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