Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/netinet/tcp_hostcache.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2002 Andre Oppermann, Internet Business Solutions AG 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. The name of the author may not be used to endorse or promote 16 * products derived from this software without specific prior written 17 * permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * The tcp_hostcache moves the tcp-specific cached metrics from the routing 34 * table to a dedicated structure indexed by the remote IP address. It keeps 35 * information on the measured TCP parameters of past TCP sessions to allow 36 * better initial start values to be used with later connections to/from the 37 * same source. Depending on the network parameters (delay, max MTU, 38 * congestion window) between local and remote sites, this can lead to 39 * significant speed-ups for new TCP connections after the first one. 40 * 41 * Due to the tcp_hostcache, all TCP-specific metrics information in the 42 * routing table have been removed. The inpcb no longer keeps a pointer to 43 * the routing entry, and protocol-initiated route cloning has been removed 44 * as well. With these changes, the routing table has gone back to being 45 * more lightwight and only carries information related to packet forwarding. 46 * 47 * tcp_hostcache is designed for multiple concurrent access in SMP 48 * environments and high contention. All bucket rows have their own lock and 49 * thus multiple lookups and modifies can be done at the same time as long as 50 * they are in different bucket rows. If a request for insertion of a new 51 * record can't be satisfied, it simply returns an empty structure. Nobody 52 * and nothing outside of tcp_hostcache.c will ever point directly to any 53 * entry in the tcp_hostcache. All communication is done in an 54 * object-oriented way and only functions of tcp_hostcache will manipulate 55 * hostcache entries. Otherwise, we are unable to achieve good behaviour in 56 * concurrent access situations. Since tcp_hostcache is only caching 57 * information, there are no fatal consequences if we either can't satisfy 58 * any particular request or have to drop/overwrite an existing entry because 59 * of bucket limit memory constrains. 60 */ 61 62 /* 63 * Many thanks to jlemon for basic structure of tcp_syncache which is being 64 * followed here. 65 */ 66 67 #include <sys/cdefs.h> 68 __FBSDID("$FreeBSD: releng/12.0/sys/netinet/tcp_hostcache.c 336676 2018-07-24 16:35:52Z andrew $"); 69 70 #include "opt_inet6.h" 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/jail.h> 75 #include <sys/kernel.h> 76 #include <sys/lock.h> 77 #include <sys/mutex.h> 78 #include <sys/malloc.h> 79 #include <sys/proc.h> 80 #include <sys/sbuf.h> 81 #include <sys/socket.h> 82 #include <sys/socketvar.h> 83 #include <sys/sysctl.h> 84 85 #include <net/if.h> 86 #include <net/if_var.h> 87 #include <net/route.h> 88 #include <net/vnet.h> 89 90 #include <netinet/in.h> 91 #include <netinet/in_systm.h> 92 #include <netinet/ip.h> 93 #include <netinet/in_var.h> 94 #include <netinet/in_pcb.h> 95 #include <netinet/ip_var.h> 96 #ifdef INET6 97 #include <netinet/ip6.h> 98 #include <netinet6/ip6_var.h> 99 #endif 100 #include <netinet/tcp.h> 101 #include <netinet/tcp_var.h> 102 #include <netinet/tcp_hostcache.h> 103 #ifdef INET6 104 #include <netinet6/tcp6_var.h> 105 #endif 106 107 #include <vm/uma.h> 108 109 /* Arbitrary values */ 110 #define TCP_HOSTCACHE_HASHSIZE 512 111 #define TCP_HOSTCACHE_BUCKETLIMIT 30 112 #define TCP_HOSTCACHE_EXPIRE 60*60 /* one hour */ 113 #define TCP_HOSTCACHE_PRUNE 5*60 /* every 5 minutes */ 114 115 VNET_DEFINE_STATIC(struct tcp_hostcache, tcp_hostcache); 116 #define V_tcp_hostcache VNET(tcp_hostcache) 117 118 VNET_DEFINE_STATIC(struct callout, tcp_hc_callout); 119 #define V_tcp_hc_callout VNET(tcp_hc_callout) 120 121 static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *); 122 static struct hc_metrics *tcp_hc_insert(struct in_conninfo *); 123 static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS); 124 static int sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS); 125 static void tcp_hc_purge_internal(int); 126 static void tcp_hc_purge(void *); 127 128 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, 129 "TCP Host cache"); 130 131 VNET_DEFINE(int, tcp_use_hostcache) = 1; 132 #define V_tcp_use_hostcache VNET(tcp_use_hostcache) 133 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 134 &VNET_NAME(tcp_use_hostcache), 0, 135 "Enable the TCP hostcache"); 136 137 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_VNET | CTLFLAG_RDTUN, 138 &VNET_NAME(tcp_hostcache.cache_limit), 0, 139 "Overall entry limit for hostcache"); 140 141 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 142 &VNET_NAME(tcp_hostcache.hashsize), 0, 143 "Size of TCP hostcache hashtable"); 144 145 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, 146 CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.bucket_limit), 0, 147 "Per-bucket hash limit for hostcache"); 148 149 SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_VNET | CTLFLAG_RD, 150 &VNET_NAME(tcp_hostcache.cache_count), 0, 151 "Current number of entries in hostcache"); 152 153 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(tcp_hostcache.expire), 0, 155 "Expire time of TCP hostcache entries"); 156 157 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_VNET | CTLFLAG_RW, 158 &VNET_NAME(tcp_hostcache.prune), 0, 159 "Time between purge runs"); 160 161 SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_VNET | CTLFLAG_RW, 162 &VNET_NAME(tcp_hostcache.purgeall), 0, 163 "Expire all entires on next purge run"); 164 165 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list, 166 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0, 167 sysctl_tcp_hc_list, "A", "List of all hostcache entries"); 168 169 SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, purgenow, 170 CTLTYPE_INT | CTLFLAG_RW, NULL, 0, 171 sysctl_tcp_hc_purgenow, "I", "Immediately purge all entries"); 172 173 static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache"); 174 175 #define HOSTCACHE_HASH(ip) \ 176 (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \ 177 V_tcp_hostcache.hashmask) 178 179 /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */ 180 #define HOSTCACHE_HASH6(ip6) \ 181 (((ip6)->s6_addr32[0] ^ \ 182 (ip6)->s6_addr32[1] ^ \ 183 (ip6)->s6_addr32[2] ^ \ 184 (ip6)->s6_addr32[3]) & \ 185 V_tcp_hostcache.hashmask) 186 187 #define THC_LOCK(lp) mtx_lock(lp) 188 #define THC_UNLOCK(lp) mtx_unlock(lp) 189 190 void 191 tcp_hc_init(void) 192 { 193 u_int cache_limit; 194 int i; 195 196 /* 197 * Initialize hostcache structures. 198 */ 199 V_tcp_hostcache.cache_count = 0; 200 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; 201 V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT; 202 V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE; 203 V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE; 204 205 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize", 206 &V_tcp_hostcache.hashsize); 207 if (!powerof2(V_tcp_hostcache.hashsize)) { 208 printf("WARNING: hostcache hash size is not a power of 2.\n"); 209 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */ 210 } 211 V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1; 212 213 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit", 214 &V_tcp_hostcache.bucket_limit); 215 216 cache_limit = V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit; 217 V_tcp_hostcache.cache_limit = cache_limit; 218 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit", 219 &V_tcp_hostcache.cache_limit); 220 if (V_tcp_hostcache.cache_limit > cache_limit) 221 V_tcp_hostcache.cache_limit = cache_limit; 222 223 /* 224 * Allocate the hash table. 225 */ 226 V_tcp_hostcache.hashbase = (struct hc_head *) 227 malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head), 228 M_HOSTCACHE, M_WAITOK | M_ZERO); 229 230 /* 231 * Initialize the hash buckets. 232 */ 233 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 234 TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket); 235 V_tcp_hostcache.hashbase[i].hch_length = 0; 236 mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry", 237 NULL, MTX_DEF); 238 } 239 240 /* 241 * Allocate the hostcache entries. 242 */ 243 V_tcp_hostcache.zone = 244 uma_zcreate("hostcache", sizeof(struct hc_metrics), 245 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 246 uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit); 247 248 /* 249 * Set up periodic cache cleanup. 250 */ 251 callout_init(&V_tcp_hc_callout, 1); 252 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 253 tcp_hc_purge, curvnet); 254 } 255 256 #ifdef VIMAGE 257 void 258 tcp_hc_destroy(void) 259 { 260 int i; 261 262 callout_drain(&V_tcp_hc_callout); 263 264 /* Purge all hc entries. */ 265 tcp_hc_purge_internal(1); 266 267 /* Free the uma zone and the allocated hash table. */ 268 uma_zdestroy(V_tcp_hostcache.zone); 269 270 for (i = 0; i < V_tcp_hostcache.hashsize; i++) 271 mtx_destroy(&V_tcp_hostcache.hashbase[i].hch_mtx); 272 free(V_tcp_hostcache.hashbase, M_HOSTCACHE); 273 } 274 #endif 275 276 /* 277 * Internal function: look up an entry in the hostcache or return NULL. 278 * 279 * If an entry has been returned, the caller becomes responsible for 280 * unlocking the bucket row after he is done reading/modifying the entry. 281 */ 282 static struct hc_metrics * 283 tcp_hc_lookup(struct in_conninfo *inc) 284 { 285 int hash; 286 struct hc_head *hc_head; 287 struct hc_metrics *hc_entry; 288 289 if (!V_tcp_use_hostcache) 290 return NULL; 291 292 KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer")); 293 294 /* 295 * Hash the foreign ip address. 296 */ 297 if (inc->inc_flags & INC_ISIPV6) 298 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 299 else 300 hash = HOSTCACHE_HASH(&inc->inc_faddr); 301 302 hc_head = &V_tcp_hostcache.hashbase[hash]; 303 304 /* 305 * Acquire lock for this bucket row; we release the lock if we don't 306 * find an entry, otherwise the caller has to unlock after he is 307 * done. 308 */ 309 THC_LOCK(&hc_head->hch_mtx); 310 311 /* 312 * Iterate through entries in bucket row looking for a match. 313 */ 314 TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) { 315 if (inc->inc_flags & INC_ISIPV6) { 316 /* XXX: check ip6_zoneid */ 317 if (memcmp(&inc->inc6_faddr, &hc_entry->ip6, 318 sizeof(inc->inc6_faddr)) == 0) 319 return hc_entry; 320 } else { 321 if (memcmp(&inc->inc_faddr, &hc_entry->ip4, 322 sizeof(inc->inc_faddr)) == 0) 323 return hc_entry; 324 } 325 } 326 327 /* 328 * We were unsuccessful and didn't find anything. 329 */ 330 THC_UNLOCK(&hc_head->hch_mtx); 331 return NULL; 332 } 333 334 /* 335 * Internal function: insert an entry into the hostcache or return NULL if 336 * unable to allocate a new one. 337 * 338 * If an entry has been returned, the caller becomes responsible for 339 * unlocking the bucket row after he is done reading/modifying the entry. 340 */ 341 static struct hc_metrics * 342 tcp_hc_insert(struct in_conninfo *inc) 343 { 344 int hash; 345 struct hc_head *hc_head; 346 struct hc_metrics *hc_entry; 347 348 if (!V_tcp_use_hostcache) 349 return NULL; 350 351 KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer")); 352 353 /* 354 * Hash the foreign ip address. 355 */ 356 if (inc->inc_flags & INC_ISIPV6) 357 hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 358 else 359 hash = HOSTCACHE_HASH(&inc->inc_faddr); 360 361 hc_head = &V_tcp_hostcache.hashbase[hash]; 362 363 /* 364 * Acquire lock for this bucket row; we release the lock if we don't 365 * find an entry, otherwise the caller has to unlock after he is 366 * done. 367 */ 368 THC_LOCK(&hc_head->hch_mtx); 369 370 /* 371 * If the bucket limit is reached, reuse the least-used element. 372 */ 373 if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit || 374 V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) { 375 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead); 376 /* 377 * At first we were dropping the last element, just to 378 * reacquire it in the next two lines again, which isn't very 379 * efficient. Instead just reuse the least used element. 380 * We may drop something that is still "in-use" but we can be 381 * "lossy". 382 * Just give up if this bucket row is empty and we don't have 383 * anything to replace. 384 */ 385 if (hc_entry == NULL) { 386 THC_UNLOCK(&hc_head->hch_mtx); 387 return NULL; 388 } 389 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q); 390 V_tcp_hostcache.hashbase[hash].hch_length--; 391 V_tcp_hostcache.cache_count--; 392 TCPSTAT_INC(tcps_hc_bucketoverflow); 393 #if 0 394 uma_zfree(V_tcp_hostcache.zone, hc_entry); 395 #endif 396 } else { 397 /* 398 * Allocate a new entry, or balk if not possible. 399 */ 400 hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT); 401 if (hc_entry == NULL) { 402 THC_UNLOCK(&hc_head->hch_mtx); 403 return NULL; 404 } 405 } 406 407 /* 408 * Initialize basic information of hostcache entry. 409 */ 410 bzero(hc_entry, sizeof(*hc_entry)); 411 if (inc->inc_flags & INC_ISIPV6) { 412 hc_entry->ip6 = inc->inc6_faddr; 413 hc_entry->ip6_zoneid = inc->inc6_zoneid; 414 } else 415 hc_entry->ip4 = inc->inc_faddr; 416 hc_entry->rmx_head = hc_head; 417 hc_entry->rmx_expire = V_tcp_hostcache.expire; 418 419 /* 420 * Put it upfront. 421 */ 422 TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q); 423 V_tcp_hostcache.hashbase[hash].hch_length++; 424 V_tcp_hostcache.cache_count++; 425 TCPSTAT_INC(tcps_hc_added); 426 427 return hc_entry; 428 } 429 430 /* 431 * External function: look up an entry in the hostcache and fill out the 432 * supplied TCP metrics structure. Fills in NULL when no entry was found or 433 * a value is not set. 434 */ 435 void 436 tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite) 437 { 438 struct hc_metrics *hc_entry; 439 440 if (!V_tcp_use_hostcache) 441 return; 442 443 /* 444 * Find the right bucket. 445 */ 446 hc_entry = tcp_hc_lookup(inc); 447 448 /* 449 * If we don't have an existing object. 450 */ 451 if (hc_entry == NULL) { 452 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite)); 453 return; 454 } 455 hc_entry->rmx_hits++; 456 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 457 458 hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu; 459 hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh; 460 hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt; 461 hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar; 462 hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd; 463 hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe; 464 hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe; 465 466 /* 467 * Unlock bucket row. 468 */ 469 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 470 } 471 472 /* 473 * External function: look up an entry in the hostcache and return the 474 * discovered path MTU. Returns 0 if no entry is found or value is not 475 * set. 476 */ 477 uint32_t 478 tcp_hc_getmtu(struct in_conninfo *inc) 479 { 480 struct hc_metrics *hc_entry; 481 uint32_t mtu; 482 483 if (!V_tcp_use_hostcache) 484 return 0; 485 486 hc_entry = tcp_hc_lookup(inc); 487 if (hc_entry == NULL) { 488 return 0; 489 } 490 hc_entry->rmx_hits++; 491 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 492 493 mtu = hc_entry->rmx_mtu; 494 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 495 return mtu; 496 } 497 498 /* 499 * External function: update the MTU value of an entry in the hostcache. 500 * Creates a new entry if none was found. 501 */ 502 void 503 tcp_hc_updatemtu(struct in_conninfo *inc, uint32_t mtu) 504 { 505 struct hc_metrics *hc_entry; 506 507 if (!V_tcp_use_hostcache) 508 return; 509 510 /* 511 * Find the right bucket. 512 */ 513 hc_entry = tcp_hc_lookup(inc); 514 515 /* 516 * If we don't have an existing object, try to insert a new one. 517 */ 518 if (hc_entry == NULL) { 519 hc_entry = tcp_hc_insert(inc); 520 if (hc_entry == NULL) 521 return; 522 } 523 hc_entry->rmx_updates++; 524 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 525 526 hc_entry->rmx_mtu = mtu; 527 528 /* 529 * Put it upfront so we find it faster next time. 530 */ 531 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 532 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 533 534 /* 535 * Unlock bucket row. 536 */ 537 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 538 } 539 540 /* 541 * External function: update the TCP metrics of an entry in the hostcache. 542 * Creates a new entry if none was found. 543 */ 544 void 545 tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml) 546 { 547 struct hc_metrics *hc_entry; 548 549 if (!V_tcp_use_hostcache) 550 return; 551 552 hc_entry = tcp_hc_lookup(inc); 553 if (hc_entry == NULL) { 554 hc_entry = tcp_hc_insert(inc); 555 if (hc_entry == NULL) 556 return; 557 } 558 hc_entry->rmx_updates++; 559 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 560 561 if (hcml->rmx_rtt != 0) { 562 if (hc_entry->rmx_rtt == 0) 563 hc_entry->rmx_rtt = hcml->rmx_rtt; 564 else 565 hc_entry->rmx_rtt = ((uint64_t)hc_entry->rmx_rtt + 566 (uint64_t)hcml->rmx_rtt) / 2; 567 TCPSTAT_INC(tcps_cachedrtt); 568 } 569 if (hcml->rmx_rttvar != 0) { 570 if (hc_entry->rmx_rttvar == 0) 571 hc_entry->rmx_rttvar = hcml->rmx_rttvar; 572 else 573 hc_entry->rmx_rttvar = ((uint64_t)hc_entry->rmx_rttvar + 574 (uint64_t)hcml->rmx_rttvar) / 2; 575 TCPSTAT_INC(tcps_cachedrttvar); 576 } 577 if (hcml->rmx_ssthresh != 0) { 578 if (hc_entry->rmx_ssthresh == 0) 579 hc_entry->rmx_ssthresh = hcml->rmx_ssthresh; 580 else 581 hc_entry->rmx_ssthresh = 582 (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2; 583 TCPSTAT_INC(tcps_cachedssthresh); 584 } 585 if (hcml->rmx_cwnd != 0) { 586 if (hc_entry->rmx_cwnd == 0) 587 hc_entry->rmx_cwnd = hcml->rmx_cwnd; 588 else 589 hc_entry->rmx_cwnd = ((uint64_t)hc_entry->rmx_cwnd + 590 (uint64_t)hcml->rmx_cwnd) / 2; 591 /* TCPSTAT_INC(tcps_cachedcwnd); */ 592 } 593 if (hcml->rmx_sendpipe != 0) { 594 if (hc_entry->rmx_sendpipe == 0) 595 hc_entry->rmx_sendpipe = hcml->rmx_sendpipe; 596 else 597 hc_entry->rmx_sendpipe = 598 ((uint64_t)hc_entry->rmx_sendpipe + 599 (uint64_t)hcml->rmx_sendpipe) /2; 600 /* TCPSTAT_INC(tcps_cachedsendpipe); */ 601 } 602 if (hcml->rmx_recvpipe != 0) { 603 if (hc_entry->rmx_recvpipe == 0) 604 hc_entry->rmx_recvpipe = hcml->rmx_recvpipe; 605 else 606 hc_entry->rmx_recvpipe = 607 ((uint64_t)hc_entry->rmx_recvpipe + 608 (uint64_t)hcml->rmx_recvpipe) /2; 609 /* TCPSTAT_INC(tcps_cachedrecvpipe); */ 610 } 611 612 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 613 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 614 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 615 } 616 617 /* 618 * Sysctl function: prints the list and values of all hostcache entries in 619 * unsorted order. 620 */ 621 static int 622 sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS) 623 { 624 const int linesize = 128; 625 struct sbuf sb; 626 int i, error; 627 struct hc_metrics *hc_entry; 628 char ip4buf[INET_ADDRSTRLEN]; 629 #ifdef INET6 630 char ip6buf[INET6_ADDRSTRLEN]; 631 #endif 632 633 if (jailed_without_vnet(curthread->td_ucred) != 0) 634 return (EPERM); 635 636 sbuf_new(&sb, NULL, linesize * (V_tcp_hostcache.cache_count + 1), 637 SBUF_INCLUDENUL); 638 639 sbuf_printf(&sb, 640 "\nIP address MTU SSTRESH RTT RTTVAR " 641 " CWND SENDPIPE RECVPIPE HITS UPD EXP\n"); 642 643 #define msec(u) (((u) + 500) / 1000) 644 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 645 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 646 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, 647 rmx_q) { 648 sbuf_printf(&sb, 649 "%-15s %5u %8u %6lums %6lums %8u %8u %8u %4lu " 650 "%4lu %4i\n", 651 hc_entry->ip4.s_addr ? 652 inet_ntoa_r(hc_entry->ip4, ip4buf) : 653 #ifdef INET6 654 ip6_sprintf(ip6buf, &hc_entry->ip6), 655 #else 656 "IPv6?", 657 #endif 658 hc_entry->rmx_mtu, 659 hc_entry->rmx_ssthresh, 660 msec((u_long)hc_entry->rmx_rtt * 661 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 662 msec((u_long)hc_entry->rmx_rttvar * 663 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE))), 664 hc_entry->rmx_cwnd, 665 hc_entry->rmx_sendpipe, 666 hc_entry->rmx_recvpipe, 667 hc_entry->rmx_hits, 668 hc_entry->rmx_updates, 669 hc_entry->rmx_expire); 670 } 671 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 672 } 673 #undef msec 674 error = sbuf_finish(&sb); 675 if (error == 0) 676 error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb)); 677 sbuf_delete(&sb); 678 return(error); 679 } 680 681 /* 682 * Caller has to make sure the curvnet is set properly. 683 */ 684 static void 685 tcp_hc_purge_internal(int all) 686 { 687 struct hc_metrics *hc_entry, *hc_next; 688 int i; 689 690 for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 691 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 692 TAILQ_FOREACH_SAFE(hc_entry, 693 &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) { 694 if (all || hc_entry->rmx_expire <= 0) { 695 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket, 696 hc_entry, rmx_q); 697 uma_zfree(V_tcp_hostcache.zone, hc_entry); 698 V_tcp_hostcache.hashbase[i].hch_length--; 699 V_tcp_hostcache.cache_count--; 700 } else 701 hc_entry->rmx_expire -= V_tcp_hostcache.prune; 702 } 703 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 704 } 705 } 706 707 /* 708 * Expire and purge (old|all) entries in the tcp_hostcache. Runs 709 * periodically from the callout. 710 */ 711 static void 712 tcp_hc_purge(void *arg) 713 { 714 CURVNET_SET((struct vnet *) arg); 715 int all = 0; 716 717 if (V_tcp_hostcache.purgeall) { 718 all = 1; 719 V_tcp_hostcache.purgeall = 0; 720 } 721 722 tcp_hc_purge_internal(all); 723 724 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 725 tcp_hc_purge, arg); 726 CURVNET_RESTORE(); 727 } 728 729 /* 730 * Expire and purge all entries in hostcache immediately. 731 */ 732 static int 733 sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS) 734 { 735 int error, val; 736 737 val = 0; 738 error = sysctl_handle_int(oidp, &val, 0, req); 739 if (error || !req->newptr) 740 return (error); 741 742 tcp_hc_purge_internal(1); 743 744 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 745 tcp_hc_purge, curvnet); 746 747 return (0); 748 }
Cache object: f86e2115a795d66d60d8ff542723083e
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]