Loading...
1/*
2 * Linux NET3: GRE over IP protocol decoder.
3 *
4 * Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/capability.h>
16#include <linux/module.h>
17#include <linux/types.h>
18#include <linux/kernel.h>
19#include <linux/slab.h>
20#include <linux/uaccess.h>
21#include <linux/skbuff.h>
22#include <linux/netdevice.h>
23#include <linux/in.h>
24#include <linux/tcp.h>
25#include <linux/udp.h>
26#include <linux/if_arp.h>
27#include <linux/if_vlan.h>
28#include <linux/init.h>
29#include <linux/in6.h>
30#include <linux/inetdevice.h>
31#include <linux/igmp.h>
32#include <linux/netfilter_ipv4.h>
33#include <linux/etherdevice.h>
34#include <linux/if_ether.h>
35
36#include <net/sock.h>
37#include <net/ip.h>
38#include <net/icmp.h>
39#include <net/protocol.h>
40#include <net/ip_tunnels.h>
41#include <net/arp.h>
42#include <net/checksum.h>
43#include <net/dsfield.h>
44#include <net/inet_ecn.h>
45#include <net/xfrm.h>
46#include <net/net_namespace.h>
47#include <net/netns/generic.h>
48#include <net/rtnetlink.h>
49#include <net/gre.h>
50#include <net/dst_metadata.h>
51#include <net/erspan.h>
52
53/*
54 Problems & solutions
55 --------------------
56
57 1. The most important issue is detecting local dead loops.
58 They would cause complete host lockup in transmit, which
59 would be "resolved" by stack overflow or, if queueing is enabled,
60 with infinite looping in net_bh.
61
62 We cannot track such dead loops during route installation,
63 it is infeasible task. The most general solutions would be
64 to keep skb->encapsulation counter (sort of local ttl),
65 and silently drop packet when it expires. It is a good
66 solution, but it supposes maintaining new variable in ALL
67 skb, even if no tunneling is used.
68
69 Current solution: xmit_recursion breaks dead loops. This is a percpu
70 counter, since when we enter the first ndo_xmit(), cpu migration is
71 forbidden. We force an exit if this counter reaches RECURSION_LIMIT
72
73 2. Networking dead loops would not kill routers, but would really
74 kill network. IP hop limit plays role of "t->recursion" in this case,
75 if we copy it from packet being encapsulated to upper header.
76 It is very good solution, but it introduces two problems:
77
78 - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
79 do not work over tunnels.
80 - traceroute does not work. I planned to relay ICMP from tunnel,
81 so that this problem would be solved and traceroute output
82 would even more informative. This idea appeared to be wrong:
83 only Linux complies to rfc1812 now (yes, guys, Linux is the only
84 true router now :-)), all routers (at least, in neighbourhood of mine)
85 return only 8 bytes of payload. It is the end.
86
87 Hence, if we want that OSPF worked or traceroute said something reasonable,
88 we should search for another solution.
89
90 One of them is to parse packet trying to detect inner encapsulation
91 made by our node. It is difficult or even impossible, especially,
92 taking into account fragmentation. TO be short, ttl is not solution at all.
93
94 Current solution: The solution was UNEXPECTEDLY SIMPLE.
95 We force DF flag on tunnels with preconfigured hop limit,
96 that is ALL. :-) Well, it does not remove the problem completely,
97 but exponential growth of network traffic is changed to linear
98 (branches, that exceed pmtu are pruned) and tunnel mtu
99 rapidly degrades to value <68, where looping stops.
100 Yes, it is not good if there exists a router in the loop,
101 which does not force DF, even when encapsulating packets have DF set.
102 But it is not our problem! Nobody could accuse us, we made
103 all that we could make. Even if it is your gated who injected
104 fatal route to network, even if it were you who configured
105 fatal static route: you are innocent. :-)
106
107 Alexey Kuznetsov.
108 */
109
110static bool log_ecn_error = true;
111module_param(log_ecn_error, bool, 0644);
112MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
113
114static struct rtnl_link_ops ipgre_link_ops __read_mostly;
115static int ipgre_tunnel_init(struct net_device *dev);
116static void erspan_build_header(struct sk_buff *skb,
117 u32 id, u32 index,
118 bool truncate, bool is_ipv4);
119
120static unsigned int ipgre_net_id __read_mostly;
121static unsigned int gre_tap_net_id __read_mostly;
122static unsigned int erspan_net_id __read_mostly;
123
124static void ipgre_err(struct sk_buff *skb, u32 info,
125 const struct tnl_ptk_info *tpi)
126{
127
128 /* All the routers (except for Linux) return only
129 8 bytes of packet payload. It means, that precise relaying of
130 ICMP in the real Internet is absolutely infeasible.
131
132 Moreover, Cisco "wise men" put GRE key to the third word
133 in GRE header. It makes impossible maintaining even soft
134 state for keyed GRE tunnels with enabled checksum. Tell
135 them "thank you".
136
137 Well, I wonder, rfc1812 was written by Cisco employee,
138 what the hell these idiots break standards established
139 by themselves???
140 */
141 struct net *net = dev_net(skb->dev);
142 struct ip_tunnel_net *itn;
143 const struct iphdr *iph;
144 const int type = icmp_hdr(skb)->type;
145 const int code = icmp_hdr(skb)->code;
146 unsigned int data_len = 0;
147 struct ip_tunnel *t;
148
149 switch (type) {
150 default:
151 case ICMP_PARAMETERPROB:
152 return;
153
154 case ICMP_DEST_UNREACH:
155 switch (code) {
156 case ICMP_SR_FAILED:
157 case ICMP_PORT_UNREACH:
158 /* Impossible event. */
159 return;
160 default:
161 /* All others are translated to HOST_UNREACH.
162 rfc2003 contains "deep thoughts" about NET_UNREACH,
163 I believe they are just ether pollution. --ANK
164 */
165 break;
166 }
167 break;
168
169 case ICMP_TIME_EXCEEDED:
170 if (code != ICMP_EXC_TTL)
171 return;
172 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
173 break;
174
175 case ICMP_REDIRECT:
176 break;
177 }
178
179 if (tpi->proto == htons(ETH_P_TEB))
180 itn = net_generic(net, gre_tap_net_id);
181 else
182 itn = net_generic(net, ipgre_net_id);
183
184 iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
185 t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
186 iph->daddr, iph->saddr, tpi->key);
187
188 if (!t)
189 return;
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;
196#endif
197
198 if (t->parms.iph.daddr == 0 ||
199 ipv4_is_multicast(t->parms.iph.daddr))
200 return;
201
202 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
203 return;
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
212static void gre_err(struct sk_buff *skb, u32 info)
213{
214 /* All the routers (except for Linux) return only
215 * 8 bytes of packet payload. It means, that precise relaying of
216 * ICMP in the real Internet is absolutely infeasible.
217 *
218 * Moreover, Cisco "wise men" put GRE key to the third word
219 * in GRE header. It makes impossible maintaining even soft
220 * state for keyed
221 * GRE tunnels with enabled checksum. Tell them "thank you".
222 *
223 * Well, I wonder, rfc1812 was written by Cisco employee,
224 * what the hell these idiots break standards established
225 * by themselves???
226 */
227
228 const struct iphdr *iph = (struct iphdr *)skb->data;
229 const int type = icmp_hdr(skb)->type;
230 const int code = icmp_hdr(skb)->code;
231 struct tnl_ptk_info tpi;
232 bool csum_err = false;
233
234 if (gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP),
235 iph->ihl * 4) < 0) {
236 if (!csum_err) /* ignore csum errors. */
237 return;
238 }
239
240 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
241 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
242 skb->dev->ifindex, 0, IPPROTO_GRE, 0);
243 return;
244 }
245 if (type == ICMP_REDIRECT) {
246 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
247 IPPROTO_GRE, 0);
248 return;
249 }
250
251 ipgre_err(skb, info, &tpi);
252}
253
254static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
255 int gre_hdr_len)
256{
257 struct net *net = dev_net(skb->dev);
258 struct metadata_dst *tun_dst = NULL;
259 struct erspan_base_hdr *ershdr;
260 struct erspan_metadata *pkt_md;
261 struct ip_tunnel_net *itn;
262 struct ip_tunnel *tunnel;
263 const struct iphdr *iph;
264 struct erspan_md2 *md2;
265 int ver;
266 int len;
267
268 itn = net_generic(net, erspan_net_id);
269 len = gre_hdr_len + sizeof(*ershdr);
270
271 /* Check based hdr len */
272 if (unlikely(!pskb_may_pull(skb, len)))
273 return PACKET_REJECT;
274
275 iph = ip_hdr(skb);
276 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
277 ver = ershdr->ver;
278
279 /* The original GRE header does not have key field,
280 * Use ERSPAN 10-bit session ID as key.
281 */
282 tpi->key = cpu_to_be32(get_session_id(ershdr));
283 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
284 tpi->flags | TUNNEL_KEY,
285 iph->saddr, iph->daddr, tpi->key);
286
287 if (tunnel) {
288 len = gre_hdr_len + erspan_hdr_len(ver);
289 if (unlikely(!pskb_may_pull(skb, len)))
290 return PACKET_REJECT;
291
292 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
293 pkt_md = (struct erspan_metadata *)(ershdr + 1);
294
295 if (__iptunnel_pull_header(skb,
296 len,
297 htons(ETH_P_TEB),
298 false, false) < 0)
299 goto drop;
300
301 if (tunnel->collect_md) {
302 struct ip_tunnel_info *info;
303 struct erspan_metadata *md;
304 __be64 tun_id;
305 __be16 flags;
306
307 tpi->flags |= TUNNEL_KEY;
308 flags = tpi->flags;
309 tun_id = key32_to_tunnel_id(tpi->key);
310
311 tun_dst = ip_tun_rx_dst(skb, flags,
312 tun_id, sizeof(*md));
313 if (!tun_dst)
314 return PACKET_REJECT;
315
316 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
317 md->version = ver;
318 md2 = &md->u.md2;
319 memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
320 ERSPAN_V2_MDSIZE);
321
322 info = &tun_dst->u.tun_info;
323 info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
324 info->options_len = sizeof(*md);
325 }
326
327 skb_reset_mac_header(skb);
328 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
329 return PACKET_RCVD;
330 }
331drop:
332 kfree_skb(skb);
333 return PACKET_RCVD;
334}
335
336static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
337 struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
338{
339 struct metadata_dst *tun_dst = NULL;
340 const struct iphdr *iph;
341 struct ip_tunnel *tunnel;
342
343 iph = ip_hdr(skb);
344 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
345 iph->saddr, iph->daddr, tpi->key);
346
347 if (tunnel) {
348 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
349 raw_proto, false) < 0)
350 goto drop;
351
352 if (tunnel->dev->type != ARPHRD_NONE)
353 skb_pop_mac_header(skb);
354 else
355 skb_reset_mac_header(skb);
356 if (tunnel->collect_md) {
357 __be16 flags;
358 __be64 tun_id;
359
360 flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
361 tun_id = key32_to_tunnel_id(tpi->key);
362 tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
363 if (!tun_dst)
364 return PACKET_REJECT;
365 }
366
367 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
368 return PACKET_RCVD;
369 }
370 return PACKET_NEXT;
371
372drop:
373 kfree_skb(skb);
374 return PACKET_RCVD;
375}
376
377static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
378 int hdr_len)
379{
380 struct net *net = dev_net(skb->dev);
381 struct ip_tunnel_net *itn;
382 int res;
383
384 if (tpi->proto == htons(ETH_P_TEB))
385 itn = net_generic(net, gre_tap_net_id);
386 else
387 itn = net_generic(net, ipgre_net_id);
388
389 res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
390 if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
391 /* ipgre tunnels in collect metadata mode should receive
392 * also ETH_P_TEB traffic.
393 */
394 itn = net_generic(net, ipgre_net_id);
395 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
396 }
397 return res;
398}
399
400static int gre_rcv(struct sk_buff *skb)
401{
402 struct tnl_ptk_info tpi;
403 bool csum_err = false;
404 int hdr_len;
405
406#ifdef CONFIG_NET_IPGRE_BROADCAST
407 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
408 /* Looped back packet, drop it! */
409 if (rt_is_output_route(skb_rtable(skb)))
410 goto drop;
411 }
412#endif
413
414 hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
415 if (hdr_len < 0)
416 goto drop;
417
418 if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
419 tpi.proto == htons(ETH_P_ERSPAN2))) {
420 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
421 return 0;
422 goto out;
423 }
424
425 if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
426 return 0;
427
428out:
429 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
430drop:
431 kfree_skb(skb);
432 return 0;
433}
434
435static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
436 const struct iphdr *tnl_params,
437 __be16 proto)
438{
439 struct ip_tunnel *tunnel = netdev_priv(dev);
440
441 if (tunnel->parms.o_flags & TUNNEL_SEQ)
442 tunnel->o_seqno++;
443
444 /* Push GRE header. */
445 gre_build_header(skb, tunnel->tun_hlen,
446 tunnel->parms.o_flags, proto, tunnel->parms.o_key,
447 htonl(tunnel->o_seqno));
448
449 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
450}
451
452static int gre_handle_offloads(struct sk_buff *skb, bool csum)
453{
454 return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
455}
456
457static struct rtable *gre_get_rt(struct sk_buff *skb,
458 struct net_device *dev,
459 struct flowi4 *fl,
460 const struct ip_tunnel_key *key)
461{
462 struct net *net = dev_net(dev);
463
464 memset(fl, 0, sizeof(*fl));
465 fl->daddr = key->u.ipv4.dst;
466 fl->saddr = key->u.ipv4.src;
467 fl->flowi4_tos = RT_TOS(key->tos);
468 fl->flowi4_mark = skb->mark;
469 fl->flowi4_proto = IPPROTO_GRE;
470
471 return ip_route_output_key(net, fl);
472}
473
474static struct rtable *prepare_fb_xmit(struct sk_buff *skb,
475 struct net_device *dev,
476 struct flowi4 *fl,
477 int tunnel_hlen)
478{
479 struct ip_tunnel_info *tun_info;
480 const struct ip_tunnel_key *key;
481 struct rtable *rt = NULL;
482 int min_headroom;
483 bool use_cache;
484 int err;
485
486 tun_info = skb_tunnel_info(skb);
487 key = &tun_info->key;
488 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
489
490 if (use_cache)
491 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl->saddr);
492 if (!rt) {
493 rt = gre_get_rt(skb, dev, fl, key);
494 if (IS_ERR(rt))
495 goto err_free_skb;
496 if (use_cache)
497 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
498 fl->saddr);
499 }
500
501 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
502 + tunnel_hlen + sizeof(struct iphdr);
503 if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
504 int head_delta = SKB_DATA_ALIGN(min_headroom -
505 skb_headroom(skb) +
506 16);
507 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
508 0, GFP_ATOMIC);
509 if (unlikely(err))
510 goto err_free_rt;
511 }
512 return rt;
513
514err_free_rt:
515 ip_rt_put(rt);
516err_free_skb:
517 kfree_skb(skb);
518 dev->stats.tx_dropped++;
519 return NULL;
520}
521
522static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
523 __be16 proto)
524{
525 struct ip_tunnel *tunnel = netdev_priv(dev);
526 struct ip_tunnel_info *tun_info;
527 const struct ip_tunnel_key *key;
528 struct rtable *rt = NULL;
529 struct flowi4 fl;
530 int tunnel_hlen;
531 __be16 df, flags;
532
533 tun_info = skb_tunnel_info(skb);
534 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
535 ip_tunnel_info_af(tun_info) != AF_INET))
536 goto err_free_skb;
537
538 key = &tun_info->key;
539 tunnel_hlen = gre_calc_hlen(key->tun_flags);
540
541 rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
542 if (!rt)
543 return;
544
545 /* Push Tunnel header. */
546 if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
547 goto err_free_rt;
548
549 flags = tun_info->key.tun_flags &
550 (TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ);
551 gre_build_header(skb, tunnel_hlen, flags, proto,
552 tunnel_id_to_key32(tun_info->key.tun_id),
553 (flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++) : 0);
554
555 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
556
557 iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
558 key->tos, key->ttl, df, false);
559 return;
560
561err_free_rt:
562 ip_rt_put(rt);
563err_free_skb:
564 kfree_skb(skb);
565 dev->stats.tx_dropped++;
566}
567
568static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev,
569 __be16 proto)
570{
571 struct ip_tunnel *tunnel = netdev_priv(dev);
572 struct ip_tunnel_info *tun_info;
573 const struct ip_tunnel_key *key;
574 struct erspan_metadata *md;
575 struct rtable *rt = NULL;
576 bool truncate = false;
577 struct flowi4 fl;
578 int tunnel_hlen;
579 int version;
580 __be16 df;
581
582 tun_info = skb_tunnel_info(skb);
583 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
584 ip_tunnel_info_af(tun_info) != AF_INET))
585 goto err_free_skb;
586
587 key = &tun_info->key;
588 md = ip_tunnel_info_opts(tun_info);
589 if (!md)
590 goto err_free_rt;
591
592 /* ERSPAN has fixed 8 byte GRE header */
593 version = md->version;
594 tunnel_hlen = 8 + erspan_hdr_len(version);
595
596 rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
597 if (!rt)
598 return;
599
600 if (gre_handle_offloads(skb, false))
601 goto err_free_rt;
602
603 if (skb->len > dev->mtu + dev->hard_header_len) {
604 pskb_trim(skb, dev->mtu + dev->hard_header_len);
605 truncate = true;
606 }
607
608 if (version == 1) {
609 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
610 ntohl(md->u.index), truncate, true);
611 } else if (version == 2) {
612 erspan_build_header_v2(skb,
613 ntohl(tunnel_id_to_key32(key->tun_id)),
614 md->u.md2.dir,
615 get_hwid(&md->u.md2),
616 truncate, true);
617 } else {
618 goto err_free_rt;
619 }
620
621 gre_build_header(skb, 8, TUNNEL_SEQ,
622 htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
623
624 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
625
626 iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
627 key->tos, key->ttl, df, false);
628 return;
629
630err_free_rt:
631 ip_rt_put(rt);
632err_free_skb:
633 kfree_skb(skb);
634 dev->stats.tx_dropped++;
635}
636
637static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
638{
639 struct ip_tunnel_info *info = skb_tunnel_info(skb);
640 struct rtable *rt;
641 struct flowi4 fl4;
642
643 if (ip_tunnel_info_af(info) != AF_INET)
644 return -EINVAL;
645
646 rt = gre_get_rt(skb, dev, &fl4, &info->key);
647 if (IS_ERR(rt))
648 return PTR_ERR(rt);
649
650 ip_rt_put(rt);
651 info->key.u.ipv4.src = fl4.saddr;
652 return 0;
653}
654
655static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
656 struct net_device *dev)
657{
658 struct ip_tunnel *tunnel = netdev_priv(dev);
659 const struct iphdr *tnl_params;
660
661 if (tunnel->collect_md) {
662 gre_fb_xmit(skb, dev, skb->protocol);
663 return NETDEV_TX_OK;
664 }
665
666 if (dev->header_ops) {
667 /* Need space for new headers */
668 if (skb_cow_head(skb, dev->needed_headroom -
669 (tunnel->hlen + sizeof(struct iphdr))))
670 goto free_skb;
671
672 tnl_params = (const struct iphdr *)skb->data;
673
674 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
675 * to gre header.
676 */
677 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
678 skb_reset_mac_header(skb);
679 } else {
680 if (skb_cow_head(skb, dev->needed_headroom))
681 goto free_skb;
682
683 tnl_params = &tunnel->parms.iph;
684 }
685
686 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
687 goto free_skb;
688
689 __gre_xmit(skb, dev, tnl_params, skb->protocol);
690 return NETDEV_TX_OK;
691
692free_skb:
693 kfree_skb(skb);
694 dev->stats.tx_dropped++;
695 return NETDEV_TX_OK;
696}
697
698static netdev_tx_t erspan_xmit(struct sk_buff *skb,
699 struct net_device *dev)
700{
701 struct ip_tunnel *tunnel = netdev_priv(dev);
702 bool truncate = false;
703
704 if (tunnel->collect_md) {
705 erspan_fb_xmit(skb, dev, skb->protocol);
706 return NETDEV_TX_OK;
707 }
708
709 if (gre_handle_offloads(skb, false))
710 goto free_skb;
711
712 if (skb_cow_head(skb, dev->needed_headroom))
713 goto free_skb;
714
715 if (skb->len > dev->mtu + dev->hard_header_len) {
716 pskb_trim(skb, dev->mtu + dev->hard_header_len);
717 truncate = true;
718 }
719
720 /* Push ERSPAN header */
721 if (tunnel->erspan_ver == 1)
722 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
723 tunnel->index,
724 truncate, true);
725 else if (tunnel->erspan_ver == 2)
726 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
727 tunnel->dir, tunnel->hwid,
728 truncate, true);
729 else
730 goto free_skb;
731
732 tunnel->parms.o_flags &= ~TUNNEL_KEY;
733 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
734 return NETDEV_TX_OK;
735
736free_skb:
737 kfree_skb(skb);
738 dev->stats.tx_dropped++;
739 return NETDEV_TX_OK;
740}
741
742static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
743 struct net_device *dev)
744{
745 struct ip_tunnel *tunnel = netdev_priv(dev);
746
747 if (tunnel->collect_md) {
748 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
749 return NETDEV_TX_OK;
750 }
751
752 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
753 goto free_skb;
754
755 if (skb_cow_head(skb, dev->needed_headroom))
756 goto free_skb;
757
758 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
759 return NETDEV_TX_OK;
760
761free_skb:
762 kfree_skb(skb);
763 dev->stats.tx_dropped++;
764 return NETDEV_TX_OK;
765}
766
767static void ipgre_link_update(struct net_device *dev, bool set_mtu)
768{
769 struct ip_tunnel *tunnel = netdev_priv(dev);
770 int len;
771
772 len = tunnel->tun_hlen;
773 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
774 len = tunnel->tun_hlen - len;
775 tunnel->hlen = tunnel->hlen + len;
776
777 dev->needed_headroom = dev->needed_headroom + len;
778 if (set_mtu)
779 dev->mtu = max_t(int, dev->mtu - len, 68);
780
781 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
782 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
783 tunnel->encap.type == TUNNEL_ENCAP_NONE) {
784 dev->features |= NETIF_F_GSO_SOFTWARE;
785 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
786 } else {
787 dev->features &= ~NETIF_F_GSO_SOFTWARE;
788 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
789 }
790 dev->features |= NETIF_F_LLTX;
791 } else {
792 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
793 dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
794 }
795}
796
797static int ipgre_tunnel_ioctl(struct net_device *dev,
798 struct ifreq *ifr, int cmd)
799{
800 struct ip_tunnel_parm p;
801 int err;
802
803 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
804 return -EFAULT;
805
806 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
807 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
808 p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
809 ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
810 return -EINVAL;
811 }
812
813 p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
814 p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
815
816 err = ip_tunnel_ioctl(dev, &p, cmd);
817 if (err)
818 return err;
819
820 if (cmd == SIOCCHGTUNNEL) {
821 struct ip_tunnel *t = netdev_priv(dev);
822
823 t->parms.i_flags = p.i_flags;
824 t->parms.o_flags = p.o_flags;
825
826 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
827 ipgre_link_update(dev, true);
828 }
829
830 p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
831 p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
832
833 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
834 return -EFAULT;
835
836 return 0;
837}
838
839/* Nice toy. Unfortunately, useless in real life :-)
840 It allows to construct virtual multiprotocol broadcast "LAN"
841 over the Internet, provided multicast routing is tuned.
842
843
844 I have no idea was this bicycle invented before me,
845 so that I had to set ARPHRD_IPGRE to a random value.
846 I have an impression, that Cisco could make something similar,
847 but this feature is apparently missing in IOS<=11.2(8).
848
849 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
850 with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
851
852 ping -t 255 224.66.66.66
853
854 If nobody answers, mbone does not work.
855
856 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
857 ip addr add 10.66.66.<somewhat>/24 dev Universe
858 ifconfig Universe up
859 ifconfig Universe add fe80::<Your_real_addr>/10
860 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
861 ftp 10.66.66.66
862 ...
863 ftp fec0:6666:6666::193.233.7.65
864 ...
865 */
866static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
867 unsigned short type,
868 const void *daddr, const void *saddr, unsigned int len)
869{
870 struct ip_tunnel *t = netdev_priv(dev);
871 struct iphdr *iph;
872 struct gre_base_hdr *greh;
873
874 iph = skb_push(skb, t->hlen + sizeof(*iph));
875 greh = (struct gre_base_hdr *)(iph+1);
876 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
877 greh->protocol = htons(type);
878
879 memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
880
881 /* Set the source hardware address. */
882 if (saddr)
883 memcpy(&iph->saddr, saddr, 4);
884 if (daddr)
885 memcpy(&iph->daddr, daddr, 4);
886 if (iph->daddr)
887 return t->hlen + sizeof(*iph);
888
889 return -(t->hlen + sizeof(*iph));
890}
891
892static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
893{
894 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
895 memcpy(haddr, &iph->saddr, 4);
896 return 4;
897}
898
899static const struct header_ops ipgre_header_ops = {
900 .create = ipgre_header,
901 .parse = ipgre_header_parse,
902};
903
904#ifdef CONFIG_NET_IPGRE_BROADCAST
905static int ipgre_open(struct net_device *dev)
906{
907 struct ip_tunnel *t = netdev_priv(dev);
908
909 if (ipv4_is_multicast(t->parms.iph.daddr)) {
910 struct flowi4 fl4;
911 struct rtable *rt;
912
913 rt = ip_route_output_gre(t->net, &fl4,
914 t->parms.iph.daddr,
915 t->parms.iph.saddr,
916 t->parms.o_key,
917 RT_TOS(t->parms.iph.tos),
918 t->parms.link);
919 if (IS_ERR(rt))
920 return -EADDRNOTAVAIL;
921 dev = rt->dst.dev;
922 ip_rt_put(rt);
923 if (!__in_dev_get_rtnl(dev))
924 return -EADDRNOTAVAIL;
925 t->mlink = dev->ifindex;
926 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
927 }
928 return 0;
929}
930
931static int ipgre_close(struct net_device *dev)
932{
933 struct ip_tunnel *t = netdev_priv(dev);
934
935 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
936 struct in_device *in_dev;
937 in_dev = inetdev_by_index(t->net, t->mlink);
938 if (in_dev)
939 ip_mc_dec_group(in_dev, t->parms.iph.daddr);
940 }
941 return 0;
942}
943#endif
944
945static const struct net_device_ops ipgre_netdev_ops = {
946 .ndo_init = ipgre_tunnel_init,
947 .ndo_uninit = ip_tunnel_uninit,
948#ifdef CONFIG_NET_IPGRE_BROADCAST
949 .ndo_open = ipgre_open,
950 .ndo_stop = ipgre_close,
951#endif
952 .ndo_start_xmit = ipgre_xmit,
953 .ndo_do_ioctl = ipgre_tunnel_ioctl,
954 .ndo_change_mtu = ip_tunnel_change_mtu,
955 .ndo_get_stats64 = ip_tunnel_get_stats64,
956 .ndo_get_iflink = ip_tunnel_get_iflink,
957};
958
959#define GRE_FEATURES (NETIF_F_SG | \
960 NETIF_F_FRAGLIST | \
961 NETIF_F_HIGHDMA | \
962 NETIF_F_HW_CSUM)
963
964static void ipgre_tunnel_setup(struct net_device *dev)
965{
966 dev->netdev_ops = &ipgre_netdev_ops;
967 dev->type = ARPHRD_IPGRE;
968 ip_tunnel_setup(dev, ipgre_net_id);
969}
970
971static void __gre_tunnel_init(struct net_device *dev)
972{
973 struct ip_tunnel *tunnel;
974 int t_hlen;
975
976 tunnel = netdev_priv(dev);
977 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
978 tunnel->parms.iph.protocol = IPPROTO_GRE;
979
980 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
981
982 t_hlen = tunnel->hlen + sizeof(struct iphdr);
983
984 dev->features |= GRE_FEATURES;
985 dev->hw_features |= GRE_FEATURES;
986
987 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
988 /* TCP offload with GRE SEQ is not supported, nor
989 * can we support 2 levels of outer headers requiring
990 * an update.
991 */
992 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
993 (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
994 dev->features |= NETIF_F_GSO_SOFTWARE;
995 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
996 }
997
998 /* Can use a lockless transmit, unless we generate
999 * output sequences
1000 */
1001 dev->features |= NETIF_F_LLTX;
1002 }
1003}
1004
1005static int ipgre_tunnel_init(struct net_device *dev)
1006{
1007 struct ip_tunnel *tunnel = netdev_priv(dev);
1008 struct iphdr *iph = &tunnel->parms.iph;
1009
1010 __gre_tunnel_init(dev);
1011
1012 memcpy(dev->dev_addr, &iph->saddr, 4);
1013 memcpy(dev->broadcast, &iph->daddr, 4);
1014
1015 dev->flags = IFF_NOARP;
1016 netif_keep_dst(dev);
1017 dev->addr_len = 4;
1018
1019 if (iph->daddr && !tunnel->collect_md) {
1020#ifdef CONFIG_NET_IPGRE_BROADCAST
1021 if (ipv4_is_multicast(iph->daddr)) {
1022 if (!iph->saddr)
1023 return -EINVAL;
1024 dev->flags = IFF_BROADCAST;
1025 dev->header_ops = &ipgre_header_ops;
1026 }
1027#endif
1028 } else if (!tunnel->collect_md) {
1029 dev->header_ops = &ipgre_header_ops;
1030 }
1031
1032 return ip_tunnel_init(dev);
1033}
1034
1035static const struct gre_protocol ipgre_protocol = {
1036 .handler = gre_rcv,
1037 .err_handler = gre_err,
1038};
1039
1040static int __net_init ipgre_init_net(struct net *net)
1041{
1042 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1043}
1044
1045static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1046{
1047 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1048}
1049
1050static struct pernet_operations ipgre_net_ops = {
1051 .init = ipgre_init_net,
1052 .exit_batch = ipgre_exit_batch_net,
1053 .id = &ipgre_net_id,
1054 .size = sizeof(struct ip_tunnel_net),
1055};
1056
1057static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1058 struct netlink_ext_ack *extack)
1059{
1060 __be16 flags;
1061
1062 if (!data)
1063 return 0;
1064
1065 flags = 0;
1066 if (data[IFLA_GRE_IFLAGS])
1067 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1068 if (data[IFLA_GRE_OFLAGS])
1069 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1070 if (flags & (GRE_VERSION|GRE_ROUTING))
1071 return -EINVAL;
1072
1073 if (data[IFLA_GRE_COLLECT_METADATA] &&
1074 data[IFLA_GRE_ENCAP_TYPE] &&
1075 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1076 return -EINVAL;
1077
1078 return 0;
1079}
1080
1081static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1082 struct netlink_ext_ack *extack)
1083{
1084 __be32 daddr;
1085
1086 if (tb[IFLA_ADDRESS]) {
1087 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1088 return -EINVAL;
1089 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1090 return -EADDRNOTAVAIL;
1091 }
1092
1093 if (!data)
1094 goto out;
1095
1096 if (data[IFLA_GRE_REMOTE]) {
1097 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1098 if (!daddr)
1099 return -EINVAL;
1100 }
1101
1102out:
1103 return ipgre_tunnel_validate(tb, data, extack);
1104}
1105
1106static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1107 struct netlink_ext_ack *extack)
1108{
1109 __be16 flags = 0;
1110 int ret;
1111
1112 if (!data)
1113 return 0;
1114
1115 ret = ipgre_tap_validate(tb, data, extack);
1116 if (ret)
1117 return ret;
1118
1119 /* ERSPAN should only have GRE sequence and key flag */
1120 if (data[IFLA_GRE_OFLAGS])
1121 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1122 if (data[IFLA_GRE_IFLAGS])
1123 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1124 if (!data[IFLA_GRE_COLLECT_METADATA] &&
1125 flags != (GRE_SEQ | GRE_KEY))
1126 return -EINVAL;
1127
1128 /* ERSPAN Session ID only has 10-bit. Since we reuse
1129 * 32-bit key field as ID, check it's range.
1130 */
1131 if (data[IFLA_GRE_IKEY] &&
1132 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1133 return -EINVAL;
1134
1135 if (data[IFLA_GRE_OKEY] &&
1136 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1137 return -EINVAL;
1138
1139 return 0;
1140}
1141
1142static int ipgre_netlink_parms(struct net_device *dev,
1143 struct nlattr *data[],
1144 struct nlattr *tb[],
1145 struct ip_tunnel_parm *parms,
1146 __u32 *fwmark)
1147{
1148 struct ip_tunnel *t = netdev_priv(dev);
1149
1150 memset(parms, 0, sizeof(*parms));
1151
1152 parms->iph.protocol = IPPROTO_GRE;
1153
1154 if (!data)
1155 return 0;
1156
1157 if (data[IFLA_GRE_LINK])
1158 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1159
1160 if (data[IFLA_GRE_IFLAGS])
1161 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1162
1163 if (data[IFLA_GRE_OFLAGS])
1164 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1165
1166 if (data[IFLA_GRE_IKEY])
1167 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1168
1169 if (data[IFLA_GRE_OKEY])
1170 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1171
1172 if (data[IFLA_GRE_LOCAL])
1173 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1174
1175 if (data[IFLA_GRE_REMOTE])
1176 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1177
1178 if (data[IFLA_GRE_TTL])
1179 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1180
1181 if (data[IFLA_GRE_TOS])
1182 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1183
1184 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1185 if (t->ignore_df)
1186 return -EINVAL;
1187 parms->iph.frag_off = htons(IP_DF);
1188 }
1189
1190 if (data[IFLA_GRE_COLLECT_METADATA]) {
1191 t->collect_md = true;
1192 if (dev->type == ARPHRD_IPGRE)
1193 dev->type = ARPHRD_NONE;
1194 }
1195
1196 if (data[IFLA_GRE_IGNORE_DF]) {
1197 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1198 && (parms->iph.frag_off & htons(IP_DF)))
1199 return -EINVAL;
1200 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1201 }
1202
1203 if (data[IFLA_GRE_FWMARK])
1204 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1205
1206 if (data[IFLA_GRE_ERSPAN_VER]) {
1207 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1208
1209 if (t->erspan_ver != 1 && t->erspan_ver != 2)
1210 return -EINVAL;
1211 }
1212
1213 if (t->erspan_ver == 1) {
1214 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1215 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1216 if (t->index & ~INDEX_MASK)
1217 return -EINVAL;
1218 }
1219 } else if (t->erspan_ver == 2) {
1220 if (data[IFLA_GRE_ERSPAN_DIR]) {
1221 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1222 if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1223 return -EINVAL;
1224 }
1225 if (data[IFLA_GRE_ERSPAN_HWID]) {
1226 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1227 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1228 return -EINVAL;
1229 }
1230 }
1231
1232 return 0;
1233}
1234
1235/* This function returns true when ENCAP attributes are present in the nl msg */
1236static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1237 struct ip_tunnel_encap *ipencap)
1238{
1239 bool ret = false;
1240
1241 memset(ipencap, 0, sizeof(*ipencap));
1242
1243 if (!data)
1244 return ret;
1245
1246 if (data[IFLA_GRE_ENCAP_TYPE]) {
1247 ret = true;
1248 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1249 }
1250
1251 if (data[IFLA_GRE_ENCAP_FLAGS]) {
1252 ret = true;
1253 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1254 }
1255
1256 if (data[IFLA_GRE_ENCAP_SPORT]) {
1257 ret = true;
1258 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1259 }
1260
1261 if (data[IFLA_GRE_ENCAP_DPORT]) {
1262 ret = true;
1263 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1264 }
1265
1266 return ret;
1267}
1268
1269static int gre_tap_init(struct net_device *dev)
1270{
1271 __gre_tunnel_init(dev);
1272 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1273 netif_keep_dst(dev);
1274
1275 return ip_tunnel_init(dev);
1276}
1277
1278static const struct net_device_ops gre_tap_netdev_ops = {
1279 .ndo_init = gre_tap_init,
1280 .ndo_uninit = ip_tunnel_uninit,
1281 .ndo_start_xmit = gre_tap_xmit,
1282 .ndo_set_mac_address = eth_mac_addr,
1283 .ndo_validate_addr = eth_validate_addr,
1284 .ndo_change_mtu = ip_tunnel_change_mtu,
1285 .ndo_get_stats64 = ip_tunnel_get_stats64,
1286 .ndo_get_iflink = ip_tunnel_get_iflink,
1287 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1288};
1289
1290static int erspan_tunnel_init(struct net_device *dev)
1291{
1292 struct ip_tunnel *tunnel = netdev_priv(dev);
1293 int t_hlen;
1294
1295 tunnel->tun_hlen = 8;
1296 tunnel->parms.iph.protocol = IPPROTO_GRE;
1297 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1298 erspan_hdr_len(tunnel->erspan_ver);
1299 t_hlen = tunnel->hlen + sizeof(struct iphdr);
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
1309static 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 = ip_tunnel_get_stats64,
1317 .ndo_get_iflink = ip_tunnel_get_iflink,
1318 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1319};
1320
1321static 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
1331bool is_gretap_dev(const struct net_device *dev)
1332{
1333 return dev->netdev_ops == &gre_tap_netdev_ops;
1334}
1335EXPORT_SYMBOL_GPL(is_gretap_dev);
1336
1337static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1338 struct nlattr *tb[], struct nlattr *data[],
1339 struct netlink_ext_ack *extack)
1340{
1341 struct ip_tunnel_parm p;
1342 struct ip_tunnel_encap ipencap;
1343 __u32 fwmark = 0;
1344 int err;
1345
1346 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1347 struct ip_tunnel *t = netdev_priv(dev);
1348 err = ip_tunnel_encap_setup(t, &ipencap);
1349
1350 if (err < 0)
1351 return err;
1352 }
1353
1354 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1355 if (err < 0)
1356 return err;
1357 return ip_tunnel_newlink(dev, tb, &p, fwmark);
1358}
1359
1360static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1361 struct nlattr *data[],
1362 struct netlink_ext_ack *extack)
1363{
1364 struct ip_tunnel *t = netdev_priv(dev);
1365 struct ip_tunnel_encap ipencap;
1366 __u32 fwmark = t->fwmark;
1367 struct ip_tunnel_parm p;
1368 int err;
1369
1370 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1371 err = ip_tunnel_encap_setup(t, &ipencap);
1372
1373 if (err < 0)
1374 return err;
1375 }
1376
1377 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1378 if (err < 0)
1379 return err;
1380
1381 err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1382 if (err < 0)
1383 return err;
1384
1385 t->parms.i_flags = p.i_flags;
1386 t->parms.o_flags = p.o_flags;
1387
1388 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1389 ipgre_link_update(dev, !tb[IFLA_MTU]);
1390
1391 return 0;
1392}
1393
1394static size_t ipgre_get_size(const struct net_device *dev)
1395{
1396 return
1397 /* IFLA_GRE_LINK */
1398 nla_total_size(4) +
1399 /* IFLA_GRE_IFLAGS */
1400 nla_total_size(2) +
1401 /* IFLA_GRE_OFLAGS */
1402 nla_total_size(2) +
1403 /* IFLA_GRE_IKEY */
1404 nla_total_size(4) +
1405 /* IFLA_GRE_OKEY */
1406 nla_total_size(4) +
1407 /* IFLA_GRE_LOCAL */
1408 nla_total_size(4) +
1409 /* IFLA_GRE_REMOTE */
1410 nla_total_size(4) +
1411 /* IFLA_GRE_TTL */
1412 nla_total_size(1) +
1413 /* IFLA_GRE_TOS */
1414 nla_total_size(1) +
1415 /* IFLA_GRE_PMTUDISC */
1416 nla_total_size(1) +
1417 /* IFLA_GRE_ENCAP_TYPE */
1418 nla_total_size(2) +
1419 /* IFLA_GRE_ENCAP_FLAGS */
1420 nla_total_size(2) +
1421 /* IFLA_GRE_ENCAP_SPORT */
1422 nla_total_size(2) +
1423 /* IFLA_GRE_ENCAP_DPORT */
1424 nla_total_size(2) +
1425 /* IFLA_GRE_COLLECT_METADATA */
1426 nla_total_size(0) +
1427 /* IFLA_GRE_IGNORE_DF */
1428 nla_total_size(1) +
1429 /* IFLA_GRE_FWMARK */
1430 nla_total_size(4) +
1431 /* IFLA_GRE_ERSPAN_INDEX */
1432 nla_total_size(4) +
1433 /* IFLA_GRE_ERSPAN_VER */
1434 nla_total_size(1) +
1435 /* IFLA_GRE_ERSPAN_DIR */
1436 nla_total_size(1) +
1437 /* IFLA_GRE_ERSPAN_HWID */
1438 nla_total_size(2) +
1439 0;
1440}
1441
1442static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1443{
1444 struct ip_tunnel *t = netdev_priv(dev);
1445 struct ip_tunnel_parm *p = &t->parms;
1446
1447 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1448 nla_put_be16(skb, IFLA_GRE_IFLAGS,
1449 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1450 nla_put_be16(skb, IFLA_GRE_OFLAGS,
1451 gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1452 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1453 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1454 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1455 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1456 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1457 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1458 nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1459 !!(p->iph.frag_off & htons(IP_DF))) ||
1460 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1461 goto nla_put_failure;
1462
1463 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1464 t->encap.type) ||
1465 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1466 t->encap.sport) ||
1467 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1468 t->encap.dport) ||
1469 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1470 t->encap.flags))
1471 goto nla_put_failure;
1472
1473 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1474 goto nla_put_failure;
1475
1476 if (t->collect_md) {
1477 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1478 goto nla_put_failure;
1479 }
1480
1481 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1482 goto nla_put_failure;
1483
1484 if (t->erspan_ver == 1) {
1485 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1486 goto nla_put_failure;
1487 } else if (t->erspan_ver == 2) {
1488 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1489 goto nla_put_failure;
1490 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1491 goto nla_put_failure;
1492 }
1493
1494 return 0;
1495
1496nla_put_failure:
1497 return -EMSGSIZE;
1498}
1499
1500static void erspan_setup(struct net_device *dev)
1501{
1502 ether_setup(dev);
1503 dev->netdev_ops = &erspan_netdev_ops;
1504 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1505 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1506 ip_tunnel_setup(dev, erspan_net_id);
1507}
1508
1509static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1510 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1511 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1512 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1513 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1514 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1515 [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1516 [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1517 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1518 [IFLA_GRE_TOS] = { .type = NLA_U8 },
1519 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 },
1520 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 },
1521 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 },
1522 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 },
1523 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 },
1524 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG },
1525 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 },
1526 [IFLA_GRE_FWMARK] = { .type = NLA_U32 },
1527 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1528 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 },
1529 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 },
1530 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 },
1531};
1532
1533static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1534 .kind = "gre",
1535 .maxtype = IFLA_GRE_MAX,
1536 .policy = ipgre_policy,
1537 .priv_size = sizeof(struct ip_tunnel),
1538 .setup = ipgre_tunnel_setup,
1539 .validate = ipgre_tunnel_validate,
1540 .newlink = ipgre_newlink,
1541 .changelink = ipgre_changelink,
1542 .dellink = ip_tunnel_dellink,
1543 .get_size = ipgre_get_size,
1544 .fill_info = ipgre_fill_info,
1545 .get_link_net = ip_tunnel_get_link_net,
1546};
1547
1548static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1549 .kind = "gretap",
1550 .maxtype = IFLA_GRE_MAX,
1551 .policy = ipgre_policy,
1552 .priv_size = sizeof(struct ip_tunnel),
1553 .setup = ipgre_tap_setup,
1554 .validate = ipgre_tap_validate,
1555 .newlink = ipgre_newlink,
1556 .changelink = ipgre_changelink,
1557 .dellink = ip_tunnel_dellink,
1558 .get_size = ipgre_get_size,
1559 .fill_info = ipgre_fill_info,
1560 .get_link_net = ip_tunnel_get_link_net,
1561};
1562
1563static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1564 .kind = "erspan",
1565 .maxtype = IFLA_GRE_MAX,
1566 .policy = ipgre_policy,
1567 .priv_size = sizeof(struct ip_tunnel),
1568 .setup = erspan_setup,
1569 .validate = erspan_validate,
1570 .newlink = ipgre_newlink,
1571 .changelink = ipgre_changelink,
1572 .dellink = ip_tunnel_dellink,
1573 .get_size = ipgre_get_size,
1574 .fill_info = ipgre_fill_info,
1575 .get_link_net = ip_tunnel_get_link_net,
1576};
1577
1578struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1579 u8 name_assign_type)
1580{
1581 struct nlattr *tb[IFLA_MAX + 1];
1582 struct net_device *dev;
1583 LIST_HEAD(list_kill);
1584 struct ip_tunnel *t;
1585 int err;
1586
1587 memset(&tb, 0, sizeof(tb));
1588
1589 dev = rtnl_create_link(net, name, name_assign_type,
1590 &ipgre_tap_ops, tb);
1591 if (IS_ERR(dev))
1592 return dev;
1593
1594 /* Configure flow based GRE device. */
1595 t = netdev_priv(dev);
1596 t->collect_md = true;
1597
1598 err = ipgre_newlink(net, dev, tb, NULL, NULL);
1599 if (err < 0) {
1600 free_netdev(dev);
1601 return ERR_PTR(err);
1602 }
1603
1604 /* openvswitch users expect packet sizes to be unrestricted,
1605 * so set the largest MTU we can.
1606 */
1607 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1608 if (err)
1609 goto out;
1610
1611 err = rtnl_configure_link(dev, NULL);
1612 if (err < 0)
1613 goto out;
1614
1615 return dev;
1616out:
1617 ip_tunnel_dellink(dev, &list_kill);
1618 unregister_netdevice_many(&list_kill);
1619 return ERR_PTR(err);
1620}
1621EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1622
1623static int __net_init ipgre_tap_init_net(struct net *net)
1624{
1625 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1626}
1627
1628static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1629{
1630 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1631}
1632
1633static struct pernet_operations ipgre_tap_net_ops = {
1634 .init = ipgre_tap_init_net,
1635 .exit_batch = ipgre_tap_exit_batch_net,
1636 .id = &gre_tap_net_id,
1637 .size = sizeof(struct ip_tunnel_net),
1638};
1639
1640static int __net_init erspan_init_net(struct net *net)
1641{
1642 return ip_tunnel_init_net(net, erspan_net_id,
1643 &erspan_link_ops, "erspan0");
1644}
1645
1646static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1647{
1648 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1649}
1650
1651static struct pernet_operations erspan_net_ops = {
1652 .init = erspan_init_net,
1653 .exit_batch = erspan_exit_batch_net,
1654 .id = &erspan_net_id,
1655 .size = sizeof(struct ip_tunnel_net),
1656};
1657
1658static int __init ipgre_init(void)
1659{
1660 int err;
1661
1662 pr_info("GRE over IPv4 tunneling driver\n");
1663
1664 err = register_pernet_device(&ipgre_net_ops);
1665 if (err < 0)
1666 return err;
1667
1668 err = register_pernet_device(&ipgre_tap_net_ops);
1669 if (err < 0)
1670 goto pnet_tap_failed;
1671
1672 err = register_pernet_device(&erspan_net_ops);
1673 if (err < 0)
1674 goto pnet_erspan_failed;
1675
1676 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1677 if (err < 0) {
1678 pr_info("%s: can't add protocol\n", __func__);
1679 goto add_proto_failed;
1680 }
1681
1682 err = rtnl_link_register(&ipgre_link_ops);
1683 if (err < 0)
1684 goto rtnl_link_failed;
1685
1686 err = rtnl_link_register(&ipgre_tap_ops);
1687 if (err < 0)
1688 goto tap_ops_failed;
1689
1690 err = rtnl_link_register(&erspan_link_ops);
1691 if (err < 0)
1692 goto erspan_link_failed;
1693
1694 return 0;
1695
1696erspan_link_failed:
1697 rtnl_link_unregister(&ipgre_tap_ops);
1698tap_ops_failed:
1699 rtnl_link_unregister(&ipgre_link_ops);
1700rtnl_link_failed:
1701 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1702add_proto_failed:
1703 unregister_pernet_device(&erspan_net_ops);
1704pnet_erspan_failed:
1705 unregister_pernet_device(&ipgre_tap_net_ops);
1706pnet_tap_failed:
1707 unregister_pernet_device(&ipgre_net_ops);
1708 return err;
1709}
1710
1711static void __exit ipgre_fini(void)
1712{
1713 rtnl_link_unregister(&ipgre_tap_ops);
1714 rtnl_link_unregister(&ipgre_link_ops);
1715 rtnl_link_unregister(&erspan_link_ops);
1716 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1717 unregister_pernet_device(&ipgre_tap_net_ops);
1718 unregister_pernet_device(&ipgre_net_ops);
1719 unregister_pernet_device(&erspan_net_ops);
1720}
1721
1722module_init(ipgre_init);
1723module_exit(ipgre_fini);
1724MODULE_LICENSE("GPL");
1725MODULE_ALIAS_RTNL_LINK("gre");
1726MODULE_ALIAS_RTNL_LINK("gretap");
1727MODULE_ALIAS_RTNL_LINK("erspan");
1728MODULE_ALIAS_NETDEV("gre0");
1729MODULE_ALIAS_NETDEV("gretap0");
1730MODULE_ALIAS_NETDEV("erspan0");
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#include <net/inet_dscp.h>
48
49/*
50 Problems & solutions
51 --------------------
52
53 1. The most important issue is detecting local dead loops.
54 They would cause complete host lockup in transmit, which
55 would be "resolved" by stack overflow or, if queueing is enabled,
56 with infinite looping in net_bh.
57
58 We cannot track such dead loops during route installation,
59 it is infeasible task. The most general solutions would be
60 to keep skb->encapsulation counter (sort of local ttl),
61 and silently drop packet when it expires. It is a good
62 solution, but it supposes maintaining new variable in ALL
63 skb, even if no tunneling is used.
64
65 Current solution: xmit_recursion breaks dead loops. This is a percpu
66 counter, since when we enter the first ndo_xmit(), cpu migration is
67 forbidden. We force an exit if this counter reaches RECURSION_LIMIT
68
69 2. Networking dead loops would not kill routers, but would really
70 kill network. IP hop limit plays role of "t->recursion" in this case,
71 if we copy it from packet being encapsulated to upper header.
72 It is very good solution, but it introduces two problems:
73
74 - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
75 do not work over tunnels.
76 - traceroute does not work. I planned to relay ICMP from tunnel,
77 so that this problem would be solved and traceroute output
78 would even more informative. This idea appeared to be wrong:
79 only Linux complies to rfc1812 now (yes, guys, Linux is the only
80 true router now :-)), all routers (at least, in neighbourhood of mine)
81 return only 8 bytes of payload. It is the end.
82
83 Hence, if we want that OSPF worked or traceroute said something reasonable,
84 we should search for another solution.
85
86 One of them is to parse packet trying to detect inner encapsulation
87 made by our node. It is difficult or even impossible, especially,
88 taking into account fragmentation. TO be short, ttl is not solution at all.
89
90 Current solution: The solution was UNEXPECTEDLY SIMPLE.
91 We force DF flag on tunnels with preconfigured hop limit,
92 that is ALL. :-) Well, it does not remove the problem completely,
93 but exponential growth of network traffic is changed to linear
94 (branches, that exceed pmtu are pruned) and tunnel mtu
95 rapidly degrades to value <68, where looping stops.
96 Yes, it is not good if there exists a router in the loop,
97 which does not force DF, even when encapsulating packets have DF set.
98 But it is not our problem! Nobody could accuse us, we made
99 all that we could make. Even if it is your gated who injected
100 fatal route to network, even if it were you who configured
101 fatal static route: you are innocent. :-)
102
103 Alexey Kuznetsov.
104 */
105
106static bool log_ecn_error = true;
107module_param(log_ecn_error, bool, 0644);
108MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
109
110static struct rtnl_link_ops ipgre_link_ops __read_mostly;
111static const struct header_ops ipgre_header_ops;
112
113static int ipgre_tunnel_init(struct net_device *dev);
114static void erspan_build_header(struct sk_buff *skb,
115 u32 id, u32 index,
116 bool truncate, bool is_ipv4);
117
118static unsigned int ipgre_net_id __read_mostly;
119static unsigned int gre_tap_net_id __read_mostly;
120static unsigned int erspan_net_id __read_mostly;
121
122static int ipgre_err(struct sk_buff *skb, u32 info,
123 const struct tnl_ptk_info *tpi)
124{
125
126 /* All the routers (except for Linux) return only
127 8 bytes of packet payload. It means, that precise relaying of
128 ICMP in the real Internet is absolutely infeasible.
129
130 Moreover, Cisco "wise men" put GRE key to the third word
131 in GRE header. It makes impossible maintaining even soft
132 state for keyed GRE tunnels with enabled checksum. Tell
133 them "thank you".
134
135 Well, I wonder, rfc1812 was written by Cisco employee,
136 what the hell these idiots break standards established
137 by themselves???
138 */
139 struct net *net = dev_net(skb->dev);
140 struct ip_tunnel_net *itn;
141 const struct iphdr *iph;
142 const int type = icmp_hdr(skb)->type;
143 const int code = icmp_hdr(skb)->code;
144 unsigned int data_len = 0;
145 struct ip_tunnel *t;
146
147 if (tpi->proto == htons(ETH_P_TEB))
148 itn = net_generic(net, gre_tap_net_id);
149 else if (tpi->proto == htons(ETH_P_ERSPAN) ||
150 tpi->proto == htons(ETH_P_ERSPAN2))
151 itn = net_generic(net, erspan_net_id);
152 else
153 itn = net_generic(net, ipgre_net_id);
154
155 iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
156 t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
157 iph->daddr, iph->saddr, tpi->key);
158
159 if (!t)
160 return -ENOENT;
161
162 switch (type) {
163 default:
164 case ICMP_PARAMETERPROB:
165 return 0;
166
167 case ICMP_DEST_UNREACH:
168 switch (code) {
169 case ICMP_SR_FAILED:
170 case ICMP_PORT_UNREACH:
171 /* Impossible event. */
172 return 0;
173 default:
174 /* All others are translated to HOST_UNREACH.
175 rfc2003 contains "deep thoughts" about NET_UNREACH,
176 I believe they are just ether pollution. --ANK
177 */
178 break;
179 }
180 break;
181
182 case ICMP_TIME_EXCEEDED:
183 if (code != ICMP_EXC_TTL)
184 return 0;
185 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
186 break;
187
188 case ICMP_REDIRECT:
189 break;
190 }
191
192#if IS_ENABLED(CONFIG_IPV6)
193 if (tpi->proto == htons(ETH_P_IPV6) &&
194 !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
195 type, data_len))
196 return 0;
197#endif
198
199 if (t->parms.iph.daddr == 0 ||
200 ipv4_is_multicast(t->parms.iph.daddr))
201 return 0;
202
203 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
204 return 0;
205
206 if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
207 t->err_count++;
208 else
209 t->err_count = 1;
210 t->err_time = jiffies;
211
212 return 0;
213}
214
215static void gre_err(struct sk_buff *skb, u32 info)
216{
217 /* All the routers (except for Linux) return only
218 * 8 bytes of packet payload. It means, that precise relaying of
219 * ICMP in the real Internet is absolutely infeasible.
220 *
221 * Moreover, Cisco "wise men" put GRE key to the third word
222 * in GRE header. It makes impossible maintaining even soft
223 * state for keyed
224 * GRE tunnels with enabled checksum. Tell them "thank you".
225 *
226 * Well, I wonder, rfc1812 was written by Cisco employee,
227 * what the hell these idiots break standards established
228 * by themselves???
229 */
230
231 const struct iphdr *iph = (struct iphdr *)skb->data;
232 const int type = icmp_hdr(skb)->type;
233 const int code = icmp_hdr(skb)->code;
234 struct tnl_ptk_info tpi;
235
236 if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP),
237 iph->ihl * 4) < 0)
238 return;
239
240 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
241 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
242 skb->dev->ifindex, IPPROTO_GRE);
243 return;
244 }
245 if (type == ICMP_REDIRECT) {
246 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex,
247 IPPROTO_GRE);
248 return;
249 }
250
251 ipgre_err(skb, info, &tpi);
252}
253
254static bool is_erspan_type1(int gre_hdr_len)
255{
256 /* Both ERSPAN type I (version 0) and type II (version 1) use
257 * protocol 0x88BE, but the type I has only 4-byte GRE header,
258 * while type II has 8-byte.
259 */
260 return gre_hdr_len == 4;
261}
262
263static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
264 int gre_hdr_len)
265{
266 struct net *net = dev_net(skb->dev);
267 struct metadata_dst *tun_dst = NULL;
268 struct erspan_base_hdr *ershdr;
269 IP_TUNNEL_DECLARE_FLAGS(flags);
270 struct ip_tunnel_net *itn;
271 struct ip_tunnel *tunnel;
272 const struct iphdr *iph;
273 struct erspan_md2 *md2;
274 int ver;
275 int len;
276
277 ip_tunnel_flags_copy(flags, tpi->flags);
278
279 itn = net_generic(net, erspan_net_id);
280 iph = ip_hdr(skb);
281 if (is_erspan_type1(gre_hdr_len)) {
282 ver = 0;
283 __set_bit(IP_TUNNEL_NO_KEY_BIT, flags);
284 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, flags,
285 iph->saddr, iph->daddr, 0);
286 } else {
287 if (unlikely(!pskb_may_pull(skb,
288 gre_hdr_len + sizeof(*ershdr))))
289 return PACKET_REJECT;
290
291 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
292 ver = ershdr->ver;
293 iph = ip_hdr(skb);
294 __set_bit(IP_TUNNEL_KEY_BIT, flags);
295 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, flags,
296 iph->saddr, iph->daddr, tpi->key);
297 }
298
299 if (tunnel) {
300 if (is_erspan_type1(gre_hdr_len))
301 len = gre_hdr_len;
302 else
303 len = gre_hdr_len + erspan_hdr_len(ver);
304
305 if (unlikely(!pskb_may_pull(skb, len)))
306 return PACKET_REJECT;
307
308 if (__iptunnel_pull_header(skb,
309 len,
310 htons(ETH_P_TEB),
311 false, false) < 0)
312 goto drop;
313
314 if (tunnel->collect_md) {
315 struct erspan_metadata *pkt_md, *md;
316 struct ip_tunnel_info *info;
317 unsigned char *gh;
318 __be64 tun_id;
319
320 __set_bit(IP_TUNNEL_KEY_BIT, tpi->flags);
321 ip_tunnel_flags_copy(flags, tpi->flags);
322 tun_id = key32_to_tunnel_id(tpi->key);
323
324 tun_dst = ip_tun_rx_dst(skb, flags,
325 tun_id, sizeof(*md));
326 if (!tun_dst)
327 return PACKET_REJECT;
328
329 /* skb can be uncloned in __iptunnel_pull_header, so
330 * old pkt_md is no longer valid and we need to reset
331 * it
332 */
333 gh = skb_network_header(skb) +
334 skb_network_header_len(skb);
335 pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len +
336 sizeof(*ershdr));
337 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
338 md->version = ver;
339 md2 = &md->u.md2;
340 memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
341 ERSPAN_V2_MDSIZE);
342
343 info = &tun_dst->u.tun_info;
344 __set_bit(IP_TUNNEL_ERSPAN_OPT_BIT,
345 info->key.tun_flags);
346 info->options_len = sizeof(*md);
347 }
348
349 skb_reset_mac_header(skb);
350 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
351 return PACKET_RCVD;
352 }
353 return PACKET_REJECT;
354
355drop:
356 kfree_skb(skb);
357 return PACKET_RCVD;
358}
359
360static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
361 struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
362{
363 struct metadata_dst *tun_dst = NULL;
364 const struct iphdr *iph;
365 struct ip_tunnel *tunnel;
366
367 iph = ip_hdr(skb);
368 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
369 iph->saddr, iph->daddr, tpi->key);
370
371 if (tunnel) {
372 const struct iphdr *tnl_params;
373
374 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
375 raw_proto, false) < 0)
376 goto drop;
377
378 /* Special case for ipgre_header_parse(), which expects the
379 * mac_header to point to the outer IP header.
380 */
381 if (tunnel->dev->header_ops == &ipgre_header_ops)
382 skb_pop_mac_header(skb);
383 else
384 skb_reset_mac_header(skb);
385
386 tnl_params = &tunnel->parms.iph;
387 if (tunnel->collect_md || tnl_params->daddr == 0) {
388 IP_TUNNEL_DECLARE_FLAGS(flags) = { };
389 __be64 tun_id;
390
391 __set_bit(IP_TUNNEL_CSUM_BIT, flags);
392 __set_bit(IP_TUNNEL_KEY_BIT, flags);
393 ip_tunnel_flags_and(flags, tpi->flags, flags);
394
395 tun_id = key32_to_tunnel_id(tpi->key);
396 tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
397 if (!tun_dst)
398 return PACKET_REJECT;
399 }
400
401 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
402 return PACKET_RCVD;
403 }
404 return PACKET_NEXT;
405
406drop:
407 kfree_skb(skb);
408 return PACKET_RCVD;
409}
410
411static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
412 int hdr_len)
413{
414 struct net *net = dev_net(skb->dev);
415 struct ip_tunnel_net *itn;
416 int res;
417
418 if (tpi->proto == htons(ETH_P_TEB))
419 itn = net_generic(net, gre_tap_net_id);
420 else
421 itn = net_generic(net, ipgre_net_id);
422
423 res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
424 if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
425 /* ipgre tunnels in collect metadata mode should receive
426 * also ETH_P_TEB traffic.
427 */
428 itn = net_generic(net, ipgre_net_id);
429 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
430 }
431 return res;
432}
433
434static int gre_rcv(struct sk_buff *skb)
435{
436 struct tnl_ptk_info tpi;
437 bool csum_err = false;
438 int hdr_len;
439
440#ifdef CONFIG_NET_IPGRE_BROADCAST
441 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
442 /* Looped back packet, drop it! */
443 if (rt_is_output_route(skb_rtable(skb)))
444 goto drop;
445 }
446#endif
447
448 hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
449 if (hdr_len < 0)
450 goto drop;
451
452 if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
453 tpi.proto == htons(ETH_P_ERSPAN2))) {
454 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
455 return 0;
456 goto out;
457 }
458
459 if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
460 return 0;
461
462out:
463 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
464drop:
465 kfree_skb(skb);
466 return 0;
467}
468
469static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
470 const struct iphdr *tnl_params,
471 __be16 proto)
472{
473 struct ip_tunnel *tunnel = netdev_priv(dev);
474 IP_TUNNEL_DECLARE_FLAGS(flags);
475
476 ip_tunnel_flags_copy(flags, tunnel->parms.o_flags);
477
478 /* Push GRE header. */
479 gre_build_header(skb, tunnel->tun_hlen,
480 flags, proto, tunnel->parms.o_key,
481 test_bit(IP_TUNNEL_SEQ_BIT, flags) ?
482 htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0);
483
484 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
485}
486
487static int gre_handle_offloads(struct sk_buff *skb, bool csum)
488{
489 return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
490}
491
492static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
493 __be16 proto)
494{
495 struct ip_tunnel *tunnel = netdev_priv(dev);
496 IP_TUNNEL_DECLARE_FLAGS(flags) = { };
497 struct ip_tunnel_info *tun_info;
498 const struct ip_tunnel_key *key;
499 int tunnel_hlen;
500
501 tun_info = skb_tunnel_info(skb);
502 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
503 ip_tunnel_info_af(tun_info) != AF_INET))
504 goto err_free_skb;
505
506 key = &tun_info->key;
507 tunnel_hlen = gre_calc_hlen(key->tun_flags);
508
509 if (skb_cow_head(skb, dev->needed_headroom))
510 goto err_free_skb;
511
512 /* Push Tunnel header. */
513 if (gre_handle_offloads(skb, test_bit(IP_TUNNEL_CSUM_BIT,
514 tunnel->parms.o_flags)))
515 goto err_free_skb;
516
517 __set_bit(IP_TUNNEL_CSUM_BIT, flags);
518 __set_bit(IP_TUNNEL_KEY_BIT, flags);
519 __set_bit(IP_TUNNEL_SEQ_BIT, flags);
520 ip_tunnel_flags_and(flags, tun_info->key.tun_flags, flags);
521
522 gre_build_header(skb, tunnel_hlen, flags, proto,
523 tunnel_id_to_key32(tun_info->key.tun_id),
524 test_bit(IP_TUNNEL_SEQ_BIT, flags) ?
525 htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0);
526
527 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
528
529 return;
530
531err_free_skb:
532 kfree_skb(skb);
533 DEV_STATS_INC(dev, tx_dropped);
534}
535
536static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev)
537{
538 struct ip_tunnel *tunnel = netdev_priv(dev);
539 IP_TUNNEL_DECLARE_FLAGS(flags) = { };
540 struct ip_tunnel_info *tun_info;
541 const struct ip_tunnel_key *key;
542 struct erspan_metadata *md;
543 bool truncate = false;
544 __be16 proto;
545 int tunnel_hlen;
546 int version;
547 int nhoff;
548
549 tun_info = skb_tunnel_info(skb);
550 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
551 ip_tunnel_info_af(tun_info) != AF_INET))
552 goto err_free_skb;
553
554 key = &tun_info->key;
555 if (!test_bit(IP_TUNNEL_ERSPAN_OPT_BIT, tun_info->key.tun_flags))
556 goto err_free_skb;
557 if (tun_info->options_len < sizeof(*md))
558 goto err_free_skb;
559 md = ip_tunnel_info_opts(tun_info);
560
561 /* ERSPAN has fixed 8 byte GRE header */
562 version = md->version;
563 tunnel_hlen = 8 + erspan_hdr_len(version);
564
565 if (skb_cow_head(skb, dev->needed_headroom))
566 goto err_free_skb;
567
568 if (gre_handle_offloads(skb, false))
569 goto err_free_skb;
570
571 if (skb->len > dev->mtu + dev->hard_header_len) {
572 if (pskb_trim(skb, dev->mtu + dev->hard_header_len))
573 goto err_free_skb;
574 truncate = true;
575 }
576
577 nhoff = skb_network_offset(skb);
578 if (skb->protocol == htons(ETH_P_IP) &&
579 (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
580 truncate = true;
581
582 if (skb->protocol == htons(ETH_P_IPV6)) {
583 int thoff;
584
585 if (skb_transport_header_was_set(skb))
586 thoff = skb_transport_offset(skb);
587 else
588 thoff = nhoff + sizeof(struct ipv6hdr);
589 if (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff)
590 truncate = true;
591 }
592
593 if (version == 1) {
594 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
595 ntohl(md->u.index), truncate, true);
596 proto = htons(ETH_P_ERSPAN);
597 } else if (version == 2) {
598 erspan_build_header_v2(skb,
599 ntohl(tunnel_id_to_key32(key->tun_id)),
600 md->u.md2.dir,
601 get_hwid(&md->u.md2),
602 truncate, true);
603 proto = htons(ETH_P_ERSPAN2);
604 } else {
605 goto err_free_skb;
606 }
607
608 __set_bit(IP_TUNNEL_SEQ_BIT, flags);
609 gre_build_header(skb, 8, flags, proto, 0,
610 htonl(atomic_fetch_inc(&tunnel->o_seqno)));
611
612 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
613
614 return;
615
616err_free_skb:
617 kfree_skb(skb);
618 DEV_STATS_INC(dev, tx_dropped);
619}
620
621static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
622{
623 struct ip_tunnel_info *info = skb_tunnel_info(skb);
624 const struct ip_tunnel_key *key;
625 struct rtable *rt;
626 struct flowi4 fl4;
627
628 if (ip_tunnel_info_af(info) != AF_INET)
629 return -EINVAL;
630
631 key = &info->key;
632 ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src,
633 tunnel_id_to_key32(key->tun_id),
634 key->tos & ~INET_ECN_MASK, dev_net(dev), 0,
635 skb->mark, skb_get_hash(skb), key->flow_flags);
636 rt = ip_route_output_key(dev_net(dev), &fl4);
637 if (IS_ERR(rt))
638 return PTR_ERR(rt);
639
640 ip_rt_put(rt);
641 info->key.u.ipv4.src = fl4.saddr;
642 return 0;
643}
644
645static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
646 struct net_device *dev)
647{
648 struct ip_tunnel *tunnel = netdev_priv(dev);
649 const struct iphdr *tnl_params;
650
651 if (!pskb_inet_may_pull(skb))
652 goto free_skb;
653
654 if (tunnel->collect_md) {
655 gre_fb_xmit(skb, dev, skb->protocol);
656 return NETDEV_TX_OK;
657 }
658
659 if (dev->header_ops) {
660 int pull_len = tunnel->hlen + sizeof(struct iphdr);
661
662 if (skb_cow_head(skb, 0))
663 goto free_skb;
664
665 if (!pskb_may_pull(skb, pull_len))
666 goto free_skb;
667
668 tnl_params = (const struct iphdr *)skb->data;
669
670 /* ip_tunnel_xmit() needs skb->data pointing to gre header. */
671 skb_pull(skb, pull_len);
672 skb_reset_mac_header(skb);
673
674 if (skb->ip_summed == CHECKSUM_PARTIAL &&
675 skb_checksum_start(skb) < skb->data)
676 goto free_skb;
677 } else {
678 if (skb_cow_head(skb, dev->needed_headroom))
679 goto free_skb;
680
681 tnl_params = &tunnel->parms.iph;
682 }
683
684 if (gre_handle_offloads(skb, test_bit(IP_TUNNEL_CSUM_BIT,
685 tunnel->parms.o_flags)))
686 goto free_skb;
687
688 __gre_xmit(skb, dev, tnl_params, skb->protocol);
689 return NETDEV_TX_OK;
690
691free_skb:
692 kfree_skb(skb);
693 DEV_STATS_INC(dev, tx_dropped);
694 return NETDEV_TX_OK;
695}
696
697static netdev_tx_t erspan_xmit(struct sk_buff *skb,
698 struct net_device *dev)
699{
700 struct ip_tunnel *tunnel = netdev_priv(dev);
701 bool truncate = false;
702 __be16 proto;
703
704 if (!pskb_inet_may_pull(skb))
705 goto free_skb;
706
707 if (tunnel->collect_md) {
708 erspan_fb_xmit(skb, dev);
709 return NETDEV_TX_OK;
710 }
711
712 if (gre_handle_offloads(skb, false))
713 goto free_skb;
714
715 if (skb_cow_head(skb, dev->needed_headroom))
716 goto free_skb;
717
718 if (skb->len > dev->mtu + dev->hard_header_len) {
719 if (pskb_trim(skb, dev->mtu + dev->hard_header_len))
720 goto free_skb;
721 truncate = true;
722 }
723
724 /* Push ERSPAN header */
725 if (tunnel->erspan_ver == 0) {
726 proto = htons(ETH_P_ERSPAN);
727 __clear_bit(IP_TUNNEL_SEQ_BIT, tunnel->parms.o_flags);
728 } else if (tunnel->erspan_ver == 1) {
729 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
730 tunnel->index,
731 truncate, true);
732 proto = htons(ETH_P_ERSPAN);
733 } else if (tunnel->erspan_ver == 2) {
734 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
735 tunnel->dir, tunnel->hwid,
736 truncate, true);
737 proto = htons(ETH_P_ERSPAN2);
738 } else {
739 goto free_skb;
740 }
741
742 __clear_bit(IP_TUNNEL_KEY_BIT, tunnel->parms.o_flags);
743 __gre_xmit(skb, dev, &tunnel->parms.iph, proto);
744 return NETDEV_TX_OK;
745
746free_skb:
747 kfree_skb(skb);
748 DEV_STATS_INC(dev, tx_dropped);
749 return NETDEV_TX_OK;
750}
751
752static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
753 struct net_device *dev)
754{
755 struct ip_tunnel *tunnel = netdev_priv(dev);
756
757 if (!pskb_inet_may_pull(skb))
758 goto free_skb;
759
760 if (tunnel->collect_md) {
761 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
762 return NETDEV_TX_OK;
763 }
764
765 if (gre_handle_offloads(skb, test_bit(IP_TUNNEL_CSUM_BIT,
766 tunnel->parms.o_flags)))
767 goto free_skb;
768
769 if (skb_cow_head(skb, dev->needed_headroom))
770 goto free_skb;
771
772 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
773 return NETDEV_TX_OK;
774
775free_skb:
776 kfree_skb(skb);
777 DEV_STATS_INC(dev, tx_dropped);
778 return NETDEV_TX_OK;
779}
780
781static void ipgre_link_update(struct net_device *dev, bool set_mtu)
782{
783 struct ip_tunnel *tunnel = netdev_priv(dev);
784 int len;
785
786 len = tunnel->tun_hlen;
787 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
788 len = tunnel->tun_hlen - len;
789 tunnel->hlen = tunnel->hlen + len;
790
791 if (dev->header_ops)
792 dev->hard_header_len += len;
793 else
794 dev->needed_headroom += len;
795
796 if (set_mtu)
797 WRITE_ONCE(dev->mtu, max_t(int, dev->mtu - len, 68));
798
799 if (test_bit(IP_TUNNEL_SEQ_BIT, tunnel->parms.o_flags) ||
800 (test_bit(IP_TUNNEL_CSUM_BIT, tunnel->parms.o_flags) &&
801 tunnel->encap.type != TUNNEL_ENCAP_NONE)) {
802 dev->features &= ~NETIF_F_GSO_SOFTWARE;
803 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
804 } else {
805 dev->features |= NETIF_F_GSO_SOFTWARE;
806 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
807 }
808}
809
810static int ipgre_tunnel_ctl(struct net_device *dev,
811 struct ip_tunnel_parm_kern *p,
812 int cmd)
813{
814 __be16 i_flags, o_flags;
815 int err;
816
817 if (!ip_tunnel_flags_is_be16_compat(p->i_flags) ||
818 !ip_tunnel_flags_is_be16_compat(p->o_flags))
819 return -EOVERFLOW;
820
821 i_flags = ip_tunnel_flags_to_be16(p->i_flags);
822 o_flags = ip_tunnel_flags_to_be16(p->o_flags);
823
824 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
825 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE ||
826 p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) ||
827 ((i_flags | o_flags) & (GRE_VERSION | GRE_ROUTING)))
828 return -EINVAL;
829 }
830
831 gre_flags_to_tnl_flags(p->i_flags, i_flags);
832 gre_flags_to_tnl_flags(p->o_flags, o_flags);
833
834 err = ip_tunnel_ctl(dev, p, cmd);
835 if (err)
836 return err;
837
838 if (cmd == SIOCCHGTUNNEL) {
839 struct ip_tunnel *t = netdev_priv(dev);
840
841 ip_tunnel_flags_copy(t->parms.i_flags, p->i_flags);
842 ip_tunnel_flags_copy(t->parms.o_flags, p->o_flags);
843
844 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
845 ipgre_link_update(dev, true);
846 }
847
848 i_flags = gre_tnl_flags_to_gre_flags(p->i_flags);
849 ip_tunnel_flags_from_be16(p->i_flags, i_flags);
850 o_flags = gre_tnl_flags_to_gre_flags(p->o_flags);
851 ip_tunnel_flags_from_be16(p->o_flags, o_flags);
852
853 return 0;
854}
855
856/* Nice toy. Unfortunately, useless in real life :-)
857 It allows to construct virtual multiprotocol broadcast "LAN"
858 over the Internet, provided multicast routing is tuned.
859
860
861 I have no idea was this bicycle invented before me,
862 so that I had to set ARPHRD_IPGRE to a random value.
863 I have an impression, that Cisco could make something similar,
864 but this feature is apparently missing in IOS<=11.2(8).
865
866 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
867 with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
868
869 ping -t 255 224.66.66.66
870
871 If nobody answers, mbone does not work.
872
873 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
874 ip addr add 10.66.66.<somewhat>/24 dev Universe
875 ifconfig Universe up
876 ifconfig Universe add fe80::<Your_real_addr>/10
877 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
878 ftp 10.66.66.66
879 ...
880 ftp fec0:6666:6666::193.233.7.65
881 ...
882 */
883static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
884 unsigned short type,
885 const void *daddr, const void *saddr, unsigned int len)
886{
887 struct ip_tunnel *t = netdev_priv(dev);
888 struct iphdr *iph;
889 struct gre_base_hdr *greh;
890
891 iph = skb_push(skb, t->hlen + sizeof(*iph));
892 greh = (struct gre_base_hdr *)(iph+1);
893 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
894 greh->protocol = htons(type);
895
896 memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
897
898 /* Set the source hardware address. */
899 if (saddr)
900 memcpy(&iph->saddr, saddr, 4);
901 if (daddr)
902 memcpy(&iph->daddr, daddr, 4);
903 if (iph->daddr)
904 return t->hlen + sizeof(*iph);
905
906 return -(t->hlen + sizeof(*iph));
907}
908
909static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
910{
911 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
912 memcpy(haddr, &iph->saddr, 4);
913 return 4;
914}
915
916static const struct header_ops ipgre_header_ops = {
917 .create = ipgre_header,
918 .parse = ipgre_header_parse,
919};
920
921#ifdef CONFIG_NET_IPGRE_BROADCAST
922static int ipgre_open(struct net_device *dev)
923{
924 struct ip_tunnel *t = netdev_priv(dev);
925
926 if (ipv4_is_multicast(t->parms.iph.daddr)) {
927 struct flowi4 fl4;
928 struct rtable *rt;
929
930 rt = ip_route_output_gre(t->net, &fl4,
931 t->parms.iph.daddr,
932 t->parms.iph.saddr,
933 t->parms.o_key,
934 t->parms.iph.tos & INET_DSCP_MASK,
935 t->parms.link);
936 if (IS_ERR(rt))
937 return -EADDRNOTAVAIL;
938 dev = rt->dst.dev;
939 ip_rt_put(rt);
940 if (!__in_dev_get_rtnl(dev))
941 return -EADDRNOTAVAIL;
942 t->mlink = dev->ifindex;
943 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
944 }
945 return 0;
946}
947
948static int ipgre_close(struct net_device *dev)
949{
950 struct ip_tunnel *t = netdev_priv(dev);
951
952 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
953 struct in_device *in_dev;
954 in_dev = inetdev_by_index(t->net, t->mlink);
955 if (in_dev)
956 ip_mc_dec_group(in_dev, t->parms.iph.daddr);
957 }
958 return 0;
959}
960#endif
961
962static const struct net_device_ops ipgre_netdev_ops = {
963 .ndo_init = ipgre_tunnel_init,
964 .ndo_uninit = ip_tunnel_uninit,
965#ifdef CONFIG_NET_IPGRE_BROADCAST
966 .ndo_open = ipgre_open,
967 .ndo_stop = ipgre_close,
968#endif
969 .ndo_start_xmit = ipgre_xmit,
970 .ndo_siocdevprivate = ip_tunnel_siocdevprivate,
971 .ndo_change_mtu = ip_tunnel_change_mtu,
972 .ndo_get_stats64 = dev_get_tstats64,
973 .ndo_get_iflink = ip_tunnel_get_iflink,
974 .ndo_tunnel_ctl = ipgre_tunnel_ctl,
975};
976
977#define GRE_FEATURES (NETIF_F_SG | \
978 NETIF_F_FRAGLIST | \
979 NETIF_F_HIGHDMA | \
980 NETIF_F_HW_CSUM)
981
982static void ipgre_tunnel_setup(struct net_device *dev)
983{
984 dev->netdev_ops = &ipgre_netdev_ops;
985 dev->type = ARPHRD_IPGRE;
986 ip_tunnel_setup(dev, ipgre_net_id);
987}
988
989static void __gre_tunnel_init(struct net_device *dev)
990{
991 struct ip_tunnel *tunnel;
992
993 tunnel = netdev_priv(dev);
994 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
995 tunnel->parms.iph.protocol = IPPROTO_GRE;
996
997 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
998 dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph);
999
1000 dev->features |= GRE_FEATURES;
1001 dev->hw_features |= GRE_FEATURES;
1002
1003 /* TCP offload with GRE SEQ is not supported, nor can we support 2
1004 * levels of outer headers requiring an update.
1005 */
1006 if (test_bit(IP_TUNNEL_SEQ_BIT, tunnel->parms.o_flags))
1007 return;
1008 if (test_bit(IP_TUNNEL_CSUM_BIT, tunnel->parms.o_flags) &&
1009 tunnel->encap.type != TUNNEL_ENCAP_NONE)
1010 return;
1011
1012 dev->features |= NETIF_F_GSO_SOFTWARE;
1013 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1014
1015 dev->lltx = true;
1016}
1017
1018static int ipgre_tunnel_init(struct net_device *dev)
1019{
1020 struct ip_tunnel *tunnel = netdev_priv(dev);
1021 struct iphdr *iph = &tunnel->parms.iph;
1022
1023 __gre_tunnel_init(dev);
1024
1025 __dev_addr_set(dev, &iph->saddr, 4);
1026 memcpy(dev->broadcast, &iph->daddr, 4);
1027
1028 dev->flags = IFF_NOARP;
1029 netif_keep_dst(dev);
1030 dev->addr_len = 4;
1031
1032 if (iph->daddr && !tunnel->collect_md) {
1033#ifdef CONFIG_NET_IPGRE_BROADCAST
1034 if (ipv4_is_multicast(iph->daddr)) {
1035 if (!iph->saddr)
1036 return -EINVAL;
1037 dev->flags = IFF_BROADCAST;
1038 dev->header_ops = &ipgre_header_ops;
1039 dev->hard_header_len = tunnel->hlen + sizeof(*iph);
1040 dev->needed_headroom = 0;
1041 }
1042#endif
1043 } else if (!tunnel->collect_md) {
1044 dev->header_ops = &ipgre_header_ops;
1045 dev->hard_header_len = tunnel->hlen + sizeof(*iph);
1046 dev->needed_headroom = 0;
1047 }
1048
1049 return ip_tunnel_init(dev);
1050}
1051
1052static const struct gre_protocol ipgre_protocol = {
1053 .handler = gre_rcv,
1054 .err_handler = gre_err,
1055};
1056
1057static int __net_init ipgre_init_net(struct net *net)
1058{
1059 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1060}
1061
1062static void __net_exit ipgre_exit_batch_rtnl(struct list_head *list_net,
1063 struct list_head *dev_to_kill)
1064{
1065 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops,
1066 dev_to_kill);
1067}
1068
1069static struct pernet_operations ipgre_net_ops = {
1070 .init = ipgre_init_net,
1071 .exit_batch_rtnl = ipgre_exit_batch_rtnl,
1072 .id = &ipgre_net_id,
1073 .size = sizeof(struct ip_tunnel_net),
1074};
1075
1076static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1077 struct netlink_ext_ack *extack)
1078{
1079 __be16 flags;
1080
1081 if (!data)
1082 return 0;
1083
1084 flags = 0;
1085 if (data[IFLA_GRE_IFLAGS])
1086 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1087 if (data[IFLA_GRE_OFLAGS])
1088 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1089 if (flags & (GRE_VERSION|GRE_ROUTING))
1090 return -EINVAL;
1091
1092 if (data[IFLA_GRE_COLLECT_METADATA] &&
1093 data[IFLA_GRE_ENCAP_TYPE] &&
1094 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1095 return -EINVAL;
1096
1097 return 0;
1098}
1099
1100static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1101 struct netlink_ext_ack *extack)
1102{
1103 __be32 daddr;
1104
1105 if (tb[IFLA_ADDRESS]) {
1106 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1107 return -EINVAL;
1108 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1109 return -EADDRNOTAVAIL;
1110 }
1111
1112 if (!data)
1113 goto out;
1114
1115 if (data[IFLA_GRE_REMOTE]) {
1116 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1117 if (!daddr)
1118 return -EINVAL;
1119 }
1120
1121out:
1122 return ipgre_tunnel_validate(tb, data, extack);
1123}
1124
1125static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1126 struct netlink_ext_ack *extack)
1127{
1128 __be16 flags = 0;
1129 int ret;
1130
1131 if (!data)
1132 return 0;
1133
1134 ret = ipgre_tap_validate(tb, data, extack);
1135 if (ret)
1136 return ret;
1137
1138 if (data[IFLA_GRE_ERSPAN_VER] &&
1139 nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0)
1140 return 0;
1141
1142 /* ERSPAN type II/III should only have GRE sequence and key flag */
1143 if (data[IFLA_GRE_OFLAGS])
1144 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1145 if (data[IFLA_GRE_IFLAGS])
1146 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1147 if (!data[IFLA_GRE_COLLECT_METADATA] &&
1148 flags != (GRE_SEQ | GRE_KEY))
1149 return -EINVAL;
1150
1151 /* ERSPAN Session ID only has 10-bit. Since we reuse
1152 * 32-bit key field as ID, check it's range.
1153 */
1154 if (data[IFLA_GRE_IKEY] &&
1155 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1156 return -EINVAL;
1157
1158 if (data[IFLA_GRE_OKEY] &&
1159 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1160 return -EINVAL;
1161
1162 return 0;
1163}
1164
1165static int ipgre_netlink_parms(struct net_device *dev,
1166 struct nlattr *data[],
1167 struct nlattr *tb[],
1168 struct ip_tunnel_parm_kern *parms,
1169 __u32 *fwmark)
1170{
1171 struct ip_tunnel *t = netdev_priv(dev);
1172
1173 memset(parms, 0, sizeof(*parms));
1174
1175 parms->iph.protocol = IPPROTO_GRE;
1176
1177 if (!data)
1178 return 0;
1179
1180 if (data[IFLA_GRE_LINK])
1181 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1182
1183 if (data[IFLA_GRE_IFLAGS])
1184 gre_flags_to_tnl_flags(parms->i_flags,
1185 nla_get_be16(data[IFLA_GRE_IFLAGS]));
1186
1187 if (data[IFLA_GRE_OFLAGS])
1188 gre_flags_to_tnl_flags(parms->o_flags,
1189 nla_get_be16(data[IFLA_GRE_OFLAGS]));
1190
1191 if (data[IFLA_GRE_IKEY])
1192 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1193
1194 if (data[IFLA_GRE_OKEY])
1195 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1196
1197 if (data[IFLA_GRE_LOCAL])
1198 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1199
1200 if (data[IFLA_GRE_REMOTE])
1201 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1202
1203 if (data[IFLA_GRE_TTL])
1204 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1205
1206 if (data[IFLA_GRE_TOS])
1207 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1208
1209 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1210 if (t->ignore_df)
1211 return -EINVAL;
1212 parms->iph.frag_off = htons(IP_DF);
1213 }
1214
1215 if (data[IFLA_GRE_COLLECT_METADATA]) {
1216 t->collect_md = true;
1217 if (dev->type == ARPHRD_IPGRE)
1218 dev->type = ARPHRD_NONE;
1219 }
1220
1221 if (data[IFLA_GRE_IGNORE_DF]) {
1222 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1223 && (parms->iph.frag_off & htons(IP_DF)))
1224 return -EINVAL;
1225 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1226 }
1227
1228 if (data[IFLA_GRE_FWMARK])
1229 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1230
1231 return 0;
1232}
1233
1234static int erspan_netlink_parms(struct net_device *dev,
1235 struct nlattr *data[],
1236 struct nlattr *tb[],
1237 struct ip_tunnel_parm_kern *parms,
1238 __u32 *fwmark)
1239{
1240 struct ip_tunnel *t = netdev_priv(dev);
1241 int err;
1242
1243 err = ipgre_netlink_parms(dev, data, tb, parms, fwmark);
1244 if (err)
1245 return err;
1246 if (!data)
1247 return 0;
1248
1249 if (data[IFLA_GRE_ERSPAN_VER]) {
1250 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1251
1252 if (t->erspan_ver > 2)
1253 return -EINVAL;
1254 }
1255
1256 if (t->erspan_ver == 1) {
1257 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1258 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1259 if (t->index & ~INDEX_MASK)
1260 return -EINVAL;
1261 }
1262 } else if (t->erspan_ver == 2) {
1263 if (data[IFLA_GRE_ERSPAN_DIR]) {
1264 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1265 if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1266 return -EINVAL;
1267 }
1268 if (data[IFLA_GRE_ERSPAN_HWID]) {
1269 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1270 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1271 return -EINVAL;
1272 }
1273 }
1274
1275 return 0;
1276}
1277
1278/* This function returns true when ENCAP attributes are present in the nl msg */
1279static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1280 struct ip_tunnel_encap *ipencap)
1281{
1282 bool ret = false;
1283
1284 memset(ipencap, 0, sizeof(*ipencap));
1285
1286 if (!data)
1287 return ret;
1288
1289 if (data[IFLA_GRE_ENCAP_TYPE]) {
1290 ret = true;
1291 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1292 }
1293
1294 if (data[IFLA_GRE_ENCAP_FLAGS]) {
1295 ret = true;
1296 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1297 }
1298
1299 if (data[IFLA_GRE_ENCAP_SPORT]) {
1300 ret = true;
1301 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1302 }
1303
1304 if (data[IFLA_GRE_ENCAP_DPORT]) {
1305 ret = true;
1306 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1307 }
1308
1309 return ret;
1310}
1311
1312static int gre_tap_init(struct net_device *dev)
1313{
1314 __gre_tunnel_init(dev);
1315 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1316 netif_keep_dst(dev);
1317
1318 return ip_tunnel_init(dev);
1319}
1320
1321static const struct net_device_ops gre_tap_netdev_ops = {
1322 .ndo_init = gre_tap_init,
1323 .ndo_uninit = ip_tunnel_uninit,
1324 .ndo_start_xmit = gre_tap_xmit,
1325 .ndo_set_mac_address = eth_mac_addr,
1326 .ndo_validate_addr = eth_validate_addr,
1327 .ndo_change_mtu = ip_tunnel_change_mtu,
1328 .ndo_get_stats64 = dev_get_tstats64,
1329 .ndo_get_iflink = ip_tunnel_get_iflink,
1330 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1331};
1332
1333static int erspan_tunnel_init(struct net_device *dev)
1334{
1335 struct ip_tunnel *tunnel = netdev_priv(dev);
1336
1337 if (tunnel->erspan_ver == 0)
1338 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */
1339 else
1340 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */
1341
1342 tunnel->parms.iph.protocol = IPPROTO_GRE;
1343 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1344 erspan_hdr_len(tunnel->erspan_ver);
1345
1346 dev->features |= GRE_FEATURES;
1347 dev->hw_features |= GRE_FEATURES;
1348 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1349 netif_keep_dst(dev);
1350
1351 return ip_tunnel_init(dev);
1352}
1353
1354static const struct net_device_ops erspan_netdev_ops = {
1355 .ndo_init = erspan_tunnel_init,
1356 .ndo_uninit = ip_tunnel_uninit,
1357 .ndo_start_xmit = erspan_xmit,
1358 .ndo_set_mac_address = eth_mac_addr,
1359 .ndo_validate_addr = eth_validate_addr,
1360 .ndo_change_mtu = ip_tunnel_change_mtu,
1361 .ndo_get_stats64 = dev_get_tstats64,
1362 .ndo_get_iflink = ip_tunnel_get_iflink,
1363 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1364};
1365
1366static void ipgre_tap_setup(struct net_device *dev)
1367{
1368 ether_setup(dev);
1369 dev->max_mtu = 0;
1370 dev->netdev_ops = &gre_tap_netdev_ops;
1371 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1372 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1373 ip_tunnel_setup(dev, gre_tap_net_id);
1374}
1375
1376static int
1377ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[])
1378{
1379 struct ip_tunnel_encap ipencap;
1380
1381 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1382 struct ip_tunnel *t = netdev_priv(dev);
1383 int err = ip_tunnel_encap_setup(t, &ipencap);
1384
1385 if (err < 0)
1386 return err;
1387 }
1388
1389 return 0;
1390}
1391
1392static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1393 struct nlattr *tb[], struct nlattr *data[],
1394 struct netlink_ext_ack *extack)
1395{
1396 struct ip_tunnel_parm_kern p;
1397 __u32 fwmark = 0;
1398 int err;
1399
1400 err = ipgre_newlink_encap_setup(dev, data);
1401 if (err)
1402 return err;
1403
1404 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1405 if (err < 0)
1406 return err;
1407 return ip_tunnel_newlink(dev, tb, &p, fwmark);
1408}
1409
1410static int erspan_newlink(struct net *src_net, struct net_device *dev,
1411 struct nlattr *tb[], struct nlattr *data[],
1412 struct netlink_ext_ack *extack)
1413{
1414 struct ip_tunnel_parm_kern p;
1415 __u32 fwmark = 0;
1416 int err;
1417
1418 err = ipgre_newlink_encap_setup(dev, data);
1419 if (err)
1420 return err;
1421
1422 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark);
1423 if (err)
1424 return err;
1425 return ip_tunnel_newlink(dev, tb, &p, fwmark);
1426}
1427
1428static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1429 struct nlattr *data[],
1430 struct netlink_ext_ack *extack)
1431{
1432 struct ip_tunnel *t = netdev_priv(dev);
1433 struct ip_tunnel_parm_kern p;
1434 __u32 fwmark = t->fwmark;
1435 int err;
1436
1437 err = ipgre_newlink_encap_setup(dev, data);
1438 if (err)
1439 return err;
1440
1441 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1442 if (err < 0)
1443 return err;
1444
1445 err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1446 if (err < 0)
1447 return err;
1448
1449 ip_tunnel_flags_copy(t->parms.i_flags, p.i_flags);
1450 ip_tunnel_flags_copy(t->parms.o_flags, p.o_flags);
1451
1452 ipgre_link_update(dev, !tb[IFLA_MTU]);
1453
1454 return 0;
1455}
1456
1457static int erspan_changelink(struct net_device *dev, struct nlattr *tb[],
1458 struct nlattr *data[],
1459 struct netlink_ext_ack *extack)
1460{
1461 struct ip_tunnel *t = netdev_priv(dev);
1462 struct ip_tunnel_parm_kern p;
1463 __u32 fwmark = t->fwmark;
1464 int err;
1465
1466 err = ipgre_newlink_encap_setup(dev, data);
1467 if (err)
1468 return err;
1469
1470 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark);
1471 if (err < 0)
1472 return err;
1473
1474 err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1475 if (err < 0)
1476 return err;
1477
1478 ip_tunnel_flags_copy(t->parms.i_flags, p.i_flags);
1479 ip_tunnel_flags_copy(t->parms.o_flags, p.o_flags);
1480
1481 return 0;
1482}
1483
1484static size_t ipgre_get_size(const struct net_device *dev)
1485{
1486 return
1487 /* IFLA_GRE_LINK */
1488 nla_total_size(4) +
1489 /* IFLA_GRE_IFLAGS */
1490 nla_total_size(2) +
1491 /* IFLA_GRE_OFLAGS */
1492 nla_total_size(2) +
1493 /* IFLA_GRE_IKEY */
1494 nla_total_size(4) +
1495 /* IFLA_GRE_OKEY */
1496 nla_total_size(4) +
1497 /* IFLA_GRE_LOCAL */
1498 nla_total_size(4) +
1499 /* IFLA_GRE_REMOTE */
1500 nla_total_size(4) +
1501 /* IFLA_GRE_TTL */
1502 nla_total_size(1) +
1503 /* IFLA_GRE_TOS */
1504 nla_total_size(1) +
1505 /* IFLA_GRE_PMTUDISC */
1506 nla_total_size(1) +
1507 /* IFLA_GRE_ENCAP_TYPE */
1508 nla_total_size(2) +
1509 /* IFLA_GRE_ENCAP_FLAGS */
1510 nla_total_size(2) +
1511 /* IFLA_GRE_ENCAP_SPORT */
1512 nla_total_size(2) +
1513 /* IFLA_GRE_ENCAP_DPORT */
1514 nla_total_size(2) +
1515 /* IFLA_GRE_COLLECT_METADATA */
1516 nla_total_size(0) +
1517 /* IFLA_GRE_IGNORE_DF */
1518 nla_total_size(1) +
1519 /* IFLA_GRE_FWMARK */
1520 nla_total_size(4) +
1521 /* IFLA_GRE_ERSPAN_INDEX */
1522 nla_total_size(4) +
1523 /* IFLA_GRE_ERSPAN_VER */
1524 nla_total_size(1) +
1525 /* IFLA_GRE_ERSPAN_DIR */
1526 nla_total_size(1) +
1527 /* IFLA_GRE_ERSPAN_HWID */
1528 nla_total_size(2) +
1529 0;
1530}
1531
1532static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1533{
1534 struct ip_tunnel *t = netdev_priv(dev);
1535 struct ip_tunnel_parm_kern *p = &t->parms;
1536 IP_TUNNEL_DECLARE_FLAGS(o_flags);
1537
1538 ip_tunnel_flags_copy(o_flags, p->o_flags);
1539
1540 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1541 nla_put_be16(skb, IFLA_GRE_IFLAGS,
1542 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1543 nla_put_be16(skb, IFLA_GRE_OFLAGS,
1544 gre_tnl_flags_to_gre_flags(o_flags)) ||
1545 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1546 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1547 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1548 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1549 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1550 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1551 nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1552 !!(p->iph.frag_off & htons(IP_DF))) ||
1553 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1554 goto nla_put_failure;
1555
1556 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1557 t->encap.type) ||
1558 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1559 t->encap.sport) ||
1560 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1561 t->encap.dport) ||
1562 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1563 t->encap.flags))
1564 goto nla_put_failure;
1565
1566 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1567 goto nla_put_failure;
1568
1569 if (t->collect_md) {
1570 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1571 goto nla_put_failure;
1572 }
1573
1574 return 0;
1575
1576nla_put_failure:
1577 return -EMSGSIZE;
1578}
1579
1580static int erspan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1581{
1582 struct ip_tunnel *t = netdev_priv(dev);
1583
1584 if (t->erspan_ver <= 2) {
1585 if (t->erspan_ver != 0 && !t->collect_md)
1586 __set_bit(IP_TUNNEL_KEY_BIT, t->parms.o_flags);
1587
1588 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1589 goto nla_put_failure;
1590
1591 if (t->erspan_ver == 1) {
1592 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1593 goto nla_put_failure;
1594 } else if (t->erspan_ver == 2) {
1595 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1596 goto nla_put_failure;
1597 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1598 goto nla_put_failure;
1599 }
1600 }
1601
1602 return ipgre_fill_info(skb, dev);
1603
1604nla_put_failure:
1605 return -EMSGSIZE;
1606}
1607
1608static void erspan_setup(struct net_device *dev)
1609{
1610 struct ip_tunnel *t = netdev_priv(dev);
1611
1612 ether_setup(dev);
1613 dev->max_mtu = 0;
1614 dev->netdev_ops = &erspan_netdev_ops;
1615 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1616 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1617 ip_tunnel_setup(dev, erspan_net_id);
1618 t->erspan_ver = 1;
1619}
1620
1621static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1622 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1623 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1624 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1625 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1626 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1627 [IFLA_GRE_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) },
1628 [IFLA_GRE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1629 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1630 [IFLA_GRE_TOS] = { .type = NLA_U8 },
1631 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 },
1632 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 },
1633 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 },
1634 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 },
1635 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 },
1636 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG },
1637 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 },
1638 [IFLA_GRE_FWMARK] = { .type = NLA_U32 },
1639 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1640 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 },
1641 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 },
1642 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 },
1643};
1644
1645static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1646 .kind = "gre",
1647 .maxtype = IFLA_GRE_MAX,
1648 .policy = ipgre_policy,
1649 .priv_size = sizeof(struct ip_tunnel),
1650 .setup = ipgre_tunnel_setup,
1651 .validate = ipgre_tunnel_validate,
1652 .newlink = ipgre_newlink,
1653 .changelink = ipgre_changelink,
1654 .dellink = ip_tunnel_dellink,
1655 .get_size = ipgre_get_size,
1656 .fill_info = ipgre_fill_info,
1657 .get_link_net = ip_tunnel_get_link_net,
1658};
1659
1660static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1661 .kind = "gretap",
1662 .maxtype = IFLA_GRE_MAX,
1663 .policy = ipgre_policy,
1664 .priv_size = sizeof(struct ip_tunnel),
1665 .setup = ipgre_tap_setup,
1666 .validate = ipgre_tap_validate,
1667 .newlink = ipgre_newlink,
1668 .changelink = ipgre_changelink,
1669 .dellink = ip_tunnel_dellink,
1670 .get_size = ipgre_get_size,
1671 .fill_info = ipgre_fill_info,
1672 .get_link_net = ip_tunnel_get_link_net,
1673};
1674
1675static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1676 .kind = "erspan",
1677 .maxtype = IFLA_GRE_MAX,
1678 .policy = ipgre_policy,
1679 .priv_size = sizeof(struct ip_tunnel),
1680 .setup = erspan_setup,
1681 .validate = erspan_validate,
1682 .newlink = erspan_newlink,
1683 .changelink = erspan_changelink,
1684 .dellink = ip_tunnel_dellink,
1685 .get_size = ipgre_get_size,
1686 .fill_info = erspan_fill_info,
1687 .get_link_net = ip_tunnel_get_link_net,
1688};
1689
1690struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1691 u8 name_assign_type)
1692{
1693 struct nlattr *tb[IFLA_MAX + 1];
1694 struct net_device *dev;
1695 LIST_HEAD(list_kill);
1696 struct ip_tunnel *t;
1697 int err;
1698
1699 memset(&tb, 0, sizeof(tb));
1700
1701 dev = rtnl_create_link(net, name, name_assign_type,
1702 &ipgre_tap_ops, tb, NULL);
1703 if (IS_ERR(dev))
1704 return dev;
1705
1706 /* Configure flow based GRE device. */
1707 t = netdev_priv(dev);
1708 t->collect_md = true;
1709
1710 err = ipgre_newlink(net, dev, tb, NULL, NULL);
1711 if (err < 0) {
1712 free_netdev(dev);
1713 return ERR_PTR(err);
1714 }
1715
1716 /* openvswitch users expect packet sizes to be unrestricted,
1717 * so set the largest MTU we can.
1718 */
1719 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1720 if (err)
1721 goto out;
1722
1723 err = rtnl_configure_link(dev, NULL, 0, NULL);
1724 if (err < 0)
1725 goto out;
1726
1727 return dev;
1728out:
1729 ip_tunnel_dellink(dev, &list_kill);
1730 unregister_netdevice_many(&list_kill);
1731 return ERR_PTR(err);
1732}
1733EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1734
1735static int __net_init ipgre_tap_init_net(struct net *net)
1736{
1737 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1738}
1739
1740static void __net_exit ipgre_tap_exit_batch_rtnl(struct list_head *list_net,
1741 struct list_head *dev_to_kill)
1742{
1743 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops,
1744 dev_to_kill);
1745}
1746
1747static struct pernet_operations ipgre_tap_net_ops = {
1748 .init = ipgre_tap_init_net,
1749 .exit_batch_rtnl = ipgre_tap_exit_batch_rtnl,
1750 .id = &gre_tap_net_id,
1751 .size = sizeof(struct ip_tunnel_net),
1752};
1753
1754static int __net_init erspan_init_net(struct net *net)
1755{
1756 return ip_tunnel_init_net(net, erspan_net_id,
1757 &erspan_link_ops, "erspan0");
1758}
1759
1760static void __net_exit erspan_exit_batch_rtnl(struct list_head *net_list,
1761 struct list_head *dev_to_kill)
1762{
1763 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops,
1764 dev_to_kill);
1765}
1766
1767static struct pernet_operations erspan_net_ops = {
1768 .init = erspan_init_net,
1769 .exit_batch_rtnl = erspan_exit_batch_rtnl,
1770 .id = &erspan_net_id,
1771 .size = sizeof(struct ip_tunnel_net),
1772};
1773
1774static int __init ipgre_init(void)
1775{
1776 int err;
1777
1778 pr_info("GRE over IPv4 tunneling driver\n");
1779
1780 err = register_pernet_device(&ipgre_net_ops);
1781 if (err < 0)
1782 return err;
1783
1784 err = register_pernet_device(&ipgre_tap_net_ops);
1785 if (err < 0)
1786 goto pnet_tap_failed;
1787
1788 err = register_pernet_device(&erspan_net_ops);
1789 if (err < 0)
1790 goto pnet_erspan_failed;
1791
1792 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1793 if (err < 0) {
1794 pr_info("%s: can't add protocol\n", __func__);
1795 goto add_proto_failed;
1796 }
1797
1798 err = rtnl_link_register(&ipgre_link_ops);
1799 if (err < 0)
1800 goto rtnl_link_failed;
1801
1802 err = rtnl_link_register(&ipgre_tap_ops);
1803 if (err < 0)
1804 goto tap_ops_failed;
1805
1806 err = rtnl_link_register(&erspan_link_ops);
1807 if (err < 0)
1808 goto erspan_link_failed;
1809
1810 return 0;
1811
1812erspan_link_failed:
1813 rtnl_link_unregister(&ipgre_tap_ops);
1814tap_ops_failed:
1815 rtnl_link_unregister(&ipgre_link_ops);
1816rtnl_link_failed:
1817 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1818add_proto_failed:
1819 unregister_pernet_device(&erspan_net_ops);
1820pnet_erspan_failed:
1821 unregister_pernet_device(&ipgre_tap_net_ops);
1822pnet_tap_failed:
1823 unregister_pernet_device(&ipgre_net_ops);
1824 return err;
1825}
1826
1827static void __exit ipgre_fini(void)
1828{
1829 rtnl_link_unregister(&ipgre_tap_ops);
1830 rtnl_link_unregister(&ipgre_link_ops);
1831 rtnl_link_unregister(&erspan_link_ops);
1832 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1833 unregister_pernet_device(&ipgre_tap_net_ops);
1834 unregister_pernet_device(&ipgre_net_ops);
1835 unregister_pernet_device(&erspan_net_ops);
1836}
1837
1838module_init(ipgre_init);
1839module_exit(ipgre_fini);
1840MODULE_DESCRIPTION("IPv4 GRE tunnels over IP library");
1841MODULE_LICENSE("GPL");
1842MODULE_ALIAS_RTNL_LINK("gre");
1843MODULE_ALIAS_RTNL_LINK("gretap");
1844MODULE_ALIAS_RTNL_LINK("erspan");
1845MODULE_ALIAS_NETDEV("gre0");
1846MODULE_ALIAS_NETDEV("gretap0");
1847MODULE_ALIAS_NETDEV("erspan0");