tcp_hostcache.c revision 181888
1226586Sdim/*- 2226586Sdim * Copyright (c) 2002 Andre Oppermann, Internet Business Solutions AG 3226586Sdim * All rights reserved. 4226586Sdim * 5226586Sdim * Redistribution and use in source and binary forms, with or without 6226586Sdim * modification, are permitted provided that the following conditions 7226586Sdim * are met: 8226586Sdim * 1. Redistributions of source code must retain the above copyright 9226586Sdim * notice, this list of conditions and the following disclaimer. 10226586Sdim * 2. Redistributions in binary form must reproduce the above copyright 11226586Sdim * notice, this list of conditions and the following disclaimer in the 12234353Sdim * documentation and/or other materials provided with the distribution. 13234353Sdim * 3. The name of the author may not be used to endorse or promote 14226586Sdim * products derived from this software without specific prior written 15226586Sdim * permission. 16226586Sdim * 17226586Sdim * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18226586Sdim * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19226586Sdim * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20226586Sdim * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21249423Sdim * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22249423Sdim * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23226586Sdim * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24226586Sdim * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25226586Sdim * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26226586Sdim * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27226586Sdim * SUCH DAMAGE. 28226586Sdim */ 29226586Sdim 30226586Sdim/* 31226586Sdim * The tcp_hostcache moves the tcp-specific cached metrics from the routing 32226586Sdim * table to a dedicated structure indexed by the remote IP address. It keeps 33226586Sdim * information on the measured TCP parameters of past TCP sessions to allow 34226586Sdim * better initial start values to be used with later connections to/from the 35226586Sdim * same source. Depending on the network parameters (delay, bandwidth, max 36243830Sdim * MTU, congestion window) between local and remote sites, this can lead to 37243830Sdim * significant speed-ups for new TCP connections after the first one. 38243830Sdim * 39243830Sdim * Due to the tcp_hostcache, all TCP-specific metrics information in the 40243830Sdim * routing table has been removed. The inpcb no longer keeps a pointer to 41243830Sdim * the routing entry, and protocol-initiated route cloning has been removed 42243830Sdim * as well. With these changes, the routing table has gone back to being 43243830Sdim * more lightwight and only carries information related to packet forwarding. 44243830Sdim * 45243830Sdim * tcp_hostcache is designed for multiple concurrent access in SMP 46243830Sdim * environments and high contention. All bucket rows have their own lock and 47243830Sdim * thus multiple lookups and modifies can be done at the same time as long as 48243830Sdim * they are in different bucket rows. If a request for insertion of a new 49243830Sdim * record can't be satisfied, it simply returns an empty structure. Nobody 50243830Sdim * and nothing outside of tcp_hostcache.c will ever point directly to any 51243830Sdim * entry in the tcp_hostcache. All communication is done in an 52243830Sdim * object-oriented way and only functions of tcp_hostcache will manipulate 53243830Sdim * hostcache entries. Otherwise, we are unable to achieve good behaviour in 54243830Sdim * concurrent access situations. Since tcp_hostcache is only caching 55243830Sdim * information, there are no fatal consequences if we either can't satisfy 56226586Sdim * any particular request or have to drop/overwrite an existing entry because 57226586Sdim * of bucket limit memory constrains. 58226586Sdim */ 59226586Sdim 60226586Sdim/* 61226586Sdim * Many thanks to jlemon for basic structure of tcp_syncache which is being 62226586Sdim * followed here. 63226586Sdim */ 64226586Sdim 65226586Sdim#include <sys/cdefs.h> 66226586Sdim__FBSDID("$FreeBSD: head/sys/netinet/tcp_hostcache.c 181888 2008-08-20 01:24:55Z julian $"); 67226586Sdim 68226586Sdim#include "opt_inet6.h" 69226586Sdim 70226586Sdim#include <sys/param.h> 71226586Sdim#include <sys/systm.h> 72226586Sdim#include <sys/kernel.h> 73226586Sdim#include <sys/lock.h> 74226586Sdim#include <sys/mutex.h> 75226586Sdim#include <sys/malloc.h> 76226586Sdim#include <sys/socket.h> 77226586Sdim#include <sys/socketvar.h> 78226586Sdim#include <sys/sysctl.h> 79226586Sdim#include <sys/vimage.h> 80226586Sdim 81226586Sdim#include <net/if.h> 82226586Sdim 83226586Sdim#include <netinet/in.h> 84226586Sdim#include <netinet/in_systm.h> 85226586Sdim#include <netinet/ip.h> 86226586Sdim#include <netinet/in_var.h> 87226586Sdim#include <netinet/in_pcb.h> 88226586Sdim#include <netinet/ip_var.h> 89226586Sdim#ifdef INET6 90226586Sdim#include <netinet/ip6.h> 91226586Sdim#include <netinet6/ip6_var.h> 92226586Sdim#endif 93226586Sdim#include <netinet/tcp.h> 94226586Sdim#include <netinet/tcp_var.h> 95226586Sdim#ifdef INET6 96243830Sdim#include <netinet6/tcp6_var.h> 97243830Sdim#endif 98243830Sdim 99243830Sdim#include <vm/uma.h> 100243830Sdim 101243830Sdim 102243830SdimTAILQ_HEAD(hc_qhead, hc_metrics); 103243830Sdim 104226586Sdimstruct hc_head { 105243830Sdim struct hc_qhead hch_bucket; 106226586Sdim u_int hch_length; 107226586Sdim struct mtx hch_mtx; 108226586Sdim}; 109226586Sdim 110226586Sdimstruct hc_metrics { 111226586Sdim /* housekeeping */ 112226586Sdim TAILQ_ENTRY(hc_metrics) rmx_q; 113226586Sdim struct hc_head *rmx_head; /* head of bucket tail queue */ 114226586Sdim struct in_addr ip4; /* IP address */ 115226586Sdim struct in6_addr ip6; /* IP6 address */ 116226586Sdim /* endpoint specific values for TCP */ 117226586Sdim u_long rmx_mtu; /* MTU for this path */ 118226586Sdim u_long rmx_ssthresh; /* outbound gateway buffer limit */ 119226586Sdim u_long rmx_rtt; /* estimated round trip time */ 120226586Sdim u_long rmx_rttvar; /* estimated rtt variance */ 121226586Sdim u_long rmx_bandwidth; /* estimated bandwidth */ 122226586Sdim u_long rmx_cwnd; /* congestion window */ 123226586Sdim u_long rmx_sendpipe; /* outbound delay-bandwidth product */ 124226586Sdim u_long rmx_recvpipe; /* inbound delay-bandwidth product */ 125226586Sdim /* TCP hostcache internal data */ 126226586Sdim int rmx_expire; /* lifetime for object */ 127226586Sdim u_long rmx_hits; /* number of hits */ 128226586Sdim u_long rmx_updates; /* number of updates */ 129226586Sdim}; 130226586Sdim 131226586Sdim/* Arbitrary values */ 132226586Sdim#define TCP_HOSTCACHE_HASHSIZE 512 133226586Sdim#define TCP_HOSTCACHE_BUCKETLIMIT 30 134226586Sdim#define TCP_HOSTCACHE_EXPIRE 60*60 /* one hour */ 135226586Sdim#define TCP_HOSTCACHE_PRUNE 5*60 /* every 5 minutes */ 136226586Sdim 137226586Sdimstruct tcp_hostcache { 138226586Sdim struct hc_head *hashbase; 139226586Sdim uma_zone_t zone; 140226586Sdim u_int hashsize; 141226586Sdim u_int hashmask; 142226586Sdim u_int bucket_limit; 143226586Sdim u_int cache_count; 144226586Sdim u_int cache_limit; 145226586Sdim int expire; 146226586Sdim int prune; 147243830Sdim int purgeall; 148226586Sdim}; 149226586Sdimstatic struct tcp_hostcache tcp_hostcache; 150226586Sdim 151226586Sdimstatic struct callout tcp_hc_callout; 152226586Sdim 153226586Sdimstatic struct hc_metrics *tcp_hc_lookup(struct in_conninfo *); 154226586Sdimstatic struct hc_metrics *tcp_hc_insert(struct in_conninfo *); 155226586Sdimstatic int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS); 156226586Sdimstatic void tcp_hc_purge(void *); 157226586Sdim 158226586SdimSYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, "TCP Host cache"); 159226586Sdim 160226586SdimSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 161226586Sdim &tcp_hostcache.cache_limit, 0, "Overall entry limit for hostcache"); 162226586Sdim 163226586SdimSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 164226586Sdim &tcp_hostcache.hashsize, 0, "Size of TCP hostcache hashtable"); 165226586Sdim 166226586SdimSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 167226586Sdim &tcp_hostcache.bucket_limit, 0, "Per-bucket hash limit for hostcache"); 168226586Sdim 169226586SdimSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD, 170249423Sdim &tcp_hostcache.cache_count, 0, "Current number of entries in hostcache"); 171226586Sdim 172226586SdimSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW, 173226586Sdim &tcp_hostcache.expire, 0, "Expire time of TCP hostcache entries"); 174226586Sdim 175226586SdimSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW, 176226586Sdim &tcp_hostcache.prune, 0, "Time between purge runs"); 177226586Sdim 178226586SdimSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW, 179226586Sdim &tcp_hostcache.purgeall, 0, "Expire all entires on next purge run"); 180226586Sdim 181226586SdimSYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list, 182226586Sdim CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0, 183226586Sdim sysctl_tcp_hc_list, "A", "List of all hostcache entries"); 184226586Sdim 185243830Sdim 186226586Sdimstatic MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache"); 187226586Sdim 188226586Sdim#define HOSTCACHE_HASH(ip) \ 189226586Sdim (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \ 190226586Sdim V_tcp_hostcache.hashmask) 191226586Sdim 192226586Sdim/* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */ 193226586Sdim#define HOSTCACHE_HASH6(ip6) \ 194226586Sdim (((ip6)->s6_addr32[0] ^ \ 195226586Sdim (ip6)->s6_addr32[1] ^ \ 196226586Sdim (ip6)->s6_addr32[2] ^ \ 197226586Sdim (ip6)->s6_addr32[3]) & \ 198226586Sdim V_tcp_hostcache.hashmask) 199226586Sdim 200226586Sdim#define THC_LOCK(lp) mtx_lock(lp) 201226586Sdim#define THC_UNLOCK(lp) mtx_unlock(lp) 202226586Sdim 203226586Sdimvoid 204226586Sdimtcp_hc_init(void) 205226586Sdim{ 206226586Sdim int i; 207226586Sdim 208226586Sdim /* 209226586Sdim * Initialize hostcache structures. 210226586Sdim */ 211226586Sdim V_tcp_hostcache.cache_count = 0; 212234353Sdim V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; 213234353Sdim V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT; 214234353Sdim V_tcp_hostcache.cache_limit = 215234353Sdim V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit; 216234353Sdim V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE; 217234353Sdim V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE; 218234353Sdim 219234353Sdim TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize", 220234353Sdim &V_tcp_hostcache.hashsize); 221234353Sdim TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit", 222234353Sdim &V_tcp_hostcache.cache_limit); 223234353Sdim TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit", 224234353Sdim &V_tcp_hostcache.bucket_limit); 225234353Sdim if (!powerof2(V_tcp_hostcache.hashsize)) { 226249423Sdim printf("WARNING: hostcache hash size is not a power of 2.\n"); 227249423Sdim V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */ 228249423Sdim } 229226586Sdim V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1; 230249423Sdim 231249423Sdim /* 232249423Sdim * Allocate the hash table. 233249423Sdim */ 234249423Sdim V_tcp_hostcache.hashbase = (struct hc_head *) 235249423Sdim malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head), 236249423Sdim M_HOSTCACHE, M_WAITOK | M_ZERO); 237249423Sdim 238249423Sdim /* 239249423Sdim * Initialize the hash buckets. 240249423Sdim */ 241226586Sdim for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 242226586Sdim TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket); 243226586Sdim 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 = 252 uma_zcreate("hostcache", sizeof(struct hc_metrics), 253 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 254 uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit); 255 256 /* 257 * Set up periodic cache cleanup. 258 */ 259 callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE); 260 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 261 tcp_hc_purge, 0); 262} 263 264/* 265 * Internal function: look up an entry in the hostcache or return NULL. 266 * 267 * If an entry has been returned, the caller becomes responsible for 268 * unlocking the bucket row after he is done reading/modifying the entry. 269 */ 270static struct hc_metrics * 271tcp_hc_lookup(struct in_conninfo *inc) 272{ 273 int hash; 274 struct hc_head *hc_head; 275 struct hc_metrics *hc_entry; 276 277 KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer")); 278 279 /* 280 * Hash the foreign ip address. 281 */ 282 if (inc->inc_isipv6) 283 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 284 else 285 hash = HOSTCACHE_HASH(&inc->inc_faddr); 286 287 hc_head = &V_tcp_hostcache.hashbase[hash]; 288 289 /* 290 * Acquire lock for this bucket row; we release the lock if we don't 291 * find an entry, otherwise the caller has to unlock after he is 292 * done. 293 */ 294 THC_LOCK(&hc_head->hch_mtx); 295 296 /* 297 * Iterate through entries in bucket row looking for a match. 298 */ 299 TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) { 300 if (inc->inc_isipv6) { 301 if (memcmp(&inc->inc6_faddr, &hc_entry->ip6, 302 sizeof(inc->inc6_faddr)) == 0) 303 return hc_entry; 304 } else { 305 if (memcmp(&inc->inc_faddr, &hc_entry->ip4, 306 sizeof(inc->inc_faddr)) == 0) 307 return hc_entry; 308 } 309 } 310 311 /* 312 * We were unsuccessful and didn't find anything. 313 */ 314 THC_UNLOCK(&hc_head->hch_mtx); 315 return NULL; 316} 317 318/* 319 * Internal function: insert an entry into the hostcache or return NULL if 320 * unable to allocate a new one. 321 * 322 * If an entry has been returned, the caller becomes responsible for 323 * unlocking the bucket row after he is done reading/modifying the entry. 324 */ 325static struct hc_metrics * 326tcp_hc_insert(struct in_conninfo *inc) 327{ 328 int hash; 329 struct hc_head *hc_head; 330 struct hc_metrics *hc_entry; 331 332 KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer")); 333 334 /* 335 * Hash the foreign ip address. 336 */ 337 if (inc->inc_isipv6) 338 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 339 else 340 hash = HOSTCACHE_HASH(&inc->inc_faddr); 341 342 hc_head = &V_tcp_hostcache.hashbase[hash]; 343 344 /* 345 * Acquire lock for this bucket row; we release the lock if we don't 346 * find an entry, otherwise the caller has to unlock after he is 347 * done. 348 */ 349 THC_LOCK(&hc_head->hch_mtx); 350 351 /* 352 * If the bucket limit is reached, reuse the least-used element. 353 */ 354 if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit || 355 V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) { 356 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead); 357 /* 358 * At first we were dropping the last element, just to 359 * reacquire it in the next two lines again, which isn't very 360 * efficient. Instead just reuse the least used element. 361 * We may drop something that is still "in-use" but we can be 362 * "lossy". 363 * Just give up if this bucket row is empty and we don't have 364 * anything to replace. 365 */ 366 if (hc_entry == NULL) { 367 THC_UNLOCK(&hc_head->hch_mtx); 368 return NULL; 369 } 370 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q); 371 V_tcp_hostcache.hashbase[hash].hch_length--; 372 V_tcp_hostcache.cache_count--; 373 V_tcpstat.tcps_hc_bucketoverflow++; 374#if 0 375 uma_zfree(V_tcp_hostcache.zone, hc_entry); 376#endif 377 } else { 378 /* 379 * Allocate a new entry, or balk if not possible. 380 */ 381 hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT); 382 if (hc_entry == NULL) { 383 THC_UNLOCK(&hc_head->hch_mtx); 384 return NULL; 385 } 386 } 387 388 /* 389 * Initialize basic information of hostcache entry. 390 */ 391 bzero(hc_entry, sizeof(*hc_entry)); 392 if (inc->inc_isipv6) 393 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6)); 394 else 395 hc_entry->ip4 = inc->inc_faddr; 396 hc_entry->rmx_head = hc_head; 397 hc_entry->rmx_expire = V_tcp_hostcache.expire; 398 399 /* 400 * Put it upfront. 401 */ 402 TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q); 403 V_tcp_hostcache.hashbase[hash].hch_length++; 404 V_tcp_hostcache.cache_count++; 405 V_tcpstat.tcps_hc_added++; 406 407 return hc_entry; 408} 409 410/* 411 * External function: look up an entry in the hostcache and fill out the 412 * supplied TCP metrics structure. Fills in NULL when no entry was found or 413 * a value is not set. 414 */ 415void 416tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite) 417{ 418 struct hc_metrics *hc_entry; 419 420 /* 421 * Find the right bucket. 422 */ 423 hc_entry = tcp_hc_lookup(inc); 424 425 /* 426 * If we don't have an existing object. 427 */ 428 if (hc_entry == NULL) { 429 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite)); 430 return; 431 } 432 hc_entry->rmx_hits++; 433 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 434 435 hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu; 436 hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh; 437 hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt; 438 hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar; 439 hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth; 440 hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd; 441 hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe; 442 hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe; 443 444 /* 445 * Unlock bucket row. 446 */ 447 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 448} 449 450/* 451 * External function: look up an entry in the hostcache and return the 452 * discovered path MTU. Returns NULL if no entry is found or value is not 453 * set. 454 */ 455u_long 456tcp_hc_getmtu(struct in_conninfo *inc) 457{ 458 struct hc_metrics *hc_entry; 459 u_long mtu; 460 461 hc_entry = tcp_hc_lookup(inc); 462 if (hc_entry == NULL) { 463 return 0; 464 } 465 hc_entry->rmx_hits++; 466 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 467 468 mtu = hc_entry->rmx_mtu; 469 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 470 return mtu; 471} 472 473/* 474 * External function: update the MTU value of an entry in the hostcache. 475 * Creates a new entry if none was found. 476 */ 477void 478tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu) 479{ 480 struct hc_metrics *hc_entry; 481 482 /* 483 * Find the right bucket. 484 */ 485 hc_entry = tcp_hc_lookup(inc); 486 487 /* 488 * If we don't have an existing object, try to insert a new one. 489 */ 490 if (hc_entry == NULL) { 491 hc_entry = tcp_hc_insert(inc); 492 if (hc_entry == NULL) 493 return; 494 } 495 hc_entry->rmx_updates++; 496 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 497 498 hc_entry->rmx_mtu = mtu; 499 500 /* 501 * Put it upfront so we find it faster next time. 502 */ 503 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 504 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 505 506 /* 507 * Unlock bucket row. 508 */ 509 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 510} 511 512/* 513 * External function: update the TCP metrics of an entry in the hostcache. 514 * Creates a new entry if none was found. 515 */ 516void 517tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml) 518{ 519 struct hc_metrics *hc_entry; 520 521 hc_entry = tcp_hc_lookup(inc); 522 if (hc_entry == NULL) { 523 hc_entry = tcp_hc_insert(inc); 524 if (hc_entry == NULL) 525 return; 526 } 527 hc_entry->rmx_updates++; 528 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 529 530 if (hcml->rmx_rtt != 0) { 531 if (hc_entry->rmx_rtt == 0) 532 hc_entry->rmx_rtt = hcml->rmx_rtt; 533 else 534 hc_entry->rmx_rtt = 535 (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2; 536 V_tcpstat.tcps_cachedrtt++; 537 } 538 if (hcml->rmx_rttvar != 0) { 539 if (hc_entry->rmx_rttvar == 0) 540 hc_entry->rmx_rttvar = hcml->rmx_rttvar; 541 else 542 hc_entry->rmx_rttvar = 543 (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2; 544 V_tcpstat.tcps_cachedrttvar++; 545 } 546 if (hcml->rmx_ssthresh != 0) { 547 if (hc_entry->rmx_ssthresh == 0) 548 hc_entry->rmx_ssthresh = hcml->rmx_ssthresh; 549 else 550 hc_entry->rmx_ssthresh = 551 (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2; 552 V_tcpstat.tcps_cachedssthresh++; 553 } 554 if (hcml->rmx_bandwidth != 0) { 555 if (hc_entry->rmx_bandwidth == 0) 556 hc_entry->rmx_bandwidth = hcml->rmx_bandwidth; 557 else 558 hc_entry->rmx_bandwidth = 559 (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2; 560 /* V_tcpstat.tcps_cachedbandwidth++; */ 561 } 562 if (hcml->rmx_cwnd != 0) { 563 if (hc_entry->rmx_cwnd == 0) 564 hc_entry->rmx_cwnd = hcml->rmx_cwnd; 565 else 566 hc_entry->rmx_cwnd = 567 (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2; 568 /* V_tcpstat.tcps_cachedcwnd++; */ 569 } 570 if (hcml->rmx_sendpipe != 0) { 571 if (hc_entry->rmx_sendpipe == 0) 572 hc_entry->rmx_sendpipe = hcml->rmx_sendpipe; 573 else 574 hc_entry->rmx_sendpipe = 575 (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2; 576 /* V_tcpstat.tcps_cachedsendpipe++; */ 577 } 578 if (hcml->rmx_recvpipe != 0) { 579 if (hc_entry->rmx_recvpipe == 0) 580 hc_entry->rmx_recvpipe = hcml->rmx_recvpipe; 581 else 582 hc_entry->rmx_recvpipe = 583 (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2; 584 /* V_tcpstat.tcps_cachedrecvpipe++; */ 585 } 586 587 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 588 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 589 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 590} 591 592/* 593 * Sysctl function: prints the list and values of all hostcache entries in 594 * unsorted order. 595 */ 596static int 597sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS) 598{ 599 int bufsize; 600 int linesize = 128; 601 char *p, *buf; 602 int len, i, error; 603 struct hc_metrics *hc_entry; 604#ifdef INET6 605 char ip6buf[INET6_ADDRSTRLEN]; 606#endif 607 608 bufsize = linesize * (V_tcp_hostcache.cache_count + 1); 609 610 p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO); 611 612 len = snprintf(p, linesize, 613 "\nIP address MTU SSTRESH RTT RTTVAR BANDWIDTH " 614 " CWND SENDPIPE RECVPIPE HITS UPD EXP\n"); 615 p += len; 616 617#define msec(u) (((u) + 500) / 1000) 618 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 619 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 620 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, 621 rmx_q) { 622 len = snprintf(p, linesize, 623 "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu " 624 "%4lu %4lu %4i\n", 625 hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) : 626#ifdef INET6 627 ip6_sprintf(ip6buf, &hc_entry->ip6), 628#else 629 "IPv6?", 630#endif 631 hc_entry->rmx_mtu, 632 hc_entry->rmx_ssthresh, 633 msec(hc_entry->rmx_rtt * 634 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 635 msec(hc_entry->rmx_rttvar * 636 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 637 hc_entry->rmx_bandwidth * 8, 638 hc_entry->rmx_cwnd, 639 hc_entry->rmx_sendpipe, 640 hc_entry->rmx_recvpipe, 641 hc_entry->rmx_hits, 642 hc_entry->rmx_updates, 643 hc_entry->rmx_expire); 644 p += len; 645 } 646 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 647 } 648#undef msec 649 error = SYSCTL_OUT(req, buf, p - buf); 650 free(buf, M_TEMP); 651 return(error); 652} 653 654/* 655 * Expire and purge (old|all) entries in the tcp_hostcache. Runs 656 * periodically from the callout. 657 */ 658static void 659tcp_hc_purge(void *arg) 660{ 661 struct hc_metrics *hc_entry, *hc_next; 662 int all = (intptr_t)arg; 663 int i; 664 665 if (V_tcp_hostcache.purgeall) { 666 all = 1; 667 V_tcp_hostcache.purgeall = 0; 668 } 669 670 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 671 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 672 TAILQ_FOREACH_SAFE(hc_entry, 673 &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) { 674 if (all || hc_entry->rmx_expire <= 0) { 675 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket, 676 hc_entry, rmx_q); 677 uma_zfree(V_tcp_hostcache.zone, hc_entry); 678 V_tcp_hostcache.hashbase[i].hch_length--; 679 V_tcp_hostcache.cache_count--; 680 } else 681 hc_entry->rmx_expire -= V_tcp_hostcache.prune; 682 } 683 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 684 } 685 686 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 687 tcp_hc_purge, arg); 688} 689