GNU Linux-libre 5.10.153-gnu1
[releases.git] / net / ipv4 / ip_gre.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *      Linux NET3:     GRE over IP protocol decoder.
4  *
5  *      Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
6  */
7
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10 #include <linux/capability.h>
11 #include <linux/module.h>
12 #include <linux/types.h>
13 #include <linux/kernel.h>
14 #include <linux/slab.h>
15 #include <linux/uaccess.h>
16 #include <linux/skbuff.h>
17 #include <linux/netdevice.h>
18 #include <linux/in.h>
19 #include <linux/tcp.h>
20 #include <linux/udp.h>
21 #include <linux/if_arp.h>
22 #include <linux/if_vlan.h>
23 #include <linux/init.h>
24 #include <linux/in6.h>
25 #include <linux/inetdevice.h>
26 #include <linux/igmp.h>
27 #include <linux/netfilter_ipv4.h>
28 #include <linux/etherdevice.h>
29 #include <linux/if_ether.h>
30
31 #include <net/sock.h>
32 #include <net/ip.h>
33 #include <net/icmp.h>
34 #include <net/protocol.h>
35 #include <net/ip_tunnels.h>
36 #include <net/arp.h>
37 #include <net/checksum.h>
38 #include <net/dsfield.h>
39 #include <net/inet_ecn.h>
40 #include <net/xfrm.h>
41 #include <net/net_namespace.h>
42 #include <net/netns/generic.h>
43 #include <net/rtnetlink.h>
44 #include <net/gre.h>
45 #include <net/dst_metadata.h>
46 #include <net/erspan.h>
47
48 /*
49    Problems & solutions
50    --------------------
51
52    1. The most important issue is detecting local dead loops.
53    They would cause complete host lockup in transmit, which
54    would be "resolved" by stack overflow or, if queueing is enabled,
55    with infinite looping in net_bh.
56
57    We cannot track such dead loops during route installation,
58    it is infeasible task. The most general solutions would be
59    to keep skb->encapsulation counter (sort of local ttl),
60    and silently drop packet when it expires. It is a good
61    solution, but it supposes maintaining new variable in ALL
62    skb, even if no tunneling is used.
63
64    Current solution: xmit_recursion breaks dead loops. This is a percpu
65    counter, since when we enter the first ndo_xmit(), cpu migration is
66    forbidden. We force an exit if this counter reaches RECURSION_LIMIT
67
68    2. Networking dead loops would not kill routers, but would really
69    kill network. IP hop limit plays role of "t->recursion" in this case,
70    if we copy it from packet being encapsulated to upper header.
71    It is very good solution, but it introduces two problems:
72
73    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
74      do not work over tunnels.
75    - traceroute does not work. I planned to relay ICMP from tunnel,
76      so that this problem would be solved and traceroute output
77      would even more informative. This idea appeared to be wrong:
78      only Linux complies to rfc1812 now (yes, guys, Linux is the only
79      true router now :-)), all routers (at least, in neighbourhood of mine)
80      return only 8 bytes of payload. It is the end.
81
82    Hence, if we want that OSPF worked or traceroute said something reasonable,
83    we should search for another solution.
84
85    One of them is to parse packet trying to detect inner encapsulation
86    made by our node. It is difficult or even impossible, especially,
87    taking into account fragmentation. TO be short, ttl is not solution at all.
88
89    Current solution: The solution was UNEXPECTEDLY SIMPLE.
90    We force DF flag on tunnels with preconfigured hop limit,
91    that is ALL. :-) Well, it does not remove the problem completely,
92    but exponential growth of network traffic is changed to linear
93    (branches, that exceed pmtu are pruned) and tunnel mtu
94    rapidly degrades to value <68, where looping stops.
95    Yes, it is not good if there exists a router in the loop,
96    which does not force DF, even when encapsulating packets have DF set.
97    But it is not our problem! Nobody could accuse us, we made
98    all that we could make. Even if it is your gated who injected
99    fatal route to network, even if it were you who configured
100    fatal static route: you are innocent. :-)
101
102    Alexey Kuznetsov.
103  */
104
105 static bool log_ecn_error = true;
106 module_param(log_ecn_error, bool, 0644);
107 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
108
109 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
110 static int ipgre_tunnel_init(struct net_device *dev);
111 static void erspan_build_header(struct sk_buff *skb,
112                                 u32 id, u32 index,
113                                 bool truncate, bool is_ipv4);
114
115 static unsigned int ipgre_net_id __read_mostly;
116 static unsigned int gre_tap_net_id __read_mostly;
117 static unsigned int erspan_net_id __read_mostly;
118
119 static int ipgre_err(struct sk_buff *skb, u32 info,
120                      const struct tnl_ptk_info *tpi)
121 {
122
123         /* All the routers (except for Linux) return only
124            8 bytes of packet payload. It means, that precise relaying of
125            ICMP in the real Internet is absolutely infeasible.
126
127            Moreover, Cisco "wise men" put GRE key to the third word
128            in GRE header. It makes impossible maintaining even soft
129            state for keyed GRE tunnels with enabled checksum. Tell
130            them "thank you".
131
132            Well, I wonder, rfc1812 was written by Cisco employee,
133            what the hell these idiots break standards established
134            by themselves???
135            */
136         struct net *net = dev_net(skb->dev);
137         struct ip_tunnel_net *itn;
138         const struct iphdr *iph;
139         const int type = icmp_hdr(skb)->type;
140         const int code = icmp_hdr(skb)->code;
141         unsigned int data_len = 0;
142         struct ip_tunnel *t;
143
144         if (tpi->proto == htons(ETH_P_TEB))
145                 itn = net_generic(net, gre_tap_net_id);
146         else if (tpi->proto == htons(ETH_P_ERSPAN) ||
147                  tpi->proto == htons(ETH_P_ERSPAN2))
148                 itn = net_generic(net, erspan_net_id);
149         else
150                 itn = net_generic(net, ipgre_net_id);
151
152         iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
153         t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
154                              iph->daddr, iph->saddr, tpi->key);
155
156         if (!t)
157                 return -ENOENT;
158
159         switch (type) {
160         default:
161         case ICMP_PARAMETERPROB:
162                 return 0;
163
164         case ICMP_DEST_UNREACH:
165                 switch (code) {
166                 case ICMP_SR_FAILED:
167                 case ICMP_PORT_UNREACH:
168                         /* Impossible event. */
169                         return 0;
170                 default:
171                         /* All others are translated to HOST_UNREACH.
172                            rfc2003 contains "deep thoughts" about NET_UNREACH,
173                            I believe they are just ether pollution. --ANK
174                          */
175                         break;
176                 }
177                 break;
178
179         case ICMP_TIME_EXCEEDED:
180                 if (code != ICMP_EXC_TTL)
181                         return 0;
182                 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
183                 break;
184
185         case ICMP_REDIRECT:
186                 break;
187         }
188
189 #if IS_ENABLED(CONFIG_IPV6)
190        if (tpi->proto == htons(ETH_P_IPV6) &&
191            !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
192                                        type, data_len))
193                return 0;
194 #endif
195
196         if (t->parms.iph.daddr == 0 ||
197             ipv4_is_multicast(t->parms.iph.daddr))
198                 return 0;
199
200         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
201                 return 0;
202
203         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
204                 t->err_count++;
205         else
206                 t->err_count = 1;
207         t->err_time = jiffies;
208
209         return 0;
210 }
211
212 static void gre_err(struct sk_buff *skb, u32 info)
213 {
214         /* All the routers (except for Linux) return only
215          * 8 bytes of packet payload. It means, that precise relaying of
216          * ICMP in the real Internet is absolutely infeasible.
217          *
218          * Moreover, Cisco "wise men" put GRE key to the third word
219          * in GRE header. It makes impossible maintaining even soft
220          * state for keyed
221          * GRE tunnels with enabled checksum. Tell them "thank you".
222          *
223          * Well, I wonder, rfc1812 was written by Cisco employee,
224          * what the hell these idiots break standards established
225          * by themselves???
226          */
227
228         const struct iphdr *iph = (struct iphdr *)skb->data;
229         const int type = icmp_hdr(skb)->type;
230         const int code = icmp_hdr(skb)->code;
231         struct tnl_ptk_info tpi;
232
233         if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP),
234                              iph->ihl * 4) < 0)
235                 return;
236
237         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
238                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
239                                  skb->dev->ifindex, IPPROTO_GRE);
240                 return;
241         }
242         if (type == ICMP_REDIRECT) {
243                 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex,
244                               IPPROTO_GRE);
245                 return;
246         }
247
248         ipgre_err(skb, info, &tpi);
249 }
250
251 static bool is_erspan_type1(int gre_hdr_len)
252 {
253         /* Both ERSPAN type I (version 0) and type II (version 1) use
254          * protocol 0x88BE, but the type I has only 4-byte GRE header,
255          * while type II has 8-byte.
256          */
257         return gre_hdr_len == 4;
258 }
259
260 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
261                       int gre_hdr_len)
262 {
263         struct net *net = dev_net(skb->dev);
264         struct metadata_dst *tun_dst = NULL;
265         struct erspan_base_hdr *ershdr;
266         struct ip_tunnel_net *itn;
267         struct ip_tunnel *tunnel;
268         const struct iphdr *iph;
269         struct erspan_md2 *md2;
270         int ver;
271         int len;
272
273         itn = net_generic(net, erspan_net_id);
274         iph = ip_hdr(skb);
275         if (is_erspan_type1(gre_hdr_len)) {
276                 ver = 0;
277                 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
278                                           tpi->flags | TUNNEL_NO_KEY,
279                                           iph->saddr, iph->daddr, 0);
280         } else {
281                 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
282                 ver = ershdr->ver;
283                 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
284                                           tpi->flags | TUNNEL_KEY,
285                                           iph->saddr, iph->daddr, tpi->key);
286         }
287
288         if (tunnel) {
289                 if (is_erspan_type1(gre_hdr_len))
290                         len = gre_hdr_len;
291                 else
292                         len = gre_hdr_len + erspan_hdr_len(ver);
293
294                 if (unlikely(!pskb_may_pull(skb, len)))
295                         return PACKET_REJECT;
296
297                 if (__iptunnel_pull_header(skb,
298                                            len,
299                                            htons(ETH_P_TEB),
300                                            false, false) < 0)
301                         goto drop;
302
303                 if (tunnel->collect_md) {
304                         struct erspan_metadata *pkt_md, *md;
305                         struct ip_tunnel_info *info;
306                         unsigned char *gh;
307                         __be64 tun_id;
308                         __be16 flags;
309
310                         tpi->flags |= TUNNEL_KEY;
311                         flags = tpi->flags;
312                         tun_id = key32_to_tunnel_id(tpi->key);
313
314                         tun_dst = ip_tun_rx_dst(skb, flags,
315                                                 tun_id, sizeof(*md));
316                         if (!tun_dst)
317                                 return PACKET_REJECT;
318
319                         /* skb can be uncloned in __iptunnel_pull_header, so
320                          * old pkt_md is no longer valid and we need to reset
321                          * it
322                          */
323                         gh = skb_network_header(skb) +
324                              skb_network_header_len(skb);
325                         pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len +
326                                                             sizeof(*ershdr));
327                         md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
328                         md->version = ver;
329                         md2 = &md->u.md2;
330                         memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
331                                                        ERSPAN_V2_MDSIZE);
332
333                         info = &tun_dst->u.tun_info;
334                         info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
335                         info->options_len = sizeof(*md);
336                 }
337
338                 skb_reset_mac_header(skb);
339                 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
340                 return PACKET_RCVD;
341         }
342         return PACKET_REJECT;
343
344 drop:
345         kfree_skb(skb);
346         return PACKET_RCVD;
347 }
348
349 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
350                        struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
351 {
352         struct metadata_dst *tun_dst = NULL;
353         const struct iphdr *iph;
354         struct ip_tunnel *tunnel;
355
356         iph = ip_hdr(skb);
357         tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
358                                   iph->saddr, iph->daddr, tpi->key);
359
360         if (tunnel) {
361                 const struct iphdr *tnl_params;
362
363                 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
364                                            raw_proto, false) < 0)
365                         goto drop;
366
367                 if (tunnel->dev->type != ARPHRD_NONE)
368                         skb_pop_mac_header(skb);
369                 else
370                         skb_reset_mac_header(skb);
371
372                 tnl_params = &tunnel->parms.iph;
373                 if (tunnel->collect_md || tnl_params->daddr == 0) {
374                         __be16 flags;
375                         __be64 tun_id;
376
377                         flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
378                         tun_id = key32_to_tunnel_id(tpi->key);
379                         tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
380                         if (!tun_dst)
381                                 return PACKET_REJECT;
382                 }
383
384                 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
385                 return PACKET_RCVD;
386         }
387         return PACKET_NEXT;
388
389 drop:
390         kfree_skb(skb);
391         return PACKET_RCVD;
392 }
393
394 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
395                      int hdr_len)
396 {
397         struct net *net = dev_net(skb->dev);
398         struct ip_tunnel_net *itn;
399         int res;
400
401         if (tpi->proto == htons(ETH_P_TEB))
402                 itn = net_generic(net, gre_tap_net_id);
403         else
404                 itn = net_generic(net, ipgre_net_id);
405
406         res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
407         if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
408                 /* ipgre tunnels in collect metadata mode should receive
409                  * also ETH_P_TEB traffic.
410                  */
411                 itn = net_generic(net, ipgre_net_id);
412                 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
413         }
414         return res;
415 }
416
417 static int gre_rcv(struct sk_buff *skb)
418 {
419         struct tnl_ptk_info tpi;
420         bool csum_err = false;
421         int hdr_len;
422
423 #ifdef CONFIG_NET_IPGRE_BROADCAST
424         if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
425                 /* Looped back packet, drop it! */
426                 if (rt_is_output_route(skb_rtable(skb)))
427                         goto drop;
428         }
429 #endif
430
431         hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
432         if (hdr_len < 0)
433                 goto drop;
434
435         if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
436                      tpi.proto == htons(ETH_P_ERSPAN2))) {
437                 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
438                         return 0;
439                 goto out;
440         }
441
442         if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
443                 return 0;
444
445 out:
446         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
447 drop:
448         kfree_skb(skb);
449         return 0;
450 }
451
452 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
453                        const struct iphdr *tnl_params,
454                        __be16 proto)
455 {
456         struct ip_tunnel *tunnel = netdev_priv(dev);
457         __be16 flags = tunnel->parms.o_flags;
458
459         /* Push GRE header. */
460         gre_build_header(skb, tunnel->tun_hlen,
461                          flags, proto, tunnel->parms.o_key,
462                          (flags & TUNNEL_SEQ) ? htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0);
463
464         ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
465 }
466
467 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
468 {
469         return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
470 }
471
472 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
473                         __be16 proto)
474 {
475         struct ip_tunnel *tunnel = netdev_priv(dev);
476         struct ip_tunnel_info *tun_info;
477         const struct ip_tunnel_key *key;
478         int tunnel_hlen;
479         __be16 flags;
480
481         tun_info = skb_tunnel_info(skb);
482         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
483                      ip_tunnel_info_af(tun_info) != AF_INET))
484                 goto err_free_skb;
485
486         key = &tun_info->key;
487         tunnel_hlen = gre_calc_hlen(key->tun_flags);
488
489         if (skb_cow_head(skb, dev->needed_headroom))
490                 goto err_free_skb;
491
492         /* Push Tunnel header. */
493         if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
494                 goto err_free_skb;
495
496         flags = tun_info->key.tun_flags &
497                 (TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ);
498         gre_build_header(skb, tunnel_hlen, flags, proto,
499                          tunnel_id_to_key32(tun_info->key.tun_id),
500                          (flags & TUNNEL_SEQ) ? htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0);
501
502         ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
503
504         return;
505
506 err_free_skb:
507         kfree_skb(skb);
508         dev->stats.tx_dropped++;
509 }
510
511 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev)
512 {
513         struct ip_tunnel *tunnel = netdev_priv(dev);
514         struct ip_tunnel_info *tun_info;
515         const struct ip_tunnel_key *key;
516         struct erspan_metadata *md;
517         bool truncate = false;
518         __be16 proto;
519         int tunnel_hlen;
520         int version;
521         int nhoff;
522
523         tun_info = skb_tunnel_info(skb);
524         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
525                      ip_tunnel_info_af(tun_info) != AF_INET))
526                 goto err_free_skb;
527
528         key = &tun_info->key;
529         if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT))
530                 goto err_free_skb;
531         if (tun_info->options_len < sizeof(*md))
532                 goto err_free_skb;
533         md = ip_tunnel_info_opts(tun_info);
534
535         /* ERSPAN has fixed 8 byte GRE header */
536         version = md->version;
537         tunnel_hlen = 8 + erspan_hdr_len(version);
538
539         if (skb_cow_head(skb, dev->needed_headroom))
540                 goto err_free_skb;
541
542         if (gre_handle_offloads(skb, false))
543                 goto err_free_skb;
544
545         if (skb->len > dev->mtu + dev->hard_header_len) {
546                 pskb_trim(skb, dev->mtu + dev->hard_header_len);
547                 truncate = true;
548         }
549
550         nhoff = skb_network_header(skb) - skb_mac_header(skb);
551         if (skb->protocol == htons(ETH_P_IP) &&
552             (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
553                 truncate = true;
554
555         if (skb->protocol == htons(ETH_P_IPV6)) {
556                 int thoff;
557
558                 if (skb_transport_header_was_set(skb))
559                         thoff = skb_transport_header(skb) - skb_mac_header(skb);
560                 else
561                         thoff = nhoff + sizeof(struct ipv6hdr);
562                 if (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff)
563                         truncate = true;
564         }
565
566         if (version == 1) {
567                 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
568                                     ntohl(md->u.index), truncate, true);
569                 proto = htons(ETH_P_ERSPAN);
570         } else if (version == 2) {
571                 erspan_build_header_v2(skb,
572                                        ntohl(tunnel_id_to_key32(key->tun_id)),
573                                        md->u.md2.dir,
574                                        get_hwid(&md->u.md2),
575                                        truncate, true);
576                 proto = htons(ETH_P_ERSPAN2);
577         } else {
578                 goto err_free_skb;
579         }
580
581         gre_build_header(skb, 8, TUNNEL_SEQ,
582                          proto, 0, htonl(atomic_fetch_inc(&tunnel->o_seqno)));
583
584         ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
585
586         return;
587
588 err_free_skb:
589         kfree_skb(skb);
590         dev->stats.tx_dropped++;
591 }
592
593 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
594 {
595         struct ip_tunnel_info *info = skb_tunnel_info(skb);
596         const struct ip_tunnel_key *key;
597         struct rtable *rt;
598         struct flowi4 fl4;
599
600         if (ip_tunnel_info_af(info) != AF_INET)
601                 return -EINVAL;
602
603         key = &info->key;
604         ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src,
605                             tunnel_id_to_key32(key->tun_id),
606                             key->tos & ~INET_ECN_MASK, 0, skb->mark,
607                             skb_get_hash(skb));
608         rt = ip_route_output_key(dev_net(dev), &fl4);
609         if (IS_ERR(rt))
610                 return PTR_ERR(rt);
611
612         ip_rt_put(rt);
613         info->key.u.ipv4.src = fl4.saddr;
614         return 0;
615 }
616
617 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
618                               struct net_device *dev)
619 {
620         struct ip_tunnel *tunnel = netdev_priv(dev);
621         const struct iphdr *tnl_params;
622
623         if (!pskb_inet_may_pull(skb))
624                 goto free_skb;
625
626         if (tunnel->collect_md) {
627                 gre_fb_xmit(skb, dev, skb->protocol);
628                 return NETDEV_TX_OK;
629         }
630
631         if (dev->header_ops) {
632                 if (skb_cow_head(skb, 0))
633                         goto free_skb;
634
635                 tnl_params = (const struct iphdr *)skb->data;
636
637                 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
638                  * to gre header.
639                  */
640                 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
641                 skb_reset_mac_header(skb);
642
643                 if (skb->ip_summed == CHECKSUM_PARTIAL &&
644                     skb_checksum_start(skb) < skb->data)
645                         goto free_skb;
646         } else {
647                 if (skb_cow_head(skb, dev->needed_headroom))
648                         goto free_skb;
649
650                 tnl_params = &tunnel->parms.iph;
651         }
652
653         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
654                 goto free_skb;
655
656         __gre_xmit(skb, dev, tnl_params, skb->protocol);
657         return NETDEV_TX_OK;
658
659 free_skb:
660         kfree_skb(skb);
661         dev->stats.tx_dropped++;
662         return NETDEV_TX_OK;
663 }
664
665 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
666                                struct net_device *dev)
667 {
668         struct ip_tunnel *tunnel = netdev_priv(dev);
669         bool truncate = false;
670         __be16 proto;
671
672         if (!pskb_inet_may_pull(skb))
673                 goto free_skb;
674
675         if (tunnel->collect_md) {
676                 erspan_fb_xmit(skb, dev);
677                 return NETDEV_TX_OK;
678         }
679
680         if (gre_handle_offloads(skb, false))
681                 goto free_skb;
682
683         if (skb_cow_head(skb, dev->needed_headroom))
684                 goto free_skb;
685
686         if (skb->len > dev->mtu + dev->hard_header_len) {
687                 pskb_trim(skb, dev->mtu + dev->hard_header_len);
688                 truncate = true;
689         }
690
691         /* Push ERSPAN header */
692         if (tunnel->erspan_ver == 0) {
693                 proto = htons(ETH_P_ERSPAN);
694                 tunnel->parms.o_flags &= ~TUNNEL_SEQ;
695         } else if (tunnel->erspan_ver == 1) {
696                 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
697                                     tunnel->index,
698                                     truncate, true);
699                 proto = htons(ETH_P_ERSPAN);
700         } else if (tunnel->erspan_ver == 2) {
701                 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
702                                        tunnel->dir, tunnel->hwid,
703                                        truncate, true);
704                 proto = htons(ETH_P_ERSPAN2);
705         } else {
706                 goto free_skb;
707         }
708
709         tunnel->parms.o_flags &= ~TUNNEL_KEY;
710         __gre_xmit(skb, dev, &tunnel->parms.iph, proto);
711         return NETDEV_TX_OK;
712
713 free_skb:
714         kfree_skb(skb);
715         dev->stats.tx_dropped++;
716         return NETDEV_TX_OK;
717 }
718
719 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
720                                 struct net_device *dev)
721 {
722         struct ip_tunnel *tunnel = netdev_priv(dev);
723
724         if (!pskb_inet_may_pull(skb))
725                 goto free_skb;
726
727         if (tunnel->collect_md) {
728                 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
729                 return NETDEV_TX_OK;
730         }
731
732         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
733                 goto free_skb;
734
735         if (skb_cow_head(skb, dev->needed_headroom))
736                 goto free_skb;
737
738         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
739         return NETDEV_TX_OK;
740
741 free_skb:
742         kfree_skb(skb);
743         dev->stats.tx_dropped++;
744         return NETDEV_TX_OK;
745 }
746
747 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
748 {
749         struct ip_tunnel *tunnel = netdev_priv(dev);
750         int len;
751
752         len = tunnel->tun_hlen;
753         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
754         len = tunnel->tun_hlen - len;
755         tunnel->hlen = tunnel->hlen + len;
756
757         if (dev->header_ops)
758                 dev->hard_header_len += len;
759         else
760                 dev->needed_headroom += len;
761
762         if (set_mtu)
763                 dev->mtu = max_t(int, dev->mtu - len, 68);
764
765         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
766                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
767                     tunnel->encap.type == TUNNEL_ENCAP_NONE) {
768                         dev->features |= NETIF_F_GSO_SOFTWARE;
769                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
770                 } else {
771                         dev->features &= ~NETIF_F_GSO_SOFTWARE;
772                         dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
773                 }
774                 dev->features |= NETIF_F_LLTX;
775         } else {
776                 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
777                 dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
778         }
779 }
780
781 static int ipgre_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p,
782                             int cmd)
783 {
784         int err;
785
786         if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
787                 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE ||
788                     p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) ||
789                     ((p->i_flags | p->o_flags) & (GRE_VERSION | GRE_ROUTING)))
790                         return -EINVAL;
791         }
792
793         p->i_flags = gre_flags_to_tnl_flags(p->i_flags);
794         p->o_flags = gre_flags_to_tnl_flags(p->o_flags);
795
796         err = ip_tunnel_ctl(dev, p, cmd);
797         if (err)
798                 return err;
799
800         if (cmd == SIOCCHGTUNNEL) {
801                 struct ip_tunnel *t = netdev_priv(dev);
802
803                 t->parms.i_flags = p->i_flags;
804                 t->parms.o_flags = p->o_flags;
805
806                 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
807                         ipgre_link_update(dev, true);
808         }
809
810         p->i_flags = gre_tnl_flags_to_gre_flags(p->i_flags);
811         p->o_flags = gre_tnl_flags_to_gre_flags(p->o_flags);
812         return 0;
813 }
814
815 /* Nice toy. Unfortunately, useless in real life :-)
816    It allows to construct virtual multiprotocol broadcast "LAN"
817    over the Internet, provided multicast routing is tuned.
818
819
820    I have no idea was this bicycle invented before me,
821    so that I had to set ARPHRD_IPGRE to a random value.
822    I have an impression, that Cisco could make something similar,
823    but this feature is apparently missing in IOS<=11.2(8).
824
825    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
826    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
827
828    ping -t 255 224.66.66.66
829
830    If nobody answers, mbone does not work.
831
832    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
833    ip addr add 10.66.66.<somewhat>/24 dev Universe
834    ifconfig Universe up
835    ifconfig Universe add fe80::<Your_real_addr>/10
836    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
837    ftp 10.66.66.66
838    ...
839    ftp fec0:6666:6666::193.233.7.65
840    ...
841  */
842 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
843                         unsigned short type,
844                         const void *daddr, const void *saddr, unsigned int len)
845 {
846         struct ip_tunnel *t = netdev_priv(dev);
847         struct iphdr *iph;
848         struct gre_base_hdr *greh;
849
850         iph = skb_push(skb, t->hlen + sizeof(*iph));
851         greh = (struct gre_base_hdr *)(iph+1);
852         greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
853         greh->protocol = htons(type);
854
855         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
856
857         /* Set the source hardware address. */
858         if (saddr)
859                 memcpy(&iph->saddr, saddr, 4);
860         if (daddr)
861                 memcpy(&iph->daddr, daddr, 4);
862         if (iph->daddr)
863                 return t->hlen + sizeof(*iph);
864
865         return -(t->hlen + sizeof(*iph));
866 }
867
868 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
869 {
870         const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
871         memcpy(haddr, &iph->saddr, 4);
872         return 4;
873 }
874
875 static const struct header_ops ipgre_header_ops = {
876         .create = ipgre_header,
877         .parse  = ipgre_header_parse,
878 };
879
880 #ifdef CONFIG_NET_IPGRE_BROADCAST
881 static int ipgre_open(struct net_device *dev)
882 {
883         struct ip_tunnel *t = netdev_priv(dev);
884
885         if (ipv4_is_multicast(t->parms.iph.daddr)) {
886                 struct flowi4 fl4;
887                 struct rtable *rt;
888
889                 rt = ip_route_output_gre(t->net, &fl4,
890                                          t->parms.iph.daddr,
891                                          t->parms.iph.saddr,
892                                          t->parms.o_key,
893                                          RT_TOS(t->parms.iph.tos),
894                                          t->parms.link);
895                 if (IS_ERR(rt))
896                         return -EADDRNOTAVAIL;
897                 dev = rt->dst.dev;
898                 ip_rt_put(rt);
899                 if (!__in_dev_get_rtnl(dev))
900                         return -EADDRNOTAVAIL;
901                 t->mlink = dev->ifindex;
902                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
903         }
904         return 0;
905 }
906
907 static int ipgre_close(struct net_device *dev)
908 {
909         struct ip_tunnel *t = netdev_priv(dev);
910
911         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
912                 struct in_device *in_dev;
913                 in_dev = inetdev_by_index(t->net, t->mlink);
914                 if (in_dev)
915                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
916         }
917         return 0;
918 }
919 #endif
920
921 static const struct net_device_ops ipgre_netdev_ops = {
922         .ndo_init               = ipgre_tunnel_init,
923         .ndo_uninit             = ip_tunnel_uninit,
924 #ifdef CONFIG_NET_IPGRE_BROADCAST
925         .ndo_open               = ipgre_open,
926         .ndo_stop               = ipgre_close,
927 #endif
928         .ndo_start_xmit         = ipgre_xmit,
929         .ndo_do_ioctl           = ip_tunnel_ioctl,
930         .ndo_change_mtu         = ip_tunnel_change_mtu,
931         .ndo_get_stats64        = ip_tunnel_get_stats64,
932         .ndo_get_iflink         = ip_tunnel_get_iflink,
933         .ndo_tunnel_ctl         = ipgre_tunnel_ctl,
934 };
935
936 #define GRE_FEATURES (NETIF_F_SG |              \
937                       NETIF_F_FRAGLIST |        \
938                       NETIF_F_HIGHDMA |         \
939                       NETIF_F_HW_CSUM)
940
941 static void ipgre_tunnel_setup(struct net_device *dev)
942 {
943         dev->netdev_ops         = &ipgre_netdev_ops;
944         dev->type               = ARPHRD_IPGRE;
945         ip_tunnel_setup(dev, ipgre_net_id);
946 }
947
948 static void __gre_tunnel_init(struct net_device *dev)
949 {
950         struct ip_tunnel *tunnel;
951
952         tunnel = netdev_priv(dev);
953         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
954         tunnel->parms.iph.protocol = IPPROTO_GRE;
955
956         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
957         dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph);
958
959         dev->features           |= GRE_FEATURES;
960         dev->hw_features        |= GRE_FEATURES;
961
962         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
963                 /* TCP offload with GRE SEQ is not supported, nor
964                  * can we support 2 levels of outer headers requiring
965                  * an update.
966                  */
967                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
968                     (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
969                         dev->features    |= NETIF_F_GSO_SOFTWARE;
970                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
971                 }
972
973                 /* Can use a lockless transmit, unless we generate
974                  * output sequences
975                  */
976                 dev->features |= NETIF_F_LLTX;
977         }
978 }
979
980 static int ipgre_tunnel_init(struct net_device *dev)
981 {
982         struct ip_tunnel *tunnel = netdev_priv(dev);
983         struct iphdr *iph = &tunnel->parms.iph;
984
985         __gre_tunnel_init(dev);
986
987         memcpy(dev->dev_addr, &iph->saddr, 4);
988         memcpy(dev->broadcast, &iph->daddr, 4);
989
990         dev->flags              = IFF_NOARP;
991         netif_keep_dst(dev);
992         dev->addr_len           = 4;
993
994         if (iph->daddr && !tunnel->collect_md) {
995 #ifdef CONFIG_NET_IPGRE_BROADCAST
996                 if (ipv4_is_multicast(iph->daddr)) {
997                         if (!iph->saddr)
998                                 return -EINVAL;
999                         dev->flags = IFF_BROADCAST;
1000                         dev->header_ops = &ipgre_header_ops;
1001                         dev->hard_header_len = tunnel->hlen + sizeof(*iph);
1002                         dev->needed_headroom = 0;
1003                 }
1004 #endif
1005         } else if (!tunnel->collect_md) {
1006                 dev->header_ops = &ipgre_header_ops;
1007                 dev->hard_header_len = tunnel->hlen + sizeof(*iph);
1008                 dev->needed_headroom = 0;
1009         }
1010
1011         return ip_tunnel_init(dev);
1012 }
1013
1014 static const struct gre_protocol ipgre_protocol = {
1015         .handler     = gre_rcv,
1016         .err_handler = gre_err,
1017 };
1018
1019 static int __net_init ipgre_init_net(struct net *net)
1020 {
1021         return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1022 }
1023
1024 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1025 {
1026         ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1027 }
1028
1029 static struct pernet_operations ipgre_net_ops = {
1030         .init = ipgre_init_net,
1031         .exit_batch = ipgre_exit_batch_net,
1032         .id   = &ipgre_net_id,
1033         .size = sizeof(struct ip_tunnel_net),
1034 };
1035
1036 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1037                                  struct netlink_ext_ack *extack)
1038 {
1039         __be16 flags;
1040
1041         if (!data)
1042                 return 0;
1043
1044         flags = 0;
1045         if (data[IFLA_GRE_IFLAGS])
1046                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1047         if (data[IFLA_GRE_OFLAGS])
1048                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1049         if (flags & (GRE_VERSION|GRE_ROUTING))
1050                 return -EINVAL;
1051
1052         if (data[IFLA_GRE_COLLECT_METADATA] &&
1053             data[IFLA_GRE_ENCAP_TYPE] &&
1054             nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1055                 return -EINVAL;
1056
1057         return 0;
1058 }
1059
1060 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1061                               struct netlink_ext_ack *extack)
1062 {
1063         __be32 daddr;
1064
1065         if (tb[IFLA_ADDRESS]) {
1066                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1067                         return -EINVAL;
1068                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1069                         return -EADDRNOTAVAIL;
1070         }
1071
1072         if (!data)
1073                 goto out;
1074
1075         if (data[IFLA_GRE_REMOTE]) {
1076                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1077                 if (!daddr)
1078                         return -EINVAL;
1079         }
1080
1081 out:
1082         return ipgre_tunnel_validate(tb, data, extack);
1083 }
1084
1085 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1086                            struct netlink_ext_ack *extack)
1087 {
1088         __be16 flags = 0;
1089         int ret;
1090
1091         if (!data)
1092                 return 0;
1093
1094         ret = ipgre_tap_validate(tb, data, extack);
1095         if (ret)
1096                 return ret;
1097
1098         if (data[IFLA_GRE_ERSPAN_VER] &&
1099             nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0)
1100                 return 0;
1101
1102         /* ERSPAN type II/III should only have GRE sequence and key flag */
1103         if (data[IFLA_GRE_OFLAGS])
1104                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1105         if (data[IFLA_GRE_IFLAGS])
1106                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1107         if (!data[IFLA_GRE_COLLECT_METADATA] &&
1108             flags != (GRE_SEQ | GRE_KEY))
1109                 return -EINVAL;
1110
1111         /* ERSPAN Session ID only has 10-bit. Since we reuse
1112          * 32-bit key field as ID, check it's range.
1113          */
1114         if (data[IFLA_GRE_IKEY] &&
1115             (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1116                 return -EINVAL;
1117
1118         if (data[IFLA_GRE_OKEY] &&
1119             (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1120                 return -EINVAL;
1121
1122         return 0;
1123 }
1124
1125 static int ipgre_netlink_parms(struct net_device *dev,
1126                                 struct nlattr *data[],
1127                                 struct nlattr *tb[],
1128                                 struct ip_tunnel_parm *parms,
1129                                 __u32 *fwmark)
1130 {
1131         struct ip_tunnel *t = netdev_priv(dev);
1132
1133         memset(parms, 0, sizeof(*parms));
1134
1135         parms->iph.protocol = IPPROTO_GRE;
1136
1137         if (!data)
1138                 return 0;
1139
1140         if (data[IFLA_GRE_LINK])
1141                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1142
1143         if (data[IFLA_GRE_IFLAGS])
1144                 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1145
1146         if (data[IFLA_GRE_OFLAGS])
1147                 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1148
1149         if (data[IFLA_GRE_IKEY])
1150                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1151
1152         if (data[IFLA_GRE_OKEY])
1153                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1154
1155         if (data[IFLA_GRE_LOCAL])
1156                 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1157
1158         if (data[IFLA_GRE_REMOTE])
1159                 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1160
1161         if (data[IFLA_GRE_TTL])
1162                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1163
1164         if (data[IFLA_GRE_TOS])
1165                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1166
1167         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1168                 if (t->ignore_df)
1169                         return -EINVAL;
1170                 parms->iph.frag_off = htons(IP_DF);
1171         }
1172
1173         if (data[IFLA_GRE_COLLECT_METADATA]) {
1174                 t->collect_md = true;
1175                 if (dev->type == ARPHRD_IPGRE)
1176                         dev->type = ARPHRD_NONE;
1177         }
1178
1179         if (data[IFLA_GRE_IGNORE_DF]) {
1180                 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1181                   && (parms->iph.frag_off & htons(IP_DF)))
1182                         return -EINVAL;
1183                 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1184         }
1185
1186         if (data[IFLA_GRE_FWMARK])
1187                 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1188
1189         return 0;
1190 }
1191
1192 static int erspan_netlink_parms(struct net_device *dev,
1193                                 struct nlattr *data[],
1194                                 struct nlattr *tb[],
1195                                 struct ip_tunnel_parm *parms,
1196                                 __u32 *fwmark)
1197 {
1198         struct ip_tunnel *t = netdev_priv(dev);
1199         int err;
1200
1201         err = ipgre_netlink_parms(dev, data, tb, parms, fwmark);
1202         if (err)
1203                 return err;
1204         if (!data)
1205                 return 0;
1206
1207         if (data[IFLA_GRE_ERSPAN_VER]) {
1208                 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1209
1210                 if (t->erspan_ver > 2)
1211                         return -EINVAL;
1212         }
1213
1214         if (t->erspan_ver == 1) {
1215                 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1216                         t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1217                         if (t->index & ~INDEX_MASK)
1218                                 return -EINVAL;
1219                 }
1220         } else if (t->erspan_ver == 2) {
1221                 if (data[IFLA_GRE_ERSPAN_DIR]) {
1222                         t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1223                         if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1224                                 return -EINVAL;
1225                 }
1226                 if (data[IFLA_GRE_ERSPAN_HWID]) {
1227                         t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1228                         if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1229                                 return -EINVAL;
1230                 }
1231         }
1232
1233         return 0;
1234 }
1235
1236 /* This function returns true when ENCAP attributes are present in the nl msg */
1237 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1238                                       struct ip_tunnel_encap *ipencap)
1239 {
1240         bool ret = false;
1241
1242         memset(ipencap, 0, sizeof(*ipencap));
1243
1244         if (!data)
1245                 return ret;
1246
1247         if (data[IFLA_GRE_ENCAP_TYPE]) {
1248                 ret = true;
1249                 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1250         }
1251
1252         if (data[IFLA_GRE_ENCAP_FLAGS]) {
1253                 ret = true;
1254                 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1255         }
1256
1257         if (data[IFLA_GRE_ENCAP_SPORT]) {
1258                 ret = true;
1259                 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1260         }
1261
1262         if (data[IFLA_GRE_ENCAP_DPORT]) {
1263                 ret = true;
1264                 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1265         }
1266
1267         return ret;
1268 }
1269
1270 static int gre_tap_init(struct net_device *dev)
1271 {
1272         __gre_tunnel_init(dev);
1273         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1274         netif_keep_dst(dev);
1275
1276         return ip_tunnel_init(dev);
1277 }
1278
1279 static const struct net_device_ops gre_tap_netdev_ops = {
1280         .ndo_init               = gre_tap_init,
1281         .ndo_uninit             = ip_tunnel_uninit,
1282         .ndo_start_xmit         = gre_tap_xmit,
1283         .ndo_set_mac_address    = eth_mac_addr,
1284         .ndo_validate_addr      = eth_validate_addr,
1285         .ndo_change_mtu         = ip_tunnel_change_mtu,
1286         .ndo_get_stats64        = ip_tunnel_get_stats64,
1287         .ndo_get_iflink         = ip_tunnel_get_iflink,
1288         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1289 };
1290
1291 static int erspan_tunnel_init(struct net_device *dev)
1292 {
1293         struct ip_tunnel *tunnel = netdev_priv(dev);
1294
1295         if (tunnel->erspan_ver == 0)
1296                 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */
1297         else
1298                 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */
1299
1300         tunnel->parms.iph.protocol = IPPROTO_GRE;
1301         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1302                        erspan_hdr_len(tunnel->erspan_ver);
1303
1304         dev->features           |= GRE_FEATURES;
1305         dev->hw_features        |= GRE_FEATURES;
1306         dev->priv_flags         |= IFF_LIVE_ADDR_CHANGE;
1307         netif_keep_dst(dev);
1308
1309         return ip_tunnel_init(dev);
1310 }
1311
1312 static const struct net_device_ops erspan_netdev_ops = {
1313         .ndo_init               = erspan_tunnel_init,
1314         .ndo_uninit             = ip_tunnel_uninit,
1315         .ndo_start_xmit         = erspan_xmit,
1316         .ndo_set_mac_address    = eth_mac_addr,
1317         .ndo_validate_addr      = eth_validate_addr,
1318         .ndo_change_mtu         = ip_tunnel_change_mtu,
1319         .ndo_get_stats64        = ip_tunnel_get_stats64,
1320         .ndo_get_iflink         = ip_tunnel_get_iflink,
1321         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1322 };
1323
1324 static void ipgre_tap_setup(struct net_device *dev)
1325 {
1326         ether_setup(dev);
1327         dev->max_mtu = 0;
1328         dev->netdev_ops = &gre_tap_netdev_ops;
1329         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1330         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1331         ip_tunnel_setup(dev, gre_tap_net_id);
1332 }
1333
1334 static int
1335 ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[])
1336 {
1337         struct ip_tunnel_encap ipencap;
1338
1339         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1340                 struct ip_tunnel *t = netdev_priv(dev);
1341                 int err = ip_tunnel_encap_setup(t, &ipencap);
1342
1343                 if (err < 0)
1344                         return err;
1345         }
1346
1347         return 0;
1348 }
1349
1350 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1351                          struct nlattr *tb[], struct nlattr *data[],
1352                          struct netlink_ext_ack *extack)
1353 {
1354         struct ip_tunnel_parm p;
1355         __u32 fwmark = 0;
1356         int err;
1357
1358         err = ipgre_newlink_encap_setup(dev, data);
1359         if (err)
1360                 return err;
1361
1362         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1363         if (err < 0)
1364                 return err;
1365         return ip_tunnel_newlink(dev, tb, &p, fwmark);
1366 }
1367
1368 static int erspan_newlink(struct net *src_net, struct net_device *dev,
1369                           struct nlattr *tb[], struct nlattr *data[],
1370                           struct netlink_ext_ack *extack)
1371 {
1372         struct ip_tunnel_parm p;
1373         __u32 fwmark = 0;
1374         int err;
1375
1376         err = ipgre_newlink_encap_setup(dev, data);
1377         if (err)
1378                 return err;
1379
1380         err = erspan_netlink_parms(dev, data, tb, &p, &fwmark);
1381         if (err)
1382                 return err;
1383         return ip_tunnel_newlink(dev, tb, &p, fwmark);
1384 }
1385
1386 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1387                             struct nlattr *data[],
1388                             struct netlink_ext_ack *extack)
1389 {
1390         struct ip_tunnel *t = netdev_priv(dev);
1391         __u32 fwmark = t->fwmark;
1392         struct ip_tunnel_parm p;
1393         int err;
1394
1395         err = ipgre_newlink_encap_setup(dev, data);
1396         if (err)
1397                 return err;
1398
1399         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1400         if (err < 0)
1401                 return err;
1402
1403         err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1404         if (err < 0)
1405                 return err;
1406
1407         t->parms.i_flags = p.i_flags;
1408         t->parms.o_flags = p.o_flags;
1409
1410         ipgre_link_update(dev, !tb[IFLA_MTU]);
1411
1412         return 0;
1413 }
1414
1415 static int erspan_changelink(struct net_device *dev, struct nlattr *tb[],
1416                              struct nlattr *data[],
1417                              struct netlink_ext_ack *extack)
1418 {
1419         struct ip_tunnel *t = netdev_priv(dev);
1420         __u32 fwmark = t->fwmark;
1421         struct ip_tunnel_parm p;
1422         int err;
1423
1424         err = ipgre_newlink_encap_setup(dev, data);
1425         if (err)
1426                 return err;
1427
1428         err = erspan_netlink_parms(dev, data, tb, &p, &fwmark);
1429         if (err < 0)
1430                 return err;
1431
1432         err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1433         if (err < 0)
1434                 return err;
1435
1436         t->parms.i_flags = p.i_flags;
1437         t->parms.o_flags = p.o_flags;
1438
1439         return 0;
1440 }
1441
1442 static size_t ipgre_get_size(const struct net_device *dev)
1443 {
1444         return
1445                 /* IFLA_GRE_LINK */
1446                 nla_total_size(4) +
1447                 /* IFLA_GRE_IFLAGS */
1448                 nla_total_size(2) +
1449                 /* IFLA_GRE_OFLAGS */
1450                 nla_total_size(2) +
1451                 /* IFLA_GRE_IKEY */
1452                 nla_total_size(4) +
1453                 /* IFLA_GRE_OKEY */
1454                 nla_total_size(4) +
1455                 /* IFLA_GRE_LOCAL */
1456                 nla_total_size(4) +
1457                 /* IFLA_GRE_REMOTE */
1458                 nla_total_size(4) +
1459                 /* IFLA_GRE_TTL */
1460                 nla_total_size(1) +
1461                 /* IFLA_GRE_TOS */
1462                 nla_total_size(1) +
1463                 /* IFLA_GRE_PMTUDISC */
1464                 nla_total_size(1) +
1465                 /* IFLA_GRE_ENCAP_TYPE */
1466                 nla_total_size(2) +
1467                 /* IFLA_GRE_ENCAP_FLAGS */
1468                 nla_total_size(2) +
1469                 /* IFLA_GRE_ENCAP_SPORT */
1470                 nla_total_size(2) +
1471                 /* IFLA_GRE_ENCAP_DPORT */
1472                 nla_total_size(2) +
1473                 /* IFLA_GRE_COLLECT_METADATA */
1474                 nla_total_size(0) +
1475                 /* IFLA_GRE_IGNORE_DF */
1476                 nla_total_size(1) +
1477                 /* IFLA_GRE_FWMARK */
1478                 nla_total_size(4) +
1479                 /* IFLA_GRE_ERSPAN_INDEX */
1480                 nla_total_size(4) +
1481                 /* IFLA_GRE_ERSPAN_VER */
1482                 nla_total_size(1) +
1483                 /* IFLA_GRE_ERSPAN_DIR */
1484                 nla_total_size(1) +
1485                 /* IFLA_GRE_ERSPAN_HWID */
1486                 nla_total_size(2) +
1487                 0;
1488 }
1489
1490 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1491 {
1492         struct ip_tunnel *t = netdev_priv(dev);
1493         struct ip_tunnel_parm *p = &t->parms;
1494         __be16 o_flags = p->o_flags;
1495
1496         if (t->erspan_ver <= 2) {
1497                 if (t->erspan_ver != 0 && !t->collect_md)
1498                         o_flags |= TUNNEL_KEY;
1499
1500                 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1501                         goto nla_put_failure;
1502
1503                 if (t->erspan_ver == 1) {
1504                         if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1505                                 goto nla_put_failure;
1506                 } else if (t->erspan_ver == 2) {
1507                         if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1508                                 goto nla_put_failure;
1509                         if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1510                                 goto nla_put_failure;
1511                 }
1512         }
1513
1514         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1515             nla_put_be16(skb, IFLA_GRE_IFLAGS,
1516                          gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1517             nla_put_be16(skb, IFLA_GRE_OFLAGS,
1518                          gre_tnl_flags_to_gre_flags(o_flags)) ||
1519             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1520             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1521             nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1522             nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1523             nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1524             nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1525             nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1526                        !!(p->iph.frag_off & htons(IP_DF))) ||
1527             nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1528                 goto nla_put_failure;
1529
1530         if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1531                         t->encap.type) ||
1532             nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1533                          t->encap.sport) ||
1534             nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1535                          t->encap.dport) ||
1536             nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1537                         t->encap.flags))
1538                 goto nla_put_failure;
1539
1540         if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1541                 goto nla_put_failure;
1542
1543         if (t->collect_md) {
1544                 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1545                         goto nla_put_failure;
1546         }
1547
1548         return 0;
1549
1550 nla_put_failure:
1551         return -EMSGSIZE;
1552 }
1553
1554 static void erspan_setup(struct net_device *dev)
1555 {
1556         struct ip_tunnel *t = netdev_priv(dev);
1557
1558         ether_setup(dev);
1559         dev->max_mtu = 0;
1560         dev->netdev_ops = &erspan_netdev_ops;
1561         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1562         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1563         ip_tunnel_setup(dev, erspan_net_id);
1564         t->erspan_ver = 1;
1565 }
1566
1567 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1568         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1569         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1570         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1571         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1572         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1573         [IFLA_GRE_LOCAL]        = { .len = sizeof_field(struct iphdr, saddr) },
1574         [IFLA_GRE_REMOTE]       = { .len = sizeof_field(struct iphdr, daddr) },
1575         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1576         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1577         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1578         [IFLA_GRE_ENCAP_TYPE]   = { .type = NLA_U16 },
1579         [IFLA_GRE_ENCAP_FLAGS]  = { .type = NLA_U16 },
1580         [IFLA_GRE_ENCAP_SPORT]  = { .type = NLA_U16 },
1581         [IFLA_GRE_ENCAP_DPORT]  = { .type = NLA_U16 },
1582         [IFLA_GRE_COLLECT_METADATA]     = { .type = NLA_FLAG },
1583         [IFLA_GRE_IGNORE_DF]    = { .type = NLA_U8 },
1584         [IFLA_GRE_FWMARK]       = { .type = NLA_U32 },
1585         [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1586         [IFLA_GRE_ERSPAN_VER]   = { .type = NLA_U8 },
1587         [IFLA_GRE_ERSPAN_DIR]   = { .type = NLA_U8 },
1588         [IFLA_GRE_ERSPAN_HWID]  = { .type = NLA_U16 },
1589 };
1590
1591 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1592         .kind           = "gre",
1593         .maxtype        = IFLA_GRE_MAX,
1594         .policy         = ipgre_policy,
1595         .priv_size      = sizeof(struct ip_tunnel),
1596         .setup          = ipgre_tunnel_setup,
1597         .validate       = ipgre_tunnel_validate,
1598         .newlink        = ipgre_newlink,
1599         .changelink     = ipgre_changelink,
1600         .dellink        = ip_tunnel_dellink,
1601         .get_size       = ipgre_get_size,
1602         .fill_info      = ipgre_fill_info,
1603         .get_link_net   = ip_tunnel_get_link_net,
1604 };
1605
1606 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1607         .kind           = "gretap",
1608         .maxtype        = IFLA_GRE_MAX,
1609         .policy         = ipgre_policy,
1610         .priv_size      = sizeof(struct ip_tunnel),
1611         .setup          = ipgre_tap_setup,
1612         .validate       = ipgre_tap_validate,
1613         .newlink        = ipgre_newlink,
1614         .changelink     = ipgre_changelink,
1615         .dellink        = ip_tunnel_dellink,
1616         .get_size       = ipgre_get_size,
1617         .fill_info      = ipgre_fill_info,
1618         .get_link_net   = ip_tunnel_get_link_net,
1619 };
1620
1621 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1622         .kind           = "erspan",
1623         .maxtype        = IFLA_GRE_MAX,
1624         .policy         = ipgre_policy,
1625         .priv_size      = sizeof(struct ip_tunnel),
1626         .setup          = erspan_setup,
1627         .validate       = erspan_validate,
1628         .newlink        = erspan_newlink,
1629         .changelink     = erspan_changelink,
1630         .dellink        = ip_tunnel_dellink,
1631         .get_size       = ipgre_get_size,
1632         .fill_info      = ipgre_fill_info,
1633         .get_link_net   = ip_tunnel_get_link_net,
1634 };
1635
1636 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1637                                         u8 name_assign_type)
1638 {
1639         struct nlattr *tb[IFLA_MAX + 1];
1640         struct net_device *dev;
1641         LIST_HEAD(list_kill);
1642         struct ip_tunnel *t;
1643         int err;
1644
1645         memset(&tb, 0, sizeof(tb));
1646
1647         dev = rtnl_create_link(net, name, name_assign_type,
1648                                &ipgre_tap_ops, tb, NULL);
1649         if (IS_ERR(dev))
1650                 return dev;
1651
1652         /* Configure flow based GRE device. */
1653         t = netdev_priv(dev);
1654         t->collect_md = true;
1655
1656         err = ipgre_newlink(net, dev, tb, NULL, NULL);
1657         if (err < 0) {
1658                 free_netdev(dev);
1659                 return ERR_PTR(err);
1660         }
1661
1662         /* openvswitch users expect packet sizes to be unrestricted,
1663          * so set the largest MTU we can.
1664          */
1665         err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1666         if (err)
1667                 goto out;
1668
1669         err = rtnl_configure_link(dev, NULL);
1670         if (err < 0)
1671                 goto out;
1672
1673         return dev;
1674 out:
1675         ip_tunnel_dellink(dev, &list_kill);
1676         unregister_netdevice_many(&list_kill);
1677         return ERR_PTR(err);
1678 }
1679 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1680
1681 static int __net_init ipgre_tap_init_net(struct net *net)
1682 {
1683         return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1684 }
1685
1686 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1687 {
1688         ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1689 }
1690
1691 static struct pernet_operations ipgre_tap_net_ops = {
1692         .init = ipgre_tap_init_net,
1693         .exit_batch = ipgre_tap_exit_batch_net,
1694         .id   = &gre_tap_net_id,
1695         .size = sizeof(struct ip_tunnel_net),
1696 };
1697
1698 static int __net_init erspan_init_net(struct net *net)
1699 {
1700         return ip_tunnel_init_net(net, erspan_net_id,
1701                                   &erspan_link_ops, "erspan0");
1702 }
1703
1704 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1705 {
1706         ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1707 }
1708
1709 static struct pernet_operations erspan_net_ops = {
1710         .init = erspan_init_net,
1711         .exit_batch = erspan_exit_batch_net,
1712         .id   = &erspan_net_id,
1713         .size = sizeof(struct ip_tunnel_net),
1714 };
1715
1716 static int __init ipgre_init(void)
1717 {
1718         int err;
1719
1720         pr_info("GRE over IPv4 tunneling driver\n");
1721
1722         err = register_pernet_device(&ipgre_net_ops);
1723         if (err < 0)
1724                 return err;
1725
1726         err = register_pernet_device(&ipgre_tap_net_ops);
1727         if (err < 0)
1728                 goto pnet_tap_failed;
1729
1730         err = register_pernet_device(&erspan_net_ops);
1731         if (err < 0)
1732                 goto pnet_erspan_failed;
1733
1734         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1735         if (err < 0) {
1736                 pr_info("%s: can't add protocol\n", __func__);
1737                 goto add_proto_failed;
1738         }
1739
1740         err = rtnl_link_register(&ipgre_link_ops);
1741         if (err < 0)
1742                 goto rtnl_link_failed;
1743
1744         err = rtnl_link_register(&ipgre_tap_ops);
1745         if (err < 0)
1746                 goto tap_ops_failed;
1747
1748         err = rtnl_link_register(&erspan_link_ops);
1749         if (err < 0)
1750                 goto erspan_link_failed;
1751
1752         return 0;
1753
1754 erspan_link_failed:
1755         rtnl_link_unregister(&ipgre_tap_ops);
1756 tap_ops_failed:
1757         rtnl_link_unregister(&ipgre_link_ops);
1758 rtnl_link_failed:
1759         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1760 add_proto_failed:
1761         unregister_pernet_device(&erspan_net_ops);
1762 pnet_erspan_failed:
1763         unregister_pernet_device(&ipgre_tap_net_ops);
1764 pnet_tap_failed:
1765         unregister_pernet_device(&ipgre_net_ops);
1766         return err;
1767 }
1768
1769 static void __exit ipgre_fini(void)
1770 {
1771         rtnl_link_unregister(&ipgre_tap_ops);
1772         rtnl_link_unregister(&ipgre_link_ops);
1773         rtnl_link_unregister(&erspan_link_ops);
1774         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1775         unregister_pernet_device(&ipgre_tap_net_ops);
1776         unregister_pernet_device(&ipgre_net_ops);
1777         unregister_pernet_device(&erspan_net_ops);
1778 }
1779
1780 module_init(ipgre_init);
1781 module_exit(ipgre_fini);
1782 MODULE_LICENSE("GPL");
1783 MODULE_ALIAS_RTNL_LINK("gre");
1784 MODULE_ALIAS_RTNL_LINK("gretap");
1785 MODULE_ALIAS_RTNL_LINK("erspan");
1786 MODULE_ALIAS_NETDEV("gre0");
1787 MODULE_ALIAS_NETDEV("gretap0");
1788 MODULE_ALIAS_NETDEV("erspan0");