]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/netinet/tcp_hostcache.c
MFV r302218: file 5.28.
[FreeBSD/FreeBSD.git] / sys / netinet / tcp_hostcache.c
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, max MTU,
36  * 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_cwnd = hc_entry->rmx_cwnd;
444         hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
445         hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
446
447         /*
448          * Unlock bucket row.
449          */
450         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
451 }
452
453 /*
454  * External function: look up an entry in the hostcache and return the
455  * discovered path MTU.  Returns NULL if no entry is found or value is not
456  * set.
457  */
458 u_long
459 tcp_hc_getmtu(struct in_conninfo *inc)
460 {
461         struct hc_metrics *hc_entry;
462         u_long mtu;
463
464         hc_entry = tcp_hc_lookup(inc);
465         if (hc_entry == NULL) {
466                 return 0;
467         }
468         hc_entry->rmx_hits++;
469         hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
470
471         mtu = hc_entry->rmx_mtu;
472         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
473         return mtu;
474 }
475
476 /*
477  * External function: update the MTU value of an entry in the hostcache.
478  * Creates a new entry if none was found.
479  */
480 void
481 tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
482 {
483         struct hc_metrics *hc_entry;
484
485         /*
486          * Find the right bucket.
487          */
488         hc_entry = tcp_hc_lookup(inc);
489
490         /*
491          * If we don't have an existing object, try to insert a new one.
492          */
493         if (hc_entry == NULL) {
494                 hc_entry = tcp_hc_insert(inc);
495                 if (hc_entry == NULL)
496                         return;
497         }
498         hc_entry->rmx_updates++;
499         hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
500
501         hc_entry->rmx_mtu = mtu;
502
503         /*
504          * Put it upfront so we find it faster next time.
505          */
506         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
507         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
508
509         /*
510          * Unlock bucket row.
511          */
512         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
513 }
514
515 /*
516  * External function: update the TCP metrics of an entry in the hostcache.
517  * Creates a new entry if none was found.
518  */
519 void
520 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
521 {
522         struct hc_metrics *hc_entry;
523
524         hc_entry = tcp_hc_lookup(inc);
525         if (hc_entry == NULL) {
526                 hc_entry = tcp_hc_insert(inc);
527                 if (hc_entry == NULL)
528                         return;
529         }
530         hc_entry->rmx_updates++;
531         hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
532
533         if (hcml->rmx_rtt != 0) {
534                 if (hc_entry->rmx_rtt == 0)
535                         hc_entry->rmx_rtt = hcml->rmx_rtt;
536                 else
537                         hc_entry->rmx_rtt =
538                             (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
539                 TCPSTAT_INC(tcps_cachedrtt);
540         }
541         if (hcml->rmx_rttvar != 0) {
542                 if (hc_entry->rmx_rttvar == 0)
543                         hc_entry->rmx_rttvar = hcml->rmx_rttvar;
544                 else
545                         hc_entry->rmx_rttvar =
546                             (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
547                 TCPSTAT_INC(tcps_cachedrttvar);
548         }
549         if (hcml->rmx_ssthresh != 0) {
550                 if (hc_entry->rmx_ssthresh == 0)
551                         hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
552                 else
553                         hc_entry->rmx_ssthresh =
554                             (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
555                 TCPSTAT_INC(tcps_cachedssthresh);
556         }
557         if (hcml->rmx_cwnd != 0) {
558                 if (hc_entry->rmx_cwnd == 0)
559                         hc_entry->rmx_cwnd = hcml->rmx_cwnd;
560                 else
561                         hc_entry->rmx_cwnd =
562                             (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
563                 /* TCPSTAT_INC(tcps_cachedcwnd); */
564         }
565         if (hcml->rmx_sendpipe != 0) {
566                 if (hc_entry->rmx_sendpipe == 0)
567                         hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
568                 else
569                         hc_entry->rmx_sendpipe =
570                             (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
571                 /* TCPSTAT_INC(tcps_cachedsendpipe); */
572         }
573         if (hcml->rmx_recvpipe != 0) {
574                 if (hc_entry->rmx_recvpipe == 0)
575                         hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
576                 else
577                         hc_entry->rmx_recvpipe =
578                             (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
579                 /* TCPSTAT_INC(tcps_cachedrecvpipe); */
580         }
581
582         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
583         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
584         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
585 }
586
587 /*
588  * Sysctl function: prints the list and values of all hostcache entries in
589  * unsorted order.
590  */
591 static int
592 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
593 {
594         const int linesize = 128;
595         struct sbuf sb;
596         int i, error;
597         struct hc_metrics *hc_entry;
598 #ifdef INET6
599         char ip6buf[INET6_ADDRSTRLEN];
600 #endif
601
602         sbuf_new(&sb, NULL, linesize * (V_tcp_hostcache.cache_count + 1),
603                 SBUF_INCLUDENUL);
604
605         sbuf_printf(&sb,
606                 "\nIP address        MTU  SSTRESH      RTT   RTTVAR "
607                 "    CWND SENDPIPE RECVPIPE HITS  UPD  EXP\n");
608
609 #define msec(u) (((u) + 500) / 1000)
610         for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
611                 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
612                 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket,
613                               rmx_q) {
614                         sbuf_printf(&sb,
615                             "%-15s %5lu %8lu %6lums %6lums %8lu %8lu %8lu %4lu "
616                             "%4lu %4i\n",
617                             hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
618 #ifdef INET6
619                                 ip6_sprintf(ip6buf, &hc_entry->ip6),
620 #else
621                                 "IPv6?",
622 #endif
623                             hc_entry->rmx_mtu,
624                             hc_entry->rmx_ssthresh,
625                             msec(hc_entry->rmx_rtt *
626                                 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
627                             msec(hc_entry->rmx_rttvar *
628                                 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE))),
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                 }
636                 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
637         }
638 #undef msec
639         error = sbuf_finish(&sb);
640         if (error == 0)
641                 error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
642         sbuf_delete(&sb);
643         return(error);
644 }
645
646 /*
647  * Caller has to make sure the curvnet is set properly.
648  */
649 static void
650 tcp_hc_purge_internal(int all)
651 {
652         struct hc_metrics *hc_entry, *hc_next;
653         int i;
654
655         for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
656                 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
657                 TAILQ_FOREACH_SAFE(hc_entry,
658                     &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) {
659                         if (all || hc_entry->rmx_expire <= 0) {
660                                 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket,
661                                               hc_entry, rmx_q);
662                                 uma_zfree(V_tcp_hostcache.zone, hc_entry);
663                                 V_tcp_hostcache.hashbase[i].hch_length--;
664                                 V_tcp_hostcache.cache_count--;
665                         } else
666                                 hc_entry->rmx_expire -= V_tcp_hostcache.prune;
667                 }
668                 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
669         }
670 }
671
672 /*
673  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
674  * periodically from the callout.
675  */
676 static void
677 tcp_hc_purge(void *arg)
678 {
679         CURVNET_SET((struct vnet *) arg);
680         int all = 0;
681
682         if (V_tcp_hostcache.purgeall) {
683                 all = 1;
684                 V_tcp_hostcache.purgeall = 0;
685         }
686
687         tcp_hc_purge_internal(all);
688
689         callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
690             tcp_hc_purge, arg);
691         CURVNET_RESTORE();
692 }
693
694 /*
695  * Expire and purge all entries in hostcache immediately.
696  */
697 static int
698 sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS)
699 {
700         int error, val;
701
702         val = 0;
703         error = sysctl_handle_int(oidp, &val, 0, req);
704         if (error || !req->newptr)
705                 return (error);
706
707         tcp_hc_purge_internal(1);
708
709         callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
710             tcp_hc_purge, curvnet);
711
712         return (0);
713 }