Home | History | Annotate | Line # | Download | only in netinet6
ip6_flow.c revision 1.31
      1 /*	$NetBSD: ip6_flow.c,v 1.31 2016/08/23 09:59:20 knakahara 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 1.31 2016/08/23 09:59:20 knakahara Exp $");
     42 
     43 #include <sys/param.h>
     44 #include <sys/systm.h>
     45 #include <sys/malloc.h>
     46 #include <sys/mbuf.h>
     47 #include <sys/domain.h>
     48 #include <sys/protosw.h>
     49 #include <sys/socket.h>
     50 #include <sys/socketvar.h>
     51 #include <sys/time.h>
     52 #include <sys/kernel.h>
     53 #include <sys/pool.h>
     54 #include <sys/sysctl.h>
     55 #include <sys/workqueue.h>
     56 #include <sys/atomic.h>
     57 
     58 #include <net/if.h>
     59 #include <net/if_dl.h>
     60 #include <net/route.h>
     61 #include <net/pfil.h>
     62 
     63 #include <netinet/in.h>
     64 #include <netinet6/in6_var.h>
     65 #include <netinet/in_systm.h>
     66 #include <netinet/ip6.h>
     67 #include <netinet6/ip6_var.h>
     68 #include <netinet6/ip6_private.h>
     69 
     70 /*
     71  * IPv6 Fast Forward caches/hashes flows from one source to destination.
     72  *
     73  * Upon a successful forward IPv6FF caches and hashes details such as the
     74  * route, source and destination. Once another packet is received matching
     75  * the source and destination the packet is forwarded straight onto if_output
     76  * using the cached details.
     77  *
     78  * Example:
     79  * ether/fddi_input -> ip6flow_fastforward -> if_output
     80  */
     81 
     82 static struct pool ip6flow_pool;
     83 
     84 TAILQ_HEAD(ip6flowhead, ip6flow);
     85 
     86 /*
     87  * We could use IPv4 defines (IPFLOW_HASHBITS) but we'll
     88  * use our own (possibly for future expansion).
     89  */
     90 #define	IP6FLOW_TIMER		(5 * PR_SLOWHZ)
     91 #define	IP6FLOW_DEFAULT_HASHSIZE	(1 << IP6FLOW_HASHBITS)
     92 
     93 /*
     94  * ip6_flow.c internal lock.
     95  * If we use softnet_lock, it would cause recursive lock.
     96  *
     97  * This is a tentative workaround.
     98  * We should make it scalable somehow in the future.
     99  */
    100 static kmutex_t ip6flow_lock;
    101 static struct ip6flowhead *ip6flowtable = NULL;
    102 static struct ip6flowhead ip6flowlist;
    103 static int ip6flow_inuse;
    104 
    105 static void ip6flow_slowtimo_work(struct work *, void *);
    106 static struct workqueue	*ip6flow_slowtimo_wq;
    107 static struct work	ip6flow_slowtimo_wk;
    108 
    109 static int sysctl_net_inet6_ip6_hashsize(SYSCTLFN_PROTO);
    110 static int sysctl_net_inet6_ip6_maxflows(SYSCTLFN_PROTO);
    111 static void ip6flow_sysctl_init(struct sysctllog **);
    112 
    113 /*
    114  * Insert an ip6flow into the list.
    115  */
    116 #define	IP6FLOW_INSERT(hashidx, ip6f) \
    117 do { \
    118 	(ip6f)->ip6f_hashidx = (hashidx); \
    119 	TAILQ_INSERT_HEAD(&ip6flowtable[(hashidx)], (ip6f), ip6f_hash); \
    120 	TAILQ_INSERT_HEAD(&ip6flowlist, (ip6f), ip6f_list); \
    121 } while (/*CONSTCOND*/ 0)
    122 
    123 /*
    124  * Remove an ip6flow from the list.
    125  */
    126 #define	IP6FLOW_REMOVE(hashidx, ip6f) \
    127 do { \
    128 	TAILQ_REMOVE(&ip6flowtable[(hashidx)], (ip6f), ip6f_hash); \
    129 	TAILQ_REMOVE(&ip6flowlist, (ip6f), ip6f_list); \
    130 } while (/*CONSTCOND*/ 0)
    131 
    132 #ifndef IP6FLOW_DEFAULT
    133 #define	IP6FLOW_DEFAULT		256
    134 #endif
    135 
    136 int ip6_maxflows = IP6FLOW_DEFAULT;
    137 int ip6_hashsize = IP6FLOW_DEFAULT_HASHSIZE;
    138 
    139 /*
    140  * Calculate hash table position.
    141  */
    142 static size_t
    143 ip6flow_hash(const struct ip6_hdr *ip6)
    144 {
    145 	size_t hash;
    146 	uint32_t dst_sum, src_sum;
    147 	size_t idx;
    148 
    149 	src_sum = ip6->ip6_src.s6_addr32[0] + ip6->ip6_src.s6_addr32[1]
    150 	    + ip6->ip6_src.s6_addr32[2] + ip6->ip6_src.s6_addr32[3];
    151 	dst_sum = ip6->ip6_dst.s6_addr32[0] + ip6->ip6_dst.s6_addr32[1]
    152 	    + ip6->ip6_dst.s6_addr32[2] + ip6->ip6_dst.s6_addr32[3];
    153 
    154 	hash = ip6->ip6_flow;
    155 
    156 	for (idx = 0; idx < 32; idx += IP6FLOW_HASHBITS)
    157 		hash += (dst_sum >> (32 - idx)) + (src_sum >> idx);
    158 
    159 	return hash & (ip6_hashsize-1);
    160 }
    161 
    162 /*
    163  * Check to see if a flow already exists - if so return it.
    164  */
    165 static struct ip6flow *
    166 ip6flow_lookup(const struct ip6_hdr *ip6)
    167 {
    168 	size_t hash;
    169 	struct ip6flow *ip6f;
    170 
    171 	KASSERT(mutex_owned(&ip6flow_lock));
    172 
    173 	hash = ip6flow_hash(ip6);
    174 
    175 	TAILQ_FOREACH(ip6f, &ip6flowtable[hash], ip6f_hash) {
    176 		if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6f->ip6f_dst)
    177 		    && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6f->ip6f_src)
    178 		    && ip6f->ip6f_flow == ip6->ip6_flow) {
    179 		    	/* A cached flow has been found. */
    180 			return ip6f;
    181 		}
    182 	}
    183 
    184 	return NULL;
    185 }
    186 
    187 void
    188 ip6flow_poolinit(void)
    189 {
    190 
    191 	pool_init(&ip6flow_pool, sizeof(struct ip6flow), 0, 0, 0, "ip6flowpl",
    192 			NULL, IPL_NET);
    193 }
    194 
    195 /*
    196  * Allocate memory and initialise lists. This function is called
    197  * from ip6_init and called there after to resize the hash table.
    198  * If a newly sized table cannot be malloc'ed we just continue
    199  * to use the old one.
    200  */
    201 static int
    202 ip6flow_init_locked(int table_size)
    203 {
    204 	struct ip6flowhead *new_table;
    205 	size_t i;
    206 
    207 	KASSERT(mutex_owned(&ip6flow_lock));
    208 
    209 	new_table = (struct ip6flowhead *)malloc(sizeof(struct ip6flowhead) *
    210 	    table_size, M_RTABLE, M_NOWAIT);
    211 
    212 	if (new_table == NULL)
    213 		return 1;
    214 
    215 	if (ip6flowtable != NULL)
    216 		free(ip6flowtable, M_RTABLE);
    217 
    218 	ip6flowtable = new_table;
    219 	ip6_hashsize = table_size;
    220 
    221 	TAILQ_INIT(&ip6flowlist);
    222 	for (i = 0; i < ip6_hashsize; i++)
    223 		TAILQ_INIT(&ip6flowtable[i]);
    224 
    225 	return 0;
    226 }
    227 
    228 int
    229 ip6flow_init(int table_size)
    230 {
    231 	int ret, error;
    232 
    233 	error = workqueue_create(&ip6flow_slowtimo_wq, "ip6flow_slowtimo",
    234 	    ip6flow_slowtimo_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
    235 	if (error != 0)
    236 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
    237 
    238 	mutex_init(&ip6flow_lock, MUTEX_DEFAULT, IPL_NONE);
    239 
    240 	mutex_enter(&ip6flow_lock);
    241 	ret = ip6flow_init_locked(table_size);
    242 	mutex_exit(&ip6flow_lock);
    243 	ip6flow_sysctl_init(NULL);
    244 
    245 	return ret;
    246 }
    247 
    248 /*
    249  * IPv6 Fast Forward routine. Attempt to forward the packet -
    250  * if any problems are found return to the main IPv6 input
    251  * routine to deal with.
    252  */
    253 int
    254 ip6flow_fastforward(struct mbuf **mp)
    255 {
    256 	struct ip6flow *ip6f;
    257 	struct ip6_hdr *ip6;
    258 	struct rtentry *rt;
    259 	struct mbuf *m;
    260 	const struct sockaddr *dst;
    261 	int error;
    262 	int ret = 0;
    263 
    264 	mutex_enter(&ip6flow_lock);
    265 
    266 	/*
    267 	 * Are we forwarding packets and have flows?
    268 	 */
    269 	if (!ip6_forwarding || ip6flow_inuse == 0)
    270 		goto out;
    271 
    272 	m = *mp;
    273 	/*
    274 	 * At least size of IPv6 Header?
    275 	 */
    276 	if (m->m_len < sizeof(struct ip6_hdr))
    277 		goto out;
    278 	/*
    279 	 * Was packet received as a link-level multicast or broadcast?
    280 	 * If so, don't try to fast forward.
    281 	 */
    282 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
    283 		goto out;
    284 
    285 	if (IP6_HDR_ALIGNED_P(mtod(m, const void *)) == 0) {
    286 		if ((m = m_copyup(m, sizeof(struct ip6_hdr),
    287 				(max_linkhdr + 3) & ~3)) == NULL) {
    288 			goto out;
    289 		}
    290 		*mp = m;
    291 	} else if (__predict_false(m->m_len < sizeof(struct ip6_hdr))) {
    292 		if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
    293 			goto out;
    294 		}
    295 		*mp = m;
    296 	}
    297 
    298 	ip6 = mtod(m, struct ip6_hdr *);
    299 
    300 	if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) {
    301 		/* Bad version. */
    302 		goto out;
    303 	}
    304 
    305 	/*
    306 	 * If we have a hop-by-hop extension we must process it.
    307 	 * We just leave this up to ip6_input to deal with.
    308 	 */
    309 	if (ip6->ip6_nxt == IPPROTO_HOPOPTS)
    310 		goto out;
    311 
    312 	/*
    313 	 * Attempt to find a flow.
    314 	 */
    315 	if ((ip6f = ip6flow_lookup(ip6)) == NULL) {
    316 		/* No flow found. */
    317 		goto out;
    318 	}
    319 
    320 	/*
    321 	 * Route and interface still up?
    322 	 */
    323 	if ((rt = rtcache_validate(&ip6f->ip6f_ro)) == NULL ||
    324 	    (rt->rt_ifp->if_flags & IFF_UP) == 0 ||
    325 	    (rt->rt_flags & RTF_BLACKHOLE) != 0)
    326 		goto out;
    327 
    328 	/*
    329 	 * Packet size greater than MTU?
    330 	 */
    331 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu) {
    332 		/* Return to main IPv6 input function. */
    333 		goto out;
    334 	}
    335 
    336 	/*
    337 	 * Clear any in-bound checksum flags for this packet.
    338 	 */
    339 	m->m_pkthdr.csum_flags = 0;
    340 
    341 	if (ip6->ip6_hlim <= IPV6_HLIMDEC)
    342 		goto out;
    343 
    344 	/* Decrement hop limit (same as TTL) */
    345 	ip6->ip6_hlim -= IPV6_HLIMDEC;
    346 
    347 	if (rt->rt_flags & RTF_GATEWAY)
    348 		dst = rt->rt_gateway;
    349 	else
    350 		dst = rtcache_getdst(&ip6f->ip6f_ro);
    351 
    352 	PRT_SLOW_ARM(ip6f->ip6f_timer, IP6FLOW_TIMER);
    353 
    354 	ip6f->ip6f_uses++;
    355 
    356 #if 0
    357 	/*
    358 	 * We use FIFO cache replacement instead of LRU the same ip_flow.c.
    359 	 */
    360 	/* move to head (LRU) for ip6flowlist. ip6flowtable does not care LRU. */
    361 	TAILQ_REMOVE(&ip6flowlist, ip6f, ip6f_list);
    362 	TAILQ_INSERT_HEAD(&ip6flowlist, ip6f, ip6f_list);
    363 #endif
    364 
    365 	/* Send on its way - straight to the interface output routine. */
    366 	if ((error = if_output_lock(rt->rt_ifp, rt->rt_ifp, m, dst, rt)) != 0) {
    367 		ip6f->ip6f_dropped++;
    368 	} else {
    369 		ip6f->ip6f_forwarded++;
    370 	}
    371 	ret = 1;
    372  out:
    373 	mutex_exit(&ip6flow_lock);
    374 	return ret;
    375 }
    376 
    377 /*
    378  * Add the IPv6 flow statistics to the main IPv6 statistics.
    379  */
    380 static void
    381 ip6flow_addstats(const struct ip6flow *ip6f)
    382 {
    383 	struct rtentry *rt;
    384 	uint64_t *ip6s;
    385 
    386 	if ((rt = rtcache_validate(&ip6f->ip6f_ro)) != NULL)
    387 		rt->rt_use += ip6f->ip6f_uses;
    388 	ip6s = IP6_STAT_GETREF();
    389 	ip6s[IP6_STAT_FASTFORWARDFLOWS] = ip6flow_inuse;
    390 	ip6s[IP6_STAT_CANTFORWARD] += ip6f->ip6f_dropped;
    391 	ip6s[IP6_STAT_ODROPPED] += ip6f->ip6f_dropped;
    392 	ip6s[IP6_STAT_TOTAL] += ip6f->ip6f_uses;
    393 	ip6s[IP6_STAT_FORWARD] += ip6f->ip6f_forwarded;
    394 	ip6s[IP6_STAT_FASTFORWARD] += ip6f->ip6f_forwarded;
    395 	IP6_STAT_PUTREF();
    396 }
    397 
    398 /*
    399  * Add statistics and free the flow.
    400  */
    401 static void
    402 ip6flow_free(struct ip6flow *ip6f)
    403 {
    404 
    405 	KASSERT(mutex_owned(&ip6flow_lock));
    406 
    407 	/*
    408 	 * Remove the flow from the hash table (at elevated IPL).
    409 	 * Once it's off the list, we can deal with it at normal
    410 	 * network IPL.
    411 	 */
    412 	IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    413 
    414 	ip6flow_inuse--;
    415 	ip6flow_addstats(ip6f);
    416 	rtcache_free(&ip6f->ip6f_ro);
    417 	pool_put(&ip6flow_pool, ip6f);
    418 }
    419 
    420 static struct ip6flow *
    421 ip6flow_reap_locked(int just_one)
    422 {
    423 	struct ip6flow *ip6f;
    424 
    425 	KASSERT(mutex_owned(&ip6flow_lock));
    426 
    427 	/*
    428 	 * This case must remove one ip6flow. Furthermore, this case is used in
    429 	 * fast path(packet processing path). So, simply remove TAILQ_LAST one.
    430 	 */
    431 	if (just_one) {
    432 		ip6f = TAILQ_LAST(&ip6flowlist, ip6flowhead);
    433 		KASSERT(ip6f != NULL);
    434 
    435 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    436 
    437 		ip6flow_addstats(ip6f);
    438 		rtcache_free(&ip6f->ip6f_ro);
    439 		return ip6f;
    440 	}
    441 
    442 	/*
    443 	 * This case is used in slow path(sysctl).
    444 	 * At first, remove invalid rtcache ip6flow, and then remove TAILQ_LAST
    445 	 * ip6flow if it is ensured least recently used by comparing last_uses.
    446 	 */
    447 	while (ip6flow_inuse > ip6_maxflows) {
    448 		struct ip6flow *maybe_ip6f = TAILQ_LAST(&ip6flowlist, ip6flowhead);
    449 
    450 		TAILQ_FOREACH(ip6f, &ip6flowlist, ip6f_list) {
    451 			/*
    452 			 * If this no longer points to a valid route -
    453 			 * reclaim it.
    454 			 */
    455 			if (rtcache_validate(&ip6f->ip6f_ro) == NULL)
    456 				goto done;
    457 			/*
    458 			 * choose the one that's been least recently
    459 			 * used or has had the least uses in the
    460 			 * last 1.5 intervals.
    461 			 */
    462 			if (ip6f->ip6f_timer < maybe_ip6f->ip6f_timer
    463 			    || ((ip6f->ip6f_timer == maybe_ip6f->ip6f_timer)
    464 				&& (ip6f->ip6f_last_uses + ip6f->ip6f_uses
    465 				    < maybe_ip6f->ip6f_last_uses + maybe_ip6f->ip6f_uses)))
    466 				maybe_ip6f = ip6f;
    467 		}
    468 		ip6f = maybe_ip6f;
    469 	    done:
    470 		/*
    471 		 * Remove the entry from the flow table
    472 		 */
    473 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    474 
    475 		rtcache_free(&ip6f->ip6f_ro);
    476 		ip6flow_inuse--;
    477 		ip6flow_addstats(ip6f);
    478 		pool_put(&ip6flow_pool, ip6f);
    479 	}
    480 	return NULL;
    481 }
    482 
    483 /*
    484  * Reap one or more flows - ip6flow_reap may remove
    485  * multiple flows if net.inet6.ip6.maxflows is reduced.
    486  */
    487 struct ip6flow *
    488 ip6flow_reap(int just_one)
    489 {
    490 	struct ip6flow *ip6f;
    491 
    492 	mutex_enter(&ip6flow_lock);
    493 	ip6f = ip6flow_reap_locked(just_one);
    494 	mutex_exit(&ip6flow_lock);
    495 	return ip6f;
    496 }
    497 
    498 static unsigned int ip6flow_work_enqueued = 0;
    499 
    500 void
    501 ip6flow_slowtimo_work(struct work *wk, void *arg)
    502 {
    503 	struct ip6flow *ip6f, *next_ip6f;
    504 
    505 	/* We can allow enqueuing another work at this point */
    506 	atomic_swap_uint(&ip6flow_work_enqueued, 0);
    507 
    508 	mutex_enter(softnet_lock);
    509 	mutex_enter(&ip6flow_lock);
    510 	KERNEL_LOCK(1, NULL);
    511 
    512 	for (ip6f = TAILQ_FIRST(&ip6flowlist); ip6f != NULL; ip6f = next_ip6f) {
    513 		next_ip6f = TAILQ_NEXT(ip6f, ip6f_list);
    514 		if (PRT_SLOW_ISEXPIRED(ip6f->ip6f_timer) ||
    515 		    rtcache_validate(&ip6f->ip6f_ro) == NULL) {
    516 			ip6flow_free(ip6f);
    517 		} else {
    518 			ip6f->ip6f_last_uses = ip6f->ip6f_uses;
    519 			ip6flow_addstats(ip6f);
    520 			ip6f->ip6f_uses = 0;
    521 			ip6f->ip6f_dropped = 0;
    522 			ip6f->ip6f_forwarded = 0;
    523 		}
    524 	}
    525 
    526 	KERNEL_UNLOCK_ONE(NULL);
    527 	mutex_exit(&ip6flow_lock);
    528 	mutex_exit(softnet_lock);
    529 }
    530 
    531 void
    532 ip6flow_slowtimo(void)
    533 {
    534 
    535 	/* Avoid enqueuing another work when one is already enqueued */
    536 	if (atomic_swap_uint(&ip6flow_work_enqueued, 1) == 1)
    537 		return;
    538 
    539 	workqueue_enqueue(ip6flow_slowtimo_wq, &ip6flow_slowtimo_wk, NULL);
    540 }
    541 
    542 /*
    543  * We have successfully forwarded a packet using the normal
    544  * IPv6 stack. Now create/update a flow.
    545  */
    546 void
    547 ip6flow_create(const struct route *ro, struct mbuf *m)
    548 {
    549 	const struct ip6_hdr *ip6;
    550 	struct ip6flow *ip6f;
    551 	size_t hash;
    552 
    553 	mutex_enter(&ip6flow_lock);
    554 
    555 	ip6 = mtod(m, const struct ip6_hdr *);
    556 
    557 	/*
    558 	 * If IPv6 Fast Forward is disabled, don't create a flow.
    559 	 * It can be disabled by setting net.inet6.ip6.maxflows to 0.
    560 	 *
    561 	 * Don't create a flow for ICMPv6 messages.
    562 	 */
    563 	if (ip6_maxflows == 0 || ip6->ip6_nxt == IPPROTO_IPV6_ICMP) {
    564 		mutex_exit(&ip6flow_lock);
    565 		return;
    566 	}
    567 
    568 	KERNEL_LOCK(1, NULL);
    569 
    570 	/*
    571 	 * See if an existing flow exists.  If so:
    572 	 *	- Remove the flow
    573 	 *	- Add flow statistics
    574 	 *	- Free the route
    575 	 *	- Reset statistics
    576 	 *
    577 	 * If a flow doesn't exist allocate a new one if
    578 	 * ip6_maxflows hasn't reached its limit. If it has
    579 	 * been reached, reap some flows.
    580 	 */
    581 	ip6f = ip6flow_lookup(ip6);
    582 	if (ip6f == NULL) {
    583 		if (ip6flow_inuse >= ip6_maxflows) {
    584 			ip6f = ip6flow_reap_locked(1);
    585 		} else {
    586 			ip6f = pool_get(&ip6flow_pool, PR_NOWAIT);
    587 			if (ip6f == NULL)
    588 				goto out;
    589 			ip6flow_inuse++;
    590 		}
    591 		memset(ip6f, 0, sizeof(*ip6f));
    592 	} else {
    593 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    594 
    595 		ip6flow_addstats(ip6f);
    596 		rtcache_free(&ip6f->ip6f_ro);
    597 		ip6f->ip6f_uses = 0;
    598 		ip6f->ip6f_last_uses = 0;
    599 		ip6f->ip6f_dropped = 0;
    600 		ip6f->ip6f_forwarded = 0;
    601 	}
    602 
    603 	/*
    604 	 * Fill in the updated/new details.
    605 	 */
    606 	rtcache_copy(&ip6f->ip6f_ro, ro);
    607 	ip6f->ip6f_dst = ip6->ip6_dst;
    608 	ip6f->ip6f_src = ip6->ip6_src;
    609 	ip6f->ip6f_flow = ip6->ip6_flow;
    610 	PRT_SLOW_ARM(ip6f->ip6f_timer, IP6FLOW_TIMER);
    611 
    612 	/*
    613 	 * Insert into the approriate bucket of the flow table.
    614 	 */
    615 	hash = ip6flow_hash(ip6);
    616 	IP6FLOW_INSERT(hash, ip6f);
    617 
    618  out:
    619 	KERNEL_UNLOCK_ONE(NULL);
    620 	mutex_exit(&ip6flow_lock);
    621 }
    622 
    623 /*
    624  * Invalidate/remove all flows - if new_size is positive we
    625  * resize the hash table.
    626  */
    627 int
    628 ip6flow_invalidate_all(int new_size)
    629 {
    630 	struct ip6flow *ip6f, *next_ip6f;
    631 	int error;
    632 
    633 	error = 0;
    634 
    635 	mutex_enter(&ip6flow_lock);
    636 
    637 	for (ip6f = TAILQ_FIRST(&ip6flowlist); ip6f != NULL; ip6f = next_ip6f) {
    638 		next_ip6f = TAILQ_NEXT(ip6f, ip6f_list);
    639 		ip6flow_free(ip6f);
    640 	}
    641 
    642 	if (new_size)
    643 		error = ip6flow_init_locked(new_size);
    644 
    645 	mutex_exit(&ip6flow_lock);
    646 
    647 	return error;
    648 }
    649 
    650 /*
    651  * sysctl helper routine for net.inet.ip6.maxflows. Since
    652  * we could reduce this value, call ip6flow_reap();
    653  */
    654 static int
    655 sysctl_net_inet6_ip6_maxflows(SYSCTLFN_ARGS)
    656 {
    657 	int error;
    658 
    659 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
    660 	if (error || newp == NULL)
    661 		return (error);
    662 
    663 	mutex_enter(softnet_lock);
    664 	KERNEL_LOCK(1, NULL);
    665 
    666 	ip6flow_reap(0);
    667 
    668 	KERNEL_UNLOCK_ONE(NULL);
    669 	mutex_exit(softnet_lock);
    670 
    671 	return (0);
    672 }
    673 
    674 static int
    675 sysctl_net_inet6_ip6_hashsize(SYSCTLFN_ARGS)
    676 {
    677 	int error, tmp;
    678 	struct sysctlnode node;
    679 
    680 	node = *rnode;
    681 	tmp = ip6_hashsize;
    682 	node.sysctl_data = &tmp;
    683 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
    684 	if (error || newp == NULL)
    685 		return (error);
    686 
    687 	if ((tmp & (tmp - 1)) == 0 && tmp != 0) {
    688 		/*
    689 		 * Can only fail due to malloc()
    690 		 */
    691 		mutex_enter(softnet_lock);
    692 		KERNEL_LOCK(1, NULL);
    693 
    694 		error = ip6flow_invalidate_all(tmp);
    695 
    696 		KERNEL_UNLOCK_ONE(NULL);
    697 		mutex_exit(softnet_lock);
    698 	} else {
    699 		/*
    700 		 * EINVAL if not a power of 2
    701 		 */
    702 		error = EINVAL;
    703 	}
    704 
    705 	return error;
    706 }
    707 
    708 static void
    709 ip6flow_sysctl_init(struct sysctllog **clog)
    710 {
    711 
    712 	sysctl_createv(clog, 0, NULL, NULL,
    713 		       CTLFLAG_PERMANENT,
    714 		       CTLTYPE_NODE, "inet6",
    715 		       SYSCTL_DESCR("PF_INET6 related settings"),
    716 		       NULL, 0, NULL, 0,
    717 		       CTL_NET, PF_INET6, CTL_EOL);
    718 	sysctl_createv(clog, 0, NULL, NULL,
    719 		       CTLFLAG_PERMANENT,
    720 		       CTLTYPE_NODE, "ip6",
    721 		       SYSCTL_DESCR("IPv6 related settings"),
    722 		       NULL, 0, NULL, 0,
    723 		       CTL_NET, PF_INET6, IPPROTO_IPV6, CTL_EOL);
    724 
    725 	sysctl_createv(clog, 0, NULL, NULL,
    726 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    727 			CTLTYPE_INT, "maxflows",
    728 			SYSCTL_DESCR("Number of flows for fast forwarding (IPv6)"),
    729 			sysctl_net_inet6_ip6_maxflows, 0, &ip6_maxflows, 0,
    730 			CTL_NET, PF_INET6, IPPROTO_IPV6,
    731 			CTL_CREATE, CTL_EOL);
    732 	sysctl_createv(clog, 0, NULL, NULL,
    733 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    734 			CTLTYPE_INT, "hashsize",
    735 			SYSCTL_DESCR("Size of hash table for fast forwarding (IPv6)"),
    736 			sysctl_net_inet6_ip6_hashsize, 0, &ip6_hashsize, 0,
    737 			CTL_NET, PF_INET6, IPPROTO_IPV6,
    738 			CTL_CREATE, CTL_EOL);
    739 }
    740