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, bandwidth, max
  36  * MTU, congestion window) between local and remote sites, this can lead to
  37  * significant speed-ups for new TCP connections after the first one.
  38  *
  39  * Due to the tcp_hostcache, all TCP-specific metrics information in the
  40  * routing table has been removed.  The inpcb no longer keeps a pointer to
  41  * the routing entry, and protocol-initiated route cloning has been removed
  42  * as well.  With these changes, the routing table has gone back to being
  43  * more lightwight and only carries information related to packet forwarding.
  44  *
  45  * tcp_hostcache is designed for multiple concurrent access in SMP
  46  * environments and high contention.  All bucket rows have their own lock and
  47  * thus multiple lookups and modifies can be done at the same time as long as
  48  * they are in different bucket rows.  If a request for insertion of a new
  49  * record can't be satisfied, it simply returns an empty structure.  Nobody
  50  * and nothing outside of tcp_hostcache.c will ever point directly to any
  51  * entry in the tcp_hostcache.  All communication is done in an
  52  * object-oriented way and only functions of tcp_hostcache will manipulate
  53  * hostcache entries.  Otherwise, we are unable to achieve good behaviour in
  54  * concurrent access situations.  Since tcp_hostcache is only caching
  55  * information, there are no fatal consequences if we either can't satisfy
  56  * any particular request or have to drop/overwrite an existing entry because
  57  * of bucket limit memory constrains.
  58  */
  59
  60 /*
  61  * Many thanks to jlemon for basic structure of tcp_syncache which is being
  62  * followed here.
  63  */
  64
  65 #include <sys/cdefs.h>
  66 __FBSDID("$FreeBSD$");
  67
  68 #include "opt_inet6.h"
  69
  70 #include <sys/param.h>
  71 #include <sys/systm.h>
  72 #include <sys/kernel.h>
  73 #include <sys/lock.h>
  74 #include <sys/mutex.h>
  75 #include <sys/malloc.h>
  76 #include <sys/socket.h>
  77 #include <sys/socketvar.h>
  78 #include <sys/sysctl.h>
  79
  80 #include <net/if.h>
  81
  82 #include <netinet/in.h>
  83 #include <netinet/in_systm.h>
  84 #include <netinet/ip.h>
  85 #include <netinet/in_var.h>
  86 #include <netinet/in_pcb.h>
  87 #include <netinet/ip_var.h>
  88 #ifdef INET6
  89 #include <netinet/ip6.h>
  90 #include <netinet6/ip6_var.h>
  91 #endif
  92 #include <netinet/tcp.h>
  93 #include <netinet/tcp_var.h>
  94 #ifdef INET6
  95 #include <netinet6/tcp6_var.h>
  96 #endif
  97
  98 #include <vm/uma.h>
  99
 100
 101 TAILQ_HEAD(hc_qhead, hc_metrics);
 102
 103 struct hc_head {
 104         struct hc_qhead hch_bucket;
 105         u_int           hch_length;
 106         struct mtx      hch_mtx;
 107 };
 108
 109 struct hc_metrics {
 110         /* housekeeping */
 111         TAILQ_ENTRY(hc_metrics) rmx_q;
 112         struct  hc_head *rmx_head; /* head of bucket tail queue */
 113         struct  in_addr ip4;    /* IP address */
 114         struct  in6_addr ip6;   /* IP6 address */
 115         /* endpoint specific values for TCP */
 116         u_long  rmx_mtu;        /* MTU for this path */
 117         u_long  rmx_ssthresh;   /* outbound gateway buffer limit */
 118         u_long  rmx_rtt;        /* estimated round trip time */
 119         u_long  rmx_rttvar;     /* estimated rtt variance */
 120         u_long  rmx_bandwidth;  /* estimated bandwidth */
 121         u_long  rmx_cwnd;       /* congestion window */
 122         u_long  rmx_sendpipe;   /* outbound delay-bandwidth product */
 123         u_long  rmx_recvpipe;   /* inbound delay-bandwidth product */
 124         /* TCP hostcache internal data */
 125         int     rmx_expire;     /* lifetime for object */
 126         u_long  rmx_hits;       /* number of hits */
 127         u_long  rmx_updates;    /* number of updates */
 128 };
 129
 130 /* Arbitrary values */
 131 #define TCP_HOSTCACHE_HASHSIZE          512
 132 #define TCP_HOSTCACHE_BUCKETLIMIT       30
 133 #define TCP_HOSTCACHE_EXPIRE            60*60   /* one hour */
 134 #define TCP_HOSTCACHE_PRUNE             5*60    /* every 5 minutes */
 135
 136 struct tcp_hostcache {
 137         struct  hc_head *hashbase;
 138         uma_zone_t zone;
 139         u_int   hashsize;
 140         u_int   hashmask;
 141         u_int   bucket_limit;
 142         u_int   cache_count;
 143         u_int   cache_limit;
 144         int     expire;
 145         int     prune;
 146         int     purgeall;
 147 };
 148 static struct tcp_hostcache tcp_hostcache;
 149
 150 static struct callout tcp_hc_callout;
 151
 152 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
 153 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
 154 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
 155 static void tcp_hc_purge(void *);
 156
 157 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, "TCP Host cache");
 158
 159 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
 160     &tcp_hostcache.cache_limit, 0, "Overall entry limit for hostcache");
 161
 162 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
 163     &tcp_hostcache.hashsize, 0, "Size of TCP hostcache hashtable");
 164
 165 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
 166     &tcp_hostcache.bucket_limit, 0, "Per-bucket hash limit for hostcache");
 167
 168 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD,
 169     &tcp_hostcache.cache_count, 0, "Current number of entries in hostcache");
 170
 171 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW,
 172     &tcp_hostcache.expire, 0, "Expire time of TCP hostcache entries");
 173
 174 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW,
 175      &tcp_hostcache.prune, 0, "Time between purge runs");
 176
 177 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW,
 178     &tcp_hostcache.purgeall, 0, "Expire all entires on next purge run");
 179
 180 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
 181     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
 182     sysctl_tcp_hc_list, "A", "List of all hostcache entries");
 183
 184
 185 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
 186
 187 #define HOSTCACHE_HASH(ip) \
 188         (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) &  \
 189           tcp_hostcache.hashmask)
 190
 191 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
 192 #define HOSTCACHE_HASH6(ip6)                            \
 193         (((ip6)->s6_addr32[0] ^                         \
 194           (ip6)->s6_addr32[1] ^                         \
 195           (ip6)->s6_addr32[2] ^                         \
 196           (ip6)->s6_addr32[3]) &                        \
 197          tcp_hostcache.hashmask)
 198
 199 #define THC_LOCK(lp)            mtx_lock(lp)
 200 #define THC_UNLOCK(lp)          mtx_unlock(lp)
 201
 202 void
 203 tcp_hc_init(void)
 204 {
 205         int i;
 206
 207         /*
 208          * Initialize hostcache structures.
 209          */
 210         tcp_hostcache.cache_count = 0;
 211         tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
 212         tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
 213         tcp_hostcache.cache_limit =
 214             tcp_hostcache.hashsize * tcp_hostcache.bucket_limit;
 215         tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
 216         tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
 217
 218         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
 219             &tcp_hostcache.hashsize);
 220         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
 221             &tcp_hostcache.cache_limit);
 222         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
 223             &tcp_hostcache.bucket_limit);
 224         if (!powerof2(tcp_hostcache.hashsize)) {
 225                 printf("WARNING: hostcache hash size is not a power of 2.\n");
 226                 tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
 227         }
 228         tcp_hostcache.hashmask = tcp_hostcache.hashsize - 1;
 229
 230         /*
 231          * Allocate the hash table.
 232          */
 233         tcp_hostcache.hashbase = (struct hc_head *)
 234             malloc(tcp_hostcache.hashsize * sizeof(struct hc_head),
 235                    M_HOSTCACHE, M_WAITOK | M_ZERO);
 236
 237         /*
 238          * Initialize the hash buckets.
 239          */
 240         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 241                 TAILQ_INIT(&tcp_hostcache.hashbase[i].hch_bucket);
 242                 tcp_hostcache.hashbase[i].hch_length = 0;
 243                 mtx_init(&tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
 244                           NULL, MTX_DEF);
 245         }
 246
 247         /*
 248          * Allocate the hostcache entries.
 249          */
 250         tcp_hostcache.zone = uma_zcreate("hostcache", sizeof(struct hc_metrics),
 251             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 252         uma_zone_set_max(tcp_hostcache.zone, tcp_hostcache.cache_limit);
 253
 254         /*
 255          * Set up periodic cache cleanup.
 256          */
 257         callout_init(&tcp_hc_callout, CALLOUT_MPSAFE);
 258         callout_reset(&tcp_hc_callout, tcp_hostcache.prune * hz, tcp_hc_purge, 0);
 259 }
 260
 261 /*
 262  * Internal function: look up an entry in the hostcache or return NULL.
 263  *
 264  * If an entry has been returned, the caller becomes responsible for
 265  * unlocking the bucket row after he is done reading/modifying the entry.
 266  */
 267 static struct hc_metrics *
 268 tcp_hc_lookup(struct in_conninfo *inc)
 269 {
 270         int hash;
 271         struct hc_head *hc_head;
 272         struct hc_metrics *hc_entry;
 273
 274         KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer"));
 275
 276         /*
 277          * Hash the foreign ip address.
 278          */
 279         if (inc->inc_flags & INC_ISIPV6)
 280                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
 281         else
 282                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
 283
 284         hc_head = &tcp_hostcache.hashbase[hash];
 285
 286         /*
 287          * Acquire lock for this bucket row; we release the lock if we don't
 288          * find an entry, otherwise the caller has to unlock after he is
 289          * done.
 290          */
 291         THC_LOCK(&hc_head->hch_mtx);
 292
 293         /*
 294          * Iterate through entries in bucket row looking for a match.
 295          */
 296         TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) {
 297                 if (inc->inc_flags & INC_ISIPV6) {
 298                         if (memcmp(&inc->inc6_faddr, &hc_entry->ip6,
 299                             sizeof(inc->inc6_faddr)) == 0)
 300                                 return hc_entry;
 301                 } else {
 302                         if (memcmp(&inc->inc_faddr, &hc_entry->ip4,
 303                             sizeof(inc->inc_faddr)) == 0)
 304                                 return hc_entry;
 305                 }
 306         }
 307
 308         /*
 309          * We were unsuccessful and didn't find anything.
 310          */
 311         THC_UNLOCK(&hc_head->hch_mtx);
 312         return NULL;
 313 }
 314
 315 /*
 316  * Internal function: insert an entry into the hostcache or return NULL if
 317  * unable to allocate a new one.
 318  *
 319  * If an entry has been returned, the caller becomes responsible for
 320  * unlocking the bucket row after he is done reading/modifying the entry.
 321  */
 322 static struct hc_metrics *
 323 tcp_hc_insert(struct in_conninfo *inc)
 324 {
 325         int hash;
 326         struct hc_head *hc_head;
 327         struct hc_metrics *hc_entry;
 328
 329         KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer"));
 330
 331         /*
 332          * Hash the foreign ip address.
 333          */
 334         if (inc->inc_flags & INC_ISIPV6)
 335                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
 336         else
 337                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
 338
 339         hc_head = &tcp_hostcache.hashbase[hash];
 340
 341         /*
 342          * Acquire lock for this bucket row; we release the lock if we don't
 343          * find an entry, otherwise the caller has to unlock after he is
 344          * done.
 345          */
 346         THC_LOCK(&hc_head->hch_mtx);
 347
 348         /*
 349          * If the bucket limit is reached, reuse the least-used element.
 350          */
 351         if (hc_head->hch_length >= tcp_hostcache.bucket_limit ||
 352             tcp_hostcache.cache_count >= tcp_hostcache.cache_limit) {
 353                 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead);
 354                 /*
 355                  * At first we were dropping the last element, just to
 356                  * reacquire it in the next two lines again, which isn't very
 357                  * efficient.  Instead just reuse the least used element.
 358                  * We may drop something that is still "in-use" but we can be
 359                  * "lossy".
 360                  * Just give up if this bucket row is empty and we don't have
 361                  * anything to replace.
 362                  */
 363                 if (hc_entry == NULL) {
 364                         THC_UNLOCK(&hc_head->hch_mtx);
 365                         return NULL;
 366                 }
 367                 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
 368                 tcp_hostcache.hashbase[hash].hch_length--;
 369                 tcp_hostcache.cache_count--;
 370                 tcpstat.tcps_hc_bucketoverflow++;
 371 #if 0
 372                 uma_zfree(tcp_hostcache.zone, hc_entry);
 373 #endif
 374         } else {
 375                 /*
 376                  * Allocate a new entry, or balk if not possible.
 377                  */
 378                 hc_entry = uma_zalloc(tcp_hostcache.zone, M_NOWAIT);
 379                 if (hc_entry == NULL) {
 380                         THC_UNLOCK(&hc_head->hch_mtx);
 381                         return NULL;
 382                 }
 383         }
 384
 385         /*
 386          * Initialize basic information of hostcache entry.
 387          */
 388         bzero(hc_entry, sizeof(*hc_entry));
 389         if (inc->inc_flags & INC_ISIPV6)
 390                 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6));
 391         else
 392                 hc_entry->ip4 = inc->inc_faddr;
 393         hc_entry->rmx_head = hc_head;
 394         hc_entry->rmx_expire = tcp_hostcache.expire;
 395
 396         /*
 397          * Put it upfront.
 398          */
 399         TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
 400         tcp_hostcache.hashbase[hash].hch_length++;
 401         tcp_hostcache.cache_count++;
 402         tcpstat.tcps_hc_added++;
 403
 404         return hc_entry;
 405 }
 406
 407 /*
 408  * External function: look up an entry in the hostcache and fill out the
 409  * supplied TCP metrics structure.  Fills in NULL when no entry was found or
 410  * a value is not set.
 411  */
 412 void
 413 tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
 414 {
 415         struct hc_metrics *hc_entry;
 416
 417         /*
 418          * Find the right bucket.
 419          */
 420         hc_entry = tcp_hc_lookup(inc);
 421
 422         /*
 423          * If we don't have an existing object.
 424          */
 425         if (hc_entry == NULL) {
 426                 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
 427                 return;
 428         }
 429         hc_entry->rmx_hits++;
 430         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 431
 432         hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
 433         hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
 434         hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
 435         hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
 436         hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth;
 437         hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
 438         hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
 439         hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
 440
 441         /*
 442          * Unlock bucket row.
 443          */
 444         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 445 }
 446
 447 /*
 448  * External function: look up an entry in the hostcache and return the
 449  * discovered path MTU.  Returns NULL if no entry is found or value is not
 450  * set.
 451  */
 452 u_long
 453 tcp_hc_getmtu(struct in_conninfo *inc)
 454 {
 455         struct hc_metrics *hc_entry;
 456         u_long mtu;
 457
 458         hc_entry = tcp_hc_lookup(inc);
 459         if (hc_entry == NULL) {
 460                 return 0;
 461         }
 462         hc_entry->rmx_hits++;
 463         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 464
 465         mtu = hc_entry->rmx_mtu;
 466         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 467         return mtu;
 468 }
 469
 470 /*
 471  * External function: update the MTU value of an entry in the hostcache.
 472  * Creates a new entry if none was found.
 473  */
 474 void
 475 tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
 476 {
 477         struct hc_metrics *hc_entry;
 478
 479         /*
 480          * Find the right bucket.
 481          */
 482         hc_entry = tcp_hc_lookup(inc);
 483
 484         /*
 485          * If we don't have an existing object, try to insert a new one.
 486          */
 487         if (hc_entry == NULL) {
 488                 hc_entry = tcp_hc_insert(inc);
 489                 if (hc_entry == NULL)
 490                         return;
 491         }
 492         hc_entry->rmx_updates++;
 493         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 494
 495         hc_entry->rmx_mtu = mtu;
 496
 497         /*
 498          * Put it upfront so we find it faster next time.
 499          */
 500         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 501         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 502
 503         /*
 504          * Unlock bucket row.
 505          */
 506         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 507 }
 508
 509 /*
 510  * External function: update the TCP metrics of an entry in the hostcache.
 511  * Creates a new entry if none was found.
 512  */
 513 void
 514 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
 515 {
 516         struct hc_metrics *hc_entry;
 517
 518         hc_entry = tcp_hc_lookup(inc);
 519         if (hc_entry == NULL) {
 520                 hc_entry = tcp_hc_insert(inc);
 521                 if (hc_entry == NULL)
 522                         return;
 523         }
 524         hc_entry->rmx_updates++;
 525         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 526
 527         if (hcml->rmx_rtt != 0) {
 528                 if (hc_entry->rmx_rtt == 0)
 529                         hc_entry->rmx_rtt = hcml->rmx_rtt;
 530                 else
 531                         hc_entry->rmx_rtt =
 532                             (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
 533                 tcpstat.tcps_cachedrtt++;
 534         }
 535         if (hcml->rmx_rttvar != 0) {
 536                 if (hc_entry->rmx_rttvar == 0)
 537                         hc_entry->rmx_rttvar = hcml->rmx_rttvar;
 538                 else
 539                         hc_entry->rmx_rttvar =
 540                             (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
 541                 tcpstat.tcps_cachedrttvar++;
 542         }
 543         if (hcml->rmx_ssthresh != 0) {
 544                 if (hc_entry->rmx_ssthresh == 0)
 545                         hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
 546                 else
 547                         hc_entry->rmx_ssthresh =
 548                             (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
 549                 tcpstat.tcps_cachedssthresh++;
 550         }
 551         if (hcml->rmx_bandwidth != 0) {
 552                 if (hc_entry->rmx_bandwidth == 0)
 553                         hc_entry->rmx_bandwidth = hcml->rmx_bandwidth;
 554                 else
 555                         hc_entry->rmx_bandwidth =
 556                             (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2;
 557                 /* tcpstat.tcps_cachedbandwidth++; */
 558         }
 559         if (hcml->rmx_cwnd != 0) {
 560                 if (hc_entry->rmx_cwnd == 0)
 561                         hc_entry->rmx_cwnd = hcml->rmx_cwnd;
 562                 else
 563                         hc_entry->rmx_cwnd =
 564                             (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
 565                 /* tcpstat.tcps_cachedcwnd++; */
 566         }
 567         if (hcml->rmx_sendpipe != 0) {
 568                 if (hc_entry->rmx_sendpipe == 0)
 569                         hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
 570                 else
 571                         hc_entry->rmx_sendpipe =
 572                             (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
 573                 /* tcpstat.tcps_cachedsendpipe++; */
 574         }
 575         if (hcml->rmx_recvpipe != 0) {
 576                 if (hc_entry->rmx_recvpipe == 0)
 577                         hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
 578                 else
 579                         hc_entry->rmx_recvpipe =
 580                             (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
 581                 /* tcpstat.tcps_cachedrecvpipe++; */
 582         }
 583
 584         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 585         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 586         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 587 }
 588
 589 /*
 590  * Sysctl function: prints the list and values of all hostcache entries in
 591  * unsorted order.
 592  */
 593 static int
 594 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
 595 {
 596         int bufsize;
 597         int linesize = 128;
 598         char *p, *buf;
 599         int len, i, error;
 600         struct hc_metrics *hc_entry;
 601 #ifdef INET6
 602         char ip6buf[INET6_ADDRSTRLEN];
 603 #endif
 604
 605         bufsize = linesize * (tcp_hostcache.cache_count + 1);
 606
 607         p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
 608
 609         len = snprintf(p, linesize,
 610                 "\nIP address        MTU  SSTRESH      RTT   RTTVAR BANDWIDTH "
 611                 "    CWND SENDPIPE RECVPIPE HITS  UPD  EXP\n");
 612         p += len;
 613
 614 #define msec(u) (((u) + 500) / 1000)
 615         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 616                 THC_LOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 617                 TAILQ_FOREACH(hc_entry, &tcp_hostcache.hashbase[i].hch_bucket,
 618                               rmx_q) {
 619                         len = snprintf(p, linesize,
 620                             "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu "
 621                             "%4lu %4lu %4i\n",
 622                             hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
 623 #ifdef INET6
 624                                 ip6_sprintf(ip6buf, &hc_entry->ip6),
 625 #else
 626                                 "IPv6?",
 627 #endif
 628                             hc_entry->rmx_mtu,
 629                             hc_entry->rmx_ssthresh,
 630                             msec(hc_entry->rmx_rtt *
 631                                 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
 632                             msec(hc_entry->rmx_rttvar *
 633                                 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
 634                             hc_entry->rmx_bandwidth * 8,
 635                             hc_entry->rmx_cwnd,
 636                             hc_entry->rmx_sendpipe,
 637                             hc_entry->rmx_recvpipe,
 638                             hc_entry->rmx_hits,
 639                             hc_entry->rmx_updates,
 640                             hc_entry->rmx_expire);
 641                         p += len;
 642                 }
 643                 THC_UNLOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 644         }
 645 #undef msec
 646         error = SYSCTL_OUT(req, buf, p - buf);
 647         free(buf, M_TEMP);
 648         return(error);
 649 }
 650
 651 /*
 652  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
 653  * periodically from the callout.
 654  */
 655 static void
 656 tcp_hc_purge(void *arg)
 657 {
 658         struct hc_metrics *hc_entry, *hc_next;
 659         int all = (intptr_t)arg;
 660         int i;
 661
 662         if (tcp_hostcache.purgeall) {
 663                 all = 1;
 664                 tcp_hostcache.purgeall = 0;
 665         }
 666
 667         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 668                 THC_LOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 669                 TAILQ_FOREACH_SAFE(hc_entry, &tcp_hostcache.hashbase[i].hch_bucket,
 670                               rmx_q, hc_next) {
 671                         if (all || hc_entry->rmx_expire <= 0) {
 672                                 TAILQ_REMOVE(&tcp_hostcache.hashbase[i].hch_bucket,
 673                                               hc_entry, rmx_q);
 674                                 uma_zfree(tcp_hostcache.zone, hc_entry);
 675                                 tcp_hostcache.hashbase[i].hch_length--;
 676                                 tcp_hostcache.cache_count--;
 677                         } else
 678                                 hc_entry->rmx_expire -= tcp_hostcache.prune;
 679                 }
 680                 THC_UNLOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 681         }
 682         callout_reset(&tcp_hc_callout, tcp_hostcache.prune * hz, tcp_hc_purge, 0);
 683 }