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 const int pull_len = tunnel->hlen + sizeof(struct iphdr); 634 635 if (skb_cow_head(skb, 0)) 636 goto free_skb; 637 638 tnl_params = (const struct iphdr *)skb->data; 639 640 if (pull_len > skb_transport_offset(skb)) 641 goto free_skb; 642 643 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing 644 * to gre header. 645 */ 646 skb_pull(skb, pull_len); 647 skb_reset_mac_header(skb); 648 } else { 649 if (skb_cow_head(skb, dev->needed_headroom)) 650 goto free_skb; 651 652 tnl_params = &tunnel->parms.iph; 653 } 654 655 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 656 goto free_skb; 657 658 __gre_xmit(skb, dev, tnl_params, skb->protocol); 659 return NETDEV_TX_OK; 660 661 free_skb: 662 kfree_skb(skb); 663 dev->stats.tx_dropped++; 664 return NETDEV_TX_OK; 665 } 666 667 static netdev_tx_t erspan_xmit(struct sk_buff *skb, 668 struct net_device *dev) 669 { 670 struct ip_tunnel *tunnel = netdev_priv(dev); 671 bool truncate = false; 672 __be16 proto; 673 674 if (!pskb_inet_may_pull(skb)) 675 goto free_skb; 676 677 if (tunnel->collect_md) { 678 erspan_fb_xmit(skb, dev); 679 return NETDEV_TX_OK; 680 } 681 682 if (gre_handle_offloads(skb, false)) 683 goto free_skb; 684 685 if (skb_cow_head(skb, dev->needed_headroom)) 686 goto free_skb; 687 688 if (skb->len > dev->mtu + dev->hard_header_len) { 689 pskb_trim(skb, dev->mtu + dev->hard_header_len); 690 truncate = true; 691 } 692 693 /* Push ERSPAN header */ 694 if (tunnel->erspan_ver == 0) { 695 proto = htons(ETH_P_ERSPAN); 696 tunnel->parms.o_flags &= ~TUNNEL_SEQ; 697 } else if (tunnel->erspan_ver == 1) { 698 erspan_build_header(skb, ntohl(tunnel->parms.o_key), 699 tunnel->index, 700 truncate, true); 701 proto = htons(ETH_P_ERSPAN); 702 } else if (tunnel->erspan_ver == 2) { 703 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key), 704 tunnel->dir, tunnel->hwid, 705 truncate, true); 706 proto = htons(ETH_P_ERSPAN2); 707 } else { 708 goto free_skb; 709 } 710 711 tunnel->parms.o_flags &= ~TUNNEL_KEY; 712 __gre_xmit(skb, dev, &tunnel->parms.iph, proto); 713 return NETDEV_TX_OK; 714 715 free_skb: 716 kfree_skb(skb); 717 dev->stats.tx_dropped++; 718 return NETDEV_TX_OK; 719 } 720 721 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb, 722 struct net_device *dev) 723 { 724 struct ip_tunnel *tunnel = netdev_priv(dev); 725 726 if (!pskb_inet_may_pull(skb)) 727 goto free_skb; 728 729 if (tunnel->collect_md) { 730 gre_fb_xmit(skb, dev, htons(ETH_P_TEB)); 731 return NETDEV_TX_OK; 732 } 733 734 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM))) 735 goto free_skb; 736 737 if (skb_cow_head(skb, dev->needed_headroom)) 738 goto free_skb; 739 740 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB)); 741 return NETDEV_TX_OK; 742 743 free_skb: 744 kfree_skb(skb); 745 dev->stats.tx_dropped++; 746 return NETDEV_TX_OK; 747 } 748 749 static void ipgre_link_update(struct net_device *dev, bool set_mtu) 750 { 751 struct ip_tunnel *tunnel = netdev_priv(dev); 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 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) { 768 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) || 769 tunnel->encap.type == TUNNEL_ENCAP_NONE) { 770 dev->features |= NETIF_F_GSO_SOFTWARE; 771 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 772 } else { 773 dev->features &= ~NETIF_F_GSO_SOFTWARE; 774 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 775 } 776 dev->features |= NETIF_F_LLTX; 777 } else { 778 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE; 779 dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE); 780 } 781 } 782 783 static int ipgre_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, 784 int cmd) 785 { 786 int err; 787 788 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) { 789 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE || 790 p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) || 791 ((p->i_flags | p->o_flags) & (GRE_VERSION | GRE_ROUTING))) 792 return -EINVAL; 793 } 794 795 p->i_flags = gre_flags_to_tnl_flags(p->i_flags); 796 p->o_flags = gre_flags_to_tnl_flags(p->o_flags); 797 798 err = ip_tunnel_ctl(dev, p, cmd); 799 if (err) 800 return err; 801 802 if (cmd == SIOCCHGTUNNEL) { 803 struct ip_tunnel *t = netdev_priv(dev); 804 805 t->parms.i_flags = p->i_flags; 806 t->parms.o_flags = p->o_flags; 807 808 if (strcmp(dev->rtnl_link_ops->kind, "erspan")) 809 ipgre_link_update(dev, true); 810 } 811 812 p->i_flags = gre_tnl_flags_to_gre_flags(p->i_flags); 813 p->o_flags = gre_tnl_flags_to_gre_flags(p->o_flags); 814 return 0; 815 } 816 817 /* Nice toy. Unfortunately, useless in real life :-) 818 It allows to construct virtual multiprotocol broadcast "LAN" 819 over the Internet, provided multicast routing is tuned. 820 821 822 I have no idea was this bicycle invented before me, 823 so that I had to set ARPHRD_IPGRE to a random value. 824 I have an impression, that Cisco could make something similar, 825 but this feature is apparently missing in IOS<=11.2(8). 826 827 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks 828 with broadcast 224.66.66.66. If you have access to mbone, play with me :-) 829 830 ping -t 255 224.66.66.66 831 832 If nobody answers, mbone does not work. 833 834 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255 835 ip addr add 10.66.66.<somewhat>/24 dev Universe 836 ifconfig Universe up 837 ifconfig Universe add fe80::<Your_real_addr>/10 838 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96 839 ftp 10.66.66.66 840 ... 841 ftp fec0:6666:6666::193.233.7.65 842 ... 843 */ 844 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, 845 unsigned short type, 846 const void *daddr, const void *saddr, unsigned int len) 847 { 848 struct ip_tunnel *t = netdev_priv(dev); 849 struct iphdr *iph; 850 struct gre_base_hdr *greh; 851 852 iph = skb_push(skb, t->hlen + sizeof(*iph)); 853 greh = (struct gre_base_hdr *)(iph+1); 854 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags); 855 greh->protocol = htons(type); 856 857 memcpy(iph, &t->parms.iph, sizeof(struct iphdr)); 858 859 /* Set the source hardware address. */ 860 if (saddr) 861 memcpy(&iph->saddr, saddr, 4); 862 if (daddr) 863 memcpy(&iph->daddr, daddr, 4); 864 if (iph->daddr) 865 return t->hlen + sizeof(*iph); 866 867 return -(t->hlen + sizeof(*iph)); 868 } 869 870 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr) 871 { 872 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb); 873 memcpy(haddr, &iph->saddr, 4); 874 return 4; 875 } 876 877 static const struct header_ops ipgre_header_ops = { 878 .create = ipgre_header, 879 .parse = ipgre_header_parse, 880 }; 881 882 #ifdef CONFIG_NET_IPGRE_BROADCAST 883 static int ipgre_open(struct net_device *dev) 884 { 885 struct ip_tunnel *t = netdev_priv(dev); 886 887 if (ipv4_is_multicast(t->parms.iph.daddr)) { 888 struct flowi4 fl4; 889 struct rtable *rt; 890 891 rt = ip_route_output_gre(t->net, &fl4, 892 t->parms.iph.daddr, 893 t->parms.iph.saddr, 894 t->parms.o_key, 895 RT_TOS(t->parms.iph.tos), 896 t->parms.link); 897 if (IS_ERR(rt)) 898 return -EADDRNOTAVAIL; 899 dev = rt->dst.dev; 900 ip_rt_put(rt); 901 if (!__in_dev_get_rtnl(dev)) 902 return -EADDRNOTAVAIL; 903 t->mlink = dev->ifindex; 904 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr); 905 } 906 return 0; 907 } 908 909 static int ipgre_close(struct net_device *dev) 910 { 911 struct ip_tunnel *t = netdev_priv(dev); 912 913 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) { 914 struct in_device *in_dev; 915 in_dev = inetdev_by_index(t->net, t->mlink); 916 if (in_dev) 917 ip_mc_dec_group(in_dev, t->parms.iph.daddr); 918 } 919 return 0; 920 } 921 #endif 922 923 static const struct net_device_ops ipgre_netdev_ops = { 924 .ndo_init = ipgre_tunnel_init, 925 .ndo_uninit = ip_tunnel_uninit, 926 #ifdef CONFIG_NET_IPGRE_BROADCAST 927 .ndo_open = ipgre_open, 928 .ndo_stop = ipgre_close, 929 #endif 930 .ndo_start_xmit = ipgre_xmit, 931 .ndo_siocdevprivate = ip_tunnel_siocdevprivate, 932 .ndo_change_mtu = ip_tunnel_change_mtu, 933 .ndo_get_stats64 = dev_get_tstats64, 934 .ndo_get_iflink = ip_tunnel_get_iflink, 935 .ndo_tunnel_ctl = ipgre_tunnel_ctl, 936 }; 937 938 #define GRE_FEATURES (NETIF_F_SG | \ 939 NETIF_F_FRAGLIST | \ 940 NETIF_F_HIGHDMA | \ 941 NETIF_F_HW_CSUM) 942 943 static void ipgre_tunnel_setup(struct net_device *dev) 944 { 945 dev->netdev_ops = &ipgre_netdev_ops; 946 dev->type = ARPHRD_IPGRE; 947 ip_tunnel_setup(dev, ipgre_net_id); 948 } 949 950 static void __gre_tunnel_init(struct net_device *dev) 951 { 952 struct ip_tunnel *tunnel; 953 954 tunnel = netdev_priv(dev); 955 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags); 956 tunnel->parms.iph.protocol = IPPROTO_GRE; 957 958 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen; 959 dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph); 960 961 dev->features |= GRE_FEATURES; 962 dev->hw_features |= GRE_FEATURES; 963 964 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) { 965 /* TCP offload with GRE SEQ is not supported, nor 966 * can we support 2 levels of outer headers requiring 967 * an update. 968 */ 969 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) || 970 (tunnel->encap.type == TUNNEL_ENCAP_NONE)) { 971 dev->features |= NETIF_F_GSO_SOFTWARE; 972 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 973 } 974 975 /* Can use a lockless transmit, unless we generate 976 * output sequences 977 */ 978 dev->features |= NETIF_F_LLTX; 979 } 980 } 981 982 static int ipgre_tunnel_init(struct net_device *dev) 983 { 984 struct ip_tunnel *tunnel = netdev_priv(dev); 985 struct iphdr *iph = &tunnel->parms.iph; 986 987 __gre_tunnel_init(dev); 988 989 memcpy(dev->dev_addr, &iph->saddr, 4); 990 memcpy(dev->broadcast, &iph->daddr, 4); 991 992 dev->flags = IFF_NOARP; 993 netif_keep_dst(dev); 994 dev->addr_len = 4; 995 996 if (iph->daddr && !tunnel->collect_md) { 997 #ifdef CONFIG_NET_IPGRE_BROADCAST 998 if (ipv4_is_multicast(iph->daddr)) { 999 if (!iph->saddr) 1000 return -EINVAL; 1001 dev->flags = IFF_BROADCAST; 1002 dev->header_ops = &ipgre_header_ops; 1003 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 1004 dev->needed_headroom = 0; 1005 } 1006 #endif 1007 } else if (!tunnel->collect_md) { 1008 dev->header_ops = &ipgre_header_ops; 1009 dev->hard_header_len = tunnel->hlen + sizeof(*iph); 1010 dev->needed_headroom = 0; 1011 } 1012 1013 return ip_tunnel_init(dev); 1014 } 1015 1016 static const struct gre_protocol ipgre_protocol = { 1017 .handler = gre_rcv, 1018 .err_handler = gre_err, 1019 }; 1020 1021 static int __net_init ipgre_init_net(struct net *net) 1022 { 1023 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL); 1024 } 1025 1026 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net) 1027 { 1028 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops); 1029 } 1030 1031 static struct pernet_operations ipgre_net_ops = { 1032 .init = ipgre_init_net, 1033 .exit_batch = ipgre_exit_batch_net, 1034 .id = &ipgre_net_id, 1035 .size = sizeof(struct ip_tunnel_net), 1036 }; 1037 1038 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[], 1039 struct netlink_ext_ack *extack) 1040 { 1041 __be16 flags; 1042 1043 if (!data) 1044 return 0; 1045 1046 flags = 0; 1047 if (data[IFLA_GRE_IFLAGS]) 1048 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1049 if (data[IFLA_GRE_OFLAGS]) 1050 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1051 if (flags & (GRE_VERSION|GRE_ROUTING)) 1052 return -EINVAL; 1053 1054 if (data[IFLA_GRE_COLLECT_METADATA] && 1055 data[IFLA_GRE_ENCAP_TYPE] && 1056 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE) 1057 return -EINVAL; 1058 1059 return 0; 1060 } 1061 1062 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[], 1063 struct netlink_ext_ack *extack) 1064 { 1065 __be32 daddr; 1066 1067 if (tb[IFLA_ADDRESS]) { 1068 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1069 return -EINVAL; 1070 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1071 return -EADDRNOTAVAIL; 1072 } 1073 1074 if (!data) 1075 goto out; 1076 1077 if (data[IFLA_GRE_REMOTE]) { 1078 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4); 1079 if (!daddr) 1080 return -EINVAL; 1081 } 1082 1083 out: 1084 return ipgre_tunnel_validate(tb, data, extack); 1085 } 1086 1087 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[], 1088 struct netlink_ext_ack *extack) 1089 { 1090 __be16 flags = 0; 1091 int ret; 1092 1093 if (!data) 1094 return 0; 1095 1096 ret = ipgre_tap_validate(tb, data, extack); 1097 if (ret) 1098 return ret; 1099 1100 if (data[IFLA_GRE_ERSPAN_VER] && 1101 nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0) 1102 return 0; 1103 1104 /* ERSPAN type II/III should only have GRE sequence and key flag */ 1105 if (data[IFLA_GRE_OFLAGS]) 1106 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); 1107 if (data[IFLA_GRE_IFLAGS]) 1108 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); 1109 if (!data[IFLA_GRE_COLLECT_METADATA] && 1110 flags != (GRE_SEQ | GRE_KEY)) 1111 return -EINVAL; 1112 1113 /* ERSPAN Session ID only has 10-bit. Since we reuse 1114 * 32-bit key field as ID, check it's range. 1115 */ 1116 if (data[IFLA_GRE_IKEY] && 1117 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK)) 1118 return -EINVAL; 1119 1120 if (data[IFLA_GRE_OKEY] && 1121 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK)) 1122 return -EINVAL; 1123 1124 return 0; 1125 } 1126 1127 static int ipgre_netlink_parms(struct net_device *dev, 1128 struct nlattr *data[], 1129 struct nlattr *tb[], 1130 struct ip_tunnel_parm *parms, 1131 __u32 *fwmark) 1132 { 1133 struct ip_tunnel *t = netdev_priv(dev); 1134 1135 memset(parms, 0, sizeof(*parms)); 1136 1137 parms->iph.protocol = IPPROTO_GRE; 1138 1139 if (!data) 1140 return 0; 1141 1142 if (data[IFLA_GRE_LINK]) 1143 parms->link = nla_get_u32(data[IFLA_GRE_LINK]); 1144 1145 if (data[IFLA_GRE_IFLAGS]) 1146 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS])); 1147 1148 if (data[IFLA_GRE_OFLAGS]) 1149 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS])); 1150 1151 if (data[IFLA_GRE_IKEY]) 1152 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]); 1153 1154 if (data[IFLA_GRE_OKEY]) 1155 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]); 1156 1157 if (data[IFLA_GRE_LOCAL]) 1158 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]); 1159 1160 if (data[IFLA_GRE_REMOTE]) 1161 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]); 1162 1163 if (data[IFLA_GRE_TTL]) 1164 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]); 1165 1166 if (data[IFLA_GRE_TOS]) 1167 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]); 1168 1169 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) { 1170 if (t->ignore_df) 1171 return -EINVAL; 1172 parms->iph.frag_off = htons(IP_DF); 1173 } 1174 1175 if (data[IFLA_GRE_COLLECT_METADATA]) { 1176 t->collect_md = true; 1177 if (dev->type == ARPHRD_IPGRE) 1178 dev->type = ARPHRD_NONE; 1179 } 1180 1181 if (data[IFLA_GRE_IGNORE_DF]) { 1182 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF]) 1183 && (parms->iph.frag_off & htons(IP_DF))) 1184 return -EINVAL; 1185 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]); 1186 } 1187 1188 if (data[IFLA_GRE_FWMARK]) 1189 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]); 1190 1191 return 0; 1192 } 1193 1194 static int erspan_netlink_parms(struct net_device *dev, 1195 struct nlattr *data[], 1196 struct nlattr *tb[], 1197 struct ip_tunnel_parm *parms, 1198 __u32 *fwmark) 1199 { 1200 struct ip_tunnel *t = netdev_priv(dev); 1201 int err; 1202 1203 err = ipgre_netlink_parms(dev, data, tb, parms, fwmark); 1204 if (err) 1205 return err; 1206 if (!data) 1207 return 0; 1208 1209 if (data[IFLA_GRE_ERSPAN_VER]) { 1210 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]); 1211 1212 if (t->erspan_ver > 2) 1213 return -EINVAL; 1214 } 1215 1216 if (t->erspan_ver == 1) { 1217 if (data[IFLA_GRE_ERSPAN_INDEX]) { 1218 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]); 1219 if (t->index & ~INDEX_MASK) 1220 return -EINVAL; 1221 } 1222 } else if (t->erspan_ver == 2) { 1223 if (data[IFLA_GRE_ERSPAN_DIR]) { 1224 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]); 1225 if (t->dir & ~(DIR_MASK >> DIR_OFFSET)) 1226 return -EINVAL; 1227 } 1228 if (data[IFLA_GRE_ERSPAN_HWID]) { 1229 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]); 1230 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET)) 1231 return -EINVAL; 1232 } 1233 } 1234 1235 return 0; 1236 } 1237 1238 /* This function returns true when ENCAP attributes are present in the nl msg */ 1239 static bool ipgre_netlink_encap_parms(struct nlattr *data[], 1240 struct ip_tunnel_encap *ipencap) 1241 { 1242 bool ret = false; 1243 1244 memset(ipencap, 0, sizeof(*ipencap)); 1245 1246 if (!data) 1247 return ret; 1248 1249 if (data[IFLA_GRE_ENCAP_TYPE]) { 1250 ret = true; 1251 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]); 1252 } 1253 1254 if (data[IFLA_GRE_ENCAP_FLAGS]) { 1255 ret = true; 1256 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]); 1257 } 1258 1259 if (data[IFLA_GRE_ENCAP_SPORT]) { 1260 ret = true; 1261 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]); 1262 } 1263 1264 if (data[IFLA_GRE_ENCAP_DPORT]) { 1265 ret = true; 1266 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]); 1267 } 1268 1269 return ret; 1270 } 1271 1272 static int gre_tap_init(struct net_device *dev) 1273 { 1274 __gre_tunnel_init(dev); 1275 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1276 netif_keep_dst(dev); 1277 1278 return ip_tunnel_init(dev); 1279 } 1280 1281 static const struct net_device_ops gre_tap_netdev_ops = { 1282 .ndo_init = gre_tap_init, 1283 .ndo_uninit = ip_tunnel_uninit, 1284 .ndo_start_xmit = gre_tap_xmit, 1285 .ndo_set_mac_address = eth_mac_addr, 1286 .ndo_validate_addr = eth_validate_addr, 1287 .ndo_change_mtu = ip_tunnel_change_mtu, 1288 .ndo_get_stats64 = dev_get_tstats64, 1289 .ndo_get_iflink = ip_tunnel_get_iflink, 1290 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1291 }; 1292 1293 static int erspan_tunnel_init(struct net_device *dev) 1294 { 1295 struct ip_tunnel *tunnel = netdev_priv(dev); 1296 1297 if (tunnel->erspan_ver == 0) 1298 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */ 1299 else 1300 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */ 1301 1302 tunnel->parms.iph.protocol = IPPROTO_GRE; 1303 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen + 1304 erspan_hdr_len(tunnel->erspan_ver); 1305 1306 dev->features |= GRE_FEATURES; 1307 dev->hw_features |= GRE_FEATURES; 1308 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1309 netif_keep_dst(dev); 1310 1311 return ip_tunnel_init(dev); 1312 } 1313 1314 static const struct net_device_ops erspan_netdev_ops = { 1315 .ndo_init = erspan_tunnel_init, 1316 .ndo_uninit = ip_tunnel_uninit, 1317 .ndo_start_xmit = erspan_xmit, 1318 .ndo_set_mac_address = eth_mac_addr, 1319 .ndo_validate_addr = eth_validate_addr, 1320 .ndo_change_mtu = ip_tunnel_change_mtu, 1321 .ndo_get_stats64 = dev_get_tstats64, 1322 .ndo_get_iflink = ip_tunnel_get_iflink, 1323 .ndo_fill_metadata_dst = gre_fill_metadata_dst, 1324 }; 1325 1326 static void ipgre_tap_setup(struct net_device *dev) 1327 { 1328 ether_setup(dev); 1329 dev->max_mtu = 0; 1330 dev->netdev_ops = &gre_tap_netdev_ops; 1331 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1332 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1333 ip_tunnel_setup(dev, gre_tap_net_id); 1334 } 1335 1336 static int 1337 ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[]) 1338 { 1339 struct ip_tunnel_encap ipencap; 1340 1341 if (ipgre_netlink_encap_parms(data, &ipencap)) { 1342 struct ip_tunnel *t = netdev_priv(dev); 1343 int err = ip_tunnel_encap_setup(t, &ipencap); 1344 1345 if (err < 0) 1346 return err; 1347 } 1348 1349 return 0; 1350 } 1351 1352 static int ipgre_newlink(struct net *src_net, struct net_device *dev, 1353 struct nlattr *tb[], struct nlattr *data[], 1354 struct netlink_ext_ack *extack) 1355 { 1356 struct ip_tunnel_parm p; 1357 __u32 fwmark = 0; 1358 int err; 1359 1360 err = ipgre_newlink_encap_setup(dev, data); 1361 if (err) 1362 return err; 1363 1364 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1365 if (err < 0) 1366 return err; 1367 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1368 } 1369 1370 static int erspan_newlink(struct net *src_net, struct net_device *dev, 1371 struct nlattr *tb[], struct nlattr *data[], 1372 struct netlink_ext_ack *extack) 1373 { 1374 struct ip_tunnel_parm p; 1375 __u32 fwmark = 0; 1376 int err; 1377 1378 err = ipgre_newlink_encap_setup(dev, data); 1379 if (err) 1380 return err; 1381 1382 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1383 if (err) 1384 return err; 1385 return ip_tunnel_newlink(dev, tb, &p, fwmark); 1386 } 1387 1388 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[], 1389 struct nlattr *data[], 1390 struct netlink_ext_ack *extack) 1391 { 1392 struct ip_tunnel *t = netdev_priv(dev); 1393 __u32 fwmark = t->fwmark; 1394 struct ip_tunnel_parm p; 1395 int err; 1396 1397 err = ipgre_newlink_encap_setup(dev, data); 1398 if (err) 1399 return err; 1400 1401 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark); 1402 if (err < 0) 1403 return err; 1404 1405 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1406 if (err < 0) 1407 return err; 1408 1409 t->parms.i_flags = p.i_flags; 1410 t->parms.o_flags = p.o_flags; 1411 1412 ipgre_link_update(dev, !tb[IFLA_MTU]); 1413 1414 return 0; 1415 } 1416 1417 static int erspan_changelink(struct net_device *dev, struct nlattr *tb[], 1418 struct nlattr *data[], 1419 struct netlink_ext_ack *extack) 1420 { 1421 struct ip_tunnel *t = netdev_priv(dev); 1422 __u32 fwmark = t->fwmark; 1423 struct ip_tunnel_parm p; 1424 int err; 1425 1426 err = ipgre_newlink_encap_setup(dev, data); 1427 if (err) 1428 return err; 1429 1430 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark); 1431 if (err < 0) 1432 return err; 1433 1434 err = ip_tunnel_changelink(dev, tb, &p, fwmark); 1435 if (err < 0) 1436 return err; 1437 1438 t->parms.i_flags = p.i_flags; 1439 t->parms.o_flags = p.o_flags; 1440 1441 return 0; 1442 } 1443 1444 static size_t ipgre_get_size(const struct net_device *dev) 1445 { 1446 return 1447 /* IFLA_GRE_LINK */ 1448 nla_total_size(4) + 1449 /* IFLA_GRE_IFLAGS */ 1450 nla_total_size(2) + 1451 /* IFLA_GRE_OFLAGS */ 1452 nla_total_size(2) + 1453 /* IFLA_GRE_IKEY */ 1454 nla_total_size(4) + 1455 /* IFLA_GRE_OKEY */ 1456 nla_total_size(4) + 1457 /* IFLA_GRE_LOCAL */ 1458 nla_total_size(4) + 1459 /* IFLA_GRE_REMOTE */ 1460 nla_total_size(4) + 1461 /* IFLA_GRE_TTL */ 1462 nla_total_size(1) + 1463 /* IFLA_GRE_TOS */ 1464 nla_total_size(1) + 1465 /* IFLA_GRE_PMTUDISC */ 1466 nla_total_size(1) + 1467 /* IFLA_GRE_ENCAP_TYPE */ 1468 nla_total_size(2) + 1469 /* IFLA_GRE_ENCAP_FLAGS */ 1470 nla_total_size(2) + 1471 /* IFLA_GRE_ENCAP_SPORT */ 1472 nla_total_size(2) + 1473 /* IFLA_GRE_ENCAP_DPORT */ 1474 nla_total_size(2) + 1475 /* IFLA_GRE_COLLECT_METADATA */ 1476 nla_total_size(0) + 1477 /* IFLA_GRE_IGNORE_DF */ 1478 nla_total_size(1) + 1479 /* IFLA_GRE_FWMARK */ 1480 nla_total_size(4) + 1481 /* IFLA_GRE_ERSPAN_INDEX */ 1482 nla_total_size(4) + 1483 /* IFLA_GRE_ERSPAN_VER */ 1484 nla_total_size(1) + 1485 /* IFLA_GRE_ERSPAN_DIR */ 1486 nla_total_size(1) + 1487 /* IFLA_GRE_ERSPAN_HWID */ 1488 nla_total_size(2) + 1489 0; 1490 } 1491 1492 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev) 1493 { 1494 struct ip_tunnel *t = netdev_priv(dev); 1495 struct ip_tunnel_parm *p = &t->parms; 1496 __be16 o_flags = p->o_flags; 1497 1498 if (t->erspan_ver <= 2) { 1499 if (t->erspan_ver != 0 && !t->collect_md) 1500 o_flags |= TUNNEL_KEY; 1501 1502 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver)) 1503 goto nla_put_failure; 1504 1505 if (t->erspan_ver == 1) { 1506 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index)) 1507 goto nla_put_failure; 1508 } else if (t->erspan_ver == 2) { 1509 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir)) 1510 goto nla_put_failure; 1511 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid)) 1512 goto nla_put_failure; 1513 } 1514 } 1515 1516 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) || 1517 nla_put_be16(skb, IFLA_GRE_IFLAGS, 1518 gre_tnl_flags_to_gre_flags(p->i_flags)) || 1519 nla_put_be16(skb, IFLA_GRE_OFLAGS, 1520 gre_tnl_flags_to_gre_flags(o_flags)) || 1521 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) || 1522 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) || 1523 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) || 1524 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) || 1525 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) || 1526 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) || 1527 nla_put_u8(skb, IFLA_GRE_PMTUDISC, 1528 !!(p->iph.frag_off & htons(IP_DF))) || 1529 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark)) 1530 goto nla_put_failure; 1531 1532 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE, 1533 t->encap.type) || 1534 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT, 1535 t->encap.sport) || 1536 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT, 1537 t->encap.dport) || 1538 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS, 1539 t->encap.flags)) 1540 goto nla_put_failure; 1541 1542 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df)) 1543 goto nla_put_failure; 1544 1545 if (t->collect_md) { 1546 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA)) 1547 goto nla_put_failure; 1548 } 1549 1550 return 0; 1551 1552 nla_put_failure: 1553 return -EMSGSIZE; 1554 } 1555 1556 static void erspan_setup(struct net_device *dev) 1557 { 1558 struct ip_tunnel *t = netdev_priv(dev); 1559 1560 ether_setup(dev); 1561 dev->max_mtu = 0; 1562 dev->netdev_ops = &erspan_netdev_ops; 1563 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1564 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1565 ip_tunnel_setup(dev, erspan_net_id); 1566 t->erspan_ver = 1; 1567 } 1568 1569 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = { 1570 [IFLA_GRE_LINK] = { .type = NLA_U32 }, 1571 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 }, 1572 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 }, 1573 [IFLA_GRE_IKEY] = { .type = NLA_U32 }, 1574 [IFLA_GRE_OKEY] = { .type = NLA_U32 }, 1575 [IFLA_GRE_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 1576 [IFLA_GRE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1577 [IFLA_GRE_TTL] = { .type = NLA_U8 }, 1578 [IFLA_GRE_TOS] = { .type = NLA_U8 }, 1579 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 }, 1580 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 }, 1581 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 }, 1582 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 }, 1583 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 }, 1584 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1585 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 }, 1586 [IFLA_GRE_FWMARK] = { .type = NLA_U32 }, 1587 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 }, 1588 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 }, 1589 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 }, 1590 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 }, 1591 }; 1592 1593 static struct rtnl_link_ops ipgre_link_ops __read_mostly = { 1594 .kind = "gre", 1595 .maxtype = IFLA_GRE_MAX, 1596 .policy = ipgre_policy, 1597 .priv_size = sizeof(struct ip_tunnel), 1598 .setup = ipgre_tunnel_setup, 1599 .validate = ipgre_tunnel_validate, 1600 .newlink = ipgre_newlink, 1601 .changelink = ipgre_changelink, 1602 .dellink = ip_tunnel_dellink, 1603 .get_size = ipgre_get_size, 1604 .fill_info = ipgre_fill_info, 1605 .get_link_net = ip_tunnel_get_link_net, 1606 }; 1607 1608 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = { 1609 .kind = "gretap", 1610 .maxtype = IFLA_GRE_MAX, 1611 .policy = ipgre_policy, 1612 .priv_size = sizeof(struct ip_tunnel), 1613 .setup = ipgre_tap_setup, 1614 .validate = ipgre_tap_validate, 1615 .newlink = ipgre_newlink, 1616 .changelink = ipgre_changelink, 1617 .dellink = ip_tunnel_dellink, 1618 .get_size = ipgre_get_size, 1619 .fill_info = ipgre_fill_info, 1620 .get_link_net = ip_tunnel_get_link_net, 1621 }; 1622 1623 static struct rtnl_link_ops erspan_link_ops __read_mostly = { 1624 .kind = "erspan", 1625 .maxtype = IFLA_GRE_MAX, 1626 .policy = ipgre_policy, 1627 .priv_size = sizeof(struct ip_tunnel), 1628 .setup = erspan_setup, 1629 .validate = erspan_validate, 1630 .newlink = erspan_newlink, 1631 .changelink = erspan_changelink, 1632 .dellink = ip_tunnel_dellink, 1633 .get_size = ipgre_get_size, 1634 .fill_info = ipgre_fill_info, 1635 .get_link_net = ip_tunnel_get_link_net, 1636 }; 1637 1638 struct net_device *gretap_fb_dev_create(struct net *net, const char *name, 1639 u8 name_assign_type) 1640 { 1641 struct nlattr *tb[IFLA_MAX + 1]; 1642 struct net_device *dev; 1643 LIST_HEAD(list_kill); 1644 struct ip_tunnel *t; 1645 int err; 1646 1647 memset(&tb, 0, sizeof(tb)); 1648 1649 dev = rtnl_create_link(net, name, name_assign_type, 1650 &ipgre_tap_ops, tb, NULL); 1651 if (IS_ERR(dev)) 1652 return dev; 1653 1654 /* Configure flow based GRE device. */ 1655 t = netdev_priv(dev); 1656 t->collect_md = true; 1657 1658 err = ipgre_newlink(net, dev, tb, NULL, NULL); 1659 if (err < 0) { 1660 free_netdev(dev); 1661 return ERR_PTR(err); 1662 } 1663 1664 /* openvswitch users expect packet sizes to be unrestricted, 1665 * so set the largest MTU we can. 1666 */ 1667 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false); 1668 if (err) 1669 goto out; 1670 1671 err = rtnl_configure_link(dev, NULL); 1672 if (err < 0) 1673 goto out; 1674 1675 return dev; 1676 out: 1677 ip_tunnel_dellink(dev, &list_kill); 1678 unregister_netdevice_many(&list_kill); 1679 return ERR_PTR(err); 1680 } 1681 EXPORT_SYMBOL_GPL(gretap_fb_dev_create); 1682 1683 static int __net_init ipgre_tap_init_net(struct net *net) 1684 { 1685 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0"); 1686 } 1687 1688 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net) 1689 { 1690 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops); 1691 } 1692 1693 static struct pernet_operations ipgre_tap_net_ops = { 1694 .init = ipgre_tap_init_net, 1695 .exit_batch = ipgre_tap_exit_batch_net, 1696 .id = &gre_tap_net_id, 1697 .size = sizeof(struct ip_tunnel_net), 1698 }; 1699 1700 static int __net_init erspan_init_net(struct net *net) 1701 { 1702 return ip_tunnel_init_net(net, erspan_net_id, 1703 &erspan_link_ops, "erspan0"); 1704 } 1705 1706 static void __net_exit erspan_exit_batch_net(struct list_head *net_list) 1707 { 1708 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops); 1709 } 1710 1711 static struct pernet_operations erspan_net_ops = { 1712 .init = erspan_init_net, 1713 .exit_batch = erspan_exit_batch_net, 1714 .id = &erspan_net_id, 1715 .size = sizeof(struct ip_tunnel_net), 1716 }; 1717 1718 static int __init ipgre_init(void) 1719 { 1720 int err; 1721 1722 pr_info("GRE over IPv4 tunneling driver\n"); 1723 1724 err = register_pernet_device(&ipgre_net_ops); 1725 if (err < 0) 1726 return err; 1727 1728 err = register_pernet_device(&ipgre_tap_net_ops); 1729 if (err < 0) 1730 goto pnet_tap_failed; 1731 1732 err = register_pernet_device(&erspan_net_ops); 1733 if (err < 0) 1734 goto pnet_erspan_failed; 1735 1736 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO); 1737 if (err < 0) { 1738 pr_info("%s: can't add protocol\n", __func__); 1739 goto add_proto_failed; 1740 } 1741 1742 err = rtnl_link_register(&ipgre_link_ops); 1743 if (err < 0) 1744 goto rtnl_link_failed; 1745 1746 err = rtnl_link_register(&ipgre_tap_ops); 1747 if (err < 0) 1748 goto tap_ops_failed; 1749 1750 err = rtnl_link_register(&erspan_link_ops); 1751 if (err < 0) 1752 goto erspan_link_failed; 1753 1754 return 0; 1755 1756 erspan_link_failed: 1757 rtnl_link_unregister(&ipgre_tap_ops); 1758 tap_ops_failed: 1759 rtnl_link_unregister(&ipgre_link_ops); 1760 rtnl_link_failed: 1761 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1762 add_proto_failed: 1763 unregister_pernet_device(&erspan_net_ops); 1764 pnet_erspan_failed: 1765 unregister_pernet_device(&ipgre_tap_net_ops); 1766 pnet_tap_failed: 1767 unregister_pernet_device(&ipgre_net_ops); 1768 return err; 1769 } 1770 1771 static void __exit ipgre_fini(void) 1772 { 1773 rtnl_link_unregister(&ipgre_tap_ops); 1774 rtnl_link_unregister(&ipgre_link_ops); 1775 rtnl_link_unregister(&erspan_link_ops); 1776 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); 1777 unregister_pernet_device(&ipgre_tap_net_ops); 1778 unregister_pernet_device(&ipgre_net_ops); 1779 unregister_pernet_device(&erspan_net_ops); 1780 } 1781 1782 module_init(ipgre_init); 1783 module_exit(ipgre_fini); 1784 MODULE_LICENSE("GPL"); 1785 MODULE_ALIAS_RTNL_LINK("gre"); 1786 MODULE_ALIAS_RTNL_LINK("gretap"); 1787 MODULE_ALIAS_RTNL_LINK("erspan"); 1788 MODULE_ALIAS_NETDEV("gre0"); 1789 MODULE_ALIAS_NETDEV("gretap0"); 1790 MODULE_ALIAS_NETDEV("erspan0"); 1791