tcp_hostcache.c revision 181803
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 has 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 181803 2008-08-17 23:27:27Z 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 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}; 149static struct tcp_hostcache tcp_hostcache; 150 151static struct callout tcp_hc_callout; 152 153static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *); 154static struct hc_metrics *tcp_hc_insert(struct in_conninfo *); 155static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS); 156static void tcp_hc_purge(void *); 157 158SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, "TCP Host cache"); 159 160SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 161 &tcp_hostcache.cache_limit, 0, "Overall entry limit for hostcache"); 162 163SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 164 &tcp_hostcache.hashsize, 0, "Size of TCP hostcache hashtable"); 165 166SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 167 &tcp_hostcache.bucket_limit, 0, "Per-bucket hash limit for hostcache"); 168 169SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD, 170 &tcp_hostcache.cache_count, 0, "Current number of entries in hostcache"); 171 172SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW, 173 &tcp_hostcache.expire, 0, "Expire time of TCP hostcache entries"); 174 175SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW, 176 &tcp_hostcache.prune, 0, "Time between purge runs"); 177 178SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW, 179 &tcp_hostcache.purgeall, 0, "Expire all entires on next purge run"); 180 181SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list, 182 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0, 183 sysctl_tcp_hc_list, "A", "List of all hostcache entries"); 184 185 186static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache"); 187 188#define HOSTCACHE_HASH(ip) \ 189 (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \ 190 V_tcp_hostcache.hashmask) 191 192/* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */ 193#define HOSTCACHE_HASH6(ip6) \ 194 (((ip6)->s6_addr32[0] ^ \ 195 (ip6)->s6_addr32[1] ^ \ 196 (ip6)->s6_addr32[2] ^ \ 197 (ip6)->s6_addr32[3]) & \ 198 V_tcp_hostcache.hashmask) 199 200#define THC_LOCK(lp) mtx_lock(lp) 201#define THC_UNLOCK(lp) mtx_unlock(lp) 202 203void 204tcp_hc_init(void) 205{ 206 int i; 207 208 /* 209 * Initialize hostcache structures. 210 */ 211 V_tcp_hostcache.cache_count = 0; 212 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; 213 V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT; 214 V_tcp_hostcache.cache_limit = 215 V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit; 216 V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE; 217 V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE; 218 219 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize", 220 &V_tcp_hostcache.hashsize); 221 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit", 222 &V_tcp_hostcache.cache_limit); 223 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit", 224 &V_tcp_hostcache.bucket_limit); 225 if (!powerof2(V_tcp_hostcache.hashsize)) { 226 printf("WARNING: hostcache hash size is not a power of 2.\n"); 227 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */ 228 } 229 V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1; 230 231 /* 232 * Allocate the hash table. 233 */ 234 V_tcp_hostcache.hashbase = (struct hc_head *) 235 malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head), 236 M_HOSTCACHE, M_WAITOK | M_ZERO); 237 238 /* 239 * Initialize the hash buckets. 240 */ 241 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 242 TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket); 243 V_tcp_hostcache.hashbase[i].hch_length = 0; 244 mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry", 245 NULL, MTX_DEF); 246 } 247 248 /* 249 * Allocate the hostcache entries. 250 */ 251 V_tcp_hostcache.zone = uma_zcreate("hostcache", sizeof(struct hc_metrics), 252 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 253 uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit); 254 255 /* 256 * Set up periodic cache cleanup. 257 */ 258 callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE); 259 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, tcp_hc_purge, 0); 260} 261 262/* 263 * Internal function: look up an entry in the hostcache or return NULL. 264 * 265 * If an entry has been returned, the caller becomes responsible for 266 * unlocking the bucket row after he is done reading/modifying the entry. 267 */ 268static struct hc_metrics * 269tcp_hc_lookup(struct in_conninfo *inc) 270{ 271 int hash; 272 struct hc_head *hc_head; 273 struct hc_metrics *hc_entry; 274 275 KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer")); 276 277 /* 278 * Hash the foreign ip address. 279 */ 280 if (inc->inc_isipv6) 281 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 282 else 283 hash = HOSTCACHE_HASH(&inc->inc_faddr); 284 285 hc_head = &V_tcp_hostcache.hashbase[hash]; 286 287 /* 288 * Acquire lock for this bucket row; we release the lock if we don't 289 * find an entry, otherwise the caller has to unlock after he is 290 * done. 291 */ 292 THC_LOCK(&hc_head->hch_mtx); 293 294 /* 295 * Iterate through entries in bucket row looking for a match. 296 */ 297 TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) { 298 if (inc->inc_isipv6) { 299 if (memcmp(&inc->inc6_faddr, &hc_entry->ip6, 300 sizeof(inc->inc6_faddr)) == 0) 301 return hc_entry; 302 } else { 303 if (memcmp(&inc->inc_faddr, &hc_entry->ip4, 304 sizeof(inc->inc_faddr)) == 0) 305 return hc_entry; 306 } 307 } 308 309 /* 310 * We were unsuccessful and didn't find anything. 311 */ 312 THC_UNLOCK(&hc_head->hch_mtx); 313 return NULL; 314} 315 316/* 317 * Internal function: insert an entry into the hostcache or return NULL if 318 * unable to allocate a new one. 319 * 320 * If an entry has been returned, the caller becomes responsible for 321 * unlocking the bucket row after he is done reading/modifying the entry. 322 */ 323static struct hc_metrics * 324tcp_hc_insert(struct in_conninfo *inc) 325{ 326 int hash; 327 struct hc_head *hc_head; 328 struct hc_metrics *hc_entry; 329 330 KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer")); 331 332 /* 333 * Hash the foreign ip address. 334 */ 335 if (inc->inc_isipv6) 336 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 337 else 338 hash = HOSTCACHE_HASH(&inc->inc_faddr); 339 340 hc_head = &V_tcp_hostcache.hashbase[hash]; 341 342 /* 343 * Acquire lock for this bucket row; we release the lock if we don't 344 * find an entry, otherwise the caller has to unlock after he is 345 * done. 346 */ 347 THC_LOCK(&hc_head->hch_mtx); 348 349 /* 350 * If the bucket limit is reached, reuse the least-used element. 351 */ 352 if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit || 353 V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) { 354 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead); 355 /* 356 * At first we were dropping the last element, just to 357 * reacquire it in the next two lines again, which isn't very 358 * efficient. Instead just reuse the least used element. 359 * We may drop something that is still "in-use" but we can be 360 * "lossy". 361 * Just give up if this bucket row is empty and we don't have 362 * anything to replace. 363 */ 364 if (hc_entry == NULL) { 365 THC_UNLOCK(&hc_head->hch_mtx); 366 return NULL; 367 } 368 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q); 369 V_tcp_hostcache.hashbase[hash].hch_length--; 370 V_tcp_hostcache.cache_count--; 371 V_tcpstat.tcps_hc_bucketoverflow++; 372#if 0 373 uma_zfree(V_tcp_hostcache.zone, hc_entry); 374#endif 375 } else { 376 /* 377 * Allocate a new entry, or balk if not possible. 378 */ 379 hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT); 380 if (hc_entry == NULL) { 381 THC_UNLOCK(&hc_head->hch_mtx); 382 return NULL; 383 } 384 } 385 386 /* 387 * Initialize basic information of hostcache entry. 388 */ 389 bzero(hc_entry, sizeof(*hc_entry)); 390 if (inc->inc_isipv6) 391 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6)); 392 else 393 hc_entry->ip4 = inc->inc_faddr; 394 hc_entry->rmx_head = hc_head; 395 hc_entry->rmx_expire = V_tcp_hostcache.expire; 396 397 /* 398 * Put it upfront. 399 */ 400 TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q); 401 V_tcp_hostcache.hashbase[hash].hch_length++; 402 V_tcp_hostcache.cache_count++; 403 V_tcpstat.tcps_hc_added++; 404 405 return hc_entry; 406} 407 408/* 409 * External function: look up an entry in the hostcache and fill out the 410 * supplied TCP metrics structure. Fills in NULL when no entry was found or 411 * a value is not set. 412 */ 413void 414tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite) 415{ 416 struct hc_metrics *hc_entry; 417 418 /* 419 * Find the right bucket. 420 */ 421 hc_entry = tcp_hc_lookup(inc); 422 423 /* 424 * If we don't have an existing object. 425 */ 426 if (hc_entry == NULL) { 427 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite)); 428 return; 429 } 430 hc_entry->rmx_hits++; 431 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 432 433 hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu; 434 hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh; 435 hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt; 436 hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar; 437 hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth; 438 hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd; 439 hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe; 440 hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe; 441 442 /* 443 * Unlock bucket row. 444 */ 445 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 446} 447 448/* 449 * External function: look up an entry in the hostcache and return the 450 * discovered path MTU. Returns NULL if no entry is found or value is not 451 * set. 452 */ 453u_long 454tcp_hc_getmtu(struct in_conninfo *inc) 455{ 456 struct hc_metrics *hc_entry; 457 u_long mtu; 458 459 hc_entry = tcp_hc_lookup(inc); 460 if (hc_entry == NULL) { 461 return 0; 462 } 463 hc_entry->rmx_hits++; 464 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 465 466 mtu = hc_entry->rmx_mtu; 467 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 468 return mtu; 469} 470 471/* 472 * External function: update the MTU value of an entry in the hostcache. 473 * Creates a new entry if none was found. 474 */ 475void 476tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu) 477{ 478 struct hc_metrics *hc_entry; 479 480 /* 481 * Find the right bucket. 482 */ 483 hc_entry = tcp_hc_lookup(inc); 484 485 /* 486 * If we don't have an existing object, try to insert a new one. 487 */ 488 if (hc_entry == NULL) { 489 hc_entry = tcp_hc_insert(inc); 490 if (hc_entry == NULL) 491 return; 492 } 493 hc_entry->rmx_updates++; 494 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 495 496 hc_entry->rmx_mtu = mtu; 497 498 /* 499 * Put it upfront so we find it faster next time. 500 */ 501 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 502 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 503 504 /* 505 * Unlock bucket row. 506 */ 507 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 508} 509 510/* 511 * External function: update the TCP metrics of an entry in the hostcache. 512 * Creates a new entry if none was found. 513 */ 514void 515tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml) 516{ 517 struct hc_metrics *hc_entry; 518 519 hc_entry = tcp_hc_lookup(inc); 520 if (hc_entry == NULL) { 521 hc_entry = tcp_hc_insert(inc); 522 if (hc_entry == NULL) 523 return; 524 } 525 hc_entry->rmx_updates++; 526 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 527 528 if (hcml->rmx_rtt != 0) { 529 if (hc_entry->rmx_rtt == 0) 530 hc_entry->rmx_rtt = hcml->rmx_rtt; 531 else 532 hc_entry->rmx_rtt = 533 (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2; 534 V_tcpstat.tcps_cachedrtt++; 535 } 536 if (hcml->rmx_rttvar != 0) { 537 if (hc_entry->rmx_rttvar == 0) 538 hc_entry->rmx_rttvar = hcml->rmx_rttvar; 539 else 540 hc_entry->rmx_rttvar = 541 (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2; 542 V_tcpstat.tcps_cachedrttvar++; 543 } 544 if (hcml->rmx_ssthresh != 0) { 545 if (hc_entry->rmx_ssthresh == 0) 546 hc_entry->rmx_ssthresh = hcml->rmx_ssthresh; 547 else 548 hc_entry->rmx_ssthresh = 549 (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2; 550 V_tcpstat.tcps_cachedssthresh++; 551 } 552 if (hcml->rmx_bandwidth != 0) { 553 if (hc_entry->rmx_bandwidth == 0) 554 hc_entry->rmx_bandwidth = hcml->rmx_bandwidth; 555 else 556 hc_entry->rmx_bandwidth = 557 (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2; 558 /* V_tcpstat.tcps_cachedbandwidth++; */ 559 } 560 if (hcml->rmx_cwnd != 0) { 561 if (hc_entry->rmx_cwnd == 0) 562 hc_entry->rmx_cwnd = hcml->rmx_cwnd; 563 else 564 hc_entry->rmx_cwnd = 565 (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2; 566 /* V_tcpstat.tcps_cachedcwnd++; */ 567 } 568 if (hcml->rmx_sendpipe != 0) { 569 if (hc_entry->rmx_sendpipe == 0) 570 hc_entry->rmx_sendpipe = hcml->rmx_sendpipe; 571 else 572 hc_entry->rmx_sendpipe = 573 (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2; 574 /* V_tcpstat.tcps_cachedsendpipe++; */ 575 } 576 if (hcml->rmx_recvpipe != 0) { 577 if (hc_entry->rmx_recvpipe == 0) 578 hc_entry->rmx_recvpipe = hcml->rmx_recvpipe; 579 else 580 hc_entry->rmx_recvpipe = 581 (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2; 582 /* V_tcpstat.tcps_cachedrecvpipe++; */ 583 } 584 585 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 586 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 587 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 588} 589 590/* 591 * Sysctl function: prints the list and values of all hostcache entries in 592 * unsorted order. 593 */ 594static int 595sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS) 596{ 597 int bufsize; 598 int linesize = 128; 599 char *p, *buf; 600 int len, i, error; 601 struct hc_metrics *hc_entry; 602#ifdef INET6 603 char ip6buf[INET6_ADDRSTRLEN]; 604#endif 605 606 bufsize = linesize * (V_tcp_hostcache.cache_count + 1); 607 608 p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO); 609 610 len = snprintf(p, linesize, 611 "\nIP address MTU SSTRESH RTT RTTVAR BANDWIDTH " 612 " CWND SENDPIPE RECVPIPE HITS UPD EXP\n"); 613 p += len; 614 615#define msec(u) (((u) + 500) / 1000) 616 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 617 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 618 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, 619 rmx_q) { 620 len = snprintf(p, linesize, 621 "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu " 622 "%4lu %4lu %4i\n", 623 hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) : 624#ifdef INET6 625 ip6_sprintf(ip6buf, &hc_entry->ip6), 626#else 627 "IPv6?", 628#endif 629 hc_entry->rmx_mtu, 630 hc_entry->rmx_ssthresh, 631 msec(hc_entry->rmx_rtt * 632 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 633 msec(hc_entry->rmx_rttvar * 634 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 635 hc_entry->rmx_bandwidth * 8, 636 hc_entry->rmx_cwnd, 637 hc_entry->rmx_sendpipe, 638 hc_entry->rmx_recvpipe, 639 hc_entry->rmx_hits, 640 hc_entry->rmx_updates, 641 hc_entry->rmx_expire); 642 p += len; 643 } 644 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 645 } 646#undef msec 647 error = SYSCTL_OUT(req, buf, p - buf); 648 free(buf, M_TEMP); 649 return(error); 650} 651 652/* 653 * Expire and purge (old|all) entries in the tcp_hostcache. Runs 654 * periodically from the callout. 655 */ 656static void 657tcp_hc_purge(void *arg) 658{ 659 struct hc_metrics *hc_entry, *hc_next; 660 int all = (intptr_t)arg; 661 int i; 662 663 if (V_tcp_hostcache.purgeall) { 664 all = 1; 665 V_tcp_hostcache.purgeall = 0; 666 } 667 668 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 669 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 670 TAILQ_FOREACH_SAFE(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, 671 rmx_q, hc_next) { 672 if (all || hc_entry->rmx_expire <= 0) { 673 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket, 674 hc_entry, rmx_q); 675 uma_zfree(V_tcp_hostcache.zone, hc_entry); 676 V_tcp_hostcache.hashbase[i].hch_length--; 677 V_tcp_hostcache.cache_count--; 678 } else 679 hc_entry->rmx_expire -= V_tcp_hostcache.prune; 680 } 681 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 682 } 683 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, tcp_hc_purge, 0); 684} 685