]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/netinet/tcp_hostcache.c
Merge branch 'releng/11.3' into releng-CDN/11.3
[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/jail.h>
73 #include <sys/kernel.h>
74 #include <sys/lock.h>
75 #include <sys/mutex.h>
76 #include <sys/malloc.h>
77 #include <sys/proc.h>
78 #include <sys/sbuf.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/sysctl.h>
82
83 #include <net/if.h>
84 #include <net/if_var.h>
85 #include <net/route.h>
86 #include <net/vnet.h>
87
88 #include <netinet/in.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/ip.h>
91 #include <netinet/in_var.h>
92 #include <netinet/in_pcb.h>
93 #include <netinet/ip_var.h>
94 #ifdef INET6
95 #include <netinet/ip6.h>
96 #include <netinet6/ip6_var.h>
97 #endif
98 #include <netinet/tcp.h>
99 #include <netinet/tcp_var.h>
100 #include <netinet/tcp_hostcache.h>
101 #ifdef INET6
102 #include <netinet6/tcp6_var.h>
103 #endif
104
105 #include <vm/uma.h>
106
107 /* Arbitrary values */
108 #define TCP_HOSTCACHE_HASHSIZE          512
109 #define TCP_HOSTCACHE_BUCKETLIMIT       30
110 #define TCP_HOSTCACHE_EXPIRE            60*60   /* one hour */
111 #define TCP_HOSTCACHE_PRUNE             5*60    /* every 5 minutes */
112
113 static VNET_DEFINE(struct tcp_hostcache, tcp_hostcache);
114 #define V_tcp_hostcache         VNET(tcp_hostcache)
115
116 static VNET_DEFINE(struct callout, tcp_hc_callout);
117 #define V_tcp_hc_callout        VNET(tcp_hc_callout)
118
119 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
120 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
121 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
122 static int sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS);
123 static void tcp_hc_purge_internal(int);
124 static void tcp_hc_purge(void *);
125
126 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0,
127     "TCP Host cache");
128
129 VNET_DEFINE(int, tcp_use_hostcache) = 1;
130 #define V_tcp_use_hostcache  VNET(tcp_use_hostcache)
131 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
132     &VNET_NAME(tcp_use_hostcache), 0,
133     "Enable the TCP hostcache");
134
135 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_VNET | CTLFLAG_RDTUN,
136     &VNET_NAME(tcp_hostcache.cache_limit), 0,
137     "Overall entry limit for hostcache");
138
139 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
140     &VNET_NAME(tcp_hostcache.hashsize), 0,
141     "Size of TCP hostcache hashtable");
142
143 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit,
144     CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.bucket_limit), 0,
145     "Per-bucket hash limit for hostcache");
146
147 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_VNET | CTLFLAG_RD,
148      &VNET_NAME(tcp_hostcache.cache_count), 0,
149     "Current number of entries in hostcache");
150
151 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_VNET | CTLFLAG_RW,
152     &VNET_NAME(tcp_hostcache.expire), 0,
153     "Expire time of TCP hostcache entries");
154
155 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_VNET | CTLFLAG_RW,
156     &VNET_NAME(tcp_hostcache.prune), 0,
157     "Time between purge runs");
158
159 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_VNET | CTLFLAG_RW,
160     &VNET_NAME(tcp_hostcache.purgeall), 0,
161     "Expire all entires on next purge run");
162
163 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
164     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
165     sysctl_tcp_hc_list, "A", "List of all hostcache entries");
166
167 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, purgenow,
168     CTLTYPE_INT | CTLFLAG_RW, NULL, 0,
169     sysctl_tcp_hc_purgenow, "I", "Immediately purge all entries");
170
171 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
172
173 #define HOSTCACHE_HASH(ip) \
174         (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) &  \
175           V_tcp_hostcache.hashmask)
176
177 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
178 #define HOSTCACHE_HASH6(ip6)                            \
179         (((ip6)->s6_addr32[0] ^                         \
180           (ip6)->s6_addr32[1] ^                         \
181           (ip6)->s6_addr32[2] ^                         \
182           (ip6)->s6_addr32[3]) &                        \
183          V_tcp_hostcache.hashmask)
184
185 #define THC_LOCK(lp)            mtx_lock(lp)
186 #define THC_UNLOCK(lp)          mtx_unlock(lp)
187
188 void
189 tcp_hc_init(void)
190 {
191         u_int cache_limit;
192         int i;
193
194         /*
195          * Initialize hostcache structures.
196          */
197         V_tcp_hostcache.cache_count = 0;
198         V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
199         V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
200         V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
201         V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
202
203         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
204             &V_tcp_hostcache.hashsize);
205         if (!powerof2(V_tcp_hostcache.hashsize)) {
206                 printf("WARNING: hostcache hash size is not a power of 2.\n");
207                 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
208         }
209         V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1;
210
211         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
212             &V_tcp_hostcache.bucket_limit);
213
214         cache_limit = V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit;
215         V_tcp_hostcache.cache_limit = cache_limit;
216         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
217             &V_tcp_hostcache.cache_limit);
218         if (V_tcp_hostcache.cache_limit > cache_limit)
219                 V_tcp_hostcache.cache_limit = cache_limit;
220
221         /*
222          * Allocate the hash table.
223          */
224         V_tcp_hostcache.hashbase = (struct hc_head *)
225             malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head),
226                    M_HOSTCACHE, M_WAITOK | M_ZERO);
227
228         /*
229          * Initialize the hash buckets.
230          */
231         for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
232                 TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket);
233                 V_tcp_hostcache.hashbase[i].hch_length = 0;
234                 mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
235                           NULL, MTX_DEF);
236         }
237
238         /*
239          * Allocate the hostcache entries.
240          */
241         V_tcp_hostcache.zone =
242             uma_zcreate("hostcache", sizeof(struct hc_metrics),
243             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
244         uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit);
245
246         /*
247          * Set up periodic cache cleanup.
248          */
249         callout_init(&V_tcp_hc_callout, 1);
250         callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
251             tcp_hc_purge, curvnet);
252 }
253
254 #ifdef VIMAGE
255 void
256 tcp_hc_destroy(void)
257 {
258         int i;
259
260         callout_drain(&V_tcp_hc_callout);
261
262         /* Purge all hc entries. */
263         tcp_hc_purge_internal(1);
264
265         /* Free the uma zone and the allocated hash table. */
266         uma_zdestroy(V_tcp_hostcache.zone);
267
268         for (i = 0; i < V_tcp_hostcache.hashsize; i++)
269                 mtx_destroy(&V_tcp_hostcache.hashbase[i].hch_mtx);
270         free(V_tcp_hostcache.hashbase, M_HOSTCACHE);
271 }
272 #endif
273
274 /*
275  * Internal function: look up an entry in the hostcache or return NULL.
276  *
277  * If an entry has been returned, the caller becomes responsible for
278  * unlocking the bucket row after he is done reading/modifying the entry.
279  */
280 static struct hc_metrics *
281 tcp_hc_lookup(struct in_conninfo *inc)
282 {
283         int hash;
284         struct hc_head *hc_head;
285         struct hc_metrics *hc_entry;
286
287         if (!V_tcp_use_hostcache)
288                 return NULL;
289
290         KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer"));
291
292         /*
293          * Hash the foreign ip address.
294          */
295         if (inc->inc_flags & INC_ISIPV6)
296                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
297         else
298                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
299
300         hc_head = &V_tcp_hostcache.hashbase[hash];
301
302         /*
303          * Acquire lock for this bucket row; we release the lock if we don't
304          * find an entry, otherwise the caller has to unlock after he is
305          * done.
306          */
307         THC_LOCK(&hc_head->hch_mtx);
308
309         /*
310          * Iterate through entries in bucket row looking for a match.
311          */
312         TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) {
313                 if (inc->inc_flags & INC_ISIPV6) {
314                         /* XXX: check ip6_zoneid */
315                         if (memcmp(&inc->inc6_faddr, &hc_entry->ip6,
316                             sizeof(inc->inc6_faddr)) == 0)
317                                 return hc_entry;
318                 } else {
319                         if (memcmp(&inc->inc_faddr, &hc_entry->ip4,
320                             sizeof(inc->inc_faddr)) == 0)
321                                 return hc_entry;
322                 }
323         }
324
325         /*
326          * We were unsuccessful and didn't find anything.
327          */
328         THC_UNLOCK(&hc_head->hch_mtx);
329         return NULL;
330 }
331
332 /*
333  * Internal function: insert an entry into the hostcache or return NULL if
334  * unable to allocate a new one.
335  *
336  * If an entry has been returned, the caller becomes responsible for
337  * unlocking the bucket row after he is done reading/modifying the entry.
338  */
339 static struct hc_metrics *
340 tcp_hc_insert(struct in_conninfo *inc)
341 {
342         int hash;
343         struct hc_head *hc_head;
344         struct hc_metrics *hc_entry;
345
346         if (!V_tcp_use_hostcache)
347                 return NULL;
348
349         KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer"));
350
351         /*
352          * Hash the foreign ip address.
353          */
354         if (inc->inc_flags & INC_ISIPV6)
355                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
356         else
357                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
358
359         hc_head = &V_tcp_hostcache.hashbase[hash];
360
361         /*
362          * Acquire lock for this bucket row; we release the lock if we don't
363          * find an entry, otherwise the caller has to unlock after he is
364          * done.
365          */
366         THC_LOCK(&hc_head->hch_mtx);
367
368         /*
369          * If the bucket limit is reached, reuse the least-used element.
370          */
371         if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit ||
372             V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) {
373                 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead);
374                 /*
375                  * At first we were dropping the last element, just to
376                  * reacquire it in the next two lines again, which isn't very
377                  * efficient.  Instead just reuse the least used element.
378                  * We may drop something that is still "in-use" but we can be
379                  * "lossy".
380                  * Just give up if this bucket row is empty and we don't have
381                  * anything to replace.
382                  */
383                 if (hc_entry == NULL) {
384                         THC_UNLOCK(&hc_head->hch_mtx);
385                         return NULL;
386                 }
387                 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
388                 V_tcp_hostcache.hashbase[hash].hch_length--;
389                 V_tcp_hostcache.cache_count--;
390                 TCPSTAT_INC(tcps_hc_bucketoverflow);
391 #if 0
392                 uma_zfree(V_tcp_hostcache.zone, hc_entry);
393 #endif
394         } else {
395                 /*
396                  * Allocate a new entry, or balk if not possible.
397                  */
398                 hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT);
399                 if (hc_entry == NULL) {
400                         THC_UNLOCK(&hc_head->hch_mtx);
401                         return NULL;
402                 }
403         }
404
405         /*
406          * Initialize basic information of hostcache entry.
407          */
408         bzero(hc_entry, sizeof(*hc_entry));
409         if (inc->inc_flags & INC_ISIPV6) {
410                 hc_entry->ip6 = inc->inc6_faddr;
411                 hc_entry->ip6_zoneid = inc->inc6_zoneid;
412         } else
413                 hc_entry->ip4 = inc->inc_faddr;
414         hc_entry->rmx_head = hc_head;
415         hc_entry->rmx_expire = V_tcp_hostcache.expire;
416
417         /*
418          * Put it upfront.
419          */
420         TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
421         V_tcp_hostcache.hashbase[hash].hch_length++;
422         V_tcp_hostcache.cache_count++;
423         TCPSTAT_INC(tcps_hc_added);
424
425         return hc_entry;
426 }
427
428 /*
429  * External function: look up an entry in the hostcache and fill out the
430  * supplied TCP metrics structure.  Fills in NULL when no entry was found or
431  * a value is not set.
432  */
433 void
434 tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
435 {
436         struct hc_metrics *hc_entry;
437
438         if (!V_tcp_use_hostcache)
439                 return;
440
441         /*
442          * Find the right bucket.
443          */
444         hc_entry = tcp_hc_lookup(inc);
445
446         /*
447          * If we don't have an existing object.
448          */
449         if (hc_entry == NULL) {
450                 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
451                 return;
452         }
453         hc_entry->rmx_hits++;
454         hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
455
456         hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
457         hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
458         hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
459         hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
460         hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
461         hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
462         hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
463
464         /*
465          * Unlock bucket row.
466          */
467         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
468 }
469
470 /*
471  * External function: look up an entry in the hostcache and return the
472  * discovered path MTU.  Returns 0 if no entry is found or value is not
473  * set.
474  */
475 u_long
476 tcp_hc_getmtu(struct in_conninfo *inc)
477 {
478         struct hc_metrics *hc_entry;
479         u_long mtu;
480
481         if (!V_tcp_use_hostcache)
482                 return 0;
483
484         hc_entry = tcp_hc_lookup(inc);
485         if (hc_entry == NULL) {
486                 return 0;
487         }
488         hc_entry->rmx_hits++;
489         hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
490
491         mtu = hc_entry->rmx_mtu;
492         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
493         return mtu;
494 }
495
496 /*
497  * External function: update the MTU value of an entry in the hostcache.
498  * Creates a new entry if none was found.
499  */
500 void
501 tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
502 {
503         struct hc_metrics *hc_entry;
504
505         if (!V_tcp_use_hostcache)
506                 return;
507
508         /*
509          * Find the right bucket.
510          */
511         hc_entry = tcp_hc_lookup(inc);
512
513         /*
514          * If we don't have an existing object, try to insert a new one.
515          */
516         if (hc_entry == NULL) {
517                 hc_entry = tcp_hc_insert(inc);
518                 if (hc_entry == NULL)
519                         return;
520         }
521         hc_entry->rmx_updates++;
522         hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
523
524         hc_entry->rmx_mtu = mtu;
525
526         /*
527          * Put it upfront so we find it faster next time.
528          */
529         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
530         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
531
532         /*
533          * Unlock bucket row.
534          */
535         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
536 }
537
538 /*
539  * External function: update the TCP metrics of an entry in the hostcache.
540  * Creates a new entry if none was found.
541  */
542 void
543 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
544 {
545         struct hc_metrics *hc_entry;
546
547         if (!V_tcp_use_hostcache)
548                 return;
549
550         hc_entry = tcp_hc_lookup(inc);
551         if (hc_entry == NULL) {
552                 hc_entry = tcp_hc_insert(inc);
553                 if (hc_entry == NULL)
554                         return;
555         }
556         hc_entry->rmx_updates++;
557         hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
558
559         if (hcml->rmx_rtt != 0) {
560                 if (hc_entry->rmx_rtt == 0)
561                         hc_entry->rmx_rtt = hcml->rmx_rtt;
562                 else
563                         hc_entry->rmx_rtt =
564                             (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
565                 TCPSTAT_INC(tcps_cachedrtt);
566         }
567         if (hcml->rmx_rttvar != 0) {
568                 if (hc_entry->rmx_rttvar == 0)
569                         hc_entry->rmx_rttvar = hcml->rmx_rttvar;
570                 else
571                         hc_entry->rmx_rttvar =
572                             (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
573                 TCPSTAT_INC(tcps_cachedrttvar);
574         }
575         if (hcml->rmx_ssthresh != 0) {
576                 if (hc_entry->rmx_ssthresh == 0)
577                         hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
578                 else
579                         hc_entry->rmx_ssthresh =
580                             (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
581                 TCPSTAT_INC(tcps_cachedssthresh);
582         }
583         if (hcml->rmx_cwnd != 0) {
584                 if (hc_entry->rmx_cwnd == 0)
585                         hc_entry->rmx_cwnd = hcml->rmx_cwnd;
586                 else
587                         hc_entry->rmx_cwnd =
588                             (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
589                 /* TCPSTAT_INC(tcps_cachedcwnd); */
590         }
591         if (hcml->rmx_sendpipe != 0) {
592                 if (hc_entry->rmx_sendpipe == 0)
593                         hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
594                 else
595                         hc_entry->rmx_sendpipe =
596                             (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
597                 /* TCPSTAT_INC(tcps_cachedsendpipe); */
598         }
599         if (hcml->rmx_recvpipe != 0) {
600                 if (hc_entry->rmx_recvpipe == 0)
601                         hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
602                 else
603                         hc_entry->rmx_recvpipe =
604                             (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
605                 /* TCPSTAT_INC(tcps_cachedrecvpipe); */
606         }
607
608         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
609         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
610         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
611 }
612
613 /*
614  * Sysctl function: prints the list and values of all hostcache entries in
615  * unsorted order.
616  */
617 static int
618 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
619 {
620         const int linesize = 128;
621         struct sbuf sb;
622         int i, error;
623         struct hc_metrics *hc_entry;
624         char ip4buf[INET_ADDRSTRLEN];
625 #ifdef INET6
626         char ip6buf[INET6_ADDRSTRLEN];
627 #endif
628
629         if (jailed_without_vnet(curthread->td_ucred) != 0)
630                 return (EPERM);
631
632         sbuf_new(&sb, NULL, linesize * (V_tcp_hostcache.cache_count + 1),
633                 SBUF_INCLUDENUL);
634
635         sbuf_printf(&sb,
636                 "\nIP address        MTU  SSTRESH      RTT   RTTVAR "
637                 "    CWND SENDPIPE RECVPIPE HITS  UPD  EXP\n");
638
639 #define msec(u) (((u) + 500) / 1000)
640         for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
641                 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
642                 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket,
643                               rmx_q) {
644                         sbuf_printf(&sb,
645                             "%-15s %5lu %8lu %6lums %6lums %8lu %8lu %8lu %4lu "
646                             "%4lu %4i\n",
647                             hc_entry->ip4.s_addr ?
648                                 inet_ntoa_r(hc_entry->ip4, ip4buf) :
649 #ifdef INET6
650                                 ip6_sprintf(ip6buf, &hc_entry->ip6),
651 #else
652                                 "IPv6?",
653 #endif
654                             hc_entry->rmx_mtu,
655                             hc_entry->rmx_ssthresh,
656                             msec(hc_entry->rmx_rtt *
657                                 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
658                             msec(hc_entry->rmx_rttvar *
659                                 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE))),
660                             hc_entry->rmx_cwnd,
661                             hc_entry->rmx_sendpipe,
662                             hc_entry->rmx_recvpipe,
663                             hc_entry->rmx_hits,
664                             hc_entry->rmx_updates,
665                             hc_entry->rmx_expire);
666                 }
667                 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
668         }
669 #undef msec
670         error = sbuf_finish(&sb);
671         if (error == 0)
672                 error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
673         sbuf_delete(&sb);
674         return(error);
675 }
676
677 /*
678  * Caller has to make sure the curvnet is set properly.
679  */
680 static void
681 tcp_hc_purge_internal(int all)
682 {
683         struct hc_metrics *hc_entry, *hc_next;
684         int i;
685
686         for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
687                 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
688                 TAILQ_FOREACH_SAFE(hc_entry,
689                     &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) {
690                         if (all || hc_entry->rmx_expire <= 0) {
691                                 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket,
692                                               hc_entry, rmx_q);
693                                 uma_zfree(V_tcp_hostcache.zone, hc_entry);
694                                 V_tcp_hostcache.hashbase[i].hch_length--;
695                                 V_tcp_hostcache.cache_count--;
696                         } else
697                                 hc_entry->rmx_expire -= V_tcp_hostcache.prune;
698                 }
699                 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
700         }
701 }
702
703 /*
704  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
705  * periodically from the callout.
706  */
707 static void
708 tcp_hc_purge(void *arg)
709 {
710         CURVNET_SET((struct vnet *) arg);
711         int all = 0;
712
713         if (V_tcp_hostcache.purgeall) {
714                 all = 1;
715                 V_tcp_hostcache.purgeall = 0;
716         }
717
718         tcp_hc_purge_internal(all);
719
720         callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
721             tcp_hc_purge, arg);
722         CURVNET_RESTORE();
723 }
724
725 /*
726  * Expire and purge all entries in hostcache immediately.
727  */
728 static int
729 sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS)
730 {
731         int error, val;
732
733         val = 0;
734         error = sysctl_handle_int(oidp, &val, 0, req);
735         if (error || !req->newptr)
736                 return (error);
737
738         tcp_hc_purge_internal(1);
739
740         callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
741             tcp_hc_purge, curvnet);
742
743         return (0);
744 }