]> Pileus Git - ~andy/linux/blob - net/ipv4/ip_gre.c
IP_GRE: Fix GRE_CSUM case.
[~andy/linux] / 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 struct sk_buff *handle_offloads(struct ip_tunnel *tunnel, struct sk_buff *skb)
739 {
740         int err;
741
742         if (skb_is_gso(skb)) {
743                 err = skb_unclone(skb, GFP_ATOMIC);
744                 if (unlikely(err))
745                         goto error;
746                 skb_shinfo(skb)->gso_type |= SKB_GSO_GRE;
747                 return skb;
748         } else if (skb->ip_summed == CHECKSUM_PARTIAL &&
749                    tunnel->parms.o_flags&GRE_CSUM) {
750                 err = skb_checksum_help(skb);
751                 if (unlikely(err))
752                         goto error;
753         } else if (skb->ip_summed != CHECKSUM_PARTIAL)
754                 skb->ip_summed = CHECKSUM_NONE;
755
756         return skb;
757
758 error:
759         kfree_skb(skb);
760         return ERR_PTR(err);
761 }
762
763 static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
764 {
765         struct pcpu_tstats *tstats = this_cpu_ptr(dev->tstats);
766         struct ip_tunnel *tunnel = netdev_priv(dev);
767         const struct iphdr  *old_iph;
768         const struct iphdr  *tiph;
769         struct flowi4 fl4;
770         u8     tos;
771         __be16 df;
772         struct rtable *rt;                      /* Route to the other host */
773         struct net_device *tdev;                /* Device to other host */
774         struct iphdr  *iph;                     /* Our new IP header */
775         unsigned int max_headroom;              /* The extra header space needed */
776         int    gre_hlen;
777         __be32 dst;
778         int    mtu;
779         u8     ttl;
780         int    err;
781         int    pkt_len;
782
783         skb = handle_offloads(tunnel, skb);
784         if (IS_ERR(skb)) {
785                 dev->stats.tx_dropped++;
786                 return NETDEV_TX_OK;
787         }
788
789         if (!skb->encapsulation) {
790                 skb_reset_inner_headers(skb);
791                 skb->encapsulation = 1;
792         }
793
794         old_iph = ip_hdr(skb);
795
796         if (dev->type == ARPHRD_ETHER)
797                 IPCB(skb)->flags = 0;
798
799         if (dev->header_ops && dev->type == ARPHRD_IPGRE) {
800                 gre_hlen = 0;
801                 if (skb->protocol == htons(ETH_P_IP))
802                         tiph = (const struct iphdr *)skb->data;
803                 else
804                         tiph = &tunnel->parms.iph;
805         } else {
806                 gre_hlen = tunnel->hlen;
807                 tiph = &tunnel->parms.iph;
808         }
809
810         if ((dst = tiph->daddr) == 0) {
811                 /* NBMA tunnel */
812
813                 if (skb_dst(skb) == NULL) {
814                         dev->stats.tx_fifo_errors++;
815                         goto tx_error;
816                 }
817
818                 if (skb->protocol == htons(ETH_P_IP)) {
819                         rt = skb_rtable(skb);
820                         dst = rt_nexthop(rt, old_iph->daddr);
821                 }
822 #if IS_ENABLED(CONFIG_IPV6)
823                 else if (skb->protocol == htons(ETH_P_IPV6)) {
824                         const struct in6_addr *addr6;
825                         struct neighbour *neigh;
826                         bool do_tx_error_icmp;
827                         int addr_type;
828
829                         neigh = dst_neigh_lookup(skb_dst(skb), &ipv6_hdr(skb)->daddr);
830                         if (neigh == NULL)
831                                 goto tx_error;
832
833                         addr6 = (const struct in6_addr *)&neigh->primary_key;
834                         addr_type = ipv6_addr_type(addr6);
835
836                         if (addr_type == IPV6_ADDR_ANY) {
837                                 addr6 = &ipv6_hdr(skb)->daddr;
838                                 addr_type = ipv6_addr_type(addr6);
839                         }
840
841                         if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
842                                 do_tx_error_icmp = true;
843                         else {
844                                 do_tx_error_icmp = false;
845                                 dst = addr6->s6_addr32[3];
846                         }
847                         neigh_release(neigh);
848                         if (do_tx_error_icmp)
849                                 goto tx_error_icmp;
850                 }
851 #endif
852                 else
853                         goto tx_error;
854         }
855
856         ttl = tiph->ttl;
857         tos = tiph->tos;
858         if (tos & 0x1) {
859                 tos &= ~0x1;
860                 if (skb->protocol == htons(ETH_P_IP))
861                         tos = old_iph->tos;
862                 else if (skb->protocol == htons(ETH_P_IPV6))
863                         tos = ipv6_get_dsfield((const struct ipv6hdr *)old_iph);
864         }
865
866         rt = ip_route_output_gre(dev_net(dev), &fl4, dst, tiph->saddr,
867                                  tunnel->parms.o_key, RT_TOS(tos),
868                                  tunnel->parms.link);
869         if (IS_ERR(rt)) {
870                 dev->stats.tx_carrier_errors++;
871                 goto tx_error;
872         }
873         tdev = rt->dst.dev;
874
875         if (tdev == dev) {
876                 ip_rt_put(rt);
877                 dev->stats.collisions++;
878                 goto tx_error;
879         }
880
881         df = tiph->frag_off;
882         if (df)
883                 mtu = dst_mtu(&rt->dst) - dev->hard_header_len - tunnel->hlen;
884         else
885                 mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
886
887         if (skb_dst(skb))
888                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
889
890         if (skb->protocol == htons(ETH_P_IP)) {
891                 df |= (old_iph->frag_off&htons(IP_DF));
892
893                 if (!skb_is_gso(skb) &&
894                     (old_iph->frag_off&htons(IP_DF)) &&
895                     mtu < ntohs(old_iph->tot_len)) {
896                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
897                         ip_rt_put(rt);
898                         goto tx_error;
899                 }
900         }
901 #if IS_ENABLED(CONFIG_IPV6)
902         else if (skb->protocol == htons(ETH_P_IPV6)) {
903                 struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
904
905                 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && mtu >= IPV6_MIN_MTU) {
906                         if ((tunnel->parms.iph.daddr &&
907                              !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
908                             rt6->rt6i_dst.plen == 128) {
909                                 rt6->rt6i_flags |= RTF_MODIFIED;
910                                 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
911                         }
912                 }
913
914                 if (!skb_is_gso(skb) &&
915                     mtu >= IPV6_MIN_MTU &&
916                     mtu < skb->len - tunnel->hlen + gre_hlen) {
917                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
918                         ip_rt_put(rt);
919                         goto tx_error;
920                 }
921         }
922 #endif
923
924         if (tunnel->err_count > 0) {
925                 if (time_before(jiffies,
926                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
927                         tunnel->err_count--;
928
929                         dst_link_failure(skb);
930                 } else
931                         tunnel->err_count = 0;
932         }
933
934         max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen + rt->dst.header_len;
935
936         if (skb_headroom(skb) < max_headroom || skb_shared(skb)||
937             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
938                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
939                 if (max_headroom > dev->needed_headroom)
940                         dev->needed_headroom = max_headroom;
941                 if (!new_skb) {
942                         ip_rt_put(rt);
943                         dev->stats.tx_dropped++;
944                         dev_kfree_skb(skb);
945                         return NETDEV_TX_OK;
946                 }
947                 if (skb->sk)
948                         skb_set_owner_w(new_skb, skb->sk);
949                 dev_kfree_skb(skb);
950                 skb = new_skb;
951                 old_iph = ip_hdr(skb);
952                 /* Warning : tiph value might point to freed memory */
953         }
954
955         skb_push(skb, gre_hlen);
956         skb_reset_network_header(skb);
957         skb_set_transport_header(skb, sizeof(*iph));
958         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
959         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
960                               IPSKB_REROUTED);
961         skb_dst_drop(skb);
962         skb_dst_set(skb, &rt->dst);
963
964         /*
965          *      Push down and install the IPIP header.
966          */
967
968         iph                     =       ip_hdr(skb);
969         iph->version            =       4;
970         iph->ihl                =       sizeof(struct iphdr) >> 2;
971         iph->frag_off           =       df;
972         iph->protocol           =       IPPROTO_GRE;
973         iph->tos                =       ipgre_ecn_encapsulate(tos, old_iph, skb);
974         iph->daddr              =       fl4.daddr;
975         iph->saddr              =       fl4.saddr;
976         iph->ttl                =       ttl;
977
978         tunnel_ip_select_ident(skb, old_iph, &rt->dst);
979
980         if (ttl == 0) {
981                 if (skb->protocol == htons(ETH_P_IP))
982                         iph->ttl = old_iph->ttl;
983 #if IS_ENABLED(CONFIG_IPV6)
984                 else if (skb->protocol == htons(ETH_P_IPV6))
985                         iph->ttl = ((const struct ipv6hdr *)old_iph)->hop_limit;
986 #endif
987                 else
988                         iph->ttl = ip4_dst_hoplimit(&rt->dst);
989         }
990
991         ((__be16 *)(iph + 1))[0] = tunnel->parms.o_flags;
992         ((__be16 *)(iph + 1))[1] = (dev->type == ARPHRD_ETHER) ?
993                                    htons(ETH_P_TEB) : skb->protocol;
994
995         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
996                 __be32 *ptr = (__be32 *)(((u8 *)iph) + tunnel->hlen - 4);
997
998                 if (tunnel->parms.o_flags&GRE_SEQ) {
999                         ++tunnel->o_seqno;
1000                         *ptr = htonl(tunnel->o_seqno);
1001                         ptr--;
1002                 }
1003                 if (tunnel->parms.o_flags&GRE_KEY) {
1004                         *ptr = tunnel->parms.o_key;
1005                         ptr--;
1006                 }
1007                 /* Skip GRE checksum if skb is getting offloaded. */
1008                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_GRE) &&
1009                     (tunnel->parms.o_flags&GRE_CSUM)) {
1010                         int offset = skb_transport_offset(skb);
1011
1012                         if (skb_has_shared_frag(skb)) {
1013                                 err = __skb_linearize(skb);
1014                                 if (err)
1015                                         goto tx_error;
1016                         }
1017
1018                         *ptr = 0;
1019                         *(__sum16 *)ptr = csum_fold(skb_checksum(skb, offset,
1020                                                                  skb->len - offset,
1021                                                                  0));
1022                 }
1023         }
1024
1025         nf_reset(skb);
1026
1027         pkt_len = skb->len - skb_transport_offset(skb);
1028         err = ip_local_out(skb);
1029         if (likely(net_xmit_eval(err) == 0)) {
1030                 u64_stats_update_begin(&tstats->syncp);
1031                 tstats->tx_bytes += pkt_len;
1032                 tstats->tx_packets++;
1033                 u64_stats_update_end(&tstats->syncp);
1034         } else {
1035                 dev->stats.tx_errors++;
1036                 dev->stats.tx_aborted_errors++;
1037         }
1038         return NETDEV_TX_OK;
1039
1040 #if IS_ENABLED(CONFIG_IPV6)
1041 tx_error_icmp:
1042         dst_link_failure(skb);
1043 #endif
1044 tx_error:
1045         dev->stats.tx_errors++;
1046         dev_kfree_skb(skb);
1047         return NETDEV_TX_OK;
1048 }
1049
1050 static int ipgre_tunnel_bind_dev(struct net_device *dev)
1051 {
1052         struct net_device *tdev = NULL;
1053         struct ip_tunnel *tunnel;
1054         const struct iphdr *iph;
1055         int hlen = LL_MAX_HEADER;
1056         int mtu = ETH_DATA_LEN;
1057         int addend = sizeof(struct iphdr) + 4;
1058
1059         tunnel = netdev_priv(dev);
1060         iph = &tunnel->parms.iph;
1061
1062         /* Guess output device to choose reasonable mtu and needed_headroom */
1063
1064         if (iph->daddr) {
1065                 struct flowi4 fl4;
1066                 struct rtable *rt;
1067
1068                 rt = ip_route_output_gre(dev_net(dev), &fl4,
1069                                          iph->daddr, iph->saddr,
1070                                          tunnel->parms.o_key,
1071                                          RT_TOS(iph->tos),
1072                                          tunnel->parms.link);
1073                 if (!IS_ERR(rt)) {
1074                         tdev = rt->dst.dev;
1075                         ip_rt_put(rt);
1076                 }
1077
1078                 if (dev->type != ARPHRD_ETHER)
1079                         dev->flags |= IFF_POINTOPOINT;
1080         }
1081
1082         if (!tdev && tunnel->parms.link)
1083                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
1084
1085         if (tdev) {
1086                 hlen = tdev->hard_header_len + tdev->needed_headroom;
1087                 mtu = tdev->mtu;
1088         }
1089         dev->iflink = tunnel->parms.link;
1090
1091         /* Precalculate GRE options length */
1092         if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
1093                 if (tunnel->parms.o_flags&GRE_CSUM)
1094                         addend += 4;
1095                 if (tunnel->parms.o_flags&GRE_KEY)
1096                         addend += 4;
1097                 if (tunnel->parms.o_flags&GRE_SEQ)
1098                         addend += 4;
1099         }
1100         dev->needed_headroom = addend + hlen;
1101         mtu -= dev->hard_header_len + addend;
1102
1103         if (mtu < 68)
1104                 mtu = 68;
1105
1106         tunnel->hlen = addend;
1107         /* TCP offload with GRE SEQ is not supported. */
1108         if (!(tunnel->parms.o_flags & GRE_SEQ)) {
1109                 /* device supports enc gso offload*/
1110                 if (tdev->hw_enc_features & NETIF_F_GRE_GSO) {
1111                         dev->features           |= NETIF_F_TSO;
1112                         dev->hw_features        |= NETIF_F_TSO;
1113                 } else {
1114                         dev->features           |= NETIF_F_GSO_SOFTWARE;
1115                         dev->hw_features        |= NETIF_F_GSO_SOFTWARE;
1116                 }
1117         }
1118
1119         return mtu;
1120 }
1121
1122 static int
1123 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
1124 {
1125         int err = 0;
1126         struct ip_tunnel_parm p;
1127         struct ip_tunnel *t;
1128         struct net *net = dev_net(dev);
1129         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1130
1131         switch (cmd) {
1132         case SIOCGETTUNNEL:
1133                 t = NULL;
1134                 if (dev == ign->fb_tunnel_dev) {
1135                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1136                                 err = -EFAULT;
1137                                 break;
1138                         }
1139                         t = ipgre_tunnel_locate(net, &p, 0);
1140                 }
1141                 if (t == NULL)
1142                         t = netdev_priv(dev);
1143                 memcpy(&p, &t->parms, sizeof(p));
1144                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1145                         err = -EFAULT;
1146                 break;
1147
1148         case SIOCADDTUNNEL:
1149         case SIOCCHGTUNNEL:
1150                 err = -EPERM;
1151                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1152                         goto done;
1153
1154                 err = -EFAULT;
1155                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1156                         goto done;
1157
1158                 err = -EINVAL;
1159                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
1160                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
1161                     ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
1162                         goto done;
1163                 if (p.iph.ttl)
1164                         p.iph.frag_off |= htons(IP_DF);
1165
1166                 if (!(p.i_flags&GRE_KEY))
1167                         p.i_key = 0;
1168                 if (!(p.o_flags&GRE_KEY))
1169                         p.o_key = 0;
1170
1171                 t = ipgre_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
1172
1173                 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1174                         if (t != NULL) {
1175                                 if (t->dev != dev) {
1176                                         err = -EEXIST;
1177                                         break;
1178                                 }
1179                         } else {
1180                                 unsigned int nflags = 0;
1181
1182                                 t = netdev_priv(dev);
1183
1184                                 if (ipv4_is_multicast(p.iph.daddr))
1185                                         nflags = IFF_BROADCAST;
1186                                 else if (p.iph.daddr)
1187                                         nflags = IFF_POINTOPOINT;
1188
1189                                 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
1190                                         err = -EINVAL;
1191                                         break;
1192                                 }
1193                                 ipgre_tunnel_unlink(ign, t);
1194                                 synchronize_net();
1195                                 t->parms.iph.saddr = p.iph.saddr;
1196                                 t->parms.iph.daddr = p.iph.daddr;
1197                                 t->parms.i_key = p.i_key;
1198                                 t->parms.o_key = p.o_key;
1199                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
1200                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
1201                                 ipgre_tunnel_link(ign, t);
1202                                 netdev_state_change(dev);
1203                         }
1204                 }
1205
1206                 if (t) {
1207                         err = 0;
1208                         if (cmd == SIOCCHGTUNNEL) {
1209                                 t->parms.iph.ttl = p.iph.ttl;
1210                                 t->parms.iph.tos = p.iph.tos;
1211                                 t->parms.iph.frag_off = p.iph.frag_off;
1212                                 if (t->parms.link != p.link) {
1213                                         t->parms.link = p.link;
1214                                         dev->mtu = ipgre_tunnel_bind_dev(dev);
1215                                         netdev_state_change(dev);
1216                                 }
1217                         }
1218                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
1219                                 err = -EFAULT;
1220                 } else
1221                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1222                 break;
1223
1224         case SIOCDELTUNNEL:
1225                 err = -EPERM;
1226                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1227                         goto done;
1228
1229                 if (dev == ign->fb_tunnel_dev) {
1230                         err = -EFAULT;
1231                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1232                                 goto done;
1233                         err = -ENOENT;
1234                         if ((t = ipgre_tunnel_locate(net, &p, 0)) == NULL)
1235                                 goto done;
1236                         err = -EPERM;
1237                         if (t == netdev_priv(ign->fb_tunnel_dev))
1238                                 goto done;
1239                         dev = t->dev;
1240                 }
1241                 unregister_netdevice(dev);
1242                 err = 0;
1243                 break;
1244
1245         default:
1246                 err = -EINVAL;
1247         }
1248
1249 done:
1250         return err;
1251 }
1252
1253 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1254 {
1255         struct ip_tunnel *tunnel = netdev_priv(dev);
1256         if (new_mtu < 68 ||
1257             new_mtu > 0xFFF8 - dev->hard_header_len - tunnel->hlen)
1258                 return -EINVAL;
1259         dev->mtu = new_mtu;
1260         return 0;
1261 }
1262
1263 /* Nice toy. Unfortunately, useless in real life :-)
1264    It allows to construct virtual multiprotocol broadcast "LAN"
1265    over the Internet, provided multicast routing is tuned.
1266
1267
1268    I have no idea was this bicycle invented before me,
1269    so that I had to set ARPHRD_IPGRE to a random value.
1270    I have an impression, that Cisco could make something similar,
1271    but this feature is apparently missing in IOS<=11.2(8).
1272
1273    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1274    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1275
1276    ping -t 255 224.66.66.66
1277
1278    If nobody answers, mbone does not work.
1279
1280    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1281    ip addr add 10.66.66.<somewhat>/24 dev Universe
1282    ifconfig Universe up
1283    ifconfig Universe add fe80::<Your_real_addr>/10
1284    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1285    ftp 10.66.66.66
1286    ...
1287    ftp fec0:6666:6666::193.233.7.65
1288    ...
1289
1290  */
1291
1292 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
1293                         unsigned short type,
1294                         const void *daddr, const void *saddr, unsigned int len)
1295 {
1296         struct ip_tunnel *t = netdev_priv(dev);
1297         struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1298         __be16 *p = (__be16 *)(iph+1);
1299
1300         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1301         p[0]            = t->parms.o_flags;
1302         p[1]            = htons(type);
1303
1304         /*
1305          *      Set the source hardware address.
1306          */
1307
1308         if (saddr)
1309                 memcpy(&iph->saddr, saddr, 4);
1310         if (daddr)
1311                 memcpy(&iph->daddr, daddr, 4);
1312         if (iph->daddr)
1313                 return t->hlen;
1314
1315         return -t->hlen;
1316 }
1317
1318 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
1319 {
1320         const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
1321         memcpy(haddr, &iph->saddr, 4);
1322         return 4;
1323 }
1324
1325 static const struct header_ops ipgre_header_ops = {
1326         .create = ipgre_header,
1327         .parse  = ipgre_header_parse,
1328 };
1329
1330 #ifdef CONFIG_NET_IPGRE_BROADCAST
1331 static int ipgre_open(struct net_device *dev)
1332 {
1333         struct ip_tunnel *t = netdev_priv(dev);
1334
1335         if (ipv4_is_multicast(t->parms.iph.daddr)) {
1336                 struct flowi4 fl4;
1337                 struct rtable *rt;
1338
1339                 rt = ip_route_output_gre(dev_net(dev), &fl4,
1340                                          t->parms.iph.daddr,
1341                                          t->parms.iph.saddr,
1342                                          t->parms.o_key,
1343                                          RT_TOS(t->parms.iph.tos),
1344                                          t->parms.link);
1345                 if (IS_ERR(rt))
1346                         return -EADDRNOTAVAIL;
1347                 dev = rt->dst.dev;
1348                 ip_rt_put(rt);
1349                 if (__in_dev_get_rtnl(dev) == NULL)
1350                         return -EADDRNOTAVAIL;
1351                 t->mlink = dev->ifindex;
1352                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1353         }
1354         return 0;
1355 }
1356
1357 static int ipgre_close(struct net_device *dev)
1358 {
1359         struct ip_tunnel *t = netdev_priv(dev);
1360
1361         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
1362                 struct in_device *in_dev;
1363                 in_dev = inetdev_by_index(dev_net(dev), t->mlink);
1364                 if (in_dev)
1365                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1366         }
1367         return 0;
1368 }
1369
1370 #endif
1371
1372 static const struct net_device_ops ipgre_netdev_ops = {
1373         .ndo_init               = ipgre_tunnel_init,
1374         .ndo_uninit             = ipgre_tunnel_uninit,
1375 #ifdef CONFIG_NET_IPGRE_BROADCAST
1376         .ndo_open               = ipgre_open,
1377         .ndo_stop               = ipgre_close,
1378 #endif
1379         .ndo_start_xmit         = ipgre_tunnel_xmit,
1380         .ndo_do_ioctl           = ipgre_tunnel_ioctl,
1381         .ndo_change_mtu         = ipgre_tunnel_change_mtu,
1382         .ndo_get_stats64        = ipgre_get_stats64,
1383 };
1384
1385 static void ipgre_dev_free(struct net_device *dev)
1386 {
1387         struct ip_tunnel *tunnel = netdev_priv(dev);
1388
1389         gro_cells_destroy(&tunnel->gro_cells);
1390         free_percpu(dev->tstats);
1391         free_netdev(dev);
1392 }
1393
1394 #define GRE_FEATURES (NETIF_F_SG |              \
1395                       NETIF_F_FRAGLIST |        \
1396                       NETIF_F_HIGHDMA |         \
1397                       NETIF_F_HW_CSUM)
1398
1399 static void ipgre_tunnel_setup(struct net_device *dev)
1400 {
1401         dev->netdev_ops         = &ipgre_netdev_ops;
1402         dev->destructor         = ipgre_dev_free;
1403
1404         dev->type               = ARPHRD_IPGRE;
1405         dev->needed_headroom    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1406         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1407         dev->flags              = IFF_NOARP;
1408         dev->iflink             = 0;
1409         dev->addr_len           = 4;
1410         dev->features           |= NETIF_F_NETNS_LOCAL;
1411         dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
1412
1413         dev->features           |= GRE_FEATURES;
1414         dev->hw_features        |= GRE_FEATURES;
1415 }
1416
1417 static int ipgre_tunnel_init(struct net_device *dev)
1418 {
1419         struct ip_tunnel *tunnel;
1420         struct iphdr *iph;
1421         int err;
1422
1423         tunnel = netdev_priv(dev);
1424         iph = &tunnel->parms.iph;
1425
1426         tunnel->dev = dev;
1427         strcpy(tunnel->parms.name, dev->name);
1428
1429         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1430         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1431
1432         if (iph->daddr) {
1433 #ifdef CONFIG_NET_IPGRE_BROADCAST
1434                 if (ipv4_is_multicast(iph->daddr)) {
1435                         if (!iph->saddr)
1436                                 return -EINVAL;
1437                         dev->flags = IFF_BROADCAST;
1438                         dev->header_ops = &ipgre_header_ops;
1439                 }
1440 #endif
1441         } else
1442                 dev->header_ops = &ipgre_header_ops;
1443
1444         dev->tstats = alloc_percpu(struct pcpu_tstats);
1445         if (!dev->tstats)
1446                 return -ENOMEM;
1447
1448         err = gro_cells_init(&tunnel->gro_cells, dev);
1449         if (err) {
1450                 free_percpu(dev->tstats);
1451                 return err;
1452         }
1453
1454         return 0;
1455 }
1456
1457 static void ipgre_fb_tunnel_init(struct net_device *dev)
1458 {
1459         struct ip_tunnel *tunnel = netdev_priv(dev);
1460         struct iphdr *iph = &tunnel->parms.iph;
1461
1462         tunnel->dev = dev;
1463         strcpy(tunnel->parms.name, dev->name);
1464
1465         iph->version            = 4;
1466         iph->protocol           = IPPROTO_GRE;
1467         iph->ihl                = 5;
1468         tunnel->hlen            = sizeof(struct iphdr) + 4;
1469
1470         dev_hold(dev);
1471 }
1472
1473
1474 static const struct gre_protocol ipgre_protocol = {
1475         .handler     = ipgre_rcv,
1476         .err_handler = ipgre_err,
1477 };
1478
1479 static void ipgre_destroy_tunnels(struct ipgre_net *ign, struct list_head *head)
1480 {
1481         int prio;
1482
1483         for (prio = 0; prio < 4; prio++) {
1484                 int h;
1485                 for (h = 0; h < HASH_SIZE; h++) {
1486                         struct ip_tunnel *t;
1487
1488                         t = rtnl_dereference(ign->tunnels[prio][h]);
1489
1490                         while (t != NULL) {
1491                                 unregister_netdevice_queue(t->dev, head);
1492                                 t = rtnl_dereference(t->next);
1493                         }
1494                 }
1495         }
1496 }
1497
1498 static int __net_init ipgre_init_net(struct net *net)
1499 {
1500         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1501         int err;
1502
1503         ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1504                                            ipgre_tunnel_setup);
1505         if (!ign->fb_tunnel_dev) {
1506                 err = -ENOMEM;
1507                 goto err_alloc_dev;
1508         }
1509         dev_net_set(ign->fb_tunnel_dev, net);
1510
1511         ipgre_fb_tunnel_init(ign->fb_tunnel_dev);
1512         ign->fb_tunnel_dev->rtnl_link_ops = &ipgre_link_ops;
1513
1514         if ((err = register_netdev(ign->fb_tunnel_dev)))
1515                 goto err_reg_dev;
1516
1517         rcu_assign_pointer(ign->tunnels_wc[0],
1518                            netdev_priv(ign->fb_tunnel_dev));
1519         return 0;
1520
1521 err_reg_dev:
1522         ipgre_dev_free(ign->fb_tunnel_dev);
1523 err_alloc_dev:
1524         return err;
1525 }
1526
1527 static void __net_exit ipgre_exit_net(struct net *net)
1528 {
1529         struct ipgre_net *ign;
1530         LIST_HEAD(list);
1531
1532         ign = net_generic(net, ipgre_net_id);
1533         rtnl_lock();
1534         ipgre_destroy_tunnels(ign, &list);
1535         unregister_netdevice_many(&list);
1536         rtnl_unlock();
1537 }
1538
1539 static struct pernet_operations ipgre_net_ops = {
1540         .init = ipgre_init_net,
1541         .exit = ipgre_exit_net,
1542         .id   = &ipgre_net_id,
1543         .size = sizeof(struct ipgre_net),
1544 };
1545
1546 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1547 {
1548         __be16 flags;
1549
1550         if (!data)
1551                 return 0;
1552
1553         flags = 0;
1554         if (data[IFLA_GRE_IFLAGS])
1555                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1556         if (data[IFLA_GRE_OFLAGS])
1557                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1558         if (flags & (GRE_VERSION|GRE_ROUTING))
1559                 return -EINVAL;
1560
1561         return 0;
1562 }
1563
1564 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1565 {
1566         __be32 daddr;
1567
1568         if (tb[IFLA_ADDRESS]) {
1569                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1570                         return -EINVAL;
1571                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1572                         return -EADDRNOTAVAIL;
1573         }
1574
1575         if (!data)
1576                 goto out;
1577
1578         if (data[IFLA_GRE_REMOTE]) {
1579                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1580                 if (!daddr)
1581                         return -EINVAL;
1582         }
1583
1584 out:
1585         return ipgre_tunnel_validate(tb, data);
1586 }
1587
1588 static void ipgre_netlink_parms(struct nlattr *data[],
1589                                 struct ip_tunnel_parm *parms)
1590 {
1591         memset(parms, 0, sizeof(*parms));
1592
1593         parms->iph.protocol = IPPROTO_GRE;
1594
1595         if (!data)
1596                 return;
1597
1598         if (data[IFLA_GRE_LINK])
1599                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1600
1601         if (data[IFLA_GRE_IFLAGS])
1602                 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1603
1604         if (data[IFLA_GRE_OFLAGS])
1605                 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1606
1607         if (data[IFLA_GRE_IKEY])
1608                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1609
1610         if (data[IFLA_GRE_OKEY])
1611                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1612
1613         if (data[IFLA_GRE_LOCAL])
1614                 parms->iph.saddr = nla_get_be32(data[IFLA_GRE_LOCAL]);
1615
1616         if (data[IFLA_GRE_REMOTE])
1617                 parms->iph.daddr = nla_get_be32(data[IFLA_GRE_REMOTE]);
1618
1619         if (data[IFLA_GRE_TTL])
1620                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1621
1622         if (data[IFLA_GRE_TOS])
1623                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1624
1625         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC]))
1626                 parms->iph.frag_off = htons(IP_DF);
1627 }
1628
1629 static int ipgre_tap_init(struct net_device *dev)
1630 {
1631         struct ip_tunnel *tunnel;
1632
1633         tunnel = netdev_priv(dev);
1634
1635         tunnel->dev = dev;
1636         strcpy(tunnel->parms.name, dev->name);
1637
1638         ipgre_tunnel_bind_dev(dev);
1639
1640         dev->tstats = alloc_percpu(struct pcpu_tstats);
1641         if (!dev->tstats)
1642                 return -ENOMEM;
1643
1644         return 0;
1645 }
1646
1647 static const struct net_device_ops ipgre_tap_netdev_ops = {
1648         .ndo_init               = ipgre_tap_init,
1649         .ndo_uninit             = ipgre_tunnel_uninit,
1650         .ndo_start_xmit         = ipgre_tunnel_xmit,
1651         .ndo_set_mac_address    = eth_mac_addr,
1652         .ndo_validate_addr      = eth_validate_addr,
1653         .ndo_change_mtu         = ipgre_tunnel_change_mtu,
1654         .ndo_get_stats64        = ipgre_get_stats64,
1655 };
1656
1657 static void ipgre_tap_setup(struct net_device *dev)
1658 {
1659
1660         ether_setup(dev);
1661
1662         dev->netdev_ops         = &ipgre_tap_netdev_ops;
1663         dev->destructor         = ipgre_dev_free;
1664
1665         dev->iflink             = 0;
1666         dev->features           |= NETIF_F_NETNS_LOCAL;
1667
1668         dev->features           |= GRE_FEATURES;
1669         dev->hw_features        |= GRE_FEATURES;
1670 }
1671
1672 static int ipgre_newlink(struct net *src_net, struct net_device *dev, struct nlattr *tb[],
1673                          struct nlattr *data[])
1674 {
1675         struct ip_tunnel *nt;
1676         struct net *net = dev_net(dev);
1677         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1678         int mtu;
1679         int err;
1680
1681         nt = netdev_priv(dev);
1682         ipgre_netlink_parms(data, &nt->parms);
1683
1684         if (ipgre_tunnel_find(net, &nt->parms, dev->type))
1685                 return -EEXIST;
1686
1687         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1688                 eth_hw_addr_random(dev);
1689
1690         mtu = ipgre_tunnel_bind_dev(dev);
1691         if (!tb[IFLA_MTU])
1692                 dev->mtu = mtu;
1693
1694         /* Can use a lockless transmit, unless we generate output sequences */
1695         if (!(nt->parms.o_flags & GRE_SEQ))
1696                 dev->features |= NETIF_F_LLTX;
1697
1698         err = register_netdevice(dev);
1699         if (err)
1700                 goto out;
1701
1702         dev_hold(dev);
1703         ipgre_tunnel_link(ign, nt);
1704
1705 out:
1706         return err;
1707 }
1708
1709 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1710                             struct nlattr *data[])
1711 {
1712         struct ip_tunnel *t, *nt;
1713         struct net *net = dev_net(dev);
1714         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1715         struct ip_tunnel_parm p;
1716         int mtu;
1717
1718         if (dev == ign->fb_tunnel_dev)
1719                 return -EINVAL;
1720
1721         nt = netdev_priv(dev);
1722         ipgre_netlink_parms(data, &p);
1723
1724         t = ipgre_tunnel_locate(net, &p, 0);
1725
1726         if (t) {
1727                 if (t->dev != dev)
1728                         return -EEXIST;
1729         } else {
1730                 t = nt;
1731
1732                 if (dev->type != ARPHRD_ETHER) {
1733                         unsigned int nflags = 0;
1734
1735                         if (ipv4_is_multicast(p.iph.daddr))
1736                                 nflags = IFF_BROADCAST;
1737                         else if (p.iph.daddr)
1738                                 nflags = IFF_POINTOPOINT;
1739
1740                         if ((dev->flags ^ nflags) &
1741                             (IFF_POINTOPOINT | IFF_BROADCAST))
1742                                 return -EINVAL;
1743                 }
1744
1745                 ipgre_tunnel_unlink(ign, t);
1746                 t->parms.iph.saddr = p.iph.saddr;
1747                 t->parms.iph.daddr = p.iph.daddr;
1748                 t->parms.i_key = p.i_key;
1749                 if (dev->type != ARPHRD_ETHER) {
1750                         memcpy(dev->dev_addr, &p.iph.saddr, 4);
1751                         memcpy(dev->broadcast, &p.iph.daddr, 4);
1752                 }
1753                 ipgre_tunnel_link(ign, t);
1754                 netdev_state_change(dev);
1755         }
1756
1757         t->parms.o_key = p.o_key;
1758         t->parms.iph.ttl = p.iph.ttl;
1759         t->parms.iph.tos = p.iph.tos;
1760         t->parms.iph.frag_off = p.iph.frag_off;
1761
1762         if (t->parms.link != p.link) {
1763                 t->parms.link = p.link;
1764                 mtu = ipgre_tunnel_bind_dev(dev);
1765                 if (!tb[IFLA_MTU])
1766                         dev->mtu = mtu;
1767                 netdev_state_change(dev);
1768         }
1769
1770         return 0;
1771 }
1772
1773 static size_t ipgre_get_size(const struct net_device *dev)
1774 {
1775         return
1776                 /* IFLA_GRE_LINK */
1777                 nla_total_size(4) +
1778                 /* IFLA_GRE_IFLAGS */
1779                 nla_total_size(2) +
1780                 /* IFLA_GRE_OFLAGS */
1781                 nla_total_size(2) +
1782                 /* IFLA_GRE_IKEY */
1783                 nla_total_size(4) +
1784                 /* IFLA_GRE_OKEY */
1785                 nla_total_size(4) +
1786                 /* IFLA_GRE_LOCAL */
1787                 nla_total_size(4) +
1788                 /* IFLA_GRE_REMOTE */
1789                 nla_total_size(4) +
1790                 /* IFLA_GRE_TTL */
1791                 nla_total_size(1) +
1792                 /* IFLA_GRE_TOS */
1793                 nla_total_size(1) +
1794                 /* IFLA_GRE_PMTUDISC */
1795                 nla_total_size(1) +
1796                 0;
1797 }
1798
1799 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1800 {
1801         struct ip_tunnel *t = netdev_priv(dev);
1802         struct ip_tunnel_parm *p = &t->parms;
1803
1804         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1805             nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1806             nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1807             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1808             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1809             nla_put_be32(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1810             nla_put_be32(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1811             nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1812             nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1813             nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1814                        !!(p->iph.frag_off & htons(IP_DF))))
1815                 goto nla_put_failure;
1816         return 0;
1817
1818 nla_put_failure:
1819         return -EMSGSIZE;
1820 }
1821
1822 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1823         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1824         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1825         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1826         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1827         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1828         [IFLA_GRE_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1829         [IFLA_GRE_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1830         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1831         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1832         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1833 };
1834
1835 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1836         .kind           = "gre",
1837         .maxtype        = IFLA_GRE_MAX,
1838         .policy         = ipgre_policy,
1839         .priv_size      = sizeof(struct ip_tunnel),
1840         .setup          = ipgre_tunnel_setup,
1841         .validate       = ipgre_tunnel_validate,
1842         .newlink        = ipgre_newlink,
1843         .changelink     = ipgre_changelink,
1844         .get_size       = ipgre_get_size,
1845         .fill_info      = ipgre_fill_info,
1846 };
1847
1848 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1849         .kind           = "gretap",
1850         .maxtype        = IFLA_GRE_MAX,
1851         .policy         = ipgre_policy,
1852         .priv_size      = sizeof(struct ip_tunnel),
1853         .setup          = ipgre_tap_setup,
1854         .validate       = ipgre_tap_validate,
1855         .newlink        = ipgre_newlink,
1856         .changelink     = ipgre_changelink,
1857         .get_size       = ipgre_get_size,
1858         .fill_info      = ipgre_fill_info,
1859 };
1860
1861 /*
1862  *      And now the modules code and kernel interface.
1863  */
1864
1865 static int __init ipgre_init(void)
1866 {
1867         int err;
1868
1869         pr_info("GRE over IPv4 tunneling driver\n");
1870
1871         err = register_pernet_device(&ipgre_net_ops);
1872         if (err < 0)
1873                 return err;
1874
1875         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1876         if (err < 0) {
1877                 pr_info("%s: can't add protocol\n", __func__);
1878                 goto add_proto_failed;
1879         }
1880
1881         err = rtnl_link_register(&ipgre_link_ops);
1882         if (err < 0)
1883                 goto rtnl_link_failed;
1884
1885         err = rtnl_link_register(&ipgre_tap_ops);
1886         if (err < 0)
1887                 goto tap_ops_failed;
1888
1889 out:
1890         return err;
1891
1892 tap_ops_failed:
1893         rtnl_link_unregister(&ipgre_link_ops);
1894 rtnl_link_failed:
1895         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1896 add_proto_failed:
1897         unregister_pernet_device(&ipgre_net_ops);
1898         goto out;
1899 }
1900
1901 static void __exit ipgre_fini(void)
1902 {
1903         rtnl_link_unregister(&ipgre_tap_ops);
1904         rtnl_link_unregister(&ipgre_link_ops);
1905         if (gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0)
1906                 pr_info("%s: can't remove protocol\n", __func__);
1907         unregister_pernet_device(&ipgre_net_ops);
1908 }
1909
1910 module_init(ipgre_init);
1911 module_exit(ipgre_fini);
1912 MODULE_LICENSE("GPL");
1913 MODULE_ALIAS_RTNL_LINK("gre");
1914 MODULE_ALIAS_RTNL_LINK("gretap");
1915 MODULE_ALIAS_NETDEV("gre0");