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     prune;
 145         int     purgeall;
 146 };
 147 static struct tcp_hostcache tcp_hostcache;
 148
 149 static struct callout tcp_hc_callout;
 150
 151 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
 152 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
 153 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
 154 static void tcp_hc_purge(void *);
 155
 156 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, "TCP Host cache");
 157
 158 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
 159     &tcp_hostcache.cache_limit, 0, "Overall entry limit for hostcache");
 160
 161 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
 162     &tcp_hostcache.hashsize, 0, "Size of TCP hostcache hashtable");
 163
 164 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
 165     &tcp_hostcache.bucket_limit, 0, "Per-bucket hash limit for hostcache");
 166
 167 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD,
 168     &tcp_hostcache.cache_count, 0, "Current number of entries in hostcache");
 169
 170 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW,
 171     &tcp_hostcache.expire, 0, "Expire time of TCP hostcache entries");
 172
 173 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW,
 174      &tcp_hostcache.prune, 0, "Time between purge runs");
 175
 176 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW,
 177     &tcp_hostcache.purgeall, 0, "Expire all entires on next purge run");
 178
 179 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
 180     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
 181     sysctl_tcp_hc_list, "A", "List of all hostcache entries");
 182
 183
 184 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
 185
 186 #define HOSTCACHE_HASH(ip) \
 187         (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) &  \
 188           tcp_hostcache.hashmask)
 189
 190 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
 191 #define HOSTCACHE_HASH6(ip6)                            \
 192         (((ip6)->s6_addr32[0] ^                         \
 193           (ip6)->s6_addr32[1] ^                         \
 194           (ip6)->s6_addr32[2] ^                         \
 195           (ip6)->s6_addr32[3]) &                        \
 196          tcp_hostcache.hashmask)
 197
 198 #define THC_LOCK(lp)            mtx_lock(lp)
 199 #define THC_UNLOCK(lp)          mtx_unlock(lp)
 200
 201 void
 202 tcp_hc_init(void)
 203 {
 204         int i;
 205
 206         /*
 207          * Initialize hostcache structures.
 208          */
 209         tcp_hostcache.cache_count = 0;
 210         tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
 211         tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
 212         tcp_hostcache.cache_limit =
 213             tcp_hostcache.hashsize * tcp_hostcache.bucket_limit;
 214         tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
 215         tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
 216
 217         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
 218             &tcp_hostcache.hashsize);
 219         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
 220             &tcp_hostcache.cache_limit);
 221         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
 222             &tcp_hostcache.bucket_limit);
 223         if (!powerof2(tcp_hostcache.hashsize)) {
 224                 printf("WARNING: hostcache hash size is not a power of 2.\n");
 225                 tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
 226         }
 227         tcp_hostcache.hashmask = tcp_hostcache.hashsize - 1;
 228
 229         /*
 230          * Allocate the hash table.
 231          */
 232         tcp_hostcache.hashbase = (struct hc_head *)
 233             malloc(tcp_hostcache.hashsize * sizeof(struct hc_head),
 234                    M_HOSTCACHE, M_WAITOK | M_ZERO);
 235
 236         /*
 237          * Initialize the hash buckets.
 238          */
 239         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 240                 TAILQ_INIT(&tcp_hostcache.hashbase[i].hch_bucket);
 241                 tcp_hostcache.hashbase[i].hch_length = 0;
 242                 mtx_init(&tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
 243                           NULL, MTX_DEF);
 244         }
 245
 246         /*
 247          * Allocate the hostcache entries.
 248          */
 249         tcp_hostcache.zone = uma_zcreate("hostcache", sizeof(struct hc_metrics),
 250             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 251         uma_zone_set_max(tcp_hostcache.zone, tcp_hostcache.cache_limit);
 252
 253         /*
 254          * Set up periodic cache cleanup.
 255          */
 256         callout_init(&tcp_hc_callout, CALLOUT_MPSAFE);
 257         callout_reset(&tcp_hc_callout, tcp_hostcache.prune * hz, tcp_hc_purge, 0);
 258 }
 259
 260 /*
 261  * Internal function: look up an entry in the hostcache or return NULL.
 262  *
 263  * If an entry has been returned, the caller becomes responsible for
 264  * unlocking the bucket row after he is done reading/modifying the entry.
 265  */
 266 static struct hc_metrics *
 267 tcp_hc_lookup(struct in_conninfo *inc)
 268 {
 269         int hash;
 270         struct hc_head *hc_head;
 271         struct hc_metrics *hc_entry;
 272
 273         KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer"));
 274
 275         /*
 276          * Hash the foreign ip address.
 277          */
 278         if (inc->inc_isipv6)
 279                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
 280         else
 281                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
 282
 283         hc_head = &tcp_hostcache.hashbase[hash];
 284
 285         /*
 286          * Acquire lock for this bucket row; we release the lock if we don't
 287          * find an entry, otherwise the caller has to unlock after he is
 288          * done.
 289          */
 290         THC_LOCK(&hc_head->hch_mtx);
 291
 292         /*
 293          * Iterate through entries in bucket row looking for a match.
 294          */
 295         TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) {
 296                 if (inc->inc_isipv6) {
 297                         if (memcmp(&inc->inc6_faddr, &hc_entry->ip6,
 298                             sizeof(inc->inc6_faddr)) == 0)
 299                                 return hc_entry;
 300                 } else {
 301                         if (memcmp(&inc->inc_faddr, &hc_entry->ip4,
 302                             sizeof(inc->inc_faddr)) == 0)
 303                                 return hc_entry;
 304                 }
 305         }
 306
 307         /*
 308          * We were unsuccessful and didn't find anything.
 309          */
 310         THC_UNLOCK(&hc_head->hch_mtx);
 311         return NULL;
 312 }
 313
 314 /*
 315  * Internal function: insert an entry into the hostcache or return NULL if
 316  * unable to allocate a new one.
 317  *
 318  * If an entry has been returned, the caller becomes responsible for
 319  * unlocking the bucket row after he is done reading/modifying the entry.
 320  */
 321 static struct hc_metrics *
 322 tcp_hc_insert(struct in_conninfo *inc)
 323 {
 324         int hash;
 325         struct hc_head *hc_head;
 326         struct hc_metrics *hc_entry;
 327
 328         KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer"));
 329
 330         /*
 331          * Hash the foreign ip address.
 332          */
 333         if (inc->inc_isipv6)
 334                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
 335         else
 336                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
 337
 338         hc_head = &tcp_hostcache.hashbase[hash];
 339
 340         /*
 341          * Acquire lock for this bucket row; we release the lock if we don't
 342          * find an entry, otherwise the caller has to unlock after he is
 343          * done.
 344          */
 345         THC_LOCK(&hc_head->hch_mtx);
 346
 347         /*
 348          * If the bucket limit is reached, reuse the least-used element.
 349          */
 350         if (hc_head->hch_length >= tcp_hostcache.bucket_limit ||
 351             tcp_hostcache.cache_count >= tcp_hostcache.cache_limit) {
 352                 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead);
 353                 /*
 354                  * At first we were dropping the last element, just to
 355                  * reacquire it in the next two lines again, which isn't very
 356                  * efficient.  Instead just reuse the least used element.
 357                  * We may drop something that is still "in-use" but we can be
 358                  * "lossy".
 359                  * Just give up if this bucket row is empty and we don't have
 360                  * anything to replace.
 361                  */
 362                 if (hc_entry == NULL) {
 363                         THC_UNLOCK(&hc_head->hch_mtx);
 364                         return NULL;
 365                 }
 366                 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
 367                 tcp_hostcache.hashbase[hash].hch_length--;
 368                 tcp_hostcache.cache_count--;
 369                 tcpstat.tcps_hc_bucketoverflow++;
 370 #if 0
 371                 uma_zfree(tcp_hostcache.zone, hc_entry);
 372 #endif
 373         } else {
 374                 /*
 375                  * Allocate a new entry, or balk if not possible.
 376                  */
 377                 hc_entry = uma_zalloc(tcp_hostcache.zone, M_NOWAIT);
 378                 if (hc_entry == NULL) {
 379                         THC_UNLOCK(&hc_head->hch_mtx);
 380                         return NULL;
 381                 }
 382         }
 383
 384         /*
 385          * Initialize basic information of hostcache entry.
 386          */
 387         bzero(hc_entry, sizeof(*hc_entry));
 388         if (inc->inc_isipv6)
 389                 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6));
 390         else
 391                 hc_entry->ip4 = inc->inc_faddr;
 392         hc_entry->rmx_head = hc_head;
 393         hc_entry->rmx_expire = tcp_hostcache.expire;
 394
 395         /*
 396          * Put it upfront.
 397          */
 398         TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
 399         tcp_hostcache.hashbase[hash].hch_length++;
 400         tcp_hostcache.cache_count++;
 401         tcpstat.tcps_hc_added++;
 402
 403         return hc_entry;
 404 }
 405
 406 /*
 407  * External function: look up an entry in the hostcache and fill out the
 408  * supplied TCP metrics structure.  Fills in NULL when no entry was found or
 409  * a value is not set.
 410  */
 411 void
 412 tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
 413 {
 414         struct hc_metrics *hc_entry;
 415
 416         /*
 417          * Find the right bucket.
 418          */
 419         hc_entry = tcp_hc_lookup(inc);
 420
 421         /*
 422          * If we don't have an existing object.
 423          */
 424         if (hc_entry == NULL) {
 425                 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
 426                 return;
 427         }
 428         hc_entry->rmx_hits++;
 429         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 430
 431         hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
 432         hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
 433         hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
 434         hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
 435         hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth;
 436         hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
 437         hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
 438         hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
 439
 440         /*
 441          * Unlock bucket row.
 442          */
 443         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 444 }
 445
 446 /*
 447  * External function: look up an entry in the hostcache and return the
 448  * discovered path MTU.  Returns NULL if no entry is found or value is not
 449  * set.
 450  */
 451 u_long
 452 tcp_hc_getmtu(struct in_conninfo *inc)
 453 {
 454         struct hc_metrics *hc_entry;
 455         u_long mtu;
 456
 457         hc_entry = tcp_hc_lookup(inc);
 458         if (hc_entry == NULL) {
 459                 return 0;
 460         }
 461         hc_entry->rmx_hits++;
 462         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 463
 464         mtu = hc_entry->rmx_mtu;
 465         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 466         return mtu;
 467 }
 468
 469 /*
 470  * External function: update the MTU value of an entry in the hostcache.
 471  * Creates a new entry if none was found.
 472  */
 473 void
 474 tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
 475 {
 476         struct hc_metrics *hc_entry;
 477
 478         /*
 479          * Find the right bucket.
 480          */
 481         hc_entry = tcp_hc_lookup(inc);
 482
 483         /*
 484          * If we don't have an existing object, try to insert a new one.
 485          */
 486         if (hc_entry == NULL) {
 487                 hc_entry = tcp_hc_insert(inc);
 488                 if (hc_entry == NULL)
 489                         return;
 490         }
 491         hc_entry->rmx_updates++;
 492         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 493
 494         hc_entry->rmx_mtu = mtu;
 495
 496         /*
 497          * Put it upfront so we find it faster next time.
 498          */
 499         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 500         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 501
 502         /*
 503          * Unlock bucket row.
 504          */
 505         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 506 }
 507
 508 /*
 509  * External function: update the TCP metrics of an entry in the hostcache.
 510  * Creates a new entry if none was found.
 511  */
 512 void
 513 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
 514 {
 515         struct hc_metrics *hc_entry;
 516
 517         hc_entry = tcp_hc_lookup(inc);
 518         if (hc_entry == NULL) {
 519                 hc_entry = tcp_hc_insert(inc);
 520                 if (hc_entry == NULL)
 521                         return;
 522         }
 523         hc_entry->rmx_updates++;
 524         hc_entry->rmx_expire = tcp_hostcache.expire; /* start over again */
 525
 526         if (hcml->rmx_rtt != 0) {
 527                 if (hc_entry->rmx_rtt == 0)
 528                         hc_entry->rmx_rtt = hcml->rmx_rtt;
 529                 else
 530                         hc_entry->rmx_rtt =
 531                             (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
 532                 tcpstat.tcps_cachedrtt++;
 533         }
 534         if (hcml->rmx_rttvar != 0) {
 535                 if (hc_entry->rmx_rttvar == 0)
 536                         hc_entry->rmx_rttvar = hcml->rmx_rttvar;
 537                 else
 538                         hc_entry->rmx_rttvar =
 539                             (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
 540                 tcpstat.tcps_cachedrttvar++;
 541         }
 542         if (hcml->rmx_ssthresh != 0) {
 543                 if (hc_entry->rmx_ssthresh == 0)
 544                         hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
 545                 else
 546                         hc_entry->rmx_ssthresh =
 547                             (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
 548                 tcpstat.tcps_cachedssthresh++;
 549         }
 550         if (hcml->rmx_bandwidth != 0) {
 551                 if (hc_entry->rmx_bandwidth == 0)
 552                         hc_entry->rmx_bandwidth = hcml->rmx_bandwidth;
 553                 else
 554                         hc_entry->rmx_bandwidth =
 555                             (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2;
 556                 /* tcpstat.tcps_cachedbandwidth++; */
 557         }
 558         if (hcml->rmx_cwnd != 0) {
 559                 if (hc_entry->rmx_cwnd == 0)
 560                         hc_entry->rmx_cwnd = hcml->rmx_cwnd;
 561                 else
 562                         hc_entry->rmx_cwnd =
 563                             (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
 564                 /* tcpstat.tcps_cachedcwnd++; */
 565         }
 566         if (hcml->rmx_sendpipe != 0) {
 567                 if (hc_entry->rmx_sendpipe == 0)
 568                         hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
 569                 else
 570                         hc_entry->rmx_sendpipe =
 571                             (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
 572                 /* tcpstat.tcps_cachedsendpipe++; */
 573         }
 574         if (hcml->rmx_recvpipe != 0) {
 575                 if (hc_entry->rmx_recvpipe == 0)
 576                         hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
 577                 else
 578                         hc_entry->rmx_recvpipe =
 579                             (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
 580                 /* tcpstat.tcps_cachedrecvpipe++; */
 581         }
 582
 583         TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 584         TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
 585         THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
 586 }
 587
 588 /*
 589  * Sysctl function: prints the list and values of all hostcache entries in
 590  * unsorted order.
 591  */
 592 static int
 593 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
 594 {
 595         int bufsize;
 596         int linesize = 128;
 597         char *p, *buf;
 598         int len, i, error;
 599         struct hc_metrics *hc_entry;
 600 #ifdef INET6
 601         char ip6buf[INET6_ADDRSTRLEN];
 602 #endif
 603
 604         bufsize = linesize * (tcp_hostcache.cache_count + 1);
 605
 606         p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
 607
 608         len = snprintf(p, linesize,
 609                 "\nIP address        MTU  SSTRESH      RTT   RTTVAR BANDWIDTH "
 610                 "    CWND SENDPIPE RECVPIPE HITS  UPD  EXP\n");
 611         p += len;
 612
 613 #define msec(u) (((u) + 500) / 1000)
 614         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 615                 THC_LOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 616                 TAILQ_FOREACH(hc_entry, &tcp_hostcache.hashbase[i].hch_bucket,
 617                               rmx_q) {
 618                         len = snprintf(p, linesize,
 619                             "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu "
 620                             "%4lu %4lu %4i\n",
 621                             hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
 622 #ifdef INET6
 623                                 ip6_sprintf(ip6buf, &hc_entry->ip6),
 624 #else
 625                                 "IPv6?",
 626 #endif
 627                             hc_entry->rmx_mtu,
 628                             hc_entry->rmx_ssthresh,
 629                             msec(hc_entry->rmx_rtt *
 630                                 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
 631                             msec(hc_entry->rmx_rttvar *
 632                                 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
 633                             hc_entry->rmx_bandwidth * 8,
 634                             hc_entry->rmx_cwnd,
 635                             hc_entry->rmx_sendpipe,
 636                             hc_entry->rmx_recvpipe,
 637                             hc_entry->rmx_hits,
 638                             hc_entry->rmx_updates,
 639                             hc_entry->rmx_expire);
 640                         p += len;
 641                 }
 642                 THC_UNLOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 643         }
 644 #undef msec
 645         error = SYSCTL_OUT(req, buf, p - buf);
 646         free(buf, M_TEMP);
 647         return(error);
 648 }
 649
 650 /*
 651  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
 652  * periodically from the callout.
 653  */
 654 static void
 655 tcp_hc_purge(void *arg)
 656 {
 657         struct hc_metrics *hc_entry, *hc_next;
 658         int all = (intptr_t)arg;
 659         int i;
 660
 661         if (tcp_hostcache.purgeall) {
 662                 all = 1;
 663                 tcp_hostcache.purgeall = 0;
 664         }
 665
 666         for (i = 0; i < tcp_hostcache.hashsize; i++) {
 667                 THC_LOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 668                 TAILQ_FOREACH_SAFE(hc_entry, &tcp_hostcache.hashbase[i].hch_bucket,
 669                               rmx_q, hc_next) {
 670                         if (all || hc_entry->rmx_expire <= 0) {
 671                                 TAILQ_REMOVE(&tcp_hostcache.hashbase[i].hch_bucket,
 672                                               hc_entry, rmx_q);
 673                                 uma_zfree(tcp_hostcache.zone, hc_entry);
 674                                 tcp_hostcache.hashbase[i].hch_length--;
 675                                 tcp_hostcache.cache_count--;
 676                         } else
 677                                 hc_entry->rmx_expire -= tcp_hostcache.prune;
 678                 }
 679                 THC_UNLOCK(&tcp_hostcache.hashbase[i].hch_mtx);
 680         }
 681         callout_reset(&tcp_hc_callout, tcp_hostcache.prune * hz, tcp_hc_purge, 0);
 682 }