blob: b057d40addec3ef2ffab44413421e502894fe60e [file] [log] [blame]
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * xfrm4_policy.c
3 *
4 * Changes:
5 * Kazunori MIYAZAWA @USAGI
6 * YOSHIFUJI Hideaki @USAGI
7 * Split up af-specific portion
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09008 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
Herbert Xu66cdb3c2007-11-13 21:37:28 -080011#include <linux/err.h>
12#include <linux/kernel.h>
Herbert Xuaabc9762005-05-03 16:27:10 -070013#include <linux/inetdevice.h>
Timo Teräscc9ff192010-11-03 04:41:38 +000014#include <linux/if_tunnel.h>
Herbert Xu45ff5a32007-11-13 21:35:32 -080015#include <net/dst.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <net/xfrm.h>
17#include <net/ip.h>
18
Linus Torvalds1da177e2005-04-16 15:20:36 -070019static struct xfrm_policy_afinfo xfrm4_policy_afinfo;
20
Alexey Dobriyanc5b3cf42008-11-25 17:51:25 -080021static struct dst_entry *xfrm4_dst_lookup(struct net *net, int tos,
22 xfrm_address_t *saddr,
Herbert Xu66cdb3c2007-11-13 21:37:28 -080023 xfrm_address_t *daddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070024{
Herbert Xu66cdb3c2007-11-13 21:37:28 -080025 struct flowi fl = {
Changli Gao58116622010-11-12 18:43:55 +000026 .fl4_dst = daddr->a4,
27 .fl4_tos = tos,
Patrick McHardya1e59ab2006-09-19 12:57:34 -070028 };
Herbert Xu66cdb3c2007-11-13 21:37:28 -080029 struct dst_entry *dst;
30 struct rtable *rt;
31 int err;
Patrick McHardya1e59ab2006-09-19 12:57:34 -070032
Herbert Xu66cdb3c2007-11-13 21:37:28 -080033 if (saddr)
34 fl.fl4_src = saddr->a4;
35
Alexey Dobriyanc5b3cf42008-11-25 17:51:25 -080036 err = __ip_route_output_key(net, &rt, &fl);
Changli Gaod8d1f302010-06-10 23:31:35 -070037 dst = &rt->dst;
Herbert Xu66cdb3c2007-11-13 21:37:28 -080038 if (err)
39 dst = ERR_PTR(err);
40 return dst;
41}
42
Alexey Dobriyanfbda33b2008-11-25 17:56:49 -080043static int xfrm4_get_saddr(struct net *net,
44 xfrm_address_t *saddr, xfrm_address_t *daddr)
Herbert Xu66cdb3c2007-11-13 21:37:28 -080045{
46 struct dst_entry *dst;
47 struct rtable *rt;
48
Alexey Dobriyanfbda33b2008-11-25 17:56:49 -080049 dst = xfrm4_dst_lookup(net, 0, NULL, daddr);
Herbert Xu66cdb3c2007-11-13 21:37:28 -080050 if (IS_ERR(dst))
51 return -EHOSTUNREACH;
52
53 rt = (struct rtable *)dst;
54 saddr->a4 = rt->rt_src;
55 dst_release(dst);
56 return 0;
Patrick McHardya1e59ab2006-09-19 12:57:34 -070057}
58
Herbert Xu25ee3282007-12-11 09:32:34 -080059static int xfrm4_get_tos(struct flowi *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060{
Ulrich Weber94e223892010-09-22 06:45:11 +000061 return IPTOS_RT_MASK & fl->fl4_tos; /* Strip ECN bits */
Herbert Xu25ee3282007-12-11 09:32:34 -080062}
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Masahide NAKAMURAa1b05142007-12-20 20:41:12 -080064static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst,
65 int nfheader_len)
66{
67 return 0;
68}
69
Herbert Xu87c1e122010-03-02 02:51:56 +000070static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
71 struct flowi *fl)
Herbert Xu25ee3282007-12-11 09:32:34 -080072{
73 struct rtable *rt = (struct rtable *)xdst->route;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Herbert Xu87c1e122010-03-02 02:51:56 +000075 xdst->u.rt.fl = *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
Herbert Xu25ee3282007-12-11 09:32:34 -080077 xdst->u.dst.dev = dev;
78 dev_hold(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Herbert Xu25ee3282007-12-11 09:32:34 -080080 xdst->u.rt.peer = rt->peer;
81 if (rt->peer)
82 atomic_inc(&rt->peer->refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Herbert Xu25ee3282007-12-11 09:32:34 -080084 /* Sheit... I remember I did this right. Apparently,
85 * it was magically lost, so this code needs audit */
86 xdst->u.rt.rt_flags = rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST |
87 RTCF_LOCAL);
88 xdst->u.rt.rt_type = rt->rt_type;
89 xdst->u.rt.rt_src = rt->rt_src;
90 xdst->u.rt.rt_dst = rt->rt_dst;
91 xdst->u.rt.rt_gateway = rt->rt_gateway;
92 xdst->u.rt.rt_spec_dst = rt->rt_spec_dst;
Miika Komu43372262007-02-06 14:27:32 -080093
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095}
96
97static void
Herbert Xud5422ef2007-12-12 10:44:16 -080098_decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700100 struct iphdr *iph = ip_hdr(skb);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700101 u8 *xprth = skb_network_header(skb) + iph->ihl * 4;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
103 memset(fl, 0, sizeof(struct flowi));
Peter Kosyh44b451f2010-07-02 07:47:55 +0000104 fl->mark = skb->mark;
105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 if (!(iph->frag_off & htons(IP_MF | IP_OFFSET))) {
107 switch (iph->protocol) {
108 case IPPROTO_UDP:
Gerrit Renkerba4e58e2006-11-27 11:10:57 -0800109 case IPPROTO_UDPLITE:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 case IPPROTO_TCP:
111 case IPPROTO_SCTP:
Patrick McHardy9e999992005-12-19 14:03:46 -0800112 case IPPROTO_DCCP:
Wei Yongjunc615c9f2009-07-02 16:57:23 +0000113 if (xprth + 4 < skb->data ||
114 pskb_may_pull(skb, xprth + 4 - skb->data)) {
Al Viro8c689a62006-11-08 00:20:21 -0800115 __be16 *ports = (__be16 *)xprth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Herbert Xud5422ef2007-12-12 10:44:16 -0800117 fl->fl_ip_sport = ports[!!reverse];
118 fl->fl_ip_dport = ports[!reverse];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 }
120 break;
121
122 case IPPROTO_ICMP:
123 if (pskb_may_pull(skb, xprth + 2 - skb->data)) {
124 u8 *icmp = xprth;
125
126 fl->fl_icmp_type = icmp[0];
127 fl->fl_icmp_code = icmp[1];
128 }
129 break;
130
131 case IPPROTO_ESP:
132 if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
Al Viro4324a172006-09-27 18:49:07 -0700133 __be32 *ehdr = (__be32 *)xprth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
135 fl->fl_ipsec_spi = ehdr[0];
136 }
137 break;
138
139 case IPPROTO_AH:
140 if (pskb_may_pull(skb, xprth + 8 - skb->data)) {
Al Viro4324a172006-09-27 18:49:07 -0700141 __be32 *ah_hdr = (__be32*)xprth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
143 fl->fl_ipsec_spi = ah_hdr[1];
144 }
145 break;
146
147 case IPPROTO_COMP:
148 if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
Al Viro4324a172006-09-27 18:49:07 -0700149 __be16 *ipcomp_hdr = (__be16 *)xprth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Alexey Dobriyan4195f812006-05-22 16:53:22 -0700151 fl->fl_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 }
153 break;
Timo Teräscc9ff192010-11-03 04:41:38 +0000154
155 case IPPROTO_GRE:
156 if (pskb_may_pull(skb, xprth + 12 - skb->data)) {
157 __be16 *greflags = (__be16 *)xprth;
158 __be32 *gre_hdr = (__be32 *)xprth;
159
160 if (greflags[0] & GRE_KEY) {
161 if (greflags[0] & GRE_CSUM)
162 gre_hdr++;
163 fl->fl_gre_key = gre_hdr[1];
164 }
165 }
166 break;
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 default:
169 fl->fl_ipsec_spi = 0;
170 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 }
173 fl->proto = iph->protocol;
Herbert Xud5422ef2007-12-12 10:44:16 -0800174 fl->fl4_dst = reverse ? iph->saddr : iph->daddr;
175 fl->fl4_src = reverse ? iph->daddr : iph->saddr;
Herbert Xu4da30892006-02-23 16:19:26 -0800176 fl->fl4_tos = iph->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
Daniel Lezcano569d3642008-01-18 03:56:57 -0800179static inline int xfrm4_garbage_collect(struct dst_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Alexey Dobriyand7c75442010-01-24 22:47:53 -0800181 struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops);
182
183 xfrm4_policy_afinfo.garbage_collect(net);
Eric Dumazetfc66f952010-10-08 06:37:34 +0000184 return (dst_entries_get_slow(ops) > ops->gc_thresh * 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185}
186
187static void xfrm4_update_pmtu(struct dst_entry *dst, u32 mtu)
188{
189 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
190 struct dst_entry *path = xdst->route;
191
192 path->ops->update_pmtu(path, mtu);
193}
194
Herbert Xuaabc9762005-05-03 16:27:10 -0700195static void xfrm4_dst_destroy(struct dst_entry *dst)
196{
197 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
198
Herbert Xued3e37d2007-10-17 21:34:46 -0700199 if (likely(xdst->u.rt.peer))
David S. Miller26db1672006-12-06 23:45:15 -0800200 inet_putpeer(xdst->u.rt.peer);
Herbert Xuaabc9762005-05-03 16:27:10 -0700201 xfrm_dst_destroy(xdst);
202}
203
204static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
205 int unregister)
206{
Herbert Xuaabc9762005-05-03 16:27:10 -0700207 if (!unregister)
208 return;
209
Herbert Xuaabc9762005-05-03 16:27:10 -0700210 xfrm_dst_ifdown(dst, dev);
211}
212
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213static struct dst_ops xfrm4_dst_ops = {
214 .family = AF_INET,
Harvey Harrison09640e62009-02-01 00:45:17 -0800215 .protocol = cpu_to_be16(ETH_P_IP),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 .gc = xfrm4_garbage_collect,
217 .update_pmtu = xfrm4_update_pmtu,
Herbert Xuaabc9762005-05-03 16:27:10 -0700218 .destroy = xfrm4_dst_destroy,
219 .ifdown = xfrm4_dst_ifdown,
Herbert Xu862b82c2007-11-13 21:43:11 -0800220 .local_out = __ip_local_out,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 .gc_thresh = 1024,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222};
223
224static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
225 .family = AF_INET,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 .dst_ops = &xfrm4_dst_ops,
227 .dst_lookup = xfrm4_dst_lookup,
Patrick McHardya1e59ab2006-09-19 12:57:34 -0700228 .get_saddr = xfrm4_get_saddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 .decode_session = _decode_session4,
Herbert Xu25ee3282007-12-11 09:32:34 -0800230 .get_tos = xfrm4_get_tos,
Masahide NAKAMURAa1b05142007-12-20 20:41:12 -0800231 .init_path = xfrm4_init_path,
Herbert Xu25ee3282007-12-11 09:32:34 -0800232 .fill_dst = xfrm4_fill_dst,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233};
234
Randy Dunlapf8167002009-08-04 20:18:33 -0700235#ifdef CONFIG_SYSCTL
Neil Hormana44a4a02009-07-27 08:22:46 +0000236static struct ctl_table xfrm4_policy_table[] = {
237 {
Neil Hormana44a4a02009-07-27 08:22:46 +0000238 .procname = "xfrm4_gc_thresh",
Alexey Dobriyand7c75442010-01-24 22:47:53 -0800239 .data = &init_net.xfrm.xfrm4_dst_ops.gc_thresh,
Neil Hormana44a4a02009-07-27 08:22:46 +0000240 .maxlen = sizeof(int),
241 .mode = 0644,
242 .proc_handler = proc_dointvec,
243 },
244 { }
245};
246
247static struct ctl_table_header *sysctl_hdr;
Randy Dunlapf8167002009-08-04 20:18:33 -0700248#endif
Neil Hormana44a4a02009-07-27 08:22:46 +0000249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250static void __init xfrm4_policy_init(void)
251{
252 xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
253}
254
255static void __exit xfrm4_policy_fini(void)
256{
Randy Dunlapf8167002009-08-04 20:18:33 -0700257#ifdef CONFIG_SYSCTL
Neil Hormana44a4a02009-07-27 08:22:46 +0000258 if (sysctl_hdr)
259 unregister_net_sysctl_table(sysctl_hdr);
Randy Dunlapf8167002009-08-04 20:18:33 -0700260#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
262}
263
Neil Hormana33bc5c2009-07-30 18:52:15 -0700264void __init xfrm4_init(int rt_max_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
Neil Hormana33bc5c2009-07-30 18:52:15 -0700266 /*
267 * Select a default value for the gc_thresh based on the main route
268 * table hash size. It seems to me the worst case scenario is when
269 * we have ipsec operating in transport mode, in which we create a
270 * dst_entry per socket. The xfrm gc algorithm starts trying to remove
271 * entries at gc_thresh, and prevents new allocations as 2*gc_thresh
272 * so lets set an initial xfrm gc_thresh value at the rt_max_size/2.
273 * That will let us store an ipsec connection per route table entry,
274 * and start cleaning when were 1/2 full
275 */
276 xfrm4_dst_ops.gc_thresh = rt_max_size/2;
Eric Dumazetfc66f952010-10-08 06:37:34 +0000277 dst_entries_init(&xfrm4_dst_ops);
Alexey Dobriyand7c75442010-01-24 22:47:53 -0800278
279 xfrm4_state_init();
280 xfrm4_policy_init();
Randy Dunlapf8167002009-08-04 20:18:33 -0700281#ifdef CONFIG_SYSCTL
Neil Hormana44a4a02009-07-27 08:22:46 +0000282 sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path,
283 xfrm4_policy_table);
Randy Dunlapf8167002009-08-04 20:18:33 -0700284#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285}
286