xref: /freebsd/sys/netinet/in_rmx.c (revision 77a1348b3c1cfe8547be49a121b56299a1e18b69)
1 /*-
2  * Copyright 1994, 1995 Massachusetts Institute of Technology
3  *
4  * Permission to use, copy, modify, and distribute this software and
5  * its documentation for any purpose and without fee is hereby
6  * granted, provided that both the above copyright notice and this
7  * permission notice appear in all copies, that both the above
8  * copyright notice and this permission notice appear in all
9  * supporting documentation, and that the name of M.I.T. not be used
10  * in advertising or publicity pertaining to distribution of the
11  * software without specific, written prior permission.  M.I.T. makes
12  * no representations about the suitability of this software for any
13  * purpose.  It is provided "as is" without express or implied
14  * warranty.
15  *
16  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
17  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
18  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
19  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
20  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
25  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
26  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_mpath.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/sysctl.h>
39 #include <sys/socket.h>
40 #include <sys/mbuf.h>
41 
42 #include <net/if.h>
43 #include <net/if_var.h>
44 #include <net/route.h>
45 #include <net/route_var.h>
46 #include <net/route/nhop.h>
47 #include <net/route/shared.h>
48 #include <net/vnet.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_var.h>
52 #include <netinet/ip.h>
53 #include <netinet/ip_icmp.h>
54 #include <netinet/ip_var.h>
55 
56 extern int	in_inithead(void **head, int off, u_int fibnum);
57 #ifdef VIMAGE
58 extern int	in_detachhead(void **head, int off);
59 #endif
60 
61 static int
62 rib4_preadd(u_int fibnum, const struct sockaddr *addr, const struct sockaddr *mask,
63     struct nhop_object *nh)
64 {
65 	const struct sockaddr_in *addr4 = (const struct sockaddr_in *)addr;
66 	uint16_t nh_type;
67 	int rt_flags;
68 
69 	/* XXX: RTF_LOCAL && RTF_MULTICAST */
70 
71 	rt_flags = nhop_get_rtflags(nh);
72 
73 	if (rt_flags & RTF_HOST) {
74 
75 		/*
76 		 * Backward compatibility:
77 		 * if the destination is broadcast,
78 		 * mark route as broadcast.
79 		 * This behavior was useful when route cloning
80 		 * was in place, so there was an explicit cloned
81 		 * route for every broadcasted address.
82 		 * Currently (2020-04) there is no kernel machinery
83 		 * to do route cloning, though someone might explicitly
84 		 * add these routes to support some cases with active-active
85 		 * load balancing. Given that, retain this support.
86 		 */
87 		if (in_broadcast(addr4->sin_addr, nh->nh_ifp)) {
88 			rt_flags |= RTF_BROADCAST;
89 			nhop_set_rtflags(nh, rt_flags);
90 			nh->nh_flags |= NHF_BROADCAST;
91 		}
92 	}
93 
94 	/*
95 	 * Check route MTU:
96 	 * inherit interface MTU if not set or
97 	 * check if MTU is too large.
98 	 */
99 	if (nh->nh_mtu == 0) {
100 		nh->nh_mtu = nh->nh_ifp->if_mtu;
101 	} else if (nh->nh_mtu > nh->nh_ifp->if_mtu)
102 		nh->nh_mtu = nh->nh_ifp->if_mtu;
103 
104 	/* Ensure that default route nhop has special flag */
105 	const struct sockaddr_in *mask4 = (const struct sockaddr_in *)mask;
106 	if ((rt_flags & RTF_HOST) == 0 && mask4->sin_addr.s_addr == 0)
107 		nh->nh_flags |= NHF_DEFAULT;
108 
109 	/* Set nhop type to basic per-AF nhop */
110 	if (nhop_get_type(nh) == 0) {
111 		if (nh->nh_flags & NHF_GATEWAY)
112 			nh_type = NH_TYPE_IPV4_ETHER_NHOP;
113 		else
114 			nh_type = NH_TYPE_IPV4_ETHER_RSLV;
115 
116 		nhop_set_type(nh, nh_type);
117 	}
118 
119 	return (0);
120 }
121 
122 /*
123  * Do what we need to do when inserting a route.
124  */
125 static struct radix_node *
126 in_addroute(void *v_arg, void *n_arg, struct radix_head *head,
127     struct radix_node *treenodes)
128 {
129 	struct rtentry *rt = (struct rtentry *)treenodes;
130 	struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
131 
132 	/*
133 	 * A little bit of help for both IP output and input:
134 	 *   For host routes, we make sure that RTF_BROADCAST
135 	 *   is set for anything that looks like a broadcast address.
136 	 *   This way, we can avoid an expensive call to in_broadcast()
137 	 *   in ip_output() most of the time (because the route passed
138 	 *   to ip_output() is almost always a host route).
139 	 *
140 	 *   We also do the same for local addresses, with the thought
141 	 *   that this might one day be used to speed up ip_input().
142 	 *
143 	 * We also mark routes to multicast addresses as such, because
144 	 * it's easy to do and might be useful (but this is much more
145 	 * dubious since it's so easy to inspect the address).
146 	 */
147 	if (rt->rt_flags & RTF_HOST) {
148 		struct epoch_tracker et;
149 		bool bcast;
150 
151 		NET_EPOCH_ENTER(et);
152 		bcast = in_broadcast(sin->sin_addr, rt->rt_ifp);
153 		NET_EPOCH_EXIT(et);
154 		if (bcast)
155 			rt->rt_flags |= RTF_BROADCAST;
156 		else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
157 		    sin->sin_addr.s_addr)
158 			rt->rt_flags |= RTF_LOCAL;
159 	}
160 	if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
161 		rt->rt_flags |= RTF_MULTICAST;
162 
163 	if (rt->rt_ifp != NULL) {
164 
165 		/*
166 		 * Check route MTU:
167 		 * inherit interface MTU if not set or
168 		 * check if MTU is too large.
169 		 */
170 		if (rt->rt_mtu == 0) {
171 			rt->rt_mtu = rt->rt_ifp->if_mtu;
172 		} else if (rt->rt_mtu > rt->rt_ifp->if_mtu)
173 			rt->rt_mtu = rt->rt_ifp->if_mtu;
174 	}
175 
176 	return (rn_addroute(v_arg, n_arg, head, treenodes));
177 }
178 
179 static int _in_rt_was_here;
180 /*
181  * Initialize our routing tree.
182  */
183 int
184 in_inithead(void **head, int off, u_int fibnum)
185 {
186 	struct rib_head *rh;
187 
188 	rh = rt_table_init(32, AF_INET, fibnum);
189 	if (rh == NULL)
190 		return (0);
191 
192 	rh->rnh_preadd = rib4_preadd;
193 	rh->rnh_addaddr = in_addroute;
194 #ifdef	RADIX_MPATH
195 	rt_mpath_init_rnh(rh);
196 #endif
197 	*head = (void *)rh;
198 
199 	if (_in_rt_was_here == 0 ) {
200 		_in_rt_was_here = 1;
201 	}
202 	return 1;
203 }
204 
205 #ifdef VIMAGE
206 int
207 in_detachhead(void **head, int off)
208 {
209 
210 	rt_table_destroy((struct rib_head *)(*head));
211 	return (1);
212 }
213 #endif
214 
215 /*
216  * This zaps old routes when the interface goes down or interface
217  * address is deleted.  In the latter case, it deletes static routes
218  * that point to this address.  If we don't do this, we may end up
219  * using the old address in the future.  The ones we always want to
220  * get rid of are things like ARP entries, since the user might down
221  * the interface, walk over to a completely different network, and
222  * plug back in.
223  */
224 struct in_ifadown_arg {
225 	struct ifaddr *ifa;
226 	int del;
227 };
228 
229 static int
230 in_ifadownkill(const struct rtentry *rt, void *xap)
231 {
232 	struct in_ifadown_arg *ap = xap;
233 
234 	if (rt->rt_ifa != ap->ifa)
235 		return (0);
236 
237 	if ((rt->rt_flags & RTF_STATIC) != 0 && ap->del == 0)
238 		return (0);
239 
240 	return (1);
241 }
242 
243 void
244 in_ifadown(struct ifaddr *ifa, int delete)
245 {
246 	struct in_ifadown_arg arg;
247 
248 	KASSERT(ifa->ifa_addr->sa_family == AF_INET,
249 	    ("%s: wrong family", __func__));
250 
251 	arg.ifa = ifa;
252 	arg.del = delete;
253 
254 	rt_foreach_fib_walk_del(AF_INET, in_ifadownkill, &arg);
255 	ifa->ifa_flags &= ~IFA_ROUTE;		/* XXXlocking? */
256 }
257 
258 /*
259  * inet versions of rt functions. These have fib extensions and
260  * for now will just reference the _fib variants.
261  * eventually this order will be reversed,
262  */
263 void
264 in_rtalloc_ign(struct route *ro, u_long ignflags, u_int fibnum)
265 {
266 	rtalloc_ign_fib(ro, ignflags, fibnum);
267 }
268 
269