tcp_hostcache.c revision 182411
1139823Simp/*- 2122922Sandre * Copyright (c) 2002 Andre Oppermann, Internet Business Solutions AG 3122922Sandre * All rights reserved. 4122922Sandre * 5122922Sandre * Redistribution and use in source and binary forms, with or without 6122922Sandre * modification, are permitted provided that the following conditions 7122922Sandre * are met: 8122922Sandre * 1. Redistributions of source code must retain the above copyright 9122922Sandre * notice, this list of conditions and the following disclaimer. 10122922Sandre * 2. Redistributions in binary form must reproduce the above copyright 11122922Sandre * notice, this list of conditions and the following disclaimer in the 12122922Sandre * documentation and/or other materials provided with the distribution. 13122922Sandre * 3. The name of the author may not be used to endorse or promote 14122922Sandre * products derived from this software without specific prior written 15122922Sandre * permission. 16122922Sandre * 17122922Sandre * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18122922Sandre * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19122922Sandre * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20122922Sandre * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21122922Sandre * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22122922Sandre * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23122922Sandre * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24122922Sandre * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25122922Sandre * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26122922Sandre * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27122922Sandre * SUCH DAMAGE. 28122922Sandre */ 29122922Sandre 30122922Sandre/* 31170030Srwatson * The tcp_hostcache moves the tcp-specific cached metrics from the routing 32170030Srwatson * table to a dedicated structure indexed by the remote IP address. It keeps 33170030Srwatson * information on the measured TCP parameters of past TCP sessions to allow 34170030Srwatson * better initial start values to be used with later connections to/from the 35170030Srwatson * same source. Depending on the network parameters (delay, bandwidth, max 36170030Srwatson * MTU, congestion window) between local and remote sites, this can lead to 37170030Srwatson * significant speed-ups for new TCP connections after the first one. 38122922Sandre * 39170030Srwatson * Due to the tcp_hostcache, all TCP-specific metrics information in the 40182411Srpaulo * routing table have been removed. The inpcb no longer keeps a pointer to 41170030Srwatson * the routing entry, and protocol-initiated route cloning has been removed 42170030Srwatson * as well. With these changes, the routing table has gone back to being 43170030Srwatson * more lightwight and only carries information related to packet forwarding. 44122922Sandre * 45170030Srwatson * tcp_hostcache is designed for multiple concurrent access in SMP 46170030Srwatson * environments and high contention. All bucket rows have their own lock and 47170030Srwatson * thus multiple lookups and modifies can be done at the same time as long as 48170030Srwatson * they are in different bucket rows. If a request for insertion of a new 49170030Srwatson * record can't be satisfied, it simply returns an empty structure. Nobody 50170030Srwatson * and nothing outside of tcp_hostcache.c will ever point directly to any 51170030Srwatson * entry in the tcp_hostcache. All communication is done in an 52170030Srwatson * object-oriented way and only functions of tcp_hostcache will manipulate 53170030Srwatson * hostcache entries. Otherwise, we are unable to achieve good behaviour in 54170030Srwatson * concurrent access situations. Since tcp_hostcache is only caching 55170030Srwatson * information, there are no fatal consequences if we either can't satisfy 56170030Srwatson * any particular request or have to drop/overwrite an existing entry because 57170030Srwatson * of bucket limit memory constrains. 58122922Sandre */ 59122922Sandre 60122922Sandre/* 61122922Sandre * Many thanks to jlemon for basic structure of tcp_syncache which is being 62122922Sandre * followed here. 63122922Sandre */ 64122922Sandre 65172467Ssilby#include <sys/cdefs.h> 66172467Ssilby__FBSDID("$FreeBSD: head/sys/netinet/tcp_hostcache.c 182411 2008-08-28 21:55:40Z rpaulo $"); 67172467Ssilby 68122922Sandre#include "opt_inet6.h" 69122922Sandre 70122922Sandre#include <sys/param.h> 71122922Sandre#include <sys/systm.h> 72122922Sandre#include <sys/kernel.h> 73122922Sandre#include <sys/lock.h> 74122922Sandre#include <sys/mutex.h> 75122922Sandre#include <sys/malloc.h> 76122922Sandre#include <sys/socket.h> 77122922Sandre#include <sys/socketvar.h> 78122922Sandre#include <sys/sysctl.h> 79181803Sbz#include <sys/vimage.h> 80122922Sandre 81122922Sandre#include <net/if.h> 82122922Sandre 83122922Sandre#include <netinet/in.h> 84122922Sandre#include <netinet/in_systm.h> 85122922Sandre#include <netinet/ip.h> 86122922Sandre#include <netinet/in_var.h> 87122922Sandre#include <netinet/in_pcb.h> 88122922Sandre#include <netinet/ip_var.h> 89122922Sandre#ifdef INET6 90122922Sandre#include <netinet/ip6.h> 91122922Sandre#include <netinet6/ip6_var.h> 92122922Sandre#endif 93122922Sandre#include <netinet/tcp.h> 94122922Sandre#include <netinet/tcp_var.h> 95122922Sandre#ifdef INET6 96122922Sandre#include <netinet6/tcp6_var.h> 97122922Sandre#endif 98122922Sandre 99122922Sandre#include <vm/uma.h> 100122922Sandre 101122922Sandre 102122922SandreTAILQ_HEAD(hc_qhead, hc_metrics); 103122922Sandre 104122922Sandrestruct hc_head { 105122922Sandre struct hc_qhead hch_bucket; 106122922Sandre u_int hch_length; 107122922Sandre struct mtx hch_mtx; 108122922Sandre}; 109122922Sandre 110122922Sandrestruct hc_metrics { 111122922Sandre /* housekeeping */ 112122922Sandre TAILQ_ENTRY(hc_metrics) rmx_q; 113122922Sandre struct hc_head *rmx_head; /* head of bucket tail queue */ 114122922Sandre struct in_addr ip4; /* IP address */ 115122922Sandre struct in6_addr ip6; /* IP6 address */ 116170030Srwatson /* endpoint specific values for TCP */ 117122922Sandre u_long rmx_mtu; /* MTU for this path */ 118122922Sandre u_long rmx_ssthresh; /* outbound gateway buffer limit */ 119122922Sandre u_long rmx_rtt; /* estimated round trip time */ 120122922Sandre u_long rmx_rttvar; /* estimated rtt variance */ 121122922Sandre u_long rmx_bandwidth; /* estimated bandwidth */ 122122922Sandre u_long rmx_cwnd; /* congestion window */ 123122922Sandre u_long rmx_sendpipe; /* outbound delay-bandwidth product */ 124122922Sandre u_long rmx_recvpipe; /* inbound delay-bandwidth product */ 125170030Srwatson /* TCP hostcache internal data */ 126122922Sandre int rmx_expire; /* lifetime for object */ 127122922Sandre u_long rmx_hits; /* number of hits */ 128122922Sandre u_long rmx_updates; /* number of updates */ 129122922Sandre}; 130122922Sandre 131122922Sandre/* Arbitrary values */ 132122922Sandre#define TCP_HOSTCACHE_HASHSIZE 512 133122922Sandre#define TCP_HOSTCACHE_BUCKETLIMIT 30 134122922Sandre#define TCP_HOSTCACHE_EXPIRE 60*60 /* one hour */ 135122922Sandre#define TCP_HOSTCACHE_PRUNE 5*60 /* every 5 minutes */ 136122922Sandre 137122922Sandrestruct tcp_hostcache { 138122922Sandre struct hc_head *hashbase; 139122922Sandre uma_zone_t zone; 140122922Sandre u_int hashsize; 141122922Sandre u_int hashmask; 142122922Sandre u_int bucket_limit; 143122922Sandre u_int cache_count; 144122922Sandre u_int cache_limit; 145122922Sandre int expire; 146170434Syar int prune; 147122922Sandre int purgeall; 148122922Sandre}; 149122922Sandrestatic struct tcp_hostcache tcp_hostcache; 150122922Sandre 151122922Sandrestatic struct callout tcp_hc_callout; 152122922Sandre 153122922Sandrestatic struct hc_metrics *tcp_hc_lookup(struct in_conninfo *); 154122922Sandrestatic struct hc_metrics *tcp_hc_insert(struct in_conninfo *); 155122922Sandrestatic int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS); 156122922Sandrestatic void tcp_hc_purge(void *); 157122922Sandre 158122922SandreSYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, "TCP Host cache"); 159122922Sandre 160122922SandreSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN, 161167784Sandre &tcp_hostcache.cache_limit, 0, "Overall entry limit for hostcache"); 162122922Sandre 163122922SandreSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN, 164167784Sandre &tcp_hostcache.hashsize, 0, "Size of TCP hostcache hashtable"); 165122922Sandre 166122922SandreSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN, 167167784Sandre &tcp_hostcache.bucket_limit, 0, "Per-bucket hash limit for hostcache"); 168122922Sandre 169122922SandreSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD, 170167784Sandre &tcp_hostcache.cache_count, 0, "Current number of entries in hostcache"); 171122922Sandre 172122922SandreSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW, 173167784Sandre &tcp_hostcache.expire, 0, "Expire time of TCP hostcache entries"); 174122922Sandre 175170434SyarSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW, 176170434Syar &tcp_hostcache.prune, 0, "Time between purge runs"); 177170434Syar 178122922SandreSYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW, 179167784Sandre &tcp_hostcache.purgeall, 0, "Expire all entires on next purge run"); 180122922Sandre 181122922SandreSYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list, 182167784Sandre CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0, 183167784Sandre sysctl_tcp_hc_list, "A", "List of all hostcache entries"); 184122922Sandre 185122922Sandre 186122922Sandrestatic MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache"); 187122922Sandre 188122922Sandre#define HOSTCACHE_HASH(ip) \ 189133874Srwatson (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \ 190181803Sbz V_tcp_hostcache.hashmask) 191122922Sandre 192122922Sandre/* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */ 193133874Srwatson#define HOSTCACHE_HASH6(ip6) \ 194122922Sandre (((ip6)->s6_addr32[0] ^ \ 195122922Sandre (ip6)->s6_addr32[1] ^ \ 196122922Sandre (ip6)->s6_addr32[2] ^ \ 197122922Sandre (ip6)->s6_addr32[3]) & \ 198181803Sbz V_tcp_hostcache.hashmask) 199122922Sandre 200122922Sandre#define THC_LOCK(lp) mtx_lock(lp) 201122922Sandre#define THC_UNLOCK(lp) mtx_unlock(lp) 202122922Sandre 203122922Sandrevoid 204122922Sandretcp_hc_init(void) 205122922Sandre{ 206122922Sandre int i; 207122922Sandre 208122922Sandre /* 209170030Srwatson * Initialize hostcache structures. 210122922Sandre */ 211181803Sbz V_tcp_hostcache.cache_count = 0; 212181803Sbz V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; 213181803Sbz V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT; 214181803Sbz V_tcp_hostcache.cache_limit = 215181803Sbz V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit; 216181803Sbz V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE; 217181803Sbz V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE; 218122922Sandre 219133874Srwatson TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize", 220181803Sbz &V_tcp_hostcache.hashsize); 221133874Srwatson TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit", 222181803Sbz &V_tcp_hostcache.cache_limit); 223133874Srwatson TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit", 224181803Sbz &V_tcp_hostcache.bucket_limit); 225181803Sbz if (!powerof2(V_tcp_hostcache.hashsize)) { 226133874Srwatson printf("WARNING: hostcache hash size is not a power of 2.\n"); 227181803Sbz V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */ 228133874Srwatson } 229181803Sbz V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1; 230122922Sandre 231122922Sandre /* 232170030Srwatson * Allocate the hash table. 233122922Sandre */ 234181803Sbz V_tcp_hostcache.hashbase = (struct hc_head *) 235181803Sbz malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head), 236122922Sandre M_HOSTCACHE, M_WAITOK | M_ZERO); 237122922Sandre 238122922Sandre /* 239170030Srwatson * Initialize the hash buckets. 240122922Sandre */ 241181803Sbz for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 242181803Sbz TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket); 243181803Sbz V_tcp_hostcache.hashbase[i].hch_length = 0; 244181803Sbz mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry", 245122922Sandre NULL, MTX_DEF); 246122922Sandre } 247122922Sandre 248122922Sandre /* 249122922Sandre * Allocate the hostcache entries. 250122922Sandre */ 251181887Sjulian V_tcp_hostcache.zone = 252181888Sjulian uma_zcreate("hostcache", sizeof(struct hc_metrics), 253181888Sjulian NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 254181803Sbz uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit); 255122922Sandre 256122922Sandre /* 257122922Sandre * Set up periodic cache cleanup. 258122922Sandre */ 259181803Sbz callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE); 260181887Sjulian callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 261181887Sjulian tcp_hc_purge, 0); 262122922Sandre} 263122922Sandre 264122922Sandre/* 265170030Srwatson * Internal function: look up an entry in the hostcache or return NULL. 266122922Sandre * 267122922Sandre * If an entry has been returned, the caller becomes responsible for 268122922Sandre * unlocking the bucket row after he is done reading/modifying the entry. 269122922Sandre */ 270122922Sandrestatic struct hc_metrics * 271122922Sandretcp_hc_lookup(struct in_conninfo *inc) 272122922Sandre{ 273122922Sandre int hash; 274122922Sandre struct hc_head *hc_head; 275122922Sandre struct hc_metrics *hc_entry; 276122922Sandre 277122922Sandre KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer")); 278122922Sandre 279122922Sandre /* 280122922Sandre * Hash the foreign ip address. 281122922Sandre */ 282122922Sandre if (inc->inc_isipv6) 283122922Sandre hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 284122922Sandre else 285122922Sandre hash = HOSTCACHE_HASH(&inc->inc_faddr); 286122922Sandre 287181803Sbz hc_head = &V_tcp_hostcache.hashbase[hash]; 288122922Sandre 289122922Sandre /* 290170030Srwatson * Acquire lock for this bucket row; we release the lock if we don't 291170030Srwatson * find an entry, otherwise the caller has to unlock after he is 292170030Srwatson * done. 293122922Sandre */ 294122922Sandre THC_LOCK(&hc_head->hch_mtx); 295122922Sandre 296122922Sandre /* 297170030Srwatson * Iterate through entries in bucket row looking for a match. 298122922Sandre */ 299122922Sandre TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) { 300122922Sandre if (inc->inc_isipv6) { 301122922Sandre if (memcmp(&inc->inc6_faddr, &hc_entry->ip6, 302122922Sandre sizeof(inc->inc6_faddr)) == 0) 303122922Sandre return hc_entry; 304122922Sandre } else { 305122922Sandre if (memcmp(&inc->inc_faddr, &hc_entry->ip4, 306122922Sandre sizeof(inc->inc_faddr)) == 0) 307122922Sandre return hc_entry; 308122922Sandre } 309122922Sandre } 310122922Sandre 311122922Sandre /* 312170030Srwatson * We were unsuccessful and didn't find anything. 313122922Sandre */ 314122922Sandre THC_UNLOCK(&hc_head->hch_mtx); 315122922Sandre return NULL; 316122922Sandre} 317122922Sandre 318122922Sandre/* 319170030Srwatson * Internal function: insert an entry into the hostcache or return NULL if 320170030Srwatson * unable to allocate a new one. 321133874Srwatson * 322122922Sandre * If an entry has been returned, the caller becomes responsible for 323122922Sandre * unlocking the bucket row after he is done reading/modifying the entry. 324122922Sandre */ 325122922Sandrestatic struct hc_metrics * 326122922Sandretcp_hc_insert(struct in_conninfo *inc) 327122922Sandre{ 328122922Sandre int hash; 329122922Sandre struct hc_head *hc_head; 330122922Sandre struct hc_metrics *hc_entry; 331122922Sandre 332122922Sandre KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer")); 333122922Sandre 334122922Sandre /* 335170030Srwatson * Hash the foreign ip address. 336122922Sandre */ 337122922Sandre if (inc->inc_isipv6) 338122922Sandre hash = HOSTCACHE_HASH6(&inc->inc6_faddr); 339122922Sandre else 340122922Sandre hash = HOSTCACHE_HASH(&inc->inc_faddr); 341122922Sandre 342181803Sbz hc_head = &V_tcp_hostcache.hashbase[hash]; 343122922Sandre 344122922Sandre /* 345170030Srwatson * Acquire lock for this bucket row; we release the lock if we don't 346170030Srwatson * find an entry, otherwise the caller has to unlock after he is 347170030Srwatson * done. 348122922Sandre */ 349122922Sandre THC_LOCK(&hc_head->hch_mtx); 350122922Sandre 351122922Sandre /* 352170030Srwatson * If the bucket limit is reached, reuse the least-used element. 353122922Sandre */ 354181803Sbz if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit || 355181803Sbz V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) { 356122922Sandre hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead); 357122922Sandre /* 358122922Sandre * At first we were dropping the last element, just to 359170030Srwatson * reacquire it in the next two lines again, which isn't very 360170030Srwatson * efficient. Instead just reuse the least used element. 361170030Srwatson * We may drop something that is still "in-use" but we can be 362170030Srwatson * "lossy". 363170405Sandre * Just give up if this bucket row is empty and we don't have 364170405Sandre * anything to replace. 365122922Sandre */ 366170405Sandre if (hc_entry == NULL) { 367170405Sandre THC_UNLOCK(&hc_head->hch_mtx); 368170405Sandre return NULL; 369170405Sandre } 370122922Sandre TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q); 371181803Sbz V_tcp_hostcache.hashbase[hash].hch_length--; 372181803Sbz V_tcp_hostcache.cache_count--; 373181803Sbz V_tcpstat.tcps_hc_bucketoverflow++; 374123028Sandre#if 0 375181803Sbz uma_zfree(V_tcp_hostcache.zone, hc_entry); 376122922Sandre#endif 377122922Sandre } else { 378122922Sandre /* 379170030Srwatson * Allocate a new entry, or balk if not possible. 380122922Sandre */ 381181803Sbz hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT); 382122922Sandre if (hc_entry == NULL) { 383122922Sandre THC_UNLOCK(&hc_head->hch_mtx); 384122922Sandre return NULL; 385122922Sandre } 386122922Sandre } 387122922Sandre 388122922Sandre /* 389170030Srwatson * Initialize basic information of hostcache entry. 390122922Sandre */ 391122922Sandre bzero(hc_entry, sizeof(*hc_entry)); 392122922Sandre if (inc->inc_isipv6) 393123113Sandre bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6)); 394122922Sandre else 395122922Sandre hc_entry->ip4 = inc->inc_faddr; 396122922Sandre hc_entry->rmx_head = hc_head; 397181803Sbz hc_entry->rmx_expire = V_tcp_hostcache.expire; 398122922Sandre 399122922Sandre /* 400170030Srwatson * Put it upfront. 401122922Sandre */ 402122922Sandre TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q); 403181803Sbz V_tcp_hostcache.hashbase[hash].hch_length++; 404181803Sbz V_tcp_hostcache.cache_count++; 405181803Sbz V_tcpstat.tcps_hc_added++; 406122922Sandre 407122922Sandre return hc_entry; 408122922Sandre} 409122922Sandre 410122922Sandre/* 411170030Srwatson * External function: look up an entry in the hostcache and fill out the 412170030Srwatson * supplied TCP metrics structure. Fills in NULL when no entry was found or 413170030Srwatson * a value is not set. 414122922Sandre */ 415122922Sandrevoid 416122922Sandretcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite) 417122922Sandre{ 418122922Sandre struct hc_metrics *hc_entry; 419122922Sandre 420122922Sandre /* 421170030Srwatson * Find the right bucket. 422122922Sandre */ 423122922Sandre hc_entry = tcp_hc_lookup(inc); 424122922Sandre 425122922Sandre /* 426170030Srwatson * If we don't have an existing object. 427122922Sandre */ 428122922Sandre if (hc_entry == NULL) { 429122922Sandre bzero(hc_metrics_lite, sizeof(*hc_metrics_lite)); 430122922Sandre return; 431122922Sandre } 432122922Sandre hc_entry->rmx_hits++; 433181803Sbz hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 434122922Sandre 435122922Sandre hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu; 436122922Sandre hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh; 437122922Sandre hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt; 438122922Sandre hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar; 439122922Sandre hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth; 440122922Sandre hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd; 441122922Sandre hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe; 442122922Sandre hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe; 443122922Sandre 444122922Sandre /* 445170030Srwatson * Unlock bucket row. 446122922Sandre */ 447122922Sandre THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 448122922Sandre} 449122922Sandre 450122922Sandre/* 451170030Srwatson * External function: look up an entry in the hostcache and return the 452170030Srwatson * discovered path MTU. Returns NULL if no entry is found or value is not 453138409Srwatson * set. 454122922Sandre */ 455122922Sandreu_long 456122922Sandretcp_hc_getmtu(struct in_conninfo *inc) 457122922Sandre{ 458122922Sandre struct hc_metrics *hc_entry; 459122922Sandre u_long mtu; 460122922Sandre 461122922Sandre hc_entry = tcp_hc_lookup(inc); 462122922Sandre if (hc_entry == NULL) { 463122922Sandre return 0; 464122922Sandre } 465122922Sandre hc_entry->rmx_hits++; 466181803Sbz hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 467122922Sandre 468122922Sandre mtu = hc_entry->rmx_mtu; 469122922Sandre THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 470122922Sandre return mtu; 471122922Sandre} 472122922Sandre 473122922Sandre/* 474170030Srwatson * External function: update the MTU value of an entry in the hostcache. 475122922Sandre * Creates a new entry if none was found. 476122922Sandre */ 477122922Sandrevoid 478122922Sandretcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu) 479122922Sandre{ 480122922Sandre struct hc_metrics *hc_entry; 481122922Sandre 482122922Sandre /* 483170030Srwatson * Find the right bucket. 484122922Sandre */ 485122922Sandre hc_entry = tcp_hc_lookup(inc); 486122922Sandre 487122922Sandre /* 488170030Srwatson * If we don't have an existing object, try to insert a new one. 489122922Sandre */ 490122922Sandre if (hc_entry == NULL) { 491122922Sandre hc_entry = tcp_hc_insert(inc); 492122922Sandre if (hc_entry == NULL) 493122922Sandre return; 494122922Sandre } 495122922Sandre hc_entry->rmx_updates++; 496181803Sbz hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 497122922Sandre 498122922Sandre hc_entry->rmx_mtu = mtu; 499122922Sandre 500122922Sandre /* 501170030Srwatson * Put it upfront so we find it faster next time. 502122922Sandre */ 503122922Sandre TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 504122922Sandre TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 505122922Sandre 506122922Sandre /* 507170030Srwatson * Unlock bucket row. 508122922Sandre */ 509122922Sandre THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 510122922Sandre} 511122922Sandre 512122922Sandre/* 513170030Srwatson * External function: update the TCP metrics of an entry in the hostcache. 514122922Sandre * Creates a new entry if none was found. 515122922Sandre */ 516122922Sandrevoid 517122922Sandretcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml) 518122922Sandre{ 519122922Sandre struct hc_metrics *hc_entry; 520122922Sandre 521122922Sandre hc_entry = tcp_hc_lookup(inc); 522122922Sandre if (hc_entry == NULL) { 523122922Sandre hc_entry = tcp_hc_insert(inc); 524122922Sandre if (hc_entry == NULL) 525122922Sandre return; 526122922Sandre } 527122922Sandre hc_entry->rmx_updates++; 528181803Sbz hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ 529122922Sandre 530122922Sandre if (hcml->rmx_rtt != 0) { 531122922Sandre if (hc_entry->rmx_rtt == 0) 532122922Sandre hc_entry->rmx_rtt = hcml->rmx_rtt; 533122922Sandre else 534122922Sandre hc_entry->rmx_rtt = 535122922Sandre (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2; 536181803Sbz V_tcpstat.tcps_cachedrtt++; 537122922Sandre } 538122922Sandre if (hcml->rmx_rttvar != 0) { 539122922Sandre if (hc_entry->rmx_rttvar == 0) 540133874Srwatson hc_entry->rmx_rttvar = hcml->rmx_rttvar; 541122922Sandre else 542122922Sandre hc_entry->rmx_rttvar = 543122922Sandre (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2; 544181803Sbz V_tcpstat.tcps_cachedrttvar++; 545122922Sandre } 546122922Sandre if (hcml->rmx_ssthresh != 0) { 547122922Sandre if (hc_entry->rmx_ssthresh == 0) 548122922Sandre hc_entry->rmx_ssthresh = hcml->rmx_ssthresh; 549122922Sandre else 550122922Sandre hc_entry->rmx_ssthresh = 551122922Sandre (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2; 552181803Sbz V_tcpstat.tcps_cachedssthresh++; 553122922Sandre } 554122922Sandre if (hcml->rmx_bandwidth != 0) { 555122922Sandre if (hc_entry->rmx_bandwidth == 0) 556122922Sandre hc_entry->rmx_bandwidth = hcml->rmx_bandwidth; 557122922Sandre else 558122922Sandre hc_entry->rmx_bandwidth = 559122922Sandre (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2; 560181803Sbz /* V_tcpstat.tcps_cachedbandwidth++; */ 561122922Sandre } 562122922Sandre if (hcml->rmx_cwnd != 0) { 563122922Sandre if (hc_entry->rmx_cwnd == 0) 564122922Sandre hc_entry->rmx_cwnd = hcml->rmx_cwnd; 565122922Sandre else 566122922Sandre hc_entry->rmx_cwnd = 567122922Sandre (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2; 568181803Sbz /* V_tcpstat.tcps_cachedcwnd++; */ 569122922Sandre } 570122922Sandre if (hcml->rmx_sendpipe != 0) { 571122922Sandre if (hc_entry->rmx_sendpipe == 0) 572122922Sandre hc_entry->rmx_sendpipe = hcml->rmx_sendpipe; 573122922Sandre else 574122922Sandre hc_entry->rmx_sendpipe = 575122922Sandre (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2; 576181803Sbz /* V_tcpstat.tcps_cachedsendpipe++; */ 577133874Srwatson } 578122922Sandre if (hcml->rmx_recvpipe != 0) { 579122922Sandre if (hc_entry->rmx_recvpipe == 0) 580122922Sandre hc_entry->rmx_recvpipe = hcml->rmx_recvpipe; 581122922Sandre else 582122922Sandre hc_entry->rmx_recvpipe = 583122922Sandre (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2; 584181803Sbz /* V_tcpstat.tcps_cachedrecvpipe++; */ 585122922Sandre } 586122922Sandre 587122922Sandre TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 588122922Sandre TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); 589122922Sandre THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); 590122922Sandre} 591122922Sandre 592122922Sandre/* 593122922Sandre * Sysctl function: prints the list and values of all hostcache entries in 594122922Sandre * unsorted order. 595122922Sandre */ 596122922Sandrestatic int 597122922Sandresysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS) 598122922Sandre{ 599122922Sandre int bufsize; 600122922Sandre int linesize = 128; 601122922Sandre char *p, *buf; 602122922Sandre int len, i, error; 603122922Sandre struct hc_metrics *hc_entry; 604165118Sbz#ifdef INET6 605165118Sbz char ip6buf[INET6_ADDRSTRLEN]; 606165118Sbz#endif 607122922Sandre 608181803Sbz bufsize = linesize * (V_tcp_hostcache.cache_count + 1); 609122922Sandre 610122922Sandre p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO); 611122922Sandre 612122922Sandre len = snprintf(p, linesize, 613122922Sandre "\nIP address MTU SSTRESH RTT RTTVAR BANDWIDTH " 614122922Sandre " CWND SENDPIPE RECVPIPE HITS UPD EXP\n"); 615122922Sandre p += len; 616122922Sandre 617122922Sandre#define msec(u) (((u) + 500) / 1000) 618181803Sbz for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 619181803Sbz THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 620181803Sbz TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, 621122922Sandre rmx_q) { 622122922Sandre len = snprintf(p, linesize, 623122922Sandre "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu " 624122922Sandre "%4lu %4lu %4i\n", 625122922Sandre hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) : 626122922Sandre#ifdef INET6 627165118Sbz ip6_sprintf(ip6buf, &hc_entry->ip6), 628122922Sandre#else 629122922Sandre "IPv6?", 630122922Sandre#endif 631122922Sandre hc_entry->rmx_mtu, 632122922Sandre hc_entry->rmx_ssthresh, 633122922Sandre msec(hc_entry->rmx_rtt * 634122922Sandre (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 635122922Sandre msec(hc_entry->rmx_rttvar * 636122922Sandre (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), 637133477Sandre hc_entry->rmx_bandwidth * 8, 638122922Sandre hc_entry->rmx_cwnd, 639122922Sandre hc_entry->rmx_sendpipe, 640122922Sandre hc_entry->rmx_recvpipe, 641122922Sandre hc_entry->rmx_hits, 642122922Sandre hc_entry->rmx_updates, 643122922Sandre hc_entry->rmx_expire); 644122922Sandre p += len; 645122922Sandre } 646181803Sbz THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 647122922Sandre } 648122922Sandre#undef msec 649122922Sandre error = SYSCTL_OUT(req, buf, p - buf); 650122922Sandre free(buf, M_TEMP); 651122922Sandre return(error); 652122922Sandre} 653122922Sandre 654122922Sandre/* 655170030Srwatson * Expire and purge (old|all) entries in the tcp_hostcache. Runs 656170030Srwatson * periodically from the callout. 657122922Sandre */ 658122922Sandrestatic void 659122922Sandretcp_hc_purge(void *arg) 660122922Sandre{ 661128574Sandre struct hc_metrics *hc_entry, *hc_next; 662122922Sandre int all = (intptr_t)arg; 663122922Sandre int i; 664122922Sandre 665181803Sbz if (V_tcp_hostcache.purgeall) { 666122922Sandre all = 1; 667181803Sbz V_tcp_hostcache.purgeall = 0; 668122922Sandre } 669122922Sandre 670181803Sbz for (i = 0; i < V_tcp_hostcache.hashsize; i++) { 671181803Sbz THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 672181887Sjulian TAILQ_FOREACH_SAFE(hc_entry, 673181888Sjulian &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) { 674122922Sandre if (all || hc_entry->rmx_expire <= 0) { 675181803Sbz TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket, 676122922Sandre hc_entry, rmx_q); 677181803Sbz uma_zfree(V_tcp_hostcache.zone, hc_entry); 678181803Sbz V_tcp_hostcache.hashbase[i].hch_length--; 679181803Sbz V_tcp_hostcache.cache_count--; 680122922Sandre } else 681181803Sbz hc_entry->rmx_expire -= V_tcp_hostcache.prune; 682122922Sandre } 683181803Sbz THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); 684122922Sandre } 685181887Sjulian 686181887Sjulian callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, 687181887Sjulian tcp_hc_purge, arg); 688122922Sandre} 689