]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - net/ipv4/ip_gre.c
8fcf0ed76548a31074293d26a671fbbd9a32845e
[~shefty/rdma-dev.git] / net / ipv4 / ip_gre.c
1 /*
2  *      Linux NET3:     GRE over IP protocol decoder.
3  *
4  *      Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
5  *
6  *      This program is free software; you can redistribute it and/or
7  *      modify it under the terms of the GNU General Public License
8  *      as published by the Free Software Foundation; either version
9  *      2 of the License, or (at your option) any later version.
10  *
11  */
12
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15 #include <linux/capability.h>
16 #include <linux/module.h>
17 #include <linux/types.h>
18 #include <linux/kernel.h>
19 #include <linux/slab.h>
20 #include <asm/uaccess.h>
21 #include <linux/skbuff.h>
22 #include <linux/netdevice.h>
23 #include <linux/in.h>
24 #include <linux/tcp.h>
25 #include <linux/udp.h>
26 #include <linux/if_arp.h>
27 #include <linux/mroute.h>
28 #include <linux/init.h>
29 #include <linux/in6.h>
30 #include <linux/inetdevice.h>
31 #include <linux/igmp.h>
32 #include <linux/netfilter_ipv4.h>
33 #include <linux/etherdevice.h>
34 #include <linux/if_ether.h>
35
36 #include <net/sock.h>
37 #include <net/ip.h>
38 #include <net/icmp.h>
39 #include <net/protocol.h>
40 #include <net/ipip.h>
41 #include <net/arp.h>
42 #include <net/checksum.h>
43 #include <net/dsfield.h>
44 #include <net/inet_ecn.h>
45 #include <net/xfrm.h>
46 #include <net/net_namespace.h>
47 #include <net/netns/generic.h>
48 #include <net/rtnetlink.h>
49 #include <net/gre.h>
50
51 #if IS_ENABLED(CONFIG_IPV6)
52 #include <net/ipv6.h>
53 #include <net/ip6_fib.h>
54 #include <net/ip6_route.h>
55 #endif
56
57 /*
58    Problems & solutions
59    --------------------
60
61    1. The most important issue is detecting local dead loops.
62    They would cause complete host lockup in transmit, which
63    would be "resolved" by stack overflow or, if queueing is enabled,
64    with infinite looping in net_bh.
65
66    We cannot track such dead loops during route installation,
67    it is infeasible task. The most general solutions would be
68    to keep skb->encapsulation counter (sort of local ttl),
69    and silently drop packet when it expires. It is a good
70    solution, but it supposes maintaining new variable in ALL
71    skb, even if no tunneling is used.
72
73    Current solution: xmit_recursion breaks dead loops. This is a percpu
74    counter, since when we enter the first ndo_xmit(), cpu migration is
75    forbidden. We force an exit if this counter reaches RECURSION_LIMIT
76
77    2. Networking dead loops would not kill routers, but would really
78    kill network. IP hop limit plays role of "t->recursion" in this case,
79    if we copy it from packet being encapsulated to upper header.
80    It is very good solution, but it introduces two problems:
81
82    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
83      do not work over tunnels.
84    - traceroute does not work. I planned to relay ICMP from tunnel,
85      so that this problem would be solved and traceroute output
86      would even more informative. This idea appeared to be wrong:
87      only Linux complies to rfc1812 now (yes, guys, Linux is the only
88      true router now :-)), all routers (at least, in neighbourhood of mine)
89      return only 8 bytes of payload. It is the end.
90
91    Hence, if we want that OSPF worked or traceroute said something reasonable,
92    we should search for another solution.
93
94    One of them is to parse packet trying to detect inner encapsulation
95    made by our node. It is difficult or even impossible, especially,
96    taking into account fragmentation. TO be short, ttl is not solution at all.
97
98    Current solution: The solution was UNEXPECTEDLY SIMPLE.
99    We force DF flag on tunnels with preconfigured hop limit,
100    that is ALL. :-) Well, it does not remove the problem completely,
101    but exponential growth of network traffic is changed to linear
102    (branches, that exceed pmtu are pruned) and tunnel mtu
103    rapidly degrades to value <68, where looping stops.
104    Yes, it is not good if there exists a router in the loop,
105    which does not force DF, even when encapsulating packets have DF set.
106    But it is not our problem! Nobody could accuse us, we made
107    all that we could make. Even if it is your gated who injected
108    fatal route to network, even if it were you who configured
109    fatal static route: you are innocent. :-)
110
111
112
113    3. Really, ipv4/ipip.c, ipv4/ip_gre.c and ipv6/sit.c contain
114    practically identical code. It would be good to glue them
115    together, but it is not very evident, how to make them modular.
116    sit is integral part of IPv6, ipip and gre are naturally modular.
117    We could extract common parts (hash table, ioctl etc)
118    to a separate module (ip_tunnel.c).
119
120    Alexey Kuznetsov.
121  */
122
123 static bool log_ecn_error = true;
124 module_param(log_ecn_error, bool, 0644);
125 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
126
127 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
128 static int ipgre_tunnel_init(struct net_device *dev);
129 static void ipgre_tunnel_setup(struct net_device *dev);
130 static int ipgre_tunnel_bind_dev(struct net_device *dev);
131
132 /* Fallback tunnel: no source, no destination, no key, no options */
133
134 #define HASH_SIZE  16
135
136 static int ipgre_net_id __read_mostly;
137 struct ipgre_net {
138         struct ip_tunnel __rcu *tunnels[4][HASH_SIZE];
139
140         struct net_device *fb_tunnel_dev;
141 };
142
143 /* Tunnel hash table */
144
145 /*
146    4 hash tables:
147
148    3: (remote,local)
149    2: (remote,*)
150    1: (*,local)
151    0: (*,*)
152
153    We require exact key match i.e. if a key is present in packet
154    it will match only tunnel with the same key; if it is not present,
155    it will match only keyless tunnel.
156
157    All keysless packets, if not matched configured keyless tunnels
158    will match fallback tunnel.
159  */
160
161 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
162
163 #define tunnels_r_l     tunnels[3]
164 #define tunnels_r       tunnels[2]
165 #define tunnels_l       tunnels[1]
166 #define tunnels_wc      tunnels[0]
167
168 static struct rtnl_link_stats64 *ipgre_get_stats64(struct net_device *dev,
169                                                    struct rtnl_link_stats64 *tot)
170 {
171         int i;
172
173         for_each_possible_cpu(i) {
174                 const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
175                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
176                 unsigned int start;
177
178                 do {
179                         start = u64_stats_fetch_begin_bh(&tstats->syncp);
180                         rx_packets = tstats->rx_packets;
181                         tx_packets = tstats->tx_packets;
182                         rx_bytes = tstats->rx_bytes;
183                         tx_bytes = tstats->tx_bytes;
184                 } while (u64_stats_fetch_retry_bh(&tstats->syncp, start));
185
186                 tot->rx_packets += rx_packets;
187                 tot->tx_packets += tx_packets;
188                 tot->rx_bytes   += rx_bytes;
189                 tot->tx_bytes   += tx_bytes;
190         }
191
192         tot->multicast = dev->stats.multicast;
193         tot->rx_crc_errors = dev->stats.rx_crc_errors;
194         tot->rx_fifo_errors = dev->stats.rx_fifo_errors;
195         tot->rx_length_errors = dev->stats.rx_length_errors;
196         tot->rx_frame_errors = dev->stats.rx_frame_errors;
197         tot->rx_errors = dev->stats.rx_errors;
198
199         tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
200         tot->tx_carrier_errors = dev->stats.tx_carrier_errors;
201         tot->tx_dropped = dev->stats.tx_dropped;
202         tot->tx_aborted_errors = dev->stats.tx_aborted_errors;
203         tot->tx_errors = dev->stats.tx_errors;
204
205         return tot;
206 }
207
208 /* Does key in tunnel parameters match packet */
209 static bool ipgre_key_match(const struct ip_tunnel_parm *p,
210                             __be16 flags, __be32 key)
211 {
212         if (p->i_flags & GRE_KEY) {
213                 if (flags & GRE_KEY)
214                         return key == p->i_key;
215                 else
216                         return false;   /* key expected, none present */
217         } else
218                 return !(flags & GRE_KEY);
219 }
220
221 /* Given src, dst and key, find appropriate for input tunnel. */
222
223 static struct ip_tunnel *ipgre_tunnel_lookup(struct net_device *dev,
224                                              __be32 remote, __be32 local,
225                                              __be16 flags, __be32 key,
226                                              __be16 gre_proto)
227 {
228         struct net *net = dev_net(dev);
229         int link = dev->ifindex;
230         unsigned int h0 = HASH(remote);
231         unsigned int h1 = HASH(key);
232         struct ip_tunnel *t, *cand = NULL;
233         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
234         int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
235                        ARPHRD_ETHER : ARPHRD_IPGRE;
236         int score, cand_score = 4;
237
238         for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
239                 if (local != t->parms.iph.saddr ||
240                     remote != t->parms.iph.daddr ||
241                     !(t->dev->flags & IFF_UP))
242                         continue;
243
244                 if (!ipgre_key_match(&t->parms, flags, key))
245                         continue;
246
247                 if (t->dev->type != ARPHRD_IPGRE &&
248                     t->dev->type != dev_type)
249                         continue;
250
251                 score = 0;
252                 if (t->parms.link != link)
253                         score |= 1;
254                 if (t->dev->type != dev_type)
255                         score |= 2;
256                 if (score == 0)
257                         return t;
258
259                 if (score < cand_score) {
260                         cand = t;
261                         cand_score = score;
262                 }
263         }
264
265         for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
266                 if (remote != t->parms.iph.daddr ||
267                     !(t->dev->flags & IFF_UP))
268                         continue;
269
270                 if (!ipgre_key_match(&t->parms, flags, key))
271                         continue;
272
273                 if (t->dev->type != ARPHRD_IPGRE &&
274                     t->dev->type != dev_type)
275                         continue;
276
277                 score = 0;
278                 if (t->parms.link != link)
279                         score |= 1;
280                 if (t->dev->type != dev_type)
281                         score |= 2;
282                 if (score == 0)
283                         return t;
284
285                 if (score < cand_score) {
286                         cand = t;
287                         cand_score = score;
288                 }
289         }
290
291         for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
292                 if ((local != t->parms.iph.saddr &&
293                      (local != t->parms.iph.daddr ||
294                       !ipv4_is_multicast(local))) ||
295                     !(t->dev->flags & IFF_UP))
296                         continue;
297
298                 if (!ipgre_key_match(&t->parms, flags, key))
299                         continue;
300
301                 if (t->dev->type != ARPHRD_IPGRE &&
302                     t->dev->type != dev_type)
303                         continue;
304
305                 score = 0;
306                 if (t->parms.link != link)
307                         score |= 1;
308                 if (t->dev->type != dev_type)
309                         score |= 2;
310                 if (score == 0)
311                         return t;
312
313                 if (score < cand_score) {
314                         cand = t;
315                         cand_score = score;
316                 }
317         }
318
319         for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
320                 if (t->parms.i_key != key ||
321                     !(t->dev->flags & IFF_UP))
322                         continue;
323
324                 if (t->dev->type != ARPHRD_IPGRE &&
325                     t->dev->type != dev_type)
326                         continue;
327
328                 score = 0;
329                 if (t->parms.link != link)
330                         score |= 1;
331                 if (t->dev->type != dev_type)
332                         score |= 2;
333                 if (score == 0)
334                         return t;
335
336                 if (score < cand_score) {
337                         cand = t;
338                         cand_score = score;
339                 }
340         }
341
342         if (cand != NULL)
343                 return cand;
344
345         dev = ign->fb_tunnel_dev;
346         if (dev->flags & IFF_UP)
347                 return netdev_priv(dev);
348
349         return NULL;
350 }
351
352 static struct ip_tunnel __rcu **__ipgre_bucket(struct ipgre_net *ign,
353                 struct ip_tunnel_parm *parms)
354 {
355         __be32 remote = parms->iph.daddr;
356         __be32 local = parms->iph.saddr;
357         __be32 key = parms->i_key;
358         unsigned int h = HASH(key);
359         int prio = 0;
360
361         if (local)
362                 prio |= 1;
363         if (remote && !ipv4_is_multicast(remote)) {
364                 prio |= 2;
365                 h ^= HASH(remote);
366         }
367
368         return &ign->tunnels[prio][h];
369 }
370
371 static inline struct ip_tunnel __rcu **ipgre_bucket(struct ipgre_net *ign,
372                 struct ip_tunnel *t)
373 {
374         return __ipgre_bucket(ign, &t->parms);
375 }
376
377 static void ipgre_tunnel_link(struct ipgre_net *ign, struct ip_tunnel *t)
378 {
379         struct ip_tunnel __rcu **tp = ipgre_bucket(ign, t);
380
381         rcu_assign_pointer(t->next, rtnl_dereference(*tp));
382         rcu_assign_pointer(*tp, t);
383 }
384
385 static void ipgre_tunnel_unlink(struct ipgre_net *ign, struct ip_tunnel *t)
386 {
387         struct ip_tunnel __rcu **tp;
388         struct ip_tunnel *iter;
389
390         for (tp = ipgre_bucket(ign, t);
391              (iter = rtnl_dereference(*tp)) != NULL;
392              tp = &iter->next) {
393                 if (t == iter) {
394                         rcu_assign_pointer(*tp, t->next);
395                         break;
396                 }
397         }
398 }
399
400 static struct ip_tunnel *ipgre_tunnel_find(struct net *net,
401                                            struct ip_tunnel_parm *parms,
402                                            int type)
403 {
404         __be32 remote = parms->iph.daddr;
405         __be32 local = parms->iph.saddr;
406         __be32 key = parms->i_key;
407         int link = parms->link;
408         struct ip_tunnel *t;
409         struct ip_tunnel __rcu **tp;
410         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
411
412         for (tp = __ipgre_bucket(ign, parms);
413              (t = rtnl_dereference(*tp)) != NULL;
414              tp = &t->next)
415                 if (local == t->parms.iph.saddr &&
416                     remote == t->parms.iph.daddr &&
417                     key == t->parms.i_key &&
418                     link == t->parms.link &&
419                     type == t->dev->type)
420                         break;
421
422         return t;
423 }
424
425 static struct ip_tunnel *ipgre_tunnel_locate(struct net *net,
426                 struct ip_tunnel_parm *parms, int create)
427 {
428         struct ip_tunnel *t, *nt;
429         struct net_device *dev;
430         char name[IFNAMSIZ];
431         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
432
433         t = ipgre_tunnel_find(net, parms, ARPHRD_IPGRE);
434         if (t || !create)
435                 return t;
436
437         if (parms->name[0])
438                 strlcpy(name, parms->name, IFNAMSIZ);
439         else
440                 strcpy(name, "gre%d");
441
442         dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup);
443         if (!dev)
444                 return NULL;
445
446         dev_net_set(dev, net);
447
448         nt = netdev_priv(dev);
449         nt->parms = *parms;
450         dev->rtnl_link_ops = &ipgre_link_ops;
451
452         dev->mtu = ipgre_tunnel_bind_dev(dev);
453
454         if (register_netdevice(dev) < 0)
455                 goto failed_free;
456
457         /* Can use a lockless transmit, unless we generate output sequences */
458         if (!(nt->parms.o_flags & GRE_SEQ))
459                 dev->features |= NETIF_F_LLTX;
460
461         dev_hold(dev);
462         ipgre_tunnel_link(ign, nt);
463         return nt;
464
465 failed_free:
466         free_netdev(dev);
467         return NULL;
468 }
469
470 static void ipgre_tunnel_uninit(struct net_device *dev)
471 {
472         struct net *net = dev_net(dev);
473         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
474
475         ipgre_tunnel_unlink(ign, netdev_priv(dev));
476         dev_put(dev);
477 }
478
479
480 static void ipgre_err(struct sk_buff *skb, u32 info)
481 {
482
483 /* All the routers (except for Linux) return only
484    8 bytes of packet payload. It means, that precise relaying of
485    ICMP in the real Internet is absolutely infeasible.
486
487    Moreover, Cisco "wise men" put GRE key to the third word
488    in GRE header. It makes impossible maintaining even soft state for keyed
489    GRE tunnels with enabled checksum. Tell them "thank you".
490
491    Well, I wonder, rfc1812 was written by Cisco employee,
492    what the hell these idiots break standards established
493    by themselves???
494  */
495
496         const struct iphdr *iph = (const struct iphdr *)skb->data;
497         __be16       *p = (__be16 *)(skb->data+(iph->ihl<<2));
498         int grehlen = (iph->ihl<<2) + 4;
499         const int type = icmp_hdr(skb)->type;
500         const int code = icmp_hdr(skb)->code;
501         struct ip_tunnel *t;
502         __be16 flags;
503         __be32 key = 0;
504
505         flags = p[0];
506         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
507                 if (flags&(GRE_VERSION|GRE_ROUTING))
508                         return;
509                 if (flags&GRE_KEY) {
510                         grehlen += 4;
511                         if (flags&GRE_CSUM)
512                                 grehlen += 4;
513                 }
514         }
515
516         /* If only 8 bytes returned, keyed message will be dropped here */
517         if (skb_headlen(skb) < grehlen)
518                 return;
519
520         if (flags & GRE_KEY)
521                 key = *(((__be32 *)p) + (grehlen / 4) - 1);
522
523         switch (type) {
524         default:
525         case ICMP_PARAMETERPROB:
526                 return;
527
528         case ICMP_DEST_UNREACH:
529                 switch (code) {
530                 case ICMP_SR_FAILED:
531                 case ICMP_PORT_UNREACH:
532                         /* Impossible event. */
533                         return;
534                 default:
535                         /* All others are translated to HOST_UNREACH.
536                            rfc2003 contains "deep thoughts" about NET_UNREACH,
537                            I believe they are just ether pollution. --ANK
538                          */
539                         break;
540                 }
541                 break;
542         case ICMP_TIME_EXCEEDED:
543                 if (code != ICMP_EXC_TTL)
544                         return;
545                 break;
546
547         case ICMP_REDIRECT:
548                 break;
549         }
550
551         t = ipgre_tunnel_lookup(skb->dev, iph->daddr, iph->saddr,
552                                 flags, key, p[1]);
553
554         if (t == NULL)
555                 return;
556
557         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
558                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
559                                  t->parms.link, 0, IPPROTO_GRE, 0);
560                 return;
561         }
562         if (type == ICMP_REDIRECT) {
563                 ipv4_redirect(skb, dev_net(skb->dev), t->parms.link, 0,
564                               IPPROTO_GRE, 0);
565                 return;
566         }
567         if (t->parms.iph.daddr == 0 ||
568             ipv4_is_multicast(t->parms.iph.daddr))
569                 return;
570
571         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
572                 return;
573
574         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
575                 t->err_count++;
576         else
577                 t->err_count = 1;
578         t->err_time = jiffies;
579 }
580
581 static inline u8
582 ipgre_ecn_encapsulate(u8 tos, const struct iphdr *old_iph, struct sk_buff *skb)
583 {
584         u8 inner = 0;
585         if (skb->protocol == htons(ETH_P_IP))
586                 inner = old_iph->tos;
587         else if (skb->protocol == htons(ETH_P_IPV6))
588                 inner = ipv6_get_dsfield((const struct ipv6hdr *)old_iph);
589         return INET_ECN_encapsulate(tos, inner);
590 }
591
592 static int ipgre_rcv(struct sk_buff *skb)
593 {
594         const struct iphdr *iph;
595         u8     *h;
596         __be16    flags;
597         __sum16   csum = 0;
598         __be32 key = 0;
599         u32    seqno = 0;
600         struct ip_tunnel *tunnel;
601         int    offset = 4;
602         __be16 gre_proto;
603         int    err;
604
605         if (!pskb_may_pull(skb, 16))
606                 goto drop;
607
608         iph = ip_hdr(skb);
609         h = skb->data;
610         flags = *(__be16 *)h;
611
612         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
613                 /* - Version must be 0.
614                    - We do not support routing headers.
615                  */
616                 if (flags&(GRE_VERSION|GRE_ROUTING))
617                         goto drop;
618
619                 if (flags&GRE_CSUM) {
620                         switch (skb->ip_summed) {
621                         case CHECKSUM_COMPLETE:
622                                 csum = csum_fold(skb->csum);
623                                 if (!csum)
624                                         break;
625                                 /* fall through */
626                         case CHECKSUM_NONE:
627                                 skb->csum = 0;
628                                 csum = __skb_checksum_complete(skb);
629                                 skb->ip_summed = CHECKSUM_COMPLETE;
630                         }
631                         offset += 4;
632                 }
633                 if (flags&GRE_KEY) {
634                         key = *(__be32 *)(h + offset);
635                         offset += 4;
636                 }
637                 if (flags&GRE_SEQ) {
638                         seqno = ntohl(*(__be32 *)(h + offset));
639                         offset += 4;
640                 }
641         }
642
643         gre_proto = *(__be16 *)(h + 2);
644
645         tunnel = ipgre_tunnel_lookup(skb->dev,
646                                      iph->saddr, iph->daddr, flags, key,
647                                      gre_proto);
648         if (tunnel) {
649                 struct pcpu_tstats *tstats;
650
651                 secpath_reset(skb);
652
653                 skb->protocol = gre_proto;
654                 /* WCCP version 1 and 2 protocol decoding.
655                  * - Change protocol to IP
656                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
657                  */
658                 if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
659                         skb->protocol = htons(ETH_P_IP);
660                         if ((*(h + offset) & 0xF0) != 0x40)
661                                 offset += 4;
662                 }
663
664                 skb->mac_header = skb->network_header;
665                 __pskb_pull(skb, offset);
666                 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
667                 skb->pkt_type = PACKET_HOST;
668 #ifdef CONFIG_NET_IPGRE_BROADCAST
669                 if (ipv4_is_multicast(iph->daddr)) {
670                         /* Looped back packet, drop it! */
671                         if (rt_is_output_route(skb_rtable(skb)))
672                                 goto drop;
673                         tunnel->dev->stats.multicast++;
674                         skb->pkt_type = PACKET_BROADCAST;
675                 }
676 #endif
677
678                 if (((flags&GRE_CSUM) && csum) ||
679                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
680                         tunnel->dev->stats.rx_crc_errors++;
681                         tunnel->dev->stats.rx_errors++;
682                         goto drop;
683                 }
684                 if (tunnel->parms.i_flags&GRE_SEQ) {
685                         if (!(flags&GRE_SEQ) ||
686                             (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
687                                 tunnel->dev->stats.rx_fifo_errors++;
688                                 tunnel->dev->stats.rx_errors++;
689                                 goto drop;
690                         }
691                         tunnel->i_seqno = seqno + 1;
692                 }
693
694                 /* Warning: All skb pointers will be invalidated! */
695                 if (tunnel->dev->type == ARPHRD_ETHER) {
696                         if (!pskb_may_pull(skb, ETH_HLEN)) {
697                                 tunnel->dev->stats.rx_length_errors++;
698                                 tunnel->dev->stats.rx_errors++;
699                                 goto drop;
700                         }
701
702                         iph = ip_hdr(skb);
703                         skb->protocol = eth_type_trans(skb, tunnel->dev);
704                         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
705                 }
706
707                 __skb_tunnel_rx(skb, tunnel->dev);
708
709                 skb_reset_network_header(skb);
710                 err = IP_ECN_decapsulate(iph, skb);
711                 if (unlikely(err)) {
712                         if (log_ecn_error)
713                                 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
714                                                      &iph->saddr, iph->tos);
715                         if (err > 1) {
716                                 ++tunnel->dev->stats.rx_frame_errors;
717                                 ++tunnel->dev->stats.rx_errors;
718                                 goto drop;
719                         }
720                 }
721
722                 tstats = this_cpu_ptr(tunnel->dev->tstats);
723                 u64_stats_update_begin(&tstats->syncp);
724                 tstats->rx_packets++;
725                 tstats->rx_bytes += skb->len;
726                 u64_stats_update_end(&tstats->syncp);
727
728                 gro_cells_receive(&tunnel->gro_cells, skb);
729                 return 0;
730         }
731         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
732
733 drop:
734         kfree_skb(skb);
735         return 0;
736 }
737
738 static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
739 {
740         struct ip_tunnel *tunnel = netdev_priv(dev);
741         const struct iphdr  *old_iph = ip_hdr(skb);
742         const struct iphdr  *tiph;
743         struct flowi4 fl4;
744         u8     tos;
745         __be16 df;
746         struct rtable *rt;                      /* Route to the other host */
747         struct net_device *tdev;                /* Device to other host */
748         struct iphdr  *iph;                     /* Our new IP header */
749         unsigned int max_headroom;              /* The extra header space needed */
750         int    gre_hlen;
751         __be32 dst;
752         int    mtu;
753
754         if (skb->ip_summed == CHECKSUM_PARTIAL &&
755             skb_checksum_help(skb))
756                 goto tx_error;
757
758         if (dev->type == ARPHRD_ETHER)
759                 IPCB(skb)->flags = 0;
760
761         if (dev->header_ops && dev->type == ARPHRD_IPGRE) {
762                 gre_hlen = 0;
763                 if (skb->protocol == htons(ETH_P_IP))
764                         tiph = (const struct iphdr *)skb->data;
765                 else
766                         tiph = &tunnel->parms.iph;
767         } else {
768                 gre_hlen = tunnel->hlen;
769                 tiph = &tunnel->parms.iph;
770         }
771
772         if ((dst = tiph->daddr) == 0) {
773                 /* NBMA tunnel */
774
775                 if (skb_dst(skb) == NULL) {
776                         dev->stats.tx_fifo_errors++;
777                         goto tx_error;
778                 }
779
780                 if (skb->protocol == htons(ETH_P_IP)) {
781                         rt = skb_rtable(skb);
782                         dst = rt_nexthop(rt, old_iph->daddr);
783                 }
784 #if IS_ENABLED(CONFIG_IPV6)
785                 else if (skb->protocol == htons(ETH_P_IPV6)) {
786                         const struct in6_addr *addr6;
787                         struct neighbour *neigh;
788                         bool do_tx_error_icmp;
789                         int addr_type;
790
791                         neigh = dst_neigh_lookup(skb_dst(skb), &ipv6_hdr(skb)->daddr);
792                         if (neigh == NULL)
793                                 goto tx_error;
794
795                         addr6 = (const struct in6_addr *)&neigh->primary_key;
796                         addr_type = ipv6_addr_type(addr6);
797
798                         if (addr_type == IPV6_ADDR_ANY) {
799                                 addr6 = &ipv6_hdr(skb)->daddr;
800                                 addr_type = ipv6_addr_type(addr6);
801                         }
802
803                         if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
804                                 do_tx_error_icmp = true;
805                         else {
806                                 do_tx_error_icmp = false;
807                                 dst = addr6->s6_addr32[3];
808                         }
809                         neigh_release(neigh);
810                         if (do_tx_error_icmp)
811                                 goto tx_error_icmp;
812                 }
813 #endif
814                 else
815                         goto tx_error;
816         }
817
818         tos = tiph->tos;
819         if (tos == 1) {
820                 tos = 0;
821                 if (skb->protocol == htons(ETH_P_IP))
822                         tos = old_iph->tos;
823                 else if (skb->protocol == htons(ETH_P_IPV6))
824                         tos = ipv6_get_dsfield((const struct ipv6hdr *)old_iph);
825         }
826
827         rt = ip_route_output_gre(dev_net(dev), &fl4, dst, tiph->saddr,
828                                  tunnel->parms.o_key, RT_TOS(tos),
829                                  tunnel->parms.link);
830         if (IS_ERR(rt)) {
831                 dev->stats.tx_carrier_errors++;
832                 goto tx_error;
833         }
834         tdev = rt->dst.dev;
835
836         if (tdev == dev) {
837                 ip_rt_put(rt);
838                 dev->stats.collisions++;
839                 goto tx_error;
840         }
841
842         df = tiph->frag_off;
843         if (df)
844                 mtu = dst_mtu(&rt->dst) - dev->hard_header_len - tunnel->hlen;
845         else
846                 mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
847
848         if (skb_dst(skb))
849                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
850
851         if (skb->protocol == htons(ETH_P_IP)) {
852                 df |= (old_iph->frag_off&htons(IP_DF));
853
854                 if ((old_iph->frag_off&htons(IP_DF)) &&
855                     mtu < ntohs(old_iph->tot_len)) {
856                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
857                         ip_rt_put(rt);
858                         goto tx_error;
859                 }
860         }
861 #if IS_ENABLED(CONFIG_IPV6)
862         else if (skb->protocol == htons(ETH_P_IPV6)) {
863                 struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
864
865                 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && mtu >= IPV6_MIN_MTU) {
866                         if ((tunnel->parms.iph.daddr &&
867                              !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
868                             rt6->rt6i_dst.plen == 128) {
869                                 rt6->rt6i_flags |= RTF_MODIFIED;
870                                 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
871                         }
872                 }
873
874                 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
875                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
876                         ip_rt_put(rt);
877                         goto tx_error;
878                 }
879         }
880 #endif
881
882         if (tunnel->err_count > 0) {
883                 if (time_before(jiffies,
884                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
885                         tunnel->err_count--;
886
887                         dst_link_failure(skb);
888                 } else
889                         tunnel->err_count = 0;
890         }
891
892         max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen + rt->dst.header_len;
893
894         if (skb_headroom(skb) < max_headroom || skb_shared(skb)||
895             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
896                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
897                 if (max_headroom > dev->needed_headroom)
898                         dev->needed_headroom = max_headroom;
899                 if (!new_skb) {
900                         ip_rt_put(rt);
901                         dev->stats.tx_dropped++;
902                         dev_kfree_skb(skb);
903                         return NETDEV_TX_OK;
904                 }
905                 if (skb->sk)
906                         skb_set_owner_w(new_skb, skb->sk);
907                 dev_kfree_skb(skb);
908                 skb = new_skb;
909                 old_iph = ip_hdr(skb);
910         }
911
912         skb_reset_transport_header(skb);
913         skb_push(skb, gre_hlen);
914         skb_reset_network_header(skb);
915         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
916         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
917                               IPSKB_REROUTED);
918         skb_dst_drop(skb);
919         skb_dst_set(skb, &rt->dst);
920
921         /*
922          *      Push down and install the IPIP header.
923          */
924
925         iph                     =       ip_hdr(skb);
926         iph->version            =       4;
927         iph->ihl                =       sizeof(struct iphdr) >> 2;
928         iph->frag_off           =       df;
929         iph->protocol           =       IPPROTO_GRE;
930         iph->tos                =       ipgre_ecn_encapsulate(tos, old_iph, skb);
931         iph->daddr              =       fl4.daddr;
932         iph->saddr              =       fl4.saddr;
933
934         if ((iph->ttl = tiph->ttl) == 0) {
935                 if (skb->protocol == htons(ETH_P_IP))
936                         iph->ttl = old_iph->ttl;
937 #if IS_ENABLED(CONFIG_IPV6)
938                 else if (skb->protocol == htons(ETH_P_IPV6))
939                         iph->ttl = ((const struct ipv6hdr *)old_iph)->hop_limit;
940 #endif
941                 else
942                         iph->ttl = ip4_dst_hoplimit(&rt->dst);
943         }
944
945         ((__be16 *)(iph + 1))[0] = tunnel->parms.o_flags;
946         ((__be16 *)(iph + 1))[1] = (dev->type == ARPHRD_ETHER) ?
947                                    htons(ETH_P_TEB) : skb->protocol;
948
949         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
950                 __be32 *ptr = (__be32 *)(((u8 *)iph) + tunnel->hlen - 4);
951
952                 if (tunnel->parms.o_flags&GRE_SEQ) {
953                         ++tunnel->o_seqno;
954                         *ptr = htonl(tunnel->o_seqno);
955                         ptr--;
956                 }
957                 if (tunnel->parms.o_flags&GRE_KEY) {
958                         *ptr = tunnel->parms.o_key;
959                         ptr--;
960                 }
961                 if (tunnel->parms.o_flags&GRE_CSUM) {
962                         *ptr = 0;
963                         *(__sum16 *)ptr = ip_compute_csum((void *)(iph+1), skb->len - sizeof(struct iphdr));
964                 }
965         }
966
967         iptunnel_xmit(skb, dev);
968         return NETDEV_TX_OK;
969
970 #if IS_ENABLED(CONFIG_IPV6)
971 tx_error_icmp:
972         dst_link_failure(skb);
973 #endif
974 tx_error:
975         dev->stats.tx_errors++;
976         dev_kfree_skb(skb);
977         return NETDEV_TX_OK;
978 }
979
980 static int ipgre_tunnel_bind_dev(struct net_device *dev)
981 {
982         struct net_device *tdev = NULL;
983         struct ip_tunnel *tunnel;
984         const struct iphdr *iph;
985         int hlen = LL_MAX_HEADER;
986         int mtu = ETH_DATA_LEN;
987         int addend = sizeof(struct iphdr) + 4;
988
989         tunnel = netdev_priv(dev);
990         iph = &tunnel->parms.iph;
991
992         /* Guess output device to choose reasonable mtu and needed_headroom */
993
994         if (iph->daddr) {
995                 struct flowi4 fl4;
996                 struct rtable *rt;
997
998                 rt = ip_route_output_gre(dev_net(dev), &fl4,
999                                          iph->daddr, iph->saddr,
1000                                          tunnel->parms.o_key,
1001                                          RT_TOS(iph->tos),
1002                                          tunnel->parms.link);
1003                 if (!IS_ERR(rt)) {
1004                         tdev = rt->dst.dev;
1005                         ip_rt_put(rt);
1006                 }
1007
1008                 if (dev->type != ARPHRD_ETHER)
1009                         dev->flags |= IFF_POINTOPOINT;
1010         }
1011
1012         if (!tdev && tunnel->parms.link)
1013                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
1014
1015         if (tdev) {
1016                 hlen = tdev->hard_header_len + tdev->needed_headroom;
1017                 mtu = tdev->mtu;
1018         }
1019         dev->iflink = tunnel->parms.link;
1020
1021         /* Precalculate GRE options length */
1022         if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
1023                 if (tunnel->parms.o_flags&GRE_CSUM)
1024                         addend += 4;
1025                 if (tunnel->parms.o_flags&GRE_KEY)
1026                         addend += 4;
1027                 if (tunnel->parms.o_flags&GRE_SEQ)
1028                         addend += 4;
1029         }
1030         dev->needed_headroom = addend + hlen;
1031         mtu -= dev->hard_header_len + addend;
1032
1033         if (mtu < 68)
1034                 mtu = 68;
1035
1036         tunnel->hlen = addend;
1037
1038         return mtu;
1039 }
1040
1041 static int
1042 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
1043 {
1044         int err = 0;
1045         struct ip_tunnel_parm p;
1046         struct ip_tunnel *t;
1047         struct net *net = dev_net(dev);
1048         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1049
1050         switch (cmd) {
1051         case SIOCGETTUNNEL:
1052                 t = NULL;
1053                 if (dev == ign->fb_tunnel_dev) {
1054                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1055                                 err = -EFAULT;
1056                                 break;
1057                         }
1058                         t = ipgre_tunnel_locate(net, &p, 0);
1059                 }
1060                 if (t == NULL)
1061                         t = netdev_priv(dev);
1062                 memcpy(&p, &t->parms, sizeof(p));
1063                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1064                         err = -EFAULT;
1065                 break;
1066
1067         case SIOCADDTUNNEL:
1068         case SIOCCHGTUNNEL:
1069                 err = -EPERM;
1070                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1071                         goto done;
1072
1073                 err = -EFAULT;
1074                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1075                         goto done;
1076
1077                 err = -EINVAL;
1078                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
1079                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
1080                     ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
1081                         goto done;
1082                 if (p.iph.ttl)
1083                         p.iph.frag_off |= htons(IP_DF);
1084
1085                 if (!(p.i_flags&GRE_KEY))
1086                         p.i_key = 0;
1087                 if (!(p.o_flags&GRE_KEY))
1088                         p.o_key = 0;
1089
1090                 t = ipgre_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
1091
1092                 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1093                         if (t != NULL) {
1094                                 if (t->dev != dev) {
1095                                         err = -EEXIST;
1096                                         break;
1097                                 }
1098                         } else {
1099                                 unsigned int nflags = 0;
1100
1101                                 t = netdev_priv(dev);
1102
1103                                 if (ipv4_is_multicast(p.iph.daddr))
1104                                         nflags = IFF_BROADCAST;
1105                                 else if (p.iph.daddr)
1106                                         nflags = IFF_POINTOPOINT;
1107
1108                                 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
1109                                         err = -EINVAL;
1110                                         break;
1111                                 }
1112                                 ipgre_tunnel_unlink(ign, t);
1113                                 synchronize_net();
1114                                 t->parms.iph.saddr = p.iph.saddr;
1115                                 t->parms.iph.daddr = p.iph.daddr;
1116                                 t->parms.i_key = p.i_key;
1117                                 t->parms.o_key = p.o_key;
1118                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
1119                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
1120                                 ipgre_tunnel_link(ign, t);
1121                                 netdev_state_change(dev);
1122                         }
1123                 }
1124
1125                 if (t) {
1126                         err = 0;
1127                         if (cmd == SIOCCHGTUNNEL) {
1128                                 t->parms.iph.ttl = p.iph.ttl;
1129                                 t->parms.iph.tos = p.iph.tos;
1130                                 t->parms.iph.frag_off = p.iph.frag_off;
1131                                 if (t->parms.link != p.link) {
1132                                         t->parms.link = p.link;
1133                                         dev->mtu = ipgre_tunnel_bind_dev(dev);
1134                                         netdev_state_change(dev);
1135                                 }
1136                         }
1137                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
1138                                 err = -EFAULT;
1139                 } else
1140                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1141                 break;
1142
1143         case SIOCDELTUNNEL:
1144                 err = -EPERM;
1145                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1146                         goto done;
1147
1148                 if (dev == ign->fb_tunnel_dev) {
1149                         err = -EFAULT;
1150                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1151                                 goto done;
1152                         err = -ENOENT;
1153                         if ((t = ipgre_tunnel_locate(net, &p, 0)) == NULL)
1154                                 goto done;
1155                         err = -EPERM;
1156                         if (t == netdev_priv(ign->fb_tunnel_dev))
1157                                 goto done;
1158                         dev = t->dev;
1159                 }
1160                 unregister_netdevice(dev);
1161                 err = 0;
1162                 break;
1163
1164         default:
1165                 err = -EINVAL;
1166         }
1167
1168 done:
1169         return err;
1170 }
1171
1172 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1173 {
1174         struct ip_tunnel *tunnel = netdev_priv(dev);
1175         if (new_mtu < 68 ||
1176             new_mtu > 0xFFF8 - dev->hard_header_len - tunnel->hlen)
1177                 return -EINVAL;
1178         dev->mtu = new_mtu;
1179         return 0;
1180 }
1181
1182 /* Nice toy. Unfortunately, useless in real life :-)
1183    It allows to construct virtual multiprotocol broadcast "LAN"
1184    over the Internet, provided multicast routing is tuned.
1185
1186
1187    I have no idea was this bicycle invented before me,
1188    so that I had to set ARPHRD_IPGRE to a random value.
1189    I have an impression, that Cisco could make something similar,
1190    but this feature is apparently missing in IOS<=11.2(8).
1191
1192    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1193    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1194
1195    ping -t 255 224.66.66.66
1196
1197    If nobody answers, mbone does not work.
1198
1199    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1200    ip addr add 10.66.66.<somewhat>/24 dev Universe
1201    ifconfig Universe up
1202    ifconfig Universe add fe80::<Your_real_addr>/10
1203    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1204    ftp 10.66.66.66
1205    ...
1206    ftp fec0:6666:6666::193.233.7.65
1207    ...
1208
1209  */
1210
1211 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
1212                         unsigned short type,
1213                         const void *daddr, const void *saddr, unsigned int len)
1214 {
1215         struct ip_tunnel *t = netdev_priv(dev);
1216         struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1217         __be16 *p = (__be16 *)(iph+1);
1218
1219         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1220         p[0]            = t->parms.o_flags;
1221         p[1]            = htons(type);
1222
1223         /*
1224          *      Set the source hardware address.
1225          */
1226
1227         if (saddr)
1228                 memcpy(&iph->saddr, saddr, 4);
1229         if (daddr)
1230                 memcpy(&iph->daddr, daddr, 4);
1231         if (iph->daddr)
1232                 return t->hlen;
1233
1234         return -t->hlen;
1235 }
1236
1237 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
1238 {
1239         const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
1240         memcpy(haddr, &iph->saddr, 4);
1241         return 4;
1242 }
1243
1244 static const struct header_ops ipgre_header_ops = {
1245         .create = ipgre_header,
1246         .parse  = ipgre_header_parse,
1247 };
1248
1249 #ifdef CONFIG_NET_IPGRE_BROADCAST
1250 static int ipgre_open(struct net_device *dev)
1251 {
1252         struct ip_tunnel *t = netdev_priv(dev);
1253
1254         if (ipv4_is_multicast(t->parms.iph.daddr)) {
1255                 struct flowi4 fl4;
1256                 struct rtable *rt;
1257
1258                 rt = ip_route_output_gre(dev_net(dev), &fl4,
1259                                          t->parms.iph.daddr,
1260                                          t->parms.iph.saddr,
1261                                          t->parms.o_key,
1262                                          RT_TOS(t->parms.iph.tos),
1263                                          t->parms.link);
1264                 if (IS_ERR(rt))
1265                         return -EADDRNOTAVAIL;
1266                 dev = rt->dst.dev;
1267                 ip_rt_put(rt);
1268                 if (__in_dev_get_rtnl(dev) == NULL)
1269                         return -EADDRNOTAVAIL;
1270                 t->mlink = dev->ifindex;
1271                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1272         }
1273         return 0;
1274 }
1275
1276 static int ipgre_close(struct net_device *dev)
1277 {
1278         struct ip_tunnel *t = netdev_priv(dev);
1279
1280         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
1281                 struct in_device *in_dev;
1282                 in_dev = inetdev_by_index(dev_net(dev), t->mlink);
1283                 if (in_dev)
1284                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1285         }
1286         return 0;
1287 }
1288
1289 #endif
1290
1291 static const struct net_device_ops ipgre_netdev_ops = {
1292         .ndo_init               = ipgre_tunnel_init,
1293         .ndo_uninit             = ipgre_tunnel_uninit,
1294 #ifdef CONFIG_NET_IPGRE_BROADCAST
1295         .ndo_open               = ipgre_open,
1296         .ndo_stop               = ipgre_close,
1297 #endif
1298         .ndo_start_xmit         = ipgre_tunnel_xmit,
1299         .ndo_do_ioctl           = ipgre_tunnel_ioctl,
1300         .ndo_change_mtu         = ipgre_tunnel_change_mtu,
1301         .ndo_get_stats64        = ipgre_get_stats64,
1302 };
1303
1304 static void ipgre_dev_free(struct net_device *dev)
1305 {
1306         struct ip_tunnel *tunnel = netdev_priv(dev);
1307
1308         gro_cells_destroy(&tunnel->gro_cells);
1309         free_percpu(dev->tstats);
1310         free_netdev(dev);
1311 }
1312
1313 #define GRE_FEATURES (NETIF_F_SG |              \
1314                       NETIF_F_FRAGLIST |        \
1315                       NETIF_F_HIGHDMA |         \
1316                       NETIF_F_HW_CSUM)
1317
1318 static void ipgre_tunnel_setup(struct net_device *dev)
1319 {
1320         dev->netdev_ops         = &ipgre_netdev_ops;
1321         dev->destructor         = ipgre_dev_free;
1322
1323         dev->type               = ARPHRD_IPGRE;
1324         dev->needed_headroom    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1325         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1326         dev->flags              = IFF_NOARP;
1327         dev->iflink             = 0;
1328         dev->addr_len           = 4;
1329         dev->features           |= NETIF_F_NETNS_LOCAL;
1330         dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
1331
1332         dev->features           |= GRE_FEATURES;
1333         dev->hw_features        |= GRE_FEATURES;
1334 }
1335
1336 static int ipgre_tunnel_init(struct net_device *dev)
1337 {
1338         struct ip_tunnel *tunnel;
1339         struct iphdr *iph;
1340         int err;
1341
1342         tunnel = netdev_priv(dev);
1343         iph = &tunnel->parms.iph;
1344
1345         tunnel->dev = dev;
1346         strcpy(tunnel->parms.name, dev->name);
1347
1348         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1349         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1350
1351         if (iph->daddr) {
1352 #ifdef CONFIG_NET_IPGRE_BROADCAST
1353                 if (ipv4_is_multicast(iph->daddr)) {
1354                         if (!iph->saddr)
1355                                 return -EINVAL;
1356                         dev->flags = IFF_BROADCAST;
1357                         dev->header_ops = &ipgre_header_ops;
1358                 }
1359 #endif
1360         } else
1361                 dev->header_ops = &ipgre_header_ops;
1362
1363         dev->tstats = alloc_percpu(struct pcpu_tstats);
1364         if (!dev->tstats)
1365                 return -ENOMEM;
1366
1367         err = gro_cells_init(&tunnel->gro_cells, dev);
1368         if (err) {
1369                 free_percpu(dev->tstats);
1370                 return err;
1371         }
1372
1373         return 0;
1374 }
1375
1376 static void ipgre_fb_tunnel_init(struct net_device *dev)
1377 {
1378         struct ip_tunnel *tunnel = netdev_priv(dev);
1379         struct iphdr *iph = &tunnel->parms.iph;
1380
1381         tunnel->dev = dev;
1382         strcpy(tunnel->parms.name, dev->name);
1383
1384         iph->version            = 4;
1385         iph->protocol           = IPPROTO_GRE;
1386         iph->ihl                = 5;
1387         tunnel->hlen            = sizeof(struct iphdr) + 4;
1388
1389         dev_hold(dev);
1390 }
1391
1392
1393 static const struct gre_protocol ipgre_protocol = {
1394         .handler     = ipgre_rcv,
1395         .err_handler = ipgre_err,
1396 };
1397
1398 static void ipgre_destroy_tunnels(struct ipgre_net *ign, struct list_head *head)
1399 {
1400         int prio;
1401
1402         for (prio = 0; prio < 4; prio++) {
1403                 int h;
1404                 for (h = 0; h < HASH_SIZE; h++) {
1405                         struct ip_tunnel *t;
1406
1407                         t = rtnl_dereference(ign->tunnels[prio][h]);
1408
1409                         while (t != NULL) {
1410                                 unregister_netdevice_queue(t->dev, head);
1411                                 t = rtnl_dereference(t->next);
1412                         }
1413                 }
1414         }
1415 }
1416
1417 static int __net_init ipgre_init_net(struct net *net)
1418 {
1419         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1420         int err;
1421
1422         ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1423                                            ipgre_tunnel_setup);
1424         if (!ign->fb_tunnel_dev) {
1425                 err = -ENOMEM;
1426                 goto err_alloc_dev;
1427         }
1428         dev_net_set(ign->fb_tunnel_dev, net);
1429
1430         ipgre_fb_tunnel_init(ign->fb_tunnel_dev);
1431         ign->fb_tunnel_dev->rtnl_link_ops = &ipgre_link_ops;
1432
1433         if ((err = register_netdev(ign->fb_tunnel_dev)))
1434                 goto err_reg_dev;
1435
1436         rcu_assign_pointer(ign->tunnels_wc[0],
1437                            netdev_priv(ign->fb_tunnel_dev));
1438         return 0;
1439
1440 err_reg_dev:
1441         ipgre_dev_free(ign->fb_tunnel_dev);
1442 err_alloc_dev:
1443         return err;
1444 }
1445
1446 static void __net_exit ipgre_exit_net(struct net *net)
1447 {
1448         struct ipgre_net *ign;
1449         LIST_HEAD(list);
1450
1451         ign = net_generic(net, ipgre_net_id);
1452         rtnl_lock();
1453         ipgre_destroy_tunnels(ign, &list);
1454         unregister_netdevice_many(&list);
1455         rtnl_unlock();
1456 }
1457
1458 static struct pernet_operations ipgre_net_ops = {
1459         .init = ipgre_init_net,
1460         .exit = ipgre_exit_net,
1461         .id   = &ipgre_net_id,
1462         .size = sizeof(struct ipgre_net),
1463 };
1464
1465 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1466 {
1467         __be16 flags;
1468
1469         if (!data)
1470                 return 0;
1471
1472         flags = 0;
1473         if (data[IFLA_GRE_IFLAGS])
1474                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1475         if (data[IFLA_GRE_OFLAGS])
1476                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1477         if (flags & (GRE_VERSION|GRE_ROUTING))
1478                 return -EINVAL;
1479
1480         return 0;
1481 }
1482
1483 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1484 {
1485         __be32 daddr;
1486
1487         if (tb[IFLA_ADDRESS]) {
1488                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1489                         return -EINVAL;
1490                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1491                         return -EADDRNOTAVAIL;
1492         }
1493
1494         if (!data)
1495                 goto out;
1496
1497         if (data[IFLA_GRE_REMOTE]) {
1498                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1499                 if (!daddr)
1500                         return -EINVAL;
1501         }
1502
1503 out:
1504         return ipgre_tunnel_validate(tb, data);
1505 }
1506
1507 static void ipgre_netlink_parms(struct nlattr *data[],
1508                                 struct ip_tunnel_parm *parms)
1509 {
1510         memset(parms, 0, sizeof(*parms));
1511
1512         parms->iph.protocol = IPPROTO_GRE;
1513
1514         if (!data)
1515                 return;
1516
1517         if (data[IFLA_GRE_LINK])
1518                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1519
1520         if (data[IFLA_GRE_IFLAGS])
1521                 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1522
1523         if (data[IFLA_GRE_OFLAGS])
1524                 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1525
1526         if (data[IFLA_GRE_IKEY])
1527                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1528
1529         if (data[IFLA_GRE_OKEY])
1530                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1531
1532         if (data[IFLA_GRE_LOCAL])
1533                 parms->iph.saddr = nla_get_be32(data[IFLA_GRE_LOCAL]);
1534
1535         if (data[IFLA_GRE_REMOTE])
1536                 parms->iph.daddr = nla_get_be32(data[IFLA_GRE_REMOTE]);
1537
1538         if (data[IFLA_GRE_TTL])
1539                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1540
1541         if (data[IFLA_GRE_TOS])
1542                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1543
1544         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC]))
1545                 parms->iph.frag_off = htons(IP_DF);
1546 }
1547
1548 static int ipgre_tap_init(struct net_device *dev)
1549 {
1550         struct ip_tunnel *tunnel;
1551
1552         tunnel = netdev_priv(dev);
1553
1554         tunnel->dev = dev;
1555         strcpy(tunnel->parms.name, dev->name);
1556
1557         ipgre_tunnel_bind_dev(dev);
1558
1559         dev->tstats = alloc_percpu(struct pcpu_tstats);
1560         if (!dev->tstats)
1561                 return -ENOMEM;
1562
1563         return 0;
1564 }
1565
1566 static const struct net_device_ops ipgre_tap_netdev_ops = {
1567         .ndo_init               = ipgre_tap_init,
1568         .ndo_uninit             = ipgre_tunnel_uninit,
1569         .ndo_start_xmit         = ipgre_tunnel_xmit,
1570         .ndo_set_mac_address    = eth_mac_addr,
1571         .ndo_validate_addr      = eth_validate_addr,
1572         .ndo_change_mtu         = ipgre_tunnel_change_mtu,
1573         .ndo_get_stats64        = ipgre_get_stats64,
1574 };
1575
1576 static void ipgre_tap_setup(struct net_device *dev)
1577 {
1578
1579         ether_setup(dev);
1580
1581         dev->netdev_ops         = &ipgre_tap_netdev_ops;
1582         dev->destructor         = ipgre_dev_free;
1583
1584         dev->iflink             = 0;
1585         dev->features           |= NETIF_F_NETNS_LOCAL;
1586 }
1587
1588 static int ipgre_newlink(struct net *src_net, struct net_device *dev, struct nlattr *tb[],
1589                          struct nlattr *data[])
1590 {
1591         struct ip_tunnel *nt;
1592         struct net *net = dev_net(dev);
1593         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1594         int mtu;
1595         int err;
1596
1597         nt = netdev_priv(dev);
1598         ipgre_netlink_parms(data, &nt->parms);
1599
1600         if (ipgre_tunnel_find(net, &nt->parms, dev->type))
1601                 return -EEXIST;
1602
1603         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1604                 eth_hw_addr_random(dev);
1605
1606         mtu = ipgre_tunnel_bind_dev(dev);
1607         if (!tb[IFLA_MTU])
1608                 dev->mtu = mtu;
1609
1610         /* Can use a lockless transmit, unless we generate output sequences */
1611         if (!(nt->parms.o_flags & GRE_SEQ))
1612                 dev->features |= NETIF_F_LLTX;
1613
1614         err = register_netdevice(dev);
1615         if (err)
1616                 goto out;
1617
1618         dev_hold(dev);
1619         ipgre_tunnel_link(ign, nt);
1620
1621 out:
1622         return err;
1623 }
1624
1625 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1626                             struct nlattr *data[])
1627 {
1628         struct ip_tunnel *t, *nt;
1629         struct net *net = dev_net(dev);
1630         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1631         struct ip_tunnel_parm p;
1632         int mtu;
1633
1634         if (dev == ign->fb_tunnel_dev)
1635                 return -EINVAL;
1636
1637         nt = netdev_priv(dev);
1638         ipgre_netlink_parms(data, &p);
1639
1640         t = ipgre_tunnel_locate(net, &p, 0);
1641
1642         if (t) {
1643                 if (t->dev != dev)
1644                         return -EEXIST;
1645         } else {
1646                 t = nt;
1647
1648                 if (dev->type != ARPHRD_ETHER) {
1649                         unsigned int nflags = 0;
1650
1651                         if (ipv4_is_multicast(p.iph.daddr))
1652                                 nflags = IFF_BROADCAST;
1653                         else if (p.iph.daddr)
1654                                 nflags = IFF_POINTOPOINT;
1655
1656                         if ((dev->flags ^ nflags) &
1657                             (IFF_POINTOPOINT | IFF_BROADCAST))
1658                                 return -EINVAL;
1659                 }
1660
1661                 ipgre_tunnel_unlink(ign, t);
1662                 t->parms.iph.saddr = p.iph.saddr;
1663                 t->parms.iph.daddr = p.iph.daddr;
1664                 t->parms.i_key = p.i_key;
1665                 if (dev->type != ARPHRD_ETHER) {
1666                         memcpy(dev->dev_addr, &p.iph.saddr, 4);
1667                         memcpy(dev->broadcast, &p.iph.daddr, 4);
1668                 }
1669                 ipgre_tunnel_link(ign, t);
1670                 netdev_state_change(dev);
1671         }
1672
1673         t->parms.o_key = p.o_key;
1674         t->parms.iph.ttl = p.iph.ttl;
1675         t->parms.iph.tos = p.iph.tos;
1676         t->parms.iph.frag_off = p.iph.frag_off;
1677
1678         if (t->parms.link != p.link) {
1679                 t->parms.link = p.link;
1680                 mtu = ipgre_tunnel_bind_dev(dev);
1681                 if (!tb[IFLA_MTU])
1682                         dev->mtu = mtu;
1683                 netdev_state_change(dev);
1684         }
1685
1686         return 0;
1687 }
1688
1689 static size_t ipgre_get_size(const struct net_device *dev)
1690 {
1691         return
1692                 /* IFLA_GRE_LINK */
1693                 nla_total_size(4) +
1694                 /* IFLA_GRE_IFLAGS */
1695                 nla_total_size(2) +
1696                 /* IFLA_GRE_OFLAGS */
1697                 nla_total_size(2) +
1698                 /* IFLA_GRE_IKEY */
1699                 nla_total_size(4) +
1700                 /* IFLA_GRE_OKEY */
1701                 nla_total_size(4) +
1702                 /* IFLA_GRE_LOCAL */
1703                 nla_total_size(4) +
1704                 /* IFLA_GRE_REMOTE */
1705                 nla_total_size(4) +
1706                 /* IFLA_GRE_TTL */
1707                 nla_total_size(1) +
1708                 /* IFLA_GRE_TOS */
1709                 nla_total_size(1) +
1710                 /* IFLA_GRE_PMTUDISC */
1711                 nla_total_size(1) +
1712                 0;
1713 }
1714
1715 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1716 {
1717         struct ip_tunnel *t = netdev_priv(dev);
1718         struct ip_tunnel_parm *p = &t->parms;
1719
1720         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1721             nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1722             nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1723             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1724             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1725             nla_put_be32(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1726             nla_put_be32(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1727             nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1728             nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1729             nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1730                        !!(p->iph.frag_off & htons(IP_DF))))
1731                 goto nla_put_failure;
1732         return 0;
1733
1734 nla_put_failure:
1735         return -EMSGSIZE;
1736 }
1737
1738 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1739         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1740         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1741         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1742         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1743         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1744         [IFLA_GRE_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1745         [IFLA_GRE_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1746         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1747         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1748         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1749 };
1750
1751 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1752         .kind           = "gre",
1753         .maxtype        = IFLA_GRE_MAX,
1754         .policy         = ipgre_policy,
1755         .priv_size      = sizeof(struct ip_tunnel),
1756         .setup          = ipgre_tunnel_setup,
1757         .validate       = ipgre_tunnel_validate,
1758         .newlink        = ipgre_newlink,
1759         .changelink     = ipgre_changelink,
1760         .get_size       = ipgre_get_size,
1761         .fill_info      = ipgre_fill_info,
1762 };
1763
1764 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1765         .kind           = "gretap",
1766         .maxtype        = IFLA_GRE_MAX,
1767         .policy         = ipgre_policy,
1768         .priv_size      = sizeof(struct ip_tunnel),
1769         .setup          = ipgre_tap_setup,
1770         .validate       = ipgre_tap_validate,
1771         .newlink        = ipgre_newlink,
1772         .changelink     = ipgre_changelink,
1773         .get_size       = ipgre_get_size,
1774         .fill_info      = ipgre_fill_info,
1775 };
1776
1777 /*
1778  *      And now the modules code and kernel interface.
1779  */
1780
1781 static int __init ipgre_init(void)
1782 {
1783         int err;
1784
1785         pr_info("GRE over IPv4 tunneling driver\n");
1786
1787         err = register_pernet_device(&ipgre_net_ops);
1788         if (err < 0)
1789                 return err;
1790
1791         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1792         if (err < 0) {
1793                 pr_info("%s: can't add protocol\n", __func__);
1794                 goto add_proto_failed;
1795         }
1796
1797         err = rtnl_link_register(&ipgre_link_ops);
1798         if (err < 0)
1799                 goto rtnl_link_failed;
1800
1801         err = rtnl_link_register(&ipgre_tap_ops);
1802         if (err < 0)
1803                 goto tap_ops_failed;
1804
1805 out:
1806         return err;
1807
1808 tap_ops_failed:
1809         rtnl_link_unregister(&ipgre_link_ops);
1810 rtnl_link_failed:
1811         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1812 add_proto_failed:
1813         unregister_pernet_device(&ipgre_net_ops);
1814         goto out;
1815 }
1816
1817 static void __exit ipgre_fini(void)
1818 {
1819         rtnl_link_unregister(&ipgre_tap_ops);
1820         rtnl_link_unregister(&ipgre_link_ops);
1821         if (gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0)
1822                 pr_info("%s: can't remove protocol\n", __func__);
1823         unregister_pernet_device(&ipgre_net_ops);
1824 }
1825
1826 module_init(ipgre_init);
1827 module_exit(ipgre_fini);
1828 MODULE_LICENSE("GPL");
1829 MODULE_ALIAS_RTNL_LINK("gre");
1830 MODULE_ALIAS_RTNL_LINK("gretap");
1831 MODULE_ALIAS_NETDEV("gre0");