xref: /freebsd/sys/netinet/tcp_hostcache.c (revision 6f9c8e5b074419423648ffb89b83fd2f257e90b7)
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/socket.h>
77 #include <sys/socketvar.h>
78 #include <sys/sysctl.h>
79 
80 #include <net/if.h>
81 #include <net/route.h>
82 #include <net/vnet.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/ip.h>
87 #include <netinet/in_var.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/ip_var.h>
90 #ifdef INET6
91 #include <netinet/ip6.h>
92 #include <netinet6/ip6_var.h>
93 #endif
94 #include <netinet/tcp.h>
95 #include <netinet/tcp_var.h>
96 #include <netinet/tcp_hostcache.h>
97 #ifdef INET6
98 #include <netinet6/tcp6_var.h>
99 #endif
100 
101 #include <vm/uma.h>
102 
103 /* Arbitrary values */
104 #define TCP_HOSTCACHE_HASHSIZE		512
105 #define TCP_HOSTCACHE_BUCKETLIMIT	30
106 #define TCP_HOSTCACHE_EXPIRE		60*60	/* one hour */
107 #define TCP_HOSTCACHE_PRUNE		5*60	/* every 5 minutes */
108 
109 static VNET_DEFINE(struct tcp_hostcache, tcp_hostcache);
110 #define	V_tcp_hostcache		VNET(tcp_hostcache)
111 
112 static VNET_DEFINE(struct callout, tcp_hc_callout);
113 #define	V_tcp_hc_callout	VNET(tcp_hc_callout)
114 
115 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
116 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
117 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
118 static void tcp_hc_purge_internal(int);
119 static void tcp_hc_purge(void *);
120 
121 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0,
122     "TCP Host cache");
123 
124 SYSCTL_VNET_UINT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
125     &VNET_NAME(tcp_hostcache.cache_limit), 0,
126     "Overall entry limit for hostcache");
127 
128 SYSCTL_VNET_UINT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
129     &VNET_NAME(tcp_hostcache.hashsize), 0,
130     "Size of TCP hostcache hashtable");
131 
132 SYSCTL_VNET_UINT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit,
133     CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.bucket_limit), 0,
134     "Per-bucket hash limit for hostcache");
135 
136 SYSCTL_VNET_UINT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD,
137      &VNET_NAME(tcp_hostcache.cache_count), 0,
138     "Current number of entries in hostcache");
139 
140 SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW,
141     &VNET_NAME(tcp_hostcache.expire), 0,
142     "Expire time of TCP hostcache entries");
143 
144 SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW,
145     &VNET_NAME(tcp_hostcache.prune), 0,
146     "Time between purge runs");
147 
148 SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW,
149     &VNET_NAME(tcp_hostcache.purgeall), 0,
150     "Expire all entires on next purge run");
151 
152 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
153     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
154     sysctl_tcp_hc_list, "A", "List of all hostcache entries");
155 
156 
157 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
158 
159 #define HOSTCACHE_HASH(ip) \
160 	(((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) &	\
161 	  V_tcp_hostcache.hashmask)
162 
163 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
164 #define HOSTCACHE_HASH6(ip6)				\
165 	(((ip6)->s6_addr32[0] ^				\
166 	  (ip6)->s6_addr32[1] ^				\
167 	  (ip6)->s6_addr32[2] ^				\
168 	  (ip6)->s6_addr32[3]) &			\
169 	 V_tcp_hostcache.hashmask)
170 
171 #define THC_LOCK(lp)		mtx_lock(lp)
172 #define THC_UNLOCK(lp)		mtx_unlock(lp)
173 
174 void
175 tcp_hc_init(void)
176 {
177 	int i;
178 
179 	/*
180 	 * Initialize hostcache structures.
181 	 */
182 	V_tcp_hostcache.cache_count = 0;
183 	V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
184 	V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
185 	V_tcp_hostcache.cache_limit =
186 	    V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit;
187 	V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
188 	V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
189 
190 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
191 	    &V_tcp_hostcache.hashsize);
192 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
193 	    &V_tcp_hostcache.cache_limit);
194 	TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
195 	    &V_tcp_hostcache.bucket_limit);
196 	if (!powerof2(V_tcp_hostcache.hashsize)) {
197 		printf("WARNING: hostcache hash size is not a power of 2.\n");
198 		V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
199 	}
200 	V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1;
201 
202 	/*
203 	 * Allocate the hash table.
204 	 */
205 	V_tcp_hostcache.hashbase = (struct hc_head *)
206 	    malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head),
207 		   M_HOSTCACHE, M_WAITOK | M_ZERO);
208 
209 	/*
210 	 * Initialize the hash buckets.
211 	 */
212 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
213 		TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket);
214 		V_tcp_hostcache.hashbase[i].hch_length = 0;
215 		mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
216 			  NULL, MTX_DEF);
217 	}
218 
219 	/*
220 	 * Allocate the hostcache entries.
221 	 */
222 	V_tcp_hostcache.zone =
223 	    uma_zcreate("hostcache", sizeof(struct hc_metrics),
224 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
225 	uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit);
226 
227 	/*
228 	 * Set up periodic cache cleanup.
229 	 */
230 	callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE);
231 	callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
232 	    tcp_hc_purge, curvnet);
233 }
234 
235 #ifdef VIMAGE
236 void
237 tcp_hc_destroy(void)
238 {
239 	int i;
240 
241 	callout_drain(&V_tcp_hc_callout);
242 
243 	/* Purge all hc entries. */
244 	tcp_hc_purge_internal(1);
245 
246 	/* Free the uma zone and the allocated hash table. */
247 	uma_zdestroy(V_tcp_hostcache.zone);
248 
249 	for (i = 0; i < V_tcp_hostcache.hashsize; i++)
250 		mtx_destroy(&V_tcp_hostcache.hashbase[i].hch_mtx);
251 	free(V_tcp_hostcache.hashbase, M_HOSTCACHE);
252 }
253 #endif
254 
255 /*
256  * Internal function: look up an entry in the hostcache or return NULL.
257  *
258  * If an entry has been returned, the caller becomes responsible for
259  * unlocking the bucket row after he is done reading/modifying the entry.
260  */
261 static struct hc_metrics *
262 tcp_hc_lookup(struct in_conninfo *inc)
263 {
264 	int hash;
265 	struct hc_head *hc_head;
266 	struct hc_metrics *hc_entry;
267 
268 	KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer"));
269 
270 	/*
271 	 * Hash the foreign ip address.
272 	 */
273 	if (inc->inc_flags & INC_ISIPV6)
274 		hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
275 	else
276 		hash = HOSTCACHE_HASH(&inc->inc_faddr);
277 
278 	hc_head = &V_tcp_hostcache.hashbase[hash];
279 
280 	/*
281 	 * Acquire lock for this bucket row; we release the lock if we don't
282 	 * find an entry, otherwise the caller has to unlock after he is
283 	 * done.
284 	 */
285 	THC_LOCK(&hc_head->hch_mtx);
286 
287 	/*
288 	 * Iterate through entries in bucket row looking for a match.
289 	 */
290 	TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) {
291 		if (inc->inc_flags & INC_ISIPV6) {
292 			if (memcmp(&inc->inc6_faddr, &hc_entry->ip6,
293 			    sizeof(inc->inc6_faddr)) == 0)
294 				return hc_entry;
295 		} else {
296 			if (memcmp(&inc->inc_faddr, &hc_entry->ip4,
297 			    sizeof(inc->inc_faddr)) == 0)
298 				return hc_entry;
299 		}
300 	}
301 
302 	/*
303 	 * We were unsuccessful and didn't find anything.
304 	 */
305 	THC_UNLOCK(&hc_head->hch_mtx);
306 	return NULL;
307 }
308 
309 /*
310  * Internal function: insert an entry into the hostcache or return NULL if
311  * unable to allocate a new one.
312  *
313  * If an entry has been returned, the caller becomes responsible for
314  * unlocking the bucket row after he is done reading/modifying the entry.
315  */
316 static struct hc_metrics *
317 tcp_hc_insert(struct in_conninfo *inc)
318 {
319 	int hash;
320 	struct hc_head *hc_head;
321 	struct hc_metrics *hc_entry;
322 
323 	KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer"));
324 
325 	/*
326 	 * Hash the foreign ip address.
327 	 */
328 	if (inc->inc_flags & INC_ISIPV6)
329 		hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
330 	else
331 		hash = HOSTCACHE_HASH(&inc->inc_faddr);
332 
333 	hc_head = &V_tcp_hostcache.hashbase[hash];
334 
335 	/*
336 	 * Acquire lock for this bucket row; we release the lock if we don't
337 	 * find an entry, otherwise the caller has to unlock after he is
338 	 * done.
339 	 */
340 	THC_LOCK(&hc_head->hch_mtx);
341 
342 	/*
343 	 * If the bucket limit is reached, reuse the least-used element.
344 	 */
345 	if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit ||
346 	    V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) {
347 		hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead);
348 		/*
349 		 * At first we were dropping the last element, just to
350 		 * reacquire it in the next two lines again, which isn't very
351 		 * efficient.  Instead just reuse the least used element.
352 		 * We may drop something that is still "in-use" but we can be
353 		 * "lossy".
354 		 * Just give up if this bucket row is empty and we don't have
355 		 * anything to replace.
356 		 */
357 		if (hc_entry == NULL) {
358 			THC_UNLOCK(&hc_head->hch_mtx);
359 			return NULL;
360 		}
361 		TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
362 		V_tcp_hostcache.hashbase[hash].hch_length--;
363 		V_tcp_hostcache.cache_count--;
364 		TCPSTAT_INC(tcps_hc_bucketoverflow);
365 #if 0
366 		uma_zfree(V_tcp_hostcache.zone, hc_entry);
367 #endif
368 	} else {
369 		/*
370 		 * Allocate a new entry, or balk if not possible.
371 		 */
372 		hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT);
373 		if (hc_entry == NULL) {
374 			THC_UNLOCK(&hc_head->hch_mtx);
375 			return NULL;
376 		}
377 	}
378 
379 	/*
380 	 * Initialize basic information of hostcache entry.
381 	 */
382 	bzero(hc_entry, sizeof(*hc_entry));
383 	if (inc->inc_flags & INC_ISIPV6)
384 		bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6));
385 	else
386 		hc_entry->ip4 = inc->inc_faddr;
387 	hc_entry->rmx_head = hc_head;
388 	hc_entry->rmx_expire = V_tcp_hostcache.expire;
389 
390 	/*
391 	 * Put it upfront.
392 	 */
393 	TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
394 	V_tcp_hostcache.hashbase[hash].hch_length++;
395 	V_tcp_hostcache.cache_count++;
396 	TCPSTAT_INC(tcps_hc_added);
397 
398 	return hc_entry;
399 }
400 
401 /*
402  * External function: look up an entry in the hostcache and fill out the
403  * supplied TCP metrics structure.  Fills in NULL when no entry was found or
404  * a value is not set.
405  */
406 void
407 tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
408 {
409 	struct hc_metrics *hc_entry;
410 
411 	/*
412 	 * Find the right bucket.
413 	 */
414 	hc_entry = tcp_hc_lookup(inc);
415 
416 	/*
417 	 * If we don't have an existing object.
418 	 */
419 	if (hc_entry == NULL) {
420 		bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
421 		return;
422 	}
423 	hc_entry->rmx_hits++;
424 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
425 
426 	hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
427 	hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
428 	hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
429 	hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
430 	hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth;
431 	hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
432 	hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
433 	hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
434 
435 	/*
436 	 * Unlock bucket row.
437 	 */
438 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
439 }
440 
441 /*
442  * External function: look up an entry in the hostcache and return the
443  * discovered path MTU.  Returns NULL if no entry is found or value is not
444  * set.
445  */
446 u_long
447 tcp_hc_getmtu(struct in_conninfo *inc)
448 {
449 	struct hc_metrics *hc_entry;
450 	u_long mtu;
451 
452 	hc_entry = tcp_hc_lookup(inc);
453 	if (hc_entry == NULL) {
454 		return 0;
455 	}
456 	hc_entry->rmx_hits++;
457 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
458 
459 	mtu = hc_entry->rmx_mtu;
460 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
461 	return mtu;
462 }
463 
464 /*
465  * External function: update the MTU value of an entry in the hostcache.
466  * Creates a new entry if none was found.
467  */
468 void
469 tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
470 {
471 	struct hc_metrics *hc_entry;
472 
473 	/*
474 	 * Find the right bucket.
475 	 */
476 	hc_entry = tcp_hc_lookup(inc);
477 
478 	/*
479 	 * If we don't have an existing object, try to insert a new one.
480 	 */
481 	if (hc_entry == NULL) {
482 		hc_entry = tcp_hc_insert(inc);
483 		if (hc_entry == NULL)
484 			return;
485 	}
486 	hc_entry->rmx_updates++;
487 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
488 
489 	hc_entry->rmx_mtu = mtu;
490 
491 	/*
492 	 * Put it upfront so we find it faster next time.
493 	 */
494 	TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
495 	TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
496 
497 	/*
498 	 * Unlock bucket row.
499 	 */
500 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
501 }
502 
503 /*
504  * External function: update the TCP metrics of an entry in the hostcache.
505  * Creates a new entry if none was found.
506  */
507 void
508 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
509 {
510 	struct hc_metrics *hc_entry;
511 
512 	hc_entry = tcp_hc_lookup(inc);
513 	if (hc_entry == NULL) {
514 		hc_entry = tcp_hc_insert(inc);
515 		if (hc_entry == NULL)
516 			return;
517 	}
518 	hc_entry->rmx_updates++;
519 	hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
520 
521 	if (hcml->rmx_rtt != 0) {
522 		if (hc_entry->rmx_rtt == 0)
523 			hc_entry->rmx_rtt = hcml->rmx_rtt;
524 		else
525 			hc_entry->rmx_rtt =
526 			    (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
527 		TCPSTAT_INC(tcps_cachedrtt);
528 	}
529 	if (hcml->rmx_rttvar != 0) {
530 	        if (hc_entry->rmx_rttvar == 0)
531 			hc_entry->rmx_rttvar = hcml->rmx_rttvar;
532 		else
533 			hc_entry->rmx_rttvar =
534 			    (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
535 		TCPSTAT_INC(tcps_cachedrttvar);
536 	}
537 	if (hcml->rmx_ssthresh != 0) {
538 		if (hc_entry->rmx_ssthresh == 0)
539 			hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
540 		else
541 			hc_entry->rmx_ssthresh =
542 			    (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
543 		TCPSTAT_INC(tcps_cachedssthresh);
544 	}
545 	if (hcml->rmx_bandwidth != 0) {
546 		if (hc_entry->rmx_bandwidth == 0)
547 			hc_entry->rmx_bandwidth = hcml->rmx_bandwidth;
548 		else
549 			hc_entry->rmx_bandwidth =
550 			    (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2;
551 		/* TCPSTAT_INC(tcps_cachedbandwidth); */
552 	}
553 	if (hcml->rmx_cwnd != 0) {
554 		if (hc_entry->rmx_cwnd == 0)
555 			hc_entry->rmx_cwnd = hcml->rmx_cwnd;
556 		else
557 			hc_entry->rmx_cwnd =
558 			    (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
559 		/* TCPSTAT_INC(tcps_cachedcwnd); */
560 	}
561 	if (hcml->rmx_sendpipe != 0) {
562 		if (hc_entry->rmx_sendpipe == 0)
563 			hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
564 		else
565 			hc_entry->rmx_sendpipe =
566 			    (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
567 		/* TCPSTAT_INC(tcps_cachedsendpipe); */
568 	}
569 	if (hcml->rmx_recvpipe != 0) {
570 		if (hc_entry->rmx_recvpipe == 0)
571 			hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
572 		else
573 			hc_entry->rmx_recvpipe =
574 			    (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
575 		/* TCPSTAT_INC(tcps_cachedrecvpipe); */
576 	}
577 
578 	TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
579 	TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
580 	THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
581 }
582 
583 /*
584  * Sysctl function: prints the list and values of all hostcache entries in
585  * unsorted order.
586  */
587 static int
588 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
589 {
590 	int bufsize;
591 	int linesize = 128;
592 	char *p, *buf;
593 	int len, i, error;
594 	struct hc_metrics *hc_entry;
595 #ifdef INET6
596 	char ip6buf[INET6_ADDRSTRLEN];
597 #endif
598 
599 	bufsize = linesize * (V_tcp_hostcache.cache_count + 1);
600 
601 	p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
602 
603 	len = snprintf(p, linesize,
604 		"\nIP address        MTU  SSTRESH      RTT   RTTVAR BANDWIDTH "
605 		"    CWND SENDPIPE RECVPIPE HITS  UPD  EXP\n");
606 	p += len;
607 
608 #define msec(u) (((u) + 500) / 1000)
609 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
610 		THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
611 		TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket,
612 			      rmx_q) {
613 			len = snprintf(p, linesize,
614 			    "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu "
615 			    "%4lu %4lu %4i\n",
616 			    hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
617 #ifdef INET6
618 				ip6_sprintf(ip6buf, &hc_entry->ip6),
619 #else
620 				"IPv6?",
621 #endif
622 			    hc_entry->rmx_mtu,
623 			    hc_entry->rmx_ssthresh,
624 			    msec(hc_entry->rmx_rtt *
625 				(RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
626 			    msec(hc_entry->rmx_rttvar *
627 				(RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
628 			    hc_entry->rmx_bandwidth * 8,
629 			    hc_entry->rmx_cwnd,
630 			    hc_entry->rmx_sendpipe,
631 			    hc_entry->rmx_recvpipe,
632 			    hc_entry->rmx_hits,
633 			    hc_entry->rmx_updates,
634 			    hc_entry->rmx_expire);
635 			p += len;
636 		}
637 		THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
638 	}
639 #undef msec
640 	error = SYSCTL_OUT(req, buf, p - buf);
641 	free(buf, M_TEMP);
642 	return(error);
643 }
644 
645 /*
646  * Caller has to make sure the curvnet is set properly.
647  */
648 static void
649 tcp_hc_purge_internal(int all)
650 {
651 	struct hc_metrics *hc_entry, *hc_next;
652 	int i;
653 
654 	for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
655 		THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
656 		TAILQ_FOREACH_SAFE(hc_entry,
657 		    &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) {
658 			if (all || hc_entry->rmx_expire <= 0) {
659 				TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket,
660 					      hc_entry, rmx_q);
661 				uma_zfree(V_tcp_hostcache.zone, hc_entry);
662 				V_tcp_hostcache.hashbase[i].hch_length--;
663 				V_tcp_hostcache.cache_count--;
664 			} else
665 				hc_entry->rmx_expire -= V_tcp_hostcache.prune;
666 		}
667 		THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
668 	}
669 }
670 
671 /*
672  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
673  * periodically from the callout.
674  */
675 static void
676 tcp_hc_purge(void *arg)
677 {
678 	CURVNET_SET((struct vnet *) arg);
679 	int all = 0;
680 
681 	if (V_tcp_hostcache.purgeall) {
682 		all = 1;
683 		V_tcp_hostcache.purgeall = 0;
684 	}
685 
686 	tcp_hc_purge_internal(all);
687 
688 	callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
689 	    tcp_hc_purge, arg);
690 	CURVNET_RESTORE();
691 }
692