tcp: fix delayed ACKs for MSS boundary condition
[linux-block.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 const struct header_ops ipgre_header_ops;
111
112 static int ipgre_tunnel_init(struct net_device *dev);
113 static void erspan_build_header(struct sk_buff *skb,
114                                 u32 id, u32 index,
115                                 bool truncate, bool is_ipv4);
116
117 static unsigned int ipgre_net_id __read_mostly;
118 static unsigned int gre_tap_net_id __read_mostly;
119 static unsigned int erspan_net_id __read_mostly;
120
121 static int ipgre_err(struct sk_buff *skb, u32 info,
122                      const struct tnl_ptk_info *tpi)
123 {
124
125         /* All the routers (except for Linux) return only
126            8 bytes of packet payload. It means, that precise relaying of
127            ICMP in the real Internet is absolutely infeasible.
128
129            Moreover, Cisco "wise men" put GRE key to the third word
130            in GRE header. It makes impossible maintaining even soft
131            state for keyed GRE tunnels with enabled checksum. Tell
132            them "thank you".
133
134            Well, I wonder, rfc1812 was written by Cisco employee,
135            what the hell these idiots break standards established
136            by themselves???
137            */
138         struct net *net = dev_net(skb->dev);
139         struct ip_tunnel_net *itn;
140         const struct iphdr *iph;
141         const int type = icmp_hdr(skb)->type;
142         const int code = icmp_hdr(skb)->code;
143         unsigned int data_len = 0;
144         struct ip_tunnel *t;
145
146         if (tpi->proto == htons(ETH_P_TEB))
147                 itn = net_generic(net, gre_tap_net_id);
148         else if (tpi->proto == htons(ETH_P_ERSPAN) ||
149                  tpi->proto == htons(ETH_P_ERSPAN2))
150                 itn = net_generic(net, erspan_net_id);
151         else
152                 itn = net_generic(net, ipgre_net_id);
153
154         iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
155         t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
156                              iph->daddr, iph->saddr, tpi->key);
157
158         if (!t)
159                 return -ENOENT;
160
161         switch (type) {
162         default:
163         case ICMP_PARAMETERPROB:
164                 return 0;
165
166         case ICMP_DEST_UNREACH:
167                 switch (code) {
168                 case ICMP_SR_FAILED:
169                 case ICMP_PORT_UNREACH:
170                         /* Impossible event. */
171                         return 0;
172                 default:
173                         /* All others are translated to HOST_UNREACH.
174                            rfc2003 contains "deep thoughts" about NET_UNREACH,
175                            I believe they are just ether pollution. --ANK
176                          */
177                         break;
178                 }
179                 break;
180
181         case ICMP_TIME_EXCEEDED:
182                 if (code != ICMP_EXC_TTL)
183                         return 0;
184                 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
185                 break;
186
187         case ICMP_REDIRECT:
188                 break;
189         }
190
191 #if IS_ENABLED(CONFIG_IPV6)
192         if (tpi->proto == htons(ETH_P_IPV6) &&
193             !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
194                                         type, data_len))
195                 return 0;
196 #endif
197
198         if (t->parms.iph.daddr == 0 ||
199             ipv4_is_multicast(t->parms.iph.daddr))
200                 return 0;
201
202         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
203                 return 0;
204
205         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
206                 t->err_count++;
207         else
208                 t->err_count = 1;
209         t->err_time = jiffies;
210
211         return 0;
212 }
213
214 static void gre_err(struct sk_buff *skb, u32 info)
215 {
216         /* All the routers (except for Linux) return only
217          * 8 bytes of packet payload. It means, that precise relaying of
218          * ICMP in the real Internet is absolutely infeasible.
219          *
220          * Moreover, Cisco "wise men" put GRE key to the third word
221          * in GRE header. It makes impossible maintaining even soft
222          * state for keyed
223          * GRE tunnels with enabled checksum. Tell them "thank you".
224          *
225          * Well, I wonder, rfc1812 was written by Cisco employee,
226          * what the hell these idiots break standards established
227          * by themselves???
228          */
229
230         const struct iphdr *iph = (struct iphdr *)skb->data;
231         const int type = icmp_hdr(skb)->type;
232         const int code = icmp_hdr(skb)->code;
233         struct tnl_ptk_info tpi;
234
235         if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP),
236                              iph->ihl * 4) < 0)
237                 return;
238
239         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
240                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
241                                  skb->dev->ifindex, IPPROTO_GRE);
242                 return;
243         }
244         if (type == ICMP_REDIRECT) {
245                 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex,
246                               IPPROTO_GRE);
247                 return;
248         }
249
250         ipgre_err(skb, info, &tpi);
251 }
252
253 static bool is_erspan_type1(int gre_hdr_len)
254 {
255         /* Both ERSPAN type I (version 0) and type II (version 1) use
256          * protocol 0x88BE, but the type I has only 4-byte GRE header,
257          * while type II has 8-byte.
258          */
259         return gre_hdr_len == 4;
260 }
261
262 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
263                       int gre_hdr_len)
264 {
265         struct net *net = dev_net(skb->dev);
266         struct metadata_dst *tun_dst = NULL;
267         struct erspan_base_hdr *ershdr;
268         struct ip_tunnel_net *itn;
269         struct ip_tunnel *tunnel;
270         const struct iphdr *iph;
271         struct erspan_md2 *md2;
272         int ver;
273         int len;
274
275         itn = net_generic(net, erspan_net_id);
276         iph = ip_hdr(skb);
277         if (is_erspan_type1(gre_hdr_len)) {
278                 ver = 0;
279                 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
280                                           tpi->flags | TUNNEL_NO_KEY,
281                                           iph->saddr, iph->daddr, 0);
282         } else {
283                 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
284                 ver = ershdr->ver;
285                 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
286                                           tpi->flags | TUNNEL_KEY,
287                                           iph->saddr, iph->daddr, tpi->key);
288         }
289
290         if (tunnel) {
291                 if (is_erspan_type1(gre_hdr_len))
292                         len = gre_hdr_len;
293                 else
294                         len = gre_hdr_len + erspan_hdr_len(ver);
295
296                 if (unlikely(!pskb_may_pull(skb, len)))
297                         return PACKET_REJECT;
298
299                 if (__iptunnel_pull_header(skb,
300                                            len,
301                                            htons(ETH_P_TEB),
302                                            false, false) < 0)
303                         goto drop;
304
305                 if (tunnel->collect_md) {
306                         struct erspan_metadata *pkt_md, *md;
307                         struct ip_tunnel_info *info;
308                         unsigned char *gh;
309                         __be64 tun_id;
310                         __be16 flags;
311
312                         tpi->flags |= TUNNEL_KEY;
313                         flags = tpi->flags;
314                         tun_id = key32_to_tunnel_id(tpi->key);
315
316                         tun_dst = ip_tun_rx_dst(skb, flags,
317                                                 tun_id, sizeof(*md));
318                         if (!tun_dst)
319                                 return PACKET_REJECT;
320
321                         /* skb can be uncloned in __iptunnel_pull_header, so
322                          * old pkt_md is no longer valid and we need to reset
323                          * it
324                          */
325                         gh = skb_network_header(skb) +
326                              skb_network_header_len(skb);
327                         pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len +
328                                                             sizeof(*ershdr));
329                         md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
330                         md->version = ver;
331                         md2 = &md->u.md2;
332                         memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
333                                                        ERSPAN_V2_MDSIZE);
334
335                         info = &tun_dst->u.tun_info;
336                         info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
337                         info->options_len = sizeof(*md);
338                 }
339
340                 skb_reset_mac_header(skb);
341                 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
342                 return PACKET_RCVD;
343         }
344         return PACKET_REJECT;
345
346 drop:
347         kfree_skb(skb);
348         return PACKET_RCVD;
349 }
350
351 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
352                        struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
353 {
354         struct metadata_dst *tun_dst = NULL;
355         const struct iphdr *iph;
356         struct ip_tunnel *tunnel;
357
358         iph = ip_hdr(skb);
359         tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
360                                   iph->saddr, iph->daddr, tpi->key);
361
362         if (tunnel) {
363                 const struct iphdr *tnl_params;
364
365                 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
366                                            raw_proto, false) < 0)
367                         goto drop;
368
369                 /* Special case for ipgre_header_parse(), which expects the
370                  * mac_header to point to the outer IP header.
371                  */
372                 if (tunnel->dev->header_ops == &ipgre_header_ops)
373                         skb_pop_mac_header(skb);
374                 else
375                         skb_reset_mac_header(skb);
376
377                 tnl_params = &tunnel->parms.iph;
378                 if (tunnel->collect_md || tnl_params->daddr == 0) {
379                         __be16 flags;
380                         __be64 tun_id;
381
382                         flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
383                         tun_id = key32_to_tunnel_id(tpi->key);
384                         tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
385                         if (!tun_dst)
386                                 return PACKET_REJECT;
387                 }
388
389                 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
390                 return PACKET_RCVD;
391         }
392         return PACKET_NEXT;
393
394 drop:
395         kfree_skb(skb);
396         return PACKET_RCVD;
397 }
398
399 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
400                      int hdr_len)
401 {
402         struct net *net = dev_net(skb->dev);
403         struct ip_tunnel_net *itn;
404         int res;
405
406         if (tpi->proto == htons(ETH_P_TEB))
407                 itn = net_generic(net, gre_tap_net_id);
408         else
409                 itn = net_generic(net, ipgre_net_id);
410
411         res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
412         if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
413                 /* ipgre tunnels in collect metadata mode should receive
414                  * also ETH_P_TEB traffic.
415                  */
416                 itn = net_generic(net, ipgre_net_id);
417                 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
418         }
419         return res;
420 }
421
422 static int gre_rcv(struct sk_buff *skb)
423 {
424         struct tnl_ptk_info tpi;
425         bool csum_err = false;
426         int hdr_len;
427
428 #ifdef CONFIG_NET_IPGRE_BROADCAST
429         if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
430                 /* Looped back packet, drop it! */
431                 if (rt_is_output_route(skb_rtable(skb)))
432                         goto drop;
433         }
434 #endif
435
436         hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
437         if (hdr_len < 0)
438                 goto drop;
439
440         if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
441                      tpi.proto == htons(ETH_P_ERSPAN2))) {
442                 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
443                         return 0;
444                 goto out;
445         }
446
447         if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
448                 return 0;
449
450 out:
451         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
452 drop:
453         kfree_skb(skb);
454         return 0;
455 }
456
457 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
458                        const struct iphdr *tnl_params,
459                        __be16 proto)
460 {
461         struct ip_tunnel *tunnel = netdev_priv(dev);
462         __be16 flags = tunnel->parms.o_flags;
463
464         /* Push GRE header. */
465         gre_build_header(skb, tunnel->tun_hlen,
466                          flags, proto, tunnel->parms.o_key,
467                          (flags & TUNNEL_SEQ) ? htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0);
468
469         ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
470 }
471
472 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
473 {
474         return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
475 }
476
477 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
478                         __be16 proto)
479 {
480         struct ip_tunnel *tunnel = netdev_priv(dev);
481         struct ip_tunnel_info *tun_info;
482         const struct ip_tunnel_key *key;
483         int tunnel_hlen;
484         __be16 flags;
485
486         tun_info = skb_tunnel_info(skb);
487         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
488                      ip_tunnel_info_af(tun_info) != AF_INET))
489                 goto err_free_skb;
490
491         key = &tun_info->key;
492         tunnel_hlen = gre_calc_hlen(key->tun_flags);
493
494         if (skb_cow_head(skb, dev->needed_headroom))
495                 goto err_free_skb;
496
497         /* Push Tunnel header. */
498         if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
499                 goto err_free_skb;
500
501         flags = tun_info->key.tun_flags &
502                 (TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ);
503         gre_build_header(skb, tunnel_hlen, flags, proto,
504                          tunnel_id_to_key32(tun_info->key.tun_id),
505                          (flags & TUNNEL_SEQ) ? htonl(atomic_fetch_inc(&tunnel->o_seqno)) : 0);
506
507         ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
508
509         return;
510
511 err_free_skb:
512         kfree_skb(skb);
513         DEV_STATS_INC(dev, tx_dropped);
514 }
515
516 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev)
517 {
518         struct ip_tunnel *tunnel = netdev_priv(dev);
519         struct ip_tunnel_info *tun_info;
520         const struct ip_tunnel_key *key;
521         struct erspan_metadata *md;
522         bool truncate = false;
523         __be16 proto;
524         int tunnel_hlen;
525         int version;
526         int nhoff;
527
528         tun_info = skb_tunnel_info(skb);
529         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
530                      ip_tunnel_info_af(tun_info) != AF_INET))
531                 goto err_free_skb;
532
533         key = &tun_info->key;
534         if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT))
535                 goto err_free_skb;
536         if (tun_info->options_len < sizeof(*md))
537                 goto err_free_skb;
538         md = ip_tunnel_info_opts(tun_info);
539
540         /* ERSPAN has fixed 8 byte GRE header */
541         version = md->version;
542         tunnel_hlen = 8 + erspan_hdr_len(version);
543
544         if (skb_cow_head(skb, dev->needed_headroom))
545                 goto err_free_skb;
546
547         if (gre_handle_offloads(skb, false))
548                 goto err_free_skb;
549
550         if (skb->len > dev->mtu + dev->hard_header_len) {
551                 if (pskb_trim(skb, dev->mtu + dev->hard_header_len))
552                         goto err_free_skb;
553                 truncate = true;
554         }
555
556         nhoff = skb_network_offset(skb);
557         if (skb->protocol == htons(ETH_P_IP) &&
558             (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
559                 truncate = true;
560
561         if (skb->protocol == htons(ETH_P_IPV6)) {
562                 int thoff;
563
564                 if (skb_transport_header_was_set(skb))
565                         thoff = skb_transport_offset(skb);
566                 else
567                         thoff = nhoff + sizeof(struct ipv6hdr);
568                 if (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff)
569                         truncate = true;
570         }
571
572         if (version == 1) {
573                 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
574                                     ntohl(md->u.index), truncate, true);
575                 proto = htons(ETH_P_ERSPAN);
576         } else if (version == 2) {
577                 erspan_build_header_v2(skb,
578                                        ntohl(tunnel_id_to_key32(key->tun_id)),
579                                        md->u.md2.dir,
580                                        get_hwid(&md->u.md2),
581                                        truncate, true);
582                 proto = htons(ETH_P_ERSPAN2);
583         } else {
584                 goto err_free_skb;
585         }
586
587         gre_build_header(skb, 8, TUNNEL_SEQ,
588                          proto, 0, htonl(atomic_fetch_inc(&tunnel->o_seqno)));
589
590         ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
591
592         return;
593
594 err_free_skb:
595         kfree_skb(skb);
596         DEV_STATS_INC(dev, tx_dropped);
597 }
598
599 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
600 {
601         struct ip_tunnel_info *info = skb_tunnel_info(skb);
602         const struct ip_tunnel_key *key;
603         struct rtable *rt;
604         struct flowi4 fl4;
605
606         if (ip_tunnel_info_af(info) != AF_INET)
607                 return -EINVAL;
608
609         key = &info->key;
610         ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src,
611                             tunnel_id_to_key32(key->tun_id),
612                             key->tos & ~INET_ECN_MASK, dev_net(dev), 0,
613                             skb->mark, skb_get_hash(skb), key->flow_flags);
614         rt = ip_route_output_key(dev_net(dev), &fl4);
615         if (IS_ERR(rt))
616                 return PTR_ERR(rt);
617
618         ip_rt_put(rt);
619         info->key.u.ipv4.src = fl4.saddr;
620         return 0;
621 }
622
623 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
624                               struct net_device *dev)
625 {
626         struct ip_tunnel *tunnel = netdev_priv(dev);
627         const struct iphdr *tnl_params;
628
629         if (!pskb_inet_may_pull(skb))
630                 goto free_skb;
631
632         if (tunnel->collect_md) {
633                 gre_fb_xmit(skb, dev, skb->protocol);
634                 return NETDEV_TX_OK;
635         }
636
637         if (dev->header_ops) {
638                 if (skb_cow_head(skb, 0))
639                         goto free_skb;
640
641                 tnl_params = (const struct iphdr *)skb->data;
642
643                 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
644                  * to gre header.
645                  */
646                 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
647                 skb_reset_mac_header(skb);
648
649                 if (skb->ip_summed == CHECKSUM_PARTIAL &&
650                     skb_checksum_start(skb) < skb->data)
651                         goto free_skb;
652         } else {
653                 if (skb_cow_head(skb, dev->needed_headroom))
654                         goto free_skb;
655
656                 tnl_params = &tunnel->parms.iph;
657         }
658
659         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
660                 goto free_skb;
661
662         __gre_xmit(skb, dev, tnl_params, skb->protocol);
663         return NETDEV_TX_OK;
664
665 free_skb:
666         kfree_skb(skb);
667         DEV_STATS_INC(dev, tx_dropped);
668         return NETDEV_TX_OK;
669 }
670
671 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
672                                struct net_device *dev)
673 {
674         struct ip_tunnel *tunnel = netdev_priv(dev);
675         bool truncate = false;
676         __be16 proto;
677
678         if (!pskb_inet_may_pull(skb))
679                 goto free_skb;
680
681         if (tunnel->collect_md) {
682                 erspan_fb_xmit(skb, dev);
683                 return NETDEV_TX_OK;
684         }
685
686         if (gre_handle_offloads(skb, false))
687                 goto free_skb;
688
689         if (skb_cow_head(skb, dev->needed_headroom))
690                 goto free_skb;
691
692         if (skb->len > dev->mtu + dev->hard_header_len) {
693                 if (pskb_trim(skb, dev->mtu + dev->hard_header_len))
694                         goto free_skb;
695                 truncate = true;
696         }
697
698         /* Push ERSPAN header */
699         if (tunnel->erspan_ver == 0) {
700                 proto = htons(ETH_P_ERSPAN);
701                 tunnel->parms.o_flags &= ~TUNNEL_SEQ;
702         } else if (tunnel->erspan_ver == 1) {
703                 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
704                                     tunnel->index,
705                                     truncate, true);
706                 proto = htons(ETH_P_ERSPAN);
707         } else if (tunnel->erspan_ver == 2) {
708                 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
709                                        tunnel->dir, tunnel->hwid,
710                                        truncate, true);
711                 proto = htons(ETH_P_ERSPAN2);
712         } else {
713                 goto free_skb;
714         }
715
716         tunnel->parms.o_flags &= ~TUNNEL_KEY;
717         __gre_xmit(skb, dev, &tunnel->parms.iph, proto);
718         return NETDEV_TX_OK;
719
720 free_skb:
721         kfree_skb(skb);
722         DEV_STATS_INC(dev, tx_dropped);
723         return NETDEV_TX_OK;
724 }
725
726 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
727                                 struct net_device *dev)
728 {
729         struct ip_tunnel *tunnel = netdev_priv(dev);
730
731         if (!pskb_inet_may_pull(skb))
732                 goto free_skb;
733
734         if (tunnel->collect_md) {
735                 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
736                 return NETDEV_TX_OK;
737         }
738
739         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
740                 goto free_skb;
741
742         if (skb_cow_head(skb, dev->needed_headroom))
743                 goto free_skb;
744
745         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
746         return NETDEV_TX_OK;
747
748 free_skb:
749         kfree_skb(skb);
750         DEV_STATS_INC(dev, tx_dropped);
751         return NETDEV_TX_OK;
752 }
753
754 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
755 {
756         struct ip_tunnel *tunnel = netdev_priv(dev);
757         __be16 flags;
758         int len;
759
760         len = tunnel->tun_hlen;
761         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
762         len = tunnel->tun_hlen - len;
763         tunnel->hlen = tunnel->hlen + len;
764
765         if (dev->header_ops)
766                 dev->hard_header_len += len;
767         else
768                 dev->needed_headroom += len;
769
770         if (set_mtu)
771                 dev->mtu = max_t(int, dev->mtu - len, 68);
772
773         flags = tunnel->parms.o_flags;
774
775         if (flags & TUNNEL_SEQ ||
776             (flags & TUNNEL_CSUM && tunnel->encap.type != TUNNEL_ENCAP_NONE)) {
777                 dev->features &= ~NETIF_F_GSO_SOFTWARE;
778                 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
779         } else {
780                 dev->features |= NETIF_F_GSO_SOFTWARE;
781                 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
782         }
783 }
784
785 static int ipgre_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p,
786                             int cmd)
787 {
788         int err;
789
790         if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
791                 if (p->iph.version != 4 || p->iph.protocol != IPPROTO_GRE ||
792                     p->iph.ihl != 5 || (p->iph.frag_off & htons(~IP_DF)) ||
793                     ((p->i_flags | p->o_flags) & (GRE_VERSION | GRE_ROUTING)))
794                         return -EINVAL;
795         }
796
797         p->i_flags = gre_flags_to_tnl_flags(p->i_flags);
798         p->o_flags = gre_flags_to_tnl_flags(p->o_flags);
799
800         err = ip_tunnel_ctl(dev, p, cmd);
801         if (err)
802                 return err;
803
804         if (cmd == SIOCCHGTUNNEL) {
805                 struct ip_tunnel *t = netdev_priv(dev);
806
807                 t->parms.i_flags = p->i_flags;
808                 t->parms.o_flags = p->o_flags;
809
810                 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
811                         ipgre_link_update(dev, true);
812         }
813
814         p->i_flags = gre_tnl_flags_to_gre_flags(p->i_flags);
815         p->o_flags = gre_tnl_flags_to_gre_flags(p->o_flags);
816         return 0;
817 }
818
819 /* Nice toy. Unfortunately, useless in real life :-)
820    It allows to construct virtual multiprotocol broadcast "LAN"
821    over the Internet, provided multicast routing is tuned.
822
823
824    I have no idea was this bicycle invented before me,
825    so that I had to set ARPHRD_IPGRE to a random value.
826    I have an impression, that Cisco could make something similar,
827    but this feature is apparently missing in IOS<=11.2(8).
828
829    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
830    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
831
832    ping -t 255 224.66.66.66
833
834    If nobody answers, mbone does not work.
835
836    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
837    ip addr add 10.66.66.<somewhat>/24 dev Universe
838    ifconfig Universe up
839    ifconfig Universe add fe80::<Your_real_addr>/10
840    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
841    ftp 10.66.66.66
842    ...
843    ftp fec0:6666:6666::193.233.7.65
844    ...
845  */
846 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
847                         unsigned short type,
848                         const void *daddr, const void *saddr, unsigned int len)
849 {
850         struct ip_tunnel *t = netdev_priv(dev);
851         struct iphdr *iph;
852         struct gre_base_hdr *greh;
853
854         iph = skb_push(skb, t->hlen + sizeof(*iph));
855         greh = (struct gre_base_hdr *)(iph+1);
856         greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
857         greh->protocol = htons(type);
858
859         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
860
861         /* Set the source hardware address. */
862         if (saddr)
863                 memcpy(&iph->saddr, saddr, 4);
864         if (daddr)
865                 memcpy(&iph->daddr, daddr, 4);
866         if (iph->daddr)
867                 return t->hlen + sizeof(*iph);
868
869         return -(t->hlen + sizeof(*iph));
870 }
871
872 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
873 {
874         const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
875         memcpy(haddr, &iph->saddr, 4);
876         return 4;
877 }
878
879 static const struct header_ops ipgre_header_ops = {
880         .create = ipgre_header,
881         .parse  = ipgre_header_parse,
882 };
883
884 #ifdef CONFIG_NET_IPGRE_BROADCAST
885 static int ipgre_open(struct net_device *dev)
886 {
887         struct ip_tunnel *t = netdev_priv(dev);
888
889         if (ipv4_is_multicast(t->parms.iph.daddr)) {
890                 struct flowi4 fl4;
891                 struct rtable *rt;
892
893                 rt = ip_route_output_gre(t->net, &fl4,
894                                          t->parms.iph.daddr,
895                                          t->parms.iph.saddr,
896                                          t->parms.o_key,
897                                          RT_TOS(t->parms.iph.tos),
898                                          t->parms.link);
899                 if (IS_ERR(rt))
900                         return -EADDRNOTAVAIL;
901                 dev = rt->dst.dev;
902                 ip_rt_put(rt);
903                 if (!__in_dev_get_rtnl(dev))
904                         return -EADDRNOTAVAIL;
905                 t->mlink = dev->ifindex;
906                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
907         }
908         return 0;
909 }
910
911 static int ipgre_close(struct net_device *dev)
912 {
913         struct ip_tunnel *t = netdev_priv(dev);
914
915         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
916                 struct in_device *in_dev;
917                 in_dev = inetdev_by_index(t->net, t->mlink);
918                 if (in_dev)
919                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
920         }
921         return 0;
922 }
923 #endif
924
925 static const struct net_device_ops ipgre_netdev_ops = {
926         .ndo_init               = ipgre_tunnel_init,
927         .ndo_uninit             = ip_tunnel_uninit,
928 #ifdef CONFIG_NET_IPGRE_BROADCAST
929         .ndo_open               = ipgre_open,
930         .ndo_stop               = ipgre_close,
931 #endif
932         .ndo_start_xmit         = ipgre_xmit,
933         .ndo_siocdevprivate     = ip_tunnel_siocdevprivate,
934         .ndo_change_mtu         = ip_tunnel_change_mtu,
935         .ndo_get_stats64        = dev_get_tstats64,
936         .ndo_get_iflink         = ip_tunnel_get_iflink,
937         .ndo_tunnel_ctl         = ipgre_tunnel_ctl,
938 };
939
940 #define GRE_FEATURES (NETIF_F_SG |              \
941                       NETIF_F_FRAGLIST |        \
942                       NETIF_F_HIGHDMA |         \
943                       NETIF_F_HW_CSUM)
944
945 static void ipgre_tunnel_setup(struct net_device *dev)
946 {
947         dev->netdev_ops         = &ipgre_netdev_ops;
948         dev->type               = ARPHRD_IPGRE;
949         ip_tunnel_setup(dev, ipgre_net_id);
950 }
951
952 static void __gre_tunnel_init(struct net_device *dev)
953 {
954         struct ip_tunnel *tunnel;
955         __be16 flags;
956
957         tunnel = netdev_priv(dev);
958         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
959         tunnel->parms.iph.protocol = IPPROTO_GRE;
960
961         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
962         dev->needed_headroom = tunnel->hlen + sizeof(tunnel->parms.iph);
963
964         dev->features           |= GRE_FEATURES | NETIF_F_LLTX;
965         dev->hw_features        |= GRE_FEATURES;
966
967         flags = tunnel->parms.o_flags;
968
969         /* TCP offload with GRE SEQ is not supported, nor can we support 2
970          * levels of outer headers requiring an update.
971          */
972         if (flags & TUNNEL_SEQ)
973                 return;
974         if (flags & TUNNEL_CSUM && tunnel->encap.type != TUNNEL_ENCAP_NONE)
975                 return;
976
977         dev->features |= NETIF_F_GSO_SOFTWARE;
978         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
979 }
980
981 static int ipgre_tunnel_init(struct net_device *dev)
982 {
983         struct ip_tunnel *tunnel = netdev_priv(dev);
984         struct iphdr *iph = &tunnel->parms.iph;
985
986         __gre_tunnel_init(dev);
987
988         __dev_addr_set(dev, &iph->saddr, 4);
989         memcpy(dev->broadcast, &iph->daddr, 4);
990
991         dev->flags              = IFF_NOARP;
992         netif_keep_dst(dev);
993         dev->addr_len           = 4;
994
995         if (iph->daddr && !tunnel->collect_md) {
996 #ifdef CONFIG_NET_IPGRE_BROADCAST
997                 if (ipv4_is_multicast(iph->daddr)) {
998                         if (!iph->saddr)
999                                 return -EINVAL;
1000                         dev->flags = IFF_BROADCAST;
1001                         dev->header_ops = &ipgre_header_ops;
1002                         dev->hard_header_len = tunnel->hlen + sizeof(*iph);
1003                         dev->needed_headroom = 0;
1004                 }
1005 #endif
1006         } else if (!tunnel->collect_md) {
1007                 dev->header_ops = &ipgre_header_ops;
1008                 dev->hard_header_len = tunnel->hlen + sizeof(*iph);
1009                 dev->needed_headroom = 0;
1010         }
1011
1012         return ip_tunnel_init(dev);
1013 }
1014
1015 static const struct gre_protocol ipgre_protocol = {
1016         .handler     = gre_rcv,
1017         .err_handler = gre_err,
1018 };
1019
1020 static int __net_init ipgre_init_net(struct net *net)
1021 {
1022         return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1023 }
1024
1025 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1026 {
1027         ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1028 }
1029
1030 static struct pernet_operations ipgre_net_ops = {
1031         .init = ipgre_init_net,
1032         .exit_batch = ipgre_exit_batch_net,
1033         .id   = &ipgre_net_id,
1034         .size = sizeof(struct ip_tunnel_net),
1035 };
1036
1037 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1038                                  struct netlink_ext_ack *extack)
1039 {
1040         __be16 flags;
1041
1042         if (!data)
1043                 return 0;
1044
1045         flags = 0;
1046         if (data[IFLA_GRE_IFLAGS])
1047                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1048         if (data[IFLA_GRE_OFLAGS])
1049                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1050         if (flags & (GRE_VERSION|GRE_ROUTING))
1051                 return -EINVAL;
1052
1053         if (data[IFLA_GRE_COLLECT_METADATA] &&
1054             data[IFLA_GRE_ENCAP_TYPE] &&
1055             nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1056                 return -EINVAL;
1057
1058         return 0;
1059 }
1060
1061 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1062                               struct netlink_ext_ack *extack)
1063 {
1064         __be32 daddr;
1065
1066         if (tb[IFLA_ADDRESS]) {
1067                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1068                         return -EINVAL;
1069                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1070                         return -EADDRNOTAVAIL;
1071         }
1072
1073         if (!data)
1074                 goto out;
1075
1076         if (data[IFLA_GRE_REMOTE]) {
1077                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1078                 if (!daddr)
1079                         return -EINVAL;
1080         }
1081
1082 out:
1083         return ipgre_tunnel_validate(tb, data, extack);
1084 }
1085
1086 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1087                            struct netlink_ext_ack *extack)
1088 {
1089         __be16 flags = 0;
1090         int ret;
1091
1092         if (!data)
1093                 return 0;
1094
1095         ret = ipgre_tap_validate(tb, data, extack);
1096         if (ret)
1097                 return ret;
1098
1099         if (data[IFLA_GRE_ERSPAN_VER] &&
1100             nla_get_u8(data[IFLA_GRE_ERSPAN_VER]) == 0)
1101                 return 0;
1102
1103         /* ERSPAN type II/III should only have GRE sequence and key flag */
1104         if (data[IFLA_GRE_OFLAGS])
1105                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1106         if (data[IFLA_GRE_IFLAGS])
1107                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1108         if (!data[IFLA_GRE_COLLECT_METADATA] &&
1109             flags != (GRE_SEQ | GRE_KEY))
1110                 return -EINVAL;
1111
1112         /* ERSPAN Session ID only has 10-bit. Since we reuse
1113          * 32-bit key field as ID, check it's range.
1114          */
1115         if (data[IFLA_GRE_IKEY] &&
1116             (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1117                 return -EINVAL;
1118
1119         if (data[IFLA_GRE_OKEY] &&
1120             (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1121                 return -EINVAL;
1122
1123         return 0;
1124 }
1125
1126 static int ipgre_netlink_parms(struct net_device *dev,
1127                                 struct nlattr *data[],
1128                                 struct nlattr *tb[],
1129                                 struct ip_tunnel_parm *parms,
1130                                 __u32 *fwmark)
1131 {
1132         struct ip_tunnel *t = netdev_priv(dev);
1133
1134         memset(parms, 0, sizeof(*parms));
1135
1136         parms->iph.protocol = IPPROTO_GRE;
1137
1138         if (!data)
1139                 return 0;
1140
1141         if (data[IFLA_GRE_LINK])
1142                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1143
1144         if (data[IFLA_GRE_IFLAGS])
1145                 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1146
1147         if (data[IFLA_GRE_OFLAGS])
1148                 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1149
1150         if (data[IFLA_GRE_IKEY])
1151                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1152
1153         if (data[IFLA_GRE_OKEY])
1154                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1155
1156         if (data[IFLA_GRE_LOCAL])
1157                 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1158
1159         if (data[IFLA_GRE_REMOTE])
1160                 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1161
1162         if (data[IFLA_GRE_TTL])
1163                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1164
1165         if (data[IFLA_GRE_TOS])
1166                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1167
1168         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1169                 if (t->ignore_df)
1170                         return -EINVAL;
1171                 parms->iph.frag_off = htons(IP_DF);
1172         }
1173
1174         if (data[IFLA_GRE_COLLECT_METADATA]) {
1175                 t->collect_md = true;
1176                 if (dev->type == ARPHRD_IPGRE)
1177                         dev->type = ARPHRD_NONE;
1178         }
1179
1180         if (data[IFLA_GRE_IGNORE_DF]) {
1181                 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1182                   && (parms->iph.frag_off & htons(IP_DF)))
1183                         return -EINVAL;
1184                 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1185         }
1186
1187         if (data[IFLA_GRE_FWMARK])
1188                 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1189
1190         return 0;
1191 }
1192
1193 static int erspan_netlink_parms(struct net_device *dev,
1194                                 struct nlattr *data[],
1195                                 struct nlattr *tb[],
1196                                 struct ip_tunnel_parm *parms,
1197                                 __u32 *fwmark)
1198 {
1199         struct ip_tunnel *t = netdev_priv(dev);
1200         int err;
1201
1202         err = ipgre_netlink_parms(dev, data, tb, parms, fwmark);
1203         if (err)
1204                 return err;
1205         if (!data)
1206                 return 0;
1207
1208         if (data[IFLA_GRE_ERSPAN_VER]) {
1209                 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1210
1211                 if (t->erspan_ver > 2)
1212                         return -EINVAL;
1213         }
1214
1215         if (t->erspan_ver == 1) {
1216                 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1217                         t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1218                         if (t->index & ~INDEX_MASK)
1219                                 return -EINVAL;
1220                 }
1221         } else if (t->erspan_ver == 2) {
1222                 if (data[IFLA_GRE_ERSPAN_DIR]) {
1223                         t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1224                         if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1225                                 return -EINVAL;
1226                 }
1227                 if (data[IFLA_GRE_ERSPAN_HWID]) {
1228                         t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1229                         if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1230                                 return -EINVAL;
1231                 }
1232         }
1233
1234         return 0;
1235 }
1236
1237 /* This function returns true when ENCAP attributes are present in the nl msg */
1238 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1239                                       struct ip_tunnel_encap *ipencap)
1240 {
1241         bool ret = false;
1242
1243         memset(ipencap, 0, sizeof(*ipencap));
1244
1245         if (!data)
1246                 return ret;
1247
1248         if (data[IFLA_GRE_ENCAP_TYPE]) {
1249                 ret = true;
1250                 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1251         }
1252
1253         if (data[IFLA_GRE_ENCAP_FLAGS]) {
1254                 ret = true;
1255                 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1256         }
1257
1258         if (data[IFLA_GRE_ENCAP_SPORT]) {
1259                 ret = true;
1260                 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1261         }
1262
1263         if (data[IFLA_GRE_ENCAP_DPORT]) {
1264                 ret = true;
1265                 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1266         }
1267
1268         return ret;
1269 }
1270
1271 static int gre_tap_init(struct net_device *dev)
1272 {
1273         __gre_tunnel_init(dev);
1274         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1275         netif_keep_dst(dev);
1276
1277         return ip_tunnel_init(dev);
1278 }
1279
1280 static const struct net_device_ops gre_tap_netdev_ops = {
1281         .ndo_init               = gre_tap_init,
1282         .ndo_uninit             = ip_tunnel_uninit,
1283         .ndo_start_xmit         = gre_tap_xmit,
1284         .ndo_set_mac_address    = eth_mac_addr,
1285         .ndo_validate_addr      = eth_validate_addr,
1286         .ndo_change_mtu         = ip_tunnel_change_mtu,
1287         .ndo_get_stats64        = dev_get_tstats64,
1288         .ndo_get_iflink         = ip_tunnel_get_iflink,
1289         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1290 };
1291
1292 static int erspan_tunnel_init(struct net_device *dev)
1293 {
1294         struct ip_tunnel *tunnel = netdev_priv(dev);
1295
1296         if (tunnel->erspan_ver == 0)
1297                 tunnel->tun_hlen = 4; /* 4-byte GRE hdr. */
1298         else
1299                 tunnel->tun_hlen = 8; /* 8-byte GRE hdr. */
1300
1301         tunnel->parms.iph.protocol = IPPROTO_GRE;
1302         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1303                        erspan_hdr_len(tunnel->erspan_ver);
1304
1305         dev->features           |= GRE_FEATURES;
1306         dev->hw_features        |= GRE_FEATURES;
1307         dev->priv_flags         |= IFF_LIVE_ADDR_CHANGE;
1308         netif_keep_dst(dev);
1309
1310         return ip_tunnel_init(dev);
1311 }
1312
1313 static const struct net_device_ops erspan_netdev_ops = {
1314         .ndo_init               = erspan_tunnel_init,
1315         .ndo_uninit             = ip_tunnel_uninit,
1316         .ndo_start_xmit         = erspan_xmit,
1317         .ndo_set_mac_address    = eth_mac_addr,
1318         .ndo_validate_addr      = eth_validate_addr,
1319         .ndo_change_mtu         = ip_tunnel_change_mtu,
1320         .ndo_get_stats64        = dev_get_tstats64,
1321         .ndo_get_iflink         = ip_tunnel_get_iflink,
1322         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1323 };
1324
1325 static void ipgre_tap_setup(struct net_device *dev)
1326 {
1327         ether_setup(dev);
1328         dev->max_mtu = 0;
1329         dev->netdev_ops = &gre_tap_netdev_ops;
1330         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1331         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1332         ip_tunnel_setup(dev, gre_tap_net_id);
1333 }
1334
1335 static int
1336 ipgre_newlink_encap_setup(struct net_device *dev, struct nlattr *data[])
1337 {
1338         struct ip_tunnel_encap ipencap;
1339
1340         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1341                 struct ip_tunnel *t = netdev_priv(dev);
1342                 int err = ip_tunnel_encap_setup(t, &ipencap);
1343
1344                 if (err < 0)
1345                         return err;
1346         }
1347
1348         return 0;
1349 }
1350
1351 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1352                          struct nlattr *tb[], struct nlattr *data[],
1353                          struct netlink_ext_ack *extack)
1354 {
1355         struct ip_tunnel_parm p;
1356         __u32 fwmark = 0;
1357         int err;
1358
1359         err = ipgre_newlink_encap_setup(dev, data);
1360         if (err)
1361                 return err;
1362
1363         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1364         if (err < 0)
1365                 return err;
1366         return ip_tunnel_newlink(dev, tb, &p, fwmark);
1367 }
1368
1369 static int erspan_newlink(struct net *src_net, struct net_device *dev,
1370                           struct nlattr *tb[], struct nlattr *data[],
1371                           struct netlink_ext_ack *extack)
1372 {
1373         struct ip_tunnel_parm p;
1374         __u32 fwmark = 0;
1375         int err;
1376
1377         err = ipgre_newlink_encap_setup(dev, data);
1378         if (err)
1379                 return err;
1380
1381         err = erspan_netlink_parms(dev, data, tb, &p, &fwmark);
1382         if (err)
1383                 return err;
1384         return ip_tunnel_newlink(dev, tb, &p, fwmark);
1385 }
1386
1387 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1388                             struct nlattr *data[],
1389                             struct netlink_ext_ack *extack)
1390 {
1391         struct ip_tunnel *t = netdev_priv(dev);
1392         __u32 fwmark = t->fwmark;
1393         struct ip_tunnel_parm p;
1394         int err;
1395
1396         err = ipgre_newlink_encap_setup(dev, data);
1397         if (err)
1398                 return err;
1399
1400         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1401         if (err < 0)
1402                 return err;
1403
1404         err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1405         if (err < 0)
1406                 return err;
1407
1408         t->parms.i_flags = p.i_flags;
1409         t->parms.o_flags = p.o_flags;
1410
1411         ipgre_link_update(dev, !tb[IFLA_MTU]);
1412
1413         return 0;
1414 }
1415
1416 static int erspan_changelink(struct net_device *dev, struct nlattr *tb[],
1417                              struct nlattr *data[],
1418                              struct netlink_ext_ack *extack)
1419 {
1420         struct ip_tunnel *t = netdev_priv(dev);
1421         __u32 fwmark = t->fwmark;
1422         struct ip_tunnel_parm p;
1423         int err;
1424
1425         err = ipgre_newlink_encap_setup(dev, data);
1426         if (err)
1427                 return err;
1428
1429         err = erspan_netlink_parms(dev, data, tb, &p, &fwmark);
1430         if (err < 0)
1431                 return err;
1432
1433         err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1434         if (err < 0)
1435                 return err;
1436
1437         t->parms.i_flags = p.i_flags;
1438         t->parms.o_flags = p.o_flags;
1439
1440         return 0;
1441 }
1442
1443 static size_t ipgre_get_size(const struct net_device *dev)
1444 {
1445         return
1446                 /* IFLA_GRE_LINK */
1447                 nla_total_size(4) +
1448                 /* IFLA_GRE_IFLAGS */
1449                 nla_total_size(2) +
1450                 /* IFLA_GRE_OFLAGS */
1451                 nla_total_size(2) +
1452                 /* IFLA_GRE_IKEY */
1453                 nla_total_size(4) +
1454                 /* IFLA_GRE_OKEY */
1455                 nla_total_size(4) +
1456                 /* IFLA_GRE_LOCAL */
1457                 nla_total_size(4) +
1458                 /* IFLA_GRE_REMOTE */
1459                 nla_total_size(4) +
1460                 /* IFLA_GRE_TTL */
1461                 nla_total_size(1) +
1462                 /* IFLA_GRE_TOS */
1463                 nla_total_size(1) +
1464                 /* IFLA_GRE_PMTUDISC */
1465                 nla_total_size(1) +
1466                 /* IFLA_GRE_ENCAP_TYPE */
1467                 nla_total_size(2) +
1468                 /* IFLA_GRE_ENCAP_FLAGS */
1469                 nla_total_size(2) +
1470                 /* IFLA_GRE_ENCAP_SPORT */
1471                 nla_total_size(2) +
1472                 /* IFLA_GRE_ENCAP_DPORT */
1473                 nla_total_size(2) +
1474                 /* IFLA_GRE_COLLECT_METADATA */
1475                 nla_total_size(0) +
1476                 /* IFLA_GRE_IGNORE_DF */
1477                 nla_total_size(1) +
1478                 /* IFLA_GRE_FWMARK */
1479                 nla_total_size(4) +
1480                 /* IFLA_GRE_ERSPAN_INDEX */
1481                 nla_total_size(4) +
1482                 /* IFLA_GRE_ERSPAN_VER */
1483                 nla_total_size(1) +
1484                 /* IFLA_GRE_ERSPAN_DIR */
1485                 nla_total_size(1) +
1486                 /* IFLA_GRE_ERSPAN_HWID */
1487                 nla_total_size(2) +
1488                 0;
1489 }
1490
1491 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1492 {
1493         struct ip_tunnel *t = netdev_priv(dev);
1494         struct ip_tunnel_parm *p = &t->parms;
1495         __be16 o_flags = p->o_flags;
1496
1497         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1498             nla_put_be16(skb, IFLA_GRE_IFLAGS,
1499                          gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1500             nla_put_be16(skb, IFLA_GRE_OFLAGS,
1501                          gre_tnl_flags_to_gre_flags(o_flags)) ||
1502             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1503             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1504             nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1505             nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1506             nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1507             nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1508             nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1509                        !!(p->iph.frag_off & htons(IP_DF))) ||
1510             nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1511                 goto nla_put_failure;
1512
1513         if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1514                         t->encap.type) ||
1515             nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1516                          t->encap.sport) ||
1517             nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1518                          t->encap.dport) ||
1519             nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1520                         t->encap.flags))
1521                 goto nla_put_failure;
1522
1523         if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1524                 goto nla_put_failure;
1525
1526         if (t->collect_md) {
1527                 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1528                         goto nla_put_failure;
1529         }
1530
1531         return 0;
1532
1533 nla_put_failure:
1534         return -EMSGSIZE;
1535 }
1536
1537 static int erspan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1538 {
1539         struct ip_tunnel *t = netdev_priv(dev);
1540
1541         if (t->erspan_ver <= 2) {
1542                 if (t->erspan_ver != 0 && !t->collect_md)
1543                         t->parms.o_flags |= TUNNEL_KEY;
1544
1545                 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1546                         goto nla_put_failure;
1547
1548                 if (t->erspan_ver == 1) {
1549                         if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1550                                 goto nla_put_failure;
1551                 } else if (t->erspan_ver == 2) {
1552                         if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1553                                 goto nla_put_failure;
1554                         if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1555                                 goto nla_put_failure;
1556                 }
1557         }
1558
1559         return ipgre_fill_info(skb, dev);
1560
1561 nla_put_failure:
1562         return -EMSGSIZE;
1563 }
1564
1565 static void erspan_setup(struct net_device *dev)
1566 {
1567         struct ip_tunnel *t = netdev_priv(dev);
1568
1569         ether_setup(dev);
1570         dev->max_mtu = 0;
1571         dev->netdev_ops = &erspan_netdev_ops;
1572         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1573         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1574         ip_tunnel_setup(dev, erspan_net_id);
1575         t->erspan_ver = 1;
1576 }
1577
1578 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1579         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1580         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1581         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1582         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1583         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1584         [IFLA_GRE_LOCAL]        = { .len = sizeof_field(struct iphdr, saddr) },
1585         [IFLA_GRE_REMOTE]       = { .len = sizeof_field(struct iphdr, daddr) },
1586         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1587         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1588         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1589         [IFLA_GRE_ENCAP_TYPE]   = { .type = NLA_U16 },
1590         [IFLA_GRE_ENCAP_FLAGS]  = { .type = NLA_U16 },
1591         [IFLA_GRE_ENCAP_SPORT]  = { .type = NLA_U16 },
1592         [IFLA_GRE_ENCAP_DPORT]  = { .type = NLA_U16 },
1593         [IFLA_GRE_COLLECT_METADATA]     = { .type = NLA_FLAG },
1594         [IFLA_GRE_IGNORE_DF]    = { .type = NLA_U8 },
1595         [IFLA_GRE_FWMARK]       = { .type = NLA_U32 },
1596         [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1597         [IFLA_GRE_ERSPAN_VER]   = { .type = NLA_U8 },
1598         [IFLA_GRE_ERSPAN_DIR]   = { .type = NLA_U8 },
1599         [IFLA_GRE_ERSPAN_HWID]  = { .type = NLA_U16 },
1600 };
1601
1602 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1603         .kind           = "gre",
1604         .maxtype        = IFLA_GRE_MAX,
1605         .policy         = ipgre_policy,
1606         .priv_size      = sizeof(struct ip_tunnel),
1607         .setup          = ipgre_tunnel_setup,
1608         .validate       = ipgre_tunnel_validate,
1609         .newlink        = ipgre_newlink,
1610         .changelink     = ipgre_changelink,
1611         .dellink        = ip_tunnel_dellink,
1612         .get_size       = ipgre_get_size,
1613         .fill_info      = ipgre_fill_info,
1614         .get_link_net   = ip_tunnel_get_link_net,
1615 };
1616
1617 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1618         .kind           = "gretap",
1619         .maxtype        = IFLA_GRE_MAX,
1620         .policy         = ipgre_policy,
1621         .priv_size      = sizeof(struct ip_tunnel),
1622         .setup          = ipgre_tap_setup,
1623         .validate       = ipgre_tap_validate,
1624         .newlink        = ipgre_newlink,
1625         .changelink     = ipgre_changelink,
1626         .dellink        = ip_tunnel_dellink,
1627         .get_size       = ipgre_get_size,
1628         .fill_info      = ipgre_fill_info,
1629         .get_link_net   = ip_tunnel_get_link_net,
1630 };
1631
1632 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1633         .kind           = "erspan",
1634         .maxtype        = IFLA_GRE_MAX,
1635         .policy         = ipgre_policy,
1636         .priv_size      = sizeof(struct ip_tunnel),
1637         .setup          = erspan_setup,
1638         .validate       = erspan_validate,
1639         .newlink        = erspan_newlink,
1640         .changelink     = erspan_changelink,
1641         .dellink        = ip_tunnel_dellink,
1642         .get_size       = ipgre_get_size,
1643         .fill_info      = erspan_fill_info,
1644         .get_link_net   = ip_tunnel_get_link_net,
1645 };
1646
1647 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1648                                         u8 name_assign_type)
1649 {
1650         struct nlattr *tb[IFLA_MAX + 1];
1651         struct net_device *dev;
1652         LIST_HEAD(list_kill);
1653         struct ip_tunnel *t;
1654         int err;
1655
1656         memset(&tb, 0, sizeof(tb));
1657
1658         dev = rtnl_create_link(net, name, name_assign_type,
1659                                &ipgre_tap_ops, tb, NULL);
1660         if (IS_ERR(dev))
1661                 return dev;
1662
1663         /* Configure flow based GRE device. */
1664         t = netdev_priv(dev);
1665         t->collect_md = true;
1666
1667         err = ipgre_newlink(net, dev, tb, NULL, NULL);
1668         if (err < 0) {
1669                 free_netdev(dev);
1670                 return ERR_PTR(err);
1671         }
1672
1673         /* openvswitch users expect packet sizes to be unrestricted,
1674          * so set the largest MTU we can.
1675          */
1676         err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1677         if (err)
1678                 goto out;
1679
1680         err = rtnl_configure_link(dev, NULL, 0, NULL);
1681         if (err < 0)
1682                 goto out;
1683
1684         return dev;
1685 out:
1686         ip_tunnel_dellink(dev, &list_kill);
1687         unregister_netdevice_many(&list_kill);
1688         return ERR_PTR(err);
1689 }
1690 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1691
1692 static int __net_init ipgre_tap_init_net(struct net *net)
1693 {
1694         return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1695 }
1696
1697 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1698 {
1699         ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1700 }
1701
1702 static struct pernet_operations ipgre_tap_net_ops = {
1703         .init = ipgre_tap_init_net,
1704         .exit_batch = ipgre_tap_exit_batch_net,
1705         .id   = &gre_tap_net_id,
1706         .size = sizeof(struct ip_tunnel_net),
1707 };
1708
1709 static int __net_init erspan_init_net(struct net *net)
1710 {
1711         return ip_tunnel_init_net(net, erspan_net_id,
1712                                   &erspan_link_ops, "erspan0");
1713 }
1714
1715 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1716 {
1717         ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1718 }
1719
1720 static struct pernet_operations erspan_net_ops = {
1721         .init = erspan_init_net,
1722         .exit_batch = erspan_exit_batch_net,
1723         .id   = &erspan_net_id,
1724         .size = sizeof(struct ip_tunnel_net),
1725 };
1726
1727 static int __init ipgre_init(void)
1728 {
1729         int err;
1730
1731         pr_info("GRE over IPv4 tunneling driver\n");
1732
1733         err = register_pernet_device(&ipgre_net_ops);
1734         if (err < 0)
1735                 return err;
1736
1737         err = register_pernet_device(&ipgre_tap_net_ops);
1738         if (err < 0)
1739                 goto pnet_tap_failed;
1740
1741         err = register_pernet_device(&erspan_net_ops);
1742         if (err < 0)
1743                 goto pnet_erspan_failed;
1744
1745         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1746         if (err < 0) {
1747                 pr_info("%s: can't add protocol\n", __func__);
1748                 goto add_proto_failed;
1749         }
1750
1751         err = rtnl_link_register(&ipgre_link_ops);
1752         if (err < 0)
1753                 goto rtnl_link_failed;
1754
1755         err = rtnl_link_register(&ipgre_tap_ops);
1756         if (err < 0)
1757                 goto tap_ops_failed;
1758
1759         err = rtnl_link_register(&erspan_link_ops);
1760         if (err < 0)
1761                 goto erspan_link_failed;
1762
1763         return 0;
1764
1765 erspan_link_failed:
1766         rtnl_link_unregister(&ipgre_tap_ops);
1767 tap_ops_failed:
1768         rtnl_link_unregister(&ipgre_link_ops);
1769 rtnl_link_failed:
1770         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1771 add_proto_failed:
1772         unregister_pernet_device(&erspan_net_ops);
1773 pnet_erspan_failed:
1774         unregister_pernet_device(&ipgre_tap_net_ops);
1775 pnet_tap_failed:
1776         unregister_pernet_device(&ipgre_net_ops);
1777         return err;
1778 }
1779
1780 static void __exit ipgre_fini(void)
1781 {
1782         rtnl_link_unregister(&ipgre_tap_ops);
1783         rtnl_link_unregister(&ipgre_link_ops);
1784         rtnl_link_unregister(&erspan_link_ops);
1785         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1786         unregister_pernet_device(&ipgre_tap_net_ops);
1787         unregister_pernet_device(&ipgre_net_ops);
1788         unregister_pernet_device(&erspan_net_ops);
1789 }
1790
1791 module_init(ipgre_init);
1792 module_exit(ipgre_fini);
1793 MODULE_LICENSE("GPL");
1794 MODULE_ALIAS_RTNL_LINK("gre");
1795 MODULE_ALIAS_RTNL_LINK("gretap");
1796 MODULE_ALIAS_RTNL_LINK("erspan");
1797 MODULE_ALIAS_NETDEV("gre0");
1798 MODULE_ALIAS_NETDEV("gretap0");
1799 MODULE_ALIAS_NETDEV("erspan0");