Home | History | Annotate | Line # | Download | only in netinet6
      1 /*	$NetBSD: ip6_flow.c,v 1.43 2024/06/29 13:00:44 riastradh Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2007 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by the 3am Software Foundry ("3am").  It was developed by Liam J. Foy
      9  * <liamjfoy (at) netbsd.org> and Matt Thomas <matt (at) netbsd.org>.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  *
     20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     30  * POSSIBILITY OF SUCH DAMAGE.
     31  *
     32  * IPv6 version was developed by Liam J. Foy. Original source existed in IPv4
     33  * format developed by Matt Thomas. Thanks to Joerg Sonnenberger, Matt
     34  * Thomas and Christos Zoulas.
     35  *
     36  * Thanks to Liverpool John Moores University, especially Dr. David Llewellyn-Jones
     37  * for providing resources (to test) and Professor Madjid Merabti.
     38  */
     39 
     40 #include <sys/cdefs.h>
     41 __KERNEL_RCSID(0, "$NetBSD: ip6_flow.c,v 1.43 2024/06/29 13:00:44 riastradh Exp $");
     42 
     43 #ifdef _KERNEL_OPT
     44 #include "opt_net_mpsafe.h"
     45 #endif
     46 
     47 #include <sys/param.h>
     48 #include <sys/systm.h>
     49 #include <sys/malloc.h>
     50 #include <sys/mbuf.h>
     51 #include <sys/socketvar.h>
     52 #include <sys/time.h>
     53 #include <sys/kernel.h>
     54 #include <sys/pool.h>
     55 #include <sys/sysctl.h>
     56 #include <sys/workqueue.h>
     57 #include <sys/atomic.h>
     58 
     59 #include <net/if.h>
     60 #include <net/if_dl.h>
     61 #include <net/route.h>
     62 #include <net/pfil.h>
     63 
     64 #include <netinet/in.h>
     65 #include <netinet6/in6_var.h>
     66 #include <netinet/in_systm.h>
     67 #include <netinet/ip6.h>
     68 #include <netinet6/ip6_var.h>
     69 #include <netinet6/ip6_private.h>
     70 
     71 /*
     72  * IPv6 Fast Forward caches/hashes flows from one source to destination.
     73  *
     74  * Upon a successful forward IPv6FF caches and hashes details such as the
     75  * route, source and destination. Once another packet is received matching
     76  * the source and destination the packet is forwarded straight onto if_output
     77  * using the cached details.
     78  *
     79  * Example:
     80  * ether/fddi_input -> ip6flow_fastforward -> if_output
     81  */
     82 
     83 static struct pool ip6flow_pool;
     84 
     85 TAILQ_HEAD(ip6flowhead, ip6flow);
     86 
     87 /*
     88  * We could use IPv4 defines (IPFLOW_HASHBITS) but we'll
     89  * use our own (possibly for future expansion).
     90  */
     91 #define	IP6FLOW_TIMER		(5 * PR_SLOWHZ)
     92 #define	IP6FLOW_DEFAULT_HASHSIZE	(1 << IP6FLOW_HASHBITS)
     93 
     94 /*
     95  * ip6_flow.c internal lock.
     96  * If we use softnet_lock, it would cause recursive lock.
     97  *
     98  * This is a tentative workaround.
     99  * We should make it scalable somehow in the future.
    100  */
    101 static kmutex_t ip6flow_lock __cacheline_aligned;
    102 static struct ip6flowhead *ip6flowtable = NULL;
    103 static struct ip6flowhead ip6flowlist;
    104 static int ip6flow_inuse __cacheline_aligned;
    105 
    106 static void ip6flow_slowtimo_work(struct work *, void *);
    107 static struct workqueue	*ip6flow_slowtimo_wq;
    108 static struct work	ip6flow_slowtimo_wk;
    109 
    110 static int sysctl_net_inet6_ip6_hashsize(SYSCTLFN_PROTO);
    111 static int sysctl_net_inet6_ip6_maxflows(SYSCTLFN_PROTO);
    112 static void ip6flow_sysctl_init(struct sysctllog **);
    113 
    114 /*
    115  * Insert an ip6flow into the list.
    116  */
    117 #define	IP6FLOW_INSERT(hashidx, ip6f) \
    118 do { \
    119 	(ip6f)->ip6f_hashidx = (hashidx); \
    120 	TAILQ_INSERT_HEAD(&ip6flowtable[(hashidx)], (ip6f), ip6f_hash); \
    121 	TAILQ_INSERT_HEAD(&ip6flowlist, (ip6f), ip6f_list); \
    122 } while (/*CONSTCOND*/ 0)
    123 
    124 /*
    125  * Remove an ip6flow from the list.
    126  */
    127 #define	IP6FLOW_REMOVE(hashidx, ip6f) \
    128 do { \
    129 	TAILQ_REMOVE(&ip6flowtable[(hashidx)], (ip6f), ip6f_hash); \
    130 	TAILQ_REMOVE(&ip6flowlist, (ip6f), ip6f_list); \
    131 } while (/*CONSTCOND*/ 0)
    132 
    133 #ifndef IP6FLOW_DEFAULT
    134 #define	IP6FLOW_DEFAULT		256
    135 #endif
    136 
    137 int ip6_maxflows = IP6FLOW_DEFAULT;
    138 int ip6_hashsize = IP6FLOW_DEFAULT_HASHSIZE;
    139 
    140 /*
    141  * Calculate hash table position.
    142  */
    143 static size_t
    144 ip6flow_hash(const struct ip6_hdr *ip6)
    145 {
    146 	size_t hash;
    147 	uint32_t dst_sum, src_sum;
    148 	size_t idx;
    149 
    150 	src_sum = ip6->ip6_src.s6_addr32[0] + ip6->ip6_src.s6_addr32[1]
    151 	    + ip6->ip6_src.s6_addr32[2] + ip6->ip6_src.s6_addr32[3];
    152 	dst_sum = ip6->ip6_dst.s6_addr32[0] + ip6->ip6_dst.s6_addr32[1]
    153 	    + ip6->ip6_dst.s6_addr32[2] + ip6->ip6_dst.s6_addr32[3];
    154 
    155 	hash = ip6->ip6_flow;
    156 
    157 	for (idx = 0; idx < 32; idx += IP6FLOW_HASHBITS)
    158 		hash += (dst_sum >> (32 - idx)) + (src_sum >> idx);
    159 
    160 	return hash & (ip6_hashsize-1);
    161 }
    162 
    163 /*
    164  * Check to see if a flow already exists - if so return it.
    165  */
    166 static struct ip6flow *
    167 ip6flow_lookup(const struct ip6_hdr *ip6)
    168 {
    169 	size_t hash;
    170 	struct ip6flow *ip6f;
    171 
    172 	KASSERT(mutex_owned(&ip6flow_lock));
    173 
    174 	hash = ip6flow_hash(ip6);
    175 
    176 	TAILQ_FOREACH(ip6f, &ip6flowtable[hash], ip6f_hash) {
    177 		if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6f->ip6f_dst)
    178 		    && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6f->ip6f_src)
    179 		    && ip6f->ip6f_flow == ip6->ip6_flow) {
    180 			/* A cached flow has been found. */
    181 			return ip6f;
    182 		}
    183 	}
    184 
    185 	return NULL;
    186 }
    187 
    188 void
    189 ip6flow_poolinit(void)
    190 {
    191 
    192 	pool_init(&ip6flow_pool, sizeof(struct ip6flow), 0, 0, 0, "ip6flowpl",
    193 			NULL, IPL_NET);
    194 }
    195 
    196 /*
    197  * Allocate memory and initialise lists. This function is called
    198  * from ip6_init and called there after to resize the hash table.
    199  * If a newly sized table cannot be malloc'ed we just continue
    200  * to use the old one.
    201  */
    202 static int
    203 ip6flow_init_locked(int table_size)
    204 {
    205 	struct ip6flowhead *new_table;
    206 	size_t i;
    207 
    208 	KASSERT(mutex_owned(&ip6flow_lock));
    209 
    210 	new_table = (struct ip6flowhead *)malloc(sizeof(struct ip6flowhead) *
    211 	    table_size, M_RTABLE, M_NOWAIT);
    212 
    213 	if (new_table == NULL)
    214 		return 1;
    215 
    216 	if (ip6flowtable != NULL)
    217 		free(ip6flowtable, M_RTABLE);
    218 
    219 	ip6flowtable = new_table;
    220 	ip6_hashsize = table_size;
    221 
    222 	TAILQ_INIT(&ip6flowlist);
    223 	for (i = 0; i < ip6_hashsize; i++)
    224 		TAILQ_INIT(&ip6flowtable[i]);
    225 
    226 	return 0;
    227 }
    228 
    229 int
    230 ip6flow_init(int table_size)
    231 {
    232 	int ret, error;
    233 
    234 	error = workqueue_create(&ip6flow_slowtimo_wq, "ip6flow",
    235 	    ip6flow_slowtimo_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
    236 	if (error != 0)
    237 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
    238 
    239 	mutex_init(&ip6flow_lock, MUTEX_DEFAULT, IPL_NONE);
    240 
    241 	mutex_enter(&ip6flow_lock);
    242 	ret = ip6flow_init_locked(table_size);
    243 	mutex_exit(&ip6flow_lock);
    244 	ip6flow_sysctl_init(NULL);
    245 
    246 	return ret;
    247 }
    248 
    249 /*
    250  * IPv6 Fast Forward routine. Attempt to forward the packet -
    251  * if any problems are found return to the main IPv6 input
    252  * routine to deal with.
    253  */
    254 int
    255 ip6flow_fastforward(struct mbuf **mp)
    256 {
    257 	struct ip6flow *ip6f;
    258 	struct ip6_hdr *ip6;
    259 	struct rtentry *rt = NULL;
    260 	struct mbuf *m;
    261 	const struct sockaddr *dst;
    262 	int error;
    263 	int ret = 0;
    264 
    265 	mutex_enter(&ip6flow_lock);
    266 
    267 	/*
    268 	 * Are we forwarding packets and have flows?
    269 	 */
    270 	if (!ip6_forwarding || ip6flow_inuse == 0)
    271 		goto out;
    272 
    273 	m = *mp;
    274 	/*
    275 	 * At least size of IPv6 Header?
    276 	 */
    277 	if (m->m_len < sizeof(struct ip6_hdr))
    278 		goto out;
    279 	/*
    280 	 * Was packet received as a link-level multicast or broadcast?
    281 	 * If so, don't try to fast forward.
    282 	 */
    283 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
    284 		goto out;
    285 
    286 	if (ACCESSIBLE_POINTER(mtod(m, const void *), struct ip6_hdr) == 0) {
    287 		if ((m = m_copyup(m, sizeof(struct ip6_hdr),
    288 		    (max_linkhdr + 3) & ~3)) == NULL) {
    289 			ret = 1;
    290 			goto out;
    291 		}
    292 		*mp = m;
    293 	}
    294 
    295 	ip6 = mtod(m, struct ip6_hdr *);
    296 
    297 	if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) {
    298 		/* Bad version. */
    299 		goto out;
    300 	}
    301 
    302 	/*
    303 	 * If we have a hop-by-hop extension we must process it.
    304 	 * We just leave this up to ip6_input to deal with.
    305 	 */
    306 	if (ip6->ip6_nxt == IPPROTO_HOPOPTS)
    307 		goto out;
    308 
    309 	/*
    310 	 * Attempt to find a flow.
    311 	 */
    312 	if ((ip6f = ip6flow_lookup(ip6)) == NULL) {
    313 		/* No flow found. */
    314 		goto out;
    315 	}
    316 
    317 	/*
    318 	 * Route and interface still up?
    319 	 */
    320 	if ((rt = rtcache_validate(&ip6f->ip6f_ro)) == NULL ||
    321 	    (rt->rt_ifp->if_flags & IFF_UP) == 0 ||
    322 	    (rt->rt_flags & RTF_BLACKHOLE) != 0)
    323 		goto out_unref;
    324 
    325 	/*
    326 	 * Packet size greater than MTU?
    327 	 */
    328 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu) {
    329 		/* Return to main IPv6 input function. */
    330 		goto out_unref;
    331 	}
    332 
    333 	/*
    334 	 * Clear any in-bound checksum flags for this packet.
    335 	 */
    336 	m->m_pkthdr.csum_flags = 0;
    337 
    338 	if (ip6->ip6_hlim <= IPV6_HLIMDEC)
    339 		goto out_unref;
    340 
    341 	/* Decrement hop limit (same as TTL) */
    342 	ip6->ip6_hlim -= IPV6_HLIMDEC;
    343 
    344 	if (rt->rt_flags & RTF_GATEWAY)
    345 		dst = rt->rt_gateway;
    346 	else
    347 		dst = rtcache_getdst(&ip6f->ip6f_ro);
    348 
    349 	PRT_SLOW_ARM(ip6f->ip6f_timer, IP6FLOW_TIMER);
    350 
    351 	ip6f->ip6f_uses++;
    352 
    353 #if 0
    354 	/*
    355 	 * We use FIFO cache replacement instead of LRU the same ip_flow.c.
    356 	 */
    357 	/* move to head (LRU) for ip6flowlist. ip6flowtable does not care LRU. */
    358 	TAILQ_REMOVE(&ip6flowlist, ip6f, ip6f_list);
    359 	TAILQ_INSERT_HEAD(&ip6flowlist, ip6f, ip6f_list);
    360 #endif
    361 
    362 	/* Send on its way - straight to the interface output routine. */
    363 	if ((error = if_output_lock(rt->rt_ifp, rt->rt_ifp, m, dst, rt)) != 0) {
    364 		ip6f->ip6f_dropped++;
    365 	} else {
    366 		ip6f->ip6f_forwarded++;
    367 	}
    368 	ret = 1;
    369 out_unref:
    370 	rtcache_unref(rt, &ip6f->ip6f_ro);
    371 out:
    372 	mutex_exit(&ip6flow_lock);
    373 	return ret;
    374 }
    375 
    376 /*
    377  * Add the IPv6 flow statistics to the main IPv6 statistics.
    378  */
    379 static void
    380 ip6flow_addstats_rt(struct rtentry *rt, struct ip6flow *ip6f)
    381 {
    382 	net_stat_ref_t ip6s;
    383 
    384 	if (rt != NULL)
    385 		rt->rt_use += ip6f->ip6f_uses;
    386 	ip6s = IP6_STAT_GETREF();
    387 	ip6s->nsr_stats[IP6_STAT_FASTFORWARDFLOWS-1] = ip6flow_inuse; /* XXX */
    388 	_NET_STATADD_REF(ip6s, IP6_STAT_CANTFORWARD, ip6f->ip6f_dropped);
    389 	_NET_STATADD_REF(ip6s, IP6_STAT_ODROPPED, ip6f->ip6f_dropped);
    390 	_NET_STATADD_REF(ip6s, IP6_STAT_TOTAL, ip6f->ip6f_uses);
    391 	_NET_STATADD_REF(ip6s, IP6_STAT_FORWARD, ip6f->ip6f_forwarded);
    392 	_NET_STATADD_REF(ip6s, IP6_STAT_FASTFORWARD, ip6f->ip6f_forwarded);
    393 	IP6_STAT_PUTREF();
    394 }
    395 
    396 static void
    397 ip6flow_addstats(struct ip6flow *ip6f)
    398 {
    399 	struct rtentry *rt;
    400 
    401 	rt = rtcache_validate(&ip6f->ip6f_ro);
    402 	ip6flow_addstats_rt(rt, ip6f);
    403 	rtcache_unref(rt, &ip6f->ip6f_ro);
    404 }
    405 
    406 /*
    407  * Add statistics and free the flow.
    408  */
    409 static void
    410 ip6flow_free(struct ip6flow *ip6f)
    411 {
    412 
    413 	KASSERT(mutex_owned(&ip6flow_lock));
    414 
    415 	/*
    416 	 * Remove the flow from the hash table (at elevated IPL).
    417 	 * Once it's off the list, we can deal with it at normal
    418 	 * network IPL.
    419 	 */
    420 	IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    421 
    422 	ip6flow_inuse--;
    423 	ip6flow_addstats(ip6f);
    424 	rtcache_free(&ip6f->ip6f_ro);
    425 	pool_put(&ip6flow_pool, ip6f);
    426 }
    427 
    428 static struct ip6flow *
    429 ip6flow_reap_locked(int just_one)
    430 {
    431 	struct ip6flow *ip6f;
    432 
    433 	KASSERT(mutex_owned(&ip6flow_lock));
    434 
    435 	/*
    436 	 * This case must remove one ip6flow. Furthermore, this case is used in
    437 	 * fast path(packet processing path). So, simply remove TAILQ_LAST one.
    438 	 */
    439 	if (just_one) {
    440 		ip6f = TAILQ_LAST(&ip6flowlist, ip6flowhead);
    441 		KASSERT(ip6f != NULL);
    442 
    443 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    444 
    445 		ip6flow_addstats(ip6f);
    446 		rtcache_free(&ip6f->ip6f_ro);
    447 		return ip6f;
    448 	}
    449 
    450 	/*
    451 	 * This case is used in slow path(sysctl).
    452 	 * At first, remove invalid rtcache ip6flow, and then remove TAILQ_LAST
    453 	 * ip6flow if it is ensured least recently used by comparing last_uses.
    454 	 */
    455 	while (ip6flow_inuse > ip6_maxflows) {
    456 		struct ip6flow *maybe_ip6f = TAILQ_LAST(&ip6flowlist, ip6flowhead);
    457 
    458 		TAILQ_FOREACH(ip6f, &ip6flowlist, ip6f_list) {
    459 			struct rtentry *rt;
    460 			/*
    461 			 * If this no longer points to a valid route -
    462 			 * reclaim it.
    463 			 */
    464 			if ((rt = rtcache_validate(&ip6f->ip6f_ro)) == NULL)
    465 				goto done;
    466 			rtcache_unref(rt, &ip6f->ip6f_ro);
    467 			/*
    468 			 * choose the one that's been least recently
    469 			 * used or has had the least uses in the
    470 			 * last 1.5 intervals.
    471 			 */
    472 			if (ip6f->ip6f_timer < maybe_ip6f->ip6f_timer
    473 			    || ((ip6f->ip6f_timer == maybe_ip6f->ip6f_timer)
    474 				&& (ip6f->ip6f_last_uses + ip6f->ip6f_uses
    475 				    < maybe_ip6f->ip6f_last_uses + maybe_ip6f->ip6f_uses)))
    476 				maybe_ip6f = ip6f;
    477 		}
    478 		ip6f = maybe_ip6f;
    479 	    done:
    480 		/*
    481 		 * Remove the entry from the flow table
    482 		 */
    483 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    484 
    485 		rtcache_free(&ip6f->ip6f_ro);
    486 		ip6flow_inuse--;
    487 		ip6flow_addstats(ip6f);
    488 		pool_put(&ip6flow_pool, ip6f);
    489 	}
    490 	return NULL;
    491 }
    492 
    493 /*
    494  * Reap one or more flows - ip6flow_reap may remove
    495  * multiple flows if net.inet6.ip6.maxflows is reduced.
    496  */
    497 struct ip6flow *
    498 ip6flow_reap(int just_one)
    499 {
    500 	struct ip6flow *ip6f;
    501 
    502 	mutex_enter(&ip6flow_lock);
    503 	ip6f = ip6flow_reap_locked(just_one);
    504 	mutex_exit(&ip6flow_lock);
    505 	return ip6f;
    506 }
    507 
    508 static unsigned int ip6flow_work_enqueued = 0;
    509 
    510 void
    511 ip6flow_slowtimo_work(struct work *wk, void *arg)
    512 {
    513 	struct ip6flow *ip6f, *next_ip6f;
    514 
    515 	/* We can allow enqueuing another work at this point */
    516 	atomic_swap_uint(&ip6flow_work_enqueued, 0);
    517 
    518 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    519 	mutex_enter(&ip6flow_lock);
    520 
    521 	for (ip6f = TAILQ_FIRST(&ip6flowlist); ip6f != NULL; ip6f = next_ip6f) {
    522 		struct rtentry *rt = NULL;
    523 		next_ip6f = TAILQ_NEXT(ip6f, ip6f_list);
    524 		if (PRT_SLOW_ISEXPIRED(ip6f->ip6f_timer) ||
    525 		    (rt = rtcache_validate(&ip6f->ip6f_ro)) == NULL) {
    526 			ip6flow_free(ip6f);
    527 		} else {
    528 			ip6f->ip6f_last_uses = ip6f->ip6f_uses;
    529 			ip6flow_addstats_rt(rt, ip6f);
    530 			ip6f->ip6f_uses = 0;
    531 			ip6f->ip6f_dropped = 0;
    532 			ip6f->ip6f_forwarded = 0;
    533 		}
    534 		rtcache_unref(rt, &ip6f->ip6f_ro);
    535 	}
    536 
    537 	mutex_exit(&ip6flow_lock);
    538 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    539 }
    540 
    541 void
    542 ip6flow_slowtimo(void)
    543 {
    544 
    545 	/* Avoid enqueuing another work when one is already enqueued */
    546 	if (atomic_swap_uint(&ip6flow_work_enqueued, 1) == 1)
    547 		return;
    548 
    549 	workqueue_enqueue(ip6flow_slowtimo_wq, &ip6flow_slowtimo_wk, NULL);
    550 }
    551 
    552 /*
    553  * We have successfully forwarded a packet using the normal
    554  * IPv6 stack. Now create/update a flow.
    555  */
    556 void
    557 ip6flow_create(struct route *ro, struct mbuf *m)
    558 {
    559 	const struct ip6_hdr *ip6;
    560 	struct ip6flow *ip6f;
    561 	size_t hash;
    562 
    563 	ip6 = mtod(m, const struct ip6_hdr *);
    564 
    565 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
    566 	mutex_enter(&ip6flow_lock);
    567 
    568 	/*
    569 	 * If IPv6 Fast Forward is disabled, don't create a flow.
    570 	 * It can be disabled by setting net.inet6.ip6.maxflows to 0.
    571 	 *
    572 	 * Don't create a flow for ICMPv6 messages.
    573 	 */
    574 	if (ip6_maxflows == 0 || ip6->ip6_nxt == IPPROTO_IPV6_ICMP)
    575 		goto out;
    576 
    577 	/*
    578 	 * See if an existing flow exists.  If so:
    579 	 *	- Remove the flow
    580 	 *	- Add flow statistics
    581 	 *	- Free the route
    582 	 *	- Reset statistics
    583 	 *
    584 	 * If a flow doesn't exist allocate a new one if
    585 	 * ip6_maxflows hasn't reached its limit. If it has
    586 	 * been reached, reap some flows.
    587 	 */
    588 	ip6f = ip6flow_lookup(ip6);
    589 	if (ip6f == NULL) {
    590 		if (ip6flow_inuse >= ip6_maxflows) {
    591 			ip6f = ip6flow_reap_locked(1);
    592 		} else {
    593 			ip6f = pool_get(&ip6flow_pool, PR_NOWAIT);
    594 			if (ip6f == NULL)
    595 				goto out;
    596 			ip6flow_inuse++;
    597 		}
    598 		memset(ip6f, 0, sizeof(*ip6f));
    599 	} else {
    600 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    601 
    602 		ip6flow_addstats(ip6f);
    603 		rtcache_free(&ip6f->ip6f_ro);
    604 		ip6f->ip6f_uses = 0;
    605 		ip6f->ip6f_last_uses = 0;
    606 		ip6f->ip6f_dropped = 0;
    607 		ip6f->ip6f_forwarded = 0;
    608 	}
    609 
    610 	/*
    611 	 * Fill in the updated/new details.
    612 	 */
    613 	rtcache_copy(&ip6f->ip6f_ro, ro);
    614 	ip6f->ip6f_dst = ip6->ip6_dst;
    615 	ip6f->ip6f_src = ip6->ip6_src;
    616 	ip6f->ip6f_flow = ip6->ip6_flow;
    617 	PRT_SLOW_ARM(ip6f->ip6f_timer, IP6FLOW_TIMER);
    618 
    619 	/*
    620 	 * Insert into the appropriate bucket of the flow table.
    621 	 */
    622 	hash = ip6flow_hash(ip6);
    623 	IP6FLOW_INSERT(hash, ip6f);
    624 
    625  out:
    626 	mutex_exit(&ip6flow_lock);
    627 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    628 }
    629 
    630 /*
    631  * Invalidate/remove all flows - if new_size is positive we
    632  * resize the hash table.
    633  */
    634 int
    635 ip6flow_invalidate_all(int new_size)
    636 {
    637 	struct ip6flow *ip6f, *next_ip6f;
    638 	int error;
    639 
    640 	error = 0;
    641 
    642 	mutex_enter(&ip6flow_lock);
    643 
    644 	for (ip6f = TAILQ_FIRST(&ip6flowlist); ip6f != NULL; ip6f = next_ip6f) {
    645 		next_ip6f = TAILQ_NEXT(ip6f, ip6f_list);
    646 		ip6flow_free(ip6f);
    647 	}
    648 
    649 	if (new_size)
    650 		error = ip6flow_init_locked(new_size);
    651 
    652 	mutex_exit(&ip6flow_lock);
    653 
    654 	return error;
    655 }
    656 
    657 /*
    658  * sysctl helper routine for net.inet.ip6.maxflows. Since
    659  * we could reduce this value, call ip6flow_reap();
    660  */
    661 static int
    662 sysctl_net_inet6_ip6_maxflows(SYSCTLFN_ARGS)
    663 {
    664 	int error;
    665 
    666 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
    667 	if (error || newp == NULL)
    668 		return (error);
    669 
    670 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    671 
    672 	ip6flow_reap(0);
    673 
    674 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    675 
    676 	return (0);
    677 }
    678 
    679 static int
    680 sysctl_net_inet6_ip6_hashsize(SYSCTLFN_ARGS)
    681 {
    682 	int error, tmp;
    683 	struct sysctlnode node;
    684 
    685 	node = *rnode;
    686 	tmp = ip6_hashsize;
    687 	node.sysctl_data = &tmp;
    688 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
    689 	if (error || newp == NULL)
    690 		return (error);
    691 
    692 	if ((tmp & (tmp - 1)) == 0 && tmp != 0) {
    693 		/*
    694 		 * Can only fail due to malloc()
    695 		 */
    696 		SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    697 		error = ip6flow_invalidate_all(tmp);
    698 		SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    699 	} else {
    700 		/*
    701 		 * EINVAL if not a power of 2
    702 		 */
    703 		error = EINVAL;
    704 	}
    705 
    706 	return error;
    707 }
    708 
    709 static void
    710 ip6flow_sysctl_init(struct sysctllog **clog)
    711 {
    712 
    713 	sysctl_createv(clog, 0, NULL, NULL,
    714 		       CTLFLAG_PERMANENT,
    715 		       CTLTYPE_NODE, "inet6",
    716 		       SYSCTL_DESCR("PF_INET6 related settings"),
    717 		       NULL, 0, NULL, 0,
    718 		       CTL_NET, PF_INET6, CTL_EOL);
    719 	sysctl_createv(clog, 0, NULL, NULL,
    720 		       CTLFLAG_PERMANENT,
    721 		       CTLTYPE_NODE, "ip6",
    722 		       SYSCTL_DESCR("IPv6 related settings"),
    723 		       NULL, 0, NULL, 0,
    724 		       CTL_NET, PF_INET6, IPPROTO_IPV6, CTL_EOL);
    725 
    726 	sysctl_createv(clog, 0, NULL, NULL,
    727 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    728 		       CTLTYPE_INT, "maxflows",
    729 		       SYSCTL_DESCR("Number of flows for fast forwarding (IPv6)"),
    730 		       sysctl_net_inet6_ip6_maxflows, 0, &ip6_maxflows, 0,
    731 		       CTL_NET, PF_INET6, IPPROTO_IPV6,
    732 		       CTL_CREATE, CTL_EOL);
    733 	sysctl_createv(clog, 0, NULL, NULL,
    734 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    735 		       CTLTYPE_INT, "hashsize",
    736 		       SYSCTL_DESCR("Size of hash table for fast forwarding (IPv6)"),
    737 		       sysctl_net_inet6_ip6_hashsize, 0, &ip6_hashsize, 0,
    738 		       CTL_NET, PF_INET6, IPPROTO_IPV6,
    739 		       CTL_CREATE, CTL_EOL);
    740 }
    741