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 have been removed.  The inpcb no longer keeps a pointer to
  41  * the routing entry, and protocol-initiated route cloning has been removed
  42  * as well.  With these changes, the routing table has gone back to being
  43  * more lightwight and only carries information related to packet forwarding.
  44  *
  45  * tcp_hostcache is designed for multiple concurrent access in SMP
  46  * environments and high contention.  All bucket rows have their own lock and
  47  * thus multiple lookups and modifies can be done at the same time as long as
  48  * they are in different bucket rows.  If a request for insertion of a new
  49  * record can't be satisfied, it simply returns an empty structure.  Nobody
  50  * and nothing outside of tcp_hostcache.c will ever point directly to any
  51  * entry in the tcp_hostcache.  All communication is done in an
  52  * object-oriented way and only functions of tcp_hostcache will manipulate
  53  * hostcache entries.  Otherwise, we are unable to achieve good behaviour in
  54  * concurrent access situations.  Since tcp_hostcache is only caching
  55  * information, there are no fatal consequences if we either can't satisfy
  56  * any particular request or have to drop/overwrite an existing entry because
  57  * of bucket limit memory constrains.
  58  */
  59
  60 /*
  61  * Many thanks to jlemon for basic structure of tcp_syncache which is being
  62  * followed here.
  63  */
  64
  65 #include <sys/cdefs.h>
  66 __FBSDID("$FreeBSD$");
  67
  68 #include "opt_inet6.h"
  69
  70 #include <sys/param.h>
  71 #include <sys/systm.h>
  72 #include <sys/kernel.h>
  73 #include <sys/lock.h>
  74 #include <sys/mutex.h>
  75 #include <sys/malloc.h>
  76 #include <sys/socket.h>
  77 #include <sys/socketvar.h>
  78 #include <sys/sysctl.h>
  79
  80 #include <net/if.h>
  81 #include <net/route.h>
  82 #include <net/vnet.h>
  83
  84 #include <netinet/in.h>
  85 #include <netinet/in_systm.h>
  86 #include <netinet/ip.h>
  87 #include <netinet/in_var.h>
  88 #include <netinet/in_pcb.h>
  89 #include <netinet/ip_var.h>
  90 #ifdef INET6
  91 #include <netinet/ip6.h>
  92 #include <netinet6/ip6_var.h>
  93 #endif
  94 #include <netinet/tcp.h>
  95 #include <netinet/tcp_var.h>
  96 #include <netinet/tcp_hostcache.h>
  97 #ifdef INET6
  98 #include <netinet6/tcp6_var.h>
  99 #endif
 100
 101 #include <vm/uma.h>
 102
 103 /* Arbitrary values */
 104 #define TCP_HOSTCACHE_HASHSIZE          512
 105 #define TCP_HOSTCACHE_BUCKETLIMIT       30
 106 #define TCP_HOSTCACHE_EXPIRE            60*60   /* one hour */
 107 #define TCP_HOSTCACHE_PRUNE             5*60    /* every 5 minutes */
 108
 109 static VNET_DEFINE(struct tcp_hostcache, tcp_hostcache);
 110 #define V_tcp_hostcache         VNET(tcp_hostcache)
 111
 112 static VNET_DEFINE(struct callout, tcp_hc_callout);
 113 #define V_tcp_hc_callout        VNET(tcp_hc_callout)
 114
 115 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
 116 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
 117 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
 118 static void tcp_hc_purge_internal(int);
 119 static void tcp_hc_purge(void *);
 120
 121 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0,
 122     "TCP Host cache");
 123
 124 SYSCTL_VNET_UINT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
 125     &VNET_NAME(tcp_hostcache.cache_limit), 0,
 126     "Overall entry limit for hostcache");
 127
 128 SYSCTL_VNET_UINT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
 129     &VNET_NAME(tcp_hostcache.hashsize), 0,
 130     "Size of TCP hostcache hashtable");
 131
 132 SYSCTL_VNET_UINT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit,
 133     CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.bucket_limit), 0,
 134     "Per-bucket hash limit for hostcache");
 135
 136 SYSCTL_VNET_UINT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD,
 137      &VNET_NAME(tcp_hostcache.cache_count), 0,
 138     "Current number of entries in hostcache");
 139
 140 SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW,
 141     &VNET_NAME(tcp_hostcache.expire), 0,
 142     "Expire time of TCP hostcache entries");
 143
 144 SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW,
 145     &VNET_NAME(tcp_hostcache.prune), 0,
 146     "Time between purge runs");
 147
 148 SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW,
 149     &VNET_NAME(tcp_hostcache.purgeall), 0,
 150     "Expire all entires on next purge run");
 151
 152 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
 153     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
 154     sysctl_tcp_hc_list, "A", "List of all hostcache entries");
 155
 156
 157 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
 158
 159 #define HOSTCACHE_HASH(ip) \
 160         (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) &  \
 161           V_tcp_hostcache.hashmask)
 162
 163 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
 164 #define HOSTCACHE_HASH6(ip6)                            \
 165         (((ip6)->s6_addr32[0] ^                         \
 166           (ip6)->s6_addr32[1] ^                         \
 167           (ip6)->s6_addr32[2] ^                         \
 168           (ip6)->s6_addr32[3]) &                        \
 169          V_tcp_hostcache.hashmask)
 170
 171 #define THC_LOCK(lp)            mtx_lock(lp)
 172 #define THC_UNLOCK(lp)          mtx_unlock(lp)
 173
 174 void
 175 tcp_hc_init(void)
 176 {
 177         u_int cache_limit;
 178         int i;
 179
 180         /*
 181          * Initialize hostcache structures.
 182          */
 183         V_tcp_hostcache.cache_count = 0;
 184         V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
 185         V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
 186         V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
 187         V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
 188
 189         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
 190             &V_tcp_hostcache.hashsize);
 191         if (!powerof2(V_tcp_hostcache.hashsize)) {
 192                 printf("WARNING: hostcache hash size is not a power of 2.\n");
 193                 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
 194         }
 195         V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1;
 196
 197         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
 198             &V_tcp_hostcache.bucket_limit);
 199
 200         cache_limit = V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit;
 201         V_tcp_hostcache.cache_limit = cache_limit;
 202         TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
 203             &V_tcp_hostcache.cache_limit);
 204         if (V_tcp_hostcache.cache_limit > cache_limit)
 205                 V_tcp_hostcache.cache_limit = cache_limit;
 206
 207         /*
 208          * Allocate the hash table.
 209          */
 210         V_tcp_hostcache.hashbase = (struct hc_head *)
 211             malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head),
 212                    M_HOSTCACHE, M_WAITOK | M_ZERO);
 213
 214         /*
 215          * Initialize the hash buckets.
 216          */
 217         for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
 218                 TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket);
 219                 V_tcp_hostcache.hashbase[i].hch_length = 0;
 220                 mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
 221                           NULL, MTX_DEF);
 222         }
 223
 224         /*
 225          * Allocate the hostcache entries.
 226          */
 227         V_tcp_hostcache.zone =
 228             uma_zcreate("hostcache", sizeof(struct hc_metrics),
 229             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 230         uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit);
 231
 232         /*
 233          * Set up periodic cache cleanup.
 234          */
 235         callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE);
 236         callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
 237             tcp_hc_purge, curvnet);
 238 }
 239
 240 #ifdef VIMAGE
 241 void
 242 tcp_hc_destroy(void)
 243 {
 244         int i;
 245
 246         callout_drain(&V_tcp_hc_callout);
 247
 248         /* Purge all hc entries. */
 249         tcp_hc_purge_internal(1);
 250
 251         /* Free the uma zone and the allocated hash table. */
 252         uma_zdestroy(V_tcp_hostcache.zone);
 253
 254         for (i = 0; i < V_tcp_hostcache.hashsize; i++)
 255                 mtx_destroy(&V_tcp_hostcache.hashbase[i].hch_mtx);
 256         free(V_tcp_hostcache.hashbase, M_HOSTCACHE);
 257 }
 258 #endif
 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_flags & INC_ISIPV6)
 279                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
 280         else
 281                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
 282
 283         hc_head = &V_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_flags & 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_flags & INC_ISIPV6)
 334                 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
 335         else
 336                 hash = HOSTCACHE_HASH(&inc->inc_faddr);
 337
 338         hc_head = &V_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 >= V_tcp_hostcache.bucket_limit ||
 351             V_tcp_hostcache.cache_count >= V_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                 V_tcp_hostcache.hashbase[hash].hch_length--;
 368                 V_tcp_hostcache.cache_count--;
 369                 TCPSTAT_INC(tcps_hc_bucketoverflow);
 370 #if 0
 371                 uma_zfree(V_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(V_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_flags & 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 = V_tcp_hostcache.expire;
 394
 395         /*
 396          * Put it upfront.
 397          */
 398         TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
 399         V_tcp_hostcache.hashbase[hash].hch_length++;
 400         V_tcp_hostcache.cache_count++;
 401         TCPSTAT_INC(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 = V_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 = V_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 = V_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 = V_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_INC(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_INC(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_INC(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_INC(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_INC(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_INC(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_INC(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 * (V_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 < V_tcp_hostcache.hashsize; i++) {
 615                 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
 616                 TAILQ_FOREACH(hc_entry, &V_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_RTTVAR_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(&V_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  * Caller has to make sure the curvnet is set properly.
 652  */
 653 static void
 654 tcp_hc_purge_internal(int all)
 655 {
 656         struct hc_metrics *hc_entry, *hc_next;
 657         int i;
 658
 659         for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
 660                 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
 661                 TAILQ_FOREACH_SAFE(hc_entry,
 662                     &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) {
 663                         if (all || hc_entry->rmx_expire <= 0) {
 664                                 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket,
 665                                               hc_entry, rmx_q);
 666                                 uma_zfree(V_tcp_hostcache.zone, hc_entry);
 667                                 V_tcp_hostcache.hashbase[i].hch_length--;
 668                                 V_tcp_hostcache.cache_count--;
 669                         } else
 670                                 hc_entry->rmx_expire -= V_tcp_hostcache.prune;
 671                 }
 672                 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
 673         }
 674 }
 675
 676 /*
 677  * Expire and purge (old|all) entries in the tcp_hostcache.  Runs
 678  * periodically from the callout.
 679  */
 680 static void
 681 tcp_hc_purge(void *arg)
 682 {
 683         CURVNET_SET((struct vnet *) arg);
 684         int all = 0;
 685
 686         if (V_tcp_hostcache.purgeall) {
 687                 all = 1;
 688                 V_tcp_hostcache.purgeall = 0;
 689         }
 690
 691         tcp_hc_purge_internal(all);
 692
 693         callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
 694             tcp_hc_purge, arg);
 695         CURVNET_RESTORE();
 696 }