1 /* $KAME: radix_mpath.c,v 1.17 2004/11/08 10:29:39 itojun Exp $ */
2
3 /*
4 * Copyright (C) 2001 WIDE Project.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the project nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 * THE AUTHORS DO NOT GUARANTEE THAT THIS SOFTWARE DOES NOT INFRINGE
31 * ANY OTHERS' INTELLECTUAL PROPERTIES. IN NO EVENT SHALL THE AUTHORS
32 * BE LIABLE FOR ANY INFRINGEMENT OF ANY OTHERS' INTELLECTUAL
33 * PROPERTIES.
34 */
35
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD: stable/10/sys/net/radix_mpath.c 265711 2014-05-08 20:41:39Z melifaro $");
38
39 #include "opt_inet.h"
40 #include "opt_inet6.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/malloc.h>
45 #include <sys/socket.h>
46 #include <sys/domain.h>
47 #include <sys/syslog.h>
48 #include <net/radix.h>
49 #include <net/radix_mpath.h>
50 #include <net/route.h>
51 #include <net/if.h>
52 #include <net/if_var.h>
53
54 /*
55 * give some jitter to hash, to avoid synchronization between routers
56 */
57 static uint32_t hashjitter;
58
59 int
rn_mpath_capable(struct radix_node_head * rnh)60 rn_mpath_capable(struct radix_node_head *rnh)
61 {
62
63 return rnh->rnh_multipath;
64 }
65
66 struct radix_node *
rn_mpath_next(struct radix_node * rn)67 rn_mpath_next(struct radix_node *rn)
68 {
69 struct radix_node *next;
70
71 if (!rn->rn_dupedkey)
72 return NULL;
73 next = rn->rn_dupedkey;
74 if (rn->rn_mask == next->rn_mask)
75 return next;
76 else
77 return NULL;
78 }
79
80 uint32_t
rn_mpath_count(struct radix_node * rn)81 rn_mpath_count(struct radix_node *rn)
82 {
83 uint32_t i = 0;
84 struct rtentry *rt;
85
86 while (rn != NULL) {
87 rt = (struct rtentry *)rn;
88 i += rt->rt_weight;
89 rn = rn_mpath_next(rn);
90 }
91 return (i);
92 }
93
94 struct rtentry *
rt_mpath_matchgate(struct rtentry * rt,struct sockaddr * gate)95 rt_mpath_matchgate(struct rtentry *rt, struct sockaddr *gate)
96 {
97 struct radix_node *rn;
98
99 if (!gate || !rt->rt_gateway)
100 return NULL;
101
102 /* beyond here, we use rn as the master copy */
103 rn = (struct radix_node *)rt;
104 do {
105 rt = (struct rtentry *)rn;
106 /*
107 * we are removing an address alias that has
108 * the same prefix as another address
109 * we need to compare the interface address because
110 * rt_gateway is a special sockadd_dl structure
111 */
112 if (rt->rt_gateway->sa_family == AF_LINK) {
113 if (!memcmp(rt->rt_ifa->ifa_addr, gate, gate->sa_len))
114 break;
115 }
116
117 /*
118 * Check for other options:
119 * 1) Routes with 'real' IPv4/IPv6 gateway
120 * 2) Loopback host routes (another AF_LINK/sockadd_dl check)
121 * */
122 if (rt->rt_gateway->sa_len == gate->sa_len &&
123 !memcmp(rt->rt_gateway, gate, gate->sa_len))
124 break;
125 } while ((rn = rn_mpath_next(rn)) != NULL);
126
127 return (struct rtentry *)rn;
128 }
129
130 /*
131 * go through the chain and unlink "rt" from the list
132 * the caller will free "rt"
133 */
134 int
rt_mpath_deldup(struct rtentry * headrt,struct rtentry * rt)135 rt_mpath_deldup(struct rtentry *headrt, struct rtentry *rt)
136 {
137 struct radix_node *t, *tt;
138
139 if (!headrt || !rt)
140 return (0);
141 t = (struct radix_node *)headrt;
142 tt = rn_mpath_next(t);
143 while (tt) {
144 if (tt == (struct radix_node *)rt) {
145 t->rn_dupedkey = tt->rn_dupedkey;
146 tt->rn_dupedkey = NULL;
147 tt->rn_flags &= ~RNF_ACTIVE;
148 tt[1].rn_flags &= ~RNF_ACTIVE;
149 return (1);
150 }
151 t = tt;
152 tt = rn_mpath_next((struct radix_node *)t);
153 }
154 return (0);
155 }
156
157 /*
158 * check if we have the same key/mask/gateway on the table already.
159 * Assume @rt rt_key host bits are cleared according to @netmask
160 */
161 int
rt_mpath_conflict(struct radix_node_head * rnh,struct rtentry * rt,struct sockaddr * netmask)162 rt_mpath_conflict(struct radix_node_head *rnh, struct rtentry *rt,
163 struct sockaddr *netmask)
164 {
165 struct radix_node *rn, *rn1;
166 struct rtentry *rt1;
167
168 rn = (struct radix_node *)rt;
169 rn1 = rnh->rnh_lookup(rt_key(rt), netmask, rnh);
170 if (!rn1 || rn1->rn_flags & RNF_ROOT)
171 return (0);
172
173 /* key/mask are the same. compare gateway for all multipaths */
174 do {
175 rt1 = (struct rtentry *)rn1;
176
177 /* sanity: no use in comparing the same thing */
178 if (rn1 == rn)
179 continue;
180
181 if (rt1->rt_gateway->sa_family == AF_LINK) {
182 if (rt1->rt_ifa->ifa_addr->sa_len != rt->rt_ifa->ifa_addr->sa_len ||
183 bcmp(rt1->rt_ifa->ifa_addr, rt->rt_ifa->ifa_addr,
184 rt1->rt_ifa->ifa_addr->sa_len))
185 continue;
186 } else {
187 if (rt1->rt_gateway->sa_len != rt->rt_gateway->sa_len ||
188 bcmp(rt1->rt_gateway, rt->rt_gateway,
189 rt1->rt_gateway->sa_len))
190 continue;
191 }
192
193 /* all key/mask/gateway are the same. conflicting entry. */
194 return (EEXIST);
195 } while ((rn1 = rn_mpath_next(rn1)) != NULL);
196
197 return (0);
198 }
199
200 void
rtalloc_mpath_fib(struct route * ro,uint32_t hash,u_int fibnum)201 rtalloc_mpath_fib(struct route *ro, uint32_t hash, u_int fibnum)
202 {
203 struct radix_node *rn0, *rn;
204 u_int32_t n;
205 struct rtentry *rt;
206 int64_t weight;
207
208 /*
209 * XXX we don't attempt to lookup cached route again; what should
210 * be done for sendto(3) case?
211 */
212 if (ro->ro_rt && ro->ro_rt->rt_ifp && (ro->ro_rt->rt_flags & RTF_UP)
213 && RT_LINK_IS_UP(ro->ro_rt->rt_ifp))
214 return;
215 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, 0, fibnum);
216
217 /* if the route does not exist or it is not multipath, don't care */
218 if (ro->ro_rt == NULL)
219 return;
220 if (rn_mpath_next((struct radix_node *)ro->ro_rt) == NULL) {
221 RT_UNLOCK(ro->ro_rt);
222 return;
223 }
224
225 /* beyond here, we use rn as the master copy */
226 rn0 = rn = (struct radix_node *)ro->ro_rt;
227 n = rn_mpath_count(rn0);
228
229 /* gw selection by Modulo-N Hash (RFC2991) XXX need improvement? */
230 hash += hashjitter;
231 hash %= n;
232 for (weight = abs((int32_t)hash), rt = ro->ro_rt;
233 weight >= rt->rt_weight && rn;
234 weight -= rt->rt_weight) {
235
236 /* stay within the multipath routes */
237 if (rn->rn_dupedkey && rn->rn_mask != rn->rn_dupedkey->rn_mask)
238 break;
239 rn = rn->rn_dupedkey;
240 rt = (struct rtentry *)rn;
241 }
242 /* XXX try filling rt_gwroute and avoid unreachable gw */
243
244 /* gw selection has failed - there must be only zero weight routes */
245 if (!rn) {
246 RT_UNLOCK(ro->ro_rt);
247 ro->ro_rt = NULL;
248 return;
249 }
250 if (ro->ro_rt != rt) {
251 RTFREE_LOCKED(ro->ro_rt);
252 ro->ro_rt = (struct rtentry *)rn;
253 RT_LOCK(ro->ro_rt);
254 RT_ADDREF(ro->ro_rt);
255
256 }
257 RT_UNLOCK(ro->ro_rt);
258 }
259
260 extern int in6_inithead(void **head, int off);
261 extern int in_inithead(void **head, int off);
262
263 #ifdef INET
264 int
rn4_mpath_inithead(void ** head,int off)265 rn4_mpath_inithead(void **head, int off)
266 {
267 struct radix_node_head *rnh;
268
269 hashjitter = arc4random();
270 if (in_inithead(head, off) == 1) {
271 rnh = (struct radix_node_head *)*head;
272 rnh->rnh_multipath = 1;
273 return 1;
274 } else
275 return 0;
276 }
277 #endif
278
279 #ifdef INET6
280 int
rn6_mpath_inithead(void ** head,int off)281 rn6_mpath_inithead(void **head, int off)
282 {
283 struct radix_node_head *rnh;
284
285 hashjitter = arc4random();
286 if (in6_inithead(head, off) == 1) {
287 rnh = (struct radix_node_head *)*head;
288 rnh->rnh_multipath = 1;
289 return 1;
290 } else
291 return 0;
292 }
293
294 #endif
295