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>
| 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>
|
114
115static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
116static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
117static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
118static void tcp_hc_purge(void *);
119
120SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0,
121 "TCP Host cache");
122
123SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
124 &VNET_NAME(tcp_hostcache.cache_limit), 0,
125 "Overall entry limit for hostcache");
126
127SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
128 &VNET_NAME(tcp_hostcache.hashsize), 0,
129 "Size of TCP hostcache hashtable");
130
131SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit,
132 CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.bucket_limit), 0,
133 "Per-bucket hash limit for hostcache");
134
135SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD,
136 &VNET_NAME(tcp_hostcache.cache_count), 0,
137 "Current number of entries in hostcache");
138
139SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW,
140 &VNET_NAME(tcp_hostcache.expire), 0,
141 "Expire time of TCP hostcache entries");
142
143SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW,
144 &VNET_NAME(tcp_hostcache.prune), 0,
145 "Time between purge runs");
146
147SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW,
148 &VNET_NAME(tcp_hostcache.purgeall), 0,
149 "Expire all entires on next purge run");
150
151SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
152 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
153 sysctl_tcp_hc_list, "A", "List of all hostcache entries");
154
155
156static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
157
158#define HOSTCACHE_HASH(ip) \
159 (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \
160 V_tcp_hostcache.hashmask)
161
162/* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
163#define HOSTCACHE_HASH6(ip6) \
164 (((ip6)->s6_addr32[0] ^ \
165 (ip6)->s6_addr32[1] ^ \
166 (ip6)->s6_addr32[2] ^ \
167 (ip6)->s6_addr32[3]) & \
168 V_tcp_hostcache.hashmask)
169
170#define THC_LOCK(lp) mtx_lock(lp)
171#define THC_UNLOCK(lp) mtx_unlock(lp)
172
173void
174tcp_hc_init(void)
175{
176 int i;
177
178 /*
179 * Initialize hostcache structures.
180 */
181 V_tcp_hostcache.cache_count = 0;
182 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
183 V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
184 V_tcp_hostcache.cache_limit =
185 V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit;
186 V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
187 V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
188
189 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
190 &V_tcp_hostcache.hashsize);
191 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
192 &V_tcp_hostcache.cache_limit);
193 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
194 &V_tcp_hostcache.bucket_limit);
195 if (!powerof2(V_tcp_hostcache.hashsize)) {
196 printf("WARNING: hostcache hash size is not a power of 2.\n");
197 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
198 }
199 V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1;
200
201 /*
202 * Allocate the hash table.
203 */
204 V_tcp_hostcache.hashbase = (struct hc_head *)
205 malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head),
206 M_HOSTCACHE, M_WAITOK | M_ZERO);
207
208 /*
209 * Initialize the hash buckets.
210 */
211 for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
212 TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket);
213 V_tcp_hostcache.hashbase[i].hch_length = 0;
214 mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
215 NULL, MTX_DEF);
216 }
217
218 /*
219 * Allocate the hostcache entries.
220 */
221 V_tcp_hostcache.zone =
222 uma_zcreate("hostcache", sizeof(struct hc_metrics),
223 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
224 uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit);
225
226 /*
227 * Set up periodic cache cleanup.
228 */
229 callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE);
230 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
231 tcp_hc_purge, curvnet);
232}
233
234#ifdef VIMAGE
235void
236tcp_hc_destroy(void)
237{
238
239 /* XXX TODO walk the hashtable and free all entries */
240
241 callout_drain(&V_tcp_hc_callout);
242}
243#endif
244
245/*
246 * Internal function: look up an entry in the hostcache or return NULL.
247 *
248 * If an entry has been returned, the caller becomes responsible for
249 * unlocking the bucket row after he is done reading/modifying the entry.
250 */
251static struct hc_metrics *
252tcp_hc_lookup(struct in_conninfo *inc)
253{
254 int hash;
255 struct hc_head *hc_head;
256 struct hc_metrics *hc_entry;
257
258 KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer"));
259
260 /*
261 * Hash the foreign ip address.
262 */
263 if (inc->inc_flags & INC_ISIPV6)
264 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
265 else
266 hash = HOSTCACHE_HASH(&inc->inc_faddr);
267
268 hc_head = &V_tcp_hostcache.hashbase[hash];
269
270 /*
271 * Acquire lock for this bucket row; we release the lock if we don't
272 * find an entry, otherwise the caller has to unlock after he is
273 * done.
274 */
275 THC_LOCK(&hc_head->hch_mtx);
276
277 /*
278 * Iterate through entries in bucket row looking for a match.
279 */
280 TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) {
281 if (inc->inc_flags & INC_ISIPV6) {
282 if (memcmp(&inc->inc6_faddr, &hc_entry->ip6,
283 sizeof(inc->inc6_faddr)) == 0)
284 return hc_entry;
285 } else {
286 if (memcmp(&inc->inc_faddr, &hc_entry->ip4,
287 sizeof(inc->inc_faddr)) == 0)
288 return hc_entry;
289 }
290 }
291
292 /*
293 * We were unsuccessful and didn't find anything.
294 */
295 THC_UNLOCK(&hc_head->hch_mtx);
296 return NULL;
297}
298
299/*
300 * Internal function: insert an entry into the hostcache or return NULL if
301 * unable to allocate a new one.
302 *
303 * If an entry has been returned, the caller becomes responsible for
304 * unlocking the bucket row after he is done reading/modifying the entry.
305 */
306static struct hc_metrics *
307tcp_hc_insert(struct in_conninfo *inc)
308{
309 int hash;
310 struct hc_head *hc_head;
311 struct hc_metrics *hc_entry;
312
313 KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer"));
314
315 /*
316 * Hash the foreign ip address.
317 */
318 if (inc->inc_flags & INC_ISIPV6)
319 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
320 else
321 hash = HOSTCACHE_HASH(&inc->inc_faddr);
322
323 hc_head = &V_tcp_hostcache.hashbase[hash];
324
325 /*
326 * Acquire lock for this bucket row; we release the lock if we don't
327 * find an entry, otherwise the caller has to unlock after he is
328 * done.
329 */
330 THC_LOCK(&hc_head->hch_mtx);
331
332 /*
333 * If the bucket limit is reached, reuse the least-used element.
334 */
335 if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit ||
336 V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) {
337 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead);
338 /*
339 * At first we were dropping the last element, just to
340 * reacquire it in the next two lines again, which isn't very
341 * efficient. Instead just reuse the least used element.
342 * We may drop something that is still "in-use" but we can be
343 * "lossy".
344 * Just give up if this bucket row is empty and we don't have
345 * anything to replace.
346 */
347 if (hc_entry == NULL) {
348 THC_UNLOCK(&hc_head->hch_mtx);
349 return NULL;
350 }
351 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
352 V_tcp_hostcache.hashbase[hash].hch_length--;
353 V_tcp_hostcache.cache_count--;
354 TCPSTAT_INC(tcps_hc_bucketoverflow);
355#if 0
356 uma_zfree(V_tcp_hostcache.zone, hc_entry);
357#endif
358 } else {
359 /*
360 * Allocate a new entry, or balk if not possible.
361 */
362 hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT);
363 if (hc_entry == NULL) {
364 THC_UNLOCK(&hc_head->hch_mtx);
365 return NULL;
366 }
367 }
368
369 /*
370 * Initialize basic information of hostcache entry.
371 */
372 bzero(hc_entry, sizeof(*hc_entry));
373 if (inc->inc_flags & INC_ISIPV6)
374 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6));
375 else
376 hc_entry->ip4 = inc->inc_faddr;
377 hc_entry->rmx_head = hc_head;
378 hc_entry->rmx_expire = V_tcp_hostcache.expire;
379
380 /*
381 * Put it upfront.
382 */
383 TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
384 V_tcp_hostcache.hashbase[hash].hch_length++;
385 V_tcp_hostcache.cache_count++;
386 TCPSTAT_INC(tcps_hc_added);
387
388 return hc_entry;
389}
390
391/*
392 * External function: look up an entry in the hostcache and fill out the
393 * supplied TCP metrics structure. Fills in NULL when no entry was found or
394 * a value is not set.
395 */
396void
397tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
398{
399 struct hc_metrics *hc_entry;
400
401 /*
402 * Find the right bucket.
403 */
404 hc_entry = tcp_hc_lookup(inc);
405
406 /*
407 * If we don't have an existing object.
408 */
409 if (hc_entry == NULL) {
410 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
411 return;
412 }
413 hc_entry->rmx_hits++;
414 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
415
416 hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
417 hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
418 hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
419 hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
420 hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth;
421 hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
422 hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
423 hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
424
425 /*
426 * Unlock bucket row.
427 */
428 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
429}
430
431/*
432 * External function: look up an entry in the hostcache and return the
433 * discovered path MTU. Returns NULL if no entry is found or value is not
434 * set.
435 */
436u_long
437tcp_hc_getmtu(struct in_conninfo *inc)
438{
439 struct hc_metrics *hc_entry;
440 u_long mtu;
441
442 hc_entry = tcp_hc_lookup(inc);
443 if (hc_entry == NULL) {
444 return 0;
445 }
446 hc_entry->rmx_hits++;
447 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
448
449 mtu = hc_entry->rmx_mtu;
450 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
451 return mtu;
452}
453
454/*
455 * External function: update the MTU value of an entry in the hostcache.
456 * Creates a new entry if none was found.
457 */
458void
459tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
460{
461 struct hc_metrics *hc_entry;
462
463 /*
464 * Find the right bucket.
465 */
466 hc_entry = tcp_hc_lookup(inc);
467
468 /*
469 * If we don't have an existing object, try to insert a new one.
470 */
471 if (hc_entry == NULL) {
472 hc_entry = tcp_hc_insert(inc);
473 if (hc_entry == NULL)
474 return;
475 }
476 hc_entry->rmx_updates++;
477 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
478
479 hc_entry->rmx_mtu = mtu;
480
481 /*
482 * Put it upfront so we find it faster next time.
483 */
484 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
485 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
486
487 /*
488 * Unlock bucket row.
489 */
490 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
491}
492
493/*
494 * External function: update the TCP metrics of an entry in the hostcache.
495 * Creates a new entry if none was found.
496 */
497void
498tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
499{
500 struct hc_metrics *hc_entry;
501
502 hc_entry = tcp_hc_lookup(inc);
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 if (hcml->rmx_rtt != 0) {
512 if (hc_entry->rmx_rtt == 0)
513 hc_entry->rmx_rtt = hcml->rmx_rtt;
514 else
515 hc_entry->rmx_rtt =
516 (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
517 TCPSTAT_INC(tcps_cachedrtt);
518 }
519 if (hcml->rmx_rttvar != 0) {
520 if (hc_entry->rmx_rttvar == 0)
521 hc_entry->rmx_rttvar = hcml->rmx_rttvar;
522 else
523 hc_entry->rmx_rttvar =
524 (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
525 TCPSTAT_INC(tcps_cachedrttvar);
526 }
527 if (hcml->rmx_ssthresh != 0) {
528 if (hc_entry->rmx_ssthresh == 0)
529 hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
530 else
531 hc_entry->rmx_ssthresh =
532 (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
533 TCPSTAT_INC(tcps_cachedssthresh);
534 }
535 if (hcml->rmx_bandwidth != 0) {
536 if (hc_entry->rmx_bandwidth == 0)
537 hc_entry->rmx_bandwidth = hcml->rmx_bandwidth;
538 else
539 hc_entry->rmx_bandwidth =
540 (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2;
541 /* TCPSTAT_INC(tcps_cachedbandwidth); */
542 }
543 if (hcml->rmx_cwnd != 0) {
544 if (hc_entry->rmx_cwnd == 0)
545 hc_entry->rmx_cwnd = hcml->rmx_cwnd;
546 else
547 hc_entry->rmx_cwnd =
548 (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
549 /* TCPSTAT_INC(tcps_cachedcwnd); */
550 }
551 if (hcml->rmx_sendpipe != 0) {
552 if (hc_entry->rmx_sendpipe == 0)
553 hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
554 else
555 hc_entry->rmx_sendpipe =
556 (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
557 /* TCPSTAT_INC(tcps_cachedsendpipe); */
558 }
559 if (hcml->rmx_recvpipe != 0) {
560 if (hc_entry->rmx_recvpipe == 0)
561 hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
562 else
563 hc_entry->rmx_recvpipe =
564 (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
565 /* TCPSTAT_INC(tcps_cachedrecvpipe); */
566 }
567
568 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
569 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
570 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
571}
572
573/*
574 * Sysctl function: prints the list and values of all hostcache entries in
575 * unsorted order.
576 */
577static int
578sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
579{
580 int bufsize;
581 int linesize = 128;
582 char *p, *buf;
583 int len, i, error;
584 struct hc_metrics *hc_entry;
585#ifdef INET6
586 char ip6buf[INET6_ADDRSTRLEN];
587#endif
588
589 bufsize = linesize * (V_tcp_hostcache.cache_count + 1);
590
591 p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
592
593 len = snprintf(p, linesize,
594 "\nIP address MTU SSTRESH RTT RTTVAR BANDWIDTH "
595 " CWND SENDPIPE RECVPIPE HITS UPD EXP\n");
596 p += len;
597
598#define msec(u) (((u) + 500) / 1000)
599 for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
600 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
601 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket,
602 rmx_q) {
603 len = snprintf(p, linesize,
604 "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu "
605 "%4lu %4lu %4i\n",
606 hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
607#ifdef INET6
608 ip6_sprintf(ip6buf, &hc_entry->ip6),
609#else
610 "IPv6?",
611#endif
612 hc_entry->rmx_mtu,
613 hc_entry->rmx_ssthresh,
614 msec(hc_entry->rmx_rtt *
615 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
616 msec(hc_entry->rmx_rttvar *
617 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
618 hc_entry->rmx_bandwidth * 8,
619 hc_entry->rmx_cwnd,
620 hc_entry->rmx_sendpipe,
621 hc_entry->rmx_recvpipe,
622 hc_entry->rmx_hits,
623 hc_entry->rmx_updates,
624 hc_entry->rmx_expire);
625 p += len;
626 }
627 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
628 }
629#undef msec
630 error = SYSCTL_OUT(req, buf, p - buf);
631 free(buf, M_TEMP);
632 return(error);
633}
634
635/*
636 * Expire and purge (old|all) entries in the tcp_hostcache. Runs
637 * periodically from the callout.
638 */
639static void
640tcp_hc_purge(void *arg)
641{
642 CURVNET_SET((struct vnet *) arg);
643 struct hc_metrics *hc_entry, *hc_next;
644 int all = 0;
645 int i;
646
647 if (V_tcp_hostcache.purgeall) {
648 all = 1;
649 V_tcp_hostcache.purgeall = 0;
650 }
651
652 for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
653 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
654 TAILQ_FOREACH_SAFE(hc_entry,
655 &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) {
656 if (all || hc_entry->rmx_expire <= 0) {
657 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket,
658 hc_entry, rmx_q);
659 uma_zfree(V_tcp_hostcache.zone, hc_entry);
660 V_tcp_hostcache.hashbase[i].hch_length--;
661 V_tcp_hostcache.cache_count--;
662 } else
663 hc_entry->rmx_expire -= V_tcp_hostcache.prune;
664 }
665 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
666 }
667
668 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
669 tcp_hc_purge, arg);
670 CURVNET_RESTORE();
671}
| 114
115static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *);
116static struct hc_metrics *tcp_hc_insert(struct in_conninfo *);
117static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS);
118static void tcp_hc_purge(void *);
119
120SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0,
121 "TCP Host cache");
122
123SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
124 &VNET_NAME(tcp_hostcache.cache_limit), 0,
125 "Overall entry limit for hostcache");
126
127SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
128 &VNET_NAME(tcp_hostcache.hashsize), 0,
129 "Size of TCP hostcache hashtable");
130
131SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit,
132 CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.bucket_limit), 0,
133 "Per-bucket hash limit for hostcache");
134
135SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_RD,
136 &VNET_NAME(tcp_hostcache.cache_count), 0,
137 "Current number of entries in hostcache");
138
139SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_RW,
140 &VNET_NAME(tcp_hostcache.expire), 0,
141 "Expire time of TCP hostcache entries");
142
143SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_RW,
144 &VNET_NAME(tcp_hostcache.prune), 0,
145 "Time between purge runs");
146
147SYSCTL_VNET_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_RW,
148 &VNET_NAME(tcp_hostcache.purgeall), 0,
149 "Expire all entires on next purge run");
150
151SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list,
152 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0,
153 sysctl_tcp_hc_list, "A", "List of all hostcache entries");
154
155
156static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache");
157
158#define HOSTCACHE_HASH(ip) \
159 (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \
160 V_tcp_hostcache.hashmask)
161
162/* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */
163#define HOSTCACHE_HASH6(ip6) \
164 (((ip6)->s6_addr32[0] ^ \
165 (ip6)->s6_addr32[1] ^ \
166 (ip6)->s6_addr32[2] ^ \
167 (ip6)->s6_addr32[3]) & \
168 V_tcp_hostcache.hashmask)
169
170#define THC_LOCK(lp) mtx_lock(lp)
171#define THC_UNLOCK(lp) mtx_unlock(lp)
172
173void
174tcp_hc_init(void)
175{
176 int i;
177
178 /*
179 * Initialize hostcache structures.
180 */
181 V_tcp_hostcache.cache_count = 0;
182 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE;
183 V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT;
184 V_tcp_hostcache.cache_limit =
185 V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit;
186 V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE;
187 V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE;
188
189 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize",
190 &V_tcp_hostcache.hashsize);
191 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit",
192 &V_tcp_hostcache.cache_limit);
193 TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit",
194 &V_tcp_hostcache.bucket_limit);
195 if (!powerof2(V_tcp_hostcache.hashsize)) {
196 printf("WARNING: hostcache hash size is not a power of 2.\n");
197 V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */
198 }
199 V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1;
200
201 /*
202 * Allocate the hash table.
203 */
204 V_tcp_hostcache.hashbase = (struct hc_head *)
205 malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head),
206 M_HOSTCACHE, M_WAITOK | M_ZERO);
207
208 /*
209 * Initialize the hash buckets.
210 */
211 for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
212 TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket);
213 V_tcp_hostcache.hashbase[i].hch_length = 0;
214 mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry",
215 NULL, MTX_DEF);
216 }
217
218 /*
219 * Allocate the hostcache entries.
220 */
221 V_tcp_hostcache.zone =
222 uma_zcreate("hostcache", sizeof(struct hc_metrics),
223 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
224 uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit);
225
226 /*
227 * Set up periodic cache cleanup.
228 */
229 callout_init(&V_tcp_hc_callout, CALLOUT_MPSAFE);
230 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
231 tcp_hc_purge, curvnet);
232}
233
234#ifdef VIMAGE
235void
236tcp_hc_destroy(void)
237{
238
239 /* XXX TODO walk the hashtable and free all entries */
240
241 callout_drain(&V_tcp_hc_callout);
242}
243#endif
244
245/*
246 * Internal function: look up an entry in the hostcache or return NULL.
247 *
248 * If an entry has been returned, the caller becomes responsible for
249 * unlocking the bucket row after he is done reading/modifying the entry.
250 */
251static struct hc_metrics *
252tcp_hc_lookup(struct in_conninfo *inc)
253{
254 int hash;
255 struct hc_head *hc_head;
256 struct hc_metrics *hc_entry;
257
258 KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer"));
259
260 /*
261 * Hash the foreign ip address.
262 */
263 if (inc->inc_flags & INC_ISIPV6)
264 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
265 else
266 hash = HOSTCACHE_HASH(&inc->inc_faddr);
267
268 hc_head = &V_tcp_hostcache.hashbase[hash];
269
270 /*
271 * Acquire lock for this bucket row; we release the lock if we don't
272 * find an entry, otherwise the caller has to unlock after he is
273 * done.
274 */
275 THC_LOCK(&hc_head->hch_mtx);
276
277 /*
278 * Iterate through entries in bucket row looking for a match.
279 */
280 TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) {
281 if (inc->inc_flags & INC_ISIPV6) {
282 if (memcmp(&inc->inc6_faddr, &hc_entry->ip6,
283 sizeof(inc->inc6_faddr)) == 0)
284 return hc_entry;
285 } else {
286 if (memcmp(&inc->inc_faddr, &hc_entry->ip4,
287 sizeof(inc->inc_faddr)) == 0)
288 return hc_entry;
289 }
290 }
291
292 /*
293 * We were unsuccessful and didn't find anything.
294 */
295 THC_UNLOCK(&hc_head->hch_mtx);
296 return NULL;
297}
298
299/*
300 * Internal function: insert an entry into the hostcache or return NULL if
301 * unable to allocate a new one.
302 *
303 * If an entry has been returned, the caller becomes responsible for
304 * unlocking the bucket row after he is done reading/modifying the entry.
305 */
306static struct hc_metrics *
307tcp_hc_insert(struct in_conninfo *inc)
308{
309 int hash;
310 struct hc_head *hc_head;
311 struct hc_metrics *hc_entry;
312
313 KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer"));
314
315 /*
316 * Hash the foreign ip address.
317 */
318 if (inc->inc_flags & INC_ISIPV6)
319 hash = HOSTCACHE_HASH6(&inc->inc6_faddr);
320 else
321 hash = HOSTCACHE_HASH(&inc->inc_faddr);
322
323 hc_head = &V_tcp_hostcache.hashbase[hash];
324
325 /*
326 * Acquire lock for this bucket row; we release the lock if we don't
327 * find an entry, otherwise the caller has to unlock after he is
328 * done.
329 */
330 THC_LOCK(&hc_head->hch_mtx);
331
332 /*
333 * If the bucket limit is reached, reuse the least-used element.
334 */
335 if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit ||
336 V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) {
337 hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead);
338 /*
339 * At first we were dropping the last element, just to
340 * reacquire it in the next two lines again, which isn't very
341 * efficient. Instead just reuse the least used element.
342 * We may drop something that is still "in-use" but we can be
343 * "lossy".
344 * Just give up if this bucket row is empty and we don't have
345 * anything to replace.
346 */
347 if (hc_entry == NULL) {
348 THC_UNLOCK(&hc_head->hch_mtx);
349 return NULL;
350 }
351 TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q);
352 V_tcp_hostcache.hashbase[hash].hch_length--;
353 V_tcp_hostcache.cache_count--;
354 TCPSTAT_INC(tcps_hc_bucketoverflow);
355#if 0
356 uma_zfree(V_tcp_hostcache.zone, hc_entry);
357#endif
358 } else {
359 /*
360 * Allocate a new entry, or balk if not possible.
361 */
362 hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT);
363 if (hc_entry == NULL) {
364 THC_UNLOCK(&hc_head->hch_mtx);
365 return NULL;
366 }
367 }
368
369 /*
370 * Initialize basic information of hostcache entry.
371 */
372 bzero(hc_entry, sizeof(*hc_entry));
373 if (inc->inc_flags & INC_ISIPV6)
374 bcopy(&inc->inc6_faddr, &hc_entry->ip6, sizeof(hc_entry->ip6));
375 else
376 hc_entry->ip4 = inc->inc_faddr;
377 hc_entry->rmx_head = hc_head;
378 hc_entry->rmx_expire = V_tcp_hostcache.expire;
379
380 /*
381 * Put it upfront.
382 */
383 TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q);
384 V_tcp_hostcache.hashbase[hash].hch_length++;
385 V_tcp_hostcache.cache_count++;
386 TCPSTAT_INC(tcps_hc_added);
387
388 return hc_entry;
389}
390
391/*
392 * External function: look up an entry in the hostcache and fill out the
393 * supplied TCP metrics structure. Fills in NULL when no entry was found or
394 * a value is not set.
395 */
396void
397tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite)
398{
399 struct hc_metrics *hc_entry;
400
401 /*
402 * Find the right bucket.
403 */
404 hc_entry = tcp_hc_lookup(inc);
405
406 /*
407 * If we don't have an existing object.
408 */
409 if (hc_entry == NULL) {
410 bzero(hc_metrics_lite, sizeof(*hc_metrics_lite));
411 return;
412 }
413 hc_entry->rmx_hits++;
414 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
415
416 hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu;
417 hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh;
418 hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt;
419 hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar;
420 hc_metrics_lite->rmx_bandwidth = hc_entry->rmx_bandwidth;
421 hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd;
422 hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe;
423 hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe;
424
425 /*
426 * Unlock bucket row.
427 */
428 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
429}
430
431/*
432 * External function: look up an entry in the hostcache and return the
433 * discovered path MTU. Returns NULL if no entry is found or value is not
434 * set.
435 */
436u_long
437tcp_hc_getmtu(struct in_conninfo *inc)
438{
439 struct hc_metrics *hc_entry;
440 u_long mtu;
441
442 hc_entry = tcp_hc_lookup(inc);
443 if (hc_entry == NULL) {
444 return 0;
445 }
446 hc_entry->rmx_hits++;
447 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
448
449 mtu = hc_entry->rmx_mtu;
450 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
451 return mtu;
452}
453
454/*
455 * External function: update the MTU value of an entry in the hostcache.
456 * Creates a new entry if none was found.
457 */
458void
459tcp_hc_updatemtu(struct in_conninfo *inc, u_long mtu)
460{
461 struct hc_metrics *hc_entry;
462
463 /*
464 * Find the right bucket.
465 */
466 hc_entry = tcp_hc_lookup(inc);
467
468 /*
469 * If we don't have an existing object, try to insert a new one.
470 */
471 if (hc_entry == NULL) {
472 hc_entry = tcp_hc_insert(inc);
473 if (hc_entry == NULL)
474 return;
475 }
476 hc_entry->rmx_updates++;
477 hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */
478
479 hc_entry->rmx_mtu = mtu;
480
481 /*
482 * Put it upfront so we find it faster next time.
483 */
484 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
485 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
486
487 /*
488 * Unlock bucket row.
489 */
490 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
491}
492
493/*
494 * External function: update the TCP metrics of an entry in the hostcache.
495 * Creates a new entry if none was found.
496 */
497void
498tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml)
499{
500 struct hc_metrics *hc_entry;
501
502 hc_entry = tcp_hc_lookup(inc);
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 if (hcml->rmx_rtt != 0) {
512 if (hc_entry->rmx_rtt == 0)
513 hc_entry->rmx_rtt = hcml->rmx_rtt;
514 else
515 hc_entry->rmx_rtt =
516 (hc_entry->rmx_rtt + hcml->rmx_rtt) / 2;
517 TCPSTAT_INC(tcps_cachedrtt);
518 }
519 if (hcml->rmx_rttvar != 0) {
520 if (hc_entry->rmx_rttvar == 0)
521 hc_entry->rmx_rttvar = hcml->rmx_rttvar;
522 else
523 hc_entry->rmx_rttvar =
524 (hc_entry->rmx_rttvar + hcml->rmx_rttvar) / 2;
525 TCPSTAT_INC(tcps_cachedrttvar);
526 }
527 if (hcml->rmx_ssthresh != 0) {
528 if (hc_entry->rmx_ssthresh == 0)
529 hc_entry->rmx_ssthresh = hcml->rmx_ssthresh;
530 else
531 hc_entry->rmx_ssthresh =
532 (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2;
533 TCPSTAT_INC(tcps_cachedssthresh);
534 }
535 if (hcml->rmx_bandwidth != 0) {
536 if (hc_entry->rmx_bandwidth == 0)
537 hc_entry->rmx_bandwidth = hcml->rmx_bandwidth;
538 else
539 hc_entry->rmx_bandwidth =
540 (hc_entry->rmx_bandwidth + hcml->rmx_bandwidth) / 2;
541 /* TCPSTAT_INC(tcps_cachedbandwidth); */
542 }
543 if (hcml->rmx_cwnd != 0) {
544 if (hc_entry->rmx_cwnd == 0)
545 hc_entry->rmx_cwnd = hcml->rmx_cwnd;
546 else
547 hc_entry->rmx_cwnd =
548 (hc_entry->rmx_cwnd + hcml->rmx_cwnd) / 2;
549 /* TCPSTAT_INC(tcps_cachedcwnd); */
550 }
551 if (hcml->rmx_sendpipe != 0) {
552 if (hc_entry->rmx_sendpipe == 0)
553 hc_entry->rmx_sendpipe = hcml->rmx_sendpipe;
554 else
555 hc_entry->rmx_sendpipe =
556 (hc_entry->rmx_sendpipe + hcml->rmx_sendpipe) /2;
557 /* TCPSTAT_INC(tcps_cachedsendpipe); */
558 }
559 if (hcml->rmx_recvpipe != 0) {
560 if (hc_entry->rmx_recvpipe == 0)
561 hc_entry->rmx_recvpipe = hcml->rmx_recvpipe;
562 else
563 hc_entry->rmx_recvpipe =
564 (hc_entry->rmx_recvpipe + hcml->rmx_recvpipe) /2;
565 /* TCPSTAT_INC(tcps_cachedrecvpipe); */
566 }
567
568 TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
569 TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q);
570 THC_UNLOCK(&hc_entry->rmx_head->hch_mtx);
571}
572
573/*
574 * Sysctl function: prints the list and values of all hostcache entries in
575 * unsorted order.
576 */
577static int
578sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS)
579{
580 int bufsize;
581 int linesize = 128;
582 char *p, *buf;
583 int len, i, error;
584 struct hc_metrics *hc_entry;
585#ifdef INET6
586 char ip6buf[INET6_ADDRSTRLEN];
587#endif
588
589 bufsize = linesize * (V_tcp_hostcache.cache_count + 1);
590
591 p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
592
593 len = snprintf(p, linesize,
594 "\nIP address MTU SSTRESH RTT RTTVAR BANDWIDTH "
595 " CWND SENDPIPE RECVPIPE HITS UPD EXP\n");
596 p += len;
597
598#define msec(u) (((u) + 500) / 1000)
599 for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
600 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
601 TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket,
602 rmx_q) {
603 len = snprintf(p, linesize,
604 "%-15s %5lu %8lu %6lums %6lums %9lu %8lu %8lu %8lu "
605 "%4lu %4lu %4i\n",
606 hc_entry->ip4.s_addr ? inet_ntoa(hc_entry->ip4) :
607#ifdef INET6
608 ip6_sprintf(ip6buf, &hc_entry->ip6),
609#else
610 "IPv6?",
611#endif
612 hc_entry->rmx_mtu,
613 hc_entry->rmx_ssthresh,
614 msec(hc_entry->rmx_rtt *
615 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
616 msec(hc_entry->rmx_rttvar *
617 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))),
618 hc_entry->rmx_bandwidth * 8,
619 hc_entry->rmx_cwnd,
620 hc_entry->rmx_sendpipe,
621 hc_entry->rmx_recvpipe,
622 hc_entry->rmx_hits,
623 hc_entry->rmx_updates,
624 hc_entry->rmx_expire);
625 p += len;
626 }
627 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
628 }
629#undef msec
630 error = SYSCTL_OUT(req, buf, p - buf);
631 free(buf, M_TEMP);
632 return(error);
633}
634
635/*
636 * Expire and purge (old|all) entries in the tcp_hostcache. Runs
637 * periodically from the callout.
638 */
639static void
640tcp_hc_purge(void *arg)
641{
642 CURVNET_SET((struct vnet *) arg);
643 struct hc_metrics *hc_entry, *hc_next;
644 int all = 0;
645 int i;
646
647 if (V_tcp_hostcache.purgeall) {
648 all = 1;
649 V_tcp_hostcache.purgeall = 0;
650 }
651
652 for (i = 0; i < V_tcp_hostcache.hashsize; i++) {
653 THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
654 TAILQ_FOREACH_SAFE(hc_entry,
655 &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) {
656 if (all || hc_entry->rmx_expire <= 0) {
657 TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket,
658 hc_entry, rmx_q);
659 uma_zfree(V_tcp_hostcache.zone, hc_entry);
660 V_tcp_hostcache.hashbase[i].hch_length--;
661 V_tcp_hostcache.cache_count--;
662 } else
663 hc_entry->rmx_expire -= V_tcp_hostcache.prune;
664 }
665 THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx);
666 }
667
668 callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz,
669 tcp_hc_purge, arg);
670 CURVNET_RESTORE();
671}
|