tcp_hostcache.c revision 194739
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: head/sys/netinet/tcp_hostcache.c 194739 2009-06-23 17:03:45Z bz $"); 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#include <sys/vimage.h> 80 81#include <net/if.h> 82#include <net/route.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#include <netinet/vinet.h> 98#ifdef INET6 99#include <netinet6/tcp6_var.h> 100#endif 101 102#include <vm/uma.h> 103 104/* Arbitrary values */ 105#define TCP_HOSTCACHE_HASHSIZE 512 106#define TCP_HOSTCACHE_BUCKETLIMIT 30 107#define TCP_HOSTCACHE_EXPIRE 60*60 /* one hour */ 108#define TCP_HOSTCACHE_PRUNE 5*60 /* every 5 minutes */ 109 110#ifdef VIMAGE_GLOBALS 111static struct tcp_hostcache tcp_hostcache; 112static struct callout tcp_hc_callout; 113#endif 114 115static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *); 116static struct hc_metrics *tcp_hc_insert(struct in_conninfo *); 117static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS); 118static void tcp_hc_purge(void *); 119 120SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, 121 "TCP Host cache"); 122 123SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, cachelimit, 124 CTLFLAG_RDTUN, tcp_hostcache.cache_limit, 0, 125 "Overall entry limit for hostcache"); 126 127SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, hashsize, 128 CTLFLAG_RDTUN, tcp_hostcache.hashsize, 0, 129 "Size of TCP hostcache hashtable"); 130 131SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, bucketlimit, 132 CTLFLAG_RDTUN, tcp_hostcache.bucket_limit, 0, 133 "Per-bucket hash limit for hostcache"); 134 135SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, count, 136 CTLFLAG_RD, tcp_hostcache.cache_count, 0, 137 "Current number of entries in hostcache"); 138 139SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, expire, 140 CTLFLAG_RW, tcp_hostcache.expire, 0, 141 "Expire time of TCP hostcache entries"); 142 143SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, prune, 144 CTLFLAG_RW, tcp_hostcache.prune, 0, "Time between purge runs"); 145 146SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, purge, 147 CTLFLAG_RW, tcp_hostcache.purgeall, 0, 148 "Expire all entires on next purge run"); 149 150SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list, 151 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0, 152 sysctl_tcp_hc_list, "A", "List of all hostcache entries"); 153 154 155static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache"); 156 157#define HOSTCACHE_HASH(ip) \ 158 (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \ 159 V_tcp_hostcache.hashmask) 160 161/* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */ 162#define HOSTCACHE_HASH6(ip6) \ 163 (((ip6)->s6_addr32[0] ^ \ 164 (ip6)->s6_addr32[1] ^ \ 165 (ip6)->s6_addr32[2] ^ \ 166 (ip6)->s6_addr32[3]) & \ 167 V_tcp_hostcache.hashmask) 168 169#define THC_LOCK(lp) mtx_lock(lp) 170#define THC_UNLOCK(lp) mtx_unlock(lp) 171 172void 173tcp_hc_init(void) 174{ 175 INIT_VNET_INET(curvnet); 176 int i; 177 178 /* 179 * Initialize hostcache structures. 180 */ 181 V_tcp_hostcache.cache_count = 0; 182 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; 183 V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT; 184 V_tcp_hostcache.cache_limit = 185 V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit; 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 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit", 192 &V_tcp_hostcache.cache_limit); 193 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit", 194 &V_tcp_hostcache.bucket_limit); 195 if (!powerof2(V_tcp_hostcache.hashsize)) { 196 printf("WARNING: hostcache hash size is not a power of 2.\n"); 197 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */ 198 } 199 V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1; 200 201 /* 202 * Allocate the hash table. 203 */ 204 V_tcp_hostcache.hashbase = (struct hc_head *) 205 malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head), 206 M_HOSTCACHE, M_WAITOK | M_ZERO); 207 208 /* 209 * Initialize the hash buckets. 210 */ 211 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 212 TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket); 213 V_tcp_hostcache.hashbase[i].hch_length = 0; 214 mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry", 215 NULL, MTX_DEF); 216 } 217 218 /* 219 * Allocate the hostcache entries. 220 */ 221 V_tcp_hostcache.zone = 222 uma_zcreate("hostcache", sizeof(struct hc_metrics), 223 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 224 uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit); 225 226 /* 227 * Set up periodic cache cleanup. 228 */ 229 callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE); 230 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 231 tcp_hc_purge, curvnet); 232} 233 234#ifdef VIMAGE 235void 236tcp_hc_destroy(void) 237{ 238 INIT_VNET_INET(curvnet); 239 240 /* XXX TODO walk the hashtable and free all entries */ 241 242 callout_drain(&V_tcp_hc_callout); 243} 244#endif 245 246/* 247 * Internal function: look up an entry in the hostcache or return NULL. 248 * 249 * If an entry has been returned, the caller becomes responsible for 250 * unlocking the bucket row after he is done reading/modifying the entry. 251 */ 252static struct hc_metrics * 253tcp_hc_lookup(struct in_conninfo *inc) 254{ 255 INIT_VNET_INET(curvnet); 256 int hash; 257 struct hc_head *hc_head; 258 struct hc_metrics *hc_entry; 259 260 KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer")); 261 262 /* 263 * Hash the foreign ip address. 264 */ 265 if (inc->inc_flags & INC_ISIPV6) 266 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 267 else 268 hash = HOSTCACHE_HASH(&inc->inc_faddr); 269 270 hc_head = &V_tcp_hostcache.hashbase[hash]; 271 272 /* 273 * Acquire lock for this bucket row; we release the lock if we don't 274 * find an entry, otherwise the caller has to unlock after he is 275 * done. 276 */ 277 THC_LOCK(&hc_head->hch_mtx); 278 279 /* 280 * Iterate through entries in bucket row looking for a match. 281 */ 282 TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) { 283 if (inc->inc_flags & INC_ISIPV6) { 284 if (memcmp(&inc->inc6_faddr, &hc_entry->ip6, 285 sizeof(inc->inc6_faddr)) == 0) 286 return hc_entry; 287 } else { 288 if (memcmp(&inc->inc_faddr, &hc_entry->ip4, 289 sizeof(inc->inc_faddr)) == 0) 290 return hc_entry; 291 } 292 } 293 294 /* 295 * We were unsuccessful and didn't find anything. 296 */ 297 THC_UNLOCK(&hc_head->hch_mtx); 298 return NULL; 299} 300 301/* 302 * Internal function: insert an entry into the hostcache or return NULL if 303 * unable to allocate a new one. 304 * 305 * If an entry has been returned, the caller becomes responsible for 306 * unlocking the bucket row after he is done reading/modifying the entry. 307 */ 308static struct hc_metrics * 309tcp_hc_insert(struct in_conninfo *inc) 310{ 311 INIT_VNET_INET(curvnet); 312 int hash; 313 struct hc_head *hc_head; 314 struct hc_metrics *hc_entry; 315 316 KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer")); 317 318 /* 319 * Hash the foreign ip address. 320 */ 321 if (inc->inc_flags & INC_ISIPV6) 322 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 323 else 324 hash = HOSTCACHE_HASH(&inc->inc_faddr); 325 326 hc_head = &V_tcp_hostcache.hashbase[hash]; 327 328 /* 329 * Acquire lock for this bucket row; we release the lock if we don't 330 * find an entry, otherwise the caller has to unlock after he is 331 * done. 332 */ 333 THC_LOCK(&hc_head->hch_mtx); 334 335 /* 336 * If the bucket limit is reached, reuse the least-used element. 337 */ 338 if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit || 339 V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) { 340 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead); 341 /* 342 * At first we were dropping the last element, just to 343 * reacquire it in the next two lines again, which isn't very 344 * efficient. Instead just reuse the least used element. 345 * We may drop something that is still "in-use" but we can be 346 * "lossy". 347 * Just give up if this bucket row is empty and we don't have 348 * anything to replace. 349 */ 350 if (hc_entry == NULL) { 351 THC_UNLOCK(&hc_head->hch_mtx); 352 return NULL; 353 } 354 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q); 355 V_tcp_hostcache.hashbase[hash].hch_length--; 356 V_tcp_hostcache.cache_count--; 357 TCPSTAT_INC(tcps_hc_bucketoverflow); 358#if 0 359 uma_zfree(V_tcp_hostcache.zone, hc_entry); 360#endif 361 } else { 362 /* 363 * Allocate a new entry, or balk if not possible. 364 */ 365 hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT); 366 if (hc_entry == NULL) { 367 THC_UNLOCK(&hc_head->hch_mtx); 368 return NULL; 369 } 370 } 371 372 /* 373 * Initialize basic information of hostcache entry. 374 */ 375 bzero(hc_entry, sizeof(*hc_entry)); 376 if (inc->inc_flags & INC_ISIPV6) 377 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6)); 378 else 379 hc_entry->ip4 = inc->inc_faddr; 380 hc_entry->rmx_head = hc_head; 381 hc_entry->rmx_expire = V_tcp_hostcache.expire; 382 383 /* 384 * Put it upfront. 385 */ 386 TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q); 387 V_tcp_hostcache.hashbase[hash].hch_length++; 388 V_tcp_hostcache.cache_count++; 389 TCPSTAT_INC(tcps_hc_added); 390 391 return hc_entry; 392} 393 394/* 395 * External function: look up an entry in the hostcache and fill out the 396 * supplied TCP metrics structure. Fills in NULL when no entry was found or 397 * a value is not set. 398 */ 399void 400tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite) 401{ 402 INIT_VNET_INET(curvnet); 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 = V_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 */ 440u_long 441tcp_hc_getmtu(struct in_conninfo *inc) 442{ 443 INIT_VNET_INET(curvnet); 444 struct hc_metrics *hc_entry; 445 u_long mtu; 446 447 hc_entry = tcp_hc_lookup(inc); 448 if (hc_entry == NULL) { 449 return 0; 450 } 451 hc_entry->rmx_hits++; 452 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 453 454 mtu = hc_entry->rmx_mtu; 455 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 456 return mtu; 457} 458 459/* 460 * External function: update the MTU value of an entry in the hostcache. 461 * Creates a new entry if none was found. 462 */ 463void 464tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu) 465{ 466 INIT_VNET_INET(curvnet); 467 struct hc_metrics *hc_entry; 468 469 /* 470 * Find the right bucket. 471 */ 472 hc_entry = tcp_hc_lookup(inc); 473 474 /* 475 * If we don't have an existing object, try to insert a new one. 476 */ 477 if (hc_entry == NULL) { 478 hc_entry = tcp_hc_insert(inc); 479 if (hc_entry == NULL) 480 return; 481 } 482 hc_entry->rmx_updates++; 483 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 484 485 hc_entry->rmx_mtu = mtu; 486 487 /* 488 * Put it upfront so we find it faster next time. 489 */ 490 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 491 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 492 493 /* 494 * Unlock bucket row. 495 */ 496 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 497} 498 499/* 500 * External function: update the TCP metrics of an entry in the hostcache. 501 * Creates a new entry if none was found. 502 */ 503void 504tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml) 505{ 506 INIT_VNET_INET(curvnet); 507 struct hc_metrics *hc_entry; 508 509 hc_entry = tcp_hc_lookup(inc); 510 if (hc_entry == NULL) { 511 hc_entry = tcp_hc_insert(inc); 512 if (hc_entry == NULL) 513 return; 514 } 515 hc_entry->rmx_updates++; 516 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 517 518 if (hcml->rmx_rtt != 0) { 519 if (hc_entry->rmx_rtt == 0) 520 hc_entry->rmx_rtt = hcml->rmx_rtt; 521 else 522 hc_entry->rmx_rtt = 523 (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2; 524 TCPSTAT_INC(tcps_cachedrtt); 525 } 526 if (hcml->rmx_rttvar != 0) { 527 if (hc_entry->rmx_rttvar == 0) 528 hc_entry->rmx_rttvar = hcml->rmx_rttvar; 529 else 530 hc_entry->rmx_rttvar = 531 (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2; 532 TCPSTAT_INC(tcps_cachedrttvar); 533 } 534 if (hcml->rmx_ssthresh != 0) { 535 if (hc_entry->rmx_ssthresh == 0) 536 hc_entry->rmx_ssthresh = hcml->rmx_ssthresh; 537 else 538 hc_entry->rmx_ssthresh = 539 (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2; 540 TCPSTAT_INC(tcps_cachedssthresh); 541 } 542 if (hcml->rmx_bandwidth != 0) { 543 if (hc_entry->rmx_bandwidth == 0) 544 hc_entry->rmx_bandwidth = hcml->rmx_bandwidth; 545 else 546 hc_entry->rmx_bandwidth = 547 (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2; 548 /* TCPSTAT_INC(tcps_cachedbandwidth); */ 549 } 550 if (hcml->rmx_cwnd != 0) { 551 if (hc_entry->rmx_cwnd == 0) 552 hc_entry->rmx_cwnd = hcml->rmx_cwnd; 553 else 554 hc_entry->rmx_cwnd = 555 (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2; 556 /* TCPSTAT_INC(tcps_cachedcwnd); */ 557 } 558 if (hcml->rmx_sendpipe != 0) { 559 if (hc_entry->rmx_sendpipe == 0) 560 hc_entry->rmx_sendpipe = hcml->rmx_sendpipe; 561 else 562 hc_entry->rmx_sendpipe = 563 (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2; 564 /* TCPSTAT_INC(tcps_cachedsendpipe); */ 565 } 566 if (hcml->rmx_recvpipe != 0) { 567 if (hc_entry->rmx_recvpipe == 0) 568 hc_entry->rmx_recvpipe = hcml->rmx_recvpipe; 569 else 570 hc_entry->rmx_recvpipe = 571 (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2; 572 /* TCPSTAT_INC(tcps_cachedrecvpipe); */ 573 } 574 575 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 576 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 577 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 578} 579 580/* 581 * Sysctl function: prints the list and values of all hostcache entries in 582 * unsorted order. 583 */ 584static int 585sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS) 586{ 587 INIT_VNET_INET(curvnet); 588 int bufsize; 589 int linesize = 128; 590 char *p, *buf; 591 int len, i, error; 592 struct hc_metrics *hc_entry; 593#ifdef INET6 594 char ip6buf[INET6_ADDRSTRLEN]; 595#endif 596 597 bufsize = linesize * (V_tcp_hostcache.cache_count + 1); 598 599 p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO); 600 601 len = snprintf(p, linesize, 602 "\nIP address MTU SSTRESH RTT RTTVAR BANDWIDTH " 603 " CWND SENDPIPE RECVPIPE HITS UPD EXP\n"); 604 p += len; 605 606#define msec(u) (((u) + 500) / 1000) 607 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 608 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 609 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, 610 rmx_q) { 611 len = snprintf(p, linesize, 612 "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu " 613 "%4lu %4lu %4i\n", 614 hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) : 615#ifdef INET6 616 ip6_sprintf(ip6buf, &hc_entry->ip6), 617#else 618 "IPv6?", 619#endif 620 hc_entry->rmx_mtu, 621 hc_entry->rmx_ssthresh, 622 msec(hc_entry->rmx_rtt * 623 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 624 msec(hc_entry->rmx_rttvar * 625 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 626 hc_entry->rmx_bandwidth * 8, 627 hc_entry->rmx_cwnd, 628 hc_entry->rmx_sendpipe, 629 hc_entry->rmx_recvpipe, 630 hc_entry->rmx_hits, 631 hc_entry->rmx_updates, 632 hc_entry->rmx_expire); 633 p += len; 634 } 635 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 636 } 637#undef msec 638 error = SYSCTL_OUT(req, buf, p - buf); 639 free(buf, M_TEMP); 640 return(error); 641} 642 643/* 644 * Expire and purge (old|all) entries in the tcp_hostcache. Runs 645 * periodically from the callout. 646 */ 647static void 648tcp_hc_purge(void *arg) 649{ 650 CURVNET_SET((struct vnet *) arg); 651 INIT_VNET_INET(curvnet); 652 struct hc_metrics *hc_entry, *hc_next; 653 int all = 0; 654 int i; 655 656 if (V_tcp_hostcache.purgeall) { 657 all = 1; 658 V_tcp_hostcache.purgeall = 0; 659 } 660 661 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 662 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 663 TAILQ_FOREACH_SAFE(hc_entry, 664 &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) { 665 if (all || hc_entry->rmx_expire <= 0) { 666 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket, 667 hc_entry, rmx_q); 668 uma_zfree(V_tcp_hostcache.zone, hc_entry); 669 V_tcp_hostcache.hashbase[i].hch_length--; 670 V_tcp_hostcache.cache_count--; 671 } else 672 hc_entry->rmx_expire -= V_tcp_hostcache.prune; 673 } 674 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 675 } 676 677 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 678 tcp_hc_purge, arg); 679 CURVNET_RESTORE(); 680} 681