tx_dropped++; 662 return NETDEV_TX_OK; 663 } 664 665 static netdev_tx_t erspan_xmit(struct sk_buff *skb, 666 struct net_device *dev) 667 { 668 struct ip_tunnel *tunnel = netdev_priv(dev); 669 bool truncate = false; 670 __be16 proto; 671 672 if (!pskb_inet_may_pull(skb)) 673 goto free_skb; 674 675 if (tunnel->collect_md) { 676 erspan_fb_xmit(skb, dev); 677 return NETDEV_TX_OK; 678 } 679 680 if (gre_handle_offloads(skb, false)) 681 goto free_skb; 682 683 if (skb_cow_head(skb, dev->needed_headroom)) 684 goto free_skb; 685 686 if (skb->len > dev->mtu + dev->hard_header_len) { 687 pskb_trim(skb, dev->mtu + dev->hard_header_len); 688 truncate = true; 689 } 690 691 /* Push ERSPAN header */ 692 if (tunnel->erspan_ver == 0) { 693 proto = htons(ETH_P_ERSPAN); 694 tunnel->parms.o_flags &= ~TUNNEL_SEQ; 695 } else if (tunnel->erspan_ver == 1) { 696 erspan_build_header(skb, ntohl(tunnel->parms.o_key), 697 tunnel->index, 698 truncate, true); 699 proto = htons(ETH_P_ERSPAN); 700 } else if (tunnel->erspan_ver == 2) { 701 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key), 702 tunnel->dir, tunnel->hwid, 703 truncate, true); 704 proto = htons(ETH_P_ERSPAN2); 705 } else { 706 goto free_skb; 707 } 708 709 tunnel->parms.o_flags &= ~TUNNEL_KEY; 710 __gre_xmit(skb, dev, &tunnel->parms.iph, proto); 711 return NETDEV_TX_OK; 712 713 free_skb: 714 kfree_skb(skb); 715 dev->stats.tx_dropped++; 716 return NETDEV_TX_OK; 717 } 718 719 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb, 720 struct net_device *dev) 721 { 722 struct ip_tunnel *tunnel = netdev_priv(dev); 723 724 if (!pskb_inet_may_pull(skb)) 725 goto free_skb; 726 727 if (tunnel->collect_md) { 728 gre_fb_xmit(skb, dev, htons(ETH_P_TEB)); 729 return NETDEV_TX_OK; 730 } 731 732 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 733 goto free_skb; 734 735 if (skb_cow_head(skb, dev->needed_headroom)) 736 goto free_skb; 737 738 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB)); 739 return NETDEV_TX_OK; 740 741 free_skb: 742 kfree_skb(skb); 743 dev->stats.tx_dropped++; 744 return NETDEV_TX_OK; 745 } 746 747 static void ipgre_link_update(struct net_device *dev, bool set_mtu) 748 { 749 struct ip_tunnel *tunnel = netdev_priv(dev); 750 __be16 flags; 751 int len; 752 753 len = tunnel->tun_hlen; 754 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 755 len = tunnel->tun_hlen - len; 756 tunnel->hlen = tunnel->hlen + len; 757 758 if (dev->header_ops) 759 dev->hard_header_len += len; 760 else 761 dev->needed_headroom += len; 762 763 if (set_mtu) 764 dev->mtu = max_t(int, dev->mtu - len, 68); 765 766 flags = tunnel->parms.o_flags; 767 768 if (flags & TUNNEL_SEQ || 769 (flags & TUNNEL_CSUM && tunnel->encap.type != TUNNEL_ENCAP_NONE)) { 770 dev->features &= ~NETIF_F_GSO_SOFTWARE; 771 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 772 } else { 773 dev->features |= NETIF_F_GSO_SOFTWARE; 774 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 775 } 776 } 777 778 static int ipgre_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, 779 int cmd) 780 { 781 int err; 782 783 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) { 784 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE || 785 p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) || 786 ((p->i_flags | p->o_flags) & (GRE_VERSION | GRE_ROUTING))) 787 return -EINVAL; 788 } 789 790 p->i_flags = gre_flags_to_tnl_flags(p->i_flags); 791 p->o_flags = gre_flags_to_tnl_flags(p->o_flags); 792 793 err = ip_tunnel_ctl(dev, p, cmd); 794 if (err) 795 return err; 796 797 if (cmd == SIOCCHGTUNNEL) { 798 struct ip_tunnel *t = netdev_priv(dev); 799 800 t->parms.i_flags = p->i_flags; 801 t->parms.o_flags = p->o_flags; 802 803 if (strcmp(dev->rtnl_link_ops->kind, "erspan")) 804 ipgre_link_update(dev, true); 805 } 806 807 p->i_flags = gre_tnl_flags_to_gre_flags(p->i_flags); 808 p->o_flags = gre_tnl_flags_to_gre_flags(p->o_flags); 809 return 0; 810 } 811 812 /* Nice toy. Unfortunately, useless in real life :-) 813 It allows to construct virtual multiprotocol broadcast "LAN" 814 over the Internet, provided multicast routing is tuned. 815 816 817 I have no idea was this bicycle invented before me, 818 so that I had to set ARPHRD_IPGRE to a random value. 819 I have an impression, that Cisco could make something similar, 820 but this feature is apparently missing in IOS<=11.2(8). 821 822 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks 823 with broadcast 224.66.66.66. If you have access to mbone, play with me :-) 824 825 ping -t 255 224.66.66.66 826 827 If nobody answers, mbone does not work. 828 829 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255 830 ip addr add 10.66.66.<somewhat>/24 dev Universe 831 ifconfig Universe up 832 ifconfig Universe add fe80::<Your_real_addr>/10 833 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96 834 ftp 10.66.66.66 835 ... 836 ftp fec0:6666:6666::193.233.7.65 837 ... 838 */ 839 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, 840 unsigned short type, 841 const void *daddr, const void *saddr, unsigned int len) 842 { 843 struct ip_tunnel *t = netdev_priv(dev); 844 struct iphdr *iph; 845 struct gre_base_hdr *greh; 846 847 iph = skb_push(skb, t->hlen + sizeof(*iph)); 848 greh = (struct gre_base_hdr *)(iph+1); 849 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags); 850 greh->protocol = htons(type); 851 852 memcpy(iph, &t->parms.iph, sizeof(struct iphdr)); 853 854 /* Set the source hardware address. */ 855 if (saddr) 856 memcpy(&iph->saddr, saddr, 4); 857 if (daddr) 858 memcpy(&iph->daddr, daddr, 4); 859 if (iph->daddr) 860 return t->hlen + sizeof(*iph); 861 862 return -(t->hlen + sizeof(*iph)); 863 } 864 865 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr) 866 { 867 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb); 868 memcpy(haddr, &iph->saddr, 4); 869 return 4; 870 } 871 872 static const struct header_ops ipgre_header_ops = { 873 .create = ipgre_header, 874 .parse = ipgre_header_parse, 875 }; 876 877 #ifdef CONFIG_NET_IPGRE_BROADCAST 878 static int ipgre_open(struct net_device *dev) 879 { 880 struct ip_tunnel *t = netdev_priv(dev); 881 882 if (ipv4_is_multicast(t->parms.iph.daddr)) { 883 struct flowi4 fl4; 884 struct rtable *rt; 885 886 rt = ip_route_output_gre(t->net, &fl4, 887 t->parms.iph.daddr, 888 t->parms.iph.saddr, 889 t->parms.o_key, 890 RT_TOS(t->parms.iph.tos), 891 t->parms.link); 892 if (IS_ERR(rt)) 893 return -EADDRNOTAVAIL; 894 dev = rt->dst.dev; 895 ip_rt_put(rt); 896 if (!__in_dev_get_rtnl(dev)) 897 return -EADDRNOTAVAIL; 898 t->mlink = dev->ifindex; 899 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr); 900 } 901 return 0; 902 } 903 904 static int ipgre_close(struct net_device *dev) 905 { 906 struct ip_tunnel *t = netdev_priv(dev); 907 908 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) { 909 struct in_device *in_dev; 910 in_dev = inetdev_by_index(t->net, t->mlink); 911 if (in_dev) 912 ip_mc_dec_group(in_dev, t->parms.iph.daddr); 913 } 914 return 0; 915 } 916 #endif 917 918 static const struct net_device_ops ipgre_netdev_ops = { 919 .ndo_init = ipgre_tunnel_init, 920 .ndo_uninit = ip_tunnel_uninit, 921 #ifdef CONFIG_NET_IPGRE_BROADCAST 922 .ndo_open = ipgre_open, 923 .ndo_stop = ipgre_close, 924 #endif 925 .ndo_start_xmit = ipgre_xmit, 926 .ndo_siocdevprivate = ip_tunnel_siocdevprivate, 927 .ndo_change_mtu = ip_tunnel_change_mtu, 928 .ndo_get_stats64 = dev_get_tstats64, 929 .ndo_get_iflink = ip_tunnel_get_iflink, 930 .ndo_tunnel_ctl = ipgre_tunnel_ctl, 931 }; 932 933 #define GRE_FEATURES (NETIF_F_SG | \ 934 NETIF_F_FRAGLIST | \ 935 NETIF_F_HIGHDMA | \ 936 NETIF_F_HW_CSUM) 937 938 static void ipgre_tunnel_setup(struct net_device *dev) 939 { 940 dev->netdev_ops = &ipgre_netdev_ops; 941 dev->type = ARPHRD_IPGRE; 942 ip_tunnel_setup(dev, ipgre_net_id); 943 } 944 945 static void __gre_tunnel_init(struct net_device *dev) 946 { 947 struct ip_tunnel *tunnel; 948 __be16 flags; 949 950 tunnel = netdev_priv(dev); 951 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 952 tunnel->parms.iph.protocol = IPPROTO_GRE; 953 954 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen; 955 dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph); 956 957 dev->features |= GRE_FEATURES | NETIF_F_LLTX; 958 dev->hw_features |= GRE_FEATURES; 959 960 flags = tunnel->parms.o_flags; 961 962 /* TCP offload with GRE SEQ is not supported, nor can we support 2 963 * levels of outer headers requiring an update. 964 */ 965 if (flags & TUNNEL_SEQ) 966 return; 967 if (flags & TUNNEL_CSUM && tunnel->encap.type != TUNNEL_ENCAP_NONE) 968 return; 969 970 dev->features |= NETIF_F_GSO_SOFTWARE; 971 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 972 } 973 974 static int ipgre_tunnel_init(struct net_device *dev) 975 { 976 struct ip_tunnel *tunnel = netdev_priv(dev); 977 struct iphdr *iph = &tunnel->parms.iph; 978 979 __gre_tunnel_init(dev); 980 981 __dev_addr_set(dev, &iph->saddr, 4); 982 memcpy(dev->broadcast, &iph->daddr, 4); 983 984 dev->flags = IFF_NOARP; 985 netif_keep_dst(dev); 986 dev->addr_len = 4; 987 988 if (iph->daddr && !tunnel->collect_md) { 989 #ifdef CONFIG_NET_IPGRE_BROADCAST 990 if (ipv4_is_multicast(iph->daddr)) { 991 if (!iph->saddr) 992 return -EINVAL; 993 dev->flags = IFF_BROADCAST; 994 dev->header_ops = &ipgre_header_ops; 995 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 996 dev->needed_headroom = 0; 997 } 998 #endif 999 } else if (!tunnel->collect_md) { 1000 dev->header_ops = &ipgre_header_ops; 1001 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 1002 dev->needed_headroom = 0; 1003 } 1004 1005 return ip_tunnel_init(dev); 1006 } 1007 1008 static const struct gre_protocol ipgre_protocol = { 1009 .handler = gre_rcv, 1010 .err_handler = gre_err, 1011 }; 1012 1013 static int __net_init ipgre_init_net(struct net *net) 1014 { 1015 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL); 1016 } 1017 1018 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net) 1019 { 1020 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops); 1021 } 1022 1023 static struct pernet_operations ipgre_net_ops = { 1024 .init = ipgre_init_net, 1025 .exit_batch = ipgre_exit_batch_net, 1026 .id = &ipgre_net_id, 1027 .size = sizeof(struct ip_tunnel_net), 1028 }; 1029 1030 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[], 1031 struct netlink_ext_ack *extack) 1032 { 1033 __be16 flags; 1034 1035 if (!data) 1036 return 0; 1037 1038 flags = 0; 1039 if (data[IFLA_GRE_IFLAGS]) 1040 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1041 if (data[IFLA_GRE_OFLAGS]) 1042 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1043 if (flags & (GRE_VERSION|GRE_ROUTING)) 1044 return -EINVAL; 1045 1046 if (data[IFLA_GRE_COLLECT_METADATA] && 1047 data[IFLA_GRE_ENCAP_TYPE] && 1048 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE) 1049 return -EINVAL; 1050 1051 return 0; 1052 } 1053 1054 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[], 1055 struct netlink_ext_ack *extack) 1056 { 1057 __be32 daddr; 1058 1059 if (tb[IFLA_ADDRESS]) { 1060 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1061 return -EINVAL; 1062 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1063 return -EADDRNOTAVAIL; 1064 } 1065 1066 if (!data) 1067 goto out; 1068 1069 if (data[IFLA_GRE_REMOTE]) { 1070 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4); 1071 if (!daddr) 1072 return -EINVAL; 1073 } 1074 1075 out: 1076 return ipgre_tunnel_validate(tb, data, extack); 1077 } 1078 1079 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[], 1080 struct netlink_ext_ack *extack) 1081 { 1082 __be16 flags = 0; 1083 int ret; 1084 1085 if (!data) 1086 return 0; 1087 1088 ret = ipgre_tap_validate(tb, data, extack); 1089 if (ret) 1090 return ret; 1091 1092 if (data[IFLA_GRE_ERSPAN_VER] && 1093 nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0) 1094 return 0; 1095 1096 /* ERSPAN type II/III should only have GRE sequence and key flag */ 1097 if (data[IFLA_GRE_OFLAGS]) 1098 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1099 if (data[IFLA_GRE_IFLAGS]) 1100 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1101 if (!data[IFLA_GRE_COLLECT_METADATA] && 1102 flags != (GRE_SEQ | GRE_KEY)) 1103 return -EINVAL; 1104 1105 /* ERSPAN Session ID only has 10-bit. Since we reuse 1106 * 32-bit key field as ID, check it's range. 1107 */ 1108 if (data[IFLA_GRE_IKEY] && 1109 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK)) 1110 return -EINVAL; 1111 1112 if (data[IFLA_GRE_OKEY] && 1113 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK)) 1114 return -EINVAL; 1115 1116 return 0; 1117 } 1118 1119 static int ipgre_netlink_parms(struct net_device *dev, 1120 struct nlattr *data[], 1121 struct nlattr *tb[], 1122 struct ip_tunnel_parm *parms, 1123 __u32 *fwmark) 1124 { 1125 struct ip_tunnel *t = netdev_priv(dev); 1126 1127 memset(parms, 0, sizeof(*parms)); 1128 1129 parms->iph.protocol = IPPROTO_GRE; 1130 1131 if (!data) 1132 return 0; 1133 1134 if (data[IFLA_GRE_LINK]) 1135 parms->link = nla_get_u32(data[IFLA_GRE_LINK]); 1136 1137 if (data[IFLA_GRE_IFLAGS]) 1138 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS])); 1139 1140 if (data[IFLA_GRE_OFLAGS]) 1141 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS])); 1142 1143 if (data[IFLA_GRE_IKEY]) 1144 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]); 1145 1146 if (data[IFLA_GRE_OKEY]) 1147 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]); 1148 1149 if (data[IFLA_GRE_LOCAL]) 1150 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]); 1151 1152 if (data[IFLA_GRE_REMOTE]) 1153 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]); 1154 1155 if (data[IFLA_GRE_TTL]) 1156 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]); 1157 1158 if (data[IFLA_GRE_TOS]) 1159 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]); 1160 1161 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) { 1162 if (t->ignore_df) 1163 return -EINVAL; 1164 parms->iph.frag_off = htons(IP_DF); 1165 } 1166 1167 if (data[IFLA_GRE_COLLECT_METADATA]) { 1168 t->collect_md = true; 1169 if (dev->type == ARPHRD_IPGRE) 1170 dev->type = ARPHRD_NONE; 1171 } 1172 1173 if (data[IFLA_GRE_IGNORE_DF]) { 1174 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF]) 1175 && (parms->iph.frag_off & htons(IP_DF))) 1176 return -EINVAL; 1177 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]); 1178 } 1179 1180 if (data[IFLA_GRE_FWMARK]) 1181 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]); 1182 1183 return 0; 1184 } 1185 1186 static int erspan_netlink_parms(struct net_device *dev, 1187 struct nlattr *data[], 1188 struct nlattr *tb[], 1189 struct ip_tunnel_parm *parms, 1190 __u32 *fwmark) 1191 { 1192 struct ip_tunnel *t = netdev_priv(dev); 1193 int err; 1194 1195 err = ipgre_netlink_parms(dev, data, tb, parms, fwmark); 1196 if (err) 1197 return err; 1198 if (!data) 1199 return 0; 1200 1201 if (data[IFLA_GRE_ERSPAN_VER]) { 1202 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]); 1203 1204 if (t->erspan_ver > 2) 1205 return -EINVAL; 1206 } 1207 1208 if (t->erspan_ver == 1) { 1209 if (data[IFLA_GRE_ERSPAN_INDEX]) { 1210 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]); 1211 if (t->index & ~INDEX_MASK) 1212 return -EINVAL; 1213 } 1214 } else if (t->erspan_ver == 2) { 1215 if (data[IFLA_GRE_ERSPAN_DIR]) { 1216 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]); 1217 if (t->dir & ~(DIR_MASK >> DIR_OFFSET)) 1218 return -EINVAL; 1219 } 1220 if (data[IFLA_GRE_ERSPAN_HWID]) { 1221 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]); 1222 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET)) 1223 return -EINVAL; 1224 } 1225 } 1226 1227 return 0; 1228 } 1229 1230 /* This function returns true when ENCAP attributes are present in the nl msg */ 1231 static bool ipgre_netlink_encap_parms(struct nlattr *data[], 1232 struct ip_tunnel_encap *ipencap) 1233 { 1234 bool ret = false; 1235 1236 memset(ipencap, 0, sizeof(*ipencap)); 1237 1238 if (!data) 1239 return ret; 1240 1241 if (data[IFLA_GRE_ENCAP_TYPE]) { 1242 ret = true; 1243 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]); 1244 } 1245 1246 if (data[IFLA_GRE_ENCAP_FLAGS]) { 1247 ret = true; 1248 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]); 1249 } 1250 1251 if (data[IFLA_GRE_ENCAP_SPORT]) { 1252 ret = true; 1253 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]); 1254 } 1255 1256 if (data[IFLA_GRE_ENCAP_DPORT]) { 1257 ret = true; 1258 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]); 1259 } 1260 1261 return ret; 1262 } 1263 1264 static int gre_tap_init(struct net_device *dev) 1265 { 1266 __gre_tunnel_init(dev); 1267 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1268 netif_keep_dst(dev); 1269 1270 return ip_tunnel_init(dev); 1271 } 1272 1273 static const struct net_device_ops gre_tap_netdev_ops = { 1274 .ndo_init = gre_tap_init, 1275 .ndo_uninit = ip_tunnel_uninit, 1276 .ndo_start_xmit = gre_tap_xmit, 1277 .ndo_set_mac_address = eth_mac_addr, 1278 .ndo_validate_addr = eth_validate_addr, 1279 .ndo_change_mtu = ip_tunnel_change_mtu, 1280 .ndo_get_stats64 = dev_get_tstats64, 1281 .ndo_get_iflink = ip_tunnel_get_iflink, 1282 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1283 }; 1284 1285 static int erspan_tunnel_init(struct net_device *dev) 1286 { 1287 struct ip_tunnel *tunnel = netdev_priv(dev); 1288 1289 if (tunnel->erspan_ver == 0) 1290 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */ 1291 else 1292 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */ 1293 1294 tunnel->parms.iph.protocol = IPPROTO_GRE; 1295 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen + 1296 erspan_hdr_len(tunnel->erspan_ver); 1297 1298 dev->features |= GRE_FEATURES; 1299 dev->hw_features |= GRE_FEATURES; 1300 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1301 netif_keep_dst(dev); 1302 1303 return ip_tunnel_init(dev); 1304 } 1305 1306 static const struct net_device_ops erspan_netdev_ops = { 1307 .ndo_init = erspan_tunnel_init, 1308 .ndo_uninit = ip_tunnel_uninit, 1309 .ndo_start_xmit = erspan_xmit, 1310 .ndo_set_mac_address = eth_mac_addr, 1311 .ndo_validate_addr = eth_validate_addr, 1312 .ndo_change_mtu = ip_tunnel_change_mtu, 1313 .ndo_get_stats64 = dev_get_tstats64, 1314 .ndo_get_iflink = ip_tunnel_get_iflink, 1315 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1316 }; 1317 1318 static void ipgre_tap_setup(struct net_device *dev) 1319 { 1320 ether_setup(dev); 1321 dev->max_mtu = 0; 1322 dev->netdev_ops = &gre_tap_netdev_ops; 1323 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1324 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1325 ip_tunnel_setup(dev, gre_tap_net_id); 1326 } 1327 1328 static int 1329 ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[]) 1330 { 1331 struct ip_tunnel_encap ipencap; 1332 1333 if (ipgre_netlink_encap_parms(data, &ipencap)) { 1334 struct ip_tunnel *t = netdev_priv(dev); 1335 int err = ip_tunnel_encap_setup(t, &ipencap); 1336 1337 if (err < 0) 1338 return err; 1339 } 1340 1341 return 0; 1342 } 1343 1344 static int ipgre_newlink(struct net *src_net, struct net_device *dev, 1345 struct nlattr *tb[], struct nlattr *data[], 1346 struct netlink_ext_ack *extack) 1347 { 1348 struct ip_tunnel_parm p; 1349 __u32 fwmark = 0; 1350 int err; 1351 1352 err = ipgre_newlink_encap_setup(dev, data); 1353 if (err) 1354 return err; 1355 1356 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1357 if (err < 0) 1358 return err; 1359 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1360 } 1361 1362 static int erspan_newlink(struct net *src_net, struct net_device *dev, 1363 struct nlattr *tb[], struct nlattr *data[], 1364 struct netlink_ext_ack *extack) 1365 { 1366 struct ip_tunnel_parm p; 1367 __u32 fwmark = 0; 1368 int err; 1369 1370 err = ipgre_newlink_encap_setup(dev, data); 1371 if (err) 1372 return err; 1373 1374 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1375 if (err) 1376 return err; 1377 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1378 } 1379 1380 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[], 1381 struct nlattr *data[], 1382 struct netlink_ext_ack *extack) 1383 { 1384 struct ip_tunnel *t = netdev_priv(dev); 1385 __u32 fwmark = t->fwmark; 1386 struct ip_tunnel_parm p; 1387 int err; 1388 1389 err = ipgre_newlink_encap_setup(dev, data); 1390 if (err) 1391 return err; 1392 1393 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1394 if (err < 0) 1395 return err; 1396 1397 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1398 if (err < 0) 1399 return err; 1400 1401 t->parms.i_flags = p.i_flags; 1402 t->parms.o_flags = p.o_flags; 1403 1404 ipgre_link_update(dev, !tb[IFLA_MTU]); 1405 1406 return 0; 1407 } 1408 1409 static int erspan_changelink(struct net_device *dev, struct nlattr *tb[], 1410 struct nlattr *data[], 1411 struct netlink_ext_ack *extack) 1412 { 1413 struct ip_tunnel *t = netdev_priv(dev); 1414 __u32 fwmark = t->fwmark; 1415 struct ip_tunnel_parm p; 1416 int err; 1417 1418 err = ipgre_newlink_encap_setup(dev, data); 1419 if (err) 1420 return err; 1421 1422 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1423 if (err < 0) 1424 return err; 1425 1426 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1427 if (err < 0) 1428 return err; 1429 1430 t->parms.i_flags = p.i_flags; 1431 t->parms.o_flags = p.o_flags; 1432 1433 return 0; 1434 } 1435 1436 static size_t ipgre_get_size(const struct net_device *dev) 1437 { 1438 return 1439 /* IFLA_GRE_LINK */ 1440 nla_total_size(4) + 1441 /* IFLA_GRE_IFLAGS */ 1442 nla_total_size(2) + 1443 /* IFLA_GRE_OFLAGS */ 1444 nla_total_size(2) + 1445 /* IFLA_GRE_IKEY */ 1446 nla_total_size(4) + 1447 /* IFLA_GRE_OKEY */ 1448 nla_total_size(4) + 1449 /* IFLA_GRE_LOCAL */ 1450 nla_total_size(4) + 1451 /* IFLA_GRE_REMOTE */ 1452 nla_total_size(4) + 1453 /* IFLA_GRE_TTL */ 1454 nla_total_size(1) + 1455 /* IFLA_GRE_TOS */ 1456 nla_total_size(1) + 1457 /* IFLA_GRE_PMTUDISC */ 1458 nla_total_size(1) + 1459 /* IFLA_GRE_ENCAP_TYPE */ 1460 nla_total_size(2) + 1461 /* IFLA_GRE_ENCAP_FLAGS */ 1462 nla_total_size(2) + 1463 /* IFLA_GRE_ENCAP_SPORT */ 1464 nla_total_size(2) + 1465 /* IFLA_GRE_ENCAP_DPORT */ 1466 nla_total_size(2) + 1467 /* IFLA_GRE_COLLECT_METADATA */ 1468 nla_total_size(0) + 1469 /* IFLA_GRE_IGNORE_DF */ 1470 nla_total_size(1) + 1471 /* IFLA_GRE_FWMARK */ 1472 nla_total_size(4) + 1473 /* IFLA_GRE_ERSPAN_INDEX */ 1474 nla_total_size(4) + 1475 /* IFLA_GRE_ERSPAN_VER */ 1476 nla_total_size(1) + 1477 /* IFLA_GRE_ERSPAN_DIR */ 1478 nla_total_size(1) + 1479 /* IFLA_GRE_ERSPAN_HWID */ 1480 nla_total_size(2) + 1481 0; 1482 } 1483 1484 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev) 1485 { 1486 struct ip_tunnel *t = netdev_priv(dev); 1487 struct ip_tunnel_parm *p = &t->parms; 1488 __be16 o_flags = p->o_flags; 1489 1490 if (t->erspan_ver <= 2) { 1491 if (t->erspan_ver != 0 && !t->collect_md) 1492 o_flags |= TUNNEL_KEY; 1493 1494 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver)) 1495 goto nla_put_failure; 1496 1497 if (t->erspan_ver == 1) { 1498 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index)) 1499 goto nla_put_failure; 1500 } else if (t->erspan_ver == 2) { 1501 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir)) 1502 goto nla_put_failure; 1503 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid)) 1504 goto nla_put_failure; 1505 } 1506 } 1507 1508 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) || 1509 nla_put_be16(skb, IFLA_GRE_IFLAGS, 1510 gre_tnl_flags_to_gre_flags(p->i_flags)) || 1511 nla_put_be16(skb, IFLA_GRE_OFLAGS, 1512 gre_tnl_flags_to_gre_flags(o_flags)) || 1513 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) || 1514 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) || 1515 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) || 1516 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) || 1517 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) || 1518 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) || 1519 nla_put_u8(skb, IFLA_GRE_PMTUDISC, 1520 !!(p->iph.frag_off & htons(IP_DF))) || 1521 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark)) 1522 goto nla_put_failure; 1523 1524 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE, 1525 t->encap.type) || 1526 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT, 1527 t->encap.sport) || 1528 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT, 1529 t->encap.dport) || 1530 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS, 1531 t->encap.flags)) 1532 goto nla_put_failure; 1533 1534 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df)) 1535 goto nla_put_failure; 1536 1537 if (t->collect_md) { 1538 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA)) 1539 goto nla_put_failure; 1540 } 1541 1542 return 0; 1543 1544 nla_put_failure: 1545 return -EMSGSIZE; 1546 } 1547 1548 static void erspan_setup(struct net_device *dev) 1549 { 1550 struct ip_tunnel *t = netdev_priv(dev); 1551 1552 ether_setup(dev); 1553 dev->max_mtu = 0; 1554 dev->netdev_ops = &erspan_netdev_ops; 1555 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1556 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1557 ip_tunnel_setup(dev, erspan_net_id); 1558 t->erspan_ver = 1; 1559 } 1560 1561 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = { 1562 [IFLA_GRE_LINK] = { .type = NLA_U32 }, 1563 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 }, 1564 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 }, 1565 [IFLA_GRE_IKEY] = { .type = NLA_U32 }, 1566 [IFLA_GRE_OKEY] = { .type = NLA_U32 }, 1567 [IFLA_GRE_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 1568 [IFLA_GRE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1569 [IFLA_GRE_TTL] = { .type = NLA_U8 }, 1570 [IFLA_GRE_TOS] = { .type = NLA_U8 }, 1571 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 }, 1572 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 }, 1573 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 }, 1574 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 }, 1575 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 }, 1576 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1577 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 }, 1578 [IFLA_GRE_FWMARK] = { .type = NLA_U32 }, 1579 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 }, 1580 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 }, 1581 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 }, 1582 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 }, 1583 }; 1584 1585 static struct rtnl_link_ops ipgre_link_ops __read_mostly = { 1586 .kind = "gre", 1587 .maxtype = IFLA_GRE_MAX, 1588 .policy = ipgre_policy, 1589 .priv_size = sizeof(struct ip_tunnel), 1590 .setup = ipgre_tunnel_setup, 1591 .validate = ipgre_tunnel_validate, 1592 .newlink = ipgre_newlink, 1593 .changelink = ipgre_changelink, 1594 .dellink = ip_tunnel_dellink, 1595 .get_size = ipgre_get_size, 1596 .fill_info = ipgre_fill_info, 1597 .get_link_net = ip_tunnel_get_link_net, 1598 }; 1599 1600 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = { 1601 .kind = "gretap", 1602 .maxtype = IFLA_GRE_MAX, 1603 .policy = ipgre_policy, 1604 .priv_size = sizeof(struct ip_tunnel), 1605 .setup = ipgre_tap_setup, 1606 .validate = ipgre_tap_validate, 1607 .newlink = ipgre_newlink, 1608 .changelink = ipgre_changelink, 1609 .dellink = ip_tunnel_dellink, 1610 .get_size = ipgre_get_size, 1611 .fill_info = ipgre_fill_info, 1612 .get_link_net = ip_tunnel_get_link_net, 1613 }; 1614 1615 static struct rtnl_link_ops erspan_link_ops __read_mostly = { 1616 .kind = "erspan", 1617 .maxtype = IFLA_GRE_MAX, 1618 .policy = ipgre_policy, 1619 .priv_size = sizeof(struct ip_tunnel), 1620 .setup = erspan_setup, 1621 .validate = erspan_validate, 1622 .newlink = erspan_newlink, 1623 .changelink = erspan_changelink, 1624 .dellink = ip_tunnel_dellink, 1625 .get_size = ipgre_get_size, 1626 .fill_info = ipgre_fill_info, 1627 .get_link_net = ip_tunnel_get_link_net, 1628 }; 1629 1630 struct net_device *gretap_fb_dev_create(struct net *net, const char *name, 1631 u8 name_assign_type) 1632 { 1633 struct nlattr *tb[IFLA_MAX + 1]; 1634 struct net_device *dev; 1635 LIST_HEAD(list_kill); 1636 struct ip_tunnel *t; 1637 int err; 1638 1639 memset(&tb, 0, sizeof(tb)); 1640 1641 dev = rtnl_create_link(net, name, name_assign_type, 1642 &ipgre_tap_ops, tb, NULL); 1643 if (IS_ERR(dev)) 1644 return dev; 1645 1646 /* Configure flow based GRE device. */ 1647 t = netdev_priv(dev); 1648 t->collect_md = true; 1649 1650 err = ipgre_newlink(net, dev, tb, NULL, NULL); 1651 if (err < 0) { 1652 free_netdev(dev); 1653 return ERR_PTR(err); 1654 } 1655 1656 /* openvswitch users expect packet sizes to be unrestricted, 1657 * so set the largest MTU we can. 1658 */ 1659 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false); 1660 if (err) 1661 goto out; 1662 1663 err = rtnl_configure_link(dev, NULL); 1664 if (err < 0) 1665 goto out; 1666 1667 return dev; 1668 out: 1669 ip_tunnel_dellink(dev, &list_kill); 1670 unregister_netdevice_many(&list_kill); 1671 return ERR_PTR(err); 1672 } 1673 EXPORT_SYMBOL_GPL(gretap_fb_dev_create); 1674 1675 static int __net_init ipgre_tap_init_net(struct net *net) 1676 { 1677 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0"); 1678 } 1679 1680 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net) 1681 { 1682 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops); 1683 } 1684 1685 static struct pernet_operations ipgre_tap_net_ops = { 1686 .init = ipgre_tap_init_net, 1687 .exit_batch = ipgre_tap_exit_batch_net, 1688 .id = &gre_tap_net_id, 1689 .size = sizeof(struct ip_tunnel_net), 1690 }; 1691 1692 static int __net_init erspan_init_net(struct net *net) 1693 { 1694 return ip_tunnel_init_net(net, erspan_net_id, 1695 &erspan_link_ops, "erspan0"); 1696 } 1697 1698 static void __net_exit erspan_exit_batch_net(struct list_head *net_list) 1699 { 1700 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops); 1701 } 1702 1703 static struct pernet_operations erspan_net_ops = { 1704 .init = erspan_init_net, 1705 .exit_batch = erspan_exit_batch_net, 1706 .id = &erspan_net_id, 1707 .size = sizeof(struct ip_tunnel_net), 1708 }; 1709 1710 static int __init ipgre_init(void) 1711 { 1712 int err; 1713 1714 pr_info("GRE over IPv4 tunneling driver\n"); 1715 1716 err = register_pernet_device(&ipgre_net_ops); 1717 if (err < 0) 1718 return err; 1719 1720 err = register_pernet_device(&ipgre_tap_net_ops); 1721 if (err < 0) 1722 goto pnet_tap_failed; 1723 1724 err = register_pernet_device(&erspan_net_ops); 1725 if (err < 0) 1726 goto pnet_erspan_failed; 1727 1728 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO); 1729 if (err < 0) { 1730 pr_info("%s: can't add protocol\n", __func__); 1731 goto add_proto_failed; 1732 } 1733 1734 err = rtnl_link_register(&ipgre_link_ops); 1735 if (err < 0) 1736 goto rtnl_link_failed; 1737 1738 err = rtnl_link_register(&ipgre_tap_ops); 1739 if (err < 0) 1740 goto tap_ops_failed; 1741 1742 err = rtnl_link_register(&erspan_link_ops); 1743 if (err < 0) 1744 goto erspan_link_failed; 1745 1746 return 0; 1747 1748 erspan_link_failed: 1749 rtnl_link_unregister(&ipgre_tap_ops); 1750 tap_ops_failed: 1751 rtnl_link_unregister(&ipgre_link_ops); 1752 rtnl_link_failed: 1753 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1754 add_proto_failed: 1755 unregister_pernet_device(&erspan_net_ops); 1756 pnet_erspan_failed: 1757 unregister_pernet_device(&ipgre_tap_net_ops); 1758 pnet_tap_failed: 1759 unregister_pernet_device(&ipgre_net_ops); 1760 return err; 1761 } 1762 1763 static void __exit ipgre_fini(void) 1764 { 1765 rtnl_link_unregister(&ipgre_tap_ops); 1766 rtnl_link_unregister(&ipgre_link_ops); 1767 rtnl_link_unregister(&erspan_link_ops); 1768 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1769 unregister_pernet_device(&ipgre_tap_net_ops); 1770 unregister_pernet_device(&ipgre_net_ops); 1771 unregister_pernet_device(&erspan_net_ops); 1772 } 1773 1774 module_init(ipgre_init); 1775 module_exit(ipgre_fini); 1776 MODULE_LICENSE("GPL"); 1777 MODULE_ALIAS_RTNL_LINK("gre"); 1778 MODULE_ALIAS_RTNL_LINK("gretap"); 1779 MODULE_ALIAS_RTNL_LINK("erspan"); 1780 MODULE_ALIAS_NETDEV("gre0"); 1781 MODULE_ALIAS_NETDEV("gretap0"); 1782 MODULE_ALIAS_NETDEV("erspan0"); 1783