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); 462 463 if (tunnel->parms.o_flags & TUNNEL_SEQ) 464 tunnel->o_seqno++; 465 466 /* Push GRE header. */ 467 gre_build_header(skb, tunnel->tun_hlen, 468 tunnel->parms.o_flags, proto, tunnel->parms.o_key, 469 htonl(tunnel->o_seqno)); 470 471 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol); 472 } 473 474 static int gre_handle_offloads(struct sk_buff *skb, bool csum) 475 { 476 return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE); 477 } 478 479 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev, 480 __be16 proto) 481 { 482 struct ip_tunnel *tunnel = netdev_priv(dev); 483 struct ip_tunnel_info *tun_info; 484 const struct ip_tunnel_key *key; 485 int tunnel_hlen; 486 __be16 flags; 487 488 tun_info = skb_tunnel_info(skb); 489 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 490 ip_tunnel_info_af(tun_info) != AF_INET)) 491 goto err_free_skb; 492 493 key = &tun_info->key; 494 tunnel_hlen = gre_calc_hlen(key->tun_flags); 495 496 if (skb_cow_head(skb, dev->needed_headroom)) 497 goto err_free_skb; 498 499 /* Push Tunnel header. */ 500 if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM))) 501 goto err_free_skb; 502 503 flags = tun_info->key.tun_flags & 504 (TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ); 505 gre_build_header(skb, tunnel_hlen, flags, proto, 506 tunnel_id_to_key32(tun_info->key.tun_id), 507 (flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++) : 0); 508 509 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen); 510 511 return; 512 513 err_free_skb: 514 kfree_skb(skb); 515 dev->stats.tx_dropped++; 516 } 517 518 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev) 519 { 520 struct ip_tunnel *tunnel = netdev_priv(dev); 521 struct ip_tunnel_info *tun_info; 522 const struct ip_tunnel_key *key; 523 struct erspan_metadata *md; 524 bool truncate = false; 525 __be16 proto; 526 int tunnel_hlen; 527 int version; 528 int nhoff; 529 int thoff; 530 531 tun_info = skb_tunnel_info(skb); 532 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 533 ip_tunnel_info_af(tun_info) != AF_INET)) 534 goto err_free_skb; 535 536 key = &tun_info->key; 537 if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT)) 538 goto err_free_skb; 539 if (tun_info->options_len < sizeof(*md)) 540 goto err_free_skb; 541 md = ip_tunnel_info_opts(tun_info); 542 543 /* ERSPAN has fixed 8 byte GRE header */ 544 version = md->version; 545 tunnel_hlen = 8 + erspan_hdr_len(version); 546 547 if (skb_cow_head(skb, dev->needed_headroom)) 548 goto err_free_skb; 549 550 if (gre_handle_offloads(skb, false)) 551 goto err_free_skb; 552 553 if (skb->len > dev->mtu + dev->hard_header_len) { 554 pskb_trim(skb, dev->mtu + dev->hard_header_len); 555 truncate = true; 556 } 557 558 nhoff = skb_network_header(skb) - skb_mac_header(skb); 559 if (skb->protocol == htons(ETH_P_IP) && 560 (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff)) 561 truncate = true; 562 563 thoff = skb_transport_header(skb) - skb_mac_header(skb); 564 if (skb->protocol == htons(ETH_P_IPV6) && 565 (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff)) 566 truncate = true; 567 568 if (version == 1) { 569 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)), 570 ntohl(md->u.index), truncate, true); 571 proto = htons(ETH_P_ERSPAN); 572 } else if (version == 2) { 573 erspan_build_header_v2(skb, 574 ntohl(tunnel_id_to_key32(key->tun_id)), 575 md->u.md2.dir, 576 get_hwid(&md->u.md2), 577 truncate, true); 578 proto = htons(ETH_P_ERSPAN2); 579 } else { 580 goto err_free_skb; 581 } 582 583 gre_build_header(skb, 8, TUNNEL_SEQ, 584 proto, 0, htonl(tunnel->o_seqno++)); 585 586 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen); 587 588 return; 589 590 err_free_skb: 591 kfree_skb(skb); 592 dev->stats.tx_dropped++; 593 } 594 595 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 596 { 597 struct ip_tunnel_info *info = skb_tunnel_info(skb); 598 const struct ip_tunnel_key *key; 599 struct rtable *rt; 600 struct flowi4 fl4; 601 602 if (ip_tunnel_info_af(info) != AF_INET) 603 return -EINVAL; 604 605 key = &info->key; 606 ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src, 607 tunnel_id_to_key32(key->tun_id), key->tos, 0, 608 skb->mark, skb_get_hash(skb)); 609 rt = ip_route_output_key(dev_net(dev), &fl4); 610 if (IS_ERR(rt)) 611 return PTR_ERR(rt); 612 613 ip_rt_put(rt); 614 info->key.u.ipv4.src = fl4.saddr; 615 return 0; 616 } 617 618 static netdev_tx_t ipgre_xmit(struct sk_buff *skb, 619 struct net_device *dev) 620 { 621 struct ip_tunnel *tunnel = netdev_priv(dev); 622 const struct iphdr *tnl_params; 623 624 if (!pskb_inet_may_pull(skb)) 625 goto free_skb; 626 627 if (tunnel->collect_md) { 628 gre_fb_xmit(skb, dev, skb->protocol); 629 return NETDEV_TX_OK; 630 } 631 632 if (dev->header_ops) { 633 if (skb_cow_head(skb, 0)) 634 goto free_skb; 635 636 tnl_params = (const struct iphdr *)skb->data; 637 638 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing 639 * to gre header. 640 */ 641 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr)); 642 skb_reset_mac_header(skb); 643 } else { 644 if (skb_cow_head(skb, dev->needed_headroom)) 645 goto free_skb; 646 647 tnl_params = &tunnel->parms.iph; 648 } 649 650 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 651 goto free_skb; 652 653 __gre_xmit(skb, dev, tnl_params, skb->protocol); 654 return NETDEV_TX_OK; 655 656 free_skb: 657 kfree_skb(skb); 658 dev->stats.tx_dropped++; 659 return NETDEV_TX_OK; 660 } 661 662 static netdev_tx_t erspan_xmit(struct sk_buff *skb, 663 struct net_device *dev) 664 { 665 struct ip_tunnel *tunnel = netdev_priv(dev); 666 bool truncate = false; 667 __be16 proto; 668 669 if (!pskb_inet_may_pull(skb)) 670 goto free_skb; 671 672 if (tunnel->collect_md) { 673 erspan_fb_xmit(skb, dev); 674 return NETDEV_TX_OK; 675 } 676 677 if (gre_handle_offloads(skb, false)) 678 goto free_skb; 679 680 if (skb_cow_head(skb, dev->needed_headroom)) 681 goto free_skb; 682 683 if (skb->len > dev->mtu + dev->hard_header_len) { 684 pskb_trim(skb, dev->mtu + dev->hard_header_len); 685 truncate = true; 686 } 687 688 /* Push ERSPAN header */ 689 if (tunnel->erspan_ver == 0) { 690 proto = htons(ETH_P_ERSPAN); 691 tunnel->parms.o_flags &= ~TUNNEL_SEQ; 692 } else if (tunnel->erspan_ver == 1) { 693 erspan_build_header(skb, ntohl(tunnel->parms.o_key), 694 tunnel->index, 695 truncate, true); 696 proto = htons(ETH_P_ERSPAN); 697 } else if (tunnel->erspan_ver == 2) { 698 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key), 699 tunnel->dir, tunnel->hwid, 700 truncate, true); 701 proto = htons(ETH_P_ERSPAN2); 702 } else { 703 goto free_skb; 704 } 705 706 tunnel->parms.o_flags &= ~TUNNEL_KEY; 707 __gre_xmit(skb, dev, &tunnel->parms.iph, proto); 708 return NETDEV_TX_OK; 709 710 free_skb: 711 kfree_skb(skb); 712 dev->stats.tx_dropped++; 713 return NETDEV_TX_OK; 714 } 715 716 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb, 717 struct net_device *dev) 718 { 719 struct ip_tunnel *tunnel = netdev_priv(dev); 720 721 if (!pskb_inet_may_pull(skb)) 722 goto free_skb; 723 724 if (tunnel->collect_md) { 725 gre_fb_xmit(skb, dev, htons(ETH_P_TEB)); 726 return NETDEV_TX_OK; 727 } 728 729 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 730 goto free_skb; 731 732 if (skb_cow_head(skb, dev->needed_headroom)) 733 goto free_skb; 734 735 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB)); 736 return NETDEV_TX_OK; 737 738 free_skb: 739 kfree_skb(skb); 740 dev->stats.tx_dropped++; 741 return NETDEV_TX_OK; 742 } 743 744 static void ipgre_link_update(struct net_device *dev, bool set_mtu) 745 { 746 struct ip_tunnel *tunnel = netdev_priv(dev); 747 int len; 748 749 len = tunnel->tun_hlen; 750 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 751 len = tunnel->tun_hlen - len; 752 tunnel->hlen = tunnel->hlen + len; 753 754 if (dev->header_ops) 755 dev->hard_header_len += len; 756 else 757 dev->needed_headroom += len; 758 759 if (set_mtu) 760 dev->mtu = max_t(int, dev->mtu - len, 68); 761 762 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) { 763 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) || 764 tunnel->encap.type == TUNNEL_ENCAP_NONE) { 765 dev->features |= NETIF_F_GSO_SOFTWARE; 766 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 767 } else { 768 dev->features &= ~NETIF_F_GSO_SOFTWARE; 769 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 770 } 771 dev->features |= NETIF_F_LLTX; 772 } else { 773 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 774 dev->features &= ~(NETIF_F_LLTX | 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_do_ioctl = ip_tunnel_ioctl, 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; 948 949 tunnel = netdev_priv(dev); 950 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 951 tunnel->parms.iph.protocol = IPPROTO_GRE; 952 953 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen; 954 dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph); 955 956 dev->features |= GRE_FEATURES; 957 dev->hw_features |= GRE_FEATURES; 958 959 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) { 960 /* TCP offload with GRE SEQ is not supported, nor 961 * can we support 2 levels of outer headers requiring 962 * an update. 963 */ 964 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) || 965 (tunnel->encap.type == TUNNEL_ENCAP_NONE)) { 966 dev->features |= NETIF_F_GSO_SOFTWARE; 967 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 968 } 969 970 /* Can use a lockless transmit, unless we generate 971 * output sequences 972 */ 973 dev->features |= NETIF_F_LLTX; 974 } 975 } 976 977 static int ipgre_tunnel_init(struct net_device *dev) 978 { 979 struct ip_tunnel *tunnel = netdev_priv(dev); 980 struct iphdr *iph = &tunnel->parms.iph; 981 982 __gre_tunnel_init(dev); 983 984 memcpy(dev->dev_addr, &iph->saddr, 4); 985 memcpy(dev->broadcast, &iph->daddr, 4); 986 987 dev->flags = IFF_NOARP; 988 netif_keep_dst(dev); 989 dev->addr_len = 4; 990 991 if (iph->daddr && !tunnel->collect_md) { 992 #ifdef CONFIG_NET_IPGRE_BROADCAST 993 if (ipv4_is_multicast(iph->daddr)) { 994 if (!iph->saddr) 995 return -EINVAL; 996 dev->flags = IFF_BROADCAST; 997 dev->header_ops = &ipgre_header_ops; 998 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 999 dev->needed_headroom = 0; 1000 } 1001 #endif 1002 } else if (!tunnel->collect_md) { 1003 dev->header_ops = &ipgre_header_ops; 1004 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 1005 dev->needed_headroom = 0; 1006 } 1007 1008 return ip_tunnel_init(dev); 1009 } 1010 1011 static const struct gre_protocol ipgre_protocol = { 1012 .handler = gre_rcv, 1013 .err_handler = gre_err, 1014 }; 1015 1016 static int __net_init ipgre_init_net(struct net *net) 1017 { 1018 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL); 1019 } 1020 1021 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net) 1022 { 1023 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops); 1024 } 1025 1026 static struct pernet_operations ipgre_net_ops = { 1027 .init = ipgre_init_net, 1028 .exit_batch = ipgre_exit_batch_net, 1029 .id = &ipgre_net_id, 1030 .size = sizeof(struct ip_tunnel_net), 1031 }; 1032 1033 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[], 1034 struct netlink_ext_ack *extack) 1035 { 1036 __be16 flags; 1037 1038 if (!data) 1039 return 0; 1040 1041 flags = 0; 1042 if (data[IFLA_GRE_IFLAGS]) 1043 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1044 if (data[IFLA_GRE_OFLAGS]) 1045 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1046 if (flags & (GRE_VERSION|GRE_ROUTING)) 1047 return -EINVAL; 1048 1049 if (data[IFLA_GRE_COLLECT_METADATA] && 1050 data[IFLA_GRE_ENCAP_TYPE] && 1051 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE) 1052 return -EINVAL; 1053 1054 return 0; 1055 } 1056 1057 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[], 1058 struct netlink_ext_ack *extack) 1059 { 1060 __be32 daddr; 1061 1062 if (tb[IFLA_ADDRESS]) { 1063 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1064 return -EINVAL; 1065 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1066 return -EADDRNOTAVAIL; 1067 } 1068 1069 if (!data) 1070 goto out; 1071 1072 if (data[IFLA_GRE_REMOTE]) { 1073 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4); 1074 if (!daddr) 1075 return -EINVAL; 1076 } 1077 1078 out: 1079 return ipgre_tunnel_validate(tb, data, extack); 1080 } 1081 1082 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[], 1083 struct netlink_ext_ack *extack) 1084 { 1085 __be16 flags = 0; 1086 int ret; 1087 1088 if (!data) 1089 return 0; 1090 1091 ret = ipgre_tap_validate(tb, data, extack); 1092 if (ret) 1093 return ret; 1094 1095 if (data[IFLA_GRE_ERSPAN_VER] && 1096 nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0) 1097 return 0; 1098 1099 /* ERSPAN type II/III should only have GRE sequence and key flag */ 1100 if (data[IFLA_GRE_OFLAGS]) 1101 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1102 if (data[IFLA_GRE_IFLAGS]) 1103 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1104 if (!data[IFLA_GRE_COLLECT_METADATA] && 1105 flags != (GRE_SEQ | GRE_KEY)) 1106 return -EINVAL; 1107 1108 /* ERSPAN Session ID only has 10-bit. Since we reuse 1109 * 32-bit key field as ID, check it's range. 1110 */ 1111 if (data[IFLA_GRE_IKEY] && 1112 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK)) 1113 return -EINVAL; 1114 1115 if (data[IFLA_GRE_OKEY] && 1116 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK)) 1117 return -EINVAL; 1118 1119 return 0; 1120 } 1121 1122 static int ipgre_netlink_parms(struct net_device *dev, 1123 struct nlattr *data[], 1124 struct nlattr *tb[], 1125 struct ip_tunnel_parm *parms, 1126 __u32 *fwmark) 1127 { 1128 struct ip_tunnel *t = netdev_priv(dev); 1129 1130 memset(parms, 0, sizeof(*parms)); 1131 1132 parms->iph.protocol = IPPROTO_GRE; 1133 1134 if (!data) 1135 return 0; 1136 1137 if (data[IFLA_GRE_LINK]) 1138 parms->link = nla_get_u32(data[IFLA_GRE_LINK]); 1139 1140 if (data[IFLA_GRE_IFLAGS]) 1141 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS])); 1142 1143 if (data[IFLA_GRE_OFLAGS]) 1144 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS])); 1145 1146 if (data[IFLA_GRE_IKEY]) 1147 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]); 1148 1149 if (data[IFLA_GRE_OKEY]) 1150 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]); 1151 1152 if (data[IFLA_GRE_LOCAL]) 1153 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]); 1154 1155 if (data[IFLA_GRE_REMOTE]) 1156 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]); 1157 1158 if (data[IFLA_GRE_TTL]) 1159 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]); 1160 1161 if (data[IFLA_GRE_TOS]) 1162 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]); 1163 1164 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) { 1165 if (t->ignore_df) 1166 return -EINVAL; 1167 parms->iph.frag_off = htons(IP_DF); 1168 } 1169 1170 if (data[IFLA_GRE_COLLECT_METADATA]) { 1171 t->collect_md = true; 1172 if (dev->type == ARPHRD_IPGRE) 1173 dev->type = ARPHRD_NONE; 1174 } 1175 1176 if (data[IFLA_GRE_IGNORE_DF]) { 1177 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF]) 1178 && (parms->iph.frag_off & htons(IP_DF))) 1179 return -EINVAL; 1180 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]); 1181 } 1182 1183 if (data[IFLA_GRE_FWMARK]) 1184 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]); 1185 1186 return 0; 1187 } 1188 1189 static int erspan_netlink_parms(struct net_device *dev, 1190 struct nlattr *data[], 1191 struct nlattr *tb[], 1192 struct ip_tunnel_parm *parms, 1193 __u32 *fwmark) 1194 { 1195 struct ip_tunnel *t = netdev_priv(dev); 1196 int err; 1197 1198 err = ipgre_netlink_parms(dev, data, tb, parms, fwmark); 1199 if (err) 1200 return err; 1201 if (!data) 1202 return 0; 1203 1204 if (data[IFLA_GRE_ERSPAN_VER]) { 1205 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]); 1206 1207 if (t->erspan_ver > 2) 1208 return -EINVAL; 1209 } 1210 1211 if (t->erspan_ver == 1) { 1212 if (data[IFLA_GRE_ERSPAN_INDEX]) { 1213 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]); 1214 if (t->index & ~INDEX_MASK) 1215 return -EINVAL; 1216 } 1217 } else if (t->erspan_ver == 2) { 1218 if (data[IFLA_GRE_ERSPAN_DIR]) { 1219 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]); 1220 if (t->dir & ~(DIR_MASK >> DIR_OFFSET)) 1221 return -EINVAL; 1222 } 1223 if (data[IFLA_GRE_ERSPAN_HWID]) { 1224 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]); 1225 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET)) 1226 return -EINVAL; 1227 } 1228 } 1229 1230 return 0; 1231 } 1232 1233 /* This function returns true when ENCAP attributes are present in the nl msg */ 1234 static bool ipgre_netlink_encap_parms(struct nlattr *data[], 1235 struct ip_tunnel_encap *ipencap) 1236 { 1237 bool ret = false; 1238 1239 memset(ipencap, 0, sizeof(*ipencap)); 1240 1241 if (!data) 1242 return ret; 1243 1244 if (data[IFLA_GRE_ENCAP_TYPE]) { 1245 ret = true; 1246 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]); 1247 } 1248 1249 if (data[IFLA_GRE_ENCAP_FLAGS]) { 1250 ret = true; 1251 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]); 1252 } 1253 1254 if (data[IFLA_GRE_ENCAP_SPORT]) { 1255 ret = true; 1256 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]); 1257 } 1258 1259 if (data[IFLA_GRE_ENCAP_DPORT]) { 1260 ret = true; 1261 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]); 1262 } 1263 1264 return ret; 1265 } 1266 1267 static int gre_tap_init(struct net_device *dev) 1268 { 1269 __gre_tunnel_init(dev); 1270 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1271 netif_keep_dst(dev); 1272 1273 return ip_tunnel_init(dev); 1274 } 1275 1276 static const struct net_device_ops gre_tap_netdev_ops = { 1277 .ndo_init = gre_tap_init, 1278 .ndo_uninit = ip_tunnel_uninit, 1279 .ndo_start_xmit = gre_tap_xmit, 1280 .ndo_set_mac_address = eth_mac_addr, 1281 .ndo_validate_addr = eth_validate_addr, 1282 .ndo_change_mtu = ip_tunnel_change_mtu, 1283 .ndo_get_stats64 = dev_get_tstats64, 1284 .ndo_get_iflink = ip_tunnel_get_iflink, 1285 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1286 }; 1287 1288 static int erspan_tunnel_init(struct net_device *dev) 1289 { 1290 struct ip_tunnel *tunnel = netdev_priv(dev); 1291 1292 if (tunnel->erspan_ver == 0) 1293 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */ 1294 else 1295 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */ 1296 1297 tunnel->parms.iph.protocol = IPPROTO_GRE; 1298 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen + 1299 erspan_hdr_len(tunnel->erspan_ver); 1300 1301 dev->features |= GRE_FEATURES; 1302 dev->hw_features |= GRE_FEATURES; 1303 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1304 netif_keep_dst(dev); 1305 1306 return ip_tunnel_init(dev); 1307 } 1308 1309 static const struct net_device_ops erspan_netdev_ops = { 1310 .ndo_init = erspan_tunnel_init, 1311 .ndo_uninit = ip_tunnel_uninit, 1312 .ndo_start_xmit = erspan_xmit, 1313 .ndo_set_mac_address = eth_mac_addr, 1314 .ndo_validate_addr = eth_validate_addr, 1315 .ndo_change_mtu = ip_tunnel_change_mtu, 1316 .ndo_get_stats64 = dev_get_tstats64, 1317 .ndo_get_iflink = ip_tunnel_get_iflink, 1318 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1319 }; 1320 1321 static void ipgre_tap_setup(struct net_device *dev) 1322 { 1323 ether_setup(dev); 1324 dev->max_mtu = 0; 1325 dev->netdev_ops = &gre_tap_netdev_ops; 1326 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1327 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1328 ip_tunnel_setup(dev, gre_tap_net_id); 1329 } 1330 1331 static int 1332 ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[]) 1333 { 1334 struct ip_tunnel_encap ipencap; 1335 1336 if (ipgre_netlink_encap_parms(data, &ipencap)) { 1337 struct ip_tunnel *t = netdev_priv(dev); 1338 int err = ip_tunnel_encap_setup(t, &ipencap); 1339 1340 if (err < 0) 1341 return err; 1342 } 1343 1344 return 0; 1345 } 1346 1347 static int ipgre_newlink(struct net *src_net, struct net_device *dev, 1348 struct nlattr *tb[], struct nlattr *data[], 1349 struct netlink_ext_ack *extack) 1350 { 1351 struct ip_tunnel_parm p; 1352 __u32 fwmark = 0; 1353 int err; 1354 1355 err = ipgre_newlink_encap_setup(dev, data); 1356 if (err) 1357 return err; 1358 1359 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1360 if (err < 0) 1361 return err; 1362 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1363 } 1364 1365 static int erspan_newlink(struct net *src_net, struct net_device *dev, 1366 struct nlattr *tb[], struct nlattr *data[], 1367 struct netlink_ext_ack *extack) 1368 { 1369 struct ip_tunnel_parm p; 1370 __u32 fwmark = 0; 1371 int err; 1372 1373 err = ipgre_newlink_encap_setup(dev, data); 1374 if (err) 1375 return err; 1376 1377 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1378 if (err) 1379 return err; 1380 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1381 } 1382 1383 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[], 1384 struct nlattr *data[], 1385 struct netlink_ext_ack *extack) 1386 { 1387 struct ip_tunnel *t = netdev_priv(dev); 1388 __u32 fwmark = t->fwmark; 1389 struct ip_tunnel_parm p; 1390 int err; 1391 1392 err = ipgre_newlink_encap_setup(dev, data); 1393 if (err) 1394 return err; 1395 1396 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1397 if (err < 0) 1398 return err; 1399 1400 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1401 if (err < 0) 1402 return err; 1403 1404 t->parms.i_flags = p.i_flags; 1405 t->parms.o_flags = p.o_flags; 1406 1407 ipgre_link_update(dev, !tb[IFLA_MTU]); 1408 1409 return 0; 1410 } 1411 1412 static int erspan_changelink(struct net_device *dev, struct nlattr *tb[], 1413 struct nlattr *data[], 1414 struct netlink_ext_ack *extack) 1415 { 1416 struct ip_tunnel *t = netdev_priv(dev); 1417 __u32 fwmark = t->fwmark; 1418 struct ip_tunnel_parm p; 1419 int err; 1420 1421 err = ipgre_newlink_encap_setup(dev, data); 1422 if (err) 1423 return err; 1424 1425 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1426 if (err < 0) 1427 return err; 1428 1429 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1430 if (err < 0) 1431 return err; 1432 1433 t->parms.i_flags = p.i_flags; 1434 t->parms.o_flags = p.o_flags; 1435 1436 return 0; 1437 } 1438 1439 static size_t ipgre_get_size(const struct net_device *dev) 1440 { 1441 return 1442 /* IFLA_GRE_LINK */ 1443 nla_total_size(4) + 1444 /* IFLA_GRE_IFLAGS */ 1445 nla_total_size(2) + 1446 /* IFLA_GRE_OFLAGS */ 1447 nla_total_size(2) + 1448 /* IFLA_GRE_IKEY */ 1449 nla_total_size(4) + 1450 /* IFLA_GRE_OKEY */ 1451 nla_total_size(4) + 1452 /* IFLA_GRE_LOCAL */ 1453 nla_total_size(4) + 1454 /* IFLA_GRE_REMOTE */ 1455 nla_total_size(4) + 1456 /* IFLA_GRE_TTL */ 1457 nla_total_size(1) + 1458 /* IFLA_GRE_TOS */ 1459 nla_total_size(1) + 1460 /* IFLA_GRE_PMTUDISC */ 1461 nla_total_size(1) + 1462 /* IFLA_GRE_ENCAP_TYPE */ 1463 nla_total_size(2) + 1464 /* IFLA_GRE_ENCAP_FLAGS */ 1465 nla_total_size(2) + 1466 /* IFLA_GRE_ENCAP_SPORT */ 1467 nla_total_size(2) + 1468 /* IFLA_GRE_ENCAP_DPORT */ 1469 nla_total_size(2) + 1470 /* IFLA_GRE_COLLECT_METADATA */ 1471 nla_total_size(0) + 1472 /* IFLA_GRE_IGNORE_DF */ 1473 nla_total_size(1) + 1474 /* IFLA_GRE_FWMARK */ 1475 nla_total_size(4) + 1476 /* IFLA_GRE_ERSPAN_INDEX */ 1477 nla_total_size(4) + 1478 /* IFLA_GRE_ERSPAN_VER */ 1479 nla_total_size(1) + 1480 /* IFLA_GRE_ERSPAN_DIR */ 1481 nla_total_size(1) + 1482 /* IFLA_GRE_ERSPAN_HWID */ 1483 nla_total_size(2) + 1484 0; 1485 } 1486 1487 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev) 1488 { 1489 struct ip_tunnel *t = netdev_priv(dev); 1490 struct ip_tunnel_parm *p = &t->parms; 1491 __be16 o_flags = p->o_flags; 1492 1493 if (t->erspan_ver <= 2) { 1494 if (t->erspan_ver != 0 && !t->collect_md) 1495 o_flags |= TUNNEL_KEY; 1496 1497 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver)) 1498 goto nla_put_failure; 1499 1500 if (t->erspan_ver == 1) { 1501 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index)) 1502 goto nla_put_failure; 1503 } else if (t->erspan_ver == 2) { 1504 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir)) 1505 goto nla_put_failure; 1506 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid)) 1507 goto nla_put_failure; 1508 } 1509 } 1510 1511 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) || 1512 nla_put_be16(skb, IFLA_GRE_IFLAGS, 1513 gre_tnl_flags_to_gre_flags(p->i_flags)) || 1514 nla_put_be16(skb, IFLA_GRE_OFLAGS, 1515 gre_tnl_flags_to_gre_flags(o_flags)) || 1516 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) || 1517 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) || 1518 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) || 1519 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) || 1520 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) || 1521 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) || 1522 nla_put_u8(skb, IFLA_GRE_PMTUDISC, 1523 !!(p->iph.frag_off & htons(IP_DF))) || 1524 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark)) 1525 goto nla_put_failure; 1526 1527 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE, 1528 t->encap.type) || 1529 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT, 1530 t->encap.sport) || 1531 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT, 1532 t->encap.dport) || 1533 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS, 1534 t->encap.flags)) 1535 goto nla_put_failure; 1536 1537 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df)) 1538 goto nla_put_failure; 1539 1540 if (t->collect_md) { 1541 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA)) 1542 goto nla_put_failure; 1543 } 1544 1545 return 0; 1546 1547 nla_put_failure: 1548 return -EMSGSIZE; 1549 } 1550 1551 static void erspan_setup(struct net_device *dev) 1552 { 1553 struct ip_tunnel *t = netdev_priv(dev); 1554 1555 ether_setup(dev); 1556 dev->max_mtu = 0; 1557 dev->netdev_ops = &erspan_netdev_ops; 1558 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1559 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1560 ip_tunnel_setup(dev, erspan_net_id); 1561 t->erspan_ver = 1; 1562 } 1563 1564 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = { 1565 [IFLA_GRE_LINK] = { .type = NLA_U32 }, 1566 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 }, 1567 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 }, 1568 [IFLA_GRE_IKEY] = { .type = NLA_U32 }, 1569 [IFLA_GRE_OKEY] = { .type = NLA_U32 }, 1570 [IFLA_GRE_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 1571 [IFLA_GRE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1572 [IFLA_GRE_TTL] = { .type = NLA_U8 }, 1573 [IFLA_GRE_TOS] = { .type = NLA_U8 }, 1574 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 }, 1575 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 }, 1576 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 }, 1577 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 }, 1578 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 }, 1579 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1580 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 }, 1581 [IFLA_GRE_FWMARK] = { .type = NLA_U32 }, 1582 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 }, 1583 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 }, 1584 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 }, 1585 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 }, 1586 }; 1587 1588 static struct rtnl_link_ops ipgre_link_ops __read_mostly = { 1589 .kind = "gre", 1590 .maxtype = IFLA_GRE_MAX, 1591 .policy = ipgre_policy, 1592 .priv_size = sizeof(struct ip_tunnel), 1593 .setup = ipgre_tunnel_setup, 1594 .validate = ipgre_tunnel_validate, 1595 .newlink = ipgre_newlink, 1596 .changelink = ipgre_changelink, 1597 .dellink = ip_tunnel_dellink, 1598 .get_size = ipgre_get_size, 1599 .fill_info = ipgre_fill_info, 1600 .get_link_net = ip_tunnel_get_link_net, 1601 }; 1602 1603 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = { 1604 .kind = "gretap", 1605 .maxtype = IFLA_GRE_MAX, 1606 .policy = ipgre_policy, 1607 .priv_size = sizeof(struct ip_tunnel), 1608 .setup = ipgre_tap_setup, 1609 .validate = ipgre_tap_validate, 1610 .newlink = ipgre_newlink, 1611 .changelink = ipgre_changelink, 1612 .dellink = ip_tunnel_dellink, 1613 .get_size = ipgre_get_size, 1614 .fill_info = ipgre_fill_info, 1615 .get_link_net = ip_tunnel_get_link_net, 1616 }; 1617 1618 static struct rtnl_link_ops erspan_link_ops __read_mostly = { 1619 .kind = "erspan", 1620 .maxtype = IFLA_GRE_MAX, 1621 .policy = ipgre_policy, 1622 .priv_size = sizeof(struct ip_tunnel), 1623 .setup = erspan_setup, 1624 .validate = erspan_validate, 1625 .newlink = erspan_newlink, 1626 .changelink = erspan_changelink, 1627 .dellink = ip_tunnel_dellink, 1628 .get_size = ipgre_get_size, 1629 .fill_info = ipgre_fill_info, 1630 .get_link_net = ip_tunnel_get_link_net, 1631 }; 1632 1633 struct net_device *gretap_fb_dev_create(struct net *net, const char *name, 1634 u8 name_assign_type) 1635 { 1636 struct nlattr *tb[IFLA_MAX + 1]; 1637 struct net_device *dev; 1638 LIST_HEAD(list_kill); 1639 struct ip_tunnel *t; 1640 int err; 1641 1642 memset(&tb, 0, sizeof(tb)); 1643 1644 dev = rtnl_create_link(net, name, name_assign_type, 1645 &ipgre_tap_ops, tb, NULL); 1646 if (IS_ERR(dev)) 1647 return dev; 1648 1649 /* Configure flow based GRE device. */ 1650 t = netdev_priv(dev); 1651 t->collect_md = true; 1652 1653 err = ipgre_newlink(net, dev, tb, NULL, NULL); 1654 if (err < 0) { 1655 free_netdev(dev); 1656 return ERR_PTR(err); 1657 } 1658 1659 /* openvswitch users expect packet sizes to be unrestricted, 1660 * so set the largest MTU we can. 1661 */ 1662 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false); 1663 if (err) 1664 goto out; 1665 1666 err = rtnl_configure_link(dev, NULL); 1667 if (err < 0) 1668 goto out; 1669 1670 return dev; 1671 out: 1672 ip_tunnel_dellink(dev, &list_kill); 1673 unregister_netdevice_many(&list_kill); 1674 return ERR_PTR(err); 1675 } 1676 EXPORT_SYMBOL_GPL(gretap_fb_dev_create); 1677 1678 static int __net_init ipgre_tap_init_net(struct net *net) 1679 { 1680 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0"); 1681 } 1682 1683 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net) 1684 { 1685 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops); 1686 } 1687 1688 static struct pernet_operations ipgre_tap_net_ops = { 1689 .init = ipgre_tap_init_net, 1690 .exit_batch = ipgre_tap_exit_batch_net, 1691 .id = &gre_tap_net_id, 1692 .size = sizeof(struct ip_tunnel_net), 1693 }; 1694 1695 static int __net_init erspan_init_net(struct net *net) 1696 { 1697 return ip_tunnel_init_net(net, erspan_net_id, 1698 &erspan_link_ops, "erspan0"); 1699 } 1700 1701 static void __net_exit erspan_exit_batch_net(struct list_head *net_list) 1702 { 1703 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops); 1704 } 1705 1706 static struct pernet_operations erspan_net_ops = { 1707 .init = erspan_init_net, 1708 .exit_batch = erspan_exit_batch_net, 1709 .id = &erspan_net_id, 1710 .size = sizeof(struct ip_tunnel_net), 1711 }; 1712 1713 static int __init ipgre_init(void) 1714 { 1715 int err; 1716 1717 pr_info("GRE over IPv4 tunneling driver\n"); 1718 1719 err = register_pernet_device(&ipgre_net_ops); 1720 if (err < 0) 1721 return err; 1722 1723 err = register_pernet_device(&ipgre_tap_net_ops); 1724 if (err < 0) 1725 goto pnet_tap_failed; 1726 1727 err = register_pernet_device(&erspan_net_ops); 1728 if (err < 0) 1729 goto pnet_erspan_failed; 1730 1731 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO); 1732 if (err < 0) { 1733 pr_info("%s: can't add protocol\n", __func__); 1734 goto add_proto_failed; 1735 } 1736 1737 err = rtnl_link_register(&ipgre_link_ops); 1738 if (err < 0) 1739 goto rtnl_link_failed; 1740 1741 err = rtnl_link_register(&ipgre_tap_ops); 1742 if (err < 0) 1743 goto tap_ops_failed; 1744 1745 err = rtnl_link_register(&erspan_link_ops); 1746 if (err < 0) 1747 goto erspan_link_failed; 1748 1749 return 0; 1750 1751 erspan_link_failed: 1752 rtnl_link_unregister(&ipgre_tap_ops); 1753 tap_ops_failed: 1754 rtnl_link_unregister(&ipgre_link_ops); 1755 rtnl_link_failed: 1756 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1757 add_proto_failed: 1758 unregister_pernet_device(&erspan_net_ops); 1759 pnet_erspan_failed: 1760 unregister_pernet_device(&ipgre_tap_net_ops); 1761 pnet_tap_failed: 1762 unregister_pernet_device(&ipgre_net_ops); 1763 return err; 1764 } 1765 1766 static void __exit ipgre_fini(void) 1767 { 1768 rtnl_link_unregister(&ipgre_tap_ops); 1769 rtnl_link_unregister(&ipgre_link_ops); 1770 rtnl_link_unregister(&erspan_link_ops); 1771 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1772 unregister_pernet_device(&ipgre_tap_net_ops); 1773 unregister_pernet_device(&ipgre_net_ops); 1774 unregister_pernet_device(&erspan_net_ops); 1775 } 1776 1777 module_init(ipgre_init); 1778 module_exit(ipgre_fini); 1779 MODULE_LICENSE("GPL"); 1780 MODULE_ALIAS_RTNL_LINK("gre"); 1781 MODULE_ALIAS_RTNL_LINK("gretap"); 1782 MODULE_ALIAS_RTNL_LINK("erspan"); 1783 MODULE_ALIAS_NETDEV("gre0"); 1784 MODULE_ALIAS_NETDEV("gretap0"); 1785 MODULE_ALIAS_NETDEV("erspan0"); 1786