1139826Simp/*-
253541Sshin * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
353541Sshin * All rights reserved.
453541Sshin *
553541Sshin * Redistribution and use in source and binary forms, with or without
653541Sshin * modification, are permitted provided that the following conditions
753541Sshin * are met:
853541Sshin * 1. Redistributions of source code must retain the above copyright
953541Sshin *    notice, this list of conditions and the following disclaimer.
1053541Sshin * 2. Redistributions in binary form must reproduce the above copyright
1153541Sshin *    notice, this list of conditions and the following disclaimer in the
1253541Sshin *    documentation and/or other materials provided with the distribution.
1353541Sshin * 3. Neither the name of the project nor the names of its contributors
1453541Sshin *    may be used to endorse or promote products derived from this software
1553541Sshin *    without specific prior written permission.
1653541Sshin *
1753541Sshin * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
1853541Sshin * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
1953541Sshin * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
2053541Sshin * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
2153541Sshin * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
2253541Sshin * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
2353541Sshin * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
2453541Sshin * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
2553541Sshin * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
2653541Sshin * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
2753541Sshin * SUCH DAMAGE.
28174510Sobrien *
29174510Sobrien *	$KAME: in6_rmx.c,v 1.11 2001/07/26 06:53:16 jinmei Exp $
3053541Sshin */
3153541Sshin
32139826Simp/*-
3353541Sshin * Copyright 1994, 1995 Massachusetts Institute of Technology
3453541Sshin *
3553541Sshin * Permission to use, copy, modify, and distribute this software and
3653541Sshin * its documentation for any purpose and without fee is hereby
3753541Sshin * granted, provided that both the above copyright notice and this
3853541Sshin * permission notice appear in all copies, that both the above
3953541Sshin * copyright notice and this permission notice appear in all
4053541Sshin * supporting documentation, and that the name of M.I.T. not be used
4153541Sshin * in advertising or publicity pertaining to distribution of the
4253541Sshin * software without specific, written prior permission.  M.I.T. makes
4353541Sshin * no representations about the suitability of this software for any
4453541Sshin * purpose.  It is provided "as is" without express or implied
4553541Sshin * warranty.
4653541Sshin *
4753541Sshin * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
4853541Sshin * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
4953541Sshin * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
5053541Sshin * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
5153541Sshin * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
5253541Sshin * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
5353541Sshin * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
5453541Sshin * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
5553541Sshin * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
5653541Sshin * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
5753541Sshin * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
5853541Sshin * SUCH DAMAGE.
5953541Sshin *
6053541Sshin */
6153541Sshin
62174510Sobrien#include <sys/cdefs.h>
63174510Sobrien__FBSDID("$FreeBSD: releng/10.3/sys/netinet6/in6_rmx.c 295389 2016-02-08 00:07:01Z bz $");
64174510Sobrien
6553541Sshin#include <sys/param.h>
6653541Sshin#include <sys/systm.h>
6753541Sshin#include <sys/kernel.h>
68185751Simp#include <sys/lock.h>
6953541Sshin#include <sys/sysctl.h>
7053541Sshin#include <sys/queue.h>
7153541Sshin#include <sys/socket.h>
7253541Sshin#include <sys/socketvar.h>
7353541Sshin#include <sys/mbuf.h>
74185747Skmacy#include <sys/rwlock.h>
7553541Sshin#include <sys/syslog.h>
76120727Ssam#include <sys/callout.h>
7753541Sshin
7853541Sshin#include <net/if.h>
79194714Sbz#include <net/route.h>
80185571Sbz
8153541Sshin#include <netinet/in.h>
8253541Sshin#include <netinet/ip_var.h>
8353541Sshin#include <netinet/in_var.h>
8453541Sshin
8562587Sitojun#include <netinet/ip6.h>
8653541Sshin#include <netinet6/ip6_var.h>
8753541Sshin
8862587Sitojun#include <netinet/icmp6.h>
89121283Sume#include <netinet6/nd6.h>
9053541Sshin
9153541Sshin#include <netinet/tcp.h>
9253541Sshin#include <netinet/tcp_seq.h>
9353541Sshin#include <netinet/tcp_timer.h>
9453541Sshin#include <netinet/tcp_var.h>
9553541Sshin
96175162Sobrienextern int	in6_inithead(void **head, int off);
97193731Szec#ifdef VIMAGE
98193731Szecextern int	in6_detachhead(void **head, int off);
99193731Szec#endif
10053541Sshin
10153541Sshin/*
10253541Sshin * Do what we need to do when inserting a route.
10353541Sshin */
10453541Sshinstatic struct radix_node *
10553541Sshinin6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
106171260Sdelphij    struct radix_node *treenodes)
10753541Sshin{
10853541Sshin	struct rtentry *rt = (struct rtentry *)treenodes;
10953541Sshin	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt);
11053541Sshin	struct radix_node *ret;
11153541Sshin
112186119Sqingli	RADIX_NODE_HEAD_WLOCK_ASSERT(head);
11353541Sshin	if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
11453541Sshin		rt->rt_flags |= RTF_MULTICAST;
11553541Sshin
11653541Sshin	/*
11753541Sshin	 * A little bit of help for both IPv6 output and input:
11853541Sshin	 *   For local addresses, we make sure that RTF_LOCAL is set,
11953541Sshin	 *   with the thought that this might one day be used to speed up
12053541Sshin	 *   ip_input().
12153541Sshin	 *
12253541Sshin	 * We also mark routes to multicast addresses as such, because
12353541Sshin	 * it's easy to do and might be useful (but this is much more
12453541Sshin	 * dubious since it's so easy to inspect the address).  (This
12553541Sshin	 * is done above.)
12653541Sshin	 *
12753541Sshin	 * XXX
12853541Sshin	 * should elaborate the code.
12953541Sshin	 */
13053541Sshin	if (rt->rt_flags & RTF_HOST) {
13153541Sshin		if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr)
13253541Sshin					->sin6_addr,
13353541Sshin				       &sin6->sin6_addr)) {
13453541Sshin			rt->rt_flags |= RTF_LOCAL;
13553541Sshin		}
13653541Sshin	}
13753541Sshin
138263478Sglebius	if (!rt->rt_mtu && rt->rt_ifp)
139263478Sglebius		rt->rt_mtu = IN6_LINKMTU(rt->rt_ifp);
14053541Sshin
14153541Sshin	ret = rn_addroute(v_arg, n_arg, head, treenodes);
142186119Sqingli	if (ret == NULL) {
14353541Sshin		struct rtentry *rt2;
14453541Sshin		/*
14553541Sshin		 * We are trying to add a net route, but can't.
14653541Sshin		 * The following case should be allowed, so we'll make a
14753541Sshin		 * special check for this:
14853541Sshin		 *	Two IPv6 addresses with the same prefix is assigned
14953541Sshin		 *	to a single interrface.
15053541Sshin		 *	# ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1)
15153541Sshin		 *	# ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2)
15253541Sshin		 *	In this case, (*1) and (*2) want to add the same
15353541Sshin		 *	net route entry, 3ffe:0501:: -> if0.
15453541Sshin		 *	This case should not raise an error.
15553541Sshin		 */
156231852Sbz		rt2 = in6_rtalloc1((struct sockaddr *)sin6, 0, RTF_RNH_LOCKED,
157231852Sbz		    rt->rt_fibnum);
15853541Sshin		if (rt2) {
159186119Sqingli			if (((rt2->rt_flags & (RTF_HOST|RTF_GATEWAY)) == 0)
16053541Sshin			 && rt2->rt_gateway
16153541Sshin			 && rt2->rt_gateway->sa_family == AF_LINK
16253541Sshin			 && rt2->rt_ifp == rt->rt_ifp) {
16353541Sshin				ret = rt2->rt_nodes;
16453541Sshin			}
165120727Ssam			RTFREE_LOCKED(rt2);
16653541Sshin		}
16753541Sshin	}
168186119Sqingli	return (ret);
16953541Sshin}
17053541Sshin
17162604SitojunSYSCTL_DECL(_net_inet6_ip6);
17262604Sitojun
173215701Sdimstatic VNET_DEFINE(int, rtq_toomany6) = 128;
174207369Sbz	/* 128 cached routes is ``too many'' */
175207369Sbz#define	V_rtq_toomany6			VNET(rtq_toomany6)
176195699SrwatsonSYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
177195699Srwatson    &VNET_NAME(rtq_toomany6) , 0, "");
178120913Sume
17953541Sshinstruct rtqk_arg {
18053541Sshin	struct radix_node_head *rnh;
18153541Sshin	int mode;
18253541Sshin	int updating;
18353541Sshin	int draining;
18453541Sshin	int killed;
18553541Sshin	int found;
18653541Sshin	time_t nextstop;
18753541Sshin};
18853541Sshin
18953541Sshin/*
19053541Sshin * Age old PMTUs.
19153541Sshin */
19253541Sshinstruct mtuex_arg {
19353541Sshin	struct radix_node_head *rnh;
19453541Sshin	time_t nextstop;
19553541Sshin};
196215701Sdimstatic VNET_DEFINE(struct callout, rtq_mtutimer);
197195727Srwatson#define	V_rtq_mtutimer			VNET(rtq_mtutimer)
198195699Srwatson
19953541Sshinstatic int
20053541Sshinin6_mtuexpire(struct radix_node *rn, void *rock)
20153541Sshin{
20253541Sshin	struct rtentry *rt = (struct rtentry *)rn;
20353541Sshin	struct mtuex_arg *ap = rock;
20453541Sshin
20553541Sshin	/* sanity */
20653541Sshin	if (!rt)
20753541Sshin		panic("rt == NULL in in6_mtuexpire");
20853541Sshin
209263478Sglebius	if (rt->rt_expire && !(rt->rt_flags & RTF_PROBEMTU)) {
210263478Sglebius		if (rt->rt_expire <= time_uptime) {
21153541Sshin			rt->rt_flags |= RTF_PROBEMTU;
21253541Sshin		} else {
213263478Sglebius			ap->nextstop = lmin(ap->nextstop, rt->rt_expire);
21453541Sshin		}
21553541Sshin	}
21653541Sshin
21753541Sshin	return 0;
21853541Sshin}
21953541Sshin
22053541Sshin#define	MTUTIMO_DEFAULT	(60*1)
22153541Sshin
22253541Sshinstatic void
223231852Sbzin6_mtutimo_one(struct radix_node_head *rnh)
22453541Sshin{
22553541Sshin	struct mtuex_arg arg;
22653541Sshin
22753541Sshin	arg.rnh = rnh;
228160123Soleg	arg.nextstop = time_uptime + MTUTIMO_DEFAULT;
229108250Shsu	RADIX_NODE_HEAD_LOCK(rnh);
23053541Sshin	rnh->rnh_walktree(rnh, in6_mtuexpire, &arg);
231108250Shsu	RADIX_NODE_HEAD_UNLOCK(rnh);
232231852Sbz}
23353541Sshin
234231852Sbzstatic void
235231852Sbzin6_mtutimo(void *rock)
236231852Sbz{
237231852Sbz	CURVNET_SET_QUIET((struct vnet *) rock);
238231852Sbz	struct radix_node_head *rnh;
239231852Sbz	struct timeval atv;
240231852Sbz	u_int fibnum;
241231852Sbz
242231852Sbz	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
243231852Sbz		rnh = rt_tables_get_rnh(fibnum, AF_INET6);
244231852Sbz		if (rnh != NULL)
245231852Sbz			in6_mtutimo_one(rnh);
246231852Sbz	}
247231852Sbz
248231852Sbz	atv.tv_sec = MTUTIMO_DEFAULT;
24953541Sshin	atv.tv_usec = 0;
250181803Sbz	callout_reset(&V_rtq_mtutimer, tvtohz(&atv), in6_mtutimo, rock);
251183550Szec	CURVNET_RESTORE();
25253541Sshin}
25353541Sshin
25453541Sshin/*
25553541Sshin * Initialize our routing tree.
256178888Sjulian * XXX MRT When off == 0, we are being called from vfs_export.c
257178888Sjulian * so just set up their table and leave. (we know what the correct
258178888Sjulian * value should be so just use that).. FIX AFTER RELENG_7 is MFC'd
259178888Sjulian * see also comments in in_inithead() vfs_export.c and domain.h
26053541Sshin */
261231852Sbzstatic VNET_DEFINE(int, _in6_rt_was_here);
262231852Sbz#define	V__in6_rt_was_here	VNET(_in6_rt_was_here)
263231852Sbz
26453541Sshinint
26553541Sshinin6_inithead(void **head, int off)
26653541Sshin{
26753541Sshin	struct radix_node_head *rnh;
26853541Sshin
269178888Sjulian	if (!rn_inithead(head, offsetof(struct sockaddr_in6, sin6_addr) << 3))
270178888Sjulian		return 0;		/* See above */
27153541Sshin
272178888Sjulian	if (off == 0)		/* See above */
273178888Sjulian		return 1;	/* only do the rest for the real thing */
27453541Sshin
27553541Sshin	rnh = *head;
27653541Sshin	rnh->rnh_addaddr = in6_addroute;
277231852Sbz
278231852Sbz	if (V__in6_rt_was_here == 0) {
279231852Sbz		callout_init(&V_rtq_mtutimer, CALLOUT_MPSAFE);
280231852Sbz		in6_mtutimo(curvnet);	/* kick off timeout first time */
281231852Sbz		V__in6_rt_was_here = 1;
282231852Sbz	}
283231852Sbz
28453541Sshin	return 1;
28553541Sshin}
286193731Szec
287193731Szec#ifdef VIMAGE
288193731Szecint
289193731Szecin6_detachhead(void **head, int off)
290193731Szec{
291193731Szec
292193731Szec	callout_drain(&V_rtq_mtutimer);
293295389Sbz	return (rn_detachhead(head));
294193731Szec}
295193731Szec#endif
296231852Sbz
297231852Sbz/*
298231852Sbz * Extended API for IPv6 FIB support.
299231852Sbz */
300231852Sbzvoid
301231852Sbzin6_rtredirect(struct sockaddr *dst, struct sockaddr *gw, struct sockaddr *nm,
302231852Sbz    int flags, struct sockaddr *src, u_int fibnum)
303231852Sbz{
304231852Sbz
305231852Sbz	rtredirect_fib(dst, gw, nm, flags, src, fibnum);
306231852Sbz}
307231852Sbz
308231852Sbzint
309231852Sbzin6_rtrequest(int req, struct sockaddr *dst, struct sockaddr *gw,
310231852Sbz    struct sockaddr *mask, int flags, struct rtentry **ret_nrt, u_int fibnum)
311231852Sbz{
312231852Sbz
313231852Sbz	return (rtrequest_fib(req, dst, gw, mask, flags, ret_nrt, fibnum));
314231852Sbz}
315231852Sbz
316231852Sbzvoid
317231852Sbzin6_rtalloc(struct route_in6 *ro, u_int fibnum)
318231852Sbz{
319231852Sbz
320231852Sbz	rtalloc_ign_fib((struct route *)ro, 0ul, fibnum);
321231852Sbz}
322231852Sbz
323231852Sbzvoid
324231852Sbzin6_rtalloc_ign(struct route_in6 *ro, u_long ignflags, u_int fibnum)
325231852Sbz{
326231852Sbz
327231852Sbz	rtalloc_ign_fib((struct route *)ro, ignflags, fibnum);
328231852Sbz}
329231852Sbz
330231852Sbzstruct rtentry *
331231852Sbzin6_rtalloc1(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum)
332231852Sbz{
333231852Sbz
334231852Sbz	return (rtalloc1_fib(dst, report, ignflags, fibnum));
335231852Sbz}
336