xref: /freebsd/sys/netinet/tcp_hostcache.c (revision 6356dba0b403daa023dec24559ab1f8e602e4f14)
1 /*-
2  * Copyright (c) 2002 Andre Oppermann, Internet Business Solutions AG
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. The name of the author may not be used to endorse or promote
14  *    products derived from this software without specific prior written
15  *    permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * The tcp_hostcache moves the tcp-specific cached metrics from the routing
32  * table to a dedicated structure indexed by the remote IP address.  It keeps
33  * information on the measured TCP parameters of past TCP sessions to allow
34  * better initial start values to be used with later connections to/from the
35  * same source.  Depending on the network parameters (delay, bandwidth, max
36  * MTU, congestion window) between local and remote sites, this can lead to
37  * significant speed-ups for new TCP connections after the first one.
38  *
39  * Due to the tcp_hostcache, all TCP-specific metrics information in the
40  * routing table has been removed.  The inpcb no longer keeps a pointer to
41  * the routing entry, and protocol-initiated route cloning has been removed
42  * as well.  With these changes, the routing table has gone back to being
43  * more lightwight and only carries information related to packet forwarding.
44  *
45  * tcp_hostcache is designed for multiple concurrent access in SMP
46  * environments and high contention.  All bucket rows have their own lock and
47  * thus multiple lookups and modifies can be done at the same time as long as
48  * they are in different bucket rows.  If a request for insertion of a new
49  * record can't be satisfied, it simply returns an empty structure.  Nobody
50  * and nothing outside of tcp_hostcache.c will ever point directly to any
51  * entry in the tcp_hostcache.  All communication is done in an
52  * object-oriented way and only functions of tcp_hostcache will manipulate
53  * hostcache entries.  Otherwise, we are unable to achieve good behaviour in
54  * concurrent access situations.  Since tcp_hostcache is only caching
55  * information, there are no fatal consequences if we either can't satisfy
56  * any particular request or have to drop/overwrite an existing entry because
57  * of bucket limit memory constrains.
58  */
59 
60 /*
61  * Many thanks to jlemon for basic structure of tcp_syncache which is being
62  * followed here.
63  */
64 
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
67 
68 #include "opt_inet6.h"
69 
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
73 #include <sys/lock.h>
74 #include <sys/mutex.h>
75 #include <sys/malloc.h>
76 #include <sys/socket.h>
77 #include <sys/socketvar.h>
78 #include <sys/sysctl.h>
79 #include <sys/vimage.h>
80 
81 #include <net/if.h>
82 
83 #include <netinet/in.h>
84 #include <netinet/in_systm.h>
85 #include <netinet/ip.h>
86 #include <netinet/in_var.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/ip_var.h>
89 #ifdef INET6
90 #include <netinet/ip6.h>
91 #include <netinet6/ip6_var.h>
92 #endif
93 #include <netinet/tcp.h>
94 #include <netinet/tcp_var.h>
95 #ifdef INET6
96 #include <netinet6/tcp6_var.h>
97 #endif
98 
99 #include <vm/uma.h>
100 
101 
102 TAILQ_HEAD(hc_qhead, hc_metrics);
103 
104 struct hc_head {
105 	struct hc_qhead	hch_bucket;
106 	u_int		hch_length;
107 	struct mtx	hch_mtx;
108 };
109 
110 struct hc_metrics {
111 	/* housekeeping */
112 	TAILQ_ENTRY(hc_metrics) rmx_q;
113 	struct	hc_head *rmx_head; /* head of bucket tail queue */
114 	struct	in_addr ip4;	/* IP address */
115 	struct	in6_addr ip6;	/* IP6 address */
116 	/* endpoint specific values for TCP */
117 	u_long	rmx_mtu;	/* MTU for this path */
118 	u_long	rmx_ssthresh;	/* outbound gateway buffer limit */
119 	u_long	rmx_rtt;	/* estimated round trip time */
120 	u_long	rmx_rttvar;	/* estimated rtt variance */
121 	u_long	rmx_bandwidth;	/* estimated bandwidth */
122 	u_long	rmx_cwnd;	/* congestion window */
123 	u_long	rmx_sendpipe;	/* outbound delay-bandwidth product */
124 	u_long	rmx_recvpipe;	/* inbound delay-bandwidth product */
125 	/* TCP hostcache internal data */
126 	int	rmx_expire;	/* lifetime for object */
127 	u_long	rmx_hits;	/* number of hits */
128 	u_long	rmx_updates;	/* number of updates */
129 };
130 
131 /* Arbitrary values */
132 #define TCP_HOSTCACHE_HASHSIZE		512
133 #define TCP_HOSTCACHE_BUCKETLIMIT	30
134 #define TCP_HOSTCACHE_EXPIRE		60*60	/* one hour */
135 #define TCP_HOSTCACHE_PRUNE		5*60	/* every 5 minutes */
136 
137 struct tcp_hostcache {
138 	struct	hc_head *hashbase;
139 	uma_zone_t zone;
140 	u_int	hashsize;
141 	u_int	hashmask;
142 	u_int	bucket_limit;
143 	u_int	cache_count;
144 	u_int	cache_limit;
145 	int	expire;
146 	int	prune;
147 	int	purgeall;
148 };
149 static struct tcp_hostcache tcp_hostcache;
150 
151 static struct callout tcp_hc_callout;
152 
153 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
154 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
155 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
156 static void tcp_hc_purge(void *);
157 
158 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, "TCP Host cache");
159 
160 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
161     &tcp_hostcache.cache_limit, 0, "Overall entry limit for hostcache");
162 
163 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
164     &tcp_hostcache.hashsize, 0, "Size of TCP hostcache hashtable");
165 
166 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
167     &tcp_hostcache.bucket_limit, 0, "Per-bucket hash limit for hostcache");
168 
169 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD,
170     &tcp_hostcache.cache_count, 0, "Current number of entries in hostcache");
171 
172 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW,
173     &tcp_hostcache.expire, 0, "Expire time of TCP hostcache entries");
174 
175 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW,
176      &tcp_hostcache.prune, 0, "Time between purge runs");
177 
178 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW,
179     &tcp_hostcache.purgeall, 0, "Expire all entires on next purge run");
180 
181 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
182     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
183     sysctl_tcp_hc_list, "A", "List of all hostcache entries");
184 
185 
186 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
187 
188 #define HOSTCACHE_HASH(ip) \
189 	(((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) &	\
190 	  V_tcp_hostcache.hashmask)
191 
192 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
193 #define HOSTCACHE_HASH6(ip6)				\
194 	(((ip6)->s6_addr32[0] ^				\
195 	  (ip6)->s6_addr32[1] ^				\
196 	  (ip6)->s6_addr32[2] ^				\
197 	  (ip6)->s6_addr32[3]) &			\
198 	 V_tcp_hostcache.hashmask)
199 
200 #define THC_LOCK(lp)		mtx_lock(lp)
201 #define THC_UNLOCK(lp)		mtx_unlock(lp)
202 
203 void
204 tcp_hc_init(void)
205 {
206 	int i;
207 
208 	/*
209 	 * Initialize hostcache structures.
210 	 */
211 	V_tcp_hostcache.cache_count = 0;
212 	V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
213 	V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
214 	V_tcp_hostcache.cache_limit =
215 	    V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit;
216 	V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
217 	V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
218 
219 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
220 	    &V_tcp_hostcache.hashsize);
221 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
222 	    &V_tcp_hostcache.cache_limit);
223 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
224 	    &V_tcp_hostcache.bucket_limit);
225 	if (!powerof2(V_tcp_hostcache.hashsize)) {
226 		printf("WARNING: hostcache hash size is not a power of 2.\n");
227 		V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
228 	}
229 	V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1;
230 
231 	/*
232 	 * Allocate the hash table.
233 	 */
234 	V_tcp_hostcache.hashbase = (struct hc_head *)
235 	    malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head),
236 		   M_HOSTCACHE, M_WAITOK | M_ZERO);
237 
238 	/*
239 	 * Initialize the hash buckets.
240 	 */
241 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
242 		TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket);
243 		V_tcp_hostcache.hashbase[i].hch_length = 0;
244 		mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
245 			  NULL, MTX_DEF);
246 	}
247 
248 	/*
249 	 * Allocate the hostcache entries.
250 	 */
251 	V_tcp_hostcache.zone =
252 	    uma_zcreate("hostcache", sizeof(struct hc_metrics),
253 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
254 	uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit);
255 
256 	/*
257 	 * Set up periodic cache cleanup.
258 	 */
259 	callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE);
260 	callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
261 	    tcp_hc_purge, 0);
262 }
263 
264 /*
265  * Internal function: look up an entry in the hostcache or return NULL.
266  *
267  * If an entry has been returned, the caller becomes responsible for
268  * unlocking the bucket row after he is done reading/modifying the entry.
269  */
270 static struct hc_metrics *
271 tcp_hc_lookup(struct in_conninfo *inc)
272 {
273 	int hash;
274 	struct hc_head *hc_head;
275 	struct hc_metrics *hc_entry;
276 
277 	KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer"));
278 
279 	/*
280 	 * Hash the foreign ip address.
281 	 */
282 	if (inc->inc_isipv6)
283 		hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
284 	else
285 		hash = HOSTCACHE_HASH(&inc->inc_faddr);
286 
287 	hc_head = &V_tcp_hostcache.hashbase[hash];
288 
289 	/*
290 	 * Acquire lock for this bucket row; we release the lock if we don't
291 	 * find an entry, otherwise the caller has to unlock after he is
292 	 * done.
293 	 */
294 	THC_LOCK(&hc_head->hch_mtx);
295 
296 	/*
297 	 * Iterate through entries in bucket row looking for a match.
298 	 */
299 	TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) {
300 		if (inc->inc_isipv6) {
301 			if (memcmp(&inc->inc6_faddr, &hc_entry->ip6,
302 			    sizeof(inc->inc6_faddr)) == 0)
303 				return hc_entry;
304 		} else {
305 			if (memcmp(&inc->inc_faddr, &hc_entry->ip4,
306 			    sizeof(inc->inc_faddr)) == 0)
307 				return hc_entry;
308 		}
309 	}
310 
311 	/*
312 	 * We were unsuccessful and didn't find anything.
313 	 */
314 	THC_UNLOCK(&hc_head->hch_mtx);
315 	return NULL;
316 }
317 
318 /*
319  * Internal function: insert an entry into the hostcache or return NULL if
320  * unable to allocate a new one.
321  *
322  * If an entry has been returned, the caller becomes responsible for
323  * unlocking the bucket row after he is done reading/modifying the entry.
324  */
325 static struct hc_metrics *
326 tcp_hc_insert(struct in_conninfo *inc)
327 {
328 	int hash;
329 	struct hc_head *hc_head;
330 	struct hc_metrics *hc_entry;
331 
332 	KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer"));
333 
334 	/*
335 	 * Hash the foreign ip address.
336 	 */
337 	if (inc->inc_isipv6)
338 		hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
339 	else
340 		hash = HOSTCACHE_HASH(&inc->inc_faddr);
341 
342 	hc_head = &V_tcp_hostcache.hashbase[hash];
343 
344 	/*
345 	 * Acquire lock for this bucket row; we release the lock if we don't
346 	 * find an entry, otherwise the caller has to unlock after he is
347 	 * done.
348 	 */
349 	THC_LOCK(&hc_head->hch_mtx);
350 
351 	/*
352 	 * If the bucket limit is reached, reuse the least-used element.
353 	 */
354 	if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit ||
355 	    V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) {
356 		hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead);
357 		/*
358 		 * At first we were dropping the last element, just to
359 		 * reacquire it in the next two lines again, which isn't very
360 		 * efficient.  Instead just reuse the least used element.
361 		 * We may drop something that is still "in-use" but we can be
362 		 * "lossy".
363 		 * Just give up if this bucket row is empty and we don't have
364 		 * anything to replace.
365 		 */
366 		if (hc_entry == NULL) {
367 			THC_UNLOCK(&hc_head->hch_mtx);
368 			return NULL;
369 		}
370 		TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
371 		V_tcp_hostcache.hashbase[hash].hch_length--;
372 		V_tcp_hostcache.cache_count--;
373 		V_tcpstat.tcps_hc_bucketoverflow++;
374 #if 0
375 		uma_zfree(V_tcp_hostcache.zone, hc_entry);
376 #endif
377 	} else {
378 		/*
379 		 * Allocate a new entry, or balk if not possible.
380 		 */
381 		hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT);
382 		if (hc_entry == NULL) {
383 			THC_UNLOCK(&hc_head->hch_mtx);
384 			return NULL;
385 		}
386 	}
387 
388 	/*
389 	 * Initialize basic information of hostcache entry.
390 	 */
391 	bzero(hc_entry, sizeof(*hc_entry));
392 	if (inc->inc_isipv6)
393 		bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6));
394 	else
395 		hc_entry->ip4 = inc->inc_faddr;
396 	hc_entry->rmx_head = hc_head;
397 	hc_entry->rmx_expire = V_tcp_hostcache.expire;
398 
399 	/*
400 	 * Put it upfront.
401 	 */
402 	TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
403 	V_tcp_hostcache.hashbase[hash].hch_length++;
404 	V_tcp_hostcache.cache_count++;
405 	V_tcpstat.tcps_hc_added++;
406 
407 	return hc_entry;
408 }
409 
410 /*
411  * External function: look up an entry in the hostcache and fill out the
412  * supplied TCP metrics structure.  Fills in NULL when no entry was found or
413  * a value is not set.
414  */
415 void
416 tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
417 {
418 	struct hc_metrics *hc_entry;
419 
420 	/*
421 	 * Find the right bucket.
422 	 */
423 	hc_entry = tcp_hc_lookup(inc);
424 
425 	/*
426 	 * If we don't have an existing object.
427 	 */
428 	if (hc_entry == NULL) {
429 		bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
430 		return;
431 	}
432 	hc_entry->rmx_hits++;
433 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
434 
435 	hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
436 	hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
437 	hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
438 	hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
439 	hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth;
440 	hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
441 	hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
442 	hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
443 
444 	/*
445 	 * Unlock bucket row.
446 	 */
447 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
448 }
449 
450 /*
451  * External function: look up an entry in the hostcache and return the
452  * discovered path MTU.  Returns NULL if no entry is found or value is not
453  * set.
454  */
455 u_long
456 tcp_hc_getmtu(struct in_conninfo *inc)
457 {
458 	struct hc_metrics *hc_entry;
459 	u_long mtu;
460 
461 	hc_entry = tcp_hc_lookup(inc);
462 	if (hc_entry == NULL) {
463 		return 0;
464 	}
465 	hc_entry->rmx_hits++;
466 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
467 
468 	mtu = hc_entry->rmx_mtu;
469 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
470 	return mtu;
471 }
472 
473 /*
474  * External function: update the MTU value of an entry in the hostcache.
475  * Creates a new entry if none was found.
476  */
477 void
478 tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
479 {
480 	struct hc_metrics *hc_entry;
481 
482 	/*
483 	 * Find the right bucket.
484 	 */
485 	hc_entry = tcp_hc_lookup(inc);
486 
487 	/*
488 	 * If we don't have an existing object, try to insert a new one.
489 	 */
490 	if (hc_entry == NULL) {
491 		hc_entry = tcp_hc_insert(inc);
492 		if (hc_entry == NULL)
493 			return;
494 	}
495 	hc_entry->rmx_updates++;
496 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
497 
498 	hc_entry->rmx_mtu = mtu;
499 
500 	/*
501 	 * Put it upfront so we find it faster next time.
502 	 */
503 	TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
504 	TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
505 
506 	/*
507 	 * Unlock bucket row.
508 	 */
509 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
510 }
511 
512 /*
513  * External function: update the TCP metrics of an entry in the hostcache.
514  * Creates a new entry if none was found.
515  */
516 void
517 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
518 {
519 	struct hc_metrics *hc_entry;
520 
521 	hc_entry = tcp_hc_lookup(inc);
522 	if (hc_entry == NULL) {
523 		hc_entry = tcp_hc_insert(inc);
524 		if (hc_entry == NULL)
525 			return;
526 	}
527 	hc_entry->rmx_updates++;
528 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
529 
530 	if (hcml->rmx_rtt != 0) {
531 		if (hc_entry->rmx_rtt == 0)
532 			hc_entry->rmx_rtt = hcml->rmx_rtt;
533 		else
534 			hc_entry->rmx_rtt =
535 			    (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
536 		V_tcpstat.tcps_cachedrtt++;
537 	}
538 	if (hcml->rmx_rttvar != 0) {
539 	        if (hc_entry->rmx_rttvar == 0)
540 			hc_entry->rmx_rttvar = hcml->rmx_rttvar;
541 		else
542 			hc_entry->rmx_rttvar =
543 			    (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
544 		V_tcpstat.tcps_cachedrttvar++;
545 	}
546 	if (hcml->rmx_ssthresh != 0) {
547 		if (hc_entry->rmx_ssthresh == 0)
548 			hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
549 		else
550 			hc_entry->rmx_ssthresh =
551 			    (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
552 		V_tcpstat.tcps_cachedssthresh++;
553 	}
554 	if (hcml->rmx_bandwidth != 0) {
555 		if (hc_entry->rmx_bandwidth == 0)
556 			hc_entry->rmx_bandwidth = hcml->rmx_bandwidth;
557 		else
558 			hc_entry->rmx_bandwidth =
559 			    (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2;
560 		/* V_tcpstat.tcps_cachedbandwidth++; */
561 	}
562 	if (hcml->rmx_cwnd != 0) {
563 		if (hc_entry->rmx_cwnd == 0)
564 			hc_entry->rmx_cwnd = hcml->rmx_cwnd;
565 		else
566 			hc_entry->rmx_cwnd =
567 			    (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
568 		/* V_tcpstat.tcps_cachedcwnd++; */
569 	}
570 	if (hcml->rmx_sendpipe != 0) {
571 		if (hc_entry->rmx_sendpipe == 0)
572 			hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
573 		else
574 			hc_entry->rmx_sendpipe =
575 			    (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
576 		/* V_tcpstat.tcps_cachedsendpipe++; */
577 	}
578 	if (hcml->rmx_recvpipe != 0) {
579 		if (hc_entry->rmx_recvpipe == 0)
580 			hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
581 		else
582 			hc_entry->rmx_recvpipe =
583 			    (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
584 		/* V_tcpstat.tcps_cachedrecvpipe++; */
585 	}
586 
587 	TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
588 	TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
589 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
590 }
591 
592 /*
593  * Sysctl function: prints the list and values of all hostcache entries in
594  * unsorted order.
595  */
596 static int
597 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
598 {
599 	int bufsize;
600 	int linesize = 128;
601 	char *p, *buf;
602 	int len, i, error;
603 	struct hc_metrics *hc_entry;
604 #ifdef INET6
605 	char ip6buf[INET6_ADDRSTRLEN];
606 #endif
607 
608 	bufsize = linesize * (V_tcp_hostcache.cache_count + 1);
609 
610 	p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
611 
612 	len = snprintf(p, linesize,
613 		"\nIP address        MTU  SSTRESH      RTT   RTTVAR BANDWIDTH "
614 		"    CWND SENDPIPE RECVPIPE HITS  UPD  EXP\n");
615 	p += len;
616 
617 #define msec(u) (((u) + 500) / 1000)
618 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
619 		THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
620 		TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket,
621 			      rmx_q) {
622 			len = snprintf(p, linesize,
623 			    "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu "
624 			    "%4lu %4lu %4i\n",
625 			    hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
626 #ifdef INET6
627 				ip6_sprintf(ip6buf, &hc_entry->ip6),
628 #else
629 				"IPv6?",
630 #endif
631 			    hc_entry->rmx_mtu,
632 			    hc_entry->rmx_ssthresh,
633 			    msec(hc_entry->rmx_rtt *
634 				(RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
635 			    msec(hc_entry->rmx_rttvar *
636 				(RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
637 			    hc_entry->rmx_bandwidth * 8,
638 			    hc_entry->rmx_cwnd,
639 			    hc_entry->rmx_sendpipe,
640 			    hc_entry->rmx_recvpipe,
641 			    hc_entry->rmx_hits,
642 			    hc_entry->rmx_updates,
643 			    hc_entry->rmx_expire);
644 			p += len;
645 		}
646 		THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
647 	}
648 #undef msec
649 	error = SYSCTL_OUT(req, buf, p - buf);
650 	free(buf, M_TEMP);
651 	return(error);
652 }
653 
654 /*
655  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
656  * periodically from the callout.
657  */
658 static void
659 tcp_hc_purge(void *arg)
660 {
661 	struct hc_metrics *hc_entry, *hc_next;
662 	int all = (intptr_t)arg;
663 	int i;
664 
665 	if (V_tcp_hostcache.purgeall) {
666 		all = 1;
667 		V_tcp_hostcache.purgeall = 0;
668 	}
669 
670 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
671 		THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
672 		TAILQ_FOREACH_SAFE(hc_entry,
673 		    &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) {
674 			if (all || hc_entry->rmx_expire <= 0) {
675 				TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket,
676 					      hc_entry, rmx_q);
677 				uma_zfree(V_tcp_hostcache.zone, hc_entry);
678 				V_tcp_hostcache.hashbase[i].hch_length--;
679 				V_tcp_hostcache.cache_count--;
680 			} else
681 				hc_entry->rmx_expire -= V_tcp_hostcache.prune;
682 		}
683 		THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
684 	}
685 
686 	callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
687 	    tcp_hc_purge, arg);
688 }
689