projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net/ipv6: allow sysctl to change link-local address generation mode
[linux-2.6-block.git] / net / ipv6 / ip6_gre.c
CommitLineData
c12b395a 1/*
2 * GRE over IPv6 protocol decoder.
3 *
4 * Authors: Dmitry Kozlov (xeb@mail.ru)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/capability.h>
16#include <linux/module.h>
17#include <linux/types.h>
18#include <linux/kernel.h>
19#include <linux/slab.h>
20#include <linux/uaccess.h>
21#include <linux/skbuff.h>
22#include <linux/netdevice.h>
23#include <linux/in.h>
24#include <linux/tcp.h>
25#include <linux/udp.h>
26#include <linux/if_arp.h>
c12b395a 27#include <linux/init.h>
28#include <linux/in6.h>
29#include <linux/inetdevice.h>
30#include <linux/igmp.h>
31#include <linux/netfilter_ipv4.h>
32#include <linux/etherdevice.h>
33#include <linux/if_ether.h>
34#include <linux/hash.h>
35#include <linux/if_tunnel.h>
36#include <linux/ip6_tunnel.h>
37
38#include <net/sock.h>
39#include <net/ip.h>
c5441932 40#include <net/ip_tunnels.h>
c12b395a 41#include <net/icmp.h>
42#include <net/protocol.h>
43#include <net/addrconf.h>
44#include <net/arp.h>
45#include <net/checksum.h>
46#include <net/dsfield.h>
47#include <net/inet_ecn.h>
48#include <net/xfrm.h>
49#include <net/net_namespace.h>
50#include <net/netns/generic.h>
51#include <net/rtnetlink.h>
52
53#include <net/ipv6.h>
54#include <net/ip6_fib.h>
55#include <net/ip6_route.h>
56#include <net/ip6_tunnel.h>
308edfdf 57#include <net/gre.h>
c12b395a 58
59
eccc1bb8 60static bool log_ecn_error = true;
61module_param(log_ecn_error, bool, 0644);
62MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
63
e87a8f24
JK		 64#define IP6_GRE_HASH_SIZE_SHIFT 5
65#define IP6_GRE_HASH_SIZE (1 << IP6_GRE_HASH_SIZE_SHIFT)
c12b395a 66
c7d03a00 67static unsigned int ip6gre_net_id __read_mostly;
c12b395a 68struct ip6gre_net {
e87a8f24 69 struct ip6_tnl __rcu *tunnels[4][IP6_GRE_HASH_SIZE];
c12b395a 70
71 struct net_device *fb_tunnel_dev;
72};
73
74static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
22f08069 75static struct rtnl_link_ops ip6gre_tap_ops __read_mostly;
c12b395a 76static int ip6gre_tunnel_init(struct net_device *dev);
77static void ip6gre_tunnel_setup(struct net_device *dev);
78static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
79static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
80
81/* Tunnel hash table */
82
83/*
84 4 hash tables:
85
86 3: (remote,local)
87 2: (remote,*)
88 1: (*,local)
89 0: (*,*)
90
91 We require exact key match i.e. if a key is present in packet
92 it will match only tunnel with the same key; if it is not present,
93 it will match only keyless tunnel.
94
95 All keysless packets, if not matched configured keyless tunnels
96 will match fallback tunnel.
97 */
98
e87a8f24 99#define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(IP6_GRE_HASH_SIZE - 1))
c12b395a 100static u32 HASH_ADDR(const struct in6_addr *addr)
101{
102 u32 hash = ipv6_addr_hash(addr);
103
e87a8f24 104 return hash_32(hash, IP6_GRE_HASH_SIZE_SHIFT);
c12b395a 105}
106
107#define tunnels_r_l tunnels[3]
108#define tunnels_r tunnels[2]
109#define tunnels_l tunnels[1]
110#define tunnels_wc tunnels[0]
c12b395a 111
c12b395a 112/* Given src, dst and key, find appropriate for input tunnel. */
113
114static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
115 const struct in6_addr *remote, const struct in6_addr *local,
116 __be32 key, __be16 gre_proto)
117{
118 struct net *net = dev_net(dev);
119 int link = dev->ifindex;
120 unsigned int h0 = HASH_ADDR(remote);
121 unsigned int h1 = HASH_KEY(key);
122 struct ip6_tnl *t, *cand = NULL;
123 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
124 int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
125 ARPHRD_ETHER : ARPHRD_IP6GRE;
126 int score, cand_score = 4;
127
e086cadc 128 for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
c12b395a 129 if (!ipv6_addr_equal(local, &t->parms.laddr) ||
130 !ipv6_addr_equal(remote, &t->parms.raddr) ||
131 key != t->parms.i_key ||
132 !(t->dev->flags & IFF_UP))
133 continue;
134
135 if (t->dev->type != ARPHRD_IP6GRE &&
136 t->dev->type != dev_type)
137 continue;
138
139 score = 0;
140 if (t->parms.link != link)
141 score |= 1;
142 if (t->dev->type != dev_type)
143 score |= 2;
144 if (score == 0)
145 return t;
146
147 if (score < cand_score) {
148 cand = t;
149 cand_score = score;
150 }
151 }
152
e086cadc 153 for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
c12b395a 154 if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
155 key != t->parms.i_key ||
156 !(t->dev->flags & IFF_UP))
157 continue;
158
159 if (t->dev->type != ARPHRD_IP6GRE &&
160 t->dev->type != dev_type)
161 continue;
162
163 score = 0;
164 if (t->parms.link != link)
165 score |= 1;
166 if (t->dev->type != dev_type)
167 score |= 2;
168 if (score == 0)
169 return t;
170
171 if (score < cand_score) {
172 cand = t;
173 cand_score = score;
174 }
175 }
176
e086cadc 177 for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
c12b395a 178 if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
179 (!ipv6_addr_equal(local, &t->parms.raddr) ||
180 !ipv6_addr_is_multicast(local))) ||
181 key != t->parms.i_key ||
182 !(t->dev->flags & IFF_UP))
183 continue;
184
185 if (t->dev->type != ARPHRD_IP6GRE &&
186 t->dev->type != dev_type)
187 continue;
188
189 score = 0;
190 if (t->parms.link != link)
191 score |= 1;
192 if (t->dev->type != dev_type)
193 score |= 2;
194 if (score == 0)
195 return t;
196
197 if (score < cand_score) {
198 cand = t;
199 cand_score = score;
200 }
201 }
202
e086cadc 203 for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
c12b395a 204 if (t->parms.i_key != key ||
205 !(t->dev->flags & IFF_UP))
206 continue;
207
208 if (t->dev->type != ARPHRD_IP6GRE &&
209 t->dev->type != dev_type)
210 continue;
211
212 score = 0;
213 if (t->parms.link != link)
214 score |= 1;
215 if (t->dev->type != dev_type)
216 score |= 2;
217 if (score == 0)
218 return t;
219
220 if (score < cand_score) {
221 cand = t;
222 cand_score = score;
223 }
224 }
225
53b24b8f 226 if (cand)
c12b395a 227 return cand;
228
229 dev = ign->fb_tunnel_dev;
230 if (dev->flags & IFF_UP)
231 return netdev_priv(dev);
232
233 return NULL;
234}
235
236static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
237 const struct __ip6_tnl_parm *p)
238{
239 const struct in6_addr *remote = &p->raddr;
240 const struct in6_addr *local = &p->laddr;
241 unsigned int h = HASH_KEY(p->i_key);
242 int prio = 0;
243
244 if (!ipv6_addr_any(local))
245 prio |= 1;
246 if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
247 prio |= 2;
248 h ^= HASH_ADDR(remote);
249 }
250
251 return &ign->tunnels[prio][h];
252}
253
254static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
255 const struct ip6_tnl *t)
256{
257 return __ip6gre_bucket(ign, &t->parms);
258}
259
260static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
261{
262 struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
263
264 rcu_assign_pointer(t->next, rtnl_dereference(*tp));
265 rcu_assign_pointer(*tp, t);
266}
267
268static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
269{
270 struct ip6_tnl __rcu **tp;
271 struct ip6_tnl *iter;
272
273 for (tp = ip6gre_bucket(ign, t);
274 (iter = rtnl_dereference(*tp)) != NULL;
275 tp = &iter->next) {
276 if (t == iter) {
277 rcu_assign_pointer(*tp, t->next);
278 break;
279 }
280 }
281}
282
283static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
284 const struct __ip6_tnl_parm *parms,
285 int type)
286{
287 const struct in6_addr *remote = &parms->raddr;
288 const struct in6_addr *local = &parms->laddr;
289 __be32 key = parms->i_key;
290 int link = parms->link;
291 struct ip6_tnl *t;
292 struct ip6_tnl __rcu **tp;
293 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
294
295 for (tp = __ip6gre_bucket(ign, parms);
296 (t = rtnl_dereference(*tp)) != NULL;
297 tp = &t->next)
298 if (ipv6_addr_equal(local, &t->parms.laddr) &&
299 ipv6_addr_equal(remote, &t->parms.raddr) &&
300 key == t->parms.i_key &&
301 link == t->parms.link &&
302 type == t->dev->type)
303 break;
304
305 return t;
306}
307
308static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
309 const struct __ip6_tnl_parm *parms, int create)
310{
311 struct ip6_tnl *t, *nt;
312 struct net_device *dev;
313 char name[IFNAMSIZ];
314 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
315
316 t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
cd0a0bd9
SK		 317 if (t && create)
318 return NULL;
c12b395a 319 if (t || !create)
320 return t;
321
322 if (parms->name[0])
323 strlcpy(name, parms->name, IFNAMSIZ);
324 else
325 strcpy(name, "ip6gre%d");
326
c835a677
TG		 327 dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
328 ip6gre_tunnel_setup);
c12b395a 329 if (!dev)
330 return NULL;
331
332 dev_net_set(dev, net);
333
334 nt = netdev_priv(dev);
335 nt->parms = *parms;
336 dev->rtnl_link_ops = &ip6gre_link_ops;
337
338 nt->dev = dev;
0bd87628 339 nt->net = dev_net(dev);
c12b395a 340 ip6gre_tnl_link_config(nt, 1);
341
342 if (register_netdevice(dev) < 0)
343 goto failed_free;
344
345 /* Can use a lockless transmit, unless we generate output sequences */
b45bd1d7 346 if (!(nt->parms.o_flags & TUNNEL_SEQ))
c12b395a 347 dev->features |= NETIF_F_LLTX;
348
349 dev_hold(dev);
350 ip6gre_tunnel_link(ign, nt);
351 return nt;
352
353failed_free:
354 free_netdev(dev);
355 return NULL;
356}
357
358static void ip6gre_tunnel_uninit(struct net_device *dev)
359{
22f08069
ND		 360 struct ip6_tnl *t = netdev_priv(dev);
361 struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
c12b395a 362
22f08069 363 ip6gre_tunnel_unlink(ign, t);
607f725f 364 dst_cache_reset(&t->dst_cache);
c12b395a 365 dev_put(dev);
366}
367
368
369static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
370 u8 type, u8 code, int offset, __be32 info)
371{
372 const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)skb->data;
b87fb39e
ED		 373 __be16 *p = (__be16 *)(skb->data + offset);
374 int grehlen = offset + 4;
c12b395a 375 struct ip6_tnl *t;
376 __be16 flags;
377
378 flags = p[0];
379 if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
380 if (flags&(GRE_VERSION|GRE_ROUTING))
381 return;
382 if (flags&GRE_KEY) {
383 grehlen += 4;
384 if (flags&GRE_CSUM)
385 grehlen += 4;
386 }
387 }
388
389 /* If only 8 bytes returned, keyed message will be dropped here */
b87fb39e 390 if (!pskb_may_pull(skb, grehlen))
c12b395a 391 return;
b87fb39e
ED		 392 ipv6h = (const struct ipv6hdr *)skb->data;
393 p = (__be16 *)(skb->data + offset);
c12b395a 394
c12b395a 395 t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
396 flags & GRE_KEY ?
397 *(((__be32 *)p) + (grehlen / 4) - 1) : 0,
398 p[1]);
63159f29 399 if (!t)
0c5794a6 400 return;
c12b395a 401
402 switch (type) {
403 __u32 teli;
404 struct ipv6_tlv_tnl_enc_lim *tel;
405 __u32 mtu;
406 case ICMPV6_DEST_UNREACH:
a46496ce
MB		 407 net_dbg_ratelimited("%s: Path to destination invalid or inactive!\n",
408 t->parms.name);
c12b395a 409 break;
410 case ICMPV6_TIME_EXCEED:
411 if (code == ICMPV6_EXC_HOPLIMIT) {
a46496ce
MB		 412 net_dbg_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
413 t->parms.name);
c12b395a 414 }
415 break;
416 case ICMPV6_PARAMPROB:
417 teli = 0;
418 if (code == ICMPV6_HDR_FIELD)
419 teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
420
d1e158e2 421 if (teli && teli == be32_to_cpu(info) - 2) {
c12b395a 422 tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
423 if (tel->encap_limit == 0) {
a46496ce
MB		 424 net_dbg_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
425 t->parms.name);
c12b395a 426 }
427 } else {
a46496ce
MB		 428 net_dbg_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
429 t->parms.name);
c12b395a 430 }
431 break;
432 case ICMPV6_PKT_TOOBIG:
d1e158e2 433 mtu = be32_to_cpu(info) - offset;
c12b395a 434 if (mtu < IPV6_MIN_MTU)
435 mtu = IPV6_MIN_MTU;
436 t->dev->mtu = mtu;
437 break;
438 }
439
440 if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
441 t->err_count++;
442 else
443 t->err_count = 1;
444 t->err_time = jiffies;
c12b395a 445}
446
308edfdf 447static int ip6gre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi)
c12b395a 448{
449 const struct ipv6hdr *ipv6h;
c12b395a 450 struct ip6_tnl *tunnel;
c12b395a 451
452 ipv6h = ipv6_hdr(skb);
c12b395a 453 tunnel = ip6gre_tunnel_lookup(skb->dev,
308edfdf
TH		 454 &ipv6h->saddr, &ipv6h->daddr, tpi->key,
455 tpi->proto);
c12b395a 456 if (tunnel) {
308edfdf 457 ip6_tnl_rcv(tunnel, skb, tpi, NULL, false);
c12b395a 458
308edfdf
TH		 459 return PACKET_RCVD;
460 }
eccc1bb8 461
308edfdf
TH		 462 return PACKET_REJECT;
463}
eccc1bb8 464
308edfdf
TH		 465static int gre_rcv(struct sk_buff *skb)
466{
467 struct tnl_ptk_info tpi;
468 bool csum_err = false;
469 int hdr_len;
eccc1bb8 470
e582615a 471 hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IPV6), 0);
f132ae7c 472 if (hdr_len < 0)
308edfdf 473 goto drop;
c12b395a 474
308edfdf
TH		 475 if (iptunnel_pull_header(skb, hdr_len, tpi.proto, false))
476 goto drop;
c12b395a 477
308edfdf 478 if (ip6gre_rcv(skb, &tpi) == PACKET_RCVD)
c12b395a 479 return 0;
c12b395a 480
308edfdf 481 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
c12b395a 482drop:
c12b395a 483 kfree_skb(skb);
484 return 0;
485}
486
b05229f4 487static int gre_handle_offloads(struct sk_buff *skb, bool csum)
c12b395a 488{
b05229f4
TH		 489 return iptunnel_handle_offloads(skb,
490 csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
c12b395a 491}
492
b05229f4
TH		 493static netdev_tx_t __gre6_xmit(struct sk_buff *skb,
494 struct net_device *dev, __u8 dsfield,
495 struct flowi6 *fl6, int encap_limit,
496 __u32 *pmtu, __be16 proto)
c12b395a 497{
c12b395a 498 struct ip6_tnl *tunnel = netdev_priv(dev);
b05229f4
TH		 499 __be16 protocol = (dev->type == ARPHRD_ETHER) ?
500 htons(ETH_P_TEB) : proto;
c12b395a 501
502 if (dev->type == ARPHRD_ETHER)
503 IPCB(skb)->flags = 0;
504
b05229f4
TH		 505 if (dev->header_ops && dev->type == ARPHRD_IP6GRE)
506 fl6->daddr = ((struct ipv6hdr *)skb->data)->daddr;
507 else
c12b395a 508 fl6->daddr = tunnel->parms.raddr;
c12b395a 509
b05229f4
TH		 510 if (tunnel->parms.o_flags & TUNNEL_SEQ)
511 tunnel->o_seqno++;
c12b395a 512
b05229f4
TH		 513 /* Push GRE header. */
514 gre_build_header(skb, tunnel->tun_hlen, tunnel->parms.o_flags,
515 protocol, tunnel->parms.o_key, htonl(tunnel->o_seqno));
c12b395a 516
b05229f4
TH		 517 return ip6_tnl_xmit(skb, dev, dsfield, fl6, encap_limit, pmtu,
518 NEXTHDR_GRE);
c12b395a 519}
520
521static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
522{
523 struct ip6_tnl *t = netdev_priv(dev);
524 const struct iphdr *iph = ip_hdr(skb);
525 int encap_limit = -1;
526 struct flowi6 fl6;
527 __u8 dsfield;
528 __u32 mtu;
529 int err;
530
5146d1f1
BH		 531 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
532
c12b395a 533 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
534 encap_limit = t->parms.encap_limit;
535
536 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
c12b395a 537
538 dsfield = ipv4_get_dsfield(iph);
539
540 if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
541 fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
542 & IPV6_TCLASS_MASK;
543 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
544 fl6.flowi6_mark = skb->mark;
545
e2d118a1
LC		 546 fl6.flowi6_uid = sock_net_uid(dev_net(dev), NULL);
547
b05229f4
TH		 548 err = gre_handle_offloads(skb, !!(t->parms.o_flags & TUNNEL_CSUM));
549 if (err)
550 return -1;
551
552 err = __gre6_xmit(skb, dev, dsfield, &fl6, encap_limit, &mtu,
553 skb->protocol);
c12b395a 554 if (err != 0) {
555 /* XXX: send ICMP error even if DF is not set. */
556 if (err == -EMSGSIZE)
557 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
558 htonl(mtu));
559 return -1;
560 }
561
562 return 0;
563}
564
565static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
566{
567 struct ip6_tnl *t = netdev_priv(dev);
568 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
569 int encap_limit = -1;
570 __u16 offset;
571 struct flowi6 fl6;
572 __u8 dsfield;
573 __u32 mtu;
574 int err;
575
576 if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
577 return -1;
578
579 offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
580 if (offset > 0) {
581 struct ipv6_tlv_tnl_enc_lim *tel;
582 tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
583 if (tel->encap_limit == 0) {
584 icmpv6_send(skb, ICMPV6_PARAMPROB,
585 ICMPV6_HDR_FIELD, offset + 2);
586 return -1;
587 }
588 encap_limit = tel->encap_limit - 1;
589 } else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
590 encap_limit = t->parms.encap_limit;
591
592 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
c12b395a 593
594 dsfield = ipv6_get_dsfield(ipv6h);
595 if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
596 fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
597 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
3308de2b 598 fl6.flowlabel |= ip6_flowlabel(ipv6h);
c12b395a 599 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
600 fl6.flowi6_mark = skb->mark;
601
e2d118a1
LC		 602 fl6.flowi6_uid = sock_net_uid(dev_net(dev), NULL);
603
b05229f4
TH		 604 if (gre_handle_offloads(skb, !!(t->parms.o_flags & TUNNEL_CSUM)))
605 return -1;
606
607 err = __gre6_xmit(skb, dev, dsfield, &fl6, encap_limit,
608 &mtu, skb->protocol);
c12b395a 609 if (err != 0) {
610 if (err == -EMSGSIZE)
611 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
612 return -1;
613 }
614
615 return 0;
616}
617
618/**
619 * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
620 * @t: the outgoing tunnel device
621 * @hdr: IPv6 header from the incoming packet
622 *
623 * Description:
624 * Avoid trivial tunneling loop by checking that tunnel exit-point
625 * doesn't match source of incoming packet.
626 *
627 * Return:
628 * 1 if conflict,
629 * 0 else
630 **/
631
632static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
633 const struct ipv6hdr *hdr)
634{
635 return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
636}
637
638static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
639{
640 struct ip6_tnl *t = netdev_priv(dev);
641 int encap_limit = -1;
642 struct flowi6 fl6;
643 __u32 mtu;
644 int err;
645
646 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
647 encap_limit = t->parms.encap_limit;
648
649 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
c12b395a 650
b05229f4
TH		 651 err = gre_handle_offloads(skb, !!(t->parms.o_flags & TUNNEL_CSUM));
652 if (err)
653 return err;
654
655 err = __gre6_xmit(skb, dev, 0, &fl6, encap_limit, &mtu, skb->protocol);
c12b395a 656
657 return err;
658}
659
660static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
661 struct net_device *dev)
662{
663 struct ip6_tnl *t = netdev_priv(dev);
664 struct net_device_stats *stats = &t->dev->stats;
665 int ret;
666
d5005140 667 if (!ip6_tnl_xmit_ctl(t, &t->parms.laddr, &t->parms.raddr))
41ab3e31 668 goto tx_err;
c12b395a 669
670 switch (skb->protocol) {
671 case htons(ETH_P_IP):
672 ret = ip6gre_xmit_ipv4(skb, dev);
673 break;
674 case htons(ETH_P_IPV6):
675 ret = ip6gre_xmit_ipv6(skb, dev);
676 break;
677 default:
678 ret = ip6gre_xmit_other(skb, dev);
679 break;
680 }
681
682 if (ret < 0)
683 goto tx_err;
684
685 return NETDEV_TX_OK;
686
687tx_err:
688 stats->tx_errors++;
689 stats->tx_dropped++;
690 kfree_skb(skb);
691 return NETDEV_TX_OK;
692}
693
694static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
695{
696 struct net_device *dev = t->dev;
697 struct __ip6_tnl_parm *p = &t->parms;
698 struct flowi6 *fl6 = &t->fl.u.ip6;
db2ec95d 699 int t_hlen;
c12b395a 700
701 if (dev->type != ARPHRD_ETHER) {
702 memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
703 memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
704 }
705
706 /* Set up flowi template */
707 fl6->saddr = p->laddr;
708 fl6->daddr = p->raddr;
709 fl6->flowi6_oif = p->link;
710 fl6->flowlabel = 0;
252f3f5a 711 fl6->flowi6_proto = IPPROTO_GRE;
c12b395a 712
713 if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
714 fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
715 if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
716 fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
717
718 p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
719 p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
720
721 if (p->flags&IP6_TNL_F_CAP_XMIT &&
722 p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
723 dev->flags |= IFF_POINTOPOINT;
724 else
725 dev->flags &= ~IFF_POINTOPOINT;
726
db2ec95d
TH		 727 t->tun_hlen = gre_calc_hlen(t->parms.o_flags);
728
1faf3d9f 729 t->hlen = t->encap_hlen + t->tun_hlen;
db2ec95d
TH		 730
731 t_hlen = t->hlen + sizeof(struct ipv6hdr);
c12b395a 732
733 if (p->flags & IP6_TNL_F_CAP_XMIT) {
734 int strict = (ipv6_addr_type(&p->raddr) &
735 (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
736
22f08069 737 struct rt6_info *rt = rt6_lookup(t->net,
c12b395a 738 &p->raddr, &p->laddr,
739 p->link, strict);
740
63159f29 741 if (!rt)
c12b395a 742 return;
743
744 if (rt->dst.dev) {
db2ec95d
TH		 745 dev->hard_header_len = rt->dst.dev->hard_header_len +
746 t_hlen;
c12b395a 747
748 if (set_mtu) {
db2ec95d 749 dev->mtu = rt->dst.dev->mtu - t_hlen;
c12b395a 750 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
751 dev->mtu -= 8;
a9e242ca
AD		 752 if (dev->type == ARPHRD_ETHER)
753 dev->mtu -= ETH_HLEN;
c12b395a 754
755 if (dev->mtu < IPV6_MIN_MTU)
756 dev->mtu = IPV6_MIN_MTU;
757 }
758 }
94e187c0 759 ip6_rt_put(rt);
c12b395a 760 }
c12b395a 761}
762
763static int ip6gre_tnl_change(struct ip6_tnl *t,
764 const struct __ip6_tnl_parm *p, int set_mtu)
765{
766 t->parms.laddr = p->laddr;
767 t->parms.raddr = p->raddr;
768 t->parms.flags = p->flags;
769 t->parms.hop_limit = p->hop_limit;
770 t->parms.encap_limit = p->encap_limit;
771 t->parms.flowinfo = p->flowinfo;
772 t->parms.link = p->link;
773 t->parms.proto = p->proto;
774 t->parms.i_key = p->i_key;
775 t->parms.o_key = p->o_key;
776 t->parms.i_flags = p->i_flags;
777 t->parms.o_flags = p->o_flags;
607f725f 778 dst_cache_reset(&t->dst_cache);
c12b395a 779 ip6gre_tnl_link_config(t, set_mtu);
780 return 0;
781}
782
783static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
784 const struct ip6_tnl_parm2 *u)
785{
786 p->laddr = u->laddr;
787 p->raddr = u->raddr;
788 p->flags = u->flags;
789 p->hop_limit = u->hop_limit;
790 p->encap_limit = u->encap_limit;
791 p->flowinfo = u->flowinfo;
792 p->link = u->link;
793 p->i_key = u->i_key;
794 p->o_key = u->o_key;
f41fe3c2
TH		 795 p->i_flags = gre_flags_to_tnl_flags(u->i_flags);
796 p->o_flags = gre_flags_to_tnl_flags(u->o_flags);
c12b395a 797 memcpy(p->name, u->name, sizeof(u->name));
798}
799
800static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
801 const struct __ip6_tnl_parm *p)
802{
803 u->proto = IPPROTO_GRE;
804 u->laddr = p->laddr;
805 u->raddr = p->raddr;
806 u->flags = p->flags;
807 u->hop_limit = p->hop_limit;
808 u->encap_limit = p->encap_limit;
809 u->flowinfo = p->flowinfo;
810 u->link = p->link;
811 u->i_key = p->i_key;
812 u->o_key = p->o_key;
f41fe3c2
TH		 813 u->i_flags = gre_tnl_flags_to_gre_flags(p->i_flags);
814 u->o_flags = gre_tnl_flags_to_gre_flags(p->o_flags);
c12b395a 815 memcpy(u->name, p->name, sizeof(u->name));
816}
817
818static int ip6gre_tunnel_ioctl(struct net_device *dev,
819 struct ifreq *ifr, int cmd)
820{
821 int err = 0;
822 struct ip6_tnl_parm2 p;
823 struct __ip6_tnl_parm p1;
22f08069
ND		 824 struct ip6_tnl *t = netdev_priv(dev);
825 struct net *net = t->net;
c12b395a 826 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
827
308edfdf
TH		 828 memset(&p1, 0, sizeof(p1));
829
c12b395a 830 switch (cmd) {
831 case SIOCGETTUNNEL:
c12b395a 832 if (dev == ign->fb_tunnel_dev) {
833 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
834 err = -EFAULT;
835 break;
836 }
837 ip6gre_tnl_parm_from_user(&p1, &p);
838 t = ip6gre_tunnel_locate(net, &p1, 0);
63159f29 839 if (!t)
22f08069 840 t = netdev_priv(dev);
c12b395a 841 }
5dbd5068 842 memset(&p, 0, sizeof(p));
c12b395a 843 ip6gre_tnl_parm_to_user(&p, &t->parms);
844 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
845 err = -EFAULT;
846 break;
847
848 case SIOCADDTUNNEL:
849 case SIOCCHGTUNNEL:
850 err = -EPERM;
af31f412 851 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
c12b395a 852 goto done;
853
854 err = -EFAULT;
855 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
856 goto done;
857
858 err = -EINVAL;
859 if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
860 goto done;
861
862 if (!(p.i_flags&GRE_KEY))
863 p.i_key = 0;
864 if (!(p.o_flags&GRE_KEY))
865 p.o_key = 0;
866
867 ip6gre_tnl_parm_from_user(&p1, &p);
868 t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
869
870 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
53b24b8f 871 if (t) {
c12b395a 872 if (t->dev != dev) {
873 err = -EEXIST;
874 break;
875 }
876 } else {
877 t = netdev_priv(dev);
878
879 ip6gre_tunnel_unlink(ign, t);
880 synchronize_net();
881 ip6gre_tnl_change(t, &p1, 1);
882 ip6gre_tunnel_link(ign, t);
883 netdev_state_change(dev);
884 }
885 }
886
887 if (t) {
888 err = 0;
889
5dbd5068 890 memset(&p, 0, sizeof(p));
c12b395a 891 ip6gre_tnl_parm_to_user(&p, &t->parms);
892 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
893 err = -EFAULT;
894 } else
895 err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
896 break;
897
898 case SIOCDELTUNNEL:
899 err = -EPERM;
af31f412 900 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
c12b395a 901 goto done;
902
903 if (dev == ign->fb_tunnel_dev) {
904 err = -EFAULT;
905 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
906 goto done;
907 err = -ENOENT;
908 ip6gre_tnl_parm_from_user(&p1, &p);
909 t = ip6gre_tunnel_locate(net, &p1, 0);
63159f29 910 if (!t)
c12b395a 911 goto done;
912 err = -EPERM;
913 if (t == netdev_priv(ign->fb_tunnel_dev))
914 goto done;
915 dev = t->dev;
916 }
917 unregister_netdevice(dev);
918 err = 0;
919 break;
920
921 default:
922 err = -EINVAL;
923 }
924
925done:
926 return err;
927}
928
c12b395a 929static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
930 unsigned short type,
931 const void *daddr, const void *saddr, unsigned int len)
932{
933 struct ip6_tnl *t = netdev_priv(dev);
934 struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
935 __be16 *p = (__be16 *)(ipv6h+1);
936
cb1ce2ef
TH		 937 ip6_flow_hdr(ipv6h, 0,
938 ip6_make_flowlabel(dev_net(dev), skb,
42240901 939 t->fl.u.ip6.flowlabel, true,
67800f9b 940 &t->fl.u.ip6));
c12b395a 941 ipv6h->hop_limit = t->parms.hop_limit;
942 ipv6h->nexthdr = NEXTHDR_GRE;
943 ipv6h->saddr = t->parms.laddr;
944 ipv6h->daddr = t->parms.raddr;
945
946 p[0] = t->parms.o_flags;
947 p[1] = htons(type);
948
949 /*
950 * Set the source hardware address.
951 */
952
953 if (saddr)
954 memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
955 if (daddr)
956 memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
957 if (!ipv6_addr_any(&ipv6h->daddr))
958 return t->hlen;
959
960 return -t->hlen;
961}
962
c12b395a 963static const struct header_ops ip6gre_header_ops = {
964 .create = ip6gre_header,
c12b395a 965};
966
967static const struct net_device_ops ip6gre_netdev_ops = {
968 .ndo_init = ip6gre_tunnel_init,
969 .ndo_uninit = ip6gre_tunnel_uninit,
970 .ndo_start_xmit = ip6gre_tunnel_xmit,
971 .ndo_do_ioctl = ip6gre_tunnel_ioctl,
b05229f4 972 .ndo_change_mtu = ip6_tnl_change_mtu,
f61dd388 973 .ndo_get_stats64 = ip_tunnel_get_stats64,
ecf2c06a 974 .ndo_get_iflink = ip6_tnl_get_iflink,
c12b395a 975};
976
977static void ip6gre_dev_free(struct net_device *dev)
978{
cdf3464e
MKL		 979 struct ip6_tnl *t = netdev_priv(dev);
980
607f725f 981 dst_cache_destroy(&t->dst_cache);
c12b395a 982 free_percpu(dev->tstats);
983 free_netdev(dev);
984}
985
986static void ip6gre_tunnel_setup(struct net_device *dev)
987{
c12b395a 988 dev->netdev_ops = &ip6gre_netdev_ops;
989 dev->destructor = ip6gre_dev_free;
990
991 dev->type = ARPHRD_IP6GRE;
b05229f4 992
c12b395a 993 dev->flags |= IFF_NOARP;
c12b395a 994 dev->addr_len = sizeof(struct in6_addr);
02875878 995 netif_keep_dst(dev);
c12b395a 996}
997
a3c119d3 998static int ip6gre_tunnel_init_common(struct net_device *dev)
c12b395a 999{
1000 struct ip6_tnl *tunnel;
cdf3464e 1001 int ret;
b05229f4 1002 int t_hlen;
c12b395a 1003
1004 tunnel = netdev_priv(dev);
1005
1006 tunnel->dev = dev;
0bd87628 1007 tunnel->net = dev_net(dev);
c12b395a 1008 strcpy(tunnel->parms.name, dev->name);
1009
a3c119d3
MKL		 1010 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1011 if (!dev->tstats)
1012 return -ENOMEM;
1013
607f725f 1014 ret = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL);
cdf3464e
MKL		 1015 if (ret) {
1016 free_percpu(dev->tstats);
1017 dev->tstats = NULL;
1018 return ret;
1019 }
1020
b05229f4 1021 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
1faf3d9f 1022 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
db2ec95d 1023 t_hlen = tunnel->hlen + sizeof(struct ipv6hdr);
b05229f4 1024
db2ec95d
TH		 1025 dev->hard_header_len = LL_MAX_HEADER + t_hlen;
1026 dev->mtu = ETH_DATA_LEN - t_hlen;
1b227e53
HY		 1027 if (dev->type == ARPHRD_ETHER)
1028 dev->mtu -= ETH_HLEN;
b05229f4
TH		 1029 if (!(tunnel->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1030 dev->mtu -= 8;
1031
a3c119d3
MKL		 1032 return 0;
1033}
1034
1035static int ip6gre_tunnel_init(struct net_device *dev)
1036{
1037 struct ip6_tnl *tunnel;
1038 int ret;
1039
1040 ret = ip6gre_tunnel_init_common(dev);
1041 if (ret)
1042 return ret;
1043
1044 tunnel = netdev_priv(dev);
1045
c12b395a 1046 memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1047 memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1048
1049 if (ipv6_addr_any(&tunnel->parms.raddr))
1050 dev->header_ops = &ip6gre_header_ops;
1051
c12b395a 1052 return 0;
1053}
1054
1055static void ip6gre_fb_tunnel_init(struct net_device *dev)
1056{
1057 struct ip6_tnl *tunnel = netdev_priv(dev);
1058
1059 tunnel->dev = dev;
0bd87628 1060 tunnel->net = dev_net(dev);
c12b395a 1061 strcpy(tunnel->parms.name, dev->name);
1062
1063 tunnel->hlen = sizeof(struct ipv6hdr) + 4;
1064
1065 dev_hold(dev);
1066}
1067
1068
1069static struct inet6_protocol ip6gre_protocol __read_mostly = {
308edfdf 1070 .handler = gre_rcv,
c12b395a 1071 .err_handler = ip6gre_err,
1072 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1073};
1074
22f08069 1075static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
c12b395a 1076{
22f08069
ND		 1077 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1078 struct net_device *dev, *aux;
c12b395a 1079 int prio;
1080
22f08069
ND		 1081 for_each_netdev_safe(net, dev, aux)
1082 if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1083 dev->rtnl_link_ops == &ip6gre_tap_ops)
1084 unregister_netdevice_queue(dev, head);
1085
c12b395a 1086 for (prio = 0; prio < 4; prio++) {
1087 int h;
e87a8f24 1088 for (h = 0; h < IP6_GRE_HASH_SIZE; h++) {
c12b395a 1089 struct ip6_tnl *t;
1090
1091 t = rtnl_dereference(ign->tunnels[prio][h]);
1092
53b24b8f 1093 while (t) {
22f08069
ND		 1094 /* If dev is in the same netns, it has already
1095 * been added to the list by the previous loop.
1096 */
1097 if (!net_eq(dev_net(t->dev), net))
1098 unregister_netdevice_queue(t->dev,
1099 head);
c12b395a 1100 t = rtnl_dereference(t->next);
1101 }
1102 }
1103 }
1104}
1105
1106static int __net_init ip6gre_init_net(struct net *net)
1107{
1108 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1109 int err;
1110
1111 ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
c835a677
TG		 1112 NET_NAME_UNKNOWN,
1113 ip6gre_tunnel_setup);
c12b395a 1114 if (!ign->fb_tunnel_dev) {
1115 err = -ENOMEM;
1116 goto err_alloc_dev;
1117 }
1118 dev_net_set(ign->fb_tunnel_dev, net);
22f08069
ND		 1119 /* FB netdevice is special: we have one, and only one per netns.
1120 * Allowing to move it to another netns is clearly unsafe.
1121 */
1122 ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1123
c12b395a 1124
1125 ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1126 ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1127
1128 err = register_netdev(ign->fb_tunnel_dev);
1129 if (err)
1130 goto err_reg_dev;
1131
1132 rcu_assign_pointer(ign->tunnels_wc[0],
1133 netdev_priv(ign->fb_tunnel_dev));
1134 return 0;
1135
1136err_reg_dev:
1137 ip6gre_dev_free(ign->fb_tunnel_dev);
1138err_alloc_dev:
1139 return err;
1140}
1141
1142static void __net_exit ip6gre_exit_net(struct net *net)
1143{
c12b395a 1144 LIST_HEAD(list);
1145
c12b395a 1146 rtnl_lock();
22f08069 1147 ip6gre_destroy_tunnels(net, &list);
c12b395a 1148 unregister_netdevice_many(&list);
1149 rtnl_unlock();
1150}
1151
1152static struct pernet_operations ip6gre_net_ops = {
1153 .init = ip6gre_init_net,
1154 .exit = ip6gre_exit_net,
1155 .id = &ip6gre_net_id,
1156 .size = sizeof(struct ip6gre_net),
1157};
1158
1159static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1160{
1161 __be16 flags;
1162
1163 if (!data)
1164 return 0;
1165
1166 flags = 0;
1167 if (data[IFLA_GRE_IFLAGS])
1168 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1169 if (data[IFLA_GRE_OFLAGS])
1170 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1171 if (flags & (GRE_VERSION|GRE_ROUTING))
1172 return -EINVAL;
1173
1174 return 0;
1175}
1176
1177static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1178{
1179 struct in6_addr daddr;
1180
1181 if (tb[IFLA_ADDRESS]) {
1182 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1183 return -EINVAL;
1184 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1185 return -EADDRNOTAVAIL;
1186 }
1187
1188 if (!data)
1189 goto out;
1190
1191 if (data[IFLA_GRE_REMOTE]) {
67b61f6c 1192 daddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
c12b395a 1193 if (ipv6_addr_any(&daddr))
1194 return -EINVAL;
1195 }
1196
1197out:
1198 return ip6gre_tunnel_validate(tb, data);
1199}
1200
1201
1202static void ip6gre_netlink_parms(struct nlattr *data[],
1203 struct __ip6_tnl_parm *parms)
1204{
1205 memset(parms, 0, sizeof(*parms));
1206
1207 if (!data)
1208 return;
1209
1210 if (data[IFLA_GRE_LINK])
1211 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1212
1213 if (data[IFLA_GRE_IFLAGS])
f41fe3c2
TH		 1214 parms->i_flags = gre_flags_to_tnl_flags(
1215 nla_get_be16(data[IFLA_GRE_IFLAGS]));
c12b395a 1216
1217 if (data[IFLA_GRE_OFLAGS])
f41fe3c2
TH		 1218 parms->o_flags = gre_flags_to_tnl_flags(
1219 nla_get_be16(data[IFLA_GRE_OFLAGS]));
c12b395a 1220
1221 if (data[IFLA_GRE_IKEY])
1222 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1223
1224 if (data[IFLA_GRE_OKEY])
1225 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1226
1227 if (data[IFLA_GRE_LOCAL])
67b61f6c 1228 parms->laddr = nla_get_in6_addr(data[IFLA_GRE_LOCAL]);
c12b395a 1229
1230 if (data[IFLA_GRE_REMOTE])
67b61f6c 1231 parms->raddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
c12b395a 1232
1233 if (data[IFLA_GRE_TTL])
1234 parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1235
1236 if (data[IFLA_GRE_ENCAP_LIMIT])
1237 parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1238
1239 if (data[IFLA_GRE_FLOWINFO])
c2675de4 1240 parms->flowinfo = nla_get_be32(data[IFLA_GRE_FLOWINFO]);
c12b395a 1241
1242 if (data[IFLA_GRE_FLAGS])
1243 parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1244}
1245
1246static int ip6gre_tap_init(struct net_device *dev)
1247{
1248 struct ip6_tnl *tunnel;
a3c119d3 1249 int ret;
c12b395a 1250
a3c119d3
MKL		 1251 ret = ip6gre_tunnel_init_common(dev);
1252 if (ret)
1253 return ret;
c12b395a 1254
0a46baaf
SC		 1255 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1256
a3c119d3 1257 tunnel = netdev_priv(dev);
c12b395a 1258
1259 ip6gre_tnl_link_config(tunnel, 1);
1260
c12b395a 1261 return 0;
1262}
1263
1264static const struct net_device_ops ip6gre_tap_netdev_ops = {
1265 .ndo_init = ip6gre_tap_init,
1266 .ndo_uninit = ip6gre_tunnel_uninit,
1267 .ndo_start_xmit = ip6gre_tunnel_xmit,
1268 .ndo_set_mac_address = eth_mac_addr,
1269 .ndo_validate_addr = eth_validate_addr,
b05229f4 1270 .ndo_change_mtu = ip6_tnl_change_mtu,
f61dd388 1271 .ndo_get_stats64 = ip_tunnel_get_stats64,
ecf2c06a 1272 .ndo_get_iflink = ip6_tnl_get_iflink,
c12b395a 1273};
1274
ac4eb009
AD		 1275#define GRE6_FEATURES (NETIF_F_SG | \
1276 NETIF_F_FRAGLIST | \
1277 NETIF_F_HIGHDMA | \
1278 NETIF_F_HW_CSUM)
1279
c12b395a 1280static void ip6gre_tap_setup(struct net_device *dev)
1281{
1282
1283 ether_setup(dev);
1284
1285 dev->netdev_ops = &ip6gre_tap_netdev_ops;
1286 dev->destructor = ip6gre_dev_free;
1287
c12b395a 1288 dev->features |= NETIF_F_NETNS_LOCAL;
d13b161c 1289 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
0a46baaf 1290 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
c12b395a 1291}
1292
1faf3d9f
TH		 1293static bool ip6gre_netlink_encap_parms(struct nlattr *data[],
1294 struct ip_tunnel_encap *ipencap)
1295{
1296 bool ret = false;
1297
1298 memset(ipencap, 0, sizeof(*ipencap));
1299
1300 if (!data)
1301 return ret;
1302
1303 if (data[IFLA_GRE_ENCAP_TYPE]) {
1304 ret = true;
1305 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1306 }
1307
1308 if (data[IFLA_GRE_ENCAP_FLAGS]) {
1309 ret = true;
1310 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1311 }
1312
1313 if (data[IFLA_GRE_ENCAP_SPORT]) {
1314 ret = true;
1315 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1316 }
1317
1318 if (data[IFLA_GRE_ENCAP_DPORT]) {
1319 ret = true;
1320 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1321 }
1322
1323 return ret;
1324}
1325
c12b395a 1326static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1327 struct nlattr *tb[], struct nlattr *data[])
1328{
1329 struct ip6_tnl *nt;
1330 struct net *net = dev_net(dev);
1331 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1faf3d9f 1332 struct ip_tunnel_encap ipencap;
c12b395a 1333 int err;
1334
1335 nt = netdev_priv(dev);
1faf3d9f
TH		 1336
1337 if (ip6gre_netlink_encap_parms(data, &ipencap)) {
1338 int err = ip6_tnl_encap_setup(nt, &ipencap);
1339
1340 if (err < 0)
1341 return err;
1342 }
1343
c12b395a 1344 ip6gre_netlink_parms(data, &nt->parms);
1345
1346 if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1347 return -EEXIST;
1348
1349 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1350 eth_hw_addr_random(dev);
1351
1352 nt->dev = dev;
0bd87628 1353 nt->net = dev_net(dev);
c12b395a 1354 ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1355
ac4eb009
AD		 1356 dev->features |= GRE6_FEATURES;
1357 dev->hw_features |= GRE6_FEATURES;
1358
b45bd1d7 1359 if (!(nt->parms.o_flags & TUNNEL_SEQ)) {
6a553681
AD		 1360 /* TCP offload with GRE SEQ is not supported, nor
1361 * can we support 2 levels of outer headers requiring
1362 * an update.
3a80e1fa 1363 */
6a553681
AD		 1364 if (!(nt->parms.o_flags & TUNNEL_CSUM) ||
1365 (nt->encap.type == TUNNEL_ENCAP_NONE)) {
1366 dev->features |= NETIF_F_GSO_SOFTWARE;
1367 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1368 }
3a80e1fa
AD		 1369
1370 /* Can use a lockless transmit, unless we generate
1371 * output sequences
1372 */
c12b395a 1373 dev->features |= NETIF_F_LLTX;
3a80e1fa 1374 }
c12b395a 1375
1376 err = register_netdevice(dev);
1377 if (err)
1378 goto out;
1379
1380 dev_hold(dev);
1381 ip6gre_tunnel_link(ign, nt);
1382
1383out:
1384 return err;
1385}
1386
1387static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1388 struct nlattr *data[])
1389{
22f08069
ND		 1390 struct ip6_tnl *t, *nt = netdev_priv(dev);
1391 struct net *net = nt->net;
c12b395a 1392 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1393 struct __ip6_tnl_parm p;
1faf3d9f 1394 struct ip_tunnel_encap ipencap;
c12b395a 1395
1396 if (dev == ign->fb_tunnel_dev)
1397 return -EINVAL;
1398
1faf3d9f
TH		 1399 if (ip6gre_netlink_encap_parms(data, &ipencap)) {
1400 int err = ip6_tnl_encap_setup(nt, &ipencap);
1401
1402 if (err < 0)
1403 return err;
1404 }
1405
c12b395a 1406 ip6gre_netlink_parms(data, &p);
1407
1408 t = ip6gre_tunnel_locate(net, &p, 0);
1409
1410 if (t) {
1411 if (t->dev != dev)
1412 return -EEXIST;
1413 } else {
1414 t = nt;
c12b395a 1415 }
1416
6a61d4db
ND		 1417 ip6gre_tunnel_unlink(ign, t);
1418 ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1419 ip6gre_tunnel_link(ign, t);
c12b395a 1420 return 0;
1421}
1422
54d63f78
ND		 1423static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1424{
1425 struct net *net = dev_net(dev);
1426 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1427
1428 if (dev != ign->fb_tunnel_dev)
1429 unregister_netdevice_queue(dev, head);
1430}
1431
c12b395a 1432static size_t ip6gre_get_size(const struct net_device *dev)
1433{
1434 return
1435 /* IFLA_GRE_LINK */
1436 nla_total_size(4) +
1437 /* IFLA_GRE_IFLAGS */
1438 nla_total_size(2) +
1439 /* IFLA_GRE_OFLAGS */
1440 nla_total_size(2) +
1441 /* IFLA_GRE_IKEY */
1442 nla_total_size(4) +
1443 /* IFLA_GRE_OKEY */
1444 nla_total_size(4) +
1445 /* IFLA_GRE_LOCAL */
a3754133 1446 nla_total_size(sizeof(struct in6_addr)) +
c12b395a 1447 /* IFLA_GRE_REMOTE */
a3754133 1448 nla_total_size(sizeof(struct in6_addr)) +
c12b395a 1449 /* IFLA_GRE_TTL */
1450 nla_total_size(1) +
c12b395a 1451 /* IFLA_GRE_ENCAP_LIMIT */
1452 nla_total_size(1) +
1453 /* IFLA_GRE_FLOWINFO */
1454 nla_total_size(4) +
1455 /* IFLA_GRE_FLAGS */
1456 nla_total_size(4) +
1faf3d9f
TH		 1457 /* IFLA_GRE_ENCAP_TYPE */
1458 nla_total_size(2) +
1459 /* IFLA_GRE_ENCAP_FLAGS */
1460 nla_total_size(2) +
1461 /* IFLA_GRE_ENCAP_SPORT */
1462 nla_total_size(2) +
1463 /* IFLA_GRE_ENCAP_DPORT */
1464 nla_total_size(2) +
c12b395a 1465 0;
1466}
1467
1468static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1469{
1470 struct ip6_tnl *t = netdev_priv(dev);
1471 struct __ip6_tnl_parm *p = &t->parms;
1472
1473 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
f41fe3c2
TH		 1474 nla_put_be16(skb, IFLA_GRE_IFLAGS,
1475 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1476 nla_put_be16(skb, IFLA_GRE_OFLAGS,
1477 gre_tnl_flags_to_gre_flags(p->o_flags)) ||
c12b395a 1478 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1479 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
930345ea
JB		 1480 nla_put_in6_addr(skb, IFLA_GRE_LOCAL, &p->laddr) ||
1481 nla_put_in6_addr(skb, IFLA_GRE_REMOTE, &p->raddr) ||
c12b395a 1482 nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
c12b395a 1483 nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1484 nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1485 nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1486 goto nla_put_failure;
1faf3d9f
TH		 1487
1488 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1489 t->encap.type) ||
1490 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1491 t->encap.sport) ||
1492 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1493 t->encap.dport) ||
1494 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1495 t->encap.flags))
1496 goto nla_put_failure;
1497
c12b395a 1498 return 0;
1499
1500nla_put_failure:
1501 return -EMSGSIZE;
1502}
1503
1504static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1505 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1506 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1507 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1508 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1509 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1510 [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1511 [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1512 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1513 [IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1514 [IFLA_GRE_FLOWINFO] = { .type = NLA_U32 },
1515 [IFLA_GRE_FLAGS] = { .type = NLA_U32 },
1faf3d9f
TH		 1516 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 },
1517 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 },
1518 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 },
1519 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 },
c12b395a 1520};
1521
1522static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1523 .kind = "ip6gre",
1524 .maxtype = IFLA_GRE_MAX,
1525 .policy = ip6gre_policy,
1526 .priv_size = sizeof(struct ip6_tnl),
1527 .setup = ip6gre_tunnel_setup,
1528 .validate = ip6gre_tunnel_validate,
1529 .newlink = ip6gre_newlink,
1530 .changelink = ip6gre_changelink,
54d63f78 1531 .dellink = ip6gre_dellink,
c12b395a 1532 .get_size = ip6gre_get_size,
1533 .fill_info = ip6gre_fill_info,
1728d4fa 1534 .get_link_net = ip6_tnl_get_link_net,
c12b395a 1535};
1536
1537static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1538 .kind = "ip6gretap",
1539 .maxtype = IFLA_GRE_MAX,
1540 .policy = ip6gre_policy,
1541 .priv_size = sizeof(struct ip6_tnl),
1542 .setup = ip6gre_tap_setup,
1543 .validate = ip6gre_tap_validate,
1544 .newlink = ip6gre_newlink,
1545 .changelink = ip6gre_changelink,
1546 .get_size = ip6gre_get_size,
1547 .fill_info = ip6gre_fill_info,
3390e397 1548 .get_link_net = ip6_tnl_get_link_net,
c12b395a 1549};
1550
1551/*
1552 * And now the modules code and kernel interface.
1553 */
1554
1555static int __init ip6gre_init(void)
1556{
1557 int err;
1558
1559 pr_info("GRE over IPv6 tunneling driver\n");
1560
1561 err = register_pernet_device(&ip6gre_net_ops);
1562 if (err < 0)
1563 return err;
1564
1565 err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1566 if (err < 0) {
1567 pr_info("%s: can't add protocol\n", __func__);
1568 goto add_proto_failed;
1569 }
1570
1571 err = rtnl_link_register(&ip6gre_link_ops);
1572 if (err < 0)
1573 goto rtnl_link_failed;
1574
1575 err = rtnl_link_register(&ip6gre_tap_ops);
1576 if (err < 0)
1577 goto tap_ops_failed;
1578
1579out:
1580 return err;
1581
1582tap_ops_failed:
1583 rtnl_link_unregister(&ip6gre_link_ops);
1584rtnl_link_failed:
1585 inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1586add_proto_failed:
1587 unregister_pernet_device(&ip6gre_net_ops);
1588 goto out;
1589}
1590
1591static void __exit ip6gre_fini(void)
1592{
1593 rtnl_link_unregister(&ip6gre_tap_ops);
1594 rtnl_link_unregister(&ip6gre_link_ops);
1595 inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1596 unregister_pernet_device(&ip6gre_net_ops);
1597}
1598
1599module_init(ip6gre_init);
1600module_exit(ip6gre_fini);
1601MODULE_LICENSE("GPL");
1602MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1603MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1604MODULE_ALIAS_RTNL_LINK("ip6gre");
5a4ee9a9 1605MODULE_ALIAS_RTNL_LINK("ip6gretap");
c12b395a 1606MODULE_ALIAS_NETDEV("ip6gre0");
Linux 6.x block layer and io_uring tree(s)