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 const int pull_len = tunnel->hlen + sizeof(struct iphdr); 633 634 if (skb_cow_head(skb, 0)) 635 goto free_skb; 636 637 tnl_params = (const struct iphdr *)skb->data; 638 639 if (pull_len > skb_transport_offset(skb)) 640 goto free_skb; 641 642 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing 643 * to gre header. 644 */ 645 skb_pull(skb, pull_len); 646 skb_reset_mac_header(skb); 647 } else { 648 if (skb_cow_head(skb, dev->needed_headroom)) 649 goto free_skb; 650 651 tnl_params = &tunnel->parms.iph; 652 } 653 654 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 655 goto free_skb; 656 657 __gre_xmit(skb, dev, tnl_params, skb->protocol); 658 return NETDEV_TX_OK; 659 660 free_skb: 661 kfree_skb(skb); 662 dev->stats.tx_dropped++; 663 return NETDEV_TX_OK; 664 } 665 666 static netdev_tx_t erspan_xmit(struct sk_buff *skb, 667 struct net_device *dev) 668 { 669 struct ip_tunnel *tunnel = netdev_priv(dev); 670 bool truncate = false; 671 __be16 proto; 672 673 if (!pskb_inet_may_pull(skb)) 674 goto free_skb; 675 676 if (tunnel->collect_md) { 677 erspan_fb_xmit(skb, dev); 678 return NETDEV_TX_OK; 679 } 680 681 if (gre_handle_offloads(skb, false)) 682 goto free_skb; 683 684 if (skb_cow_head(skb, dev->needed_headroom)) 685 goto free_skb; 686 687 if (skb->len > dev->mtu + dev->hard_header_len) { 688 pskb_trim(skb, dev->mtu + dev->hard_header_len); 689 truncate = true; 690 } 691 692 /* Push ERSPAN header */ 693 if (tunnel->erspan_ver == 0) { 694 proto = htons(ETH_P_ERSPAN); 695 tunnel->parms.o_flags &= ~TUNNEL_SEQ; 696 } else if (tunnel->erspan_ver == 1) { 697 erspan_build_header(skb, ntohl(tunnel->parms.o_key), 698 tunnel->index, 699 truncate, true); 700 proto = htons(ETH_P_ERSPAN); 701 } else if (tunnel->erspan_ver == 2) { 702 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key), 703 tunnel->dir, tunnel->hwid, 704 truncate, true); 705 proto = htons(ETH_P_ERSPAN2); 706 } else { 707 goto free_skb; 708 } 709 710 tunnel->parms.o_flags &= ~TUNNEL_KEY; 711 __gre_xmit(skb, dev, &tunnel->parms.iph, proto); 712 return NETDEV_TX_OK; 713 714 free_skb: 715 kfree_skb(skb); 716 dev->stats.tx_dropped++; 717 return NETDEV_TX_OK; 718 } 719 720 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb, 721 struct net_device *dev) 722 { 723 struct ip_tunnel *tunnel = netdev_priv(dev); 724 725 if (!pskb_inet_may_pull(skb)) 726 goto free_skb; 727 728 if (tunnel->collect_md) { 729 gre_fb_xmit(skb, dev, htons(ETH_P_TEB)); 730 return NETDEV_TX_OK; 731 } 732 733 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 734 goto free_skb; 735 736 if (skb_cow_head(skb, dev->needed_headroom)) 737 goto free_skb; 738 739 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB)); 740 return NETDEV_TX_OK; 741 742 free_skb: 743 kfree_skb(skb); 744 dev->stats.tx_dropped++; 745 return NETDEV_TX_OK; 746 } 747 748 static void ipgre_link_update(struct net_device *dev, bool set_mtu) 749 { 750 struct ip_tunnel *tunnel = netdev_priv(dev); 751 __be16 flags; 752 int len; 753 754 len = tunnel->tun_hlen; 755 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 756 len = tunnel->tun_hlen - len; 757 tunnel->hlen = tunnel->hlen + len; 758 759 if (dev->header_ops) 760 dev->hard_header_len += len; 761 else 762 dev->needed_headroom += len; 763 764 if (set_mtu) 765 dev->mtu = max_t(int, dev->mtu - len, 68); 766 767 flags = tunnel->parms.o_flags; 768 769 if (flags & TUNNEL_SEQ || 770 (flags & TUNNEL_CSUM && tunnel->encap.type != TUNNEL_ENCAP_NONE)) { 771 dev->features &= ~NETIF_F_GSO_SOFTWARE; 772 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 773 } else { 774 dev->features |= NETIF_F_GSO_SOFTWARE; 775 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 776 } 777 } 778 779 static int ipgre_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, 780 int cmd) 781 { 782 int err; 783 784 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) { 785 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE || 786 p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) || 787 ((p->i_flags | p->o_flags) & (GRE_VERSION | GRE_ROUTING))) 788 return -EINVAL; 789 } 790 791 p->i_flags = gre_flags_to_tnl_flags(p->i_flags); 792 p->o_flags = gre_flags_to_tnl_flags(p->o_flags); 793 794 err = ip_tunnel_ctl(dev, p, cmd); 795 if (err) 796 return err; 797 798 if (cmd == SIOCCHGTUNNEL) { 799 struct ip_tunnel *t = netdev_priv(dev); 800 801 t->parms.i_flags = p->i_flags; 802 t->parms.o_flags = p->o_flags; 803 804 if (strcmp(dev->rtnl_link_ops->kind, "erspan")) 805 ipgre_link_update(dev, true); 806 } 807 808 p->i_flags = gre_tnl_flags_to_gre_flags(p->i_flags); 809 p->o_flags = gre_tnl_flags_to_gre_flags(p->o_flags); 810 return 0; 811 } 812 813 /* Nice toy. Unfortunately, useless in real life :-) 814 It allows to construct virtual multiprotocol broadcast "LAN" 815 over the Internet, provided multicast routing is tuned. 816 817 818 I have no idea was this bicycle invented before me, 819 so that I had to set ARPHRD_IPGRE to a random value. 820 I have an impression, that Cisco could make something similar, 821 but this feature is apparently missing in IOS<=11.2(8). 822 823 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks 824 with broadcast 224.66.66.66. If you have access to mbone, play with me :-) 825 826 ping -t 255 224.66.66.66 827 828 If nobody answers, mbone does not work. 829 830 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255 831 ip addr add 10.66.66.<somewhat>/24 dev Universe 832 ifconfig Universe up 833 ifconfig Universe add fe80::<Your_real_addr>/10 834 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96 835 ftp 10.66.66.66 836 ... 837 ftp fec0:6666:6666::193.233.7.65 838 ... 839 */ 840 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, 841 unsigned short type, 842 const void *daddr, const void *saddr, unsigned int len) 843 { 844 struct ip_tunnel *t = netdev_priv(dev); 845 struct iphdr *iph; 846 struct gre_base_hdr *greh; 847 848 iph = skb_push(skb, t->hlen + sizeof(*iph)); 849 greh = (struct gre_base_hdr *)(iph+1); 850 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags); 851 greh->protocol = htons(type); 852 853 memcpy(iph, &t->parms.iph, sizeof(struct iphdr)); 854 855 /* Set the source hardware address. */ 856 if (saddr) 857 memcpy(&iph->saddr, saddr, 4); 858 if (daddr) 859 memcpy(&iph->daddr, daddr, 4); 860 if (iph->daddr) 861 return t->hlen + sizeof(*iph); 862 863 return -(t->hlen + sizeof(*iph)); 864 } 865 866 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr) 867 { 868 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb); 869 memcpy(haddr, &iph->saddr, 4); 870 return 4; 871 } 872 873 static const struct header_ops ipgre_header_ops = { 874 .create = ipgre_header, 875 .parse = ipgre_header_parse, 876 }; 877 878 #ifdef CONFIG_NET_IPGRE_BROADCAST 879 static int ipgre_open(struct net_device *dev) 880 { 881 struct ip_tunnel *t = netdev_priv(dev); 882 883 if (ipv4_is_multicast(t->parms.iph.daddr)) { 884 struct flowi4 fl4; 885 struct rtable *rt; 886 887 rt = ip_route_output_gre(t->net, &fl4, 888 t->parms.iph.daddr, 889 t->parms.iph.saddr, 890 t->parms.o_key, 891 RT_TOS(t->parms.iph.tos), 892 t->parms.link); 893 if (IS_ERR(rt)) 894 return -EADDRNOTAVAIL; 895 dev = rt->dst.dev; 896 ip_rt_put(rt); 897 if (!__in_dev_get_rtnl(dev)) 898 return -EADDRNOTAVAIL; 899 t->mlink = dev->ifindex; 900 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr); 901 } 902 return 0; 903 } 904 905 static int ipgre_close(struct net_device *dev) 906 { 907 struct ip_tunnel *t = netdev_priv(dev); 908 909 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) { 910 struct in_device *in_dev; 911 in_dev = inetdev_by_index(t->net, t->mlink); 912 if (in_dev) 913 ip_mc_dec_group(in_dev, t->parms.iph.daddr); 914 } 915 return 0; 916 } 917 #endif 918 919 static const struct net_device_ops ipgre_netdev_ops = { 920 .ndo_init = ipgre_tunnel_init, 921 .ndo_uninit = ip_tunnel_uninit, 922 #ifdef CONFIG_NET_IPGRE_BROADCAST 923 .ndo_open = ipgre_open, 924 .ndo_stop = ipgre_close, 925 #endif 926 .ndo_start_xmit = ipgre_xmit, 927 .ndo_siocdevprivate = ip_tunnel_siocdevprivate, 928 .ndo_change_mtu = ip_tunnel_change_mtu, 929 .ndo_get_stats64 = dev_get_tstats64, 930 .ndo_get_iflink = ip_tunnel_get_iflink, 931 .ndo_tunnel_ctl = ipgre_tunnel_ctl, 932 }; 933 934 #define GRE_FEATURES (NETIF_F_SG | \ 935 NETIF_F_FRAGLIST | \ 936 NETIF_F_HIGHDMA | \ 937 NETIF_F_HW_CSUM) 938 939 static void ipgre_tunnel_setup(struct net_device *dev) 940 { 941 dev->netdev_ops = &ipgre_netdev_ops; 942 dev->type = ARPHRD_IPGRE; 943 ip_tunnel_setup(dev, ipgre_net_id); 944 } 945 946 static void __gre_tunnel_init(struct net_device *dev) 947 { 948 struct ip_tunnel *tunnel; 949 __be16 flags; 950 951 tunnel = netdev_priv(dev); 952 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 953 tunnel->parms.iph.protocol = IPPROTO_GRE; 954 955 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen; 956 dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph); 957 958 dev->features |= GRE_FEATURES | NETIF_F_LLTX; 959 dev->hw_features |= GRE_FEATURES; 960 961 flags = tunnel->parms.o_flags; 962 963 /* TCP offload with GRE SEQ is not supported, nor can we support 2 964 * levels of outer headers requiring an update. 965 */ 966 if (flags & TUNNEL_SEQ) 967 return; 968 if (flags & TUNNEL_CSUM && tunnel->encap.type != TUNNEL_ENCAP_NONE) 969 return; 970 971 dev->features |= NETIF_F_GSO_SOFTWARE; 972 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 973 } 974 975 static int ipgre_tunnel_init(struct net_device *dev) 976 { 977 struct ip_tunnel *tunnel = netdev_priv(dev); 978 struct iphdr *iph = &tunnel->parms.iph; 979 980 __gre_tunnel_init(dev); 981 982 __dev_addr_set(dev, &iph->saddr, 4); 983 memcpy(dev->broadcast, &iph->daddr, 4); 984 985 dev->flags = IFF_NOARP; 986 netif_keep_dst(dev); 987 dev->addr_len = 4; 988 989 if (iph->daddr && !tunnel->collect_md) { 990 #ifdef CONFIG_NET_IPGRE_BROADCAST 991 if (ipv4_is_multicast(iph->daddr)) { 992 if (!iph->saddr) 993 return -EINVAL; 994 dev->flags = IFF_BROADCAST; 995 dev->header_ops = &ipgre_header_ops; 996 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 997 dev->needed_headroom = 0; 998 } 999 #endif 1000 } else if (!tunnel->collect_md) { 1001 dev->header_ops = &ipgre_header_ops; 1002 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 1003 dev->needed_headroom = 0; 1004 } 1005 1006 return ip_tunnel_init(dev); 1007 } 1008 1009 static const struct gre_protocol ipgre_protocol = { 1010 .handler = gre_rcv, 1011 .err_handler = gre_err, 1012 }; 1013 1014 static int __net_init ipgre_init_net(struct net *net) 1015 { 1016 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL); 1017 } 1018 1019 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net) 1020 { 1021 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops); 1022 } 1023 1024 static struct pernet_operations ipgre_net_ops = { 1025 .init = ipgre_init_net, 1026 .exit_batch = ipgre_exit_batch_net, 1027 .id = &ipgre_net_id, 1028 .size = sizeof(struct ip_tunnel_net), 1029 }; 1030 1031 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[], 1032 struct netlink_ext_ack *extack) 1033 { 1034 __be16 flags; 1035 1036 if (!data) 1037 return 0; 1038 1039 flags = 0; 1040 if (data[IFLA_GRE_IFLAGS]) 1041 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1042 if (data[IFLA_GRE_OFLAGS]) 1043 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1044 if (flags & (GRE_VERSION|GRE_ROUTING)) 1045 return -EINVAL; 1046 1047 if (data[IFLA_GRE_COLLECT_METADATA] && 1048 data[IFLA_GRE_ENCAP_TYPE] && 1049 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE) 1050 return -EINVAL; 1051 1052 return 0; 1053 } 1054 1055 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[], 1056 struct netlink_ext_ack *extack) 1057 { 1058 __be32 daddr; 1059 1060 if (tb[IFLA_ADDRESS]) { 1061 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1062 return -EINVAL; 1063 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1064 return -EADDRNOTAVAIL; 1065 } 1066 1067 if (!data) 1068 goto out; 1069 1070 if (data[IFLA_GRE_REMOTE]) { 1071 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4); 1072 if (!daddr) 1073 return -EINVAL; 1074 } 1075 1076 out: 1077 return ipgre_tunnel_validate(tb, data, extack); 1078 } 1079 1080 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[], 1081 struct netlink_ext_ack *extack) 1082 { 1083 __be16 flags = 0; 1084 int ret; 1085 1086 if (!data) 1087 return 0; 1088 1089 ret = ipgre_tap_validate(tb, data, extack); 1090 if (ret) 1091 return ret; 1092 1093 if (data[IFLA_GRE_ERSPAN_VER] && 1094 nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0) 1095 return 0; 1096 1097 /* ERSPAN type II/III should only have GRE sequence and key flag */ 1098 if (data[IFLA_GRE_OFLAGS]) 1099 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1100 if (data[IFLA_GRE_IFLAGS]) 1101 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1102 if (!data[IFLA_GRE_COLLECT_METADATA] && 1103 flags != (GRE_SEQ | GRE_KEY)) 1104 return -EINVAL; 1105 1106 /* ERSPAN Session ID only has 10-bit. Since we reuse 1107 * 32-bit key field as ID, check it's range. 1108 */ 1109 if (data[IFLA_GRE_IKEY] && 1110 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK)) 1111 return -EINVAL; 1112 1113 if (data[IFLA_GRE_OKEY] && 1114 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK)) 1115 return -EINVAL; 1116 1117 return 0; 1118 } 1119 1120 static int ipgre_netlink_parms(struct net_device *dev, 1121 struct nlattr *data[], 1122 struct nlattr *tb[], 1123 struct ip_tunnel_parm *parms, 1124 __u32 *fwmark) 1125 { 1126 struct ip_tunnel *t = netdev_priv(dev); 1127 1128 memset(parms, 0, sizeof(*parms)); 1129 1130 parms->iph.protocol = IPPROTO_GRE; 1131 1132 if (!data) 1133 return 0; 1134 1135 if (data[IFLA_GRE_LINK]) 1136 parms->link = nla_get_u32(data[IFLA_GRE_LINK]); 1137 1138 if (data[IFLA_GRE_IFLAGS]) 1139 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS])); 1140 1141 if (data[IFLA_GRE_OFLAGS]) 1142 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS])); 1143 1144 if (data[IFLA_GRE_IKEY]) 1145 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]); 1146 1147 if (data[IFLA_GRE_OKEY]) 1148 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]); 1149 1150 if (data[IFLA_GRE_LOCAL]) 1151 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]); 1152 1153 if (data[IFLA_GRE_REMOTE]) 1154 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]); 1155 1156 if (data[IFLA_GRE_TTL]) 1157 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]); 1158 1159 if (data[IFLA_GRE_TOS]) 1160 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]); 1161 1162 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) { 1163 if (t->ignore_df) 1164 return -EINVAL; 1165 parms->iph.frag_off = htons(IP_DF); 1166 } 1167 1168 if (data[IFLA_GRE_COLLECT_METADATA]) { 1169 t->collect_md = true; 1170 if (dev->type == ARPHRD_IPGRE) 1171 dev->type = ARPHRD_NONE; 1172 } 1173 1174 if (data[IFLA_GRE_IGNORE_DF]) { 1175 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF]) 1176 && (parms->iph.frag_off & htons(IP_DF))) 1177 return -EINVAL; 1178 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]); 1179 } 1180 1181 if (data[IFLA_GRE_FWMARK]) 1182 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]); 1183 1184 return 0; 1185 } 1186 1187 static int erspan_netlink_parms(struct net_device *dev, 1188 struct nlattr *data[], 1189 struct nlattr *tb[], 1190 struct ip_tunnel_parm *parms, 1191 __u32 *fwmark) 1192 { 1193 struct ip_tunnel *t = netdev_priv(dev); 1194 int err; 1195 1196 err = ipgre_netlink_parms(dev, data, tb, parms, fwmark); 1197 if (err) 1198 return err; 1199 if (!data) 1200 return 0; 1201 1202 if (data[IFLA_GRE_ERSPAN_VER]) { 1203 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]); 1204 1205 if (t->erspan_ver > 2) 1206 return -EINVAL; 1207 } 1208 1209 if (t->erspan_ver == 1) { 1210 if (data[IFLA_GRE_ERSPAN_INDEX]) { 1211 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]); 1212 if (t->index & ~INDEX_MASK) 1213 return -EINVAL; 1214 } 1215 } else if (t->erspan_ver == 2) { 1216 if (data[IFLA_GRE_ERSPAN_DIR]) { 1217 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]); 1218 if (t->dir & ~(DIR_MASK >> DIR_OFFSET)) 1219 return -EINVAL; 1220 } 1221 if (data[IFLA_GRE_ERSPAN_HWID]) { 1222 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]); 1223 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET)) 1224 return -EINVAL; 1225 } 1226 } 1227 1228 return 0; 1229 } 1230 1231 /* This function returns true when ENCAP attributes are present in the nl msg */ 1232 static bool ipgre_netlink_encap_parms(struct nlattr *data[], 1233 struct ip_tunnel_encap *ipencap) 1234 { 1235 bool ret = false; 1236 1237 memset(ipencap, 0, sizeof(*ipencap)); 1238 1239 if (!data) 1240 return ret; 1241 1242 if (data[IFLA_GRE_ENCAP_TYPE]) { 1243 ret = true; 1244 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]); 1245 } 1246 1247 if (data[IFLA_GRE_ENCAP_FLAGS]) { 1248 ret = true; 1249 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]); 1250 } 1251 1252 if (data[IFLA_GRE_ENCAP_SPORT]) { 1253 ret = true; 1254 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]); 1255 } 1256 1257 if (data[IFLA_GRE_ENCAP_DPORT]) { 1258 ret = true; 1259 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]); 1260 } 1261 1262 return ret; 1263 } 1264 1265 static int gre_tap_init(struct net_device *dev) 1266 { 1267 __gre_tunnel_init(dev); 1268 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1269 netif_keep_dst(dev); 1270 1271 return ip_tunnel_init(dev); 1272 } 1273 1274 static const struct net_device_ops gre_tap_netdev_ops = { 1275 .ndo_init = gre_tap_init, 1276 .ndo_uninit = ip_tunnel_uninit, 1277 .ndo_start_xmit = gre_tap_xmit, 1278 .ndo_set_mac_address = eth_mac_addr, 1279 .ndo_validate_addr = eth_validate_addr, 1280 .ndo_change_mtu = ip_tunnel_change_mtu, 1281 .ndo_get_stats64 = dev_get_tstats64, 1282 .ndo_get_iflink = ip_tunnel_get_iflink, 1283 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1284 }; 1285 1286 static int erspan_tunnel_init(struct net_device *dev) 1287 { 1288 struct ip_tunnel *tunnel = netdev_priv(dev); 1289 1290 if (tunnel->erspan_ver == 0) 1291 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */ 1292 else 1293 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */ 1294 1295 tunnel->parms.iph.protocol = IPPROTO_GRE; 1296 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen + 1297 erspan_hdr_len(tunnel->erspan_ver); 1298 1299 dev->features |= GRE_FEATURES; 1300 dev->hw_features |= GRE_FEATURES; 1301 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1302 netif_keep_dst(dev); 1303 1304 return ip_tunnel_init(dev); 1305 } 1306 1307 static const struct net_device_ops erspan_netdev_ops = { 1308 .ndo_init = erspan_tunnel_init, 1309 .ndo_uninit = ip_tunnel_uninit, 1310 .ndo_start_xmit = erspan_xmit, 1311 .ndo_set_mac_address = eth_mac_addr, 1312 .ndo_validate_addr = eth_validate_addr, 1313 .ndo_change_mtu = ip_tunnel_change_mtu, 1314 .ndo_get_stats64 = dev_get_tstats64, 1315 .ndo_get_iflink = ip_tunnel_get_iflink, 1316 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1317 }; 1318 1319 static void ipgre_tap_setup(struct net_device *dev) 1320 { 1321 ether_setup(dev); 1322 dev->max_mtu = 0; 1323 dev->netdev_ops = &gre_tap_netdev_ops; 1324 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1325 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1326 ip_tunnel_setup(dev, gre_tap_net_id); 1327 } 1328 1329 static int 1330 ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[]) 1331 { 1332 struct ip_tunnel_encap ipencap; 1333 1334 if (ipgre_netlink_encap_parms(data, &ipencap)) { 1335 struct ip_tunnel *t = netdev_priv(dev); 1336 int err = ip_tunnel_encap_setup(t, &ipencap); 1337 1338 if (err < 0) 1339 return err; 1340 } 1341 1342 return 0; 1343 } 1344 1345 static int ipgre_newlink(struct net *src_net, struct net_device *dev, 1346 struct nlattr *tb[], struct nlattr *data[], 1347 struct netlink_ext_ack *extack) 1348 { 1349 struct ip_tunnel_parm p; 1350 __u32 fwmark = 0; 1351 int err; 1352 1353 err = ipgre_newlink_encap_setup(dev, data); 1354 if (err) 1355 return err; 1356 1357 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1358 if (err < 0) 1359 return err; 1360 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1361 } 1362 1363 static int erspan_newlink(struct net *src_net, struct net_device *dev, 1364 struct nlattr *tb[], struct nlattr *data[], 1365 struct netlink_ext_ack *extack) 1366 { 1367 struct ip_tunnel_parm p; 1368 __u32 fwmark = 0; 1369 int err; 1370 1371 err = ipgre_newlink_encap_setup(dev, data); 1372 if (err) 1373 return err; 1374 1375 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1376 if (err) 1377 return err; 1378 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1379 } 1380 1381 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[], 1382 struct nlattr *data[], 1383 struct netlink_ext_ack *extack) 1384 { 1385 struct ip_tunnel *t = netdev_priv(dev); 1386 __u32 fwmark = t->fwmark; 1387 struct ip_tunnel_parm p; 1388 int err; 1389 1390 err = ipgre_newlink_encap_setup(dev, data); 1391 if (err) 1392 return err; 1393 1394 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1395 if (err < 0) 1396 return err; 1397 1398 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1399 if (err < 0) 1400 return err; 1401 1402 t->parms.i_flags = p.i_flags; 1403 t->parms.o_flags = p.o_flags; 1404 1405 ipgre_link_update(dev, !tb[IFLA_MTU]); 1406 1407 return 0; 1408 } 1409 1410 static int erspan_changelink(struct net_device *dev, struct nlattr *tb[], 1411 struct nlattr *data[], 1412 struct netlink_ext_ack *extack) 1413 { 1414 struct ip_tunnel *t = netdev_priv(dev); 1415 __u32 fwmark = t->fwmark; 1416 struct ip_tunnel_parm p; 1417 int err; 1418 1419 err = ipgre_newlink_encap_setup(dev, data); 1420 if (err) 1421 return err; 1422 1423 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1424 if (err < 0) 1425 return err; 1426 1427 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1428 if (err < 0) 1429 return err; 1430 1431 t->parms.i_flags = p.i_flags; 1432 t->parms.o_flags = p.o_flags; 1433 1434 return 0; 1435 } 1436 1437 static size_t ipgre_get_size(const struct net_device *dev) 1438 { 1439 return 1440 /* IFLA_GRE_LINK */ 1441 nla_total_size(4) + 1442 /* IFLA_GRE_IFLAGS */ 1443 nla_total_size(2) + 1444 /* IFLA_GRE_OFLAGS */ 1445 nla_total_size(2) + 1446 /* IFLA_GRE_IKEY */ 1447 nla_total_size(4) + 1448 /* IFLA_GRE_OKEY */ 1449 nla_total_size(4) + 1450 /* IFLA_GRE_LOCAL */ 1451 nla_total_size(4) + 1452 /* IFLA_GRE_REMOTE */ 1453 nla_total_size(4) + 1454 /* IFLA_GRE_TTL */ 1455 nla_total_size(1) + 1456 /* IFLA_GRE_TOS */ 1457 nla_total_size(1) + 1458 /* IFLA_GRE_PMTUDISC */ 1459 nla_total_size(1) + 1460 /* IFLA_GRE_ENCAP_TYPE */ 1461 nla_total_size(2) + 1462 /* IFLA_GRE_ENCAP_FLAGS */ 1463 nla_total_size(2) + 1464 /* IFLA_GRE_ENCAP_SPORT */ 1465 nla_total_size(2) + 1466 /* IFLA_GRE_ENCAP_DPORT */ 1467 nla_total_size(2) + 1468 /* IFLA_GRE_COLLECT_METADATA */ 1469 nla_total_size(0) + 1470 /* IFLA_GRE_IGNORE_DF */ 1471 nla_total_size(1) + 1472 /* IFLA_GRE_FWMARK */ 1473 nla_total_size(4) + 1474 /* IFLA_GRE_ERSPAN_INDEX */ 1475 nla_total_size(4) + 1476 /* IFLA_GRE_ERSPAN_VER */ 1477 nla_total_size(1) + 1478 /* IFLA_GRE_ERSPAN_DIR */ 1479 nla_total_size(1) + 1480 /* IFLA_GRE_ERSPAN_HWID */ 1481 nla_total_size(2) + 1482 0; 1483 } 1484 1485 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev) 1486 { 1487 struct ip_tunnel *t = netdev_priv(dev); 1488 struct ip_tunnel_parm *p = &t->parms; 1489 __be16 o_flags = p->o_flags; 1490 1491 if (t->erspan_ver <= 2) { 1492 if (t->erspan_ver != 0 && !t->collect_md) 1493 o_flags |= TUNNEL_KEY; 1494 1495 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver)) 1496 goto nla_put_failure; 1497 1498 if (t->erspan_ver == 1) { 1499 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index)) 1500 goto nla_put_failure; 1501 } else if (t->erspan_ver == 2) { 1502 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir)) 1503 goto nla_put_failure; 1504 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid)) 1505 goto nla_put_failure; 1506 } 1507 } 1508 1509 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) || 1510 nla_put_be16(skb, IFLA_GRE_IFLAGS, 1511 gre_tnl_flags_to_gre_flags(p->i_flags)) || 1512 nla_put_be16(skb, IFLA_GRE_OFLAGS, 1513 gre_tnl_flags_to_gre_flags(o_flags)) || 1514 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) || 1515 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) || 1516 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) || 1517 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) || 1518 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) || 1519 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) || 1520 nla_put_u8(skb, IFLA_GRE_PMTUDISC, 1521 !!(p->iph.frag_off & htons(IP_DF))) || 1522 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark)) 1523 goto nla_put_failure; 1524 1525 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE, 1526 t->encap.type) || 1527 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT, 1528 t->encap.sport) || 1529 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT, 1530 t->encap.dport) || 1531 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS, 1532 t->encap.flags)) 1533 goto nla_put_failure; 1534 1535 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df)) 1536 goto nla_put_failure; 1537 1538 if (t->collect_md) { 1539 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA)) 1540 goto nla_put_failure; 1541 } 1542 1543 return 0; 1544 1545 nla_put_failure: 1546 return -EMSGSIZE; 1547 } 1548 1549 static void erspan_setup(struct net_device *dev) 1550 { 1551 struct ip_tunnel *t = netdev_priv(dev); 1552 1553 ether_setup(dev); 1554 dev->max_mtu = 0; 1555 dev->netdev_ops = &erspan_netdev_ops; 1556 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1557 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1558 ip_tunnel_setup(dev, erspan_net_id); 1559 t->erspan_ver = 1; 1560 } 1561 1562 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = { 1563 [IFLA_GRE_LINK] = { .type = NLA_U32 }, 1564 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 }, 1565 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 }, 1566 [IFLA_GRE_IKEY] = { .type = NLA_U32 }, 1567 [IFLA_GRE_OKEY] = { .type = NLA_U32 }, 1568 [IFLA_GRE_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 1569 [IFLA_GRE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1570 [IFLA_GRE_TTL] = { .type = NLA_U8 }, 1571 [IFLA_GRE_TOS] = { .type = NLA_U8 }, 1572 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 }, 1573 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 }, 1574 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 }, 1575 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 }, 1576 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 }, 1577 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1578 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 }, 1579 [IFLA_GRE_FWMARK] = { .type = NLA_U32 }, 1580 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 }, 1581 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 }, 1582 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 }, 1583 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 }, 1584 }; 1585 1586 static struct rtnl_link_ops ipgre_link_ops __read_mostly = { 1587 .kind = "gre", 1588 .maxtype = IFLA_GRE_MAX, 1589 .policy = ipgre_policy, 1590 .priv_size = sizeof(struct ip_tunnel), 1591 .setup = ipgre_tunnel_setup, 1592 .validate = ipgre_tunnel_validate, 1593 .newlink = ipgre_newlink, 1594 .changelink = ipgre_changelink, 1595 .dellink = ip_tunnel_dellink, 1596 .get_size = ipgre_get_size, 1597 .fill_info = ipgre_fill_info, 1598 .get_link_net = ip_tunnel_get_link_net, 1599 }; 1600 1601 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = { 1602 .kind = "gretap", 1603 .maxtype = IFLA_GRE_MAX, 1604 .policy = ipgre_policy, 1605 .priv_size = sizeof(struct ip_tunnel), 1606 .setup = ipgre_tap_setup, 1607 .validate = ipgre_tap_validate, 1608 .newlink = ipgre_newlink, 1609 .changelink = ipgre_changelink, 1610 .dellink = ip_tunnel_dellink, 1611 .get_size = ipgre_get_size, 1612 .fill_info = ipgre_fill_info, 1613 .get_link_net = ip_tunnel_get_link_net, 1614 }; 1615 1616 static struct rtnl_link_ops erspan_link_ops __read_mostly = { 1617 .kind = "erspan", 1618 .maxtype = IFLA_GRE_MAX, 1619 .policy = ipgre_policy, 1620 .priv_size = sizeof(struct ip_tunnel), 1621 .setup = erspan_setup, 1622 .validate = erspan_validate, 1623 .newlink = erspan_newlink, 1624 .changelink = erspan_changelink, 1625 .dellink = ip_tunnel_dellink, 1626 .get_size = ipgre_get_size, 1627 .fill_info = ipgre_fill_info, 1628 .get_link_net = ip_tunnel_get_link_net, 1629 }; 1630 1631 struct net_device *gretap_fb_dev_create(struct net *net, const char *name, 1632 u8 name_assign_type) 1633 { 1634 struct nlattr *tb[IFLA_MAX + 1]; 1635 struct net_device *dev; 1636 LIST_HEAD(list_kill); 1637 struct ip_tunnel *t; 1638 int err; 1639 1640 memset(&tb, 0, sizeof(tb)); 1641 1642 dev = rtnl_create_link(net, name, name_assign_type, 1643 &ipgre_tap_ops, tb, NULL); 1644 if (IS_ERR(dev)) 1645 return dev; 1646 1647 /* Configure flow based GRE device. */ 1648 t = netdev_priv(dev); 1649 t->collect_md = true; 1650 1651 err = ipgre_newlink(net, dev, tb, NULL, NULL); 1652 if (err < 0) { 1653 free_netdev(dev); 1654 return ERR_PTR(err); 1655 } 1656 1657 /* openvswitch users expect packet sizes to be unrestricted, 1658 * so set the largest MTU we can. 1659 */ 1660 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false); 1661 if (err) 1662 goto out; 1663 1664 err = rtnl_configure_link(dev, NULL); 1665 if (err < 0) 1666 goto out; 1667 1668 return dev; 1669 out: 1670 ip_tunnel_dellink(dev, &list_kill); 1671 unregister_netdevice_many(&list_kill); 1672 return ERR_PTR(err); 1673 } 1674 EXPORT_SYMBOL_GPL(gretap_fb_dev_create); 1675 1676 static int __net_init ipgre_tap_init_net(struct net *net) 1677 { 1678 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0"); 1679 } 1680 1681 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net) 1682 { 1683 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops); 1684 } 1685 1686 static struct pernet_operations ipgre_tap_net_ops = { 1687 .init = ipgre_tap_init_net, 1688 .exit_batch = ipgre_tap_exit_batch_net, 1689 .id = &gre_tap_net_id, 1690 .size = sizeof(struct ip_tunnel_net), 1691 }; 1692 1693 static int __net_init erspan_init_net(struct net *net) 1694 { 1695 return ip_tunnel_init_net(net, erspan_net_id, 1696 &erspan_link_ops, "erspan0"); 1697 } 1698 1699 static void __net_exit erspan_exit_batch_net(struct list_head *net_list) 1700 { 1701 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops); 1702 } 1703 1704 static struct pernet_operations erspan_net_ops = { 1705 .init = erspan_init_net, 1706 .exit_batch = erspan_exit_batch_net, 1707 .id = &erspan_net_id, 1708 .size = sizeof(struct ip_tunnel_net), 1709 }; 1710 1711 static int __init ipgre_init(void) 1712 { 1713 int err; 1714 1715 pr_info("GRE over IPv4 tunneling driver\n"); 1716 1717 err = register_pernet_device(&ipgre_net_ops); 1718 if (err < 0) 1719 return err; 1720 1721 err = register_pernet_device(&ipgre_tap_net_ops); 1722 if (err < 0) 1723 goto pnet_tap_failed; 1724 1725 err = register_pernet_device(&erspan_net_ops); 1726 if (err < 0) 1727 goto pnet_erspan_failed; 1728 1729 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO); 1730 if (err < 0) { 1731 pr_info("%s: can't add protocol\n", __func__); 1732 goto add_proto_failed; 1733 } 1734 1735 err = rtnl_link_register(&ipgre_link_ops); 1736 if (err < 0) 1737 goto rtnl_link_failed; 1738 1739 err = rtnl_link_register(&ipgre_tap_ops); 1740 if (err < 0) 1741 goto tap_ops_failed; 1742 1743 err = rtnl_link_register(&erspan_link_ops); 1744 if (err < 0) 1745 goto erspan_link_failed; 1746 1747 return 0; 1748 1749 erspan_link_failed: 1750 rtnl_link_unregister(&ipgre_tap_ops); 1751 tap_ops_failed: 1752 rtnl_link_unregister(&ipgre_link_ops); 1753 rtnl_link_failed: 1754 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1755 add_proto_failed: 1756 unregister_pernet_device(&erspan_net_ops); 1757 pnet_erspan_failed: 1758 unregister_pernet_device(&ipgre_tap_net_ops); 1759 pnet_tap_failed: 1760 unregister_pernet_device(&ipgre_net_ops); 1761 return err; 1762 } 1763 1764 static void __exit ipgre_fini(void) 1765 { 1766 rtnl_link_unregister(&ipgre_tap_ops); 1767 rtnl_link_unregister(&ipgre_link_ops); 1768 rtnl_link_unregister(&erspan_link_ops); 1769 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1770 unregister_pernet_device(&ipgre_tap_net_ops); 1771 unregister_pernet_device(&ipgre_net_ops); 1772 unregister_pernet_device(&erspan_net_ops); 1773 } 1774 1775 module_init(ipgre_init); 1776 module_exit(ipgre_fini); 1777 MODULE_LICENSE("GPL"); 1778 MODULE_ALIAS_RTNL_LINK("gre"); 1779 MODULE_ALIAS_RTNL_LINK("gretap"); 1780 MODULE_ALIAS_RTNL_LINK("erspan"); 1781 MODULE_ALIAS_NETDEV("gre0"); 1782 MODULE_ALIAS_NETDEV("gretap0"); 1783 MODULE_ALIAS_NETDEV("erspan0"); 1784