Loading...
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");
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 <asm/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/mroute.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/ipip.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
51#if IS_ENABLED(CONFIG_IPV6)
52#include <net/ipv6.h>
53#include <net/ip6_fib.h>
54#include <net/ip6_route.h>
55#endif
56
57/*
58 Problems & solutions
59 --------------------
60
61 1. The most important issue is detecting local dead loops.
62 They would cause complete host lockup in transmit, which
63 would be "resolved" by stack overflow or, if queueing is enabled,
64 with infinite looping in net_bh.
65
66 We cannot track such dead loops during route installation,
67 it is infeasible task. The most general solutions would be
68 to keep skb->encapsulation counter (sort of local ttl),
69 and silently drop packet when it expires. It is a good
70 solution, but it supposes maintaining new variable in ALL
71 skb, even if no tunneling is used.
72
73 Current solution: xmit_recursion breaks dead loops. This is a percpu
74 counter, since when we enter the first ndo_xmit(), cpu migration is
75 forbidden. We force an exit if this counter reaches RECURSION_LIMIT
76
77 2. Networking dead loops would not kill routers, but would really
78 kill network. IP hop limit plays role of "t->recursion" in this case,
79 if we copy it from packet being encapsulated to upper header.
80 It is very good solution, but it introduces two problems:
81
82 - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
83 do not work over tunnels.
84 - traceroute does not work. I planned to relay ICMP from tunnel,
85 so that this problem would be solved and traceroute output
86 would even more informative. This idea appeared to be wrong:
87 only Linux complies to rfc1812 now (yes, guys, Linux is the only
88 true router now :-)), all routers (at least, in neighbourhood of mine)
89 return only 8 bytes of payload. It is the end.
90
91 Hence, if we want that OSPF worked or traceroute said something reasonable,
92 we should search for another solution.
93
94 One of them is to parse packet trying to detect inner encapsulation
95 made by our node. It is difficult or even impossible, especially,
96 taking into account fragmentation. TO be short, ttl is not solution at all.
97
98 Current solution: The solution was UNEXPECTEDLY SIMPLE.
99 We force DF flag on tunnels with preconfigured hop limit,
100 that is ALL. :-) Well, it does not remove the problem completely,
101 but exponential growth of network traffic is changed to linear
102 (branches, that exceed pmtu are pruned) and tunnel mtu
103 rapidly degrades to value <68, where looping stops.
104 Yes, it is not good if there exists a router in the loop,
105 which does not force DF, even when encapsulating packets have DF set.
106 But it is not our problem! Nobody could accuse us, we made
107 all that we could make. Even if it is your gated who injected
108 fatal route to network, even if it were you who configured
109 fatal static route: you are innocent. :-)
110
111
112
113 3. Really, ipv4/ipip.c, ipv4/ip_gre.c and ipv6/sit.c contain
114 practically identical code. It would be good to glue them
115 together, but it is not very evident, how to make them modular.
116 sit is integral part of IPv6, ipip and gre are naturally modular.
117 We could extract common parts (hash table, ioctl etc)
118 to a separate module (ip_tunnel.c).
119
120 Alexey Kuznetsov.
121 */
122
123static struct rtnl_link_ops ipgre_link_ops __read_mostly;
124static int ipgre_tunnel_init(struct net_device *dev);
125static void ipgre_tunnel_setup(struct net_device *dev);
126static int ipgre_tunnel_bind_dev(struct net_device *dev);
127
128/* Fallback tunnel: no source, no destination, no key, no options */
129
130#define HASH_SIZE 16
131
132static int ipgre_net_id __read_mostly;
133struct ipgre_net {
134 struct ip_tunnel __rcu *tunnels[4][HASH_SIZE];
135
136 struct net_device *fb_tunnel_dev;
137};
138
139/* Tunnel hash table */
140
141/*
142 4 hash tables:
143
144 3: (remote,local)
145 2: (remote,*)
146 1: (*,local)
147 0: (*,*)
148
149 We require exact key match i.e. if a key is present in packet
150 it will match only tunnel with the same key; if it is not present,
151 it will match only keyless tunnel.
152
153 All keysless packets, if not matched configured keyless tunnels
154 will match fallback tunnel.
155 */
156
157#define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
158
159#define tunnels_r_l tunnels[3]
160#define tunnels_r tunnels[2]
161#define tunnels_l tunnels[1]
162#define tunnels_wc tunnels[0]
163/*
164 * Locking : hash tables are protected by RCU and RTNL
165 */
166
167#define for_each_ip_tunnel_rcu(start) \
168 for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
169
170/* often modified stats are per cpu, other are shared (netdev->stats) */
171struct pcpu_tstats {
172 u64 rx_packets;
173 u64 rx_bytes;
174 u64 tx_packets;
175 u64 tx_bytes;
176 struct u64_stats_sync syncp;
177};
178
179static struct rtnl_link_stats64 *ipgre_get_stats64(struct net_device *dev,
180 struct rtnl_link_stats64 *tot)
181{
182 int i;
183
184 for_each_possible_cpu(i) {
185 const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
186 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
187 unsigned int start;
188
189 do {
190 start = u64_stats_fetch_begin_bh(&tstats->syncp);
191 rx_packets = tstats->rx_packets;
192 tx_packets = tstats->tx_packets;
193 rx_bytes = tstats->rx_bytes;
194 tx_bytes = tstats->tx_bytes;
195 } while (u64_stats_fetch_retry_bh(&tstats->syncp, start));
196
197 tot->rx_packets += rx_packets;
198 tot->tx_packets += tx_packets;
199 tot->rx_bytes += rx_bytes;
200 tot->tx_bytes += tx_bytes;
201 }
202
203 tot->multicast = dev->stats.multicast;
204 tot->rx_crc_errors = dev->stats.rx_crc_errors;
205 tot->rx_fifo_errors = dev->stats.rx_fifo_errors;
206 tot->rx_length_errors = dev->stats.rx_length_errors;
207 tot->rx_errors = dev->stats.rx_errors;
208 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
209 tot->tx_carrier_errors = dev->stats.tx_carrier_errors;
210 tot->tx_dropped = dev->stats.tx_dropped;
211 tot->tx_aborted_errors = dev->stats.tx_aborted_errors;
212 tot->tx_errors = dev->stats.tx_errors;
213
214 return tot;
215}
216
217/* Given src, dst and key, find appropriate for input tunnel. */
218
219static struct ip_tunnel *ipgre_tunnel_lookup(struct net_device *dev,
220 __be32 remote, __be32 local,
221 __be32 key, __be16 gre_proto)
222{
223 struct net *net = dev_net(dev);
224 int link = dev->ifindex;
225 unsigned int h0 = HASH(remote);
226 unsigned int h1 = HASH(key);
227 struct ip_tunnel *t, *cand = NULL;
228 struct ipgre_net *ign = net_generic(net, ipgre_net_id);
229 int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
230 ARPHRD_ETHER : ARPHRD_IPGRE;
231 int score, cand_score = 4;
232
233 for_each_ip_tunnel_rcu(ign->tunnels_r_l[h0 ^ h1]) {
234 if (local != t->parms.iph.saddr ||
235 remote != t->parms.iph.daddr ||
236 key != t->parms.i_key ||
237 !(t->dev->flags & IFF_UP))
238 continue;
239
240 if (t->dev->type != ARPHRD_IPGRE &&
241 t->dev->type != dev_type)
242 continue;
243
244 score = 0;
245 if (t->parms.link != link)
246 score |= 1;
247 if (t->dev->type != dev_type)
248 score |= 2;
249 if (score == 0)
250 return t;
251
252 if (score < cand_score) {
253 cand = t;
254 cand_score = score;
255 }
256 }
257
258 for_each_ip_tunnel_rcu(ign->tunnels_r[h0 ^ h1]) {
259 if (remote != t->parms.iph.daddr ||
260 key != t->parms.i_key ||
261 !(t->dev->flags & IFF_UP))
262 continue;
263
264 if (t->dev->type != ARPHRD_IPGRE &&
265 t->dev->type != dev_type)
266 continue;
267
268 score = 0;
269 if (t->parms.link != link)
270 score |= 1;
271 if (t->dev->type != dev_type)
272 score |= 2;
273 if (score == 0)
274 return t;
275
276 if (score < cand_score) {
277 cand = t;
278 cand_score = score;
279 }
280 }
281
282 for_each_ip_tunnel_rcu(ign->tunnels_l[h1]) {
283 if ((local != t->parms.iph.saddr &&
284 (local != t->parms.iph.daddr ||
285 !ipv4_is_multicast(local))) ||
286 key != t->parms.i_key ||
287 !(t->dev->flags & IFF_UP))
288 continue;
289
290 if (t->dev->type != ARPHRD_IPGRE &&
291 t->dev->type != dev_type)
292 continue;
293
294 score = 0;
295 if (t->parms.link != link)
296 score |= 1;
297 if (t->dev->type != dev_type)
298 score |= 2;
299 if (score == 0)
300 return t;
301
302 if (score < cand_score) {
303 cand = t;
304 cand_score = score;
305 }
306 }
307
308 for_each_ip_tunnel_rcu(ign->tunnels_wc[h1]) {
309 if (t->parms.i_key != key ||
310 !(t->dev->flags & IFF_UP))
311 continue;
312
313 if (t->dev->type != ARPHRD_IPGRE &&
314 t->dev->type != dev_type)
315 continue;
316
317 score = 0;
318 if (t->parms.link != link)
319 score |= 1;
320 if (t->dev->type != dev_type)
321 score |= 2;
322 if (score == 0)
323 return t;
324
325 if (score < cand_score) {
326 cand = t;
327 cand_score = score;
328 }
329 }
330
331 if (cand != NULL)
332 return cand;
333
334 dev = ign->fb_tunnel_dev;
335 if (dev->flags & IFF_UP)
336 return netdev_priv(dev);
337
338 return NULL;
339}
340
341static struct ip_tunnel __rcu **__ipgre_bucket(struct ipgre_net *ign,
342 struct ip_tunnel_parm *parms)
343{
344 __be32 remote = parms->iph.daddr;
345 __be32 local = parms->iph.saddr;
346 __be32 key = parms->i_key;
347 unsigned int h = HASH(key);
348 int prio = 0;
349
350 if (local)
351 prio |= 1;
352 if (remote && !ipv4_is_multicast(remote)) {
353 prio |= 2;
354 h ^= HASH(remote);
355 }
356
357 return &ign->tunnels[prio][h];
358}
359
360static inline struct ip_tunnel __rcu **ipgre_bucket(struct ipgre_net *ign,
361 struct ip_tunnel *t)
362{
363 return __ipgre_bucket(ign, &t->parms);
364}
365
366static void ipgre_tunnel_link(struct ipgre_net *ign, struct ip_tunnel *t)
367{
368 struct ip_tunnel __rcu **tp = ipgre_bucket(ign, t);
369
370 rcu_assign_pointer(t->next, rtnl_dereference(*tp));
371 rcu_assign_pointer(*tp, t);
372}
373
374static void ipgre_tunnel_unlink(struct ipgre_net *ign, struct ip_tunnel *t)
375{
376 struct ip_tunnel __rcu **tp;
377 struct ip_tunnel *iter;
378
379 for (tp = ipgre_bucket(ign, t);
380 (iter = rtnl_dereference(*tp)) != NULL;
381 tp = &iter->next) {
382 if (t == iter) {
383 rcu_assign_pointer(*tp, t->next);
384 break;
385 }
386 }
387}
388
389static struct ip_tunnel *ipgre_tunnel_find(struct net *net,
390 struct ip_tunnel_parm *parms,
391 int type)
392{
393 __be32 remote = parms->iph.daddr;
394 __be32 local = parms->iph.saddr;
395 __be32 key = parms->i_key;
396 int link = parms->link;
397 struct ip_tunnel *t;
398 struct ip_tunnel __rcu **tp;
399 struct ipgre_net *ign = net_generic(net, ipgre_net_id);
400
401 for (tp = __ipgre_bucket(ign, parms);
402 (t = rtnl_dereference(*tp)) != NULL;
403 tp = &t->next)
404 if (local == t->parms.iph.saddr &&
405 remote == t->parms.iph.daddr &&
406 key == t->parms.i_key &&
407 link == t->parms.link &&
408 type == t->dev->type)
409 break;
410
411 return t;
412}
413
414static struct ip_tunnel *ipgre_tunnel_locate(struct net *net,
415 struct ip_tunnel_parm *parms, int create)
416{
417 struct ip_tunnel *t, *nt;
418 struct net_device *dev;
419 char name[IFNAMSIZ];
420 struct ipgre_net *ign = net_generic(net, ipgre_net_id);
421
422 t = ipgre_tunnel_find(net, parms, ARPHRD_IPGRE);
423 if (t || !create)
424 return t;
425
426 if (parms->name[0])
427 strlcpy(name, parms->name, IFNAMSIZ);
428 else
429 strcpy(name, "gre%d");
430
431 dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup);
432 if (!dev)
433 return NULL;
434
435 dev_net_set(dev, net);
436
437 nt = netdev_priv(dev);
438 nt->parms = *parms;
439 dev->rtnl_link_ops = &ipgre_link_ops;
440
441 dev->mtu = ipgre_tunnel_bind_dev(dev);
442
443 if (register_netdevice(dev) < 0)
444 goto failed_free;
445
446 /* Can use a lockless transmit, unless we generate output sequences */
447 if (!(nt->parms.o_flags & GRE_SEQ))
448 dev->features |= NETIF_F_LLTX;
449
450 dev_hold(dev);
451 ipgre_tunnel_link(ign, nt);
452 return nt;
453
454failed_free:
455 free_netdev(dev);
456 return NULL;
457}
458
459static void ipgre_tunnel_uninit(struct net_device *dev)
460{
461 struct net *net = dev_net(dev);
462 struct ipgre_net *ign = net_generic(net, ipgre_net_id);
463
464 ipgre_tunnel_unlink(ign, netdev_priv(dev));
465 dev_put(dev);
466}
467
468
469static void ipgre_err(struct sk_buff *skb, u32 info)
470{
471
472/* All the routers (except for Linux) return only
473 8 bytes of packet payload. It means, that precise relaying of
474 ICMP in the real Internet is absolutely infeasible.
475
476 Moreover, Cisco "wise men" put GRE key to the third word
477 in GRE header. It makes impossible maintaining even soft state for keyed
478 GRE tunnels with enabled checksum. Tell them "thank you".
479
480 Well, I wonder, rfc1812 was written by Cisco employee,
481 what the hell these idiots break standards established
482 by themselves???
483 */
484
485 const struct iphdr *iph = (const struct iphdr *)skb->data;
486 __be16 *p = (__be16 *)(skb->data+(iph->ihl<<2));
487 int grehlen = (iph->ihl<<2) + 4;
488 const int type = icmp_hdr(skb)->type;
489 const int code = icmp_hdr(skb)->code;
490 struct ip_tunnel *t;
491 __be16 flags;
492
493 flags = p[0];
494 if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
495 if (flags&(GRE_VERSION|GRE_ROUTING))
496 return;
497 if (flags&GRE_KEY) {
498 grehlen += 4;
499 if (flags&GRE_CSUM)
500 grehlen += 4;
501 }
502 }
503
504 /* If only 8 bytes returned, keyed message will be dropped here */
505 if (skb_headlen(skb) < grehlen)
506 return;
507
508 switch (type) {
509 default:
510 case ICMP_PARAMETERPROB:
511 return;
512
513 case ICMP_DEST_UNREACH:
514 switch (code) {
515 case ICMP_SR_FAILED:
516 case ICMP_PORT_UNREACH:
517 /* Impossible event. */
518 return;
519 case ICMP_FRAG_NEEDED:
520 /* Soft state for pmtu is maintained by IP core. */
521 return;
522 default:
523 /* All others are translated to HOST_UNREACH.
524 rfc2003 contains "deep thoughts" about NET_UNREACH,
525 I believe they are just ether pollution. --ANK
526 */
527 break;
528 }
529 break;
530 case ICMP_TIME_EXCEEDED:
531 if (code != ICMP_EXC_TTL)
532 return;
533 break;
534 }
535
536 rcu_read_lock();
537 t = ipgre_tunnel_lookup(skb->dev, iph->daddr, iph->saddr,
538 flags & GRE_KEY ?
539 *(((__be32 *)p) + (grehlen / 4) - 1) : 0,
540 p[1]);
541 if (t == NULL || t->parms.iph.daddr == 0 ||
542 ipv4_is_multicast(t->parms.iph.daddr))
543 goto out;
544
545 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
546 goto out;
547
548 if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
549 t->err_count++;
550 else
551 t->err_count = 1;
552 t->err_time = jiffies;
553out:
554 rcu_read_unlock();
555}
556
557static inline void ipgre_ecn_decapsulate(const struct iphdr *iph, struct sk_buff *skb)
558{
559 if (INET_ECN_is_ce(iph->tos)) {
560 if (skb->protocol == htons(ETH_P_IP)) {
561 IP_ECN_set_ce(ip_hdr(skb));
562 } else if (skb->protocol == htons(ETH_P_IPV6)) {
563 IP6_ECN_set_ce(ipv6_hdr(skb));
564 }
565 }
566}
567
568static inline u8
569ipgre_ecn_encapsulate(u8 tos, const struct iphdr *old_iph, struct sk_buff *skb)
570{
571 u8 inner = 0;
572 if (skb->protocol == htons(ETH_P_IP))
573 inner = old_iph->tos;
574 else if (skb->protocol == htons(ETH_P_IPV6))
575 inner = ipv6_get_dsfield((const struct ipv6hdr *)old_iph);
576 return INET_ECN_encapsulate(tos, inner);
577}
578
579static int ipgre_rcv(struct sk_buff *skb)
580{
581 const struct iphdr *iph;
582 u8 *h;
583 __be16 flags;
584 __sum16 csum = 0;
585 __be32 key = 0;
586 u32 seqno = 0;
587 struct ip_tunnel *tunnel;
588 int offset = 4;
589 __be16 gre_proto;
590
591 if (!pskb_may_pull(skb, 16))
592 goto drop_nolock;
593
594 iph = ip_hdr(skb);
595 h = skb->data;
596 flags = *(__be16 *)h;
597
598 if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
599 /* - Version must be 0.
600 - We do not support routing headers.
601 */
602 if (flags&(GRE_VERSION|GRE_ROUTING))
603 goto drop_nolock;
604
605 if (flags&GRE_CSUM) {
606 switch (skb->ip_summed) {
607 case CHECKSUM_COMPLETE:
608 csum = csum_fold(skb->csum);
609 if (!csum)
610 break;
611 /* fall through */
612 case CHECKSUM_NONE:
613 skb->csum = 0;
614 csum = __skb_checksum_complete(skb);
615 skb->ip_summed = CHECKSUM_COMPLETE;
616 }
617 offset += 4;
618 }
619 if (flags&GRE_KEY) {
620 key = *(__be32 *)(h + offset);
621 offset += 4;
622 }
623 if (flags&GRE_SEQ) {
624 seqno = ntohl(*(__be32 *)(h + offset));
625 offset += 4;
626 }
627 }
628
629 gre_proto = *(__be16 *)(h + 2);
630
631 rcu_read_lock();
632 if ((tunnel = ipgre_tunnel_lookup(skb->dev,
633 iph->saddr, iph->daddr, key,
634 gre_proto))) {
635 struct pcpu_tstats *tstats;
636
637 secpath_reset(skb);
638
639 skb->protocol = gre_proto;
640 /* WCCP version 1 and 2 protocol decoding.
641 * - Change protocol to IP
642 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
643 */
644 if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
645 skb->protocol = htons(ETH_P_IP);
646 if ((*(h + offset) & 0xF0) != 0x40)
647 offset += 4;
648 }
649
650 skb->mac_header = skb->network_header;
651 __pskb_pull(skb, offset);
652 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
653 skb->pkt_type = PACKET_HOST;
654#ifdef CONFIG_NET_IPGRE_BROADCAST
655 if (ipv4_is_multicast(iph->daddr)) {
656 /* Looped back packet, drop it! */
657 if (rt_is_output_route(skb_rtable(skb)))
658 goto drop;
659 tunnel->dev->stats.multicast++;
660 skb->pkt_type = PACKET_BROADCAST;
661 }
662#endif
663
664 if (((flags&GRE_CSUM) && csum) ||
665 (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
666 tunnel->dev->stats.rx_crc_errors++;
667 tunnel->dev->stats.rx_errors++;
668 goto drop;
669 }
670 if (tunnel->parms.i_flags&GRE_SEQ) {
671 if (!(flags&GRE_SEQ) ||
672 (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
673 tunnel->dev->stats.rx_fifo_errors++;
674 tunnel->dev->stats.rx_errors++;
675 goto drop;
676 }
677 tunnel->i_seqno = seqno + 1;
678 }
679
680 /* Warning: All skb pointers will be invalidated! */
681 if (tunnel->dev->type == ARPHRD_ETHER) {
682 if (!pskb_may_pull(skb, ETH_HLEN)) {
683 tunnel->dev->stats.rx_length_errors++;
684 tunnel->dev->stats.rx_errors++;
685 goto drop;
686 }
687
688 iph = ip_hdr(skb);
689 skb->protocol = eth_type_trans(skb, tunnel->dev);
690 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
691 }
692
693 tstats = this_cpu_ptr(tunnel->dev->tstats);
694 u64_stats_update_begin(&tstats->syncp);
695 tstats->rx_packets++;
696 tstats->rx_bytes += skb->len;
697 u64_stats_update_end(&tstats->syncp);
698
699 __skb_tunnel_rx(skb, tunnel->dev);
700
701 skb_reset_network_header(skb);
702 ipgre_ecn_decapsulate(iph, skb);
703
704 netif_rx(skb);
705
706 rcu_read_unlock();
707 return 0;
708 }
709 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
710
711drop:
712 rcu_read_unlock();
713drop_nolock:
714 kfree_skb(skb);
715 return 0;
716}
717
718static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
719{
720 struct ip_tunnel *tunnel = netdev_priv(dev);
721 struct pcpu_tstats *tstats;
722 const struct iphdr *old_iph = ip_hdr(skb);
723 const struct iphdr *tiph;
724 struct flowi4 fl4;
725 u8 tos;
726 __be16 df;
727 struct rtable *rt; /* Route to the other host */
728 struct net_device *tdev; /* Device to other host */
729 struct iphdr *iph; /* Our new IP header */
730 unsigned int max_headroom; /* The extra header space needed */
731 int gre_hlen;
732 __be32 dst;
733 int mtu;
734
735 if (dev->type == ARPHRD_ETHER)
736 IPCB(skb)->flags = 0;
737
738 if (dev->header_ops && dev->type == ARPHRD_IPGRE) {
739 gre_hlen = 0;
740 tiph = (const struct iphdr *)skb->data;
741 } else {
742 gre_hlen = tunnel->hlen;
743 tiph = &tunnel->parms.iph;
744 }
745
746 if ((dst = tiph->daddr) == 0) {
747 /* NBMA tunnel */
748
749 if (skb_dst(skb) == NULL) {
750 dev->stats.tx_fifo_errors++;
751 goto tx_error;
752 }
753
754 if (skb->protocol == htons(ETH_P_IP)) {
755 rt = skb_rtable(skb);
756 dst = rt->rt_gateway;
757 }
758#if IS_ENABLED(CONFIG_IPV6)
759 else if (skb->protocol == htons(ETH_P_IPV6)) {
760 const struct in6_addr *addr6;
761 struct neighbour *neigh;
762 bool do_tx_error_icmp;
763 int addr_type;
764
765 neigh = dst_neigh_lookup(skb_dst(skb), &ipv6_hdr(skb)->daddr);
766 if (neigh == NULL)
767 goto tx_error;
768
769 addr6 = (const struct in6_addr *)&neigh->primary_key;
770 addr_type = ipv6_addr_type(addr6);
771
772 if (addr_type == IPV6_ADDR_ANY) {
773 addr6 = &ipv6_hdr(skb)->daddr;
774 addr_type = ipv6_addr_type(addr6);
775 }
776
777 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
778 do_tx_error_icmp = true;
779 else {
780 do_tx_error_icmp = false;
781 dst = addr6->s6_addr32[3];
782 }
783 neigh_release(neigh);
784 if (do_tx_error_icmp)
785 goto tx_error_icmp;
786 }
787#endif
788 else
789 goto tx_error;
790 }
791
792 tos = tiph->tos;
793 if (tos == 1) {
794 tos = 0;
795 if (skb->protocol == htons(ETH_P_IP))
796 tos = old_iph->tos;
797 else if (skb->protocol == htons(ETH_P_IPV6))
798 tos = ipv6_get_dsfield((const struct ipv6hdr *)old_iph);
799 }
800
801 rt = ip_route_output_gre(dev_net(dev), &fl4, dst, tiph->saddr,
802 tunnel->parms.o_key, RT_TOS(tos),
803 tunnel->parms.link);
804 if (IS_ERR(rt)) {
805 dev->stats.tx_carrier_errors++;
806 goto tx_error;
807 }
808 tdev = rt->dst.dev;
809
810 if (tdev == dev) {
811 ip_rt_put(rt);
812 dev->stats.collisions++;
813 goto tx_error;
814 }
815
816 df = tiph->frag_off;
817 if (df)
818 mtu = dst_mtu(&rt->dst) - dev->hard_header_len - tunnel->hlen;
819 else
820 mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
821
822 if (skb_dst(skb))
823 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
824
825 if (skb->protocol == htons(ETH_P_IP)) {
826 df |= (old_iph->frag_off&htons(IP_DF));
827
828 if ((old_iph->frag_off&htons(IP_DF)) &&
829 mtu < ntohs(old_iph->tot_len)) {
830 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
831 ip_rt_put(rt);
832 goto tx_error;
833 }
834 }
835#if IS_ENABLED(CONFIG_IPV6)
836 else if (skb->protocol == htons(ETH_P_IPV6)) {
837 struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
838
839 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && mtu >= IPV6_MIN_MTU) {
840 if ((tunnel->parms.iph.daddr &&
841 !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
842 rt6->rt6i_dst.plen == 128) {
843 rt6->rt6i_flags |= RTF_MODIFIED;
844 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
845 }
846 }
847
848 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
849 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
850 ip_rt_put(rt);
851 goto tx_error;
852 }
853 }
854#endif
855
856 if (tunnel->err_count > 0) {
857 if (time_before(jiffies,
858 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
859 tunnel->err_count--;
860
861 dst_link_failure(skb);
862 } else
863 tunnel->err_count = 0;
864 }
865
866 max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen + rt->dst.header_len;
867
868 if (skb_headroom(skb) < max_headroom || skb_shared(skb)||
869 (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
870 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
871 if (max_headroom > dev->needed_headroom)
872 dev->needed_headroom = max_headroom;
873 if (!new_skb) {
874 ip_rt_put(rt);
875 dev->stats.tx_dropped++;
876 dev_kfree_skb(skb);
877 return NETDEV_TX_OK;
878 }
879 if (skb->sk)
880 skb_set_owner_w(new_skb, skb->sk);
881 dev_kfree_skb(skb);
882 skb = new_skb;
883 old_iph = ip_hdr(skb);
884 }
885
886 skb_reset_transport_header(skb);
887 skb_push(skb, gre_hlen);
888 skb_reset_network_header(skb);
889 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
890 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
891 IPSKB_REROUTED);
892 skb_dst_drop(skb);
893 skb_dst_set(skb, &rt->dst);
894
895 /*
896 * Push down and install the IPIP header.
897 */
898
899 iph = ip_hdr(skb);
900 iph->version = 4;
901 iph->ihl = sizeof(struct iphdr) >> 2;
902 iph->frag_off = df;
903 iph->protocol = IPPROTO_GRE;
904 iph->tos = ipgre_ecn_encapsulate(tos, old_iph, skb);
905 iph->daddr = fl4.daddr;
906 iph->saddr = fl4.saddr;
907
908 if ((iph->ttl = tiph->ttl) == 0) {
909 if (skb->protocol == htons(ETH_P_IP))
910 iph->ttl = old_iph->ttl;
911#if IS_ENABLED(CONFIG_IPV6)
912 else if (skb->protocol == htons(ETH_P_IPV6))
913 iph->ttl = ((const struct ipv6hdr *)old_iph)->hop_limit;
914#endif
915 else
916 iph->ttl = ip4_dst_hoplimit(&rt->dst);
917 }
918
919 ((__be16 *)(iph + 1))[0] = tunnel->parms.o_flags;
920 ((__be16 *)(iph + 1))[1] = (dev->type == ARPHRD_ETHER) ?
921 htons(ETH_P_TEB) : skb->protocol;
922
923 if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
924 __be32 *ptr = (__be32 *)(((u8 *)iph) + tunnel->hlen - 4);
925
926 if (tunnel->parms.o_flags&GRE_SEQ) {
927 ++tunnel->o_seqno;
928 *ptr = htonl(tunnel->o_seqno);
929 ptr--;
930 }
931 if (tunnel->parms.o_flags&GRE_KEY) {
932 *ptr = tunnel->parms.o_key;
933 ptr--;
934 }
935 if (tunnel->parms.o_flags&GRE_CSUM) {
936 *ptr = 0;
937 *(__sum16 *)ptr = ip_compute_csum((void *)(iph+1), skb->len - sizeof(struct iphdr));
938 }
939 }
940
941 nf_reset(skb);
942 tstats = this_cpu_ptr(dev->tstats);
943 __IPTUNNEL_XMIT(tstats, &dev->stats);
944 return NETDEV_TX_OK;
945
946#if IS_ENABLED(CONFIG_IPV6)
947tx_error_icmp:
948 dst_link_failure(skb);
949#endif
950tx_error:
951 dev->stats.tx_errors++;
952 dev_kfree_skb(skb);
953 return NETDEV_TX_OK;
954}
955
956static int ipgre_tunnel_bind_dev(struct net_device *dev)
957{
958 struct net_device *tdev = NULL;
959 struct ip_tunnel *tunnel;
960 const struct iphdr *iph;
961 int hlen = LL_MAX_HEADER;
962 int mtu = ETH_DATA_LEN;
963 int addend = sizeof(struct iphdr) + 4;
964
965 tunnel = netdev_priv(dev);
966 iph = &tunnel->parms.iph;
967
968 /* Guess output device to choose reasonable mtu and needed_headroom */
969
970 if (iph->daddr) {
971 struct flowi4 fl4;
972 struct rtable *rt;
973
974 rt = ip_route_output_gre(dev_net(dev), &fl4,
975 iph->daddr, iph->saddr,
976 tunnel->parms.o_key,
977 RT_TOS(iph->tos),
978 tunnel->parms.link);
979 if (!IS_ERR(rt)) {
980 tdev = rt->dst.dev;
981 ip_rt_put(rt);
982 }
983
984 if (dev->type != ARPHRD_ETHER)
985 dev->flags |= IFF_POINTOPOINT;
986 }
987
988 if (!tdev && tunnel->parms.link)
989 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
990
991 if (tdev) {
992 hlen = tdev->hard_header_len + tdev->needed_headroom;
993 mtu = tdev->mtu;
994 }
995 dev->iflink = tunnel->parms.link;
996
997 /* Precalculate GRE options length */
998 if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
999 if (tunnel->parms.o_flags&GRE_CSUM)
1000 addend += 4;
1001 if (tunnel->parms.o_flags&GRE_KEY)
1002 addend += 4;
1003 if (tunnel->parms.o_flags&GRE_SEQ)
1004 addend += 4;
1005 }
1006 dev->needed_headroom = addend + hlen;
1007 mtu -= dev->hard_header_len + addend;
1008
1009 if (mtu < 68)
1010 mtu = 68;
1011
1012 tunnel->hlen = addend;
1013
1014 return mtu;
1015}
1016
1017static int
1018ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
1019{
1020 int err = 0;
1021 struct ip_tunnel_parm p;
1022 struct ip_tunnel *t;
1023 struct net *net = dev_net(dev);
1024 struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1025
1026 switch (cmd) {
1027 case SIOCGETTUNNEL:
1028 t = NULL;
1029 if (dev == ign->fb_tunnel_dev) {
1030 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1031 err = -EFAULT;
1032 break;
1033 }
1034 t = ipgre_tunnel_locate(net, &p, 0);
1035 }
1036 if (t == NULL)
1037 t = netdev_priv(dev);
1038 memcpy(&p, &t->parms, sizeof(p));
1039 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1040 err = -EFAULT;
1041 break;
1042
1043 case SIOCADDTUNNEL:
1044 case SIOCCHGTUNNEL:
1045 err = -EPERM;
1046 if (!capable(CAP_NET_ADMIN))
1047 goto done;
1048
1049 err = -EFAULT;
1050 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1051 goto done;
1052
1053 err = -EINVAL;
1054 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
1055 p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
1056 ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
1057 goto done;
1058 if (p.iph.ttl)
1059 p.iph.frag_off |= htons(IP_DF);
1060
1061 if (!(p.i_flags&GRE_KEY))
1062 p.i_key = 0;
1063 if (!(p.o_flags&GRE_KEY))
1064 p.o_key = 0;
1065
1066 t = ipgre_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
1067
1068 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1069 if (t != NULL) {
1070 if (t->dev != dev) {
1071 err = -EEXIST;
1072 break;
1073 }
1074 } else {
1075 unsigned int nflags = 0;
1076
1077 t = netdev_priv(dev);
1078
1079 if (ipv4_is_multicast(p.iph.daddr))
1080 nflags = IFF_BROADCAST;
1081 else if (p.iph.daddr)
1082 nflags = IFF_POINTOPOINT;
1083
1084 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
1085 err = -EINVAL;
1086 break;
1087 }
1088 ipgre_tunnel_unlink(ign, t);
1089 synchronize_net();
1090 t->parms.iph.saddr = p.iph.saddr;
1091 t->parms.iph.daddr = p.iph.daddr;
1092 t->parms.i_key = p.i_key;
1093 t->parms.o_key = p.o_key;
1094 memcpy(dev->dev_addr, &p.iph.saddr, 4);
1095 memcpy(dev->broadcast, &p.iph.daddr, 4);
1096 ipgre_tunnel_link(ign, t);
1097 netdev_state_change(dev);
1098 }
1099 }
1100
1101 if (t) {
1102 err = 0;
1103 if (cmd == SIOCCHGTUNNEL) {
1104 t->parms.iph.ttl = p.iph.ttl;
1105 t->parms.iph.tos = p.iph.tos;
1106 t->parms.iph.frag_off = p.iph.frag_off;
1107 if (t->parms.link != p.link) {
1108 t->parms.link = p.link;
1109 dev->mtu = ipgre_tunnel_bind_dev(dev);
1110 netdev_state_change(dev);
1111 }
1112 }
1113 if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
1114 err = -EFAULT;
1115 } else
1116 err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1117 break;
1118
1119 case SIOCDELTUNNEL:
1120 err = -EPERM;
1121 if (!capable(CAP_NET_ADMIN))
1122 goto done;
1123
1124 if (dev == ign->fb_tunnel_dev) {
1125 err = -EFAULT;
1126 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1127 goto done;
1128 err = -ENOENT;
1129 if ((t = ipgre_tunnel_locate(net, &p, 0)) == NULL)
1130 goto done;
1131 err = -EPERM;
1132 if (t == netdev_priv(ign->fb_tunnel_dev))
1133 goto done;
1134 dev = t->dev;
1135 }
1136 unregister_netdevice(dev);
1137 err = 0;
1138 break;
1139
1140 default:
1141 err = -EINVAL;
1142 }
1143
1144done:
1145 return err;
1146}
1147
1148static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1149{
1150 struct ip_tunnel *tunnel = netdev_priv(dev);
1151 if (new_mtu < 68 ||
1152 new_mtu > 0xFFF8 - dev->hard_header_len - tunnel->hlen)
1153 return -EINVAL;
1154 dev->mtu = new_mtu;
1155 return 0;
1156}
1157
1158/* Nice toy. Unfortunately, useless in real life :-)
1159 It allows to construct virtual multiprotocol broadcast "LAN"
1160 over the Internet, provided multicast routing is tuned.
1161
1162
1163 I have no idea was this bicycle invented before me,
1164 so that I had to set ARPHRD_IPGRE to a random value.
1165 I have an impression, that Cisco could make something similar,
1166 but this feature is apparently missing in IOS<=11.2(8).
1167
1168 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1169 with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1170
1171 ping -t 255 224.66.66.66
1172
1173 If nobody answers, mbone does not work.
1174
1175 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1176 ip addr add 10.66.66.<somewhat>/24 dev Universe
1177 ifconfig Universe up
1178 ifconfig Universe add fe80::<Your_real_addr>/10
1179 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1180 ftp 10.66.66.66
1181 ...
1182 ftp fec0:6666:6666::193.233.7.65
1183 ...
1184
1185 */
1186
1187static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
1188 unsigned short type,
1189 const void *daddr, const void *saddr, unsigned int len)
1190{
1191 struct ip_tunnel *t = netdev_priv(dev);
1192 struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1193 __be16 *p = (__be16 *)(iph+1);
1194
1195 memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1196 p[0] = t->parms.o_flags;
1197 p[1] = htons(type);
1198
1199 /*
1200 * Set the source hardware address.
1201 */
1202
1203 if (saddr)
1204 memcpy(&iph->saddr, saddr, 4);
1205 if (daddr)
1206 memcpy(&iph->daddr, daddr, 4);
1207 if (iph->daddr)
1208 return t->hlen;
1209
1210 return -t->hlen;
1211}
1212
1213static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
1214{
1215 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
1216 memcpy(haddr, &iph->saddr, 4);
1217 return 4;
1218}
1219
1220static const struct header_ops ipgre_header_ops = {
1221 .create = ipgre_header,
1222 .parse = ipgre_header_parse,
1223};
1224
1225#ifdef CONFIG_NET_IPGRE_BROADCAST
1226static int ipgre_open(struct net_device *dev)
1227{
1228 struct ip_tunnel *t = netdev_priv(dev);
1229
1230 if (ipv4_is_multicast(t->parms.iph.daddr)) {
1231 struct flowi4 fl4;
1232 struct rtable *rt;
1233
1234 rt = ip_route_output_gre(dev_net(dev), &fl4,
1235 t->parms.iph.daddr,
1236 t->parms.iph.saddr,
1237 t->parms.o_key,
1238 RT_TOS(t->parms.iph.tos),
1239 t->parms.link);
1240 if (IS_ERR(rt))
1241 return -EADDRNOTAVAIL;
1242 dev = rt->dst.dev;
1243 ip_rt_put(rt);
1244 if (__in_dev_get_rtnl(dev) == NULL)
1245 return -EADDRNOTAVAIL;
1246 t->mlink = dev->ifindex;
1247 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1248 }
1249 return 0;
1250}
1251
1252static int ipgre_close(struct net_device *dev)
1253{
1254 struct ip_tunnel *t = netdev_priv(dev);
1255
1256 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
1257 struct in_device *in_dev;
1258 in_dev = inetdev_by_index(dev_net(dev), t->mlink);
1259 if (in_dev)
1260 ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1261 }
1262 return 0;
1263}
1264
1265#endif
1266
1267static const struct net_device_ops ipgre_netdev_ops = {
1268 .ndo_init = ipgre_tunnel_init,
1269 .ndo_uninit = ipgre_tunnel_uninit,
1270#ifdef CONFIG_NET_IPGRE_BROADCAST
1271 .ndo_open = ipgre_open,
1272 .ndo_stop = ipgre_close,
1273#endif
1274 .ndo_start_xmit = ipgre_tunnel_xmit,
1275 .ndo_do_ioctl = ipgre_tunnel_ioctl,
1276 .ndo_change_mtu = ipgre_tunnel_change_mtu,
1277 .ndo_get_stats64 = ipgre_get_stats64,
1278};
1279
1280static void ipgre_dev_free(struct net_device *dev)
1281{
1282 free_percpu(dev->tstats);
1283 free_netdev(dev);
1284}
1285
1286static void ipgre_tunnel_setup(struct net_device *dev)
1287{
1288 dev->netdev_ops = &ipgre_netdev_ops;
1289 dev->destructor = ipgre_dev_free;
1290
1291 dev->type = ARPHRD_IPGRE;
1292 dev->needed_headroom = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1293 dev->mtu = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1294 dev->flags = IFF_NOARP;
1295 dev->iflink = 0;
1296 dev->addr_len = 4;
1297 dev->features |= NETIF_F_NETNS_LOCAL;
1298 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1299}
1300
1301static int ipgre_tunnel_init(struct net_device *dev)
1302{
1303 struct ip_tunnel *tunnel;
1304 struct iphdr *iph;
1305
1306 tunnel = netdev_priv(dev);
1307 iph = &tunnel->parms.iph;
1308
1309 tunnel->dev = dev;
1310 strcpy(tunnel->parms.name, dev->name);
1311
1312 memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1313 memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1314
1315 if (iph->daddr) {
1316#ifdef CONFIG_NET_IPGRE_BROADCAST
1317 if (ipv4_is_multicast(iph->daddr)) {
1318 if (!iph->saddr)
1319 return -EINVAL;
1320 dev->flags = IFF_BROADCAST;
1321 dev->header_ops = &ipgre_header_ops;
1322 }
1323#endif
1324 } else
1325 dev->header_ops = &ipgre_header_ops;
1326
1327 dev->tstats = alloc_percpu(struct pcpu_tstats);
1328 if (!dev->tstats)
1329 return -ENOMEM;
1330
1331 return 0;
1332}
1333
1334static void ipgre_fb_tunnel_init(struct net_device *dev)
1335{
1336 struct ip_tunnel *tunnel = netdev_priv(dev);
1337 struct iphdr *iph = &tunnel->parms.iph;
1338
1339 tunnel->dev = dev;
1340 strcpy(tunnel->parms.name, dev->name);
1341
1342 iph->version = 4;
1343 iph->protocol = IPPROTO_GRE;
1344 iph->ihl = 5;
1345 tunnel->hlen = sizeof(struct iphdr) + 4;
1346
1347 dev_hold(dev);
1348}
1349
1350
1351static const struct gre_protocol ipgre_protocol = {
1352 .handler = ipgre_rcv,
1353 .err_handler = ipgre_err,
1354};
1355
1356static void ipgre_destroy_tunnels(struct ipgre_net *ign, struct list_head *head)
1357{
1358 int prio;
1359
1360 for (prio = 0; prio < 4; prio++) {
1361 int h;
1362 for (h = 0; h < HASH_SIZE; h++) {
1363 struct ip_tunnel *t;
1364
1365 t = rtnl_dereference(ign->tunnels[prio][h]);
1366
1367 while (t != NULL) {
1368 unregister_netdevice_queue(t->dev, head);
1369 t = rtnl_dereference(t->next);
1370 }
1371 }
1372 }
1373}
1374
1375static int __net_init ipgre_init_net(struct net *net)
1376{
1377 struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1378 int err;
1379
1380 ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1381 ipgre_tunnel_setup);
1382 if (!ign->fb_tunnel_dev) {
1383 err = -ENOMEM;
1384 goto err_alloc_dev;
1385 }
1386 dev_net_set(ign->fb_tunnel_dev, net);
1387
1388 ipgre_fb_tunnel_init(ign->fb_tunnel_dev);
1389 ign->fb_tunnel_dev->rtnl_link_ops = &ipgre_link_ops;
1390
1391 if ((err = register_netdev(ign->fb_tunnel_dev)))
1392 goto err_reg_dev;
1393
1394 rcu_assign_pointer(ign->tunnels_wc[0],
1395 netdev_priv(ign->fb_tunnel_dev));
1396 return 0;
1397
1398err_reg_dev:
1399 ipgre_dev_free(ign->fb_tunnel_dev);
1400err_alloc_dev:
1401 return err;
1402}
1403
1404static void __net_exit ipgre_exit_net(struct net *net)
1405{
1406 struct ipgre_net *ign;
1407 LIST_HEAD(list);
1408
1409 ign = net_generic(net, ipgre_net_id);
1410 rtnl_lock();
1411 ipgre_destroy_tunnels(ign, &list);
1412 unregister_netdevice_many(&list);
1413 rtnl_unlock();
1414}
1415
1416static struct pernet_operations ipgre_net_ops = {
1417 .init = ipgre_init_net,
1418 .exit = ipgre_exit_net,
1419 .id = &ipgre_net_id,
1420 .size = sizeof(struct ipgre_net),
1421};
1422
1423static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1424{
1425 __be16 flags;
1426
1427 if (!data)
1428 return 0;
1429
1430 flags = 0;
1431 if (data[IFLA_GRE_IFLAGS])
1432 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1433 if (data[IFLA_GRE_OFLAGS])
1434 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1435 if (flags & (GRE_VERSION|GRE_ROUTING))
1436 return -EINVAL;
1437
1438 return 0;
1439}
1440
1441static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1442{
1443 __be32 daddr;
1444
1445 if (tb[IFLA_ADDRESS]) {
1446 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1447 return -EINVAL;
1448 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1449 return -EADDRNOTAVAIL;
1450 }
1451
1452 if (!data)
1453 goto out;
1454
1455 if (data[IFLA_GRE_REMOTE]) {
1456 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1457 if (!daddr)
1458 return -EINVAL;
1459 }
1460
1461out:
1462 return ipgre_tunnel_validate(tb, data);
1463}
1464
1465static void ipgre_netlink_parms(struct nlattr *data[],
1466 struct ip_tunnel_parm *parms)
1467{
1468 memset(parms, 0, sizeof(*parms));
1469
1470 parms->iph.protocol = IPPROTO_GRE;
1471
1472 if (!data)
1473 return;
1474
1475 if (data[IFLA_GRE_LINK])
1476 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1477
1478 if (data[IFLA_GRE_IFLAGS])
1479 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1480
1481 if (data[IFLA_GRE_OFLAGS])
1482 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1483
1484 if (data[IFLA_GRE_IKEY])
1485 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1486
1487 if (data[IFLA_GRE_OKEY])
1488 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1489
1490 if (data[IFLA_GRE_LOCAL])
1491 parms->iph.saddr = nla_get_be32(data[IFLA_GRE_LOCAL]);
1492
1493 if (data[IFLA_GRE_REMOTE])
1494 parms->iph.daddr = nla_get_be32(data[IFLA_GRE_REMOTE]);
1495
1496 if (data[IFLA_GRE_TTL])
1497 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1498
1499 if (data[IFLA_GRE_TOS])
1500 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1501
1502 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC]))
1503 parms->iph.frag_off = htons(IP_DF);
1504}
1505
1506static int ipgre_tap_init(struct net_device *dev)
1507{
1508 struct ip_tunnel *tunnel;
1509
1510 tunnel = netdev_priv(dev);
1511
1512 tunnel->dev = dev;
1513 strcpy(tunnel->parms.name, dev->name);
1514
1515 ipgre_tunnel_bind_dev(dev);
1516
1517 dev->tstats = alloc_percpu(struct pcpu_tstats);
1518 if (!dev->tstats)
1519 return -ENOMEM;
1520
1521 return 0;
1522}
1523
1524static const struct net_device_ops ipgre_tap_netdev_ops = {
1525 .ndo_init = ipgre_tap_init,
1526 .ndo_uninit = ipgre_tunnel_uninit,
1527 .ndo_start_xmit = ipgre_tunnel_xmit,
1528 .ndo_set_mac_address = eth_mac_addr,
1529 .ndo_validate_addr = eth_validate_addr,
1530 .ndo_change_mtu = ipgre_tunnel_change_mtu,
1531 .ndo_get_stats64 = ipgre_get_stats64,
1532};
1533
1534static void ipgre_tap_setup(struct net_device *dev)
1535{
1536
1537 ether_setup(dev);
1538
1539 dev->netdev_ops = &ipgre_tap_netdev_ops;
1540 dev->destructor = ipgre_dev_free;
1541
1542 dev->iflink = 0;
1543 dev->features |= NETIF_F_NETNS_LOCAL;
1544}
1545
1546static int ipgre_newlink(struct net *src_net, struct net_device *dev, struct nlattr *tb[],
1547 struct nlattr *data[])
1548{
1549 struct ip_tunnel *nt;
1550 struct net *net = dev_net(dev);
1551 struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1552 int mtu;
1553 int err;
1554
1555 nt = netdev_priv(dev);
1556 ipgre_netlink_parms(data, &nt->parms);
1557
1558 if (ipgre_tunnel_find(net, &nt->parms, dev->type))
1559 return -EEXIST;
1560
1561 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1562 eth_hw_addr_random(dev);
1563
1564 mtu = ipgre_tunnel_bind_dev(dev);
1565 if (!tb[IFLA_MTU])
1566 dev->mtu = mtu;
1567
1568 /* Can use a lockless transmit, unless we generate output sequences */
1569 if (!(nt->parms.o_flags & GRE_SEQ))
1570 dev->features |= NETIF_F_LLTX;
1571
1572 err = register_netdevice(dev);
1573 if (err)
1574 goto out;
1575
1576 dev_hold(dev);
1577 ipgre_tunnel_link(ign, nt);
1578
1579out:
1580 return err;
1581}
1582
1583static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1584 struct nlattr *data[])
1585{
1586 struct ip_tunnel *t, *nt;
1587 struct net *net = dev_net(dev);
1588 struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1589 struct ip_tunnel_parm p;
1590 int mtu;
1591
1592 if (dev == ign->fb_tunnel_dev)
1593 return -EINVAL;
1594
1595 nt = netdev_priv(dev);
1596 ipgre_netlink_parms(data, &p);
1597
1598 t = ipgre_tunnel_locate(net, &p, 0);
1599
1600 if (t) {
1601 if (t->dev != dev)
1602 return -EEXIST;
1603 } else {
1604 t = nt;
1605
1606 if (dev->type != ARPHRD_ETHER) {
1607 unsigned int nflags = 0;
1608
1609 if (ipv4_is_multicast(p.iph.daddr))
1610 nflags = IFF_BROADCAST;
1611 else if (p.iph.daddr)
1612 nflags = IFF_POINTOPOINT;
1613
1614 if ((dev->flags ^ nflags) &
1615 (IFF_POINTOPOINT | IFF_BROADCAST))
1616 return -EINVAL;
1617 }
1618
1619 ipgre_tunnel_unlink(ign, t);
1620 t->parms.iph.saddr = p.iph.saddr;
1621 t->parms.iph.daddr = p.iph.daddr;
1622 t->parms.i_key = p.i_key;
1623 if (dev->type != ARPHRD_ETHER) {
1624 memcpy(dev->dev_addr, &p.iph.saddr, 4);
1625 memcpy(dev->broadcast, &p.iph.daddr, 4);
1626 }
1627 ipgre_tunnel_link(ign, t);
1628 netdev_state_change(dev);
1629 }
1630
1631 t->parms.o_key = p.o_key;
1632 t->parms.iph.ttl = p.iph.ttl;
1633 t->parms.iph.tos = p.iph.tos;
1634 t->parms.iph.frag_off = p.iph.frag_off;
1635
1636 if (t->parms.link != p.link) {
1637 t->parms.link = p.link;
1638 mtu = ipgre_tunnel_bind_dev(dev);
1639 if (!tb[IFLA_MTU])
1640 dev->mtu = mtu;
1641 netdev_state_change(dev);
1642 }
1643
1644 return 0;
1645}
1646
1647static size_t ipgre_get_size(const struct net_device *dev)
1648{
1649 return
1650 /* IFLA_GRE_LINK */
1651 nla_total_size(4) +
1652 /* IFLA_GRE_IFLAGS */
1653 nla_total_size(2) +
1654 /* IFLA_GRE_OFLAGS */
1655 nla_total_size(2) +
1656 /* IFLA_GRE_IKEY */
1657 nla_total_size(4) +
1658 /* IFLA_GRE_OKEY */
1659 nla_total_size(4) +
1660 /* IFLA_GRE_LOCAL */
1661 nla_total_size(4) +
1662 /* IFLA_GRE_REMOTE */
1663 nla_total_size(4) +
1664 /* IFLA_GRE_TTL */
1665 nla_total_size(1) +
1666 /* IFLA_GRE_TOS */
1667 nla_total_size(1) +
1668 /* IFLA_GRE_PMTUDISC */
1669 nla_total_size(1) +
1670 0;
1671}
1672
1673static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1674{
1675 struct ip_tunnel *t = netdev_priv(dev);
1676 struct ip_tunnel_parm *p = &t->parms;
1677
1678 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1679 nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1680 nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1681 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1682 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1683 nla_put_be32(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1684 nla_put_be32(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1685 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1686 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1687 nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1688 !!(p->iph.frag_off & htons(IP_DF))))
1689 goto nla_put_failure;
1690 return 0;
1691
1692nla_put_failure:
1693 return -EMSGSIZE;
1694}
1695
1696static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1697 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1698 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1699 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1700 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1701 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1702 [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1703 [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1704 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1705 [IFLA_GRE_TOS] = { .type = NLA_U8 },
1706 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 },
1707};
1708
1709static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1710 .kind = "gre",
1711 .maxtype = IFLA_GRE_MAX,
1712 .policy = ipgre_policy,
1713 .priv_size = sizeof(struct ip_tunnel),
1714 .setup = ipgre_tunnel_setup,
1715 .validate = ipgre_tunnel_validate,
1716 .newlink = ipgre_newlink,
1717 .changelink = ipgre_changelink,
1718 .get_size = ipgre_get_size,
1719 .fill_info = ipgre_fill_info,
1720};
1721
1722static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1723 .kind = "gretap",
1724 .maxtype = IFLA_GRE_MAX,
1725 .policy = ipgre_policy,
1726 .priv_size = sizeof(struct ip_tunnel),
1727 .setup = ipgre_tap_setup,
1728 .validate = ipgre_tap_validate,
1729 .newlink = ipgre_newlink,
1730 .changelink = ipgre_changelink,
1731 .get_size = ipgre_get_size,
1732 .fill_info = ipgre_fill_info,
1733};
1734
1735/*
1736 * And now the modules code and kernel interface.
1737 */
1738
1739static int __init ipgre_init(void)
1740{
1741 int err;
1742
1743 pr_info("GRE over IPv4 tunneling driver\n");
1744
1745 err = register_pernet_device(&ipgre_net_ops);
1746 if (err < 0)
1747 return err;
1748
1749 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1750 if (err < 0) {
1751 pr_info("%s: can't add protocol\n", __func__);
1752 goto add_proto_failed;
1753 }
1754
1755 err = rtnl_link_register(&ipgre_link_ops);
1756 if (err < 0)
1757 goto rtnl_link_failed;
1758
1759 err = rtnl_link_register(&ipgre_tap_ops);
1760 if (err < 0)
1761 goto tap_ops_failed;
1762
1763out:
1764 return err;
1765
1766tap_ops_failed:
1767 rtnl_link_unregister(&ipgre_link_ops);
1768rtnl_link_failed:
1769 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1770add_proto_failed:
1771 unregister_pernet_device(&ipgre_net_ops);
1772 goto out;
1773}
1774
1775static void __exit ipgre_fini(void)
1776{
1777 rtnl_link_unregister(&ipgre_tap_ops);
1778 rtnl_link_unregister(&ipgre_link_ops);
1779 if (gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0)
1780 pr_info("%s: can't remove protocol\n", __func__);
1781 unregister_pernet_device(&ipgre_net_ops);
1782}
1783
1784module_init(ipgre_init);
1785module_exit(ipgre_fini);
1786MODULE_LICENSE("GPL");
1787MODULE_ALIAS_RTNL_LINK("gre");
1788MODULE_ALIAS_RTNL_LINK("gretap");
1789MODULE_ALIAS_NETDEV("gre0");