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