Home | History | Annotate | Line # | Download | only in netinet6
ip6_flow.c revision 1.19.6.4
      1 /*	$NetBSD: ip6_flow.c,v 1.19.6.4 2017/12/03 11:39:04 jdolecek 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 1.19.6.4 2017/12/03 11:39:04 jdolecek 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;
    102 static struct ip6flowhead *ip6flowtable = NULL;
    103 static struct ip6flowhead ip6flowlist;
    104 static int ip6flow_inuse;
    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_slowtimo",
    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 (IP6_HDR_ALIGNED_P(mtod(m, const void *)) == 0) {
    287 		if ((m = m_copyup(m, sizeof(struct ip6_hdr),
    288 				(max_linkhdr + 3) & ~3)) == NULL) {
    289 			goto out;
    290 		}
    291 		*mp = m;
    292 	} else if (__predict_false(m->m_len < sizeof(struct ip6_hdr))) {
    293 		if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
    294 			goto out;
    295 		}
    296 		*mp = m;
    297 	}
    298 
    299 	ip6 = mtod(m, struct ip6_hdr *);
    300 
    301 	if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) {
    302 		/* Bad version. */
    303 		goto out;
    304 	}
    305 
    306 	/*
    307 	 * If we have a hop-by-hop extension we must process it.
    308 	 * We just leave this up to ip6_input to deal with.
    309 	 */
    310 	if (ip6->ip6_nxt == IPPROTO_HOPOPTS)
    311 		goto out;
    312 
    313 	/*
    314 	 * Attempt to find a flow.
    315 	 */
    316 	if ((ip6f = ip6flow_lookup(ip6)) == NULL) {
    317 		/* No flow found. */
    318 		goto out;
    319 	}
    320 
    321 	/*
    322 	 * Route and interface still up?
    323 	 */
    324 	if ((rt = rtcache_validate(&ip6f->ip6f_ro)) == NULL ||
    325 	    (rt->rt_ifp->if_flags & IFF_UP) == 0 ||
    326 	    (rt->rt_flags & RTF_BLACKHOLE) != 0)
    327 		goto out_unref;
    328 
    329 	/*
    330 	 * Packet size greater than MTU?
    331 	 */
    332 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu) {
    333 		/* Return to main IPv6 input function. */
    334 		goto out_unref;
    335 	}
    336 
    337 	/*
    338 	 * Clear any in-bound checksum flags for this packet.
    339 	 */
    340 	m->m_pkthdr.csum_flags = 0;
    341 
    342 	if (ip6->ip6_hlim <= IPV6_HLIMDEC)
    343 		goto out_unref;
    344 
    345 	/* Decrement hop limit (same as TTL) */
    346 	ip6->ip6_hlim -= IPV6_HLIMDEC;
    347 
    348 	if (rt->rt_flags & RTF_GATEWAY)
    349 		dst = rt->rt_gateway;
    350 	else
    351 		dst = rtcache_getdst(&ip6f->ip6f_ro);
    352 
    353 	PRT_SLOW_ARM(ip6f->ip6f_timer, IP6FLOW_TIMER);
    354 
    355 	ip6f->ip6f_uses++;
    356 
    357 #if 0
    358 	/*
    359 	 * We use FIFO cache replacement instead of LRU the same ip_flow.c.
    360 	 */
    361 	/* move to head (LRU) for ip6flowlist. ip6flowtable does not care LRU. */
    362 	TAILQ_REMOVE(&ip6flowlist, ip6f, ip6f_list);
    363 	TAILQ_INSERT_HEAD(&ip6flowlist, ip6f, ip6f_list);
    364 #endif
    365 
    366 	/* Send on its way - straight to the interface output routine. */
    367 	if ((error = if_output_lock(rt->rt_ifp, rt->rt_ifp, m, dst, rt)) != 0) {
    368 		ip6f->ip6f_dropped++;
    369 	} else {
    370 		ip6f->ip6f_forwarded++;
    371 	}
    372 	ret = 1;
    373 out_unref:
    374 	rtcache_unref(rt, &ip6f->ip6f_ro);
    375 out:
    376 	mutex_exit(&ip6flow_lock);
    377 	return ret;
    378 }
    379 
    380 /*
    381  * Add the IPv6 flow statistics to the main IPv6 statistics.
    382  */
    383 static void
    384 ip6flow_addstats_rt(struct rtentry *rt, struct ip6flow *ip6f)
    385 {
    386 	uint64_t *ip6s;
    387 
    388 	if (rt != NULL)
    389 		rt->rt_use += ip6f->ip6f_uses;
    390 	ip6s = IP6_STAT_GETREF();
    391 	ip6s[IP6_STAT_FASTFORWARDFLOWS] = ip6flow_inuse;
    392 	ip6s[IP6_STAT_CANTFORWARD] += ip6f->ip6f_dropped;
    393 	ip6s[IP6_STAT_ODROPPED] += ip6f->ip6f_dropped;
    394 	ip6s[IP6_STAT_TOTAL] += ip6f->ip6f_uses;
    395 	ip6s[IP6_STAT_FORWARD] += ip6f->ip6f_forwarded;
    396 	ip6s[IP6_STAT_FASTFORWARD] += ip6f->ip6f_forwarded;
    397 	IP6_STAT_PUTREF();
    398 }
    399 
    400 static void
    401 ip6flow_addstats(struct ip6flow *ip6f)
    402 {
    403 	struct rtentry *rt;
    404 
    405 	rt = rtcache_validate(&ip6f->ip6f_ro);
    406 	ip6flow_addstats_rt(rt, ip6f);
    407 	rtcache_unref(rt, &ip6f->ip6f_ro);
    408 }
    409 
    410 /*
    411  * Add statistics and free the flow.
    412  */
    413 static void
    414 ip6flow_free(struct ip6flow *ip6f)
    415 {
    416 
    417 	KASSERT(mutex_owned(&ip6flow_lock));
    418 
    419 	/*
    420 	 * Remove the flow from the hash table (at elevated IPL).
    421 	 * Once it's off the list, we can deal with it at normal
    422 	 * network IPL.
    423 	 */
    424 	IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    425 
    426 	ip6flow_inuse--;
    427 	ip6flow_addstats(ip6f);
    428 	rtcache_free(&ip6f->ip6f_ro);
    429 	pool_put(&ip6flow_pool, ip6f);
    430 }
    431 
    432 static struct ip6flow *
    433 ip6flow_reap_locked(int just_one)
    434 {
    435 	struct ip6flow *ip6f;
    436 
    437 	KASSERT(mutex_owned(&ip6flow_lock));
    438 
    439 	/*
    440 	 * This case must remove one ip6flow. Furthermore, this case is used in
    441 	 * fast path(packet processing path). So, simply remove TAILQ_LAST one.
    442 	 */
    443 	if (just_one) {
    444 		ip6f = TAILQ_LAST(&ip6flowlist, ip6flowhead);
    445 		KASSERT(ip6f != NULL);
    446 
    447 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    448 
    449 		ip6flow_addstats(ip6f);
    450 		rtcache_free(&ip6f->ip6f_ro);
    451 		return ip6f;
    452 	}
    453 
    454 	/*
    455 	 * This case is used in slow path(sysctl).
    456 	 * At first, remove invalid rtcache ip6flow, and then remove TAILQ_LAST
    457 	 * ip6flow if it is ensured least recently used by comparing last_uses.
    458 	 */
    459 	while (ip6flow_inuse > ip6_maxflows) {
    460 		struct ip6flow *maybe_ip6f = TAILQ_LAST(&ip6flowlist, ip6flowhead);
    461 
    462 		TAILQ_FOREACH(ip6f, &ip6flowlist, ip6f_list) {
    463 			struct rtentry *rt;
    464 			/*
    465 			 * If this no longer points to a valid route -
    466 			 * reclaim it.
    467 			 */
    468 			if ((rt = rtcache_validate(&ip6f->ip6f_ro)) == NULL)
    469 				goto done;
    470 			rtcache_unref(rt, &ip6f->ip6f_ro);
    471 			/*
    472 			 * choose the one that's been least recently
    473 			 * used or has had the least uses in the
    474 			 * last 1.5 intervals.
    475 			 */
    476 			if (ip6f->ip6f_timer < maybe_ip6f->ip6f_timer
    477 			    || ((ip6f->ip6f_timer == maybe_ip6f->ip6f_timer)
    478 				&& (ip6f->ip6f_last_uses + ip6f->ip6f_uses
    479 				    < maybe_ip6f->ip6f_last_uses + maybe_ip6f->ip6f_uses)))
    480 				maybe_ip6f = ip6f;
    481 		}
    482 		ip6f = maybe_ip6f;
    483 	    done:
    484 		/*
    485 		 * Remove the entry from the flow table
    486 		 */
    487 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    488 
    489 		rtcache_free(&ip6f->ip6f_ro);
    490 		ip6flow_inuse--;
    491 		ip6flow_addstats(ip6f);
    492 		pool_put(&ip6flow_pool, ip6f);
    493 	}
    494 	return NULL;
    495 }
    496 
    497 /*
    498  * Reap one or more flows - ip6flow_reap may remove
    499  * multiple flows if net.inet6.ip6.maxflows is reduced.
    500  */
    501 struct ip6flow *
    502 ip6flow_reap(int just_one)
    503 {
    504 	struct ip6flow *ip6f;
    505 
    506 	mutex_enter(&ip6flow_lock);
    507 	ip6f = ip6flow_reap_locked(just_one);
    508 	mutex_exit(&ip6flow_lock);
    509 	return ip6f;
    510 }
    511 
    512 static unsigned int ip6flow_work_enqueued = 0;
    513 
    514 void
    515 ip6flow_slowtimo_work(struct work *wk, void *arg)
    516 {
    517 	struct ip6flow *ip6f, *next_ip6f;
    518 
    519 	/* We can allow enqueuing another work at this point */
    520 	atomic_swap_uint(&ip6flow_work_enqueued, 0);
    521 
    522 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    523 	mutex_enter(&ip6flow_lock);
    524 
    525 	for (ip6f = TAILQ_FIRST(&ip6flowlist); ip6f != NULL; ip6f = next_ip6f) {
    526 		struct rtentry *rt = NULL;
    527 		next_ip6f = TAILQ_NEXT(ip6f, ip6f_list);
    528 		if (PRT_SLOW_ISEXPIRED(ip6f->ip6f_timer) ||
    529 		    (rt = rtcache_validate(&ip6f->ip6f_ro)) == NULL) {
    530 			ip6flow_free(ip6f);
    531 		} else {
    532 			ip6f->ip6f_last_uses = ip6f->ip6f_uses;
    533 			ip6flow_addstats_rt(rt, ip6f);
    534 			ip6f->ip6f_uses = 0;
    535 			ip6f->ip6f_dropped = 0;
    536 			ip6f->ip6f_forwarded = 0;
    537 		}
    538 		rtcache_unref(rt, &ip6f->ip6f_ro);
    539 	}
    540 
    541 	mutex_exit(&ip6flow_lock);
    542 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    543 }
    544 
    545 void
    546 ip6flow_slowtimo(void)
    547 {
    548 
    549 	/* Avoid enqueuing another work when one is already enqueued */
    550 	if (atomic_swap_uint(&ip6flow_work_enqueued, 1) == 1)
    551 		return;
    552 
    553 	workqueue_enqueue(ip6flow_slowtimo_wq, &ip6flow_slowtimo_wk, NULL);
    554 }
    555 
    556 /*
    557  * We have successfully forwarded a packet using the normal
    558  * IPv6 stack. Now create/update a flow.
    559  */
    560 void
    561 ip6flow_create(struct route *ro, struct mbuf *m)
    562 {
    563 	const struct ip6_hdr *ip6;
    564 	struct ip6flow *ip6f;
    565 	size_t hash;
    566 
    567 	ip6 = mtod(m, const struct ip6_hdr *);
    568 
    569 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
    570 	mutex_enter(&ip6flow_lock);
    571 
    572 	/*
    573 	 * If IPv6 Fast Forward is disabled, don't create a flow.
    574 	 * It can be disabled by setting net.inet6.ip6.maxflows to 0.
    575 	 *
    576 	 * Don't create a flow for ICMPv6 messages.
    577 	 */
    578 	if (ip6_maxflows == 0 || ip6->ip6_nxt == IPPROTO_IPV6_ICMP)
    579 		goto out;
    580 
    581 	/*
    582 	 * See if an existing flow exists.  If so:
    583 	 *	- Remove the flow
    584 	 *	- Add flow statistics
    585 	 *	- Free the route
    586 	 *	- Reset statistics
    587 	 *
    588 	 * If a flow doesn't exist allocate a new one if
    589 	 * ip6_maxflows hasn't reached its limit. If it has
    590 	 * been reached, reap some flows.
    591 	 */
    592 	ip6f = ip6flow_lookup(ip6);
    593 	if (ip6f == NULL) {
    594 		if (ip6flow_inuse >= ip6_maxflows) {
    595 			ip6f = ip6flow_reap_locked(1);
    596 		} else {
    597 			ip6f = pool_get(&ip6flow_pool, PR_NOWAIT);
    598 			if (ip6f == NULL)
    599 				goto out;
    600 			ip6flow_inuse++;
    601 		}
    602 		memset(ip6f, 0, sizeof(*ip6f));
    603 	} else {
    604 		IP6FLOW_REMOVE(ip6f->ip6f_hashidx, ip6f);
    605 
    606 		ip6flow_addstats(ip6f);
    607 		rtcache_free(&ip6f->ip6f_ro);
    608 		ip6f->ip6f_uses = 0;
    609 		ip6f->ip6f_last_uses = 0;
    610 		ip6f->ip6f_dropped = 0;
    611 		ip6f->ip6f_forwarded = 0;
    612 	}
    613 
    614 	/*
    615 	 * Fill in the updated/new details.
    616 	 */
    617 	rtcache_copy(&ip6f->ip6f_ro, ro);
    618 	ip6f->ip6f_dst = ip6->ip6_dst;
    619 	ip6f->ip6f_src = ip6->ip6_src;
    620 	ip6f->ip6f_flow = ip6->ip6_flow;
    621 	PRT_SLOW_ARM(ip6f->ip6f_timer, IP6FLOW_TIMER);
    622 
    623 	/*
    624 	 * Insert into the approriate bucket of the flow table.
    625 	 */
    626 	hash = ip6flow_hash(ip6);
    627 	IP6FLOW_INSERT(hash, ip6f);
    628 
    629  out:
    630 	mutex_exit(&ip6flow_lock);
    631 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    632 }
    633 
    634 /*
    635  * Invalidate/remove all flows - if new_size is positive we
    636  * resize the hash table.
    637  */
    638 int
    639 ip6flow_invalidate_all(int new_size)
    640 {
    641 	struct ip6flow *ip6f, *next_ip6f;
    642 	int error;
    643 
    644 	error = 0;
    645 
    646 	mutex_enter(&ip6flow_lock);
    647 
    648 	for (ip6f = TAILQ_FIRST(&ip6flowlist); ip6f != NULL; ip6f = next_ip6f) {
    649 		next_ip6f = TAILQ_NEXT(ip6f, ip6f_list);
    650 		ip6flow_free(ip6f);
    651 	}
    652 
    653 	if (new_size)
    654 		error = ip6flow_init_locked(new_size);
    655 
    656 	mutex_exit(&ip6flow_lock);
    657 
    658 	return error;
    659 }
    660 
    661 /*
    662  * sysctl helper routine for net.inet.ip6.maxflows. Since
    663  * we could reduce this value, call ip6flow_reap();
    664  */
    665 static int
    666 sysctl_net_inet6_ip6_maxflows(SYSCTLFN_ARGS)
    667 {
    668 	int error;
    669 
    670 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
    671 	if (error || newp == NULL)
    672 		return (error);
    673 
    674 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    675 
    676 	ip6flow_reap(0);
    677 
    678 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    679 
    680 	return (0);
    681 }
    682 
    683 static int
    684 sysctl_net_inet6_ip6_hashsize(SYSCTLFN_ARGS)
    685 {
    686 	int error, tmp;
    687 	struct sysctlnode node;
    688 
    689 	node = *rnode;
    690 	tmp = ip6_hashsize;
    691 	node.sysctl_data = &tmp;
    692 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
    693 	if (error || newp == NULL)
    694 		return (error);
    695 
    696 	if ((tmp & (tmp - 1)) == 0 && tmp != 0) {
    697 		/*
    698 		 * Can only fail due to malloc()
    699 		 */
    700 		SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    701 		error = ip6flow_invalidate_all(tmp);
    702 		SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    703 	} else {
    704 		/*
    705 		 * EINVAL if not a power of 2
    706 		 */
    707 		error = EINVAL;
    708 	}
    709 
    710 	return error;
    711 }
    712 
    713 static void
    714 ip6flow_sysctl_init(struct sysctllog **clog)
    715 {
    716 
    717 	sysctl_createv(clog, 0, NULL, NULL,
    718 		       CTLFLAG_PERMANENT,
    719 		       CTLTYPE_NODE, "inet6",
    720 		       SYSCTL_DESCR("PF_INET6 related settings"),
    721 		       NULL, 0, NULL, 0,
    722 		       CTL_NET, PF_INET6, CTL_EOL);
    723 	sysctl_createv(clog, 0, NULL, NULL,
    724 		       CTLFLAG_PERMANENT,
    725 		       CTLTYPE_NODE, "ip6",
    726 		       SYSCTL_DESCR("IPv6 related settings"),
    727 		       NULL, 0, NULL, 0,
    728 		       CTL_NET, PF_INET6, IPPROTO_IPV6, CTL_EOL);
    729 
    730 	sysctl_createv(clog, 0, NULL, NULL,
    731 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    732 			CTLTYPE_INT, "maxflows",
    733 			SYSCTL_DESCR("Number of flows for fast forwarding (IPv6)"),
    734 			sysctl_net_inet6_ip6_maxflows, 0, &ip6_maxflows, 0,
    735 			CTL_NET, PF_INET6, IPPROTO_IPV6,
    736 			CTL_CREATE, CTL_EOL);
    737 	sysctl_createv(clog, 0, NULL, NULL,
    738 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    739 			CTLTYPE_INT, "hashsize",
    740 			SYSCTL_DESCR("Size of hash table for fast forwarding (IPv6)"),
    741 			sysctl_net_inet6_ip6_hashsize, 0, &ip6_hashsize, 0,
    742 			CTL_NET, PF_INET6, IPPROTO_IPV6,
    743 			CTL_CREATE, CTL_EOL);
    744 }
    745