projects/linux·blob master · refs

1845 lines · 47.8 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Linux NET3: GRE over IP protocol decoder.
4 *
5 * Authors: Alexey Kuznetsov ([email protected])
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/flow.h>
32#include <net/sock.h>
33#include <net/ip.h>
34#include <net/icmp.h>
35#include <net/protocol.h>
36#include <net/ip_tunnels.h>
37#include <net/arp.h>
38#include <net/checksum.h>
39#include <net/dsfield.h>
40#include <net/inet_ecn.h>
41#include <net/xfrm.h>
42#include <net/net_namespace.h>
43#include <net/netns/generic.h>
44#include <net/rtnetlink.h>
45#include <net/gre.h>
46#include <net/dst_metadata.h>
47#include <net/erspan.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 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 break;
185
186 case ICMP_REDIRECT:
187 break;
188 }
189
190#if IS_ENABLED(CONFIG_IPV6)
191 if (tpi->proto == htons(ETH_P_IPV6)) {
192 unsigned int data_len = 0;
193
194 if (type == ICMP_TIME_EXCEEDED)
195 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
196
197 if (!ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
198 type, data_len))
199 return 0;
200 }
201#endif
202
203 if (t->parms.iph.daddr == 0 ||
204 ipv4_is_multicast(t->parms.iph.daddr))
205 return 0;
206
207 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
208 return 0;
209
210 if (time_before(jiffies, READ_ONCE(t->err_time) + IPTUNNEL_ERR_TIMEO))
211 WRITE_ONCE(t->err_count, READ_ONCE(t->err_count) + 1);
212 else
213 WRITE_ONCE(t->err_count, 1);
214 WRITE_ONCE(t->err_time, jiffies);
215
216 return 0;
217}
218
219static void gre_err(struct sk_buff *skb, u32 info)
220{
221 /* All the routers (except for Linux) return only
222 * 8 bytes of packet payload. It means, that precise relaying of
223 * ICMP in the real Internet is absolutely infeasible.
224 *
225 * Moreover, Cisco "wise men" put GRE key to the third word
226 * in GRE header. It makes impossible maintaining even soft
227 * state for keyed
228 * GRE tunnels with enabled checksum. Tell them "thank you".
229 *
230 * Well, I wonder, rfc1812 was written by Cisco employee,
231 * what the hell these idiots break standards established
232 * by themselves???
233 */
234
235 const struct iphdr *iph = (struct iphdr *)skb->data;
236 const int type = icmp_hdr(skb)->type;
237 const int code = icmp_hdr(skb)->code;
238 struct tnl_ptk_info tpi;
239
240 if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP),
241 iph->ihl * 4) < 0)
242 return;
243
244 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
245 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
246 skb->dev->ifindex, IPPROTO_GRE);
247 return;
248 }
249 if (type == ICMP_REDIRECT) {
250 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex,
251 IPPROTO_GRE);
252 return;
253 }
254
255 ipgre_err(skb, info, &tpi);
256}
257
258static bool is_erspan_type1(int gre_hdr_len)
259{
260 /* Both ERSPAN type I (version 0) and type II (version 1) use
261 * protocol 0x88BE, but the type I has only 4-byte GRE header,
262 * while type II has 8-byte.
263 */
264 return gre_hdr_len == 4;
265}
266
267static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
268 int gre_hdr_len)
269{
270 struct net *net = dev_net(skb->dev);
271 struct metadata_dst *tun_dst = NULL;
272 struct erspan_base_hdr *ershdr;
273 IP_TUNNEL_DECLARE_FLAGS(flags);
274 struct ip_tunnel_net *itn;
275 struct ip_tunnel *tunnel;
276 const struct iphdr *iph;
277 struct erspan_md2 *md2;
278 int ver;
279 int len;
280
281 ip_tunnel_flags_copy(flags, tpi->flags);
282
283 itn = net_generic(net, erspan_net_id);
284 iph = ip_hdr(skb);
285 if (is_erspan_type1(gre_hdr_len)) {
286 ver = 0;
287 __set_bit(IP_TUNNEL_NO_KEY_BIT, flags);
288 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, flags,
289 iph->saddr, iph->daddr, 0);
290 } else {
291 if (unlikely(!pskb_may_pull(skb,
292 gre_hdr_len + sizeof(*ershdr))))
293 return PACKET_REJECT;
294
295 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
296 ver = ershdr->ver;
297 iph = ip_hdr(skb);
298 __set_bit(IP_TUNNEL_KEY_BIT, flags);
299 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, flags,
300 iph->saddr, iph->daddr, tpi->key);
301 }
302
303 if (tunnel) {
304 if (is_erspan_type1(gre_hdr_len))
305 len = gre_hdr_len;
306 else
307 len = gre_hdr_len + erspan_hdr_len(ver);
308
309 if (unlikely(!pskb_may_pull(skb, len)))
310 return PACKET_REJECT;
311
312 if (__iptunnel_pull_header(skb,
313 len,
314 htons(ETH_P_TEB),
315 false, false) < 0)
316 goto drop;
317
318 if (tunnel->collect_md) {
319 struct erspan_metadata *pkt_md, *md;
320 struct ip_tunnel_info *info;
321 unsigned char *gh;
322 __be64 tun_id;
323
324 __set_bit(IP_TUNNEL_KEY_BIT, tpi->flags);
325 ip_tunnel_flags_copy(flags, tpi->flags);
326 tun_id = key32_to_tunnel_id(tpi->key);
327
328 tun_dst = ip_tun_rx_dst(skb, flags,
329 tun_id, sizeof(*md));
330 if (!tun_dst)
331 return PACKET_REJECT;
332
333 /* MUST set options_len before referencing options */
334 info = &tun_dst->u.tun_info;
335 info->options_len = sizeof(*md);
336
337 /* skb can be uncloned in __iptunnel_pull_header, so
338 * old pkt_md is no longer valid and we need to reset
339 * it
340 */
341 gh = skb_network_header(skb) +
342 skb_network_header_len(skb);
343 pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len +
344 sizeof(*ershdr));
345 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
346 md->version = ver;
347 md2 = &md->u.md2;
348 memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
349 ERSPAN_V2_MDSIZE);
350
351 __set_bit(IP_TUNNEL_ERSPAN_OPT_BIT,
352 info->key.tun_flags);
353 }
354
355 skb_reset_mac_header(skb);
356 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
357 return PACKET_RCVD;
358 }
359 return PACKET_REJECT;
360
361drop:
362 kfree_skb(skb);
363 return PACKET_RCVD;
364}
365
366static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
367 struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
368{
369 struct metadata_dst *tun_dst = NULL;
370 const struct iphdr *iph;
371 struct ip_tunnel *tunnel;
372
373 iph = ip_hdr(skb);
374 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
375 iph->saddr, iph->daddr, tpi->key);
376
377 if (tunnel) {
378 const struct iphdr *tnl_params;
379
380 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
381 raw_proto, false) < 0)
382 goto drop;
383
384 /* Special case for ipgre_header_parse(), which expects the
385 * mac_header to point to the outer IP header.
386 */
387 if (tunnel->dev->header_ops == &ipgre_header_ops)
388 skb_pop_mac_header(skb);
389 else
390 skb_reset_mac_header(skb);
391
392 tnl_params = &tunnel->parms.iph;
393 if (tunnel->collect_md || tnl_params->daddr == 0) {
394 IP_TUNNEL_DECLARE_FLAGS(flags) = { };
395 __be64 tun_id;
396
397 __set_bit(IP_TUNNEL_CSUM_BIT, flags);
398 __set_bit(IP_TUNNEL_KEY_BIT, flags);
399 ip_tunnel_flags_and(flags, tpi->flags, flags);
400
401 tun_id = key32_to_tunnel_id(tpi->key);
402 tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
403 if (!tun_dst)
404 return PACKET_REJECT;
405 }
406
407 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
408 return PACKET_RCVD;
409 }
410 return PACKET_NEXT;
411
412drop:
413 kfree_skb(skb);
414 return PACKET_RCVD;
415}
416
417static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
418 int hdr_len)
419{
420 struct net *net = dev_net(skb->dev);
421 struct ip_tunnel_net *itn;
422 int res;
423
424 if (tpi->proto == htons(ETH_P_TEB))
425 itn = net_generic(net, gre_tap_net_id);
426 else
427 itn = net_generic(net, ipgre_net_id);
428
429 res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
430 if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
431 /* ipgre tunnels in collect metadata mode should receive
432 * also ETH_P_TEB traffic.
433 */
434 itn = net_generic(net, ipgre_net_id);
435 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
436 }
437 return res;
438}
439
440static int gre_rcv(struct sk_buff *skb)
441{
442 struct tnl_ptk_info tpi;
443 bool csum_err = false;
444 int hdr_len;
445
446#ifdef CONFIG_NET_IPGRE_BROADCAST
447 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
448 /* Looped back packet, drop it! */
449 if (rt_is_output_route(skb_rtable(skb)))
450 goto drop;
451 }
452#endif
453
454 hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
455 if (hdr_len < 0)
456 goto drop;
457
458 if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
459 tpi.proto == htons(ETH_P_ERSPAN2))) {
460 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
461 return 0;
462 goto out;
463 }
464
465 if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
466 return 0;
467
468out:
469 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
470drop:
471 dev_core_stats_rx_dropped_inc(skb->dev);
472 kfree_skb(skb);
473 return 0;
474}
475
476static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
477 const struct iphdr *tnl_params,
478 __be16 proto, const unsigned long *flags)
479{
480 struct ip_tunnel *tunnel = netdev_priv(dev);
481
482 /* Push GRE header. */
483 gre_build_header(skb, tunnel->tun_hlen,
484 flags, proto, tunnel->parms.o_key,
485 test_bit(IP_TUNNEL_SEQ_BIT, flags) ?
486 htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0);
487
488 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
489}
490
491static int gre_handle_offloads(struct sk_buff *skb, bool csum)
492{
493 return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
494}
495
496static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
497 __be16 proto)
498{
499 struct ip_tunnel *tunnel = netdev_priv(dev);
500 IP_TUNNEL_DECLARE_FLAGS(flags) = { };
501 struct ip_tunnel_info *tun_info;
502 const struct ip_tunnel_key *key;
503 int tunnel_hlen;
504
505 tun_info = skb_tunnel_info(skb);
506 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
507 ip_tunnel_info_af(tun_info) != AF_INET))
508 goto err_free_skb;
509
510 key = &tun_info->key;
511 tunnel_hlen = gre_calc_hlen(key->tun_flags);
512
513 if (skb_cow_head(skb, dev->needed_headroom))
514 goto err_free_skb;
515
516 /* Push Tunnel header. */
517 if (gre_handle_offloads(skb, test_bit(IP_TUNNEL_CSUM_BIT,
518 tunnel->parms.o_flags)))
519 goto err_free_skb;
520
521 __set_bit(IP_TUNNEL_CSUM_BIT, flags);
522 __set_bit(IP_TUNNEL_KEY_BIT, flags);
523 __set_bit(IP_TUNNEL_SEQ_BIT, flags);
524 ip_tunnel_flags_and(flags, tun_info->key.tun_flags, flags);
525
526 gre_build_header(skb, tunnel_hlen, flags, proto,
527 tunnel_id_to_key32(tun_info->key.tun_id),
528 test_bit(IP_TUNNEL_SEQ_BIT, flags) ?
529 htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0);
530
531 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
532
533 return;
534
535err_free_skb:
536 kfree_skb(skb);
537 DEV_STATS_INC(dev, tx_dropped);
538}
539
540static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev)
541{
542 struct ip_tunnel *tunnel = netdev_priv(dev);
543 IP_TUNNEL_DECLARE_FLAGS(flags) = { };
544 struct ip_tunnel_info *tun_info;
545 const struct ip_tunnel_key *key;
546 struct erspan_metadata *md;
547 bool truncate = false;
548 __be16 proto;
549 int tunnel_hlen;
550 int version;
551 int nhoff;
552
553 tun_info = skb_tunnel_info(skb);
554 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
555 ip_tunnel_info_af(tun_info) != AF_INET))
556 goto err_free_skb;
557
558 key = &tun_info->key;
559 if (!test_bit(IP_TUNNEL_ERSPAN_OPT_BIT, tun_info->key.tun_flags))
560 goto err_free_skb;
561 if (tun_info->options_len < sizeof(*md))
562 goto err_free_skb;
563 md = ip_tunnel_info_opts(tun_info);
564
565 /* ERSPAN has fixed 8 byte GRE header */
566 version = md->version;
567 tunnel_hlen = 8 + erspan_hdr_len(version);
568
569 if (skb_cow_head(skb, dev->needed_headroom))
570 goto err_free_skb;
571
572 if (gre_handle_offloads(skb, false))
573 goto err_free_skb;
574
575 if (skb->len > dev->mtu + dev->hard_header_len) {
576 if (pskb_trim(skb, dev->mtu + dev->hard_header_len))
577 goto err_free_skb;
578 truncate = true;
579 }
580
581 nhoff = skb_network_offset(skb);
582 if (skb->protocol == htons(ETH_P_IP) &&
583 (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
584 truncate = true;
585
586 if (skb->protocol == htons(ETH_P_IPV6)) {
587 int thoff;
588
589 if (skb_transport_header_was_set(skb))
590 thoff = skb_transport_offset(skb);
591 else
592 thoff = nhoff + sizeof(struct ipv6hdr);
593 if (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff)
594 truncate = true;
595 }
596
597 if (version == 1) {
598 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
599 ntohl(md->u.index), truncate, true);
600 proto = htons(ETH_P_ERSPAN);
601 } else if (version == 2) {
602 erspan_build_header_v2(skb,
603 ntohl(tunnel_id_to_key32(key->tun_id)),
604 md->u.md2.dir,
605 get_hwid(&md->u.md2),
606 truncate, true);
607 proto = htons(ETH_P_ERSPAN2);
608 } else {
609 goto err_free_skb;
610 }
611
612 __set_bit(IP_TUNNEL_SEQ_BIT, flags);
613 gre_build_header(skb, 8, flags, proto, 0,
614 htonl(atomic_fetch_inc(&tunnel->o_seqno)));
615
616 ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
617
618 return;
619
620err_free_skb:
621 kfree_skb(skb);
622 DEV_STATS_INC(dev, tx_dropped);
623}
624
625static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
626{
627 struct ip_tunnel_info *info = skb_tunnel_info(skb);
628 const struct ip_tunnel_key *key;
629 struct rtable *rt;
630 struct flowi4 fl4;
631
632 if (ip_tunnel_info_af(info) != AF_INET)
633 return -EINVAL;
634
635 key = &info->key;
636 ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src,
637 tunnel_id_to_key32(key->tun_id),
638 key->tos & ~INET_ECN_MASK, dev_net(dev), 0,
639 skb->mark, skb_get_hash(skb), key->flow_flags);
640 rt = ip_route_output_key(dev_net(dev), &fl4);
641 if (IS_ERR(rt))
642 return PTR_ERR(rt);
643
644 ip_rt_put(rt);
645 info->key.u.ipv4.src = fl4.saddr;
646 return 0;
647}
648
649static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
650 struct net_device *dev)
651{
652 struct ip_tunnel *tunnel = netdev_priv(dev);
653 IP_TUNNEL_DECLARE_FLAGS(flags);
654 const struct iphdr *tnl_params;
655
656 if (!pskb_inet_may_pull(skb))
657 goto free_skb;
658
659 if (tunnel->collect_md) {
660 gre_fb_xmit(skb, dev, skb->protocol);
661 return NETDEV_TX_OK;
662 }
663
664 if (dev->header_ops) {
665 int pull_len = tunnel->hlen + sizeof(struct iphdr);
666
667 if (skb_cow_head(skb, 0))
668 goto free_skb;
669
670 if (!pskb_may_pull(skb, pull_len))
671 goto free_skb;
672
673 tnl_params = (const struct iphdr *)skb->data;
674
675 /* ip_tunnel_xmit() needs skb->data pointing to gre header. */
676 skb_pull(skb, pull_len);
677 skb_reset_mac_header(skb);
678
679 if (skb->ip_summed == CHECKSUM_PARTIAL &&
680 skb_checksum_start(skb) < skb->data)
681 goto free_skb;
682 } else {
683 if (skb_cow_head(skb, dev->needed_headroom))
684 goto free_skb;
685
686 tnl_params = &tunnel->parms.iph;
687 }
688
689 ip_tunnel_flags_copy(flags, tunnel->parms.o_flags);
690
691 if (gre_handle_offloads(skb, test_bit(IP_TUNNEL_CSUM_BIT, flags)))
692 goto free_skb;
693
694 __gre_xmit(skb, dev, tnl_params, skb->protocol, flags);
695 return NETDEV_TX_OK;
696
697free_skb:
698 kfree_skb(skb);
699 DEV_STATS_INC(dev, tx_dropped);
700 return NETDEV_TX_OK;
701}
702
703static netdev_tx_t erspan_xmit(struct sk_buff *skb,
704 struct net_device *dev)
705{
706 struct ip_tunnel *tunnel = netdev_priv(dev);
707 IP_TUNNEL_DECLARE_FLAGS(flags);
708 bool truncate = false;
709 __be16 proto;
710
711 if (!pskb_inet_may_pull(skb))
712 goto free_skb;
713
714 if (tunnel->collect_md) {
715 erspan_fb_xmit(skb, dev);
716 return NETDEV_TX_OK;
717 }
718
719 if (gre_handle_offloads(skb, false))
720 goto free_skb;
721
722 if (skb_cow_head(skb, dev->needed_headroom))
723 goto free_skb;
724
725 if (skb->len > dev->mtu + dev->hard_header_len) {
726 if (pskb_trim(skb, dev->mtu + dev->hard_header_len))
727 goto free_skb;
728 truncate = true;
729 }
730
731 ip_tunnel_flags_copy(flags, tunnel->parms.o_flags);
732
733 /* Push ERSPAN header */
734 if (tunnel->erspan_ver == 0) {
735 proto = htons(ETH_P_ERSPAN);
736 __clear_bit(IP_TUNNEL_SEQ_BIT, flags);
737 } else if (tunnel->erspan_ver == 1) {
738 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
739 tunnel->index,
740 truncate, true);
741 proto = htons(ETH_P_ERSPAN);
742 } else if (tunnel->erspan_ver == 2) {
743 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
744 tunnel->dir, tunnel->hwid,
745 truncate, true);
746 proto = htons(ETH_P_ERSPAN2);
747 } else {
748 goto free_skb;
749 }
750
751 __clear_bit(IP_TUNNEL_KEY_BIT, flags);
752 __gre_xmit(skb, dev, &tunnel->parms.iph, proto, flags);
753 return NETDEV_TX_OK;
754
755free_skb:
756 kfree_skb(skb);
757 DEV_STATS_INC(dev, tx_dropped);
758 return NETDEV_TX_OK;
759}
760
761static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
762 struct net_device *dev)
763{
764 struct ip_tunnel *tunnel = netdev_priv(dev);
765 IP_TUNNEL_DECLARE_FLAGS(flags);
766
767 if (!pskb_inet_may_pull(skb))
768 goto free_skb;
769
770 if (tunnel->collect_md) {
771 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
772 return NETDEV_TX_OK;
773 }
774
775 ip_tunnel_flags_copy(flags, tunnel->parms.o_flags);
776
777 if (gre_handle_offloads(skb, test_bit(IP_TUNNEL_CSUM_BIT, flags)))
778 goto free_skb;
779
780 if (skb_cow_head(skb, dev->needed_headroom))
781 goto free_skb;
782
783 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB), flags);
784 return NETDEV_TX_OK;
785
786free_skb:
787 kfree_skb(skb);
788 DEV_STATS_INC(dev, tx_dropped);
789 return NETDEV_TX_OK;
790}
791
792static void ipgre_link_update(struct net_device *dev, bool set_mtu)
793{
794 struct ip_tunnel *tunnel = netdev_priv(dev);
795 int len;
796
797 len = tunnel->tun_hlen;
798 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
799 len = tunnel->tun_hlen - len;
800 tunnel->hlen = tunnel->hlen + len;
801
802 if (dev->header_ops)
803 dev->hard_header_len += len;
804 else
805 dev->needed_headroom += len;
806
807 if (set_mtu)
808 WRITE_ONCE(dev->mtu, max_t(int, dev->mtu - len, 68));
809
810 if (test_bit(IP_TUNNEL_SEQ_BIT, tunnel->parms.o_flags) ||
811 (test_bit(IP_TUNNEL_CSUM_BIT, tunnel->parms.o_flags) &&
812 tunnel->encap.type != TUNNEL_ENCAP_NONE)) {
813 dev->features &= ~NETIF_F_GSO_SOFTWARE;
814 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
815 } else {
816 dev->features |= NETIF_F_GSO_SOFTWARE;
817 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
818 }
819}
820
821static int ipgre_tunnel_ctl(struct net_device *dev,
822 struct ip_tunnel_parm_kern *p,
823 int cmd)
824{
825 __be16 i_flags, o_flags;
826 int err;
827
828 if (!ip_tunnel_flags_is_be16_compat(p->i_flags) ||
829 !ip_tunnel_flags_is_be16_compat(p->o_flags))
830 return -EOVERFLOW;
831
832 i_flags = ip_tunnel_flags_to_be16(p->i_flags);
833 o_flags = ip_tunnel_flags_to_be16(p->o_flags);
834
835 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
836 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE ||
837 p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) ||
838 ((i_flags | o_flags) & (GRE_VERSION | GRE_ROUTING)))
839 return -EINVAL;
840 }
841
842 gre_flags_to_tnl_flags(p->i_flags, i_flags);
843 gre_flags_to_tnl_flags(p->o_flags, o_flags);
844
845 err = ip_tunnel_ctl(dev, p, cmd);
846 if (err)
847 return err;
848
849 if (cmd == SIOCCHGTUNNEL) {
850 struct ip_tunnel *t = netdev_priv(dev);
851
852 ip_tunnel_flags_copy(t->parms.i_flags, p->i_flags);
853 ip_tunnel_flags_copy(t->parms.o_flags, p->o_flags);
854
855 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
856 ipgre_link_update(dev, true);
857 }
858
859 i_flags = gre_tnl_flags_to_gre_flags(p->i_flags);
860 ip_tunnel_flags_from_be16(p->i_flags, i_flags);
861 o_flags = gre_tnl_flags_to_gre_flags(p->o_flags);
862 ip_tunnel_flags_from_be16(p->o_flags, o_flags);
863
864 return 0;
865}
866
867/* Nice toy. Unfortunately, useless in real life :-)
868 It allows to construct virtual multiprotocol broadcast "LAN"
869 over the Internet, provided multicast routing is tuned.
870
871
872 I have no idea was this bicycle invented before me,
873 so that I had to set ARPHRD_IPGRE to a random value.
874 I have an impression, that Cisco could make something similar,
875 but this feature is apparently missing in IOS<=11.2(8).
876
877 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
878 with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
879
880 ping -t 255 224.66.66.66
881
882 If nobody answers, mbone does not work.
883
884 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
885 ip addr add 10.66.66.<somewhat>/24 dev Universe
886 ifconfig Universe up
887 ifconfig Universe add fe80::<Your_real_addr>/10
888 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
889 ftp 10.66.66.66
890 ...
891 ftp fec0:6666:6666::193.233.7.65
892 ...
893 */
894static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
895 unsigned short type,
896 const void *daddr, const void *saddr, unsigned int len)
897{
898 struct ip_tunnel *t = netdev_priv(dev);
899 struct gre_base_hdr *greh;
900 struct iphdr *iph;
901 int needed;
902
903 needed = t->hlen + sizeof(*iph);
904 if (skb_headroom(skb) < needed &&
905 pskb_expand_head(skb, HH_DATA_ALIGN(needed - skb_headroom(skb)),
906 0, GFP_ATOMIC))
907 return -needed;
908
909 iph = skb_push(skb, needed);
910 greh = (struct gre_base_hdr *)(iph+1);
911 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
912 greh->protocol = htons(type);
913
914 memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
915
916 /* Set the source hardware address. */
917 if (saddr)
918 memcpy(&iph->saddr, saddr, 4);
919 if (daddr)
920 memcpy(&iph->daddr, daddr, 4);
921 if (iph->daddr)
922 return t->hlen + sizeof(*iph);
923
924 return -(t->hlen + sizeof(*iph));
925}
926
927static int ipgre_header_parse(const struct sk_buff *skb, const struct net_device *dev,
928 unsigned char *haddr)
929{
930 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
931 memcpy(haddr, &iph->saddr, 4);
932 return 4;
933}
934
935static const struct header_ops ipgre_header_ops = {
936 .create = ipgre_header,
937 .parse = ipgre_header_parse,
938};
939
940#ifdef CONFIG_NET_IPGRE_BROADCAST
941static int ipgre_open(struct net_device *dev)
942{
943 struct ip_tunnel *t = netdev_priv(dev);
944
945 if (ipv4_is_multicast(t->parms.iph.daddr)) {
946 struct flowi4 fl4 = {
947 .flowi4_oif = t->parms.link,
948 .flowi4_dscp = ip4h_dscp(&t->parms.iph),
949 .flowi4_scope = RT_SCOPE_UNIVERSE,
950 .flowi4_proto = IPPROTO_GRE,
951 .saddr = t->parms.iph.saddr,
952 .daddr = t->parms.iph.daddr,
953 .fl4_gre_key = t->parms.o_key,
954 };
955 struct rtable *rt;
956
957 rt = ip_route_output_key(t->net, &fl4);
958 if (IS_ERR(rt))
959 return -EADDRNOTAVAIL;
960 dev = rt->dst.dev;
961 ip_rt_put(rt);
962 if (!__in_dev_get_rtnl(dev))
963 return -EADDRNOTAVAIL;
964 t->mlink = dev->ifindex;
965 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
966 }
967 return 0;
968}
969
970static int ipgre_close(struct net_device *dev)
971{
972 struct ip_tunnel *t = netdev_priv(dev);
973
974 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
975 struct in_device *in_dev;
976 in_dev = inetdev_by_index(t->net, t->mlink);
977 if (in_dev)
978 ip_mc_dec_group(in_dev, t->parms.iph.daddr);
979 }
980 return 0;
981}
982#endif
983
984static const struct net_device_ops ipgre_netdev_ops = {
985 .ndo_init = ipgre_tunnel_init,
986 .ndo_uninit = ip_tunnel_uninit,
987#ifdef CONFIG_NET_IPGRE_BROADCAST
988 .ndo_open = ipgre_open,
989 .ndo_stop = ipgre_close,
990#endif
991 .ndo_start_xmit = ipgre_xmit,
992 .ndo_siocdevprivate = ip_tunnel_siocdevprivate,
993 .ndo_change_mtu = ip_tunnel_change_mtu,
994 .ndo_get_stats64 = dev_get_tstats64,
995 .ndo_get_iflink = ip_tunnel_get_iflink,
996 .ndo_tunnel_ctl = ipgre_tunnel_ctl,
997};
998
999#define GRE_FEATURES (NETIF_F_SG | \
1000 NETIF_F_FRAGLIST | \
1001 NETIF_F_HIGHDMA | \
1002 NETIF_F_HW_CSUM)
1003
1004static void ipgre_tunnel_setup(struct net_device *dev)
1005{
1006 dev->netdev_ops = &ipgre_netdev_ops;
1007 dev->type = ARPHRD_IPGRE;
1008 ip_tunnel_setup(dev, ipgre_net_id);
1009}
1010
1011static void __gre_tunnel_init(struct net_device *dev)
1012{
1013 struct ip_tunnel *tunnel;
1014
1015 tunnel = netdev_priv(dev);
1016 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
1017 tunnel->parms.iph.protocol = IPPROTO_GRE;
1018
1019 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
1020 dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph);
1021
1022 dev->features |= GRE_FEATURES;
1023 dev->hw_features |= GRE_FEATURES;
1024
1025 dev->lltx = true;
1026
1027 /* TCP offload with GRE SEQ is not supported, nor can we support 2
1028 * levels of outer headers requiring an update.
1029 */
1030 if (test_bit(IP_TUNNEL_SEQ_BIT, tunnel->parms.o_flags))
1031 return;
1032 if (test_bit(IP_TUNNEL_CSUM_BIT, tunnel->parms.o_flags) &&
1033 tunnel->encap.type != TUNNEL_ENCAP_NONE)
1034 return;
1035
1036 dev->features |= NETIF_F_GSO_SOFTWARE;
1037 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1038}
1039
1040static int ipgre_tunnel_init(struct net_device *dev)
1041{
1042 struct ip_tunnel *tunnel = netdev_priv(dev);
1043 struct iphdr *iph = &tunnel->parms.iph;
1044
1045 __gre_tunnel_init(dev);
1046
1047 __dev_addr_set(dev, &iph->saddr, 4);
1048 memcpy(dev->broadcast, &iph->daddr, 4);
1049
1050 dev->flags = IFF_NOARP;
1051 netif_keep_dst(dev);
1052 dev->addr_len = 4;
1053
1054 if (iph->daddr && !tunnel->collect_md) {
1055#ifdef CONFIG_NET_IPGRE_BROADCAST
1056 if (ipv4_is_multicast(iph->daddr)) {
1057 if (!iph->saddr)
1058 return -EINVAL;
1059 dev->flags = IFF_BROADCAST;
1060 dev->header_ops = &ipgre_header_ops;
1061 dev->hard_header_len = tunnel->hlen + sizeof(*iph);
1062 dev->needed_headroom = 0;
1063 }
1064#endif
1065 } else if (!tunnel->collect_md) {
1066 dev->header_ops = &ipgre_header_ops;
1067 dev->hard_header_len = tunnel->hlen + sizeof(*iph);
1068 dev->needed_headroom = 0;
1069 }
1070
1071 return ip_tunnel_init(dev);
1072}
1073
1074static const struct gre_protocol ipgre_protocol = {
1075 .handler = gre_rcv,
1076 .err_handler = gre_err,
1077};
1078
1079static int __net_init ipgre_init_net(struct net *net)
1080{
1081 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1082}
1083
1084static void __net_exit ipgre_exit_rtnl(struct net *net,
1085 struct list_head *dev_to_kill)
1086{
1087 ip_tunnel_delete_net(net, ipgre_net_id, &ipgre_link_ops, dev_to_kill);
1088}
1089
1090static struct pernet_operations ipgre_net_ops = {
1091 .init = ipgre_init_net,
1092 .exit_rtnl = ipgre_exit_rtnl,
1093 .id = &ipgre_net_id,
1094 .size = sizeof(struct ip_tunnel_net),
1095};
1096
1097static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1098 struct netlink_ext_ack *extack)
1099{
1100 __be16 flags;
1101
1102 if (!data)
1103 return 0;
1104
1105 flags = 0;
1106 if (data[IFLA_GRE_IFLAGS])
1107 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1108 if (data[IFLA_GRE_OFLAGS])
1109 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1110 if (flags & (GRE_VERSION|GRE_ROUTING))
1111 return -EINVAL;
1112
1113 if (data[IFLA_GRE_COLLECT_METADATA] &&
1114 data[IFLA_GRE_ENCAP_TYPE] &&
1115 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1116 return -EINVAL;
1117
1118 return 0;
1119}
1120
1121static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1122 struct netlink_ext_ack *extack)
1123{
1124 __be32 daddr;
1125
1126 if (tb[IFLA_ADDRESS]) {
1127 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1128 return -EINVAL;
1129 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1130 return -EADDRNOTAVAIL;
1131 }
1132
1133 if (!data)
1134 goto out;
1135
1136 if (data[IFLA_GRE_REMOTE]) {
1137 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1138 if (!daddr)
1139 return -EINVAL;
1140 }
1141
1142out:
1143 return ipgre_tunnel_validate(tb, data, extack);
1144}
1145
1146static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1147 struct netlink_ext_ack *extack)
1148{
1149 __be16 flags = 0;
1150 int ret;
1151
1152 if (!data)
1153 return 0;
1154
1155 ret = ipgre_tap_validate(tb, data, extack);
1156 if (ret)
1157 return ret;
1158
1159 if (data[IFLA_GRE_ERSPAN_VER] &&
1160 nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0)
1161 return 0;
1162
1163 /* ERSPAN type II/III should only have GRE sequence and key flag */
1164 if (data[IFLA_GRE_OFLAGS])
1165 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1166 if (data[IFLA_GRE_IFLAGS])
1167 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1168 if (!data[IFLA_GRE_COLLECT_METADATA] &&
1169 flags != (GRE_SEQ | GRE_KEY))
1170 return -EINVAL;
1171
1172 /* ERSPAN Session ID only has 10-bit. Since we reuse
1173 * 32-bit key field as ID, check it's range.
1174 */
1175 if (data[IFLA_GRE_IKEY] &&
1176 (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1177 return -EINVAL;
1178
1179 if (data[IFLA_GRE_OKEY] &&
1180 (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1181 return -EINVAL;
1182
1183 return 0;
1184}
1185
1186static int ipgre_netlink_parms(struct net_device *dev,
1187 struct nlattr *data[],
1188 struct nlattr *tb[],
1189 struct ip_tunnel_parm_kern *parms,
1190 __u32 *fwmark)
1191{
1192 struct ip_tunnel *t = netdev_priv(dev);
1193
1194 memset(parms, 0, sizeof(*parms));
1195
1196 parms->iph.protocol = IPPROTO_GRE;
1197
1198 if (!data)
1199 return 0;
1200
1201 if (data[IFLA_GRE_LINK])
1202 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1203
1204 if (data[IFLA_GRE_IFLAGS])
1205 gre_flags_to_tnl_flags(parms->i_flags,
1206 nla_get_be16(data[IFLA_GRE_IFLAGS]));
1207
1208 if (data[IFLA_GRE_OFLAGS])
1209 gre_flags_to_tnl_flags(parms->o_flags,
1210 nla_get_be16(data[IFLA_GRE_OFLAGS]));
1211
1212 if (data[IFLA_GRE_IKEY])
1213 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1214
1215 if (data[IFLA_GRE_OKEY])
1216 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1217
1218 if (data[IFLA_GRE_LOCAL])
1219 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1220
1221 if (data[IFLA_GRE_REMOTE])
1222 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1223
1224 if (data[IFLA_GRE_TTL])
1225 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1226
1227 if (data[IFLA_GRE_TOS])
1228 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1229
1230 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1231 if (t->ignore_df)
1232 return -EINVAL;
1233 parms->iph.frag_off = htons(IP_DF);
1234 }
1235
1236 if (data[IFLA_GRE_COLLECT_METADATA]) {
1237 t->collect_md = true;
1238 if (dev->type == ARPHRD_IPGRE)
1239 dev->type = ARPHRD_NONE;
1240 }
1241
1242 if (data[IFLA_GRE_IGNORE_DF]) {
1243 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1244 && (parms->iph.frag_off & htons(IP_DF)))
1245 return -EINVAL;
1246 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1247 }
1248
1249 if (data[IFLA_GRE_FWMARK])
1250 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1251
1252 return 0;
1253}
1254
1255static int erspan_netlink_parms(struct net_device *dev,
1256 struct nlattr *data[],
1257 struct nlattr *tb[],
1258 struct ip_tunnel_parm_kern *parms,
1259 __u32 *fwmark)
1260{
1261 struct ip_tunnel *t = netdev_priv(dev);
1262 int err;
1263
1264 err = ipgre_netlink_parms(dev, data, tb, parms, fwmark);
1265 if (err)
1266 return err;
1267 if (!data)
1268 return 0;
1269
1270 if (data[IFLA_GRE_ERSPAN_VER]) {
1271 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1272
1273 if (t->erspan_ver > 2)
1274 return -EINVAL;
1275 }
1276
1277 if (t->erspan_ver == 1) {
1278 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1279 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1280 if (t->index & ~INDEX_MASK)
1281 return -EINVAL;
1282 }
1283 } else if (t->erspan_ver == 2) {
1284 if (data[IFLA_GRE_ERSPAN_DIR]) {
1285 t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1286 if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1287 return -EINVAL;
1288 }
1289 if (data[IFLA_GRE_ERSPAN_HWID]) {
1290 t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1291 if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1292 return -EINVAL;
1293 }
1294 }
1295
1296 return 0;
1297}
1298
1299/* This function returns true when ENCAP attributes are present in the nl msg */
1300static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1301 struct ip_tunnel_encap *ipencap)
1302{
1303 bool ret = false;
1304
1305 memset(ipencap, 0, sizeof(*ipencap));
1306
1307 if (!data)
1308 return ret;
1309
1310 if (data[IFLA_GRE_ENCAP_TYPE]) {
1311 ret = true;
1312 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1313 }
1314
1315 if (data[IFLA_GRE_ENCAP_FLAGS]) {
1316 ret = true;
1317 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1318 }
1319
1320 if (data[IFLA_GRE_ENCAP_SPORT]) {
1321 ret = true;
1322 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1323 }
1324
1325 if (data[IFLA_GRE_ENCAP_DPORT]) {
1326 ret = true;
1327 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1328 }
1329
1330 return ret;
1331}
1332
1333static int gre_tap_init(struct net_device *dev)
1334{
1335 __gre_tunnel_init(dev);
1336 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1337 netif_keep_dst(dev);
1338
1339 return ip_tunnel_init(dev);
1340}
1341
1342static const struct net_device_ops gre_tap_netdev_ops = {
1343 .ndo_init = gre_tap_init,
1344 .ndo_uninit = ip_tunnel_uninit,
1345 .ndo_start_xmit = gre_tap_xmit,
1346 .ndo_set_mac_address = eth_mac_addr,
1347 .ndo_validate_addr = eth_validate_addr,
1348 .ndo_change_mtu = ip_tunnel_change_mtu,
1349 .ndo_get_stats64 = dev_get_tstats64,
1350 .ndo_get_iflink = ip_tunnel_get_iflink,
1351 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1352};
1353
1354static int erspan_tunnel_init(struct net_device *dev)
1355{
1356 struct ip_tunnel *tunnel = netdev_priv(dev);
1357
1358 if (tunnel->erspan_ver == 0)
1359 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */
1360 else
1361 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */
1362
1363 tunnel->parms.iph.protocol = IPPROTO_GRE;
1364 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1365 erspan_hdr_len(tunnel->erspan_ver);
1366
1367 dev->features |= GRE_FEATURES;
1368 dev->hw_features |= GRE_FEATURES;
1369 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1370 netif_keep_dst(dev);
1371
1372 return ip_tunnel_init(dev);
1373}
1374
1375static const struct net_device_ops erspan_netdev_ops = {
1376 .ndo_init = erspan_tunnel_init,
1377 .ndo_uninit = ip_tunnel_uninit,
1378 .ndo_start_xmit = erspan_xmit,
1379 .ndo_set_mac_address = eth_mac_addr,
1380 .ndo_validate_addr = eth_validate_addr,
1381 .ndo_change_mtu = ip_tunnel_change_mtu,
1382 .ndo_get_stats64 = dev_get_tstats64,
1383 .ndo_get_iflink = ip_tunnel_get_iflink,
1384 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1385};
1386
1387static void ipgre_tap_setup(struct net_device *dev)
1388{
1389 ether_setup(dev);
1390 dev->max_mtu = 0;
1391 dev->netdev_ops = &gre_tap_netdev_ops;
1392 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1393 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1394 ip_tunnel_setup(dev, gre_tap_net_id);
1395}
1396
1397static int
1398ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[])
1399{
1400 struct ip_tunnel_encap ipencap;
1401
1402 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1403 struct ip_tunnel *t = netdev_priv(dev);
1404 int err = ip_tunnel_encap_setup(t, &ipencap);
1405
1406 if (err < 0)
1407 return err;
1408 }
1409
1410 return 0;
1411}
1412
1413static int ipgre_newlink(struct net_device *dev,
1414 struct rtnl_newlink_params *params,
1415 struct netlink_ext_ack *extack)
1416{
1417 struct nlattr **data = params->data;
1418 struct nlattr **tb = params->tb;
1419 struct ip_tunnel_parm_kern p;
1420 __u32 fwmark = 0;
1421 int err;
1422
1423 err = ipgre_newlink_encap_setup(dev, data);
1424 if (err)
1425 return err;
1426
1427 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1428 if (err < 0)
1429 return err;
1430 return ip_tunnel_newlink(params->link_net ? : dev_net(dev), dev, tb, &p,
1431 fwmark);
1432}
1433
1434static int erspan_newlink(struct net_device *dev,
1435 struct rtnl_newlink_params *params,
1436 struct netlink_ext_ack *extack)
1437{
1438 struct nlattr **data = params->data;
1439 struct nlattr **tb = params->tb;
1440 struct ip_tunnel_parm_kern p;
1441 __u32 fwmark = 0;
1442 int err;
1443
1444 err = ipgre_newlink_encap_setup(dev, data);
1445 if (err)
1446 return err;
1447
1448 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark);
1449 if (err)
1450 return err;
1451 return ip_tunnel_newlink(params->link_net ? : dev_net(dev), dev, tb, &p,
1452 fwmark);
1453}
1454
1455static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1456 struct nlattr *data[],
1457 struct netlink_ext_ack *extack)
1458{
1459 struct ip_tunnel *t = netdev_priv(dev);
1460 struct ip_tunnel_parm_kern p;
1461 __u32 fwmark = t->fwmark;
1462 int err;
1463
1464 if (!rtnl_dev_link_net_capable(dev, t->net))
1465 return -EPERM;
1466
1467 if (data && data[IFLA_GRE_COLLECT_METADATA] && !t->collect_md) {
1468 NL_SET_ERR_MSG(extack,
1469 "Enabling collect_md on an existing device is not supported");
1470 return -EOPNOTSUPP;
1471 }
1472
1473 err = ipgre_newlink_encap_setup(dev, data);
1474 if (err)
1475 return err;
1476
1477 err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1478 if (err < 0)
1479 return err;
1480
1481 err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1482 if (err < 0)
1483 return err;
1484
1485 ip_tunnel_flags_copy(t->parms.i_flags, p.i_flags);
1486 ip_tunnel_flags_copy(t->parms.o_flags, p.o_flags);
1487
1488 ipgre_link_update(dev, !tb[IFLA_MTU]);
1489
1490 return 0;
1491}
1492
1493static int erspan_changelink(struct net_device *dev, struct nlattr *tb[],
1494 struct nlattr *data[],
1495 struct netlink_ext_ack *extack)
1496{
1497 struct ip_tunnel *t = netdev_priv(dev);
1498 struct ip_tunnel_parm_kern p;
1499 __u32 fwmark = t->fwmark;
1500 int err;
1501
1502 if (!rtnl_dev_link_net_capable(dev, t->net))
1503 return -EPERM;
1504
1505 if (data && data[IFLA_GRE_COLLECT_METADATA] && !t->collect_md) {
1506 NL_SET_ERR_MSG(extack,
1507 "Enabling collect_md on an existing device is not supported");
1508 return -EOPNOTSUPP;
1509 }
1510
1511 err = ipgre_newlink_encap_setup(dev, data);
1512 if (err)
1513 return err;
1514
1515 err = erspan_netlink_parms(dev, data, tb, &p, &fwmark);
1516 if (err < 0)
1517 return err;
1518
1519 err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1520 if (err < 0)
1521 return err;
1522
1523 ip_tunnel_flags_copy(t->parms.i_flags, p.i_flags);
1524 ip_tunnel_flags_copy(t->parms.o_flags, p.o_flags);
1525
1526 return 0;
1527}
1528
1529static size_t ipgre_get_size(const struct net_device *dev)
1530{
1531 return
1532 /* IFLA_GRE_LINK */
1533 nla_total_size(4) +
1534 /* IFLA_GRE_IFLAGS */
1535 nla_total_size(2) +
1536 /* IFLA_GRE_OFLAGS */
1537 nla_total_size(2) +
1538 /* IFLA_GRE_IKEY */
1539 nla_total_size(4) +
1540 /* IFLA_GRE_OKEY */
1541 nla_total_size(4) +
1542 /* IFLA_GRE_LOCAL */
1543 nla_total_size(4) +
1544 /* IFLA_GRE_REMOTE */
1545 nla_total_size(4) +
1546 /* IFLA_GRE_TTL */
1547 nla_total_size(1) +
1548 /* IFLA_GRE_TOS */
1549 nla_total_size(1) +
1550 /* IFLA_GRE_PMTUDISC */
1551 nla_total_size(1) +
1552 /* IFLA_GRE_ENCAP_TYPE */
1553 nla_total_size(2) +
1554 /* IFLA_GRE_ENCAP_FLAGS */
1555 nla_total_size(2) +
1556 /* IFLA_GRE_ENCAP_SPORT */
1557 nla_total_size(2) +
1558 /* IFLA_GRE_ENCAP_DPORT */
1559 nla_total_size(2) +
1560 /* IFLA_GRE_COLLECT_METADATA */
1561 nla_total_size(0) +
1562 /* IFLA_GRE_IGNORE_DF */
1563 nla_total_size(1) +
1564 /* IFLA_GRE_FWMARK */
1565 nla_total_size(4) +
1566 /* IFLA_GRE_ERSPAN_INDEX */
1567 nla_total_size(4) +
1568 /* IFLA_GRE_ERSPAN_VER */
1569 nla_total_size(1) +
1570 /* IFLA_GRE_ERSPAN_DIR */
1571 nla_total_size(1) +
1572 /* IFLA_GRE_ERSPAN_HWID */
1573 nla_total_size(2) +
1574 0;
1575}
1576
1577static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1578{
1579 const struct ip_tunnel *t = netdev_priv(dev);
1580 const struct ip_tunnel_parm_kern *p = &t->parms;
1581 IP_TUNNEL_DECLARE_FLAGS(o_flags);
1582
1583 ip_tunnel_flags_copy(o_flags, p->o_flags);
1584
1585 if (t->erspan_ver != 0 && !t->collect_md)
1586 __set_bit(IP_TUNNEL_KEY_BIT, o_flags);
1587
1588 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1589 nla_put_be16(skb, IFLA_GRE_IFLAGS,
1590 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1591 nla_put_be16(skb, IFLA_GRE_OFLAGS,
1592 gre_tnl_flags_to_gre_flags(o_flags)) ||
1593 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1594 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1595 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1596 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1597 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1598 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1599 nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1600 !!(p->iph.frag_off & htons(IP_DF))) ||
1601 nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1602 goto nla_put_failure;
1603
1604 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1605 t->encap.type) ||
1606 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1607 t->encap.sport) ||
1608 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1609 t->encap.dport) ||
1610 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1611 t->encap.flags))
1612 goto nla_put_failure;
1613
1614 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1615 goto nla_put_failure;
1616
1617 if (t->collect_md) {
1618 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1619 goto nla_put_failure;
1620 }
1621
1622 return 0;
1623
1624nla_put_failure:
1625 return -EMSGSIZE;
1626}
1627
1628static int erspan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1629{
1630 const struct ip_tunnel *t = netdev_priv(dev);
1631
1632 if (t->erspan_ver <= 2) {
1633 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1634 goto nla_put_failure;
1635
1636 if (t->erspan_ver == 1) {
1637 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1638 goto nla_put_failure;
1639 } else if (t->erspan_ver == 2) {
1640 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1641 goto nla_put_failure;
1642 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1643 goto nla_put_failure;
1644 }
1645 }
1646
1647 return ipgre_fill_info(skb, dev);
1648
1649nla_put_failure:
1650 return -EMSGSIZE;
1651}
1652
1653static void erspan_setup(struct net_device *dev)
1654{
1655 struct ip_tunnel *t = netdev_priv(dev);
1656
1657 ether_setup(dev);
1658 dev->max_mtu = 0;
1659 dev->netdev_ops = &erspan_netdev_ops;
1660 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1661 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1662 ip_tunnel_setup(dev, erspan_net_id);
1663 t->erspan_ver = 1;
1664}
1665
1666static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1667 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1668 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1669 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1670 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1671 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1672 [IFLA_GRE_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) },
1673 [IFLA_GRE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1674 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1675 [IFLA_GRE_TOS] = { .type = NLA_U8 },
1676 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 },
1677 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 },
1678 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 },
1679 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 },
1680 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 },
1681 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG },
1682 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 },
1683 [IFLA_GRE_FWMARK] = { .type = NLA_U32 },
1684 [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1685 [IFLA_GRE_ERSPAN_VER] = { .type = NLA_U8 },
1686 [IFLA_GRE_ERSPAN_DIR] = { .type = NLA_U8 },
1687 [IFLA_GRE_ERSPAN_HWID] = { .type = NLA_U16 },
1688};
1689
1690static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1691 .kind = "gre",
1692 .maxtype = IFLA_GRE_MAX,
1693 .policy = ipgre_policy,
1694 .priv_size = sizeof(struct ip_tunnel),
1695 .setup = ipgre_tunnel_setup,
1696 .validate = ipgre_tunnel_validate,
1697 .newlink = ipgre_newlink,
1698 .changelink = ipgre_changelink,
1699 .dellink = ip_tunnel_dellink,
1700 .get_size = ipgre_get_size,
1701 .fill_info = ipgre_fill_info,
1702 .get_link_net = ip_tunnel_get_link_net,
1703};
1704
1705static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1706 .kind = "gretap",
1707 .maxtype = IFLA_GRE_MAX,
1708 .policy = ipgre_policy,
1709 .priv_size = sizeof(struct ip_tunnel),
1710 .setup = ipgre_tap_setup,
1711 .validate = ipgre_tap_validate,
1712 .newlink = ipgre_newlink,
1713 .changelink = ipgre_changelink,
1714 .dellink = ip_tunnel_dellink,
1715 .get_size = ipgre_get_size,
1716 .fill_info = ipgre_fill_info,
1717 .get_link_net = ip_tunnel_get_link_net,
1718};
1719
1720static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1721 .kind = "erspan",
1722 .maxtype = IFLA_GRE_MAX,
1723 .policy = ipgre_policy,
1724 .priv_size = sizeof(struct ip_tunnel),
1725 .setup = erspan_setup,
1726 .validate = erspan_validate,
1727 .newlink = erspan_newlink,
1728 .changelink = erspan_changelink,
1729 .dellink = ip_tunnel_dellink,
1730 .get_size = ipgre_get_size,
1731 .fill_info = erspan_fill_info,
1732 .get_link_net = ip_tunnel_get_link_net,
1733};
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_rtnl(struct net *net,
1741 struct list_head *dev_to_kill)
1742{
1743 ip_tunnel_delete_net(net, gre_tap_net_id, &ipgre_tap_ops, dev_to_kill);
1744}
1745
1746static struct pernet_operations ipgre_tap_net_ops = {
1747 .init = ipgre_tap_init_net,
1748 .exit_rtnl = ipgre_tap_exit_rtnl,
1749 .id = &gre_tap_net_id,
1750 .size = sizeof(struct ip_tunnel_net),
1751};
1752
1753static int __net_init erspan_init_net(struct net *net)
1754{
1755 return ip_tunnel_init_net(net, erspan_net_id,
1756 &erspan_link_ops, "erspan0");
1757}
1758
1759static void __net_exit erspan_exit_rtnl(struct net *net,
1760 struct list_head *dev_to_kill)
1761{
1762 ip_tunnel_delete_net(net, erspan_net_id, &erspan_link_ops, dev_to_kill);
1763}
1764
1765static struct pernet_operations erspan_net_ops = {
1766 .init = erspan_init_net,
1767 .exit_rtnl = erspan_exit_rtnl,
1768 .id = &erspan_net_id,
1769 .size = sizeof(struct ip_tunnel_net),
1770};
1771
1772static int __init ipgre_init(void)
1773{
1774 int err;
1775
1776 pr_info("GRE over IPv4 tunneling driver\n");
1777
1778 err = register_pernet_device(&ipgre_net_ops);
1779 if (err < 0)
1780 return err;
1781
1782 err = register_pernet_device(&ipgre_tap_net_ops);
1783 if (err < 0)
1784 goto pnet_tap_failed;
1785
1786 err = register_pernet_device(&erspan_net_ops);
1787 if (err < 0)
1788 goto pnet_erspan_failed;
1789
1790 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1791 if (err < 0) {
1792 pr_info("%s: can't add protocol\n", __func__);
1793 goto add_proto_failed;
1794 }
1795
1796 err = rtnl_link_register(&ipgre_link_ops);
1797 if (err < 0)
1798 goto rtnl_link_failed;
1799
1800 err = rtnl_link_register(&ipgre_tap_ops);
1801 if (err < 0)
1802 goto tap_ops_failed;
1803
1804 err = rtnl_link_register(&erspan_link_ops);
1805 if (err < 0)
1806 goto erspan_link_failed;
1807
1808 return 0;
1809
1810erspan_link_failed:
1811 rtnl_link_unregister(&ipgre_tap_ops);
1812tap_ops_failed:
1813 rtnl_link_unregister(&ipgre_link_ops);
1814rtnl_link_failed:
1815 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1816add_proto_failed:
1817 unregister_pernet_device(&erspan_net_ops);
1818pnet_erspan_failed:
1819 unregister_pernet_device(&ipgre_tap_net_ops);
1820pnet_tap_failed:
1821 unregister_pernet_device(&ipgre_net_ops);
1822 return err;
1823}
1824
1825static void __exit ipgre_fini(void)
1826{
1827 rtnl_link_unregister(&ipgre_tap_ops);
1828 rtnl_link_unregister(&ipgre_link_ops);
1829 rtnl_link_unregister(&erspan_link_ops);
1830 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1831 unregister_pernet_device(&ipgre_tap_net_ops);
1832 unregister_pernet_device(&ipgre_net_ops);
1833 unregister_pernet_device(&erspan_net_ops);
1834}
1835
1836module_init(ipgre_init);
1837module_exit(ipgre_fini);
1838MODULE_DESCRIPTION("IPv4 GRE tunnels over IP library");
1839MODULE_LICENSE("GPL");
1840MODULE_ALIAS_RTNL_LINK("gre");
1841MODULE_ALIAS_RTNL_LINK("gretap");
1842MODULE_ALIAS_RTNL_LINK("erspan");
1843MODULE_ALIAS_NETDEV("gre0");
1844MODULE_ALIAS_NETDEV("gretap0");
1845MODULE_ALIAS_NETDEV("erspan0");