xref: /freebsd/sys/netinet/tcp_hostcache.c (revision 7d8f797b725e3efc0a4256554654780df83c456c)
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 have 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/sbuf.h>
77 #include <sys/socket.h>
78 #include <sys/socketvar.h>
79 #include <sys/sysctl.h>
80 
81 #include <net/if.h>
82 #include <net/if_var.h>
83 #include <net/route.h>
84 #include <net/vnet.h>
85 
86 #include <netinet/in.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/ip.h>
89 #include <netinet/in_var.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/ip_var.h>
92 #ifdef INET6
93 #include <netinet/ip6.h>
94 #include <netinet6/ip6_var.h>
95 #endif
96 #include <netinet/tcp.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet/tcp_hostcache.h>
99 #ifdef INET6
100 #include <netinet6/tcp6_var.h>
101 #endif
102 
103 #include <vm/uma.h>
104 
105 /* Arbitrary values */
106 #define TCP_HOSTCACHE_HASHSIZE		512
107 #define TCP_HOSTCACHE_BUCKETLIMIT	30
108 #define TCP_HOSTCACHE_EXPIRE		60*60	/* one hour */
109 #define TCP_HOSTCACHE_PRUNE		5*60	/* every 5 minutes */
110 
111 static VNET_DEFINE(struct tcp_hostcache, tcp_hostcache);
112 #define	V_tcp_hostcache		VNET(tcp_hostcache)
113 
114 static VNET_DEFINE(struct callout, tcp_hc_callout);
115 #define	V_tcp_hc_callout	VNET(tcp_hc_callout)
116 
117 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
118 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
119 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
120 static int sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS);
121 static void tcp_hc_purge_internal(int);
122 static void tcp_hc_purge(void *);
123 
124 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0,
125     "TCP Host cache");
126 
127 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_VNET | CTLFLAG_RDTUN,
128     &VNET_NAME(tcp_hostcache.cache_limit), 0,
129     "Overall entry limit for hostcache");
130 
131 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
132     &VNET_NAME(tcp_hostcache.hashsize), 0,
133     "Size of TCP hostcache hashtable");
134 
135 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit,
136     CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.bucket_limit), 0,
137     "Per-bucket hash limit for hostcache");
138 
139 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_VNET | CTLFLAG_RD,
140      &VNET_NAME(tcp_hostcache.cache_count), 0,
141     "Current number of entries in hostcache");
142 
143 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_VNET | CTLFLAG_RW,
144     &VNET_NAME(tcp_hostcache.expire), 0,
145     "Expire time of TCP hostcache entries");
146 
147 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_VNET | CTLFLAG_RW,
148     &VNET_NAME(tcp_hostcache.prune), 0,
149     "Time between purge runs");
150 
151 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_VNET | CTLFLAG_RW,
152     &VNET_NAME(tcp_hostcache.purgeall), 0,
153     "Expire all entires on next purge run");
154 
155 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
156     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
157     sysctl_tcp_hc_list, "A", "List of all hostcache entries");
158 
159 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, purgenow,
160     CTLTYPE_INT | CTLFLAG_RW, NULL, 0,
161     sysctl_tcp_hc_purgenow, "I", "Immediately purge all entries");
162 
163 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
164 
165 #define HOSTCACHE_HASH(ip) \
166 	(((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) &	\
167 	  V_tcp_hostcache.hashmask)
168 
169 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
170 #define HOSTCACHE_HASH6(ip6)				\
171 	(((ip6)->s6_addr32[0] ^				\
172 	  (ip6)->s6_addr32[1] ^				\
173 	  (ip6)->s6_addr32[2] ^				\
174 	  (ip6)->s6_addr32[3]) &			\
175 	 V_tcp_hostcache.hashmask)
176 
177 #define THC_LOCK(lp)		mtx_lock(lp)
178 #define THC_UNLOCK(lp)		mtx_unlock(lp)
179 
180 void
181 tcp_hc_init(void)
182 {
183 	u_int cache_limit;
184 	int i;
185 
186 	/*
187 	 * Initialize hostcache structures.
188 	 */
189 	V_tcp_hostcache.cache_count = 0;
190 	V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
191 	V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
192 	V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
193 	V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
194 
195 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
196 	    &V_tcp_hostcache.hashsize);
197 	if (!powerof2(V_tcp_hostcache.hashsize)) {
198 		printf("WARNING: hostcache hash size is not a power of 2.\n");
199 		V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
200 	}
201 	V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1;
202 
203 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
204 	    &V_tcp_hostcache.bucket_limit);
205 
206 	cache_limit = V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit;
207 	V_tcp_hostcache.cache_limit = cache_limit;
208 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
209 	    &V_tcp_hostcache.cache_limit);
210 	if (V_tcp_hostcache.cache_limit > cache_limit)
211 		V_tcp_hostcache.cache_limit = cache_limit;
212 
213 	/*
214 	 * Allocate the hash table.
215 	 */
216 	V_tcp_hostcache.hashbase = (struct hc_head *)
217 	    malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head),
218 		   M_HOSTCACHE, M_WAITOK | M_ZERO);
219 
220 	/*
221 	 * Initialize the hash buckets.
222 	 */
223 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
224 		TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket);
225 		V_tcp_hostcache.hashbase[i].hch_length = 0;
226 		mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
227 			  NULL, MTX_DEF);
228 	}
229 
230 	/*
231 	 * Allocate the hostcache entries.
232 	 */
233 	V_tcp_hostcache.zone =
234 	    uma_zcreate("hostcache", sizeof(struct hc_metrics),
235 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
236 	uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit);
237 
238 	/*
239 	 * Set up periodic cache cleanup.
240 	 */
241 	callout_init(&V_tcp_hc_callout, 1);
242 	callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
243 	    tcp_hc_purge, curvnet);
244 }
245 
246 #ifdef VIMAGE
247 void
248 tcp_hc_destroy(void)
249 {
250 	int i;
251 
252 	callout_drain(&V_tcp_hc_callout);
253 
254 	/* Purge all hc entries. */
255 	tcp_hc_purge_internal(1);
256 
257 	/* Free the uma zone and the allocated hash table. */
258 	uma_zdestroy(V_tcp_hostcache.zone);
259 
260 	for (i = 0; i < V_tcp_hostcache.hashsize; i++)
261 		mtx_destroy(&V_tcp_hostcache.hashbase[i].hch_mtx);
262 	free(V_tcp_hostcache.hashbase, M_HOSTCACHE);
263 }
264 #endif
265 
266 /*
267  * Internal function: look up an entry in the hostcache or return NULL.
268  *
269  * If an entry has been returned, the caller becomes responsible for
270  * unlocking the bucket row after he is done reading/modifying the entry.
271  */
272 static struct hc_metrics *
273 tcp_hc_lookup(struct in_conninfo *inc)
274 {
275 	int hash;
276 	struct hc_head *hc_head;
277 	struct hc_metrics *hc_entry;
278 
279 	KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer"));
280 
281 	/*
282 	 * Hash the foreign ip address.
283 	 */
284 	if (inc->inc_flags & INC_ISIPV6)
285 		hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
286 	else
287 		hash = HOSTCACHE_HASH(&inc->inc_faddr);
288 
289 	hc_head = &V_tcp_hostcache.hashbase[hash];
290 
291 	/*
292 	 * Acquire lock for this bucket row; we release the lock if we don't
293 	 * find an entry, otherwise the caller has to unlock after he is
294 	 * done.
295 	 */
296 	THC_LOCK(&hc_head->hch_mtx);
297 
298 	/*
299 	 * Iterate through entries in bucket row looking for a match.
300 	 */
301 	TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) {
302 		if (inc->inc_flags & INC_ISIPV6) {
303 			/* XXX: check ip6_zoneid */
304 			if (memcmp(&inc->inc6_faddr, &hc_entry->ip6,
305 			    sizeof(inc->inc6_faddr)) == 0)
306 				return hc_entry;
307 		} else {
308 			if (memcmp(&inc->inc_faddr, &hc_entry->ip4,
309 			    sizeof(inc->inc_faddr)) == 0)
310 				return hc_entry;
311 		}
312 	}
313 
314 	/*
315 	 * We were unsuccessful and didn't find anything.
316 	 */
317 	THC_UNLOCK(&hc_head->hch_mtx);
318 	return NULL;
319 }
320 
321 /*
322  * Internal function: insert an entry into the hostcache or return NULL if
323  * unable to allocate a new one.
324  *
325  * If an entry has been returned, the caller becomes responsible for
326  * unlocking the bucket row after he is done reading/modifying the entry.
327  */
328 static struct hc_metrics *
329 tcp_hc_insert(struct in_conninfo *inc)
330 {
331 	int hash;
332 	struct hc_head *hc_head;
333 	struct hc_metrics *hc_entry;
334 
335 	KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer"));
336 
337 	/*
338 	 * Hash the foreign ip address.
339 	 */
340 	if (inc->inc_flags & INC_ISIPV6)
341 		hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
342 	else
343 		hash = HOSTCACHE_HASH(&inc->inc_faddr);
344 
345 	hc_head = &V_tcp_hostcache.hashbase[hash];
346 
347 	/*
348 	 * Acquire lock for this bucket row; we release the lock if we don't
349 	 * find an entry, otherwise the caller has to unlock after he is
350 	 * done.
351 	 */
352 	THC_LOCK(&hc_head->hch_mtx);
353 
354 	/*
355 	 * If the bucket limit is reached, reuse the least-used element.
356 	 */
357 	if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit ||
358 	    V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) {
359 		hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead);
360 		/*
361 		 * At first we were dropping the last element, just to
362 		 * reacquire it in the next two lines again, which isn't very
363 		 * efficient.  Instead just reuse the least used element.
364 		 * We may drop something that is still "in-use" but we can be
365 		 * "lossy".
366 		 * Just give up if this bucket row is empty and we don't have
367 		 * anything to replace.
368 		 */
369 		if (hc_entry == NULL) {
370 			THC_UNLOCK(&hc_head->hch_mtx);
371 			return NULL;
372 		}
373 		TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
374 		V_tcp_hostcache.hashbase[hash].hch_length--;
375 		V_tcp_hostcache.cache_count--;
376 		TCPSTAT_INC(tcps_hc_bucketoverflow);
377 #if 0
378 		uma_zfree(V_tcp_hostcache.zone, hc_entry);
379 #endif
380 	} else {
381 		/*
382 		 * Allocate a new entry, or balk if not possible.
383 		 */
384 		hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT);
385 		if (hc_entry == NULL) {
386 			THC_UNLOCK(&hc_head->hch_mtx);
387 			return NULL;
388 		}
389 	}
390 
391 	/*
392 	 * Initialize basic information of hostcache entry.
393 	 */
394 	bzero(hc_entry, sizeof(*hc_entry));
395 	if (inc->inc_flags & INC_ISIPV6) {
396 		hc_entry->ip6 = inc->inc6_faddr;
397 		hc_entry->ip6_zoneid = inc->inc6_zoneid;
398 	} else
399 		hc_entry->ip4 = inc->inc_faddr;
400 	hc_entry->rmx_head = hc_head;
401 	hc_entry->rmx_expire = V_tcp_hostcache.expire;
402 
403 	/*
404 	 * Put it upfront.
405 	 */
406 	TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
407 	V_tcp_hostcache.hashbase[hash].hch_length++;
408 	V_tcp_hostcache.cache_count++;
409 	TCPSTAT_INC(tcps_hc_added);
410 
411 	return hc_entry;
412 }
413 
414 /*
415  * External function: look up an entry in the hostcache and fill out the
416  * supplied TCP metrics structure.  Fills in NULL when no entry was found or
417  * a value is not set.
418  */
419 void
420 tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
421 {
422 	struct hc_metrics *hc_entry;
423 
424 	/*
425 	 * Find the right bucket.
426 	 */
427 	hc_entry = tcp_hc_lookup(inc);
428 
429 	/*
430 	 * If we don't have an existing object.
431 	 */
432 	if (hc_entry == NULL) {
433 		bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
434 		return;
435 	}
436 	hc_entry->rmx_hits++;
437 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
438 
439 	hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
440 	hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
441 	hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
442 	hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
443 	hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth;
444 	hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
445 	hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
446 	hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
447 
448 	/*
449 	 * Unlock bucket row.
450 	 */
451 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
452 }
453 
454 /*
455  * External function: look up an entry in the hostcache and return the
456  * discovered path MTU.  Returns NULL if no entry is found or value is not
457  * set.
458  */
459 u_long
460 tcp_hc_getmtu(struct in_conninfo *inc)
461 {
462 	struct hc_metrics *hc_entry;
463 	u_long mtu;
464 
465 	hc_entry = tcp_hc_lookup(inc);
466 	if (hc_entry == NULL) {
467 		return 0;
468 	}
469 	hc_entry->rmx_hits++;
470 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
471 
472 	mtu = hc_entry->rmx_mtu;
473 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
474 	return mtu;
475 }
476 
477 /*
478  * External function: update the MTU value of an entry in the hostcache.
479  * Creates a new entry if none was found.
480  */
481 void
482 tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
483 {
484 	struct hc_metrics *hc_entry;
485 
486 	/*
487 	 * Find the right bucket.
488 	 */
489 	hc_entry = tcp_hc_lookup(inc);
490 
491 	/*
492 	 * If we don't have an existing object, try to insert a new one.
493 	 */
494 	if (hc_entry == NULL) {
495 		hc_entry = tcp_hc_insert(inc);
496 		if (hc_entry == NULL)
497 			return;
498 	}
499 	hc_entry->rmx_updates++;
500 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
501 
502 	hc_entry->rmx_mtu = mtu;
503 
504 	/*
505 	 * Put it upfront so we find it faster next time.
506 	 */
507 	TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
508 	TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
509 
510 	/*
511 	 * Unlock bucket row.
512 	 */
513 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
514 }
515 
516 /*
517  * External function: update the TCP metrics of an entry in the hostcache.
518  * Creates a new entry if none was found.
519  */
520 void
521 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
522 {
523 	struct hc_metrics *hc_entry;
524 
525 	hc_entry = tcp_hc_lookup(inc);
526 	if (hc_entry == NULL) {
527 		hc_entry = tcp_hc_insert(inc);
528 		if (hc_entry == NULL)
529 			return;
530 	}
531 	hc_entry->rmx_updates++;
532 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
533 
534 	if (hcml->rmx_rtt != 0) {
535 		if (hc_entry->rmx_rtt == 0)
536 			hc_entry->rmx_rtt = hcml->rmx_rtt;
537 		else
538 			hc_entry->rmx_rtt =
539 			    (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
540 		TCPSTAT_INC(tcps_cachedrtt);
541 	}
542 	if (hcml->rmx_rttvar != 0) {
543 	        if (hc_entry->rmx_rttvar == 0)
544 			hc_entry->rmx_rttvar = hcml->rmx_rttvar;
545 		else
546 			hc_entry->rmx_rttvar =
547 			    (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
548 		TCPSTAT_INC(tcps_cachedrttvar);
549 	}
550 	if (hcml->rmx_ssthresh != 0) {
551 		if (hc_entry->rmx_ssthresh == 0)
552 			hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
553 		else
554 			hc_entry->rmx_ssthresh =
555 			    (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
556 		TCPSTAT_INC(tcps_cachedssthresh);
557 	}
558 	if (hcml->rmx_bandwidth != 0) {
559 		if (hc_entry->rmx_bandwidth == 0)
560 			hc_entry->rmx_bandwidth = hcml->rmx_bandwidth;
561 		else
562 			hc_entry->rmx_bandwidth =
563 			    (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2;
564 		/* TCPSTAT_INC(tcps_cachedbandwidth); */
565 	}
566 	if (hcml->rmx_cwnd != 0) {
567 		if (hc_entry->rmx_cwnd == 0)
568 			hc_entry->rmx_cwnd = hcml->rmx_cwnd;
569 		else
570 			hc_entry->rmx_cwnd =
571 			    (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
572 		/* TCPSTAT_INC(tcps_cachedcwnd); */
573 	}
574 	if (hcml->rmx_sendpipe != 0) {
575 		if (hc_entry->rmx_sendpipe == 0)
576 			hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
577 		else
578 			hc_entry->rmx_sendpipe =
579 			    (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
580 		/* TCPSTAT_INC(tcps_cachedsendpipe); */
581 	}
582 	if (hcml->rmx_recvpipe != 0) {
583 		if (hc_entry->rmx_recvpipe == 0)
584 			hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
585 		else
586 			hc_entry->rmx_recvpipe =
587 			    (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
588 		/* TCPSTAT_INC(tcps_cachedrecvpipe); */
589 	}
590 
591 	TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
592 	TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
593 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
594 }
595 
596 /*
597  * Sysctl function: prints the list and values of all hostcache entries in
598  * unsorted order.
599  */
600 static int
601 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
602 {
603 	const int linesize = 128;
604 	struct sbuf sb;
605 	int i, error;
606 	struct hc_metrics *hc_entry;
607 #ifdef INET6
608 	char ip6buf[INET6_ADDRSTRLEN];
609 #endif
610 
611 	sbuf_new(&sb, NULL, linesize * (V_tcp_hostcache.cache_count + 1),
612 		SBUF_INCLUDENUL);
613 
614 	sbuf_printf(&sb,
615 	        "\nIP address        MTU  SSTRESH      RTT   RTTVAR BANDWIDTH "
616 		"    CWND SENDPIPE RECVPIPE HITS  UPD  EXP\n");
617 
618 #define msec(u) (((u) + 500) / 1000)
619 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
620 		THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
621 		TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket,
622 			      rmx_q) {
623 			sbuf_printf(&sb,
624 			    "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu "
625 			    "%4lu %4lu %4i\n",
626 			    hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
627 #ifdef INET6
628 				ip6_sprintf(ip6buf, &hc_entry->ip6),
629 #else
630 				"IPv6?",
631 #endif
632 			    hc_entry->rmx_mtu,
633 			    hc_entry->rmx_ssthresh,
634 			    msec(hc_entry->rmx_rtt *
635 				(RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
636 			    msec(hc_entry->rmx_rttvar *
637 				(RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE))),
638 			    hc_entry->rmx_bandwidth * 8,
639 			    hc_entry->rmx_cwnd,
640 			    hc_entry->rmx_sendpipe,
641 			    hc_entry->rmx_recvpipe,
642 			    hc_entry->rmx_hits,
643 			    hc_entry->rmx_updates,
644 			    hc_entry->rmx_expire);
645 		}
646 		THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
647 	}
648 #undef msec
649 	error = sbuf_finish(&sb);
650 	if (error == 0)
651 		error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
652 	sbuf_delete(&sb);
653 	return(error);
654 }
655 
656 /*
657  * Caller has to make sure the curvnet is set properly.
658  */
659 static void
660 tcp_hc_purge_internal(int all)
661 {
662 	struct hc_metrics *hc_entry, *hc_next;
663 	int i;
664 
665 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
666 		THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
667 		TAILQ_FOREACH_SAFE(hc_entry,
668 		    &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) {
669 			if (all || hc_entry->rmx_expire <= 0) {
670 				TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket,
671 					      hc_entry, rmx_q);
672 				uma_zfree(V_tcp_hostcache.zone, hc_entry);
673 				V_tcp_hostcache.hashbase[i].hch_length--;
674 				V_tcp_hostcache.cache_count--;
675 			} else
676 				hc_entry->rmx_expire -= V_tcp_hostcache.prune;
677 		}
678 		THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
679 	}
680 }
681 
682 /*
683  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
684  * periodically from the callout.
685  */
686 static void
687 tcp_hc_purge(void *arg)
688 {
689 	CURVNET_SET((struct vnet *) arg);
690 	int all = 0;
691 
692 	if (V_tcp_hostcache.purgeall) {
693 		all = 1;
694 		V_tcp_hostcache.purgeall = 0;
695 	}
696 
697 	tcp_hc_purge_internal(all);
698 
699 	callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
700 	    tcp_hc_purge, arg);
701 	CURVNET_RESTORE();
702 }
703 
704 /*
705  * Expire and purge all entries in hostcache immediately.
706  */
707 static int
708 sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS)
709 {
710 	int error, val;
711 
712 	val = 0;
713 	error = sysctl_handle_int(oidp, &val, 0, req);
714 	if (error || !req->newptr)
715 		return (error);
716 
717 	tcp_hc_purge_internal(1);
718 
719 	callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
720 	    tcp_hc_purge, curvnet);
721 
722 	return (0);
723 }
724