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// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Linux NET3: GRE over IP protocol decoder.
4 *
5 * Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/capability.h>
11#include <linux/module.h>
12#include <linux/types.h>
13#include <linux/kernel.h>
14#include <linux/slab.h>
15#include <linux/uaccess.h>
16#include <linux/skbuff.h>
17#include <linux/netdevice.h>
18#include <linux/in.h>
19#include <linux/tcp.h>
20#include <linux/udp.h>
21#include <linux/if_arp.h>
22#include <linux/if_vlan.h>
23#include <linux/init.h>
24#include <linux/in6.h>
25#include <linux/inetdevice.h>
26#include <linux/igmp.h>
27#include <linux/netfilter_ipv4.h>
28#include <linux/etherdevice.h>
29#include <linux/if_ether.h>
30
31#include <net/sock.h>
32#include <net/ip.h>
33#include <net/icmp.h>
34#include <net/protocol.h>
35#include <net/ip_tunnels.h>
36#include <net/arp.h>
37#include <net/checksum.h>
38#include <net/dsfield.h>
39#include <net/inet_ecn.h>
40#include <net/xfrm.h>
41#include <net/net_namespace.h>
42#include <net/netns/generic.h>
43#include <net/rtnetlink.h>
44#include <net/gre.h>
45#include <net/dst_metadata.h>
46#include <net/erspan.h>
47
48/*
49 Problems & solutions
50 --------------------
51
52 1. The most important issue is detecting local dead loops.
53 They would cause complete host lockup in transmit, which
54 would be "resolved" by stack overflow or, if queueing is enabled,
55 with infinite looping in net_bh.
56
57 We cannot track such dead loops during route installation,
58 it is infeasible task. The most general solutions would be
59 to keep skb->encapsulation counter (sort of local ttl),
60 and silently drop packet when it expires. It is a good
61 solution, but it supposes maintaining new variable in ALL
62 skb, even if no tunneling is used.
63
64 Current solution: xmit_recursion breaks dead loops. This is a percpu
65 counter, since when we enter the first ndo_xmit(), cpu migration is
66 forbidden. We force an exit if this counter reaches RECURSION_LIMIT
67
68 2. Networking dead loops would not kill routers, but would really
69 kill network. IP hop limit plays role of "t->recursion" in this case,
70 if we copy it from packet being encapsulated to upper header.
71 It is very good solution, but it introduces two problems:
72
73 - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
74 do not work over tunnels.
75 - traceroute does not work. I planned to relay ICMP from tunnel,
76 so that this problem would be solved and traceroute output
77 would even more informative. This idea appeared to be wrong:
78 only Linux complies to rfc1812 now (yes, guys, Linux is the only
79 true router now :-)), all routers (at least, in neighbourhood of mine)
80 return only 8 bytes of payload. It is the end.
81
82 Hence, if we want that OSPF worked or traceroute said something reasonable,
83 we should search for another solution.
84
85 One of them is to parse packet trying to detect inner encapsulation
86 made by our node. It is difficult or even impossible, especially,
87 taking into account fragmentation. TO be short, ttl is not solution at all.
88
89 Current solution: The solution was UNEXPECTEDLY SIMPLE.
90 We force DF flag on tunnels with preconfigured hop limit,
91 that is ALL. :-) Well, it does not remove the problem completely,
92 but exponential growth of network traffic is changed to linear
93 (branches, that exceed pmtu are pruned) and tunnel mtu
94 rapidly degrades to value <68, where looping stops.
95 Yes, it is not good if there exists a router in the loop,
96 which does not force DF, even when encapsulating packets have DF set.
97 But it is not our problem! Nobody could accuse us, we made
98 all that we could make. Even if it is your gated who injected
99 fatal route to network, even if it were you who configured
100 fatal static route: you are innocent. :-)
101
102 Alexey Kuznetsov.
103 */
104
105static bool log_ecn_error = true;
106module_param(log_ecn_error, bool, 0644);
107MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
108
109static struct rtnl_link_ops ipgre_link_ops __read_mostly;
110static int ipgre_tunnel_init(struct net_device *dev);
111static void erspan_build_header(struct sk_buff *skb,
112 u32 id, u32 index,
113 bool truncate, bool is_ipv4);
114
115static unsigned int ipgre_net_id __read_mostly;
116static unsigned int gre_tap_net_id __read_mostly;
117static unsigned int erspan_net_id __read_mostly;
118
119static int ipgre_err(struct sk_buff *skb, u32 info,
120 const struct tnl_ptk_info *tpi)
121{
122
123 /* All the routers (except for Linux) return only
124 8 bytes of packet payload. It means, that precise relaying of
125 ICMP in the real Internet is absolutely infeasible.
126
127 Moreover, Cisco "wise men" put GRE key to the third word
128 in GRE header. It makes impossible maintaining even soft
129 state for keyed GRE tunnels with enabled checksum. Tell
130 them "thank you".
131
132 Well, I wonder, rfc1812 was written by Cisco employee,
133 what the hell these idiots break standards established
134 by themselves???
135 */
136 struct net *net = dev_net(skb->dev);
137 struct ip_tunnel_net *itn;
138 const struct iphdr *iph;
139 const int type = icmp_hdr(skb)->type;
140 const int code = icmp_hdr(skb)->code;
141 unsigned int data_len = 0;
142 struct ip_tunnel *t;
143
144 if (tpi->proto == htons(ETH_P_TEB))
145 itn = net_generic(net, gre_tap_net_id);
146 else if (tpi->proto == htons(ETH_P_ERSPAN) ||
147 tpi->proto == htons(ETH_P_ERSPAN2))
148 itn = net_generic(net, erspan_net_id);
149 else
150 itn = net_generic(net, ipgre_net_id);
151
152 iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
153 t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
154 iph->daddr, iph->saddr, tpi->key);
155
156 if (!t)
157 return -ENOENT;
158
159 switch (type) {
160 default:
161 case ICMP_PARAMETERPROB:
162 return 0;
163
164 case ICMP_DEST_UNREACH:
165 switch (code) {
166 case ICMP_SR_FAILED:
167 case ICMP_PORT_UNREACH:
168 /* Impossible event. */
169 return 0;
170 default:
171 /* All others are translated to HOST_UNREACH.
172 rfc2003 contains "deep thoughts" about NET_UNREACH,
173 I believe they are just ether pollution. --ANK
174 */
175 break;
176 }
177 break;
178
179 case ICMP_TIME_EXCEEDED:
180 if (code != ICMP_EXC_TTL)
181 return 0;
182 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
183 break;
184
185 case ICMP_REDIRECT:
186 break;
187 }
188
189#if IS_ENABLED(CONFIG_IPV6)
190 if (tpi->proto == htons(ETH_P_IPV6) &&
191 !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
192 type, data_len))
193 return 0;
194#endif
195
196 if (t->parms.iph.daddr == 0 ||
197 ipv4_is_multicast(t->parms.iph.daddr))
198 return 0;
199
200 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
201 return 0;
202
203 if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
204 t->err_count++;
205 else
206 t->err_count = 1;
207 t->err_time = jiffies;
208
209 return 0;
210}
211
212static void gre_err(struct sk_buff *skb, u32 info)
213{
214 /* All the routers (except for Linux) return only
215 * 8 bytes of packet payload. It means, that precise relaying of
216 * ICMP in the real Internet is absolutely infeasible.
217 *
218 * Moreover, Cisco "wise men" put GRE key to the third word
219 * in GRE header. It makes impossible maintaining even soft
220 * state for keyed
221 * GRE tunnels with enabled checksum. Tell them "thank you".
222 *
223 * Well, I wonder, rfc1812 was written by Cisco employee,
224 * what the hell these idiots break standards established
225 * by themselves???
226 */
227
228 const struct iphdr *iph = (struct iphdr *)skb->data;
229 const int type = icmp_hdr(skb)->type;
230 const int code = icmp_hdr(skb)->code;
231 struct tnl_ptk_info tpi;
232
233 if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP),
234 iph->ihl * 4) < 0)
235 return;
236
237 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
238 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
239 skb->dev->ifindex, IPPROTO_GRE);
240 return;
241 }
242 if (type == ICMP_REDIRECT) {
243 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex,
244 IPPROTO_GRE);
245 return;
246 }
247
248 ipgre_err(skb, info, &tpi);
249}
250
251static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
252 int gre_hdr_len)
253{
254 struct net *net = dev_net(skb->dev);
255 struct metadata_dst *tun_dst = NULL;
256 struct erspan_base_hdr *ershdr;
257 struct ip_tunnel_net *itn;
258 struct ip_tunnel *tunnel;
259 const struct iphdr *iph;
260 struct erspan_md2 *md2;
261 int ver;
262 int len;
263
264 itn = net_generic(net, erspan_net_id);
265
266 iph = ip_hdr(skb);
267 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
268 ver = ershdr->ver;
269
270 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
271 tpi->flags | TUNNEL_KEY,
272 iph->saddr, iph->daddr, tpi->key);
273
274 if (tunnel) {
275 len = gre_hdr_len + erspan_hdr_len(ver);
276 if (unlikely(!pskb_may_pull(skb, len)))
277 return PACKET_REJECT;
278
279 if (__iptunnel_pull_header(skb,
280 len,
281 htons(ETH_P_TEB),
282 false, false) < 0)
283 goto drop;
284
285 if (tunnel->collect_md) {
286 struct erspan_metadata *pkt_md, *md;
287 struct ip_tunnel_info *info;
288 unsigned char *gh;
289 __be64 tun_id;
290 __be16 flags;
291
292 tpi->flags |= TUNNEL_KEY;
293 flags = tpi->flags;
294 tun_id = key32_to_tunnel_id(tpi->key);
295
296 tun_dst = ip_tun_rx_dst(skb, flags,
297 tun_id, sizeof(*md));
298 if (!tun_dst)
299 return PACKET_REJECT;
300
301 /* skb can be uncloned in __iptunnel_pull_header, so
302 * old pkt_md is no longer valid and we need to reset
303 * it
304 */
305 gh = skb_network_header(skb) +
306 skb_network_header_len(skb);
307 pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len +
308 sizeof(*ershdr));
309 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
310 md->version = ver;
311 md2 = &md->u.md2;
312 memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
313 ERSPAN_V2_MDSIZE);
314
315 info = &tun_dst->u.tun_info;
316 info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
317 info->options_len = sizeof(*md);
318 }
319
320 skb_reset_mac_header(skb);
321 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
322 return PACKET_RCVD;
323 }
324 return PACKET_REJECT;
325
326drop:
327 kfree_skb(skb);
328 return PACKET_RCVD;
329}
330
331static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
332 struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
333{
334 struct metadata_dst *tun_dst = NULL;
335 const struct iphdr *iph;
336 struct ip_tunnel *tunnel;
337
338 iph = ip_hdr(skb);
339 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
340 iph->saddr, iph->daddr, tpi->key);
341
342 if (tunnel) {
343 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
344 raw_proto, false) < 0)
345 goto drop;
346
347 if (tunnel->dev->type != ARPHRD_NONE)
348 skb_pop_mac_header(skb);
349 else
350 skb_reset_mac_header(skb);
351 if (tunnel->collect_md) {
352 __be16 flags;
353 __be64 tun_id;
354
355 flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
356 tun_id = key32_to_tunnel_id(tpi->key);
357 tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
358 if (!tun_dst)
359 return PACKET_REJECT;
360 }
361
362 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
363 return PACKET_RCVD;
364 }
365 return PACKET_NEXT;
366
367drop:
368 kfree_skb(skb);
369 return PACKET_RCVD;
370}
371
372static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
373 int hdr_len)
374{
375 struct net *net = dev_net(skb->dev);
376 struct ip_tunnel_net *itn;
377 int res;
378
379 if (tpi->proto == htons(ETH_P_TEB))
380 itn = net_generic(net, gre_tap_net_id);
381 else
382 itn = net_generic(net, ipgre_net_id);
383
384 res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
385 if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
386 /* ipgre tunnels in collect metadata mode should receive
387 * also ETH_P_TEB traffic.
388 */
389 itn = net_generic(net, ipgre_net_id);
390 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
391 }
392 return res;
393}
394
395static int gre_rcv(struct sk_buff *skb)
396{
397 struct tnl_ptk_info tpi;
398 bool csum_err = false;
399 int hdr_len;
400
401#ifdef CONFIG_NET_IPGRE_BROADCAST
402 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
403 /* Looped back packet, drop it! */
404 if (rt_is_output_route(skb_rtable(skb)))
405 goto drop;
406 }
407#endif
408
409 hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
410 if (hdr_len < 0)
411 goto drop;
412
413 if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
414 tpi.proto == htons(ETH_P_ERSPAN2))) {
415 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
416 return 0;
417 goto out;
418 }
419
420 if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
421 return 0;
422
423out:
424 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
425drop:
426 kfree_skb(skb);
427 return 0;
428}
429
430static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
431 const struct iphdr *tnl_params,
432 __be16 proto)
433{
434 struct ip_tunnel *tunnel = netdev_priv(dev);
435
436 if (tunnel->parms.o_flags & TUNNEL_SEQ)
437 tunnel->o_seqno++;
438
439 /* Push GRE header. */
440 gre_build_header(skb, tunnel->tun_hlen,
441 tunnel->parms.o_flags, proto, tunnel->parms.o_key,
442 htonl(tunnel->o_seqno));
443
444 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
445}
446
447static int gre_handle_offloads(struct sk_buff *skb, bool csum)
448{
449 return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
450}
451
452static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
453 __be16 proto)
454{
455 struct ip_tunnel *tunnel = netdev_priv(dev);
456 struct ip_tunnel_info *tun_info;
457 const struct ip_tunnel_key *key;
458 int tunnel_hlen;
459 __be16 flags;
460
461 tun_info = skb_tunnel_info(skb);
462 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
463 ip_tunnel_info_af(tun_info) != AF_INET))
464 goto err_free_skb;
465
466 key = &tun_info->key;
467 tunnel_hlen = gre_calc_hlen(key->tun_flags);
468
469 if (skb_cow_head(skb, dev->needed_headroom))
470 goto err_free_skb;
471
472 /* Push Tunnel header. */
473 if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
474 goto err_free_skb;
475
476 flags = tun_info->key.tun_flags &
477 (TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ);
478 gre_build_header(skb, tunnel_hlen, flags, proto,
479 tunnel_id_to_key32(tun_info->key.tun_id),
480 (flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++) : 0);
481
482 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
483
484 return;
485
486err_free_skb:
487 kfree_skb(skb);
488 dev->stats.tx_dropped++;
489}
490
491static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev)
492{
493 struct ip_tunnel *tunnel = netdev_priv(dev);
494 struct ip_tunnel_info *tun_info;
495 const struct ip_tunnel_key *key;
496 struct erspan_metadata *md;
497 bool truncate = false;
498 __be16 proto;
499 int tunnel_hlen;
500 int version;
501 int nhoff;
502 int thoff;
503
504 tun_info = skb_tunnel_info(skb);
505 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
506 ip_tunnel_info_af(tun_info) != AF_INET))
507 goto err_free_skb;
508
509 key = &tun_info->key;
510 if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT))
511 goto err_free_skb;
512 if (tun_info->options_len < sizeof(*md))
513 goto err_free_skb;
514 md = ip_tunnel_info_opts(tun_info);
515
516 /* ERSPAN has fixed 8 byte GRE header */
517 version = md->version;
518 tunnel_hlen = 8 + erspan_hdr_len(version);
519
520 if (skb_cow_head(skb, dev->needed_headroom))
521 goto err_free_skb;
522
523 if (gre_handle_offloads(skb, false))
524 goto err_free_skb;
525
526 if (skb->len > dev->mtu + dev->hard_header_len) {
527 pskb_trim(skb, dev->mtu + dev->hard_header_len);
528 truncate = true;
529 }
530
531 nhoff = skb_network_header(skb) - skb_mac_header(skb);
532 if (skb->protocol == htons(ETH_P_IP) &&
533 (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
534 truncate = true;
535
536 thoff = skb_transport_header(skb) - skb_mac_header(skb);
537 if (skb->protocol == htons(ETH_P_IPV6) &&
538 (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff))
539 truncate = true;
540
541 if (version == 1) {
542 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
543 ntohl(md->u.index), truncate, true);
544 proto = htons(ETH_P_ERSPAN);
545 } else if (version == 2) {
546 erspan_build_header_v2(skb,
547 ntohl(tunnel_id_to_key32(key->tun_id)),
548 md->u.md2.dir,
549 get_hwid(&md->u.md2),
550 truncate, true);
551 proto = htons(ETH_P_ERSPAN2);
552 } else {
553 goto err_free_skb;
554 }
555
556 gre_build_header(skb, 8, TUNNEL_SEQ,
557 proto, 0, htonl(tunnel->o_seqno++));
558
559 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
560
561 return;
562
563err_free_skb:
564 kfree_skb(skb);
565 dev->stats.tx_dropped++;
566}
567
568static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
569{
570 struct ip_tunnel_info *info = skb_tunnel_info(skb);
571 const struct ip_tunnel_key *key;
572 struct rtable *rt;
573 struct flowi4 fl4;
574
575 if (ip_tunnel_info_af(info) != AF_INET)
576 return -EINVAL;
577
578 key = &info->key;
579 ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src,
580 tunnel_id_to_key32(key->tun_id), key->tos, 0,
581 skb->mark, skb_get_hash(skb));
582 rt = ip_route_output_key(dev_net(dev), &fl4);
583 if (IS_ERR(rt))
584 return PTR_ERR(rt);
585
586 ip_rt_put(rt);
587 info->key.u.ipv4.src = fl4.saddr;
588 return 0;
589}
590
591static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
592 struct net_device *dev)
593{
594 struct ip_tunnel *tunnel = netdev_priv(dev);
595 const struct iphdr *tnl_params;
596
597 if (!pskb_inet_may_pull(skb))
598 goto free_skb;
599
600 if (tunnel->collect_md) {
601 gre_fb_xmit(skb, dev, skb->protocol);
602 return NETDEV_TX_OK;
603 }
604
605 if (dev->header_ops) {
606 /* Need space for new headers */
607 if (skb_cow_head(skb, dev->needed_headroom -
608 (tunnel->hlen + sizeof(struct iphdr))))
609 goto free_skb;
610
611 tnl_params = (const struct iphdr *)skb->data;
612
613 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
614 * to gre header.
615 */
616 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
617 skb_reset_mac_header(skb);
618 } else {
619 if (skb_cow_head(skb, dev->needed_headroom))
620 goto free_skb;
621
622 tnl_params = &tunnel->parms.iph;
623 }
624
625 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
626 goto free_skb;
627
628 __gre_xmit(skb, dev, tnl_params, skb->protocol);
629 return NETDEV_TX_OK;
630
631free_skb:
632 kfree_skb(skb);
633 dev->stats.tx_dropped++;
634 return NETDEV_TX_OK;
635}
636
637static netdev_tx_t erspan_xmit(struct sk_buff *skb,
638 struct net_device *dev)
639{
640 struct ip_tunnel *tunnel = netdev_priv(dev);
641 bool truncate = false;
642 __be16 proto;
643
644 if (!pskb_inet_may_pull(skb))
645 goto free_skb;
646
647 if (tunnel->collect_md) {
648 erspan_fb_xmit(skb, dev);
649 return NETDEV_TX_OK;
650 }
651
652 if (gre_handle_offloads(skb, false))
653 goto free_skb;
654
655 if (skb_cow_head(skb, dev->needed_headroom))
656 goto free_skb;
657
658 if (skb->len > dev->mtu + dev->hard_header_len) {
659 pskb_trim(skb, dev->mtu + dev->hard_header_len);
660 truncate = true;
661 }
662
663 /* Push ERSPAN header */
664 if (tunnel->erspan_ver == 1) {
665 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
666 tunnel->index,
667 truncate, true);
668 proto = htons(ETH_P_ERSPAN);
669 } else if (tunnel->erspan_ver == 2) {
670 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
671 tunnel->dir, tunnel->hwid,
672 truncate, true);
673 proto = htons(ETH_P_ERSPAN2);
674 } else {
675 goto free_skb;
676 }
677
678 tunnel->parms.o_flags &= ~TUNNEL_KEY;
679 __gre_xmit(skb, dev, &tunnel->parms.iph, proto);
680 return NETDEV_TX_OK;
681
682free_skb:
683 kfree_skb(skb);
684 dev->stats.tx_dropped++;
685 return NETDEV_TX_OK;
686}
687
688static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
689 struct net_device *dev)
690{
691 struct ip_tunnel *tunnel = netdev_priv(dev);
692
693 if (!pskb_inet_may_pull(skb))
694 goto free_skb;
695
696 if (tunnel->collect_md) {
697 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
698 return NETDEV_TX_OK;
699 }
700
701 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
702 goto free_skb;
703
704 if (skb_cow_head(skb, dev->needed_headroom))
705 goto free_skb;
706
707 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
708 return NETDEV_TX_OK;
709
710free_skb:
711 kfree_skb(skb);
712 dev->stats.tx_dropped++;
713 return NETDEV_TX_OK;
714}
715
716static void ipgre_link_update(struct net_device *dev, bool set_mtu)
717{
718 struct ip_tunnel *tunnel = netdev_priv(dev);
719 int len;
720
721 len = tunnel->tun_hlen;
722 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
723 len = tunnel->tun_hlen - len;
724 tunnel->hlen = tunnel->hlen + len;
725
726 dev->needed_headroom = dev->needed_headroom + len;
727 if (set_mtu)
728 dev->mtu = max_t(int, dev->mtu - len, 68);
729
730 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
731 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
732 tunnel->encap.type == TUNNEL_ENCAP_NONE) {
733 dev->features |= NETIF_F_GSO_SOFTWARE;
734 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
735 } else {
736 dev->features &= ~NETIF_F_GSO_SOFTWARE;
737 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
738 }
739 dev->features |= NETIF_F_LLTX;
740 } else {
741 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
742 dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
743 }
744}
745
746static int ipgre_tunnel_ioctl(struct net_device *dev,
747 struct ifreq *ifr, int cmd)
748{
749 struct ip_tunnel_parm p;
750 int err;
751
752 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
753 return -EFAULT;
754
755 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
756 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
757 p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
758 ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
759 return -EINVAL;
760 }
761
762 p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
763 p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
764
765 err = ip_tunnel_ioctl(dev, &p, cmd);
766 if (err)
767 return err;
768
769 if (cmd == SIOCCHGTUNNEL) {
770 struct ip_tunnel *t = netdev_priv(dev);
771
772 t->parms.i_flags = p.i_flags;
773 t->parms.o_flags = p.o_flags;
774
775 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
776 ipgre_link_update(dev, true);
777 }
778
779 p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
780 p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
781
782 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
783 return -EFAULT;
784
785 return 0;
786}
787
788/* Nice toy. Unfortunately, useless in real life :-)
789 It allows to construct virtual multiprotocol broadcast "LAN"
790 over the Internet, provided multicast routing is tuned.
791
792
793 I have no idea was this bicycle invented before me,
794 so that I had to set ARPHRD_IPGRE to a random value.
795 I have an impression, that Cisco could make something similar,
796 but this feature is apparently missing in IOS<=11.2(8).
797
798 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
799 with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
800
801 ping -t 255 224.66.66.66
802
803 If nobody answers, mbone does not work.
804
805 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
806 ip addr add 10.66.66.<somewhat>/24 dev Universe
807 ifconfig Universe up
808 ifconfig Universe add fe80::<Your_real_addr>/10
809 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
810 ftp 10.66.66.66
811 ...
812 ftp fec0:6666:6666::193.233.7.65
813 ...
814 */
815static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
816 unsigned short type,
817 const void *daddr, const void *saddr, unsigned int len)
818{
819 struct ip_tunnel *t = netdev_priv(dev);
820 struct iphdr *iph;
821 struct gre_base_hdr *greh;
822
823 iph = skb_push(skb, t->hlen + sizeof(*iph));
824 greh = (struct gre_base_hdr *)(iph+1);
825 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
826 greh->protocol = htons(type);
827
828 memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
829
830 /* Set the source hardware address. */
831 if (saddr)
832 memcpy(&iph->saddr, saddr, 4);
833 if (daddr)
834 memcpy(&iph->daddr, daddr, 4);
835 if (iph->daddr)
836 return t->hlen + sizeof(*iph);
837
838 return -(t->hlen + sizeof(*iph));
839}
840
841static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
842{
843 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
844 memcpy(haddr, &iph->saddr, 4);
845 return 4;
846}
847
848static const struct header_ops ipgre_header_ops = {
849 .create = ipgre_header,
850 .parse = ipgre_header_parse,
851};
852
853#ifdef CONFIG_NET_IPGRE_BROADCAST
854static int ipgre_open(struct net_device *dev)
855{
856 struct ip_tunnel *t = netdev_priv(dev);
857
858 if (ipv4_is_multicast(t->parms.iph.daddr)) {
859 struct flowi4 fl4;
860 struct rtable *rt;
861
862 rt = ip_route_output_gre(t->net, &fl4,
863 t->parms.iph.daddr,
864 t->parms.iph.saddr,
865 t->parms.o_key,
866 RT_TOS(t->parms.iph.tos),
867 t->parms.link);
868 if (IS_ERR(rt))
869 return -EADDRNOTAVAIL;
870 dev = rt->dst.dev;
871 ip_rt_put(rt);
872 if (!__in_dev_get_rtnl(dev))
873 return -EADDRNOTAVAIL;
874 t->mlink = dev->ifindex;
875 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
876 }
877 return 0;
878}
879
880static int ipgre_close(struct net_device *dev)
881{
882 struct ip_tunnel *t = netdev_priv(dev);
883
884 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
885 struct in_device *in_dev;
886 in_dev = inetdev_by_index(t->net, t->mlink);
887 if (in_dev)
888 ip_mc_dec_group(in_dev, t->parms.iph.daddr);
889 }
890 return 0;
891}
892#endif
893
894static const struct net_device_ops ipgre_netdev_ops = {
895 .ndo_init = ipgre_tunnel_init,
896 .ndo_uninit = ip_tunnel_uninit,
897#ifdef CONFIG_NET_IPGRE_BROADCAST
898 .ndo_open = ipgre_open,
899 .ndo_stop = ipgre_close,
900#endif
901 .ndo_start_xmit = ipgre_xmit,
902 .ndo_do_ioctl = ipgre_tunnel_ioctl,
903 .ndo_change_mtu = ip_tunnel_change_mtu,
904 .ndo_get_stats64 = ip_tunnel_get_stats64,
905 .ndo_get_iflink = ip_tunnel_get_iflink,
906};
907
908#define GRE_FEATURES (NETIF_F_SG | \
909 NETIF_F_FRAGLIST | \
910 NETIF_F_HIGHDMA | \
911 NETIF_F_HW_CSUM)
912
913static void ipgre_tunnel_setup(struct net_device *dev)
914{
915 dev->netdev_ops = &ipgre_netdev_ops;
916 dev->type = ARPHRD_IPGRE;
917 ip_tunnel_setup(dev, ipgre_net_id);
918}
919
920static void __gre_tunnel_init(struct net_device *dev)
921{
922 struct ip_tunnel *tunnel;
923
924 tunnel = netdev_priv(dev);
925 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
926 tunnel->parms.iph.protocol = IPPROTO_GRE;
927
928 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
929
930 dev->features |= GRE_FEATURES;
931 dev->hw_features |= GRE_FEATURES;
932
933 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
934 /* TCP offload with GRE SEQ is not supported, nor
935 * can we support 2 levels of outer headers requiring
936 * an update.
937 */
938 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
939 (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
940 dev->features |= NETIF_F_GSO_SOFTWARE;
941 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
942 }
943
944 /* Can use a lockless transmit, unless we generate
945 * output sequences
946 */
947 dev->features |= NETIF_F_LLTX;
948 }
949}
950
951static int ipgre_tunnel_init(struct net_device *dev)
952{
953 struct ip_tunnel *tunnel = netdev_priv(dev);
954 struct iphdr *iph = &tunnel->parms.iph;
955
956 __gre_tunnel_init(dev);
957
958 memcpy(dev->dev_addr, &iph->saddr, 4);
959 memcpy(dev->broadcast, &iph->daddr, 4);
960
961 dev->flags = IFF_NOARP;
962 netif_keep_dst(dev);
963 dev->addr_len = 4;
964
965 if (iph->daddr && !tunnel->collect_md) {
966#ifdef CONFIG_NET_IPGRE_BROADCAST
967 if (ipv4_is_multicast(iph->daddr)) {
968 if (!iph->saddr)
969 return -EINVAL;
970 dev->flags = IFF_BROADCAST;
971 dev->header_ops = &ipgre_header_ops;
972 }
973#endif
974 } else if (!tunnel->collect_md) {
975 dev->header_ops = &ipgre_header_ops;
976 }
977
978 return ip_tunnel_init(dev);
979}
980
981static const struct gre_protocol ipgre_protocol = {
982 .handler = gre_rcv,
983 .err_handler = gre_err,
984};
985
986static int __net_init ipgre_init_net(struct net *net)
987{
988 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
989}
990
991static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
992{
993 ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
994}
995
996static struct pernet_operations ipgre_net_ops = {
997 .init = ipgre_init_net,
998 .exit_batch = ipgre_exit_batch_net,
999 .id = &ipgre_net_id,
1000 .size = sizeof(struct ip_tunnel_net),
1001};
1002
1003static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1004 struct netlink_ext_ack *extack)
1005{
1006 __be16 flags;
1007
1008 if (!data)
1009 return 0;
1010
1011 flags = 0;
1012 if (data[IFLA_GRE_IFLAGS])
1013 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1014 if (data[IFLA_GRE_OFLAGS])
1015 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1016 if (flags & (GRE_VERSION|GRE_ROUTING))
1017 return -EINVAL;
1018
1019 if (data[IFLA_GRE_COLLECT_METADATA] &&
1020 data[IFLA_GRE_ENCAP_TYPE] &&
1021 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1022 return -EINVAL;
1023
1024 return 0;
1025}
1026
1027static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1028 struct netlink_ext_ack *extack)
1029{
1030 __be32 daddr;
1031
1032 if (tb[IFLA_ADDRESS]) {
1033 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1034 return -EINVAL;
1035 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1036 return -EADDRNOTAVAIL;
1037 }
1038
1039 if (!data)
1040 goto out;
1041
1042 if (data[IFLA_GRE_REMOTE]) {
1043 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1044 if (!daddr)
1045 return -EINVAL;
1046 }
1047
1048out:
1049 return ipgre_tunnel_validate(tb, data, extack);
1050}
1051
1052static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1053 struct netlink_ext_ack *extack)
1054{
1055 __be16 flags = 0;
1056 int ret;
1057
1058 if (!data)
1059 return 0;
1060
1061 ret = ipgre_tap_validate(tb, data, extack);
1062 if (ret)
1063 return ret;
1064
1065 /* ERSPAN should only have GRE sequence and key flag */
1066 if (data[IFLA_GRE_OFLAGS])
1067 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1068 if (data[IFLA_GRE_IFLAGS])
1069 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1070 if (!data[IFLA_GRE_COLLECT_METADATA] &&
1071 flags != (GRE_SEQ | GRE_KEY))
1072 return -EINVAL;
1073
1074 /* ERSPAN Session ID only has 10-bit. Since we reuse
1075 * 32-bit key field as ID, check it's range.
1076 */
1077 if (data[IFLA_GRE_IKEY] &&
1078 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1079 return -EINVAL;
1080
1081 if (data[IFLA_GRE_OKEY] &&
1082 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1083 return -EINVAL;
1084
1085 return 0;
1086}
1087
1088static int ipgre_netlink_parms(struct net_device *dev,
1089 struct nlattr *data[],
1090 struct nlattr *tb[],
1091 struct ip_tunnel_parm *parms,
1092 __u32 *fwmark)
1093{
1094 struct ip_tunnel *t = netdev_priv(dev);
1095
1096 memset(parms, 0, sizeof(*parms));
1097
1098 parms->iph.protocol = IPPROTO_GRE;
1099
1100 if (!data)
1101 return 0;
1102
1103 if (data[IFLA_GRE_LINK])
1104 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1105
1106 if (data[IFLA_GRE_IFLAGS])
1107 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1108
1109 if (data[IFLA_GRE_OFLAGS])
1110 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1111
1112 if (data[IFLA_GRE_IKEY])
1113 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1114
1115 if (data[IFLA_GRE_OKEY])
1116 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1117
1118 if (data[IFLA_GRE_LOCAL])
1119 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1120
1121 if (data[IFLA_GRE_REMOTE])
1122 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1123
1124 if (data[IFLA_GRE_TTL])
1125 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1126
1127 if (data[IFLA_GRE_TOS])
1128 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1129
1130 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1131 if (t->ignore_df)
1132 return -EINVAL;
1133 parms->iph.frag_off = htons(IP_DF);
1134 }
1135
1136 if (data[IFLA_GRE_COLLECT_METADATA]) {
1137 t->collect_md = true;
1138 if (dev->type == ARPHRD_IPGRE)
1139 dev->type = ARPHRD_NONE;
1140 }
1141
1142 if (data[IFLA_GRE_IGNORE_DF]) {
1143 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1144 && (parms->iph.frag_off & htons(IP_DF)))
1145 return -EINVAL;
1146 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1147 }
1148
1149 if (data[IFLA_GRE_FWMARK])
1150 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1151
1152 if (data[IFLA_GRE_ERSPAN_VER]) {
1153 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1154
1155 if (t->erspan_ver != 1 && t->erspan_ver != 2)
1156 return -EINVAL;
1157 }
1158
1159 if (t->erspan_ver == 1) {
1160 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1161 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1162 if (t->index & ~INDEX_MASK)
1163 return -EINVAL;
1164 }
1165 } else if (t->erspan_ver == 2) {
1166 if (data[IFLA_GRE_ERSPAN_DIR]) {
1167 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1168 if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1169 return -EINVAL;
1170 }
1171 if (data[IFLA_GRE_ERSPAN_HWID]) {
1172 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1173 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1174 return -EINVAL;
1175 }
1176 }
1177
1178 return 0;
1179}
1180
1181/* This function returns true when ENCAP attributes are present in the nl msg */
1182static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1183 struct ip_tunnel_encap *ipencap)
1184{
1185 bool ret = false;
1186
1187 memset(ipencap, 0, sizeof(*ipencap));
1188
1189 if (!data)
1190 return ret;
1191
1192 if (data[IFLA_GRE_ENCAP_TYPE]) {
1193 ret = true;
1194 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1195 }
1196
1197 if (data[IFLA_GRE_ENCAP_FLAGS]) {
1198 ret = true;
1199 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1200 }
1201
1202 if (data[IFLA_GRE_ENCAP_SPORT]) {
1203 ret = true;
1204 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1205 }
1206
1207 if (data[IFLA_GRE_ENCAP_DPORT]) {
1208 ret = true;
1209 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1210 }
1211
1212 return ret;
1213}
1214
1215static int gre_tap_init(struct net_device *dev)
1216{
1217 __gre_tunnel_init(dev);
1218 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1219 netif_keep_dst(dev);
1220
1221 return ip_tunnel_init(dev);
1222}
1223
1224static const struct net_device_ops gre_tap_netdev_ops = {
1225 .ndo_init = gre_tap_init,
1226 .ndo_uninit = ip_tunnel_uninit,
1227 .ndo_start_xmit = gre_tap_xmit,
1228 .ndo_set_mac_address = eth_mac_addr,
1229 .ndo_validate_addr = eth_validate_addr,
1230 .ndo_change_mtu = ip_tunnel_change_mtu,
1231 .ndo_get_stats64 = ip_tunnel_get_stats64,
1232 .ndo_get_iflink = ip_tunnel_get_iflink,
1233 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1234};
1235
1236static int erspan_tunnel_init(struct net_device *dev)
1237{
1238 struct ip_tunnel *tunnel = netdev_priv(dev);
1239
1240 tunnel->tun_hlen = 8;
1241 tunnel->parms.iph.protocol = IPPROTO_GRE;
1242 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1243 erspan_hdr_len(tunnel->erspan_ver);
1244
1245 dev->features |= GRE_FEATURES;
1246 dev->hw_features |= GRE_FEATURES;
1247 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1248 netif_keep_dst(dev);
1249
1250 return ip_tunnel_init(dev);
1251}
1252
1253static const struct net_device_ops erspan_netdev_ops = {
1254 .ndo_init = erspan_tunnel_init,
1255 .ndo_uninit = ip_tunnel_uninit,
1256 .ndo_start_xmit = erspan_xmit,
1257 .ndo_set_mac_address = eth_mac_addr,
1258 .ndo_validate_addr = eth_validate_addr,
1259 .ndo_change_mtu = ip_tunnel_change_mtu,
1260 .ndo_get_stats64 = ip_tunnel_get_stats64,
1261 .ndo_get_iflink = ip_tunnel_get_iflink,
1262 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1263};
1264
1265static void ipgre_tap_setup(struct net_device *dev)
1266{
1267 ether_setup(dev);
1268 dev->max_mtu = 0;
1269 dev->netdev_ops = &gre_tap_netdev_ops;
1270 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1271 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1272 ip_tunnel_setup(dev, gre_tap_net_id);
1273}
1274
1275static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1276 struct nlattr *tb[], struct nlattr *data[],
1277 struct netlink_ext_ack *extack)
1278{
1279 struct ip_tunnel_parm p;
1280 struct ip_tunnel_encap ipencap;
1281 __u32 fwmark = 0;
1282 int err;
1283
1284 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1285 struct ip_tunnel *t = netdev_priv(dev);
1286 err = ip_tunnel_encap_setup(t, &ipencap);
1287
1288 if (err < 0)
1289 return err;
1290 }
1291
1292 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1293 if (err < 0)
1294 return err;
1295 return ip_tunnel_newlink(dev, tb, &p, fwmark);
1296}
1297
1298static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1299 struct nlattr *data[],
1300 struct netlink_ext_ack *extack)
1301{
1302 struct ip_tunnel *t = netdev_priv(dev);
1303 struct ip_tunnel_encap ipencap;
1304 __u32 fwmark = t->fwmark;
1305 struct ip_tunnel_parm p;
1306 int err;
1307
1308 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1309 err = ip_tunnel_encap_setup(t, &ipencap);
1310
1311 if (err < 0)
1312 return err;
1313 }
1314
1315 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1316 if (err < 0)
1317 return err;
1318
1319 err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1320 if (err < 0)
1321 return err;
1322
1323 t->parms.i_flags = p.i_flags;
1324 t->parms.o_flags = p.o_flags;
1325
1326 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1327 ipgre_link_update(dev, !tb[IFLA_MTU]);
1328
1329 return 0;
1330}
1331
1332static size_t ipgre_get_size(const struct net_device *dev)
1333{
1334 return
1335 /* IFLA_GRE_LINK */
1336 nla_total_size(4) +
1337 /* IFLA_GRE_IFLAGS */
1338 nla_total_size(2) +
1339 /* IFLA_GRE_OFLAGS */
1340 nla_total_size(2) +
1341 /* IFLA_GRE_IKEY */
1342 nla_total_size(4) +
1343 /* IFLA_GRE_OKEY */
1344 nla_total_size(4) +
1345 /* IFLA_GRE_LOCAL */
1346 nla_total_size(4) +
1347 /* IFLA_GRE_REMOTE */
1348 nla_total_size(4) +
1349 /* IFLA_GRE_TTL */
1350 nla_total_size(1) +
1351 /* IFLA_GRE_TOS */
1352 nla_total_size(1) +
1353 /* IFLA_GRE_PMTUDISC */
1354 nla_total_size(1) +
1355 /* IFLA_GRE_ENCAP_TYPE */
1356 nla_total_size(2) +
1357 /* IFLA_GRE_ENCAP_FLAGS */
1358 nla_total_size(2) +
1359 /* IFLA_GRE_ENCAP_SPORT */
1360 nla_total_size(2) +
1361 /* IFLA_GRE_ENCAP_DPORT */
1362 nla_total_size(2) +
1363 /* IFLA_GRE_COLLECT_METADATA */
1364 nla_total_size(0) +
1365 /* IFLA_GRE_IGNORE_DF */
1366 nla_total_size(1) +
1367 /* IFLA_GRE_FWMARK */
1368 nla_total_size(4) +
1369 /* IFLA_GRE_ERSPAN_INDEX */
1370 nla_total_size(4) +
1371 /* IFLA_GRE_ERSPAN_VER */
1372 nla_total_size(1) +
1373 /* IFLA_GRE_ERSPAN_DIR */
1374 nla_total_size(1) +
1375 /* IFLA_GRE_ERSPAN_HWID */
1376 nla_total_size(2) +
1377 0;
1378}
1379
1380static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1381{
1382 struct ip_tunnel *t = netdev_priv(dev);
1383 struct ip_tunnel_parm *p = &t->parms;
1384 __be16 o_flags = p->o_flags;
1385
1386 if (t->erspan_ver == 1 || t->erspan_ver == 2) {
1387 if (!t->collect_md)
1388 o_flags |= TUNNEL_KEY;
1389
1390 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1391 goto nla_put_failure;
1392
1393 if (t->erspan_ver == 1) {
1394 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1395 goto nla_put_failure;
1396 } else {
1397 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1398 goto nla_put_failure;
1399 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1400 goto nla_put_failure;
1401 }
1402 }
1403
1404 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1405 nla_put_be16(skb, IFLA_GRE_IFLAGS,
1406 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1407 nla_put_be16(skb, IFLA_GRE_OFLAGS,
1408 gre_tnl_flags_to_gre_flags(o_flags)) ||
1409 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1410 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1411 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1412 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1413 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1414 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1415 nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1416 !!(p->iph.frag_off & htons(IP_DF))) ||
1417 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1418 goto nla_put_failure;
1419
1420 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1421 t->encap.type) ||
1422 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1423 t->encap.sport) ||
1424 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1425 t->encap.dport) ||
1426 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1427 t->encap.flags))
1428 goto nla_put_failure;
1429
1430 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1431 goto nla_put_failure;
1432
1433 if (t->collect_md) {
1434 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1435 goto nla_put_failure;
1436 }
1437
1438 return 0;
1439
1440nla_put_failure:
1441 return -EMSGSIZE;
1442}
1443
1444static void erspan_setup(struct net_device *dev)
1445{
1446 struct ip_tunnel *t = netdev_priv(dev);
1447
1448 ether_setup(dev);
1449 dev->max_mtu = 0;
1450 dev->netdev_ops = &erspan_netdev_ops;
1451 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1452 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1453 ip_tunnel_setup(dev, erspan_net_id);
1454 t->erspan_ver = 1;
1455}
1456
1457static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1458 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1459 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1460 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1461 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1462 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1463 [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1464 [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1465 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1466 [IFLA_GRE_TOS] = { .type = NLA_U8 },
1467 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 },
1468 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 },
1469 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 },
1470 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 },
1471 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 },
1472 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG },
1473 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 },
1474 [IFLA_GRE_FWMARK] = { .type = NLA_U32 },
1475 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1476 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 },
1477 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 },
1478 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 },
1479};
1480
1481static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1482 .kind = "gre",
1483 .maxtype = IFLA_GRE_MAX,
1484 .policy = ipgre_policy,
1485 .priv_size = sizeof(struct ip_tunnel),
1486 .setup = ipgre_tunnel_setup,
1487 .validate = ipgre_tunnel_validate,
1488 .newlink = ipgre_newlink,
1489 .changelink = ipgre_changelink,
1490 .dellink = ip_tunnel_dellink,
1491 .get_size = ipgre_get_size,
1492 .fill_info = ipgre_fill_info,
1493 .get_link_net = ip_tunnel_get_link_net,
1494};
1495
1496static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1497 .kind = "gretap",
1498 .maxtype = IFLA_GRE_MAX,
1499 .policy = ipgre_policy,
1500 .priv_size = sizeof(struct ip_tunnel),
1501 .setup = ipgre_tap_setup,
1502 .validate = ipgre_tap_validate,
1503 .newlink = ipgre_newlink,
1504 .changelink = ipgre_changelink,
1505 .dellink = ip_tunnel_dellink,
1506 .get_size = ipgre_get_size,
1507 .fill_info = ipgre_fill_info,
1508 .get_link_net = ip_tunnel_get_link_net,
1509};
1510
1511static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1512 .kind = "erspan",
1513 .maxtype = IFLA_GRE_MAX,
1514 .policy = ipgre_policy,
1515 .priv_size = sizeof(struct ip_tunnel),
1516 .setup = erspan_setup,
1517 .validate = erspan_validate,
1518 .newlink = ipgre_newlink,
1519 .changelink = ipgre_changelink,
1520 .dellink = ip_tunnel_dellink,
1521 .get_size = ipgre_get_size,
1522 .fill_info = ipgre_fill_info,
1523 .get_link_net = ip_tunnel_get_link_net,
1524};
1525
1526struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1527 u8 name_assign_type)
1528{
1529 struct nlattr *tb[IFLA_MAX + 1];
1530 struct net_device *dev;
1531 LIST_HEAD(list_kill);
1532 struct ip_tunnel *t;
1533 int err;
1534
1535 memset(&tb, 0, sizeof(tb));
1536
1537 dev = rtnl_create_link(net, name, name_assign_type,
1538 &ipgre_tap_ops, tb, NULL);
1539 if (IS_ERR(dev))
1540 return dev;
1541
1542 /* Configure flow based GRE device. */
1543 t = netdev_priv(dev);
1544 t->collect_md = true;
1545
1546 err = ipgre_newlink(net, dev, tb, NULL, NULL);
1547 if (err < 0) {
1548 free_netdev(dev);
1549 return ERR_PTR(err);
1550 }
1551
1552 /* openvswitch users expect packet sizes to be unrestricted,
1553 * so set the largest MTU we can.
1554 */
1555 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1556 if (err)
1557 goto out;
1558
1559 err = rtnl_configure_link(dev, NULL);
1560 if (err < 0)
1561 goto out;
1562
1563 return dev;
1564out:
1565 ip_tunnel_dellink(dev, &list_kill);
1566 unregister_netdevice_many(&list_kill);
1567 return ERR_PTR(err);
1568}
1569EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1570
1571static int __net_init ipgre_tap_init_net(struct net *net)
1572{
1573 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1574}
1575
1576static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1577{
1578 ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1579}
1580
1581static struct pernet_operations ipgre_tap_net_ops = {
1582 .init = ipgre_tap_init_net,
1583 .exit_batch = ipgre_tap_exit_batch_net,
1584 .id = &gre_tap_net_id,
1585 .size = sizeof(struct ip_tunnel_net),
1586};
1587
1588static int __net_init erspan_init_net(struct net *net)
1589{
1590 return ip_tunnel_init_net(net, erspan_net_id,
1591 &erspan_link_ops, "erspan0");
1592}
1593
1594static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1595{
1596 ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1597}
1598
1599static struct pernet_operations erspan_net_ops = {
1600 .init = erspan_init_net,
1601 .exit_batch = erspan_exit_batch_net,
1602 .id = &erspan_net_id,
1603 .size = sizeof(struct ip_tunnel_net),
1604};
1605
1606static int __init ipgre_init(void)
1607{
1608 int err;
1609
1610 pr_info("GRE over IPv4 tunneling driver\n");
1611
1612 err = register_pernet_device(&ipgre_net_ops);
1613 if (err < 0)
1614 return err;
1615
1616 err = register_pernet_device(&ipgre_tap_net_ops);
1617 if (err < 0)
1618 goto pnet_tap_failed;
1619
1620 err = register_pernet_device(&erspan_net_ops);
1621 if (err < 0)
1622 goto pnet_erspan_failed;
1623
1624 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1625 if (err < 0) {
1626 pr_info("%s: can't add protocol\n", __func__);
1627 goto add_proto_failed;
1628 }
1629
1630 err = rtnl_link_register(&ipgre_link_ops);
1631 if (err < 0)
1632 goto rtnl_link_failed;
1633
1634 err = rtnl_link_register(&ipgre_tap_ops);
1635 if (err < 0)
1636 goto tap_ops_failed;
1637
1638 err = rtnl_link_register(&erspan_link_ops);
1639 if (err < 0)
1640 goto erspan_link_failed;
1641
1642 return 0;
1643
1644erspan_link_failed:
1645 rtnl_link_unregister(&ipgre_tap_ops);
1646tap_ops_failed:
1647 rtnl_link_unregister(&ipgre_link_ops);
1648rtnl_link_failed:
1649 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1650add_proto_failed:
1651 unregister_pernet_device(&erspan_net_ops);
1652pnet_erspan_failed:
1653 unregister_pernet_device(&ipgre_tap_net_ops);
1654pnet_tap_failed:
1655 unregister_pernet_device(&ipgre_net_ops);
1656 return err;
1657}
1658
1659static void __exit ipgre_fini(void)
1660{
1661 rtnl_link_unregister(&ipgre_tap_ops);
1662 rtnl_link_unregister(&ipgre_link_ops);
1663 rtnl_link_unregister(&erspan_link_ops);
1664 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1665 unregister_pernet_device(&ipgre_tap_net_ops);
1666 unregister_pernet_device(&ipgre_net_ops);
1667 unregister_pernet_device(&erspan_net_ops);
1668}
1669
1670module_init(ipgre_init);
1671module_exit(ipgre_fini);
1672MODULE_LICENSE("GPL");
1673MODULE_ALIAS_RTNL_LINK("gre");
1674MODULE_ALIAS_RTNL_LINK("gretap");
1675MODULE_ALIAS_RTNL_LINK("erspan");
1676MODULE_ALIAS_NETDEV("gre0");
1677MODULE_ALIAS_NETDEV("gretap0");
1678MODULE_ALIAS_NETDEV("erspan0");