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  * $FreeBSD$
  30  */
  31
  32 /*
  33  * The tcp_hostcache moves the tcp-specific cached metrics from the routing
  34  * table to a dedicated structure indexed by the remote IP address.  It keeps
  35  * information on the measured TCP parameters of past TCP sessions to allow
  36  * better initial start values to be used with later connections to/from the
  37  * same source.  Depending on the network parameters (delay, bandwidth, max
  38  * MTU, congestion window) between local and remote sites, this can lead to
  39  * significant speed-ups for new TCP connections after the first one.
  40  *
  41  * Due to the tcp_hostcache, all TCP-specific metrics information in the
  42  * routing table has been removed.  The inpcb no longer keeps a pointer to
  43  * the routing entry, and protocol-initiated route cloning has been removed
  44  * as well.  With these changes, the routing table has gone back to being
  45  * more lightwight and only carries information related to packet forwarding.
  46  *
  47  * tcp_hostcache is designed for multiple concurrent access in SMP
  48  * environments and high contention.  All bucket rows have their own lock and
  49  * thus multiple lookups and modifies can be done at the same time as long as
  50  * they are in different bucket rows.  If a request for insertion of a new
  51  * record can't be satisfied, it simply returns an empty structure.  Nobody
  52  * and nothing outside of tcp_hostcache.c will ever point directly to any
  53  * entry in the tcp_hostcache.  All communication is done in an
  54  * object-oriented way and only functions of tcp_hostcache will manipulate
  55  * hostcache entries.  Otherwise, we are unable to achieve good behaviour in
  56  * concurrent access situations.  Since tcp_hostcache is only caching
  57  * information, there are no fatal consequences if we either can't satisfy
  58  * any particular request or have to drop/overwrite an existing entry because
  59  * of bucket limit memory constrains.
  60  */
  61
  62 /*
  63  * Many thanks to jlemon for basic structure of tcp_syncache which is being
  64  * followed here.
  65  */
  66
  67 #include "opt_inet6.h"
  68
  69 #include <sys/param.h>
  70 #include <sys/systm.h>
  71 #include <sys/kernel.h>
  72 #include <sys/lock.h>
  73 #include <sys/mutex.h>
  74 #include <sys/malloc.h>
  75 #include <sys/socket.h>
  76 #include <sys/socketvar.h>
  77 #include <sys/sysctl.h>
  78
  79 #include <net/if.h>
  80
  81 #include <netinet/in.h>
  82 #include <netinet/in_systm.h>
  83 #include <netinet/ip.h>
  84 #include <netinet/in_var.h>
  85 #include <netinet/in_pcb.h>
  86 #include <netinet/ip_var.h>
  87 #ifdef INET6
  88 #include <netinet/ip6.h>
  89 #include <netinet6/ip6_var.h>
  90 #endif
  91 #include <netinet/tcp.h>
  92 #include <netinet/tcp_var.h>
  93 #ifdef INET6
  94 #include <netinet6/tcp6_var.h>
  95 #endif
  96
  97 #include <vm/uma.h>
  98
  99
 100 TAILQ_HEAD(hc_qhead, hc_metrics);
 101
 102 struct hc_head {
 103         struct hc_qhead hch_bucket;
 104         u_int           hch_length;
 105         struct mtx      hch_mtx;
 106 };
 107
 108 struct hc_metrics {
 109         /* housekeeping */
 110         TAILQ_ENTRY(hc_metrics) rmx_q;
 111         struct  hc_head *rmx_head; /* head of bucket tail queue */
 112         struct  in_addr ip4;    /* IP address */
 113         struct  in6_addr ip6;   /* IP6 address */
 114         /* endpoint specific values for TCP */
 115         u_long  rmx_mtu;        /* MTU for this path */
 116         u_long  rmx_ssthresh;   /* outbound gateway buffer limit */
 117         u_long  rmx_rtt;        /* estimated round trip time */
 118         u_long  rmx_rttvar;     /* estimated rtt variance */
 119         u_long  rmx_bandwidth;  /* estimated bandwidth */
 120         u_long  rmx_cwnd;       /* congestion window */
 121         u_long  rmx_sendpipe;   /* outbound delay-bandwidth product */
 122         u_long  rmx_recvpipe;   /* inbound delay-bandwidth product */
 123         /* TCP hostcache internal data */
 124         int     rmx_expire;     /* lifetime for object */
 125         u_long  rmx_hits;       /* number of hits */
 126         u_long  rmx_updates;    /* number of updates */
 127 };
 128
 129 /* Arbitrary values */
 130 #define TCP_HOSTCACHE_HASHSIZE          512
 131 #define TCP_HOSTCACHE_BUCKETLIMIT       30
 132 #define TCP_HOSTCACHE_EXPIRE            60*60   /* one hour */
 133 #define TCP_HOSTCACHE_PRUNE             5*60    /* every 5 minutes */
 134
 135 struct tcp_hostcache {
 136         struct  hc_head *hashbase;
 137         uma_zone_t zone;
 138         u_int   hashsize;
 139         u_int   hashmask;
 140         u_int   bucket_limit;
 141         u_int   cache_count;
 142         u_int   cache_limit;
 143         int     expire;
 144         int     purgeall;
 145 };
 146 static struct tcp_hostcache tcp_hostcache;
 147
 148 static struct callout tcp_hc_callout;
 149
 150 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
 151 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
 152 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
 153 static void tcp_hc_purge(void *);
 154
 155 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, "TCP Host cache");
 156
 157 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
 158     &tcp_hostcache.cache_limit, 0, "Overall entry limit for hostcache");
 159
 160 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
 161     &tcp_hostcache.hashsize, 0, "Size of TCP hostcache hashtable");
 162
 163 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
 164     &tcp_hostcache.bucket_limit, 0, "Per-bucket hash limit for hostcache");
 165
 166 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD,
 167     &tcp_hostcache.cache_count, 0, "Current number of entries in hostcache");
 168
 169 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW,
 170     &tcp_hostcache.expire, 0, "Expire time of TCP hostcache entries");
 171
 172 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW,
 173     &tcp_hostcache.purgeall, 0, "Expire all entires on next purge run");
 174
 175 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
 176     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
 177     sysctl_tcp_hc_list, "A", "List of all hostcache entries");
 178
 179
 180 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
 181
 182 #define HOSTCACHE_HASH(ip) \
 183         (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) &  \
 184           tcp_hostcache.hashmask)
 185
 186 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
 187 #define HOSTCACHE_HASH6(ip6)                            \
 188         (((ip6)->s6_addr32[0] ^                         \
 189           (ip6)->s6_addr32[1] ^                         \
 190           (ip6)->s6_addr32[2] ^                         \
 191           (ip6)->s6_addr32[3]) &                        \
 192          tcp_hostcache.hashmask)
 193
 194 #define THC_LOCK(lp)            mtx_lock(lp)
 195 #define THC_UNLOCK(lp)          mtx_unlock(lp)
 196
 197 void
 198 tcp_hc_init(void)
 199 {
 200         int i;
 201
 202         /*
 203          * Initialize hostcache structures.
 204          */
 205         tcp_hostcache.cache_count = 0;
 206         tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
 207         tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
 208         tcp_hostcache.cache_limit =
 209             tcp_hostcache.hashsize * tcp_hostcache.bucket_limit;
 210         tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
 211
 212         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
 213             &tcp_hostcache.hashsize);
 214         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
 215             &tcp_hostcache.cache_limit);
 216         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
 217             &tcp_hostcache.bucket_limit);
 218         if (!powerof2(tcp_hostcache.hashsize)) {
 219                 printf("WARNING: hostcache hash size is not a power of 2.\n");
 220                 tcp_hostcache.hashsize = 512;   /* safe default */
 221         }
 222         tcp_hostcache.hashmask = tcp_hostcache.hashsize - 1;
 223
 224         /*
 225          * Allocate the hash table.
 226          */
 227         tcp_hostcache.hashbase = (struct hc_head *)
 228             malloc(tcp_hostcache.hashsize * sizeof(struct hc_head),
 229                    M_HOSTCACHE, M_WAITOK | M_ZERO);
 230
 231         /*
 232          * Initialize the hash buckets.
 233          */
 234         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 235                 TAILQ_INIT(&tcp_hostcache.hashbase[i].hch_bucket);
 236                 tcp_hostcache.hashbase[i].hch_length = 0;
 237                 mtx_init(&tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
 238                           NULL, MTX_DEF);
 239         }
 240
 241         /*
 242          * Allocate the hostcache entries.
 243          */
 244         tcp_hostcache.zone = uma_zcreate("hostcache", sizeof(struct hc_metrics),
 245             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 246         uma_zone_set_max(tcp_hostcache.zone, tcp_hostcache.cache_limit);
 247
 248         /*
 249          * Set up periodic cache cleanup.
 250          */
 251         callout_init(&tcp_hc_callout, CALLOUT_MPSAFE);
 252         callout_reset(&tcp_hc_callout, TCP_HOSTCACHE_PRUNE * hz, tcp_hc_purge, 0);
 253 }
 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_isipv6)
 274                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
 275         else
 276                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
 277
 278         hc_head = &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_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_isipv6)
 329                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
 330         else
 331                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
 332
 333         hc_head = &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 >= tcp_hostcache.bucket_limit ||
 346             tcp_hostcache.cache_count >= 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                  */
 355                 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
 356                 tcp_hostcache.hashbase[hash].hch_length--;
 357                 tcp_hostcache.cache_count--;
 358                 tcpstat.tcps_hc_bucketoverflow++;
 359 #if 0
 360                 uma_zfree(tcp_hostcache.zone, hc_entry);
 361 #endif
 362         } else {
 363                 /*
 364                  * Allocate a new entry, or balk if not possible.
 365                  */
 366                 hc_entry = uma_zalloc(tcp_hostcache.zone, M_NOWAIT);
 367                 if (hc_entry == NULL) {
 368                         THC_UNLOCK(&hc_head->hch_mtx);
 369                         return NULL;
 370                 }
 371         }
 372
 373         /*
 374          * Initialize basic information of hostcache entry.
 375          */
 376         bzero(hc_entry, sizeof(*hc_entry));
 377         if (inc->inc_isipv6)
 378                 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6));
 379         else
 380                 hc_entry->ip4 = inc->inc_faddr;
 381         hc_entry->rmx_head = hc_head;
 382         hc_entry->rmx_expire = tcp_hostcache.expire;
 383
 384         /*
 385          * Put it upfront.
 386          */
 387         TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
 388         tcp_hostcache.hashbase[hash].hch_length++;
 389         tcp_hostcache.cache_count++;
 390         tcpstat.tcps_hc_added++;
 391
 392         return hc_entry;
 393 }
 394
 395 /*
 396  * External function: look up an entry in the hostcache and fill out the
 397  * supplied TCP metrics structure.  Fills in NULL when no entry was found or
 398  * a value is not set.
 399  */
 400 void
 401 tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
 402 {
 403         struct hc_metrics *hc_entry;
 404
 405         /*
 406          * Find the right bucket.
 407          */
 408         hc_entry = tcp_hc_lookup(inc);
 409
 410         /*
 411          * If we don't have an existing object.
 412          */
 413         if (hc_entry == NULL) {
 414                 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
 415                 return;
 416         }
 417         hc_entry->rmx_hits++;
 418         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 419
 420         hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
 421         hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
 422         hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
 423         hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
 424         hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth;
 425         hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
 426         hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
 427         hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
 428
 429         /*
 430          * Unlock bucket row.
 431          */
 432         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 433 }
 434
 435 /*
 436  * External function: look up an entry in the hostcache and return the
 437  * discovered path MTU.  Returns NULL if no entry is found or value is not
 438  * set.
 439  */
 440 u_long
 441 tcp_hc_getmtu(struct in_conninfo *inc)
 442 {
 443         struct hc_metrics *hc_entry;
 444         u_long mtu;
 445
 446         hc_entry = tcp_hc_lookup(inc);
 447         if (hc_entry == NULL) {
 448                 return 0;
 449         }
 450         hc_entry->rmx_hits++;
 451         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 452
 453         mtu = hc_entry->rmx_mtu;
 454         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 455         return mtu;
 456 }
 457
 458 /*
 459  * External function: update the MTU value of an entry in the hostcache.
 460  * Creates a new entry if none was found.
 461  */
 462 void
 463 tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
 464 {
 465         struct hc_metrics *hc_entry;
 466
 467         /*
 468          * Find the right bucket.
 469          */
 470         hc_entry = tcp_hc_lookup(inc);
 471
 472         /*
 473          * If we don't have an existing object, try to insert a new one.
 474          */
 475         if (hc_entry == NULL) {
 476                 hc_entry = tcp_hc_insert(inc);
 477                 if (hc_entry == NULL)
 478                         return;
 479         }
 480         hc_entry->rmx_updates++;
 481         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 482
 483         hc_entry->rmx_mtu = mtu;
 484
 485         /*
 486          * Put it upfront so we find it faster next time.
 487          */
 488         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 489         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 490
 491         /*
 492          * Unlock bucket row.
 493          */
 494         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 495 }
 496
 497 /*
 498  * External function: update the TCP metrics of an entry in the hostcache.
 499  * Creates a new entry if none was found.
 500  */
 501 void
 502 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
 503 {
 504         struct hc_metrics *hc_entry;
 505
 506         hc_entry = tcp_hc_lookup(inc);
 507         if (hc_entry == NULL) {
 508                 hc_entry = tcp_hc_insert(inc);
 509                 if (hc_entry == NULL)
 510                         return;
 511         }
 512         hc_entry->rmx_updates++;
 513         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 514
 515         if (hcml->rmx_rtt != 0) {
 516                 if (hc_entry->rmx_rtt == 0)
 517                         hc_entry->rmx_rtt = hcml->rmx_rtt;
 518                 else
 519                         hc_entry->rmx_rtt =
 520                             (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
 521                 tcpstat.tcps_cachedrtt++;
 522         }
 523         if (hcml->rmx_rttvar != 0) {
 524                 if (hc_entry->rmx_rttvar == 0)
 525                         hc_entry->rmx_rttvar = hcml->rmx_rttvar;
 526                 else
 527                         hc_entry->rmx_rttvar =
 528                             (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
 529                 tcpstat.tcps_cachedrttvar++;
 530         }
 531         if (hcml->rmx_ssthresh != 0) {
 532                 if (hc_entry->rmx_ssthresh == 0)
 533                         hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
 534                 else
 535                         hc_entry->rmx_ssthresh =
 536                             (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
 537                 tcpstat.tcps_cachedssthresh++;
 538         }
 539         if (hcml->rmx_bandwidth != 0) {
 540                 if (hc_entry->rmx_bandwidth == 0)
 541                         hc_entry->rmx_bandwidth = hcml->rmx_bandwidth;
 542                 else
 543                         hc_entry->rmx_bandwidth =
 544                             (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2;
 545                 /* tcpstat.tcps_cachedbandwidth++; */
 546         }
 547         if (hcml->rmx_cwnd != 0) {
 548                 if (hc_entry->rmx_cwnd == 0)
 549                         hc_entry->rmx_cwnd = hcml->rmx_cwnd;
 550                 else
 551                         hc_entry->rmx_cwnd =
 552                             (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
 553                 /* tcpstat.tcps_cachedcwnd++; */
 554         }
 555         if (hcml->rmx_sendpipe != 0) {
 556                 if (hc_entry->rmx_sendpipe == 0)
 557                         hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
 558                 else
 559                         hc_entry->rmx_sendpipe =
 560                             (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
 561                 /* tcpstat.tcps_cachedsendpipe++; */
 562         }
 563         if (hcml->rmx_recvpipe != 0) {
 564                 if (hc_entry->rmx_recvpipe == 0)
 565                         hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
 566                 else
 567                         hc_entry->rmx_recvpipe =
 568                             (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
 569                 /* tcpstat.tcps_cachedrecvpipe++; */
 570         }
 571
 572         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 573         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 574         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 575 }
 576
 577 /*
 578  * Sysctl function: prints the list and values of all hostcache entries in
 579  * unsorted order.
 580  */
 581 static int
 582 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
 583 {
 584         int bufsize;
 585         int linesize = 128;
 586         char *p, *buf;
 587         int len, i, error;
 588         struct hc_metrics *hc_entry;
 589 #ifdef INET6
 590         char ip6buf[INET6_ADDRSTRLEN];
 591 #endif
 592
 593         bufsize = linesize * (tcp_hostcache.cache_count + 1);
 594
 595         p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
 596
 597         len = snprintf(p, linesize,
 598                 "\nIP address        MTU  SSTRESH      RTT   RTTVAR BANDWIDTH "
 599                 "    CWND SENDPIPE RECVPIPE HITS  UPD  EXP\n");
 600         p += len;
 601
 602 #define msec(u) (((u) + 500) / 1000)
 603         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 604                 THC_LOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 605                 TAILQ_FOREACH(hc_entry, &tcp_hostcache.hashbase[i].hch_bucket,
 606                               rmx_q) {
 607                         len = snprintf(p, linesize,
 608                             "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu "
 609                             "%4lu %4lu %4i\n",
 610                             hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
 611 #ifdef INET6
 612                                 ip6_sprintf(ip6buf, &hc_entry->ip6),
 613 #else
 614                                 "IPv6?",
 615 #endif
 616                             hc_entry->rmx_mtu,
 617                             hc_entry->rmx_ssthresh,
 618                             msec(hc_entry->rmx_rtt *
 619                                 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
 620                             msec(hc_entry->rmx_rttvar *
 621                                 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
 622                             hc_entry->rmx_bandwidth * 8,
 623                             hc_entry->rmx_cwnd,
 624                             hc_entry->rmx_sendpipe,
 625                             hc_entry->rmx_recvpipe,
 626                             hc_entry->rmx_hits,
 627                             hc_entry->rmx_updates,
 628                             hc_entry->rmx_expire);
 629                         p += len;
 630                 }
 631                 THC_UNLOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 632         }
 633 #undef msec
 634         error = SYSCTL_OUT(req, buf, p - buf);
 635         free(buf, M_TEMP);
 636         return(error);
 637 }
 638
 639 /*
 640  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
 641  * periodically from the callout.
 642  */
 643 static void
 644 tcp_hc_purge(void *arg)
 645 {
 646         struct hc_metrics *hc_entry, *hc_next;
 647         int all = (intptr_t)arg;
 648         int i;
 649
 650         if (tcp_hostcache.purgeall) {
 651                 all = 1;
 652                 tcp_hostcache.purgeall = 0;
 653         }
 654
 655         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 656                 THC_LOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 657                 TAILQ_FOREACH_SAFE(hc_entry, &tcp_hostcache.hashbase[i].hch_bucket,
 658                               rmx_q, hc_next) {
 659                         if (all || hc_entry->rmx_expire <= 0) {
 660                                 TAILQ_REMOVE(&tcp_hostcache.hashbase[i].hch_bucket,
 661                                               hc_entry, rmx_q);
 662                                 uma_zfree(tcp_hostcache.zone, hc_entry);
 663                                 tcp_hostcache.hashbase[i].hch_length--;
 664                                 tcp_hostcache.cache_count--;
 665                         } else
 666                                 hc_entry->rmx_expire -= TCP_HOSTCACHE_PRUNE;
 667                 }
 668                 THC_UNLOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 669         }
 670         callout_reset(&tcp_hc_callout, TCP_HOSTCACHE_PRUNE * hz, tcp_hc_purge, 0);
 671 }