tcp_hostcache.c revision 185088
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 185088 2008-11-19 09:39:34Z zec $"); 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 83#include <netinet/in.h> 84#include <netinet/in_systm.h> 85#include <netinet/ip.h> 86#include <netinet/in_var.h> 87#include <netinet/in_pcb.h> 88#include <netinet/ip_var.h> 89#ifdef INET6 90#include <netinet/ip6.h> 91#include <netinet6/ip6_var.h> 92#endif 93#include <netinet/tcp.h> 94#include <netinet/tcp_var.h> 95#ifdef INET6 96#include <netinet6/tcp6_var.h> 97#endif 98 99#include <vm/uma.h> 100 101 102TAILQ_HEAD(hc_qhead, hc_metrics); 103 104struct hc_head { 105 struct hc_qhead hch_bucket; 106 u_int hch_length; 107 struct mtx hch_mtx; 108}; 109 110struct hc_metrics { 111 /* housekeeping */ 112 TAILQ_ENTRY(hc_metrics) rmx_q; 113 struct hc_head *rmx_head; /* head of bucket tail queue */ 114 struct in_addr ip4; /* IP address */ 115 struct in6_addr ip6; /* IP6 address */ 116 /* endpoint specific values for TCP */ 117 u_long rmx_mtu; /* MTU for this path */ 118 u_long rmx_ssthresh; /* outbound gateway buffer limit */ 119 u_long rmx_rtt; /* estimated round trip time */ 120 u_long rmx_rttvar; /* estimated rtt variance */ 121 u_long rmx_bandwidth; /* estimated bandwidth */ 122 u_long rmx_cwnd; /* congestion window */ 123 u_long rmx_sendpipe; /* outbound delay-bandwidth product */ 124 u_long rmx_recvpipe; /* inbound delay-bandwidth product */ 125 /* TCP hostcache internal data */ 126 int rmx_expire; /* lifetime for object */ 127 u_long rmx_hits; /* number of hits */ 128 u_long rmx_updates; /* number of updates */ 129}; 130 131/* Arbitrary values */ 132#define TCP_HOSTCACHE_HASHSIZE 512 133#define TCP_HOSTCACHE_BUCKETLIMIT 30 134#define TCP_HOSTCACHE_EXPIRE 60*60 /* one hour */ 135#define TCP_HOSTCACHE_PRUNE 5*60 /* every 5 minutes */ 136 137struct tcp_hostcache { 138 struct hc_head *hashbase; 139 uma_zone_t zone; 140 u_int hashsize; 141 u_int hashmask; 142 u_int bucket_limit; 143 u_int cache_count; 144 u_int cache_limit; 145 int expire; 146 int prune; 147 int purgeall; 148}; 149 150#ifdef VIMAGE_GLOBALS 151static struct tcp_hostcache tcp_hostcache; 152static struct callout tcp_hc_callout; 153#endif 154 155static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *); 156static struct hc_metrics *tcp_hc_insert(struct in_conninfo *); 157static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS); 158static void tcp_hc_purge(void *); 159 160SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, 161 "TCP Host cache"); 162 163SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, cachelimit, 164 CTLFLAG_RDTUN, tcp_hostcache.cache_limit, 0, 165 "Overall entry limit for hostcache"); 166 167SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, hashsize, 168 CTLFLAG_RDTUN, tcp_hostcache.hashsize, 0, 169 "Size of TCP hostcache hashtable"); 170 171SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, bucketlimit, 172 CTLFLAG_RDTUN, tcp_hostcache.bucket_limit, 0, 173 "Per-bucket hash limit for hostcache"); 174 175SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, count, 176 CTLFLAG_RD, tcp_hostcache.cache_count, 0, 177 "Current number of entries in hostcache"); 178 179SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, expire, 180 CTLFLAG_RW, tcp_hostcache.expire, 0, 181 "Expire time of TCP hostcache entries"); 182 183SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, prune, 184 CTLFLAG_RW, tcp_hostcache.prune, 0, "Time between purge runs"); 185 186SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_hostcache, OID_AUTO, purge, 187 CTLFLAG_RW, tcp_hostcache.purgeall, 0, 188 "Expire all entires on next purge run"); 189 190SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list, 191 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0, 192 sysctl_tcp_hc_list, "A", "List of all hostcache entries"); 193 194 195static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache"); 196 197#define HOSTCACHE_HASH(ip) \ 198 (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \ 199 V_tcp_hostcache.hashmask) 200 201/* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */ 202#define HOSTCACHE_HASH6(ip6) \ 203 (((ip6)->s6_addr32[0] ^ \ 204 (ip6)->s6_addr32[1] ^ \ 205 (ip6)->s6_addr32[2] ^ \ 206 (ip6)->s6_addr32[3]) & \ 207 V_tcp_hostcache.hashmask) 208 209#define THC_LOCK(lp) mtx_lock(lp) 210#define THC_UNLOCK(lp) mtx_unlock(lp) 211 212void 213tcp_hc_init(void) 214{ 215 INIT_VNET_INET(curvnet); 216 int i; 217 218 /* 219 * Initialize hostcache structures. 220 */ 221 V_tcp_hostcache.cache_count = 0; 222 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; 223 V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT; 224 V_tcp_hostcache.cache_limit = 225 V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit; 226 V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE; 227 V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE; 228 229 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize", 230 &V_tcp_hostcache.hashsize); 231 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit", 232 &V_tcp_hostcache.cache_limit); 233 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit", 234 &V_tcp_hostcache.bucket_limit); 235 if (!powerof2(V_tcp_hostcache.hashsize)) { 236 printf("WARNING: hostcache hash size is not a power of 2.\n"); 237 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */ 238 } 239 V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1; 240 241 /* 242 * Allocate the hash table. 243 */ 244 V_tcp_hostcache.hashbase = (struct hc_head *) 245 malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head), 246 M_HOSTCACHE, M_WAITOK | M_ZERO); 247 248 /* 249 * Initialize the hash buckets. 250 */ 251 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 252 TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket); 253 V_tcp_hostcache.hashbase[i].hch_length = 0; 254 mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry", 255 NULL, MTX_DEF); 256 } 257 258 /* 259 * Allocate the hostcache entries. 260 */ 261 V_tcp_hostcache.zone = 262 uma_zcreate("hostcache", sizeof(struct hc_metrics), 263 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 264 uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit); 265 266 /* 267 * Set up periodic cache cleanup. 268 */ 269 callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE); 270 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 271 tcp_hc_purge, 0); 272} 273 274/* 275 * Internal function: look up an entry in the hostcache or return NULL. 276 * 277 * If an entry has been returned, the caller becomes responsible for 278 * unlocking the bucket row after he is done reading/modifying the entry. 279 */ 280static struct hc_metrics * 281tcp_hc_lookup(struct in_conninfo *inc) 282{ 283 INIT_VNET_INET(curvnet); 284 int hash; 285 struct hc_head *hc_head; 286 struct hc_metrics *hc_entry; 287 288 KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer")); 289 290 /* 291 * Hash the foreign ip address. 292 */ 293 if (inc->inc_isipv6) 294 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 295 else 296 hash = HOSTCACHE_HASH(&inc->inc_faddr); 297 298 hc_head = &V_tcp_hostcache.hashbase[hash]; 299 300 /* 301 * Acquire lock for this bucket row; we release the lock if we don't 302 * find an entry, otherwise the caller has to unlock after he is 303 * done. 304 */ 305 THC_LOCK(&hc_head->hch_mtx); 306 307 /* 308 * Iterate through entries in bucket row looking for a match. 309 */ 310 TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) { 311 if (inc->inc_isipv6) { 312 if (memcmp(&inc->inc6_faddr, &hc_entry->ip6, 313 sizeof(inc->inc6_faddr)) == 0) 314 return hc_entry; 315 } else { 316 if (memcmp(&inc->inc_faddr, &hc_entry->ip4, 317 sizeof(inc->inc_faddr)) == 0) 318 return hc_entry; 319 } 320 } 321 322 /* 323 * We were unsuccessful and didn't find anything. 324 */ 325 THC_UNLOCK(&hc_head->hch_mtx); 326 return NULL; 327} 328 329/* 330 * Internal function: insert an entry into the hostcache or return NULL if 331 * unable to allocate a new one. 332 * 333 * If an entry has been returned, the caller becomes responsible for 334 * unlocking the bucket row after he is done reading/modifying the entry. 335 */ 336static struct hc_metrics * 337tcp_hc_insert(struct in_conninfo *inc) 338{ 339 INIT_VNET_INET(curvnet); 340 int hash; 341 struct hc_head *hc_head; 342 struct hc_metrics *hc_entry; 343 344 KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer")); 345 346 /* 347 * Hash the foreign ip address. 348 */ 349 if (inc->inc_isipv6) 350 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 351 else 352 hash = HOSTCACHE_HASH(&inc->inc_faddr); 353 354 hc_head = &V_tcp_hostcache.hashbase[hash]; 355 356 /* 357 * Acquire lock for this bucket row; we release the lock if we don't 358 * find an entry, otherwise the caller has to unlock after he is 359 * done. 360 */ 361 THC_LOCK(&hc_head->hch_mtx); 362 363 /* 364 * If the bucket limit is reached, reuse the least-used element. 365 */ 366 if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit || 367 V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) { 368 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead); 369 /* 370 * At first we were dropping the last element, just to 371 * reacquire it in the next two lines again, which isn't very 372 * efficient. Instead just reuse the least used element. 373 * We may drop something that is still "in-use" but we can be 374 * "lossy". 375 * Just give up if this bucket row is empty and we don't have 376 * anything to replace. 377 */ 378 if (hc_entry == NULL) { 379 THC_UNLOCK(&hc_head->hch_mtx); 380 return NULL; 381 } 382 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q); 383 V_tcp_hostcache.hashbase[hash].hch_length--; 384 V_tcp_hostcache.cache_count--; 385 V_tcpstat.tcps_hc_bucketoverflow++; 386#if 0 387 uma_zfree(V_tcp_hostcache.zone, hc_entry); 388#endif 389 } else { 390 /* 391 * Allocate a new entry, or balk if not possible. 392 */ 393 hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT); 394 if (hc_entry == NULL) { 395 THC_UNLOCK(&hc_head->hch_mtx); 396 return NULL; 397 } 398 } 399 400 /* 401 * Initialize basic information of hostcache entry. 402 */ 403 bzero(hc_entry, sizeof(*hc_entry)); 404 if (inc->inc_isipv6) 405 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6)); 406 else 407 hc_entry->ip4 = inc->inc_faddr; 408 hc_entry->rmx_head = hc_head; 409 hc_entry->rmx_expire = V_tcp_hostcache.expire; 410 411 /* 412 * Put it upfront. 413 */ 414 TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q); 415 V_tcp_hostcache.hashbase[hash].hch_length++; 416 V_tcp_hostcache.cache_count++; 417 V_tcpstat.tcps_hc_added++; 418 419 return hc_entry; 420} 421 422/* 423 * External function: look up an entry in the hostcache and fill out the 424 * supplied TCP metrics structure. Fills in NULL when no entry was found or 425 * a value is not set. 426 */ 427void 428tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite) 429{ 430 INIT_VNET_INET(curvnet); 431 struct hc_metrics *hc_entry; 432 433 /* 434 * Find the right bucket. 435 */ 436 hc_entry = tcp_hc_lookup(inc); 437 438 /* 439 * If we don't have an existing object. 440 */ 441 if (hc_entry == NULL) { 442 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite)); 443 return; 444 } 445 hc_entry->rmx_hits++; 446 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 447 448 hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu; 449 hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh; 450 hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt; 451 hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar; 452 hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth; 453 hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd; 454 hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe; 455 hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe; 456 457 /* 458 * Unlock bucket row. 459 */ 460 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 461} 462 463/* 464 * External function: look up an entry in the hostcache and return the 465 * discovered path MTU. Returns NULL if no entry is found or value is not 466 * set. 467 */ 468u_long 469tcp_hc_getmtu(struct in_conninfo *inc) 470{ 471 INIT_VNET_INET(curvnet); 472 struct hc_metrics *hc_entry; 473 u_long mtu; 474 475 hc_entry = tcp_hc_lookup(inc); 476 if (hc_entry == NULL) { 477 return 0; 478 } 479 hc_entry->rmx_hits++; 480 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 481 482 mtu = hc_entry->rmx_mtu; 483 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 484 return mtu; 485} 486 487/* 488 * External function: update the MTU value of an entry in the hostcache. 489 * Creates a new entry if none was found. 490 */ 491void 492tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu) 493{ 494 INIT_VNET_INET(curvnet); 495 struct hc_metrics *hc_entry; 496 497 /* 498 * Find the right bucket. 499 */ 500 hc_entry = tcp_hc_lookup(inc); 501 502 /* 503 * If we don't have an existing object, try to insert a new one. 504 */ 505 if (hc_entry == NULL) { 506 hc_entry = tcp_hc_insert(inc); 507 if (hc_entry == NULL) 508 return; 509 } 510 hc_entry->rmx_updates++; 511 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 512 513 hc_entry->rmx_mtu = mtu; 514 515 /* 516 * Put it upfront so we find it faster next time. 517 */ 518 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 519 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 520 521 /* 522 * Unlock bucket row. 523 */ 524 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 525} 526 527/* 528 * External function: update the TCP metrics of an entry in the hostcache. 529 * Creates a new entry if none was found. 530 */ 531void 532tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml) 533{ 534 INIT_VNET_INET(curvnet); 535 struct hc_metrics *hc_entry; 536 537 hc_entry = tcp_hc_lookup(inc); 538 if (hc_entry == NULL) { 539 hc_entry = tcp_hc_insert(inc); 540 if (hc_entry == NULL) 541 return; 542 } 543 hc_entry->rmx_updates++; 544 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 545 546 if (hcml->rmx_rtt != 0) { 547 if (hc_entry->rmx_rtt == 0) 548 hc_entry->rmx_rtt = hcml->rmx_rtt; 549 else 550 hc_entry->rmx_rtt = 551 (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2; 552 V_tcpstat.tcps_cachedrtt++; 553 } 554 if (hcml->rmx_rttvar != 0) { 555 if (hc_entry->rmx_rttvar == 0) 556 hc_entry->rmx_rttvar = hcml->rmx_rttvar; 557 else 558 hc_entry->rmx_rttvar = 559 (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2; 560 V_tcpstat.tcps_cachedrttvar++; 561 } 562 if (hcml->rmx_ssthresh != 0) { 563 if (hc_entry->rmx_ssthresh == 0) 564 hc_entry->rmx_ssthresh = hcml->rmx_ssthresh; 565 else 566 hc_entry->rmx_ssthresh = 567 (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2; 568 V_tcpstat.tcps_cachedssthresh++; 569 } 570 if (hcml->rmx_bandwidth != 0) { 571 if (hc_entry->rmx_bandwidth == 0) 572 hc_entry->rmx_bandwidth = hcml->rmx_bandwidth; 573 else 574 hc_entry->rmx_bandwidth = 575 (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2; 576 /* V_tcpstat.tcps_cachedbandwidth++; */ 577 } 578 if (hcml->rmx_cwnd != 0) { 579 if (hc_entry->rmx_cwnd == 0) 580 hc_entry->rmx_cwnd = hcml->rmx_cwnd; 581 else 582 hc_entry->rmx_cwnd = 583 (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2; 584 /* V_tcpstat.tcps_cachedcwnd++; */ 585 } 586 if (hcml->rmx_sendpipe != 0) { 587 if (hc_entry->rmx_sendpipe == 0) 588 hc_entry->rmx_sendpipe = hcml->rmx_sendpipe; 589 else 590 hc_entry->rmx_sendpipe = 591 (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2; 592 /* V_tcpstat.tcps_cachedsendpipe++; */ 593 } 594 if (hcml->rmx_recvpipe != 0) { 595 if (hc_entry->rmx_recvpipe == 0) 596 hc_entry->rmx_recvpipe = hcml->rmx_recvpipe; 597 else 598 hc_entry->rmx_recvpipe = 599 (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2; 600 /* V_tcpstat.tcps_cachedrecvpipe++; */ 601 } 602 603 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 604 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 605 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 606} 607 608/* 609 * Sysctl function: prints the list and values of all hostcache entries in 610 * unsorted order. 611 */ 612static int 613sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS) 614{ 615 INIT_VNET_INET(curvnet); 616 int bufsize; 617 int linesize = 128; 618 char *p, *buf; 619 int len, i, error; 620 struct hc_metrics *hc_entry; 621#ifdef INET6 622 char ip6buf[INET6_ADDRSTRLEN]; 623#endif 624 625 bufsize = linesize * (V_tcp_hostcache.cache_count + 1); 626 627 p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO); 628 629 len = snprintf(p, linesize, 630 "\nIP address MTU SSTRESH RTT RTTVAR BANDWIDTH " 631 " CWND SENDPIPE RECVPIPE HITS UPD EXP\n"); 632 p += len; 633 634#define msec(u) (((u) + 500) / 1000) 635 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 636 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 637 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, 638 rmx_q) { 639 len = snprintf(p, linesize, 640 "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu " 641 "%4lu %4lu %4i\n", 642 hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) : 643#ifdef INET6 644 ip6_sprintf(ip6buf, &hc_entry->ip6), 645#else 646 "IPv6?", 647#endif 648 hc_entry->rmx_mtu, 649 hc_entry->rmx_ssthresh, 650 msec(hc_entry->rmx_rtt * 651 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 652 msec(hc_entry->rmx_rttvar * 653 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 654 hc_entry->rmx_bandwidth * 8, 655 hc_entry->rmx_cwnd, 656 hc_entry->rmx_sendpipe, 657 hc_entry->rmx_recvpipe, 658 hc_entry->rmx_hits, 659 hc_entry->rmx_updates, 660 hc_entry->rmx_expire); 661 p += len; 662 } 663 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 664 } 665#undef msec 666 error = SYSCTL_OUT(req, buf, p - buf); 667 free(buf, M_TEMP); 668 return(error); 669} 670 671/* 672 * Expire and purge (old|all) entries in the tcp_hostcache. Runs 673 * periodically from the callout. 674 */ 675static void 676tcp_hc_purge(void *arg) 677{ 678 INIT_VNET_INET(curvnet); 679 struct hc_metrics *hc_entry, *hc_next; 680 int all = (intptr_t)arg; 681 int i; 682 683 if (V_tcp_hostcache.purgeall) { 684 all = 1; 685 V_tcp_hostcache.purgeall = 0; 686 } 687 688 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 689 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 690 TAILQ_FOREACH_SAFE(hc_entry, 691 &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) { 692 if (all || hc_entry->rmx_expire <= 0) { 693 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket, 694 hc_entry, rmx_q); 695 uma_zfree(V_tcp_hostcache.zone, hc_entry); 696 V_tcp_hostcache.hashbase[i].hch_length--; 697 V_tcp_hostcache.cache_count--; 698 } else 699 hc_entry->rmx_expire -= V_tcp_hostcache.prune; 700 } 701 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 702 } 703 704 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 705 tcp_hc_purge, arg); 706} 707