Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 2 | * Linux NET3: GRE over IP protocol decoder. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
| 4 | * Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru) |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | */ |
| 12 | |
Randy Dunlap | 4fc268d | 2006-01-11 12:17:47 -0800 | [diff] [blame] | 13 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/module.h> |
| 15 | #include <linux/types.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | #include <linux/kernel.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 17 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <asm/uaccess.h> |
| 19 | #include <linux/skbuff.h> |
| 20 | #include <linux/netdevice.h> |
| 21 | #include <linux/in.h> |
| 22 | #include <linux/tcp.h> |
| 23 | #include <linux/udp.h> |
| 24 | #include <linux/if_arp.h> |
| 25 | #include <linux/mroute.h> |
| 26 | #include <linux/init.h> |
| 27 | #include <linux/in6.h> |
| 28 | #include <linux/inetdevice.h> |
| 29 | #include <linux/igmp.h> |
| 30 | #include <linux/netfilter_ipv4.h> |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 31 | #include <linux/etherdevice.h> |
Kris Katterjohn | 46f25df | 2006-01-05 16:35:42 -0800 | [diff] [blame] | 32 | #include <linux/if_ether.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | |
| 34 | #include <net/sock.h> |
| 35 | #include <net/ip.h> |
| 36 | #include <net/icmp.h> |
| 37 | #include <net/protocol.h> |
| 38 | #include <net/ipip.h> |
| 39 | #include <net/arp.h> |
| 40 | #include <net/checksum.h> |
| 41 | #include <net/dsfield.h> |
| 42 | #include <net/inet_ecn.h> |
| 43 | #include <net/xfrm.h> |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 44 | #include <net/net_namespace.h> |
| 45 | #include <net/netns/generic.h> |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 46 | #include <net/rtnetlink.h> |
Dmitry Kozlov | 00959ad | 2010-08-21 23:05:39 -0700 | [diff] [blame] | 47 | #include <net/gre.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | |
Eric Dumazet | 842c74b | 2010-09-20 10:06:12 -0700 | [diff] [blame] | 49 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | #include <net/ipv6.h> |
| 51 | #include <net/ip6_fib.h> |
| 52 | #include <net/ip6_route.h> |
| 53 | #endif |
| 54 | |
| 55 | /* |
| 56 | Problems & solutions |
| 57 | -------------------- |
| 58 | |
| 59 | 1. The most important issue is detecting local dead loops. |
| 60 | They would cause complete host lockup in transmit, which |
| 61 | would be "resolved" by stack overflow or, if queueing is enabled, |
| 62 | with infinite looping in net_bh. |
| 63 | |
| 64 | We cannot track such dead loops during route installation, |
| 65 | it is infeasible task. The most general solutions would be |
| 66 | to keep skb->encapsulation counter (sort of local ttl), |
Eric Dumazet | 6d0722a | 2010-09-29 23:35:10 -0700 | [diff] [blame] | 67 | and silently drop packet when it expires. It is a good |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | solution, but it supposes maintaing new variable in ALL |
| 69 | skb, even if no tunneling is used. |
| 70 | |
Eric Dumazet | 6d0722a | 2010-09-29 23:35:10 -0700 | [diff] [blame] | 71 | Current solution: xmit_recursion breaks dead loops. This is a percpu |
| 72 | counter, since when we enter the first ndo_xmit(), cpu migration is |
| 73 | forbidden. We force an exit if this counter reaches RECURSION_LIMIT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | |
| 75 | 2. Networking dead loops would not kill routers, but would really |
| 76 | kill network. IP hop limit plays role of "t->recursion" in this case, |
| 77 | if we copy it from packet being encapsulated to upper header. |
| 78 | It is very good solution, but it introduces two problems: |
| 79 | |
| 80 | - Routing protocols, using packets with ttl=1 (OSPF, RIP2), |
| 81 | do not work over tunnels. |
| 82 | - traceroute does not work. I planned to relay ICMP from tunnel, |
| 83 | so that this problem would be solved and traceroute output |
| 84 | would even more informative. This idea appeared to be wrong: |
| 85 | only Linux complies to rfc1812 now (yes, guys, Linux is the only |
| 86 | true router now :-)), all routers (at least, in neighbourhood of mine) |
| 87 | return only 8 bytes of payload. It is the end. |
| 88 | |
| 89 | Hence, if we want that OSPF worked or traceroute said something reasonable, |
| 90 | we should search for another solution. |
| 91 | |
| 92 | One of them is to parse packet trying to detect inner encapsulation |
| 93 | made by our node. It is difficult or even impossible, especially, |
| 94 | taking into account fragmentation. TO be short, tt is not solution at all. |
| 95 | |
| 96 | Current solution: The solution was UNEXPECTEDLY SIMPLE. |
| 97 | We force DF flag on tunnels with preconfigured hop limit, |
| 98 | that is ALL. :-) Well, it does not remove the problem completely, |
| 99 | but exponential growth of network traffic is changed to linear |
| 100 | (branches, that exceed pmtu are pruned) and tunnel mtu |
| 101 | fastly degrades to value <68, where looping stops. |
| 102 | Yes, it is not good if there exists a router in the loop, |
| 103 | which does not force DF, even when encapsulating packets have DF set. |
| 104 | But it is not our problem! Nobody could accuse us, we made |
| 105 | all that we could make. Even if it is your gated who injected |
| 106 | fatal route to network, even if it were you who configured |
| 107 | fatal static route: you are innocent. :-) |
| 108 | |
| 109 | |
| 110 | |
| 111 | 3. Really, ipv4/ipip.c, ipv4/ip_gre.c and ipv6/sit.c contain |
| 112 | practically identical code. It would be good to glue them |
| 113 | together, but it is not very evident, how to make them modular. |
| 114 | sit is integral part of IPv6, ipip and gre are naturally modular. |
| 115 | We could extract common parts (hash table, ioctl etc) |
| 116 | to a separate module (ip_tunnel.c). |
| 117 | |
| 118 | Alexey Kuznetsov. |
| 119 | */ |
| 120 | |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 121 | static struct rtnl_link_ops ipgre_link_ops __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | static int ipgre_tunnel_init(struct net_device *dev); |
| 123 | static void ipgre_tunnel_setup(struct net_device *dev); |
Herbert Xu | 42aa916 | 2008-10-09 11:59:32 -0700 | [diff] [blame] | 124 | static int ipgre_tunnel_bind_dev(struct net_device *dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | |
| 126 | /* Fallback tunnel: no source, no destination, no key, no options */ |
| 127 | |
Pavel Emelyanov | eb8ce74 | 2008-04-16 01:10:26 -0700 | [diff] [blame] | 128 | #define HASH_SIZE 16 |
| 129 | |
Eric Dumazet | f99189b | 2009-11-17 10:42:49 +0000 | [diff] [blame] | 130 | static int ipgre_net_id __read_mostly; |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 131 | struct ipgre_net { |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 132 | struct ip_tunnel __rcu *tunnels[4][HASH_SIZE]; |
Pavel Emelyanov | eb8ce74 | 2008-04-16 01:10:26 -0700 | [diff] [blame] | 133 | |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 134 | struct net_device *fb_tunnel_dev; |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 135 | }; |
| 136 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | /* Tunnel hash table */ |
| 138 | |
| 139 | /* |
| 140 | 4 hash tables: |
| 141 | |
| 142 | 3: (remote,local) |
| 143 | 2: (remote,*) |
| 144 | 1: (*,local) |
| 145 | 0: (*,*) |
| 146 | |
| 147 | We require exact key match i.e. if a key is present in packet |
| 148 | it will match only tunnel with the same key; if it is not present, |
| 149 | it will match only keyless tunnel. |
| 150 | |
| 151 | All keysless packets, if not matched configured keyless tunnels |
| 152 | will match fallback tunnel. |
| 153 | */ |
| 154 | |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 155 | #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | |
Pavel Emelyanov | eb8ce74 | 2008-04-16 01:10:26 -0700 | [diff] [blame] | 157 | #define tunnels_r_l tunnels[3] |
| 158 | #define tunnels_r tunnels[2] |
| 159 | #define tunnels_l tunnels[1] |
| 160 | #define tunnels_wc tunnels[0] |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 161 | /* |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 162 | * Locking : hash tables are protected by RCU and RTNL |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 163 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 165 | #define for_each_ip_tunnel_rcu(start) \ |
| 166 | for (t = rcu_dereference(start); t; t = rcu_dereference(t->next)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 168 | /* often modified stats are per cpu, other are shared (netdev->stats) */ |
| 169 | struct pcpu_tstats { |
| 170 | unsigned long rx_packets; |
| 171 | unsigned long rx_bytes; |
| 172 | unsigned long tx_packets; |
| 173 | unsigned long tx_bytes; |
| 174 | }; |
| 175 | |
| 176 | static struct net_device_stats *ipgre_get_stats(struct net_device *dev) |
| 177 | { |
| 178 | struct pcpu_tstats sum = { 0 }; |
| 179 | int i; |
| 180 | |
| 181 | for_each_possible_cpu(i) { |
| 182 | const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i); |
| 183 | |
| 184 | sum.rx_packets += tstats->rx_packets; |
| 185 | sum.rx_bytes += tstats->rx_bytes; |
| 186 | sum.tx_packets += tstats->tx_packets; |
| 187 | sum.tx_bytes += tstats->tx_bytes; |
| 188 | } |
| 189 | dev->stats.rx_packets = sum.rx_packets; |
| 190 | dev->stats.rx_bytes = sum.rx_bytes; |
| 191 | dev->stats.tx_packets = sum.tx_packets; |
| 192 | dev->stats.tx_bytes = sum.tx_bytes; |
| 193 | return &dev->stats; |
| 194 | } |
| 195 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | /* Given src, dst and key, find appropriate for input tunnel. */ |
| 197 | |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 198 | static struct ip_tunnel * ipgre_tunnel_lookup(struct net_device *dev, |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 199 | __be32 remote, __be32 local, |
| 200 | __be32 key, __be16 gre_proto) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | { |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 202 | struct net *net = dev_net(dev); |
| 203 | int link = dev->ifindex; |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 204 | unsigned int h0 = HASH(remote); |
| 205 | unsigned int h1 = HASH(key); |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 206 | struct ip_tunnel *t, *cand = NULL; |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 207 | struct ipgre_net *ign = net_generic(net, ipgre_net_id); |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 208 | int dev_type = (gre_proto == htons(ETH_P_TEB)) ? |
| 209 | ARPHRD_ETHER : ARPHRD_IPGRE; |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 210 | int score, cand_score = 4; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 212 | for_each_ip_tunnel_rcu(ign->tunnels_r_l[h0 ^ h1]) { |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 213 | if (local != t->parms.iph.saddr || |
| 214 | remote != t->parms.iph.daddr || |
| 215 | key != t->parms.i_key || |
| 216 | !(t->dev->flags & IFF_UP)) |
| 217 | continue; |
| 218 | |
| 219 | if (t->dev->type != ARPHRD_IPGRE && |
| 220 | t->dev->type != dev_type) |
| 221 | continue; |
| 222 | |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 223 | score = 0; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 224 | if (t->parms.link != link) |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 225 | score |= 1; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 226 | if (t->dev->type != dev_type) |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 227 | score |= 2; |
| 228 | if (score == 0) |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 229 | return t; |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 230 | |
| 231 | if (score < cand_score) { |
| 232 | cand = t; |
| 233 | cand_score = score; |
| 234 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | } |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 236 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 237 | for_each_ip_tunnel_rcu(ign->tunnels_r[h0 ^ h1]) { |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 238 | if (remote != t->parms.iph.daddr || |
| 239 | key != t->parms.i_key || |
| 240 | !(t->dev->flags & IFF_UP)) |
| 241 | continue; |
| 242 | |
| 243 | if (t->dev->type != ARPHRD_IPGRE && |
| 244 | t->dev->type != dev_type) |
| 245 | continue; |
| 246 | |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 247 | score = 0; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 248 | if (t->parms.link != link) |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 249 | score |= 1; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 250 | if (t->dev->type != dev_type) |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 251 | score |= 2; |
| 252 | if (score == 0) |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 253 | return t; |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 254 | |
| 255 | if (score < cand_score) { |
| 256 | cand = t; |
| 257 | cand_score = score; |
| 258 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 259 | } |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 260 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 261 | for_each_ip_tunnel_rcu(ign->tunnels_l[h1]) { |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 262 | if ((local != t->parms.iph.saddr && |
| 263 | (local != t->parms.iph.daddr || |
| 264 | !ipv4_is_multicast(local))) || |
| 265 | key != t->parms.i_key || |
| 266 | !(t->dev->flags & IFF_UP)) |
| 267 | continue; |
| 268 | |
| 269 | if (t->dev->type != ARPHRD_IPGRE && |
| 270 | t->dev->type != dev_type) |
| 271 | continue; |
| 272 | |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 273 | score = 0; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 274 | if (t->parms.link != link) |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 275 | score |= 1; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 276 | if (t->dev->type != dev_type) |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 277 | score |= 2; |
| 278 | if (score == 0) |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 279 | return t; |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 280 | |
| 281 | if (score < cand_score) { |
| 282 | cand = t; |
| 283 | cand_score = score; |
| 284 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | } |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 286 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 287 | for_each_ip_tunnel_rcu(ign->tunnels_wc[h1]) { |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 288 | if (t->parms.i_key != key || |
| 289 | !(t->dev->flags & IFF_UP)) |
| 290 | continue; |
| 291 | |
| 292 | if (t->dev->type != ARPHRD_IPGRE && |
| 293 | t->dev->type != dev_type) |
| 294 | continue; |
| 295 | |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 296 | score = 0; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 297 | if (t->parms.link != link) |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 298 | score |= 1; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 299 | if (t->dev->type != dev_type) |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 300 | score |= 2; |
| 301 | if (score == 0) |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 302 | return t; |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 303 | |
| 304 | if (score < cand_score) { |
| 305 | cand = t; |
| 306 | cand_score = score; |
| 307 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | } |
| 309 | |
Timo Teras | afcf124 | 2009-01-26 20:56:10 -0800 | [diff] [blame] | 310 | if (cand != NULL) |
| 311 | return cand; |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 312 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 313 | dev = ign->fb_tunnel_dev; |
| 314 | if (dev->flags & IFF_UP) |
| 315 | return netdev_priv(dev); |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 316 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | return NULL; |
| 318 | } |
| 319 | |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 320 | static struct ip_tunnel __rcu **__ipgre_bucket(struct ipgre_net *ign, |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 321 | struct ip_tunnel_parm *parms) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | { |
YOSHIFUJI Hideaki | 5056a1e | 2007-04-24 20:44:48 +0900 | [diff] [blame] | 323 | __be32 remote = parms->iph.daddr; |
| 324 | __be32 local = parms->iph.saddr; |
| 325 | __be32 key = parms->i_key; |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 326 | unsigned int h = HASH(key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 327 | int prio = 0; |
| 328 | |
| 329 | if (local) |
| 330 | prio |= 1; |
Joe Perches | f97c1e0 | 2007-12-16 13:45:43 -0800 | [diff] [blame] | 331 | if (remote && !ipv4_is_multicast(remote)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | prio |= 2; |
| 333 | h ^= HASH(remote); |
| 334 | } |
| 335 | |
Pavel Emelyanov | eb8ce74 | 2008-04-16 01:10:26 -0700 | [diff] [blame] | 336 | return &ign->tunnels[prio][h]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 337 | } |
| 338 | |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 339 | static inline struct ip_tunnel __rcu **ipgre_bucket(struct ipgre_net *ign, |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 340 | struct ip_tunnel *t) |
YOSHIFUJI Hideaki | 5056a1e | 2007-04-24 20:44:48 +0900 | [diff] [blame] | 341 | { |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 342 | return __ipgre_bucket(ign, &t->parms); |
YOSHIFUJI Hideaki | 5056a1e | 2007-04-24 20:44:48 +0900 | [diff] [blame] | 343 | } |
| 344 | |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 345 | static void ipgre_tunnel_link(struct ipgre_net *ign, struct ip_tunnel *t) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | { |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 347 | struct ip_tunnel __rcu **tp = ipgre_bucket(ign, t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 349 | rcu_assign_pointer(t->next, rtnl_dereference(*tp)); |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 350 | rcu_assign_pointer(*tp, t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | } |
| 352 | |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 353 | static void ipgre_tunnel_unlink(struct ipgre_net *ign, struct ip_tunnel *t) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | { |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 355 | struct ip_tunnel __rcu **tp; |
| 356 | struct ip_tunnel *iter; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 357 | |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 358 | for (tp = ipgre_bucket(ign, t); |
| 359 | (iter = rtnl_dereference(*tp)) != NULL; |
| 360 | tp = &iter->next) { |
| 361 | if (t == iter) { |
| 362 | rcu_assign_pointer(*tp, t->next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 363 | break; |
| 364 | } |
| 365 | } |
| 366 | } |
| 367 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 368 | static struct ip_tunnel *ipgre_tunnel_find(struct net *net, |
| 369 | struct ip_tunnel_parm *parms, |
| 370 | int type) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | { |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 372 | __be32 remote = parms->iph.daddr; |
| 373 | __be32 local = parms->iph.saddr; |
| 374 | __be32 key = parms->i_key; |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 375 | int link = parms->link; |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 376 | struct ip_tunnel *t; |
| 377 | struct ip_tunnel __rcu **tp; |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 378 | struct ipgre_net *ign = net_generic(net, ipgre_net_id); |
| 379 | |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 380 | for (tp = __ipgre_bucket(ign, parms); |
| 381 | (t = rtnl_dereference(*tp)) != NULL; |
| 382 | tp = &t->next) |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 383 | if (local == t->parms.iph.saddr && |
| 384 | remote == t->parms.iph.daddr && |
| 385 | key == t->parms.i_key && |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 386 | link == t->parms.link && |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 387 | type == t->dev->type) |
| 388 | break; |
| 389 | |
| 390 | return t; |
| 391 | } |
| 392 | |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 393 | static struct ip_tunnel *ipgre_tunnel_locate(struct net *net, |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 394 | struct ip_tunnel_parm *parms, int create) |
| 395 | { |
| 396 | struct ip_tunnel *t, *nt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 397 | struct net_device *dev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | char name[IFNAMSIZ]; |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 399 | struct ipgre_net *ign = net_generic(net, ipgre_net_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 401 | t = ipgre_tunnel_find(net, parms, ARPHRD_IPGRE); |
| 402 | if (t || !create) |
| 403 | return t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | |
| 405 | if (parms->name[0]) |
| 406 | strlcpy(name, parms->name, IFNAMSIZ); |
Pavel Emelyanov | 34cc7ba | 2008-02-23 20:19:20 -0800 | [diff] [blame] | 407 | else |
stephen hemminger | 407d6fc | 2010-11-29 09:47:47 +0000 | [diff] [blame] | 408 | strcpy(name, "gre%d"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | |
| 410 | dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup); |
| 411 | if (!dev) |
stephen hemminger | 407d6fc | 2010-11-29 09:47:47 +0000 | [diff] [blame] | 412 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | |
Pavel Emelyanov | 0b67ece | 2008-04-16 01:11:13 -0700 | [diff] [blame] | 414 | dev_net_set(dev, net); |
| 415 | |
Pavel Emelyanov | b37d428b | 2008-02-26 23:51:04 -0800 | [diff] [blame] | 416 | if (strchr(name, '%')) { |
| 417 | if (dev_alloc_name(dev, name) < 0) |
| 418 | goto failed_free; |
| 419 | } |
| 420 | |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 421 | nt = netdev_priv(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | nt->parms = *parms; |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 423 | dev->rtnl_link_ops = &ipgre_link_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 424 | |
Herbert Xu | 42aa916 | 2008-10-09 11:59:32 -0700 | [diff] [blame] | 425 | dev->mtu = ipgre_tunnel_bind_dev(dev); |
| 426 | |
Pavel Emelyanov | b37d428b | 2008-02-26 23:51:04 -0800 | [diff] [blame] | 427 | if (register_netdevice(dev) < 0) |
| 428 | goto failed_free; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 429 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | dev_hold(dev); |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 431 | ipgre_tunnel_link(ign, nt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | return nt; |
| 433 | |
Pavel Emelyanov | b37d428b | 2008-02-26 23:51:04 -0800 | [diff] [blame] | 434 | failed_free: |
| 435 | free_netdev(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | return NULL; |
| 437 | } |
| 438 | |
| 439 | static void ipgre_tunnel_uninit(struct net_device *dev) |
| 440 | { |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 441 | struct net *net = dev_net(dev); |
| 442 | struct ipgre_net *ign = net_generic(net, ipgre_net_id); |
| 443 | |
| 444 | ipgre_tunnel_unlink(ign, netdev_priv(dev)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | dev_put(dev); |
| 446 | } |
| 447 | |
| 448 | |
| 449 | static void ipgre_err(struct sk_buff *skb, u32 info) |
| 450 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | |
Rami Rosen | 071f92d | 2008-05-21 17:47:54 -0700 | [diff] [blame] | 452 | /* All the routers (except for Linux) return only |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | 8 bytes of packet payload. It means, that precise relaying of |
| 454 | ICMP in the real Internet is absolutely infeasible. |
| 455 | |
| 456 | Moreover, Cisco "wise men" put GRE key to the third word |
| 457 | in GRE header. It makes impossible maintaining even soft state for keyed |
| 458 | GRE tunnels with enabled checksum. Tell them "thank you". |
| 459 | |
| 460 | Well, I wonder, rfc1812 was written by Cisco employee, |
| 461 | what the hell these idiots break standrads established |
| 462 | by themself??? |
| 463 | */ |
| 464 | |
Jianjun Kong | 6ed2533 | 2008-11-03 00:25:16 -0800 | [diff] [blame] | 465 | struct iphdr *iph = (struct iphdr *)skb->data; |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 466 | __be16 *p = (__be16*)(skb->data+(iph->ihl<<2)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | int grehlen = (iph->ihl<<2) + 4; |
Arnaldo Carvalho de Melo | 88c7664 | 2007-03-13 14:43:18 -0300 | [diff] [blame] | 468 | const int type = icmp_hdr(skb)->type; |
| 469 | const int code = icmp_hdr(skb)->code; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | struct ip_tunnel *t; |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 471 | __be16 flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | |
| 473 | flags = p[0]; |
| 474 | if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) { |
| 475 | if (flags&(GRE_VERSION|GRE_ROUTING)) |
| 476 | return; |
| 477 | if (flags&GRE_KEY) { |
| 478 | grehlen += 4; |
| 479 | if (flags&GRE_CSUM) |
| 480 | grehlen += 4; |
| 481 | } |
| 482 | } |
| 483 | |
| 484 | /* If only 8 bytes returned, keyed message will be dropped here */ |
| 485 | if (skb_headlen(skb) < grehlen) |
| 486 | return; |
| 487 | |
| 488 | switch (type) { |
| 489 | default: |
| 490 | case ICMP_PARAMETERPROB: |
| 491 | return; |
| 492 | |
| 493 | case ICMP_DEST_UNREACH: |
| 494 | switch (code) { |
| 495 | case ICMP_SR_FAILED: |
| 496 | case ICMP_PORT_UNREACH: |
| 497 | /* Impossible event. */ |
| 498 | return; |
| 499 | case ICMP_FRAG_NEEDED: |
| 500 | /* Soft state for pmtu is maintained by IP core. */ |
| 501 | return; |
| 502 | default: |
| 503 | /* All others are translated to HOST_UNREACH. |
| 504 | rfc2003 contains "deep thoughts" about NET_UNREACH, |
| 505 | I believe they are just ether pollution. --ANK |
| 506 | */ |
| 507 | break; |
| 508 | } |
| 509 | break; |
| 510 | case ICMP_TIME_EXCEEDED: |
| 511 | if (code != ICMP_EXC_TTL) |
| 512 | return; |
| 513 | break; |
| 514 | } |
| 515 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 516 | rcu_read_lock(); |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 517 | t = ipgre_tunnel_lookup(skb->dev, iph->daddr, iph->saddr, |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 518 | flags & GRE_KEY ? |
| 519 | *(((__be32 *)p) + (grehlen / 4) - 1) : 0, |
| 520 | p[1]); |
Joe Perches | f97c1e0 | 2007-12-16 13:45:43 -0800 | [diff] [blame] | 521 | if (t == NULL || t->parms.iph.daddr == 0 || |
| 522 | ipv4_is_multicast(t->parms.iph.daddr)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 523 | goto out; |
| 524 | |
| 525 | if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED) |
| 526 | goto out; |
| 527 | |
Wei Yongjun | da6185d8 | 2009-02-24 23:34:48 -0800 | [diff] [blame] | 528 | if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | t->err_count++; |
| 530 | else |
| 531 | t->err_count = 1; |
| 532 | t->err_time = jiffies; |
| 533 | out: |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 534 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 535 | } |
| 536 | |
| 537 | static inline void ipgre_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb) |
| 538 | { |
| 539 | if (INET_ECN_is_ce(iph->tos)) { |
| 540 | if (skb->protocol == htons(ETH_P_IP)) { |
Arnaldo Carvalho de Melo | eddc9ec | 2007-04-20 22:47:35 -0700 | [diff] [blame] | 541 | IP_ECN_set_ce(ip_hdr(skb)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | } else if (skb->protocol == htons(ETH_P_IPV6)) { |
Arnaldo Carvalho de Melo | 0660e03 | 2007-04-25 17:54:47 -0700 | [diff] [blame] | 543 | IP6_ECN_set_ce(ipv6_hdr(skb)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | } |
| 545 | } |
| 546 | } |
| 547 | |
| 548 | static inline u8 |
| 549 | ipgre_ecn_encapsulate(u8 tos, struct iphdr *old_iph, struct sk_buff *skb) |
| 550 | { |
| 551 | u8 inner = 0; |
| 552 | if (skb->protocol == htons(ETH_P_IP)) |
| 553 | inner = old_iph->tos; |
| 554 | else if (skb->protocol == htons(ETH_P_IPV6)) |
| 555 | inner = ipv6_get_dsfield((struct ipv6hdr *)old_iph); |
| 556 | return INET_ECN_encapsulate(tos, inner); |
| 557 | } |
| 558 | |
| 559 | static int ipgre_rcv(struct sk_buff *skb) |
| 560 | { |
| 561 | struct iphdr *iph; |
| 562 | u8 *h; |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 563 | __be16 flags; |
Al Viro | d3bc23e | 2006-11-14 21:24:49 -0800 | [diff] [blame] | 564 | __sum16 csum = 0; |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 565 | __be32 key = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | u32 seqno = 0; |
| 567 | struct ip_tunnel *tunnel; |
| 568 | int offset = 4; |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 569 | __be16 gre_proto; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | |
| 571 | if (!pskb_may_pull(skb, 16)) |
| 572 | goto drop_nolock; |
| 573 | |
Arnaldo Carvalho de Melo | eddc9ec | 2007-04-20 22:47:35 -0700 | [diff] [blame] | 574 | iph = ip_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | h = skb->data; |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 576 | flags = *(__be16*)h; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | |
| 578 | if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) { |
| 579 | /* - Version must be 0. |
| 580 | - We do not support routing headers. |
| 581 | */ |
| 582 | if (flags&(GRE_VERSION|GRE_ROUTING)) |
| 583 | goto drop_nolock; |
| 584 | |
| 585 | if (flags&GRE_CSUM) { |
Herbert Xu | fb286bb | 2005-11-10 13:01:24 -0800 | [diff] [blame] | 586 | switch (skb->ip_summed) { |
Patrick McHardy | 84fa793 | 2006-08-29 16:44:56 -0700 | [diff] [blame] | 587 | case CHECKSUM_COMPLETE: |
Al Viro | d3bc23e | 2006-11-14 21:24:49 -0800 | [diff] [blame] | 588 | csum = csum_fold(skb->csum); |
Herbert Xu | fb286bb | 2005-11-10 13:01:24 -0800 | [diff] [blame] | 589 | if (!csum) |
| 590 | break; |
| 591 | /* fall through */ |
| 592 | case CHECKSUM_NONE: |
| 593 | skb->csum = 0; |
| 594 | csum = __skb_checksum_complete(skb); |
Patrick McHardy | 84fa793 | 2006-08-29 16:44:56 -0700 | [diff] [blame] | 595 | skb->ip_summed = CHECKSUM_COMPLETE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 596 | } |
| 597 | offset += 4; |
| 598 | } |
| 599 | if (flags&GRE_KEY) { |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 600 | key = *(__be32*)(h + offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 601 | offset += 4; |
| 602 | } |
| 603 | if (flags&GRE_SEQ) { |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 604 | seqno = ntohl(*(__be32*)(h + offset)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 605 | offset += 4; |
| 606 | } |
| 607 | } |
| 608 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 609 | gre_proto = *(__be16 *)(h + 2); |
| 610 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 611 | rcu_read_lock(); |
Timo Teras | 749c10f | 2009-01-19 17:22:12 -0800 | [diff] [blame] | 612 | if ((tunnel = ipgre_tunnel_lookup(skb->dev, |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 613 | iph->saddr, iph->daddr, key, |
| 614 | gre_proto))) { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 615 | struct pcpu_tstats *tstats; |
Pavel Emelyanov | addd68e | 2008-05-21 14:14:22 -0700 | [diff] [blame] | 616 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | secpath_reset(skb); |
| 618 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 619 | skb->protocol = gre_proto; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | /* WCCP version 1 and 2 protocol decoding. |
| 621 | * - Change protocol to IP |
| 622 | * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header |
| 623 | */ |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 624 | if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) { |
YOSHIFUJI Hideaki | 496c98d | 2006-10-10 19:41:21 -0700 | [diff] [blame] | 625 | skb->protocol = htons(ETH_P_IP); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 626 | if ((*(h + offset) & 0xF0) != 0x40) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | offset += 4; |
| 628 | } |
| 629 | |
Timo Teras | 1d06916 | 2007-12-20 00:10:33 -0800 | [diff] [blame] | 630 | skb->mac_header = skb->network_header; |
Arnaldo Carvalho de Melo | 4209fb6 | 2007-03-10 18:42:03 -0300 | [diff] [blame] | 631 | __pskb_pull(skb, offset); |
Arnaldo Carvalho de Melo | 9c70220 | 2007-04-25 18:04:18 -0700 | [diff] [blame] | 632 | skb_postpull_rcsum(skb, skb_transport_header(skb), offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 633 | skb->pkt_type = PACKET_HOST; |
| 634 | #ifdef CONFIG_NET_IPGRE_BROADCAST |
Joe Perches | f97c1e0 | 2007-12-16 13:45:43 -0800 | [diff] [blame] | 635 | if (ipv4_is_multicast(iph->daddr)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | /* Looped back packet, drop it! */ |
David S. Miller | c753796 | 2010-11-11 17:07:48 -0800 | [diff] [blame] | 637 | if (rt_is_output_route(skb_rtable(skb))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | goto drop; |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 639 | tunnel->dev->stats.multicast++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 640 | skb->pkt_type = PACKET_BROADCAST; |
| 641 | } |
| 642 | #endif |
| 643 | |
| 644 | if (((flags&GRE_CSUM) && csum) || |
| 645 | (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 646 | tunnel->dev->stats.rx_crc_errors++; |
| 647 | tunnel->dev->stats.rx_errors++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | goto drop; |
| 649 | } |
| 650 | if (tunnel->parms.i_flags&GRE_SEQ) { |
| 651 | if (!(flags&GRE_SEQ) || |
| 652 | (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 653 | tunnel->dev->stats.rx_fifo_errors++; |
| 654 | tunnel->dev->stats.rx_errors++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | goto drop; |
| 656 | } |
| 657 | tunnel->i_seqno = seqno + 1; |
| 658 | } |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 659 | |
| 660 | /* Warning: All skb pointers will be invalidated! */ |
| 661 | if (tunnel->dev->type == ARPHRD_ETHER) { |
| 662 | if (!pskb_may_pull(skb, ETH_HLEN)) { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 663 | tunnel->dev->stats.rx_length_errors++; |
| 664 | tunnel->dev->stats.rx_errors++; |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 665 | goto drop; |
| 666 | } |
| 667 | |
| 668 | iph = ip_hdr(skb); |
| 669 | skb->protocol = eth_type_trans(skb, tunnel->dev); |
| 670 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); |
| 671 | } |
| 672 | |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 673 | tstats = this_cpu_ptr(tunnel->dev->tstats); |
| 674 | tstats->rx_packets++; |
| 675 | tstats->rx_bytes += skb->len; |
| 676 | |
| 677 | __skb_tunnel_rx(skb, tunnel->dev); |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 678 | |
| 679 | skb_reset_network_header(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | ipgre_ecn_decapsulate(iph, skb); |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 681 | |
Eric Dumazet | caf586e | 2010-09-30 21:06:55 +0000 | [diff] [blame] | 682 | netif_rx(skb); |
Eric Dumazet | 8990f46 | 2010-09-20 00:12:11 +0000 | [diff] [blame] | 683 | |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 684 | rcu_read_unlock(); |
Eric Dumazet | 8990f46 | 2010-09-20 00:12:11 +0000 | [diff] [blame] | 685 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | } |
Herbert Xu | 45af08b | 2006-04-05 22:31:19 -0700 | [diff] [blame] | 687 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | |
| 689 | drop: |
Eric Dumazet | 8d5b2c0 | 2009-10-23 06:14:38 +0000 | [diff] [blame] | 690 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 691 | drop_nolock: |
| 692 | kfree_skb(skb); |
Eric Dumazet | a02cec2 | 2010-09-22 20:43:57 +0000 | [diff] [blame] | 693 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | } |
| 695 | |
Stephen Hemminger | 6fef4c0 | 2009-08-31 19:50:41 +0000 | [diff] [blame] | 696 | static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | { |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 698 | struct ip_tunnel *tunnel = netdev_priv(dev); |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 699 | struct pcpu_tstats *tstats; |
Arnaldo Carvalho de Melo | eddc9ec | 2007-04-20 22:47:35 -0700 | [diff] [blame] | 700 | struct iphdr *old_iph = ip_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 701 | struct iphdr *tiph; |
| 702 | u8 tos; |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 703 | __be16 df; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | struct rtable *rt; /* Route to the other host */ |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 705 | struct net_device *tdev; /* Device to other host */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | struct iphdr *iph; /* Our new IP header */ |
Chuck Lever | c2636b4 | 2007-10-23 21:07:32 -0700 | [diff] [blame] | 707 | unsigned int max_headroom; /* The extra header space needed */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 708 | int gre_hlen; |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 709 | __be32 dst; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 710 | int mtu; |
| 711 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 712 | if (dev->type == ARPHRD_ETHER) |
| 713 | IPCB(skb)->flags = 0; |
| 714 | |
| 715 | if (dev->header_ops && dev->type == ARPHRD_IPGRE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | gre_hlen = 0; |
Jianjun Kong | 6ed2533 | 2008-11-03 00:25:16 -0800 | [diff] [blame] | 717 | tiph = (struct iphdr *)skb->data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | } else { |
| 719 | gre_hlen = tunnel->hlen; |
| 720 | tiph = &tunnel->parms.iph; |
| 721 | } |
| 722 | |
| 723 | if ((dst = tiph->daddr) == 0) { |
| 724 | /* NBMA tunnel */ |
| 725 | |
Eric Dumazet | adf3090 | 2009-06-02 05:19:30 +0000 | [diff] [blame] | 726 | if (skb_dst(skb) == NULL) { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 727 | dev->stats.tx_fifo_errors++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | goto tx_error; |
| 729 | } |
| 730 | |
| 731 | if (skb->protocol == htons(ETH_P_IP)) { |
Eric Dumazet | 511c3f9 | 2009-06-02 05:14:27 +0000 | [diff] [blame] | 732 | rt = skb_rtable(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 733 | if ((dst = rt->rt_gateway) == 0) |
| 734 | goto tx_error_icmp; |
| 735 | } |
Eric Dumazet | 842c74b | 2010-09-20 10:06:12 -0700 | [diff] [blame] | 736 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 737 | else if (skb->protocol == htons(ETH_P_IPV6)) { |
| 738 | struct in6_addr *addr6; |
| 739 | int addr_type; |
Eric Dumazet | adf3090 | 2009-06-02 05:19:30 +0000 | [diff] [blame] | 740 | struct neighbour *neigh = skb_dst(skb)->neighbour; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 741 | |
| 742 | if (neigh == NULL) |
| 743 | goto tx_error; |
| 744 | |
Jianjun Kong | 6ed2533 | 2008-11-03 00:25:16 -0800 | [diff] [blame] | 745 | addr6 = (struct in6_addr *)&neigh->primary_key; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 746 | addr_type = ipv6_addr_type(addr6); |
| 747 | |
| 748 | if (addr_type == IPV6_ADDR_ANY) { |
Arnaldo Carvalho de Melo | 0660e03 | 2007-04-25 17:54:47 -0700 | [diff] [blame] | 749 | addr6 = &ipv6_hdr(skb)->daddr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 750 | addr_type = ipv6_addr_type(addr6); |
| 751 | } |
| 752 | |
| 753 | if ((addr_type & IPV6_ADDR_COMPATv4) == 0) |
| 754 | goto tx_error_icmp; |
| 755 | |
| 756 | dst = addr6->s6_addr32[3]; |
| 757 | } |
| 758 | #endif |
| 759 | else |
| 760 | goto tx_error; |
| 761 | } |
| 762 | |
| 763 | tos = tiph->tos; |
Andreas Jaggi | ee686ca | 2009-07-14 09:35:59 -0700 | [diff] [blame] | 764 | if (tos == 1) { |
| 765 | tos = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 766 | if (skb->protocol == htons(ETH_P_IP)) |
| 767 | tos = old_iph->tos; |
Stephen Hemminger | dd4ba83 | 2010-07-08 21:35:58 -0700 | [diff] [blame] | 768 | else if (skb->protocol == htons(ETH_P_IPV6)) |
| 769 | tos = ipv6_get_dsfield((struct ipv6hdr *)old_iph); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | } |
| 771 | |
| 772 | { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 773 | struct flowi fl = { |
| 774 | .oif = tunnel->parms.link, |
Changli Gao | 5811662 | 2010-11-12 18:43:55 +0000 | [diff] [blame] | 775 | .fl4_dst = dst, |
| 776 | .fl4_src = tiph->saddr, |
| 777 | .fl4_tos = RT_TOS(tos), |
Steffen Klassert | 946bf5e | 2011-02-11 11:21:57 -0800 | [diff] [blame] | 778 | .proto = IPPROTO_GRE, |
Timo Teräs | cc9ff19 | 2010-11-03 04:41:38 +0000 | [diff] [blame] | 779 | .fl_gre_key = tunnel->parms.o_key |
| 780 | }; |
David S. Miller | b23dd4f | 2011-03-02 14:31:35 -0800 | [diff] [blame^] | 781 | rt = ip_route_output_key(dev_net(dev), &fl); |
| 782 | if (IS_ERR(rt)) { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 783 | dev->stats.tx_carrier_errors++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 784 | goto tx_error; |
| 785 | } |
| 786 | } |
Changli Gao | d8d1f30 | 2010-06-10 23:31:35 -0700 | [diff] [blame] | 787 | tdev = rt->dst.dev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 788 | |
| 789 | if (tdev == dev) { |
| 790 | ip_rt_put(rt); |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 791 | dev->stats.collisions++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | goto tx_error; |
| 793 | } |
| 794 | |
| 795 | df = tiph->frag_off; |
| 796 | if (df) |
Changli Gao | d8d1f30 | 2010-06-10 23:31:35 -0700 | [diff] [blame] | 797 | mtu = dst_mtu(&rt->dst) - dev->hard_header_len - tunnel->hlen; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | else |
Eric Dumazet | adf3090 | 2009-06-02 05:19:30 +0000 | [diff] [blame] | 799 | mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | |
Eric Dumazet | adf3090 | 2009-06-02 05:19:30 +0000 | [diff] [blame] | 801 | if (skb_dst(skb)) |
| 802 | skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 803 | |
| 804 | if (skb->protocol == htons(ETH_P_IP)) { |
| 805 | df |= (old_iph->frag_off&htons(IP_DF)); |
| 806 | |
| 807 | if ((old_iph->frag_off&htons(IP_DF)) && |
| 808 | mtu < ntohs(old_iph->tot_len)) { |
| 809 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); |
| 810 | ip_rt_put(rt); |
| 811 | goto tx_error; |
| 812 | } |
| 813 | } |
Eric Dumazet | 842c74b | 2010-09-20 10:06:12 -0700 | [diff] [blame] | 814 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | else if (skb->protocol == htons(ETH_P_IPV6)) { |
Eric Dumazet | adf3090 | 2009-06-02 05:19:30 +0000 | [diff] [blame] | 816 | struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 817 | |
Eric Dumazet | adf3090 | 2009-06-02 05:19:30 +0000 | [diff] [blame] | 818 | if (rt6 && mtu < dst_mtu(skb_dst(skb)) && mtu >= IPV6_MIN_MTU) { |
Joe Perches | f97c1e0 | 2007-12-16 13:45:43 -0800 | [diff] [blame] | 819 | if ((tunnel->parms.iph.daddr && |
| 820 | !ipv4_is_multicast(tunnel->parms.iph.daddr)) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 821 | rt6->rt6i_dst.plen == 128) { |
| 822 | rt6->rt6i_flags |= RTF_MODIFIED; |
David S. Miller | defb351 | 2010-12-08 21:16:57 -0800 | [diff] [blame] | 823 | dst_metric_set(skb_dst(skb), RTAX_MTU, mtu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 824 | } |
| 825 | } |
| 826 | |
| 827 | if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) { |
Alexey Dobriyan | 3ffe533 | 2010-02-18 08:25:24 +0000 | [diff] [blame] | 828 | icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 829 | ip_rt_put(rt); |
| 830 | goto tx_error; |
| 831 | } |
| 832 | } |
| 833 | #endif |
| 834 | |
| 835 | if (tunnel->err_count > 0) { |
Wei Yongjun | da6185d8 | 2009-02-24 23:34:48 -0800 | [diff] [blame] | 836 | if (time_before(jiffies, |
| 837 | tunnel->err_time + IPTUNNEL_ERR_TIMEO)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | tunnel->err_count--; |
| 839 | |
| 840 | dst_link_failure(skb); |
| 841 | } else |
| 842 | tunnel->err_count = 0; |
| 843 | } |
| 844 | |
Changli Gao | d8d1f30 | 2010-06-10 23:31:35 -0700 | [diff] [blame] | 845 | max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen + rt->dst.header_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | |
Patrick McHardy | cfbba49 | 2007-07-09 15:33:40 -0700 | [diff] [blame] | 847 | if (skb_headroom(skb) < max_headroom || skb_shared(skb)|| |
| 848 | (skb_cloned(skb) && !skb_clone_writable(skb, 0))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 849 | struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom); |
Timo Teräs | 243aad8 | 2010-03-20 02:27:58 +0000 | [diff] [blame] | 850 | if (max_headroom > dev->needed_headroom) |
| 851 | dev->needed_headroom = max_headroom; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | if (!new_skb) { |
| 853 | ip_rt_put(rt); |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 854 | dev->stats.tx_dropped++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | dev_kfree_skb(skb); |
Patrick McHardy | 6ed1065 | 2009-06-23 06:03:08 +0000 | [diff] [blame] | 856 | return NETDEV_TX_OK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 857 | } |
| 858 | if (skb->sk) |
| 859 | skb_set_owner_w(new_skb, skb->sk); |
| 860 | dev_kfree_skb(skb); |
| 861 | skb = new_skb; |
Arnaldo Carvalho de Melo | eddc9ec | 2007-04-20 22:47:35 -0700 | [diff] [blame] | 862 | old_iph = ip_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | } |
| 864 | |
Herbert Xu | 64194c3 | 2008-10-09 12:03:17 -0700 | [diff] [blame] | 865 | skb_reset_transport_header(skb); |
Arnaldo Carvalho de Melo | e2d1bca | 2007-04-10 20:46:21 -0700 | [diff] [blame] | 866 | skb_push(skb, gre_hlen); |
| 867 | skb_reset_network_header(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 868 | memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); |
Patrick McHardy | 48d5cad | 2006-02-15 15:10:22 -0800 | [diff] [blame] | 869 | IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED | |
| 870 | IPSKB_REROUTED); |
Eric Dumazet | adf3090 | 2009-06-02 05:19:30 +0000 | [diff] [blame] | 871 | skb_dst_drop(skb); |
Changli Gao | d8d1f30 | 2010-06-10 23:31:35 -0700 | [diff] [blame] | 872 | skb_dst_set(skb, &rt->dst); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 873 | |
| 874 | /* |
| 875 | * Push down and install the IPIP header. |
| 876 | */ |
| 877 | |
Arnaldo Carvalho de Melo | eddc9ec | 2007-04-20 22:47:35 -0700 | [diff] [blame] | 878 | iph = ip_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 879 | iph->version = 4; |
| 880 | iph->ihl = sizeof(struct iphdr) >> 2; |
| 881 | iph->frag_off = df; |
| 882 | iph->protocol = IPPROTO_GRE; |
| 883 | iph->tos = ipgre_ecn_encapsulate(tos, old_iph, skb); |
| 884 | iph->daddr = rt->rt_dst; |
| 885 | iph->saddr = rt->rt_src; |
| 886 | |
| 887 | if ((iph->ttl = tiph->ttl) == 0) { |
| 888 | if (skb->protocol == htons(ETH_P_IP)) |
| 889 | iph->ttl = old_iph->ttl; |
Eric Dumazet | 842c74b | 2010-09-20 10:06:12 -0700 | [diff] [blame] | 890 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 891 | else if (skb->protocol == htons(ETH_P_IPV6)) |
Jianjun Kong | 6ed2533 | 2008-11-03 00:25:16 -0800 | [diff] [blame] | 892 | iph->ttl = ((struct ipv6hdr *)old_iph)->hop_limit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | #endif |
| 894 | else |
David S. Miller | 323e126 | 2010-12-12 21:55:08 -0800 | [diff] [blame] | 895 | iph->ttl = ip4_dst_hoplimit(&rt->dst); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | } |
| 897 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 898 | ((__be16 *)(iph + 1))[0] = tunnel->parms.o_flags; |
| 899 | ((__be16 *)(iph + 1))[1] = (dev->type == ARPHRD_ETHER) ? |
| 900 | htons(ETH_P_TEB) : skb->protocol; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 901 | |
| 902 | if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) { |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 903 | __be32 *ptr = (__be32*)(((u8*)iph) + tunnel->hlen - 4); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 904 | |
| 905 | if (tunnel->parms.o_flags&GRE_SEQ) { |
| 906 | ++tunnel->o_seqno; |
| 907 | *ptr = htonl(tunnel->o_seqno); |
| 908 | ptr--; |
| 909 | } |
| 910 | if (tunnel->parms.o_flags&GRE_KEY) { |
| 911 | *ptr = tunnel->parms.o_key; |
| 912 | ptr--; |
| 913 | } |
| 914 | if (tunnel->parms.o_flags&GRE_CSUM) { |
| 915 | *ptr = 0; |
Al Viro | 5f92a73 | 2006-11-14 21:36:54 -0800 | [diff] [blame] | 916 | *(__sum16*)ptr = ip_compute_csum((void*)(iph+1), skb->len - sizeof(struct iphdr)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 917 | } |
| 918 | } |
| 919 | |
| 920 | nf_reset(skb); |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 921 | tstats = this_cpu_ptr(dev->tstats); |
| 922 | __IPTUNNEL_XMIT(tstats, &dev->stats); |
Patrick McHardy | 6ed1065 | 2009-06-23 06:03:08 +0000 | [diff] [blame] | 923 | return NETDEV_TX_OK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | |
| 925 | tx_error_icmp: |
| 926 | dst_link_failure(skb); |
| 927 | |
| 928 | tx_error: |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 929 | dev->stats.tx_errors++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 930 | dev_kfree_skb(skb); |
Patrick McHardy | 6ed1065 | 2009-06-23 06:03:08 +0000 | [diff] [blame] | 931 | return NETDEV_TX_OK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 932 | } |
| 933 | |
Herbert Xu | 42aa916 | 2008-10-09 11:59:32 -0700 | [diff] [blame] | 934 | static int ipgre_tunnel_bind_dev(struct net_device *dev) |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 935 | { |
| 936 | struct net_device *tdev = NULL; |
| 937 | struct ip_tunnel *tunnel; |
| 938 | struct iphdr *iph; |
| 939 | int hlen = LL_MAX_HEADER; |
| 940 | int mtu = ETH_DATA_LEN; |
| 941 | int addend = sizeof(struct iphdr) + 4; |
| 942 | |
| 943 | tunnel = netdev_priv(dev); |
| 944 | iph = &tunnel->parms.iph; |
| 945 | |
Herbert Xu | c95b819 | 2008-10-09 11:58:54 -0700 | [diff] [blame] | 946 | /* Guess output device to choose reasonable mtu and needed_headroom */ |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 947 | |
| 948 | if (iph->daddr) { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 949 | struct flowi fl = { |
| 950 | .oif = tunnel->parms.link, |
Changli Gao | 5811662 | 2010-11-12 18:43:55 +0000 | [diff] [blame] | 951 | .fl4_dst = iph->daddr, |
| 952 | .fl4_src = iph->saddr, |
| 953 | .fl4_tos = RT_TOS(iph->tos), |
Timo Teräs | cc9ff19 | 2010-11-03 04:41:38 +0000 | [diff] [blame] | 954 | .proto = IPPROTO_GRE, |
| 955 | .fl_gre_key = tunnel->parms.o_key |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 956 | }; |
David S. Miller | b23dd4f | 2011-03-02 14:31:35 -0800 | [diff] [blame^] | 957 | struct rtable *rt = ip_route_output_key(dev_net(dev), &fl); |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 958 | |
David S. Miller | b23dd4f | 2011-03-02 14:31:35 -0800 | [diff] [blame^] | 959 | if (!IS_ERR(rt)) { |
Changli Gao | d8d1f30 | 2010-06-10 23:31:35 -0700 | [diff] [blame] | 960 | tdev = rt->dst.dev; |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 961 | ip_rt_put(rt); |
| 962 | } |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 963 | |
| 964 | if (dev->type != ARPHRD_ETHER) |
| 965 | dev->flags |= IFF_POINTOPOINT; |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 966 | } |
| 967 | |
| 968 | if (!tdev && tunnel->parms.link) |
Pavel Emelyanov | 9663552 | 2008-04-16 01:10:44 -0700 | [diff] [blame] | 969 | tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link); |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 970 | |
| 971 | if (tdev) { |
Herbert Xu | c95b819 | 2008-10-09 11:58:54 -0700 | [diff] [blame] | 972 | hlen = tdev->hard_header_len + tdev->needed_headroom; |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 973 | mtu = tdev->mtu; |
| 974 | } |
| 975 | dev->iflink = tunnel->parms.link; |
| 976 | |
| 977 | /* Precalculate GRE options length */ |
| 978 | if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) { |
| 979 | if (tunnel->parms.o_flags&GRE_CSUM) |
| 980 | addend += 4; |
| 981 | if (tunnel->parms.o_flags&GRE_KEY) |
| 982 | addend += 4; |
| 983 | if (tunnel->parms.o_flags&GRE_SEQ) |
| 984 | addend += 4; |
| 985 | } |
Herbert Xu | c95b819 | 2008-10-09 11:58:54 -0700 | [diff] [blame] | 986 | dev->needed_headroom = addend + hlen; |
Tom Goff | 8cdb045 | 2009-08-14 16:33:56 -0700 | [diff] [blame] | 987 | mtu -= dev->hard_header_len + addend; |
Herbert Xu | 42aa916 | 2008-10-09 11:59:32 -0700 | [diff] [blame] | 988 | |
| 989 | if (mtu < 68) |
| 990 | mtu = 68; |
| 991 | |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 992 | tunnel->hlen = addend; |
| 993 | |
Herbert Xu | 42aa916 | 2008-10-09 11:59:32 -0700 | [diff] [blame] | 994 | return mtu; |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 995 | } |
| 996 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 997 | static int |
| 998 | ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd) |
| 999 | { |
| 1000 | int err = 0; |
| 1001 | struct ip_tunnel_parm p; |
| 1002 | struct ip_tunnel *t; |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 1003 | struct net *net = dev_net(dev); |
| 1004 | struct ipgre_net *ign = net_generic(net, ipgre_net_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | |
| 1006 | switch (cmd) { |
| 1007 | case SIOCGETTUNNEL: |
| 1008 | t = NULL; |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1009 | if (dev == ign->fb_tunnel_dev) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1010 | if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) { |
| 1011 | err = -EFAULT; |
| 1012 | break; |
| 1013 | } |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 1014 | t = ipgre_tunnel_locate(net, &p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | } |
| 1016 | if (t == NULL) |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 1017 | t = netdev_priv(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1018 | memcpy(&p, &t->parms, sizeof(p)); |
| 1019 | if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p))) |
| 1020 | err = -EFAULT; |
| 1021 | break; |
| 1022 | |
| 1023 | case SIOCADDTUNNEL: |
| 1024 | case SIOCCHGTUNNEL: |
| 1025 | err = -EPERM; |
| 1026 | if (!capable(CAP_NET_ADMIN)) |
| 1027 | goto done; |
| 1028 | |
| 1029 | err = -EFAULT; |
| 1030 | if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) |
| 1031 | goto done; |
| 1032 | |
| 1033 | err = -EINVAL; |
| 1034 | if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE || |
| 1035 | p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) || |
| 1036 | ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))) |
| 1037 | goto done; |
| 1038 | if (p.iph.ttl) |
| 1039 | p.iph.frag_off |= htons(IP_DF); |
| 1040 | |
| 1041 | if (!(p.i_flags&GRE_KEY)) |
| 1042 | p.i_key = 0; |
| 1043 | if (!(p.o_flags&GRE_KEY)) |
| 1044 | p.o_key = 0; |
| 1045 | |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 1046 | t = ipgre_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1047 | |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1048 | if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1049 | if (t != NULL) { |
| 1050 | if (t->dev != dev) { |
| 1051 | err = -EEXIST; |
| 1052 | break; |
| 1053 | } |
| 1054 | } else { |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 1055 | unsigned int nflags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1056 | |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 1057 | t = netdev_priv(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | |
Joe Perches | f97c1e0 | 2007-12-16 13:45:43 -0800 | [diff] [blame] | 1059 | if (ipv4_is_multicast(p.iph.daddr)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | nflags = IFF_BROADCAST; |
| 1061 | else if (p.iph.daddr) |
| 1062 | nflags = IFF_POINTOPOINT; |
| 1063 | |
| 1064 | if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) { |
| 1065 | err = -EINVAL; |
| 1066 | break; |
| 1067 | } |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 1068 | ipgre_tunnel_unlink(ign, t); |
Pavel Emelyanov | 74b0b85 | 2010-10-27 05:43:53 +0000 | [diff] [blame] | 1069 | synchronize_net(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1070 | t->parms.iph.saddr = p.iph.saddr; |
| 1071 | t->parms.iph.daddr = p.iph.daddr; |
| 1072 | t->parms.i_key = p.i_key; |
| 1073 | t->parms.o_key = p.o_key; |
| 1074 | memcpy(dev->dev_addr, &p.iph.saddr, 4); |
| 1075 | memcpy(dev->broadcast, &p.iph.daddr, 4); |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 1076 | ipgre_tunnel_link(ign, t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | netdev_state_change(dev); |
| 1078 | } |
| 1079 | } |
| 1080 | |
| 1081 | if (t) { |
| 1082 | err = 0; |
| 1083 | if (cmd == SIOCCHGTUNNEL) { |
| 1084 | t->parms.iph.ttl = p.iph.ttl; |
| 1085 | t->parms.iph.tos = p.iph.tos; |
| 1086 | t->parms.iph.frag_off = p.iph.frag_off; |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 1087 | if (t->parms.link != p.link) { |
| 1088 | t->parms.link = p.link; |
Herbert Xu | 42aa916 | 2008-10-09 11:59:32 -0700 | [diff] [blame] | 1089 | dev->mtu = ipgre_tunnel_bind_dev(dev); |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 1090 | netdev_state_change(dev); |
| 1091 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1092 | } |
| 1093 | if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p))) |
| 1094 | err = -EFAULT; |
| 1095 | } else |
| 1096 | err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT); |
| 1097 | break; |
| 1098 | |
| 1099 | case SIOCDELTUNNEL: |
| 1100 | err = -EPERM; |
| 1101 | if (!capable(CAP_NET_ADMIN)) |
| 1102 | goto done; |
| 1103 | |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1104 | if (dev == ign->fb_tunnel_dev) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | err = -EFAULT; |
| 1106 | if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) |
| 1107 | goto done; |
| 1108 | err = -ENOENT; |
Pavel Emelyanov | f57e7d5 | 2008-04-16 01:09:22 -0700 | [diff] [blame] | 1109 | if ((t = ipgre_tunnel_locate(net, &p, 0)) == NULL) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1110 | goto done; |
| 1111 | err = -EPERM; |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1112 | if (t == netdev_priv(ign->fb_tunnel_dev)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | goto done; |
| 1114 | dev = t->dev; |
| 1115 | } |
Stephen Hemminger | 22f8cde | 2007-02-07 00:09:58 -0800 | [diff] [blame] | 1116 | unregister_netdevice(dev); |
| 1117 | err = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | break; |
| 1119 | |
| 1120 | default: |
| 1121 | err = -EINVAL; |
| 1122 | } |
| 1123 | |
| 1124 | done: |
| 1125 | return err; |
| 1126 | } |
| 1127 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1128 | static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu) |
| 1129 | { |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 1130 | struct ip_tunnel *tunnel = netdev_priv(dev); |
Herbert Xu | c95b819 | 2008-10-09 11:58:54 -0700 | [diff] [blame] | 1131 | if (new_mtu < 68 || |
| 1132 | new_mtu > 0xFFF8 - dev->hard_header_len - tunnel->hlen) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | return -EINVAL; |
| 1134 | dev->mtu = new_mtu; |
| 1135 | return 0; |
| 1136 | } |
| 1137 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1138 | /* Nice toy. Unfortunately, useless in real life :-) |
| 1139 | It allows to construct virtual multiprotocol broadcast "LAN" |
| 1140 | over the Internet, provided multicast routing is tuned. |
| 1141 | |
| 1142 | |
| 1143 | I have no idea was this bicycle invented before me, |
| 1144 | so that I had to set ARPHRD_IPGRE to a random value. |
| 1145 | I have an impression, that Cisco could make something similar, |
| 1146 | but this feature is apparently missing in IOS<=11.2(8). |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 1147 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1148 | I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks |
| 1149 | with broadcast 224.66.66.66. If you have access to mbone, play with me :-) |
| 1150 | |
| 1151 | ping -t 255 224.66.66.66 |
| 1152 | |
| 1153 | If nobody answers, mbone does not work. |
| 1154 | |
| 1155 | ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255 |
| 1156 | ip addr add 10.66.66.<somewhat>/24 dev Universe |
| 1157 | ifconfig Universe up |
| 1158 | ifconfig Universe add fe80::<Your_real_addr>/10 |
| 1159 | ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96 |
| 1160 | ftp 10.66.66.66 |
| 1161 | ... |
| 1162 | ftp fec0:6666:6666::193.233.7.65 |
| 1163 | ... |
| 1164 | |
| 1165 | */ |
| 1166 | |
Stephen Hemminger | 3b04ddd | 2007-10-09 01:40:57 -0700 | [diff] [blame] | 1167 | static int ipgre_header(struct sk_buff *skb, struct net_device *dev, |
| 1168 | unsigned short type, |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 1169 | const void *daddr, const void *saddr, unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1170 | { |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 1171 | struct ip_tunnel *t = netdev_priv(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1172 | struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen); |
Al Viro | d5a0a1e | 2006-11-08 00:23:14 -0800 | [diff] [blame] | 1173 | __be16 *p = (__be16*)(iph+1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1174 | |
| 1175 | memcpy(iph, &t->parms.iph, sizeof(struct iphdr)); |
| 1176 | p[0] = t->parms.o_flags; |
| 1177 | p[1] = htons(type); |
| 1178 | |
| 1179 | /* |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 1180 | * Set the source hardware address. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1181 | */ |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 1182 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1183 | if (saddr) |
| 1184 | memcpy(&iph->saddr, saddr, 4); |
Timo Teräs | 6d55cb9 | 2010-03-03 04:01:13 +0000 | [diff] [blame] | 1185 | if (daddr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1186 | memcpy(&iph->daddr, daddr, 4); |
Timo Teräs | 6d55cb9 | 2010-03-03 04:01:13 +0000 | [diff] [blame] | 1187 | if (iph->daddr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1188 | return t->hlen; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 1189 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1190 | return -t->hlen; |
| 1191 | } |
| 1192 | |
Timo Teras | 6a5f44d | 2007-10-23 20:31:53 -0700 | [diff] [blame] | 1193 | static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr) |
| 1194 | { |
Jianjun Kong | 6ed2533 | 2008-11-03 00:25:16 -0800 | [diff] [blame] | 1195 | struct iphdr *iph = (struct iphdr *) skb_mac_header(skb); |
Timo Teras | 6a5f44d | 2007-10-23 20:31:53 -0700 | [diff] [blame] | 1196 | memcpy(haddr, &iph->saddr, 4); |
| 1197 | return 4; |
| 1198 | } |
| 1199 | |
Stephen Hemminger | 3b04ddd | 2007-10-09 01:40:57 -0700 | [diff] [blame] | 1200 | static const struct header_ops ipgre_header_ops = { |
| 1201 | .create = ipgre_header, |
Timo Teras | 6a5f44d | 2007-10-23 20:31:53 -0700 | [diff] [blame] | 1202 | .parse = ipgre_header_parse, |
Stephen Hemminger | 3b04ddd | 2007-10-09 01:40:57 -0700 | [diff] [blame] | 1203 | }; |
| 1204 | |
Timo Teras | 6a5f44d | 2007-10-23 20:31:53 -0700 | [diff] [blame] | 1205 | #ifdef CONFIG_NET_IPGRE_BROADCAST |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1206 | static int ipgre_open(struct net_device *dev) |
| 1207 | { |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 1208 | struct ip_tunnel *t = netdev_priv(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1209 | |
Joe Perches | f97c1e0 | 2007-12-16 13:45:43 -0800 | [diff] [blame] | 1210 | if (ipv4_is_multicast(t->parms.iph.daddr)) { |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1211 | struct flowi fl = { |
| 1212 | .oif = t->parms.link, |
Changli Gao | 5811662 | 2010-11-12 18:43:55 +0000 | [diff] [blame] | 1213 | .fl4_dst = t->parms.iph.daddr, |
| 1214 | .fl4_src = t->parms.iph.saddr, |
| 1215 | .fl4_tos = RT_TOS(t->parms.iph.tos), |
Timo Teräs | cc9ff19 | 2010-11-03 04:41:38 +0000 | [diff] [blame] | 1216 | .proto = IPPROTO_GRE, |
| 1217 | .fl_gre_key = t->parms.o_key |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1218 | }; |
David S. Miller | b23dd4f | 2011-03-02 14:31:35 -0800 | [diff] [blame^] | 1219 | struct rtable *rt = ip_route_output_key(dev_net(dev), &fl); |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1220 | |
David S. Miller | b23dd4f | 2011-03-02 14:31:35 -0800 | [diff] [blame^] | 1221 | if (IS_ERR(rt)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | return -EADDRNOTAVAIL; |
Changli Gao | d8d1f30 | 2010-06-10 23:31:35 -0700 | [diff] [blame] | 1223 | dev = rt->dst.dev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | ip_rt_put(rt); |
Herbert Xu | e5ed639 | 2005-10-03 14:35:55 -0700 | [diff] [blame] | 1225 | if (__in_dev_get_rtnl(dev) == NULL) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1226 | return -EADDRNOTAVAIL; |
| 1227 | t->mlink = dev->ifindex; |
Herbert Xu | e5ed639 | 2005-10-03 14:35:55 -0700 | [diff] [blame] | 1228 | ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | } |
| 1230 | return 0; |
| 1231 | } |
| 1232 | |
| 1233 | static int ipgre_close(struct net_device *dev) |
| 1234 | { |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 1235 | struct ip_tunnel *t = netdev_priv(dev); |
Stephen Hemminger | b8c26a3 | 2008-11-20 20:34:29 -0800 | [diff] [blame] | 1236 | |
Joe Perches | f97c1e0 | 2007-12-16 13:45:43 -0800 | [diff] [blame] | 1237 | if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) { |
Denis V. Lunev | 7fee0ca | 2008-01-21 17:32:38 -0800 | [diff] [blame] | 1238 | struct in_device *in_dev; |
YOSHIFUJI Hideaki | c346dca | 2008-03-25 21:47:49 +0900 | [diff] [blame] | 1239 | in_dev = inetdev_by_index(dev_net(dev), t->mlink); |
Eric Dumazet | 8723e1b | 2010-10-19 00:39:26 +0000 | [diff] [blame] | 1240 | if (in_dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | ip_mc_dec_group(in_dev, t->parms.iph.daddr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | } |
| 1243 | return 0; |
| 1244 | } |
| 1245 | |
| 1246 | #endif |
| 1247 | |
Stephen Hemminger | b8c26a3 | 2008-11-20 20:34:29 -0800 | [diff] [blame] | 1248 | static const struct net_device_ops ipgre_netdev_ops = { |
| 1249 | .ndo_init = ipgre_tunnel_init, |
| 1250 | .ndo_uninit = ipgre_tunnel_uninit, |
| 1251 | #ifdef CONFIG_NET_IPGRE_BROADCAST |
| 1252 | .ndo_open = ipgre_open, |
| 1253 | .ndo_stop = ipgre_close, |
| 1254 | #endif |
| 1255 | .ndo_start_xmit = ipgre_tunnel_xmit, |
| 1256 | .ndo_do_ioctl = ipgre_tunnel_ioctl, |
| 1257 | .ndo_change_mtu = ipgre_tunnel_change_mtu, |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1258 | .ndo_get_stats = ipgre_get_stats, |
Stephen Hemminger | b8c26a3 | 2008-11-20 20:34:29 -0800 | [diff] [blame] | 1259 | }; |
| 1260 | |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1261 | static void ipgre_dev_free(struct net_device *dev) |
| 1262 | { |
| 1263 | free_percpu(dev->tstats); |
| 1264 | free_netdev(dev); |
| 1265 | } |
| 1266 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | static void ipgre_tunnel_setup(struct net_device *dev) |
| 1268 | { |
Stephen Hemminger | b8c26a3 | 2008-11-20 20:34:29 -0800 | [diff] [blame] | 1269 | dev->netdev_ops = &ipgre_netdev_ops; |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1270 | dev->destructor = ipgre_dev_free; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1271 | |
| 1272 | dev->type = ARPHRD_IPGRE; |
Herbert Xu | c95b819 | 2008-10-09 11:58:54 -0700 | [diff] [blame] | 1273 | dev->needed_headroom = LL_MAX_HEADER + sizeof(struct iphdr) + 4; |
Kris Katterjohn | 46f25df | 2006-01-05 16:35:42 -0800 | [diff] [blame] | 1274 | dev->mtu = ETH_DATA_LEN - sizeof(struct iphdr) - 4; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1275 | dev->flags = IFF_NOARP; |
| 1276 | dev->iflink = 0; |
| 1277 | dev->addr_len = 4; |
Pavel Emelyanov | 0b67ece | 2008-04-16 01:11:13 -0700 | [diff] [blame] | 1278 | dev->features |= NETIF_F_NETNS_LOCAL; |
Eric Dumazet | 108bfa8 | 2009-05-28 22:35:10 +0000 | [diff] [blame] | 1279 | dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | } |
| 1281 | |
| 1282 | static int ipgre_tunnel_init(struct net_device *dev) |
| 1283 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1284 | struct ip_tunnel *tunnel; |
| 1285 | struct iphdr *iph; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1286 | |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 1287 | tunnel = netdev_priv(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1288 | iph = &tunnel->parms.iph; |
| 1289 | |
| 1290 | tunnel->dev = dev; |
| 1291 | strcpy(tunnel->parms.name, dev->name); |
| 1292 | |
| 1293 | memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4); |
| 1294 | memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4); |
| 1295 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1296 | if (iph->daddr) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1297 | #ifdef CONFIG_NET_IPGRE_BROADCAST |
Joe Perches | f97c1e0 | 2007-12-16 13:45:43 -0800 | [diff] [blame] | 1298 | if (ipv4_is_multicast(iph->daddr)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1299 | if (!iph->saddr) |
| 1300 | return -EINVAL; |
| 1301 | dev->flags = IFF_BROADCAST; |
Stephen Hemminger | 3b04ddd | 2007-10-09 01:40:57 -0700 | [diff] [blame] | 1302 | dev->header_ops = &ipgre_header_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1303 | } |
| 1304 | #endif |
Michal Schmidt | ee34c1e | 2007-12-13 09:46:32 -0800 | [diff] [blame] | 1305 | } else |
Timo Teras | 6a5f44d | 2007-10-23 20:31:53 -0700 | [diff] [blame] | 1306 | dev->header_ops = &ipgre_header_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1307 | |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1308 | dev->tstats = alloc_percpu(struct pcpu_tstats); |
| 1309 | if (!dev->tstats) |
| 1310 | return -ENOMEM; |
| 1311 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1312 | return 0; |
| 1313 | } |
| 1314 | |
Stephen Hemminger | b8c26a3 | 2008-11-20 20:34:29 -0800 | [diff] [blame] | 1315 | static void ipgre_fb_tunnel_init(struct net_device *dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | { |
Patrick McHardy | 2941a48 | 2006-01-08 22:05:26 -0800 | [diff] [blame] | 1317 | struct ip_tunnel *tunnel = netdev_priv(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1318 | struct iphdr *iph = &tunnel->parms.iph; |
| 1319 | |
| 1320 | tunnel->dev = dev; |
| 1321 | strcpy(tunnel->parms.name, dev->name); |
| 1322 | |
| 1323 | iph->version = 4; |
| 1324 | iph->protocol = IPPROTO_GRE; |
| 1325 | iph->ihl = 5; |
| 1326 | tunnel->hlen = sizeof(struct iphdr) + 4; |
| 1327 | |
| 1328 | dev_hold(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | } |
| 1330 | |
| 1331 | |
Dmitry Kozlov | 00959ad | 2010-08-21 23:05:39 -0700 | [diff] [blame] | 1332 | static const struct gre_protocol ipgre_protocol = { |
| 1333 | .handler = ipgre_rcv, |
| 1334 | .err_handler = ipgre_err, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1335 | }; |
| 1336 | |
Eric Dumazet | eef6dd6 | 2009-10-27 07:07:16 +0000 | [diff] [blame] | 1337 | static void ipgre_destroy_tunnels(struct ipgre_net *ign, struct list_head *head) |
Pavel Emelyanov | eb8ce74 | 2008-04-16 01:10:26 -0700 | [diff] [blame] | 1338 | { |
| 1339 | int prio; |
| 1340 | |
| 1341 | for (prio = 0; prio < 4; prio++) { |
| 1342 | int h; |
| 1343 | for (h = 0; h < HASH_SIZE; h++) { |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 1344 | struct ip_tunnel *t; |
| 1345 | |
| 1346 | t = rtnl_dereference(ign->tunnels[prio][h]); |
Eric Dumazet | eef6dd6 | 2009-10-27 07:07:16 +0000 | [diff] [blame] | 1347 | |
| 1348 | while (t != NULL) { |
| 1349 | unregister_netdevice_queue(t->dev, head); |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 1350 | t = rtnl_dereference(t->next); |
Eric Dumazet | eef6dd6 | 2009-10-27 07:07:16 +0000 | [diff] [blame] | 1351 | } |
Pavel Emelyanov | eb8ce74 | 2008-04-16 01:10:26 -0700 | [diff] [blame] | 1352 | } |
| 1353 | } |
| 1354 | } |
| 1355 | |
Alexey Dobriyan | 2c8c1e7 | 2010-01-17 03:35:32 +0000 | [diff] [blame] | 1356 | static int __net_init ipgre_init_net(struct net *net) |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1357 | { |
Eric W. Biederman | cfb8fbf | 2009-11-29 15:46:13 +0000 | [diff] [blame] | 1358 | struct ipgre_net *ign = net_generic(net, ipgre_net_id); |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1359 | int err; |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1360 | |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1361 | ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0", |
| 1362 | ipgre_tunnel_setup); |
| 1363 | if (!ign->fb_tunnel_dev) { |
| 1364 | err = -ENOMEM; |
| 1365 | goto err_alloc_dev; |
| 1366 | } |
Alexey Dobriyan | be77e59 | 2008-11-23 17:26:26 -0800 | [diff] [blame] | 1367 | dev_net_set(ign->fb_tunnel_dev, net); |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1368 | |
Stephen Hemminger | b8c26a3 | 2008-11-20 20:34:29 -0800 | [diff] [blame] | 1369 | ipgre_fb_tunnel_init(ign->fb_tunnel_dev); |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1370 | ign->fb_tunnel_dev->rtnl_link_ops = &ipgre_link_ops; |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1371 | |
| 1372 | if ((err = register_netdev(ign->fb_tunnel_dev))) |
| 1373 | goto err_reg_dev; |
| 1374 | |
Eric Dumazet | 3285ee3 | 2010-10-30 16:21:28 -0700 | [diff] [blame] | 1375 | rcu_assign_pointer(ign->tunnels_wc[0], |
| 1376 | netdev_priv(ign->fb_tunnel_dev)); |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1377 | return 0; |
| 1378 | |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1379 | err_reg_dev: |
Eric Dumazet | 3285ee3 | 2010-10-30 16:21:28 -0700 | [diff] [blame] | 1380 | ipgre_dev_free(ign->fb_tunnel_dev); |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1381 | err_alloc_dev: |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1382 | return err; |
| 1383 | } |
| 1384 | |
Alexey Dobriyan | 2c8c1e7 | 2010-01-17 03:35:32 +0000 | [diff] [blame] | 1385 | static void __net_exit ipgre_exit_net(struct net *net) |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1386 | { |
| 1387 | struct ipgre_net *ign; |
Eric Dumazet | eef6dd6 | 2009-10-27 07:07:16 +0000 | [diff] [blame] | 1388 | LIST_HEAD(list); |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1389 | |
| 1390 | ign = net_generic(net, ipgre_net_id); |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1391 | rtnl_lock(); |
Eric Dumazet | eef6dd6 | 2009-10-27 07:07:16 +0000 | [diff] [blame] | 1392 | ipgre_destroy_tunnels(ign, &list); |
| 1393 | unregister_netdevice_many(&list); |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1394 | rtnl_unlock(); |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1395 | } |
| 1396 | |
| 1397 | static struct pernet_operations ipgre_net_ops = { |
| 1398 | .init = ipgre_init_net, |
| 1399 | .exit = ipgre_exit_net, |
Eric W. Biederman | cfb8fbf | 2009-11-29 15:46:13 +0000 | [diff] [blame] | 1400 | .id = &ipgre_net_id, |
| 1401 | .size = sizeof(struct ipgre_net), |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1402 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1404 | static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[]) |
| 1405 | { |
| 1406 | __be16 flags; |
| 1407 | |
| 1408 | if (!data) |
| 1409 | return 0; |
| 1410 | |
| 1411 | flags = 0; |
| 1412 | if (data[IFLA_GRE_IFLAGS]) |
| 1413 | flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]); |
| 1414 | if (data[IFLA_GRE_OFLAGS]) |
| 1415 | flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]); |
| 1416 | if (flags & (GRE_VERSION|GRE_ROUTING)) |
| 1417 | return -EINVAL; |
| 1418 | |
| 1419 | return 0; |
| 1420 | } |
| 1421 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1422 | static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[]) |
| 1423 | { |
| 1424 | __be32 daddr; |
| 1425 | |
| 1426 | if (tb[IFLA_ADDRESS]) { |
| 1427 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) |
| 1428 | return -EINVAL; |
| 1429 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) |
| 1430 | return -EADDRNOTAVAIL; |
| 1431 | } |
| 1432 | |
| 1433 | if (!data) |
| 1434 | goto out; |
| 1435 | |
| 1436 | if (data[IFLA_GRE_REMOTE]) { |
| 1437 | memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4); |
| 1438 | if (!daddr) |
| 1439 | return -EINVAL; |
| 1440 | } |
| 1441 | |
| 1442 | out: |
| 1443 | return ipgre_tunnel_validate(tb, data); |
| 1444 | } |
| 1445 | |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1446 | static void ipgre_netlink_parms(struct nlattr *data[], |
| 1447 | struct ip_tunnel_parm *parms) |
| 1448 | { |
Herbert Xu | 7bb82d9 | 2008-10-11 12:20:15 -0700 | [diff] [blame] | 1449 | memset(parms, 0, sizeof(*parms)); |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1450 | |
| 1451 | parms->iph.protocol = IPPROTO_GRE; |
| 1452 | |
| 1453 | if (!data) |
| 1454 | return; |
| 1455 | |
| 1456 | if (data[IFLA_GRE_LINK]) |
| 1457 | parms->link = nla_get_u32(data[IFLA_GRE_LINK]); |
| 1458 | |
| 1459 | if (data[IFLA_GRE_IFLAGS]) |
| 1460 | parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]); |
| 1461 | |
| 1462 | if (data[IFLA_GRE_OFLAGS]) |
| 1463 | parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]); |
| 1464 | |
| 1465 | if (data[IFLA_GRE_IKEY]) |
| 1466 | parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]); |
| 1467 | |
| 1468 | if (data[IFLA_GRE_OKEY]) |
| 1469 | parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]); |
| 1470 | |
| 1471 | if (data[IFLA_GRE_LOCAL]) |
Patrick McHardy | 4d74f8b | 2008-10-10 12:11:06 -0700 | [diff] [blame] | 1472 | parms->iph.saddr = nla_get_be32(data[IFLA_GRE_LOCAL]); |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1473 | |
| 1474 | if (data[IFLA_GRE_REMOTE]) |
Patrick McHardy | 4d74f8b | 2008-10-10 12:11:06 -0700 | [diff] [blame] | 1475 | parms->iph.daddr = nla_get_be32(data[IFLA_GRE_REMOTE]); |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1476 | |
| 1477 | if (data[IFLA_GRE_TTL]) |
| 1478 | parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]); |
| 1479 | |
| 1480 | if (data[IFLA_GRE_TOS]) |
| 1481 | parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]); |
| 1482 | |
| 1483 | if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) |
| 1484 | parms->iph.frag_off = htons(IP_DF); |
| 1485 | } |
| 1486 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1487 | static int ipgre_tap_init(struct net_device *dev) |
| 1488 | { |
| 1489 | struct ip_tunnel *tunnel; |
| 1490 | |
| 1491 | tunnel = netdev_priv(dev); |
| 1492 | |
| 1493 | tunnel->dev = dev; |
| 1494 | strcpy(tunnel->parms.name, dev->name); |
| 1495 | |
| 1496 | ipgre_tunnel_bind_dev(dev); |
| 1497 | |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1498 | dev->tstats = alloc_percpu(struct pcpu_tstats); |
| 1499 | if (!dev->tstats) |
| 1500 | return -ENOMEM; |
| 1501 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1502 | return 0; |
| 1503 | } |
| 1504 | |
Stephen Hemminger | b8c26a3 | 2008-11-20 20:34:29 -0800 | [diff] [blame] | 1505 | static const struct net_device_ops ipgre_tap_netdev_ops = { |
| 1506 | .ndo_init = ipgre_tap_init, |
| 1507 | .ndo_uninit = ipgre_tunnel_uninit, |
| 1508 | .ndo_start_xmit = ipgre_tunnel_xmit, |
| 1509 | .ndo_set_mac_address = eth_mac_addr, |
| 1510 | .ndo_validate_addr = eth_validate_addr, |
| 1511 | .ndo_change_mtu = ipgre_tunnel_change_mtu, |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1512 | .ndo_get_stats = ipgre_get_stats, |
Stephen Hemminger | b8c26a3 | 2008-11-20 20:34:29 -0800 | [diff] [blame] | 1513 | }; |
| 1514 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1515 | static void ipgre_tap_setup(struct net_device *dev) |
| 1516 | { |
| 1517 | |
| 1518 | ether_setup(dev); |
| 1519 | |
Herbert Xu | 2e9526b | 2009-10-30 05:51:48 +0000 | [diff] [blame] | 1520 | dev->netdev_ops = &ipgre_tap_netdev_ops; |
Eric Dumazet | e985aad | 2010-09-27 03:57:11 +0000 | [diff] [blame] | 1521 | dev->destructor = ipgre_dev_free; |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1522 | |
| 1523 | dev->iflink = 0; |
| 1524 | dev->features |= NETIF_F_NETNS_LOCAL; |
| 1525 | } |
| 1526 | |
Eric W. Biederman | 81adee4 | 2009-11-08 00:53:51 -0800 | [diff] [blame] | 1527 | static int ipgre_newlink(struct net *src_net, struct net_device *dev, struct nlattr *tb[], |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1528 | struct nlattr *data[]) |
| 1529 | { |
| 1530 | struct ip_tunnel *nt; |
| 1531 | struct net *net = dev_net(dev); |
| 1532 | struct ipgre_net *ign = net_generic(net, ipgre_net_id); |
| 1533 | int mtu; |
| 1534 | int err; |
| 1535 | |
| 1536 | nt = netdev_priv(dev); |
| 1537 | ipgre_netlink_parms(data, &nt->parms); |
| 1538 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1539 | if (ipgre_tunnel_find(net, &nt->parms, dev->type)) |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1540 | return -EEXIST; |
| 1541 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1542 | if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS]) |
| 1543 | random_ether_addr(dev->dev_addr); |
| 1544 | |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1545 | mtu = ipgre_tunnel_bind_dev(dev); |
| 1546 | if (!tb[IFLA_MTU]) |
| 1547 | dev->mtu = mtu; |
| 1548 | |
Eric Dumazet | b790e01 | 2010-09-27 23:05:47 +0000 | [diff] [blame] | 1549 | /* Can use a lockless transmit, unless we generate output sequences */ |
| 1550 | if (!(nt->parms.o_flags & GRE_SEQ)) |
| 1551 | dev->features |= NETIF_F_LLTX; |
| 1552 | |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1553 | err = register_netdevice(dev); |
| 1554 | if (err) |
| 1555 | goto out; |
| 1556 | |
| 1557 | dev_hold(dev); |
| 1558 | ipgre_tunnel_link(ign, nt); |
| 1559 | |
| 1560 | out: |
| 1561 | return err; |
| 1562 | } |
| 1563 | |
| 1564 | static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[], |
| 1565 | struct nlattr *data[]) |
| 1566 | { |
| 1567 | struct ip_tunnel *t, *nt; |
| 1568 | struct net *net = dev_net(dev); |
| 1569 | struct ipgre_net *ign = net_generic(net, ipgre_net_id); |
| 1570 | struct ip_tunnel_parm p; |
| 1571 | int mtu; |
| 1572 | |
| 1573 | if (dev == ign->fb_tunnel_dev) |
| 1574 | return -EINVAL; |
| 1575 | |
| 1576 | nt = netdev_priv(dev); |
| 1577 | ipgre_netlink_parms(data, &p); |
| 1578 | |
| 1579 | t = ipgre_tunnel_locate(net, &p, 0); |
| 1580 | |
| 1581 | if (t) { |
| 1582 | if (t->dev != dev) |
| 1583 | return -EEXIST; |
| 1584 | } else { |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1585 | t = nt; |
| 1586 | |
Herbert Xu | 2e9526b | 2009-10-30 05:51:48 +0000 | [diff] [blame] | 1587 | if (dev->type != ARPHRD_ETHER) { |
Eric Dumazet | 1507850 | 2010-09-15 11:07:53 +0000 | [diff] [blame] | 1588 | unsigned int nflags = 0; |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1589 | |
Herbert Xu | 2e9526b | 2009-10-30 05:51:48 +0000 | [diff] [blame] | 1590 | if (ipv4_is_multicast(p.iph.daddr)) |
| 1591 | nflags = IFF_BROADCAST; |
| 1592 | else if (p.iph.daddr) |
| 1593 | nflags = IFF_POINTOPOINT; |
| 1594 | |
| 1595 | if ((dev->flags ^ nflags) & |
| 1596 | (IFF_POINTOPOINT | IFF_BROADCAST)) |
| 1597 | return -EINVAL; |
| 1598 | } |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1599 | |
| 1600 | ipgre_tunnel_unlink(ign, t); |
| 1601 | t->parms.iph.saddr = p.iph.saddr; |
| 1602 | t->parms.iph.daddr = p.iph.daddr; |
| 1603 | t->parms.i_key = p.i_key; |
Herbert Xu | 2e9526b | 2009-10-30 05:51:48 +0000 | [diff] [blame] | 1604 | if (dev->type != ARPHRD_ETHER) { |
| 1605 | memcpy(dev->dev_addr, &p.iph.saddr, 4); |
| 1606 | memcpy(dev->broadcast, &p.iph.daddr, 4); |
| 1607 | } |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1608 | ipgre_tunnel_link(ign, t); |
| 1609 | netdev_state_change(dev); |
| 1610 | } |
| 1611 | |
| 1612 | t->parms.o_key = p.o_key; |
| 1613 | t->parms.iph.ttl = p.iph.ttl; |
| 1614 | t->parms.iph.tos = p.iph.tos; |
| 1615 | t->parms.iph.frag_off = p.iph.frag_off; |
| 1616 | |
| 1617 | if (t->parms.link != p.link) { |
| 1618 | t->parms.link = p.link; |
| 1619 | mtu = ipgre_tunnel_bind_dev(dev); |
| 1620 | if (!tb[IFLA_MTU]) |
| 1621 | dev->mtu = mtu; |
| 1622 | netdev_state_change(dev); |
| 1623 | } |
| 1624 | |
| 1625 | return 0; |
| 1626 | } |
| 1627 | |
| 1628 | static size_t ipgre_get_size(const struct net_device *dev) |
| 1629 | { |
| 1630 | return |
| 1631 | /* IFLA_GRE_LINK */ |
| 1632 | nla_total_size(4) + |
| 1633 | /* IFLA_GRE_IFLAGS */ |
| 1634 | nla_total_size(2) + |
| 1635 | /* IFLA_GRE_OFLAGS */ |
| 1636 | nla_total_size(2) + |
| 1637 | /* IFLA_GRE_IKEY */ |
| 1638 | nla_total_size(4) + |
| 1639 | /* IFLA_GRE_OKEY */ |
| 1640 | nla_total_size(4) + |
| 1641 | /* IFLA_GRE_LOCAL */ |
| 1642 | nla_total_size(4) + |
| 1643 | /* IFLA_GRE_REMOTE */ |
| 1644 | nla_total_size(4) + |
| 1645 | /* IFLA_GRE_TTL */ |
| 1646 | nla_total_size(1) + |
| 1647 | /* IFLA_GRE_TOS */ |
| 1648 | nla_total_size(1) + |
| 1649 | /* IFLA_GRE_PMTUDISC */ |
| 1650 | nla_total_size(1) + |
| 1651 | 0; |
| 1652 | } |
| 1653 | |
| 1654 | static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev) |
| 1655 | { |
| 1656 | struct ip_tunnel *t = netdev_priv(dev); |
| 1657 | struct ip_tunnel_parm *p = &t->parms; |
| 1658 | |
| 1659 | NLA_PUT_U32(skb, IFLA_GRE_LINK, p->link); |
| 1660 | NLA_PUT_BE16(skb, IFLA_GRE_IFLAGS, p->i_flags); |
| 1661 | NLA_PUT_BE16(skb, IFLA_GRE_OFLAGS, p->o_flags); |
Patrick McHardy | ba9e64b | 2008-10-10 12:10:30 -0700 | [diff] [blame] | 1662 | NLA_PUT_BE32(skb, IFLA_GRE_IKEY, p->i_key); |
| 1663 | NLA_PUT_BE32(skb, IFLA_GRE_OKEY, p->o_key); |
Patrick McHardy | 4d74f8b | 2008-10-10 12:11:06 -0700 | [diff] [blame] | 1664 | NLA_PUT_BE32(skb, IFLA_GRE_LOCAL, p->iph.saddr); |
| 1665 | NLA_PUT_BE32(skb, IFLA_GRE_REMOTE, p->iph.daddr); |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1666 | NLA_PUT_U8(skb, IFLA_GRE_TTL, p->iph.ttl); |
| 1667 | NLA_PUT_U8(skb, IFLA_GRE_TOS, p->iph.tos); |
| 1668 | NLA_PUT_U8(skb, IFLA_GRE_PMTUDISC, !!(p->iph.frag_off & htons(IP_DF))); |
| 1669 | |
| 1670 | return 0; |
| 1671 | |
| 1672 | nla_put_failure: |
| 1673 | return -EMSGSIZE; |
| 1674 | } |
| 1675 | |
| 1676 | static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = { |
| 1677 | [IFLA_GRE_LINK] = { .type = NLA_U32 }, |
| 1678 | [IFLA_GRE_IFLAGS] = { .type = NLA_U16 }, |
| 1679 | [IFLA_GRE_OFLAGS] = { .type = NLA_U16 }, |
| 1680 | [IFLA_GRE_IKEY] = { .type = NLA_U32 }, |
| 1681 | [IFLA_GRE_OKEY] = { .type = NLA_U32 }, |
Patrick McHardy | 4d74f8b | 2008-10-10 12:11:06 -0700 | [diff] [blame] | 1682 | [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) }, |
| 1683 | [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) }, |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1684 | [IFLA_GRE_TTL] = { .type = NLA_U8 }, |
| 1685 | [IFLA_GRE_TOS] = { .type = NLA_U8 }, |
| 1686 | [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 }, |
| 1687 | }; |
| 1688 | |
| 1689 | static struct rtnl_link_ops ipgre_link_ops __read_mostly = { |
| 1690 | .kind = "gre", |
| 1691 | .maxtype = IFLA_GRE_MAX, |
| 1692 | .policy = ipgre_policy, |
| 1693 | .priv_size = sizeof(struct ip_tunnel), |
| 1694 | .setup = ipgre_tunnel_setup, |
| 1695 | .validate = ipgre_tunnel_validate, |
| 1696 | .newlink = ipgre_newlink, |
| 1697 | .changelink = ipgre_changelink, |
| 1698 | .get_size = ipgre_get_size, |
| 1699 | .fill_info = ipgre_fill_info, |
| 1700 | }; |
| 1701 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1702 | static struct rtnl_link_ops ipgre_tap_ops __read_mostly = { |
| 1703 | .kind = "gretap", |
| 1704 | .maxtype = IFLA_GRE_MAX, |
| 1705 | .policy = ipgre_policy, |
| 1706 | .priv_size = sizeof(struct ip_tunnel), |
| 1707 | .setup = ipgre_tap_setup, |
| 1708 | .validate = ipgre_tap_validate, |
| 1709 | .newlink = ipgre_newlink, |
| 1710 | .changelink = ipgre_changelink, |
| 1711 | .get_size = ipgre_get_size, |
| 1712 | .fill_info = ipgre_fill_info, |
| 1713 | }; |
| 1714 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1715 | /* |
| 1716 | * And now the modules code and kernel interface. |
| 1717 | */ |
| 1718 | |
| 1719 | static int __init ipgre_init(void) |
| 1720 | { |
| 1721 | int err; |
| 1722 | |
| 1723 | printk(KERN_INFO "GRE over IPv4 tunneling driver\n"); |
| 1724 | |
Eric W. Biederman | cfb8fbf | 2009-11-29 15:46:13 +0000 | [diff] [blame] | 1725 | err = register_pernet_device(&ipgre_net_ops); |
Pavel Emelyanov | 59a4c75 | 2008-04-16 01:08:53 -0700 | [diff] [blame] | 1726 | if (err < 0) |
Alexey Dobriyan | c2892f0 | 2010-02-16 07:57:44 +0000 | [diff] [blame] | 1727 | return err; |
| 1728 | |
Dmitry Kozlov | 00959ad | 2010-08-21 23:05:39 -0700 | [diff] [blame] | 1729 | err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO); |
Alexey Dobriyan | c2892f0 | 2010-02-16 07:57:44 +0000 | [diff] [blame] | 1730 | if (err < 0) { |
| 1731 | printk(KERN_INFO "ipgre init: can't add protocol\n"); |
| 1732 | goto add_proto_failed; |
| 1733 | } |
Pavel Emelyanov | 7daa000 | 2008-04-16 01:10:05 -0700 | [diff] [blame] | 1734 | |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1735 | err = rtnl_link_register(&ipgre_link_ops); |
| 1736 | if (err < 0) |
| 1737 | goto rtnl_link_failed; |
| 1738 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1739 | err = rtnl_link_register(&ipgre_tap_ops); |
| 1740 | if (err < 0) |
| 1741 | goto tap_ops_failed; |
| 1742 | |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1743 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1744 | return err; |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1745 | |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1746 | tap_ops_failed: |
| 1747 | rtnl_link_unregister(&ipgre_link_ops); |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1748 | rtnl_link_failed: |
Dmitry Kozlov | 00959ad | 2010-08-21 23:05:39 -0700 | [diff] [blame] | 1749 | gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO); |
Alexey Dobriyan | c2892f0 | 2010-02-16 07:57:44 +0000 | [diff] [blame] | 1750 | add_proto_failed: |
| 1751 | unregister_pernet_device(&ipgre_net_ops); |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1752 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | } |
| 1754 | |
Alexey Kuznetsov | db44575 | 2005-07-30 17:46:44 -0700 | [diff] [blame] | 1755 | static void __exit ipgre_fini(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1756 | { |
Herbert Xu | e1a8000 | 2008-10-09 12:00:17 -0700 | [diff] [blame] | 1757 | rtnl_link_unregister(&ipgre_tap_ops); |
Herbert Xu | c19e654 | 2008-10-09 11:59:55 -0700 | [diff] [blame] | 1758 | rtnl_link_unregister(&ipgre_link_ops); |
Dmitry Kozlov | 00959ad | 2010-08-21 23:05:39 -0700 | [diff] [blame] | 1759 | if (gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1760 | printk(KERN_INFO "ipgre close: can't remove protocol\n"); |
Alexey Dobriyan | c2892f0 | 2010-02-16 07:57:44 +0000 | [diff] [blame] | 1761 | unregister_pernet_device(&ipgre_net_ops); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1762 | } |
| 1763 | |
| 1764 | module_init(ipgre_init); |
| 1765 | module_exit(ipgre_fini); |
| 1766 | MODULE_LICENSE("GPL"); |
Patrick McHardy | 4d74f8b | 2008-10-10 12:11:06 -0700 | [diff] [blame] | 1767 | MODULE_ALIAS_RTNL_LINK("gre"); |
| 1768 | MODULE_ALIAS_RTNL_LINK("gretap"); |
stephen hemminger | 4da6a73 | 2010-11-29 09:47:48 +0000 | [diff] [blame] | 1769 | MODULE_ALIAS("gre0"); |